diff -urN linux-2.6.30.orig/Documentation/Changes linux-2.6.30/Documentation/Changes --- linux-2.6.30.orig/Documentation/Changes 2009-03-24 00:12:14.000000000 +0100 +++ linux-2.6.30/Documentation/Changes 2009-06-22 16:08:11.000000000 +0200 @@ -36,6 +36,7 @@ o e2fsprogs 1.41.4 # e2fsck -V o jfsutils 1.1.3 # fsck.jfs -V o reiserfsprogs 3.6.3 # reiserfsck -V 2>&1|grep reiserfsprogs +o reiser4progs 1.0.0 # fsck.reiser4 -V o xfsprogs 2.6.0 # xfs_db -V o squashfs-tools 4.0 # mksquashfs -version o btrfs-progs 0.18 # btrfsck @@ -147,6 +148,13 @@ versions of mkreiserfs, resize_reiserfs, debugreiserfs and reiserfsck. These utils work on both i386 and alpha platforms. +Reiser4progs +------------ + +The reiser4progs package contains utilities for the reiser4 file system. +Detailed instructions are provided in the README file located at: +. + Xfsprogs -------- @@ -325,6 +333,10 @@ ------------- o +Reiser4progs +------------ +o + Xfsprogs -------- o diff -urN linux-2.6.30.orig/Documentation/filesystems/reiser4.txt linux-2.6.30/Documentation/filesystems/reiser4.txt --- linux-2.6.30.orig/Documentation/filesystems/reiser4.txt 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/Documentation/filesystems/reiser4.txt 2009-06-22 16:08:11.000000000 +0200 @@ -0,0 +1,75 @@ +Reiser4 filesystem +================== +Reiser4 is a file system based on dancing tree algorithms, and is +described at http://www.namesys.com + + +References +========== +web page http://namesys.com/v4/v4.html +source code ftp://ftp.namesys.com/pub/reiser4-for-2.6/ +userland tools ftp://ftp.namesys.com/pub/reiser4progs/ +install page http://www.namesys.com/install_v4.html + +Compile options +=============== +Enable reiser4 debug mode + This checks everything imaginable while reiser4 + runs + +Mount options +============= +tmgr.atom_max_size=N + Atoms containing more than N blocks will be forced to commit. + N is decimal. + Default is nr_free_pagecache_pages() / 2 at mount time. + +tmgr.atom_max_age=N + Atoms older than N seconds will be forced to commit. N is decimal. + Default is 600. + +tmgr.atom_max_flushers=N + Limit of concurrent flushers for one atom. 0 means no limit. + Default is 0. + +tree.cbk_cache.nr_slots=N + Number of slots in the cbk cache. + +flush.relocate_threshold=N + If flush finds more than N adjacent dirty leaf-level blocks it + will force them to be relocated. + Default is 64. + +flush.relocate_distance=N + If flush finds can find a block allocation closer than at most + N from the preceder it will relocate to that position. + Default is 64. + +flush.scan_maxnodes=N + The maximum number of nodes to scan left on a level during + flush. + Default is 10000. + +optimal_io_size=N + Preferred IO size. This value is used to set st_blksize of + struct stat. + Default is 65536. + +bsdgroups + Turn on BSD-style gid assignment. + +32bittimes + By default file in reiser4 have 64 bit timestamps. Files + created when filesystem is mounted with 32bittimes mount + option will get 32 bit timestamps. + +mtflush + Turn off concurrent flushing. + +nopseudo + Disable pseudo files support. See + http://namesys.com/v4/pseudo.html for more about pseudo files. + +dont_load_bitmap + Don't load all bitmap blocks at mount time, it is useful for + machines with tiny RAM and large disks. diff -urN linux-2.6.30.orig/fs/fs-writeback.c linux-2.6.30/fs/fs-writeback.c --- linux-2.6.30.orig/fs/fs-writeback.c 2009-06-23 00:20:39.000000000 +0200 +++ linux-2.6.30/fs/fs-writeback.c 2009-06-22 16:08:13.000000000 +0200 @@ -593,7 +593,10 @@ static void sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc) { - generic_sync_sb_inodes(sb, wbc); + if (sb->s_op->sync_inodes) + sb->s_op->sync_inodes(sb, wbc); + else + generic_sync_sb_inodes(sb, wbc); } /* diff -urN linux-2.6.30.orig/fs/Kconfig linux-2.6.30/fs/Kconfig --- linux-2.6.30.orig/fs/Kconfig 2009-06-23 00:20:39.000000000 +0200 +++ linux-2.6.30/fs/Kconfig 2009-06-22 16:08:13.000000000 +0200 @@ -27,6 +27,7 @@ default y if EXT4_FS=y && EXT4_FS_XATTR default m if EXT2_FS_XATTR || EXT3_FS_XATTR || EXT4_FS_XATTR +source "fs/reiser4/Kconfig" source "fs/reiserfs/Kconfig" source "fs/jfs/Kconfig" diff -urN linux-2.6.30.orig/fs/Makefile linux-2.6.30/fs/Makefile --- linux-2.6.30.orig/fs/Makefile 2009-06-23 00:20:39.000000000 +0200 +++ linux-2.6.30/fs/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -65,6 +65,7 @@ # Do not add any filesystems before this line obj-$(CONFIG_FSCACHE) += fscache/ obj-$(CONFIG_REISERFS_FS) += reiserfs/ +obj-$(CONFIG_REISER4_FS) += reiser4/ obj-$(CONFIG_EXT3_FS) += ext3/ # Before ext2 so root fs can be ext3 obj-$(CONFIG_EXT2_FS) += ext2/ # We place ext4 after ext2 so plain ext2 root fs's are mounted using ext2 diff -urN linux-2.6.30.orig/fs/reiser4/as_ops.c linux-2.6.30/fs/reiser4/as_ops.c --- linux-2.6.30.orig/fs/reiser4/as_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/as_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,337 @@ +/* Copyright 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Interface to VFS. Reiser4 address_space_operations are defined here. */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "plugin/file/file.h" +#include "plugin/security/perm.h" +#include "plugin/disk_format/disk_format.h" +#include "plugin/plugin.h" +#include "plugin/plugin_set.h" +#include "plugin/object.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree.h" +#include "vfs_ops.h" +#include "inode.h" +#include "page_cache.h" +#include "ktxnmgrd.h" +#include "super.h" +#include "reiser4.h" +#include "entd.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* address space operations */ + +/** + * reiser4_set_page_dirty - set dirty bit, tag in page tree, dirty accounting + * @page: page to be dirtied + * + * Operation of struct address_space_operations. This implementation is used by + * unix and cryptcompress file plugins. + * + * This is called when reiser4 page gets dirtied outside of reiser4, for + * example, when dirty bit is moved from pte to physical page. + * + * Tags page in the mapping's page tree with special tag so that it is possible + * to do all the reiser4 specific work wrt dirty pages (jnode creation, + * capturing by an atom) later because it can not be done in the contexts where + * set_page_dirty is called. + */ +int reiser4_set_page_dirty(struct page *page) +{ + /* this page can be unformatted only */ + assert("vs-1734", (page->mapping && + page->mapping->host && + reiser4_get_super_fake(page->mapping->host->i_sb) != + page->mapping->host && + reiser4_get_cc_fake(page->mapping->host->i_sb) != + page->mapping->host && + reiser4_get_bitmap_fake(page->mapping->host->i_sb) != + page->mapping->host)); + return __set_page_dirty_nobuffers(page); +} + +/* ->invalidatepage method for reiser4 */ + +/* + * this is called for each truncated page from + * truncate_inode_pages()->truncate_{complete,partial}_page(). + * + * At the moment of call, page is under lock, and outstanding io (if any) has + * completed. + */ + +/** + * reiser4_invalidatepage + * @page: page to invalidate + * @offset: starting offset for partial invalidation + * + */ +void reiser4_invalidatepage(struct page *page, unsigned long offset) +{ + int ret = 0; + reiser4_context *ctx; + struct inode *inode; + jnode *node; + + /* + * This is called to truncate file's page. + * + * Originally, reiser4 implemented truncate in a standard way + * (vmtruncate() calls ->invalidatepage() on all truncated pages + * first, then file system ->truncate() call-back is invoked). + * + * This lead to the problem when ->invalidatepage() was called on a + * page with jnode that was captured into atom in ASTAGE_PRE_COMMIT + * process. That is, truncate was bypassing transactions. To avoid + * this, try_capture_page_to_invalidate() call was added here. + * + * After many troubles with vmtruncate() based truncate (including + * races with flush, tail conversion, etc.) it was re-written in the + * top-to-bottom style: items are killed in reiser4_cut_tree_object() + * and pages belonging to extent are invalidated in kill_hook_extent(). + * So probably now additional call to capture is not needed here. + */ + + assert("nikita-3137", PageLocked(page)); + assert("nikita-3138", !PageWriteback(page)); + inode = page->mapping->host; + + /* + * ->invalidatepage() should only be called for the unformatted + * jnodes. Destruction of all other types of jnodes is performed + * separately. But, during some corner cases (like handling errors + * during mount) it is simpler to let ->invalidatepage to be called on + * them. Check for this, and do nothing. + */ + if (reiser4_get_super_fake(inode->i_sb) == inode) + return; + if (reiser4_get_cc_fake(inode->i_sb) == inode) + return; + if (reiser4_get_bitmap_fake(inode->i_sb) == inode) + return; + assert("vs-1426", PagePrivate(page)); + assert("vs-1427", + page->mapping == jnode_get_mapping(jnode_by_page(page))); + assert("", jprivate(page) != NULL); + assert("", ergo(inode_file_plugin(inode) != + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID), + offset == 0)); + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return; + + node = jprivate(page); + spin_lock_jnode(node); + if (!(node->state & ((1 << JNODE_DIRTY) | (1 << JNODE_FLUSH_QUEUED) | + (1 << JNODE_WRITEBACK) | (1 << JNODE_OVRWR)))) { + /* there is not need to capture */ + jref(node); + JF_SET(node, JNODE_HEARD_BANSHEE); + page_clear_jnode(page, node); + reiser4_uncapture_jnode(node); + unhash_unformatted_jnode(node); + jput(node); + reiser4_exit_context(ctx); + return; + } + spin_unlock_jnode(node); + + /* capture page being truncated. */ + ret = try_capture_page_to_invalidate(page); + if (ret != 0) + warning("nikita-3141", "Cannot capture: %i", ret); + + if (offset == 0) { + /* remove jnode from transaction and detach it from page. */ + jref(node); + JF_SET(node, JNODE_HEARD_BANSHEE); + /* page cannot be detached from jnode concurrently, because it + * is locked */ + reiser4_uncapture_page(page); + + /* this detaches page from jnode, so that jdelete will not try + * to lock page which is already locked */ + spin_lock_jnode(node); + page_clear_jnode(page, node); + spin_unlock_jnode(node); + unhash_unformatted_jnode(node); + + jput(node); + } + + reiser4_exit_context(ctx); +} + +/* help function called from reiser4_releasepage(). It returns true if jnode + * can be detached from its page and page released. */ +int jnode_is_releasable(jnode * node/* node to check */) +{ + assert("nikita-2781", node != NULL); + assert_spin_locked(&(node->guard)); + assert_spin_locked(&(node->load)); + + /* is some thread is currently using jnode page, later cannot be + * detached */ + if (atomic_read(&node->d_count) != 0) + return 0; + + assert("vs-1214", !jnode_is_loaded(node)); + + /* + * can only release page if real block number is assigned to it. Simple + * check for ->atom wouldn't do, because it is possible for node to be + * clean, not it atom yet, and still having fake block number. For + * example, node just created in jinit_new(). + */ + if (reiser4_blocknr_is_fake(jnode_get_block(node))) + return 0; + + /* + * pages prepared for write can not be released anyway, so avoid + * detaching jnode from the page + */ + if (JF_ISSET(node, JNODE_WRITE_PREPARED)) + return 0; + + /* + * dirty jnode cannot be released. It can however be submitted to disk + * as part of early flushing, but only after getting flush-prepped. + */ + if (JF_ISSET(node, JNODE_DIRTY)) + return 0; + + /* overwrite set is only written by log writer. */ + if (JF_ISSET(node, JNODE_OVRWR)) + return 0; + + /* jnode is already under writeback */ + if (JF_ISSET(node, JNODE_WRITEBACK)) + return 0; + + /* don't flush bitmaps or journal records */ + if (!jnode_is_znode(node) && !jnode_is_unformatted(node)) + return 0; + + return 1; +} + +/* + * ->releasepage method for reiser4 + * + * This is called by VM scanner when it comes across clean page. What we have + * to do here is to check whether page can really be released (freed that is) + * and if so, detach jnode from it and remove page from the page cache. + * + * Check for releasability is done by releasable() function. + */ +int reiser4_releasepage(struct page *page, gfp_t gfp UNUSED_ARG) +{ + jnode *node; + + assert("nikita-2257", PagePrivate(page)); + assert("nikita-2259", PageLocked(page)); + assert("nikita-2892", !PageWriteback(page)); + assert("nikita-3019", reiser4_schedulable()); + + /* NOTE-NIKITA: this can be called in the context of reiser4 call. It + is not clear what to do in this case. A lot of deadlocks seems be + possible. */ + + node = jnode_by_page(page); + assert("nikita-2258", node != NULL); + assert("reiser4-4", page->mapping != NULL); + assert("reiser4-5", page->mapping->host != NULL); + + if (PageDirty(page)) + return 0; + + /* extra page reference is used by reiser4 to protect + * jnode<->page link from this ->releasepage(). */ + if (page_count(page) > 3) + return 0; + + /* releasable() needs jnode lock, because it looks at the jnode fields + * and we need jload_lock here to avoid races with jload(). */ + spin_lock_jnode(node); + spin_lock(&(node->load)); + if (jnode_is_releasable(node)) { + struct address_space *mapping; + + mapping = page->mapping; + jref(node); + /* there is no need to synchronize against + * jnode_extent_write() here, because pages seen by + * jnode_extent_write() are !releasable(). */ + page_clear_jnode(page, node); + spin_unlock(&(node->load)); + spin_unlock_jnode(node); + + /* we are under memory pressure so release jnode also. */ + jput(node); + + return 1; + } else { + spin_unlock(&(node->load)); + spin_unlock_jnode(node); + assert("nikita-3020", reiser4_schedulable()); + return 0; + } +} + +int reiser4_readpage(struct file *file, struct page *page) +{ + assert("edward-1533", PageLocked(page)); + assert("edward-1534", !PageUptodate(page)); + assert("edward-1535", page->mapping && page->mapping->host); + + return inode_file_plugin(page->mapping->host)->readpage(file, page); +} + +int reiser4_readpages(struct file *file, struct address_space *mapping, + struct list_head *pages, unsigned nr_pages) +{ + return inode_file_plugin(mapping->host)->readpages(file, mapping, + pages, nr_pages); +} + +int reiser4_writepages(struct address_space *mapping, + struct writeback_control *wbc) +{ + return inode_file_plugin(mapping->host)->writepages(mapping, wbc); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/block_alloc.c linux-2.6.30/fs/reiser4/block_alloc.c --- linux-2.6.30.orig/fs/reiser4/block_alloc.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/block_alloc.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1142 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by +reiser4/README */ + +#include "debug.h" +#include "dformat.h" +#include "plugin/plugin.h" +#include "txnmgr.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree.h" +#include "super.h" + +#include /* for __u?? */ +#include /* for struct super_block */ +#include + +/* THE REISER4 DISK SPACE RESERVATION SCHEME. */ + +/* We need to be able to reserve enough disk space to ensure that an atomic + operation will have enough disk space to flush (see flush.c and + http://namesys.com/v4/v4.html) and commit it once it is started. + + In our design a call for reserving disk space may fail but not an actual + block allocation. + + All free blocks, already allocated blocks, and all kinds of reserved blocks + are counted in different per-fs block counters. + + A reiser4 super block's set of block counters currently is: + + free -- free blocks, + used -- already allocated blocks, + + grabbed -- initially reserved for performing an fs operation, those blocks + are taken from free blocks, then grabbed disk space leaks from grabbed + blocks counter to other counters like "fake allocated", "flush + reserved", "used", the rest of not used grabbed space is returned to + free space at the end of fs operation; + + fake allocated -- counts all nodes without real disk block numbers assigned, + we have separate accounting for formatted and unformatted + nodes (for easier debugging); + + flush reserved -- disk space needed for flushing and committing an atom. + Each dirty already allocated block could be written as a + part of atom's overwrite set or as a part of atom's + relocate set. In both case one additional block is needed, + it is used as a wandered block if we do overwrite or as a + new location for a relocated block. + + In addition, blocks in some states are counted on per-thread and per-atom + basis. A reiser4 context has a counter of blocks grabbed by this transaction + and the sb's grabbed blocks counter is a sum of grabbed blocks counter values + of each reiser4 context. Each reiser4 atom has a counter of "flush reserved" + blocks, which are reserved for flush processing and atom commit. */ + +/* AN EXAMPLE: suppose we insert new item to the reiser4 tree. We estimate + number of blocks to grab for most expensive case of balancing when the leaf + node we insert new item to gets split and new leaf node is allocated. + + So, we need to grab blocks for + + 1) one block for possible dirtying the node we insert an item to. That block + would be used for node relocation at flush time or for allocating of a + wandered one, it depends what will be a result (what set, relocate or + overwrite the node gets assigned to) of the node processing by the flush + algorithm. + + 2) one block for either allocating a new node, or dirtying of right or left + clean neighbor, only one case may happen. + + VS-FIXME-HANS: why can only one case happen? I would expect to see dirtying + of left neighbor, right neighbor, current node, and creation of new node. + Have I forgotten something? email me. + + These grabbed blocks are counted in both reiser4 context "grabbed blocks" + counter and in the fs-wide one (both ctx->grabbed_blocks and + sbinfo->blocks_grabbed get incremented by 2), sb's free blocks counter is + decremented by 2. + + Suppose both two blocks were spent for dirtying of an already allocated clean + node (one block went from "grabbed" to "flush reserved") and for new block + allocating (one block went from "grabbed" to "fake allocated formatted"). + + Inserting of a child pointer to the parent node caused parent node to be + split, the balancing code takes care about this grabbing necessary space + immediately by calling reiser4_grab with BA_RESERVED flag set which means + "can use the 5% reserved disk space". + + At this moment insertion completes and grabbed blocks (if they were not used) + should be returned to the free space counter. + + However the atom life-cycle is not completed. The atom had one "flush + reserved" block added by our insertion and the new fake allocated node is + counted as a "fake allocated formatted" one. The atom has to be fully + processed by flush before commit. Suppose that the flush moved the first, + already allocated node to the atom's overwrite list, the new fake allocated + node, obviously, went into the atom relocate set. The reiser4 flush + allocates the new node using one unit from "fake allocated formatted" + counter, the log writer uses one from "flush reserved" for wandered block + allocation. + + And, it is not the end. When the wandered block is deallocated after the + atom gets fully played (see wander.c for term description), the disk space + occupied for it is returned to free blocks. */ + +/* BLOCK NUMBERS */ + +/* Any reiser4 node has a block number assigned to it. We use these numbers for + indexing in hash tables, so if a block has not yet been assigned a location + on disk we need to give it a temporary fake block number. + + Current implementation of reiser4 uses 64-bit integers for block numbers. We + use highest bit in 64-bit block number to distinguish fake and real block + numbers. So, only 63 bits may be used to addressing of real device + blocks. That "fake" block numbers space is divided into subspaces of fake + block numbers for data blocks and for shadow (working) bitmap blocks. + + Fake block numbers for data blocks are generated by a cyclic counter, which + gets incremented after each real block allocation. We assume that it is + impossible to overload this counter during one transaction life. */ + +/* Initialize a blocknr hint. */ +void reiser4_blocknr_hint_init(reiser4_blocknr_hint * hint) +{ + memset(hint, 0, sizeof(reiser4_blocknr_hint)); +} + +/* Release any resources of a blocknr hint. */ +void reiser4_blocknr_hint_done(reiser4_blocknr_hint * hint UNUSED_ARG) +{ +/* No resources should be freed in current blocknr_hint implementation. */ +} + +/* see above for explanation of fake block number. */ +/* Audited by: green(2002.06.11) */ +int reiser4_blocknr_is_fake(const reiser4_block_nr * da) +{ + /* The reason for not simply returning result of '&' operation is that + while return value is (possibly 32bit) int, the reiser4_block_nr is + at least 64 bits long, and high bit (which is the only possible + non zero bit after the masking) would be stripped off */ + return (*da & REISER4_FAKE_BLOCKNR_BIT_MASK) ? 1 : 0; +} + +/* Static functions for / block counters + arithmetic. Mostly, they are isolated to not to code same assertions in + several places. */ +static void sub_from_ctx_grabbed(reiser4_context * ctx, __u64 count) +{ + BUG_ON(ctx->grabbed_blocks < count); + assert("zam-527", ctx->grabbed_blocks >= count); + ctx->grabbed_blocks -= count; +} + +static void add_to_ctx_grabbed(reiser4_context * ctx, __u64 count) +{ + ctx->grabbed_blocks += count; +} + +static void sub_from_sb_grabbed(reiser4_super_info_data * sbinfo, __u64 count) +{ + assert("zam-525", sbinfo->blocks_grabbed >= count); + sbinfo->blocks_grabbed -= count; +} + +/* Decrease the counter of block reserved for flush in super block. */ +static void +sub_from_sb_flush_reserved(reiser4_super_info_data * sbinfo, __u64 count) +{ + assert("vpf-291", sbinfo->blocks_flush_reserved >= count); + sbinfo->blocks_flush_reserved -= count; +} + +static void +sub_from_sb_fake_allocated(reiser4_super_info_data * sbinfo, __u64 count, + reiser4_ba_flags_t flags) +{ + if (flags & BA_FORMATTED) { + assert("zam-806", sbinfo->blocks_fake_allocated >= count); + sbinfo->blocks_fake_allocated -= count; + } else { + assert("zam-528", + sbinfo->blocks_fake_allocated_unformatted >= count); + sbinfo->blocks_fake_allocated_unformatted -= count; + } +} + +static void sub_from_sb_used(reiser4_super_info_data * sbinfo, __u64 count) +{ + assert("zam-530", + sbinfo->blocks_used >= count + sbinfo->min_blocks_used); + sbinfo->blocks_used -= count; +} + +static void +sub_from_cluster_reserved(reiser4_super_info_data * sbinfo, __u64 count) +{ + assert("edward-501", sbinfo->blocks_clustered >= count); + sbinfo->blocks_clustered -= count; +} + +/* Increase the counter of block reserved for flush in atom. */ +static void add_to_atom_flush_reserved_nolock(txn_atom * atom, __u32 count) +{ + assert("zam-772", atom != NULL); + assert_spin_locked(&(atom->alock)); + atom->flush_reserved += count; +} + +/* Decrease the counter of block reserved for flush in atom. */ +static void sub_from_atom_flush_reserved_nolock(txn_atom * atom, __u32 count) +{ + assert("zam-774", atom != NULL); + assert_spin_locked(&(atom->alock)); + assert("nikita-2790", atom->flush_reserved >= count); + atom->flush_reserved -= count; +} + +/* super block has 6 counters: free, used, grabbed, fake allocated + (formatted and unformatted) and flush reserved. Their sum must be + number of blocks on a device. This function checks this */ +int reiser4_check_block_counters(const struct super_block *super) +{ + __u64 sum; + + sum = reiser4_grabbed_blocks(super) + reiser4_free_blocks(super) + + reiser4_data_blocks(super) + reiser4_fake_allocated(super) + + reiser4_fake_allocated_unformatted(super) + reiser4_flush_reserved(super) + + reiser4_clustered_blocks(super); + if (reiser4_block_count(super) != sum) { + printk("super block counters: " + "used %llu, free %llu, " + "grabbed %llu, fake allocated (formatetd %llu, unformatted %llu), " + "reserved %llu, clustered %llu, sum %llu, must be (block count) %llu\n", + (unsigned long long)reiser4_data_blocks(super), + (unsigned long long)reiser4_free_blocks(super), + (unsigned long long)reiser4_grabbed_blocks(super), + (unsigned long long)reiser4_fake_allocated(super), + (unsigned long long) + reiser4_fake_allocated_unformatted(super), + (unsigned long long)reiser4_flush_reserved(super), + (unsigned long long)reiser4_clustered_blocks(super), + (unsigned long long)sum, + (unsigned long long)reiser4_block_count(super)); + return 0; + } + return 1; +} + +/* Adjust "working" free blocks counter for number of blocks we are going to + allocate. Record number of grabbed blocks in fs-wide and per-thread + counters. This function should be called before bitmap scanning or + allocating fake block numbers + + @super -- pointer to reiser4 super block; + @count -- number of blocks we reserve; + + @return -- 0 if success, -ENOSPC, if all + free blocks are preserved or already allocated. +*/ + +static int +reiser4_grab(reiser4_context * ctx, __u64 count, reiser4_ba_flags_t flags) +{ + __u64 free_blocks; + int ret = 0, use_reserved = flags & BA_RESERVED; + reiser4_super_info_data *sbinfo; + + assert("vs-1276", ctx == get_current_context()); + + /* Do not grab anything on ro-mounted fs. */ + if (rofs_super(ctx->super)) { + ctx->grab_enabled = 0; + return 0; + } + + sbinfo = get_super_private(ctx->super); + + spin_lock_reiser4_super(sbinfo); + + free_blocks = sbinfo->blocks_free; + + if ((use_reserved && free_blocks < count) || + (!use_reserved && free_blocks < count + sbinfo->blocks_reserved)) { + ret = RETERR(-ENOSPC); + goto unlock_and_ret; + } + + add_to_ctx_grabbed(ctx, count); + + sbinfo->blocks_grabbed += count; + sbinfo->blocks_free -= count; + +#if REISER4_DEBUG + if (ctx->grabbed_initially == 0) + ctx->grabbed_initially = count; +#endif + + assert("nikita-2986", reiser4_check_block_counters(ctx->super)); + + /* disable grab space in current context */ + ctx->grab_enabled = 0; + +unlock_and_ret: + spin_unlock_reiser4_super(sbinfo); + + return ret; +} + +int reiser4_grab_space(__u64 count, reiser4_ba_flags_t flags) +{ + int ret; + reiser4_context *ctx; + + assert("nikita-2964", ergo(flags & BA_CAN_COMMIT, + lock_stack_isclean(get_current_lock_stack + ()))); + ctx = get_current_context(); + if (!(flags & BA_FORCE) && !is_grab_enabled(ctx)) + return 0; + + ret = reiser4_grab(ctx, count, flags); + if (ret == -ENOSPC) { + + /* Trying to commit the all transactions if BA_CAN_COMMIT flag + present */ + if (flags & BA_CAN_COMMIT) { + txnmgr_force_commit_all(ctx->super, 0); + ctx->grab_enabled = 1; + ret = reiser4_grab(ctx, count, flags); + } + } + /* + * allocation from reserved pool cannot fail. This is severe error. + */ + assert("nikita-3005", ergo(flags & BA_RESERVED, ret == 0)); + return ret; +} + +/* + * SPACE RESERVED FOR UNLINK/TRUNCATE + * + * Unlink and truncate require space in transaction (to update stat data, at + * least). But we don't want rm(1) to fail with "No space on device" error. + * + * Solution is to reserve 5% of disk space for truncates and + * unlinks. Specifically, normal space grabbing requests don't grab space from + * reserved area. Only requests with BA_RESERVED bit in flags are allowed to + * drain it. Per super block delete mutex is used to allow only one + * thread at a time to grab from reserved area. + * + * Grabbing from reserved area should always be performed with BA_CAN_COMMIT + * flag. + * + */ + +int reiser4_grab_reserved(struct super_block *super, + __u64 count, reiser4_ba_flags_t flags) +{ + reiser4_super_info_data *sbinfo = get_super_private(super); + + assert("nikita-3175", flags & BA_CAN_COMMIT); + + /* Check the delete mutex already taken by us, we assume that + * reading of machine word is atomic. */ + if (sbinfo->delete_mutex_owner == current) { + if (reiser4_grab_space + (count, (flags | BA_RESERVED) & ~BA_CAN_COMMIT)) { + warning("zam-1003", + "nested call of grab_reserved fails count=(%llu)", + (unsigned long long)count); + reiser4_release_reserved(super); + return RETERR(-ENOSPC); + } + return 0; + } + + if (reiser4_grab_space(count, flags)) { + mutex_lock(&sbinfo->delete_mutex); + assert("nikita-2929", sbinfo->delete_mutex_owner == NULL); + sbinfo->delete_mutex_owner = current; + + if (reiser4_grab_space(count, flags | BA_RESERVED)) { + warning("zam-833", + "reserved space is not enough (%llu)", + (unsigned long long)count); + reiser4_release_reserved(super); + return RETERR(-ENOSPC); + } + } + return 0; +} + +void reiser4_release_reserved(struct super_block *super) +{ + reiser4_super_info_data *info; + + info = get_super_private(super); + if (info->delete_mutex_owner == current) { + info->delete_mutex_owner = NULL; + mutex_unlock(&info->delete_mutex); + } +} + +static reiser4_super_info_data *grabbed2fake_allocated_head(int count) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + sub_from_ctx_grabbed(ctx, count); + + sbinfo = get_super_private(ctx->super); + spin_lock_reiser4_super(sbinfo); + + sub_from_sb_grabbed(sbinfo, count); + /* return sbinfo locked */ + return sbinfo; +} + +/* is called after @count fake block numbers are allocated and pointer to + those blocks are inserted into tree. */ +static void grabbed2fake_allocated_formatted(void) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = grabbed2fake_allocated_head(1); + sbinfo->blocks_fake_allocated++; + + assert("vs-922", reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +/** + * grabbed2fake_allocated_unformatted + * @count: + * + */ +static void grabbed2fake_allocated_unformatted(int count) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = grabbed2fake_allocated_head(count); + sbinfo->blocks_fake_allocated_unformatted += count; + + assert("vs-9221", reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +void grabbed2cluster_reserved(int count) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + sub_from_ctx_grabbed(ctx, count); + + sbinfo = get_super_private(ctx->super); + spin_lock_reiser4_super(sbinfo); + + sub_from_sb_grabbed(sbinfo, count); + sbinfo->blocks_clustered += count; + + assert("edward-504", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +void cluster_reserved2grabbed(int count) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + + sbinfo = get_super_private(ctx->super); + spin_lock_reiser4_super(sbinfo); + + sub_from_cluster_reserved(sbinfo, count); + sbinfo->blocks_grabbed += count; + + assert("edward-505", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); + add_to_ctx_grabbed(ctx, count); +} + +void cluster_reserved2free(int count) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + cluster_reserved2grabbed(count); + grabbed2free(ctx, sbinfo, count); +} + +static DEFINE_SPINLOCK(fake_lock); +static reiser4_block_nr fake_gen = 0; + +/** + * assign_fake_blocknr + * @blocknr: + * @count: + * + * Obtain a fake block number for new node which will be used to refer to + * this newly allocated node until real allocation is done. + */ +static void assign_fake_blocknr(reiser4_block_nr *blocknr, int count) +{ + spin_lock(&fake_lock); + *blocknr = fake_gen; + fake_gen += count; + spin_unlock(&fake_lock); + + BUG_ON(*blocknr & REISER4_BLOCKNR_STATUS_BIT_MASK); + /**blocknr &= ~REISER4_BLOCKNR_STATUS_BIT_MASK;*/ + *blocknr |= REISER4_UNALLOCATED_STATUS_VALUE; + assert("zam-394", zlook(current_tree, blocknr) == NULL); +} + +int assign_fake_blocknr_formatted(reiser4_block_nr * blocknr) +{ + assign_fake_blocknr(blocknr, 1); + grabbed2fake_allocated_formatted(); + return 0; +} + +/** + * fake_blocknrs_unformatted + * @count: number of fake numbers to get + * + * Allocates @count fake block numbers which will be assigned to jnodes + */ +reiser4_block_nr fake_blocknr_unformatted(int count) +{ + reiser4_block_nr blocknr; + + assign_fake_blocknr(&blocknr, count); + grabbed2fake_allocated_unformatted(count); + + return blocknr; +} + +/* adjust sb block counters, if real (on-disk) block allocation immediately + follows grabbing of free disk space. */ +static void grabbed2used(reiser4_context *ctx, reiser4_super_info_data *sbinfo, + __u64 count) +{ + sub_from_ctx_grabbed(ctx, count); + + spin_lock_reiser4_super(sbinfo); + + sub_from_sb_grabbed(sbinfo, count); + sbinfo->blocks_used += count; + + assert("nikita-2679", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +/* adjust sb block counters when @count unallocated blocks get mapped to disk */ +static void fake_allocated2used(reiser4_super_info_data *sbinfo, __u64 count, + reiser4_ba_flags_t flags) +{ + spin_lock_reiser4_super(sbinfo); + + sub_from_sb_fake_allocated(sbinfo, count, flags); + sbinfo->blocks_used += count; + + assert("nikita-2680", + reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +static void flush_reserved2used(txn_atom * atom, __u64 count) +{ + reiser4_super_info_data *sbinfo; + + assert("zam-787", atom != NULL); + assert_spin_locked(&(atom->alock)); + + sub_from_atom_flush_reserved_nolock(atom, (__u32) count); + + sbinfo = get_current_super_private(); + spin_lock_reiser4_super(sbinfo); + + sub_from_sb_flush_reserved(sbinfo, count); + sbinfo->blocks_used += count; + + assert("zam-789", + reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +/* update the per fs blocknr hint default value. */ +void +update_blocknr_hint_default(const struct super_block *s, + const reiser4_block_nr * block) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + + assert("nikita-3342", !reiser4_blocknr_is_fake(block)); + + spin_lock_reiser4_super(sbinfo); + if (*block < sbinfo->block_count) { + sbinfo->blocknr_hint_default = *block; + } else { + warning("zam-676", + "block number %llu is too large to be used in a blocknr hint\n", + (unsigned long long)*block); + dump_stack(); + DEBUGON(1); + } + spin_unlock_reiser4_super(sbinfo); +} + +/* get current value of the default blocknr hint. */ +void get_blocknr_hint_default(reiser4_block_nr * result) +{ + reiser4_super_info_data *sbinfo = get_current_super_private(); + + spin_lock_reiser4_super(sbinfo); + *result = sbinfo->blocknr_hint_default; + assert("zam-677", *result < sbinfo->block_count); + spin_unlock_reiser4_super(sbinfo); +} + +/* Allocate "real" disk blocks by calling a proper space allocation plugin + * method. Blocks are allocated in one contiguous disk region. The plugin + * independent part accounts blocks by subtracting allocated amount from grabbed + * or fake block counter and add the same amount to the counter of allocated + * blocks. + * + * @hint -- a reiser4 blocknr hint object which contains further block + * allocation hints and parameters (search start, a stage of block + * which will be mapped to disk, etc.), + * @blk -- an out parameter for the beginning of the allocated region, + * @len -- in/out parameter, it should contain the maximum number of allocated + * blocks, after block allocation completes, it contains the length of + * allocated disk region. + * @flags -- see reiser4_ba_flags_t description. + * + * @return -- 0 if success, error code otherwise. + */ +int +reiser4_alloc_blocks(reiser4_blocknr_hint * hint, reiser4_block_nr * blk, + reiser4_block_nr * len, reiser4_ba_flags_t flags) +{ + __u64 needed = *len; + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + int ret; + + assert("zam-986", hint != NULL); + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + /* For write-optimized data we use default search start value, which is + * close to last write location. */ + if (flags & BA_USE_DEFAULT_SEARCH_START) + get_blocknr_hint_default(&hint->blk); + + /* VITALY: allocator should grab this for internal/tx-lists/similar + only. */ +/* VS-FIXME-HANS: why is this comment above addressed to vitaly (from vitaly)?*/ + if (hint->block_stage == BLOCK_NOT_COUNTED) { + ret = reiser4_grab_space_force(*len, flags); + if (ret != 0) + return ret; + } + + ret = + sa_alloc_blocks(reiser4_get_space_allocator(ctx->super), + hint, (int)needed, blk, len); + + if (!ret) { + assert("zam-680", *blk < reiser4_block_count(ctx->super)); + assert("zam-681", + *blk + *len <= reiser4_block_count(ctx->super)); + + if (flags & BA_PERMANENT) { + /* we assume that current atom exists at this moment */ + txn_atom *atom = get_current_atom_locked(); + atom->nr_blocks_allocated += *len; + spin_unlock_atom(atom); + } + + switch (hint->block_stage) { + case BLOCK_NOT_COUNTED: + case BLOCK_GRABBED: + grabbed2used(ctx, sbinfo, *len); + break; + case BLOCK_UNALLOCATED: + fake_allocated2used(sbinfo, *len, flags); + break; + case BLOCK_FLUSH_RESERVED: + { + txn_atom *atom = get_current_atom_locked(); + flush_reserved2used(atom, *len); + spin_unlock_atom(atom); + } + break; + default: + impossible("zam-531", "wrong block stage"); + } + } else { + assert("zam-821", + ergo(hint->max_dist == 0 + && !hint->backward, ret != -ENOSPC)); + if (hint->block_stage == BLOCK_NOT_COUNTED) + grabbed2free(ctx, sbinfo, needed); + } + + return ret; +} + +/* used -> fake_allocated -> grabbed -> free */ + +/* adjust sb block counters when @count unallocated blocks get unmapped from + disk */ +static void +used2fake_allocated(reiser4_super_info_data * sbinfo, __u64 count, + int formatted) +{ + spin_lock_reiser4_super(sbinfo); + + if (formatted) + sbinfo->blocks_fake_allocated += count; + else + sbinfo->blocks_fake_allocated_unformatted += count; + + sub_from_sb_used(sbinfo, count); + + assert("nikita-2681", + reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +static void +used2flush_reserved(reiser4_super_info_data * sbinfo, txn_atom * atom, + __u64 count, reiser4_ba_flags_t flags UNUSED_ARG) +{ + assert("nikita-2791", atom != NULL); + assert_spin_locked(&(atom->alock)); + + add_to_atom_flush_reserved_nolock(atom, (__u32) count); + + spin_lock_reiser4_super(sbinfo); + + sbinfo->blocks_flush_reserved += count; + /*add_to_sb_flush_reserved(sbinfo, count); */ + sub_from_sb_used(sbinfo, count); + + assert("nikita-2681", + reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +/* disk space, virtually used by fake block numbers is counted as "grabbed" + again. */ +static void +fake_allocated2grabbed(reiser4_context * ctx, reiser4_super_info_data * sbinfo, + __u64 count, reiser4_ba_flags_t flags) +{ + add_to_ctx_grabbed(ctx, count); + + spin_lock_reiser4_super(sbinfo); + + assert("nikita-2682", reiser4_check_block_counters(ctx->super)); + + sbinfo->blocks_grabbed += count; + sub_from_sb_fake_allocated(sbinfo, count, flags & BA_FORMATTED); + + assert("nikita-2683", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +void fake_allocated2free(__u64 count, reiser4_ba_flags_t flags) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + fake_allocated2grabbed(ctx, sbinfo, count, flags); + grabbed2free(ctx, sbinfo, count); +} + +void grabbed2free_mark(__u64 mark) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + assert("nikita-3007", (__s64) mark >= 0); + assert("nikita-3006", ctx->grabbed_blocks >= mark); + grabbed2free(ctx, sbinfo, ctx->grabbed_blocks - mark); +} + +/** + * grabbed2free - adjust grabbed and free block counters + * @ctx: context to update grabbed block counter of + * @sbinfo: super block to update grabbed and free block counters of + * @count: number of blocks to adjust counters by + * + * Decreases context's and per filesystem's counters of grabbed + * blocks. Increases per filesystem's counter of free blocks. + */ +void grabbed2free(reiser4_context *ctx, reiser4_super_info_data *sbinfo, + __u64 count) +{ + sub_from_ctx_grabbed(ctx, count); + + spin_lock_reiser4_super(sbinfo); + + sub_from_sb_grabbed(sbinfo, count); + sbinfo->blocks_free += count; + assert("nikita-2684", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +void grabbed2flush_reserved_nolock(txn_atom * atom, __u64 count) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + assert("vs-1095", atom); + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + sub_from_ctx_grabbed(ctx, count); + + add_to_atom_flush_reserved_nolock(atom, count); + + spin_lock_reiser4_super(sbinfo); + + sbinfo->blocks_flush_reserved += count; + sub_from_sb_grabbed(sbinfo, count); + + assert("vpf-292", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +void grabbed2flush_reserved(__u64 count) +{ + txn_atom *atom = get_current_atom_locked(); + + grabbed2flush_reserved_nolock(atom, count); + + spin_unlock_atom(atom); +} + +void flush_reserved2grabbed(txn_atom * atom, __u64 count) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + assert("nikita-2788", atom != NULL); + assert_spin_locked(&(atom->alock)); + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + add_to_ctx_grabbed(ctx, count); + + sub_from_atom_flush_reserved_nolock(atom, (__u32) count); + + spin_lock_reiser4_super(sbinfo); + + sbinfo->blocks_grabbed += count; + sub_from_sb_flush_reserved(sbinfo, count); + + assert("vpf-292", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +/** + * all_grabbed2free - releases all blocks grabbed in context + * + * Decreases context's and super block's grabbed block counters by number of + * blocks grabbed by current context and increases super block's free block + * counter correspondingly. + */ +void all_grabbed2free(void) +{ + reiser4_context *ctx = get_current_context(); + + grabbed2free(ctx, get_super_private(ctx->super), ctx->grabbed_blocks); +} + +/* adjust sb block counters if real (on-disk) blocks do not become unallocated + after freeing, @count blocks become "grabbed". */ +static void +used2grabbed(reiser4_context * ctx, reiser4_super_info_data * sbinfo, + __u64 count) +{ + add_to_ctx_grabbed(ctx, count); + + spin_lock_reiser4_super(sbinfo); + + sbinfo->blocks_grabbed += count; + sub_from_sb_used(sbinfo, count); + + assert("nikita-2685", reiser4_check_block_counters(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); +} + +/* this used to be done through used2grabbed and grabbed2free*/ +static void used2free(reiser4_super_info_data * sbinfo, __u64 count) +{ + spin_lock_reiser4_super(sbinfo); + + sbinfo->blocks_free += count; + sub_from_sb_used(sbinfo, count); + + assert("nikita-2685", + reiser4_check_block_counters(reiser4_get_current_sb())); + + spin_unlock_reiser4_super(sbinfo); +} + +#if REISER4_DEBUG + +/* check "allocated" state of given block range */ +static void +reiser4_check_blocks(const reiser4_block_nr * start, + const reiser4_block_nr * len, int desired) +{ + sa_check_blocks(start, len, desired); +} + +/* check "allocated" state of given block */ +void reiser4_check_block(const reiser4_block_nr * block, int desired) +{ + const reiser4_block_nr one = 1; + + reiser4_check_blocks(block, &one, desired); +} + +#endif + +/* Blocks deallocation function may do an actual deallocation through space + plugin allocation or store deleted block numbers in atom's delete_set data + structure depend on @defer parameter. */ + +/* if BA_DEFER bit is not turned on, @target_stage means the stage of blocks + which will be deleted from WORKING bitmap. They might be just unmapped from + disk, or freed but disk space is still grabbed by current thread, or these + blocks must not be counted in any reiser4 sb block counters, + see block_stage_t comment */ + +/* BA_FORMATTED bit is only used when BA_DEFER in not present: it is used to + distinguish blocks allocated for unformatted and formatted nodes */ + +int +reiser4_dealloc_blocks(const reiser4_block_nr * start, + const reiser4_block_nr * len, + block_stage_t target_stage, reiser4_ba_flags_t flags) +{ + txn_atom *atom = NULL; + int ret; + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + if (REISER4_DEBUG) { + assert("zam-431", *len != 0); + assert("zam-432", *start != 0); + assert("zam-558", !reiser4_blocknr_is_fake(start)); + + spin_lock_reiser4_super(sbinfo); + assert("zam-562", *start < sbinfo->block_count); + spin_unlock_reiser4_super(sbinfo); + } + + if (flags & BA_DEFER) { + blocknr_set_entry *bsep = NULL; + + /* storing deleted block numbers in a blocknr set + datastructure for further actual deletion */ + do { + atom = get_current_atom_locked(); + assert("zam-430", atom != NULL); + + ret = + blocknr_set_add_extent(atom, &atom->delete_set, + &bsep, start, len); + + if (ret == -ENOMEM) + return ret; + + /* This loop might spin at most two times */ + } while (ret == -E_REPEAT); + + assert("zam-477", ret == 0); + assert("zam-433", atom != NULL); + + spin_unlock_atom(atom); + + } else { + assert("zam-425", get_current_super_private() != NULL); + sa_dealloc_blocks(reiser4_get_space_allocator(ctx->super), + *start, *len); + + if (flags & BA_PERMANENT) { + /* These blocks were counted as allocated, we have to + * revert it back if allocation is discarded. */ + txn_atom *atom = get_current_atom_locked(); + atom->nr_blocks_allocated -= *len; + spin_unlock_atom(atom); + } + + switch (target_stage) { + case BLOCK_NOT_COUNTED: + assert("vs-960", flags & BA_FORMATTED); + /* VITALY: This is what was grabbed for + internal/tx-lists/similar only */ + used2free(sbinfo, *len); + break; + + case BLOCK_GRABBED: + used2grabbed(ctx, sbinfo, *len); + break; + + case BLOCK_UNALLOCATED: + used2fake_allocated(sbinfo, *len, flags & BA_FORMATTED); + break; + + case BLOCK_FLUSH_RESERVED:{ + txn_atom *atom; + + atom = get_current_atom_locked(); + used2flush_reserved(sbinfo, atom, *len, + flags & BA_FORMATTED); + spin_unlock_atom(atom); + break; + } + default: + impossible("zam-532", "wrong block stage"); + } + } + + return 0; +} + +/* wrappers for block allocator plugin methods */ +int reiser4_pre_commit_hook(void) +{ + assert("zam-502", get_current_super_private() != NULL); + sa_pre_commit_hook(); + return 0; +} + +/* an actor which applies delete set to block allocator data */ +static int +apply_dset(txn_atom * atom UNUSED_ARG, const reiser4_block_nr * a, + const reiser4_block_nr * b, void *data UNUSED_ARG) +{ + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + __u64 len = 1; + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + assert("zam-877", atom->stage >= ASTAGE_PRE_COMMIT); + assert("zam-552", sbinfo != NULL); + + if (b != NULL) + len = *b; + + if (REISER4_DEBUG) { + spin_lock_reiser4_super(sbinfo); + + assert("zam-554", *a < reiser4_block_count(ctx->super)); + assert("zam-555", *a + len <= reiser4_block_count(ctx->super)); + + spin_unlock_reiser4_super(sbinfo); + } + + sa_dealloc_blocks(&sbinfo->space_allocator, *a, len); + /* adjust sb block counters */ + used2free(sbinfo, len); + return 0; +} + +void reiser4_post_commit_hook(void) +{ + txn_atom *atom; + + atom = get_current_atom_locked(); + assert("zam-452", atom->stage == ASTAGE_POST_COMMIT); + spin_unlock_atom(atom); + + /* do the block deallocation which was deferred + until commit is done */ + blocknr_set_iterator(atom, &atom->delete_set, apply_dset, NULL, 1); + + assert("zam-504", get_current_super_private() != NULL); + sa_post_commit_hook(); +} + +void reiser4_post_write_back_hook(void) +{ + assert("zam-504", get_current_super_private() != NULL); + + sa_post_commit_hook(); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/block_alloc.h linux-2.6.30/fs/reiser4/block_alloc.h --- linux-2.6.30.orig/fs/reiser4/block_alloc.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/block_alloc.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,177 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#if !defined(__FS_REISER4_BLOCK_ALLOC_H__) +#define __FS_REISER4_BLOCK_ALLOC_H__ + +#include "dformat.h" +#include "forward.h" + +#include /* for __u?? */ +#include + +/* Mask when is applied to given block number shows is that block number is a + fake one */ +#define REISER4_FAKE_BLOCKNR_BIT_MASK 0x8000000000000000ULL +/* Mask which isolates a type of object this fake block number was assigned + to */ +#define REISER4_BLOCKNR_STATUS_BIT_MASK 0xC000000000000000ULL + +/*result after applying the REISER4_BLOCKNR_STATUS_BIT_MASK should be compared + against these two values to understand is the object unallocated or bitmap + shadow object (WORKING BITMAP block, look at the plugin/space/bitmap.c) */ +#define REISER4_UNALLOCATED_STATUS_VALUE 0xC000000000000000ULL +#define REISER4_BITMAP_BLOCKS_STATUS_VALUE 0x8000000000000000ULL + +/* specification how block allocation was counted in sb block counters */ +typedef enum { + BLOCK_NOT_COUNTED = 0, /* reiser4 has no info about this block yet */ + BLOCK_GRABBED = 1, /* free space grabbed for further allocation + of this block */ + BLOCK_FLUSH_RESERVED = 2, /* block is reserved for flush needs. */ + BLOCK_UNALLOCATED = 3, /* block is used for existing in-memory object + ( unallocated formatted or unformatted + node) */ + BLOCK_ALLOCATED = 4 /* block is mapped to disk, real on-disk block + number assigned */ +} block_stage_t; + +/* a hint for block allocator */ +struct reiser4_blocknr_hint { + /* FIXME: I think we want to add a longterm lock on the bitmap block + here. This is to prevent jnode_flush() calls from interleaving + allocations on the same bitmap, once a hint is established. */ + + /* search start hint */ + reiser4_block_nr blk; + /* if not zero, it is a region size we search for free blocks in */ + reiser4_block_nr max_dist; + /* level for allocation, may be useful have branch-level and higher + write-optimized. */ + tree_level level; + /* block allocator assumes that blocks, which will be mapped to disk, + are in this specified block_stage */ + block_stage_t block_stage; + /* If direction = 1 allocate blocks in backward direction from the end + * of disk to the beginning of disk. */ + unsigned int backward:1; + +}; + +/* These flags control block allocation/deallocation behavior */ +enum reiser4_ba_flags { + /* do allocatations from reserved (5%) area */ + BA_RESERVED = (1 << 0), + + /* block allocator can do commit trying to recover free space */ + BA_CAN_COMMIT = (1 << 1), + + /* if operation will be applied to formatted block */ + BA_FORMATTED = (1 << 2), + + /* defer actual block freeing until transaction commit */ + BA_DEFER = (1 << 3), + + /* allocate blocks for permanent fs objects (formatted or unformatted), + not wandered of log blocks */ + BA_PERMANENT = (1 << 4), + + /* grab space even it was disabled */ + BA_FORCE = (1 << 5), + + /* use default start value for free blocks search. */ + BA_USE_DEFAULT_SEARCH_START = (1 << 6) +}; + +typedef enum reiser4_ba_flags reiser4_ba_flags_t; + +extern void reiser4_blocknr_hint_init(reiser4_blocknr_hint * hint); +extern void reiser4_blocknr_hint_done(reiser4_blocknr_hint * hint); +extern void update_blocknr_hint_default(const struct super_block *, + const reiser4_block_nr *); +extern void get_blocknr_hint_default(reiser4_block_nr *); + +extern reiser4_block_nr reiser4_fs_reserved_space(struct super_block *super); + +int assign_fake_blocknr_formatted(reiser4_block_nr *); +reiser4_block_nr fake_blocknr_unformatted(int); + +/* free -> grabbed -> fake_allocated -> used */ + +int reiser4_grab_space(__u64 count, reiser4_ba_flags_t flags); +void all_grabbed2free(void); +void grabbed2free(reiser4_context * , reiser4_super_info_data * , __u64 count); +void fake_allocated2free(__u64 count, reiser4_ba_flags_t flags); +void grabbed2flush_reserved_nolock(txn_atom * atom, __u64 count); +void grabbed2flush_reserved(__u64 count); +int reiser4_alloc_blocks(reiser4_blocknr_hint * hint, + reiser4_block_nr * start, + reiser4_block_nr * len, reiser4_ba_flags_t flags); +int reiser4_dealloc_blocks(const reiser4_block_nr *, + const reiser4_block_nr *, + block_stage_t, reiser4_ba_flags_t flags); + +static inline int reiser4_alloc_block(reiser4_blocknr_hint * hint, + reiser4_block_nr * start, + reiser4_ba_flags_t flags) +{ + reiser4_block_nr one = 1; + return reiser4_alloc_blocks(hint, start, &one, flags); +} + +static inline int reiser4_dealloc_block(const reiser4_block_nr * block, + block_stage_t stage, + reiser4_ba_flags_t flags) +{ + const reiser4_block_nr one = 1; + return reiser4_dealloc_blocks(block, &one, stage, flags); +} + +#define reiser4_grab_space_force(count, flags) \ + reiser4_grab_space(count, flags | BA_FORCE) + +extern void grabbed2free_mark(__u64 mark); +extern int reiser4_grab_reserved(struct super_block *, + __u64, reiser4_ba_flags_t); +extern void reiser4_release_reserved(struct super_block *super); + +/* grabbed -> fake_allocated */ + +/* fake_allocated -> used */ + +/* used -> fake_allocated -> grabbed -> free */ + +extern void flush_reserved2grabbed(txn_atom * atom, __u64 count); + +extern int reiser4_blocknr_is_fake(const reiser4_block_nr * da); + +extern void grabbed2cluster_reserved(int count); +extern void cluster_reserved2grabbed(int count); +extern void cluster_reserved2free(int count); + +extern int reiser4_check_block_counters(const struct super_block *); + +#if REISER4_DEBUG + +extern void reiser4_check_block(const reiser4_block_nr *, int); + +#else + +# define reiser4_check_block(beg, val) noop + +#endif + +extern int reiser4_pre_commit_hook(void); +extern void reiser4_post_commit_hook(void); +extern void reiser4_post_write_back_hook(void); + +#endif /* __FS_REISER4_BLOCK_ALLOC_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/blocknrset.c linux-2.6.30/fs/reiser4/blocknrset.c --- linux-2.6.30.orig/fs/reiser4/blocknrset.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/blocknrset.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,371 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by +reiser4/README */ + +/* This file contains code for various block number sets used by the atom to + track the deleted set and wandered block mappings. */ + +#include "debug.h" +#include "dformat.h" +#include "txnmgr.h" +#include "context.h" + +#include + +/* The proposed data structure for storing unordered block number sets is a + list of elements, each of which contains an array of block number or/and + array of block number pairs. That element called blocknr_set_entry is used + to store block numbers from the beginning and for extents from the end of + the data field (char data[...]). The ->nr_blocks and ->nr_pairs fields + count numbers of blocks and extents. + + +------------------- blocknr_set_entry->data ------------------+ + |block1|block2| ... ... |pair3|pair2|pair1| + +------------------------------------------------------------+ + + When current blocknr_set_entry is full, allocate a new one. */ + +/* Usage examples: blocknr sets are used in reiser4 for storing atom's delete + * set (single blocks and block extents), in that case blocknr pair represent an + * extent; atom's wandered map is also stored as a blocknr set, blocknr pairs + * there represent a (real block) -> (wandered block) mapping. */ + +/* Protection: blocknr sets belong to reiser4 atom, and + * their modifications are performed with the atom lock held */ + +/* The total size of a blocknr_set_entry. */ +#define BLOCKNR_SET_ENTRY_SIZE 128 + +/* The number of blocks that can fit the blocknr data area. */ +#define BLOCKNR_SET_ENTRIES_NUMBER \ + ((BLOCKNR_SET_ENTRY_SIZE - \ + 2 * sizeof(unsigned) - \ + sizeof(struct list_head)) / \ + sizeof(reiser4_block_nr)) + +/* An entry of the blocknr_set */ +struct blocknr_set_entry { + unsigned nr_singles; + unsigned nr_pairs; + struct list_head link; + reiser4_block_nr entries[BLOCKNR_SET_ENTRIES_NUMBER]; +}; + +/* A pair of blocks as recorded in the blocknr_set_entry data. */ +struct blocknr_pair { + reiser4_block_nr a; + reiser4_block_nr b; +}; + +/* Return the number of blocknr slots available in a blocknr_set_entry. */ +/* Audited by: green(2002.06.11) */ +static unsigned bse_avail(blocknr_set_entry * bse) +{ + unsigned used = bse->nr_singles + 2 * bse->nr_pairs; + + assert("jmacd-5088", BLOCKNR_SET_ENTRIES_NUMBER >= used); + cassert(sizeof(blocknr_set_entry) == BLOCKNR_SET_ENTRY_SIZE); + + return BLOCKNR_SET_ENTRIES_NUMBER - used; +} + +/* Initialize a blocknr_set_entry. */ +static void bse_init(blocknr_set_entry *bse) +{ + bse->nr_singles = 0; + bse->nr_pairs = 0; + INIT_LIST_HEAD(&bse->link); +} + +/* Allocate and initialize a blocknr_set_entry. */ +/* Audited by: green(2002.06.11) */ +static blocknr_set_entry *bse_alloc(void) +{ + blocknr_set_entry *e; + + if ((e = (blocknr_set_entry *) kmalloc(sizeof(blocknr_set_entry), + reiser4_ctx_gfp_mask_get())) == NULL) + return NULL; + + bse_init(e); + + return e; +} + +/* Free a blocknr_set_entry. */ +/* Audited by: green(2002.06.11) */ +static void bse_free(blocknr_set_entry * bse) +{ + kfree(bse); +} + +/* Add a block number to a blocknr_set_entry */ +/* Audited by: green(2002.06.11) */ +static void +bse_put_single(blocknr_set_entry * bse, const reiser4_block_nr * block) +{ + assert("jmacd-5099", bse_avail(bse) >= 1); + + bse->entries[bse->nr_singles++] = *block; +} + +/* Get a pair of block numbers */ +/* Audited by: green(2002.06.11) */ +static inline struct blocknr_pair *bse_get_pair(blocknr_set_entry * bse, + unsigned pno) +{ + assert("green-1", BLOCKNR_SET_ENTRIES_NUMBER >= 2 * (pno + 1)); + + return (struct blocknr_pair *) (bse->entries + + BLOCKNR_SET_ENTRIES_NUMBER - + 2 * (pno + 1)); +} + +/* Add a pair of block numbers to a blocknr_set_entry */ +/* Audited by: green(2002.06.11) */ +static void +bse_put_pair(blocknr_set_entry * bse, const reiser4_block_nr * a, + const reiser4_block_nr * b) +{ + struct blocknr_pair *pair; + + assert("jmacd-5100", bse_avail(bse) >= 2 && a != NULL && b != NULL); + + pair = bse_get_pair(bse, bse->nr_pairs++); + + pair->a = *a; + pair->b = *b; +} + +/* Add either a block or pair of blocks to the block number set. The first + blocknr (@a) must be non-NULL. If @b is NULL a single blocknr is added, if + @b is non-NULL a pair is added. The block number set belongs to atom, and + the call is made with the atom lock held. There may not be enough space in + the current blocknr_set_entry. If new_bsep points to a non-NULL + blocknr_set_entry then it will be added to the blocknr_set and new_bsep + will be set to NULL. If new_bsep contains NULL then the atom lock will be + released and a new bse will be allocated in new_bsep. E_REPEAT will be + returned with the atom unlocked for the operation to be tried again. If + the operation succeeds, 0 is returned. If new_bsep is non-NULL and not + used during the call, it will be freed automatically. */ +static int blocknr_set_add(txn_atom *atom, struct list_head *bset, + blocknr_set_entry **new_bsep, const reiser4_block_nr *a, + const reiser4_block_nr *b) +{ + blocknr_set_entry *bse; + unsigned entries_needed; + + assert("jmacd-5101", a != NULL); + + entries_needed = (b == NULL) ? 1 : 2; + if (list_empty(bset) || + bse_avail(list_entry(bset->next, blocknr_set_entry, link)) < entries_needed) { + /* See if a bse was previously allocated. */ + if (*new_bsep == NULL) { + spin_unlock_atom(atom); + *new_bsep = bse_alloc(); + return (*new_bsep != NULL) ? -E_REPEAT : + RETERR(-ENOMEM); + } + + /* Put it on the head of the list. */ + list_add(&((*new_bsep)->link), bset); + + *new_bsep = NULL; + } + + /* Add the single or pair. */ + bse = list_entry(bset->next, blocknr_set_entry, link); + if (b == NULL) { + bse_put_single(bse, a); + } else { + bse_put_pair(bse, a, b); + } + + /* If new_bsep is non-NULL then there was an allocation race, free this + copy. */ + if (*new_bsep != NULL) { + bse_free(*new_bsep); + *new_bsep = NULL; + } + + return 0; +} + +/* Add an extent to the block set. If the length is 1, it is treated as a + single block (e.g., reiser4_set_add_block). */ +/* Audited by: green(2002.06.11) */ +/* Auditor note: Entire call chain cannot hold any spinlocks, because + kmalloc might schedule. The only exception is atom spinlock, which is + properly freed. */ +int +blocknr_set_add_extent(txn_atom * atom, + struct list_head *bset, + blocknr_set_entry ** new_bsep, + const reiser4_block_nr * start, + const reiser4_block_nr * len) +{ + assert("jmacd-5102", start != NULL && len != NULL && *len > 0); + return blocknr_set_add(atom, bset, new_bsep, start, + *len == 1 ? NULL : len); +} + +/* Add a block pair to the block set. It adds exactly a pair, which is checked + * by an assertion that both arguments are not null.*/ +/* Audited by: green(2002.06.11) */ +/* Auditor note: Entire call chain cannot hold any spinlocks, because + kmalloc might schedule. The only exception is atom spinlock, which is + properly freed. */ +int +blocknr_set_add_pair(txn_atom * atom, + struct list_head *bset, + blocknr_set_entry ** new_bsep, const reiser4_block_nr * a, + const reiser4_block_nr * b) +{ + assert("jmacd-5103", a != NULL && b != NULL); + return blocknr_set_add(atom, bset, new_bsep, a, b); +} + +/* Initialize a blocknr_set. */ +void blocknr_set_init(struct list_head *bset) +{ + INIT_LIST_HEAD(bset); +} + +/* Release the entries of a blocknr_set. */ +void blocknr_set_destroy(struct list_head *bset) +{ + blocknr_set_entry *bse; + + while (!list_empty(bset)) { + bse = list_entry(bset->next, blocknr_set_entry, link); + list_del_init(&bse->link); + bse_free(bse); + } +} + +/* Merge blocknr_set entries out of @from into @into. */ +/* Audited by: green(2002.06.11) */ +/* Auditor comments: This merge does not know if merged sets contain + blocks pairs (As for wandered sets) or extents, so it cannot really merge + overlapping ranges if there is some. So I believe it may lead to + some blocks being presented several times in one blocknr_set. To help + debugging such problems it might help to check for duplicate entries on + actual processing of this set. Testing this kind of stuff right here is + also complicated by the fact that these sets are not sorted and going + through whole set on each element addition is going to be CPU-heavy task */ +void blocknr_set_merge(struct list_head *from, struct list_head *into) +{ + blocknr_set_entry *bse_into = NULL; + + /* If @from is empty, no work to perform. */ + if (list_empty(from)) + return; + /* If @into is not empty, try merging partial-entries. */ + if (!list_empty(into)) { + + /* Neither set is empty, pop the front to members and try to + combine them. */ + blocknr_set_entry *bse_from; + unsigned into_avail; + + bse_into = list_entry(into->next, blocknr_set_entry, link); + list_del_init(&bse_into->link); + bse_from = list_entry(from->next, blocknr_set_entry, link); + list_del_init(&bse_from->link); + + /* Combine singles. */ + for (into_avail = bse_avail(bse_into); + into_avail != 0 && bse_from->nr_singles != 0; + into_avail -= 1) { + bse_put_single(bse_into, + &bse_from->entries[--bse_from-> + nr_singles]); + } + + /* Combine pairs. */ + for (; into_avail > 1 && bse_from->nr_pairs != 0; + into_avail -= 2) { + struct blocknr_pair *pair = + bse_get_pair(bse_from, --bse_from->nr_pairs); + bse_put_pair(bse_into, &pair->a, &pair->b); + } + + /* If bse_from is empty, delete it now. */ + if (bse_avail(bse_from) == BLOCKNR_SET_ENTRIES_NUMBER) { + bse_free(bse_from); + } else { + /* Otherwise, bse_into is full or nearly full (e.g., + it could have one slot avail and bse_from has one + pair left). Push it back onto the list. bse_from + becomes bse_into, which will be the new partial. */ + list_add(&bse_into->link, into); + bse_into = bse_from; + } + } + + /* Splice lists together. */ + list_splice_init(from, into->prev); + + /* Add the partial entry back to the head of the list. */ + if (bse_into != NULL) + list_add(&bse_into->link, into); +} + +/* Iterate over all blocknr set elements. */ +int blocknr_set_iterator(txn_atom *atom, struct list_head *bset, + blocknr_set_actor_f actor, void *data, int delete) +{ + + blocknr_set_entry *entry; + + assert("zam-429", atom != NULL); + assert("zam-430", atom_is_protected(atom)); + assert("zam-431", bset != 0); + assert("zam-432", actor != NULL); + + entry = list_entry(bset->next, blocknr_set_entry, link); + while (bset != &entry->link) { + blocknr_set_entry *tmp = list_entry(entry->link.next, blocknr_set_entry, link); + unsigned int i; + int ret; + + for (i = 0; i < entry->nr_singles; i++) { + ret = actor(atom, &entry->entries[i], NULL, data); + + /* We can't break a loop if delete flag is set. */ + if (ret != 0 && !delete) + return ret; + } + + for (i = 0; i < entry->nr_pairs; i++) { + struct blocknr_pair *ab; + + ab = bse_get_pair(entry, i); + + ret = actor(atom, &ab->a, &ab->b, data); + + if (ret != 0 && !delete) + return ret; + } + + if (delete) { + list_del(&entry->link); + bse_free(entry); + } + + entry = tmp; + } + + return 0; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/carry.c linux-2.6.30/fs/reiser4/carry.c --- linux-2.6.30.orig/fs/reiser4/carry.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/carry.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1398 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ +/* Functions to "carry" tree modification(s) upward. */ +/* Tree is modified one level at a time. As we modify a level we accumulate a + set of changes that need to be propagated to the next level. We manage + node locking such that any searches that collide with carrying are + restarted, from the root if necessary. + + Insertion of a new item may result in items being moved among nodes and + this requires the delimiting key to be updated at the least common parent + of the nodes modified to preserve search tree invariants. Also, insertion + may require allocation of a new node. A pointer to the new node has to be + inserted into some node on the parent level, etc. + + Tree carrying is meant to be analogous to arithmetic carrying. + + A carry operation is always associated with some node (&carry_node). + + Carry process starts with some initial set of operations to be performed + and an initial set of already locked nodes. Operations are performed one + by one. Performing each single operation has following possible effects: + + - content of carry node associated with operation is modified + - new carry nodes are locked and involved into carry process on this level + - new carry operations are posted to the next level + + After all carry operations on this level are done, process is repeated for + the accumulated sequence on carry operations for the next level. This + starts by trying to lock (in left to right order) all carry nodes + associated with carry operations on the parent level. After this, we decide + whether more nodes are required on the left of already locked set. If so, + all locks taken on the parent level are released, new carry nodes are + added, and locking process repeats. + + It may happen that balancing process fails owing to unrecoverable error on + some of upper levels of a tree (possible causes are io error, failure to + allocate new node, etc.). In this case we should unmount the filesystem, + rebooting if it is the root, and possibly advise the use of fsck. + + USAGE: + + int some_tree_operation( znode *node, ... ) + { + // Allocate on a stack pool of carry objects: operations and nodes. + // Most carry processes will only take objects from here, without + // dynamic allocation. + +I feel uneasy about this pool. It adds to code complexity, I understand why it +exists, but.... -Hans + + carry_pool pool; + carry_level lowest_level; + carry_op *op; + + init_carry_pool( &pool ); + init_carry_level( &lowest_level, &pool ); + + // operation may be one of: + // COP_INSERT --- insert new item into node + // COP_CUT --- remove part of or whole node + // COP_PASTE --- increase size of item + // COP_DELETE --- delete pointer from parent node + // COP_UPDATE --- update delimiting key in least + // common ancestor of two + + op = reiser4_post_carry( &lowest_level, operation, node, 0 ); + if( IS_ERR( op ) || ( op == NULL ) ) { + handle error + } else { + // fill in remaining fields in @op, according to carry.h:carry_op + result = carry(&lowest_level, NULL); + } + done_carry_pool(&pool); + } + + When you are implementing node plugin method that participates in carry + (shifting, insertion, deletion, etc.), do the following: + + int foo_node_method(znode * node, ..., carry_level * todo) + { + carry_op *op; + + .... + + // note, that last argument to reiser4_post_carry() is non-null + // here, because @op is to be applied to the parent of @node, rather + // than to the @node itself as in the previous case. + + op = node_post_carry(todo, operation, node, 1); + // fill in remaining fields in @op, according to carry.h:carry_op + + .... + + } + + BATCHING: + + One of the main advantages of level-by-level balancing implemented here is + ability to batch updates on a parent level and to peform them more + efficiently as a result. + + Description To Be Done (TBD). + + DIFFICULTIES AND SUBTLE POINTS: + + 1. complex plumbing is required, because: + + a. effective allocation through pools is needed + + b. target of operation is not exactly known when operation is + posted. This is worked around through bitfields in &carry_node and + logic in lock_carry_node() + + c. of interaction with locking code: node should be added into sibling + list when pointer to it is inserted into its parent, which is some time + after node was created. Between these moments, node is somewhat in + suspended state and is only registered in the carry lists + + 2. whole balancing logic is implemented here, in particular, insertion + logic is coded in make_space(). + + 3. special cases like insertion (reiser4_add_tree_root()) or deletion + (reiser4_kill_tree_root()) of tree root and morphing of paste into insert + (insert_paste()) have to be handled. + + 4. there is non-trivial interdependency between allocation of new nodes + and almost everything else. This is mainly due to the (1.c) above. I shall + write about this later. + +*/ + +#include "forward.h" +#include "debug.h" +#include "key.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "plugin/item/extent.h" +#include "plugin/node/node.h" +#include "jnode.h" +#include "znode.h" +#include "tree_mod.h" +#include "tree_walk.h" +#include "block_alloc.h" +#include "pool.h" +#include "tree.h" +#include "carry.h" +#include "carry_ops.h" +#include "super.h" +#include "reiser4.h" + +#include + +/* level locking/unlocking */ +static int lock_carry_level(carry_level * level); +static void unlock_carry_level(carry_level * level, int failure); +static void done_carry_level(carry_level * level); +static void unlock_carry_node(carry_level * level, carry_node * node, int fail); + +int lock_carry_node(carry_level * level, carry_node * node); +int lock_carry_node_tail(carry_node * node); + +/* carry processing proper */ +static int carry_on_level(carry_level * doing, carry_level * todo); + +static carry_op *add_op(carry_level * level, pool_ordering order, + carry_op * reference); + +/* handlers for carry operations. */ + +static void fatal_carry_error(carry_level * doing, int ecode); +static int add_new_root(carry_level * level, carry_node * node, znode * fake); + +static void print_level(const char *prefix, carry_level * level); + +#if REISER4_DEBUG +typedef enum { + CARRY_TODO, + CARRY_DOING +} carry_queue_state; +static int carry_level_invariant(carry_level * level, carry_queue_state state); +#endif + +/* main entry point for tree balancing. + + Tree carry performs operations from @doing and while doing so accumulates + information about operations to be performed on the next level ("carried" + to the parent level). Carried operations are performed, causing possibly + more operations to be carried upward etc. carry() takes care about + locking and pinning znodes while operating on them. + + For usage, see comment at the top of fs/reiser4/carry.c + +*/ +int reiser4_carry(carry_level * doing /* set of carry operations to be + * performed */ , + carry_level * done /* set of nodes, already performed + * at the previous level. + * NULL in most cases */) +{ + int result = 0; + /* queue of new requests */ + carry_level *todo; + ON_DEBUG(STORE_COUNTERS); + + assert("nikita-888", doing != NULL); + BUG_ON(done != NULL); + + todo = doing + 1; + init_carry_level(todo, doing->pool); + + /* queue of requests preformed on the previous level */ + done = todo + 1; + init_carry_level(done, doing->pool); + + /* iterate until there is nothing more to do */ + while (result == 0 && doing->ops_num > 0) { + carry_level *tmp; + + /* at this point @done is locked. */ + /* repeat lock/do/unlock while + + (1) lock_carry_level() fails due to deadlock avoidance, or + + (2) carry_on_level() decides that more nodes have to + be involved. + + (3) some unexpected error occurred while balancing on the + upper levels. In this case all changes are rolled back. + + */ + while (1) { + result = lock_carry_level(doing); + if (result == 0) { + /* perform operations from @doing and + accumulate new requests in @todo */ + result = carry_on_level(doing, todo); + if (result == 0) + break; + else if (result != -E_REPEAT || + !doing->restartable) { + warning("nikita-1043", + "Fatal error during carry: %i", + result); + print_level("done", done); + print_level("doing", doing); + print_level("todo", todo); + /* do some rough stuff like aborting + all pending transcrashes and thus + pushing tree back to the consistent + state. Alternatvely, just panic. + */ + fatal_carry_error(doing, result); + return result; + } + } else if (result != -E_REPEAT) { + fatal_carry_error(doing, result); + return result; + } + unlock_carry_level(doing, 1); + } + /* at this point @done can be safely unlocked */ + done_carry_level(done); + + /* cyclically shift queues */ + tmp = done; + done = doing; + doing = todo; + todo = tmp; + init_carry_level(todo, doing->pool); + + /* give other threads chance to run */ + reiser4_preempt_point(); + } + done_carry_level(done); + + /* all counters, but x_refs should remain the same. x_refs can change + owing to transaction manager */ + ON_DEBUG(CHECK_COUNTERS); + return result; +} + +/* perform carry operations on given level. + + Optimizations proposed by pooh: + + (1) don't lock all nodes from queue at the same time. Lock nodes lazily as + required; + + (2) unlock node if there are no more operations to be performed upon it and + node didn't add any operation to @todo. This can be implemented by + attaching to each node two counters: counter of operaions working on this + node and counter and operations carried upward from this node. + +*/ +static int carry_on_level(carry_level * doing /* queue of carry operations to + * do on this level */ , + carry_level * todo /* queue where new carry + * operations to be performed on + * the * parent level are + * accumulated during @doing + * processing. */ ) +{ + int result; + int (*f) (carry_op *, carry_level *, carry_level *); + carry_op *op; + carry_op *tmp_op; + + assert("nikita-1034", doing != NULL); + assert("nikita-1035", todo != NULL); + + /* @doing->nodes are locked. */ + + /* This function can be split into two phases: analysis and modification + + Analysis calculates precisely what items should be moved between + nodes. This information is gathered in some structures attached to + each carry_node in a @doing queue. Analysis also determines whether + new nodes are to be allocated etc. + + After analysis is completed, actual modification is performed. Here + we can take advantage of "batch modification": if there are several + operations acting on the same node, modifications can be performed + more efficiently when batched together. + + Above is an optimization left for the future. + */ + /* Important, but delayed optimization: it's possible to batch + operations together and perform them more efficiently as a + result. For example, deletion of several neighboring items from a + node can be converted to a single ->cut() operation. + + Before processing queue, it should be scanned and "mergeable" + operations merged. + */ + result = 0; + for_all_ops(doing, op, tmp_op) { + carry_opcode opcode; + + assert("nikita-1041", op != NULL); + opcode = op->op; + assert("nikita-1042", op->op < COP_LAST_OP); + f = op_dispatch_table[op->op].handler; + result = f(op, doing, todo); + /* locking can fail with -E_REPEAT. Any different error is fatal + and will be handled by fatal_carry_error() sledgehammer. + */ + if (result != 0) + break; + } + if (result == 0) { + carry_plugin_info info; + carry_node *scan; + carry_node *tmp_scan; + + info.doing = doing; + info.todo = todo; + + assert("nikita-3002", + carry_level_invariant(doing, CARRY_DOING)); + for_all_nodes(doing, scan, tmp_scan) { + znode *node; + + node = reiser4_carry_real(scan); + assert("nikita-2547", node != NULL); + if (node_is_empty(node)) { + result = + node_plugin_by_node(node)-> + prepare_removal(node, &info); + if (result != 0) + break; + } + } + } + return result; +} + +/* post carry operation + + This is main function used by external carry clients: node layout plugins + and tree operations to create new carry operation to be performed on some + level. + + New operation will be included in the @level queue. To actually perform it, + call carry( level, ... ). This function takes write lock on @node. Carry + manages all its locks by itself, don't worry about this. + + This function adds operation and node at the end of the queue. It is up to + caller to guarantee proper ordering of node queue. + +*/ +carry_op * reiser4_post_carry(carry_level * level /* queue where new operation + * is to be posted at */ , + carry_opcode op /* opcode of operation */ , + znode * node /* node on which this operation + * will operate */ , + int apply_to_parent_p /* whether operation will + * operate directly on @node + * or on it parent. */) +{ + carry_op *result; + carry_node *child; + + assert("nikita-1046", level != NULL); + assert("nikita-1788", znode_is_write_locked(node)); + + result = add_op(level, POOLO_LAST, NULL); + if (IS_ERR(result)) + return result; + child = reiser4_add_carry(level, POOLO_LAST, NULL); + if (IS_ERR(child)) { + reiser4_pool_free(&level->pool->op_pool, &result->header); + return (carry_op *) child; + } + result->node = child; + result->op = op; + child->parent = apply_to_parent_p; + if (ZF_ISSET(node, JNODE_ORPHAN)) + child->left_before = 1; + child->node = node; + return result; +} + +/* initialize carry queue */ +void init_carry_level(carry_level * level /* level to initialize */ , + carry_pool * pool /* pool @level will allocate objects + * from */ ) +{ + assert("nikita-1045", level != NULL); + assert("nikita-967", pool != NULL); + + memset(level, 0, sizeof *level); + level->pool = pool; + + INIT_LIST_HEAD(&level->nodes); + INIT_LIST_HEAD(&level->ops); +} + +/* allocate carry pool and initialize pools within queue */ +carry_pool *init_carry_pool(int size) +{ + carry_pool *pool; + + assert("", size >= sizeof(carry_pool) + 3 * sizeof(carry_level)); + pool = kmalloc(size, reiser4_ctx_gfp_mask_get()); + if (pool == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + + reiser4_init_pool(&pool->op_pool, sizeof(carry_op), CARRIES_POOL_SIZE, + (char *)pool->op); + reiser4_init_pool(&pool->node_pool, sizeof(carry_node), + NODES_LOCKED_POOL_SIZE, (char *)pool->node); + return pool; +} + +/* finish with queue pools */ +void done_carry_pool(carry_pool * pool/* pool to destroy */) +{ + reiser4_done_pool(&pool->op_pool); + reiser4_done_pool(&pool->node_pool); + kfree(pool); +} + +/* add new carry node to the @level. + + Returns pointer to the new carry node allocated from pool. It's up to + callers to maintain proper order in the @level. Assumption is that if carry + nodes on one level are already sorted and modifications are peroformed from + left to right, carry nodes added on the parent level will be ordered + automatically. To control ordering use @order and @reference parameters. + +*/ +carry_node *reiser4_add_carry_skip(carry_level * level /* &carry_level to add + * node to */ , + pool_ordering order /* where to insert: + * at the beginning of + * @level, + * before @reference, + * after @reference, + * at the end of @level + */ , + carry_node * reference/* reference node for + * insertion */) +{ + ON_DEBUG(carry_node * orig_ref = reference); + + if (order == POOLO_BEFORE) { + reference = find_left_carry(reference, level); + if (reference == NULL) + reference = list_entry(level->nodes.next, carry_node, + header.level_linkage); + else + reference = list_entry(reference->header.level_linkage.next, + carry_node, header.level_linkage); + } else if (order == POOLO_AFTER) { + reference = find_right_carry(reference, level); + if (reference == NULL) + reference = list_entry(level->nodes.prev, carry_node, + header.level_linkage); + else + reference = list_entry(reference->header.level_linkage.prev, + carry_node, header.level_linkage); + } + assert("nikita-2209", + ergo(orig_ref != NULL, + reiser4_carry_real(reference) == + reiser4_carry_real(orig_ref))); + return reiser4_add_carry(level, order, reference); +} + +carry_node *reiser4_add_carry(carry_level * level, /* carry_level to add + node to */ + pool_ordering order, /* where to insert: + * at the beginning of + * @level; + * before @reference; + * after @reference; + * at the end of @level + */ + carry_node * reference /* reference node for + * insertion */) +{ + carry_node *result; + + result = + (carry_node *) reiser4_add_obj(&level->pool->node_pool, + &level->nodes, + order, &reference->header); + if (!IS_ERR(result) && (result != NULL)) + ++level->nodes_num; + return result; +} + +/** + * add new carry operation to the @level. + * + * Returns pointer to the new carry operations allocated from pool. It's up to + * callers to maintain proper order in the @level. To control ordering use + * @order and @reference parameters. + */ +static carry_op *add_op(carry_level * level, /* &carry_level to add node to */ + pool_ordering order, /* where to insert: + * at the beginning of @level; + * before @reference; + * after @reference; + * at the end of @level */ + carry_op * reference /* reference node for insertion */) +{ + carry_op *result; + + result = + (carry_op *) reiser4_add_obj(&level->pool->op_pool, &level->ops, + order, &reference->header); + if (!IS_ERR(result) && (result != NULL)) + ++level->ops_num; + return result; +} + +/** + * Return node on the right of which @node was created. + * + * Each node is created on the right of some existing node (or it is new root, + * which is special case not handled here). + * + * @node is new node created on some level, but not yet inserted into its + * parent, it has corresponding bit (JNODE_ORPHAN) set in zstate. + */ +static carry_node *find_begetting_brother(carry_node * node,/* node to start + search from */ + carry_level * kin UNUSED_ARG + /* level to scan */) +{ + carry_node *scan; + + assert("nikita-1614", node != NULL); + assert("nikita-1615", kin != NULL); + assert("nikita-1616", LOCK_CNT_GTZ(rw_locked_tree)); + assert("nikita-1619", ergo(reiser4_carry_real(node) != NULL, + ZF_ISSET(reiser4_carry_real(node), + JNODE_ORPHAN))); + for (scan = node;; + scan = list_entry(scan->header.level_linkage.prev, carry_node, + header.level_linkage)) { + assert("nikita-1617", &kin->nodes != &scan->header.level_linkage); + if ((scan->node != node->node) && + !ZF_ISSET(scan->node, JNODE_ORPHAN)) { + assert("nikita-1618", reiser4_carry_real(scan) != NULL); + break; + } + } + return scan; +} + +static cmp_t +carry_node_cmp(carry_level * level, carry_node * n1, carry_node * n2) +{ + assert("nikita-2199", n1 != NULL); + assert("nikita-2200", n2 != NULL); + + if (n1 == n2) + return EQUAL_TO; + while (1) { + n1 = carry_node_next(n1); + if (carry_node_end(level, n1)) + return GREATER_THAN; + if (n1 == n2) + return LESS_THAN; + } + impossible("nikita-2201", "End of level reached"); +} + +carry_node *find_carry_node(carry_level * level, const znode * node) +{ + carry_node *scan; + carry_node *tmp_scan; + + assert("nikita-2202", level != NULL); + assert("nikita-2203", node != NULL); + + for_all_nodes(level, scan, tmp_scan) { + if (reiser4_carry_real(scan) == node) + return scan; + } + return NULL; +} + +znode *reiser4_carry_real(const carry_node * node) +{ + assert("nikita-3061", node != NULL); + + return node->lock_handle.node; +} + +carry_node *insert_carry_node(carry_level * doing, carry_level * todo, + const znode * node) +{ + carry_node *base; + carry_node *scan; + carry_node *tmp_scan; + carry_node *proj; + + base = find_carry_node(doing, node); + assert("nikita-2204", base != NULL); + + for_all_nodes(todo, scan, tmp_scan) { + proj = find_carry_node(doing, scan->node); + assert("nikita-2205", proj != NULL); + if (carry_node_cmp(doing, proj, base) != LESS_THAN) + break; + } + return scan; +} + +static carry_node *add_carry_atplace(carry_level * doing, carry_level * todo, + znode * node) +{ + carry_node *reference; + + assert("nikita-2994", doing != NULL); + assert("nikita-2995", todo != NULL); + assert("nikita-2996", node != NULL); + + reference = insert_carry_node(doing, todo, node); + assert("nikita-2997", reference != NULL); + + return reiser4_add_carry(todo, POOLO_BEFORE, reference); +} + +/* like reiser4_post_carry(), but designed to be called from node plugin + methods. This function is different from reiser4_post_carry() in that it + finds proper place to insert node in the queue. */ +carry_op *node_post_carry(carry_plugin_info * info /* carry parameters + * passed down to node + * plugin */ , + carry_opcode op /* opcode of operation */ , + znode * node /* node on which this + * operation will operate */ , + int apply_to_parent_p /* whether operation will + * operate directly on @node + * or on it parent. */ ) +{ + carry_op *result; + carry_node *child; + + assert("nikita-2207", info != NULL); + assert("nikita-2208", info->todo != NULL); + + if (info->doing == NULL) + return reiser4_post_carry(info->todo, op, node, + apply_to_parent_p); + + result = add_op(info->todo, POOLO_LAST, NULL); + if (IS_ERR(result)) + return result; + child = add_carry_atplace(info->doing, info->todo, node); + if (IS_ERR(child)) { + reiser4_pool_free(&info->todo->pool->op_pool, &result->header); + return (carry_op *) child; + } + result->node = child; + result->op = op; + child->parent = apply_to_parent_p; + if (ZF_ISSET(node, JNODE_ORPHAN)) + child->left_before = 1; + child->node = node; + return result; +} + +/* lock all carry nodes in @level */ +static int lock_carry_level(carry_level * level/* level to lock */) +{ + int result; + carry_node *node; + carry_node *tmp_node; + + assert("nikita-881", level != NULL); + assert("nikita-2229", carry_level_invariant(level, CARRY_TODO)); + + /* lock nodes from left to right */ + result = 0; + for_all_nodes(level, node, tmp_node) { + result = lock_carry_node(level, node); + if (result != 0) + break; + } + return result; +} + +/* Synchronize delimiting keys between @node and its left neighbor. + + To reduce contention on dk key and simplify carry code, we synchronize + delimiting keys only when carry ultimately leaves tree level (carrying + changes upward) and unlocks nodes at this level. + + This function first finds left neighbor of @node and then updates left + neighbor's right delimiting key to conincide with least key in @node. + +*/ + +ON_DEBUG(extern atomic_t delim_key_version; + ) + +static void sync_dkeys(znode * spot/* node to update */) +{ + reiser4_key pivot; + reiser4_tree *tree; + + assert("nikita-1610", spot != NULL); + assert("nikita-1612", LOCK_CNT_NIL(rw_locked_dk)); + + tree = znode_get_tree(spot); + read_lock_tree(tree); + write_lock_dk(tree); + + assert("nikita-2192", znode_is_loaded(spot)); + + /* sync left delimiting key of @spot with key in its leftmost item */ + if (node_is_empty(spot)) + pivot = *znode_get_rd_key(spot); + else + leftmost_key_in_node(spot, &pivot); + + znode_set_ld_key(spot, &pivot); + + /* there can be sequence of empty nodes pending removal on the left of + @spot. Scan them and update their left and right delimiting keys to + match left delimiting key of @spot. Also, update right delimiting + key of first non-empty left neighbor. + */ + while (1) { + if (!ZF_ISSET(spot, JNODE_LEFT_CONNECTED)) + break; + + spot = spot->left; + if (spot == NULL) + break; + + znode_set_rd_key(spot, &pivot); + /* don't sink into the domain of another balancing */ + if (!znode_is_write_locked(spot)) + break; + if (ZF_ISSET(spot, JNODE_HEARD_BANSHEE)) + znode_set_ld_key(spot, &pivot); + else + break; + } + + write_unlock_dk(tree); + read_unlock_tree(tree); +} + +/* unlock all carry nodes in @level */ +static void unlock_carry_level(carry_level * level /* level to unlock */ , + int failure /* true if unlocking owing to + * failure */ ) +{ + carry_node *node; + carry_node *tmp_node; + + assert("nikita-889", level != NULL); + + if (!failure) { + znode *spot; + + spot = NULL; + /* update delimiting keys */ + for_all_nodes(level, node, tmp_node) { + if (reiser4_carry_real(node) != spot) { + spot = reiser4_carry_real(node); + sync_dkeys(spot); + } + } + } + + /* nodes can be unlocked in arbitrary order. In preemptible + environment it's better to unlock in reverse order of locking, + though. + */ + for_all_nodes_back(level, node, tmp_node) { + /* all allocated nodes should be already linked to their + parents at this moment. */ + assert("nikita-1631", + ergo(!failure, !ZF_ISSET(reiser4_carry_real(node), + JNODE_ORPHAN))); + ON_DEBUG(check_dkeys(reiser4_carry_real(node))); + unlock_carry_node(level, node, failure); + } + level->new_root = NULL; +} + +/* finish with @level + + Unlock nodes and release all allocated resources */ +static void done_carry_level(carry_level * level/* level to finish */) +{ + carry_node *node; + carry_node *tmp_node; + carry_op *op; + carry_op *tmp_op; + + assert("nikita-1076", level != NULL); + + unlock_carry_level(level, 0); + for_all_nodes(level, node, tmp_node) { + assert("nikita-2113", list_empty_careful(&node->lock_handle.locks_link)); + assert("nikita-2114", list_empty_careful(&node->lock_handle.owners_link)); + reiser4_pool_free(&level->pool->node_pool, &node->header); + } + for_all_ops(level, op, tmp_op) + reiser4_pool_free(&level->pool->op_pool, &op->header); +} + +/* helper function to complete locking of carry node + + Finish locking of carry node. There are several ways in which new carry + node can be added into carry level and locked. Normal is through + lock_carry_node(), but also from find_{left|right}_neighbor(). This + function factors out common final part of all locking scenarios. It + supposes that @node -> lock_handle is lock handle for lock just taken and + fills ->real_node from this lock handle. + +*/ +int lock_carry_node_tail(carry_node * node/* node to complete locking of */) +{ + assert("nikita-1052", node != NULL); + assert("nikita-1187", reiser4_carry_real(node) != NULL); + assert("nikita-1188", !node->unlock); + + node->unlock = 1; + /* Load node content into memory and install node plugin by + looking at the node header. + + Most of the time this call is cheap because the node is + already in memory. + + Corresponding zrelse() is in unlock_carry_node() + */ + return zload(reiser4_carry_real(node)); +} + +/* lock carry node + + "Resolve" node to real znode, lock it and mark as locked. + This requires recursive locking of znodes. + + When operation is posted to the parent level, node it will be applied to is + not yet known. For example, when shifting data between two nodes, + delimiting has to be updated in parent or parents of nodes involved. But + their parents is not yet locked and, moreover said nodes can be reparented + by concurrent balancing. + + To work around this, carry operation is applied to special "carry node" + rather than to the znode itself. Carry node consists of some "base" or + "reference" znode and flags indicating how to get to the target of carry + operation (->real_node field of carry_node) from base. + +*/ +int lock_carry_node(carry_level * level /* level @node is in */ , + carry_node * node/* node to lock */) +{ + int result; + znode *reference_point; + lock_handle lh; + lock_handle tmp_lh; + reiser4_tree *tree; + + assert("nikita-887", level != NULL); + assert("nikita-882", node != NULL); + + result = 0; + reference_point = node->node; + init_lh(&lh); + init_lh(&tmp_lh); + if (node->left_before) { + /* handling of new nodes, allocated on the previous level: + + some carry ops were propably posted from the new node, but + this node neither has parent pointer set, nor is + connected. This will be done in ->create_hook() for + internal item. + + No then less, parent of new node has to be locked. To do + this, first go to the "left" in the carry order. This + depends on the decision to always allocate new node on the + right of existing one. + + Loop handles case when multiple nodes, all orphans, were + inserted. + + Strictly speaking, taking tree lock is not necessary here, + because all nodes scanned by loop in + find_begetting_brother() are write-locked by this thread, + and thus, their sibling linkage cannot change. + + */ + tree = znode_get_tree(reference_point); + read_lock_tree(tree); + reference_point = find_begetting_brother(node, level)->node; + read_unlock_tree(tree); + assert("nikita-1186", reference_point != NULL); + } + if (node->parent && (result == 0)) { + result = + reiser4_get_parent(&tmp_lh, reference_point, + ZNODE_WRITE_LOCK); + if (result != 0) { + ; /* nothing */ + } else if (znode_get_level(tmp_lh.node) == 0) { + assert("nikita-1347", znode_above_root(tmp_lh.node)); + result = add_new_root(level, node, tmp_lh.node); + if (result == 0) { + reference_point = level->new_root; + move_lh(&lh, &node->lock_handle); + } + } else if ((level->new_root != NULL) + && (level->new_root != + znode_parent_nolock(reference_point))) { + /* parent of node exists, but this level aready + created different new root, so */ + warning("nikita-1109", + /* it should be "radicis", but tradition is + tradition. do banshees read latin? */ + "hodie natus est radici frater"); + result = -EIO; + } else { + move_lh(&lh, &tmp_lh); + reference_point = lh.node; + } + } + if (node->left && (result == 0)) { + assert("nikita-1183", node->parent); + assert("nikita-883", reference_point != NULL); + result = + reiser4_get_left_neighbor(&tmp_lh, reference_point, + ZNODE_WRITE_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (result == 0) { + done_lh(&lh); + move_lh(&lh, &tmp_lh); + reference_point = lh.node; + } + } + if (!node->parent && !node->left && !node->left_before) { + result = + longterm_lock_znode(&lh, reference_point, ZNODE_WRITE_LOCK, + ZNODE_LOCK_HIPRI); + } + if (result == 0) { + move_lh(&node->lock_handle, &lh); + result = lock_carry_node_tail(node); + } + done_lh(&tmp_lh); + done_lh(&lh); + return result; +} + +/* release a lock on &carry_node. + + Release if necessary lock on @node. This opearion is pair of + lock_carry_node() and is idempotent: you can call it more than once on the + same node. + +*/ +static void +unlock_carry_node(carry_level * level, + carry_node * node /* node to be released */ , + int failure /* 0 if node is unlocked due + * to some error */ ) +{ + znode *real_node; + + assert("nikita-884", node != NULL); + + real_node = reiser4_carry_real(node); + /* pair to zload() in lock_carry_node_tail() */ + zrelse(real_node); + if (node->unlock && (real_node != NULL)) { + assert("nikita-899", real_node == node->lock_handle.node); + longterm_unlock_znode(&node->lock_handle); + } + if (failure) { + if (node->deallocate && (real_node != NULL)) { + /* free node in bitmap + + Prepare node for removal. Last zput() will finish + with it. + */ + ZF_SET(real_node, JNODE_HEARD_BANSHEE); + } + if (node->free) { + assert("nikita-2177", + list_empty_careful(&node->lock_handle.locks_link)); + assert("nikita-2112", + list_empty_careful(&node->lock_handle.owners_link)); + reiser4_pool_free(&level->pool->node_pool, + &node->header); + } + } +} + +/* fatal_carry_error() - all-catching error handling function + + It is possible that carry faces unrecoverable error, like unability to + insert pointer at the internal level. Our simple solution is just panic in + this situation. More sophisticated things like attempt to remount + file-system as read-only can be implemented without much difficlties. + + It is believed, that: + + 1. in stead of panicking, all current transactions can be aborted rolling + system back to the consistent state. + +Umm, if you simply panic without doing anything more at all, then all current +transactions are aborted and the system is rolled back to a consistent state, +by virtue of the design of the transactional mechanism. Well, wait, let's be +precise. If an internal node is corrupted on disk due to hardware failure, +then there may be no consistent state that can be rolled back to, so instead +we should say that it will rollback the transactions, which barring other +factors means rolling back to a consistent state. + +# Nikita: there is a subtle difference between panic and aborting +# transactions: machine doesn't reboot. Processes aren't killed. Processes +# don't using reiser4 (not that we care about such processes), or using other +# reiser4 mounts (about them we do care) will simply continue to run. With +# some luck, even application using aborted file system can survive: it will +# get some error, like EBADF, from each file descriptor on failed file system, +# but applications that do care about tolerance will cope with this (squid +# will). + +It would be a nice feature though to support rollback without rebooting +followed by remount, but this can wait for later versions. + + 2. once isolated transactions will be implemented it will be possible to + roll back offending transaction. + +2. is additional code complexity of inconsistent value (it implies that a +broken tree should be kept in operation), so we must think about it more +before deciding if it should be done. -Hans + +*/ +static void fatal_carry_error(carry_level * doing UNUSED_ARG /* carry level + * where + * unrecoverable + * error + * occurred */ , + int ecode/* error code */) +{ + assert("nikita-1230", doing != NULL); + assert("nikita-1231", ecode < 0); + + reiser4_panic("nikita-1232", "Carry failed: %i", ecode); +} + +/** + * Add new root to the tree + * + * This function itself only manages changes in carry structures and delegates + * all hard work (allocation of znode for new root, changes of parent and + * sibling pointers) to the reiser4_add_tree_root(). + * + * Locking: old tree root is locked by carry at this point. Fake znode is also + * locked. + */ +static int add_new_root(carry_level * level,/* carry level in context of which + * operation is performed */ + carry_node * node, /* carry node for existing root */ + znode * fake /* "fake" znode already locked by + * us */) +{ + int result; + + assert("nikita-1104", level != NULL); + assert("nikita-1105", node != NULL); + + assert("nikita-1403", znode_is_write_locked(node->node)); + assert("nikita-1404", znode_is_write_locked(fake)); + + /* trying to create new root. */ + /* @node is root and it's already locked by us. This + means that nobody else can be trying to add/remove + tree root right now. + */ + if (level->new_root == NULL) + level->new_root = reiser4_add_tree_root(node->node, fake); + if (!IS_ERR(level->new_root)) { + assert("nikita-1210", znode_is_root(level->new_root)); + node->deallocate = 1; + result = + longterm_lock_znode(&node->lock_handle, level->new_root, + ZNODE_WRITE_LOCK, ZNODE_LOCK_LOPRI); + if (result == 0) + zput(level->new_root); + } else { + result = PTR_ERR(level->new_root); + level->new_root = NULL; + } + return result; +} + +/* allocate new znode and add the operation that inserts the + pointer to it into the parent node into the todo level + + Allocate new znode, add it into carry queue and post into @todo queue + request to add pointer to new node into its parent. + + This is carry related routing that calls reiser4_new_node() to allocate new + node. +*/ +carry_node *add_new_znode(znode * brother /* existing left neighbor of new + * node */ , + carry_node * ref /* carry node after which new + * carry node is to be inserted + * into queue. This affects + * locking. */ , + carry_level * doing /* carry queue where new node is + * to be added */ , + carry_level * todo /* carry queue where COP_INSERT + * operation to add pointer to + * new node will ne added */ ) +{ + carry_node *fresh; + znode *new_znode; + carry_op *add_pointer; + carry_plugin_info info; + + assert("nikita-1048", brother != NULL); + assert("nikita-1049", todo != NULL); + + /* There is a lot of possible variations here: to what parent + new node will be attached and where. For simplicity, always + do the following: + + (1) new node and @brother will have the same parent. + + (2) new node is added on the right of @brother + + */ + + fresh = reiser4_add_carry_skip(doing, + ref ? POOLO_AFTER : POOLO_LAST, ref); + if (IS_ERR(fresh)) + return fresh; + + fresh->deallocate = 1; + fresh->free = 1; + + new_znode = reiser4_new_node(brother, znode_get_level(brother)); + if (IS_ERR(new_znode)) + /* @fresh will be deallocated automatically by error + handling code in the caller. */ + return (carry_node *) new_znode; + + /* new_znode returned znode with x_count 1. Caller has to decrease + it. make_space() does. */ + + ZF_SET(new_znode, JNODE_ORPHAN); + fresh->node = new_znode; + + while (ZF_ISSET(reiser4_carry_real(ref), JNODE_ORPHAN)) { + ref = carry_node_prev(ref); + assert("nikita-1606", !carry_node_end(doing, ref)); + } + + info.todo = todo; + info.doing = doing; + add_pointer = node_post_carry(&info, COP_INSERT, + reiser4_carry_real(ref), 1); + if (IS_ERR(add_pointer)) { + /* no need to deallocate @new_znode here: it will be + deallocated during carry error handling. */ + return (carry_node *) add_pointer; + } + + add_pointer->u.insert.type = COPT_CHILD; + add_pointer->u.insert.child = fresh; + add_pointer->u.insert.brother = brother; + /* initially new node spawns empty key range */ + write_lock_dk(znode_get_tree(brother)); + znode_set_ld_key(new_znode, + znode_set_rd_key(new_znode, + znode_get_rd_key(brother))); + write_unlock_dk(znode_get_tree(brother)); + return fresh; +} + +/* DEBUGGING FUNCTIONS. + + Probably we also should leave them on even when + debugging is turned off to print dumps at errors. +*/ +#if REISER4_DEBUG +static int carry_level_invariant(carry_level * level, carry_queue_state state) +{ + carry_node *node; + carry_node *tmp_node; + + if (level == NULL) + return 0; + + if (level->track_type != 0 && + level->track_type != CARRY_TRACK_NODE && + level->track_type != CARRY_TRACK_CHANGE) + return 0; + + /* check that nodes are in ascending order */ + for_all_nodes(level, node, tmp_node) { + znode *left; + znode *right; + + reiser4_key lkey; + reiser4_key rkey; + + if (node != carry_node_front(level)) { + if (state == CARRY_TODO) { + right = node->node; + left = carry_node_prev(node)->node; + } else { + right = reiser4_carry_real(node); + left = reiser4_carry_real(carry_node_prev(node)); + } + if (right == NULL || left == NULL) + continue; + if (node_is_empty(right) || node_is_empty(left)) + continue; + if (!keyle(leftmost_key_in_node(left, &lkey), + leftmost_key_in_node(right, &rkey))) { + warning("", "wrong key order"); + return 0; + } + } + } + return 1; +} +#endif + +/* get symbolic name for boolean */ +static const char *tf(int boolean/* truth value */) +{ + return boolean ? "t" : "f"; +} + +/* symbolic name for carry operation */ +static const char *carry_op_name(carry_opcode op/* carry opcode */) +{ + switch (op) { + case COP_INSERT: + return "COP_INSERT"; + case COP_DELETE: + return "COP_DELETE"; + case COP_CUT: + return "COP_CUT"; + case COP_PASTE: + return "COP_PASTE"; + case COP_UPDATE: + return "COP_UPDATE"; + case COP_EXTENT: + return "COP_EXTENT"; + case COP_INSERT_FLOW: + return "COP_INSERT_FLOW"; + default:{ + /* not mt safe, but who cares? */ + static char buf[20]; + + sprintf(buf, "unknown op: %x", op); + return buf; + } + } +} + +/* dump information about carry node */ +static void print_carry(const char *prefix /* prefix to print */ , + carry_node * node/* node to print */) +{ + if (node == NULL) { + printk("%s: null\n", prefix); + return; + } + printk + ("%s: %p parent: %s, left: %s, unlock: %s, free: %s, dealloc: %s\n", + prefix, node, tf(node->parent), tf(node->left), tf(node->unlock), + tf(node->free), tf(node->deallocate)); +} + +/* dump information about carry operation */ +static void print_op(const char *prefix /* prefix to print */ , + carry_op * op/* operation to print */) +{ + if (op == NULL) { + printk("%s: null\n", prefix); + return; + } + printk("%s: %p carry_opcode: %s\n", prefix, op, carry_op_name(op->op)); + print_carry("\tnode", op->node); + switch (op->op) { + case COP_INSERT: + case COP_PASTE: + print_coord("\tcoord", + op->u.insert.d ? op->u.insert.d->coord : NULL, 0); + reiser4_print_key("\tkey", + op->u.insert.d ? op->u.insert.d->key : NULL); + print_carry("\tchild", op->u.insert.child); + break; + case COP_DELETE: + print_carry("\tchild", op->u.delete.child); + break; + case COP_CUT: + if (op->u.cut_or_kill.is_cut) { + print_coord("\tfrom", + op->u.cut_or_kill.u.kill->params.from, 0); + print_coord("\tto", op->u.cut_or_kill.u.kill->params.to, + 0); + } else { + print_coord("\tfrom", + op->u.cut_or_kill.u.cut->params.from, 0); + print_coord("\tto", op->u.cut_or_kill.u.cut->params.to, + 0); + } + break; + case COP_UPDATE: + print_carry("\tleft", op->u.update.left); + break; + default: + /* do nothing */ + break; + } +} + +/* dump information about all nodes and operations in a @level */ +static void print_level(const char *prefix /* prefix to print */ , + carry_level * level/* level to print */) +{ + carry_node *node; + carry_node *tmp_node; + carry_op *op; + carry_op *tmp_op; + + if (level == NULL) { + printk("%s: null\n", prefix); + return; + } + printk("%s: %p, restartable: %s\n", + prefix, level, tf(level->restartable)); + + for_all_nodes(level, node, tmp_node) + print_carry("\tcarry node", node); + for_all_ops(level, op, tmp_op) + print_op("\tcarry op", op); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/carry.h linux-2.6.30/fs/reiser4/carry.h --- linux-2.6.30.orig/fs/reiser4/carry.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/carry.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,445 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Functions and data types to "carry" tree modification(s) upward. + See fs/reiser4/carry.c for details. */ + +#if !defined(__FS_REISER4_CARRY_H__) +#define __FS_REISER4_CARRY_H__ + +#include "forward.h" +#include "debug.h" +#include "pool.h" +#include "znode.h" + +#include + +/* &carry_node - "location" of carry node. + + "location" of node that is involved or going to be involved into + carry process. Node where operation will be carried to on the + parent level cannot be recorded explicitly. Operation will be carried + usually to the parent of some node (where changes are performed at + the current level) or, to the left neighbor of its parent. But while + modifications are performed at the current level, parent may + change. So, we have to allow some indirection (or, positevly, + flexibility) in locating carry nodes. + +*/ +typedef struct carry_node { + /* pool linkage */ + struct reiser4_pool_header header; + + /* base node from which real_node is calculated. See + fs/reiser4/carry.c:lock_carry_node(). */ + znode *node; + + /* how to get ->real_node */ + /* to get ->real_node obtain parent of ->node */ + __u32 parent:1; + /* to get ->real_node obtain left neighbor of parent of + ->node */ + __u32 left:1; + __u32 left_before:1; + + /* locking */ + + /* this node was locked by carry process and should be + unlocked when carry leaves a level */ + __u32 unlock:1; + + /* disk block for this node was allocated by carry process and + should be deallocated when carry leaves a level */ + __u32 deallocate:1; + /* this carry node was allocated by carry process and should be + freed when carry leaves a level */ + __u32 free:1; + + /* type of lock we want to take on this node */ + lock_handle lock_handle; +} carry_node; + +/* &carry_opcode - elementary operations that can be carried upward + + Operations that carry() can handle. This list is supposed to be + expanded. + + Each carry operation (cop) is handled by appropriate function defined + in fs/reiser4/carry.c. For example COP_INSERT is handled by + fs/reiser4/carry.c:carry_insert() etc. These functions in turn + call plugins of nodes affected by operation to modify nodes' content + and to gather operations to be performed on the next level. + +*/ +typedef enum { + /* insert new item into node. */ + COP_INSERT, + /* delete pointer from parent node */ + COP_DELETE, + /* remove part of or whole node. */ + COP_CUT, + /* increase size of item. */ + COP_PASTE, + /* insert extent (that is sequence of unformatted nodes). */ + COP_EXTENT, + /* update delimiting key in least common ancestor of two + nodes. This is performed when items are moved between two + nodes. + */ + COP_UPDATE, + /* insert flow */ + COP_INSERT_FLOW, + COP_LAST_OP, +} carry_opcode; + +#define CARRY_FLOW_NEW_NODES_LIMIT 20 + +/* mode (or subtype) of COP_{INSERT|PASTE} operation. Specifies how target + item is determined. */ +typedef enum { + /* target item is one containing pointer to the ->child node */ + COPT_CHILD, + /* target item is given explicitly by @coord */ + COPT_ITEM_DATA, + /* target item is given by key */ + COPT_KEY, + /* see insert_paste_common() for more comments on this. */ + COPT_PASTE_RESTARTED, +} cop_insert_pos_type; + +/* flags to cut and delete */ +typedef enum { + /* don't kill node even if it became completely empty as results of + * cut. This is needed for eottl handling. See carry_extent() for + * details. */ + DELETE_RETAIN_EMPTY = (1 << 0) +} cop_delete_flag; + +/* + * carry() implements "lock handle tracking" feature. + * + * Callers supply carry with node where to perform initial operation and lock + * handle on this node. Trying to optimize node utilization carry may actually + * move insertion point to different node. Callers expect that lock handle + * will rebe transferred to the new node also. + * + */ +typedef enum { + /* transfer lock handle along with insertion point */ + CARRY_TRACK_CHANGE = 1, + /* acquire new lock handle to the node where insertion point is. This + * is used when carry() client doesn't initially possess lock handle + * on the insertion point node, for example, by extent insertion + * code. See carry_extent(). */ + CARRY_TRACK_NODE = 2 +} carry_track_type; + +/* data supplied to COP_{INSERT|PASTE} by callers */ +typedef struct carry_insert_data { + /* position where new item is to be inserted */ + coord_t *coord; + /* new item description */ + reiser4_item_data * data; + /* key of new item */ + const reiser4_key * key; +} carry_insert_data; + +/* cut and kill are similar, so carry_cut_data and carry_kill_data share the + below structure of parameters */ +struct cut_kill_params { + /* coord where cut starts (inclusive) */ + coord_t *from; + /* coord where cut stops (inclusive, this item/unit will also be + * cut) */ + coord_t *to; + /* starting key. This is necessary when item and unit pos don't + * uniquely identify what portion or tree to remove. For example, this + * indicates what portion of extent unit will be affected. */ + const reiser4_key * from_key; + /* exclusive stop key */ + const reiser4_key * to_key; + /* if this is not NULL, smallest actually removed key is stored + * here. */ + reiser4_key *smallest_removed; + /* kill_node_content() is called for file truncate */ + int truncate; +}; + +struct carry_cut_data { + struct cut_kill_params params; +}; + +struct carry_kill_data { + struct cut_kill_params params; + /* parameter to be passed to the ->kill_hook() method of item + * plugin */ + /*void *iplug_params; *//* FIXME: unused currently */ + /* if not NULL---inode whose items are being removed. This is needed + * for ->kill_hook() of extent item to update VM structures when + * removing pages. */ + struct inode *inode; + /* sibling list maintenance is complicated by existence of eottl. When + * eottl whose left and right neighbors are formatted leaves is + * removed, one has to connect said leaves in the sibling list. This + * cannot be done when extent removal is just started as locking rules + * require sibling list update to happen atomically with removal of + * extent item. Therefore: 1. pointers to left and right neighbors + * have to be passed down to the ->kill_hook() of extent item, and + * 2. said neighbors have to be locked. */ + lock_handle *left; + lock_handle *right; + /* flags modifying behavior of kill. Currently, it may have + DELETE_RETAIN_EMPTY set. */ + unsigned flags; + char *buf; +}; + +/* &carry_tree_op - operation to "carry" upward. + + Description of an operation we want to "carry" to the upper level of + a tree: e.g, when we insert something and there is not enough space + we allocate a new node and "carry" the operation of inserting a + pointer to the new node to the upper level, on removal of empty node, + we carry up operation of removing appropriate entry from parent. + + There are two types of carry ops: when adding or deleting node we + node at the parent level where appropriate modification has to be + performed is known in advance. When shifting items between nodes + (split, merge), delimiting key should be changed in the least common + parent of the nodes involved that is not known in advance. + + For the operations of the first type we store in &carry_op pointer to + the &carry_node at the parent level. For the operation of the second + type we store &carry_node or parents of the left and right nodes + modified and keep track of them upward until they coincide. + +*/ +typedef struct carry_op { + /* pool linkage */ + struct reiser4_pool_header header; + carry_opcode op; + /* node on which operation is to be performed: + + for insert, paste: node where new item is to be inserted + + for delete: node where pointer is to be deleted + + for cut: node to cut from + + for update: node where delimiting key is to be modified + + for modify: parent of modified node + + */ + carry_node *node; + union { + struct { + /* (sub-)type of insertion/paste. Taken from + cop_insert_pos_type. */ + __u8 type; + /* various operation flags. Taken from + cop_insert_flag. */ + __u8 flags; + carry_insert_data *d; + carry_node *child; + znode *brother; + } insert, paste, extent; + + struct { + int is_cut; + union { + carry_kill_data *kill; + carry_cut_data *cut; + } u; + } cut_or_kill; + + struct { + carry_node *left; + } update; + struct { + /* changed child */ + carry_node *child; + /* bitmask of changes. See &cop_modify_flag */ + __u32 flag; + } modify; + struct { + /* flags to deletion operation. Are taken from + cop_delete_flag */ + __u32 flags; + /* child to delete from parent. If this is + NULL, delete op->node. */ + carry_node *child; + } delete; + struct { + /* various operation flags. Taken from + cop_insert_flag. */ + __u32 flags; + flow_t *flow; + coord_t *insert_point; + reiser4_item_data *data; + /* flow insertion is limited by number of new blocks + added in that operation which do not get any data + but part of flow. This limit is set by macro + CARRY_FLOW_NEW_NODES_LIMIT. This field stores number + of nodes added already during one carry_flow */ + int new_nodes; + } insert_flow; + } u; +} carry_op; + +/* &carry_op_pool - preallocated pool of carry operations, and nodes */ +typedef struct carry_pool { + carry_op op[CARRIES_POOL_SIZE]; + struct reiser4_pool op_pool; + carry_node node[NODES_LOCKED_POOL_SIZE]; + struct reiser4_pool node_pool; +} carry_pool; + +/* &carry_tree_level - carry process on given level + + Description of balancing process on the given level. + + No need for locking here, as carry_tree_level is essentially per + thread thing (for now). + +*/ +struct carry_level { + /* this level may be restarted */ + __u32 restartable:1; + /* list of carry nodes on this level, ordered by key order */ + struct list_head nodes; + struct list_head ops; + /* pool where new objects are allocated from */ + carry_pool *pool; + int ops_num; + int nodes_num; + /* new root created on this level, if any */ + znode *new_root; + /* This is set by caller (insert_by_key(), rreiser4_esize_item(), etc.) + when they want ->tracked to automagically wander to the node where + insertion point moved after insert or paste. + */ + carry_track_type track_type; + /* lock handle supplied by user that we are tracking. See + above. */ + lock_handle *tracked; +}; + +/* information carry passes to plugin methods that may add new operations to + the @todo queue */ +struct carry_plugin_info { + carry_level *doing; + carry_level *todo; +}; + +int reiser4_carry(carry_level * doing, carry_level * done); + +carry_node *reiser4_add_carry(carry_level * level, pool_ordering order, + carry_node * reference); +carry_node *reiser4_add_carry_skip(carry_level * level, pool_ordering order, + carry_node * reference); + +extern carry_node *insert_carry_node(carry_level * doing, + carry_level * todo, const znode * node); + +extern carry_pool *init_carry_pool(int); +extern void done_carry_pool(carry_pool * pool); + +extern void init_carry_level(carry_level * level, carry_pool * pool); + +extern carry_op *reiser4_post_carry(carry_level * level, carry_opcode op, + znode * node, int apply_to_parent); +extern carry_op *node_post_carry(carry_plugin_info * info, carry_opcode op, + znode * node, int apply_to_parent_p); + +carry_node *add_new_znode(znode * brother, carry_node * reference, + carry_level * doing, carry_level * todo); + +carry_node *find_carry_node(carry_level * level, const znode * node); + +extern znode *reiser4_carry_real(const carry_node * node); + +/* helper macros to iterate over carry queues */ + +#define carry_node_next(node) \ + list_entry((node)->header.level_linkage.next, carry_node, \ + header.level_linkage) + +#define carry_node_prev(node) \ + list_entry((node)->header.level_linkage.prev, carry_node, \ + header.level_linkage) + +#define carry_node_front(level) \ + list_entry((level)->nodes.next, carry_node, header.level_linkage) + +#define carry_node_back(level) \ + list_entry((level)->nodes.prev, carry_node, header.level_linkage) + +#define carry_node_end(level, node) \ + (&(level)->nodes == &(node)->header.level_linkage) + +/* macro to iterate over all operations in a @level */ +#define for_all_ops(level /* carry level (of type carry_level *) */, \ + op /* pointer to carry operation, modified by loop (of \ + * type carry_op *) */, \ + tmp /* pointer to carry operation (of type carry_op *), \ + * used to make iterator stable in the face of \ + * deletions from the level */ ) \ +for (op = list_entry(level->ops.next, carry_op, header.level_linkage), \ + tmp = list_entry(op->header.level_linkage.next, carry_op, header.level_linkage); \ + &op->header.level_linkage != &level->ops; \ + op = tmp, \ + tmp = list_entry(op->header.level_linkage.next, carry_op, header.level_linkage)) + +#if 0 +for (op = (carry_op *) pool_level_list_front(&level->ops), \ + tmp = (carry_op *) pool_level_list_next(&op->header) ; \ + !pool_level_list_end(&level->ops, &op->header) ; \ + op = tmp, tmp = (carry_op *) pool_level_list_next(&op->header)) +#endif + +/* macro to iterate over all nodes in a @level */ \ +#define for_all_nodes(level /* carry level (of type carry_level *) */, \ + node /* pointer to carry node, modified by loop (of \ + * type carry_node *) */, \ + tmp /* pointer to carry node (of type carry_node *), \ + * used to make iterator stable in the face of * \ + * deletions from the level */ ) \ +for (node = list_entry(level->nodes.next, carry_node, header.level_linkage), \ + tmp = list_entry(node->header.level_linkage.next, carry_node, header.level_linkage); \ + &node->header.level_linkage != &level->nodes; \ + node = tmp, \ + tmp = list_entry(node->header.level_linkage.next, carry_node, header.level_linkage)) + +#if 0 +for (node = carry_node_front(level), \ + tmp = carry_node_next(node) ; !carry_node_end(level, node) ; \ + node = tmp, tmp = carry_node_next(node)) +#endif + +/* macro to iterate over all nodes in a @level in reverse order + + This is used, because nodes are unlocked in reversed order of locking */ +#define for_all_nodes_back(level /* carry level (of type carry_level *) */, \ + node /* pointer to carry node, modified by loop \ + * (of type carry_node *) */, \ + tmp /* pointer to carry node (of type carry_node \ + * *), used to make iterator stable in the \ + * face of deletions from the level */ ) \ +for (node = carry_node_back(level), \ + tmp = carry_node_prev(node) ; !carry_node_end(level, node) ; \ + node = tmp, tmp = carry_node_prev(node)) + +/* __FS_REISER4_CARRY_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/carry_ops.c linux-2.6.30/fs/reiser4/carry_ops.c --- linux-2.6.30.orig/fs/reiser4/carry_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/carry_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,2132 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* implementation of carry operations */ + +#include "forward.h" +#include "debug.h" +#include "key.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "plugin/node/node.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree_walk.h" +#include "pool.h" +#include "tree_mod.h" +#include "carry.h" +#include "carry_ops.h" +#include "tree.h" +#include "super.h" +#include "reiser4.h" + +#include +#include + +static int carry_shift_data(sideof side, coord_t *insert_coord, znode * node, + carry_level * doing, carry_level * todo, + unsigned int including_insert_coord_p); + +extern int lock_carry_node(carry_level * level, carry_node * node); +extern int lock_carry_node_tail(carry_node * node); + +/* find left neighbor of a carry node + + Look for left neighbor of @node and add it to the @doing queue. See + comments in the body. + +*/ +static carry_node *find_left_neighbor(carry_op * op /* node to find left + * neighbor of */ , + carry_level * doing/* level to scan */) +{ + int result; + carry_node *node; + carry_node *left; + int flags; + reiser4_tree *tree; + + node = op->node; + + tree = current_tree; + read_lock_tree(tree); + /* first, check whether left neighbor is already in a @doing queue */ + if (reiser4_carry_real(node)->left != NULL) { + /* NOTE: there is locking subtlety here. Look into + * find_right_neighbor() for more info */ + if (find_carry_node(doing, + reiser4_carry_real(node)->left) != NULL) { + read_unlock_tree(tree); + left = node; + do { + left = list_entry(left->header.level_linkage.prev, + carry_node, header.level_linkage); + assert("nikita-3408", !carry_node_end(doing, + left)); + } while (reiser4_carry_real(left) == + reiser4_carry_real(node)); + return left; + } + } + read_unlock_tree(tree); + + left = reiser4_add_carry_skip(doing, POOLO_BEFORE, node); + if (IS_ERR(left)) + return left; + + left->node = node->node; + left->free = 1; + + flags = GN_TRY_LOCK; + if (!op->u.insert.flags & COPI_LOAD_LEFT) + flags |= GN_NO_ALLOC; + + /* then, feeling lucky, peek left neighbor in the cache. */ + result = reiser4_get_left_neighbor(&left->lock_handle, + reiser4_carry_real(node), + ZNODE_WRITE_LOCK, flags); + if (result == 0) { + /* ok, node found and locked. */ + result = lock_carry_node_tail(left); + if (result != 0) + left = ERR_PTR(result); + } else if (result == -E_NO_NEIGHBOR || result == -ENOENT) { + /* node is leftmost node in a tree, or neighbor wasn't in + cache, or there is an extent on the left. */ + reiser4_pool_free(&doing->pool->node_pool, &left->header); + left = NULL; + } else if (doing->restartable) { + /* if left neighbor is locked, and level is restartable, add + new node to @doing and restart. */ + assert("nikita-913", node->parent != 0); + assert("nikita-914", node->node != NULL); + left->left = 1; + left->free = 0; + left = ERR_PTR(-E_REPEAT); + } else { + /* left neighbor is locked, level cannot be restarted. Just + ignore left neighbor. */ + reiser4_pool_free(&doing->pool->node_pool, &left->header); + left = NULL; + } + return left; +} + +/* find right neighbor of a carry node + + Look for right neighbor of @node and add it to the @doing queue. See + comments in the body. + +*/ +static carry_node *find_right_neighbor(carry_op * op /* node to find right + * neighbor of */ , + carry_level * doing/* level to scan */) +{ + int result; + carry_node *node; + carry_node *right; + lock_handle lh; + int flags; + reiser4_tree *tree; + + init_lh(&lh); + + node = op->node; + + tree = current_tree; + read_lock_tree(tree); + /* first, check whether right neighbor is already in a @doing queue */ + if (reiser4_carry_real(node)->right != NULL) { + /* + * Tree lock is taken here anyway, because, even if _outcome_ + * of (find_carry_node() != NULL) doesn't depends on + * concurrent updates to ->right, find_carry_node() cannot + * work with second argument NULL. Hence, following comment is + * of historic importance only. + * + * Subtle: + * + * Q: why don't we need tree lock here, looking for the right + * neighbor? + * + * A: even if value of node->real_node->right were changed + * during find_carry_node() execution, outcome of execution + * wouldn't change, because (in short) other thread cannot add + * elements to the @doing, and if node->real_node->right + * already was in @doing, value of node->real_node->right + * couldn't change, because node cannot be inserted between + * locked neighbors. + */ + if (find_carry_node(doing, + reiser4_carry_real(node)->right) != NULL) { + read_unlock_tree(tree); + /* + * What we are doing here (this is also applicable to + * the find_left_neighbor()). + * + * tree_walk.c code requires that insertion of a + * pointer to a child, modification of parent pointer + * in the child, and insertion of the child into + * sibling list are atomic (see + * plugin/item/internal.c:create_hook_internal()). + * + * carry allocates new node long before pointer to it + * is inserted into parent and, actually, long before + * parent is even known. Such allocated-but-orphaned + * nodes are only trackable through carry level lists. + * + * Situation that is handled here is following: @node + * has valid ->right pointer, but there is + * allocated-but-orphaned node in the carry queue that + * is logically between @node and @node->right. Here + * we are searching for it. Critical point is that + * this is only possible if @node->right is also in + * the carry queue (this is checked above), because + * this is the only way new orphaned node could be + * inserted between them (before inserting new node, + * make_space() first tries to shift to the right, so, + * right neighbor will be locked and queued). + * + */ + right = node; + do { + right = list_entry(right->header.level_linkage.next, + carry_node, header.level_linkage); + assert("nikita-3408", !carry_node_end(doing, + right)); + } while (reiser4_carry_real(right) == + reiser4_carry_real(node)); + return right; + } + } + read_unlock_tree(tree); + + flags = GN_CAN_USE_UPPER_LEVELS; + if (!op->u.insert.flags & COPI_LOAD_RIGHT) + flags = GN_NO_ALLOC; + + /* then, try to lock right neighbor */ + init_lh(&lh); + result = reiser4_get_right_neighbor(&lh, + reiser4_carry_real(node), + ZNODE_WRITE_LOCK, flags); + if (result == 0) { + /* ok, node found and locked. */ + right = reiser4_add_carry_skip(doing, POOLO_AFTER, node); + if (!IS_ERR(right)) { + right->node = lh.node; + move_lh(&right->lock_handle, &lh); + right->free = 1; + result = lock_carry_node_tail(right); + if (result != 0) + right = ERR_PTR(result); + } + } else if ((result == -E_NO_NEIGHBOR) || (result == -ENOENT)) { + /* node is rightmost node in a tree, or neighbor wasn't in + cache, or there is an extent on the right. */ + right = NULL; + } else + right = ERR_PTR(result); + done_lh(&lh); + return right; +} + +/* how much free space in a @node is needed for @op + + How much space in @node is required for completion of @op, where @op is + insert or paste operation. +*/ +static unsigned int space_needed_for_op(znode * node /* znode data are + * inserted or + * pasted in */ , + carry_op * op /* carry + operation */ ) +{ + assert("nikita-919", op != NULL); + + switch (op->op) { + default: + impossible("nikita-1701", "Wrong opcode"); + case COP_INSERT: + return space_needed(node, NULL, op->u.insert.d->data, 1); + case COP_PASTE: + return space_needed(node, op->u.insert.d->coord, + op->u.insert.d->data, 0); + } +} + +/* how much space in @node is required to insert or paste @data at + @coord. */ +unsigned int space_needed(const znode * node /* node data are inserted or + * pasted in */ , + const coord_t *coord /* coord where data are + * inserted or pasted + * at */ , + const reiser4_item_data * data /* data to insert or + * paste */ , + int insertion/* non-0 is inserting, 0---paste */) +{ + int result; + item_plugin *iplug; + + assert("nikita-917", node != NULL); + assert("nikita-918", node_plugin_by_node(node) != NULL); + assert("vs-230", !insertion || (coord == NULL)); + + result = 0; + iplug = data->iplug; + if (iplug->b.estimate != NULL) { + /* ask item plugin how much space is needed to insert this + item */ + result += iplug->b.estimate(insertion ? NULL : coord, data); + } else { + /* reasonable default */ + result += data->length; + } + if (insertion) { + node_plugin *nplug; + + nplug = node->nplug; + /* and add node overhead */ + if (nplug->item_overhead != NULL) + result += nplug->item_overhead(node, NULL); + } + return result; +} + +/* find &coord in parent where pointer to new child is to be stored. */ +static int find_new_child_coord(carry_op * op /* COP_INSERT carry operation to + * insert pointer to new + * child */ ) +{ + int result; + znode *node; + znode *child; + + assert("nikita-941", op != NULL); + assert("nikita-942", op->op == COP_INSERT); + + node = reiser4_carry_real(op->node); + assert("nikita-943", node != NULL); + assert("nikita-944", node_plugin_by_node(node) != NULL); + + child = reiser4_carry_real(op->u.insert.child); + result = + find_new_child_ptr(node, child, op->u.insert.brother, + op->u.insert.d->coord); + + build_child_ptr_data(child, op->u.insert.d->data); + return result; +} + +/* additional amount of free space in @node required to complete @op */ +static int free_space_shortage(znode * node /* node to check */ , + carry_op * op/* operation being performed */) +{ + assert("nikita-1061", node != NULL); + assert("nikita-1062", op != NULL); + + switch (op->op) { + default: + impossible("nikita-1702", "Wrong opcode"); + case COP_INSERT: + case COP_PASTE: + return space_needed_for_op(node, op) - znode_free_space(node); + case COP_EXTENT: + /* when inserting extent shift data around until insertion + point is utmost in the node. */ + if (coord_wrt(op->u.insert.d->coord) == COORD_INSIDE) + return +1; + else + return -1; + } +} + +/* helper function: update node pointer in operation after insertion + point was probably shifted into @target. */ +static znode *sync_op(carry_op * op, carry_node * target) +{ + znode *insertion_node; + + /* reget node from coord: shift might move insertion coord to + the neighbor */ + insertion_node = op->u.insert.d->coord->node; + /* if insertion point was actually moved into new node, + update carry node pointer in operation. */ + if (insertion_node != reiser4_carry_real(op->node)) { + op->node = target; + assert("nikita-2540", + reiser4_carry_real(target) == insertion_node); + } + assert("nikita-2541", + reiser4_carry_real(op->node) == op->u.insert.d->coord->node); + return insertion_node; +} + +/* + * complete make_space() call: update tracked lock handle if necessary. See + * comments for fs/reiser4/carry.h:carry_track_type + */ +static int +make_space_tail(carry_op * op, carry_level * doing, znode * orig_node) +{ + int result; + carry_track_type tracking; + znode *node; + + tracking = doing->track_type; + node = op->u.insert.d->coord->node; + + if (tracking == CARRY_TRACK_NODE || + (tracking == CARRY_TRACK_CHANGE && node != orig_node)) { + /* inserting or pasting into node different from + original. Update lock handle supplied by caller. */ + assert("nikita-1417", doing->tracked != NULL); + done_lh(doing->tracked); + init_lh(doing->tracked); + result = longterm_lock_znode(doing->tracked, node, + ZNODE_WRITE_LOCK, + ZNODE_LOCK_HIPRI); + } else + result = 0; + return result; +} + +/* This is insertion policy function. It shifts data to the left and right + neighbors of insertion coord and allocates new nodes until there is enough + free space to complete @op. + + See comments in the body. + + Assumes that the node format favors insertions at the right end of the node + as node40 does. + + See carry_flow() on detail about flow insertion +*/ +static int make_space(carry_op * op /* carry operation, insert or paste */ , + carry_level * doing /* current carry queue */ , + carry_level * todo/* carry queue on the parent level */) +{ + znode *node; + int result; + int not_enough_space; + int blk_alloc; + znode *orig_node; + __u32 flags; + + coord_t *coord; + + assert("nikita-890", op != NULL); + assert("nikita-891", todo != NULL); + assert("nikita-892", + op->op == COP_INSERT || + op->op == COP_PASTE || op->op == COP_EXTENT); + assert("nikita-1607", + reiser4_carry_real(op->node) == op->u.insert.d->coord->node); + + flags = op->u.insert.flags; + + /* NOTE check that new node can only be allocated after checking left + * and right neighbors. This is necessary for proper work of + * find_{left,right}_neighbor(). */ + assert("nikita-3410", ergo(flags & COPI_DONT_ALLOCATE, + flags & COPI_DONT_SHIFT_LEFT)); + assert("nikita-3411", ergo(flags & COPI_DONT_ALLOCATE, + flags & COPI_DONT_SHIFT_RIGHT)); + + coord = op->u.insert.d->coord; + orig_node = node = coord->node; + + assert("nikita-908", node != NULL); + assert("nikita-909", node_plugin_by_node(node) != NULL); + + result = 0; + /* If there is not enough space in a node, try to shift something to + the left neighbor. This is a bit tricky, as locking to the left is + low priority. This is handled by restart logic in carry(). + */ + not_enough_space = free_space_shortage(node, op); + if (not_enough_space <= 0) + /* it is possible that carry was called when there actually + was enough space in the node. For example, when inserting + leftmost item so that delimiting keys have to be updated. + */ + return make_space_tail(op, doing, orig_node); + if (!(flags & COPI_DONT_SHIFT_LEFT)) { + carry_node *left; + /* make note in statistics of an attempt to move + something into the left neighbor */ + left = find_left_neighbor(op, doing); + if (unlikely(IS_ERR(left))) { + if (PTR_ERR(left) == -E_REPEAT) + return -E_REPEAT; + else { + /* some error other than restart request + occurred. This shouldn't happen. Issue a + warning and continue as if left neighbor + weren't existing. + */ + warning("nikita-924", + "Error accessing left neighbor: %li", + PTR_ERR(left)); + } + } else if (left != NULL) { + + /* shift everything possible on the left of and + including insertion coord into the left neighbor */ + result = carry_shift_data(LEFT_SIDE, coord, + reiser4_carry_real(left), + doing, todo, + flags & COPI_GO_LEFT); + + /* reget node from coord: shift_left() might move + insertion coord to the left neighbor */ + node = sync_op(op, left); + + not_enough_space = free_space_shortage(node, op); + /* There is not enough free space in @node, but + may be, there is enough free space in + @left. Various balancing decisions are valid here. + The same for the shifiting to the right. + */ + } + } + /* If there still is not enough space, shift to the right */ + if (not_enough_space > 0 && !(flags & COPI_DONT_SHIFT_RIGHT)) { + carry_node *right; + + right = find_right_neighbor(op, doing); + if (IS_ERR(right)) { + warning("nikita-1065", + "Error accessing right neighbor: %li", + PTR_ERR(right)); + } else if (right != NULL) { + /* node containing insertion point, and its right + neighbor node are write locked by now. + + shift everything possible on the right of but + excluding insertion coord into the right neighbor + */ + result = carry_shift_data(RIGHT_SIDE, coord, + reiser4_carry_real(right), + doing, todo, + flags & COPI_GO_RIGHT); + /* reget node from coord: shift_right() might move + insertion coord to the right neighbor */ + node = sync_op(op, right); + not_enough_space = free_space_shortage(node, op); + } + } + /* If there is still not enough space, allocate new node(s). + + We try to allocate new blocks if COPI_DONT_ALLOCATE is not set in + the carry operation flags (currently this is needed during flush + only). + */ + for (blk_alloc = 0; + not_enough_space > 0 && result == 0 && blk_alloc < 2 && + !(flags & COPI_DONT_ALLOCATE); ++blk_alloc) { + carry_node *fresh; /* new node we are allocating */ + coord_t coord_shadow; /* remembered insertion point before + * shifting data into new node */ + carry_node *node_shadow; /* remembered insertion node + * before shifting */ + unsigned int gointo; /* whether insertion point should move + * into newly allocated node */ + + /* allocate new node on the right of @node. Znode and disk + fake block number for new node are allocated. + + add_new_znode() posts carry operation COP_INSERT with + COPT_CHILD option to the parent level to add + pointer to newly created node to its parent. + + Subtle point: if several new nodes are required to complete + insertion operation at this level, they will be inserted + into their parents in the order of creation, which means + that @node will be valid "cookie" at the time of insertion. + + */ + fresh = add_new_znode(node, op->node, doing, todo); + if (IS_ERR(fresh)) + return PTR_ERR(fresh); + + /* Try to shift into new node. */ + result = lock_carry_node(doing, fresh); + zput(reiser4_carry_real(fresh)); + if (result != 0) { + warning("nikita-947", + "Cannot lock new node: %i", result); + return result; + } + + /* both nodes are write locked by now. + + shift everything possible on the right of and + including insertion coord into the right neighbor. + */ + coord_dup(&coord_shadow, op->u.insert.d->coord); + node_shadow = op->node; + /* move insertion point into newly created node if: + + . insertion point is rightmost in the source node, or + . this is not the first node we are allocating in a row. + */ + gointo = + (blk_alloc > 0) || + coord_is_after_rightmost(op->u.insert.d->coord); + + if (gointo && + op->op == COP_PASTE && + coord_is_existing_item(op->u.insert.d->coord) && + is_solid_item((item_plugin_by_coord(op->u.insert.d->coord)))) { + /* paste into solid (atomic) item, which can contain + only one unit, so we need to shift it right, where + insertion point supposed to be */ + + assert("edward-1444", op->u.insert.d->data->iplug == + item_plugin_by_id(STATIC_STAT_DATA_ID)); + assert("edward-1445", + op->u.insert.d->data->length > + node_plugin_by_node(coord->node)->free_space + (coord->node)); + + op->u.insert.d->coord->between = BEFORE_UNIT; + } + + result = carry_shift_data(RIGHT_SIDE, coord, + reiser4_carry_real(fresh), + doing, todo, gointo); + /* if insertion point was actually moved into new node, + update carry node pointer in operation. */ + node = sync_op(op, fresh); + not_enough_space = free_space_shortage(node, op); + if ((not_enough_space > 0) && (node != coord_shadow.node)) { + /* there is not enough free in new node. Shift + insertion point back to the @shadow_node so that + next new node would be inserted between + @shadow_node and @fresh. + */ + coord_normalize(&coord_shadow); + coord_dup(coord, &coord_shadow); + node = coord->node; + op->node = node_shadow; + if (1 || (flags & COPI_STEP_BACK)) { + /* still not enough space?! Maybe there is + enough space in the source node (i.e., node + data are moved from) now. + */ + not_enough_space = + free_space_shortage(node, op); + } + } + } + if (not_enough_space > 0) { + if (!(flags & COPI_DONT_ALLOCATE)) + warning("nikita-948", "Cannot insert new item"); + result = -E_NODE_FULL; + } + assert("nikita-1622", ergo(result == 0, + reiser4_carry_real(op->node) == coord->node)); + assert("nikita-2616", coord == op->u.insert.d->coord); + if (result == 0) + result = make_space_tail(op, doing, orig_node); + return result; +} + +/* insert_paste_common() - common part of insert and paste operations + + This function performs common part of COP_INSERT and COP_PASTE. + + There are two ways in which insertion/paste can be requested: + + . by directly supplying reiser4_item_data. In this case, op -> + u.insert.type is set to COPT_ITEM_DATA. + + . by supplying child pointer to which is to inserted into parent. In this + case op -> u.insert.type == COPT_CHILD. + + . by supplying key of new item/unit. This is currently only used during + extent insertion + + This is required, because when new node is allocated we don't know at what + position pointer to it is to be stored in the parent. Actually, we don't + even know what its parent will be, because parent can be re-balanced + concurrently and new node re-parented, and because parent can be full and + pointer to the new node will go into some other node. + + insert_paste_common() resolves pointer to child node into position in the + parent by calling find_new_child_coord(), that fills + reiser4_item_data. After this, insertion/paste proceeds uniformly. + + Another complication is with finding free space during pasting. It may + happen that while shifting items to the neighbors and newly allocated + nodes, insertion coord can no longer be in the item we wanted to paste + into. At this point, paste becomes (morphs) into insert. Moreover free + space analysis has to be repeated, because amount of space required for + insertion is different from that of paste (item header overhead, etc). + + This function "unifies" different insertion modes (by resolving child + pointer or key into insertion coord), and then calls make_space() to free + enough space in the node by shifting data to the left and right and by + allocating new nodes if necessary. Carry operation knows amount of space + required for its completion. After enough free space is obtained, caller of + this function (carry_{insert,paste,etc.}) performs actual insertion/paste + by calling item plugin method. + +*/ +static int insert_paste_common(carry_op * op /* carry operation being + * performed */ , + carry_level * doing /* current carry level */ , + carry_level * todo /* next carry level */ , + carry_insert_data * cdata /* pointer to + * cdata */ , + coord_t *coord /* insertion/paste coord */ , + reiser4_item_data * data /* data to be + * inserted/pasted */ ) +{ + assert("nikita-981", op != NULL); + assert("nikita-980", todo != NULL); + assert("nikita-979", (op->op == COP_INSERT) || (op->op == COP_PASTE) + || (op->op == COP_EXTENT)); + + if (op->u.insert.type == COPT_PASTE_RESTARTED) { + /* nothing to do. Fall through to make_space(). */ + ; + } else if (op->u.insert.type == COPT_KEY) { + node_search_result intra_node; + znode *node; + /* Problem with doing batching at the lowest level, is that + operations here are given by coords where modification is + to be performed, and one modification can invalidate coords + of all following operations. + + So, we are implementing yet another type for operation that + will use (the only) "locator" stable across shifting of + data between nodes, etc.: key (COPT_KEY). + + This clause resolves key to the coord in the node. + + But node can change also. Probably some pieces have to be + added to the lock_carry_node(), to lock node by its key. + + */ + /* NOTE-NIKITA Lookup bias is fixed to FIND_EXACT. Complain + if you need something else. */ + op->u.insert.d->coord = coord; + node = reiser4_carry_real(op->node); + intra_node = node_plugin_by_node(node)->lookup + (node, op->u.insert.d->key, FIND_EXACT, + op->u.insert.d->coord); + if ((intra_node != NS_FOUND) && (intra_node != NS_NOT_FOUND)) { + warning("nikita-1715", "Intra node lookup failure: %i", + intra_node); + return intra_node; + } + } else if (op->u.insert.type == COPT_CHILD) { + /* if we are asked to insert pointer to the child into + internal node, first convert pointer to the child into + coord within parent node. + */ + znode *child; + int result; + + op->u.insert.d = cdata; + op->u.insert.d->coord = coord; + op->u.insert.d->data = data; + op->u.insert.d->coord->node = reiser4_carry_real(op->node); + result = find_new_child_coord(op); + child = reiser4_carry_real(op->u.insert.child); + if (result != NS_NOT_FOUND) { + warning("nikita-993", + "Cannot find a place for child pointer: %i", + result); + return result; + } + /* This only happens when we did multiple insertions at + the previous level, trying to insert single item and + it so happened, that insertion of pointers to all new + nodes before this one already caused parent node to + split (may be several times). + + I am going to come up with better solution. + + You are not expected to understand this. + -- v6root/usr/sys/ken/slp.c + + Basically, what happens here is the following: carry came + to the parent level and is about to insert internal item + pointing to the child node that it just inserted in the + level below. Position where internal item is to be inserted + was found by find_new_child_coord() above, but node of the + current carry operation (that is, parent node of child + inserted on the previous level), was determined earlier in + the lock_carry_level/lock_carry_node. It could so happen + that other carry operations already performed on the parent + level already split parent node, so that insertion point + moved into another node. Handle this by creating new carry + node for insertion point if necessary. + */ + if (reiser4_carry_real(op->node) != + op->u.insert.d->coord->node) { + pool_ordering direction; + znode *z1; + znode *z2; + reiser4_key k1; + reiser4_key k2; + + /* + * determine in what direction insertion point + * moved. Do this by comparing delimiting keys. + */ + z1 = op->u.insert.d->coord->node; + z2 = reiser4_carry_real(op->node); + if (keyle(leftmost_key_in_node(z1, &k1), + leftmost_key_in_node(z2, &k2))) + /* insertion point moved to the left */ + direction = POOLO_BEFORE; + else + /* insertion point moved to the right */ + direction = POOLO_AFTER; + + op->node = reiser4_add_carry_skip(doing, + direction, op->node); + if (IS_ERR(op->node)) + return PTR_ERR(op->node); + op->node->node = op->u.insert.d->coord->node; + op->node->free = 1; + result = lock_carry_node(doing, op->node); + if (result != 0) + return result; + } + + /* + * set up key of an item being inserted: we are inserting + * internal item and its key is (by the very definition of + * search tree) is leftmost key in the child node. + */ + write_lock_dk(znode_get_tree(child)); + op->u.insert.d->key = leftmost_key_in_node(child, + znode_get_ld_key(child)); + write_unlock_dk(znode_get_tree(child)); + op->u.insert.d->data->arg = op->u.insert.brother; + } else { + assert("vs-243", op->u.insert.d->coord != NULL); + op->u.insert.d->coord->node = reiser4_carry_real(op->node); + } + + /* find free space. */ + return make_space(op, doing, todo); +} + +/* handle carry COP_INSERT operation. + + Insert new item into node. New item can be given in one of two ways: + + - by passing &tree_coord and &reiser4_item_data as part of @op. This is + only applicable at the leaf/twig level. + + - by passing a child node pointer to which is to be inserted by this + operation. + +*/ +static int carry_insert(carry_op * op /* operation to perform */ , + carry_level * doing /* queue of operations @op + * is part of */ , + carry_level * todo /* queue where new operations + * are accumulated */ ) +{ + znode *node; + carry_insert_data cdata; + coord_t coord; + reiser4_item_data data; + carry_plugin_info info; + int result; + + assert("nikita-1036", op != NULL); + assert("nikita-1037", todo != NULL); + assert("nikita-1038", op->op == COP_INSERT); + + coord_init_zero(&coord); + + /* perform common functionality of insert and paste. */ + result = insert_paste_common(op, doing, todo, &cdata, &coord, &data); + if (result != 0) + return result; + + node = op->u.insert.d->coord->node; + assert("nikita-1039", node != NULL); + assert("nikita-1040", node_plugin_by_node(node) != NULL); + + assert("nikita-949", + space_needed_for_op(node, op) <= znode_free_space(node)); + + /* ask node layout to create new item. */ + info.doing = doing; + info.todo = todo; + result = node_plugin_by_node(node)->create_item + (op->u.insert.d->coord, op->u.insert.d->key, op->u.insert.d->data, + &info); + doing->restartable = 0; + znode_make_dirty(node); + + return result; +} + +/* + * Flow insertion code. COP_INSERT_FLOW is special tree operation that is + * supplied with a "flow" (that is, a stream of data) and inserts it into tree + * by slicing into multiple items. + */ + +#define flow_insert_point(op) ((op)->u.insert_flow.insert_point) +#define flow_insert_flow(op) ((op)->u.insert_flow.flow) +#define flow_insert_data(op) ((op)->u.insert_flow.data) + +static size_t item_data_overhead(carry_op * op) +{ + if (flow_insert_data(op)->iplug->b.estimate == NULL) + return 0; + return (flow_insert_data(op)->iplug->b. + estimate(NULL /* estimate insertion */ , flow_insert_data(op)) - + flow_insert_data(op)->length); +} + +/* FIXME-VS: this is called several times during one make_flow_for_insertion + and it will always return the same result. Some optimization could be made + by calculating this value once at the beginning and passing it around. That + would reduce some flexibility in future changes +*/ +static int can_paste(coord_t *, const reiser4_key *, const reiser4_item_data *); +static size_t flow_insertion_overhead(carry_op * op) +{ + znode *node; + size_t insertion_overhead; + + node = flow_insert_point(op)->node; + insertion_overhead = 0; + if (node->nplug->item_overhead && + !can_paste(flow_insert_point(op), &flow_insert_flow(op)->key, + flow_insert_data(op))) + insertion_overhead = + node->nplug->item_overhead(node, NULL) + + item_data_overhead(op); + return insertion_overhead; +} + +/* how many bytes of flow does fit to the node */ +static int what_can_fit_into_node(carry_op * op) +{ + size_t free, overhead; + + overhead = flow_insertion_overhead(op); + free = znode_free_space(flow_insert_point(op)->node); + if (free <= overhead) + return 0; + free -= overhead; + /* FIXME: flow->length is loff_t only to not get overflowed in case of + expandign truncate */ + if (free < op->u.insert_flow.flow->length) + return free; + return (int)op->u.insert_flow.flow->length; +} + +/* in make_space_for_flow_insertion we need to check either whether whole flow + fits into a node or whether minimal fraction of flow fits into a node */ +static int enough_space_for_whole_flow(carry_op * op) +{ + return (unsigned)what_can_fit_into_node(op) == + op->u.insert_flow.flow->length; +} + +#define MIN_FLOW_FRACTION 1 +static int enough_space_for_min_flow_fraction(carry_op * op) +{ + assert("vs-902", coord_is_after_rightmost(flow_insert_point(op))); + + return what_can_fit_into_node(op) >= MIN_FLOW_FRACTION; +} + +/* this returns 0 if left neighbor was obtained successfully and everything + upto insertion point including it were shifted and left neighbor still has + some free space to put minimal fraction of flow into it */ +static int +make_space_by_shift_left(carry_op * op, carry_level * doing, carry_level * todo) +{ + carry_node *left; + znode *orig; + + left = find_left_neighbor(op, doing); + if (unlikely(IS_ERR(left))) { + warning("vs-899", + "make_space_by_shift_left: " + "error accessing left neighbor: %li", PTR_ERR(left)); + return 1; + } + if (left == NULL) + /* left neighbor either does not exist or is unformatted + node */ + return 1; + + orig = flow_insert_point(op)->node; + /* try to shift content of node @orig from its head upto insert point + including insertion point into the left neighbor */ + carry_shift_data(LEFT_SIDE, flow_insert_point(op), + reiser4_carry_real(left), doing, todo, + 1/* including insert point */); + if (reiser4_carry_real(left) != flow_insert_point(op)->node) { + /* insertion point did not move */ + return 1; + } + + /* insertion point is set after last item in the node */ + assert("vs-900", coord_is_after_rightmost(flow_insert_point(op))); + + if (!enough_space_for_min_flow_fraction(op)) { + /* insertion point node does not have enough free space to put + even minimal portion of flow into it, therefore, move + insertion point back to orig node (before first item) */ + coord_init_before_first_item(flow_insert_point(op), orig); + return 1; + } + + /* part of flow is to be written to the end of node */ + op->node = left; + return 0; +} + +/* this returns 0 if right neighbor was obtained successfully and everything to + the right of insertion point was shifted to it and node got enough free + space to put minimal fraction of flow into it */ +static int +make_space_by_shift_right(carry_op * op, carry_level * doing, + carry_level * todo) +{ + carry_node *right; + + right = find_right_neighbor(op, doing); + if (unlikely(IS_ERR(right))) { + warning("nikita-1065", "shift_right_excluding_insert_point: " + "error accessing right neighbor: %li", PTR_ERR(right)); + return 1; + } + if (right) { + /* shift everything possible on the right of but excluding + insertion coord into the right neighbor */ + carry_shift_data(RIGHT_SIDE, flow_insert_point(op), + reiser4_carry_real(right), doing, todo, + 0/* not including insert point */); + } else { + /* right neighbor either does not exist or is unformatted + node */ + ; + } + if (coord_is_after_rightmost(flow_insert_point(op))) { + if (enough_space_for_min_flow_fraction(op)) { + /* part of flow is to be written to the end of node */ + return 0; + } + } + + /* new node is to be added if insert point node did not get enough + space for whole flow */ + return 1; +} + +/* this returns 0 when insert coord is set at the node end and fraction of flow + fits into that node */ +static int +make_space_by_new_nodes(carry_op * op, carry_level * doing, carry_level * todo) +{ + int result; + znode *node; + carry_node *new; + + node = flow_insert_point(op)->node; + + if (op->u.insert_flow.new_nodes == CARRY_FLOW_NEW_NODES_LIMIT) + return RETERR(-E_NODE_FULL); + /* add new node after insert point node */ + new = add_new_znode(node, op->node, doing, todo); + if (unlikely(IS_ERR(new))) + return PTR_ERR(new); + result = lock_carry_node(doing, new); + zput(reiser4_carry_real(new)); + if (unlikely(result)) + return result; + op->u.insert_flow.new_nodes++; + if (!coord_is_after_rightmost(flow_insert_point(op))) { + carry_shift_data(RIGHT_SIDE, flow_insert_point(op), + reiser4_carry_real(new), doing, todo, + 0/* not including insert point */); + assert("vs-901", + coord_is_after_rightmost(flow_insert_point(op))); + + if (enough_space_for_min_flow_fraction(op)) + return 0; + if (op->u.insert_flow.new_nodes == CARRY_FLOW_NEW_NODES_LIMIT) + return RETERR(-E_NODE_FULL); + + /* add one more new node */ + new = add_new_znode(node, op->node, doing, todo); + if (unlikely(IS_ERR(new))) + return PTR_ERR(new); + result = lock_carry_node(doing, new); + zput(reiser4_carry_real(new)); + if (unlikely(result)) + return result; + op->u.insert_flow.new_nodes++; + } + + /* move insertion point to new node */ + coord_init_before_first_item(flow_insert_point(op), + reiser4_carry_real(new)); + op->node = new; + return 0; +} + +static int +make_space_for_flow_insertion(carry_op * op, carry_level * doing, + carry_level * todo) +{ + __u32 flags = op->u.insert_flow.flags; + + if (enough_space_for_whole_flow(op)) { + /* whole flow fits into insert point node */ + return 0; + } + + if (!(flags & COPI_DONT_SHIFT_LEFT) + && (make_space_by_shift_left(op, doing, todo) == 0)) { + /* insert point is shifted to left neighbor of original insert + point node and is set after last unit in that node. It has + enough space to fit at least minimal fraction of flow. */ + return 0; + } + + if (enough_space_for_whole_flow(op)) { + /* whole flow fits into insert point node */ + return 0; + } + + if (!(flags & COPI_DONT_SHIFT_RIGHT) + && (make_space_by_shift_right(op, doing, todo) == 0)) { + /* insert point is still set to the same node, but there is + nothing to the right of insert point. */ + return 0; + } + + if (enough_space_for_whole_flow(op)) { + /* whole flow fits into insert point node */ + return 0; + } + + return make_space_by_new_nodes(op, doing, todo); +} + +/* implements COP_INSERT_FLOW operation */ +static int +carry_insert_flow(carry_op * op, carry_level * doing, carry_level * todo) +{ + int result; + flow_t *f; + coord_t *insert_point; + node_plugin *nplug; + carry_plugin_info info; + znode *orig_node; + lock_handle *orig_lh; + + f = op->u.insert_flow.flow; + result = 0; + + /* carry system needs this to work */ + info.doing = doing; + info.todo = todo; + + orig_node = flow_insert_point(op)->node; + orig_lh = doing->tracked; + + while (f->length) { + result = make_space_for_flow_insertion(op, doing, todo); + if (result) + break; + + insert_point = flow_insert_point(op); + nplug = node_plugin_by_node(insert_point->node); + + /* compose item data for insertion/pasting */ + flow_insert_data(op)->data = f->data; + flow_insert_data(op)->length = what_can_fit_into_node(op); + + if (can_paste(insert_point, &f->key, flow_insert_data(op))) { + /* insert point is set to item of file we are writing to + and we have to append to it */ + assert("vs-903", insert_point->between == AFTER_UNIT); + nplug->change_item_size(insert_point, + flow_insert_data(op)->length); + flow_insert_data(op)->iplug->b.paste(insert_point, + flow_insert_data + (op), &info); + } else { + /* new item must be inserted */ + pos_in_node_t new_pos; + flow_insert_data(op)->length += item_data_overhead(op); + + /* FIXME-VS: this is because node40_create_item changes + insert_point for obscure reasons */ + switch (insert_point->between) { + case AFTER_ITEM: + new_pos = insert_point->item_pos + 1; + break; + case EMPTY_NODE: + new_pos = 0; + break; + case BEFORE_ITEM: + assert("vs-905", insert_point->item_pos == 0); + new_pos = 0; + break; + default: + impossible("vs-906", + "carry_insert_flow: invalid coord"); + new_pos = 0; + break; + } + + nplug->create_item(insert_point, &f->key, + flow_insert_data(op), &info); + coord_set_item_pos(insert_point, new_pos); + } + coord_init_after_item_end(insert_point); + doing->restartable = 0; + znode_make_dirty(insert_point->node); + + move_flow_forward(f, (unsigned)flow_insert_data(op)->length); + } + + if (orig_node != flow_insert_point(op)->node) { + /* move lock to new insert point */ + done_lh(orig_lh); + init_lh(orig_lh); + result = + longterm_lock_znode(orig_lh, flow_insert_point(op)->node, + ZNODE_WRITE_LOCK, ZNODE_LOCK_HIPRI); + } + + return result; +} + +/* implements COP_DELETE operation + + Remove pointer to @op -> u.delete.child from it's parent. + + This function also handles killing of a tree root is last pointer from it + was removed. This is complicated by our handling of "twig" level: root on + twig level is never killed. + +*/ +static int carry_delete(carry_op * op /* operation to be performed */ , + carry_level * doing UNUSED_ARG /* current carry + * level */ , + carry_level * todo/* next carry level */) +{ + int result; + coord_t coord; + coord_t coord2; + znode *parent; + znode *child; + carry_plugin_info info; + reiser4_tree *tree; + + /* + * This operation is called to delete internal item pointing to the + * child node that was removed by carry from the tree on the previous + * tree level. + */ + + assert("nikita-893", op != NULL); + assert("nikita-894", todo != NULL); + assert("nikita-895", op->op == COP_DELETE); + + coord_init_zero(&coord); + coord_init_zero(&coord2); + + parent = reiser4_carry_real(op->node); + child = op->u.delete.child ? + reiser4_carry_real(op->u.delete.child) : op->node->node; + tree = znode_get_tree(child); + read_lock_tree(tree); + + /* + * @parent was determined when carry entered parent level + * (lock_carry_level/lock_carry_node). Since then, actual parent of + * @child node could change due to other carry operations performed on + * the parent level. Check for this. + */ + + if (znode_parent(child) != parent) { + /* NOTE-NIKITA add stat counter for this. */ + parent = znode_parent(child); + assert("nikita-2581", find_carry_node(doing, parent)); + } + read_unlock_tree(tree); + + assert("nikita-1213", znode_get_level(parent) > LEAF_LEVEL); + + /* Twig level horrors: tree should be of height at least 2. So, last + pointer from the root at twig level is preserved even if child is + empty. This is ugly, but so it was architectured. + */ + + if (znode_is_root(parent) && + znode_get_level(parent) <= REISER4_MIN_TREE_HEIGHT && + node_num_items(parent) == 1) { + /* Delimiting key manipulations. */ + write_lock_dk(tree); + znode_set_ld_key(child, znode_set_ld_key(parent, reiser4_min_key())); + znode_set_rd_key(child, znode_set_rd_key(parent, reiser4_max_key())); + ZF_SET(child, JNODE_DKSET); + write_unlock_dk(tree); + + /* @child escaped imminent death! */ + ZF_CLR(child, JNODE_HEARD_BANSHEE); + return 0; + } + + /* convert child pointer to the coord_t */ + result = find_child_ptr(parent, child, &coord); + if (result != NS_FOUND) { + warning("nikita-994", "Cannot find child pointer: %i", result); + print_coord_content("coord", &coord); + return result; + } + + coord_dup(&coord2, &coord); + info.doing = doing; + info.todo = todo; + { + /* + * Actually kill internal item: prepare structure with + * arguments for ->cut_and_kill() method... + */ + + struct carry_kill_data kdata; + kdata.params.from = &coord; + kdata.params.to = &coord2; + kdata.params.from_key = NULL; + kdata.params.to_key = NULL; + kdata.params.smallest_removed = NULL; + kdata.params.truncate = 1; + kdata.flags = op->u.delete.flags; + kdata.inode = NULL; + kdata.left = NULL; + kdata.right = NULL; + kdata.buf = NULL; + /* ... and call it. */ + result = node_plugin_by_node(parent)->cut_and_kill(&kdata, + &info); + } + doing->restartable = 0; + + /* check whether root should be killed violently */ + if (znode_is_root(parent) && + /* don't kill roots at and lower than twig level */ + znode_get_level(parent) > REISER4_MIN_TREE_HEIGHT && + node_num_items(parent) == 1) + result = reiser4_kill_tree_root(coord.node); + + return result < 0 ? : 0; +} + +/* implements COP_CUT opration + + Cuts part or whole content of node. + +*/ +static int carry_cut(carry_op * op /* operation to be performed */ , + carry_level * doing /* current carry level */ , + carry_level * todo/* next carry level */) +{ + int result; + carry_plugin_info info; + node_plugin *nplug; + + assert("nikita-896", op != NULL); + assert("nikita-897", todo != NULL); + assert("nikita-898", op->op == COP_CUT); + + info.doing = doing; + info.todo = todo; + + nplug = node_plugin_by_node(reiser4_carry_real(op->node)); + if (op->u.cut_or_kill.is_cut) + result = nplug->cut(op->u.cut_or_kill.u.cut, &info); + else + result = nplug->cut_and_kill(op->u.cut_or_kill.u.kill, &info); + + doing->restartable = 0; + return result < 0 ? : 0; +} + +/* helper function for carry_paste(): returns true if @op can be continued as + paste */ +static int +can_paste(coord_t *icoord, const reiser4_key * key, + const reiser4_item_data * data) +{ + coord_t circa; + item_plugin *new_iplug; + item_plugin *old_iplug; + int result = 0; /* to keep gcc shut */ + + assert("", icoord->between != AT_UNIT); + + /* obviously, one cannot paste when node is empty---there is nothing + to paste into. */ + if (node_is_empty(icoord->node)) + return 0; + /* if insertion point is at the middle of the item, then paste */ + if (!coord_is_between_items(icoord)) + return 1; + coord_dup(&circa, icoord); + circa.between = AT_UNIT; + + old_iplug = item_plugin_by_coord(&circa); + new_iplug = data->iplug; + + /* check whether we can paste to the item @icoord is "at" when we + ignore ->between field */ + if (old_iplug == new_iplug && item_can_contain_key(&circa, key, data)) + result = 1; + else if (icoord->between == BEFORE_UNIT + || icoord->between == BEFORE_ITEM) { + /* otherwise, try to glue to the item at the left, if any */ + coord_dup(&circa, icoord); + if (coord_set_to_left(&circa)) { + result = 0; + coord_init_before_item(icoord); + } else { + old_iplug = item_plugin_by_coord(&circa); + result = (old_iplug == new_iplug) + && item_can_contain_key(icoord, key, data); + if (result) { + coord_dup(icoord, &circa); + icoord->between = AFTER_UNIT; + } + } + } else if (icoord->between == AFTER_UNIT + || icoord->between == AFTER_ITEM) { + coord_dup(&circa, icoord); + /* otherwise, try to glue to the item at the right, if any */ + if (coord_set_to_right(&circa)) { + result = 0; + coord_init_after_item(icoord); + } else { + int (*cck) (const coord_t *, const reiser4_key *, + const reiser4_item_data *); + + old_iplug = item_plugin_by_coord(&circa); + + cck = old_iplug->b.can_contain_key; + if (cck == NULL) + /* item doesn't define ->can_contain_key + method? So it is not expandable. */ + result = 0; + else { + result = (old_iplug == new_iplug) + && cck(&circa /*icoord */ , key, data); + if (result) { + coord_dup(icoord, &circa); + icoord->between = BEFORE_UNIT; + } + } + } + } else + impossible("nikita-2513", "Nothing works"); + if (result) { + if (icoord->between == BEFORE_ITEM) { + assert("vs-912", icoord->unit_pos == 0); + icoord->between = BEFORE_UNIT; + } else if (icoord->between == AFTER_ITEM) { + coord_init_after_item_end(icoord); + } + } + return result; +} + +/* implements COP_PASTE operation + + Paste data into existing item. This is complicated by the fact that after + we shifted something to the left or right neighbors trying to free some + space, item we were supposed to paste into can be in different node than + insertion coord. If so, we are no longer doing paste, but insert. See + comments in insert_paste_common(). + +*/ +static int carry_paste(carry_op * op /* operation to be performed */ , + carry_level * doing UNUSED_ARG /* current carry + * level */ , + carry_level * todo/* next carry level */) +{ + znode *node; + carry_insert_data cdata; + coord_t dcoord; + reiser4_item_data data; + int result; + int real_size; + item_plugin *iplug; + carry_plugin_info info; + coord_t *coord; + + assert("nikita-982", op != NULL); + assert("nikita-983", todo != NULL); + assert("nikita-984", op->op == COP_PASTE); + + coord_init_zero(&dcoord); + + result = insert_paste_common(op, doing, todo, &cdata, &dcoord, &data); + if (result != 0) + return result; + + coord = op->u.insert.d->coord; + + /* handle case when op -> u.insert.coord doesn't point to the item + of required type. restart as insert. */ + if (!can_paste(coord, op->u.insert.d->key, op->u.insert.d->data)) { + op->op = COP_INSERT; + op->u.insert.type = COPT_PASTE_RESTARTED; + result = op_dispatch_table[COP_INSERT].handler(op, doing, todo); + + return result; + } + + node = coord->node; + iplug = item_plugin_by_coord(coord); + assert("nikita-992", iplug != NULL); + + assert("nikita-985", node != NULL); + assert("nikita-986", node_plugin_by_node(node) != NULL); + + assert("nikita-987", + space_needed_for_op(node, op) <= znode_free_space(node)); + + assert("nikita-1286", coord_is_existing_item(coord)); + + /* + * if item is expanded as a result of this operation, we should first + * change item size, than call ->b.paste item method. If item is + * shrunk, it should be done other way around: first call ->b.paste + * method, then reduce item size. + */ + + real_size = space_needed_for_op(node, op); + if (real_size > 0) + node->nplug->change_item_size(coord, real_size); + + doing->restartable = 0; + info.doing = doing; + info.todo = todo; + + result = iplug->b.paste(coord, op->u.insert.d->data, &info); + + if (real_size < 0) + node->nplug->change_item_size(coord, real_size); + + /* if we pasted at the beginning of the item, update item's key. */ + if (coord->unit_pos == 0 && coord->between != AFTER_UNIT) + node->nplug->update_item_key(coord, op->u.insert.d->key, &info); + + znode_make_dirty(node); + return result; +} + +/* handle carry COP_EXTENT operation. */ +static int carry_extent(carry_op * op /* operation to perform */ , + carry_level * doing /* queue of operations @op + * is part of */ , + carry_level * todo /* queue where new operations + * are accumulated */ ) +{ + znode *node; + carry_insert_data cdata; + coord_t coord; + reiser4_item_data data; + carry_op *delete_dummy; + carry_op *insert_extent; + int result; + carry_plugin_info info; + + assert("nikita-1751", op != NULL); + assert("nikita-1752", todo != NULL); + assert("nikita-1753", op->op == COP_EXTENT); + + /* extent insertion overview: + + extents live on the TWIG LEVEL, which is level one above the leaf + one. This complicates extent insertion logic somewhat: it may + happen (and going to happen all the time) that in logical key + ordering extent has to be placed between items I1 and I2, located + at the leaf level, but I1 and I2 are in the same formatted leaf + node N1. To insert extent one has to + + (1) reach node N1 and shift data between N1, its neighbors and + possibly newly allocated nodes until I1 and I2 fall into different + nodes. Since I1 and I2 are still neighboring items in logical key + order, they will be necessary utmost items in their respective + nodes. + + (2) After this new extent item is inserted into node on the twig + level. + + Fortunately this process can reuse almost all code from standard + insertion procedure (viz. make_space() and insert_paste_common()), + due to the following observation: make_space() only shifts data up + to and excluding or including insertion point. It never + "over-moves" through insertion point. Thus, one can use + make_space() to perform step (1). All required for this is just to + instruct free_space_shortage() to keep make_space() shifting data + until insertion point is at the node border. + + */ + + /* perform common functionality of insert and paste. */ + result = insert_paste_common(op, doing, todo, &cdata, &coord, &data); + if (result != 0) + return result; + + node = op->u.extent.d->coord->node; + assert("nikita-1754", node != NULL); + assert("nikita-1755", node_plugin_by_node(node) != NULL); + assert("nikita-1700", coord_wrt(op->u.extent.d->coord) != COORD_INSIDE); + + /* NOTE-NIKITA add some checks here. Not assertions, -EIO. Check that + extent fits between items. */ + + info.doing = doing; + info.todo = todo; + + /* there is another complication due to placement of extents on the + twig level: extents are "rigid" in the sense that key-range + occupied by extent cannot grow indefinitely to the right as it is + for the formatted leaf nodes. Because of this when search finds two + adjacent extents on the twig level, it has to "drill" to the leaf + level, creating new node. Here we are removing this node. + */ + if (node_is_empty(node)) { + delete_dummy = node_post_carry(&info, COP_DELETE, node, 1); + if (IS_ERR(delete_dummy)) + return PTR_ERR(delete_dummy); + delete_dummy->u.delete.child = NULL; + delete_dummy->u.delete.flags = DELETE_RETAIN_EMPTY; + ZF_SET(node, JNODE_HEARD_BANSHEE); + } + + /* proceed with inserting extent item into parent. We are definitely + inserting rather than pasting if we get that far. */ + insert_extent = node_post_carry(&info, COP_INSERT, node, 1); + if (IS_ERR(insert_extent)) + /* @delete_dummy will be automatically destroyed on the level + exiting */ + return PTR_ERR(insert_extent); + /* NOTE-NIKITA insertion by key is simplest option here. Another + possibility is to insert on the left or right of already existing + item. + */ + insert_extent->u.insert.type = COPT_KEY; + insert_extent->u.insert.d = op->u.extent.d; + assert("nikita-1719", op->u.extent.d->key != NULL); + insert_extent->u.insert.d->data->arg = op->u.extent.d->coord; + insert_extent->u.insert.flags = + znode_get_tree(node)->carry.new_extent_flags; + + /* + * if carry was asked to track lock handle we should actually track + * lock handle on the twig node rather than on the leaf where + * operation was started from. Transfer tracked lock handle. + */ + if (doing->track_type) { + assert("nikita-3242", doing->tracked != NULL); + assert("nikita-3244", todo->tracked == NULL); + todo->tracked = doing->tracked; + todo->track_type = CARRY_TRACK_NODE; + doing->tracked = NULL; + doing->track_type = 0; + } + + return 0; +} + +/* update key in @parent between pointers to @left and @right. + + Find coords of @left and @right and update delimiting key between them. + This is helper function called by carry_update(). Finds position of + internal item involved. Updates item key. Updates delimiting keys of child + nodes involved. +*/ +static int update_delimiting_key(znode * parent /* node key is updated + * in */ , + znode * left /* child of @parent */ , + znode * right /* child of @parent */ , + carry_level * doing /* current carry + * level */ , + carry_level * todo /* parent carry + * level */ , + const char **error_msg /* place to + * store error + * message */ ) +{ + coord_t left_pos; + coord_t right_pos; + int result; + reiser4_key ldkey; + carry_plugin_info info; + + assert("nikita-1177", right != NULL); + /* find position of right left child in a parent */ + result = find_child_ptr(parent, right, &right_pos); + if (result != NS_FOUND) { + *error_msg = "Cannot find position of right child"; + return result; + } + + if ((left != NULL) && !coord_is_leftmost_unit(&right_pos)) { + /* find position of the left child in a parent */ + result = find_child_ptr(parent, left, &left_pos); + if (result != NS_FOUND) { + *error_msg = "Cannot find position of left child"; + return result; + } + assert("nikita-1355", left_pos.node != NULL); + } else + left_pos.node = NULL; + + /* check that they are separated by exactly one key and are basically + sane */ + if (REISER4_DEBUG) { + if ((left_pos.node != NULL) + && !coord_is_existing_unit(&left_pos)) { + *error_msg = "Left child is bastard"; + return RETERR(-EIO); + } + if (!coord_is_existing_unit(&right_pos)) { + *error_msg = "Right child is bastard"; + return RETERR(-EIO); + } + if (left_pos.node != NULL && + !coord_are_neighbors(&left_pos, &right_pos)) { + *error_msg = "Children are not direct siblings"; + return RETERR(-EIO); + } + } + *error_msg = NULL; + + info.doing = doing; + info.todo = todo; + + /* + * If child node is not empty, new key of internal item is a key of + * leftmost item in the child node. If the child is empty, take its + * right delimiting key as a new key of the internal item. Precise key + * in the latter case is not important per se, because the child (and + * the internal item) are going to be killed shortly anyway, but we + * have to preserve correct order of keys in the parent node. + */ + + if (!ZF_ISSET(right, JNODE_HEARD_BANSHEE)) + leftmost_key_in_node(right, &ldkey); + else { + read_lock_dk(znode_get_tree(parent)); + ldkey = *znode_get_rd_key(right); + read_unlock_dk(znode_get_tree(parent)); + } + node_plugin_by_node(parent)->update_item_key(&right_pos, &ldkey, &info); + doing->restartable = 0; + znode_make_dirty(parent); + return 0; +} + +/* implements COP_UPDATE opration + + Update delimiting keys. + +*/ +static int carry_update(carry_op * op /* operation to be performed */ , + carry_level * doing /* current carry level */ , + carry_level * todo/* next carry level */) +{ + int result; + carry_node *missing UNUSED_ARG; + znode *left; + znode *right; + carry_node *lchild; + carry_node *rchild; + const char *error_msg; + reiser4_tree *tree; + + /* + * This operation is called to update key of internal item. This is + * necessary when carry shifted of cut data on the child + * level. Arguments of this operation are: + * + * @right --- child node. Operation should update key of internal + * item pointing to @right. + * + * @left --- left neighbor of @right. This parameter is optional. + */ + + assert("nikita-902", op != NULL); + assert("nikita-903", todo != NULL); + assert("nikita-904", op->op == COP_UPDATE); + + lchild = op->u.update.left; + rchild = op->node; + + if (lchild != NULL) { + assert("nikita-1001", lchild->parent); + assert("nikita-1003", !lchild->left); + left = reiser4_carry_real(lchild); + } else + left = NULL; + + tree = znode_get_tree(rchild->node); + read_lock_tree(tree); + right = znode_parent(rchild->node); + read_unlock_tree(tree); + + if (right != NULL) { + result = update_delimiting_key(right, + lchild ? lchild->node : NULL, + rchild->node, + doing, todo, &error_msg); + } else { + error_msg = "Cannot find node to update key in"; + result = RETERR(-EIO); + } + /* operation will be reposted to the next level by the + ->update_item_key() method of node plugin, if necessary. */ + + if (result != 0) { + warning("nikita-999", "Error updating delimiting key: %s (%i)", + error_msg ? : "", result); + } + return result; +} + +/* move items from @node during carry */ +static int carry_shift_data(sideof side /* in what direction to move data */ , + coord_t *insert_coord /* coord where new item + * is to be inserted */, + znode * node /* node which data are moved from */ , + carry_level * doing /* active carry queue */ , + carry_level * todo /* carry queue where new + * operations are to be put + * in */ , + unsigned int including_insert_coord_p + /* true if @insertion_coord can be moved */ ) +{ + int result; + znode *source; + carry_plugin_info info; + node_plugin *nplug; + + source = insert_coord->node; + + info.doing = doing; + info.todo = todo; + + nplug = node_plugin_by_node(node); + result = nplug->shift(insert_coord, node, + (side == LEFT_SIDE) ? SHIFT_LEFT : SHIFT_RIGHT, 0, + (int)including_insert_coord_p, &info); + /* the only error ->shift() method of node plugin can return is + -ENOMEM due to carry node/operation allocation. */ + assert("nikita-915", result >= 0 || result == -ENOMEM); + if (result > 0) { + /* + * if some number of bytes was actually shifted, mark nodes + * dirty, and carry level as non-restartable. + */ + doing->restartable = 0; + znode_make_dirty(source); + znode_make_dirty(node); + } + + assert("nikita-2077", coord_check(insert_coord)); + return 0; +} + +typedef carry_node *(*carry_iterator) (carry_node * node); +static carry_node *find_dir_carry(carry_node * node, carry_level * level, + carry_iterator iterator); + +static carry_node *pool_level_list_prev(carry_node *node) +{ + return list_entry(node->header.level_linkage.prev, carry_node, header.level_linkage); +} + +/* look for the left neighbor of given carry node in a carry queue. + + This is used by find_left_neighbor(), but I am not sure that this + really gives any advantage. More statistics required. + +*/ +carry_node *find_left_carry(carry_node * node /* node to find left neighbor + * of */ , + carry_level * level/* level to scan */) +{ + return find_dir_carry(node, level, + (carry_iterator) pool_level_list_prev); +} + +static carry_node *pool_level_list_next(carry_node *node) +{ + return list_entry(node->header.level_linkage.next, carry_node, header.level_linkage); +} + +/* look for the right neighbor of given carry node in a + carry queue. + + This is used by find_right_neighbor(), but I am not sure that this + really gives any advantage. More statistics required. + +*/ +carry_node *find_right_carry(carry_node * node /* node to find right neighbor + * of */ , + carry_level * level/* level to scan */) +{ + return find_dir_carry(node, level, + (carry_iterator) pool_level_list_next); +} + +/* look for the left or right neighbor of given carry node in a carry + queue. + + Helper function used by find_{left|right}_carry(). +*/ +static carry_node *find_dir_carry(carry_node * node /* node to start + * scanning from */ , + carry_level * level /* level to scan */ , + carry_iterator iterator /* operation to + * move to the + * next node */) +{ + carry_node *neighbor; + + assert("nikita-1059", node != NULL); + assert("nikita-1060", level != NULL); + + /* scan list of carry nodes on this list dir-ward, skipping all + carry nodes referencing the same znode. */ + neighbor = node; + while (1) { + neighbor = iterator(neighbor); + if (carry_node_end(level, neighbor)) + /* list head is reached */ + return NULL; + if (reiser4_carry_real(neighbor) != reiser4_carry_real(node)) + return neighbor; + } +} + +/* + * Memory reservation estimation. + * + * Carry process proceeds through tree levels upwards. Carry assumes that it + * takes tree in consistent state (e.g., that search tree invariants hold), + * and leaves tree consistent after it finishes. This means that when some + * error occurs carry cannot simply return if there are pending carry + * operations. Generic solution for this problem is carry-undo either as + * transaction manager feature (requiring checkpoints and isolation), or + * through some carry specific mechanism. + * + * Our current approach is to panic if carry hits an error while tree is + * inconsistent. Unfortunately -ENOMEM can easily be triggered. To work around + * this "memory reservation" mechanism was added. + * + * Memory reservation is implemented by perthread-pages.diff patch from + * core-patches. Its API is defined in + * + * int perthread_pages_reserve(int nrpages, gfp_t gfp); + * void perthread_pages_release(int nrpages); + * int perthread_pages_count(void); + * + * carry estimates its worst case memory requirements at the entry, reserved + * enough memory, and released unused pages before returning. + * + * Code below estimates worst case memory requirements for a given carry + * queue. This is dome by summing worst case memory requirements for each + * operation in the queue. + * + */ + +/* + * Memory memory requirements of many operations depends on the tree + * height. For example, item insertion requires new node to be inserted at + * each tree level in the worst case. What tree height should be used for + * estimation? Current tree height is wrong, because tree height can change + * between the time when estimation was done and the time when operation is + * actually performed. Maximal possible tree height (REISER4_MAX_ZTREE_HEIGHT) + * is also not desirable, because it would lead to the huge over-estimation + * all the time. Plausible solution is "capped tree height": if current tree + * height is less than some TREE_HEIGHT_CAP constant, capped tree height is + * TREE_HEIGHT_CAP, otherwise it's current tree height. Idea behind this is + * that if tree height is TREE_HEIGHT_CAP or larger, it's extremely unlikely + * to be increased even more during short interval of time. + */ +#define TREE_HEIGHT_CAP (5) + +/* return capped tree height for the @tree. See comment above. */ +static int cap_tree_height(reiser4_tree * tree) +{ + return max_t(int, tree->height, TREE_HEIGHT_CAP); +} + +/* return capped tree height for the current tree. */ +static int capped_height(void) +{ + return cap_tree_height(current_tree); +} + +/* return number of pages required to store given number of bytes */ +static int bytes_to_pages(int bytes) +{ + return (bytes + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; +} + +/* how many pages are required to allocate znodes during item insertion. */ +static int carry_estimate_znodes(void) +{ + /* + * Note, that there we have some problem here: there is no way to + * reserve pages specifically for the given slab. This means that + * these pages can be hijacked for some other end. + */ + + /* in the worst case we need 3 new znode on each tree level */ + return bytes_to_pages(capped_height() * sizeof(znode) * 3); +} + +/* + * how many pages are required to load bitmaps. One bitmap per level. + */ +static int carry_estimate_bitmaps(void) +{ + if (reiser4_is_set(reiser4_get_current_sb(), REISER4_DONT_LOAD_BITMAP)) { + int bytes; + + bytes = capped_height() * (0 + /* bnode should be added, but + * it is private to bitmap.c, + * skip for now. */ + 2 * sizeof(jnode)); + /* working and commit jnodes */ + return bytes_to_pages(bytes) + 2; /* and their contents */ + } else + /* bitmaps were pre-loaded during mount */ + return 0; +} + +/* worst case item insertion memory requirements */ +static int carry_estimate_insert(carry_op * op, carry_level * level) +{ + return carry_estimate_bitmaps() + carry_estimate_znodes() + 1 + + /* new atom */ + capped_height() + /* new block on each level */ + 1 + /* and possibly extra new block at the leaf level */ + 3; /* loading of leaves into memory */ +} + +/* worst case item deletion memory requirements */ +static int carry_estimate_delete(carry_op * op, carry_level * level) +{ + return carry_estimate_bitmaps() + carry_estimate_znodes() + 1 + + /* new atom */ + 3; /* loading of leaves into memory */ +} + +/* worst case tree cut memory requirements */ +static int carry_estimate_cut(carry_op * op, carry_level * level) +{ + return carry_estimate_bitmaps() + carry_estimate_znodes() + 1 + + /* new atom */ + 3; /* loading of leaves into memory */ +} + +/* worst case memory requirements of pasting into item */ +static int carry_estimate_paste(carry_op * op, carry_level * level) +{ + return carry_estimate_bitmaps() + carry_estimate_znodes() + 1 + + /* new atom */ + capped_height() + /* new block on each level */ + 1 + /* and possibly extra new block at the leaf level */ + 3; /* loading of leaves into memory */ +} + +/* worst case memory requirements of extent insertion */ +static int carry_estimate_extent(carry_op * op, carry_level * level) +{ + return carry_estimate_insert(op, level) + /* insert extent */ + carry_estimate_delete(op, level); /* kill leaf */ +} + +/* worst case memory requirements of key update */ +static int carry_estimate_update(carry_op * op, carry_level * level) +{ + return 0; +} + +/* worst case memory requirements of flow insertion */ +static int carry_estimate_insert_flow(carry_op * op, carry_level * level) +{ + int newnodes; + + newnodes = min(bytes_to_pages(op->u.insert_flow.flow->length), + CARRY_FLOW_NEW_NODES_LIMIT); + /* + * roughly estimate insert_flow as a sequence of insertions. + */ + return newnodes * carry_estimate_insert(op, level); +} + +/* This is dispatch table for carry operations. It can be trivially + abstracted into useful plugin: tunable balancing policy is a good + thing. */ +carry_op_handler op_dispatch_table[COP_LAST_OP] = { + [COP_INSERT] = { + .handler = carry_insert, + .estimate = carry_estimate_insert} + , + [COP_DELETE] = { + .handler = carry_delete, + .estimate = carry_estimate_delete} + , + [COP_CUT] = { + .handler = carry_cut, + .estimate = carry_estimate_cut} + , + [COP_PASTE] = { + .handler = carry_paste, + .estimate = carry_estimate_paste} + , + [COP_EXTENT] = { + .handler = carry_extent, + .estimate = carry_estimate_extent} + , + [COP_UPDATE] = { + .handler = carry_update, + .estimate = carry_estimate_update} + , + [COP_INSERT_FLOW] = { + .handler = carry_insert_flow, + .estimate = carry_estimate_insert_flow} +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/carry_ops.h linux-2.6.30/fs/reiser4/carry_ops.h --- linux-2.6.30.orig/fs/reiser4/carry_ops.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/carry_ops.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,43 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* implementation of carry operations. See carry_ops.c for details. */ + +#if !defined(__CARRY_OPS_H__) +#define __CARRY_OPS_H__ + +#include "forward.h" +#include "znode.h" +#include "carry.h" + +/* carry operation handlers */ +typedef struct carry_op_handler { + /* perform operation */ + int (*handler) (carry_op * op, carry_level * doing, carry_level * todo); + /* estimate memory requirements for @op */ + int (*estimate) (carry_op * op, carry_level * level); +} carry_op_handler; + +/* This is dispatch table for carry operations. It can be trivially + abstracted into useful plugin: tunable balancing policy is a good + thing. */ +extern carry_op_handler op_dispatch_table[COP_LAST_OP]; + +unsigned int space_needed(const znode * node, const coord_t *coord, + const reiser4_item_data * data, int inserting); +extern carry_node *find_left_carry(carry_node * node, carry_level * level); +extern carry_node *find_right_carry(carry_node * node, carry_level * level); + +/* __CARRY_OPS_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/context.c linux-2.6.30/fs/reiser4/context.c --- linux-2.6.30.orig/fs/reiser4/context.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/context.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,289 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Manipulation of reiser4_context */ + +/* + * global context used during system call. Variable of this type is allocated + * on the stack at the beginning of the reiser4 part of the system call and + * pointer to it is stored in the current->fs_context. This allows us to avoid + * passing pointer to current transaction and current lockstack (both in + * one-to-one mapping with threads) all over the call chain. + * + * It's kind of like those global variables the prof used to tell you not to + * use in CS1, except thread specific.;-) Nikita, this was a good idea. + * + * In some situations it is desirable to have ability to enter reiser4_context + * more than once for the same thread (nested contexts). For example, there + * are some functions that can be called either directly from VFS/VM or from + * already active reiser4 context (->writepage, for example). + * + * In such situations "child" context acts like dummy: all activity is + * actually performed in the top level context, and get_current_context() + * always returns top level context. + * Of course, reiser4_init_context()/reiser4_done_context() have to be properly + * nested any way. + * + * Note that there is an important difference between reiser4 uses + * ->fs_context and the way other file systems use it. Other file systems + * (ext3 and reiserfs) use ->fs_context only for the duration of _transaction_ + * (this is why ->fs_context was initially called ->journal_info). This means, + * that when ext3 or reiserfs finds that ->fs_context is not NULL on the entry + * to the file system, they assume that some transaction is already underway, + * and usually bail out, because starting nested transaction would most likely + * lead to the deadlock. This gives false positives with reiser4, because we + * set ->fs_context before starting transaction. + */ + +#include "debug.h" +#include "super.h" +#include "context.h" + +#include /* balance_dirty_pages() */ +#include + +static void _reiser4_init_context(reiser4_context * context, + struct super_block *super) +{ + memset(context, 0, sizeof(*context)); + + context->super = super; + context->magic = context_magic; + context->outer = current->journal_info; + current->journal_info = (void *)context; + context->nr_children = 0; + context->gfp_mask = GFP_KERNEL; + + init_lock_stack(&context->stack); + + reiser4_txn_begin(context); + + /* initialize head of tap list */ + INIT_LIST_HEAD(&context->taps); +#if REISER4_DEBUG + context->task = current; +#endif + grab_space_enable(); +} + +/* initialize context and bind it to the current thread + + This function should be called at the beginning of reiser4 part of + syscall. +*/ +reiser4_context * reiser4_init_context(struct super_block *super) +{ + reiser4_context *context; + + assert("nikita-2662", !in_interrupt() && !in_irq()); + assert("nikita-3357", super != NULL); + assert("nikita-3358", super->s_op == NULL || is_reiser4_super(super)); + + context = get_current_context_check(); + if (context && context->super == super) { + context = (reiser4_context *) current->journal_info; + context->nr_children++; + return context; + } + + context = kmalloc(sizeof(*context), GFP_KERNEL); + if (context == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + + _reiser4_init_context(context, super); + return context; +} + +/* this is used in scan_mgr which is called with spinlock held and in + reiser4_fill_super magic */ +void init_stack_context(reiser4_context *context, struct super_block *super) +{ + assert("nikita-2662", !in_interrupt() && !in_irq()); + assert("nikita-3357", super != NULL); + assert("nikita-3358", super->s_op == NULL || is_reiser4_super(super)); + assert("vs-12", !is_in_reiser4_context()); + + _reiser4_init_context(context, super); + context->on_stack = 1; + return; +} + +/* cast lock stack embedded into reiser4 context up to its container */ +reiser4_context *get_context_by_lock_stack(lock_stack * owner) +{ + return container_of(owner, reiser4_context, stack); +} + +/* true if there is already _any_ reiser4 context for the current thread */ +int is_in_reiser4_context(void) +{ + reiser4_context *ctx; + + ctx = current->journal_info; + return ctx != NULL && ((unsigned long)ctx->magic) == context_magic; +} + +/* + * call balance dirty pages for the current context. + * + * File system is expected to call balance_dirty_pages_ratelimited() whenever + * it dirties a page. reiser4 does this for unformatted nodes (that is, during + * write---this covers vast majority of all dirty traffic), but we cannot do + * this immediately when formatted node is dirtied, because long term lock is + * usually held at that time. To work around this, dirtying of formatted node + * simply increases ->nr_marked_dirty counter in the current reiser4 + * context. When we are about to leave this context, + * balance_dirty_pages_ratelimited() is called, if necessary. + * + * This introduces another problem: sometimes we do not want to run + * balance_dirty_pages_ratelimited() when leaving a context, for example + * because some important lock (like ->i_mutex on the parent directory) is + * held. To achieve this, ->nobalance flag can be set in the current context. + */ +static void balance_dirty_pages_at(reiser4_context *context) +{ + reiser4_super_info_data *sbinfo = get_super_private(context->super); + + /* + * call balance_dirty_pages_ratelimited() to process formatted nodes + * dirtied during this system call. Do that only if we are not in mount + * and there were nodes dirtied in this context and we are not in + * writepage (to avoid deadlock) and not in pdflush + */ + if (sbinfo != NULL && sbinfo->fake != NULL && + context->nr_marked_dirty != 0 && + !(current->flags & PF_MEMALLOC) && + !current_is_pdflush()) + balance_dirty_pages_ratelimited(sbinfo->fake->i_mapping); +} + +/* release resources associated with context. + + This function should be called at the end of "session" with reiser4, + typically just before leaving reiser4 driver back to VFS. + + This is good place to put some degugging consistency checks, like that + thread released all locks and closed transcrash etc. + +*/ +static void reiser4_done_context(reiser4_context * context) + /* context being released */ +{ + assert("nikita-860", context != NULL); + assert("nikita-859", context->magic == context_magic); + assert("vs-646", (reiser4_context *) current->journal_info == context); + assert("zam-686", !in_interrupt() && !in_irq()); + + /* only do anything when leaving top-level reiser4 context. All nested + * contexts are just dummies. */ + if (context->nr_children == 0) { + assert("jmacd-673", context->trans == NULL); + assert("jmacd-1002", lock_stack_isclean(&context->stack)); + assert("nikita-1936", reiser4_no_counters_are_held()); + assert("nikita-2626", list_empty_careful(reiser4_taps_list())); + assert("zam-1004", ergo(get_super_private(context->super), + get_super_private(context->super)->delete_mutex_owner != + current)); + + /* release all grabbed but as yet unused blocks */ + if (context->grabbed_blocks != 0) + all_grabbed2free(); + + /* + * synchronize against longterm_unlock_znode(): + * wake_up_requestor() wakes up requestors without holding + * zlock (otherwise they will immediately bump into that lock + * after wake up on another CPU). To work around (rare) + * situation where requestor has been woken up asynchronously + * and managed to run until completion (and destroy its + * context and lock stack) before wake_up_requestor() called + * wake_up() on it, wake_up_requestor() synchronize on lock + * stack spin lock. It has actually been observed that spin + * lock _was_ locked at this point, because + * wake_up_requestor() took interrupt. + */ + spin_lock_stack(&context->stack); + spin_unlock_stack(&context->stack); + + assert("zam-684", context->nr_children == 0); + /* restore original ->fs_context value */ + current->journal_info = context->outer; + if (context->on_stack == 0) + kfree(context); + } else { + context->nr_children--; +#if REISER4_DEBUG + assert("zam-685", context->nr_children >= 0); +#endif + } +} + +/* + * exit reiser4 context. Call balance_dirty_pages_at() if necessary. Close + * transaction. Call done_context() to do context related book-keeping. + */ +void reiser4_exit_context(reiser4_context * context) +{ + assert("nikita-3021", reiser4_schedulable()); + + if (context->nr_children == 0) { + if (!context->nobalance) { + reiser4_txn_restart(context); + balance_dirty_pages_at(context); + } + + /* if filesystem is mounted with -o sync or -o dirsync - commit + transaction. FIXME: TXNH_DONT_COMMIT is used to avoid + commiting on exit_context when inode semaphore is held and + to have ktxnmgrd to do commit instead to get better + concurrent filesystem accesses. But, when one mounts with -o + sync, he cares more about reliability than about + performance. So, for now we have this simple mount -o sync + support. */ + if (context->super->s_flags & (MS_SYNCHRONOUS | MS_DIRSYNC)) { + txn_atom *atom; + + atom = get_current_atom_locked_nocheck(); + if (atom) { + atom->flags |= ATOM_FORCE_COMMIT; + context->trans->flags &= ~TXNH_DONT_COMMIT; + spin_unlock_atom(atom); + } + } + reiser4_txn_end(context); + } + reiser4_done_context(context); +} + +void reiser4_ctx_gfp_mask_set(void) +{ + reiser4_context *ctx; + + ctx = get_current_context(); + if (ctx->entd == 0 && + list_empty(&ctx->stack.locks) && + ctx->trans->atom == NULL) + ctx->gfp_mask = GFP_KERNEL; + else + ctx->gfp_mask = GFP_NOFS; +} + +void reiser4_ctx_gfp_mask_force(gfp_t mask) +{ + reiser4_context *ctx; + ctx = get_current_context(); + + assert("edward-1454", ctx != NULL); + + ctx->gfp_mask = mask; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/context.h linux-2.6.30/fs/reiser4/context.h --- linux-2.6.30.orig/fs/reiser4/context.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/context.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,228 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Reiser4 context. See context.c for details. */ + +#if !defined( __REISER4_CONTEXT_H__ ) +#define __REISER4_CONTEXT_H__ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "tap.h" +#include "lock.h" + +#include /* for __u?? */ +#include /* for struct super_block */ +#include +#include /* for struct task_struct */ + +/* reiser4 per-thread context */ +struct reiser4_context { + /* magic constant. For identification of reiser4 contexts. */ + __u32 magic; + + /* current lock stack. See lock.[ch]. This is where list of all + locks taken by current thread is kept. This is also used in + deadlock detection. */ + lock_stack stack; + + /* current transcrash. */ + txn_handle *trans; + /* transaction handle embedded into reiser4_context. ->trans points + * here by default. */ + txn_handle trans_in_ctx; + + /* super block we are working with. To get the current tree + use &get_super_private (reiser4_get_current_sb ())->tree. */ + struct super_block *super; + + /* parent fs activation */ + struct fs_activation *outer; + + /* per-thread grabbed (for further allocation) blocks counter */ + reiser4_block_nr grabbed_blocks; + + /* list of taps currently monitored. See tap.c */ + struct list_head taps; + + /* grabbing space is enabled */ + unsigned int grab_enabled:1; + /* should be set when we are write dirty nodes to disk in jnode_flush or + * reiser4_write_logs() */ + unsigned int writeout_mode:1; + /* true, if current thread is an ent thread */ + unsigned int entd:1; + /* true, if balance_dirty_pages() should not be run when leaving this + * context. This is used to avoid lengthly balance_dirty_pages() + * operation when holding some important resource, like directory + * ->i_mutex */ + unsigned int nobalance:1; + + /* this bit is used on reiser4_done_context to decide whether context is + kmalloc-ed and has to be kfree-ed */ + unsigned int on_stack:1; + + /* count non-trivial jnode_set_dirty() calls */ + unsigned long nr_marked_dirty; + + /* reiser4_sync_inodes calls (via generic_sync_sb_inodes) + * reiser4_writepages for each of dirty inodes. Reiser4_writepages + * captures pages. When number of pages captured in one + * reiser4_sync_inodes reaches some threshold - some atoms get + * flushed */ + int nr_captured; + int nr_children; /* number of child contexts */ +#if REISER4_DEBUG + /* debugging information about reiser4 locks held by the current + * thread */ + reiser4_lock_cnt_info locks; + struct task_struct *task; /* so we can easily find owner of the stack */ + + /* + * disk space grabbing debugging support + */ + /* how many disk blocks were grabbed by the first call to + * reiser4_grab_space() in this context */ + reiser4_block_nr grabbed_initially; + + /* list of all threads doing flush currently */ + struct list_head flushers_link; + /* information about last error encountered by reiser4 */ + err_site err; +#endif + void *vp; + gfp_t gfp_mask; +}; + +extern reiser4_context *get_context_by_lock_stack(lock_stack *); + +/* Debugging helps. */ +#if REISER4_DEBUG +extern void print_contexts(void); +#endif + +#define current_tree (&(get_super_private(reiser4_get_current_sb())->tree)) +#define current_blocksize reiser4_get_current_sb()->s_blocksize +#define current_blocksize_bits reiser4_get_current_sb()->s_blocksize_bits + +extern reiser4_context *reiser4_init_context(struct super_block *); +extern void init_stack_context(reiser4_context *, struct super_block *); +extern void reiser4_exit_context(reiser4_context *); + +/* magic constant we store in reiser4_context allocated at the stack. Used to + catch accesses to staled or uninitialized contexts. */ +#define context_magic ((__u32) 0x4b1b5d0b) + +extern int is_in_reiser4_context(void); + +/* + * return reiser4_context for the thread @tsk + */ +static inline reiser4_context *get_context(const struct task_struct *tsk) +{ + assert("vs-1682", + ((reiser4_context *) tsk->journal_info)->magic == context_magic); + return (reiser4_context *) tsk->journal_info; +} + +/* + * return reiser4 context of the current thread, or NULL if there is none. + */ +static inline reiser4_context *get_current_context_check(void) +{ + if (is_in_reiser4_context()) + return get_context(current); + else + return NULL; +} + +static inline reiser4_context *get_current_context(void); /* __attribute__((const)); */ + +/* return context associated with current thread */ +static inline reiser4_context *get_current_context(void) +{ + return get_context(current); +} + +static inline gfp_t reiser4_ctx_gfp_mask_get(void) +{ + reiser4_context *ctx; + + ctx = get_current_context_check(); + return (ctx == NULL) ? GFP_KERNEL : ctx->gfp_mask; +} + +void reiser4_ctx_gfp_mask_set(void); +void reiser4_ctx_gfp_mask_force (gfp_t mask); + +/* + * true if current thread is in the write-out mode. Thread enters write-out + * mode during jnode_flush and reiser4_write_logs(). + */ +static inline int is_writeout_mode(void) +{ + return get_current_context()->writeout_mode; +} + +/* + * enter write-out mode + */ +static inline void writeout_mode_enable(void) +{ + assert("zam-941", !get_current_context()->writeout_mode); + get_current_context()->writeout_mode = 1; +} + +/* + * leave write-out mode + */ +static inline void writeout_mode_disable(void) +{ + assert("zam-942", get_current_context()->writeout_mode); + get_current_context()->writeout_mode = 0; +} + +static inline void grab_space_enable(void) +{ + get_current_context()->grab_enabled = 1; +} + +static inline void grab_space_disable(void) +{ + get_current_context()->grab_enabled = 0; +} + +static inline void grab_space_set_enabled(int enabled) +{ + get_current_context()->grab_enabled = enabled; +} + +static inline int is_grab_enabled(reiser4_context * ctx) +{ + return ctx->grab_enabled; +} + +/* mark transaction handle in @ctx as TXNH_DONT_COMMIT, so that no commit or + * flush would be performed when it is closed. This is necessary when handle + * has to be closed under some coarse semaphore, like i_mutex of + * directory. Commit will be performed by ktxnmgrd. */ +static inline void context_set_commit_async(reiser4_context * context) +{ + context->nobalance = 1; + context->trans->flags |= TXNH_DONT_COMMIT; +} + +/* __REISER4_CONTEXT_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/coord.c linux-2.6.30/fs/reiser4/coord.c --- linux-2.6.30.orig/fs/reiser4/coord.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/coord.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,928 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "tree.h" +#include "plugin/item/item.h" +#include "znode.h" +#include "coord.h" + +/* Internal constructor. */ +static inline void +coord_init_values(coord_t *coord, const znode * node, pos_in_node_t item_pos, + pos_in_node_t unit_pos, between_enum between) +{ + coord->node = (znode *) node; + coord_set_item_pos(coord, item_pos); + coord->unit_pos = unit_pos; + coord->between = between; + ON_DEBUG(coord->plug_v = 0); + ON_DEBUG(coord->body_v = 0); + + /*ON_TRACE (TRACE_COORDS, "init coord %p node %p: %u %u %s\n", coord, + node, item_pos, unit_pos, coord_tween_tostring (between)); */ +} + +/* after shifting of node content, coord previously set properly may become + invalid, try to "normalize" it. */ +void coord_normalize(coord_t *coord) +{ + znode *node; + + node = coord->node; + assert("vs-683", node); + + coord_clear_iplug(coord); + + if (node_is_empty(node)) { + coord_init_first_unit(coord, node); + } else if ((coord->between == AFTER_ITEM) + || (coord->between == AFTER_UNIT)) { + return; + } else if (coord->item_pos == coord_num_items(coord) + && coord->between == BEFORE_ITEM) { + coord_dec_item_pos(coord); + coord->between = AFTER_ITEM; + } else if (coord->unit_pos == coord_num_units(coord) + && coord->between == BEFORE_UNIT) { + coord->unit_pos--; + coord->between = AFTER_UNIT; + } else if (coord->item_pos == coord_num_items(coord) + && coord->unit_pos == 0 && coord->between == BEFORE_UNIT) { + coord_dec_item_pos(coord); + coord->unit_pos = 0; + coord->between = AFTER_ITEM; + } +} + +/* Copy a coordinate. */ +void coord_dup(coord_t *coord, const coord_t *old_coord) +{ + assert("jmacd-9800", coord_check(old_coord)); + coord_dup_nocheck(coord, old_coord); +} + +/* Copy a coordinate without check. Useful when old_coord->node is not + loaded. As in cbk_tree_lookup -> connect_znode -> connect_one_side */ +void coord_dup_nocheck(coord_t *coord, const coord_t *old_coord) +{ + coord->node = old_coord->node; + coord_set_item_pos(coord, old_coord->item_pos); + coord->unit_pos = old_coord->unit_pos; + coord->between = old_coord->between; + coord->iplugid = old_coord->iplugid; + ON_DEBUG(coord->plug_v = old_coord->plug_v); + ON_DEBUG(coord->body_v = old_coord->body_v); +} + +/* Initialize an invalid coordinate. */ +void coord_init_invalid(coord_t *coord, const znode * node) +{ + coord_init_values(coord, node, 0, 0, INVALID_COORD); +} + +void coord_init_first_unit_nocheck(coord_t *coord, const znode * node) +{ + coord_init_values(coord, node, 0, 0, AT_UNIT); +} + +/* Initialize a coordinate to point at the first unit of the first item. If the + node is empty, it is positioned at the EMPTY_NODE. */ +void coord_init_first_unit(coord_t *coord, const znode * node) +{ + int is_empty = node_is_empty(node); + + coord_init_values(coord, node, 0, 0, (is_empty ? EMPTY_NODE : AT_UNIT)); + + assert("jmacd-9801", coord_check(coord)); +} + +/* Initialize a coordinate to point at the last unit of the last item. If the + node is empty, it is positioned at the EMPTY_NODE. */ +void coord_init_last_unit(coord_t *coord, const znode * node) +{ + int is_empty = node_is_empty(node); + + coord_init_values(coord, node, + (is_empty ? 0 : node_num_items(node) - 1), 0, + (is_empty ? EMPTY_NODE : AT_UNIT)); + if (!is_empty) + coord->unit_pos = coord_last_unit_pos(coord); + assert("jmacd-9802", coord_check(coord)); +} + +/* Initialize a coordinate to before the first item. If the node is empty, it is + positioned at the EMPTY_NODE. */ +void coord_init_before_first_item(coord_t *coord, const znode * node) +{ + int is_empty = node_is_empty(node); + + coord_init_values(coord, node, 0, 0, + (is_empty ? EMPTY_NODE : BEFORE_UNIT)); + + assert("jmacd-9803", coord_check(coord)); +} + +/* Initialize a coordinate to after the last item. If the node is empty, it is + positioned at the EMPTY_NODE. */ +void coord_init_after_last_item(coord_t *coord, const znode * node) +{ + int is_empty = node_is_empty(node); + + coord_init_values(coord, node, + (is_empty ? 0 : node_num_items(node) - 1), 0, + (is_empty ? EMPTY_NODE : AFTER_ITEM)); + + assert("jmacd-9804", coord_check(coord)); +} + +/* Initialize a coordinate to after last unit in the item. Coord must be set + already to existing item */ +void coord_init_after_item_end(coord_t *coord) +{ + coord->between = AFTER_UNIT; + coord->unit_pos = coord_last_unit_pos(coord); +} + +/* Initialize a coordinate to before the item. Coord must be set already to + existing item */ +void coord_init_before_item(coord_t *coord) +{ + coord->unit_pos = 0; + coord->between = BEFORE_ITEM; +} + +/* Initialize a coordinate to after the item. Coord must be set already to + existing item */ +void coord_init_after_item(coord_t *coord) +{ + coord->unit_pos = 0; + coord->between = AFTER_ITEM; +} + +/* Initialize a coordinate by 0s. Used in places where init_coord was used and + it was not clear how actually */ +void coord_init_zero(coord_t *coord) +{ + memset(coord, 0, sizeof(*coord)); +} + +/* Return the number of units at the present item. + Asserts coord_is_existing_item(). */ +unsigned coord_num_units(const coord_t *coord) +{ + assert("jmacd-9806", coord_is_existing_item(coord)); + + return item_plugin_by_coord(coord)->b.nr_units(coord); +} + +/* Returns true if the coord was initializewd by coord_init_invalid (). */ +/* Audited by: green(2002.06.15) */ +int coord_is_invalid(const coord_t *coord) +{ + return coord->between == INVALID_COORD; +} + +/* Returns true if the coordinate is positioned at an existing item, not before + or after an item. It may be placed at, before, or after any unit within the + item, whether existing or not. */ +int coord_is_existing_item(const coord_t *coord) +{ + switch (coord->between) { + case EMPTY_NODE: + case BEFORE_ITEM: + case AFTER_ITEM: + case INVALID_COORD: + return 0; + + case BEFORE_UNIT: + case AT_UNIT: + case AFTER_UNIT: + return coord->item_pos < coord_num_items(coord); + } + + impossible("jmacd-9900", "unreachable coord: %p", coord); + return 0; +} + +/* Returns true if the coordinate is positioned at an existing unit, not before + or after a unit. */ +/* Audited by: green(2002.06.15) */ +int coord_is_existing_unit(const coord_t *coord) +{ + switch (coord->between) { + case EMPTY_NODE: + case BEFORE_UNIT: + case AFTER_UNIT: + case BEFORE_ITEM: + case AFTER_ITEM: + case INVALID_COORD: + return 0; + + case AT_UNIT: + return (coord->item_pos < coord_num_items(coord) + && coord->unit_pos < coord_num_units(coord)); + } + + impossible("jmacd-9902", "unreachable"); + return 0; +} + +/* Returns true if the coordinate is positioned at the first unit of the first + item. Not true for empty nodes nor coordinates positioned before the first + item. */ +/* Audited by: green(2002.06.15) */ +int coord_is_leftmost_unit(const coord_t *coord) +{ + return (coord->between == AT_UNIT && coord->item_pos == 0 + && coord->unit_pos == 0); +} + +#if REISER4_DEBUG +/* For assertions only, checks for a valid coordinate. */ +int coord_check(const coord_t *coord) +{ + if (coord->node == NULL) + return 0; + if (znode_above_root(coord->node)) + return 1; + + switch (coord->between) { + default: + case INVALID_COORD: + return 0; + case EMPTY_NODE: + if (!node_is_empty(coord->node)) + return 0; + return coord->item_pos == 0 && coord->unit_pos == 0; + + case BEFORE_UNIT: + case AFTER_UNIT: + if (node_is_empty(coord->node) && (coord->item_pos == 0) + && (coord->unit_pos == 0)) + return 1; + case AT_UNIT: + break; + case AFTER_ITEM: + case BEFORE_ITEM: + /* before/after item should not set unit_pos. */ + if (coord->unit_pos != 0) + return 0; + break; + } + + if (coord->item_pos >= node_num_items(coord->node)) + return 0; + + /* FIXME-VS: we are going to check unit_pos. This makes no sense when + between is set either AFTER_ITEM or BEFORE_ITEM */ + if (coord->between == AFTER_ITEM || coord->between == BEFORE_ITEM) + return 1; + + if (coord_is_iplug_set(coord) && + coord->unit_pos > + item_plugin_by_coord(coord)->b.nr_units(coord) - 1) + return 0; + return 1; +} +#endif + +/* Adjust coordinate boundaries based on the number of items prior to + coord_next/prev. Returns 1 if the new position is does not exist. */ +static int coord_adjust_items(coord_t *coord, unsigned items, int is_next) +{ + /* If the node is invalid, leave it. */ + if (coord->between == INVALID_COORD) + return 1; + + /* If the node is empty, set it appropriately. */ + if (items == 0) { + coord->between = EMPTY_NODE; + coord_set_item_pos(coord, 0); + coord->unit_pos = 0; + return 1; + } + + /* If it was empty and it no longer is, set to BEFORE/AFTER_ITEM. */ + if (coord->between == EMPTY_NODE) { + coord->between = (is_next ? BEFORE_ITEM : AFTER_ITEM); + coord_set_item_pos(coord, 0); + coord->unit_pos = 0; + return 0; + } + + /* If the item_pos is out-of-range, set it appropriatly. */ + if (coord->item_pos >= items) { + coord->between = AFTER_ITEM; + coord_set_item_pos(coord, items - 1); + coord->unit_pos = 0; + /* If is_next, return 1 (can't go any further). */ + return is_next; + } + + return 0; +} + +/* Advances the coordinate by one unit to the right. If empty, no change. If + coord_is_rightmost_unit, advances to AFTER THE LAST ITEM. Returns 0 if new + position is an existing unit. */ +int coord_next_unit(coord_t *coord) +{ + unsigned items = coord_num_items(coord); + + if (coord_adjust_items(coord, items, 1) == 1) + return 1; + + switch (coord->between) { + case BEFORE_UNIT: + /* Now it is positioned at the same unit. */ + coord->between = AT_UNIT; + return 0; + + case AFTER_UNIT: + case AT_UNIT: + /* If it was at or after a unit and there are more units in this + item, advance to the next one. */ + if (coord->unit_pos < coord_last_unit_pos(coord)) { + coord->unit_pos += 1; + coord->between = AT_UNIT; + return 0; + } + + /* Otherwise, it is crossing an item boundary and treated as if + it was after the current item. */ + coord->between = AFTER_ITEM; + coord->unit_pos = 0; + /* FALLTHROUGH */ + + case AFTER_ITEM: + /* Check for end-of-node. */ + if (coord->item_pos == items - 1) + return 1; + + coord_inc_item_pos(coord); + coord->unit_pos = 0; + coord->between = AT_UNIT; + return 0; + + case BEFORE_ITEM: + /* The adjust_items checks ensure that we are valid here. */ + coord->unit_pos = 0; + coord->between = AT_UNIT; + return 0; + + case INVALID_COORD: + case EMPTY_NODE: + /* Handled in coord_adjust_items(). */ + break; + } + + impossible("jmacd-9902", "unreachable"); + return 0; +} + +/* Advances the coordinate by one item to the right. If empty, no change. If + coord_is_rightmost_unit, advances to AFTER THE LAST ITEM. Returns 0 if new + position is an existing item. */ +int coord_next_item(coord_t *coord) +{ + unsigned items = coord_num_items(coord); + + if (coord_adjust_items(coord, items, 1) == 1) + return 1; + + switch (coord->between) { + case AFTER_UNIT: + case AT_UNIT: + case BEFORE_UNIT: + case AFTER_ITEM: + /* Check for end-of-node. */ + if (coord->item_pos == items - 1) { + coord->between = AFTER_ITEM; + coord->unit_pos = 0; + coord_clear_iplug(coord); + return 1; + } + + /* Anywhere in an item, go to the next one. */ + coord->between = AT_UNIT; + coord_inc_item_pos(coord); + coord->unit_pos = 0; + return 0; + + case BEFORE_ITEM: + /* The out-of-range check ensures that we are valid here. */ + coord->unit_pos = 0; + coord->between = AT_UNIT; + return 0; + case INVALID_COORD: + case EMPTY_NODE: + /* Handled in coord_adjust_items(). */ + break; + } + + impossible("jmacd-9903", "unreachable"); + return 0; +} + +/* Advances the coordinate by one unit to the left. If empty, no change. If + coord_is_leftmost_unit, advances to BEFORE THE FIRST ITEM. Returns 0 if new + position is an existing unit. */ +int coord_prev_unit(coord_t *coord) +{ + unsigned items = coord_num_items(coord); + + if (coord_adjust_items(coord, items, 0) == 1) + return 1; + + switch (coord->between) { + case AT_UNIT: + case BEFORE_UNIT: + if (coord->unit_pos > 0) { + coord->unit_pos -= 1; + coord->between = AT_UNIT; + return 0; + } + + if (coord->item_pos == 0) { + coord->between = BEFORE_ITEM; + return 1; + } + + coord_dec_item_pos(coord); + coord->unit_pos = coord_last_unit_pos(coord); + coord->between = AT_UNIT; + return 0; + + case AFTER_UNIT: + /* What if unit_pos is out-of-range? */ + assert("jmacd-5442", + coord->unit_pos <= coord_last_unit_pos(coord)); + coord->between = AT_UNIT; + return 0; + + case BEFORE_ITEM: + if (coord->item_pos == 0) + return 1; + + coord_dec_item_pos(coord); + /* FALLTHROUGH */ + + case AFTER_ITEM: + coord->between = AT_UNIT; + coord->unit_pos = coord_last_unit_pos(coord); + return 0; + + case INVALID_COORD: + case EMPTY_NODE: + break; + } + + impossible("jmacd-9904", "unreachable"); + return 0; +} + +/* Advances the coordinate by one item to the left. If empty, no change. If + coord_is_leftmost_unit, advances to BEFORE THE FIRST ITEM. Returns 0 if new + position is an existing item. */ +int coord_prev_item(coord_t *coord) +{ + unsigned items = coord_num_items(coord); + + if (coord_adjust_items(coord, items, 0) == 1) + return 1; + + switch (coord->between) { + case AT_UNIT: + case AFTER_UNIT: + case BEFORE_UNIT: + case BEFORE_ITEM: + + if (coord->item_pos == 0) { + coord->between = BEFORE_ITEM; + coord->unit_pos = 0; + return 1; + } + + coord_dec_item_pos(coord); + coord->unit_pos = 0; + coord->between = AT_UNIT; + return 0; + + case AFTER_ITEM: + coord->between = AT_UNIT; + coord->unit_pos = 0; + return 0; + + case INVALID_COORD: + case EMPTY_NODE: + break; + } + + impossible("jmacd-9905", "unreachable"); + return 0; +} + +/* Calls either coord_init_first_unit or coord_init_last_unit depending on + sideof argument. */ +void coord_init_sideof_unit(coord_t *coord, const znode * node, sideof dir) +{ + assert("jmacd-9821", dir == LEFT_SIDE || dir == RIGHT_SIDE); + if (dir == LEFT_SIDE) { + coord_init_first_unit(coord, node); + } else { + coord_init_last_unit(coord, node); + } +} + +/* Calls either coord_is_before_leftmost or coord_is_after_rightmost depending + on sideof argument. */ +/* Audited by: green(2002.06.15) */ +int coord_is_after_sideof_unit(coord_t *coord, sideof dir) +{ + assert("jmacd-9822", dir == LEFT_SIDE || dir == RIGHT_SIDE); + if (dir == LEFT_SIDE) { + return coord_is_before_leftmost(coord); + } else { + return coord_is_after_rightmost(coord); + } +} + +/* Calls either coord_next_unit or coord_prev_unit depending on sideof argument. + */ +/* Audited by: green(2002.06.15) */ +int coord_sideof_unit(coord_t *coord, sideof dir) +{ + assert("jmacd-9823", dir == LEFT_SIDE || dir == RIGHT_SIDE); + if (dir == LEFT_SIDE) { + return coord_prev_unit(coord); + } else { + return coord_next_unit(coord); + } +} + +#if REISER4_DEBUG +int coords_equal(const coord_t *c1, const coord_t *c2) +{ + assert("nikita-2840", c1 != NULL); + assert("nikita-2841", c2 != NULL); + + return + c1->node == c2->node && + c1->item_pos == c2->item_pos && + c1->unit_pos == c2->unit_pos && c1->between == c2->between; +} +#endif /* REISER4_DEBUG */ + +/* If coord_is_after_rightmost return NCOORD_ON_THE_RIGHT, if + coord_is_after_leftmost return NCOORD_ON_THE_LEFT, otherwise return + NCOORD_INSIDE. */ +/* Audited by: green(2002.06.15) */ +coord_wrt_node coord_wrt(const coord_t *coord) +{ + if (coord_is_before_leftmost(coord)) + return COORD_ON_THE_LEFT; + + if (coord_is_after_rightmost(coord)) + return COORD_ON_THE_RIGHT; + + return COORD_INSIDE; +} + +/* Returns true if the coordinate is positioned after the last item or after the + last unit of the last item or it is an empty node. */ +/* Audited by: green(2002.06.15) */ +int coord_is_after_rightmost(const coord_t *coord) +{ + assert("jmacd-7313", coord_check(coord)); + + switch (coord->between) { + case INVALID_COORD: + case AT_UNIT: + case BEFORE_UNIT: + case BEFORE_ITEM: + return 0; + + case EMPTY_NODE: + return 1; + + case AFTER_ITEM: + return (coord->item_pos == node_num_items(coord->node) - 1); + + case AFTER_UNIT: + return ((coord->item_pos == node_num_items(coord->node) - 1) && + coord->unit_pos == coord_last_unit_pos(coord)); + } + + impossible("jmacd-9908", "unreachable"); + return 0; +} + +/* Returns true if the coordinate is positioned before the first item or it is + an empty node. */ +int coord_is_before_leftmost(const coord_t *coord) +{ + /* FIXME-VS: coord_check requires node to be loaded whereas it is not + necessary to check if coord is set before leftmost + assert ("jmacd-7313", coord_check (coord)); */ + switch (coord->between) { + case INVALID_COORD: + case AT_UNIT: + case AFTER_ITEM: + case AFTER_UNIT: + return 0; + + case EMPTY_NODE: + return 1; + + case BEFORE_ITEM: + case BEFORE_UNIT: + return (coord->item_pos == 0) && (coord->unit_pos == 0); + } + + impossible("jmacd-9908", "unreachable"); + return 0; +} + +/* Returns true if the coordinate is positioned after a item, before a item, + after the last unit of an item, before the first unit of an item, or at an + empty node. */ +/* Audited by: green(2002.06.15) */ +int coord_is_between_items(const coord_t *coord) +{ + assert("jmacd-7313", coord_check(coord)); + + switch (coord->between) { + case INVALID_COORD: + case AT_UNIT: + return 0; + + case AFTER_ITEM: + case BEFORE_ITEM: + case EMPTY_NODE: + return 1; + + case BEFORE_UNIT: + return coord->unit_pos == 0; + + case AFTER_UNIT: + return coord->unit_pos == coord_last_unit_pos(coord); + } + + impossible("jmacd-9908", "unreachable"); + return 0; +} + +#if REISER4_DEBUG +/* Returns true if the coordinates are positioned at adjacent units, regardless + of before-after or item boundaries. */ +int coord_are_neighbors(coord_t *c1, coord_t *c2) +{ + coord_t *left; + coord_t *right; + + assert("nikita-1241", c1 != NULL); + assert("nikita-1242", c2 != NULL); + assert("nikita-1243", c1->node == c2->node); + assert("nikita-1244", coord_is_existing_unit(c1)); + assert("nikita-1245", coord_is_existing_unit(c2)); + + left = right = NULL; + switch (coord_compare(c1, c2)) { + case COORD_CMP_ON_LEFT: + left = c1; + right = c2; + break; + case COORD_CMP_ON_RIGHT: + left = c2; + right = c1; + break; + case COORD_CMP_SAME: + return 0; + default: + wrong_return_value("nikita-1246", "compare_coords()"); + } + assert("vs-731", left && right); + if (left->item_pos == right->item_pos) { + return left->unit_pos + 1 == right->unit_pos; + } else if (left->item_pos + 1 == right->item_pos) { + return (left->unit_pos == coord_last_unit_pos(left)) + && (right->unit_pos == 0); + } else { + return 0; + } +} +#endif /* REISER4_DEBUG */ + +/* Assuming two coordinates are positioned in the same node, return + COORD_CMP_ON_RIGHT, COORD_CMP_ON_LEFT, or COORD_CMP_SAME depending on c1's + position relative to c2. */ +/* Audited by: green(2002.06.15) */ +coord_cmp coord_compare(coord_t *c1, coord_t *c2) +{ + assert("vs-209", c1->node == c2->node); + assert("vs-194", coord_is_existing_unit(c1) + && coord_is_existing_unit(c2)); + + if (c1->item_pos > c2->item_pos) + return COORD_CMP_ON_RIGHT; + if (c1->item_pos < c2->item_pos) + return COORD_CMP_ON_LEFT; + if (c1->unit_pos > c2->unit_pos) + return COORD_CMP_ON_RIGHT; + if (c1->unit_pos < c2->unit_pos) + return COORD_CMP_ON_LEFT; + return COORD_CMP_SAME; +} + +/* If the coordinate is between items, shifts it to the right. Returns 0 on + success and non-zero if there is no position to the right. */ +int coord_set_to_right(coord_t *coord) +{ + unsigned items = coord_num_items(coord); + + if (coord_adjust_items(coord, items, 1) == 1) + return 1; + + switch (coord->between) { + case AT_UNIT: + return 0; + + case BEFORE_ITEM: + case BEFORE_UNIT: + coord->between = AT_UNIT; + return 0; + + case AFTER_UNIT: + if (coord->unit_pos < coord_last_unit_pos(coord)) { + coord->unit_pos += 1; + coord->between = AT_UNIT; + return 0; + } else { + + coord->unit_pos = 0; + + if (coord->item_pos == items - 1) { + coord->between = AFTER_ITEM; + return 1; + } + + coord_inc_item_pos(coord); + coord->between = AT_UNIT; + return 0; + } + + case AFTER_ITEM: + if (coord->item_pos == items - 1) + return 1; + + coord_inc_item_pos(coord); + coord->unit_pos = 0; + coord->between = AT_UNIT; + return 0; + + case EMPTY_NODE: + return 1; + + case INVALID_COORD: + break; + } + + impossible("jmacd-9920", "unreachable"); + return 0; +} + +/* If the coordinate is between items, shifts it to the left. Returns 0 on + success and non-zero if there is no position to the left. */ +int coord_set_to_left(coord_t *coord) +{ + unsigned items = coord_num_items(coord); + + if (coord_adjust_items(coord, items, 0) == 1) + return 1; + + switch (coord->between) { + case AT_UNIT: + return 0; + + case AFTER_UNIT: + coord->between = AT_UNIT; + return 0; + + case AFTER_ITEM: + coord->between = AT_UNIT; + coord->unit_pos = coord_last_unit_pos(coord); + return 0; + + case BEFORE_UNIT: + if (coord->unit_pos > 0) { + coord->unit_pos -= 1; + coord->between = AT_UNIT; + return 0; + } else { + + if (coord->item_pos == 0) { + coord->between = BEFORE_ITEM; + return 1; + } + + coord->unit_pos = coord_last_unit_pos(coord); + coord_dec_item_pos(coord); + coord->between = AT_UNIT; + return 0; + } + + case BEFORE_ITEM: + if (coord->item_pos == 0) + return 1; + + coord_dec_item_pos(coord); + coord->unit_pos = coord_last_unit_pos(coord); + coord->between = AT_UNIT; + return 0; + + case EMPTY_NODE: + return 1; + + case INVALID_COORD: + break; + } + + impossible("jmacd-9920", "unreachable"); + return 0; +} + +static const char *coord_tween_tostring(between_enum n) +{ + switch (n) { + case BEFORE_UNIT: + return "before unit"; + case BEFORE_ITEM: + return "before item"; + case AT_UNIT: + return "at unit"; + case AFTER_UNIT: + return "after unit"; + case AFTER_ITEM: + return "after item"; + case EMPTY_NODE: + return "empty node"; + case INVALID_COORD: + return "invalid"; + default: + { + static char buf[30]; + + sprintf(buf, "unknown: %i", n); + return buf; + } + } +} + +void print_coord(const char *mes, const coord_t *coord, int node) +{ + if (coord == NULL) { + printk("%s: null\n", mes); + return; + } + printk("%s: item_pos = %d, unit_pos %d, tween=%s, iplug=%d\n", + mes, coord->item_pos, coord->unit_pos, + coord_tween_tostring(coord->between), coord->iplugid); +} + +int +item_utmost_child_real_block(const coord_t *coord, sideof side, + reiser4_block_nr * blk) +{ + return item_plugin_by_coord(coord)->f.utmost_child_real_block(coord, + side, + blk); +} + +int item_utmost_child(const coord_t *coord, sideof side, jnode ** child) +{ + return item_plugin_by_coord(coord)->f.utmost_child(coord, side, child); +} + +/* @count bytes of flow @f got written, update correspondingly f->length, + f->data and f->key */ +void move_flow_forward(flow_t *f, unsigned count) +{ + if (f->data) + f->data += count; + f->length -= count; + set_key_offset(&f->key, get_key_offset(&f->key) + count); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/coord.h linux-2.6.30/fs/reiser4/coord.h --- linux-2.6.30.orig/fs/reiser4/coord.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/coord.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,399 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Coords */ + +#if !defined(__REISER4_COORD_H__) +#define __REISER4_COORD_H__ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" + +/* insertions happen between coords in the tree, so we need some means + of specifying the sense of betweenness. */ +typedef enum { + BEFORE_UNIT, /* Note: we/init_coord depends on this value being zero. */ + AT_UNIT, + AFTER_UNIT, + BEFORE_ITEM, + AFTER_ITEM, + INVALID_COORD, + EMPTY_NODE, +} between_enum; + +/* location of coord w.r.t. its node */ +typedef enum { + COORD_ON_THE_LEFT = -1, + COORD_ON_THE_RIGHT = +1, + COORD_INSIDE = 0 +} coord_wrt_node; + +typedef enum { + COORD_CMP_SAME = 0, COORD_CMP_ON_LEFT = -1, COORD_CMP_ON_RIGHT = +1 +} coord_cmp; + +struct coord { + /* node in a tree */ + /* 0 */ znode *node; + + /* position of item within node */ + /* 4 */ pos_in_node_t item_pos; + /* position of unit within item */ + /* 6 */ pos_in_node_t unit_pos; + /* optimization: plugin of item is stored in coord_t. Until this was + implemented, item_plugin_by_coord() was major CPU consumer. ->iplugid + is invalidated (set to 0xff) on each modification of ->item_pos, + and all such modifications are funneled through coord_*_item_pos() + functions below. + */ + /* 8 */ char iplugid; + /* position of coord w.r.t. to neighboring items and/or units. + Values are taken from &between_enum above. + */ + /* 9 */ char between; + /* padding. It will be added by the compiler anyway to conform to the + * C language alignment requirements. We keep it here to be on the + * safe side and to have a clear picture of the memory layout of this + * structure. */ + /* 10 */ __u16 pad; + /* 12 */ int offset; +#if REISER4_DEBUG + unsigned long plug_v; + unsigned long body_v; +#endif +}; + +#define INVALID_PLUGID ((char)((1 << 8) - 1)) +#define INVALID_OFFSET -1 + +static inline void coord_clear_iplug(coord_t *coord) +{ + assert("nikita-2835", coord != NULL); + coord->iplugid = INVALID_PLUGID; + coord->offset = INVALID_OFFSET; +} + +static inline int coord_is_iplug_set(const coord_t *coord) +{ + assert("nikita-2836", coord != NULL); + return coord->iplugid != INVALID_PLUGID; +} + +static inline void coord_set_item_pos(coord_t *coord, pos_in_node_t pos) +{ + assert("nikita-2478", coord != NULL); + coord->item_pos = pos; + coord_clear_iplug(coord); +} + +static inline void coord_dec_item_pos(coord_t *coord) +{ + assert("nikita-2480", coord != NULL); + --coord->item_pos; + coord_clear_iplug(coord); +} + +static inline void coord_inc_item_pos(coord_t *coord) +{ + assert("nikita-2481", coord != NULL); + ++coord->item_pos; + coord_clear_iplug(coord); +} + +static inline void coord_add_item_pos(coord_t *coord, int delta) +{ + assert("nikita-2482", coord != NULL); + coord->item_pos += delta; + coord_clear_iplug(coord); +} + +static inline void coord_invalid_item_pos(coord_t *coord) +{ + assert("nikita-2832", coord != NULL); + coord->item_pos = (unsigned short)~0; + coord_clear_iplug(coord); +} + +/* Reverse a direction. */ +static inline sideof sideof_reverse(sideof side) +{ + return side == LEFT_SIDE ? RIGHT_SIDE : LEFT_SIDE; +} + +/* NOTE: There is a somewhat odd mixture of the following opposed terms: + + "first" and "last" + "next" and "prev" + "before" and "after" + "leftmost" and "rightmost" + + But I think the chosen names are decent the way they are. +*/ + +/* COORD INITIALIZERS */ + +/* Initialize an invalid coordinate. */ +extern void coord_init_invalid(coord_t *coord, const znode * node); + +extern void coord_init_first_unit_nocheck(coord_t *coord, const znode * node); + +/* Initialize a coordinate to point at the first unit of the first item. If the + node is empty, it is positioned at the EMPTY_NODE. */ +extern void coord_init_first_unit(coord_t *coord, const znode * node); + +/* Initialize a coordinate to point at the last unit of the last item. If the + node is empty, it is positioned at the EMPTY_NODE. */ +extern void coord_init_last_unit(coord_t *coord, const znode * node); + +/* Initialize a coordinate to before the first item. If the node is empty, it is + positioned at the EMPTY_NODE. */ +extern void coord_init_before_first_item(coord_t *coord, const znode * node); + +/* Initialize a coordinate to after the last item. If the node is empty, it is + positioned at the EMPTY_NODE. */ +extern void coord_init_after_last_item(coord_t *coord, const znode * node); + +/* Initialize a coordinate to after last unit in the item. Coord must be set + already to existing item */ +void coord_init_after_item_end(coord_t *coord); + +/* Initialize a coordinate to before the item. Coord must be set already to + existing item */ +void coord_init_before_item(coord_t *); +/* Initialize a coordinate to after the item. Coord must be set already to + existing item */ +void coord_init_after_item(coord_t *); + +/* Calls either coord_init_first_unit or coord_init_last_unit depending on + sideof argument. */ +extern void coord_init_sideof_unit(coord_t *coord, const znode * node, + sideof dir); + +/* Initialize a coordinate by 0s. Used in places where init_coord was used and + it was not clear how actually + FIXME-VS: added by vs (2002, june, 8) */ +extern void coord_init_zero(coord_t *coord); + +/* COORD METHODS */ + +/* after shifting of node content, coord previously set properly may become + invalid, try to "normalize" it. */ +void coord_normalize(coord_t *coord); + +/* Copy a coordinate. */ +extern void coord_dup(coord_t *coord, const coord_t *old_coord); + +/* Copy a coordinate without check. */ +void coord_dup_nocheck(coord_t *coord, const coord_t *old_coord); + +unsigned coord_num_units(const coord_t *coord); + +/* Return the last valid unit number at the present item (i.e., + coord_num_units() - 1). */ +static inline unsigned coord_last_unit_pos(const coord_t *coord) +{ + return coord_num_units(coord) - 1; +} + +#if REISER4_DEBUG +/* For assertions only, checks for a valid coordinate. */ +extern int coord_check(const coord_t *coord); + +extern unsigned long znode_times_locked(const znode * z); + +static inline void coord_update_v(coord_t *coord) +{ + coord->plug_v = coord->body_v = znode_times_locked(coord->node); +} +#endif + +extern int coords_equal(const coord_t *c1, const coord_t *c2); + +extern void print_coord(const char *mes, const coord_t *coord, int print_node); + +/* If coord_is_after_rightmost return NCOORD_ON_THE_RIGHT, if + coord_is_after_leftmost return NCOORD_ON_THE_LEFT, otherwise return + NCOORD_INSIDE. */ +extern coord_wrt_node coord_wrt(const coord_t *coord); + +/* Returns true if the coordinates are positioned at adjacent units, regardless + of before-after or item boundaries. */ +extern int coord_are_neighbors(coord_t *c1, coord_t *c2); + +/* Assuming two coordinates are positioned in the same node, return + NCOORD_CMP_ON_RIGHT, NCOORD_CMP_ON_LEFT, or NCOORD_CMP_SAME depending on c1's + position relative to c2. */ +extern coord_cmp coord_compare(coord_t *c1, coord_t *c2); + +/* COORD PREDICATES */ + +/* Returns true if the coord was initializewd by coord_init_invalid (). */ +extern int coord_is_invalid(const coord_t *coord); + +/* Returns true if the coordinate is positioned at an existing item, not before + or after an item. It may be placed at, before, or after any unit within the + item, whether existing or not. If this is true you can call methods of the + item plugin. */ +extern int coord_is_existing_item(const coord_t *coord); + +/* Returns true if the coordinate is positioned after a item, before a item, + after the last unit of an item, before the first unit of an item, or at an + empty node. */ +extern int coord_is_between_items(const coord_t *coord); + +/* Returns true if the coordinate is positioned at an existing unit, not before + or after a unit. */ +extern int coord_is_existing_unit(const coord_t *coord); + +/* Returns true if the coordinate is positioned at an empty node. */ +extern int coord_is_empty(const coord_t *coord); + +/* Returns true if the coordinate is positioned at the first unit of the first + item. Not true for empty nodes nor coordinates positioned before the first + item. */ +extern int coord_is_leftmost_unit(const coord_t *coord); + +/* Returns true if the coordinate is positioned after the last item or after the + last unit of the last item or it is an empty node. */ +extern int coord_is_after_rightmost(const coord_t *coord); + +/* Returns true if the coordinate is positioned before the first item or it is + an empty node. */ +extern int coord_is_before_leftmost(const coord_t *coord); + +/* Calls either coord_is_before_leftmost or coord_is_after_rightmost depending + on sideof argument. */ +extern int coord_is_after_sideof_unit(coord_t *coord, sideof dir); + +/* COORD MODIFIERS */ + +/* Advances the coordinate by one unit to the right. If empty, no change. If + coord_is_rightmost_unit, advances to AFTER THE LAST ITEM. Returns 0 if new + position is an existing unit. */ +extern int coord_next_unit(coord_t *coord); + +/* Advances the coordinate by one item to the right. If empty, no change. If + coord_is_rightmost_unit, advances to AFTER THE LAST ITEM. Returns 0 if new + position is an existing item. */ +extern int coord_next_item(coord_t *coord); + +/* Advances the coordinate by one unit to the left. If empty, no change. If + coord_is_leftmost_unit, advances to BEFORE THE FIRST ITEM. Returns 0 if new + position is an existing unit. */ +extern int coord_prev_unit(coord_t *coord); + +/* Advances the coordinate by one item to the left. If empty, no change. If + coord_is_leftmost_unit, advances to BEFORE THE FIRST ITEM. Returns 0 if new + position is an existing item. */ +extern int coord_prev_item(coord_t *coord); + +/* If the coordinate is between items, shifts it to the right. Returns 0 on + success and non-zero if there is no position to the right. */ +extern int coord_set_to_right(coord_t *coord); + +/* If the coordinate is between items, shifts it to the left. Returns 0 on + success and non-zero if there is no position to the left. */ +extern int coord_set_to_left(coord_t *coord); + +/* If the coordinate is at an existing unit, set to after that unit. Returns 0 + on success and non-zero if the unit did not exist. */ +extern int coord_set_after_unit(coord_t *coord); + +/* Calls either coord_next_unit or coord_prev_unit depending on sideof + argument. */ +extern int coord_sideof_unit(coord_t *coord, sideof dir); + +/* iterate over all units in @node */ +#define for_all_units(coord, node) \ + for (coord_init_before_first_item((coord), (node)) ; \ + coord_next_unit(coord) == 0 ;) + +/* iterate over all items in @node */ +#define for_all_items(coord, node) \ + for (coord_init_before_first_item((coord), (node)) ; \ + coord_next_item(coord) == 0 ;) + +/* COORD/ITEM METHODS */ + +extern int item_utmost_child_real_block(const coord_t *coord, sideof side, + reiser4_block_nr * blk); +extern int item_utmost_child(const coord_t *coord, sideof side, + jnode ** child); + +/* a flow is a sequence of bytes being written to or read from the tree. The + tree will slice the flow into items while storing it into nodes, but all of + that is hidden from anything outside the tree. */ + +struct flow { + reiser4_key key; /* key of start of flow's sequence of bytes */ + loff_t length; /* length of flow's sequence of bytes */ + char *data; /* start of flow's sequence of bytes */ + int user; /* if 1 data is user space, 0 - kernel space */ + rw_op op; /* NIKITA-FIXME-HANS: comment is where? */ +}; + +void move_flow_forward(flow_t *f, unsigned count); + +/* &reiser4_item_data - description of data to be inserted or pasted + + Q: articulate the reasons for the difference between this and flow. + + A: Becides flow we insert into tree other things: stat data, directory + entry, etc. To insert them into tree one has to provide this structure. If + one is going to insert flow - he can use insert_flow, where this structure + does not have to be created +*/ +struct reiser4_item_data { + /* actual data to be inserted. If NULL, ->create_item() will not + do xmemcpy itself, leaving this up to the caller. This can + save some amount of unnecessary memory copying, for example, + during insertion of stat data. + + */ + char *data; + /* 1 if 'char * data' contains pointer to user space and 0 if it is + kernel space */ + int user; + /* amount of data we are going to insert or paste */ + int length; + /* "Arg" is opaque data that is passed down to the + ->create_item() method of node layout, which in turn + hands it to the ->create_hook() of item being created. This + arg is currently used by: + + . ->create_hook() of internal item + (fs/reiser4/plugin/item/internal.c:internal_create_hook()), + . ->paste() method of directory item. + . ->create_hook() of extent item + + For internal item, this is left "brother" of new node being + inserted and it is used to add new node into sibling list + after parent to it was just inserted into parent. + + While ->arg does look somewhat of unnecessary compication, + it actually saves a lot of headache in many places, because + all data necessary to insert or paste new data into tree are + collected in one place, and this eliminates a lot of extra + argument passing and storing everywhere. + + */ + void *arg; + /* plugin of item we are inserting */ + item_plugin *iplug; +}; + +/* __REISER4_COORD_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/debug.c linux-2.6.30/fs/reiser4/debug.c --- linux-2.6.30.orig/fs/reiser4/debug.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/debug.c 2009-06-22 18:11:49.000000000 +0200 @@ -0,0 +1,308 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Debugging facilities. */ + +/* + * This file contains generic debugging functions used by reiser4. Roughly + * following: + * + * panicking: reiser4_do_panic(), reiser4_print_prefix(). + * + * locking: + * reiser4_schedulable(), reiser4_lock_counters(), print_lock_counters(), + * reiser4_no_counters_are_held(), reiser4_commit_check_locks() + * + * error code monitoring (see comment before RETERR macro): + * reiser4_return_err(), reiser4_report_err(). + * + * stack back-tracing: fill_backtrace() + * + * miscellaneous: reiser4_preempt_point(), call_on_each_assert(), + * reiser4_debugtrap(). + * + */ + +#include "reiser4.h" +#include "context.h" +#include "super.h" +#include "txnmgr.h" +#include "znode.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#if 0 +#if REISER4_DEBUG +static void reiser4_report_err(void); +#else +#define reiser4_report_err() noop +#endif +#endif /* 0 */ + +/* + * global buffer where message given to reiser4_panic is formatted. + */ +static char panic_buf[REISER4_PANIC_MSG_BUFFER_SIZE]; + +/* + * lock protecting consistency of panic_buf under concurrent panics + */ +static DEFINE_SPINLOCK(panic_guard); + +/* Your best friend. Call it on each occasion. This is called by + fs/reiser4/debug.h:reiser4_panic(). */ +void reiser4_do_panic(const char *format/* format string */ , ... /* rest */) +{ + static int in_panic = 0; + va_list args; + + /* + * check for recursive panic. + */ + if (in_panic == 0) { + in_panic = 1; + + spin_lock(&panic_guard); + va_start(args, format); + vsnprintf(panic_buf, sizeof(panic_buf), format, args); + va_end(args); + printk(KERN_EMERG "reiser4 panicked cowardly: %s", panic_buf); + spin_unlock(&panic_guard); + + /* + * if kernel debugger is configured---drop in. Early dropping + * into kgdb is not always convenient, because panic message + * is not yet printed most of the times. But: + * + * (1) message can be extracted from printk_buf[] + * (declared static inside of printk()), and + * + * (2) sometimes serial/kgdb combo dies while printing + * long panic message, so it's more prudent to break into + * debugger earlier. + * + */ + DEBUGON(1); + } + /* to make gcc happy about noreturn attribute */ + panic("%s", panic_buf); +} + +#if 0 +void +reiser4_print_prefix(const char *level, int reperr, const char *mid, + const char *function, const char *file, int lineno) +{ + const char *comm; + int pid; + + if (unlikely(in_interrupt() || in_irq())) { + comm = "interrupt"; + pid = 0; + } else { + comm = current->comm; + pid = current->pid; + } + printk("%sreiser4[%.16s(%i)]: %s (%s:%i)[%s]:\n", + level, comm, pid, function, file, lineno, mid); + if (reperr) + reiser4_report_err(); +} +#endif /* 0 */ + +/* Preemption point: this should be called periodically during long running + operations (carry, allocate, and squeeze are best examples) */ +int reiser4_preempt_point(void) +{ + assert("nikita-3008", reiser4_schedulable()); + cond_resched(); + return signal_pending(current); +} + +#if REISER4_DEBUG +/* Debugging aid: return struct where information about locks taken by current + thread is accumulated. This can be used to formulate lock ordering + constraints and various assertions. + +*/ +reiser4_lock_cnt_info *reiser4_lock_counters(void) +{ + reiser4_context *ctx = get_current_context(); + assert("jmacd-1123", ctx != NULL); + return &ctx->locks; +} + +/* + * print human readable information about locks held by the reiser4 context. + */ +static void print_lock_counters(const char *prefix, + const reiser4_lock_cnt_info * info) +{ + printk("%s: jnode: %i, tree: %i (r:%i,w:%i), dk: %i (r:%i,w:%i)\n" + "jload: %i, " + "txnh: %i, atom: %i, stack: %i, txnmgr: %i, " + "ktxnmgrd: %i, fq: %i\n" + "inode: %i, " + "cbk_cache: %i (r:%i,w%i), " + "eflush: %i, " + "zlock: %i,\n" + "spin: %i, long: %i inode_sem: (r:%i,w:%i)\n" + "d: %i, x: %i, t: %i\n", prefix, + info->spin_locked_jnode, + info->rw_locked_tree, info->read_locked_tree, + info->write_locked_tree, + info->rw_locked_dk, info->read_locked_dk, info->write_locked_dk, + info->spin_locked_jload, + info->spin_locked_txnh, + info->spin_locked_atom, info->spin_locked_stack, + info->spin_locked_txnmgr, info->spin_locked_ktxnmgrd, + info->spin_locked_fq, + info->spin_locked_inode, + info->rw_locked_cbk_cache, + info->read_locked_cbk_cache, + info->write_locked_cbk_cache, + info->spin_locked_super_eflush, + info->spin_locked_zlock, + info->spin_locked, + info->long_term_locked_znode, + info->inode_sem_r, info->inode_sem_w, + info->d_refs, info->x_refs, info->t_refs); +} + +/* check that no spinlocks are held */ +int reiser4_schedulable(void) +{ + if (get_current_context_check() != NULL) { + if (!LOCK_CNT_NIL(spin_locked)) { + print_lock_counters("in atomic", reiser4_lock_counters()); + return 0; + } + } + might_sleep(); + return 1; +} +/* + * return true, iff no locks are held. + */ +int reiser4_no_counters_are_held(void) +{ + reiser4_lock_cnt_info *counters; + + counters = reiser4_lock_counters(); + return + (counters->spin_locked_zlock == 0) && + (counters->spin_locked_jnode == 0) && + (counters->rw_locked_tree == 0) && + (counters->read_locked_tree == 0) && + (counters->write_locked_tree == 0) && + (counters->rw_locked_dk == 0) && + (counters->read_locked_dk == 0) && + (counters->write_locked_dk == 0) && + (counters->spin_locked_txnh == 0) && + (counters->spin_locked_atom == 0) && + (counters->spin_locked_stack == 0) && + (counters->spin_locked_txnmgr == 0) && + (counters->spin_locked_inode == 0) && + (counters->spin_locked == 0) && + (counters->long_term_locked_znode == 0) && + (counters->inode_sem_r == 0) && + (counters->inode_sem_w == 0) && (counters->d_refs == 0); +} + +/* + * return true, iff transaction commit can be done under locks held by the + * current thread. + */ +int reiser4_commit_check_locks(void) +{ + reiser4_lock_cnt_info *counters; + int inode_sem_r; + int inode_sem_w; + int result; + + /* + * inode's read/write semaphore is the only reiser4 lock that can be + * held during commit. + */ + + counters = reiser4_lock_counters(); + inode_sem_r = counters->inode_sem_r; + inode_sem_w = counters->inode_sem_w; + + counters->inode_sem_r = counters->inode_sem_w = 0; + result = reiser4_no_counters_are_held(); + counters->inode_sem_r = inode_sem_r; + counters->inode_sem_w = inode_sem_w; + return result; +} + +/* + * fill "error site" in the current reiser4 context. See comment before RETERR + * macro for more details. + */ +void reiser4_return_err(int code, const char *file, int line) +{ + if (code < 0 && is_in_reiser4_context()) { + reiser4_context *ctx = get_current_context(); + + if (ctx != NULL) { + ctx->err.code = code; + ctx->err.file = file; + ctx->err.line = line; + } + } +} + +#if 0 +/* + * report error information recorder by reiser4_return_err(). + */ +static void reiser4_report_err(void) +{ + reiser4_context *ctx = get_current_context_check(); + + if (ctx != NULL) { + if (ctx->err.code != 0) { + printk("code: %i at %s:%i\n", + ctx->err.code, ctx->err.file, ctx->err.line); + } + } +} +#endif /* 0 */ + +#endif /* REISER4_DEBUG */ + +#if KERNEL_DEBUGGER + +/* + * this functions just drops into kernel debugger. It is a convenient place to + * put breakpoint in. + */ +void reiser4_debugtrap(void) +{ + /* do nothing. Put break point here. */ +#if defined(CONFIG_KGDB) && !defined(CONFIG_REISER4_FS_MODULE) + extern void kgdb_breakpoint(void); + kgdb_breakpoint(); +#endif +} +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/debug.h linux-2.6.30/fs/reiser4/debug.h --- linux-2.6.30.orig/fs/reiser4/debug.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/debug.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,351 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Declarations of debug macros. */ + +#if !defined(__FS_REISER4_DEBUG_H__) +#define __FS_REISER4_DEBUG_H__ + +#include "forward.h" +#include "reiser4.h" + +/* generic function to produce formatted output, decorating it with + whatever standard prefixes/postfixes we want. "Fun" is a function + that will be actually called, can be printk, panic etc. + This is for use by other debugging macros, not by users. */ +#define DCALL(lev, fun, reperr, label, format, ...) \ +({ \ + fun(lev "reiser4[%.16s(%i)]: %s (%s:%i)[%s]:\n" format "\n" , \ + current->comm, current->pid, __FUNCTION__, \ + __FILE__, __LINE__, label, ## __VA_ARGS__); \ +}) + +/* + * cause kernel to crash + */ +#define reiser4_panic(mid, format, ...) \ + DCALL("", reiser4_do_panic, 1, mid, format , ## __VA_ARGS__) + +/* print message with indication of current process, file, line and + function */ +#define reiser4_log(label, format, ...) \ + DCALL(KERN_DEBUG, printk, 0, label, format , ## __VA_ARGS__) + +/* Assertion checked during compilation. + If "cond" is false (0) we get duplicate case label in switch. + Use this to check something like famous + cassert (sizeof(struct reiserfs_journal_commit) == 4096) ; + in 3.x journal.c. If cassertion fails you get compiler error, + so no "maintainer-id". +*/ +#define cassert(cond) ({ switch (-1) { case (cond): case 0: break; } }) + +#define noop do {; } while (0) + +#if REISER4_DEBUG +/* version of info that only actually prints anything when _d_ebugging + is on */ +#define dinfo(format, ...) printk(format , ## __VA_ARGS__) +/* macro to catch logical errors. Put it into `default' clause of + switch() statement. */ +#define impossible(label, format, ...) \ + reiser4_panic(label, "impossible: " format , ## __VA_ARGS__) +/* assert assures that @cond is true. If it is not, reiser4_panic() is + called. Use this for checking logical consistency and _never_ call + this to check correctness of external data: disk blocks and user-input . */ +#define assert(label, cond) \ +({ \ + /* call_on_each_assert(); */ \ + if (cond) { \ + /* put negated check to avoid using !(cond) that would lose \ + * warnings for things like assert(a = b); */ \ + ; \ + } else { \ + DEBUGON(1); \ + reiser4_panic(label, "assertion failed: %s", #cond); \ + } \ +}) + +/* like assertion, but @expr is evaluated even if REISER4_DEBUG is off. */ +#define check_me(label, expr) assert(label, (expr)) + +#define ON_DEBUG(exp) exp + +extern int reiser4_schedulable(void); +extern void call_on_each_assert(void); + +#else + +#define dinfo(format, args...) noop +#define impossible(label, format, args...) noop +#define assert(label, cond) noop +#define check_me(label, expr) ((void) (expr)) +#define ON_DEBUG(exp) +#define reiser4_schedulable() might_sleep() + +/* REISER4_DEBUG */ +#endif + +#if REISER4_DEBUG +/* per-thread information about lock acquired by this thread. Used by lock + * ordering checking in spin_macros.h */ +typedef struct reiser4_lock_cnt_info { + int rw_locked_tree; + int read_locked_tree; + int write_locked_tree; + + int rw_locked_dk; + int read_locked_dk; + int write_locked_dk; + + int rw_locked_cbk_cache; + int read_locked_cbk_cache; + int write_locked_cbk_cache; + + int spin_locked_zlock; + int spin_locked_jnode; + int spin_locked_jload; + int spin_locked_txnh; + int spin_locked_atom; + int spin_locked_stack; + int spin_locked_txnmgr; + int spin_locked_ktxnmgrd; + int spin_locked_fq; + int spin_locked_inode; + int spin_locked_super_eflush; + int spin_locked; + int long_term_locked_znode; + + int inode_sem_r; + int inode_sem_w; + + int d_refs; + int x_refs; + int t_refs; +} reiser4_lock_cnt_info; + +extern struct reiser4_lock_cnt_info *reiser4_lock_counters(void); +#define IN_CONTEXT(a, b) (is_in_reiser4_context() ? (a) : (b)) + +/* increment lock-counter @counter, if present */ +#define LOCK_CNT_INC(counter) \ + IN_CONTEXT(++(reiser4_lock_counters()->counter), 0) + +/* decrement lock-counter @counter, if present */ +#define LOCK_CNT_DEC(counter) \ + IN_CONTEXT(--(reiser4_lock_counters()->counter), 0) + +/* check that lock-counter is zero. This is for use in assertions */ +#define LOCK_CNT_NIL(counter) \ + IN_CONTEXT(reiser4_lock_counters()->counter == 0, 1) + +/* check that lock-counter is greater than zero. This is for use in + * assertions */ +#define LOCK_CNT_GTZ(counter) \ + IN_CONTEXT(reiser4_lock_counters()->counter > 0, 1) +#define LOCK_CNT_LT(counter,n) \ + IN_CONTEXT(reiser4_lock_counters()->counter < n, 1) + +#else /* REISER4_DEBUG */ + +/* no-op versions on the above */ + +typedef struct reiser4_lock_cnt_info { +} reiser4_lock_cnt_info; + +#define reiser4_lock_counters() ((reiser4_lock_cnt_info *)NULL) +#define LOCK_CNT_INC(counter) noop +#define LOCK_CNT_DEC(counter) noop +#define LOCK_CNT_NIL(counter) (1) +#define LOCK_CNT_GTZ(counter) (1) +#define LOCK_CNT_LT(counter, n) (1) + +#endif /* REISER4_DEBUG */ + +#define assert_spin_not_locked(lock) BUG_ON(0) +#define assert_rw_write_locked(lock) BUG_ON(0) +#define assert_rw_read_locked(lock) BUG_ON(0) +#define assert_rw_locked(lock) BUG_ON(0) +#define assert_rw_not_write_locked(lock) BUG_ON(0) +#define assert_rw_not_read_locked(lock) BUG_ON(0) +#define assert_rw_not_locked(lock) BUG_ON(0) + +/* flags controlling debugging behavior. Are set through debug_flags=N mount + option. */ +typedef enum { + /* print a lot of information during panic. When this is on all jnodes + * are listed. This can be *very* large output. Usually you don't want + * this. Especially over serial line. */ + REISER4_VERBOSE_PANIC = 0x00000001, + /* print a lot of information during umount */ + REISER4_VERBOSE_UMOUNT = 0x00000002, + /* print gathered statistics on umount */ + REISER4_STATS_ON_UMOUNT = 0x00000004, + /* check node consistency */ + REISER4_CHECK_NODE = 0x00000008 +} reiser4_debug_flags; + +extern int is_in_reiser4_context(void); + +/* + * evaluate expression @e only if with reiser4 context + */ +#define ON_CONTEXT(e) do { \ + if (is_in_reiser4_context()) { \ + e; \ + } } while (0) + +/* + * evaluate expression @e only when within reiser4_context and debugging is + * on. + */ +#define ON_DEBUG_CONTEXT(e) ON_DEBUG(ON_CONTEXT(e)) + +/* + * complain about unexpected function result and crash. Used in "default" + * branches of switch statements and alike to assert that invalid results are + * not silently ignored. + */ +#define wrong_return_value(label, function) \ + impossible(label, "wrong return value from " function) + +/* Issue different types of reiser4 messages to the console */ +#define warning(label, format, ...) \ + DCALL(KERN_WARNING, \ + printk, 1, label, "WARNING: " format , ## __VA_ARGS__) +#define notice(label, format, ...) \ + DCALL(KERN_NOTICE, \ + printk, 1, label, "NOTICE: " format , ## __VA_ARGS__) + +/* mark not yet implemented functionality */ +#define not_yet(label, format, ...) \ + reiser4_panic(label, "NOT YET IMPLEMENTED: " format , ## __VA_ARGS__) + +extern void reiser4_do_panic(const char *format, ...) + __attribute__ ((noreturn, format(printf, 1, 2))); + +extern int reiser4_preempt_point(void); +extern void reiser4_print_stats(void); + +#if REISER4_DEBUG +extern int reiser4_no_counters_are_held(void); +extern int reiser4_commit_check_locks(void); +#else +#define reiser4_no_counters_are_held() (1) +#define reiser4_commit_check_locks() (1) +#endif + +/* true if @i is power-of-two. Useful for rate-limited warnings, etc. */ +#define IS_POW(i) \ +({ \ + typeof(i) __i; \ + \ + __i = (i); \ + !(__i & (__i - 1)); \ +}) + +#define KERNEL_DEBUGGER (1) + +#if KERNEL_DEBUGGER + +extern void reiser4_debugtrap(void); + +/* + * Check condition @cond and drop into kernel debugger (kgdb) if it's true. If + * kgdb is not compiled in, do nothing. + */ +#define DEBUGON(cond) \ +({ \ + if (unlikely(cond)) \ + reiser4_debugtrap(); \ +}) +#else +#define DEBUGON(cond) noop +#endif + +/* + * Error code tracing facility. (Idea is borrowed from XFS code.) + * + * Suppose some strange and/or unexpected code is returned from some function + * (for example, write(2) returns -EEXIST). It is possible to place a + * breakpoint in the reiser4_write(), but it is too late here. How to find out + * in what particular place -EEXIST was generated first? + * + * In reiser4 all places where actual error codes are produced (that is, + * statements of the form + * + * return -EFOO; // (1), or + * + * result = -EFOO; // (2) + * + * are replaced with + * + * return RETERR(-EFOO); // (1a), and + * + * result = RETERR(-EFOO); // (2a) respectively + * + * RETERR() macro fills a backtrace in reiser4_context. This back-trace is + * printed in error and warning messages. Moreover, it's possible to put a + * conditional breakpoint in reiser4_return_err (low-level function called + * by RETERR() to do the actual work) to break into debugger immediately + * when particular error happens. + * + */ + +#if REISER4_DEBUG + +/* + * data-type to store information about where error happened ("error site"). + */ +typedef struct err_site { + int code; /* error code */ + const char *file; /* source file, filled by __FILE__ */ + int line; /* source file line, filled by __LINE__ */ +} err_site; + +extern void reiser4_return_err(int code, const char *file, int line); + +/* + * fill &get_current_context()->err_site with error information. + */ +#define RETERR(code) \ +({ \ + typeof(code) __code; \ + \ + __code = (code); \ + reiser4_return_err(__code, __FILE__, __LINE__); \ + __code; \ +}) + +#else + +/* + * no-op versions of the above + */ + +typedef struct err_site { +} err_site; +#define RETERR(code) code +#endif + +#if REISER4_LARGE_KEY +/* + * conditionally compile arguments only if REISER4_LARGE_KEY is on. + */ +#define ON_LARGE_KEY(...) __VA_ARGS__ +#else +#define ON_LARGE_KEY(...) +#endif + +/* __FS_REISER4_DEBUG_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/dformat.h linux-2.6.30/fs/reiser4/dformat.h --- linux-2.6.30.orig/fs/reiser4/dformat.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/dformat.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,71 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Formats of on-disk data and conversion functions. */ + +/* put all item formats in the files describing the particular items, + our model is, everything you need to do to add an item to reiser4, + (excepting the changes to the plugin that uses the item which go + into the file defining that plugin), you put into one file. */ +/* Data on disk are stored in little-endian format. + To declare fields of on-disk structures, use d8, d16, d32 and d64. + d??tocpu() and cputod??() to convert. */ + +#if !defined(__FS_REISER4_DFORMAT_H__) +#define __FS_REISER4_DFORMAT_H__ + +#include +#include +#include + +typedef __u8 d8; +typedef __le16 d16; +typedef __le32 d32; +typedef __le64 d64; + +#define PACKED __attribute__((packed)) + +/* data-type for block number */ +typedef __u64 reiser4_block_nr; + +/* data-type for block number on disk, disk format */ +typedef __le64 reiser4_dblock_nr; + +/** + * disk_addr_eq - compare disk addresses + * @b1: pointer to block number ot compare + * @b2: pointer to block number ot compare + * + * Returns true if if disk addresses are the same + */ +static inline int disk_addr_eq(const reiser4_block_nr * b1, + const reiser4_block_nr * b2) +{ + assert("nikita-1033", b1 != NULL); + assert("nikita-1266", b2 != NULL); + + return !memcmp(b1, b2, sizeof *b1); +} + +/* structure of master reiser4 super block */ +typedef struct reiser4_master_sb { + char magic[16]; /* "ReIsEr4" */ + __le16 disk_plugin_id; /* id of disk layout plugin */ + __le16 blocksize; + char uuid[16]; /* unique id */ + char label[16]; /* filesystem label */ + __le64 diskmap; /* location of the diskmap. 0 if not present */ +} reiser4_master_sb; + +/* __FS_REISER4_DFORMAT_H__ */ +#endif + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/dscale.c linux-2.6.30/fs/reiser4/dscale.c --- linux-2.6.30.orig/fs/reiser4/dscale.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/dscale.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,192 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Scalable on-disk integers */ + +/* + * Various on-disk structures contain integer-like structures. Stat-data + * contain [yes, "data" is plural, check the dictionary] file size, link + * count; extent unit contains extent width etc. To accommodate for general + * case enough space is reserved to keep largest possible value. 64 bits in + * all cases above. But in overwhelming majority of cases numbers actually + * stored in these fields will be comparatively small and reserving 8 bytes is + * a waste of precious disk bandwidth. + * + * Scalable integers are one way to solve this problem. dscale_write() + * function stores __u64 value in the given area consuming from 1 to 9 bytes, + * depending on the magnitude of the value supplied. dscale_read() reads value + * previously stored by dscale_write(). + * + * dscale_write() produces format not completely unlike of UTF: two highest + * bits of the first byte are used to store "tag". One of 4 possible tag + * values is chosen depending on the number being encoded: + * + * 0 ... 0x3f => 0 [table 1] + * 0x40 ... 0x3fff => 1 + * 0x4000 ... 0x3fffffff => 2 + * 0x40000000 ... 0xffffffffffffffff => 3 + * + * (see dscale_range() function) + * + * Values in the range 0x40000000 ... 0xffffffffffffffff require 8 full bytes + * to be stored, so in this case there is no place in the first byte to store + * tag. For such values tag is stored in an extra 9th byte. + * + * As _highest_ bits are used for the test (which is natural) scaled integers + * are stored in BIG-ENDIAN format in contrast with the rest of reiser4 which + * uses LITTLE-ENDIAN. + * + */ + +#include "debug.h" +#include "dscale.h" + +/* return tag of scaled integer stored at @address */ +static int gettag(const unsigned char *address) +{ + /* tag is stored in two highest bits */ + return (*address) >> 6; +} + +/* clear tag from value. Clear tag embedded into @value. */ +static void cleartag(__u64 *value, int tag) +{ + /* + * W-w-what ?! + * + * Actually, this is rather simple: @value passed here was read by + * dscale_read(), converted from BIG-ENDIAN, and padded to __u64 by + * zeroes. Tag is still stored in the highest (arithmetically) + * non-zero bits of @value, but relative position of tag within __u64 + * depends on @tag. + * + * For example if @tag is 0, it's stored 2 highest bits of lowest + * byte, and its offset (counting from lowest bit) is 8 - 2 == 6 bits. + * + * If tag is 1, it's stored in two highest bits of 2nd lowest byte, + * and it's offset if (2 * 8) - 2 == 14 bits. + * + * See table 1 above for details. + * + * All these cases are captured by the formula: + */ + *value &= ~(3 << (((1 << tag) << 3) - 2)); + /* + * That is, clear two (3 == 0t11) bits at the offset + * + * 8 * (2 ^ tag) - 2, + * + * that is, two highest bits of (2 ^ tag)-th byte of @value. + */ +} + +/* return tag for @value. See table 1 above for details. */ +static int dscale_range(__u64 value) +{ + if (value > 0x3fffffff) + return 3; + if (value > 0x3fff) + return 2; + if (value > 0x3f) + return 1; + return 0; +} + +/* restore value stored at @adderss by dscale_write() and return number of + * bytes consumed */ +int dscale_read(unsigned char *address, __u64 *value) +{ + int tag; + + /* read tag */ + tag = gettag(address); + switch (tag) { + case 3: + /* In this case tag is stored in an extra byte, skip this byte + * and decode value stored in the next 8 bytes.*/ + *value = __be64_to_cpu(get_unaligned((__be64 *)(address + 1))); + /* worst case: 8 bytes for value itself plus one byte for + * tag. */ + return 9; + case 0: + *value = get_unaligned(address); + break; + case 1: + *value = __be16_to_cpu(get_unaligned((__be16 *)address)); + break; + case 2: + *value = __be32_to_cpu(get_unaligned((__be32 *)address)); + break; + default: + return RETERR(-EIO); + } + /* clear tag embedded into @value */ + cleartag(value, tag); + /* number of bytes consumed is (2 ^ tag)---see table 1. */ + return 1 << tag; +} + +/* number of bytes consumed */ +int dscale_bytes_to_read(unsigned char *address) +{ + int tag; + + tag = gettag(address); + switch (tag) { + case 0: + case 1: + case 2: + return 1 << tag; + case 3: + return 9; + default: + return RETERR(-EIO); + } +} + +/* store @value at @address and return number of bytes consumed */ +int dscale_write(unsigned char *address, __u64 value) +{ + int tag; + int shift; + __be64 v; + unsigned char *valarr; + + tag = dscale_range(value); + v = __cpu_to_be64(value); + valarr = (unsigned char *)&v; + shift = (tag == 3) ? 1 : 0; + memcpy(address + shift, valarr + sizeof v - (1 << tag), 1 << tag); + *address |= (tag << 6); + return shift + (1 << tag); +} + +/* number of bytes required to store @value */ +int dscale_bytes_to_write(__u64 value) +{ + int bytes; + + bytes = 1 << dscale_range(value); + if (bytes == 8) + ++bytes; + return bytes; +} + +/* returns true if @value and @other require the same number of bytes to be + * stored. Used by detect when data structure (like stat-data) has to be + * expanded or contracted. */ +int dscale_fit(__u64 value, __u64 other) +{ + return dscale_range(value) == dscale_range(other); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/dscale.h linux-2.6.30/fs/reiser4/dscale.h --- linux-2.6.30.orig/fs/reiser4/dscale.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/dscale.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,28 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Scalable on-disk integers. See dscale.h for details. */ + +#if !defined(__FS_REISER4_DSCALE_H__) +#define __FS_REISER4_DSCALE_H__ + +#include "dformat.h" + +extern int dscale_read(unsigned char *address, __u64 *value); +extern int dscale_write(unsigned char *address, __u64 value); +extern int dscale_bytes_to_read(unsigned char *address); +extern int dscale_bytes_to_write(__u64 value); +extern int dscale_fit(__u64 value, __u64 other); + +/* __FS_REISER4_DSCALE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/entd.c linux-2.6.30/fs/reiser4/entd.c --- linux-2.6.30.orig/fs/reiser4/entd.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/entd.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,335 @@ +/* Copyright 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Ent daemon. */ + +#include "debug.h" +#include "txnmgr.h" +#include "tree.h" +#include "entd.h" +#include "super.h" +#include "context.h" +#include "reiser4.h" +#include "vfs_ops.h" +#include "page_cache.h" +#include "inode.h" + +#include /* struct task_struct */ +#include +#include +#include +#include /* INITIAL_JIFFIES */ +#include /* bdi_write_congested */ +#include +#include +#include + +#define DEF_PRIORITY 12 +#define MAX_ENTD_ITERS 10 + +static void entd_flush(struct super_block *, struct wbq *); +static int entd(void *arg); + +/* + * set ->comm field of end thread to make its state visible to the user level + */ +#define entd_set_comm(state) \ + snprintf(current->comm, sizeof(current->comm), \ + "ent:%s%s", super->s_id, (state)) + +/** + * reiser4_init_entd - initialize entd context and start kernel daemon + * @super: super block to start ent thread for + * + * Creates entd contexts, starts kernel thread and waits until it + * initializes. + */ +int reiser4_init_entd(struct super_block *super) +{ + entd_context *ctx; + + assert("nikita-3104", super != NULL); + + ctx = get_entd_context(super); + + memset(ctx, 0, sizeof *ctx); + spin_lock_init(&ctx->guard); + init_waitqueue_head(&ctx->wait); +#if REISER4_DEBUG + INIT_LIST_HEAD(&ctx->flushers_list); +#endif + /* lists of writepage requests */ + INIT_LIST_HEAD(&ctx->todo_list); + INIT_LIST_HEAD(&ctx->done_list); + /* start entd */ + ctx->tsk = kthread_run(entd, super, "ent:%s", super->s_id); + if (IS_ERR(ctx->tsk)) + return PTR_ERR(ctx->tsk); + return 0; +} + +static void put_wbq(struct wbq *rq) +{ + iput(rq->mapping->host); + complete(&rq->completion); +} + +/* ent should be locked */ +static struct wbq *__get_wbq(entd_context * ent) +{ + struct wbq *wbq; + + if (list_empty(&ent->todo_list)) + return NULL; + + ent->nr_todo_reqs--; + wbq = list_entry(ent->todo_list.next, struct wbq, link); + list_del_init(&wbq->link); + return wbq; +} + +/* ent thread function */ +static int entd(void *arg) +{ + struct super_block *super; + entd_context *ent; + int done = 0; + + super = arg; + /* do_fork() just copies task_struct into the new + thread. ->fs_context shouldn't be copied of course. This shouldn't + be a problem for the rest of the code though. + */ + current->journal_info = NULL; + + ent = get_entd_context(super); + + while (!done) { + try_to_freeze(); + + spin_lock(&ent->guard); + while (ent->nr_todo_reqs != 0) { + struct wbq *rq; + + assert("", list_empty(&ent->done_list)); + + /* take request from the queue head */ + rq = __get_wbq(ent); + assert("", rq != NULL); + ent->cur_request = rq; + spin_unlock(&ent->guard); + + entd_set_comm("!"); + entd_flush(super, rq); + + put_wbq(rq); + + /* + * wakeup all requestors and iput their inodes + */ + spin_lock(&ent->guard); + while (!list_empty(&ent->done_list)) { + rq = list_entry(ent->done_list.next, struct wbq, link); + list_del_init(&rq->link); + ent->nr_done_reqs--; + spin_unlock(&ent->guard); + assert("", rq->written == 1); + put_wbq(rq); + spin_lock(&ent->guard); + } + } + spin_unlock(&ent->guard); + + entd_set_comm("."); + + { + DEFINE_WAIT(__wait); + + do { + prepare_to_wait(&ent->wait, &__wait, TASK_INTERRUPTIBLE); + if (kthread_should_stop()) { + done = 1; + break; + } + if (ent->nr_todo_reqs != 0) + break; + schedule(); + } while (0); + finish_wait(&ent->wait, &__wait); + } + } + BUG_ON(ent->nr_todo_reqs != 0); + return 0; +} + +/** + * reiser4_done_entd - stop entd kernel thread + * @super: super block to stop ent thread for + * + * It is called on umount. Sends stop signal to entd and wait until it handles + * it. + */ +void reiser4_done_entd(struct super_block *super) +{ + entd_context *ent; + + assert("nikita-3103", super != NULL); + + ent = get_entd_context(super); + assert("zam-1055", ent->tsk != NULL); + kthread_stop(ent->tsk); +} + +/* called at the beginning of jnode_flush to register flusher thread with ent + * daemon */ +void reiser4_enter_flush(struct super_block *super) +{ + entd_context *ent; + + assert("zam-1029", super != NULL); + ent = get_entd_context(super); + + assert("zam-1030", ent != NULL); + + spin_lock(&ent->guard); + ent->flushers++; +#if REISER4_DEBUG + list_add(&get_current_context()->flushers_link, &ent->flushers_list); +#endif + spin_unlock(&ent->guard); +} + +/* called at the end of jnode_flush */ +void reiser4_leave_flush(struct super_block *super) +{ + entd_context *ent; + int wake_up_ent; + + assert("zam-1027", super != NULL); + ent = get_entd_context(super); + + assert("zam-1028", ent != NULL); + + spin_lock(&ent->guard); + ent->flushers--; + wake_up_ent = (ent->flushers == 0 && ent->nr_todo_reqs != 0); +#if REISER4_DEBUG + list_del_init(&get_current_context()->flushers_link); +#endif + spin_unlock(&ent->guard); + if (wake_up_ent) + wake_up_process(ent->tsk); +} + +#define ENTD_CAPTURE_APAGE_BURST SWAP_CLUSTER_MAX + +static void entd_flush(struct super_block *super, struct wbq *rq) +{ + reiser4_context ctx; + int tmp; + + init_stack_context(&ctx, super); + ctx.entd = 1; + ctx.gfp_mask = GFP_NOFS; + + rq->wbc->range_start = page_offset(rq->page); + rq->wbc->range_end = rq->wbc->range_start + + (ENTD_CAPTURE_APAGE_BURST << PAGE_CACHE_SHIFT); + tmp = rq->wbc->nr_to_write; + rq->mapping->a_ops->writepages(rq->mapping, rq->wbc); + + if (rq->wbc->nr_to_write > 0) { + rq->wbc->range_start = 0; + rq->wbc->range_end = LLONG_MAX; + generic_sync_sb_inodes(super, rq->wbc); + } + rq->wbc->nr_to_write = ENTD_CAPTURE_APAGE_BURST; + reiser4_writeout(super, rq->wbc); + + context_set_commit_async(&ctx); + reiser4_exit_context(&ctx); +} + +/** + * write_page_by_ent - ask entd thread to flush this page as part of slum + * @page: page to be written + * @wbc: writeback control passed to reiser4_writepage + * + * Creates a request, puts it on entd list of requests, wakeups entd if + * necessary, waits until entd completes with the request. + */ +int write_page_by_ent(struct page *page, struct writeback_control *wbc) +{ + struct super_block *sb; + struct inode *inode; + entd_context *ent; + struct wbq rq; + + assert("", PageLocked(page)); + assert("", page->mapping != NULL); + + sb = page->mapping->host->i_sb; + ent = get_entd_context(sb); + assert("", ent && ent->done == 0); + + /* + * we are going to unlock page and ask ent thread to write the + * page. Re-dirty page before unlocking so that if ent thread fails to + * write it - it will remain dirty + */ + set_page_dirty_notag(page); + + /* + * pin inode in memory, unlock page, entd_flush will iput. We can not + * iput here becasue we can not allow delete_inode to be called here + */ + inode = igrab(page->mapping->host); + unlock_page(page); + if (inode == NULL) + /* inode is getting freed */ + return 0; + + /* init wbq */ + INIT_LIST_HEAD(&rq.link); + rq.magic = WBQ_MAGIC; + rq.wbc = wbc; + rq.page = page; + rq.mapping = inode->i_mapping; + rq.node = NULL; + rq.written = 0; + init_completion(&rq.completion); + + /* add request to entd's list of writepage requests */ + spin_lock(&ent->guard); + ent->nr_todo_reqs++; + list_add_tail(&rq.link, &ent->todo_list); + if (ent->nr_todo_reqs == 1) + wake_up_process(ent->tsk); + + spin_unlock(&ent->guard); + + /* wait until entd finishes */ + wait_for_completion(&rq.completion); + + if (rq.written) + /* Eventually ENTD has written the page to disk. */ + return 0; + return 0; +} + +int wbq_available(void) +{ + struct super_block *sb = reiser4_get_current_sb(); + entd_context *ent = get_entd_context(sb); + return ent->nr_todo_reqs; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/entd.h linux-2.6.30/fs/reiser4/entd.h --- linux-2.6.30.orig/fs/reiser4/entd.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/entd.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,90 @@ +/* Copyright 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Ent daemon. */ + +#ifndef __ENTD_H__ +#define __ENTD_H__ + +#include "context.h" + +#include +#include +#include +#include +#include /* for struct task_struct */ + +#define WBQ_MAGIC 0x7876dc76 + +/* write-back request. */ +struct wbq { + int magic; + struct list_head link; /* list head of this list is in entd context */ + struct writeback_control *wbc; + struct page *page; + struct address_space *mapping; + struct completion completion; + jnode *node; /* set if ent thread captured requested page */ + int written; /* set if ent thread wrote requested page */ +}; + +/* ent-thread context. This is used to synchronize starting/stopping ent + * threads. */ +typedef struct entd_context { + /* wait queue that ent thread waits on for more work. It's + * signaled by write_page_by_ent(). */ + wait_queue_head_t wait; + /* spinlock protecting other fields */ + spinlock_t guard; + /* ent thread */ + struct task_struct *tsk; + /* set to indicate that ent thread should leave. */ + int done; + /* counter of active flushers */ + int flushers; + /* + * when reiser4_writepage asks entd to write a page - it adds struct + * wbq to this list + */ + struct list_head todo_list; + /* number of elements on the above list */ + int nr_todo_reqs; + + struct wbq *cur_request; + /* + * when entd writes a page it moves write-back request from todo_list + * to done_list. This list is used at the end of entd iteration to + * wakeup requestors and iput inodes. + */ + struct list_head done_list; + /* number of elements on the above list */ + int nr_done_reqs; + +#if REISER4_DEBUG + /* list of all active flushers */ + struct list_head flushers_list; +#endif +} entd_context; + +extern int reiser4_init_entd(struct super_block *); +extern void reiser4_done_entd(struct super_block *); + +extern void reiser4_enter_flush(struct super_block *); +extern void reiser4_leave_flush(struct super_block *); + +extern int write_page_by_ent(struct page *, struct writeback_control *); +extern int wbq_available(void); +extern void ent_writes_page(struct super_block *, struct page *); + +extern jnode *get_jnode_by_wbq(struct super_block *, struct wbq *); +/* __ENTD_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/eottl.c linux-2.6.30/fs/reiser4/eottl.c --- linux-2.6.30.orig/fs/reiser4/eottl.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/eottl.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,510 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +#include "forward.h" +#include "debug.h" +#include "key.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "plugin/node/node.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree_walk.h" +#include "tree_mod.h" +#include "carry.h" +#include "tree.h" +#include "super.h" + +#include /* for __u?? */ + +/* + * Extents on the twig level (EOTTL) handling. + * + * EOTTL poses some problems to the tree traversal, that are better explained + * by example. + * + * Suppose we have block B1 on the twig level with the following items: + * + * 0. internal item I0 with key (0:0:0:0) (locality, key-type, object-id, + * offset) + * 1. extent item E1 with key (1:4:100:0), having 10 blocks of 4k each + * 2. internal item I2 with key (10:0:0:0) + * + * We are trying to insert item with key (5:0:0:0). Lookup finds node B1, and + * then intra-node lookup is done. This lookup finished on the E1, because the + * key we are looking for is larger than the key of E1 and is smaller than key + * the of I2. + * + * Here search is stuck. + * + * After some thought it is clear what is wrong here: extents on the twig level + * break some basic property of the *search* tree (on the pretext, that they + * restore property of balanced tree). + * + * Said property is the following: if in the internal node of the search tree + * we have [ ... Key1 Pointer Key2 ... ] then, all data that are or will be + * keyed in the tree with the Key such that Key1 <= Key < Key2 are accessible + * through the Pointer. + * + * This is not true, when Pointer is Extent-Pointer, simply because extent + * cannot expand indefinitely to the right to include any item with + * + * Key1 <= Key <= Key2. + * + * For example, our E1 extent is only responsible for the data with keys + * + * (1:4:100:0) <= key <= (1:4:100:0xffffffffffffffff), and + * + * so, key range + * + * ( (1:4:100:0xffffffffffffffff), (10:0:0:0) ) + * + * is orphaned: there is no way to get there from the tree root. + * + * In other words, extent pointers are different than normal child pointers as + * far as search tree is concerned, and this creates such problems. + * + * Possible solution for this problem is to insert our item into node pointed + * to by I2. There are some problems through: + * + * (1) I2 can be in a different node. + * (2) E1 can be immediately followed by another extent E2. + * + * (1) is solved by calling reiser4_get_right_neighbor() and accounting + * for locks/coords as necessary. + * + * (2) is more complex. Solution here is to insert new empty leaf node and + * insert internal item between E1 and E2 pointing to said leaf node. This is + * further complicated by possibility that E2 is in a different node, etc. + * + * Problems: + * + * (1) if there was internal item I2 immediately on the right of an extent E1 + * we and we decided to insert new item S1 into node N2 pointed to by I2, then + * key of S1 will be less than smallest key in the N2. Normally, search key + * checks that key we are looking for is in the range of keys covered by the + * node key is being looked in. To work around of this situation, while + * preserving useful consistency check new flag CBK_TRUST_DK was added to the + * cbk falgs bitmask. This flag is automatically set on entrance to the + * coord_by_key() and is only cleared when we are about to enter situation + * described above. + * + * (2) If extent E1 is immediately followed by another extent E2 and we are + * searching for the key that is between E1 and E2 we only have to insert new + * empty leaf node when coord_by_key was called for insertion, rather than just + * for lookup. To distinguish these cases, new flag CBK_FOR_INSERT was added to + * the cbk falgs bitmask. This flag is automatically set by coord_by_key calls + * performed by insert_by_key() and friends. + * + * (3) Insertion of new empty leaf node (possibly) requires balancing. In any + * case it requires modification of node content which is only possible under + * write lock. It may well happen that we only have read lock on the node where + * new internal pointer is to be inserted (common case: lookup of non-existent + * stat-data that fells between two extents). If only read lock is held, tree + * traversal is restarted with lock_level modified so that next time we hit + * this problem, write lock will be held. Once we have write lock, balancing + * will be performed. + */ + +/** + * is_next_item_internal - check whether next item is internal + * @coord: coordinate of extent item in twig node + * @key: search key + * @lh: twig node lock handle + * + * Looks at the unit next to @coord. If it is an internal one - 1 is returned, + * @coord is set to that unit. If that unit is in right neighbor, @lh is moved + * to that node, @coord is set to its first unit. If next item is not internal + * or does not exist then 0 is returned, @coord and @lh are left unchanged. 2 + * is returned if search restart has to be done. + */ +static int +is_next_item_internal(coord_t *coord, const reiser4_key * key, + lock_handle * lh) +{ + coord_t next; + lock_handle rn; + int result; + + coord_dup(&next, coord); + if (coord_next_unit(&next) == 0) { + /* next unit is in this node */ + if (item_is_internal(&next)) { + coord_dup(coord, &next); + return 1; + } + assert("vs-3", item_is_extent(&next)); + return 0; + } + + /* + * next unit either does not exist or is in right neighbor. If it is in + * right neighbor we have to check right delimiting key because + * concurrent thread could get their first and insert item with a key + * smaller than @key + */ + read_lock_dk(current_tree); + result = keycmp(key, znode_get_rd_key(coord->node)); + read_unlock_dk(current_tree); + assert("vs-6", result != EQUAL_TO); + if (result == GREATER_THAN) + return 2; + + /* lock right neighbor */ + init_lh(&rn); + result = reiser4_get_right_neighbor(&rn, coord->node, + znode_is_wlocked(coord->node) ? + ZNODE_WRITE_LOCK : ZNODE_READ_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (result == -E_NO_NEIGHBOR) { + /* we are on the rightmost edge of the tree */ + done_lh(&rn); + return 0; + } + + if (result) { + assert("vs-4", result < 0); + done_lh(&rn); + return result; + } + + /* + * check whether concurrent thread managed to insert item with a key + * smaller than @key + */ + read_lock_dk(current_tree); + result = keycmp(key, znode_get_ld_key(rn.node)); + read_unlock_dk(current_tree); + assert("vs-6", result != EQUAL_TO); + if (result == GREATER_THAN) { + done_lh(&rn); + return 2; + } + + result = zload(rn.node); + if (result) { + assert("vs-5", result < 0); + done_lh(&rn); + return result; + } + + coord_init_first_unit(&next, rn.node); + if (item_is_internal(&next)) { + /* + * next unit is in right neighbor and it is an unit of internal + * item. Unlock coord->node. Move @lh to right neighbor. @coord + * is set to the first unit of right neighbor. + */ + coord_dup(coord, &next); + zrelse(rn.node); + done_lh(lh); + move_lh(lh, &rn); + return 1; + } + + /* + * next unit is unit of extent item. Return without chaning @lh and + * @coord. + */ + assert("vs-6", item_is_extent(&next)); + zrelse(rn.node); + done_lh(&rn); + return 0; +} + +/** + * rd_key - calculate key of an item next to the given one + * @coord: position in a node + * @key: storage for result key + * + * @coord is set between items or after the last item in a node. Calculate key + * of item to the right of @coord. + */ +static reiser4_key *rd_key(const coord_t *coord, reiser4_key *key) +{ + coord_t dup; + + assert("nikita-2281", coord_is_between_items(coord)); + coord_dup(&dup, coord); + + if (coord_set_to_right(&dup) == 0) + /* next item is in this node. Return its key. */ + unit_key_by_coord(&dup, key); + else { + /* + * next item either does not exist or is in right + * neighbor. Return znode's right delimiting key. + */ + read_lock_dk(current_tree); + *key = *znode_get_rd_key(coord->node); + read_unlock_dk(current_tree); + } + return key; +} + +/** + * add_empty_leaf - insert empty leaf between two extents + * @insert_coord: position in twig node between two extents + * @lh: twig node lock handle + * @key: left delimiting key of new node + * @rdkey: right delimiting key of new node + * + * Inserts empty leaf node between two extent items. It is necessary when we + * have to insert an item on leaf level between two extents (items on the twig + * level). + */ +static int +add_empty_leaf(coord_t *insert_coord, lock_handle *lh, + const reiser4_key *key, const reiser4_key *rdkey) +{ + int result; + carry_pool *pool; + carry_level *todo; + reiser4_item_data *item; + carry_insert_data *cdata; + carry_op *op; + znode *node; + reiser4_tree *tree; + + assert("vs-49827", znode_contains_key_lock(insert_coord->node, key)); + tree = znode_get_tree(insert_coord->node); + node = reiser4_new_node(insert_coord->node, LEAF_LEVEL); + if (IS_ERR(node)) + return PTR_ERR(node); + + /* setup delimiting keys for node being inserted */ + write_lock_dk(tree); + znode_set_ld_key(node, key); + znode_set_rd_key(node, rdkey); + ON_DEBUG(node->creator = current); + ON_DEBUG(node->first_key = *key); + write_unlock_dk(tree); + + ZF_SET(node, JNODE_ORPHAN); + + /* + * allocate carry_pool, 3 carry_level-s, reiser4_item_data and + * carry_insert_data + */ + pool = init_carry_pool(sizeof(*pool) + 3 * sizeof(*todo) + + sizeof(*item) + sizeof(*cdata)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + todo = (carry_level *) (pool + 1); + init_carry_level(todo, pool); + + item = (reiser4_item_data *) (todo + 3); + cdata = (carry_insert_data *) (item + 1); + + op = reiser4_post_carry(todo, COP_INSERT, insert_coord->node, 0); + if (!IS_ERR(op)) { + cdata->coord = insert_coord; + cdata->key = key; + cdata->data = item; + op->u.insert.d = cdata; + op->u.insert.type = COPT_ITEM_DATA; + build_child_ptr_data(node, item); + item->arg = NULL; + /* have @insert_coord to be set at inserted item after + insertion is done */ + todo->track_type = CARRY_TRACK_CHANGE; + todo->tracked = lh; + + result = reiser4_carry(todo, NULL); + if (result == 0) { + /* + * pin node in memory. This is necessary for + * znode_make_dirty() below. + */ + result = zload(node); + if (result == 0) { + lock_handle local_lh; + + /* + * if we inserted new child into tree we have + * to mark it dirty so that flush will be able + * to process it. + */ + init_lh(&local_lh); + result = longterm_lock_znode(&local_lh, node, + ZNODE_WRITE_LOCK, + ZNODE_LOCK_LOPRI); + if (result == 0) { + znode_make_dirty(node); + + /* + * when internal item pointing to @node + * was inserted into twig node + * create_hook_internal did not connect + * it properly because its right + * neighbor was not known. Do it + * here + */ + write_lock_tree(tree); + assert("nikita-3312", + znode_is_right_connected(node)); + assert("nikita-2984", + node->right == NULL); + ZF_CLR(node, JNODE_RIGHT_CONNECTED); + write_unlock_tree(tree); + result = + connect_znode(insert_coord, node); + ON_DEBUG(if (result == 0) check_dkeys(node);); + + done_lh(lh); + move_lh(lh, &local_lh); + assert("vs-1676", node_is_empty(node)); + coord_init_first_unit(insert_coord, + node); + } else { + warning("nikita-3136", + "Cannot lock child"); + } + done_lh(&local_lh); + zrelse(node); + } + } + } else + result = PTR_ERR(op); + zput(node); + done_carry_pool(pool); + return result; +} + +/** + * handle_eottl - handle extent-on-the-twig-level cases in tree traversal + * @h: search handle + * @outcome: flag saying whether search has to restart or is done + * + * Handles search on twig level. If this function completes search itself then + * it returns 1. If search has to go one level down then 0 is returned. If + * error happens then LOOKUP_DONE is returned via @outcome and error code is + * saved in @h->result. + */ +int handle_eottl(cbk_handle *h, int *outcome) +{ + int result; + reiser4_key key; + coord_t *coord; + + coord = h->coord; + + if (h->level != TWIG_LEVEL || + (coord_is_existing_item(coord) && item_is_internal(coord))) { + /* Continue to traverse tree downward. */ + return 0; + } + + /* + * make sure that @h->coord is set to twig node and that it is either + * set to extent item or after extent item + */ + assert("vs-356", h->level == TWIG_LEVEL); + assert("vs-357", ({ + coord_t lcoord; + coord_dup(&lcoord, coord); + check_me("vs-733", coord_set_to_left(&lcoord) == 0); + item_is_extent(&lcoord); + } + )); + + if (*outcome == NS_FOUND) { + /* we have found desired key on twig level in extent item */ + h->result = CBK_COORD_FOUND; + *outcome = LOOKUP_DONE; + return 1; + } + + if (!(h->flags & CBK_FOR_INSERT)) { + /* tree traversal is not for insertion. Just return + CBK_COORD_NOTFOUND. */ + h->result = CBK_COORD_NOTFOUND; + *outcome = LOOKUP_DONE; + return 1; + } + + /* take a look at the item to the right of h -> coord */ + result = is_next_item_internal(coord, h->key, h->active_lh); + if (unlikely(result < 0)) { + h->error = "get_right_neighbor failed"; + h->result = result; + *outcome = LOOKUP_DONE; + return 1; + } + if (result == 0) { + /* + * item to the right is also an extent one. Allocate a new node + * and insert pointer to it after item h -> coord. + * + * This is a result of extents being located at the twig + * level. For explanation, see comment just above + * is_next_item_internal(). + */ + znode *loaded; + + if (cbk_lock_mode(h->level, h) != ZNODE_WRITE_LOCK) { + /* + * we got node read locked, restart coord_by_key to + * have write lock on twig level + */ + h->lock_level = TWIG_LEVEL; + h->lock_mode = ZNODE_WRITE_LOCK; + *outcome = LOOKUP_REST; + return 1; + } + + loaded = coord->node; + result = + add_empty_leaf(coord, h->active_lh, h->key, + rd_key(coord, &key)); + if (result) { + h->error = "could not add empty leaf"; + h->result = result; + *outcome = LOOKUP_DONE; + return 1; + } + /* added empty leaf is locked (h->active_lh), its parent node + is unlocked, h->coord is set as EMPTY */ + assert("vs-13", coord->between == EMPTY_NODE); + assert("vs-14", znode_is_write_locked(coord->node)); + assert("vs-15", + WITH_DATA(coord->node, node_is_empty(coord->node))); + assert("vs-16", jnode_is_leaf(ZJNODE(coord->node))); + assert("vs-17", coord->node == h->active_lh->node); + *outcome = LOOKUP_DONE; + h->result = CBK_COORD_NOTFOUND; + return 1; + } else if (result == 1) { + /* + * this is special case mentioned in the comment on + * tree.h:cbk_flags. We have found internal item immediately on + * the right of extent, and we are going to insert new item + * there. Key of item we are going to insert is smaller than + * leftmost key in the node pointed to by said internal item + * (otherwise search wouldn't come to the extent in the first + * place). + * + * This is a result of extents being located at the twig + * level. For explanation, see comment just above + * is_next_item_internal(). + */ + h->flags &= ~CBK_TRUST_DK; + } else { + assert("vs-8", result == 2); + *outcome = LOOKUP_REST; + return 1; + } + assert("vs-362", WITH_DATA(coord->node, item_is_internal(coord))); + return 0; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/estimate.c linux-2.6.30/fs/reiser4/estimate.c --- linux-2.6.30.orig/fs/reiser4/estimate.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/estimate.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,129 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +#include "debug.h" +#include "dformat.h" +#include "tree.h" +#include "carry.h" +#include "inode.h" +#include "plugin/cluster.h" +#include "plugin/item/ctail.h" + +/* This returns how many nodes might get dirty and added nodes if @children + nodes are dirtied + + Amount of internals which will get dirty or get allocated we estimate as 5% + of the childs + 1 balancing. 1 balancing is 2 neighbours, 2 new blocks and + the current block on the leaf level, 2 neighbour nodes + the current (or 1 + neighbour and 1 new and the current) on twig level, 2 neighbour nodes on + upper levels and 1 for a new root. So 5 for leaf level, 3 for twig level, + 2 on upper + 1 for root. + + Do not calculate the current node of the lowest level here - this is overhead + only. + + children is almost always 1 here. Exception is flow insertion +*/ +static reiser4_block_nr +max_balance_overhead(reiser4_block_nr childen, tree_level tree_height) +{ + reiser4_block_nr ten_percent; + + ten_percent = ((103 * childen) >> 10); + + /* If we have too many balancings at the time, tree height can raise on + more then 1. Assume that if tree_height is 5, it can raise on 1 only. + */ + return ((tree_height < 5 ? 5 : tree_height) * 2 + (4 + ten_percent)); +} + +/* this returns maximal possible number of nodes which can be modified plus + number of new nodes which can be required to perform insertion of one item + into the tree */ +/* it is only called when tree height changes, or gets initialized */ +reiser4_block_nr calc_estimate_one_insert(tree_level height) +{ + return 1 + max_balance_overhead(1, height); +} + +reiser4_block_nr estimate_one_insert_item(reiser4_tree * tree) +{ + return tree->estimate_one_insert; +} + +/* this returns maximal possible number of nodes which can be modified plus + number of new nodes which can be required to perform insertion of one unit + into an item in the tree */ +reiser4_block_nr estimate_one_insert_into_item(reiser4_tree * tree) +{ + /* estimate insert into item just like item insertion */ + return tree->estimate_one_insert; +} + +reiser4_block_nr estimate_one_item_removal(reiser4_tree * tree) +{ + /* on item removal reiser4 does not try to pack nodes more complact, so, + only one node may be dirtied on leaf level */ + return tree->estimate_one_insert; +} + +/* on leaf level insert_flow may add CARRY_FLOW_NEW_NODES_LIMIT new nodes and + dirty 3 existing nodes (insert point and both its neighbors). + Max_balance_overhead should estimate number of blocks which may change/get + added on internal levels */ +reiser4_block_nr estimate_insert_flow(tree_level height) +{ + return 3 + CARRY_FLOW_NEW_NODES_LIMIT + max_balance_overhead(3 + + CARRY_FLOW_NEW_NODES_LIMIT, + height); +} + +/* returnes max number of nodes can be occupied by disk cluster */ +static reiser4_block_nr estimate_cluster(struct inode *inode, int unprepped) +{ + int per_cluster; + per_cluster = (unprepped ? 1 : cluster_nrpages(inode)); + return 3 + per_cluster + + max_balance_overhead(3 + per_cluster, + REISER4_MAX_ZTREE_HEIGHT); +} + +/* how many nodes might get dirty and added + during insertion of a disk cluster */ +reiser4_block_nr estimate_insert_cluster(struct inode *inode) +{ + return estimate_cluster(inode, 1); /* 24 */ +} + +/* how many nodes might get dirty and added + during update of a (prepped or unprepped) disk cluster */ +reiser4_block_nr estimate_update_cluster(struct inode *inode) +{ + return estimate_cluster(inode, 0); /* 44, for 64K-cluster */ +} + +/* How many nodes occupied by a disk cluster might get dirty. + Note that this estimation is not precise (i.e. disk cluster + can occupy more nodes). + Q: Why we don't use precise estimation? + A: 1.Because precise estimation is fairly bad: 65536 nodes + for 64K logical cluster, it means 256M of dead space on + a partition + 2.It is a very rare case when disk cluster occupies more + nodes then this estimation returns. +*/ +reiser4_block_nr estimate_dirty_cluster(struct inode *inode) +{ + return cluster_nrpages(inode) + 4; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/export_ops.c linux-2.6.30/fs/reiser4/export_ops.c --- linux-2.6.30.orig/fs/reiser4/export_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/export_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,328 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "inode.h" +#include "plugin/plugin.h" + +/* + * Supported file-handle types + */ +typedef enum { + FH_WITH_PARENT = 0x10, /* file handle with parent */ + FH_WITHOUT_PARENT = 0x11 /* file handle without parent */ +} reiser4_fhtype; + +#define NFSERROR (255) + +/* initialize place-holder for object */ +static void object_on_wire_init(reiser4_object_on_wire *o) +{ + o->plugin = NULL; +} + +/* finish with @o */ +static void object_on_wire_done(reiser4_object_on_wire *o) +{ + if (o->plugin != NULL) + o->plugin->wire.done(o); +} + +/* + * read serialized object identity from @addr and store information about + * object in @obj. This is dual to encode_inode(). + */ +static char *decode_inode(struct super_block *s, char *addr, + reiser4_object_on_wire * obj) +{ + file_plugin *fplug; + + /* identifier of object plugin is stored in the first two bytes, + * followed by... */ + fplug = file_plugin_by_disk_id(reiser4_get_tree(s), (d16 *) addr); + if (fplug != NULL) { + addr += sizeof(d16); + obj->plugin = fplug; + assert("nikita-3520", fplug->wire.read != NULL); + /* plugin specific encoding of object identity. */ + addr = fplug->wire.read(addr, obj); + } else + addr = ERR_PTR(RETERR(-EINVAL)); + return addr; +} + +static struct dentry *reiser4_get_dentry(struct super_block *super, + void *data); +/** + * reiser4_decode_fh: decode on-wire object - helper function + * for fh_to_dentry, fh_to_parent export operations; + * @super: super block; + * @addr: onwire object to be decoded; + * + * Returns dentry referring to the object being decoded. + */ +static struct dentry *reiser4_decode_fh(struct super_block * super, + char * addr) +{ + reiser4_object_on_wire object; + + object_on_wire_init(&object); + + addr = decode_inode(super, addr, &object); + if (!IS_ERR(addr)) { + struct dentry *d; + d = reiser4_get_dentry(super, &object); + if (d != NULL && !IS_ERR(d)) + /* FIXME check for -ENOMEM */ + reiser4_get_dentry_fsdata(d)->stateless = 1; + addr = (char *)d; + } + object_on_wire_done(&object); + return (void *)addr; +} + +static struct dentry *reiser4_fh_to_dentry(struct super_block *sb, + struct fid *fid, + int fh_len, int fh_type) +{ + reiser4_context *ctx; + struct dentry *d; + + assert("edward-1536", + fh_type == FH_WITH_PARENT || fh_type == FH_WITHOUT_PARENT); + + ctx = reiser4_init_context(sb); + if (IS_ERR(ctx)) + return (struct dentry *)ctx; + + d = reiser4_decode_fh(sb, (char *)fid->raw); + + reiser4_exit_context(ctx); + return d; +} + +static struct dentry *reiser4_fh_to_parent(struct super_block *sb, + struct fid *fid, + int fh_len, int fh_type) +{ + char * addr; + struct dentry * d; + reiser4_context *ctx; + file_plugin *fplug; + + if (fh_type == FH_WITHOUT_PARENT) + return NULL; + assert("edward-1537", fh_type == FH_WITH_PARENT); + + ctx = reiser4_init_context(sb); + if (IS_ERR(ctx)) + return (struct dentry *)ctx; + addr = (char *)fid->raw; + /* extract 2-bytes file plugin id */ + fplug = file_plugin_by_disk_id(reiser4_get_tree(sb), (d16 *)addr); + if (fplug == NULL) { + d = ERR_PTR(RETERR(-EINVAL)); + goto exit; + } + addr += sizeof(d16); + /* skip previously encoded object */ + addr = fplug->wire.read(addr, NULL /* skip */); + if (IS_ERR(addr)) { + d = (struct dentry *)addr; + goto exit; + } + /* @extract and decode parent object */ + d = reiser4_decode_fh(sb, addr); + exit: + reiser4_exit_context(ctx); + return d; +} + +/* + * Object serialization support. + * + * To support knfsd file system provides export_operations that are used to + * construct and interpret NFS file handles. As a generalization of this, + * reiser4 object plugins have serialization support: it provides methods to + * create on-wire representation of identity of reiser4 object, and + * re-create/locate object given its on-wire identity. + * + */ + +/* + * return number of bytes that on-wire representation of @inode's identity + * consumes. + */ +static int encode_inode_size(struct inode *inode) +{ + assert("nikita-3514", inode != NULL); + assert("nikita-3515", inode_file_plugin(inode) != NULL); + assert("nikita-3516", inode_file_plugin(inode)->wire.size != NULL); + + return inode_file_plugin(inode)->wire.size(inode) + sizeof(d16); +} + +/* + * store on-wire representation of @inode's identity at the area beginning at + * @start. + */ +static char *encode_inode(struct inode *inode, char *start) +{ + assert("nikita-3517", inode != NULL); + assert("nikita-3518", inode_file_plugin(inode) != NULL); + assert("nikita-3519", inode_file_plugin(inode)->wire.write != NULL); + + /* + * first, store two-byte identifier of object plugin, then + */ + save_plugin_id(file_plugin_to_plugin(inode_file_plugin(inode)), + (d16 *) start); + start += sizeof(d16); + /* + * call plugin to serialize object's identity + */ + return inode_file_plugin(inode)->wire.write(inode, start); +} + +/* this returns number of 32 bit long numbers encoded in @lenp. 255 is + * returned if file handle can not be stored */ +/** + * reiser4_encode_fh - encode_fh of export operations + * @dentry: + * @fh: + * @lenp: + * @need_parent: + * + */ +static int +reiser4_encode_fh(struct dentry *dentry, __u32 *fh, int *lenp, + int need_parent) +{ + struct inode *inode; + struct inode *parent; + char *addr; + int need; + int delta; + int result; + reiser4_context *ctx; + + /* + * knfsd asks as to serialize object in @dentry, and, optionally its + * parent (if need_parent != 0). + * + * encode_inode() and encode_inode_size() is used to build + * representation of object and its parent. All hard work is done by + * object plugins. + */ + inode = dentry->d_inode; + parent = dentry->d_parent->d_inode; + + addr = (char *)fh; + + need = encode_inode_size(inode); + if (need < 0) + return NFSERROR; + if (need_parent) { + delta = encode_inode_size(parent); + if (delta < 0) + return NFSERROR; + need += delta; + } + + ctx = reiser4_init_context(dentry->d_inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + if (need <= sizeof(__u32) * (*lenp)) { + addr = encode_inode(inode, addr); + if (need_parent) + addr = encode_inode(parent, addr); + + /* store in lenp number of 32bit words required for file + * handle. */ + *lenp = (need + sizeof(__u32) - 1) >> 2; + result = need_parent ? FH_WITH_PARENT : FH_WITHOUT_PARENT; + } else + /* no enough space in file handle */ + result = NFSERROR; + reiser4_exit_context(ctx); + return result; +} + +/** + * reiser4_get_dentry_parent - get_parent of export operations + * @child: + * + */ +static struct dentry *reiser4_get_dentry_parent(struct dentry *child) +{ + struct inode *dir; + dir_plugin *dplug; + struct dentry *result; + reiser4_context *ctx; + + assert("nikita-3527", child != NULL); + + dir = child->d_inode; + assert("nikita-3529", dir != NULL); + + ctx = reiser4_init_context(dir->i_sb); + if (IS_ERR(ctx)) + return (void *)ctx; + + dplug = inode_dir_plugin(dir); + assert("nikita-3531", ergo(dplug != NULL, dplug->get_parent != NULL)); + + if (unlikely(dplug == NULL)) { + reiser4_exit_context(ctx); + return ERR_PTR(RETERR(-ENOTDIR)); + } + result = dplug->get_parent(dir); + reiser4_exit_context(ctx); + return result; +} + +/** + * reiser4_get_dentry - get_dentry of export operations + * @super: + * @data: + * + * + */ +static struct dentry *reiser4_get_dentry(struct super_block *super, void *data) +{ + reiser4_object_on_wire *o; + + assert("nikita-3522", super != NULL); + assert("nikita-3523", data != NULL); + /* + * this is only supposed to be called by + * + * reiser4_decode_fh->find_exported_dentry + * + * so, reiser4_context should be here already. + */ + assert("nikita-3526", is_in_reiser4_context()); + + o = (reiser4_object_on_wire *)data; + assert("nikita-3524", o->plugin != NULL); + assert("nikita-3525", o->plugin->wire.get != NULL); + + return o->plugin->wire.get(super, o); +} + +struct export_operations reiser4_export_operations = { + .encode_fh = reiser4_encode_fh, + .fh_to_dentry = reiser4_fh_to_dentry, + .fh_to_parent = reiser4_fh_to_parent, + .get_parent = reiser4_get_dentry_parent, +}; + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/flush.c linux-2.6.30/fs/reiser4/flush.c --- linux-2.6.30.orig/fs/reiser4/flush.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/flush.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,3703 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* The design document for this file is at http://www.namesys.com/v4/v4.html. */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "plugin/plugin.h" +#include "plugin/object.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree_walk.h" +#include "carry.h" +#include "tree.h" +#include "vfs_ops.h" +#include "inode.h" +#include "page_cache.h" +#include "wander.h" +#include "super.h" +#include "entd.h" +#include "reiser4.h" +#include "flush.h" +#include "writeout.h" + +#include +#include /* for struct super_block */ +#include /* for struct page */ +#include /* for struct bio */ +#include +#include + +/* IMPLEMENTATION NOTES */ + +/* PARENT-FIRST: Some terminology: A parent-first traversal is a way of + assigning a total order to the nodes of the tree in which the parent is + placed before its children, which are ordered (recursively) in left-to-right + order. When we speak of a "parent-first preceder", it describes the node that + "came before in forward parent-first order". When we speak of a "parent-first + follower", it describes the node that "comes next in parent-first order" + (alternatively the node that "came before in reverse parent-first order"). + + The following pseudo-code prints the nodes of a tree in forward parent-first + order: + + void parent_first (node) + { + print_node (node); + if (node->level > leaf) { + for (i = 0; i < num_children; i += 1) { + parent_first (node->child[i]); + } + } + } +*/ + +/* JUST WHAT ARE WE TRYING TO OPTIMIZE, HERE? The idea is to optimize block + allocation so that a left-to-right scan of the tree's data (i.e., the leaves + in left-to-right order) can be accomplished with sequential reads, which + results in reading nodes in their parent-first order. This is a + read-optimization aspect of the flush algorithm, and there is also a + write-optimization aspect, which is that we wish to make large sequential + writes to the disk by allocating or reallocating blocks so that they can be + written in sequence. Sometimes the read-optimization and write-optimization + goals conflict with each other, as we discuss in more detail below. +*/ + +/* STATE BITS: The flush code revolves around the state of the jnodes it covers. + Here are the relevant jnode->state bits and their relevence to flush: + + JNODE_DIRTY: If a node is dirty, it must be flushed. But in order to be + written it must be allocated first. In order to be considered allocated, + the jnode must have exactly one of { JNODE_OVRWR, JNODE_RELOC } set. These + two bits are exclusive, and all dirtied jnodes eventually have one of these + bits set during each transaction. + + JNODE_CREATED: The node was freshly created in its transaction and has no + previous block address, so it is unconditionally assigned to be relocated, + although this is mainly for code-convenience. It is not being 'relocated' + from anything, but in almost every regard it is treated as part of the + relocate set. The JNODE_CREATED bit remains set even after JNODE_RELOC is + set, so the actual relocate can be distinguished from the + created-and-allocated set easily: relocate-set members (belonging to the + preserve-set) have (JNODE_RELOC) set and created-set members which have no + previous location to preserve have (JNODE_RELOC | JNODE_CREATED) set. + + JNODE_OVRWR: The node belongs to atom's overwrite set. The flush algorithm + made the decision to maintain the pre-existing location for this node and + it will be written to the wandered-log. + + JNODE_RELOC: The flush algorithm made the decision to relocate this block + (if it was not created, see note above). A block with JNODE_RELOC set is + eligible for early-flushing and may be submitted during flush_empty_queues. + When the JNODE_RELOC bit is set on a znode, the parent node's internal item + is modified and the znode is rehashed. + + JNODE_SQUEEZABLE: Before shifting everything left, the flush algorithm + scans the node and calls plugin->f.squeeze() method for its items. By this + technology we update disk clusters of cryptcompress objects. Also if + leftmost point that was found by flush scan has this flag (races with + write(), rare case) the flush algorythm makes the decision to pass it to + squalloc() in spite of its flushprepped status for squeezing, not for + repeated allocation. + + JNODE_FLUSH_QUEUED: This bit is set when a call to flush enters the jnode + into its flush queue. This means the jnode is not on any clean or dirty + list, instead it is moved to one of the flush queue (see flush_queue.h) + object private list. This prevents multiple concurrent flushes from + attempting to start flushing from the same node. + + (DEAD STATE BIT) JNODE_FLUSH_BUSY: This bit was set during the bottom-up + squeeze-and-allocate on a node while its children are actively being + squeezed and allocated. This flag was created to avoid submitting a write + request for a node while its children are still being allocated and + squeezed. Then flush queue was re-implemented to allow unlimited number of + nodes be queued. This flag support was commented out in source code because + we decided that there was no reason to submit queued nodes before + jnode_flush() finishes. However, current code calls fq_write() during a + slum traversal and may submit "busy nodes" to disk. Probably we can + re-enable the JNODE_FLUSH_BUSY bit support in future. + + With these state bits, we describe a test used frequently in the code below, + jnode_is_flushprepped()(and the spin-lock-taking jnode_check_flushprepped()). + The test for "flushprepped" returns true if any of the following are true: + + - The node is not dirty + - The node has JNODE_RELOC set + - The node has JNODE_OVRWR set + + If either the node is not dirty or it has already been processed by flush + (and assigned JNODE_OVRWR or JNODE_RELOC), then it is prepped. If + jnode_is_flushprepped() returns true then flush has work to do on that node. +*/ + +/* FLUSH_PREP_ONCE_PER_TRANSACTION: Within a single transaction a node is never + flushprepped twice (unless an explicit call to flush_unprep is made as + described in detail below). For example a node is dirtied, allocated, and + then early-flushed to disk and set clean. Before the transaction commits, the + page is dirtied again and, due to memory pressure, the node is flushed again. + The flush algorithm will not relocate the node to a new disk location, it + will simply write it to the same, previously relocated position again. +*/ + +/* THE BOTTOM-UP VS. TOP-DOWN ISSUE: This code implements a bottom-up algorithm + where we start at a leaf node and allocate in parent-first order by iterating + to the right. At each step of the iteration, we check for the right neighbor. + Before advancing to the right neighbor, we check if the current position and + the right neighbor share the same parent. If they do not share the same + parent, the parent is allocated before the right neighbor. + + This process goes recursively up the tree and squeeze nodes level by level as + long as the right neighbor and the current position have different parents, + then it allocates the right-neighbors-with-different-parents on the way back + down. This process is described in more detail in + flush_squalloc_changed_ancestor and the recursive function + squalloc_one_changed_ancestor. But the purpose here is not to discuss the + specifics of the bottom-up approach as it is to contrast the bottom-up and + top-down approaches. + + The top-down algorithm was implemented earlier (April-May 2002). In the + top-down approach, we find a starting point by scanning left along each level + past dirty nodes, then going up and repeating the process until the left node + and the parent node are clean. We then perform a parent-first traversal from + the starting point, which makes allocating in parent-first order trivial. + After one subtree has been allocated in this manner, we move to the right, + try moving upward, then repeat the parent-first traversal. + + Both approaches have problems that need to be addressed. Both are + approximately the same amount of code, but the bottom-up approach has + advantages in the order it acquires locks which, at the very least, make it + the better approach. At first glance each one makes the other one look + simpler, so it is important to remember a few of the problems with each one. + + Main problem with the top-down approach: When you encounter a clean child + during the parent-first traversal, what do you do? You would like to avoid + searching through a large tree of nodes just to find a few dirty leaves at + the bottom, and there is not an obvious solution. One of the advantages of + the top-down approach is that during the parent-first traversal you check + every child of a parent to see if it is dirty. In this way, the top-down + approach easily handles the main problem of the bottom-up approach: + unallocated children. + + The unallocated children problem is that before writing a node to disk we + must make sure that all of its children are allocated. Otherwise, the writing + the node means extra I/O because the node will have to be written again when + the child is finally allocated. + + WE HAVE NOT YET ELIMINATED THE UNALLOCATED CHILDREN PROBLEM. Except for bugs, + this should not cause any file system corruption, it only degrades I/O + performance because a node may be written when it is sure to be written at + least one more time in the same transaction when the remaining children are + allocated. What follows is a description of how we will solve the problem. +*/ + +/* HANDLING UNALLOCATED CHILDREN: During flush we may allocate a parent node, + then proceeding in parent first order, allocate some of its left-children, + then encounter a clean child in the middle of the parent. We do not allocate + the clean child, but there may remain unallocated (dirty) children to the + right of the clean child. If we were to stop flushing at this moment and + write everything to disk, the parent might still contain unallocated + children. + + We could try to allocate all the descendents of every node that we allocate, + but this is not necessary. Doing so could result in allocating the entire + tree: if the root node is allocated then every unallocated node would have to + be allocated before flushing. Actually, we do not have to write a node just + because we allocate it. It is possible to allocate but not write a node + during flush, when it still has unallocated children. However, this approach + is probably not optimal for the following reason. + + The flush algorithm is designed to allocate nodes in parent-first order in an + attempt to optimize reads that occur in the same order. Thus we are + read-optimizing for a left-to-right scan through all the leaves in the + system, and we are hoping to write-optimize at the same time because those + nodes will be written together in batch. What happens, however, if we assign + a block number to a node in its read-optimized order but then avoid writing + it because it has unallocated children? In that situation, we lose out on the + write-optimization aspect because a node will have to be written again to the + its location on the device, later, which likely means seeking back to that + location. + + So there are tradeoffs. We can choose either: + + A. Allocate all unallocated children to preserve both write-optimization and + read-optimization, but this is not always desirable because it may mean + having to allocate and flush very many nodes at once. + + B. Defer writing nodes with unallocated children, keep their read-optimized + locations, but sacrifice write-optimization because those nodes will be + written again. + + C. Defer writing nodes with unallocated children, but do not keep their + read-optimized locations. Instead, choose to write-optimize them later, when + they are written. To facilitate this, we "undo" the read-optimized allocation + that was given to the node so that later it can be write-optimized, thus + "unpreparing" the flush decision. This is a case where we disturb the + FLUSH_PREP_ONCE_PER_TRANSACTION rule described above. By a call to + flush_unprep() we will: if the node was wandered, unset the JNODE_OVRWR bit; + if the node was relocated, unset the JNODE_RELOC bit, non-deferred-deallocate + its block location, and set the JNODE_CREATED bit, effectively setting the + node back to an unallocated state. + + We will take the following approach in v4.0: for twig nodes we will always + finish allocating unallocated children (A). For nodes with (level > TWIG) + we will defer writing and choose write-optimization (C). + + To summarize, there are several parts to a solution that avoids the problem + with unallocated children: + + FIXME-ZAM: Still no one approach is implemented to eliminate the + "UNALLOCATED CHILDREN" problem because there was an experiment which was done + showed that we have 1-2 nodes with unallocated children for thousands of + written nodes. The experiment was simple like coping/deletion of linux kernel + sources. However the problem can arise in more complex tests. I think we have + jnode_io_hook to insert a check for unallocated children and see what kind of + problem we have. + + 1. When flush reaches a stopping point (e.g. a clean node) it should continue + calling squeeze-and-allocate on any remaining unallocated children. + FIXME: Difficulty to implement: should be simple -- amounts to adding a while + loop to jnode_flush, see comments in that function. + + 2. When flush reaches flush_empty_queue(), some of the (level > TWIG) nodes + may still have unallocated children. If the twig level has unallocated + children it is an assertion failure. If a higher-level node has unallocated + children, then it should be explicitly de-allocated by a call to + flush_unprep(). + FIXME: Difficulty to implement: should be simple. + + 3. (CPU-Optimization) Checking whether a node has unallocated children may + consume more CPU cycles than we would like, and it is possible (but medium + complexity) to optimize this somewhat in the case where large sub-trees are + flushed. The following observation helps: if both the left- and + right-neighbor of a node are processed by the flush algorithm then the node + itself is guaranteed to have all of its children allocated. However, the cost + of this check may not be so expensive after all: it is not needed for leaves + and flush can guarantee this property for twigs. That leaves only (level > + TWIG) nodes that have to be checked, so this optimization only helps if at + least three (level > TWIG) nodes are flushed in one pass, and the savings + will be very small unless there are many more (level > TWIG) nodes. But if + there are many (level > TWIG) nodes then the number of blocks being written + will be very large, so the savings may be insignificant. That said, the idea + is to maintain both the left and right edges of nodes that are processed in + flush. When flush_empty_queue() is called, a relatively simple test will + tell whether the (level > TWIG) node is on the edge. If it is on the edge, + the slow check is necessary, but if it is in the interior then it can be + assumed to have all of its children allocated. FIXME: medium complexity to + implement, but simple to verify given that we must have a slow check anyway. + + 4. (Optional) This part is optional, not for v4.0--flush should work + independently of whether this option is used or not. Called RAPID_SCAN, the + idea is to amend the left-scan operation to take unallocated children into + account. Normally, the left-scan operation goes left as long as adjacent + nodes are dirty up until some large maximum value (FLUSH_SCAN_MAXNODES) at + which point it stops and begins flushing. But scan-left may stop at a + position where there are unallocated children to the left with the same + parent. When RAPID_SCAN is enabled, the ordinary scan-left operation stops + after FLUSH_RELOCATE_THRESHOLD, which is much smaller than + FLUSH_SCAN_MAXNODES, then procedes with a rapid scan. The rapid scan skips + all the interior children of a node--if the leftmost child of a twig is + dirty, check its left neighbor (the rightmost child of the twig to the left). + If the left neighbor of the leftmost child is also dirty, then continue the + scan at the left twig and repeat. This option will cause flush to allocate + more twigs in a single pass, but it also has the potential to write many more + nodes than would otherwise be written without the RAPID_SCAN option. + RAPID_SCAN was partially implemented, code removed August 12, 2002 by JMACD. +*/ + +/* FLUSH CALLED ON NON-LEAF LEVEL. Most of our design considerations assume that + the starting point for flush is a leaf node, but actually the flush code + cares very little about whether or not this is true. It is possible that all + the leaf nodes are flushed and dirty parent nodes still remain, in which case + jnode_flush() is called on a non-leaf argument. Flush doesn't care--it treats + the argument node as if it were a leaf, even when it is not. This is a simple + approach, and there may be a more optimal policy but until a problem with + this approach is discovered, simplest is probably best. + + NOTE: In this case, the ordering produced by flush is parent-first only if + you ignore the leaves. This is done as a matter of simplicity and there is + only one (shaky) justification. When an atom commits, it flushes all leaf + level nodes first, followed by twigs, and so on. With flushing done in this + order, if flush is eventually called on a non-leaf node it means that + (somehow) we reached a point where all leaves are clean and only internal + nodes need to be flushed. If that it the case, then it means there were no + leaves that were the parent-first preceder/follower of the parent. This is + expected to be a rare case, which is why we do nothing special about it. + However, memory pressure may pass an internal node to flush when there are + still dirty leaf nodes that need to be flushed, which could prove our + original assumptions "inoperative". If this needs to be fixed, then + scan_left/right should have special checks for the non-leaf levels. For + example, instead of passing from a node to the left neighbor, it should pass + from the node to the left neighbor's rightmost descendent (if dirty). + +*/ + +/* UNIMPLEMENTED AS YET: REPACKING AND RESIZING. We walk the tree in 4MB-16MB + chunks, dirtying everything and putting it into a transaction. We tell the + allocator to allocate the blocks as far as possible towards one end of the + logical device--the left (starting) end of the device if we are walking from + left to right, the right end of the device if we are walking from right to + left. We then make passes in alternating directions, and as we do this the + device becomes sorted such that tree order and block number order fully + correlate. + + Resizing is done by shifting everything either all the way to the left or all + the way to the right, and then reporting the last block. +*/ + +/* RELOCATE DECISIONS: The code makes a decision to relocate in several places. + This descibes the policy from the highest level: + + The FLUSH_RELOCATE_THRESHOLD parameter: If we count this many consecutive + nodes on the leaf level during flush-scan (right, left), then we + unconditionally decide to relocate leaf nodes. + + Otherwise, there are two contexts in which we make a decision to relocate: + + 1. The REVERSE PARENT-FIRST context: Implemented in reverse_relocate_test(). + During the initial stages of flush, after scan-right completes, we want to + ask the question: should we relocate this leaf node and thus dirty the parent + node. Then if the node is a leftmost child its parent is its own parent-first + preceder, thus we repeat the question at the next level up, and so on. In + these cases we are moving in the reverse-parent first direction. + + There is another case which is considered the reverse direction, which comes + at the end of a twig in reverse_relocate_end_of_twig(). As we finish + processing a twig we may reach a point where there is a clean twig to the + right with a dirty leftmost child. In this case, we may wish to relocate the + child by testing if it should be relocated relative to its parent. + + 2. The FORWARD PARENT-FIRST context: Testing for forward relocation is done + in allocate_znode. What distinguishes the forward parent-first case from the + reverse-parent first case is that the preceder has already been allocated in + the forward case, whereas in the reverse case we don't know what the preceder + is until we finish "going in reverse". That simplifies the forward case + considerably, and there we actually use the block allocator to determine + whether, e.g., a block closer to the preceder is available. +*/ + +/* SQUEEZE_LEFT_EDGE: Unimplemented idea for future consideration. The idea is, + once we finish scan-left and find a starting point, if the parent's left + neighbor is dirty then squeeze the parent's left neighbor and the parent. + This may change the flush-starting-node's parent. Repeat until the child's + parent is stable. If the child is a leftmost child, repeat this left-edge + squeezing operation at the next level up. Note that we cannot allocate + extents during this or they will be out of parent-first order. There is also + some difficult coordinate maintenence issues. We can't do a tree search to + find coordinates again (because we hold locks), we have to determine them + from the two nodes being squeezed. Looks difficult, but has potential to + increase space utilization. */ + +/* Flush-scan helper functions. */ +static void scan_init(flush_scan * scan); +static void scan_done(flush_scan * scan); + +/* Flush-scan algorithm. */ +static int scan_left(flush_scan * scan, flush_scan * right, jnode * node, + unsigned limit); +static int scan_right(flush_scan * scan, jnode * node, unsigned limit); +static int scan_common(flush_scan * scan, flush_scan * other); +static int scan_formatted(flush_scan * scan); +static int scan_unformatted(flush_scan * scan, flush_scan * other); +static int scan_by_coord(flush_scan * scan); + +/* Initial flush-point ancestor allocation. */ +static int alloc_pos_and_ancestors(flush_pos_t *pos); +static int alloc_one_ancestor(const coord_t *coord, flush_pos_t *pos); +static int set_preceder(const coord_t *coord_in, flush_pos_t *pos); + +/* Main flush algorithm. + Note on abbreviation: "squeeze and allocate" == "squalloc". */ +static int squalloc(flush_pos_t *pos); + +/* Flush squeeze implementation. */ +static int squeeze_right_non_twig(znode * left, znode * right); +static int shift_one_internal_unit(znode * left, znode * right); + +/* Flush reverse parent-first relocation routines. */ +static int reverse_relocate_if_close_enough(const reiser4_block_nr * pblk, + const reiser4_block_nr * nblk); +static int reverse_relocate_test(jnode * node, const coord_t *parent_coord, + flush_pos_t *pos); +static int reverse_relocate_check_dirty_parent(jnode * node, + const coord_t *parent_coord, + flush_pos_t *pos); + +/* Flush allocate write-queueing functions: */ +static int allocate_znode(znode * node, const coord_t *parent_coord, + flush_pos_t *pos); +static int allocate_znode_update(znode * node, const coord_t *parent_coord, + flush_pos_t *pos); +static int lock_parent_and_allocate_znode(znode *, flush_pos_t *); + +/* Flush helper functions: */ +static int jnode_lock_parent_coord(jnode * node, + coord_t *coord, + lock_handle * parent_lh, + load_count * parent_zh, + znode_lock_mode mode, int try); +static int neighbor_in_slum(znode * node, lock_handle * right_lock, sideof side, + znode_lock_mode mode, int check_dirty, int expected); +static int znode_same_parents(znode * a, znode * b); + +static int znode_check_flushprepped(znode * node) +{ + return jnode_check_flushprepped(ZJNODE(node)); +} + +/* Flush position functions */ +static void pos_init(flush_pos_t *pos); +static int pos_valid(flush_pos_t *pos); +static void pos_done(flush_pos_t *pos); +static int pos_stop(flush_pos_t *pos); + +/* check that @org is first jnode extent unit, if extent is unallocated, + * because all jnodes of unallocated extent are dirty and of the same atom. */ +#define checkchild(scan) \ +assert("nikita-3435", \ + ergo(scan->direction == LEFT_SIDE && \ + (scan->parent_coord.node->level == TWIG_LEVEL) && \ + jnode_is_unformatted(scan->node) && \ + extent_is_unallocated(&scan->parent_coord), \ + extent_unit_index(&scan->parent_coord) == index_jnode(scan->node))) + +/* This flush_cnt variable is used to track the number of concurrent flush + operations, useful for debugging. It is initialized in txnmgr.c out of + laziness (because flush has no static initializer function...) */ +ON_DEBUG(atomic_t flush_cnt; + ) + +/* check fs backing device for write congestion */ +static int check_write_congestion(void) +{ + struct super_block *sb; + struct backing_dev_info *bdi; + + sb = reiser4_get_current_sb(); + bdi = reiser4_get_super_fake(sb)->i_mapping->backing_dev_info; + return bdi_write_congested(bdi); +} + +/* conditionally write flush queue */ +static int write_prepped_nodes(flush_pos_t *pos) +{ + int ret; + + assert("zam-831", pos); + assert("zam-832", pos->fq); + + if (!(pos->flags & JNODE_FLUSH_WRITE_BLOCKS)) + return 0; + + if (check_write_congestion()) + return 0; + + ret = reiser4_write_fq(pos->fq, pos->nr_written, + WRITEOUT_SINGLE_STREAM | WRITEOUT_FOR_PAGE_RECLAIM); + return ret; +} + +/* Proper release all flush pos. resources then move flush position to new + locked node */ +static void move_flush_pos(flush_pos_t *pos, lock_handle * new_lock, + load_count * new_load, const coord_t *new_coord) +{ + assert("zam-857", new_lock->node == new_load->node); + + if (new_coord) { + assert("zam-858", new_coord->node == new_lock->node); + coord_dup(&pos->coord, new_coord); + } else { + coord_init_first_unit(&pos->coord, new_lock->node); + } + + if (pos->child) { + jput(pos->child); + pos->child = NULL; + } + + move_load_count(&pos->load, new_load); + done_lh(&pos->lock); + move_lh(&pos->lock, new_lock); +} + +/* delete empty node which link from the parent still exists. */ +static int delete_empty_node(znode * node) +{ + reiser4_key smallest_removed; + + assert("zam-1019", node != NULL); + assert("zam-1020", node_is_empty(node)); + assert("zam-1023", znode_is_wlocked(node)); + + return reiser4_delete_node(node, &smallest_removed, NULL, 1); +} + +/* Prepare flush position for alloc_pos_and_ancestors() and squalloc() */ +static int prepare_flush_pos(flush_pos_t *pos, jnode * org) +{ + int ret; + load_count load; + lock_handle lock; + + init_lh(&lock); + init_load_count(&load); + + if (jnode_is_znode(org)) { + ret = longterm_lock_znode(&lock, JZNODE(org), + ZNODE_WRITE_LOCK, ZNODE_LOCK_HIPRI); + if (ret) + return ret; + + ret = incr_load_count_znode(&load, JZNODE(org)); + if (ret) + return ret; + + pos->state = + (jnode_get_level(org) == + LEAF_LEVEL) ? POS_ON_LEAF : POS_ON_INTERNAL; + move_flush_pos(pos, &lock, &load, NULL); + } else { + coord_t parent_coord; + ret = jnode_lock_parent_coord(org, &parent_coord, &lock, + &load, ZNODE_WRITE_LOCK, 0); + if (ret) + goto done; + if (!item_is_extent(&parent_coord)) { + /* file was converted to tail, org became HB, we found + internal item */ + ret = -EAGAIN; + goto done; + } + + pos->state = POS_ON_EPOINT; + move_flush_pos(pos, &lock, &load, &parent_coord); + pos->child = jref(org); + if (extent_is_unallocated(&parent_coord) + && extent_unit_index(&parent_coord) != index_jnode(org)) { + /* @org is not first child of its parent unit. This may + happen because longerm lock of its parent node was + released between scan_left and scan_right. For now + work around this having flush to repeat */ + ret = -EAGAIN; + } + } + +done: + done_load_count(&load); + done_lh(&lock); + return ret; +} + +/* TODO LIST (no particular order): */ +/* I have labelled most of the legitimate FIXME comments in this file with + letters to indicate which issue they relate to. There are a few miscellaneous + FIXMEs with specific names mentioned instead that need to be + inspected/resolved. */ +/* B. There is an issue described in reverse_relocate_test having to do with an + imprecise is_preceder? check having to do with partially-dirty extents. The + code that sets preceder hints and computes the preceder is basically + untested. Careful testing needs to be done that preceder calculations are + done correctly, since if it doesn't affect correctness we will not catch this + stuff during regular testing. */ +/* C. EINVAL, E_DEADLOCK, E_NO_NEIGHBOR, ENOENT handling. It is unclear which of + these are considered expected but unlikely conditions. Flush currently + returns 0 (i.e., success but no progress, i.e., restart) whenever it receives + any of these in jnode_flush(). Many of the calls that may produce one of + these return values (i.e., longterm_lock_znode, reiser4_get_parent, + reiser4_get_neighbor, ...) check some of these values themselves and, for + instance, stop flushing instead of resulting in a restart. If any of these + results are true error conditions then flush will go into a busy-loop, as we + noticed during testing when a corrupt tree caused find_child_ptr to return + ENOENT. It needs careful thought and testing of corner conditions. +*/ +/* D. Atomicity of flush_prep against deletion and flush concurrency. Suppose a + created block is assigned a block number then early-flushed to disk. It is + dirtied again and flush is called again. Concurrently, that block is deleted, + and the de-allocation of its block number does not need to be deferred, since + it is not part of the preserve set (i.e., it didn't exist before the + transaction). I think there may be a race condition where flush writes the + dirty, created block after the non-deferred deallocated block number is + re-allocated, making it possible to write deleted data on top of non-deleted + data. Its just a theory, but it needs to be thought out. */ +/* F. bio_alloc() failure is not handled gracefully. */ +/* G. Unallocated children. */ +/* H. Add a WANDERED_LIST to the atom to clarify the placement of wandered + blocks. */ +/* I. Rename flush-scan to scan-point, (flush-pos to flush-point?) */ + +/* JNODE_FLUSH: MAIN ENTRY POINT */ +/* This is the main entry point for flushing a jnode and its dirty neighborhood + (dirty neighborhood is named "slum"). Jnode_flush() is called if reiser4 has + to write dirty blocks to disk, it happens when Linux VM decides to reduce + number of dirty pages or as a part of transaction commit. + + Our objective here is to prep and flush the slum the jnode belongs to. We + want to squish the slum together, and allocate the nodes in it as we squish + because allocation of children affects squishing of parents. + + The "argument" @node tells flush where to start. From there, flush finds the + left edge of the slum, and calls squalloc (in which nodes are squeezed and + allocated). To find a "better place" to start squalloc first we perform a + flush_scan. + + Flush-scanning may be performed in both left and right directions, but for + different purposes. When scanning to the left, we are searching for a node + that precedes a sequence of parent-first-ordered nodes which we will then + flush in parent-first order. During flush-scanning, we also take the + opportunity to count the number of consecutive leaf nodes. If this number is + past some threshold (FLUSH_RELOCATE_THRESHOLD), then we make a decision to + reallocate leaf nodes (thus favoring write-optimization). + + Since the flush argument node can be anywhere in a sequence of dirty leaves, + there may also be dirty nodes to the right of the argument. If the scan-left + operation does not count at least FLUSH_RELOCATE_THRESHOLD nodes then we + follow it with a right-scan operation to see whether there is, in fact, + enough nodes to meet the relocate threshold. Each right- and left-scan + operation uses a single flush_scan object. + + After left-scan and possibly right-scan, we prepare a flush_position object + with the starting flush point or parent coordinate, which was determined + using scan-left. + + Next we call the main flush routine, squalloc, which iterates along the leaf + level, squeezing and allocating nodes (and placing them into the flush + queue). + + After squalloc returns we take extra steps to ensure that all the children + of the final twig node are allocated--this involves repeating squalloc + until we finish at a twig with no unallocated children. + + Finally, we call flush_empty_queue to submit write-requests to disk. If we + encounter any above-twig nodes during flush_empty_queue that still have + unallocated children, we flush_unprep them. + + Flush treats several "failure" cases as non-failures, essentially causing + them to start over. E_DEADLOCK is one example. + FIXME:(C) EINVAL, E_NO_NEIGHBOR, ENOENT: these should probably be handled + properly rather than restarting, but there are a bunch of cases to audit. +*/ + +static int +jnode_flush(jnode * node, long nr_to_write, long *nr_written, + flush_queue_t *fq, int flags) +{ + long ret = 0; + flush_scan *right_scan; + flush_scan *left_scan; + flush_pos_t *flush_pos; + int todo; + struct super_block *sb; + reiser4_super_info_data *sbinfo; + jnode *leftmost_in_slum = NULL; + + assert("jmacd-76619", lock_stack_isclean(get_current_lock_stack())); + assert("nikita-3022", reiser4_schedulable()); + + assert("nikita-3185", + get_current_super_private()->delete_mutex_owner != current); + + /* allocate right_scan, left_scan and flush_pos */ + right_scan = + kmalloc(2 * sizeof(*right_scan) + sizeof(*flush_pos), + reiser4_ctx_gfp_mask_get()); + if (right_scan == NULL) + return RETERR(-ENOMEM); + left_scan = right_scan + 1; + flush_pos = (flush_pos_t *) (left_scan + 1); + + sb = reiser4_get_current_sb(); + sbinfo = get_super_private(sb); + + /* Flush-concurrency debug code */ +#if REISER4_DEBUG + atomic_inc(&flush_cnt); +#endif + + reiser4_enter_flush(sb); + + /* Initialize a flush position. */ + pos_init(flush_pos); + + flush_pos->nr_written = nr_written; + flush_pos->fq = fq; + flush_pos->flags = flags; + flush_pos->nr_to_write = nr_to_write; + + scan_init(right_scan); + scan_init(left_scan); + + /* First scan left and remember the leftmost scan position. If the + leftmost position is unformatted we remember its parent_coord. We + scan until counting FLUSH_SCAN_MAXNODES. + + If starting @node is unformatted, at the beginning of left scan its + parent (twig level node, containing extent item) will be long term + locked and lock handle will be stored in the + @right_scan->parent_lock. This lock is used to start the rightward + scan without redoing the tree traversal (necessary to find parent) + and, hence, is kept during leftward scan. As a result, we have to + use try-lock when taking long term locks during the leftward scan. + */ + ret = scan_left(left_scan, right_scan, + node, sbinfo->flush.scan_maxnodes); + if (ret != 0) + goto failed; + + leftmost_in_slum = jref(left_scan->node); + scan_done(left_scan); + + /* Then possibly go right to decide if we will use a policy of + relocating leaves. This is only done if we did not scan past (and + count) enough nodes during the leftward scan. If we do scan right, + we only care to go far enough to establish that at least + FLUSH_RELOCATE_THRESHOLD number of nodes are being flushed. The scan + limit is the difference between left_scan.count and the threshold. */ + + todo = sbinfo->flush.relocate_threshold - left_scan->count; + /* scan right is inherently deadlock prone, because we are + * (potentially) holding a lock on the twig node at this moment. + * FIXME: this is incorrect comment: lock is not held */ + if (todo > 0) { + ret = scan_right(right_scan, node, (unsigned)todo); + if (ret != 0) + goto failed; + } + + /* Only the right-scan count is needed, release any rightward locks + right away. */ + scan_done(right_scan); + + /* ... and the answer is: we should relocate leaf nodes if at least + FLUSH_RELOCATE_THRESHOLD nodes were found. */ + flush_pos->leaf_relocate = JF_ISSET(node, JNODE_REPACK) || + (left_scan->count + right_scan->count >= + sbinfo->flush.relocate_threshold); + + /* Funny business here. We set the 'point' in the flush_position at + prior to starting squalloc regardless of whether the first point is + formatted or unformatted. Without this there would be an invariant, + in the rest of the code, that if the flush_position is unformatted + then flush_position->point is NULL and + flush_position->parent_{lock,coord} is set, and if the flush_position + is formatted then flush_position->point is non-NULL and no parent + info is set. + + This seems lazy, but it makes the initial calls to + reverse_relocate_test (which ask "is it the pos->point the leftmost + child of its parent") much easier because we know the first child + already. Nothing is broken by this, but the reasoning is subtle. + Holding an extra reference on a jnode during flush can cause us to + see nodes with HEARD_BANSHEE during squalloc, because nodes are not + removed from sibling lists until they have zero reference count. + Flush would never observe a HEARD_BANSHEE node on the left-edge of + flush, nodes are only deleted to the right. So if nothing is broken, + why fix it? + + NOTE-NIKITA actually, flush can meet HEARD_BANSHEE node at any + point and in any moment, because of the concurrent file system + activity (for example, truncate). */ + + /* Check jnode state after flush_scan completed. Having a lock on this + node or its parent (in case of unformatted) helps us in case of + concurrent flushing. */ + if (jnode_check_flushprepped(leftmost_in_slum) + && !jnode_convertible(leftmost_in_slum)) { + ret = 0; + goto failed; + } + + /* Now setup flush_pos using scan_left's endpoint. */ + ret = prepare_flush_pos(flush_pos, leftmost_in_slum); + if (ret) + goto failed; + + if (znode_get_level(flush_pos->coord.node) == LEAF_LEVEL + && node_is_empty(flush_pos->coord.node)) { + znode *empty = flush_pos->coord.node; + + assert("zam-1022", !ZF_ISSET(empty, JNODE_HEARD_BANSHEE)); + ret = delete_empty_node(empty); + goto failed; + } + + if (jnode_check_flushprepped(leftmost_in_slum) + && !jnode_convertible(leftmost_in_slum)) { + ret = 0; + goto failed; + } + + /* Set pos->preceder and (re)allocate pos and its ancestors if it is + needed */ + ret = alloc_pos_and_ancestors(flush_pos); + if (ret) + goto failed; + + /* Do the main rightward-bottom-up squeeze and allocate loop. */ + ret = squalloc(flush_pos); + pos_stop(flush_pos); + if (ret) + goto failed; + + /* FIXME_NFQUCMPD: Here, handle the twig-special case for unallocated + children. First, the pos_stop() and pos_valid() routines should be + modified so that pos_stop() sets a flush_position->stop flag to 1 + without releasing the current position immediately--instead release + it in pos_done(). This is a better implementation than the current + one anyway. + + It is not clear that all fields of the flush_position should not be + released, but at the very least the parent_lock, parent_coord, and + parent_load should remain held because they are hold the last twig + when pos_stop() is called. + + When we reach this point in the code, if the parent_coord is set to + after the last item then we know that flush reached the end of a twig + (and according to the new flush queueing design, we will return now). + If parent_coord is not past the last item, we should check if the + current twig has any unallocated children to the right (we are not + concerned with unallocated children to the left--in that case the + twig itself should not have been allocated). If the twig has + unallocated children to the right, set the parent_coord to that + position and then repeat the call to squalloc. + + Testing for unallocated children may be defined in two ways: if any + internal item has a fake block number, it is unallocated; if any + extent item is unallocated then all of its children are unallocated. + But there is a more aggressive approach: if there are any dirty + children of the twig to the right of the current position, we may + wish to relocate those nodes now. Checking for potential relocation + is more expensive as it requires knowing whether there are any dirty + children that are not unallocated. The extent_needs_allocation should + be used after setting the correct preceder. + + When we reach the end of a twig at this point in the code, if the + flush can continue (when the queue is ready) it will need some + information on the future starting point. That should be stored away + in the flush_handle using a seal, I believe. Holding a jref() on the + future starting point may break other code that deletes that node. + */ + + /* FIXME_NFQUCMPD: Also, we don't want to do any flushing when flush is + called above the twig level. If the VM calls flush above the twig + level, do nothing and return (but figure out why this happens). The + txnmgr should be modified to only flush its leaf-level dirty list. + This will do all the necessary squeeze and allocate steps but leave + unallocated branches and possibly unallocated twigs (when the twig's + leftmost child is not dirty). After flushing the leaf level, the + remaining unallocated nodes should be given write-optimized + locations. (Possibly, the remaining unallocated twigs should be + allocated just before their leftmost child.) + */ + + /* Any failure reaches this point. */ +failed: + + switch (ret) { + case -E_REPEAT: + case -EINVAL: + case -E_DEADLOCK: + case -E_NO_NEIGHBOR: + case -ENOENT: + /* FIXME(C): Except for E_DEADLOCK, these should probably be + handled properly in each case. They already are handled in + many cases. */ + /* Something bad happened, but difficult to avoid... Try again! + */ + ret = 0; + } + + if (leftmost_in_slum) + jput(leftmost_in_slum); + + pos_done(flush_pos); + scan_done(left_scan); + scan_done(right_scan); + kfree(right_scan); + + ON_DEBUG(atomic_dec(&flush_cnt)); + + reiser4_leave_flush(sb); + + return ret; +} + +/* The reiser4 flush subsystem can be turned into "rapid flush mode" means that + * flusher should submit all prepped nodes immediately without keeping them in + * flush queues for long time. The reason for rapid flush mode is to free + * memory as fast as possible. */ + +#if REISER4_USE_RAPID_FLUSH + +/** + * submit all prepped nodes if rapid flush mode is set, + * turn rapid flush mode off. + */ + +static int rapid_flush(flush_pos_t *pos) +{ + if (!wbq_available()) + return 0; + + return write_prepped_nodes(pos); +} + +#else + +#define rapid_flush(pos) (0) + +#endif /* REISER4_USE_RAPID_FLUSH */ + +static jnode *find_flush_start_jnode(jnode *start, txn_atom * atom, + flush_queue_t *fq, int *nr_queued, + int flags) +{ + jnode * node; + + if (start != NULL) { + spin_lock_jnode(start); + if (!jnode_is_flushprepped(start)) { + assert("zam-1056", start->atom == atom); + node = start; + goto enter; + } + spin_unlock_jnode(start); + } + /* + * In this loop we process all already prepped (RELOC or OVRWR) and + * dirtied again nodes. The atom spin lock is not released until all + * dirty nodes processed or not prepped node found in the atom dirty + * lists. + */ + while ((node = find_first_dirty_jnode(atom, flags))) { + spin_lock_jnode(node); +enter: + assert("zam-881", JF_ISSET(node, JNODE_DIRTY)); + assert("zam-898", !JF_ISSET(node, JNODE_OVRWR)); + + if (JF_ISSET(node, JNODE_WRITEBACK)) { + /* move node to the end of atom's writeback list */ + list_move_tail(&node->capture_link, ATOM_WB_LIST(atom)); + + /* + * jnode is not necessarily on dirty list: if it was + * dirtied when it was on flush queue - it does not get + * moved to dirty list + */ + ON_DEBUG(count_jnode(atom, node, NODE_LIST(node), + WB_LIST, 1)); + + } else if (jnode_is_znode(node) + && znode_above_root(JZNODE(node))) { + /* + * A special case for znode-above-root. The above-root + * (fake) znode is captured and dirtied when the tree + * height changes or when the root node is relocated. + * This causes atoms to fuse so that changes at the root + * are serialized. However, this node is never flushed. + * This special case used to be in lock.c to prevent the + * above-root node from ever being captured, but now + * that it is captured we simply prevent it from + * flushing. The log-writer code relies on this to + * properly log superblock modifications of the tree + * height. + */ + jnode_make_wander_nolock(node); + } else if (JF_ISSET(node, JNODE_RELOC)) { + queue_jnode(fq, node); + ++(*nr_queued); + } else + break; + + spin_unlock_jnode(node); + } + return node; +} + +/* Flush some nodes of current atom, usually slum, return -E_REPEAT if there are + * more nodes to flush, return 0 if atom's dirty lists empty and keep current + * atom locked, return other errors as they are. */ +int +flush_current_atom(int flags, long nr_to_write, long *nr_submitted, + txn_atom ** atom, jnode *start) +{ + reiser4_super_info_data *sinfo = get_current_super_private(); + flush_queue_t *fq = NULL; + jnode *node; + int nr_queued; + int ret; + + assert("zam-889", atom != NULL && *atom != NULL); + assert_spin_locked(&((*atom)->alock)); + assert("zam-892", get_current_context()->trans->atom == *atom); + + nr_to_write = LONG_MAX; + while (1) { + ret = reiser4_fq_by_atom(*atom, &fq); + if (ret != -E_REPEAT) + break; + *atom = get_current_atom_locked(); + } + if (ret) + return ret; + + assert_spin_locked(&((*atom)->alock)); + + /* parallel flushers limit */ + if (sinfo->tmgr.atom_max_flushers != 0) { + while ((*atom)->nr_flushers >= sinfo->tmgr.atom_max_flushers) { + /* An reiser4_atom_send_event() call is inside + reiser4_fq_put_nolock() which is called when flush is + finished and nr_flushers is decremented. */ + reiser4_atom_wait_event(*atom); + *atom = get_current_atom_locked(); + } + } + + /* count ourself as a flusher */ + (*atom)->nr_flushers++; + + writeout_mode_enable(); + + nr_queued = 0; + node = find_flush_start_jnode(start, *atom, fq, &nr_queued, flags); + + if (node == NULL) { + if (nr_queued == 0) { + (*atom)->nr_flushers--; + reiser4_fq_put_nolock(fq); + reiser4_atom_send_event(*atom); + /* current atom remains locked */ + writeout_mode_disable(); + return 0; + } + spin_unlock_atom(*atom); + } else { + jref(node); + BUG_ON((*atom)->super != node->tree->super); + spin_unlock_atom(*atom); + spin_unlock_jnode(node); + BUG_ON(nr_to_write == 0); + ret = jnode_flush(node, nr_to_write, nr_submitted, fq, flags); + jput(node); + } + + ret = + reiser4_write_fq(fq, nr_submitted, + WRITEOUT_SINGLE_STREAM | WRITEOUT_FOR_PAGE_RECLAIM); + + *atom = get_current_atom_locked(); + (*atom)->nr_flushers--; + reiser4_fq_put_nolock(fq); + reiser4_atom_send_event(*atom); + spin_unlock_atom(*atom); + + writeout_mode_disable(); + + if (ret == 0) + ret = -E_REPEAT; + + return ret; +} + +/* REVERSE PARENT-FIRST RELOCATION POLICIES */ + +/* This implements the is-it-close-enough-to-its-preceder? test for relocation + in the reverse parent-first relocate context. Here all we know is the + preceder and the block number. Since we are going in reverse, the preceder + may still be relocated as well, so we can't ask the block allocator "is there + a closer block available to relocate?" here. In the _forward_ parent-first + relocate context (not here) we actually call the block allocator to try and + find a closer location. */ +static int +reverse_relocate_if_close_enough(const reiser4_block_nr * pblk, + const reiser4_block_nr * nblk) +{ + reiser4_block_nr dist; + + assert("jmacd-7710", *pblk != 0 && *nblk != 0); + assert("jmacd-7711", !reiser4_blocknr_is_fake(pblk)); + assert("jmacd-7712", !reiser4_blocknr_is_fake(nblk)); + + /* Distance is the absolute value. */ + dist = (*pblk > *nblk) ? (*pblk - *nblk) : (*nblk - *pblk); + + /* If the block is less than FLUSH_RELOCATE_DISTANCE blocks away from + its preceder block, do not relocate. */ + if (dist <= get_current_super_private()->flush.relocate_distance) + return 0; + + return 1; +} + +/* This function is a predicate that tests for relocation. Always called in the + reverse-parent-first context, when we are asking whether the current node + should be relocated in order to expand the flush by dirtying the parent level + (and thus proceeding to flush that level). When traversing in the forward + parent-first direction (not here), relocation decisions are handled in two + places: allocate_znode() and extent_needs_allocation(). */ +static int +reverse_relocate_test(jnode * node, const coord_t *parent_coord, + flush_pos_t *pos) +{ + reiser4_block_nr pblk = 0; + reiser4_block_nr nblk = 0; + + assert("jmacd-8989", !jnode_is_root(node)); + + /* + * This function is called only from the + * reverse_relocate_check_dirty_parent() and only if the parent + * node is clean. This implies that the parent has the real (i.e., not + * fake) block number, and, so does the child, because otherwise the + * parent would be dirty. + */ + + /* New nodes are treated as if they are being relocated. */ + if (JF_ISSET(node, JNODE_CREATED) || + (pos->leaf_relocate && jnode_get_level(node) == LEAF_LEVEL)) + return 1; + + /* Find the preceder. FIXME(B): When the child is an unformatted, + previously existing node, the coord may be leftmost even though the + child is not the parent-first preceder of the parent. If the first + dirty node appears somewhere in the middle of the first extent unit, + this preceder calculation is wrong. + Needs more logic in here. */ + if (coord_is_leftmost_unit(parent_coord)) { + pblk = *znode_get_block(parent_coord->node); + } else { + pblk = pos->preceder.blk; + } + check_preceder(pblk); + + /* If (pblk == 0) then the preceder isn't allocated or isn't known: + relocate. */ + if (pblk == 0) + return 1; + + nblk = *jnode_get_block(node); + + if (reiser4_blocknr_is_fake(&nblk)) + /* child is unallocated, mark parent dirty */ + return 1; + + return reverse_relocate_if_close_enough(&pblk, &nblk); +} + +/* This function calls reverse_relocate_test to make a reverse-parent-first + relocation decision and then, if yes, it marks the parent dirty. */ +static int +reverse_relocate_check_dirty_parent(jnode * node, const coord_t *parent_coord, + flush_pos_t *pos) +{ + int ret; + + if (!JF_ISSET(ZJNODE(parent_coord->node), JNODE_DIRTY)) { + + ret = reverse_relocate_test(node, parent_coord, pos); + if (ret < 0) + return ret; + + /* FIXME-ZAM + if parent is already relocated - we do not want to grab space, + right? */ + if (ret == 1) { + int grabbed; + + grabbed = get_current_context()->grabbed_blocks; + if (reiser4_grab_space_force((__u64) 1, BA_RESERVED) != + 0) + reiser4_panic("umka-1250", + "No space left during flush."); + + assert("jmacd-18923", + znode_is_write_locked(parent_coord->node)); + znode_make_dirty(parent_coord->node); + grabbed2free_mark(grabbed); + } + } + + return 0; +} + +/* INITIAL ALLOCATE ANCESTORS STEP (REVERSE PARENT-FIRST ALLOCATION BEFORE + FORWARD PARENT-FIRST LOOP BEGINS) */ + +/* Get the leftmost child for given coord. */ +static int get_leftmost_child_of_unit(const coord_t *coord, jnode ** child) +{ + int ret; + + ret = item_utmost_child(coord, LEFT_SIDE, child); + + if (ret) + return ret; + + if (IS_ERR(*child)) + return PTR_ERR(*child); + + return 0; +} + +/* This step occurs after the left- and right-scans are completed, before + starting the forward parent-first traversal. Here we attempt to allocate + ancestors of the starting flush point, which means continuing in the reverse + parent-first direction to the parent, grandparent, and so on (as long as the + child is a leftmost child). This routine calls a recursive process, + alloc_one_ancestor, which does the real work, except there is special-case + handling here for the first ancestor, which may be a twig. At each level + (here and alloc_one_ancestor), we check for relocation and then, if the child + is a leftmost child, repeat at the next level. On the way back down (the + recursion), we allocate the ancestors in parent-first order. */ +static int alloc_pos_and_ancestors(flush_pos_t *pos) +{ + int ret = 0; + lock_handle plock; + load_count pload; + coord_t pcoord; + + if (znode_check_flushprepped(pos->lock.node)) + return 0; + + coord_init_invalid(&pcoord, NULL); + init_lh(&plock); + init_load_count(&pload); + + if (pos->state == POS_ON_EPOINT) { + /* a special case for pos on twig level, where we already have + a lock on parent node. */ + /* The parent may not be dirty, in which case we should decide + whether to relocate the child now. If decision is made to + relocate the child, the parent is marked dirty. */ + ret = + reverse_relocate_check_dirty_parent(pos->child, &pos->coord, + pos); + if (ret) + goto exit; + + /* FIXME_NFQUCMPD: We only need to allocate the twig (if child + is leftmost) and the leaf/child, so recursion is not needed. + Levels above the twig will be allocated for + write-optimization before the transaction commits. */ + + /* Do the recursive step, allocating zero or more of our + * ancestors. */ + ret = alloc_one_ancestor(&pos->coord, pos); + + } else { + if (!znode_is_root(pos->lock.node)) { + /* all formatted nodes except tree root */ + ret = + reiser4_get_parent(&plock, pos->lock.node, + ZNODE_WRITE_LOCK); + if (ret) + goto exit; + + ret = incr_load_count_znode(&pload, plock.node); + if (ret) + goto exit; + + ret = + find_child_ptr(plock.node, pos->lock.node, &pcoord); + if (ret) + goto exit; + + ret = + reverse_relocate_check_dirty_parent(ZJNODE + (pos->lock. + node), &pcoord, + pos); + if (ret) + goto exit; + + ret = alloc_one_ancestor(&pcoord, pos); + if (ret) + goto exit; + } + + ret = allocate_znode(pos->lock.node, &pcoord, pos); + } +exit: + done_load_count(&pload); + done_lh(&plock); + return ret; +} + +/* This is the recursive step described in alloc_pos_and_ancestors, above. + Ignoring the call to set_preceder, which is the next function described, this + checks if the child is a leftmost child and returns if it is not. If the + child is a leftmost child it checks for relocation, possibly dirtying the + parent. Then it performs the recursive step. */ +static int alloc_one_ancestor(const coord_t *coord, flush_pos_t *pos) +{ + int ret = 0; + lock_handle alock; + load_count aload; + coord_t acoord; + + /* As we ascend at the left-edge of the region to flush, take this + opportunity at the twig level to find our parent-first preceder + unless we have already set it. */ + if (pos->preceder.blk == 0) { + ret = set_preceder(coord, pos); + if (ret != 0) + return ret; + } + + /* If the ancestor is clean or already allocated, or if the child is not + a leftmost child, stop going up, even leaving coord->node not + flushprepped. */ + if (znode_check_flushprepped(coord->node) + || !coord_is_leftmost_unit(coord)) + return 0; + + init_lh(&alock); + init_load_count(&aload); + coord_init_invalid(&acoord, NULL); + + /* Only ascend to the next level if it is a leftmost child, but + write-lock the parent in case we will relocate the child. */ + if (!znode_is_root(coord->node)) { + + ret = + jnode_lock_parent_coord(ZJNODE(coord->node), &acoord, + &alock, &aload, ZNODE_WRITE_LOCK, + 0); + if (ret != 0) { + /* FIXME(C): check EINVAL, E_DEADLOCK */ + goto exit; + } + + ret = + reverse_relocate_check_dirty_parent(ZJNODE(coord->node), + &acoord, pos); + if (ret != 0) + goto exit; + + /* Recursive call. */ + if (!znode_check_flushprepped(acoord.node)) { + ret = alloc_one_ancestor(&acoord, pos); + if (ret) + goto exit; + } + } + + /* Note: we call allocate with the parent write-locked (except at the + root) in case we relocate the child, in which case it will modify the + parent during this call. */ + ret = allocate_znode(coord->node, &acoord, pos); + +exit: + done_load_count(&aload); + done_lh(&alock); + return ret; +} + +/* During the reverse parent-first alloc_pos_and_ancestors process described + above there is a call to this function at the twig level. During + alloc_pos_and_ancestors we may ask: should this node be relocated (in reverse + parent-first context)? We repeat this process as long as the child is the + leftmost child, eventually reaching an ancestor of the flush point that is + not a leftmost child. The preceder of that ancestors, which is not a leftmost + child, is actually on the leaf level. The preceder of that block is the + left-neighbor of the flush point. The preceder of that block is the rightmost + child of the twig on the left. So, when alloc_pos_and_ancestors passes upward + through the twig level, it stops momentarily to remember the block of the + rightmost child of the twig on the left and sets it to the flush_position's + preceder_hint. + + There is one other place where we may set the flush_position's preceder hint, + which is during scan-left. +*/ +static int set_preceder(const coord_t *coord_in, flush_pos_t *pos) +{ + int ret; + coord_t coord; + lock_handle left_lock; + load_count left_load; + + coord_dup(&coord, coord_in); + + init_lh(&left_lock); + init_load_count(&left_load); + + /* FIXME(B): Same FIXME as in "Find the preceder" in + reverse_relocate_test. coord_is_leftmost_unit is not the right test + if the unformatted child is in the middle of the first extent unit.*/ + if (!coord_is_leftmost_unit(&coord)) { + coord_prev_unit(&coord); + } else { + ret = + reiser4_get_left_neighbor(&left_lock, coord.node, + ZNODE_READ_LOCK, GN_SAME_ATOM); + if (ret) { + /* If we fail for any reason it doesn't matter because + the preceder is only a hint. We are low-priority at + this point, so this must be the case. */ + if (ret == -E_REPEAT || ret == -E_NO_NEIGHBOR || + ret == -ENOENT || ret == -EINVAL + || ret == -E_DEADLOCK) + ret = 0; + goto exit; + } + + ret = incr_load_count_znode(&left_load, left_lock.node); + if (ret) + goto exit; + + coord_init_last_unit(&coord, left_lock.node); + } + + ret = + item_utmost_child_real_block(&coord, RIGHT_SIDE, + &pos->preceder.blk); +exit: + check_preceder(pos->preceder.blk); + done_load_count(&left_load); + done_lh(&left_lock); + return ret; +} + +/* MAIN SQUEEZE AND ALLOCATE LOOP (THREE BIG FUNCTIONS) */ + +/* This procedure implements the outer loop of the flush algorithm. To put this + in context, here is the general list of steps taken by the flush routine as a + whole: + + 1. Scan-left + 2. Scan-right (maybe) + 3. Allocate initial flush position and its ancestors + 4. + 5. + 6. + + This procedure implements the loop in steps 4 through 6 in the above listing. + + Step 4: if the current flush position is an extent item (position on the twig + level), it allocates the extent (allocate_extent_item_in_place) then shifts + to the next coordinate. If the next coordinate's leftmost child needs + flushprep, we will continue. If the next coordinate is an internal item, we + descend back to the leaf level, otherwise we repeat a step #4 (labeled + ALLOC_EXTENTS below). If the "next coordinate" brings us past the end of the + twig level, then we call reverse_relocate_end_of_twig to possibly dirty the + next (right) twig, prior to step #5 which moves to the right. + + Step 5: calls squalloc_changed_ancestors, which initiates a recursive call up + the tree to allocate any ancestors of the next-right flush position that are + not also ancestors of the current position. Those ancestors (in top-down + order) are the next in parent-first order. We squeeze adjacent nodes on the + way up until the right node and current node share the same parent, then + allocate on the way back down. Finally, this step sets the flush position to + the next-right node. Then repeat steps 4 and 5. +*/ + +/* SQUEEZE CODE */ + +/* squalloc_right_twig helper function, cut a range of extent items from + cut node to->node from the beginning up to coord @to. */ +static int squalloc_right_twig_cut(coord_t *to, reiser4_key * to_key, + znode * left) +{ + coord_t from; + reiser4_key from_key; + + coord_init_first_unit(&from, to->node); + item_key_by_coord(&from, &from_key); + + return cut_node_content(&from, to, &from_key, to_key, NULL); +} + +/* Copy as much of the leading extents from @right to @left, allocating + unallocated extents as they are copied. Returns SQUEEZE_TARGET_FULL or + SQUEEZE_SOURCE_EMPTY when no more can be shifted. If the next item is an + internal item it calls shift_one_internal_unit and may then return + SUBTREE_MOVED. */ +static int squeeze_right_twig(znode * left, znode * right, flush_pos_t *pos) +{ + int ret = SUBTREE_MOVED; + coord_t coord; /* used to iterate over items */ + reiser4_key stop_key; + + assert("jmacd-2008", !node_is_empty(right)); + coord_init_first_unit(&coord, right); + + /* FIXME: can be optimized to cut once */ + while (!node_is_empty(coord.node) && item_is_extent(&coord)) { + ON_DEBUG(void *vp); + + assert("vs-1468", coord_is_leftmost_unit(&coord)); + ON_DEBUG(vp = shift_check_prepare(left, coord.node)); + + /* stop_key is used to find what was copied and what to cut */ + stop_key = *reiser4_min_key(); + ret = squalloc_extent(left, &coord, pos, &stop_key); + if (ret != SQUEEZE_CONTINUE) { + ON_DEBUG(kfree(vp)); + break; + } + assert("vs-1465", !keyeq(&stop_key, reiser4_min_key())); + + /* Helper function to do the cutting. */ + set_key_offset(&stop_key, get_key_offset(&stop_key) - 1); + check_me("vs-1466", + squalloc_right_twig_cut(&coord, &stop_key, left) == 0); + + ON_DEBUG(shift_check(vp, left, coord.node)); + } + + if (node_is_empty(coord.node)) + ret = SQUEEZE_SOURCE_EMPTY; + + if (ret == SQUEEZE_TARGET_FULL) + goto out; + + if (node_is_empty(right)) { + /* The whole right node was copied into @left. */ + assert("vs-464", ret == SQUEEZE_SOURCE_EMPTY); + goto out; + } + + coord_init_first_unit(&coord, right); + + if (!item_is_internal(&coord)) { + /* we do not want to squeeze anything else to left neighbor + because "slum" is over */ + ret = SQUEEZE_TARGET_FULL; + goto out; + } + assert("jmacd-433", item_is_internal(&coord)); + + /* Shift an internal unit. The child must be allocated before shifting + any more extents, so we stop here. */ + ret = shift_one_internal_unit(left, right); + +out: + assert("jmacd-8612", ret < 0 || ret == SQUEEZE_TARGET_FULL + || ret == SUBTREE_MOVED || ret == SQUEEZE_SOURCE_EMPTY); + + if (ret == SQUEEZE_TARGET_FULL) { + /* We submit prepped nodes here and expect that this @left twig + * will not be modified again during this jnode_flush() call. */ + int ret1; + + /* NOTE: seems like io is done under long term locks. */ + ret1 = write_prepped_nodes(pos); + if (ret1 < 0) + return ret1; + } + + return ret; +} + +#if REISER4_DEBUG +static void item_convert_invariant(flush_pos_t *pos) +{ + assert("edward-1225", coord_is_existing_item(&pos->coord)); + if (chaining_data_present(pos)) { + item_plugin *iplug = item_convert_plug(pos); + + assert("edward-1000", + iplug == item_plugin_by_coord(&pos->coord)); + assert("edward-1001", iplug->f.convert != NULL); + } else + assert("edward-1226", pos->child == NULL); +} +#else + +#define item_convert_invariant(pos) noop + +#endif + +/* Scan node items starting from the first one and apply for each + item its flush ->convert() method (if any). This method may + resize/kill the item so the tree will be changed. +*/ +static int convert_node(flush_pos_t *pos, znode * node) +{ + int ret = 0; + item_plugin *iplug; + + assert("edward-304", pos != NULL); + assert("edward-305", pos->child == NULL); + assert("edward-475", znode_convertible(node)); + assert("edward-669", znode_is_wlocked(node)); + assert("edward-1210", !node_is_empty(node)); + + if (znode_get_level(node) != LEAF_LEVEL) + /* unsupported */ + goto exit; + + coord_init_first_unit(&pos->coord, node); + + while (1) { + ret = 0; + coord_set_to_left(&pos->coord); + item_convert_invariant(pos); + + iplug = item_plugin_by_coord(&pos->coord); + assert("edward-844", iplug != NULL); + + if (iplug->f.convert) { + ret = iplug->f.convert(pos); + if (ret) + goto exit; + } + assert("edward-307", pos->child == NULL); + + if (coord_next_item(&pos->coord)) { + /* node is over */ + + if (!chaining_data_present(pos)) + /* finished this node */ + break; + if (should_chain_next_node(pos)) { + /* go to next node */ + move_chaining_data(pos, 0/* to next node */); + break; + } + /* repeat this node */ + move_chaining_data(pos, 1/* this node */); + continue; + } + /* Node is not over. + Check if there is attached convert data. + If so roll one item position back and repeat + on this node + */ + if (chaining_data_present(pos)) { + + if (iplug != item_plugin_by_coord(&pos->coord)) + set_item_convert_count(pos, 0); + + ret = coord_prev_item(&pos->coord); + assert("edward-1003", !ret); + + move_chaining_data(pos, 1/* this node */); + } + } + JF_CLR(ZJNODE(node), JNODE_CONVERTIBLE); + znode_make_dirty(node); +exit: + assert("edward-1004", !ret); + return ret; +} + +/* Squeeze and allocate the right neighbor. This is called after @left and + its current children have been squeezed and allocated already. This + procedure's job is to squeeze and items from @right to @left. + + If at the leaf level, use the shift_everything_left memcpy-optimized + version of shifting (squeeze_right_leaf). + + If at the twig level, extents are allocated as they are shifted from @right + to @left (squalloc_right_twig). + + At any other level, shift one internal item and return to the caller + (squalloc_parent_first) so that the shifted-subtree can be processed in + parent-first order. + + When unit of internal item is moved, squeezing stops and SUBTREE_MOVED is + returned. When all content of @right is squeezed, SQUEEZE_SOURCE_EMPTY is + returned. If nothing can be moved into @left anymore, SQUEEZE_TARGET_FULL + is returned. +*/ + +static int squeeze_right_neighbor(flush_pos_t *pos, znode * left, + znode * right) +{ + int ret; + + /* FIXME it is possible to see empty hasn't-heard-banshee node in a + * tree owing to error (for example, ENOSPC) in write */ + /* assert("jmacd-9321", !node_is_empty(left)); */ + assert("jmacd-9322", !node_is_empty(right)); + assert("jmacd-9323", znode_get_level(left) == znode_get_level(right)); + + switch (znode_get_level(left)) { + case TWIG_LEVEL: + /* Shift with extent allocating until either an internal item + is encountered or everything is shifted or no free space + left in @left */ + ret = squeeze_right_twig(left, right, pos); + break; + + default: + /* All other levels can use shift_everything until we implement + per-item flush plugins. */ + ret = squeeze_right_non_twig(left, right); + break; + } + + assert("jmacd-2011", (ret < 0 || + ret == SQUEEZE_SOURCE_EMPTY + || ret == SQUEEZE_TARGET_FULL + || ret == SUBTREE_MOVED)); + return ret; +} + +static int squeeze_right_twig_and_advance_coord(flush_pos_t *pos, + znode * right) +{ + int ret; + + ret = squeeze_right_twig(pos->lock.node, right, pos); + if (ret < 0) + return ret; + if (ret > 0) { + coord_init_after_last_item(&pos->coord, pos->lock.node); + return ret; + } + + coord_init_last_unit(&pos->coord, pos->lock.node); + return 0; +} + +/* forward declaration */ +static int squalloc_upper_levels(flush_pos_t *, znode *, znode *); + +/* do a fast check for "same parents" condition before calling + * squalloc_upper_levels() */ +static inline int check_parents_and_squalloc_upper_levels(flush_pos_t *pos, + znode * left, + znode * right) +{ + if (znode_same_parents(left, right)) + return 0; + + return squalloc_upper_levels(pos, left, right); +} + +/* Check whether the parent of given @right node needs to be processes + ((re)allocated) prior to processing of the child. If @left and @right do not + share at least the parent of the @right is after the @left but before the + @right in parent-first order, we have to (re)allocate it before the @right + gets (re)allocated. */ +static int squalloc_upper_levels(flush_pos_t *pos, znode * left, znode * right) +{ + int ret; + + lock_handle left_parent_lock; + lock_handle right_parent_lock; + + load_count left_parent_load; + load_count right_parent_load; + + init_lh(&left_parent_lock); + init_lh(&right_parent_lock); + + init_load_count(&left_parent_load); + init_load_count(&right_parent_load); + + ret = reiser4_get_parent(&left_parent_lock, left, ZNODE_WRITE_LOCK); + if (ret) + goto out; + + ret = reiser4_get_parent(&right_parent_lock, right, ZNODE_WRITE_LOCK); + if (ret) + goto out; + + /* Check for same parents */ + if (left_parent_lock.node == right_parent_lock.node) + goto out; + + if (znode_check_flushprepped(right_parent_lock.node)) { + /* Keep parent-first order. In the order, the right parent node + stands before the @right node. If it is already allocated, + we set the preceder (next block search start point) to its + block number, @right node should be allocated after it. + + However, preceder is set only if the right parent is on twig + level. The explanation is the following: new branch nodes are + allocated over already allocated children while the tree + grows, it is difficult to keep tree ordered, we assume that + only leaves and twings are correctly allocated. So, only + twigs are used as a preceder for allocating of the rest of + the slum. */ + if (znode_get_level(right_parent_lock.node) == TWIG_LEVEL) { + pos->preceder.blk = + *znode_get_block(right_parent_lock.node); + check_preceder(pos->preceder.blk); + } + goto out; + } + + ret = incr_load_count_znode(&left_parent_load, left_parent_lock.node); + if (ret) + goto out; + + ret = incr_load_count_znode(&right_parent_load, right_parent_lock.node); + if (ret) + goto out; + + ret = + squeeze_right_neighbor(pos, left_parent_lock.node, + right_parent_lock.node); + /* We stop if error. We stop if some items/units were shifted (ret == 0) + * and thus @right changed its parent. It means we have not process + * right_parent node prior to processing of @right. Positive return + * values say that shifting items was not happen because of "empty + * source" or "target full" conditions. */ + if (ret <= 0) + goto out; + + /* parent(@left) and parent(@right) may have different parents also. We + * do a recursive call for checking that. */ + ret = + check_parents_and_squalloc_upper_levels(pos, left_parent_lock.node, + right_parent_lock.node); + if (ret) + goto out; + + /* allocate znode when going down */ + ret = lock_parent_and_allocate_znode(right_parent_lock.node, pos); + +out: + done_load_count(&left_parent_load); + done_load_count(&right_parent_load); + + done_lh(&left_parent_lock); + done_lh(&right_parent_lock); + + return ret; +} + +/* Check the leftmost child "flushprepped" status, also returns true if child + * node was not found in cache. */ +static int leftmost_child_of_unit_check_flushprepped(const coord_t *coord) +{ + int ret; + int prepped; + + jnode *child; + + ret = get_leftmost_child_of_unit(coord, &child); + + if (ret) + return ret; + + if (child) { + prepped = jnode_check_flushprepped(child); + jput(child); + } else { + /* We consider not existing child as a node which slum + processing should not continue to. Not cached node is clean, + so it is flushprepped. */ + prepped = 1; + } + + return prepped; +} + +/* (re)allocate znode with automated getting parent node */ +static int lock_parent_and_allocate_znode(znode * node, flush_pos_t *pos) +{ + int ret; + lock_handle parent_lock; + load_count parent_load; + coord_t pcoord; + + assert("zam-851", znode_is_write_locked(node)); + + init_lh(&parent_lock); + init_load_count(&parent_load); + + ret = reiser4_get_parent(&parent_lock, node, ZNODE_WRITE_LOCK); + if (ret) + goto out; + + ret = incr_load_count_znode(&parent_load, parent_lock.node); + if (ret) + goto out; + + ret = find_child_ptr(parent_lock.node, node, &pcoord); + if (ret) + goto out; + + ret = allocate_znode(node, &pcoord, pos); + +out: + done_load_count(&parent_load); + done_lh(&parent_lock); + return ret; +} + +/* Process nodes on leaf level until unformatted node or rightmost node in the + * slum reached. */ +static int handle_pos_on_formatted(flush_pos_t *pos) +{ + int ret; + lock_handle right_lock; + load_count right_load; + + init_lh(&right_lock); + init_load_count(&right_load); + + if (should_convert_node(pos, pos->lock.node)) { + ret = convert_node(pos, pos->lock.node); + if (ret) + return ret; + } + + while (1) { + int expected; + expected = should_convert_next_node(pos); + ret = neighbor_in_slum(pos->lock.node, &right_lock, RIGHT_SIDE, + ZNODE_WRITE_LOCK, !expected, expected); + if (ret) { + if (expected) + warning("edward-1495", + "Expected neighbor not found (ret = %d). Fsck?", + ret); + break; + } + + /* we don't prep(allocate) nodes for flushing twice. This can be + * suboptimal, or it can be optimal. For now we choose to live + * with the risk that it will be suboptimal because it would be + * quite complex to code it to be smarter. */ + if (znode_check_flushprepped(right_lock.node) + && !znode_convertible(right_lock.node)) { + assert("edward-1005", !should_convert_next_node(pos)); + pos_stop(pos); + break; + } + + ret = incr_load_count_znode(&right_load, right_lock.node); + if (ret) + break; + if (should_convert_node(pos, right_lock.node)) { + ret = convert_node(pos, right_lock.node); + if (ret) + break; + if (node_is_empty(right_lock.node)) { + /* node became empty after converting, repeat */ + done_load_count(&right_load); + done_lh(&right_lock); + continue; + } + } + + /* squeeze _before_ going upward. */ + ret = + squeeze_right_neighbor(pos, pos->lock.node, + right_lock.node); + if (ret < 0) + break; + + if (znode_check_flushprepped(right_lock.node)) { + if (should_convert_next_node(pos)) { + /* in spite of flushprepped status of the node, + its right slum neighbor should be converted*/ + assert("edward-953", convert_data(pos)); + assert("edward-954", item_convert_data(pos)); + + if (node_is_empty(right_lock.node)) { + done_load_count(&right_load); + done_lh(&right_lock); + } else + move_flush_pos(pos, &right_lock, + &right_load, NULL); + continue; + } + pos_stop(pos); + break; + } + + if (node_is_empty(right_lock.node)) { + /* repeat if right node was squeezed completely */ + done_load_count(&right_load); + done_lh(&right_lock); + continue; + } + + /* parent(right_lock.node) has to be processed before + * (right_lock.node) due to "parent-first" allocation order. */ + ret = + check_parents_and_squalloc_upper_levels(pos, pos->lock.node, + right_lock.node); + if (ret) + break; + /* (re)allocate _after_ going upward */ + ret = lock_parent_and_allocate_znode(right_lock.node, pos); + if (ret) + break; + if (should_terminate_squalloc(pos)) { + set_item_convert_count(pos, 0); + break; + } + + /* advance the flush position to the right neighbor */ + move_flush_pos(pos, &right_lock, &right_load, NULL); + + ret = rapid_flush(pos); + if (ret) + break; + } + check_convert_info(pos); + done_load_count(&right_load); + done_lh(&right_lock); + + /* This function indicates via pos whether to stop or go to twig or + * continue on current level. */ + return ret; + +} + +/* Process nodes on leaf level until unformatted node or rightmost node in the + * slum reached. */ +static int handle_pos_on_leaf(flush_pos_t *pos) +{ + int ret; + + assert("zam-845", pos->state == POS_ON_LEAF); + + ret = handle_pos_on_formatted(pos); + + if (ret == -E_NO_NEIGHBOR) { + /* cannot get right neighbor, go process extents. */ + pos->state = POS_TO_TWIG; + return 0; + } + + return ret; +} + +/* Process slum on level > 1 */ +static int handle_pos_on_internal(flush_pos_t *pos) +{ + assert("zam-850", pos->state == POS_ON_INTERNAL); + return handle_pos_on_formatted(pos); +} + +/* check whether squalloc should stop before processing given extent */ +static int squalloc_extent_should_stop(flush_pos_t *pos) +{ + assert("zam-869", item_is_extent(&pos->coord)); + + /* pos->child is a jnode handle_pos_on_extent() should start with in + * stead of the first child of the first extent unit. */ + if (pos->child) { + int prepped; + + assert("vs-1383", jnode_is_unformatted(pos->child)); + prepped = jnode_check_flushprepped(pos->child); + pos->pos_in_unit = + jnode_get_index(pos->child) - + extent_unit_index(&pos->coord); + assert("vs-1470", + pos->pos_in_unit < extent_unit_width(&pos->coord)); + assert("nikita-3434", + ergo(extent_is_unallocated(&pos->coord), + pos->pos_in_unit == 0)); + jput(pos->child); + pos->child = NULL; + + return prepped; + } + + pos->pos_in_unit = 0; + if (extent_is_unallocated(&pos->coord)) + return 0; + + return leftmost_child_of_unit_check_flushprepped(&pos->coord); +} + +/* Handle the case when regular reiser4 tree (znodes connected one to its + * neighbors by sibling pointers) is interrupted on leaf level by one or more + * unformatted nodes. By having a lock on twig level and use extent code + * routines to process unformatted nodes we swim around an irregular part of + * reiser4 tree. */ +static int handle_pos_on_twig(flush_pos_t *pos) +{ + int ret; + + assert("zam-844", pos->state == POS_ON_EPOINT); + assert("zam-843", item_is_extent(&pos->coord)); + + /* We decide should we continue slum processing with current extent + unit: if leftmost child of current extent unit is flushprepped + (i.e. clean or already processed by flush) we stop squalloc(). There + is a fast check for unallocated extents which we assume contain all + not flushprepped nodes. */ + /* FIXME: Here we implement simple check, we are only looking on the + leftmost child. */ + ret = squalloc_extent_should_stop(pos); + if (ret != 0) { + pos_stop(pos); + return ret; + } + + while (pos_valid(pos) && coord_is_existing_unit(&pos->coord) + && item_is_extent(&pos->coord)) { + ret = reiser4_alloc_extent(pos); + if (ret) + break; + coord_next_unit(&pos->coord); + } + + if (coord_is_after_rightmost(&pos->coord)) { + pos->state = POS_END_OF_TWIG; + return 0; + } + if (item_is_internal(&pos->coord)) { + pos->state = POS_TO_LEAF; + return 0; + } + + assert("zam-860", item_is_extent(&pos->coord)); + + /* "slum" is over */ + pos->state = POS_INVALID; + return 0; +} + +/* When we about to return flush position from twig to leaf level we can process + * the right twig node or move position to the leaf. This processes right twig + * if it is possible and jump to leaf level if not. */ +static int handle_pos_end_of_twig(flush_pos_t *pos) +{ + int ret; + lock_handle right_lock; + load_count right_load; + coord_t at_right; + jnode *child = NULL; + + assert("zam-848", pos->state == POS_END_OF_TWIG); + assert("zam-849", coord_is_after_rightmost(&pos->coord)); + + init_lh(&right_lock); + init_load_count(&right_load); + + /* We get a lock on the right twig node even it is not dirty because + * slum continues or discontinues on leaf level not on next twig. This + * lock on the right twig is needed for getting its leftmost child. */ + ret = + reiser4_get_right_neighbor(&right_lock, pos->lock.node, + ZNODE_WRITE_LOCK, GN_SAME_ATOM); + if (ret) + goto out; + + ret = incr_load_count_znode(&right_load, right_lock.node); + if (ret) + goto out; + + /* right twig could be not dirty */ + if (JF_ISSET(ZJNODE(right_lock.node), JNODE_DIRTY)) { + /* If right twig node is dirty we always attempt to squeeze it + * content to the left... */ +became_dirty: + ret = + squeeze_right_twig_and_advance_coord(pos, right_lock.node); + if (ret <= 0) { + /* pos->coord is on internal item, go to leaf level, or + * we have an error which will be caught in squalloc() + */ + pos->state = POS_TO_LEAF; + goto out; + } + + /* If right twig was squeezed completely we wave to re-lock + * right twig. now it is done through the top-level squalloc + * routine. */ + if (node_is_empty(right_lock.node)) + goto out; + + /* ... and prep it if it is not yet prepped */ + if (!znode_check_flushprepped(right_lock.node)) { + /* As usual, process parent before ... */ + ret = + check_parents_and_squalloc_upper_levels(pos, + pos->lock. + node, + right_lock. + node); + if (ret) + goto out; + + /* ... processing the child */ + ret = + lock_parent_and_allocate_znode(right_lock.node, + pos); + if (ret) + goto out; + } + } else { + coord_init_first_unit(&at_right, right_lock.node); + + /* check first child of next twig, should we continue there ? */ + ret = get_leftmost_child_of_unit(&at_right, &child); + if (ret || child == NULL || jnode_check_flushprepped(child)) { + pos_stop(pos); + goto out; + } + + /* check clean twig for possible relocation */ + if (!znode_check_flushprepped(right_lock.node)) { + ret = + reverse_relocate_check_dirty_parent(child, + &at_right, pos); + if (ret) + goto out; + if (JF_ISSET(ZJNODE(right_lock.node), JNODE_DIRTY)) + goto became_dirty; + } + } + + assert("zam-875", znode_check_flushprepped(right_lock.node)); + + /* Update the preceder by a block number of just processed right twig + * node. The code above could miss the preceder updating because + * allocate_znode() could not be called for this node. */ + pos->preceder.blk = *znode_get_block(right_lock.node); + check_preceder(pos->preceder.blk); + + coord_init_first_unit(&at_right, right_lock.node); + assert("zam-868", coord_is_existing_unit(&at_right)); + + pos->state = item_is_extent(&at_right) ? POS_ON_EPOINT : POS_TO_LEAF; + move_flush_pos(pos, &right_lock, &right_load, &at_right); + +out: + done_load_count(&right_load); + done_lh(&right_lock); + + if (child) + jput(child); + + return ret; +} + +/* Move the pos->lock to leaf node pointed by pos->coord, check should we + * continue there. */ +static int handle_pos_to_leaf(flush_pos_t *pos) +{ + int ret; + lock_handle child_lock; + load_count child_load; + jnode *child; + + assert("zam-846", pos->state == POS_TO_LEAF); + assert("zam-847", item_is_internal(&pos->coord)); + + init_lh(&child_lock); + init_load_count(&child_load); + + ret = get_leftmost_child_of_unit(&pos->coord, &child); + if (ret) + return ret; + if (child == NULL) { + pos_stop(pos); + return 0; + } + + if (jnode_check_flushprepped(child)) { + pos->state = POS_INVALID; + goto out; + } + + ret = + longterm_lock_znode(&child_lock, JZNODE(child), ZNODE_WRITE_LOCK, + ZNODE_LOCK_LOPRI); + if (ret) + goto out; + + ret = incr_load_count_znode(&child_load, JZNODE(child)); + if (ret) + goto out; + + ret = allocate_znode(JZNODE(child), &pos->coord, pos); + if (ret) + goto out; + + /* move flush position to leaf level */ + pos->state = POS_ON_LEAF; + move_flush_pos(pos, &child_lock, &child_load, NULL); + + if (node_is_empty(JZNODE(child))) { + ret = delete_empty_node(JZNODE(child)); + pos->state = POS_INVALID; + } +out: + done_load_count(&child_load); + done_lh(&child_lock); + jput(child); + + return ret; +} + +/* move pos from leaf to twig, and move lock from leaf to twig. */ +/* Move pos->lock to upper (twig) level */ +static int handle_pos_to_twig(flush_pos_t *pos) +{ + int ret; + + lock_handle parent_lock; + load_count parent_load; + coord_t pcoord; + + assert("zam-852", pos->state == POS_TO_TWIG); + + init_lh(&parent_lock); + init_load_count(&parent_load); + + ret = + reiser4_get_parent(&parent_lock, pos->lock.node, ZNODE_WRITE_LOCK); + if (ret) + goto out; + + ret = incr_load_count_znode(&parent_load, parent_lock.node); + if (ret) + goto out; + + ret = find_child_ptr(parent_lock.node, pos->lock.node, &pcoord); + if (ret) + goto out; + + assert("zam-870", item_is_internal(&pcoord)); + coord_next_item(&pcoord); + + if (coord_is_after_rightmost(&pcoord)) + pos->state = POS_END_OF_TWIG; + else if (item_is_extent(&pcoord)) + pos->state = POS_ON_EPOINT; + else { + /* Here we understand that getting -E_NO_NEIGHBOR in + * handle_pos_on_leaf() was because of just a reaching edge of + * slum */ + pos_stop(pos); + goto out; + } + + move_flush_pos(pos, &parent_lock, &parent_load, &pcoord); + +out: + done_load_count(&parent_load); + done_lh(&parent_lock); + + return ret; +} + +typedef int (*pos_state_handle_t) (flush_pos_t *); +static pos_state_handle_t flush_pos_handlers[] = { + /* process formatted nodes on leaf level, keep lock on a leaf node */ + [POS_ON_LEAF] = handle_pos_on_leaf, + /* process unformatted nodes, keep lock on twig node, pos->coord points + * to extent currently being processed */ + [POS_ON_EPOINT] = handle_pos_on_twig, + /* move a lock from leaf node to its parent for further processing of + unformatted nodes */ + [POS_TO_TWIG] = handle_pos_to_twig, + /* move a lock from twig to leaf level when a processing of unformatted + * nodes finishes, pos->coord points to the leaf node we jump to */ + [POS_TO_LEAF] = handle_pos_to_leaf, + /* after processing last extent in the twig node, attempting to shift + * items from the twigs right neighbor and process them while shifting*/ + [POS_END_OF_TWIG] = handle_pos_end_of_twig, + /* process formatted nodes on internal level, keep lock on an internal + node */ + [POS_ON_INTERNAL] = handle_pos_on_internal +}; + +/* Advance flush position horizontally, prepare for flushing ((re)allocate, + * squeeze, encrypt) nodes and their ancestors in "parent-first" order */ +static int squalloc(flush_pos_t *pos) +{ + int ret = 0; + + /* maybe needs to be made a case statement with handle_pos_on_leaf as + * first case, for greater CPU efficiency? Measure and see.... -Hans */ + while (pos_valid(pos)) { + ret = flush_pos_handlers[pos->state] (pos); + if (ret < 0) + break; + + ret = rapid_flush(pos); + if (ret) + break; + } + + /* any positive value or -E_NO_NEIGHBOR are legal return codes for + handle_pos* routines, -E_NO_NEIGHBOR means that slum edge was + reached */ + if (ret > 0 || ret == -E_NO_NEIGHBOR) + ret = 0; + + return ret; +} + +static void update_ldkey(znode * node) +{ + reiser4_key ldkey; + + assert_rw_write_locked(&(znode_get_tree(node)->dk_lock)); + if (node_is_empty(node)) + return; + + znode_set_ld_key(node, leftmost_key_in_node(node, &ldkey)); +} + +/* this is to be called after calling of shift node's method to shift data from + @right to @left. It sets left delimiting keys of @left and @right to keys of + first items of @left and @right correspondingly and sets right delimiting key + of @left to first key of @right */ +static void update_znode_dkeys(znode * left, znode * right) +{ + assert_rw_write_locked(&(znode_get_tree(right)->dk_lock)); + assert("vs-1629", (znode_is_write_locked(left) && + znode_is_write_locked(right))); + + /* we need to update left delimiting of left if it was empty before + shift */ + update_ldkey(left); + update_ldkey(right); + if (node_is_empty(right)) + znode_set_rd_key(left, znode_get_rd_key(right)); + else + znode_set_rd_key(left, znode_get_ld_key(right)); +} + +/* try to shift everything from @right to @left. If everything was shifted - + @right is removed from the tree. Result is the number of bytes shifted. */ +static int +shift_everything_left(znode * right, znode * left, carry_level * todo) +{ + coord_t from; + node_plugin *nplug; + carry_plugin_info info; + + coord_init_after_last_item(&from, right); + + nplug = node_plugin_by_node(right); + info.doing = NULL; + info.todo = todo; + return nplug->shift(&from, left, SHIFT_LEFT, + 1 /* delete @right if it becomes empty */ , + 1 + /* move coord @from to node @left if everything will + be shifted */ + , + &info); +} + +/* Shift as much as possible from @right to @left using the memcpy-optimized + shift_everything_left. @left and @right are formatted neighboring nodes on + leaf level. */ +static int squeeze_right_non_twig(znode * left, znode * right) +{ + int ret; + carry_pool *pool; + carry_level *todo; + + assert("nikita-2246", znode_get_level(left) == znode_get_level(right)); + + if (!JF_ISSET(ZJNODE(left), JNODE_DIRTY) || + !JF_ISSET(ZJNODE(right), JNODE_DIRTY)) + return SQUEEZE_TARGET_FULL; + + pool = init_carry_pool(sizeof(*pool) + 3 * sizeof(*todo)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + todo = (carry_level *) (pool + 1); + init_carry_level(todo, pool); + + ret = shift_everything_left(right, left, todo); + if (ret > 0) { + /* something was shifted */ + reiser4_tree *tree; + __u64 grabbed; + + znode_make_dirty(left); + znode_make_dirty(right); + + /* update delimiting keys of nodes which participated in + shift. FIXME: it would be better to have this in shift + node's operation. But it can not be done there. Nobody + remembers why, though */ + tree = znode_get_tree(left); + write_lock_dk(tree); + update_znode_dkeys(left, right); + write_unlock_dk(tree); + + /* Carry is called to update delimiting key and, maybe, to + remove empty node. */ + grabbed = get_current_context()->grabbed_blocks; + ret = reiser4_grab_space_force(tree->height, BA_RESERVED); + assert("nikita-3003", ret == 0); /* reserved space is + exhausted. Ask Hans. */ + ret = reiser4_carry(todo, NULL/* previous level */); + grabbed2free_mark(grabbed); + } else { + /* Shifting impossible, we return appropriate result code */ + ret = + node_is_empty(right) ? SQUEEZE_SOURCE_EMPTY : + SQUEEZE_TARGET_FULL; + } + + done_carry_pool(pool); + + return ret; +} + +#if REISER4_DEBUG +static int sibling_link_is_ok(const znode *left, const znode *right) +{ + int result; + + read_lock_tree(znode_get_tree(left)); + result = (left->right == right && left == right->left); + read_unlock_tree(znode_get_tree(left)); + return result; +} +#endif + +/* Shift first unit of first item if it is an internal one. Return + SQUEEZE_TARGET_FULL if it fails to shift an item, otherwise return + SUBTREE_MOVED. */ +static int shift_one_internal_unit(znode * left, znode * right) +{ + int ret; + carry_pool *pool; + carry_level *todo; + coord_t *coord; + carry_plugin_info *info; + int size, moved; + + assert("nikita-2247", znode_get_level(left) == znode_get_level(right)); + assert("nikita-2435", znode_is_write_locked(left)); + assert("nikita-2436", znode_is_write_locked(right)); + assert("nikita-2434", sibling_link_is_ok(left, right)); + + pool = init_carry_pool(sizeof(*pool) + 3 * sizeof(*todo) + + sizeof(*coord) + sizeof(*info) +#if REISER4_DEBUG + + sizeof(*coord) + 2 * sizeof(reiser4_key) +#endif + ); + if (IS_ERR(pool)) + return PTR_ERR(pool); + todo = (carry_level *) (pool + 1); + init_carry_level(todo, pool); + + coord = (coord_t *) (todo + 3); + coord_init_first_unit(coord, right); + info = (carry_plugin_info *) (coord + 1); + +#if REISER4_DEBUG + if (!node_is_empty(left)) { + coord_t *last; + reiser4_key *right_key; + reiser4_key *left_key; + + last = (coord_t *) (info + 1); + right_key = (reiser4_key *) (last + 1); + left_key = right_key + 1; + coord_init_last_unit(last, left); + + assert("nikita-2463", + keyle(item_key_by_coord(last, left_key), + item_key_by_coord(coord, right_key))); + } +#endif + + assert("jmacd-2007", item_is_internal(coord)); + + size = item_length_by_coord(coord); + info->todo = todo; + info->doing = NULL; + + ret = node_plugin_by_node(left)->shift(coord, left, SHIFT_LEFT, + 1 + /* delete @right if it becomes + empty */ + , + 0 + /* do not move coord @coord to + node @left */ + , + info); + + /* If shift returns positive, then we shifted the item. */ + assert("vs-423", ret <= 0 || size == ret); + moved = (ret > 0); + + if (moved) { + /* something was moved */ + reiser4_tree *tree; + int grabbed; + + znode_make_dirty(left); + znode_make_dirty(right); + tree = znode_get_tree(left); + write_lock_dk(tree); + update_znode_dkeys(left, right); + write_unlock_dk(tree); + + /* reserve space for delimiting keys after shifting */ + grabbed = get_current_context()->grabbed_blocks; + ret = reiser4_grab_space_force(tree->height, BA_RESERVED); + assert("nikita-3003", ret == 0); /* reserved space is + exhausted. Ask Hans. */ + + ret = reiser4_carry(todo, NULL/* previous level */); + grabbed2free_mark(grabbed); + } + + done_carry_pool(pool); + + if (ret != 0) { + /* Shift or carry operation failed. */ + assert("jmacd-7325", ret < 0); + return ret; + } + + return moved ? SUBTREE_MOVED : SQUEEZE_TARGET_FULL; +} + +/* Make the final relocate/wander decision during forward parent-first squalloc + for a znode. For unformatted nodes this is done in + plugin/item/extent.c:extent_needs_allocation(). */ +static int +allocate_znode_loaded(znode * node, + const coord_t *parent_coord, flush_pos_t *pos) +{ + int ret; + reiser4_super_info_data *sbinfo = get_current_super_private(); + /* FIXME(D): We have the node write-locked and should have checked for ! + allocated() somewhere before reaching this point, but there can be a + race, so this assertion is bogus. */ + assert("jmacd-7987", !jnode_check_flushprepped(ZJNODE(node))); + assert("jmacd-7988", znode_is_write_locked(node)); + assert("jmacd-7989", coord_is_invalid(parent_coord) + || znode_is_write_locked(parent_coord->node)); + + if (ZF_ISSET(node, JNODE_REPACK) || ZF_ISSET(node, JNODE_CREATED) || + znode_is_root(node) || + /* We have enough nodes to relocate no matter what. */ + (pos->leaf_relocate != 0 && znode_get_level(node) == LEAF_LEVEL)) { + /* No need to decide with new nodes, they are treated the same + as relocate. If the root node is dirty, relocate. */ + if (pos->preceder.blk == 0) { + /* preceder is unknown and we have decided to relocate + node -- using of default value for search start is + better than search from block #0. */ + get_blocknr_hint_default(&pos->preceder.blk); + check_preceder(pos->preceder.blk); + } + + goto best_reloc; + + } else if (pos->preceder.blk == 0) { + /* If we don't know the preceder, leave it where it is. */ + jnode_make_wander(ZJNODE(node)); + } else { + /* Make a decision based on block distance. */ + reiser4_block_nr dist; + reiser4_block_nr nblk = *znode_get_block(node); + + assert("jmacd-6172", !reiser4_blocknr_is_fake(&nblk)); + assert("jmacd-6173", !reiser4_blocknr_is_fake(&pos->preceder.blk)); + assert("jmacd-6174", pos->preceder.blk != 0); + + if (pos->preceder.blk == nblk - 1) { + /* Ideal. */ + jnode_make_wander(ZJNODE(node)); + } else { + + dist = + (nblk < + pos->preceder.blk) ? (pos->preceder.blk - + nblk) : (nblk - + pos->preceder.blk); + + /* See if we can find a closer block + (forward direction only). */ + pos->preceder.max_dist = + min((reiser4_block_nr) sbinfo->flush. + relocate_distance, dist); + pos->preceder.level = znode_get_level(node); + + ret = allocate_znode_update(node, parent_coord, pos); + + pos->preceder.max_dist = 0; + + if (ret && (ret != -ENOSPC)) + return ret; + + if (ret == 0) { + /* Got a better allocation. */ + znode_make_reloc(node, pos->fq); + } else if (dist < sbinfo->flush.relocate_distance) { + /* The present allocation is good enough. */ + jnode_make_wander(ZJNODE(node)); + } else { + /* Otherwise, try to relocate to the best + position. */ +best_reloc: + ret = + allocate_znode_update(node, parent_coord, + pos); + if (ret != 0) + return ret; + + /* set JNODE_RELOC bit _after_ node gets + allocated */ + znode_make_reloc(node, pos->fq); + } + } + } + + /* This is the new preceder. */ + pos->preceder.blk = *znode_get_block(node); + check_preceder(pos->preceder.blk); + pos->alloc_cnt += 1; + + assert("jmacd-4277", !reiser4_blocknr_is_fake(&pos->preceder.blk)); + + return 0; +} + +static int +allocate_znode(znode * node, const coord_t *parent_coord, flush_pos_t *pos) +{ + /* + * perform znode allocation with znode pinned in memory to avoid races + * with asynchronous emergency flush (which plays with + * JNODE_FLUSH_RESERVED bit). + */ + return WITH_DATA(node, allocate_znode_loaded(node, parent_coord, pos)); +} + +/* A subroutine of allocate_znode, this is called first to see if there is a + close position to relocate to. It may return ENOSPC if there is no close + position. If there is no close position it may not relocate. This takes care + of updating the parent node with the relocated block address. */ +static int +allocate_znode_update(znode * node, const coord_t *parent_coord, + flush_pos_t *pos) +{ + int ret; + reiser4_block_nr blk; + lock_handle uber_lock; + int flush_reserved_used = 0; + int grabbed; + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + init_lh(&uber_lock); + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + grabbed = ctx->grabbed_blocks; + + /* discard e-flush allocation */ + ret = zload(node); + if (ret) + return ret; + + if (ZF_ISSET(node, JNODE_CREATED)) { + assert("zam-816", reiser4_blocknr_is_fake(znode_get_block(node))); + pos->preceder.block_stage = BLOCK_UNALLOCATED; + } else { + pos->preceder.block_stage = BLOCK_GRABBED; + + /* The disk space for relocating the @node is already reserved + * in "flush reserved" counter if @node is leaf, otherwise we + * grab space using BA_RESERVED (means grab space from whole + * disk not from only 95%). */ + if (znode_get_level(node) == LEAF_LEVEL) { + /* + * earlier (during do_jnode_make_dirty()) we decided + * that @node can possibly go into overwrite set and + * reserved block for its wandering location. + */ + txn_atom *atom = get_current_atom_locked(); + assert("nikita-3449", + ZF_ISSET(node, JNODE_FLUSH_RESERVED)); + flush_reserved2grabbed(atom, (__u64) 1); + spin_unlock_atom(atom); + /* + * we are trying to move node into relocate + * set. Allocation of relocated position "uses" + * reserved block. + */ + ZF_CLR(node, JNODE_FLUSH_RESERVED); + flush_reserved_used = 1; + } else { + ret = reiser4_grab_space_force((__u64) 1, BA_RESERVED); + if (ret != 0) + goto exit; + } + } + + /* We may do not use 5% of reserved disk space here and flush will not + pack tightly. */ + ret = reiser4_alloc_block(&pos->preceder, &blk, + BA_FORMATTED | BA_PERMANENT); + if (ret) + goto exit; + + if (!ZF_ISSET(node, JNODE_CREATED) && + (ret = + reiser4_dealloc_block(znode_get_block(node), 0, + BA_DEFER | BA_FORMATTED))) + goto exit; + + if (likely(!znode_is_root(node))) { + item_plugin *iplug; + + iplug = item_plugin_by_coord(parent_coord); + assert("nikita-2954", iplug->f.update != NULL); + iplug->f.update(parent_coord, &blk); + + znode_make_dirty(parent_coord->node); + + } else { + reiser4_tree *tree = znode_get_tree(node); + znode *uber; + + /* We take a longterm lock on the fake node in order to change + the root block number. This may cause atom fusion. */ + ret = get_uber_znode(tree, ZNODE_WRITE_LOCK, ZNODE_LOCK_HIPRI, + &uber_lock); + /* The fake node cannot be deleted, and we must have priority + here, and may not be confused with ENOSPC. */ + assert("jmacd-74412", + ret != -EINVAL && ret != -E_DEADLOCK && ret != -ENOSPC); + + if (ret) + goto exit; + + uber = uber_lock.node; + + write_lock_tree(tree); + tree->root_block = blk; + write_unlock_tree(tree); + + znode_make_dirty(uber); + } + + ret = znode_rehash(node, &blk); +exit: + if (ret) { + /* Get flush reserved block back if something fails, because + * callers assume that on error block wasn't relocated and its + * flush reserved block wasn't used. */ + if (flush_reserved_used) { + /* + * ok, we failed to move node into relocate + * set. Restore status quo. + */ + grabbed2flush_reserved((__u64) 1); + ZF_SET(node, JNODE_FLUSH_RESERVED); + } + } + zrelse(node); + done_lh(&uber_lock); + grabbed2free_mark(grabbed); + return ret; +} + +/* JNODE INTERFACE */ + +/* Lock a node (if formatted) and then get its parent locked, set the child's + coordinate in the parent. If the child is the root node, the above_root + znode is returned but the coord is not set. This function may cause atom + fusion, but it is only used for read locks (at this point) and therefore + fusion only occurs when the parent is already dirty. */ +/* Hans adds this note: remember to ask how expensive this operation is vs. + storing parent pointer in jnodes. */ +static int +jnode_lock_parent_coord(jnode * node, + coord_t *coord, + lock_handle * parent_lh, + load_count * parent_zh, + znode_lock_mode parent_mode, int try) +{ + int ret; + + assert("edward-53", jnode_is_unformatted(node) || jnode_is_znode(node)); + assert("edward-54", jnode_is_unformatted(node) + || znode_is_any_locked(JZNODE(node))); + + if (!jnode_is_znode(node)) { + reiser4_key key; + tree_level stop_level = TWIG_LEVEL; + lookup_bias bias = FIND_EXACT; + + assert("edward-168", !(jnode_get_type(node) == JNODE_BITMAP)); + + /* The case when node is not znode, but can have parent coord + (unformatted node, node which represents cluster page, + etc..). Generate a key for the appropriate entry, search + in the tree using coord_by_key, which handles locking for + us. */ + + /* + * nothing is locked at this moment, so, nothing prevents + * concurrent truncate from removing jnode from inode. To + * prevent this spin-lock jnode. jnode can be truncated just + * after call to the jnode_build_key(), but this is ok, + * because coord_by_key() will just fail to find appropriate + * extent. + */ + spin_lock_jnode(node); + if (!JF_ISSET(node, JNODE_HEARD_BANSHEE)) { + jnode_build_key(node, &key); + ret = 0; + } else + ret = RETERR(-ENOENT); + spin_unlock_jnode(node); + + if (ret != 0) + return ret; + + if (jnode_is_cluster_page(node)) + stop_level = LEAF_LEVEL; + + assert("jmacd-1812", coord != NULL); + + ret = coord_by_key(jnode_get_tree(node), &key, coord, parent_lh, + parent_mode, bias, stop_level, stop_level, + CBK_UNIQUE, NULL/*ra_info */); + switch (ret) { + case CBK_COORD_NOTFOUND: + assert("edward-1038", + ergo(jnode_is_cluster_page(node), + JF_ISSET(node, JNODE_HEARD_BANSHEE))); + if (!JF_ISSET(node, JNODE_HEARD_BANSHEE)) + warning("nikita-3177", "Parent not found"); + return ret; + case CBK_COORD_FOUND: + if (coord->between != AT_UNIT) { + /* FIXME: comment needed */ + done_lh(parent_lh); + if (!JF_ISSET(node, JNODE_HEARD_BANSHEE)) { + warning("nikita-3178", + "Found but not happy: %i", + coord->between); + } + return RETERR(-ENOENT); + } + ret = incr_load_count_znode(parent_zh, parent_lh->node); + if (ret != 0) + return ret; + /* if (jnode_is_cluster_page(node)) { + races with write() are possible + check_child_cluster (parent_lh->node); + } + */ + break; + default: + return ret; + } + + } else { + int flags; + znode *z; + + z = JZNODE(node); + /* Formatted node case: */ + assert("jmacd-2061", !znode_is_root(z)); + + flags = GN_ALLOW_NOT_CONNECTED; + if (try) + flags |= GN_TRY_LOCK; + + ret = + reiser4_get_parent_flags(parent_lh, z, parent_mode, flags); + if (ret != 0) + /* -E_REPEAT is ok here, it is handled by the caller. */ + return ret; + + /* Make the child's position "hint" up-to-date. (Unless above + root, which caller must check.) */ + if (coord != NULL) { + + ret = incr_load_count_znode(parent_zh, parent_lh->node); + if (ret != 0) { + warning("jmacd-976812386", + "incr_load_count_znode failed: %d", + ret); + return ret; + } + + ret = find_child_ptr(parent_lh->node, z, coord); + if (ret != 0) { + warning("jmacd-976812", + "find_child_ptr failed: %d", ret); + return ret; + } + } + } + + return 0; +} + +/* Get the (locked) next neighbor of a znode which is dirty and a member of the + same atom. If there is no next neighbor or the neighbor is not in memory or + if there is a neighbor but it is not dirty or not in the same atom, + -E_NO_NEIGHBOR is returned. In some cases the slum may include nodes which + are not dirty, if so @check_dirty should be 0 */ +static int neighbor_in_slum(znode * node, /* starting point */ + lock_handle * lock, /* lock on starting point */ + sideof side, /* left or right direction we + seek the next node in */ + znode_lock_mode mode, /* kind of lock we want */ + int check_dirty, /* true if the neighbor should + be dirty */ + int use_upper_levels /* get neighbor by going though + upper levels */) +{ + int ret; + int flags; + + assert("jmacd-6334", znode_is_connected(node)); + + flags = GN_SAME_ATOM | (side == LEFT_SIDE ? GN_GO_LEFT : 0); + if (use_upper_levels) + flags |= GN_CAN_USE_UPPER_LEVELS; + + ret = reiser4_get_neighbor(lock, node, mode, flags); + if (ret) { + /* May return -ENOENT or -E_NO_NEIGHBOR. */ + /* FIXME(C): check EINVAL, E_DEADLOCK */ + if (ret == -ENOENT) + ret = RETERR(-E_NO_NEIGHBOR); + return ret; + } + if (!check_dirty) + return 0; + /* Check dirty bit of locked znode, no races here */ + if (JF_ISSET(ZJNODE(lock->node), JNODE_DIRTY)) + return 0; + + done_lh(lock); + return RETERR(-E_NO_NEIGHBOR); +} + +/* Return true if two znodes have the same parent. This is called with both + nodes write-locked (for squeezing) so no tree lock is needed. */ +static int znode_same_parents(znode * a, znode * b) +{ + int result; + + assert("jmacd-7011", znode_is_write_locked(a)); + assert("jmacd-7012", znode_is_write_locked(b)); + + /* We lock the whole tree for this check.... I really don't like whole + * tree locks... -Hans */ + read_lock_tree(znode_get_tree(a)); + result = (znode_parent(a) == znode_parent(b)); + read_unlock_tree(znode_get_tree(a)); + return result; +} + +/* FLUSH SCAN */ + +/* Initialize the flush_scan data structure. */ +static void scan_init(flush_scan * scan) +{ + memset(scan, 0, sizeof(*scan)); + init_lh(&scan->node_lock); + init_lh(&scan->parent_lock); + init_load_count(&scan->parent_load); + init_load_count(&scan->node_load); + coord_init_invalid(&scan->parent_coord, NULL); +} + +/* Release any resources held by the flush scan, e.g. release locks, + free memory, etc. */ +static void scan_done(flush_scan * scan) +{ + done_load_count(&scan->node_load); + if (scan->node != NULL) { + jput(scan->node); + scan->node = NULL; + } + done_load_count(&scan->parent_load); + done_lh(&scan->parent_lock); + done_lh(&scan->node_lock); +} + +/* Returns true if flush scanning is finished. */ +int reiser4_scan_finished(flush_scan * scan) +{ + return scan->stop || (scan->direction == RIGHT_SIDE && + scan->count >= scan->max_count); +} + +/* Return true if the scan should continue to the @tonode. True if the node + meets the same_slum_check condition. If not, deref the "left" node and stop + the scan. */ +int reiser4_scan_goto(flush_scan * scan, jnode * tonode) +{ + int go = same_slum_check(scan->node, tonode, 1, 0); + + if (!go) { + scan->stop = 1; + jput(tonode); + } + + return go; +} + +/* Set the current scan->node, refcount it, increment count by the @add_count + (number to count, e.g., skipped unallocated nodes), deref previous current, + and copy the current parent coordinate. */ +int +scan_set_current(flush_scan * scan, jnode * node, unsigned add_count, + const coord_t *parent) +{ + /* Release the old references, take the new reference. */ + done_load_count(&scan->node_load); + + if (scan->node != NULL) + jput(scan->node); + scan->node = node; + scan->count += add_count; + + /* This next stmt is somewhat inefficient. The reiser4_scan_extent() + code could delay this update step until it finishes and update the + parent_coord only once. It did that before, but there was a bug and + this was the easiest way to make it correct. */ + if (parent != NULL) + coord_dup(&scan->parent_coord, parent); + + /* Failure may happen at the incr_load_count call, but the caller can + assume the reference is safely taken. */ + return incr_load_count_jnode(&scan->node_load, node); +} + +/* Return true if scanning in the leftward direction. */ +int reiser4_scanning_left(flush_scan * scan) +{ + return scan->direction == LEFT_SIDE; +} + +/* Performs leftward scanning starting from either kind of node. Counts the + starting node. The right-scan object is passed in for the left-scan in order + to copy the parent of an unformatted starting position. This way we avoid + searching for the unformatted node's parent when scanning in each direction. + If we search for the parent once it is set in both scan objects. The limit + parameter tells flush-scan when to stop. + + Rapid scanning is used only during scan_left, where we are interested in + finding the 'leftpoint' where we begin flushing. We are interested in + stopping at the left child of a twig that does not have a dirty left + neighbour. THIS IS A SPECIAL CASE. The problem is finding a way to flush only + those nodes without unallocated children, and it is difficult to solve in the + bottom-up flushing algorithm we are currently using. The problem can be + solved by scanning left at every level as we go upward, but this would + basically bring us back to using a top-down allocation strategy, which we + already tried (see BK history from May 2002), and has a different set of + problems. The top-down strategy makes avoiding unallocated children easier, + but makes it difficult to propertly flush dirty children with clean parents + that would otherwise stop the top-down flush, only later to dirty the parent + once the children are flushed. So we solve the problem in the bottom-up + algorithm with a special case for twigs and leaves only. + + The first step in solving the problem is this rapid leftward scan. After we + determine that there are at least enough nodes counted to qualify for + FLUSH_RELOCATE_THRESHOLD we are no longer interested in the exact count, we + are only interested in finding the best place to start the flush. + + We could choose one of two possibilities: + + 1. Stop at the leftmost child (of a twig) that does not have a dirty left + neighbor. This requires checking one leaf per rapid-scan twig + + 2. Stop at the leftmost child (of a twig) where there are no dirty children + of the twig to the left. This requires checking possibly all of the in-memory + children of each twig during the rapid scan. + + For now we implement the first policy. +*/ +static int +scan_left(flush_scan * scan, flush_scan * right, jnode * node, unsigned limit) +{ + int ret = 0; + + scan->max_count = limit; + scan->direction = LEFT_SIDE; + + ret = scan_set_current(scan, jref(node), 1, NULL); + if (ret != 0) + return ret; + + ret = scan_common(scan, right); + if (ret != 0) + return ret; + + /* Before rapid scanning, we need a lock on scan->node so that we can + get its parent, only if formatted. */ + if (jnode_is_znode(scan->node)) { + ret = longterm_lock_znode(&scan->node_lock, JZNODE(scan->node), + ZNODE_WRITE_LOCK, ZNODE_LOCK_LOPRI); + } + + /* Rapid_scan would go here (with limit set to FLUSH_RELOCATE_THRESHOLD) + */ + return ret; +} + +/* Performs rightward scanning... Does not count the starting node. The limit + parameter is described in scan_left. If the starting node is unformatted then + the parent_coord was already set during scan_left. The rapid_after parameter + is not used during right-scanning. + + scan_right is only called if the scan_left operation does not count at least + FLUSH_RELOCATE_THRESHOLD nodes for flushing. Otherwise, the limit parameter + is set to the difference between scan-left's count and + FLUSH_RELOCATE_THRESHOLD, meaning scan-right counts as high as + FLUSH_RELOCATE_THRESHOLD and then stops. */ +static int scan_right(flush_scan * scan, jnode * node, unsigned limit) +{ + int ret; + + scan->max_count = limit; + scan->direction = RIGHT_SIDE; + + ret = scan_set_current(scan, jref(node), 0, NULL); + if (ret != 0) + return ret; + + return scan_common(scan, NULL); +} + +/* Common code to perform left or right scanning. */ +static int scan_common(flush_scan * scan, flush_scan * other) +{ + int ret; + + assert("nikita-2376", scan->node != NULL); + assert("edward-54", jnode_is_unformatted(scan->node) + || jnode_is_znode(scan->node)); + + /* Special case for starting at an unformatted node. Optimization: we + only want to search for the parent (which requires a tree traversal) + once. Obviously, we shouldn't have to call it once for the left scan + and once for the right scan. For this reason, if we search for the + parent during scan-left we then duplicate the coord/lock/load into + the scan-right object. */ + if (jnode_is_unformatted(scan->node)) { + ret = scan_unformatted(scan, other); + if (ret != 0) + return ret; + } + /* This loop expects to start at a formatted position and performs + chaining of formatted regions */ + while (!reiser4_scan_finished(scan)) { + + ret = scan_formatted(scan); + if (ret != 0) + return ret; + } + + return 0; +} + +static int scan_unformatted(flush_scan * scan, flush_scan * other) +{ + int ret = 0; + int try = 0; + + if (!coord_is_invalid(&scan->parent_coord)) + goto scan; + + /* set parent coord from */ + if (!jnode_is_unformatted(scan->node)) { + /* formatted position */ + + lock_handle lock; + assert("edward-301", jnode_is_znode(scan->node)); + init_lh(&lock); + + /* + * when flush starts from unformatted node, first thing it + * does is tree traversal to find formatted parent of starting + * node. This parent is then kept lock across scans to the + * left and to the right. This means that during scan to the + * left we cannot take left-ward lock, because this is + * dead-lock prone. So, if we are scanning to the left and + * there is already lock held by this thread, + * jnode_lock_parent_coord() should use try-lock. + */ + try = reiser4_scanning_left(scan) + && !lock_stack_isclean(get_current_lock_stack()); + /* Need the node locked to get the parent lock, We have to + take write lock since there is at least one call path + where this znode is already write-locked by us. */ + ret = + longterm_lock_znode(&lock, JZNODE(scan->node), + ZNODE_WRITE_LOCK, + reiser4_scanning_left(scan) ? + ZNODE_LOCK_LOPRI : + ZNODE_LOCK_HIPRI); + if (ret != 0) + /* EINVAL or E_DEADLOCK here mean... try again! At this + point we've scanned too far and can't back out, just + start over. */ + return ret; + + ret = jnode_lock_parent_coord(scan->node, + &scan->parent_coord, + &scan->parent_lock, + &scan->parent_load, + ZNODE_WRITE_LOCK, try); + + /* FIXME(C): check EINVAL, E_DEADLOCK */ + done_lh(&lock); + if (ret == -E_REPEAT) { + scan->stop = 1; + return 0; + } + if (ret) + return ret; + + } else { + /* unformatted position */ + + ret = + jnode_lock_parent_coord(scan->node, &scan->parent_coord, + &scan->parent_lock, + &scan->parent_load, + ZNODE_WRITE_LOCK, try); + + if (IS_CBKERR(ret)) + return ret; + + if (ret == CBK_COORD_NOTFOUND) + /* FIXME(C): check EINVAL, E_DEADLOCK */ + return ret; + + /* parent was found */ + assert("jmacd-8661", other != NULL); + /* Duplicate the reference into the other flush_scan. */ + coord_dup(&other->parent_coord, &scan->parent_coord); + copy_lh(&other->parent_lock, &scan->parent_lock); + copy_load_count(&other->parent_load, &scan->parent_load); + } +scan: + return scan_by_coord(scan); +} + +/* Performs left- or rightward scanning starting from a formatted node. Follow + left pointers under tree lock as long as: + + - node->left/right is non-NULL + - node->left/right is connected, dirty + - node->left/right belongs to the same atom + - scan has not reached maximum count +*/ +static int scan_formatted(flush_scan * scan) +{ + int ret; + znode *neighbor = NULL; + + assert("jmacd-1401", !reiser4_scan_finished(scan)); + + do { + znode *node = JZNODE(scan->node); + + /* Node should be connected, but if not stop the scan. */ + if (!znode_is_connected(node)) { + scan->stop = 1; + break; + } + + /* Lock the tree, check-for and reference the next sibling. */ + read_lock_tree(znode_get_tree(node)); + + /* It may be that a node is inserted or removed between a node + and its left sibling while the tree lock is released, but the + flush-scan count does not need to be precise. Thus, we + release the tree lock as soon as we get the neighboring node. + */ + neighbor = + reiser4_scanning_left(scan) ? node->left : node->right; + if (neighbor != NULL) + zref(neighbor); + + read_unlock_tree(znode_get_tree(node)); + + /* If neighbor is NULL at the leaf level, need to check for an + unformatted sibling using the parent--break in any case. */ + if (neighbor == NULL) + break; + + /* Check the condition for going left, break if it is not met. + This also releases (jputs) the neighbor if false. */ + if (!reiser4_scan_goto(scan, ZJNODE(neighbor))) + break; + + /* Advance the flush_scan state to the left, repeat. */ + ret = scan_set_current(scan, ZJNODE(neighbor), 1, NULL); + if (ret != 0) + return ret; + + } while (!reiser4_scan_finished(scan)); + + /* If neighbor is NULL then we reached the end of a formatted region, or + else the sibling is out of memory, now check for an extent to the + left (as long as LEAF_LEVEL). */ + if (neighbor != NULL || jnode_get_level(scan->node) != LEAF_LEVEL + || reiser4_scan_finished(scan)) { + scan->stop = 1; + return 0; + } + /* Otherwise, calls scan_by_coord for the right(left)most item of the + left(right) neighbor on the parent level, then possibly continue. */ + + coord_init_invalid(&scan->parent_coord, NULL); + return scan_unformatted(scan, NULL); +} + +/* NOTE-EDWARD: + This scans adjacent items of the same type and calls scan flush plugin for + each one. Performs left(right)ward scanning starting from a (possibly) + unformatted node. If we start from unformatted node, then we continue only if + the next neighbor is also unformatted. When called from scan_formatted, we + skip first iteration (to make sure that right(left)most item of the + left(right) neighbor on the parent level is of the same type and set + appropriate coord). */ +static int scan_by_coord(flush_scan * scan) +{ + int ret = 0; + int scan_this_coord; + lock_handle next_lock; + load_count next_load; + coord_t next_coord; + jnode *child; + item_plugin *iplug; + + init_lh(&next_lock); + init_load_count(&next_load); + scan_this_coord = (jnode_is_unformatted(scan->node) ? 1 : 0); + + /* set initial item id */ + iplug = item_plugin_by_coord(&scan->parent_coord); + + for (; !reiser4_scan_finished(scan); scan_this_coord = 1) { + if (scan_this_coord) { + /* Here we expect that unit is scannable. it would not + * be so due to race with extent->tail conversion. */ + if (iplug->f.scan == NULL) { + scan->stop = 1; + ret = -E_REPEAT; + /* skip the check at the end. */ + goto race; + } + + ret = iplug->f.scan(scan); + if (ret != 0) + goto exit; + + if (reiser4_scan_finished(scan)) { + checkchild(scan); + break; + } + } else { + /* the same race against truncate as above is possible + * here, it seems */ + + /* NOTE-JMACD: In this case, apply the same end-of-node + logic but don't scan the first coordinate. */ + assert("jmacd-1231", + item_is_internal(&scan->parent_coord)); + } + + if (iplug->f.utmost_child == NULL + || znode_get_level(scan->parent_coord.node) != TWIG_LEVEL) { + /* stop this coord and continue on parrent level */ + ret = + scan_set_current(scan, + ZJNODE(zref + (scan->parent_coord.node)), + 1, NULL); + if (ret != 0) + goto exit; + break; + } + + /* Either way, the invariant is that scan->parent_coord is set + to the parent of scan->node. Now get the next unit. */ + coord_dup(&next_coord, &scan->parent_coord); + coord_sideof_unit(&next_coord, scan->direction); + + /* If off-the-end of the twig, try the next twig. */ + if (coord_is_after_sideof_unit(&next_coord, scan->direction)) { + /* We take the write lock because we may start flushing + * from this coordinate. */ + ret = neighbor_in_slum(next_coord.node, + &next_lock, + scan->direction, + ZNODE_WRITE_LOCK, + 1 /* check dirty */, + 0 /* don't go though upper + levels */); + if (ret == -E_NO_NEIGHBOR) { + scan->stop = 1; + ret = 0; + break; + } + + if (ret != 0) + goto exit; + + ret = incr_load_count_znode(&next_load, next_lock.node); + if (ret != 0) + goto exit; + + coord_init_sideof_unit(&next_coord, next_lock.node, + sideof_reverse(scan->direction)); + } + + iplug = item_plugin_by_coord(&next_coord); + + /* Get the next child. */ + ret = + iplug->f.utmost_child(&next_coord, + sideof_reverse(scan->direction), + &child); + if (ret != 0) + goto exit; + /* If the next child is not in memory, or, item_utmost_child + failed (due to race with unlink, most probably), stop + here. */ + if (child == NULL || IS_ERR(child)) { + scan->stop = 1; + checkchild(scan); + break; + } + + assert("nikita-2374", jnode_is_unformatted(child) + || jnode_is_znode(child)); + + /* See if it is dirty, part of the same atom. */ + if (!reiser4_scan_goto(scan, child)) { + checkchild(scan); + break; + } + + /* If so, make this child current. */ + ret = scan_set_current(scan, child, 1, &next_coord); + if (ret != 0) + goto exit; + + /* Now continue. If formatted we release the parent lock and + return, then proceed. */ + if (jnode_is_znode(child)) + break; + + /* Otherwise, repeat the above loop with next_coord. */ + if (next_load.node != NULL) { + done_lh(&scan->parent_lock); + move_lh(&scan->parent_lock, &next_lock); + move_load_count(&scan->parent_load, &next_load); + } + } + + assert("jmacd-6233", + reiser4_scan_finished(scan) || jnode_is_znode(scan->node)); +exit: + checkchild(scan); +race: /* skip the above check */ + if (jnode_is_znode(scan->node)) { + done_lh(&scan->parent_lock); + done_load_count(&scan->parent_load); + } + + done_load_count(&next_load); + done_lh(&next_lock); + return ret; +} + +/* FLUSH POS HELPERS */ + +/* Initialize the fields of a flush_position. */ +static void pos_init(flush_pos_t *pos) +{ + memset(pos, 0, sizeof *pos); + + pos->state = POS_INVALID; + coord_init_invalid(&pos->coord, NULL); + init_lh(&pos->lock); + init_load_count(&pos->load); + + reiser4_blocknr_hint_init(&pos->preceder); +} + +/* The flush loop inside squalloc periodically checks pos_valid to determine + when "enough flushing" has been performed. This will return true until one + of the following conditions is met: + + 1. the number of flush-queued nodes has reached the kernel-supplied + "int *nr_to_flush" parameter, meaning we have flushed as many blocks as the + kernel requested. When flushing to commit, this parameter is NULL. + + 2. pos_stop() is called because squalloc discovers that the "next" node in + the flush order is either non-existant, not dirty, or not in the same atom. +*/ + +static int pos_valid(flush_pos_t *pos) +{ + return pos->state != POS_INVALID; +} + +/* Release any resources of a flush_position. Called when jnode_flush + finishes. */ +static void pos_done(flush_pos_t *pos) +{ + pos_stop(pos); + reiser4_blocknr_hint_done(&pos->preceder); + if (convert_data(pos)) + free_convert_data(pos); +} + +/* Reset the point and parent. Called during flush subroutines to terminate the + squalloc loop. */ +static int pos_stop(flush_pos_t *pos) +{ + pos->state = POS_INVALID; + done_lh(&pos->lock); + done_load_count(&pos->load); + coord_init_invalid(&pos->coord, NULL); + + if (pos->child) { + jput(pos->child); + pos->child = NULL; + } + + return 0; +} + +/* Return the flush_position's block allocator hint. */ +reiser4_blocknr_hint *reiser4_pos_hint(flush_pos_t *pos) +{ + return &pos->preceder; +} + +flush_queue_t *reiser4_pos_fq(flush_pos_t *pos) +{ + return pos->fq; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 90 + LocalWords: preceder + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/flush.h linux-2.6.30/fs/reiser4/flush.h --- linux-2.6.30.orig/fs/reiser4/flush.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/flush.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,300 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* DECLARATIONS: */ + +#if !defined(__REISER4_FLUSH_H__) +#define __REISER4_FLUSH_H__ + +#include "plugin/cluster.h" + +/* The flush_scan data structure maintains the state of an in-progress + flush-scan on a single level of the tree. A flush-scan is used for counting + the number of adjacent nodes to flush, which is used to determine whether we + should relocate, and it is also used to find a starting point for flush. A + flush-scan object can scan in both right and left directions via the + scan_left() and scan_right() interfaces. The right- and left-variations are + similar but perform different functions. When scanning left we (optionally + perform rapid scanning and then) longterm-lock the endpoint node. When + scanning right we are simply counting the number of adjacent, dirty nodes. */ +struct flush_scan { + + /* The current number of nodes scanned on this level. */ + unsigned count; + + /* There may be a maximum number of nodes for a scan on any single + level. When going leftward, max_count is determined by + FLUSH_SCAN_MAXNODES (see reiser4.h) */ + unsigned max_count; + + /* Direction: Set to one of the sideof enumeration: + { LEFT_SIDE, RIGHT_SIDE }. */ + sideof direction; + + /* Initially @stop is set to false then set true once some condition + stops the search (e.g., we found a clean node before reaching + max_count or we found a node belonging to another atom). */ + int stop; + + /* The current scan position. If @node is non-NULL then its reference + count has been incremented to reflect this reference. */ + jnode *node; + + /* A handle for zload/zrelse of current scan position node. */ + load_count node_load; + + /* During left-scan, if the final position (a.k.a. endpoint node) is + formatted the node is locked using this lock handle. The endpoint + needs to be locked for transfer to the flush_position object after + scanning finishes. */ + lock_handle node_lock; + + /* When the position is unformatted, its parent, coordinate, and parent + zload/zrelse handle. */ + lock_handle parent_lock; + coord_t parent_coord; + load_count parent_load; + + /* The block allocator preceder hint. Sometimes flush_scan determines + what the preceder is and if so it sets it here, after which it is + copied into the flush_position. Otherwise, the preceder is computed + later. */ + reiser4_block_nr preceder_blk; +}; + +struct convert_item_info { + dc_item_stat d_cur; /* disk cluster state of the current item */ + dc_item_stat d_next; /* disk cluster state of the next slum item */ + int cluster_shift; /* disk cluster shift */ + flow_t flow; /* disk cluster data */ +}; + +struct convert_info { + int count; /* for squalloc terminating */ + item_plugin *iplug; /* current item plugin */ + struct convert_item_info *itm; /* current item info */ + struct cluster_handle clust; /* transform cluster */ +}; + +typedef enum flush_position_state { + POS_INVALID, /* Invalid or stopped pos, do not continue slum + * processing */ + POS_ON_LEAF, /* pos points to already prepped, locked + * formatted node at leaf level */ + POS_ON_EPOINT, /* pos keeps a lock on twig level, "coord" field + * is used to traverse unformatted nodes */ + POS_TO_LEAF, /* pos is being moved to leaf level */ + POS_TO_TWIG, /* pos is being moved to twig level */ + POS_END_OF_TWIG, /* special case of POS_ON_TWIG, when coord is + * after rightmost unit of the current twig */ + POS_ON_INTERNAL /* same as POS_ON_LEAF, but points to internal + * node */ +} flushpos_state_t; + +/* An encapsulation of the current flush point and all the parameters that are + passed through the entire squeeze-and-allocate stage of the flush routine. + A single flush_position object is constructed after left- and right-scanning + finishes. */ +struct flush_position { + flushpos_state_t state; + + coord_t coord; /* coord to traverse unformatted nodes */ + lock_handle lock; /* current lock we hold */ + load_count load; /* load status for current locked formatted node + */ + jnode *child; /* for passing a reference to unformatted child + * across pos state changes */ + + reiser4_blocknr_hint preceder; /* The flush 'hint' state. */ + int leaf_relocate; /* True if enough leaf-level nodes were + * found to suggest a relocate policy. */ + int alloc_cnt; /* The number of nodes allocated during squeeze + and allococate. */ + int prep_or_free_cnt; /* The number of nodes prepared for write + (allocate) or squeezed and freed. */ + flush_queue_t *fq; + long *nr_written; /* number of nodes submitted to disk */ + int flags; /* a copy of jnode_flush flags argument */ + + znode *prev_twig; /* previous parent pointer value, used to catch + * processing of new twig node */ + struct convert_info *sq; /* convert info */ + + unsigned long pos_in_unit; /* for extents only. Position + within an extent unit of first + jnode of slum */ + long nr_to_write; /* number of unformatted nodes to handle on + flush */ +}; + +static inline int item_convert_count(flush_pos_t *pos) +{ + return pos->sq->count; +} +static inline void inc_item_convert_count(flush_pos_t *pos) +{ + pos->sq->count++; +} +static inline void set_item_convert_count(flush_pos_t *pos, int count) +{ + pos->sq->count = count; +} +static inline item_plugin *item_convert_plug(flush_pos_t *pos) +{ + return pos->sq->iplug; +} + +static inline struct convert_info *convert_data(flush_pos_t *pos) +{ + return pos->sq; +} + +static inline struct convert_item_info *item_convert_data(flush_pos_t *pos) +{ + assert("edward-955", convert_data(pos)); + return pos->sq->itm; +} + +static inline struct tfm_cluster *tfm_cluster_sq(flush_pos_t *pos) +{ + return &pos->sq->clust.tc; +} + +static inline struct tfm_stream *tfm_stream_sq(flush_pos_t *pos, + tfm_stream_id id) +{ + assert("edward-854", pos->sq != NULL); + return get_tfm_stream(tfm_cluster_sq(pos), id); +} + +static inline int chaining_data_present(flush_pos_t *pos) +{ + return convert_data(pos) && item_convert_data(pos); +} + +/* Returns true if next node contains next item of the disk cluster + so item convert data should be moved to the right slum neighbor. +*/ +static inline int should_chain_next_node(flush_pos_t *pos) +{ + int result = 0; + + assert("edward-1007", chaining_data_present(pos)); + + switch (item_convert_data(pos)->d_next) { + case DC_CHAINED_ITEM: + result = 1; + break; + case DC_AFTER_CLUSTER: + break; + default: + impossible("edward-1009", "bad state of next slum item"); + } + return result; +} + +/* update item state in a disk cluster to assign conversion mode */ +static inline void +move_chaining_data(flush_pos_t *pos, int this_node/* where is next item */) +{ + + assert("edward-1010", chaining_data_present(pos)); + + if (this_node == 0) { + /* next item is on the right neighbor */ + assert("edward-1011", + item_convert_data(pos)->d_cur == DC_FIRST_ITEM || + item_convert_data(pos)->d_cur == DC_CHAINED_ITEM); + assert("edward-1012", + item_convert_data(pos)->d_next == DC_CHAINED_ITEM); + + item_convert_data(pos)->d_cur = DC_CHAINED_ITEM; + item_convert_data(pos)->d_next = DC_INVALID_STATE; + } else { + /* next item is on the same node */ + assert("edward-1013", + item_convert_data(pos)->d_cur == DC_FIRST_ITEM || + item_convert_data(pos)->d_cur == DC_CHAINED_ITEM); + assert("edward-1227", + item_convert_data(pos)->d_next == DC_AFTER_CLUSTER || + item_convert_data(pos)->d_next == DC_INVALID_STATE); + + item_convert_data(pos)->d_cur = DC_AFTER_CLUSTER; + item_convert_data(pos)->d_next = DC_INVALID_STATE; + } +} + +static inline int should_convert_node(flush_pos_t *pos, znode * node) +{ + return znode_convertible(node); +} + +/* true if there is attached convert item info */ +static inline int should_convert_next_node(flush_pos_t *pos) +{ + return convert_data(pos) && item_convert_data(pos); +} + +#define SQUALLOC_THRESHOLD 256 + +static inline int should_terminate_squalloc(flush_pos_t *pos) +{ + return convert_data(pos) && + !item_convert_data(pos) && + item_convert_count(pos) >= SQUALLOC_THRESHOLD; +} + +#if 1 +#define check_convert_info(pos) \ +do { \ + if (unlikely(should_convert_next_node(pos))) { \ + warning("edward-1006", "unprocessed chained data"); \ + printk("d_cur = %d, d_next = %d, flow.len = %llu\n", \ + item_convert_data(pos)->d_cur, \ + item_convert_data(pos)->d_next, \ + item_convert_data(pos)->flow.length); \ + } \ +} while (0) +#else +#define check_convert_info(pos) +#endif /* REISER4_DEBUG */ + +void free_convert_data(flush_pos_t *pos); +/* used in extent.c */ +int scan_set_current(flush_scan * scan, jnode * node, unsigned add_size, + const coord_t *parent); +int reiser4_scan_finished(flush_scan * scan); +int reiser4_scanning_left(flush_scan * scan); +int reiser4_scan_goto(flush_scan * scan, jnode * tonode); +txn_atom *atom_locked_by_fq(flush_queue_t *fq); +int reiser4_alloc_extent(flush_pos_t *flush_pos); +squeeze_result squalloc_extent(znode *left, const coord_t *, flush_pos_t *, + reiser4_key *stop_key); +extern int reiser4_init_fqs(void); +extern void reiser4_done_fqs(void); + +#if REISER4_DEBUG + +extern void reiser4_check_fq(const txn_atom *atom); +extern atomic_t flush_cnt; + +#define check_preceder(blk) \ +assert("nikita-2588", blk < reiser4_block_count(reiser4_get_current_sb())); +extern void check_pos(flush_pos_t *pos); +#else +#define check_preceder(b) noop +#define check_pos(pos) noop +#endif + +/* __REISER4_FLUSH_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 90 + LocalWords: preceder + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/flush_queue.c linux-2.6.30/fs/reiser4/flush_queue.c --- linux-2.6.30.orig/fs/reiser4/flush_queue.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/flush_queue.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,678 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +#include "debug.h" +#include "super.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "page_cache.h" +#include "wander.h" +#include "vfs_ops.h" +#include "writeout.h" +#include "flush.h" + +#include +#include +#include +#include +#include + +/* A flush queue object is an accumulator for keeping jnodes prepared + by the jnode_flush() function for writing to disk. Those "queued" jnodes are + kept on the flush queue until memory pressure or atom commit asks + flush queues to write some or all from their jnodes. */ + +/* + LOCKING: + + fq->guard spin lock protects fq->atom pointer and nothing else. fq->prepped + list protected by atom spin lock. fq->prepped list uses the following + locking: + + two ways to protect fq->prepped list for read-only list traversal: + + 1. atom spin-lock atom. + 2. fq is IN_USE, atom->nr_running_queues increased. + + and one for list modification: + + 1. atom is spin-locked and one condition is true: fq is IN_USE or + atom->nr_running_queues == 0. + + The deadlock-safe order for flush queues and atoms is: first lock atom, then + lock flush queue, then lock jnode. +*/ + +#define fq_in_use(fq) ((fq)->state & FQ_IN_USE) +#define fq_ready(fq) (!fq_in_use(fq)) + +#define mark_fq_in_use(fq) do { (fq)->state |= FQ_IN_USE; } while (0) +#define mark_fq_ready(fq) do { (fq)->state &= ~FQ_IN_USE; } while (0) + +/* get lock on atom from locked flush queue object */ +static txn_atom *atom_locked_by_fq_nolock(flush_queue_t *fq) +{ + /* This code is similar to jnode_get_atom(), look at it for the + * explanation. */ + txn_atom *atom; + + assert_spin_locked(&(fq->guard)); + + while (1) { + atom = fq->atom; + if (atom == NULL) + break; + + if (spin_trylock_atom(atom)) + break; + + atomic_inc(&atom->refcount); + spin_unlock(&(fq->guard)); + spin_lock_atom(atom); + spin_lock(&(fq->guard)); + + if (fq->atom == atom) { + atomic_dec(&atom->refcount); + break; + } + + spin_unlock(&(fq->guard)); + atom_dec_and_unlock(atom); + spin_lock(&(fq->guard)); + } + + return atom; +} + +txn_atom *atom_locked_by_fq(flush_queue_t *fq) +{ + txn_atom *atom; + + spin_lock(&(fq->guard)); + atom = atom_locked_by_fq_nolock(fq); + spin_unlock(&(fq->guard)); + return atom; +} + +static void init_fq(flush_queue_t *fq) +{ + memset(fq, 0, sizeof *fq); + + atomic_set(&fq->nr_submitted, 0); + + INIT_LIST_HEAD(ATOM_FQ_LIST(fq)); + + init_waitqueue_head(&fq->wait); + spin_lock_init(&fq->guard); +} + +/* slab for flush queues */ +static struct kmem_cache *fq_slab; + +/** + * reiser4_init_fqs - create flush queue cache + * + * Initializes slab cache of flush queues. It is part of reiser4 module + * initialization. + */ +int reiser4_init_fqs(void) +{ + fq_slab = kmem_cache_create("fq", + sizeof(flush_queue_t), + 0, SLAB_HWCACHE_ALIGN, NULL); + if (fq_slab == NULL) + return RETERR(-ENOMEM); + return 0; +} + +/** + * reiser4_done_fqs - delete flush queue cache + * + * This is called on reiser4 module unloading or system shutdown. + */ +void reiser4_done_fqs(void) +{ + destroy_reiser4_cache(&fq_slab); +} + +/* create new flush queue object */ +static flush_queue_t *create_fq(gfp_t gfp) +{ + flush_queue_t *fq; + + fq = kmem_cache_alloc(fq_slab, gfp); + if (fq) + init_fq(fq); + + return fq; +} + +/* adjust atom's and flush queue's counters of queued nodes */ +static void count_enqueued_node(flush_queue_t *fq) +{ + ON_DEBUG(fq->atom->num_queued++); +} + +static void count_dequeued_node(flush_queue_t *fq) +{ + assert("zam-993", fq->atom->num_queued > 0); + ON_DEBUG(fq->atom->num_queued--); +} + +/* attach flush queue object to the atom */ +static void attach_fq(txn_atom *atom, flush_queue_t *fq) +{ + assert_spin_locked(&(atom->alock)); + list_add(&fq->alink, &atom->flush_queues); + fq->atom = atom; + ON_DEBUG(atom->nr_flush_queues++); +} + +static void detach_fq(flush_queue_t *fq) +{ + assert_spin_locked(&(fq->atom->alock)); + + spin_lock(&(fq->guard)); + list_del_init(&fq->alink); + assert("vs-1456", fq->atom->nr_flush_queues > 0); + ON_DEBUG(fq->atom->nr_flush_queues--); + fq->atom = NULL; + spin_unlock(&(fq->guard)); +} + +/* destroy flush queue object */ +static void done_fq(flush_queue_t *fq) +{ + assert("zam-763", list_empty_careful(ATOM_FQ_LIST(fq))); + assert("zam-766", atomic_read(&fq->nr_submitted) == 0); + + kmem_cache_free(fq_slab, fq); +} + +/* */ +static void mark_jnode_queued(flush_queue_t *fq, jnode * node) +{ + JF_SET(node, JNODE_FLUSH_QUEUED); + count_enqueued_node(fq); +} + +/* Putting jnode into the flush queue. Both atom and jnode should be + spin-locked. */ +void queue_jnode(flush_queue_t *fq, jnode * node) +{ + assert_spin_locked(&(node->guard)); + assert("zam-713", node->atom != NULL); + assert_spin_locked(&(node->atom->alock)); + assert("zam-716", fq->atom != NULL); + assert("zam-717", fq->atom == node->atom); + assert("zam-907", fq_in_use(fq)); + + assert("zam-714", JF_ISSET(node, JNODE_DIRTY)); + assert("zam-826", JF_ISSET(node, JNODE_RELOC)); + assert("vs-1481", !JF_ISSET(node, JNODE_FLUSH_QUEUED)); + assert("vs-1481", NODE_LIST(node) != FQ_LIST); + + mark_jnode_queued(fq, node); + list_move_tail(&node->capture_link, ATOM_FQ_LIST(fq)); + + ON_DEBUG(count_jnode(node->atom, node, NODE_LIST(node), + FQ_LIST, 1)); +} + +/* repeatable process for waiting io completion on a flush queue object */ +static int wait_io(flush_queue_t *fq, int *nr_io_errors) +{ + assert("zam-738", fq->atom != NULL); + assert_spin_locked(&(fq->atom->alock)); + assert("zam-736", fq_in_use(fq)); + assert("zam-911", list_empty_careful(ATOM_FQ_LIST(fq))); + + if (atomic_read(&fq->nr_submitted) != 0) { + struct super_block *super; + + spin_unlock_atom(fq->atom); + + assert("nikita-3013", reiser4_schedulable()); + + super = reiser4_get_current_sb(); + + /* FIXME: this is instead of blk_run_queues() */ + blk_run_address_space(reiser4_get_super_fake(super)->i_mapping); + + if (!(super->s_flags & MS_RDONLY)) + wait_event(fq->wait, + atomic_read(&fq->nr_submitted) == 0); + + /* Ask the caller to re-acquire the locks and call this + function again. Note: this technique is commonly used in + the txnmgr code. */ + return -E_REPEAT; + } + + *nr_io_errors += atomic_read(&fq->nr_errors); + return 0; +} + +/* wait on I/O completion, re-submit dirty nodes to write */ +static int finish_fq(flush_queue_t *fq, int *nr_io_errors) +{ + int ret; + txn_atom *atom = fq->atom; + + assert("zam-801", atom != NULL); + assert_spin_locked(&(atom->alock)); + assert("zam-762", fq_in_use(fq)); + + ret = wait_io(fq, nr_io_errors); + if (ret) + return ret; + + detach_fq(fq); + done_fq(fq); + + reiser4_atom_send_event(atom); + + return 0; +} + +/* wait for all i/o for given atom to be completed, actually do one iteration + on that and return -E_REPEAT if there more iterations needed */ +static int finish_all_fq(txn_atom * atom, int *nr_io_errors) +{ + flush_queue_t *fq; + + assert_spin_locked(&(atom->alock)); + + if (list_empty_careful(&atom->flush_queues)) + return 0; + + list_for_each_entry(fq, &atom->flush_queues, alink) { + if (fq_ready(fq)) { + int ret; + + mark_fq_in_use(fq); + assert("vs-1247", fq->owner == NULL); + ON_DEBUG(fq->owner = current); + ret = finish_fq(fq, nr_io_errors); + + if (*nr_io_errors) + reiser4_handle_error(); + + if (ret) { + reiser4_fq_put(fq); + return ret; + } + + spin_unlock_atom(atom); + + return -E_REPEAT; + } + } + + /* All flush queues are in use; atom remains locked */ + return -EBUSY; +} + +/* wait all i/o for current atom */ +int current_atom_finish_all_fq(void) +{ + txn_atom *atom; + int nr_io_errors = 0; + int ret = 0; + + do { + while (1) { + atom = get_current_atom_locked(); + ret = finish_all_fq(atom, &nr_io_errors); + if (ret != -EBUSY) + break; + reiser4_atom_wait_event(atom); + } + } while (ret == -E_REPEAT); + + /* we do not need locked atom after this function finishes, SUCCESS or + -EBUSY are two return codes when atom remains locked after + finish_all_fq */ + if (!ret) + spin_unlock_atom(atom); + + assert_spin_not_locked(&(atom->alock)); + + if (ret) + return ret; + + if (nr_io_errors) + return RETERR(-EIO); + + return 0; +} + +/* change node->atom field for all jnode from given list */ +static void +scan_fq_and_update_atom_ref(struct list_head *list, txn_atom *atom) +{ + jnode *cur; + + list_for_each_entry(cur, list, capture_link) { + spin_lock_jnode(cur); + cur->atom = atom; + spin_unlock_jnode(cur); + } +} + +/* support for atom fusion operation */ +void reiser4_fuse_fq(txn_atom *to, txn_atom *from) +{ + flush_queue_t *fq; + + assert_spin_locked(&(to->alock)); + assert_spin_locked(&(from->alock)); + + list_for_each_entry(fq, &from->flush_queues, alink) { + scan_fq_and_update_atom_ref(ATOM_FQ_LIST(fq), to); + spin_lock(&(fq->guard)); + fq->atom = to; + spin_unlock(&(fq->guard)); + } + + list_splice_init(&from->flush_queues, to->flush_queues.prev); + +#if REISER4_DEBUG + to->num_queued += from->num_queued; + to->nr_flush_queues += from->nr_flush_queues; + from->nr_flush_queues = 0; +#endif +} + +#if REISER4_DEBUG +int atom_fq_parts_are_clean(txn_atom * atom) +{ + assert("zam-915", atom != NULL); + return list_empty_careful(&atom->flush_queues); +} +#endif +/* Bio i/o completion routine for reiser4 write operations. */ +static void +end_io_handler(struct bio *bio, int err) +{ + int i; + int nr_errors = 0; + flush_queue_t *fq; + + assert("zam-958", bio->bi_rw & WRITE); + + if (err == -EOPNOTSUPP) + set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); + + /* we expect that bio->private is set to NULL or fq object which is used + * for synchronization and error counting. */ + fq = bio->bi_private; + /* Check all elements of io_vec for correct write completion. */ + for (i = 0; i < bio->bi_vcnt; i += 1) { + struct page *pg = bio->bi_io_vec[i].bv_page; + + if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { + SetPageError(pg); + nr_errors++; + } + + { + /* jnode WRITEBACK ("write is in progress bit") is + * atomically cleared here. */ + jnode *node; + + assert("zam-736", pg != NULL); + assert("zam-736", PagePrivate(pg)); + node = jprivate(pg); + + JF_CLR(node, JNODE_WRITEBACK); + } + + end_page_writeback(pg); + page_cache_release(pg); + } + + if (fq) { + /* count i/o error in fq object */ + atomic_add(nr_errors, &fq->nr_errors); + + /* If all write requests registered in this "fq" are done we up + * the waiter. */ + if (atomic_sub_and_test(bio->bi_vcnt, &fq->nr_submitted)) + wake_up(&fq->wait); + } + + bio_put(bio); +} + +/* Count I/O requests which will be submitted by @bio in given flush queues + @fq */ +void add_fq_to_bio(flush_queue_t *fq, struct bio *bio) +{ + bio->bi_private = fq; + bio->bi_end_io = end_io_handler; + + if (fq) + atomic_add(bio->bi_vcnt, &fq->nr_submitted); +} + +/* Move all queued nodes out from @fq->prepped list. */ +static void release_prepped_list(flush_queue_t *fq) +{ + txn_atom *atom; + + assert("zam-904", fq_in_use(fq)); + atom = atom_locked_by_fq(fq); + + while (!list_empty(ATOM_FQ_LIST(fq))) { + jnode *cur; + + cur = list_entry(ATOM_FQ_LIST(fq)->next, jnode, capture_link); + list_del_init(&cur->capture_link); + + count_dequeued_node(fq); + spin_lock_jnode(cur); + assert("nikita-3154", !JF_ISSET(cur, JNODE_OVRWR)); + assert("nikita-3154", JF_ISSET(cur, JNODE_RELOC)); + assert("nikita-3154", JF_ISSET(cur, JNODE_FLUSH_QUEUED)); + JF_CLR(cur, JNODE_FLUSH_QUEUED); + + if (JF_ISSET(cur, JNODE_DIRTY)) { + list_add_tail(&cur->capture_link, + ATOM_DIRTY_LIST(atom, + jnode_get_level(cur))); + ON_DEBUG(count_jnode(atom, cur, FQ_LIST, + DIRTY_LIST, 1)); + } else { + list_add_tail(&cur->capture_link, + ATOM_CLEAN_LIST(atom)); + ON_DEBUG(count_jnode(atom, cur, FQ_LIST, + CLEAN_LIST, 1)); + } + + spin_unlock_jnode(cur); + } + + if (--atom->nr_running_queues == 0) + reiser4_atom_send_event(atom); + + spin_unlock_atom(atom); +} + +/* Submit write requests for nodes on the already filled flush queue @fq. + + @fq: flush queue object which contains jnodes we can (and will) write. + @return: number of submitted blocks (>=0) if success, otherwise -- an error + code (<0). */ +int reiser4_write_fq(flush_queue_t *fq, long *nr_submitted, int flags) +{ + int ret; + txn_atom *atom; + + while (1) { + atom = atom_locked_by_fq(fq); + assert("zam-924", atom); + /* do not write fq in parallel. */ + if (atom->nr_running_queues == 0 + || !(flags & WRITEOUT_SINGLE_STREAM)) + break; + reiser4_atom_wait_event(atom); + } + + atom->nr_running_queues++; + spin_unlock_atom(atom); + + ret = write_jnode_list(ATOM_FQ_LIST(fq), fq, nr_submitted, flags); + release_prepped_list(fq); + + return ret; +} + +/* Getting flush queue object for exclusive use by one thread. May require + several iterations which is indicated by -E_REPEAT return code. + + This function does not contain code for obtaining an atom lock because an + atom lock is obtained by different ways in different parts of reiser4, + usually it is current atom, but we need a possibility for getting fq for the + atom of given jnode. */ +static int fq_by_atom_gfp(txn_atom *atom, flush_queue_t **new_fq, gfp_t gfp) +{ + flush_queue_t *fq; + + assert_spin_locked(&(atom->alock)); + + fq = list_entry(atom->flush_queues.next, flush_queue_t, alink); + while (&atom->flush_queues != &fq->alink) { + spin_lock(&(fq->guard)); + + if (fq_ready(fq)) { + mark_fq_in_use(fq); + assert("vs-1246", fq->owner == NULL); + ON_DEBUG(fq->owner = current); + spin_unlock(&(fq->guard)); + + if (*new_fq) + done_fq(*new_fq); + + *new_fq = fq; + + return 0; + } + + spin_unlock(&(fq->guard)); + + fq = list_entry(fq->alink.next, flush_queue_t, alink); + } + + /* Use previously allocated fq object */ + if (*new_fq) { + mark_fq_in_use(*new_fq); + assert("vs-1248", (*new_fq)->owner == 0); + ON_DEBUG((*new_fq)->owner = current); + attach_fq(atom, *new_fq); + + return 0; + } + + spin_unlock_atom(atom); + + *new_fq = create_fq(gfp); + + if (*new_fq == NULL) + return RETERR(-ENOMEM); + + return RETERR(-E_REPEAT); +} + +int reiser4_fq_by_atom(txn_atom * atom, flush_queue_t **new_fq) +{ + return fq_by_atom_gfp(atom, new_fq, reiser4_ctx_gfp_mask_get()); +} + +/* A wrapper around reiser4_fq_by_atom for getting a flush queue + object for current atom, if success fq->atom remains locked. */ +flush_queue_t *get_fq_for_current_atom(void) +{ + flush_queue_t *fq = NULL; + txn_atom *atom; + int ret; + + do { + atom = get_current_atom_locked(); + ret = reiser4_fq_by_atom(atom, &fq); + } while (ret == -E_REPEAT); + + if (ret) + return ERR_PTR(ret); + return fq; +} + +/* Releasing flush queue object after exclusive use */ +void reiser4_fq_put_nolock(flush_queue_t *fq) +{ + assert("zam-747", fq->atom != NULL); + assert("zam-902", list_empty_careful(ATOM_FQ_LIST(fq))); + mark_fq_ready(fq); + assert("vs-1245", fq->owner == current); + ON_DEBUG(fq->owner = NULL); +} + +void reiser4_fq_put(flush_queue_t *fq) +{ + txn_atom *atom; + + spin_lock(&(fq->guard)); + atom = atom_locked_by_fq_nolock(fq); + + assert("zam-746", atom != NULL); + + reiser4_fq_put_nolock(fq); + reiser4_atom_send_event(atom); + + spin_unlock(&(fq->guard)); + spin_unlock_atom(atom); +} + +/* A part of atom object initialization related to the embedded flush queue + list head */ + +void init_atom_fq_parts(txn_atom *atom) +{ + INIT_LIST_HEAD(&atom->flush_queues); +} + +#if REISER4_DEBUG + +void reiser4_check_fq(const txn_atom *atom) +{ + /* check number of nodes on all atom's flush queues */ + flush_queue_t *fq; + int count; + struct list_head *pos; + + count = 0; + list_for_each_entry(fq, &atom->flush_queues, alink) { + spin_lock(&(fq->guard)); + /* calculate number of jnodes on fq' list of prepped jnodes */ + list_for_each(pos, ATOM_FQ_LIST(fq)) + count++; + spin_unlock(&(fq->guard)); + } + if (count != atom->fq) + warning("", "fq counter %d, real %d\n", atom->fq, count); + +} + +#endif + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/forward.h linux-2.6.30/fs/reiser4/forward.h --- linux-2.6.30.orig/fs/reiser4/forward.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/forward.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,256 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Forward declarations. Thank you Kernighan. */ + +#if !defined(__REISER4_FORWARD_H__) +#define __REISER4_FORWARD_H__ + +#include +#include + +typedef struct zlock zlock; +typedef struct lock_stack lock_stack; +typedef struct lock_handle lock_handle; +typedef struct znode znode; +typedef struct flow flow_t; +typedef struct coord coord_t; +typedef struct tree_access_pointer tap_t; +typedef struct reiser4_object_create_data reiser4_object_create_data; +typedef union reiser4_plugin reiser4_plugin; +typedef __u16 reiser4_plugin_id; +typedef __u64 reiser4_plugin_groups; +typedef struct item_plugin item_plugin; +typedef struct jnode_plugin jnode_plugin; +typedef struct reiser4_item_data reiser4_item_data; +typedef union reiser4_key reiser4_key; +typedef struct reiser4_tree reiser4_tree; +typedef struct carry_cut_data carry_cut_data; +typedef struct carry_kill_data carry_kill_data; +typedef struct carry_tree_op carry_tree_op; +typedef struct carry_tree_node carry_tree_node; +typedef struct carry_plugin_info carry_plugin_info; +typedef struct reiser4_journal reiser4_journal; +typedef struct txn_atom txn_atom; +typedef struct txn_handle txn_handle; +typedef struct txn_mgr txn_mgr; +typedef struct reiser4_dir_entry_desc reiser4_dir_entry_desc; +typedef struct reiser4_context reiser4_context; +typedef struct carry_level carry_level; +typedef struct blocknr_set_entry blocknr_set_entry; +/* super_block->s_fs_info points to this */ +typedef struct reiser4_super_info_data reiser4_super_info_data; +/* next two objects are fields of reiser4_super_info_data */ +typedef struct reiser4_oid_allocator reiser4_oid_allocator; +typedef struct reiser4_space_allocator reiser4_space_allocator; + +typedef struct flush_scan flush_scan; +typedef struct flush_position flush_pos_t; + +typedef unsigned short pos_in_node_t; +#define MAX_POS_IN_NODE 65535 + +typedef struct jnode jnode; +typedef struct reiser4_blocknr_hint reiser4_blocknr_hint; + +typedef struct uf_coord uf_coord_t; +typedef struct hint hint_t; + +typedef struct ktxnmgrd_context ktxnmgrd_context; + +struct inode; +struct page; +struct file; +struct dentry; +struct super_block; + +/* return values of coord_by_key(). cbk == coord_by_key */ +typedef enum { + CBK_COORD_FOUND = 0, + CBK_COORD_NOTFOUND = -ENOENT, +} lookup_result; + +/* results of lookup with directory file */ +typedef enum { + FILE_NAME_FOUND = 0, + FILE_NAME_NOTFOUND = -ENOENT, + FILE_IO_ERROR = -EIO, /* FIXME: it seems silly to have special OOM, + IO_ERROR return codes for each search. */ + FILE_OOM = -ENOMEM /* FIXME: it seems silly to have special OOM, + IO_ERROR return codes for each search. */ +} file_lookup_result; + +/* behaviors of lookup. If coord we are looking for is actually in a tree, + both coincide. */ +typedef enum { + /* search exactly for the coord with key given */ + FIND_EXACT, + /* search for coord with the maximal key not greater than one + given */ + FIND_MAX_NOT_MORE_THAN /*LEFT_SLANT_BIAS */ +} lookup_bias; + +typedef enum { + /* number of leaf level of the tree + The fake root has (tree_level=0). */ + LEAF_LEVEL = 1, + + /* number of level one above leaf level of the tree. + + It is supposed that internal tree used by reiser4 to store file + system data and meta data will have height 2 initially (when + created by mkfs). + */ + TWIG_LEVEL = 2, +} tree_level; + +/* The "real" maximum ztree height is the 0-origin size of any per-level + array, since the zero'th level is not used. */ +#define REAL_MAX_ZTREE_HEIGHT (REISER4_MAX_ZTREE_HEIGHT-LEAF_LEVEL) + +/* enumeration of possible mutual position of item and coord. This enum is + return type of ->is_in_item() item plugin method which see. */ +typedef enum { + /* coord is on the left of an item */ + IP_ON_THE_LEFT, + /* coord is inside item */ + IP_INSIDE, + /* coord is inside item, but to the right of the rightmost unit of + this item */ + IP_RIGHT_EDGE, + /* coord is on the right of an item */ + IP_ON_THE_RIGHT +} interposition; + +/* type of lock to acquire on znode before returning it to caller */ +typedef enum { + ZNODE_NO_LOCK = 0, + ZNODE_READ_LOCK = 1, + ZNODE_WRITE_LOCK = 2, +} znode_lock_mode; + +/* type of lock request */ +typedef enum { + ZNODE_LOCK_LOPRI = 0, + ZNODE_LOCK_HIPRI = (1 << 0), + + /* By setting the ZNODE_LOCK_NONBLOCK flag in a lock request the call to + longterm_lock_znode will not sleep waiting for the lock to become + available. If the lock is unavailable, reiser4_znode_lock will + immediately return the value -E_REPEAT. */ + ZNODE_LOCK_NONBLOCK = (1 << 1), + /* An option for longterm_lock_znode which prevents atom fusion */ + ZNODE_LOCK_DONT_FUSE = (1 << 2) +} znode_lock_request; + +typedef enum { READ_OP = 0, WRITE_OP = 1 } rw_op; + +/* used to specify direction of shift. These must be -1 and 1 */ +typedef enum { + SHIFT_LEFT = 1, + SHIFT_RIGHT = -1 +} shift_direction; + +typedef enum { + LEFT_SIDE, + RIGHT_SIDE +} sideof; + +#define round_up(value, order) \ + ((typeof(value))(((long) (value) + (order) - 1U) & \ + ~((order) - 1))) + +/* values returned by squalloc_right_neighbor and its auxiliary functions */ +typedef enum { + /* unit of internal item is moved */ + SUBTREE_MOVED = 0, + /* nothing else can be squeezed into left neighbor */ + SQUEEZE_TARGET_FULL = 1, + /* all content of node is squeezed into its left neighbor */ + SQUEEZE_SOURCE_EMPTY = 2, + /* one more item is copied (this is only returned by + allocate_and_copy_extent to squalloc_twig)) */ + SQUEEZE_CONTINUE = 3 +} squeeze_result; + +/* Do not change items ids. If you do - there will be format change */ +typedef enum { + STATIC_STAT_DATA_ID = 0x0, + SIMPLE_DIR_ENTRY_ID = 0x1, + COMPOUND_DIR_ID = 0x2, + NODE_POINTER_ID = 0x3, + EXTENT_POINTER_ID = 0x5, + FORMATTING_ID = 0x6, + CTAIL_ID = 0x7, + BLACK_BOX_ID = 0x8, + LAST_ITEM_ID = 0x9 +} item_id; + +/* Flags passed to jnode_flush() to allow it to distinguish default settings + based on whether commit() was called or VM memory pressure was applied. */ +typedef enum { + /* submit flush queue to disk at jnode_flush completion */ + JNODE_FLUSH_WRITE_BLOCKS = 1, + + /* flush is called for commit */ + JNODE_FLUSH_COMMIT = 2, + /* not implemented */ + JNODE_FLUSH_MEMORY_FORMATTED = 4, + + /* not implemented */ + JNODE_FLUSH_MEMORY_UNFORMATTED = 8, +} jnode_flush_flags; + +/* Flags to insert/paste carry operations. Currently they only used in + flushing code, but in future, they can be used to optimize for repetitive + accesses. */ +typedef enum { + /* carry is not allowed to shift data to the left when trying to find + free space */ + COPI_DONT_SHIFT_LEFT = (1 << 0), + /* carry is not allowed to shift data to the right when trying to find + free space */ + COPI_DONT_SHIFT_RIGHT = (1 << 1), + /* carry is not allowed to allocate new node(s) when trying to find + free space */ + COPI_DONT_ALLOCATE = (1 << 2), + /* try to load left neighbor if its not in a cache */ + COPI_LOAD_LEFT = (1 << 3), + /* try to load right neighbor if its not in a cache */ + COPI_LOAD_RIGHT = (1 << 4), + /* shift insertion point to the left neighbor */ + COPI_GO_LEFT = (1 << 5), + /* shift insertion point to the right neighbor */ + COPI_GO_RIGHT = (1 << 6), + /* try to step back into original node if insertion into new node + fails after shifting data there. */ + COPI_STEP_BACK = (1 << 7) +} cop_insert_flag; + +typedef enum { + SAFE_UNLINK, /* safe-link for unlink */ + SAFE_TRUNCATE /* safe-link for truncate */ +} reiser4_safe_link_t; + +/* this is to show on which list of atom jnode is */ +typedef enum { + NOT_CAPTURED, + DIRTY_LIST, + CLEAN_LIST, + FQ_LIST, + WB_LIST, + OVRWR_LIST +} atom_list; + +/* __REISER4_FORWARD_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/fsdata.c linux-2.6.30/fs/reiser4/fsdata.c --- linux-2.6.30.orig/fs/reiser4/fsdata.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/fsdata.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,804 @@ +/* Copyright 2001, 2002, 2003, 2004, 2005 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "fsdata.h" +#include "inode.h" + + +/* cache or dir_cursors */ +static struct kmem_cache *d_cursor_cache; + +/* list of unused cursors */ +static LIST_HEAD(cursor_cache); + +/* number of cursors in list of ununsed cursors */ +static unsigned long d_cursor_unused = 0; + +/* spinlock protecting manipulations with dir_cursor's hash table and lists */ +DEFINE_SPINLOCK(d_lock); + +static reiser4_file_fsdata *create_fsdata(struct file *file); +static int file_is_stateless(struct file *file); +static void free_fsdata(reiser4_file_fsdata *fsdata); +static void kill_cursor(dir_cursor *); + +/** + * d_cursor_shrink - shrink callback for cache of dir_cursor-s + * @nr: number of objects to free + * @mask: GFP mask + * + * Shrinks d_cursor_cache. Scan LRU list of unused cursors, freeing requested + * number. Return number of still freeable cursors. + */ +static int d_cursor_shrink(int nr, gfp_t mask) +{ + if (nr != 0) { + dir_cursor *scan; + int killed; + + killed = 0; + spin_lock(&d_lock); + while (!list_empty(&cursor_cache)) { + scan = list_entry(cursor_cache.next, dir_cursor, alist); + assert("nikita-3567", scan->ref == 0); + kill_cursor(scan); + ++killed; + --nr; + if (nr == 0) + break; + } + spin_unlock(&d_lock); + } + return d_cursor_unused; +} + +/* + * actually, d_cursors are "priceless", because there is no way to + * recover information stored in them. On the other hand, we don't + * want to consume all kernel memory by them. As a compromise, just + * assign higher "seeks" value to d_cursor cache, so that it will be + * shrunk only if system is really tight on memory. + */ +static struct shrinker d_cursor_shrinker = { + .shrink = d_cursor_shrink, + .seeks = DEFAULT_SEEKS << 3, +}; + +/** + * reiser4_init_d_cursor - create d_cursor cache + * + * Initializes slab cache of d_cursors. It is part of reiser4 module + * initialization. + */ +int reiser4_init_d_cursor(void) +{ + d_cursor_cache = kmem_cache_create("d_cursor", sizeof(dir_cursor), 0, + SLAB_HWCACHE_ALIGN, NULL); + if (d_cursor_cache == NULL) + return RETERR(-ENOMEM); + + register_shrinker(&d_cursor_shrinker); + return 0; +} + +/** + * reiser4_done_d_cursor - delete d_cursor cache and d_cursor shrinker + * + * This is called on reiser4 module unloading or system shutdown. + */ +void reiser4_done_d_cursor(void) +{ + unregister_shrinker(&d_cursor_shrinker); + + destroy_reiser4_cache(&d_cursor_cache); +} + +#define D_CURSOR_TABLE_SIZE (256) + +static inline unsigned long +d_cursor_hash(d_cursor_hash_table * table, const struct d_cursor_key *key) +{ + assert("nikita-3555", IS_POW(D_CURSOR_TABLE_SIZE)); + return (key->oid + key->cid) & (D_CURSOR_TABLE_SIZE - 1); +} + +static inline int d_cursor_eq(const struct d_cursor_key *k1, + const struct d_cursor_key *k2) +{ + return k1->cid == k2->cid && k1->oid == k2->oid; +} + +/* + * define functions to manipulate reiser4 super block's hash table of + * dir_cursors + */ +#define KMALLOC(size) kmalloc((size), reiser4_ctx_gfp_mask_get()) +#define KFREE(ptr, size) kfree(ptr) +TYPE_SAFE_HASH_DEFINE(d_cursor, + dir_cursor, + struct d_cursor_key, + key, hash, d_cursor_hash, d_cursor_eq); +#undef KFREE +#undef KMALLOC + +/** + * reiser4_init_super_d_info - initialize per-super-block d_cursor resources + * @super: super block to initialize + * + * Initializes per-super-block d_cursor's hash table and radix tree. It is part + * of mount. + */ +int reiser4_init_super_d_info(struct super_block *super) +{ + struct d_cursor_info *p; + + p = &get_super_private(super)->d_info; + + INIT_RADIX_TREE(&p->tree, reiser4_ctx_gfp_mask_get()); + return d_cursor_hash_init(&p->table, D_CURSOR_TABLE_SIZE); +} + +/** + * reiser4_done_super_d_info - release per-super-block d_cursor resources + * @super: super block being umounted + * + * It is called on umount. Kills all directory cursors attached to suoer block. + */ +void reiser4_done_super_d_info(struct super_block *super) +{ + struct d_cursor_info *d_info; + dir_cursor *cursor, *next; + + d_info = &get_super_private(super)->d_info; + for_all_in_htable(&d_info->table, d_cursor, cursor, next) + kill_cursor(cursor); + + BUG_ON(d_info->tree.rnode != NULL); + d_cursor_hash_done(&d_info->table); +} + +/** + * kill_cursor - free dir_cursor and reiser4_file_fsdata attached to it + * @cursor: cursor to free + * + * Removes reiser4_file_fsdata attached to @cursor from readdir list of + * reiser4_inode, frees that reiser4_file_fsdata. Removes @cursor from from + * indices, hash table, list of unused cursors and frees it. + */ +static void kill_cursor(dir_cursor *cursor) +{ + unsigned long index; + + assert("nikita-3566", cursor->ref == 0); + assert("nikita-3572", cursor->fsdata != NULL); + + index = (unsigned long)cursor->key.oid; + list_del_init(&cursor->fsdata->dir.linkage); + free_fsdata(cursor->fsdata); + cursor->fsdata = NULL; + + if (list_empty_careful(&cursor->list)) + /* this is last cursor for a file. Kill radix-tree entry */ + radix_tree_delete(&cursor->info->tree, index); + else { + void **slot; + + /* + * there are other cursors for the same oid. + */ + + /* + * if radix tree point to the cursor being removed, re-target + * radix tree slot to the next cursor in the (non-empty as was + * checked above) element of the circular list of all cursors + * for this oid. + */ + slot = radix_tree_lookup_slot(&cursor->info->tree, index); + assert("nikita-3571", *slot != NULL); + if (*slot == cursor) + *slot = list_entry(cursor->list.next, dir_cursor, list); + /* remove cursor from circular list */ + list_del_init(&cursor->list); + } + /* remove cursor from the list of unused cursors */ + list_del_init(&cursor->alist); + /* remove cursor from the hash table */ + d_cursor_hash_remove(&cursor->info->table, cursor); + /* and free it */ + kmem_cache_free(d_cursor_cache, cursor); + --d_cursor_unused; +} + +/* possible actions that can be performed on all cursors for the given file */ +enum cursor_action { + /* + * load all detached state: this is called when stat-data is loaded + * from the disk to recover information about all pending readdirs + */ + CURSOR_LOAD, + /* + * detach all state from inode, leaving it in the cache. This is called + * when inode is removed form the memory by memory pressure + */ + CURSOR_DISPOSE, + /* + * detach cursors from the inode, and free them. This is called when + * inode is destroyed + */ + CURSOR_KILL +}; + +/* + * return d_cursor data for the file system @inode is in. + */ +static inline struct d_cursor_info *d_info(struct inode *inode) +{ + return &get_super_private(inode->i_sb)->d_info; +} + +/* + * lookup d_cursor in the per-super-block radix tree. + */ +static inline dir_cursor *lookup(struct d_cursor_info *info, + unsigned long index) +{ + return (dir_cursor *) radix_tree_lookup(&info->tree, index); +} + +/* + * attach @cursor to the radix tree. There may be multiple cursors for the + * same oid, they are chained into circular list. + */ +static void bind_cursor(dir_cursor * cursor, unsigned long index) +{ + dir_cursor *head; + + head = lookup(cursor->info, index); + if (head == NULL) { + /* this is the first cursor for this index */ + INIT_LIST_HEAD(&cursor->list); + radix_tree_insert(&cursor->info->tree, index, cursor); + } else { + /* some cursor already exists. Chain ours */ + list_add(&cursor->list, &head->list); + } +} + +/* + * detach fsdata (if detachable) from file descriptor, and put cursor on the + * "unused" list. Called when file descriptor is not longer in active use. + */ +static void clean_fsdata(struct file *file) +{ + dir_cursor *cursor; + reiser4_file_fsdata *fsdata; + + assert("nikita-3570", file_is_stateless(file)); + + fsdata = (reiser4_file_fsdata *) file->private_data; + if (fsdata != NULL) { + cursor = fsdata->cursor; + if (cursor != NULL) { + spin_lock(&d_lock); + --cursor->ref; + if (cursor->ref == 0) { + list_add_tail(&cursor->alist, &cursor_cache); + ++d_cursor_unused; + } + spin_unlock(&d_lock); + file->private_data = NULL; + } + } +} + +/* + * global counter used to generate "client ids". These ids are encoded into + * high bits of fpos. + */ +static __u32 cid_counter = 0; +#define CID_SHIFT (20) +#define CID_MASK (0xfffffull) + +static void free_file_fsdata_nolock(struct file *); + +/** + * insert_cursor - allocate file_fsdata, insert cursor to tree and hash table + * @cursor: + * @file: + * @inode: + * + * Allocates reiser4_file_fsdata, attaches it to @cursor, inserts cursor to + * reiser4 super block's hash table and radix tree. + add detachable readdir + * state to the @f + */ +static int insert_cursor(dir_cursor *cursor, struct file *file, + struct inode *inode) +{ + int result; + reiser4_file_fsdata *fsdata; + + memset(cursor, 0, sizeof *cursor); + + /* this is either first call to readdir, or rewind. Anyway, create new + * cursor. */ + fsdata = create_fsdata(NULL); + if (fsdata != NULL) { + result = radix_tree_preload(reiser4_ctx_gfp_mask_get()); + if (result == 0) { + struct d_cursor_info *info; + oid_t oid; + + info = d_info(inode); + oid = get_inode_oid(inode); + /* cid occupies higher 12 bits of f->f_pos. Don't + * allow it to become negative: this confuses + * nfsd_readdir() */ + cursor->key.cid = (++cid_counter) & 0x7ff; + cursor->key.oid = oid; + cursor->fsdata = fsdata; + cursor->info = info; + cursor->ref = 1; + + spin_lock_inode(inode); + /* install cursor as @f's private_data, discarding old + * one if necessary */ +#if REISER4_DEBUG + if (file->private_data) + warning("", "file has fsdata already"); +#endif + clean_fsdata(file); + free_file_fsdata_nolock(file); + file->private_data = fsdata; + fsdata->cursor = cursor; + spin_unlock_inode(inode); + spin_lock(&d_lock); + /* insert cursor into hash table */ + d_cursor_hash_insert(&info->table, cursor); + /* and chain it into radix-tree */ + bind_cursor(cursor, (unsigned long)oid); + spin_unlock(&d_lock); + radix_tree_preload_end(); + file->f_pos = ((__u64) cursor->key.cid) << CID_SHIFT; + } + } else + result = RETERR(-ENOMEM); + return result; +} + +/** + * process_cursors - do action on each cursor attached to inode + * @inode: + * @act: action to do + * + * Finds all cursors of @inode in reiser4's super block radix tree of cursors + * and performs action specified by @act on each of cursors. + */ +static void process_cursors(struct inode *inode, enum cursor_action act) +{ + oid_t oid; + dir_cursor *start; + struct list_head *head; + reiser4_context *ctx; + struct d_cursor_info *info; + + /* this can be called by + * + * kswapd->...->prune_icache->..reiser4_destroy_inode + * + * without reiser4_context + */ + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) { + warning("vs-23", "failed to init context"); + return; + } + + assert("nikita-3558", inode != NULL); + + info = d_info(inode); + oid = get_inode_oid(inode); + spin_lock_inode(inode); + head = get_readdir_list(inode); + spin_lock(&d_lock); + /* find any cursor for this oid: reference to it is hanging of radix + * tree */ + start = lookup(info, (unsigned long)oid); + if (start != NULL) { + dir_cursor *scan; + reiser4_file_fsdata *fsdata; + + /* process circular list of cursors for this oid */ + scan = start; + do { + dir_cursor *next; + + next = list_entry(scan->list.next, dir_cursor, list); + fsdata = scan->fsdata; + assert("nikita-3557", fsdata != NULL); + if (scan->key.oid == oid) { + switch (act) { + case CURSOR_DISPOSE: + list_del_init(&fsdata->dir.linkage); + break; + case CURSOR_LOAD: + list_add(&fsdata->dir.linkage, head); + break; + case CURSOR_KILL: + kill_cursor(scan); + break; + } + } + if (scan == next) + /* last cursor was just killed */ + break; + scan = next; + } while (scan != start); + } + spin_unlock(&d_lock); + /* check that we killed 'em all */ + assert("nikita-3568", + ergo(act == CURSOR_KILL, + list_empty_careful(get_readdir_list(inode)))); + assert("nikita-3569", + ergo(act == CURSOR_KILL, lookup(info, oid) == NULL)); + spin_unlock_inode(inode); + reiser4_exit_context(ctx); +} + +/** + * reiser4_dispose_cursors - removes cursors from inode's list + * @inode: inode to dispose cursors of + * + * For each of cursors corresponding to @inode - removes reiser4_file_fsdata + * attached to cursor from inode's readdir list. This is called when inode is + * removed from the memory by memory pressure. + */ +void reiser4_dispose_cursors(struct inode *inode) +{ + process_cursors(inode, CURSOR_DISPOSE); +} + +/** + * reiser4_load_cursors - attach cursors to inode + * @inode: inode to load cursors to + * + * For each of cursors corresponding to @inode - attaches reiser4_file_fsdata + * attached to cursor to inode's readdir list. This is done when inode is + * loaded into memory. + */ +void reiser4_load_cursors(struct inode *inode) +{ + process_cursors(inode, CURSOR_LOAD); +} + +/** + * reiser4_kill_cursors - kill all inode cursors + * @inode: inode to kill cursors of + * + * Frees all cursors for this inode. This is called when inode is destroyed. + */ +void reiser4_kill_cursors(struct inode *inode) +{ + process_cursors(inode, CURSOR_KILL); +} + +/** + * file_is_stateless - + * @file: + * + * true, if file descriptor @f is created by NFS server by "demand" to serve + * one file system operation. This means that there may be "detached state" + * for underlying inode. + */ +static int file_is_stateless(struct file *file) +{ + return reiser4_get_dentry_fsdata(file->f_dentry)->stateless; +} + +/** + * reiser4_get_dir_fpos - + * @dir: + * + * Calculates ->fpos from user-supplied cookie. Normally it is dir->f_pos, but + * in the case of stateless directory operation (readdir-over-nfs), client id + * was encoded in the high bits of cookie and should me masked off. + */ +loff_t reiser4_get_dir_fpos(struct file *dir) +{ + if (file_is_stateless(dir)) + return dir->f_pos & CID_MASK; + else + return dir->f_pos; +} + +/** + * reiser4_attach_fsdata - try to attach fsdata + * @file: + * @inode: + * + * Finds or creates cursor for readdir-over-nfs. + */ +int reiser4_attach_fsdata(struct file *file, struct inode *inode) +{ + loff_t pos; + int result; + dir_cursor *cursor; + + /* + * we are serialized by inode->i_mutex + */ + if (!file_is_stateless(file)) + return 0; + + pos = file->f_pos; + result = 0; + if (pos == 0) { + /* + * first call to readdir (or rewind to the beginning of + * directory) + */ + cursor = kmem_cache_alloc(d_cursor_cache, + reiser4_ctx_gfp_mask_get()); + if (cursor != NULL) + result = insert_cursor(cursor, file, inode); + else + result = RETERR(-ENOMEM); + } else { + /* try to find existing cursor */ + struct d_cursor_key key; + + key.cid = pos >> CID_SHIFT; + key.oid = get_inode_oid(inode); + spin_lock(&d_lock); + cursor = d_cursor_hash_find(&d_info(inode)->table, &key); + if (cursor != NULL) { + /* cursor was found */ + if (cursor->ref == 0) { + /* move it from unused list */ + list_del_init(&cursor->alist); + --d_cursor_unused; + } + ++cursor->ref; + } + spin_unlock(&d_lock); + if (cursor != NULL) { + spin_lock_inode(inode); + assert("nikita-3556", cursor->fsdata->back == NULL); + clean_fsdata(file); + free_file_fsdata_nolock(file); + file->private_data = cursor->fsdata; + spin_unlock_inode(inode); + } + } + return result; +} + +/** + * reiser4_detach_fsdata - ??? + * @file: + * + * detach fsdata, if necessary + */ +void reiser4_detach_fsdata(struct file *file) +{ + struct inode *inode; + + if (!file_is_stateless(file)) + return; + + inode = file->f_dentry->d_inode; + spin_lock_inode(inode); + clean_fsdata(file); + spin_unlock_inode(inode); +} + +/* slab for reiser4_dentry_fsdata */ +static struct kmem_cache *dentry_fsdata_cache; + +/** + * reiser4_init_dentry_fsdata - create cache of dentry_fsdata + * + * Initializes slab cache of structures attached to denty->d_fsdata. It is + * part of reiser4 module initialization. + */ +int reiser4_init_dentry_fsdata(void) +{ + dentry_fsdata_cache = kmem_cache_create("dentry_fsdata", + sizeof(struct reiser4_dentry_fsdata), + 0, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT, + NULL); + if (dentry_fsdata_cache == NULL) + return RETERR(-ENOMEM); + return 0; +} + +/** + * reiser4_done_dentry_fsdata - delete cache of dentry_fsdata + * + * This is called on reiser4 module unloading or system shutdown. + */ +void reiser4_done_dentry_fsdata(void) +{ + destroy_reiser4_cache(&dentry_fsdata_cache); +} + +/** + * reiser4_get_dentry_fsdata - get fs-specific dentry data + * @dentry: queried dentry + * + * Allocates if necessary and returns per-dentry data that we attach to each + * dentry. + */ +struct reiser4_dentry_fsdata *reiser4_get_dentry_fsdata(struct dentry *dentry) +{ + assert("nikita-1365", dentry != NULL); + + if (dentry->d_fsdata == NULL) { + dentry->d_fsdata = kmem_cache_alloc(dentry_fsdata_cache, + reiser4_ctx_gfp_mask_get()); + if (dentry->d_fsdata == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + memset(dentry->d_fsdata, 0, + sizeof(struct reiser4_dentry_fsdata)); + } + return dentry->d_fsdata; +} + +/** + * reiser4_free_dentry_fsdata - detach and free dentry_fsdata + * @dentry: dentry to free fsdata of + * + * Detaches and frees fs-specific dentry data + */ +void reiser4_free_dentry_fsdata(struct dentry *dentry) +{ + if (dentry->d_fsdata != NULL) { + kmem_cache_free(dentry_fsdata_cache, dentry->d_fsdata); + dentry->d_fsdata = NULL; + } +} + +/* slab for reiser4_file_fsdata */ +static struct kmem_cache *file_fsdata_cache; + +/** + * reiser4_init_file_fsdata - create cache of reiser4_file_fsdata + * + * Initializes slab cache of structures attached to file->private_data. It is + * part of reiser4 module initialization. + */ +int reiser4_init_file_fsdata(void) +{ + file_fsdata_cache = kmem_cache_create("file_fsdata", + sizeof(reiser4_file_fsdata), + 0, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT, NULL); + if (file_fsdata_cache == NULL) + return RETERR(-ENOMEM); + return 0; +} + +/** + * reiser4_done_file_fsdata - delete cache of reiser4_file_fsdata + * + * This is called on reiser4 module unloading or system shutdown. + */ +void reiser4_done_file_fsdata(void) +{ + destroy_reiser4_cache(&file_fsdata_cache); +} + +/** + * create_fsdata - allocate and initialize reiser4_file_fsdata + * @file: what to create file_fsdata for, may be NULL + * + * Allocates and initializes reiser4_file_fsdata structure. + */ +static reiser4_file_fsdata *create_fsdata(struct file *file) +{ + reiser4_file_fsdata *fsdata; + + fsdata = kmem_cache_alloc(file_fsdata_cache, + reiser4_ctx_gfp_mask_get()); + if (fsdata != NULL) { + memset(fsdata, 0, sizeof *fsdata); + fsdata->ra1.max_window_size = VM_MAX_READAHEAD * 1024; + fsdata->back = file; + INIT_LIST_HEAD(&fsdata->dir.linkage); + } + return fsdata; +} + +/** + * free_fsdata - free reiser4_file_fsdata + * @fsdata: object to free + * + * Dual to create_fsdata(). Free reiser4_file_fsdata. + */ +static void free_fsdata(reiser4_file_fsdata *fsdata) +{ + BUG_ON(fsdata == NULL); + kmem_cache_free(file_fsdata_cache, fsdata); +} + +/** + * reiser4_get_file_fsdata - get fs-specific file data + * @file: queried file + * + * Returns fs-specific data of @file. If it is NULL, allocates it and attaches + * to @file. + */ +reiser4_file_fsdata *reiser4_get_file_fsdata(struct file *file) +{ + assert("nikita-1603", file != NULL); + + if (file->private_data == NULL) { + reiser4_file_fsdata *fsdata; + struct inode *inode; + + fsdata = create_fsdata(file); + if (fsdata == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + + inode = file->f_dentry->d_inode; + spin_lock_inode(inode); + if (file->private_data == NULL) { + file->private_data = fsdata; + fsdata = NULL; + } + spin_unlock_inode(inode); + if (fsdata != NULL) + /* other thread initialized ->fsdata */ + kmem_cache_free(file_fsdata_cache, fsdata); + } + assert("nikita-2665", file->private_data != NULL); + return file->private_data; +} + +/** + * free_file_fsdata_nolock - detach and free reiser4_file_fsdata + * @file: + * + * Detaches reiser4_file_fsdata from @file, removes reiser4_file_fsdata from + * readdir list, frees if it is not linked to d_cursor object. + */ +static void free_file_fsdata_nolock(struct file *file) +{ + reiser4_file_fsdata *fsdata; + + assert("", spin_inode_is_locked(file->f_dentry->d_inode)); + fsdata = file->private_data; + if (fsdata != NULL) { + list_del_init(&fsdata->dir.linkage); + if (fsdata->cursor == NULL) + free_fsdata(fsdata); + } + file->private_data = NULL; +} + +/** + * reiser4_free_file_fsdata - detach from struct file and free reiser4_file_fsdata + * @file: + * + * Spinlocks inode and calls free_file_fsdata_nolock to do the work. + */ +void reiser4_free_file_fsdata(struct file *file) +{ + spin_lock_inode(file->f_dentry->d_inode); + free_file_fsdata_nolock(file); + spin_unlock_inode(file->f_dentry->d_inode); +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/fsdata.h linux-2.6.30/fs/reiser4/fsdata.h --- linux-2.6.30.orig/fs/reiser4/fsdata.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/fsdata.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,205 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +#if !defined(__REISER4_FSDATA_H__) +#define __REISER4_FSDATA_H__ + +#include "debug.h" +#include "kassign.h" +#include "seal.h" +#include "type_safe_hash.h" +#include "plugin/file/file.h" +#include "readahead.h" + +/* + * comment about reiser4_dentry_fsdata + * + * + */ + +/* + * locking: fields of per file descriptor readdir_pos and ->f_pos are + * protected by ->i_mutex on inode. Under this lock following invariant + * holds: + * + * file descriptor is "looking" at the entry_no-th directory entry from + * the beginning of directory. This entry has key dir_entry_key and is + * pos-th entry with duplicate-key sequence. + * + */ + +/* logical position within directory */ +struct dir_pos { + /* key of directory entry (actually, part of a key sufficient to + identify directory entry) */ + de_id dir_entry_key; + /* ordinal number of directory entry among all entries with the same + key. (Starting from 0.) */ + unsigned pos; +}; + +struct readdir_pos { + /* f_pos corresponding to this readdir position */ + __u64 fpos; + /* logical position within directory */ + struct dir_pos position; + /* logical number of directory entry within + directory */ + __u64 entry_no; +}; + +/* + * this is used to speed up lookups for directory entry: on initial call to + * ->lookup() seal and coord of directory entry (if found, that is) are stored + * in struct dentry and reused later to avoid tree traversals. + */ +struct de_location { + /* seal covering directory entry */ + seal_t entry_seal; + /* coord of directory entry */ + coord_t entry_coord; + /* ordinal number of directory entry among all entries with the same + key. (Starting from 0.) */ + int pos; +}; + +/** + * reiser4_dentry_fsdata - reiser4-specific data attached to dentries + * + * This is allocated dynamically and released in d_op->d_release() + * + * Currently it only contains cached location (hint) of directory entry, but + * it is expected that other information will be accumulated here. + */ +struct reiser4_dentry_fsdata { + /* + * here will go fields filled by ->lookup() to speedup next + * create/unlink, like blocknr of znode with stat-data, or key of + * stat-data. + */ + struct de_location dec; + int stateless; /* created through reiser4_decode_fh, needs + * special treatment in readdir. */ +}; + +extern int reiser4_init_dentry_fsdata(void); +extern void reiser4_done_dentry_fsdata(void); +extern struct reiser4_dentry_fsdata *reiser4_get_dentry_fsdata(struct dentry *); +extern void reiser4_free_dentry_fsdata(struct dentry *dentry); + +/** + * reiser4_file_fsdata - reiser4-specific data attached to file->private_data + * + * This is allocated dynamically and released in inode->i_fop->release + */ +typedef struct reiser4_file_fsdata { + /* + * pointer back to the struct file which this reiser4_file_fsdata is + * part of + */ + struct file *back; + /* detached cursor for stateless readdir. */ + struct dir_cursor *cursor; + /* + * We need both directory and regular file parts here, because there + * are file system objects that are files and directories. + */ + struct { + /* + * position in directory. It is updated each time directory is + * modified + */ + struct readdir_pos readdir; + /* head of this list is reiser4_inode->lists.readdir_list */ + struct list_head linkage; + } dir; + /* hints to speed up operations with regular files: read and write. */ + struct { + hint_t hint; + } reg; + struct reiser4_file_ra_state ra1; + +} reiser4_file_fsdata; + +extern int reiser4_init_file_fsdata(void); +extern void reiser4_done_file_fsdata(void); +extern reiser4_file_fsdata *reiser4_get_file_fsdata(struct file *); +extern void reiser4_free_file_fsdata(struct file *); + +/* + * d_cursor is reiser4_file_fsdata not attached to struct file. d_cursors are + * used to address problem reiser4 has with readdir accesses via NFS. See + * plugin/file_ops_readdir.c for more details. + */ +struct d_cursor_key{ + __u16 cid; + __u64 oid; +}; + +/* + * define structures d_cursor_hash_table d_cursor_hash_link which are used to + * maintain hash table of dir_cursor-s in reiser4's super block + */ +typedef struct dir_cursor dir_cursor; +TYPE_SAFE_HASH_DECLARE(d_cursor, dir_cursor); + +struct dir_cursor { + int ref; + reiser4_file_fsdata *fsdata; + + /* link to reiser4 super block hash table of cursors */ + d_cursor_hash_link hash; + + /* + * this is to link cursors to reiser4 super block's radix tree of + * cursors if there are more than one cursor of the same objectid + */ + struct list_head list; + struct d_cursor_key key; + struct d_cursor_info *info; + /* list of unused cursors */ + struct list_head alist; +}; + +extern int reiser4_init_d_cursor(void); +extern void reiser4_done_d_cursor(void); + +extern int reiser4_init_super_d_info(struct super_block *); +extern void reiser4_done_super_d_info(struct super_block *); + +extern loff_t reiser4_get_dir_fpos(struct file *); +extern int reiser4_attach_fsdata(struct file *, struct inode *); +extern void reiser4_detach_fsdata(struct file *); + +/* these are needed for "stateless" readdir. See plugin/file_ops_readdir.c for + more details */ +void reiser4_dispose_cursors(struct inode *inode); +void reiser4_load_cursors(struct inode *inode); +void reiser4_kill_cursors(struct inode *inode); +void reiser4_adjust_dir_file(struct inode *dir, const struct dentry *de, + int offset, int adj); + +/* + * this structure is embedded to reise4_super_info_data. It maintains d_cursors + * (detached readdir state). See plugin/file_ops_readdir.c for more details. + */ +struct d_cursor_info { + d_cursor_hash_table table; + struct radix_tree_root tree; +}; + +/* spinlock protecting readdir cursors */ +extern spinlock_t d_lock; + +/* __REISER4_FSDATA_H__ */ +#endif + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/init_super.c linux-2.6.30/fs/reiser4/init_super.c --- linux-2.6.30.orig/fs/reiser4/init_super.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/init_super.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,751 @@ +/* Copyright by Hans Reiser, 2003 */ + +#include "super.h" +#include "inode.h" +#include "plugin/plugin_set.h" + +#include + +/** + * init_fs_info - allocate reiser4 specific super block + * @super: super block of filesystem + * + * Allocates and initialize reiser4_super_info_data, attaches it to + * super->s_fs_info, initializes structures maintaining d_cursor-s. + */ +int reiser4_init_fs_info(struct super_block *super) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = kzalloc(sizeof(reiser4_super_info_data), + reiser4_ctx_gfp_mask_get()); + if (!sbinfo) + return RETERR(-ENOMEM); + + super->s_fs_info = sbinfo; + super->s_op = NULL; + + ON_DEBUG(INIT_LIST_HEAD(&sbinfo->all_jnodes)); + ON_DEBUG(spin_lock_init(&sbinfo->all_guard)); + + mutex_init(&sbinfo->delete_mutex); + spin_lock_init(&(sbinfo->guard)); + + /* initialize per-super-block d_cursor resources */ + reiser4_init_super_d_info(super); + + return 0; +} + +/** + * reiser4_done_fs_info - free reiser4 specific super block + * @super: super block of filesystem + * + * Performs some sanity checks, releases structures maintaining d_cursor-s, + * frees reiser4_super_info_data. + */ +void reiser4_done_fs_info(struct super_block *super) +{ + assert("zam-990", super->s_fs_info != NULL); + + /* release per-super-block d_cursor resources */ + reiser4_done_super_d_info(super); + + /* make sure that there are not jnodes already */ + assert("", list_empty(&get_super_private(super)->all_jnodes)); + assert("", get_current_context()->trans->atom == NULL); + reiser4_check_block_counters(super); + kfree(super->s_fs_info); + super->s_fs_info = NULL; +} + +/* type of option parseable by parse_option() */ +typedef enum { + /* value of option is arbitrary string */ + OPT_STRING, + + /* + * option specifies bit in a bitmask. When option is set - bit in + * sbinfo->fs_flags is set. Examples are bsdgroups, 32bittimes, mtflush, + * dont_load_bitmap, atomic_write. + */ + OPT_BIT, + + /* + * value of option should conform to sprintf() format. Examples are + * tmgr.atom_max_size=N, tmgr.atom_max_age=N + */ + OPT_FORMAT, + + /* + * option can take one of predefined values. Example is onerror=panic or + * onerror=remount-ro + */ + OPT_ONEOF, +} opt_type_t; + +#if 0 +struct opt_bitmask_bit { + const char *bit_name; + int bit_nr; +}; +#endif + +/* description of option parseable by parse_option() */ +struct opt_desc { + /* option name. + + parsed portion of string has a form "name=value". + */ + const char *name; + /* type of option */ + opt_type_t type; + union { + /* where to store value of string option (type == OPT_STRING) */ + char **string; + /* description of bits for bit option (type == OPT_BIT) */ + struct { + int nr; + void *addr; + } bit; + /* description of format and targets for format option (type + == OPT_FORMAT) */ + struct { + const char *format; + int nr_args; + void *arg1; + void *arg2; + void *arg3; + void *arg4; + } f; + struct { + int *result; + const char *list[10]; + } oneof; + struct { + void *addr; + int nr_bits; + /* struct opt_bitmask_bit *bits; */ + } bitmask; + } u; +}; + +/** + * parse_option - parse one option + * @opt_strin: starting point of parsing + * @opt: option description + * + * foo=bar, + * ^ ^ ^ + * | | +-- replaced to '\0' + * | +-- val_start + * +-- opt_string + * Figures out option type and handles option correspondingly. + */ +static int parse_option(char *opt_string, struct opt_desc *opt) +{ + char *val_start; + int result; + const char *err_msg; + + /* NOTE-NIKITA think about using lib/cmdline.c functions here. */ + + val_start = strchr(opt_string, '='); + if (val_start != NULL) { + *val_start = '\0'; + ++val_start; + } + + err_msg = NULL; + result = 0; + switch (opt->type) { + case OPT_STRING: + if (val_start == NULL) { + err_msg = "String arg missing"; + result = RETERR(-EINVAL); + } else + *opt->u.string = val_start; + break; + case OPT_BIT: + if (val_start != NULL) + err_msg = "Value ignored"; + else + set_bit(opt->u.bit.nr, opt->u.bit.addr); + break; + case OPT_FORMAT: + if (val_start == NULL) { + err_msg = "Formatted arg missing"; + result = RETERR(-EINVAL); + break; + } + if (sscanf(val_start, opt->u.f.format, + opt->u.f.arg1, opt->u.f.arg2, opt->u.f.arg3, + opt->u.f.arg4) != opt->u.f.nr_args) { + err_msg = "Wrong conversion"; + result = RETERR(-EINVAL); + } + break; + case OPT_ONEOF: + { + int i = 0; + + if (val_start == NULL) { + err_msg = "Value is missing"; + result = RETERR(-EINVAL); + break; + } + err_msg = "Wrong option value"; + result = RETERR(-EINVAL); + while (opt->u.oneof.list[i]) { + if (!strcmp(opt->u.oneof.list[i], val_start)) { + result = 0; + err_msg = NULL; + *opt->u.oneof.result = i; + break; + } + i++; + } + break; + } + default: + wrong_return_value("nikita-2100", "opt -> type"); + break; + } + if (err_msg != NULL) { + warning("nikita-2496", "%s when parsing option \"%s%s%s\"", + err_msg, opt->name, val_start ? "=" : "", + val_start ? : ""); + } + return result; +} + +/** + * parse_options - parse reiser4 mount options + * @opt_string: starting point + * @opts: array of option description + * @nr_opts: number of elements in @opts + * + * Parses comma separated list of reiser4 mount options. + */ +static int parse_options(char *opt_string, struct opt_desc *opts, int nr_opts) +{ + int result; + + result = 0; + while ((result == 0) && opt_string && *opt_string) { + int j; + char *next; + + next = strchr(opt_string, ','); + if (next != NULL) { + *next = '\0'; + ++next; + } + for (j = 0; j < nr_opts; ++j) { + if (!strncmp(opt_string, opts[j].name, + strlen(opts[j].name))) { + result = parse_option(opt_string, &opts[j]); + break; + } + } + if (j == nr_opts) { + warning("nikita-2307", "Unrecognized option: \"%s\"", + opt_string); + /* traditionally, -EINVAL is returned on wrong mount + option */ + result = RETERR(-EINVAL); + } + opt_string = next; + } + return result; +} + +#define NUM_OPT(label, fmt, addr) \ + { \ + .name = (label), \ + .type = OPT_FORMAT, \ + .u = { \ + .f = { \ + .format = (fmt), \ + .nr_args = 1, \ + .arg1 = (addr), \ + .arg2 = NULL, \ + .arg3 = NULL, \ + .arg4 = NULL \ + } \ + } \ + } + +#define SB_FIELD_OPT(field, fmt) NUM_OPT(#field, fmt, &sbinfo->field) + +#define BIT_OPT(label, bitnr) \ + { \ + .name = label, \ + .type = OPT_BIT, \ + .u = { \ + .bit = { \ + .nr = bitnr, \ + .addr = &sbinfo->fs_flags \ + } \ + } \ + } + +#define MAX_NR_OPTIONS (30) + +/** + * reiser4_init_super_data - initialize reiser4 private super block + * @super: super block to initialize + * @opt_string: list of reiser4 mount options + * + * Sets various reiser4 parameters to default values. Parses mount options and + * overwrites default settings. + */ +int reiser4_init_super_data(struct super_block *super, char *opt_string) +{ + int result; + struct opt_desc *opts, *p; + reiser4_super_info_data *sbinfo = get_super_private(super); + + /* initialize super, export, dentry operations */ + sbinfo->ops.super = reiser4_super_operations; + sbinfo->ops.export = reiser4_export_operations; + sbinfo->ops.dentry = reiser4_dentry_operations; + super->s_op = &sbinfo->ops.super; + super->s_export_op = &sbinfo->ops.export; + + /* initialize transaction manager parameters to default values */ + sbinfo->tmgr.atom_max_size = totalram_pages / 4; + sbinfo->tmgr.atom_max_age = REISER4_ATOM_MAX_AGE / HZ; + sbinfo->tmgr.atom_min_size = 256; + sbinfo->tmgr.atom_max_flushers = ATOM_MAX_FLUSHERS; + + /* initialize cbk cache parameter */ + sbinfo->tree.cbk_cache.nr_slots = CBK_CACHE_SLOTS; + + /* initialize flush parameters */ + sbinfo->flush.relocate_threshold = FLUSH_RELOCATE_THRESHOLD; + sbinfo->flush.relocate_distance = FLUSH_RELOCATE_DISTANCE; + sbinfo->flush.written_threshold = FLUSH_WRITTEN_THRESHOLD; + sbinfo->flush.scan_maxnodes = FLUSH_SCAN_MAXNODES; + + sbinfo->optimal_io_size = REISER4_OPTIMAL_IO_SIZE; + + /* preliminary tree initializations */ + sbinfo->tree.super = super; + sbinfo->tree.carry.new_node_flags = REISER4_NEW_NODE_FLAGS; + sbinfo->tree.carry.new_extent_flags = REISER4_NEW_EXTENT_FLAGS; + sbinfo->tree.carry.paste_flags = REISER4_PASTE_FLAGS; + sbinfo->tree.carry.insert_flags = REISER4_INSERT_FLAGS; + rwlock_init(&(sbinfo->tree.tree_lock)); + spin_lock_init(&(sbinfo->tree.epoch_lock)); + + /* initialize default readahead params */ + sbinfo->ra_params.max = num_physpages / 4; + sbinfo->ra_params.flags = 0; + + /* allocate memory for structure describing reiser4 mount options */ + opts = kmalloc(sizeof(struct opt_desc) * MAX_NR_OPTIONS, + reiser4_ctx_gfp_mask_get()); + if (opts == NULL) + return RETERR(-ENOMEM); + + /* initialize structure describing reiser4 mount options */ + p = opts; + +#if REISER4_DEBUG +# define OPT_ARRAY_CHECK if ((p) > (opts) + MAX_NR_OPTIONS) { \ + warning("zam-1046", "opt array is overloaded"); break; \ + } +#else +# define OPT_ARRAY_CHECK noop +#endif + +#define PUSH_OPT(...) \ +do { \ + struct opt_desc o = __VA_ARGS__; \ + OPT_ARRAY_CHECK; \ + *p ++ = o; \ +} while (0) + +#define PUSH_SB_FIELD_OPT(field, format) PUSH_OPT(SB_FIELD_OPT(field, format)) +#define PUSH_BIT_OPT(name, bit) PUSH_OPT(BIT_OPT(name, bit)) + + /* + * tmgr.atom_max_size=N + * Atoms containing more than N blocks will be forced to commit. N is + * decimal. + */ + PUSH_SB_FIELD_OPT(tmgr.atom_max_size, "%u"); + /* + * tmgr.atom_max_age=N + * Atoms older than N seconds will be forced to commit. N is decimal. + */ + PUSH_SB_FIELD_OPT(tmgr.atom_max_age, "%u"); + /* + * tmgr.atom_min_size=N + * In committing an atom to free dirty pages, force the atom less than + * N in size to fuse with another one. + */ + PUSH_SB_FIELD_OPT(tmgr.atom_min_size, "%u"); + /* + * tmgr.atom_max_flushers=N + * limit of concurrent flushers for one atom. 0 means no limit. + */ + PUSH_SB_FIELD_OPT(tmgr.atom_max_flushers, "%u"); + /* + * tree.cbk_cache_slots=N + * Number of slots in the cbk cache. + */ + PUSH_SB_FIELD_OPT(tree.cbk_cache.nr_slots, "%u"); + /* + * If flush finds more than FLUSH_RELOCATE_THRESHOLD adjacent dirty + * leaf-level blocks it will force them to be relocated. + */ + PUSH_SB_FIELD_OPT(flush.relocate_threshold, "%u"); + /* + * If flush finds can find a block allocation closer than at most + * FLUSH_RELOCATE_DISTANCE from the preceder it will relocate to that + * position. + */ + PUSH_SB_FIELD_OPT(flush.relocate_distance, "%u"); + /* + * If we have written this much or more blocks before encountering busy + * jnode in flush list - abort flushing hoping that next time we get + * called this jnode will be clean already, and we will save some + * seeks. + */ + PUSH_SB_FIELD_OPT(flush.written_threshold, "%u"); + /* The maximum number of nodes to scan left on a level during flush. */ + PUSH_SB_FIELD_OPT(flush.scan_maxnodes, "%u"); + /* preferred IO size */ + PUSH_SB_FIELD_OPT(optimal_io_size, "%u"); + /* carry flags used for insertion of new nodes */ + PUSH_SB_FIELD_OPT(tree.carry.new_node_flags, "%u"); + /* carry flags used for insertion of new extents */ + PUSH_SB_FIELD_OPT(tree.carry.new_extent_flags, "%u"); + /* carry flags used for paste operations */ + PUSH_SB_FIELD_OPT(tree.carry.paste_flags, "%u"); + /* carry flags used for insert operations */ + PUSH_SB_FIELD_OPT(tree.carry.insert_flags, "%u"); + +#ifdef CONFIG_REISER4_BADBLOCKS + /* + * Alternative master superblock location in case if it's original + * location is not writeable/accessable. This is offset in BYTES. + */ + PUSH_SB_FIELD_OPT(altsuper, "%lu"); +#endif + + /* turn on BSD-style gid assignment */ + PUSH_BIT_OPT("bsdgroups", REISER4_BSD_GID); + /* turn on 32 bit times */ + PUSH_BIT_OPT("32bittimes", REISER4_32_BIT_TIMES); + /* + * Don't load all bitmap blocks at mount time, it is useful for + * machines with tiny RAM and large disks. + */ + PUSH_BIT_OPT("dont_load_bitmap", REISER4_DONT_LOAD_BITMAP); + /* disable transaction commits during write() */ + PUSH_BIT_OPT("atomic_write", REISER4_ATOMIC_WRITE); + /* disable use of write barriers in the reiser4 log writer. */ + PUSH_BIT_OPT("no_write_barrier", REISER4_NO_WRITE_BARRIER); + + PUSH_OPT( + { + /* + * tree traversal readahead parameters: + * -o readahead:MAXNUM:FLAGS + * MAXNUM - max number fo nodes to request readahead for: -1UL + * will set it to max_sane_readahead() + * FLAGS - combination of bits: RA_ADJCENT_ONLY, RA_ALL_LEVELS, + * CONTINUE_ON_PRESENT + */ + .name = "readahead", + .type = OPT_FORMAT, + .u = { + .f = { + .format = "%u:%u", + .nr_args = 2, + .arg1 = &sbinfo->ra_params.max, + .arg2 = &sbinfo->ra_params.flags, + .arg3 = NULL, + .arg4 = NULL + } + } + } + ); + + /* What to do in case of fs error */ + PUSH_OPT( + { + .name = "onerror", + .type = OPT_ONEOF, + .u = { + .oneof = { + .result = &sbinfo->onerror, + .list = { + "panic", "remount-ro", NULL + }, + } + } + } + ); + + /* modify default settings to values set by mount options */ + result = parse_options(opt_string, opts, p - opts); + kfree(opts); + if (result != 0) + return result; + + /* correct settings to sanity values */ + sbinfo->tmgr.atom_max_age *= HZ; + if (sbinfo->tmgr.atom_max_age <= 0) + /* overflow */ + sbinfo->tmgr.atom_max_age = REISER4_ATOM_MAX_AGE; + + /* round optimal io size up to 512 bytes */ + sbinfo->optimal_io_size >>= VFS_BLKSIZE_BITS; + sbinfo->optimal_io_size <<= VFS_BLKSIZE_BITS; + if (sbinfo->optimal_io_size == 0) { + warning("nikita-2497", "optimal_io_size is too small"); + return RETERR(-EINVAL); + } + return result; +} + +/** + * reiser4_init_read_super - read reiser4 master super block + * @super: super block to fill + * @silent: if 0 - print warnings + * + * Reads reiser4 master super block either from predefined location or from + * location specified by altsuper mount option, initializes disk format plugin. + */ +int reiser4_init_read_super(struct super_block *super, int silent) +{ + struct buffer_head *super_bh; + struct reiser4_master_sb *master_sb; + reiser4_super_info_data *sbinfo = get_super_private(super); + unsigned long blocksize; + + read_super_block: +#ifdef CONFIG_REISER4_BADBLOCKS + if (sbinfo->altsuper) + /* + * read reiser4 master super block at position specified by + * mount option + */ + super_bh = sb_bread(super, + (sector_t)(sbinfo->altsuper / super->s_blocksize)); + else +#endif + /* read reiser4 master super block at 16-th 4096 block */ + super_bh = sb_bread(super, + (sector_t)(REISER4_MAGIC_OFFSET / super->s_blocksize)); + if (!super_bh) + return RETERR(-EIO); + + master_sb = (struct reiser4_master_sb *)super_bh->b_data; + /* check reiser4 magic string */ + if (!strncmp(master_sb->magic, REISER4_SUPER_MAGIC_STRING, + sizeof(REISER4_SUPER_MAGIC_STRING))) { + /* reiser4 master super block contains filesystem blocksize */ + blocksize = le16_to_cpu(get_unaligned(&master_sb->blocksize)); + + if (blocksize != PAGE_CACHE_SIZE) { + /* + * currenly reiser4's blocksize must be equal to + * pagesize + */ + if (!silent) + warning("nikita-2609", + "%s: wrong block size %ld\n", super->s_id, + blocksize); + brelse(super_bh); + return RETERR(-EINVAL); + } + if (blocksize != super->s_blocksize) { + /* + * filesystem uses different blocksize. Reread master + * super block with correct blocksize + */ + brelse(super_bh); + if (!sb_set_blocksize(super, (int)blocksize)) + return RETERR(-EINVAL); + goto read_super_block; + } + + sbinfo->df_plug = + disk_format_plugin_by_id( + le16_to_cpu(get_unaligned(&master_sb->disk_plugin_id))); + if (sbinfo->df_plug == NULL) { + if (!silent) + warning("nikita-26091", + "%s: unknown disk format plugin %d\n", + super->s_id, + le16_to_cpu(get_unaligned(&master_sb->disk_plugin_id))); + brelse(super_bh); + return RETERR(-EINVAL); + } + sbinfo->diskmap_block = le64_to_cpu(get_unaligned(&master_sb->diskmap)); + brelse(super_bh); + return 0; + } + + /* there is no reiser4 on the device */ + if (!silent) + warning("nikita-2608", + "%s: wrong master super block magic", super->s_id); + brelse(super_bh); + return RETERR(-EINVAL); +} + +static struct { + reiser4_plugin_type type; + reiser4_plugin_id id; +} default_plugins[PSET_LAST] = { + [PSET_FILE] = { + .type = REISER4_FILE_PLUGIN_TYPE, + .id = UNIX_FILE_PLUGIN_ID + }, + [PSET_DIR] = { + .type = REISER4_DIR_PLUGIN_TYPE, + .id = HASHED_DIR_PLUGIN_ID + }, + [PSET_HASH] = { + .type = REISER4_HASH_PLUGIN_TYPE, + .id = R5_HASH_ID + }, + [PSET_FIBRATION] = { + .type = REISER4_FIBRATION_PLUGIN_TYPE, + .id = FIBRATION_DOT_O + }, + [PSET_PERM] = { + .type = REISER4_PERM_PLUGIN_TYPE, + .id = NULL_PERM_ID + }, + [PSET_FORMATTING] = { + .type = REISER4_FORMATTING_PLUGIN_TYPE, + .id = SMALL_FILE_FORMATTING_ID + }, + [PSET_SD] = { + .type = REISER4_ITEM_PLUGIN_TYPE, + .id = STATIC_STAT_DATA_ID + }, + [PSET_DIR_ITEM] = { + .type = REISER4_ITEM_PLUGIN_TYPE, + .id = COMPOUND_DIR_ID + }, + [PSET_CIPHER] = { + .type = REISER4_CIPHER_PLUGIN_TYPE, + .id = NONE_CIPHER_ID + }, + [PSET_DIGEST] = { + .type = REISER4_DIGEST_PLUGIN_TYPE, + .id = SHA256_32_DIGEST_ID + }, + [PSET_COMPRESSION] = { + .type = REISER4_COMPRESSION_PLUGIN_TYPE, + .id = LZO1_COMPRESSION_ID + }, + [PSET_COMPRESSION_MODE] = { + .type = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .id = CONVX_COMPRESSION_MODE_ID + }, + [PSET_CLUSTER] = { + .type = REISER4_CLUSTER_PLUGIN_TYPE, + .id = CLUSTER_64K_ID + }, + [PSET_CREATE] = { + .type = REISER4_FILE_PLUGIN_TYPE, + .id = UNIX_FILE_PLUGIN_ID + } +}; + +/* access to default plugin table */ +reiser4_plugin *get_default_plugin(pset_member memb) +{ + return plugin_by_id(default_plugins[memb].type, + default_plugins[memb].id); +} + +/** + * reiser4_init_root_inode - obtain inode of root directory + * @super: super block of filesystem + * + * Obtains inode of root directory (reading it from disk), initializes plugin + * set it was not initialized. + */ +int reiser4_init_root_inode(struct super_block *super) +{ + reiser4_super_info_data *sbinfo = get_super_private(super); + struct inode *inode; + int result = 0; + + inode = reiser4_iget(super, sbinfo->df_plug->root_dir_key(super), 0); + if (IS_ERR(inode)) + return RETERR(PTR_ERR(inode)); + + super->s_root = d_alloc_root(inode); + if (!super->s_root) { + iput(inode); + return RETERR(-ENOMEM); + } + + super->s_root->d_op = &sbinfo->ops.dentry; + + if (!is_inode_loaded(inode)) { + pset_member memb; + plugin_set *pset; + + pset = reiser4_inode_data(inode)->pset; + for (memb = 0; memb < PSET_LAST; ++memb) { + + if (aset_get(pset, memb) != NULL) + continue; + + result = grab_plugin_pset(inode, NULL, memb); + if (result != 0) + break; + + reiser4_inode_clr_flag(inode, REISER4_SDLEN_KNOWN); + } + + if (result == 0) { + if (REISER4_DEBUG) { + for (memb = 0; memb < PSET_LAST; ++memb) + assert("nikita-3500", + aset_get(pset, memb) != NULL); + } + } else + warning("nikita-3448", "Cannot set plugins of root: %i", + result); + reiser4_iget_complete(inode); + + /* As the default pset kept in the root dir may has been changed + (length is unknown), call update_sd. */ + if (!reiser4_inode_get_flag(inode, REISER4_SDLEN_KNOWN)) { + result = reiser4_grab_space( + inode_file_plugin(inode)->estimate.update(inode), + BA_CAN_COMMIT); + + if (result == 0) + result = reiser4_update_sd(inode); + + all_grabbed2free(); + } + } + + super->s_maxbytes = MAX_LFS_FILESIZE; + return result; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/inode.c linux-2.6.30/fs/reiser4/inode.c --- linux-2.6.30.orig/fs/reiser4/inode.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/inode.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,711 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Inode specific operations. */ + +#include "forward.h" +#include "debug.h" +#include "key.h" +#include "kassign.h" +#include "coord.h" +#include "seal.h" +#include "dscale.h" +#include "plugin/item/item.h" +#include "plugin/security/perm.h" +#include "plugin/plugin.h" +#include "plugin/object.h" +#include "znode.h" +#include "vfs_ops.h" +#include "inode.h" +#include "super.h" +#include "reiser4.h" + +#include /* for struct super_block, address_space */ + +/* return reiser4 internal tree which inode belongs to */ +/* Audited by: green(2002.06.17) */ +reiser4_tree *reiser4_tree_by_inode(const struct inode *inode/* inode queried*/) +{ + assert("nikita-256", inode != NULL); + assert("nikita-257", inode->i_sb != NULL); + return reiser4_get_tree(inode->i_sb); +} + +/* return reiser4-specific inode flags */ +static inline unsigned long *inode_flags(const struct inode *const inode) +{ + assert("nikita-2842", inode != NULL); + return &reiser4_inode_data(inode)->flags; +} + +/* set reiser4-specific flag @f in @inode */ +void reiser4_inode_set_flag(struct inode *inode, reiser4_file_plugin_flags f) +{ + assert("nikita-2248", inode != NULL); + set_bit((int)f, inode_flags(inode)); +} + +/* clear reiser4-specific flag @f in @inode */ +void reiser4_inode_clr_flag(struct inode *inode, reiser4_file_plugin_flags f) +{ + assert("nikita-2250", inode != NULL); + clear_bit((int)f, inode_flags(inode)); +} + +/* true if reiser4-specific flag @f is set in @inode */ +int reiser4_inode_get_flag(const struct inode *inode, + reiser4_file_plugin_flags f) +{ + assert("nikita-2251", inode != NULL); + return test_bit((int)f, inode_flags(inode)); +} + +/* convert oid to inode number */ +ino_t oid_to_ino(oid_t oid) +{ + return (ino_t) oid; +} + +/* convert oid to user visible inode number */ +ino_t oid_to_uino(oid_t oid) +{ + /* reiser4 object is uniquely identified by oid which is 64 bit + quantity. Kernel in-memory inode is indexed (in the hash table) by + 32 bit i_ino field, but this is not a problem, because there is a + way to further distinguish inodes with identical inode numbers + (find_actor supplied to iget()). + + But user space expects unique 32 bit inode number. Obviously this + is impossible. Work-around is to somehow hash oid into user visible + inode number. + */ + oid_t max_ino = (ino_t) ~0; + + if (REISER4_INO_IS_OID || (oid <= max_ino)) + return oid; + else + /* this is remotely similar to algorithm used to find next pid + to use for process: after wrap-around start from some + offset rather than from 0. Idea is that there are some long + living objects with which we don't want to collide. + */ + return REISER4_UINO_SHIFT + ((oid - max_ino) & (max_ino >> 1)); +} + +/* check that "inode" is on reiser4 file-system */ +int is_reiser4_inode(const struct inode *inode/* inode queried */) +{ + return inode != NULL && is_reiser4_super(inode->i_sb); +} + +/* Maximal length of a name that can be stored in directory @inode. + + This is used in check during file creation and lookup. */ +int reiser4_max_filename_len(const struct inode *inode/* inode queried */) +{ + assert("nikita-287", is_reiser4_inode(inode)); + assert("nikita-1710", inode_dir_item_plugin(inode)); + if (inode_dir_item_plugin(inode)->s.dir.max_name_len) + return inode_dir_item_plugin(inode)->s.dir.max_name_len(inode); + else + return 255; +} + +#if REISER4_USE_COLLISION_LIMIT +/* Maximal number of hash collisions for this directory. */ +int max_hash_collisions(const struct inode *dir/* inode queried */) +{ + assert("nikita-1711", dir != NULL); + return reiser4_inode_data(dir)->plugin.max_collisions; +} +#endif /* REISER4_USE_COLLISION_LIMIT */ + +/* Install file, inode, and address_space operation on @inode, depending on + its mode. */ +int setup_inode_ops(struct inode *inode /* inode to intialize */ , + reiser4_object_create_data * data /* parameters to create + * object */ ) +{ + reiser4_super_info_data *sinfo; + file_plugin *fplug; + dir_plugin *dplug; + + fplug = inode_file_plugin(inode); + dplug = inode_dir_plugin(inode); + + sinfo = get_super_private(inode->i_sb); + + switch (inode->i_mode & S_IFMT) { + case S_IFSOCK: + case S_IFBLK: + case S_IFCHR: + case S_IFIFO: + { + dev_t rdev; /* to keep gcc happy */ + + assert("vs-46", fplug != NULL); + /* ugly hack with rdev */ + if (data == NULL) { + rdev = inode->i_rdev; + inode->i_rdev = 0; + } else + rdev = data->rdev; + inode->i_blocks = 0; + assert("vs-42", fplug->h.id == SPECIAL_FILE_PLUGIN_ID); + inode->i_op = file_plugins[fplug->h.id].inode_ops; + /* initialize inode->i_fop and inode->i_rdev for block + and char devices */ + init_special_inode(inode, inode->i_mode, rdev); + /* all address space operations are null */ + inode->i_mapping->a_ops = + file_plugins[fplug->h.id].as_ops; + break; + } + case S_IFLNK: + assert("vs-46", fplug != NULL); + assert("vs-42", fplug->h.id == SYMLINK_FILE_PLUGIN_ID); + inode->i_op = file_plugins[fplug->h.id].inode_ops; + inode->i_fop = NULL; + /* all address space operations are null */ + inode->i_mapping->a_ops = file_plugins[fplug->h.id].as_ops; + break; + case S_IFDIR: + assert("vs-46", dplug != NULL); + assert("vs-43", (dplug->h.id == HASHED_DIR_PLUGIN_ID || + dplug->h.id == SEEKABLE_HASHED_DIR_PLUGIN_ID)); + inode->i_op = dir_plugins[dplug->h.id].inode_ops; + inode->i_fop = dir_plugins[dplug->h.id].file_ops; + inode->i_mapping->a_ops = dir_plugins[dplug->h.id].as_ops; + break; + case S_IFREG: + assert("vs-46", fplug != NULL); + assert("vs-43", (fplug->h.id == UNIX_FILE_PLUGIN_ID || + fplug->h.id == CRYPTCOMPRESS_FILE_PLUGIN_ID)); + inode->i_op = file_plugins[fplug->h.id].inode_ops; + inode->i_fop = file_plugins[fplug->h.id].file_ops; + inode->i_mapping->a_ops = file_plugins[fplug->h.id].as_ops; + break; + default: + warning("nikita-291", "wrong file mode: %o for %llu", + inode->i_mode, + (unsigned long long)get_inode_oid(inode)); + reiser4_make_bad_inode(inode); + return RETERR(-EINVAL); + } + return 0; +} + +/* Initialize inode from disk data. Called with inode locked. + Return inode locked. */ +static int init_inode(struct inode *inode /* inode to intialise */ , + coord_t *coord/* coord of stat data */) +{ + int result; + item_plugin *iplug; + void *body; + int length; + reiser4_inode *state; + + assert("nikita-292", coord != NULL); + assert("nikita-293", inode != NULL); + + coord_clear_iplug(coord); + result = zload(coord->node); + if (result) + return result; + iplug = item_plugin_by_coord(coord); + body = item_body_by_coord(coord); + length = item_length_by_coord(coord); + + assert("nikita-295", iplug != NULL); + assert("nikita-296", body != NULL); + assert("nikita-297", length > 0); + + /* inode is under I_LOCK now */ + + state = reiser4_inode_data(inode); + /* call stat-data plugin method to load sd content into inode */ + result = iplug->s.sd.init_inode(inode, body, length); + set_plugin(&state->pset, PSET_SD, item_plugin_to_plugin(iplug)); + if (result == 0) { + result = setup_inode_ops(inode, NULL); + if (result == 0 && inode->i_sb->s_root && + inode->i_sb->s_root->d_inode) + result = finish_pset(inode); + } + zrelse(coord->node); + return result; +} + +/* read `inode' from the disk. This is what was previously in + reiserfs_read_inode2(). + + Must be called with inode locked. Return inode still locked. +*/ +static int read_inode(struct inode *inode /* inode to read from disk */ , + const reiser4_key * key /* key of stat data */ , + int silent) +{ + int result; + lock_handle lh; + reiser4_inode *info; + coord_t coord; + + assert("nikita-298", inode != NULL); + assert("nikita-1945", !is_inode_loaded(inode)); + + info = reiser4_inode_data(inode); + assert("nikita-300", info->locality_id != 0); + + coord_init_zero(&coord); + init_lh(&lh); + /* locate stat-data in a tree and return znode locked */ + result = lookup_sd(inode, ZNODE_READ_LOCK, &coord, &lh, key, silent); + assert("nikita-301", !is_inode_loaded(inode)); + if (result == 0) { + /* use stat-data plugin to load sd into inode. */ + result = init_inode(inode, &coord); + if (result == 0) { + /* initialize stat-data seal */ + spin_lock_inode(inode); + reiser4_seal_init(&info->sd_seal, &coord, key); + info->sd_coord = coord; + spin_unlock_inode(inode); + + /* call file plugin's method to initialize plugin + * specific part of inode */ + if (inode_file_plugin(inode)->init_inode_data) + inode_file_plugin(inode)->init_inode_data(inode, + NULL, + 0); + /* load detached directory cursors for stateless + * directory readers (NFS). */ + reiser4_load_cursors(inode); + + /* Check the opened inode for consistency. */ + result = + get_super_private(inode->i_sb)->df_plug-> + check_open(inode); + } + } + /* lookup_sd() doesn't release coord because we want znode + stay read-locked while stat-data fields are accessed in + init_inode() */ + done_lh(&lh); + + if (result != 0) + reiser4_make_bad_inode(inode); + return result; +} + +/* initialise new reiser4 inode being inserted into hash table. */ +static int init_locked_inode(struct inode *inode /* new inode */ , + void *opaque /* key of stat data passed to + * the iget5_locked as cookie */) +{ + reiser4_key *key; + + assert("nikita-1995", inode != NULL); + assert("nikita-1996", opaque != NULL); + key = opaque; + set_inode_oid(inode, get_key_objectid(key)); + reiser4_inode_data(inode)->locality_id = get_key_locality(key); + return 0; +} + +/* reiser4_inode_find_actor() - "find actor" supplied by reiser4 to + iget5_locked(). + + This function is called by iget5_locked() to distinguish reiser4 inodes + having the same inode numbers. Such inodes can only exist due to some error + condition. One of them should be bad. Inodes with identical inode numbers + (objectids) are distinguished by their packing locality. + +*/ +static int reiser4_inode_find_actor(struct inode *inode /* inode from hash table + * to check */ , + void *opaque /* "cookie" passed to + * iget5_locked(). This + * is stat-data key */) +{ + reiser4_key *key; + + key = opaque; + return + /* oid is unique, so first term is enough, actually. */ + get_inode_oid(inode) == get_key_objectid(key) && + /* + * also, locality should be checked, but locality is stored in + * the reiser4-specific part of the inode, and actor can be + * called against arbitrary inode that happened to be in this + * hash chain. Hence we first have to check that this is + * reiser4 inode at least. is_reiser4_inode() is probably too + * early to call, as inode may have ->i_op not yet + * initialised. + */ + is_reiser4_super(inode->i_sb) && + /* + * usually objectid is unique, but pseudo files use counter to + * generate objectid. All pseudo files are placed into special + * (otherwise unused) locality. + */ + reiser4_inode_data(inode)->locality_id == get_key_locality(key); +} + +/* hook for kmem_cache_create */ +void loading_init_once(reiser4_inode * info) +{ + mutex_init(&info->loading); +} + +/* for reiser4_alloc_inode */ +void loading_alloc(reiser4_inode * info) +{ + assert("vs-1717", !mutex_is_locked(&info->loading)); +} + +/* for reiser4_destroy */ +void loading_destroy(reiser4_inode * info) +{ + assert("vs-1717a", !mutex_is_locked(&info->loading)); +} + +static void loading_begin(reiser4_inode * info) +{ + mutex_lock(&info->loading); +} + +static void loading_end(reiser4_inode * info) +{ + mutex_unlock(&info->loading); +} + +/** + * reiser4_iget - obtain inode via iget5_locked, read from disk if necessary + * @super: super block of filesystem + * @key: key of inode's stat-data + * @silent: + * + * This is our helper function a la iget(). This is be called by + * lookup_common() and reiser4_read_super(). Return inode locked or error + * encountered. + */ +struct inode *reiser4_iget(struct super_block *super, const reiser4_key *key, + int silent) +{ + struct inode *inode; + int result; + reiser4_inode *info; + + assert("nikita-302", super != NULL); + assert("nikita-303", key != NULL); + + result = 0; + + /* call iget(). Our ->read_inode() is dummy, so this will either + find inode in cache or return uninitialised inode */ + inode = iget5_locked(super, + (unsigned long)get_key_objectid(key), + reiser4_inode_find_actor, + init_locked_inode, (reiser4_key *) key); + if (inode == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + if (is_bad_inode(inode)) { + warning("nikita-304", "Bad inode found"); + reiser4_print_key("key", key); + iput(inode); + return ERR_PTR(RETERR(-EIO)); + } + + info = reiser4_inode_data(inode); + + /* Reiser4 inode state bit REISER4_LOADED is used to distinguish fully + loaded and initialized inode from just allocated inode. If + REISER4_LOADED bit is not set, reiser4_iget() completes loading under + info->loading. The place in reiser4 which uses not initialized inode + is the reiser4 repacker, see repacker-related functions in + plugin/item/extent.c */ + if (!is_inode_loaded(inode)) { + loading_begin(info); + if (!is_inode_loaded(inode)) { + /* locking: iget5_locked returns locked inode */ + assert("nikita-1941", !is_inode_loaded(inode)); + assert("nikita-1949", + reiser4_inode_find_actor(inode, + (reiser4_key *) key)); + /* now, inode has objectid as ->i_ino and locality in + reiser4-specific part. This is enough for + read_inode() to read stat data from the disk */ + result = read_inode(inode, key, silent); + } else + loading_end(info); + } + + if (inode->i_state & I_NEW) + unlock_new_inode(inode); + + if (is_bad_inode(inode)) { + assert("vs-1717", result != 0); + loading_end(info); + iput(inode); + inode = ERR_PTR(result); + } else if (REISER4_DEBUG) { + reiser4_key found_key; + + assert("vs-1717", result == 0); + build_sd_key(inode, &found_key); + if (!keyeq(&found_key, key)) { + warning("nikita-305", "Wrong key in sd"); + reiser4_print_key("sought for", key); + reiser4_print_key("found", &found_key); + } + if (inode->i_nlink == 0) { + warning("nikita-3559", "Unlinked inode found: %llu\n", + (unsigned long long)get_inode_oid(inode)); + } + } + return inode; +} + +/* reiser4_iget() may return not fully initialized inode, this function should + * be called after one completes reiser4 inode initializing. */ +void reiser4_iget_complete(struct inode *inode) +{ + assert("zam-988", is_reiser4_inode(inode)); + + if (!is_inode_loaded(inode)) { + reiser4_inode_set_flag(inode, REISER4_LOADED); + loading_end(reiser4_inode_data(inode)); + } +} + +void reiser4_make_bad_inode(struct inode *inode) +{ + assert("nikita-1934", inode != NULL); + + /* clear LOADED bit */ + reiser4_inode_clr_flag(inode, REISER4_LOADED); + make_bad_inode(inode); + return; +} + +file_plugin *inode_file_plugin(const struct inode *inode) +{ + assert("nikita-1997", inode != NULL); + return reiser4_inode_data(inode)->pset->file; +} + +dir_plugin *inode_dir_plugin(const struct inode *inode) +{ + assert("nikita-1998", inode != NULL); + return reiser4_inode_data(inode)->pset->dir; +} + +formatting_plugin *inode_formatting_plugin(const struct inode *inode) +{ + assert("nikita-2000", inode != NULL); + return reiser4_inode_data(inode)->pset->formatting; +} + +hash_plugin *inode_hash_plugin(const struct inode *inode) +{ + assert("nikita-2001", inode != NULL); + return reiser4_inode_data(inode)->pset->hash; +} + +fibration_plugin *inode_fibration_plugin(const struct inode *inode) +{ + assert("nikita-2001", inode != NULL); + return reiser4_inode_data(inode)->pset->fibration; +} + +cipher_plugin *inode_cipher_plugin(const struct inode *inode) +{ + assert("edward-36", inode != NULL); + return reiser4_inode_data(inode)->pset->cipher; +} + +compression_plugin *inode_compression_plugin(const struct inode *inode) +{ + assert("edward-37", inode != NULL); + return reiser4_inode_data(inode)->pset->compression; +} + +compression_mode_plugin *inode_compression_mode_plugin(const struct inode * + inode) +{ + assert("edward-1330", inode != NULL); + return reiser4_inode_data(inode)->pset->compression_mode; +} + +cluster_plugin *inode_cluster_plugin(const struct inode *inode) +{ + assert("edward-1328", inode != NULL); + return reiser4_inode_data(inode)->pset->cluster; +} + +file_plugin *inode_create_plugin(const struct inode *inode) +{ + assert("edward-1329", inode != NULL); + return reiser4_inode_data(inode)->pset->create; +} + +digest_plugin *inode_digest_plugin(const struct inode *inode) +{ + assert("edward-86", inode != NULL); + return reiser4_inode_data(inode)->pset->digest; +} + +item_plugin *inode_sd_plugin(const struct inode *inode) +{ + assert("vs-534", inode != NULL); + return reiser4_inode_data(inode)->pset->sd; +} + +item_plugin *inode_dir_item_plugin(const struct inode *inode) +{ + assert("vs-534", inode != NULL); + return reiser4_inode_data(inode)->pset->dir_item; +} + +file_plugin *child_create_plugin(const struct inode *inode) +{ + assert("edward-1329", inode != NULL); + return reiser4_inode_data(inode)->hset->create; +} + +void inode_set_extension(struct inode *inode, sd_ext_bits ext) +{ + reiser4_inode *state; + + assert("nikita-2716", inode != NULL); + assert("nikita-2717", ext < LAST_SD_EXTENSION); + assert("nikita-3491", spin_inode_is_locked(inode)); + + state = reiser4_inode_data(inode); + state->extmask |= 1 << ext; + /* force re-calculation of stat-data length on next call to + update_sd(). */ + reiser4_inode_clr_flag(inode, REISER4_SDLEN_KNOWN); +} + +void inode_clr_extension(struct inode *inode, sd_ext_bits ext) +{ + reiser4_inode *state; + + assert("vpf-1926", inode != NULL); + assert("vpf-1927", ext < LAST_SD_EXTENSION); + assert("vpf-1928", spin_inode_is_locked(inode)); + + state = reiser4_inode_data(inode); + state->extmask &= ~(1 << ext); + /* force re-calculation of stat-data length on next call to + update_sd(). */ + reiser4_inode_clr_flag(inode, REISER4_SDLEN_KNOWN); +} + +void inode_check_scale_nolock(struct inode *inode, __u64 old, __u64 new) +{ + assert("edward-1287", inode != NULL); + if (!dscale_fit(old, new)) + reiser4_inode_clr_flag(inode, REISER4_SDLEN_KNOWN); + return; +} + +void inode_check_scale(struct inode *inode, __u64 old, __u64 new) +{ + assert("nikita-2875", inode != NULL); + spin_lock_inode(inode); + inode_check_scale_nolock(inode, old, new); + spin_unlock_inode(inode); +} + +/* + * initialize ->ordering field of inode. This field defines how file stat-data + * and body is ordered within a tree with respect to other objects within the + * same parent directory. + */ +void +init_inode_ordering(struct inode *inode, + reiser4_object_create_data * crd, int create) +{ + reiser4_key key; + + if (create) { + struct inode *parent; + + parent = crd->parent; + assert("nikita-3224", inode_dir_plugin(parent) != NULL); + inode_dir_plugin(parent)->build_entry_key(parent, + &crd->dentry->d_name, + &key); + } else { + coord_t *coord; + + coord = &reiser4_inode_data(inode)->sd_coord; + coord_clear_iplug(coord); + /* safe to use ->sd_coord, because node is under long term + * lock */ + WITH_DATA(coord->node, item_key_by_coord(coord, &key)); + } + + set_inode_ordering(inode, get_key_ordering(&key)); +} + +znode *inode_get_vroot(struct inode *inode) +{ + reiser4_block_nr blk; + znode *result; + + spin_lock_inode(inode); + blk = reiser4_inode_data(inode)->vroot; + spin_unlock_inode(inode); + if (!disk_addr_eq(&UBER_TREE_ADDR, &blk)) + result = zlook(reiser4_tree_by_inode(inode), &blk); + else + result = NULL; + return result; +} + +void inode_set_vroot(struct inode *inode, znode *vroot) +{ + spin_lock_inode(inode); + reiser4_inode_data(inode)->vroot = *znode_get_block(vroot); + spin_unlock_inode(inode); +} + +#if REISER4_DEBUG + +void reiser4_inode_invariant(const struct inode *inode) +{ + assert("nikita-3077", spin_inode_is_locked(inode)); +} + +int inode_has_no_jnodes(reiser4_inode * r4_inode) +{ + return jnode_tree_by_reiser4_inode(r4_inode)->rnode == NULL && + r4_inode->nr_jnodes == 0; +} + +#endif + +/* true if directory is empty (only contains dot and dotdot) */ +/* FIXME: shouldn't it be dir plugin method? */ +int is_dir_empty(const struct inode *dir) +{ + assert("nikita-1976", dir != NULL); + + /* rely on our method to maintain directory i_size being equal to the + number of entries. */ + return dir->i_size <= 2 ? 0 : RETERR(-ENOTEMPTY); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/inode.h linux-2.6.30/fs/reiser4/inode.h --- linux-2.6.30.orig/fs/reiser4/inode.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/inode.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,453 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ + +/* Inode functions. */ + +#if !defined(__REISER4_INODE_H__) +#define __REISER4_INODE_H__ + +#include "forward.h" +#include "debug.h" +#include "key.h" +#include "seal.h" +#include "plugin/plugin.h" +#include "plugin/file/cryptcompress.h" +#include "plugin/file/file.h" +#include "plugin/dir/dir.h" +#include "plugin/plugin_set.h" +#include "plugin/security/perm.h" +#include "vfs_ops.h" +#include "jnode.h" +#include "fsdata.h" + +#include /* for __u?? , ino_t */ +#include /* for struct super_block, struct + * rw_semaphore, etc */ +#include +#include + +/* reiser4-specific inode flags. They are "transient" and are not + supposed to be stored on disk. Used to trace "state" of + inode +*/ +typedef enum { + /* this is light-weight inode, inheriting some state from its + parent */ + REISER4_LIGHT_WEIGHT = 0, + /* stat data wasn't yet created */ + REISER4_NO_SD = 1, + /* internal immutable flag. Currently is only used + to avoid race condition during file creation. + See comment in create_object(). */ + REISER4_IMMUTABLE = 2, + /* inode was read from storage */ + REISER4_LOADED = 3, + /* this bit is set for symlinks. inode->i_private points to target + name of symlink. */ + REISER4_GENERIC_PTR_USED = 4, + /* set if size of stat-data item for this inode is known. If this is + * set we can avoid recalculating size of stat-data on each update. */ + REISER4_SDLEN_KNOWN = 5, + /* reiser4_inode->crypt points to the crypto stat */ + REISER4_CRYPTO_STAT_LOADED = 6, + /* cryptcompress_inode_data points to the secret key */ + REISER4_SECRET_KEY_INSTALLED = 7, + /* File (possibly) has pages corresponding to the tail items, that + * were created by ->readpage. It is set by mmap_unix_file() and + * sendfile_unix_file(). This bit is inspected by write_unix_file and + * kill-hook of tail items. It is never cleared once set. This bit is + * modified and inspected under i_mutex. */ + REISER4_HAS_MMAP = 8, + REISER4_PART_MIXED = 9, + REISER4_PART_IN_CONV = 10, + /* This flag indicates that file plugin conversion is in progress */ + REISER4_FILE_CONV_IN_PROGRESS = 11 +} reiser4_file_plugin_flags; + +/* state associated with each inode. + reiser4 inode. + + NOTE-NIKITA In 2.5 kernels it is not necessary that all file-system inodes + be of the same size. File-system allocates inodes by itself through + s_op->allocate_inode() method. So, it is possible to adjust size of inode + at the time of its creation. + + Invariants involving parts of this data-type: + + [inode->eflushed] + +*/ + +typedef struct reiser4_inode reiser4_inode; +/* return pointer to reiser4-specific part of inode */ +static inline reiser4_inode *reiser4_inode_data(const struct inode *inode + /* inode queried */ ); + +#if BITS_PER_LONG == 64 + +#define REISER4_INO_IS_OID (1) +typedef struct {; +} oid_hi_t; + +/* BITS_PER_LONG == 64 */ +#else + +#define REISER4_INO_IS_OID (0) +typedef __u32 oid_hi_t; + +/* BITS_PER_LONG == 64 */ +#endif + +struct reiser4_inode { + /* spin lock protecting fields of this structure. */ + spinlock_t guard; + /* main plugin set that control the file + (see comments in plugin/plugin_set.c) */ + plugin_set *pset; + /* plugin set for inheritance + (see comments in plugin/plugin_set.c) */ + plugin_set *hset; + /* high 32 bits of object id */ + oid_hi_t oid_hi; + /* seal for stat-data */ + seal_t sd_seal; + /* locality id for this file */ + oid_t locality_id; +#if REISER4_LARGE_KEY + __u64 ordering; +#endif + /* coord of stat-data in sealed node */ + coord_t sd_coord; + /* bit-mask of stat-data extentions used by this file */ + __u64 extmask; + /* bitmask of non-default plugins for this inode */ + __u16 plugin_mask; + /* bitmask of set heir plugins for this inode. */ + __u16 heir_mask; + union { + struct list_head readdir_list; + struct list_head not_used; + } lists; + /* per-inode flags. Filled by values of reiser4_file_plugin_flags */ + unsigned long flags; + union { + /* fields specific to unix_file plugin */ + struct unix_file_info unix_file_info; + /* fields specific to cryptcompress file plugin */ + struct cryptcompress_info cryptcompress_info; + } file_plugin_data; + + /* this semaphore is to serialize readers and writers of @pset->file + * when file plugin conversion is enabled + */ + struct rw_semaphore conv_sem; + + /* tree of jnodes. Phantom jnodes (ones not attched to any atom) are + tagged in that tree by EFLUSH_TAG_ANONYMOUS */ + struct radix_tree_root jnodes_tree; +#if REISER4_DEBUG + /* number of unformatted node jnodes of this file in jnode hash table */ + unsigned long nr_jnodes; +#endif + + /* block number of virtual root for this object. See comment above + * fs/reiser4/search.c:handle_vroot() */ + reiser4_block_nr vroot; + struct mutex loading; +}; + +void loading_init_once(reiser4_inode *); +void loading_alloc(reiser4_inode *); +void loading_destroy(reiser4_inode *); + +struct reiser4_inode_object { + /* private part */ + reiser4_inode p; + /* generic fields not specific to reiser4, but used by VFS */ + struct inode vfs_inode; +}; + +/* return pointer to the reiser4 specific portion of @inode */ +static inline reiser4_inode *reiser4_inode_data(const struct inode *inode + /* inode queried */ ) +{ + assert("nikita-254", inode != NULL); + return &container_of(inode, struct reiser4_inode_object, vfs_inode)->p; +} + +static inline struct inode *inode_by_reiser4_inode(const reiser4_inode * + r4_inode /* inode queried */ + ) +{ + return &container_of(r4_inode, struct reiser4_inode_object, + p)->vfs_inode; +} + +/* + * reiser4 inodes are identified by 64bit object-id (oid_t), but in struct + * inode ->i_ino field is of type ino_t (long) that can be either 32 or 64 + * bits. + * + * If ->i_ino is 32 bits we store remaining 32 bits in reiser4 specific part + * of inode, otherwise whole oid is stored in i_ino. + * + * Wrappers below ([sg]et_inode_oid()) are used to hide this difference. + */ + +#define OID_HI_SHIFT (sizeof(ino_t) * 8) + +#if REISER4_INO_IS_OID + +static inline oid_t get_inode_oid(const struct inode *inode) +{ + return inode->i_ino; +} + +static inline void set_inode_oid(struct inode *inode, oid_t oid) +{ + inode->i_ino = oid; +} + +/* REISER4_INO_IS_OID */ +#else + +static inline oid_t get_inode_oid(const struct inode *inode) +{ + return + ((__u64) reiser4_inode_data(inode)->oid_hi << OID_HI_SHIFT) | + inode->i_ino; +} + +static inline void set_inode_oid(struct inode *inode, oid_t oid) +{ + assert("nikita-2519", inode != NULL); + inode->i_ino = (ino_t) (oid); + reiser4_inode_data(inode)->oid_hi = (oid) >> OID_HI_SHIFT; + assert("nikita-2521", get_inode_oid(inode) == (oid)); +} + +/* REISER4_INO_IS_OID */ +#endif + +static inline oid_t get_inode_locality(const struct inode *inode) +{ + return reiser4_inode_data(inode)->locality_id; +} + +#if REISER4_LARGE_KEY +static inline __u64 get_inode_ordering(const struct inode *inode) +{ + return reiser4_inode_data(inode)->ordering; +} + +static inline void set_inode_ordering(const struct inode *inode, __u64 ordering) +{ + reiser4_inode_data(inode)->ordering = ordering; +} + +#else + +#define get_inode_ordering(inode) (0) +#define set_inode_ordering(inode, val) noop + +#endif + +/* return inode in which @uf_info is embedded */ +static inline struct inode * +unix_file_info_to_inode(const struct unix_file_info *uf_info) +{ + return &container_of(uf_info, struct reiser4_inode_object, + p.file_plugin_data.unix_file_info)->vfs_inode; +} + +extern ino_t oid_to_ino(oid_t oid) __attribute__ ((const)); +extern ino_t oid_to_uino(oid_t oid) __attribute__ ((const)); + +extern reiser4_tree *reiser4_tree_by_inode(const struct inode *inode); + +#if REISER4_DEBUG +extern void reiser4_inode_invariant(const struct inode *inode); +extern int inode_has_no_jnodes(reiser4_inode *); +#else +#define reiser4_inode_invariant(inode) noop +#endif + +static inline int spin_inode_is_locked(const struct inode *inode) +{ + assert_spin_locked(&reiser4_inode_data(inode)->guard); + return 1; +} + +/** + * spin_lock_inode - lock reiser4_inode' embedded spinlock + * @inode: inode to lock + * + * In debug mode it checks that lower priority locks are not held and + * increments reiser4_context's lock counters on which lock ordering checking + * is based. + */ +static inline void spin_lock_inode(struct inode *inode) +{ + assert("", LOCK_CNT_NIL(spin_locked)); + /* check lock ordering */ + assert_spin_not_locked(&d_lock); + + spin_lock(&reiser4_inode_data(inode)->guard); + + LOCK_CNT_INC(spin_locked_inode); + LOCK_CNT_INC(spin_locked); + + reiser4_inode_invariant(inode); +} + +/** + * spin_unlock_inode - unlock reiser4_inode' embedded spinlock + * @inode: inode to unlock + * + * In debug mode it checks that spinlock is held and decrements + * reiser4_context's lock counters on which lock ordering checking is based. + */ +static inline void spin_unlock_inode(struct inode *inode) +{ + assert_spin_locked(&reiser4_inode_data(inode)->guard); + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_inode)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + reiser4_inode_invariant(inode); + + LOCK_CNT_DEC(spin_locked_inode); + LOCK_CNT_DEC(spin_locked); + + spin_unlock(&reiser4_inode_data(inode)->guard); +} + +extern znode *inode_get_vroot(struct inode *inode); +extern void inode_set_vroot(struct inode *inode, znode * vroot); + +extern int reiser4_max_filename_len(const struct inode *inode); +extern int max_hash_collisions(const struct inode *dir); +extern void reiser4_unlock_inode(struct inode *inode); +extern int is_reiser4_inode(const struct inode *inode); +extern int setup_inode_ops(struct inode *inode, reiser4_object_create_data *); +extern struct inode *reiser4_iget(struct super_block *super, + const reiser4_key * key, int silent); +extern void reiser4_iget_complete(struct inode *inode); +extern void reiser4_inode_set_flag(struct inode *inode, + reiser4_file_plugin_flags f); +extern void reiser4_inode_clr_flag(struct inode *inode, + reiser4_file_plugin_flags f); +extern int reiser4_inode_get_flag(const struct inode *inode, + reiser4_file_plugin_flags f); + +/* has inode been initialized? */ +static inline int +is_inode_loaded(const struct inode *inode/* inode queried */) +{ + assert("nikita-1120", inode != NULL); + return reiser4_inode_get_flag(inode, REISER4_LOADED); +} + +extern file_plugin *inode_file_plugin(const struct inode *inode); +extern dir_plugin *inode_dir_plugin(const struct inode *inode); +extern formatting_plugin *inode_formatting_plugin(const struct inode *inode); +extern hash_plugin *inode_hash_plugin(const struct inode *inode); +extern fibration_plugin *inode_fibration_plugin(const struct inode *inode); +extern cipher_plugin *inode_cipher_plugin(const struct inode *inode); +extern digest_plugin *inode_digest_plugin(const struct inode *inode); +extern compression_plugin *inode_compression_plugin(const struct inode *inode); +extern compression_mode_plugin *inode_compression_mode_plugin(const struct inode + *inode); +extern cluster_plugin *inode_cluster_plugin(const struct inode *inode); +extern file_plugin *inode_create_plugin(const struct inode *inode); +extern item_plugin *inode_sd_plugin(const struct inode *inode); +extern item_plugin *inode_dir_item_plugin(const struct inode *inode); +extern file_plugin *child_create_plugin(const struct inode *inode); + +extern void reiser4_make_bad_inode(struct inode *inode); + +extern void inode_set_extension(struct inode *inode, sd_ext_bits ext); +extern void inode_clr_extension(struct inode *inode, sd_ext_bits ext); +extern void inode_check_scale(struct inode *inode, __u64 old, __u64 new); +extern void inode_check_scale_nolock(struct inode *inode, __u64 old, __u64 new); + +#define INODE_SET_SIZE(i, value) \ +({ \ + struct inode *__i; \ + typeof(value) __v; \ + \ + __i = (i); \ + __v = (value); \ + inode_check_scale(__i, __i->i_size, __v); \ + i_size_write(__i, __v); \ +}) + +/* + * update field @field in inode @i to contain value @value. + */ +#define INODE_SET_FIELD(i, field, value) \ +({ \ + struct inode *__i; \ + typeof(value) __v; \ + \ + __i = (i); \ + __v = (value); \ + inode_check_scale(__i, __i->field, __v); \ + __i->field = __v; \ +}) + +#define INODE_INC_FIELD(i, field) \ +({ \ + struct inode *__i; \ + \ + __i = (i); \ + inode_check_scale(__i, __i->field, __i->field + 1); \ + ++ __i->field; \ +}) + +#define INODE_DEC_FIELD(i, field) \ +({ \ + struct inode *__i; \ + \ + __i = (i); \ + inode_check_scale(__i, __i->field, __i->field - 1); \ + -- __i->field; \ +}) + +/* See comment before reiser4_readdir_common() for description. */ +static inline struct list_head *get_readdir_list(const struct inode *inode) +{ + return &reiser4_inode_data(inode)->lists.readdir_list; +} + +extern void init_inode_ordering(struct inode *inode, + reiser4_object_create_data * crd, int create); + +static inline struct radix_tree_root *jnode_tree_by_inode(struct inode *inode) +{ + return &reiser4_inode_data(inode)->jnodes_tree; +} + +static inline struct radix_tree_root *jnode_tree_by_reiser4_inode(reiser4_inode + *r4_inode) +{ + return &r4_inode->jnodes_tree; +} + +#if REISER4_DEBUG +extern void print_inode(const char *prefix, const struct inode *i); +#endif + +int is_dir_empty(const struct inode *); + +/* __REISER4_INODE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/ioctl.h linux-2.6.30/fs/reiser4/ioctl.h --- linux-2.6.30.orig/fs/reiser4/ioctl.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/ioctl.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,41 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +#if !defined(__REISER4_IOCTL_H__) +#define __REISER4_IOCTL_H__ + +#include + +/* + * ioctl(2) command used to "unpack" reiser4 file, that is, convert it into + * extents and fix in this state. This is used by applications that rely on + * + * . files being block aligned, and + * + * . files never migrating on disk + * + * for example, boot loaders (LILO) need this. + * + * This ioctl should be used as + * + * result = ioctl(fd, REISER4_IOC_UNPACK); + * + * File behind fd descriptor will be converted to the extents (if necessary), + * and its stat-data will be updated so that it will never be converted back + * into tails again. + */ +#define REISER4_IOC_UNPACK _IOW(0xCD, 1, long) + +/* __REISER4_IOCTL_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/jnode.c linux-2.6.30/fs/reiser4/jnode.c --- linux-2.6.30.orig/fs/reiser4/jnode.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/jnode.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1923 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ +/* Jnode manipulation functions. */ +/* Jnode is entity used to track blocks with data and meta-data in reiser4. + + In particular, jnodes are used to track transactional information + associated with each block. Each znode contains jnode as ->zjnode field. + + Jnode stands for either Josh or Journal node. +*/ + +/* + * Taxonomy. + * + * Jnode represents block containing data or meta-data. There are jnodes + * for: + * + * unformatted blocks (jnodes proper). There are plans, however to + * have a handle per extent unit rather than per each unformatted + * block, because there are so many of them. + * + * For bitmaps. Each bitmap is actually represented by two jnodes--one + * for working and another for "commit" data, together forming bnode. + * + * For io-heads. These are used by log writer. + * + * For formatted nodes (znode). See comment at the top of znode.c for + * details specific to the formatted nodes (znodes). + * + * Node data. + * + * Jnode provides access to the data of node it represents. Data are + * stored in a page. Page is kept in a page cache. This means, that jnodes + * are highly interconnected with page cache and VM internals. + * + * jnode has a pointer to page (->pg) containing its data. Pointer to data + * themselves is cached in ->data field to avoid frequent calls to + * page_address(). + * + * jnode and page are attached to each other by jnode_attach_page(). This + * function places pointer to jnode in set_page_private(), sets PG_private + * flag and increments page counter. + * + * Opposite operation is performed by page_clear_jnode(). + * + * jnode->pg is protected by jnode spin lock, and page->private is + * protected by page lock. See comment at the top of page_cache.c for + * more. + * + * page can be detached from jnode for two reasons: + * + * . jnode is removed from a tree (file is truncated, of formatted + * node is removed by balancing). + * + * . during memory pressure, VM calls ->releasepage() method + * (reiser4_releasepage()) to evict page from memory. + * + * (there, of course, is also umount, but this is special case we are not + * concerned with here). + * + * To protect jnode page from eviction, one calls jload() function that + * "pins" page in memory (loading it if necessary), increments + * jnode->d_count, and kmap()s page. Page is unpinned through call to + * jrelse(). + * + * Jnode life cycle. + * + * jnode is created, placed in hash table, and, optionally, in per-inode + * radix tree. Page can be attached to jnode, pinned, released, etc. + * + * When jnode is captured into atom its reference counter is + * increased. While being part of an atom, jnode can be "early + * flushed". This means that as part of flush procedure, jnode is placed + * into "relocate set", and its page is submitted to the disk. After io + * completes, page can be detached, then loaded again, re-dirtied, etc. + * + * Thread acquired reference to jnode by calling jref() and releases it by + * jput(). When last reference is removed, jnode is still retained in + * memory (cached) if it has page attached, _unless_ it is scheduled for + * destruction (has JNODE_HEARD_BANSHEE bit set). + * + * Tree read-write lock was used as "existential" lock for jnodes. That is, + * jnode->x_count could be changed from 0 to 1 only under tree write lock, + * that is, tree lock protected unreferenced jnodes stored in the hash + * table, from recycling. + * + * This resulted in high contention on tree lock, because jref()/jput() is + * frequent operation. To ameliorate this problem, RCU is used: when jput() + * is just about to release last reference on jnode it sets JNODE_RIP bit + * on it, and then proceed with jnode destruction (removing jnode from hash + * table, cbk_cache, detaching page, etc.). All places that change jnode + * reference counter from 0 to 1 (jlookup(), zlook(), zget(), and + * cbk_cache_scan_slots()) check for JNODE_RIP bit (this is done by + * jnode_rip_check() function), and pretend that nothing was found in hash + * table if bit is set. + * + * jput defers actual return of jnode into slab cache to some later time + * (by call_rcu()), this guarantees that other threads can safely continue + * working with JNODE_RIP-ped jnode. + * + */ + +#include "reiser4.h" +#include "debug.h" +#include "dformat.h" +#include "jnode.h" +#include "plugin/plugin_header.h" +#include "plugin/plugin.h" +#include "txnmgr.h" +/*#include "jnode.h"*/ +#include "znode.h" +#include "tree.h" +#include "tree_walk.h" +#include "super.h" +#include "inode.h" +#include "page_cache.h" + +#include /* UML needs this for PAGE_OFFSET */ +#include +#include +#include +#include +#include /* for struct address_space */ +#include /* for inode_lock */ + +static struct kmem_cache *_jnode_slab = NULL; + +static void jnode_set_type(jnode * node, jnode_type type); +static int jdelete(jnode * node); +static int jnode_try_drop(jnode * node); + +#if REISER4_DEBUG +static int jnode_invariant(jnode * node, int tlocked, int jlocked); +#endif + +/* true if valid page is attached to jnode */ +static inline int jnode_is_parsed(jnode * node) +{ + return JF_ISSET(node, JNODE_PARSED); +} + +/* hash table support */ + +/* compare two jnode keys for equality. Used by hash-table macros */ +static inline int jnode_key_eq(const struct jnode_key *k1, + const struct jnode_key *k2) +{ + assert("nikita-2350", k1 != NULL); + assert("nikita-2351", k2 != NULL); + + return (k1->index == k2->index && k1->objectid == k2->objectid); +} + +/* Hash jnode by its key (inode plus offset). Used by hash-table macros */ +static inline __u32 jnode_key_hashfn(j_hash_table * table, + const struct jnode_key *key) +{ + assert("nikita-2352", key != NULL); + assert("nikita-3346", IS_POW(table->_buckets)); + + /* yes, this is remarkable simply (where not stupid) hash function. */ + return (key->objectid + key->index) & (table->_buckets - 1); +} + +/* The hash table definition */ +#define KMALLOC(size) reiser4_vmalloc(size) +#define KFREE(ptr, size) vfree(ptr) +TYPE_SAFE_HASH_DEFINE(j, jnode, struct jnode_key, key.j, link.j, + jnode_key_hashfn, jnode_key_eq); +#undef KFREE +#undef KMALLOC + +/* call this to initialise jnode hash table */ +int jnodes_tree_init(reiser4_tree * tree/* tree to initialise jnodes for */) +{ + assert("nikita-2359", tree != NULL); + return j_hash_init(&tree->jhash_table, 16384); +} + +/* call this to destroy jnode hash table. This is called during umount. */ +int jnodes_tree_done(reiser4_tree * tree/* tree to destroy jnodes for */) +{ + j_hash_table *jtable; + jnode *node; + jnode *next; + + assert("nikita-2360", tree != NULL); + + /* + * Scan hash table and free all jnodes. + */ + jtable = &tree->jhash_table; + if (jtable->_table) { + for_all_in_htable(jtable, j, node, next) { + assert("nikita-2361", !atomic_read(&node->x_count)); + jdrop(node); + } + + j_hash_done(&tree->jhash_table); + } + return 0; +} + +/** + * init_jnodes - create jnode cache + * + * Initializes slab cache jnodes. It is part of reiser4 module initialization. + */ +int init_jnodes(void) +{ + assert("umka-168", _jnode_slab == NULL); + + _jnode_slab = kmem_cache_create("jnode", sizeof(jnode), 0, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT, NULL); + if (_jnode_slab == NULL) + return RETERR(-ENOMEM); + + return 0; +} + +/** + * done_znodes - delete znode cache + * + * This is called on reiser4 module unloading or system shutdown. + */ +void done_jnodes(void) +{ + destroy_reiser4_cache(&_jnode_slab); +} + +/* Initialize a jnode. */ +void jnode_init(jnode * node, reiser4_tree * tree, jnode_type type) +{ + assert("umka-175", node != NULL); + + memset(node, 0, sizeof(jnode)); + ON_DEBUG(node->magic = JMAGIC); + jnode_set_type(node, type); + atomic_set(&node->d_count, 0); + atomic_set(&node->x_count, 0); + spin_lock_init(&node->guard); + spin_lock_init(&node->load); + node->atom = NULL; + node->tree = tree; + INIT_LIST_HEAD(&node->capture_link); + + ASSIGN_NODE_LIST(node, NOT_CAPTURED); + + INIT_RCU_HEAD(&node->rcu); + +#if REISER4_DEBUG + { + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(tree->super); + spin_lock_irq(&sbinfo->all_guard); + list_add(&node->jnodes, &sbinfo->all_jnodes); + spin_unlock_irq(&sbinfo->all_guard); + } +#endif +} + +#if REISER4_DEBUG +/* + * Remove jnode from ->all_jnodes list. + */ +static void jnode_done(jnode * node, reiser4_tree * tree) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(tree->super); + + spin_lock_irq(&sbinfo->all_guard); + assert("nikita-2422", !list_empty(&node->jnodes)); + list_del_init(&node->jnodes); + spin_unlock_irq(&sbinfo->all_guard); +} +#endif + +/* return already existing jnode of page */ +jnode *jnode_by_page(struct page *pg) +{ + assert("nikita-2066", pg != NULL); + assert("nikita-2400", PageLocked(pg)); + assert("nikita-2068", PagePrivate(pg)); + assert("nikita-2067", jprivate(pg) != NULL); + return jprivate(pg); +} + +/* exported functions to allocate/free jnode objects outside this file */ +jnode *jalloc(void) +{ + jnode *jal = kmem_cache_alloc(_jnode_slab, reiser4_ctx_gfp_mask_get()); + return jal; +} + +/* return jnode back to the slab allocator */ +inline void jfree(jnode * node) +{ + assert("zam-449", node != NULL); + + assert("nikita-2663", (list_empty_careful(&node->capture_link) && + NODE_LIST(node) == NOT_CAPTURED)); + assert("nikita-3222", list_empty(&node->jnodes)); + assert("nikita-3221", jnode_page(node) == NULL); + + /* not yet phash_jnode_destroy(node); */ + + kmem_cache_free(_jnode_slab, node); +} + +/* + * This function is supplied as RCU callback. It actually frees jnode when + * last reference to it is gone. + */ +static void jnode_free_actor(struct rcu_head *head) +{ + jnode *node; + jnode_type jtype; + + node = container_of(head, jnode, rcu); + jtype = jnode_get_type(node); + + ON_DEBUG(jnode_done(node, jnode_get_tree(node))); + + switch (jtype) { + case JNODE_IO_HEAD: + case JNODE_BITMAP: + case JNODE_UNFORMATTED_BLOCK: + jfree(node); + break; + case JNODE_FORMATTED_BLOCK: + zfree(JZNODE(node)); + break; + case JNODE_INODE: + default: + wrong_return_value("nikita-3197", "Wrong jnode type"); + } +} + +/* + * Free a jnode. Post a callback to be executed later through RCU when all + * references to @node are released. + */ +static inline void jnode_free(jnode * node, jnode_type jtype) +{ + if (jtype != JNODE_INODE) { + /*assert("nikita-3219", list_empty(&node->rcu.list)); */ + call_rcu(&node->rcu, jnode_free_actor); + } else + jnode_list_remove(node); +} + +/* allocate new unformatted jnode */ +static jnode *jnew_unformatted(void) +{ + jnode *jal; + + jal = jalloc(); + if (jal == NULL) + return NULL; + + jnode_init(jal, current_tree, JNODE_UNFORMATTED_BLOCK); + jal->key.j.mapping = NULL; + jal->key.j.index = (unsigned long)-1; + jal->key.j.objectid = 0; + return jal; +} + +/* look for jnode with given mapping and offset within hash table */ +jnode *jlookup(reiser4_tree * tree, oid_t objectid, unsigned long index) +{ + struct jnode_key jkey; + jnode *node; + + assert("nikita-2353", tree != NULL); + + jkey.objectid = objectid; + jkey.index = index; + + /* + * hash table is _not_ protected by any lock during lookups. All we + * have to do is to disable preemption to keep RCU happy. + */ + + rcu_read_lock(); + node = j_hash_find(&tree->jhash_table, &jkey); + if (node != NULL) { + /* protect @node from recycling */ + jref(node); + assert("nikita-2955", jnode_invariant(node, 0, 0)); + node = jnode_rip_check(tree, node); + } + rcu_read_unlock(); + return node; +} + +/* per inode radix tree of jnodes is protected by tree's read write spin lock */ +static jnode *jfind_nolock(struct address_space *mapping, unsigned long index) +{ + assert("vs-1694", mapping->host != NULL); + + return radix_tree_lookup(jnode_tree_by_inode(mapping->host), index); +} + +jnode *jfind(struct address_space *mapping, unsigned long index) +{ + reiser4_tree *tree; + jnode *node; + + assert("vs-1694", mapping->host != NULL); + tree = reiser4_tree_by_inode(mapping->host); + + read_lock_tree(tree); + node = jfind_nolock(mapping, index); + if (node != NULL) + jref(node); + read_unlock_tree(tree); + return node; +} + +static void inode_attach_jnode(jnode * node) +{ + struct inode *inode; + reiser4_inode *info; + struct radix_tree_root *rtree; + + assert_rw_write_locked(&(jnode_get_tree(node)->tree_lock)); + assert("zam-1043", node->key.j.mapping != NULL); + inode = node->key.j.mapping->host; + info = reiser4_inode_data(inode); + rtree = jnode_tree_by_reiser4_inode(info); + if (rtree->rnode == NULL) { + /* prevent inode from being pruned when it has jnodes attached + to it */ + spin_lock_irq(&inode->i_data.tree_lock); + inode->i_data.nrpages++; + spin_unlock_irq(&inode->i_data.tree_lock); + } + assert("zam-1049", equi(rtree->rnode != NULL, info->nr_jnodes != 0)); + check_me("zam-1045", + !radix_tree_insert(rtree, node->key.j.index, node)); + ON_DEBUG(info->nr_jnodes++); +} + +static void inode_detach_jnode(jnode * node) +{ + struct inode *inode; + reiser4_inode *info; + struct radix_tree_root *rtree; + + assert_rw_write_locked(&(jnode_get_tree(node)->tree_lock)); + assert("zam-1044", node->key.j.mapping != NULL); + inode = node->key.j.mapping->host; + info = reiser4_inode_data(inode); + rtree = jnode_tree_by_reiser4_inode(info); + + assert("zam-1051", info->nr_jnodes != 0); + assert("zam-1052", rtree->rnode != NULL); + ON_DEBUG(info->nr_jnodes--); + + /* delete jnode from inode's radix tree of jnodes */ + check_me("zam-1046", radix_tree_delete(rtree, node->key.j.index)); + if (rtree->rnode == NULL) { + /* inode can be pruned now */ + spin_lock_irq(&inode->i_data.tree_lock); + inode->i_data.nrpages--; + spin_unlock_irq(&inode->i_data.tree_lock); + } +} + +/* put jnode into hash table (where they can be found by flush who does not know + mapping) and to inode's tree of jnodes (where they can be found (hopefully + faster) in places where mapping is known). Currently it is used by + fs/reiser4/plugin/item/extent_file_ops.c:index_extent_jnode when new jnode is + created */ +static void +hash_unformatted_jnode(jnode * node, struct address_space *mapping, + unsigned long index) +{ + j_hash_table *jtable; + + assert("vs-1446", jnode_is_unformatted(node)); + assert("vs-1442", node->key.j.mapping == 0); + assert("vs-1443", node->key.j.objectid == 0); + assert("vs-1444", node->key.j.index == (unsigned long)-1); + assert_rw_write_locked(&(jnode_get_tree(node)->tree_lock)); + + node->key.j.mapping = mapping; + node->key.j.objectid = get_inode_oid(mapping->host); + node->key.j.index = index; + + jtable = &jnode_get_tree(node)->jhash_table; + + /* race with some other thread inserting jnode into the hash table is + * impossible, because we keep the page lock. */ + /* + * following assertion no longer holds because of RCU: it is possible + * jnode is in the hash table, but with JNODE_RIP bit set. + */ + /* assert("nikita-3211", j_hash_find(jtable, &node->key.j) == NULL); */ + j_hash_insert_rcu(jtable, node); + inode_attach_jnode(node); +} + +static void unhash_unformatted_node_nolock(jnode * node) +{ + assert("vs-1683", node->key.j.mapping != NULL); + assert("vs-1684", + node->key.j.objectid == + get_inode_oid(node->key.j.mapping->host)); + + /* remove jnode from hash-table */ + j_hash_remove_rcu(&node->tree->jhash_table, node); + inode_detach_jnode(node); + node->key.j.mapping = NULL; + node->key.j.index = (unsigned long)-1; + node->key.j.objectid = 0; + +} + +/* remove jnode from hash table and from inode's tree of jnodes. This is used in + reiser4_invalidatepage and in kill_hook_extent -> truncate_inode_jnodes -> + reiser4_uncapture_jnode */ +void unhash_unformatted_jnode(jnode * node) +{ + assert("vs-1445", jnode_is_unformatted(node)); + + write_lock_tree(node->tree); + unhash_unformatted_node_nolock(node); + write_unlock_tree(node->tree); +} + +/* + * search hash table for a jnode with given oid and index. If not found, + * allocate new jnode, insert it, and also insert into radix tree for the + * given inode/mapping. + */ +static jnode *find_get_jnode(reiser4_tree * tree, + struct address_space *mapping, + oid_t oid, unsigned long index) +{ + jnode *result; + jnode *shadow; + int preload; + + result = jnew_unformatted(); + + if (unlikely(result == NULL)) + return ERR_PTR(RETERR(-ENOMEM)); + + preload = radix_tree_preload(reiser4_ctx_gfp_mask_get()); + if (preload != 0) + return ERR_PTR(preload); + + write_lock_tree(tree); + shadow = jfind_nolock(mapping, index); + if (likely(shadow == NULL)) { + /* add new jnode to hash table and inode's radix tree of + * jnodes */ + jref(result); + hash_unformatted_jnode(result, mapping, index); + } else { + /* jnode is found in inode's radix tree of jnodes */ + jref(shadow); + jnode_free(result, JNODE_UNFORMATTED_BLOCK); + assert("vs-1498", shadow->key.j.mapping == mapping); + result = shadow; + } + write_unlock_tree(tree); + + assert("nikita-2955", + ergo(result != NULL, jnode_invariant(result, 0, 0))); + radix_tree_preload_end(); + return result; +} + +/* jget() (a la zget() but for unformatted nodes). Returns (and possibly + creates) jnode corresponding to page @pg. jnode is attached to page and + inserted into jnode hash-table. */ +static jnode *do_jget(reiser4_tree * tree, struct page *pg) +{ + /* + * There are two ways to create jnode: starting with pre-existing page + * and without page. + * + * When page already exists, jnode is created + * (jnode_of_page()->do_jget()) under page lock. This is done in + * ->writepage(), or when capturing anonymous page dirtied through + * mmap. + * + * Jnode without page is created by index_extent_jnode(). + * + */ + + jnode *result; + oid_t oid = get_inode_oid(pg->mapping->host); + + assert("umka-176", pg != NULL); + assert("nikita-2394", PageLocked(pg)); + + result = jprivate(pg); + if (likely(result != NULL)) + return jref(result); + + tree = reiser4_tree_by_page(pg); + + /* check hash-table first */ + result = jfind(pg->mapping, pg->index); + if (unlikely(result != NULL)) { + spin_lock_jnode(result); + jnode_attach_page(result, pg); + spin_unlock_jnode(result); + result->key.j.mapping = pg->mapping; + return result; + } + + /* since page is locked, jnode should be allocated with GFP_NOFS flag */ + reiser4_ctx_gfp_mask_force(GFP_NOFS); + result = find_get_jnode(tree, pg->mapping, oid, pg->index); + if (unlikely(IS_ERR(result))) + return result; + /* attach jnode to page */ + spin_lock_jnode(result); + jnode_attach_page(result, pg); + spin_unlock_jnode(result); + return result; +} + +/* + * return jnode for @pg, creating it if necessary. + */ +jnode *jnode_of_page(struct page *pg) +{ + jnode *result; + + assert("umka-176", pg != NULL); + assert("nikita-2394", PageLocked(pg)); + + result = do_jget(reiser4_tree_by_page(pg), pg); + + if (REISER4_DEBUG && !IS_ERR(result)) { + assert("nikita-3210", result == jprivate(pg)); + assert("nikita-2046", jnode_page(jprivate(pg)) == pg); + if (jnode_is_unformatted(jprivate(pg))) { + assert("nikita-2364", + jprivate(pg)->key.j.index == pg->index); + assert("nikita-2367", + jprivate(pg)->key.j.mapping == pg->mapping); + assert("nikita-2365", + jprivate(pg)->key.j.objectid == + get_inode_oid(pg->mapping->host)); + assert("vs-1200", + jprivate(pg)->key.j.objectid == + pg->mapping->host->i_ino); + assert("nikita-2356", + jnode_is_unformatted(jnode_by_page(pg))); + } + assert("nikita-2956", jnode_invariant(jprivate(pg), 0, 0)); + } + return result; +} + +/* attach page to jnode: set ->pg pointer in jnode, and ->private one in the + * page.*/ +void jnode_attach_page(jnode * node, struct page *pg) +{ + assert("nikita-2060", node != NULL); + assert("nikita-2061", pg != NULL); + + assert("nikita-2050", jprivate(pg) == 0ul); + assert("nikita-2393", !PagePrivate(pg)); + assert("vs-1741", node->pg == NULL); + + assert("nikita-2396", PageLocked(pg)); + assert_spin_locked(&(node->guard)); + + page_cache_get(pg); + set_page_private(pg, (unsigned long)node); + node->pg = pg; + SetPagePrivate(pg); +} + +/* Dual to jnode_attach_page: break a binding between page and jnode */ +void page_clear_jnode(struct page *page, jnode * node) +{ + assert("nikita-2424", page != NULL); + assert("nikita-2425", PageLocked(page)); + assert("nikita-2426", node != NULL); + assert_spin_locked(&(node->guard)); + assert("nikita-2428", PagePrivate(page)); + + assert("nikita-3551", !PageWriteback(page)); + + JF_CLR(node, JNODE_PARSED); + set_page_private(page, 0ul); + ClearPagePrivate(page); + node->pg = NULL; + page_cache_release(page); +} + +#if 0 +/* it is only used in one place to handle error */ +void +page_detach_jnode(struct page *page, struct address_space *mapping, + unsigned long index) +{ + assert("nikita-2395", page != NULL); + + lock_page(page); + if ((page->mapping == mapping) && (page->index == index) + && PagePrivate(page)) { + jnode *node; + + node = jprivate(page); + spin_lock_jnode(node); + page_clear_jnode(page, node); + spin_unlock_jnode(node); + } + unlock_page(page); +} +#endif /* 0 */ + +/* return @node page locked. + + Locking ordering requires that one first takes page lock and afterwards + spin lock on node attached to this page. Sometimes it is necessary to go in + the opposite direction. This is done through standard trylock-and-release + loop. +*/ +static struct page *jnode_lock_page(jnode * node) +{ + struct page *page; + + assert("nikita-2052", node != NULL); + assert("nikita-2401", LOCK_CNT_NIL(spin_locked_jnode)); + + while (1) { + + spin_lock_jnode(node); + page = jnode_page(node); + if (page == NULL) + break; + + /* no need to page_cache_get( page ) here, because page cannot + be evicted from memory without detaching it from jnode and + this requires spin lock on jnode that we already hold. + */ + if (trylock_page(page)) { + /* We won a lock on jnode page, proceed. */ + break; + } + + /* Page is locked by someone else. */ + page_cache_get(page); + spin_unlock_jnode(node); + wait_on_page_locked(page); + /* it is possible that page was detached from jnode and + returned to the free pool, or re-assigned while we were + waiting on locked bit. This will be rechecked on the next + loop iteration. + */ + page_cache_release(page); + + /* try again */ + } + return page; +} + +/* + * is JNODE_PARSED bit is not set, call ->parse() method of jnode, to verify + * validness of jnode content. + */ +static inline int jparse(jnode * node) +{ + int result; + + assert("nikita-2466", node != NULL); + + spin_lock_jnode(node); + if (likely(!jnode_is_parsed(node))) { + result = jnode_ops(node)->parse(node); + if (likely(result == 0)) + JF_SET(node, JNODE_PARSED); + } else + result = 0; + spin_unlock_jnode(node); + return result; +} + +/* Lock a page attached to jnode, create and attach page to jnode if it had no + * one. */ +static struct page *jnode_get_page_locked(jnode * node, gfp_t gfp_flags) +{ + struct page *page; + + spin_lock_jnode(node); + page = jnode_page(node); + + if (page == NULL) { + spin_unlock_jnode(node); + page = find_or_create_page(jnode_get_mapping(node), + jnode_get_index(node), gfp_flags); + if (page == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + } else { + if (trylock_page(page)) { + spin_unlock_jnode(node); + return page; + } + page_cache_get(page); + spin_unlock_jnode(node); + lock_page(page); + assert("nikita-3134", page->mapping == jnode_get_mapping(node)); + } + + spin_lock_jnode(node); + if (!jnode_page(node)) + jnode_attach_page(node, page); + spin_unlock_jnode(node); + + page_cache_release(page); + assert("zam-894", jnode_page(node) == page); + return page; +} + +/* Start read operation for jnode's page if page is not up-to-date. */ +static int jnode_start_read(jnode * node, struct page *page) +{ + assert("zam-893", PageLocked(page)); + + if (PageUptodate(page)) { + unlock_page(page); + return 0; + } + return reiser4_page_io(page, node, READ, reiser4_ctx_gfp_mask_get()); +} + +#if REISER4_DEBUG +static void check_jload(jnode * node, struct page *page) +{ + if (jnode_is_znode(node)) { + node40_header *nh; + znode *z; + + z = JZNODE(node); + if (znode_is_any_locked(z)) { + nh = (node40_header *) kmap(page); + /* this only works for node40-only file systems. For + * debugging. */ + assert("nikita-3253", + z->nr_items == le16_to_cpu(get_unaligned(&nh->nr_items))); + kunmap(page); + } + assert("nikita-3565", znode_invariant(z)); + } +} +#else +#define check_jload(node, page) noop +#endif + +/* prefetch jnode to speed up next call to jload. Call this when you are going + * to call jload() shortly. This will bring appropriate portion of jnode into + * CPU cache. */ +void jload_prefetch(jnode * node) +{ + prefetchw(&node->x_count); +} + +/* load jnode's data into memory */ +int jload_gfp(jnode * node /* node to load */ , + gfp_t gfp_flags /* allocation flags */ , + int do_kmap/* true if page should be kmapped */) +{ + struct page *page; + int result = 0; + int parsed; + + assert("nikita-3010", reiser4_schedulable()); + + prefetchw(&node->pg); + + /* taking d-reference implies taking x-reference. */ + jref(node); + + /* + * acquiring d-reference to @jnode and check for JNODE_PARSED bit + * should be atomic, otherwise there is a race against + * reiser4_releasepage(). + */ + spin_lock(&(node->load)); + add_d_ref(node); + parsed = jnode_is_parsed(node); + spin_unlock(&(node->load)); + + if (unlikely(!parsed)) { + page = jnode_get_page_locked(node, gfp_flags); + if (unlikely(IS_ERR(page))) { + result = PTR_ERR(page); + goto failed; + } + + result = jnode_start_read(node, page); + if (unlikely(result != 0)) + goto failed; + + wait_on_page_locked(page); + if (unlikely(!PageUptodate(page))) { + result = RETERR(-EIO); + goto failed; + } + + if (do_kmap) + node->data = kmap(page); + + result = jparse(node); + if (unlikely(result != 0)) { + if (do_kmap) + kunmap(page); + goto failed; + } + check_jload(node, page); + } else { + page = jnode_page(node); + check_jload(node, page); + if (do_kmap) + node->data = kmap(page); + } + + if (!is_writeout_mode()) + /* We do not mark pages active if jload is called as a part of + * jnode_flush() or reiser4_write_logs(). Both jnode_flush() + * and write_logs() add no value to cached data, there is no + * sense to mark pages as active when they go to disk, it just + * confuses vm scanning routines because clean page could be + * moved out from inactive list as a result of this + * mark_page_accessed() call. */ + mark_page_accessed(page); + + return 0; + +failed: + jrelse_tail(node); + return result; + +} + +/* start asynchronous reading for given jnode's page. */ +int jstartio(jnode * node) +{ + struct page *page; + + page = jnode_get_page_locked(node, reiser4_ctx_gfp_mask_get()); + if (IS_ERR(page)) + return PTR_ERR(page); + + return jnode_start_read(node, page); +} + +/* Initialize a node by calling appropriate plugin instead of reading + * node from disk as in jload(). */ +int jinit_new(jnode * node, gfp_t gfp_flags) +{ + struct page *page; + int result; + + jref(node); + add_d_ref(node); + + page = jnode_get_page_locked(node, gfp_flags); + if (IS_ERR(page)) { + result = PTR_ERR(page); + goto failed; + } + + SetPageUptodate(page); + unlock_page(page); + + node->data = kmap(page); + + if (!jnode_is_parsed(node)) { + jnode_plugin *jplug = jnode_ops(node); + spin_lock_jnode(node); + result = jplug->init(node); + spin_unlock_jnode(node); + if (result) { + kunmap(page); + goto failed; + } + JF_SET(node, JNODE_PARSED); + } + + return 0; + +failed: + jrelse(node); + return result; +} + +/* release a reference to jnode acquired by jload(), decrement ->d_count */ +void jrelse_tail(jnode * node/* jnode to release references to */) +{ + assert("nikita-489", atomic_read(&node->d_count) > 0); + atomic_dec(&node->d_count); + /* release reference acquired in jload_gfp() or jinit_new() */ + jput(node); + if (jnode_is_unformatted(node) || jnode_is_znode(node)) + LOCK_CNT_DEC(d_refs); +} + +/* drop reference to node data. When last reference is dropped, data are + unloaded. */ +void jrelse(jnode * node/* jnode to release references to */) +{ + struct page *page; + + assert("nikita-487", node != NULL); + assert_spin_not_locked(&(node->guard)); + + page = jnode_page(node); + if (likely(page != NULL)) { + /* + * it is safe not to lock jnode here, because at this point + * @node->d_count is greater than zero (if jrelse() is used + * correctly, that is). JNODE_PARSED may be not set yet, if, + * for example, we got here as a result of error handling path + * in jload(). Anyway, page cannot be detached by + * reiser4_releasepage(). truncate will invalidate page + * regardless, but this should not be a problem. + */ + kunmap(page); + } + jrelse_tail(node); +} + +/* called from jput() to wait for io completion */ +static void jnode_finish_io(jnode * node) +{ + struct page *page; + + assert("nikita-2922", node != NULL); + + spin_lock_jnode(node); + page = jnode_page(node); + if (page != NULL) { + page_cache_get(page); + spin_unlock_jnode(node); + wait_on_page_writeback(page); + page_cache_release(page); + } else + spin_unlock_jnode(node); +} + +/* + * This is called by jput() when last reference to jnode is released. This is + * separate function, because we want fast path of jput() to be inline and, + * therefore, small. + */ +void jput_final(jnode * node) +{ + int r_i_p; + + /* A fast check for keeping node in cache. We always keep node in cache + * if its page is present and node was not marked for deletion */ + if (jnode_page(node) != NULL && !JF_ISSET(node, JNODE_HEARD_BANSHEE)) { + rcu_read_unlock(); + return; + } + r_i_p = !JF_TEST_AND_SET(node, JNODE_RIP); + /* + * if r_i_p is true, we were first to set JNODE_RIP on this node. In + * this case it is safe to access node after unlock. + */ + rcu_read_unlock(); + if (r_i_p) { + jnode_finish_io(node); + if (JF_ISSET(node, JNODE_HEARD_BANSHEE)) + /* node is removed from the tree. */ + jdelete(node); + else + jnode_try_drop(node); + } + /* if !r_i_p some other thread is already killing it */ +} + +int jwait_io(jnode * node, int rw) +{ + struct page *page; + int result; + + assert("zam-447", node != NULL); + assert("zam-448", jnode_page(node) != NULL); + + page = jnode_page(node); + + result = 0; + if (rw == READ) { + wait_on_page_locked(page); + } else { + assert("nikita-2227", rw == WRITE); + wait_on_page_writeback(page); + } + if (PageError(page)) + result = RETERR(-EIO); + + return result; +} + +/* + * jnode types and plugins. + * + * jnode by itself is a "base type". There are several different jnode + * flavors, called "jnode types" (see jnode_type for a list). Sometimes code + * has to do different things based on jnode type. In the standard reiser4 way + * this is done by having jnode plugin (see fs/reiser4/plugin.h:jnode_plugin). + * + * Functions below deal with jnode types and define methods of jnode plugin. + * + */ + +/* set jnode type. This is done during jnode initialization. */ +static void jnode_set_type(jnode * node, jnode_type type) +{ + static unsigned long type_to_mask[] = { + [JNODE_UNFORMATTED_BLOCK] = 1, + [JNODE_FORMATTED_BLOCK] = 0, + [JNODE_BITMAP] = 2, + [JNODE_IO_HEAD] = 6, + [JNODE_INODE] = 4 + }; + + assert("zam-647", type < LAST_JNODE_TYPE); + assert("nikita-2815", !jnode_is_loaded(node)); + assert("nikita-3386", node->state == 0); + + node->state |= (type_to_mask[type] << JNODE_TYPE_1); +} + +/* ->init() method of jnode plugin for jnodes that don't require plugin + * specific initialization. */ +static int init_noinit(jnode * node UNUSED_ARG) +{ + return 0; +} + +/* ->parse() method of jnode plugin for jnodes that don't require plugin + * specific pasring. */ +static int parse_noparse(jnode * node UNUSED_ARG) +{ + return 0; +} + +/* ->mapping() method for unformatted jnode */ +struct address_space *mapping_jnode(const jnode * node) +{ + struct address_space *map; + + assert("nikita-2713", node != NULL); + + /* mapping is stored in jnode */ + + map = node->key.j.mapping; + assert("nikita-2714", map != NULL); + assert("nikita-2897", is_reiser4_inode(map->host)); + assert("nikita-2715", get_inode_oid(map->host) == node->key.j.objectid); + return map; +} + +/* ->index() method for unformatted jnodes */ +unsigned long index_jnode(const jnode * node) +{ + /* index is stored in jnode */ + return node->key.j.index; +} + +/* ->remove() method for unformatted jnodes */ +static inline void remove_jnode(jnode * node, reiser4_tree * tree) +{ + /* remove jnode from hash table and radix tree */ + if (node->key.j.mapping) + unhash_unformatted_node_nolock(node); +} + +/* ->mapping() method for znodes */ +static struct address_space *mapping_znode(const jnode * node) +{ + /* all znodes belong to fake inode */ + return reiser4_get_super_fake(jnode_get_tree(node)->super)->i_mapping; +} + +/* ->index() method for znodes */ +static unsigned long index_znode(const jnode * node) +{ + unsigned long addr; + assert("nikita-3317", (1 << znode_shift_order) < sizeof(znode)); + + /* index of znode is just its address (shifted) */ + addr = (unsigned long)node; + return (addr - PAGE_OFFSET) >> znode_shift_order; +} + +/* ->mapping() method for bitmap jnode */ +static struct address_space *mapping_bitmap(const jnode * node) +{ + /* all bitmap blocks belong to special bitmap inode */ + return get_super_private(jnode_get_tree(node)->super)->bitmap-> + i_mapping; +} + +/* ->index() method for jnodes that are indexed by address */ +static unsigned long index_is_address(const jnode * node) +{ + unsigned long ind; + + ind = (unsigned long)node; + return ind - PAGE_OFFSET; +} + +/* resolve race with jput */ +jnode *jnode_rip_sync(reiser4_tree *tree, jnode *node) +{ + /* + * This is used as part of RCU-based jnode handling. + * + * jlookup(), zlook(), zget(), and cbk_cache_scan_slots() have to work + * with unreferenced jnodes (ones with ->x_count == 0). Hash table is + * not protected during this, so concurrent thread may execute + * zget-set-HEARD_BANSHEE-zput, or somehow else cause jnode to be + * freed in jput_final(). To avoid such races, jput_final() sets + * JNODE_RIP on jnode (under tree lock). All places that work with + * unreferenced jnodes call this function. It checks for JNODE_RIP bit + * (first without taking tree lock), and if this bit is set, released + * reference acquired by the current thread and returns NULL. + * + * As a result, if jnode is being concurrently freed, NULL is returned + * and caller should pretend that jnode wasn't found in the first + * place. + * + * Otherwise it's safe to release "rcu-read-lock" and continue with + * jnode. + */ + if (unlikely(JF_ISSET(node, JNODE_RIP))) { + read_lock_tree(tree); + if (JF_ISSET(node, JNODE_RIP)) { + dec_x_ref(node); + node = NULL; + } + read_unlock_tree(tree); + } + return node; +} + +reiser4_key *jnode_build_key(const jnode * node, reiser4_key * key) +{ + struct inode *inode; + item_plugin *iplug; + loff_t off; + + assert("nikita-3092", node != NULL); + assert("nikita-3093", key != NULL); + assert("nikita-3094", jnode_is_unformatted(node)); + + off = ((loff_t) index_jnode(node)) << PAGE_CACHE_SHIFT; + inode = mapping_jnode(node)->host; + + if (node->parent_item_id != 0) + iplug = item_plugin_by_id(node->parent_item_id); + else + iplug = NULL; + + if (iplug != NULL && iplug->f.key_by_offset) + iplug->f.key_by_offset(inode, off, key); + else { + file_plugin *fplug; + + fplug = inode_file_plugin(inode); + assert("zam-1007", fplug != NULL); + assert("zam-1008", fplug->key_by_inode != NULL); + + fplug->key_by_inode(inode, off, key); + } + + return key; +} + +/* ->parse() method for formatted nodes */ +static int parse_znode(jnode * node) +{ + return zparse(JZNODE(node)); +} + +/* ->delete() method for formatted nodes */ +static void delete_znode(jnode * node, reiser4_tree * tree) +{ + znode *z; + + assert_rw_write_locked(&(tree->tree_lock)); + assert("vs-898", JF_ISSET(node, JNODE_HEARD_BANSHEE)); + + z = JZNODE(node); + assert("vs-899", z->c_count == 0); + + /* delete znode from sibling list. */ + sibling_list_remove(z); + + znode_remove(z, tree); +} + +/* ->remove() method for formatted nodes */ +static int remove_znode(jnode * node, reiser4_tree * tree) +{ + znode *z; + + assert_rw_write_locked(&(tree->tree_lock)); + z = JZNODE(node); + + if (z->c_count == 0) { + /* detach znode from sibling list. */ + sibling_list_drop(z); + /* this is called with tree spin-lock held, so call + znode_remove() directly (rather than znode_lock_remove()). */ + znode_remove(z, tree); + return 0; + } + return RETERR(-EBUSY); +} + +/* ->init() method for formatted nodes */ +static int init_znode(jnode * node) +{ + znode *z; + + z = JZNODE(node); + /* call node plugin to do actual initialization */ + return z->nplug->init(z); +} + +/* ->clone() method for formatted nodes */ +static jnode *clone_formatted(jnode * node) +{ + znode *clone; + + assert("vs-1430", jnode_is_znode(node)); + clone = zalloc(reiser4_ctx_gfp_mask_get()); + if (clone == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + zinit(clone, NULL, current_tree); + jnode_set_block(ZJNODE(clone), jnode_get_block(node)); + /* ZJNODE(clone)->key.z is not initialized */ + clone->level = JZNODE(node)->level; + + return ZJNODE(clone); +} + +/* jplug->clone for unformatted nodes */ +static jnode *clone_unformatted(jnode * node) +{ + jnode *clone; + + assert("vs-1431", jnode_is_unformatted(node)); + clone = jalloc(); + if (clone == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + + jnode_init(clone, current_tree, JNODE_UNFORMATTED_BLOCK); + jnode_set_block(clone, jnode_get_block(node)); + + return clone; + +} + +/* + * Setup jnode plugin methods for various jnode types. + */ +jnode_plugin jnode_plugins[LAST_JNODE_TYPE] = { + [JNODE_UNFORMATTED_BLOCK] = { + .h = { + .type_id = REISER4_JNODE_PLUGIN_TYPE, + .id = JNODE_UNFORMATTED_BLOCK, + .pops = NULL, + .label = "unformatted", + .desc = "unformatted node", + .linkage = {NULL, NULL} + }, + .init = init_noinit, + .parse = parse_noparse, + .mapping = mapping_jnode, + .index = index_jnode, + .clone = clone_unformatted + }, + [JNODE_FORMATTED_BLOCK] = { + .h = { + .type_id = REISER4_JNODE_PLUGIN_TYPE, + .id = JNODE_FORMATTED_BLOCK, + .pops = NULL, + .label = "formatted", + .desc = "formatted tree node", + .linkage = {NULL, NULL} + }, + .init = init_znode, + .parse = parse_znode, + .mapping = mapping_znode, + .index = index_znode, + .clone = clone_formatted + }, + [JNODE_BITMAP] = { + .h = { + .type_id = REISER4_JNODE_PLUGIN_TYPE, + .id = JNODE_BITMAP, + .pops = NULL, + .label = "bitmap", + .desc = "bitmap node", + .linkage = {NULL, NULL} + }, + .init = init_noinit, + .parse = parse_noparse, + .mapping = mapping_bitmap, + .index = index_is_address, + .clone = NULL + }, + [JNODE_IO_HEAD] = { + .h = { + .type_id = REISER4_JNODE_PLUGIN_TYPE, + .id = JNODE_IO_HEAD, + .pops = NULL, + .label = "io head", + .desc = "io head", + .linkage = {NULL, NULL} + }, + .init = init_noinit, + .parse = parse_noparse, + .mapping = mapping_bitmap, + .index = index_is_address, + .clone = NULL + }, + [JNODE_INODE] = { + .h = { + .type_id = REISER4_JNODE_PLUGIN_TYPE, + .id = JNODE_INODE, + .pops = NULL, + .label = "inode", + .desc = "inode's builtin jnode", + .linkage = {NULL, NULL} + }, + .init = NULL, + .parse = NULL, + .mapping = NULL, + .index = NULL, + .clone = NULL + } +}; + +/* + * jnode destruction. + * + * Thread may use a jnode after it acquired a reference to it. References are + * counted in ->x_count field. Reference protects jnode from being + * recycled. This is different from protecting jnode data (that are stored in + * jnode page) from being evicted from memory. Data are protected by jload() + * and released by jrelse(). + * + * If thread already possesses a reference to the jnode it can acquire another + * one through jref(). Initial reference is obtained (usually) by locating + * jnode in some indexing structure that depends on jnode type: formatted + * nodes are kept in global hash table, where they are indexed by block + * number, and also in the cbk cache. Unformatted jnodes are also kept in hash + * table, which is indexed by oid and offset within file, and in per-inode + * radix tree. + * + * Reference to jnode is released by jput(). If last reference is released, + * jput_final() is called. This function determines whether jnode has to be + * deleted (this happens when corresponding node is removed from the file + * system, jnode is marked with JNODE_HEARD_BANSHEE bit in this case), or it + * should be just "removed" (deleted from memory). + * + * Jnode destruction is signally delicate dance because of locking and RCU. + */ + +/* + * Returns true if jnode cannot be removed right now. This check is called + * under tree lock. If it returns true, jnode is irrevocably committed to be + * deleted/removed. + */ +static inline int jnode_is_busy(const jnode * node, jnode_type jtype) +{ + /* if other thread managed to acquire a reference to this jnode, don't + * free it. */ + if (atomic_read(&node->x_count) > 0) + return 1; + /* also, don't free znode that has children in memory */ + if (jtype == JNODE_FORMATTED_BLOCK && JZNODE(node)->c_count > 0) + return 1; + return 0; +} + +/* + * this is called as part of removing jnode. Based on jnode type, call + * corresponding function that removes jnode from indices and returns it back + * to the appropriate slab (through RCU). + */ +static inline void +jnode_remove(jnode * node, jnode_type jtype, reiser4_tree * tree) +{ + switch (jtype) { + case JNODE_UNFORMATTED_BLOCK: + remove_jnode(node, tree); + break; + case JNODE_IO_HEAD: + case JNODE_BITMAP: + break; + case JNODE_INODE: + break; + case JNODE_FORMATTED_BLOCK: + remove_znode(node, tree); + break; + default: + wrong_return_value("nikita-3196", "Wrong jnode type"); + } +} + +/* + * this is called as part of deleting jnode. Based on jnode type, call + * corresponding function that removes jnode from indices and returns it back + * to the appropriate slab (through RCU). + * + * This differs from jnode_remove() only for formatted nodes---for them + * sibling list handling is different for removal and deletion. + */ +static inline void +jnode_delete(jnode * node, jnode_type jtype, reiser4_tree * tree UNUSED_ARG) +{ + switch (jtype) { + case JNODE_UNFORMATTED_BLOCK: + remove_jnode(node, tree); + break; + case JNODE_IO_HEAD: + case JNODE_BITMAP: + break; + case JNODE_FORMATTED_BLOCK: + delete_znode(node, tree); + break; + case JNODE_INODE: + default: + wrong_return_value("nikita-3195", "Wrong jnode type"); + } +} + +#if REISER4_DEBUG +/* + * remove jnode from the debugging list of all jnodes hanging off super-block. + */ +void jnode_list_remove(jnode * node) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(jnode_get_tree(node)->super); + + spin_lock_irq(&sbinfo->all_guard); + assert("nikita-2422", !list_empty(&node->jnodes)); + list_del_init(&node->jnodes); + spin_unlock_irq(&sbinfo->all_guard); +} +#endif + +/* + * this is called by jput_final() to remove jnode when last reference to it is + * released. + */ +static int jnode_try_drop(jnode * node) +{ + int result; + reiser4_tree *tree; + jnode_type jtype; + + assert("nikita-2491", node != NULL); + assert("nikita-2583", JF_ISSET(node, JNODE_RIP)); + + tree = jnode_get_tree(node); + jtype = jnode_get_type(node); + + spin_lock_jnode(node); + write_lock_tree(tree); + /* + * if jnode has a page---leave it alone. Memory pressure will + * eventually kill page and jnode. + */ + if (jnode_page(node) != NULL) { + write_unlock_tree(tree); + spin_unlock_jnode(node); + JF_CLR(node, JNODE_RIP); + return RETERR(-EBUSY); + } + + /* re-check ->x_count under tree lock. */ + result = jnode_is_busy(node, jtype); + if (result == 0) { + assert("nikita-2582", !JF_ISSET(node, JNODE_HEARD_BANSHEE)); + assert("jmacd-511/b", atomic_read(&node->d_count) == 0); + + spin_unlock_jnode(node); + /* no page and no references---despatch him. */ + jnode_remove(node, jtype, tree); + write_unlock_tree(tree); + jnode_free(node, jtype); + } else { + /* busy check failed: reference was acquired by concurrent + * thread. */ + write_unlock_tree(tree); + spin_unlock_jnode(node); + JF_CLR(node, JNODE_RIP); + } + return result; +} + +/* jdelete() -- Delete jnode from the tree and file system */ +static int jdelete(jnode * node/* jnode to finish with */) +{ + struct page *page; + int result; + reiser4_tree *tree; + jnode_type jtype; + + assert("nikita-467", node != NULL); + assert("nikita-2531", JF_ISSET(node, JNODE_RIP)); + + jtype = jnode_get_type(node); + + page = jnode_lock_page(node); + assert_spin_locked(&(node->guard)); + + tree = jnode_get_tree(node); + + write_lock_tree(tree); + /* re-check ->x_count under tree lock. */ + result = jnode_is_busy(node, jtype); + if (likely(!result)) { + assert("nikita-2123", JF_ISSET(node, JNODE_HEARD_BANSHEE)); + assert("jmacd-511", atomic_read(&node->d_count) == 0); + + /* detach page */ + if (page != NULL) { + /* + * FIXME this is racy against jnode_extent_write(). + */ + page_clear_jnode(page, node); + } + spin_unlock_jnode(node); + /* goodbye */ + jnode_delete(node, jtype, tree); + write_unlock_tree(tree); + jnode_free(node, jtype); + /* @node is no longer valid pointer */ + if (page != NULL) + reiser4_drop_page(page); + } else { + /* busy check failed: reference was acquired by concurrent + * thread. */ + JF_CLR(node, JNODE_RIP); + write_unlock_tree(tree); + spin_unlock_jnode(node); + if (page != NULL) + unlock_page(page); + } + return result; +} + +/* drop jnode on the floor. + + Return value: + + -EBUSY: failed to drop jnode, because there are still references to it + + 0: successfully dropped jnode + +*/ +static int jdrop_in_tree(jnode * node, reiser4_tree * tree) +{ + struct page *page; + jnode_type jtype; + int result; + + assert("zam-602", node != NULL); + assert_rw_not_read_locked(&(tree->tree_lock)); + assert_rw_not_write_locked(&(tree->tree_lock)); + assert("nikita-2403", !JF_ISSET(node, JNODE_HEARD_BANSHEE)); + + jtype = jnode_get_type(node); + + page = jnode_lock_page(node); + assert_spin_locked(&(node->guard)); + + write_lock_tree(tree); + + /* re-check ->x_count under tree lock. */ + result = jnode_is_busy(node, jtype); + if (!result) { + assert("nikita-2488", page == jnode_page(node)); + assert("nikita-2533", atomic_read(&node->d_count) == 0); + if (page != NULL) { + assert("nikita-2126", !PageDirty(page)); + assert("nikita-2127", PageUptodate(page)); + assert("nikita-2181", PageLocked(page)); + page_clear_jnode(page, node); + } + spin_unlock_jnode(node); + jnode_remove(node, jtype, tree); + write_unlock_tree(tree); + jnode_free(node, jtype); + if (page != NULL) + reiser4_drop_page(page); + } else { + /* busy check failed: reference was acquired by concurrent + * thread. */ + JF_CLR(node, JNODE_RIP); + write_unlock_tree(tree); + spin_unlock_jnode(node); + if (page != NULL) + unlock_page(page); + } + return result; +} + +/* This function frees jnode "if possible". In particular, [dcx]_count has to + be 0 (where applicable). */ +void jdrop(jnode * node) +{ + jdrop_in_tree(node, jnode_get_tree(node)); +} + +/* IO head jnode implementation; The io heads are simple j-nodes with limited + functionality (these j-nodes are not in any hash table) just for reading + from and writing to disk. */ + +jnode *reiser4_alloc_io_head(const reiser4_block_nr * block) +{ + jnode *jal = jalloc(); + + if (jal != NULL) { + jnode_init(jal, current_tree, JNODE_IO_HEAD); + jnode_set_block(jal, block); + } + + jref(jal); + + return jal; +} + +void reiser4_drop_io_head(jnode * node) +{ + assert("zam-648", jnode_get_type(node) == JNODE_IO_HEAD); + + jput(node); + jdrop(node); +} + +/* protect keep jnode data from reiser4_releasepage() */ +void pin_jnode_data(jnode * node) +{ + assert("zam-671", jnode_page(node) != NULL); + page_cache_get(jnode_page(node)); +} + +/* make jnode data free-able again */ +void unpin_jnode_data(jnode * node) +{ + assert("zam-672", jnode_page(node) != NULL); + page_cache_release(jnode_page(node)); +} + +struct address_space *jnode_get_mapping(const jnode * node) +{ + assert("nikita-3162", node != NULL); + return jnode_ops(node)->mapping(node); +} + +#if REISER4_DEBUG +/* debugging aid: jnode invariant */ +int jnode_invariant_f(const jnode * node, char const **msg) +{ +#define _ergo(ant, con) \ + ((*msg) = "{" #ant "} ergo {" #con "}", ergo((ant), (con))) +#define _check(exp) ((*msg) = #exp, (exp)) + + return _check(node != NULL) && + /* [jnode-queued] */ + /* only relocated node can be queued, except that when znode + * is being deleted, its JNODE_RELOC bit is cleared */ + _ergo(JF_ISSET(node, JNODE_FLUSH_QUEUED), + JF_ISSET(node, JNODE_RELOC) || + JF_ISSET(node, JNODE_HEARD_BANSHEE)) && + _check(node->jnodes.prev != NULL) && + _check(node->jnodes.next != NULL) && + /* [jnode-dirty] invariant */ + /* dirty inode is part of atom */ + _ergo(JF_ISSET(node, JNODE_DIRTY), node->atom != NULL) && + /* [jnode-oid] invariant */ + /* for unformatted node ->objectid and ->mapping fields are + * consistent */ + _ergo(jnode_is_unformatted(node) && node->key.j.mapping != NULL, + node->key.j.objectid == + get_inode_oid(node->key.j.mapping->host)) && + /* [jnode-atom-valid] invariant */ + /* node atom has valid state */ + _ergo(node->atom != NULL, node->atom->stage != ASTAGE_INVALID) && + /* [jnode-page-binding] invariant */ + /* if node points to page, it points back to node */ + _ergo(node->pg != NULL, jprivate(node->pg) == node) && + /* [jnode-refs] invariant */ + /* only referenced jnode can be loaded */ + _check(atomic_read(&node->x_count) >= atomic_read(&node->d_count)); + +} + +static const char *jnode_type_name(jnode_type type) +{ + switch (type) { + case JNODE_UNFORMATTED_BLOCK: + return "unformatted"; + case JNODE_FORMATTED_BLOCK: + return "formatted"; + case JNODE_BITMAP: + return "bitmap"; + case JNODE_IO_HEAD: + return "io head"; + case JNODE_INODE: + return "inode"; + case LAST_JNODE_TYPE: + return "last"; + default:{ + static char unknown[30]; + + sprintf(unknown, "unknown %i", type); + return unknown; + } + } +} + +#define jnode_state_name(node, flag) \ + (JF_ISSET((node), (flag)) ? ((#flag "|")+6) : "") + +/* debugging aid: output human readable information about @node */ +static void info_jnode(const char *prefix /* prefix to print */ , + const jnode * node/* node to print */) +{ + assert("umka-068", prefix != NULL); + + if (node == NULL) { + printk("%s: null\n", prefix); + return; + } + + printk + ("%s: %p: state: %lx: [%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s], level: %i," + " block: %s, d_count: %d, x_count: %d, " + "pg: %p, atom: %p, lock: %i:%i, type: %s, ", prefix, node, + node->state, + jnode_state_name(node, JNODE_PARSED), + jnode_state_name(node, JNODE_HEARD_BANSHEE), + jnode_state_name(node, JNODE_LEFT_CONNECTED), + jnode_state_name(node, JNODE_RIGHT_CONNECTED), + jnode_state_name(node, JNODE_ORPHAN), + jnode_state_name(node, JNODE_CREATED), + jnode_state_name(node, JNODE_RELOC), + jnode_state_name(node, JNODE_OVRWR), + jnode_state_name(node, JNODE_DIRTY), + jnode_state_name(node, JNODE_IS_DYING), + jnode_state_name(node, JNODE_RIP), + jnode_state_name(node, JNODE_MISSED_IN_CAPTURE), + jnode_state_name(node, JNODE_WRITEBACK), + jnode_state_name(node, JNODE_NEW), + jnode_state_name(node, JNODE_DKSET), + jnode_state_name(node, JNODE_REPACK), + jnode_state_name(node, JNODE_CLUSTER_PAGE), + jnode_get_level(node), sprint_address(jnode_get_block(node)), + atomic_read(&node->d_count), atomic_read(&node->x_count), + jnode_page(node), node->atom, 0, 0, + jnode_type_name(jnode_get_type(node))); + if (jnode_is_unformatted(node)) { + printk("inode: %llu, index: %lu, ", + node->key.j.objectid, node->key.j.index); + } +} + +/* debugging aid: check znode invariant and panic if it doesn't hold */ +static int jnode_invariant(jnode * node, int tlocked, int jlocked) +{ + char const *failed_msg; + int result; + reiser4_tree *tree; + + tree = jnode_get_tree(node); + + assert("umka-063312", node != NULL); + assert("umka-064321", tree != NULL); + + if (!jlocked && !tlocked) + spin_lock_jnode((jnode *) node); + if (!tlocked) + read_lock_tree(jnode_get_tree(node)); + result = jnode_invariant_f(node, &failed_msg); + if (!result) { + info_jnode("corrupted node", node); + warning("jmacd-555", "Condition %s failed", failed_msg); + } + if (!tlocked) + read_unlock_tree(jnode_get_tree(node)); + if (!jlocked && !tlocked) + spin_unlock_jnode((jnode *) node); + return result; +} + +#endif /* REISER4_DEBUG */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/jnode.h linux-2.6.30/fs/reiser4/jnode.h --- linux-2.6.30.orig/fs/reiser4/jnode.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/jnode.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,704 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Declaration of jnode. See jnode.c for details. */ + +#ifndef __JNODE_H__ +#define __JNODE_H__ + +#include "forward.h" +#include "type_safe_hash.h" +#include "txnmgr.h" +#include "key.h" +#include "debug.h" +#include "dformat.h" +#include "page_cache.h" +#include "context.h" + +#include "plugin/plugin.h" + +#include +#include +#include +#include +#include +#include +#include + +/* declare hash table of jnodes (jnodes proper, that is, unformatted + nodes) */ +TYPE_SAFE_HASH_DECLARE(j, jnode); + +/* declare hash table of znodes */ +TYPE_SAFE_HASH_DECLARE(z, znode); + +struct jnode_key { + __u64 objectid; + unsigned long index; + struct address_space *mapping; +}; + +/* + Jnode is the "base class" of other nodes in reiser4. It is also happens to + be exactly the node we use for unformatted tree nodes. + + Jnode provides following basic functionality: + + . reference counting and indexing. + + . integration with page cache. Jnode has ->pg reference to which page can + be attached. + + . interface to transaction manager. It is jnode that is kept in transaction + manager lists, attached to atoms, etc. (NOTE-NIKITA one may argue that this + means, there should be special type of jnode for inode.) + + Locking: + + Spin lock: the following fields are protected by the per-jnode spin lock: + + ->state + ->atom + ->capture_link + + Following fields are protected by the global tree lock: + + ->link + ->key.z (content of ->key.z is only changed in znode_rehash()) + ->key.j + + Atomic counters + + ->x_count + ->d_count + + ->pg, and ->data are protected by spin lock for unused jnode and are + immutable for used jnode (one for which fs/reiser4/vfs_ops.c:releasable() + is false). + + ->tree is immutable after creation + + Unclear + + ->blocknr: should be under jnode spin-lock, but current interface is based + on passing of block address. + + If you ever need to spin lock two nodes at once, do this in "natural" + memory order: lock znode with lower address first. (See lock_two_nodes().) + + Invariants involving this data-type: + + [jnode-dirty] + [jnode-refs] + [jnode-oid] + [jnode-queued] + [jnode-atom-valid] + [jnode-page-binding] +*/ + +struct jnode { +#if REISER4_DEBUG +#define JMAGIC 0x52654973 /* "ReIs" */ + int magic; +#endif + /* FIRST CACHE LINE (16 bytes): data used by jload */ + + /* jnode's state: bitwise flags from the reiser4_jnode_state enum. */ + /* 0 */ unsigned long state; + + /* lock, protecting jnode's fields. */ + /* 4 */ spinlock_t load; + + /* counter of references to jnode itself. Increased on jref(). + Decreased on jput(). + */ + /* 8 */ atomic_t x_count; + + /* counter of references to jnode's data. Pin data page(s) in + memory while this is greater than 0. Increased on jload(). + Decreased on jrelse(). + */ + /* 12 */ atomic_t d_count; + + /* SECOND CACHE LINE: data used by hash table lookups */ + + /* 16 */ union { + /* znodes are hashed by block number */ + reiser4_block_nr z; + /* unformatted nodes are hashed by mapping plus offset */ + struct jnode_key j; + } key; + + /* THIRD CACHE LINE */ + + /* 32 */ union { + /* pointers to maintain hash-table */ + z_hash_link z; + j_hash_link j; + } link; + + /* pointer to jnode page. */ + /* 36 */ struct page *pg; + /* pointer to node itself. This is page_address(node->pg) when page is + attached to the jnode + */ + /* 40 */ void *data; + + /* 44 */ reiser4_tree *tree; + + /* FOURTH CACHE LINE: atom related fields */ + + /* 48 */ spinlock_t guard; + + /* atom the block is in, if any */ + /* 52 */ txn_atom *atom; + + /* capture list */ + /* 56 */ struct list_head capture_link; + + /* FIFTH CACHE LINE */ + + /* 64 */ struct rcu_head rcu; + /* crosses cache line */ + + /* SIXTH CACHE LINE */ + + /* the real blocknr (where io is going to/from) */ + /* 80 */ reiser4_block_nr blocknr; + /* Parent item type, unformatted and CRC need it for + * offset => key conversion. */ + /* NOTE: this parent_item_id looks like jnode type. */ + /* 88 */ reiser4_plugin_id parent_item_id; + /* 92 */ +#if REISER4_DEBUG + /* list of all jnodes for debugging purposes. */ + struct list_head jnodes; + /* how many times this jnode was written in one transaction */ + int written; + /* this indicates which atom's list the jnode is on */ + atom_list list; +#endif +} __attribute__ ((aligned(16))); + +/* + * jnode types. Enumeration of existing jnode types. + */ +typedef enum { + JNODE_UNFORMATTED_BLOCK, /* unformatted block */ + JNODE_FORMATTED_BLOCK, /* formatted block, znode */ + JNODE_BITMAP, /* bitmap */ + JNODE_IO_HEAD, /* jnode representing a block in the + * wandering log */ + JNODE_INODE, /* jnode embedded into inode */ + LAST_JNODE_TYPE +} jnode_type; + +/* jnode states */ +typedef enum { + /* jnode's page is loaded and data checked */ + JNODE_PARSED = 0, + /* node was deleted, not all locks on it were released. This + node is empty and is going to be removed from the tree + shortly. */ + JNODE_HEARD_BANSHEE = 1, + /* left sibling pointer is valid */ + JNODE_LEFT_CONNECTED = 2, + /* right sibling pointer is valid */ + JNODE_RIGHT_CONNECTED = 3, + + /* znode was just created and doesn't yet have a pointer from + its parent */ + JNODE_ORPHAN = 4, + + /* this node was created by its transaction and has not been assigned + a block address. */ + JNODE_CREATED = 5, + + /* this node is currently relocated */ + JNODE_RELOC = 6, + /* this node is currently wandered */ + JNODE_OVRWR = 7, + + /* this znode has been modified */ + JNODE_DIRTY = 8, + + /* znode lock is being invalidated */ + JNODE_IS_DYING = 9, + + /* THIS PLACE IS INTENTIONALLY LEFT BLANK */ + + /* jnode is queued for flushing. */ + JNODE_FLUSH_QUEUED = 12, + + /* In the following bits jnode type is encoded. */ + JNODE_TYPE_1 = 13, + JNODE_TYPE_2 = 14, + JNODE_TYPE_3 = 15, + + /* jnode is being destroyed */ + JNODE_RIP = 16, + + /* znode was not captured during locking (it might so be because + ->level != LEAF_LEVEL and lock_mode == READ_LOCK) */ + JNODE_MISSED_IN_CAPTURE = 17, + + /* write is in progress */ + JNODE_WRITEBACK = 18, + + /* FIXME: now it is used by crypto-compress plugin only */ + JNODE_NEW = 19, + + /* delimiting keys are already set for this znode. */ + JNODE_DKSET = 20, + + /* when this bit is set page and jnode can not be disconnected */ + JNODE_WRITE_PREPARED = 21, + + JNODE_CLUSTER_PAGE = 22, + /* Jnode is marked for repacking, that means the reiser4 flush and the + * block allocator should process this node special way */ + JNODE_REPACK = 23, + /* node should be converted by flush in squalloc phase */ + JNODE_CONVERTIBLE = 24, + /* + * When jnode is dirtied for the first time in given transaction, + * do_jnode_make_dirty() checks whether this jnode can possible became + * member of overwrite set. If so, this bit is set, and one block is + * reserved in the ->flush_reserved space of atom. + * + * This block is "used" (and JNODE_FLUSH_RESERVED bit is cleared) when + * + * (1) flush decides that we want this block to go into relocate + * set after all. + * + * (2) wandering log is allocated (by log writer) + * + * (3) extent is allocated + * + */ + JNODE_FLUSH_RESERVED = 29 +} reiser4_jnode_state; + +/* Macros for accessing the jnode state. */ + +static inline void JF_CLR(jnode * j, int f) +{ + assert("unknown-1", j->magic == JMAGIC); + clear_bit(f, &j->state); +} +static inline int JF_ISSET(const jnode * j, int f) +{ + assert("unknown-2", j->magic == JMAGIC); + return test_bit(f, &((jnode *) j)->state); +} +static inline void JF_SET(jnode * j, int f) +{ + assert("unknown-3", j->magic == JMAGIC); + set_bit(f, &j->state); +} + +static inline int JF_TEST_AND_SET(jnode * j, int f) +{ + assert("unknown-4", j->magic == JMAGIC); + return test_and_set_bit(f, &j->state); +} + +static inline void spin_lock_jnode(jnode *node) +{ + /* check that spinlocks of lower priorities are not held */ + assert("", (LOCK_CNT_NIL(rw_locked_tree) && + LOCK_CNT_NIL(spin_locked_txnh) && + LOCK_CNT_NIL(spin_locked_zlock) && + LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_LT(spin_locked_jnode, 2))); + + spin_lock(&(node->guard)); + + LOCK_CNT_INC(spin_locked_jnode); + LOCK_CNT_INC(spin_locked); +} + +static inline void spin_unlock_jnode(jnode *node) +{ + assert_spin_locked(&(node->guard)); + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_jnode)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(spin_locked_jnode); + LOCK_CNT_DEC(spin_locked); + + spin_unlock(&(node->guard)); +} + +static inline int jnode_is_in_deleteset(const jnode * node) +{ + return JF_ISSET(node, JNODE_RELOC); +} + +extern int init_jnodes(void); +extern void done_jnodes(void); + +/* Jnode routines */ +extern jnode *jalloc(void); +extern void jfree(jnode * node) NONNULL; +extern jnode *jclone(jnode *); +extern jnode *jlookup(reiser4_tree * tree, + oid_t objectid, unsigned long ind) NONNULL; +extern jnode *jfind(struct address_space *, unsigned long index) NONNULL; +extern jnode *jnode_by_page(struct page *pg) NONNULL; +extern jnode *jnode_of_page(struct page *pg) NONNULL; +void jnode_attach_page(jnode * node, struct page *pg); + +void unhash_unformatted_jnode(jnode *); +extern jnode *page_next_jnode(jnode * node) NONNULL; +extern void jnode_init(jnode * node, reiser4_tree * tree, jnode_type) NONNULL; +extern void jnode_make_dirty(jnode * node) NONNULL; +extern void jnode_make_clean(jnode * node) NONNULL; +extern void jnode_make_wander_nolock(jnode * node) NONNULL; +extern void jnode_make_wander(jnode *) NONNULL; +extern void znode_make_reloc(znode * , flush_queue_t *) NONNULL; +extern void unformatted_make_reloc(jnode *, flush_queue_t *) NONNULL; +extern struct address_space *jnode_get_mapping(const jnode * node) NONNULL; + +/** + * jnode_get_block + * @node: jnode to query + * + */ +static inline const reiser4_block_nr *jnode_get_block(const jnode *node) +{ + assert("nikita-528", node != NULL); + + return &node->blocknr; +} + +/** + * jnode_set_block + * @node: jnode to update + * @blocknr: new block nr + */ +static inline void jnode_set_block(jnode *node, const reiser4_block_nr *blocknr) +{ + assert("nikita-2020", node != NULL); + assert("umka-055", blocknr != NULL); + node->blocknr = *blocknr; +} + + +/* block number for IO. Usually this is the same as jnode_get_block(), unless + * jnode was emergency flushed---then block number chosen by eflush is + * used. */ +static inline const reiser4_block_nr *jnode_get_io_block(jnode * node) +{ + assert("nikita-2768", node != NULL); + assert_spin_locked(&(node->guard)); + + return jnode_get_block(node); +} + +/* Jnode flush interface. */ +extern reiser4_blocknr_hint *reiser4_pos_hint(flush_pos_t *pos); +extern flush_queue_t *reiser4_pos_fq(flush_pos_t *pos); + +/* FIXME-VS: these are used in plugin/item/extent.c */ + +/* does extent_get_block have to be called */ +#define jnode_mapped(node) JF_ISSET (node, JNODE_MAPPED) +#define jnode_set_mapped(node) JF_SET (node, JNODE_MAPPED) + +/* the node should be converted during flush squalloc phase */ +#define jnode_convertible(node) JF_ISSET (node, JNODE_CONVERTIBLE) +#define jnode_set_convertible(node) JF_SET (node, JNODE_CONVERTIBLE) + +/* Macros to convert from jnode to znode, znode to jnode. These are macros + because C doesn't allow overloading of const prototypes. */ +#define ZJNODE(x) (&(x)->zjnode) +#define JZNODE(x) \ +({ \ + typeof(x) __tmp_x; \ + \ + __tmp_x = (x); \ + assert("jmacd-1300", jnode_is_znode(__tmp_x)); \ + (znode*) __tmp_x; \ +}) + +extern int jnodes_tree_init(reiser4_tree * tree); +extern int jnodes_tree_done(reiser4_tree * tree); + +#if REISER4_DEBUG + +extern int znode_is_any_locked(const znode * node); +extern void jnode_list_remove(jnode * node); + +#else + +#define jnode_list_remove(node) noop + +#endif + +int znode_is_root(const znode * node) NONNULL; + +/* bump reference counter on @node */ +static inline void add_x_ref(jnode * node/* node to increase x_count of */) +{ + assert("nikita-1911", node != NULL); + + atomic_inc(&node->x_count); + LOCK_CNT_INC(x_refs); +} + +static inline void dec_x_ref(jnode * node) +{ + assert("nikita-3215", node != NULL); + assert("nikita-3216", atomic_read(&node->x_count) > 0); + + atomic_dec(&node->x_count); + assert("nikita-3217", LOCK_CNT_GTZ(x_refs)); + LOCK_CNT_DEC(x_refs); +} + +/* jref() - increase counter of references to jnode/znode (x_count) */ +static inline jnode *jref(jnode * node) +{ + assert("jmacd-508", (node != NULL) && !IS_ERR(node)); + add_x_ref(node); + return node; +} + +/* get the page of jnode */ +static inline struct page *jnode_page(const jnode * node) +{ + return node->pg; +} + +/* return pointer to jnode data */ +static inline char *jdata(const jnode * node) +{ + assert("nikita-1415", node != NULL); + assert("nikita-3198", jnode_page(node) != NULL); + return node->data; +} + +static inline int jnode_is_loaded(const jnode * node) +{ + assert("zam-506", node != NULL); + return atomic_read(&node->d_count) > 0; +} + +extern void page_clear_jnode(struct page *page, jnode * node) NONNULL; + +static inline void jnode_set_reloc(jnode * node) +{ + assert("nikita-2431", node != NULL); + assert("nikita-2432", !JF_ISSET(node, JNODE_OVRWR)); + JF_SET(node, JNODE_RELOC); +} + +/* jload/jwrite/junload give a bread/bwrite/brelse functionality for jnodes */ + +extern int jload_gfp(jnode *, gfp_t, int do_kmap) NONNULL; + +static inline int jload(jnode *node) +{ + return jload_gfp(node, reiser4_ctx_gfp_mask_get(), 1); +} + +extern int jinit_new(jnode *, gfp_t) NONNULL; +extern int jstartio(jnode *) NONNULL; + +extern void jdrop(jnode *) NONNULL; +extern int jwait_io(jnode *, int rw) NONNULL; + +void jload_prefetch(jnode *); + +extern jnode *reiser4_alloc_io_head(const reiser4_block_nr * block) NONNULL; +extern void reiser4_drop_io_head(jnode * node) NONNULL; + +static inline reiser4_tree *jnode_get_tree(const jnode * node) +{ + assert("nikita-2691", node != NULL); + return node->tree; +} + +extern void pin_jnode_data(jnode *); +extern void unpin_jnode_data(jnode *); + +static inline jnode_type jnode_get_type(const jnode * node) +{ + static const unsigned long state_mask = + (1 << JNODE_TYPE_1) | (1 << JNODE_TYPE_2) | (1 << JNODE_TYPE_3); + + static jnode_type mask_to_type[] = { + /* JNODE_TYPE_3 : JNODE_TYPE_2 : JNODE_TYPE_1 */ + + /* 000 */ + [0] = JNODE_FORMATTED_BLOCK, + /* 001 */ + [1] = JNODE_UNFORMATTED_BLOCK, + /* 010 */ + [2] = JNODE_BITMAP, + /* 011 */ + [3] = LAST_JNODE_TYPE, /*invalid */ + /* 100 */ + [4] = JNODE_INODE, + /* 101 */ + [5] = LAST_JNODE_TYPE, + /* 110 */ + [6] = JNODE_IO_HEAD, + /* 111 */ + [7] = LAST_JNODE_TYPE, /* invalid */ + }; + + return mask_to_type[(node->state & state_mask) >> JNODE_TYPE_1]; +} + +/* returns true if node is a znode */ +static inline int jnode_is_znode(const jnode * node) +{ + return jnode_get_type(node) == JNODE_FORMATTED_BLOCK; +} + +static inline int jnode_is_flushprepped(jnode * node) +{ + assert("jmacd-78212", node != NULL); + assert_spin_locked(&(node->guard)); + return !JF_ISSET(node, JNODE_DIRTY) || JF_ISSET(node, JNODE_RELOC) || + JF_ISSET(node, JNODE_OVRWR); +} + +/* Return true if @node has already been processed by the squeeze and allocate + process. This implies the block address has been finalized for the + duration of this atom (or it is clean and will remain in place). If this + returns true you may use the block number as a hint. */ +static inline int jnode_check_flushprepped(jnode * node) +{ + int result; + + /* It must be clean or relocated or wandered. New allocations are set + * to relocate. */ + spin_lock_jnode(node); + result = jnode_is_flushprepped(node); + spin_unlock_jnode(node); + return result; +} + +/* returns true if node is unformatted */ +static inline int jnode_is_unformatted(const jnode * node) +{ + assert("jmacd-0123", node != NULL); + return jnode_get_type(node) == JNODE_UNFORMATTED_BLOCK; +} + +/* returns true if node represents a cluster cache page */ +static inline int jnode_is_cluster_page(const jnode * node) +{ + assert("edward-50", node != NULL); + return (JF_ISSET(node, JNODE_CLUSTER_PAGE)); +} + +/* returns true is node is builtin inode's jnode */ +static inline int jnode_is_inode(const jnode * node) +{ + assert("vs-1240", node != NULL); + return jnode_get_type(node) == JNODE_INODE; +} + +static inline jnode_plugin *jnode_ops_of(const jnode_type type) +{ + assert("nikita-2367", type < LAST_JNODE_TYPE); + return jnode_plugin_by_id((reiser4_plugin_id) type); +} + +static inline jnode_plugin *jnode_ops(const jnode * node) +{ + assert("nikita-2366", node != NULL); + + return jnode_ops_of(jnode_get_type(node)); +} + +/* Get the index of a block. */ +static inline unsigned long jnode_get_index(jnode * node) +{ + return jnode_ops(node)->index(node); +} + +/* return true if "node" is the root */ +static inline int jnode_is_root(const jnode * node) +{ + return jnode_is_znode(node) && znode_is_root(JZNODE(node)); +} + +extern struct address_space *mapping_jnode(const jnode * node); +extern unsigned long index_jnode(const jnode * node); + +static inline void jput(jnode * node); +extern void jput_final(jnode * node); + +/* bump data counter on @node */ +static inline void add_d_ref(jnode * node/* node to increase d_count of */) +{ + assert("nikita-1962", node != NULL); + + atomic_inc(&node->d_count); + if (jnode_is_unformatted(node) || jnode_is_znode(node)) + LOCK_CNT_INC(d_refs); +} + +/* jput() - decrement x_count reference counter on znode. + + Count may drop to 0, jnode stays in cache until memory pressure causes the + eviction of its page. The c_count variable also ensures that children are + pressured out of memory before the parent. The jnode remains hashed as + long as the VM allows its page to stay in memory. +*/ +static inline void jput(jnode * node) +{ + assert("jmacd-509", node != NULL); + assert("jmacd-510", atomic_read(&node->x_count) > 0); + assert("zam-926", reiser4_schedulable()); + LOCK_CNT_DEC(x_refs); + + rcu_read_lock(); + /* + * we don't need any kind of lock here--jput_final() uses RCU. + */ + if (unlikely(atomic_dec_and_test(&node->x_count))) + jput_final(node); + else + rcu_read_unlock(); + assert("nikita-3473", reiser4_schedulable()); +} + +extern void jrelse(jnode * node); +extern void jrelse_tail(jnode * node); + +extern jnode *jnode_rip_sync(reiser4_tree * t, jnode * node); + +/* resolve race with jput */ +static inline jnode *jnode_rip_check(reiser4_tree * tree, jnode * node) +{ + if (unlikely(JF_ISSET(node, JNODE_RIP))) + node = jnode_rip_sync(tree, node); + return node; +} + +extern reiser4_key *jnode_build_key(const jnode *node, reiser4_key * key); + +#if REISER4_DEBUG +extern int jnode_invariant_f(const jnode *node, char const **msg); +#endif + +extern jnode_plugin jnode_plugins[LAST_JNODE_TYPE]; + +/* __JNODE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/kassign.c linux-2.6.30/fs/reiser4/kassign.c --- linux-2.6.30.orig/fs/reiser4/kassign.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/kassign.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,677 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Key assignment policy implementation */ + +/* + * In reiser4 every piece of file system data and meta-data has a key. Keys + * are used to store information in and retrieve it from reiser4 internal + * tree. In addition to this, keys define _ordering_ of all file system + * information: things having close keys are placed into the same or + * neighboring (in the tree order) nodes of the tree. As our block allocator + * tries to respect tree order (see flush.c), keys also define order in which + * things are laid out on the disk, and hence, affect performance directly. + * + * Obviously, assignment of keys to data and meta-data should be consistent + * across whole file system. Algorithm that calculates a key for a given piece + * of data or meta-data is referred to as "key assignment". + * + * Key assignment is too expensive to be implemented as a plugin (that is, + * with an ability to support different key assignment schemas in the same + * compiled kernel image). As a compromise, all key-assignment functions and + * data-structures are collected in this single file, so that modifications to + * key assignment algorithm can be localized. Additional changes may be + * required in key.[ch]. + * + * Current default reiser4 key assignment algorithm is dubbed "Plan A". As one + * may guess, there is "Plan B" too. + * + */ + +/* + * Additional complication with key assignment implementation is a requirement + * to support different key length. + */ + +/* + * KEY ASSIGNMENT: PLAN A, LONG KEYS. + * + * DIRECTORY ITEMS + * + * | 60 | 4 | 7 |1| 56 | 64 | 64 | + * +--------------+---+---+-+-------------+------------------+-----------------+ + * | dirid | 0 | F |H| prefix-1 | prefix-2 | prefix-3/hash | + * +--------------+---+---+-+-------------+------------------+-----------------+ + * | | | | | + * | 8 bytes | 8 bytes | 8 bytes | 8 bytes | + * + * dirid objectid of directory this item is for + * + * F fibration, see fs/reiser4/plugin/fibration.[ch] + * + * H 1 if last 8 bytes of the key contain hash, + * 0 if last 8 bytes of the key contain prefix-3 + * + * prefix-1 first 7 characters of file name. + * Padded by zeroes if name is not long enough. + * + * prefix-2 next 8 characters of the file name. + * + * prefix-3 next 8 characters of the file name. + * + * hash hash of the rest of file name (i.e., portion of file + * name not included into prefix-1 and prefix-2). + * + * File names shorter than 23 (== 7 + 8 + 8) characters are completely encoded + * in the key. Such file names are called "short". They are distinguished by H + * bit set 0 in the key. + * + * Other file names are "long". For long name, H bit is 1, and first 15 (== 7 + * + 8) characters are encoded in prefix-1 and prefix-2 portions of the + * key. Last 8 bytes of the key are occupied by hash of the remaining + * characters of the name. + * + * This key assignment reaches following important goals: + * + * (1) directory entries are sorted in approximately lexicographical + * order. + * + * (2) collisions (when multiple directory items have the same key), while + * principally unavoidable in a tree with fixed length keys, are rare. + * + * STAT DATA + * + * | 60 | 4 | 64 | 4 | 60 | 64 | + * +--------------+---+-----------------+---+--------------+-----------------+ + * | locality id | 1 | ordering | 0 | objectid | 0 | + * +--------------+---+-----------------+---+--------------+-----------------+ + * | | | | | + * | 8 bytes | 8 bytes | 8 bytes | 8 bytes | + * + * locality id object id of a directory where first name was created for + * the object + * + * ordering copy of second 8-byte portion of the key of directory + * entry for the first name of this object. Ordering has a form + * { + * fibration :7; + * h :1; + * prefix1 :56; + * } + * see description of key for directory entry above. + * + * objectid object id for this object + * + * This key assignment policy is designed to keep stat-data in the same order + * as corresponding directory items, thus speeding up readdir/stat types of + * workload. + * + * FILE BODY + * + * | 60 | 4 | 64 | 4 | 60 | 64 | + * +--------------+---+-----------------+---+--------------+-----------------+ + * | locality id | 4 | ordering | 0 | objectid | offset | + * +--------------+---+-----------------+---+--------------+-----------------+ + * | | | | | + * | 8 bytes | 8 bytes | 8 bytes | 8 bytes | + * + * locality id object id of a directory where first name was created for + * the object + * + * ordering the same as in the key of stat-data for this object + * + * objectid object id for this object + * + * offset logical offset from the beginning of this file. + * Measured in bytes. + * + * + * KEY ASSIGNMENT: PLAN A, SHORT KEYS. + * + * DIRECTORY ITEMS + * + * | 60 | 4 | 7 |1| 56 | 64 | + * +--------------+---+---+-+-------------+-----------------+ + * | dirid | 0 | F |H| prefix-1 | prefix-2/hash | + * +--------------+---+---+-+-------------+-----------------+ + * | | | | + * | 8 bytes | 8 bytes | 8 bytes | + * + * dirid objectid of directory this item is for + * + * F fibration, see fs/reiser4/plugin/fibration.[ch] + * + * H 1 if last 8 bytes of the key contain hash, + * 0 if last 8 bytes of the key contain prefix-2 + * + * prefix-1 first 7 characters of file name. + * Padded by zeroes if name is not long enough. + * + * prefix-2 next 8 characters of the file name. + * + * hash hash of the rest of file name (i.e., portion of file + * name not included into prefix-1). + * + * File names shorter than 15 (== 7 + 8) characters are completely encoded in + * the key. Such file names are called "short". They are distinguished by H + * bit set in the key. + * + * Other file names are "long". For long name, H bit is 0, and first 7 + * characters are encoded in prefix-1 portion of the key. Last 8 bytes of the + * key are occupied by hash of the remaining characters of the name. + * + * STAT DATA + * + * | 60 | 4 | 4 | 60 | 64 | + * +--------------+---+---+--------------+-----------------+ + * | locality id | 1 | 0 | objectid | 0 | + * +--------------+---+---+--------------+-----------------+ + * | | | | + * | 8 bytes | 8 bytes | 8 bytes | + * + * locality id object id of a directory where first name was created for + * the object + * + * objectid object id for this object + * + * FILE BODY + * + * | 60 | 4 | 4 | 60 | 64 | + * +--------------+---+---+--------------+-----------------+ + * | locality id | 4 | 0 | objectid | offset | + * +--------------+---+---+--------------+-----------------+ + * | | | | + * | 8 bytes | 8 bytes | 8 bytes | + * + * locality id object id of a directory where first name was created for + * the object + * + * objectid object id for this object + * + * offset logical offset from the beginning of this file. + * Measured in bytes. + * + * + */ + +#include "debug.h" +#include "key.h" +#include "kassign.h" +#include "vfs_ops.h" +#include "inode.h" +#include "super.h" +#include "dscale.h" + +#include /* for __u?? */ +#include /* for struct super_block, etc */ + +/* bitmask for H bit (see comment at the beginning of this file */ +static const __u64 longname_mark = 0x0100000000000000ull; +/* bitmask for F and H portions of the key. */ +static const __u64 fibration_mask = 0xff00000000000000ull; + +/* return true if name is not completely encoded in @key */ +int is_longname_key(const reiser4_key * key) +{ + __u64 highpart; + + assert("nikita-2863", key != NULL); + if (get_key_type(key) != KEY_FILE_NAME_MINOR) + reiser4_print_key("oops", key); + assert("nikita-2864", get_key_type(key) == KEY_FILE_NAME_MINOR); + + if (REISER4_LARGE_KEY) + highpart = get_key_ordering(key); + else + highpart = get_key_objectid(key); + + return (highpart & longname_mark) ? 1 : 0; +} + +/* return true if @name is too long to be completely encoded in the key */ +int is_longname(const char *name UNUSED_ARG, int len) +{ + if (REISER4_LARGE_KEY) + return len > 23; + else + return len > 15; +} + +/* code ascii string into __u64. + + Put characters of @name into result (@str) one after another starting + from @start_idx-th highest (arithmetically) byte. This produces + endian-safe encoding. memcpy(2) will not do. + +*/ +static __u64 pack_string(const char *name /* string to encode */ , + int start_idx /* highest byte in result from + * which to start encoding */ ) +{ + unsigned i; + __u64 str; + + str = 0; + for (i = 0; (i < sizeof str - start_idx) && name[i]; ++i) { + str <<= 8; + str |= (unsigned char)name[i]; + } + str <<= (sizeof str - i - start_idx) << 3; + return str; +} + +/* opposite to pack_string(). Takes value produced by pack_string(), restores + * string encoded in it and stores result in @buf */ +char *reiser4_unpack_string(__u64 value, char *buf) +{ + do { + *buf = value >> (64 - 8); + if (*buf) + ++buf; + value <<= 8; + } while (value != 0); + *buf = 0; + return buf; +} + +/* obtain name encoded in @key and store it in @buf */ +char *extract_name_from_key(const reiser4_key * key, char *buf) +{ + char *c; + + assert("nikita-2868", !is_longname_key(key)); + + c = buf; + if (REISER4_LARGE_KEY) { + c = reiser4_unpack_string(get_key_ordering(key) & + ~fibration_mask, c); + c = reiser4_unpack_string(get_key_fulloid(key), c); + } else + c = reiser4_unpack_string(get_key_fulloid(key) & + ~fibration_mask, c); + reiser4_unpack_string(get_key_offset(key), c); + return buf; +} + +/** + * complete_entry_key - calculate entry key by name + * @dir: directory where entry is (or will be) in + * @name: name to calculate key of + * @len: lenth of name + * @result: place to store result in + * + * Sets fields of entry key @result which depend on file name. + * When REISER4_LARGE_KEY is defined three fields of @result are set: ordering, + * objectid and offset. Otherwise, objectid and offset are set. + */ +void complete_entry_key(const struct inode *dir, const char *name, + int len, reiser4_key *result) +{ +#if REISER4_LARGE_KEY + __u64 ordering; + __u64 objectid; + __u64 offset; + + assert("nikita-1139", dir != NULL); + assert("nikita-1142", result != NULL); + assert("nikita-2867", strlen(name) == len); + + /* + * key allocation algorithm for directory entries in case of large + * keys: + * + * If name is not longer than 7 + 8 + 8 = 23 characters, put first 7 + * characters into ordering field of key, next 8 charactes (if any) + * into objectid field of key and next 8 ones (of any) into offset + * field of key + * + * If file name is longer than 23 characters, put first 7 characters + * into key's ordering, next 8 to objectid and hash of remaining + * characters into offset field. + * + * To distinguish above cases, in latter set up unused high bit in + * ordering field. + */ + + /* [0-6] characters to ordering */ + ordering = pack_string(name, 1); + if (len > 7) { + /* [7-14] characters to objectid */ + objectid = pack_string(name + 7, 0); + if (len > 15) { + if (len <= 23) { + /* [15-23] characters to offset */ + offset = pack_string(name + 15, 0); + } else { + /* note in a key the fact that offset contains + * hash */ + ordering |= longname_mark; + + /* offset is the hash of the file name's tail */ + offset = inode_hash_plugin(dir)->hash(name + 15, + len - 15); + } + } else { + offset = 0ull; + } + } else { + objectid = 0ull; + offset = 0ull; + } + + assert("nikita-3480", inode_fibration_plugin(dir) != NULL); + ordering |= inode_fibration_plugin(dir)->fibre(dir, name, len); + + set_key_ordering(result, ordering); + set_key_fulloid(result, objectid); + set_key_offset(result, offset); + return; + +#else + __u64 objectid; + __u64 offset; + + assert("nikita-1139", dir != NULL); + assert("nikita-1142", result != NULL); + assert("nikita-2867", strlen(name) == len); + + /* + * key allocation algorithm for directory entries in case of not large + * keys: + * + * If name is not longer than 7 + 8 = 15 characters, put first 7 + * characters into objectid field of key, next 8 charactes (if any) + * into offset field of key + * + * If file name is longer than 15 characters, put first 7 characters + * into key's objectid, and hash of remaining characters into offset + * field. + * + * To distinguish above cases, in latter set up unused high bit in + * objectid field. + */ + + /* [0-6] characters to objectid */ + objectid = pack_string(name, 1); + if (len > 7) { + if (len <= 15) { + /* [7-14] characters to offset */ + offset = pack_string(name + 7, 0); + } else { + /* note in a key the fact that offset contains hash. */ + objectid |= longname_mark; + + /* offset is the hash of the file name. */ + offset = inode_hash_plugin(dir)->hash(name + 7, + len - 7); + } + } else + offset = 0ull; + + assert("nikita-3480", inode_fibration_plugin(dir) != NULL); + objectid |= inode_fibration_plugin(dir)->fibre(dir, name, len); + + set_key_fulloid(result, objectid); + set_key_offset(result, offset); + return; +#endif /* ! REISER4_LARGE_KEY */ +} + +/* true, if @key is the key of "." */ +int is_dot_key(const reiser4_key * key/* key to check */) +{ + assert("nikita-1717", key != NULL); + assert("nikita-1718", get_key_type(key) == KEY_FILE_NAME_MINOR); + return + (get_key_ordering(key) == 0ull) && + (get_key_objectid(key) == 0ull) && (get_key_offset(key) == 0ull); +} + +/* build key for stat-data. + + return key of stat-data of this object. This should became sd plugin + method in the future. For now, let it be here. + +*/ +reiser4_key *build_sd_key(const struct inode *target /* inode of an object */ , + reiser4_key * result /* resulting key of @target + stat-data */ ) +{ + assert("nikita-261", result != NULL); + + reiser4_key_init(result); + set_key_locality(result, reiser4_inode_data(target)->locality_id); + set_key_ordering(result, get_inode_ordering(target)); + set_key_objectid(result, get_inode_oid(target)); + set_key_type(result, KEY_SD_MINOR); + set_key_offset(result, (__u64) 0); + return result; +} + +/* encode part of key into &obj_key_id + + This encodes into @id part of @key sufficient to restore @key later, + given that latter is key of object (key of stat-data). + + See &obj_key_id +*/ +int build_obj_key_id(const reiser4_key * key /* key to encode */ , + obj_key_id * id/* id where key is encoded in */) +{ + assert("nikita-1151", key != NULL); + assert("nikita-1152", id != NULL); + + memcpy(id, key, sizeof *id); + return 0; +} + +/* encode reference to @obj in @id. + + This is like build_obj_key_id() above, but takes inode as parameter. */ +int build_inode_key_id(const struct inode *obj /* object to build key of */ , + obj_key_id * id/* result */) +{ + reiser4_key sdkey; + + assert("nikita-1166", obj != NULL); + assert("nikita-1167", id != NULL); + + build_sd_key(obj, &sdkey); + build_obj_key_id(&sdkey, id); + return 0; +} + +/* decode @id back into @key + + Restore key of object stat-data from @id. This is dual to + build_obj_key_id() above. +*/ +int extract_key_from_id(const obj_key_id * id /* object key id to extract key + * from */ , + reiser4_key * key/* result */) +{ + assert("nikita-1153", id != NULL); + assert("nikita-1154", key != NULL); + + reiser4_key_init(key); + memcpy(key, id, sizeof *id); + return 0; +} + +/* extract objectid of directory from key of directory entry within said + directory. + */ +oid_t extract_dir_id_from_key(const reiser4_key * de_key /* key of + * directory + * entry */ ) +{ + assert("nikita-1314", de_key != NULL); + return get_key_locality(de_key); +} + +/* encode into @id key of directory entry. + + Encode into @id information sufficient to later distinguish directory + entries within the same directory. This is not whole key, because all + directory entries within directory item share locality which is equal + to objectid of their directory. + +*/ +int build_de_id(const struct inode *dir /* inode of directory */ , + const struct qstr *name /* name to be given to @obj by + * directory entry being + * constructed */ , + de_id * id/* short key of directory entry */) +{ + reiser4_key key; + + assert("nikita-1290", dir != NULL); + assert("nikita-1292", id != NULL); + + /* NOTE-NIKITA this is suboptimal. */ + inode_dir_plugin(dir)->build_entry_key(dir, name, &key); + return build_de_id_by_key(&key, id); +} + +/* encode into @id key of directory entry. + + Encode into @id information sufficient to later distinguish directory + entries within the same directory. This is not whole key, because all + directory entries within directory item share locality which is equal + to objectid of their directory. + +*/ +int build_de_id_by_key(const reiser4_key * entry_key /* full key of directory + * entry */ , + de_id * id/* short key of directory entry */) +{ + memcpy(id, ((__u64 *) entry_key) + 1, sizeof *id); + return 0; +} + +/* restore from @id key of directory entry. + + Function dual to build_de_id(): given @id and locality, build full + key of directory entry within directory item. + +*/ +int extract_key_from_de_id(const oid_t locality /* locality of directory + * entry */ , + const de_id * id /* directory entry id */ , + reiser4_key * key/* result */) +{ + /* no need to initialise key here: all fields are overwritten */ + memcpy(((__u64 *) key) + 1, id, sizeof *id); + set_key_locality(key, locality); + set_key_type(key, KEY_FILE_NAME_MINOR); + return 0; +} + +/* compare two &de_id's */ +cmp_t de_id_cmp(const de_id * id1 /* first &de_id to compare */ , + const de_id * id2/* second &de_id to compare */) +{ + /* NOTE-NIKITA ugly implementation */ + reiser4_key k1; + reiser4_key k2; + + extract_key_from_de_id((oid_t) 0, id1, &k1); + extract_key_from_de_id((oid_t) 0, id2, &k2); + return keycmp(&k1, &k2); +} + +/* compare &de_id with key */ +cmp_t de_id_key_cmp(const de_id * id /* directory entry id to compare */ , + const reiser4_key * key/* key to compare */) +{ + cmp_t result; + reiser4_key *k1; + + k1 = (reiser4_key *) (((unsigned long)id) - sizeof key->el[0]); + result = KEY_DIFF_EL(k1, key, 1); + if (result == EQUAL_TO) { + result = KEY_DIFF_EL(k1, key, 2); + if (REISER4_LARGE_KEY && result == EQUAL_TO) + result = KEY_DIFF_EL(k1, key, 3); + } + return result; +} + +/* + * return number of bytes necessary to encode @inode identity. + */ +int inode_onwire_size(const struct inode *inode) +{ + int result; + + result = dscale_bytes_to_write(get_inode_oid(inode)); + result += dscale_bytes_to_write(get_inode_locality(inode)); + + /* + * ordering is large (it usually has highest bits set), so it makes + * little sense to dscale it. + */ + if (REISER4_LARGE_KEY) + result += sizeof(get_inode_ordering(inode)); + return result; +} + +/* + * encode @inode identity at @start + */ +char *build_inode_onwire(const struct inode *inode, char *start) +{ + start += dscale_write(start, get_inode_locality(inode)); + start += dscale_write(start, get_inode_oid(inode)); + + if (REISER4_LARGE_KEY) { + put_unaligned(cpu_to_le64(get_inode_ordering(inode)), (__le64 *)start); + start += sizeof(get_inode_ordering(inode)); + } + return start; +} + +/* + * extract key that was previously encoded by build_inode_onwire() at @addr + */ +char *extract_obj_key_id_from_onwire(char *addr, obj_key_id * key_id) +{ + __u64 val; + + addr += dscale_read(addr, &val); + val = (val << KEY_LOCALITY_SHIFT) | KEY_SD_MINOR; + put_unaligned(cpu_to_le64(val), (__le64 *)key_id->locality); + addr += dscale_read(addr, &val); + put_unaligned(cpu_to_le64(val), (__le64 *)key_id->objectid); +#if REISER4_LARGE_KEY + memcpy(&key_id->ordering, addr, sizeof key_id->ordering); + addr += sizeof key_id->ordering; +#endif + return addr; +} + +/* + * skip a key that was previously encoded by build_inode_onwire() at @addr + * FIXME: handle IO errors. + */ +char * locate_obj_key_id_onwire(char * addr) +{ + /* locality */ + addr += dscale_bytes_to_read(addr); + /* objectid */ + addr += dscale_bytes_to_read(addr); +#if REISER4_LARGE_KEY + addr += sizeof ((obj_key_id *)0)->ordering; +#endif + return addr; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/kassign.h linux-2.6.30/fs/reiser4/kassign.h --- linux-2.6.30.orig/fs/reiser4/kassign.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/kassign.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,111 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Key assignment policy interface. See kassign.c for details. */ + +#if !defined(__KASSIGN_H__) +#define __KASSIGN_H__ + +#include "forward.h" +#include "key.h" +#include "dformat.h" + +#include /* for __u?? */ +#include /* for struct super_block, etc */ +#include /* for struct qstr */ + +/* key assignment functions */ + +/* Information from which key of file stat-data can be uniquely + restored. This depends on key assignment policy for + stat-data. Currently it's enough to store object id and locality id + (60+60==120) bits, because minor packing locality and offset of + stat-data key are always known constants: KEY_SD_MINOR and 0 + respectively. For simplicity 4 bits are wasted in each id, and just + two 64 bit integers are stored. + + This field has to be byte-aligned, because we don't want to waste + space in directory entries. There is another side of a coin of + course: we waste CPU and bus bandwidth in stead, by copying data back + and forth. + + Next optimization: &obj_key_id is mainly used to address stat data from + directory entries. Under the assumption that majority of files only have + only name (one hard link) from *the* parent directory it seems reasonable + to only store objectid of stat data and take its locality from key of + directory item. + + This requires some flag to be added to the &obj_key_id to distinguish + between these two cases. Remaining bits in flag byte are then asking to be + used to store file type. + + This optimization requires changes in directory item handling code. + +*/ +typedef struct obj_key_id { + d8 locality[sizeof(__u64)]; + ON_LARGE_KEY(d8 ordering[sizeof(__u64)]; + ) + d8 objectid[sizeof(__u64)]; +} +obj_key_id; + +/* Information sufficient to uniquely identify directory entry within + compressed directory item. + + For alignment issues see &obj_key_id above. +*/ +typedef struct de_id { + ON_LARGE_KEY(d8 ordering[sizeof(__u64)];) + d8 objectid[sizeof(__u64)]; + d8 offset[sizeof(__u64)]; +} +de_id; + +extern int inode_onwire_size(const struct inode *obj); +extern char *build_inode_onwire(const struct inode *obj, char *area); +extern char *locate_obj_key_id_onwire(char *area); +extern char *extract_obj_key_id_from_onwire(char *area, obj_key_id * key_id); + +extern int build_inode_key_id(const struct inode *obj, obj_key_id * id); +extern int extract_key_from_id(const obj_key_id * id, reiser4_key * key); +extern int build_obj_key_id(const reiser4_key * key, obj_key_id * id); +extern oid_t extract_dir_id_from_key(const reiser4_key * de_key); +extern int build_de_id(const struct inode *dir, const struct qstr *name, + de_id * id); +extern int build_de_id_by_key(const reiser4_key * entry_key, de_id * id); +extern int extract_key_from_de_id(const oid_t locality, const de_id * id, + reiser4_key * key); +extern cmp_t de_id_cmp(const de_id * id1, const de_id * id2); +extern cmp_t de_id_key_cmp(const de_id * id, const reiser4_key * key); + +extern int build_readdir_key_common(struct file *dir, reiser4_key * result); +extern void build_entry_key_common(const struct inode *dir, + const struct qstr *name, + reiser4_key * result); +extern void build_entry_key_stable_entry(const struct inode *dir, + const struct qstr *name, + reiser4_key * result); +extern int is_dot_key(const reiser4_key * key); +extern reiser4_key *build_sd_key(const struct inode *target, + reiser4_key * result); + +extern int is_longname_key(const reiser4_key * key); +extern int is_longname(const char *name, int len); +extern char *extract_name_from_key(const reiser4_key * key, char *buf); +extern char *reiser4_unpack_string(__u64 value, char *buf); +extern void complete_entry_key(const struct inode *dir, const char *name, + int len, reiser4_key *result); + +/* __KASSIGN_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/Kconfig linux-2.6.30/fs/reiser4/Kconfig --- linux-2.6.30.orig/fs/reiser4/Kconfig 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/Kconfig 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,34 @@ +config REISER4_FS + tristate "Reiser4 (EXPERIMENTAL)" + depends on EXPERIMENTAL + select ZLIB_INFLATE + select ZLIB_DEFLATE + select LZO_COMPRESS + select LZO_DECOMPRESS + select CRYPTO + help + Reiser4 is a filesystem that performs all filesystem operations + as atomic transactions, which means that it either performs a + write, or it does not, and in the event of a crash it does not + partially perform it or corrupt it. + + It stores files in dancing trees, which are like balanced trees but + faster. It packs small files together so that they share blocks + without wasting space. This means you can use it to store really + small files. It also means that it saves you disk space. It avoids + hassling you with anachronisms like having a maximum number of + inodes, and wasting space if you use less than that number. + + Reiser4 is a distinct filesystem type from reiserfs (V3). + It's therefore not possible to use reiserfs file systems + with reiser4. + + To learn more about reiser4, go to http://www.namesys.com + +config REISER4_DEBUG + bool "Enable reiser4 debug mode" + depends on REISER4_FS + help + Don't use this unless you are debugging reiser4. + + If unsure, say N. diff -urN linux-2.6.30.orig/fs/reiser4/key.c linux-2.6.30/fs/reiser4/key.c --- linux-2.6.30.orig/fs/reiser4/key.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/key.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,138 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Key manipulations. */ + +#include "debug.h" +#include "key.h" +#include "super.h" +#include "reiser4.h" + +#include /* for __u?? */ + +/* Minimal possible key: all components are zero. It is presumed that this is + independent of key scheme. */ +static const reiser4_key MINIMAL_KEY = { + .el = { + 0ull, + ON_LARGE_KEY(0ull,) + 0ull, + 0ull + } +}; + +/* Maximal possible key: all components are ~0. It is presumed that this is + independent of key scheme. */ +static const reiser4_key MAXIMAL_KEY = { + .el = { + __constant_cpu_to_le64(~0ull), + ON_LARGE_KEY(__constant_cpu_to_le64(~0ull),) + __constant_cpu_to_le64(~0ull), + __constant_cpu_to_le64(~0ull) + } +}; + +/* Initialize key. */ +void reiser4_key_init(reiser4_key * key/* key to init */) +{ + assert("nikita-1169", key != NULL); + memset(key, 0, sizeof *key); +} + +/* minimal possible key in the tree. Return pointer to the static storage. */ +const reiser4_key * reiser4_min_key(void) +{ + return &MINIMAL_KEY; +} + +/* maximum possible key in the tree. Return pointer to the static storage. */ +const reiser4_key * reiser4_max_key(void) +{ + return &MAXIMAL_KEY; +} + +#if REISER4_DEBUG +/* debugging aid: print symbolic name of key type */ +static const char *type_name(unsigned int key_type/* key type */) +{ + switch (key_type) { + case KEY_FILE_NAME_MINOR: + return "file name"; + case KEY_SD_MINOR: + return "stat data"; + case KEY_ATTR_NAME_MINOR: + return "attr name"; + case KEY_ATTR_BODY_MINOR: + return "attr body"; + case KEY_BODY_MINOR: + return "file body"; + default: + return "unknown"; + } +} + +/* debugging aid: print human readable information about key */ +void reiser4_print_key(const char *prefix /* prefix to print */ , + const reiser4_key * key/* key to print */) +{ + /* turn bold on */ + /* printf ("\033[1m"); */ + if (key == NULL) + printk("%s: null key\n", prefix); + else { + if (REISER4_LARGE_KEY) + printk("%s: (%Lx:%x:%Lx:%Lx:%Lx:%Lx)", prefix, + get_key_locality(key), + get_key_type(key), + get_key_ordering(key), + get_key_band(key), + get_key_objectid(key), get_key_offset(key)); + else + printk("%s: (%Lx:%x:%Lx:%Lx:%Lx)", prefix, + get_key_locality(key), + get_key_type(key), + get_key_band(key), + get_key_objectid(key), get_key_offset(key)); + /* + * if this is a key of directory entry, try to decode part of + * a name stored in the key, and output it. + */ + if (get_key_type(key) == KEY_FILE_NAME_MINOR) { + char buf[DE_NAME_BUF_LEN]; + char *c; + + c = buf; + c = reiser4_unpack_string(get_key_ordering(key), c); + reiser4_unpack_string(get_key_fulloid(key), c); + printk("[%s", buf); + if (is_longname_key(key)) + /* + * only part of the name is stored in the key. + */ + printk("...]\n"); + else { + /* + * whole name is stored in the key. + */ + reiser4_unpack_string(get_key_offset(key), buf); + printk("%s]\n", buf); + } + } else { + printk("[%s]\n", type_name(get_key_type(key))); + } + } + /* turn bold off */ + /* printf ("\033[m\017"); */ +} + +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/key.h linux-2.6.30/fs/reiser4/key.h --- linux-2.6.30.orig/fs/reiser4/key.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/key.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,392 @@ +/* Copyright 2000, 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Declarations of key-related data-structures and operations on keys. */ + +#if !defined(__REISER4_KEY_H__) +#define __REISER4_KEY_H__ + +#include "dformat.h" +#include "forward.h" +#include "debug.h" + +#include /* for __u?? */ + +/* Operations on keys in reiser4 tree */ + +/* No access to any of these fields shall be done except via a + wrapping macro/function, and that wrapping macro/function shall + convert to little endian order. Compare keys will consider cpu byte order. */ + +/* A storage layer implementation difference between a regular unix file body + and its attributes is in the typedef below which causes all of the attributes + of a file to be near in key to all of the other attributes for all of the + files within that directory, and not near to the file itself. It is + interesting to consider whether this is the wrong approach, and whether there + should be no difference at all. For current usage patterns this choice is + probably the right one. */ + +/* possible values for minor packing locality (4 bits required) */ +typedef enum { + /* file name */ + KEY_FILE_NAME_MINOR = 0, + /* stat-data */ + KEY_SD_MINOR = 1, + /* file attribute name */ + KEY_ATTR_NAME_MINOR = 2, + /* file attribute value */ + KEY_ATTR_BODY_MINOR = 3, + /* file body (tail or extent) */ + KEY_BODY_MINOR = 4, +} key_minor_locality; + +/* Everything stored in the tree has a unique key, which means that the tree is + (logically) fully ordered by key. Physical order is determined by dynamic + heuristics that attempt to reflect key order when allocating available space, + and by the repacker. It is stylistically better to put aggregation + information into the key. Thus, if you want to segregate extents from tails, + it is better to give them distinct minor packing localities rather than + changing block_alloc.c to check the node type when deciding where to allocate + the node. + + The need to randomly displace new directories and large files disturbs this + symmetry unfortunately. However, it should be noted that this is a need that + is not clearly established given the existence of a repacker. Also, in our + current implementation tails have a different minor packing locality from + extents, and no files have both extents and tails, so maybe symmetry can be + had without performance cost after all. Symmetry is what we ship for now.... +*/ + +/* Arbitrary major packing localities can be assigned to objects using + the reiser4(filenameA/..packing<=some_number) system call. + + In reiser4, the creat() syscall creates a directory + + whose default flow (that which is referred to if the directory is + read as a file) is the traditional unix file body. + + whose directory plugin is the 'filedir' + + whose major packing locality is that of the parent of the object created. + + The static_stat item is a particular commonly used directory + compression (the one for normal unix files). + + The filedir plugin checks to see if the static_stat item exists. + There is a unique key for static_stat. If yes, then it uses the + static_stat item for all of the values that it contains. The + static_stat item contains a flag for each stat it contains which + indicates whether one should look outside the static_stat item for its + contents. +*/ + +/* offset of fields in reiser4_key. Value of each element of this enum + is index within key (thought as array of __u64's) where this field + is. */ +typedef enum { + /* major "locale", aka dirid. Sits in 1st element */ + KEY_LOCALITY_INDEX = 0, + /* minor "locale", aka item type. Sits in 1st element */ + KEY_TYPE_INDEX = 0, + ON_LARGE_KEY(KEY_ORDERING_INDEX,) + /* "object band". Sits in 2nd element */ + KEY_BAND_INDEX, + /* objectid. Sits in 2nd element */ + KEY_OBJECTID_INDEX = KEY_BAND_INDEX, + /* full objectid. Sits in 2nd element */ + KEY_FULLOID_INDEX = KEY_BAND_INDEX, + /* Offset. Sits in 3rd element */ + KEY_OFFSET_INDEX, + /* Name hash. Sits in 3rd element */ + KEY_HASH_INDEX = KEY_OFFSET_INDEX, + KEY_CACHELINE_END = KEY_OFFSET_INDEX, + KEY_LAST_INDEX +} reiser4_key_field_index; + +/* key in reiser4 internal "balanced" tree. It is just array of three + 64bit integers in disk byte order (little-endian by default). This + array is actually indexed by reiser4_key_field. Each __u64 within + this array is called "element". Logical key component encoded within + elements are called "fields". + + We declare this as union with second component dummy to suppress + inconvenient array<->pointer casts implied in C. */ +union reiser4_key { + __le64 el[KEY_LAST_INDEX]; + int pad; +}; + +/* bitmasks showing where within reiser4_key particular key is stored. */ +/* major locality occupies higher 60 bits of the first element */ +#define KEY_LOCALITY_MASK 0xfffffffffffffff0ull + +/* minor locality occupies lower 4 bits of the first element */ +#define KEY_TYPE_MASK 0xfull + +/* controversial band occupies higher 4 bits of the 2nd element */ +#define KEY_BAND_MASK 0xf000000000000000ull + +/* objectid occupies lower 60 bits of the 2nd element */ +#define KEY_OBJECTID_MASK 0x0fffffffffffffffull + +/* full 64bit objectid*/ +#define KEY_FULLOID_MASK 0xffffffffffffffffull + +/* offset is just 3rd L.M.Nt itself */ +#define KEY_OFFSET_MASK 0xffffffffffffffffull + +/* ordering is whole second element */ +#define KEY_ORDERING_MASK 0xffffffffffffffffull + +/* how many bits key element should be shifted to left to get particular field + */ +typedef enum { + KEY_LOCALITY_SHIFT = 4, + KEY_TYPE_SHIFT = 0, + KEY_BAND_SHIFT = 60, + KEY_OBJECTID_SHIFT = 0, + KEY_FULLOID_SHIFT = 0, + KEY_OFFSET_SHIFT = 0, + KEY_ORDERING_SHIFT = 0, +} reiser4_key_field_shift; + +static inline __u64 +get_key_el(const reiser4_key * key, reiser4_key_field_index off) +{ + assert("nikita-753", key != NULL); + assert("nikita-754", off < KEY_LAST_INDEX); + return le64_to_cpu(get_unaligned(&key->el[off])); +} + +static inline void +set_key_el(reiser4_key * key, reiser4_key_field_index off, __u64 value) +{ + assert("nikita-755", key != NULL); + assert("nikita-756", off < KEY_LAST_INDEX); + put_unaligned(cpu_to_le64(value), &key->el[off]); +} + +/* macro to define getter and setter functions for field F with type T */ +#define DEFINE_KEY_FIELD(L, U, T) \ +static inline T get_key_ ## L(const reiser4_key *key) \ +{ \ + assert("nikita-750", key != NULL); \ + return (T) (get_key_el(key, KEY_ ## U ## _INDEX) & \ + KEY_ ## U ## _MASK) >> KEY_ ## U ## _SHIFT; \ +} \ + \ +static inline void set_key_ ## L(reiser4_key * key, T loc) \ +{ \ + __u64 el; \ + \ + assert("nikita-752", key != NULL); \ + \ + el = get_key_el(key, KEY_ ## U ## _INDEX); \ + /* clear field bits in the key */ \ + el &= ~KEY_ ## U ## _MASK; \ + /* actually it should be \ + \ + el |= ( loc << KEY_ ## U ## _SHIFT ) & KEY_ ## U ## _MASK; \ + \ + but we trust user to never pass values that wouldn't fit \ + into field. Clearing extra bits is one operation, but this \ + function is time-critical. \ + But check this in assertion. */ \ + assert("nikita-759", ((loc << KEY_ ## U ## _SHIFT) & \ + ~KEY_ ## U ## _MASK) == 0); \ + el |= (loc << KEY_ ## U ## _SHIFT); \ + set_key_el(key, KEY_ ## U ## _INDEX, el); \ +} + +typedef __u64 oid_t; + +/* define get_key_locality(), set_key_locality() */ +DEFINE_KEY_FIELD(locality, LOCALITY, oid_t); +/* define get_key_type(), set_key_type() */ +DEFINE_KEY_FIELD(type, TYPE, key_minor_locality); +/* define get_key_band(), set_key_band() */ +DEFINE_KEY_FIELD(band, BAND, __u64); +/* define get_key_objectid(), set_key_objectid() */ +DEFINE_KEY_FIELD(objectid, OBJECTID, oid_t); +/* define get_key_fulloid(), set_key_fulloid() */ +DEFINE_KEY_FIELD(fulloid, FULLOID, oid_t); +/* define get_key_offset(), set_key_offset() */ +DEFINE_KEY_FIELD(offset, OFFSET, __u64); +#if (REISER4_LARGE_KEY) +/* define get_key_ordering(), set_key_ordering() */ +DEFINE_KEY_FIELD(ordering, ORDERING, __u64); +#else +static inline __u64 get_key_ordering(const reiser4_key * key) +{ + return 0; +} + +static inline void set_key_ordering(reiser4_key * key, __u64 val) +{ +} +#endif + +/* key comparison result */ +typedef enum { LESS_THAN = -1, /* if first key is less than second */ + EQUAL_TO = 0, /* if keys are equal */ + GREATER_THAN = +1 /* if first key is greater than second */ +} cmp_t; + +void reiser4_key_init(reiser4_key * key); + +/* minimal possible key in the tree. Return pointer to the static storage. */ +extern const reiser4_key *reiser4_min_key(void); +extern const reiser4_key *reiser4_max_key(void); + +/* helper macro for keycmp() */ +#define KEY_DIFF(k1, k2, field) \ +({ \ + typeof(get_key_ ## field(k1)) f1; \ + typeof(get_key_ ## field(k2)) f2; \ + \ + f1 = get_key_ ## field(k1); \ + f2 = get_key_ ## field(k2); \ + \ + (f1 < f2) ? LESS_THAN : ((f1 == f2) ? EQUAL_TO : GREATER_THAN); \ +}) + +/* helper macro for keycmp() */ +#define KEY_DIFF_EL(k1, k2, off) \ +({ \ + __u64 e1; \ + __u64 e2; \ + \ + e1 = get_key_el(k1, off); \ + e2 = get_key_el(k2, off); \ + \ + (e1 < e2) ? LESS_THAN : ((e1 == e2) ? EQUAL_TO : GREATER_THAN); \ +}) + +/* compare `k1' and `k2'. This function is a heart of "key allocation + policy". All you need to implement new policy is to add yet another + clause here. */ +static inline cmp_t keycmp(const reiser4_key * k1 /* first key to compare */ , + const reiser4_key * k2/* second key to compare */) +{ + cmp_t result; + + /* + * This function is the heart of reiser4 tree-routines. Key comparison + * is among most heavily used operations in the file system. + */ + + assert("nikita-439", k1 != NULL); + assert("nikita-440", k2 != NULL); + + /* there is no actual branch here: condition is compile time constant + * and constant folding and propagation ensures that only one branch + * is actually compiled in. */ + + if (REISER4_PLANA_KEY_ALLOCATION) { + /* if physical order of fields in a key is identical + with logical order, we can implement key comparison + as three 64bit comparisons. */ + /* logical order of fields in plan-a: + locality->type->objectid->offset. */ + /* compare locality and type at once */ + result = KEY_DIFF_EL(k1, k2, 0); + if (result == EQUAL_TO) { + /* compare objectid (and band if it's there) */ + result = KEY_DIFF_EL(k1, k2, 1); + /* compare offset */ + if (result == EQUAL_TO) { + result = KEY_DIFF_EL(k1, k2, 2); + if (REISER4_LARGE_KEY && result == EQUAL_TO) + result = KEY_DIFF_EL(k1, k2, 3); + } + } + } else if (REISER4_3_5_KEY_ALLOCATION) { + result = KEY_DIFF(k1, k2, locality); + if (result == EQUAL_TO) { + result = KEY_DIFF(k1, k2, objectid); + if (result == EQUAL_TO) { + result = KEY_DIFF(k1, k2, type); + if (result == EQUAL_TO) + result = KEY_DIFF(k1, k2, offset); + } + } + } else + impossible("nikita-441", "Unknown key allocation scheme!"); + return result; +} + +/* true if @k1 equals @k2 */ +static inline int keyeq(const reiser4_key * k1 /* first key to compare */ , + const reiser4_key * k2/* second key to compare */) +{ + assert("nikita-1879", k1 != NULL); + assert("nikita-1880", k2 != NULL); + return !memcmp(k1, k2, sizeof *k1); +} + +/* true if @k1 is less than @k2 */ +static inline int keylt(const reiser4_key * k1 /* first key to compare */ , + const reiser4_key * k2/* second key to compare */) +{ + assert("nikita-1952", k1 != NULL); + assert("nikita-1953", k2 != NULL); + return keycmp(k1, k2) == LESS_THAN; +} + +/* true if @k1 is less than or equal to @k2 */ +static inline int keyle(const reiser4_key * k1 /* first key to compare */ , + const reiser4_key * k2/* second key to compare */) +{ + assert("nikita-1954", k1 != NULL); + assert("nikita-1955", k2 != NULL); + return keycmp(k1, k2) != GREATER_THAN; +} + +/* true if @k1 is greater than @k2 */ +static inline int keygt(const reiser4_key * k1 /* first key to compare */ , + const reiser4_key * k2/* second key to compare */) +{ + assert("nikita-1959", k1 != NULL); + assert("nikita-1960", k2 != NULL); + return keycmp(k1, k2) == GREATER_THAN; +} + +/* true if @k1 is greater than or equal to @k2 */ +static inline int keyge(const reiser4_key * k1 /* first key to compare */ , + const reiser4_key * k2/* second key to compare */) +{ + assert("nikita-1956", k1 != NULL); + assert("nikita-1957", k2 != NULL); /* October 4: sputnik launched + * November 3: Laika */ + return keycmp(k1, k2) != LESS_THAN; +} + +static inline void prefetchkey(reiser4_key * key) +{ + prefetch(key); + prefetch(&key->el[KEY_CACHELINE_END]); +} + +/* (%Lx:%x:%Lx:%Lx:%Lx:%Lx) = + 1 + 16 + 1 + 1 + 1 + 1 + 1 + 16 + 1 + 16 + 1 + 16 + 1 */ +/* size of a buffer suitable to hold human readable key representation */ +#define KEY_BUF_LEN (80) + +#if REISER4_DEBUG +extern void reiser4_print_key(const char *prefix, const reiser4_key * key); +#else +#define reiser4_print_key(p, k) noop +#endif + +/* __FS_REISERFS_KEY_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/ktxnmgrd.c linux-2.6.30/fs/reiser4/ktxnmgrd.c --- linux-2.6.30.orig/fs/reiser4/ktxnmgrd.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/ktxnmgrd.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,215 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* Transaction manager daemon. */ + +/* + * ktxnmgrd is a kernel daemon responsible for committing transactions. It is + * needed/important for the following reasons: + * + * 1. in reiser4 atom is not committed immediately when last transaction + * handle closes, unless atom is either too old or too large (see + * atom_should_commit()). This is done to avoid committing too frequently. + * because: + * + * 2. sometimes we don't want to commit atom when closing last transaction + * handle even if it is old and fat enough. For example, because we are at + * this point under directory semaphore, and committing would stall all + * accesses to this directory. + * + * ktxnmgrd binds its time sleeping on condition variable. When is awakes + * either due to (tunable) timeout or because it was explicitly woken up by + * call to ktxnmgrd_kick(), it scans list of all atoms and commits ones + * eligible. + * + */ + +#include "debug.h" +#include "txnmgr.h" +#include "tree.h" +#include "ktxnmgrd.h" +#include "super.h" +#include "reiser4.h" + +#include /* for struct task_struct */ +#include +#include +#include +#include +#include +#include + +static int scan_mgr(struct super_block *); + +/* + * change current->comm so that ps, top, and friends will see changed + * state. This serves no useful purpose whatsoever, but also costs nothing. May + * be it will make lonely system administrator feeling less alone at 3 A.M. + */ +#define set_comm(state) \ + snprintf(current->comm, sizeof(current->comm), \ + "%s:%s:%s", __FUNCTION__, (super)->s_id, (state)) + +/** + * ktxnmgrd - kernel txnmgr daemon + * @arg: pointer to super block + * + * The background transaction manager daemon, started as a kernel thread during + * reiser4 initialization. + */ +static int ktxnmgrd(void *arg) +{ + struct super_block *super; + ktxnmgrd_context *ctx; + txn_mgr *mgr; + int done = 0; + + super = arg; + mgr = &get_super_private(super)->tmgr; + + /* + * do_fork() just copies task_struct into the new thread. ->fs_context + * shouldn't be copied of course. This shouldn't be a problem for the + * rest of the code though. + */ + current->journal_info = NULL; + ctx = mgr->daemon; + while (1) { + try_to_freeze(); + set_comm("wait"); + { + DEFINE_WAIT(__wait); + + prepare_to_wait(&ctx->wait, &__wait, + TASK_INTERRUPTIBLE); + if (kthread_should_stop()) + done = 1; + else + schedule_timeout(ctx->timeout); + finish_wait(&ctx->wait, &__wait); + } + if (done) + break; + set_comm("run"); + spin_lock(&ctx->guard); + /* + * wait timed out or ktxnmgrd was woken up by explicit request + * to commit something. Scan list of atoms in txnmgr and look + * for too old atoms. + */ + do { + ctx->rescan = 0; + scan_mgr(super); + spin_lock(&ctx->guard); + if (ctx->rescan) { + /* + * the list could be modified while ctx + * spinlock was released, we have to repeat + * scanning from the beginning + */ + break; + } + } while (ctx->rescan); + spin_unlock(&ctx->guard); + } + return 0; +} + +#undef set_comm + +/** + * reiser4_init_ktxnmgrd - initialize ktxnmgrd context and start kernel daemon + * @super: pointer to super block + * + * Allocates and initializes ktxnmgrd_context, attaches it to transaction + * manager. Starts kernel txnmgr daemon. This is called on mount. + */ +int reiser4_init_ktxnmgrd(struct super_block *super) +{ + txn_mgr *mgr; + ktxnmgrd_context *ctx; + + mgr = &get_super_private(super)->tmgr; + + assert("zam-1014", mgr->daemon == NULL); + + ctx = kzalloc(sizeof(ktxnmgrd_context), reiser4_ctx_gfp_mask_get()); + if (!ctx) + return RETERR(-ENOMEM); + + assert("nikita-2442", ctx != NULL); + + init_waitqueue_head(&ctx->wait); + + /*kcond_init(&ctx->startup);*/ + spin_lock_init(&ctx->guard); + ctx->timeout = REISER4_TXNMGR_TIMEOUT; + ctx->rescan = 1; + mgr->daemon = ctx; + + ctx->tsk = kthread_run(ktxnmgrd, super, "ktxnmgrd"); + if (IS_ERR(ctx->tsk)) { + int ret = PTR_ERR(ctx->tsk); + mgr->daemon = NULL; + kfree(ctx); + return RETERR(ret); + } + return 0; +} + +void ktxnmgrd_kick(txn_mgr *mgr) +{ + assert("nikita-3234", mgr != NULL); + assert("nikita-3235", mgr->daemon != NULL); + wake_up(&mgr->daemon->wait); +} + +int is_current_ktxnmgrd(void) +{ + return (get_current_super_private()->tmgr.daemon->tsk == current); +} + +/** + * scan_mgr - commit atoms which are to be committed + * @super: super block to commit atoms of + * + * Commits old atoms. + */ +static int scan_mgr(struct super_block *super) +{ + int ret; + reiser4_context ctx; + + init_stack_context(&ctx, super); + + ret = commit_some_atoms(&get_super_private(super)->tmgr); + + reiser4_exit_context(&ctx); + return ret; +} + +/** + * reiser4_done_ktxnmgrd - stop kernel thread and frees ktxnmgrd context + * @mgr: + * + * This is called on umount. Stops ktxnmgrd and free t + */ +void reiser4_done_ktxnmgrd(struct super_block *super) +{ + txn_mgr *mgr; + + mgr = &get_super_private(super)->tmgr; + assert("zam-1012", mgr->daemon != NULL); + + kthread_stop(mgr->daemon->tsk); + kfree(mgr->daemon); + mgr->daemon = NULL; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/ktxnmgrd.h linux-2.6.30/fs/reiser4/ktxnmgrd.h --- linux-2.6.30.orig/fs/reiser4/ktxnmgrd.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/ktxnmgrd.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,52 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Transaction manager daemon. See ktxnmgrd.c for comments. */ + +#ifndef __KTXNMGRD_H__ +#define __KTXNMGRD_H__ + +#include "txnmgr.h" + +#include +#include +#include +#include +#include +#include /* for struct task_struct */ + +/* in this structure all data necessary to start up, shut down and communicate + * with ktxnmgrd are kept. */ +struct ktxnmgrd_context { + /* wait queue head on which ktxnmgrd sleeps */ + wait_queue_head_t wait; + /* spin lock protecting all fields of this structure */ + spinlock_t guard; + /* timeout of sleeping on ->wait */ + signed long timeout; + /* kernel thread running ktxnmgrd */ + struct task_struct *tsk; + /* list of all file systems served by this ktxnmgrd */ + struct list_head queue; + /* should ktxnmgrd repeat scanning of atoms? */ + unsigned int rescan:1; +}; + +extern int reiser4_init_ktxnmgrd(struct super_block *); +extern void reiser4_done_ktxnmgrd(struct super_block *); + +extern void ktxnmgrd_kick(txn_mgr * mgr); +extern int is_current_ktxnmgrd(void); + +/* __KTXNMGRD_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/lock.c linux-2.6.30/fs/reiser4/lock.c --- linux-2.6.30.orig/fs/reiser4/lock.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/lock.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1237 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Traditional deadlock avoidance is achieved by acquiring all locks in a single + order. V4 balances the tree from the bottom up, and searches the tree from + the top down, and that is really the way we want it, so tradition won't work + for us. + + Instead we have two lock orderings, a high priority lock ordering, and a low + priority lock ordering. Each node in the tree has a lock in its znode. + + Suppose we have a set of processes which lock (R/W) tree nodes. Each process + has a set (maybe empty) of already locked nodes ("process locked set"). Each + process may have a pending lock request to a node locked by another process. + Note: we lock and unlock, but do not transfer locks: it is possible + transferring locks instead would save some bus locking.... + + Deadlock occurs when we have a loop constructed from process locked sets and + lock request vectors. + + NOTE: The reiser4 "tree" is a tree on disk, but its cached representation in + memory is extended with "znodes" with which we connect nodes with their left + and right neighbors using sibling pointers stored in the znodes. When we + perform balancing operations we often go from left to right and from right to + left. + + +-P1-+ +-P3-+ + |+--+| V1 |+--+| + ||N1|| -------> ||N3|| + |+--+| |+--+| + +----+ +----+ + ^ | + |V2 |V3 + | v + +---------P2---------+ + |+--+ +--+| + ||N2| -------- |N4|| + |+--+ +--+| + +--------------------+ + + We solve this by ensuring that only low priority processes lock in top to + bottom order and from right to left, and high priority processes lock from + bottom to top and left to right. + + ZAM-FIXME-HANS: order not just node locks in this way, order atom locks, and + kill those damn busy loops. + ANSWER(ZAM): atom locks (which are introduced by ASTAGE_CAPTURE_WAIT atom + stage) cannot be ordered that way. There are no rules what nodes can belong + to the atom and what nodes cannot. We cannot define what is right or left + direction, what is top or bottom. We can take immediate parent or side + neighbor of one node, but nobody guarantees that, say, left neighbor node is + not a far right neighbor for other nodes from the same atom. It breaks + deadlock avoidance rules and hi-low priority locking cannot be applied for + atom locks. + + How does it help to avoid deadlocks ? + + Suppose we have a deadlock with n processes. Processes from one priority + class never deadlock because they take locks in one consistent + order. + + So, any possible deadlock loop must have low priority as well as high + priority processes. There are no other lock priority levels except low and + high. We know that any deadlock loop contains at least one node locked by a + low priority process and requested by a high priority process. If this + situation is caught and resolved it is sufficient to avoid deadlocks. + + V4 DEADLOCK PREVENTION ALGORITHM IMPLEMENTATION. + + The deadlock prevention algorithm is based on comparing + priorities of node owners (processes which keep znode locked) and + requesters (processes which want to acquire a lock on znode). We + implement a scheme where low-priority owners yield locks to + high-priority requesters. We created a signal passing system that + is used to ask low-priority processes to yield one or more locked + znodes. + + The condition when a znode needs to change its owners is described by the + following formula: + + ############################################# + # # + # (number of high-priority requesters) > 0 # + # AND # + # (numbers of high-priority owners) == 0 # + # # + ############################################# + + Note that a low-priority process delays node releasing if another + high-priority process owns this node. So, slightly more strictly speaking, + to have a deadlock capable cycle you must have a loop in which a high + priority process is waiting on a low priority process to yield a node, which + is slightly different from saying a high priority process is waiting on a + node owned by a low priority process. + + It is enough to avoid deadlocks if we prevent any low-priority process from + falling asleep if its locked set contains a node which satisfies the + deadlock condition. + + That condition is implicitly or explicitly checked in all places where new + high-priority requests may be added or removed from node request queue or + high-priority process takes or releases a lock on node. The main + goal of these checks is to never lose the moment when node becomes "has + wrong owners" and send "must-yield-this-lock" signals to its low-pri owners + at that time. + + The information about received signals is stored in the per-process + structure (lock stack) and analyzed before a low-priority process goes to + sleep but after a "fast" attempt to lock a node fails. Any signal wakes + sleeping process up and forces him to re-check lock status and received + signal info. If "must-yield-this-lock" signals were received the locking + primitive (longterm_lock_znode()) fails with -E_DEADLOCK error code. + + V4 LOCKING DRAWBACKS + + If we have already balanced on one level, and we are propagating our changes + upward to a higher level, it could be very messy to surrender all locks on + the lower level because we put so much computational work into it, and + reverting them to their state before they were locked might be very complex. + We also don't want to acquire all locks before performing balancing because + that would either be almost as much work as the balancing, or it would be + too conservative and lock too much. We want balancing to be done only at + high priority. Yet, we might want to go to the left one node and use some + of its empty space... So we make one attempt at getting the node to the left + using try_lock, and if it fails we do without it, because we didn't really + need it, it was only a nice to have. + + LOCK STRUCTURES DESCRIPTION + + The following data structures are used in the reiser4 locking + implementation: + + All fields related to long-term locking are stored in znode->lock. + + The lock stack is a per thread object. It owns all znodes locked by the + thread. One znode may be locked by several threads in case of read lock or + one znode may be write locked by one thread several times. The special link + objects (lock handles) support n<->m relation between znodes and lock + owners. + + + + +---------+ +---------+ + | LS1 | | LS2 | + +---------+ +---------+ + ^ ^ + |---------------+ +----------+ + v v v v + +---------+ +---------+ +---------+ +---------+ + | LH1 | | LH2 | | LH3 | | LH4 | + +---------+ +---------+ +---------+ +---------+ + ^ ^ ^ ^ + | +------------+ | + v v v + +---------+ +---------+ +---------+ + | Z1 | | Z2 | | Z3 | + +---------+ +---------+ +---------+ + + Thread 1 locked znodes Z1 and Z2, thread 2 locked znodes Z2 and Z3. The + picture above shows that lock stack LS1 has a list of 2 lock handles LH1 and + LH2, lock stack LS2 has a list with lock handles LH3 and LH4 on it. Znode + Z1 is locked by only one thread, znode has only one lock handle LH1 on its + list, similar situation is for Z3 which is locked by the thread 2 only. Z2 + is locked (for read) twice by different threads and two lock handles are on + its list. Each lock handle represents a single relation of a locking of a + znode by a thread. Locking of a znode is an establishing of a locking + relation between the lock stack and the znode by adding of a new lock handle + to a list of lock handles, the lock stack. The lock stack links all lock + handles for all znodes locked by the lock stack. The znode list groups all + lock handles for all locks stacks which locked the znode. + + Yet another relation may exist between znode and lock owners. If lock + procedure cannot immediately take lock on an object it adds the lock owner + on special `requestors' list belongs to znode. That list represents a + queue of pending lock requests. Because one lock owner may request only + only one lock object at a time, it is a 1->n relation between lock objects + and a lock owner implemented as it is described above. Full information + (priority, pointers to lock and link objects) about each lock request is + stored in lock owner structure in `request' field. + + SHORT_TERM LOCKING + + This is a list of primitive operations over lock stacks / lock handles / + znodes and locking descriptions for them. + + 1. locking / unlocking which is done by two list insertion/deletion, one + to/from znode's list of lock handles, another one is to/from lock stack's + list of lock handles. The first insertion is protected by + znode->lock.guard spinlock. The list owned by the lock stack can be + modified only by thread who owns the lock stack and nobody else can + modify/read it. There is nothing to be protected by a spinlock or + something else. + + 2. adding/removing a lock request to/from znode requesters list. The rule is + that znode->lock.guard spinlock should be taken for this. + + 3. we can traverse list of lock handles and use references to lock stacks who + locked given znode if znode->lock.guard spinlock is taken. + + 4. If a lock stack is associated with a znode as a lock requestor or lock + owner its existence is guaranteed by znode->lock.guard spinlock. Some its + (lock stack's) fields should be protected from being accessed in parallel + by two or more threads. Please look at lock_stack structure definition + for the info how those fields are protected. */ + +/* Znode lock and capturing intertwining. */ +/* In current implementation we capture formatted nodes before locking + them. Take a look on longterm lock znode, reiser4_try_capture() request + precedes locking requests. The longterm_lock_znode function unconditionally + captures znode before even checking of locking conditions. + + Another variant is to capture znode after locking it. It was not tested, but + at least one deadlock condition is supposed to be there. One thread has + locked a znode (Node-1) and calls reiser4_try_capture() for it. + reiser4_try_capture() sleeps because znode's atom has CAPTURE_WAIT state. + Second thread is a flushing thread, its current atom is the atom Node-1 + belongs to. Second thread wants to lock Node-1 and sleeps because Node-1 + is locked by the first thread. The described situation is a deadlock. */ + +#include "debug.h" +#include "txnmgr.h" +#include "znode.h" +#include "jnode.h" +#include "tree.h" +#include "plugin/node/node.h" +#include "super.h" + +#include + +#if REISER4_DEBUG +static int request_is_deadlock_safe(znode * , znode_lock_mode, + znode_lock_request); +#endif + +/* Returns a lock owner associated with current thread */ +lock_stack *get_current_lock_stack(void) +{ + return &get_current_context()->stack; +} + +/* Wakes up all low priority owners informing them about possible deadlock */ +static void wake_up_all_lopri_owners(znode * node) +{ + lock_handle *handle; + + assert_spin_locked(&(node->lock.guard)); + list_for_each_entry(handle, &node->lock.owners, owners_link) { + assert("nikita-1832", handle->node == node); + /* count this signal in owner->nr_signaled */ + if (!handle->signaled) { + handle->signaled = 1; + atomic_inc(&handle->owner->nr_signaled); + /* Wake up a single process */ + reiser4_wake_up(handle->owner); + } + } +} + +/* Adds a lock to a lock owner, which means creating a link to the lock and + putting the link into the two lists all links are on (the doubly linked list + that forms the lock_stack, and the doubly linked list of links attached + to a lock. +*/ +static inline void +link_object(lock_handle * handle, lock_stack * owner, znode * node) +{ + assert("jmacd-810", handle->owner == NULL); + assert_spin_locked(&(node->lock.guard)); + + handle->owner = owner; + handle->node = node; + + assert("reiser4-4", + ergo(list_empty_careful(&owner->locks), owner->nr_locks == 0)); + + /* add lock handle to the end of lock_stack's list of locks */ + list_add_tail(&handle->locks_link, &owner->locks); + ON_DEBUG(owner->nr_locks++); + reiser4_ctx_gfp_mask_set(); + + /* add lock handle to the head of znode's list of owners */ + list_add(&handle->owners_link, &node->lock.owners); + handle->signaled = 0; +} + +/* Breaks a relation between a lock and its owner */ +static inline void unlink_object(lock_handle * handle) +{ + assert("zam-354", handle->owner != NULL); + assert("nikita-1608", handle->node != NULL); + assert_spin_locked(&(handle->node->lock.guard)); + assert("nikita-1829", handle->owner == get_current_lock_stack()); + assert("reiser4-5", handle->owner->nr_locks > 0); + + /* remove lock handle from lock_stack's list of locks */ + list_del(&handle->locks_link); + ON_DEBUG(handle->owner->nr_locks--); + reiser4_ctx_gfp_mask_set(); + assert("reiser4-6", + ergo(list_empty_careful(&handle->owner->locks), + handle->owner->nr_locks == 0)); + /* remove lock handle from znode's list of owners */ + list_del(&handle->owners_link); + /* indicates that lock handle is free now */ + handle->node = NULL; +#if REISER4_DEBUG + INIT_LIST_HEAD(&handle->locks_link); + INIT_LIST_HEAD(&handle->owners_link); + handle->owner = NULL; +#endif +} + +/* Actually locks an object knowing that we are able to do this */ +static void lock_object(lock_stack * owner) +{ + struct lock_request *request; + znode *node; + + request = &owner->request; + node = request->node; + assert_spin_locked(&(node->lock.guard)); + if (request->mode == ZNODE_READ_LOCK) { + node->lock.nr_readers++; + } else { + /* check that we don't switched from read to write lock */ + assert("nikita-1840", node->lock.nr_readers <= 0); + /* We allow recursive locking; a node can be locked several + times for write by same process */ + node->lock.nr_readers--; + } + + link_object(request->handle, owner, node); + + if (owner->curpri) + node->lock.nr_hipri_owners++; +} + +/* Check for recursive write locking */ +static int recursive(lock_stack * owner) +{ + int ret; + znode *node; + lock_handle *lh; + + node = owner->request.node; + + /* Owners list is not empty for a locked node */ + assert("zam-314", !list_empty_careful(&node->lock.owners)); + assert("nikita-1841", owner == get_current_lock_stack()); + assert_spin_locked(&(node->lock.guard)); + + lh = list_entry(node->lock.owners.next, lock_handle, owners_link); + ret = (lh->owner == owner); + + /* Recursive read locking should be done usual way */ + assert("zam-315", !ret || owner->request.mode == ZNODE_WRITE_LOCK); + /* mixing of read/write locks is not allowed */ + assert("zam-341", !ret || znode_is_wlocked(node)); + + return ret; +} + +#if REISER4_DEBUG +/* Returns true if the lock is held by the calling thread. */ +int znode_is_any_locked(const znode * node) +{ + lock_handle *handle; + lock_stack *stack; + int ret; + + if (!znode_is_locked(node)) + return 0; + + stack = get_current_lock_stack(); + + spin_lock_stack(stack); + + ret = 0; + + list_for_each_entry(handle, &stack->locks, locks_link) { + if (handle->node == node) { + ret = 1; + break; + } + } + + spin_unlock_stack(stack); + + return ret; +} + +#endif + +/* Returns true if a write lock is held by the calling thread. */ +int znode_is_write_locked(const znode * node) +{ + lock_stack *stack; + lock_handle *handle; + + assert("jmacd-8765", node != NULL); + + if (!znode_is_wlocked(node)) + return 0; + + stack = get_current_lock_stack(); + + /* + * When znode is write locked, all owner handles point to the same lock + * stack. Get pointer to lock stack from the first lock handle from + * znode's owner list + */ + handle = list_entry(node->lock.owners.next, lock_handle, owners_link); + + return (handle->owner == stack); +} + +/* This "deadlock" condition is the essential part of reiser4 locking + implementation. This condition is checked explicitly by calling + check_deadlock_condition() or implicitly in all places where znode lock + state (set of owners and request queue) is changed. Locking code is + designed to use this condition to trigger procedure of passing object from + low priority owner(s) to high priority one(s). + + The procedure results in passing an event (setting lock_handle->signaled + flag) and counting this event in nr_signaled field of owner's lock stack + object and wakeup owner's process. +*/ +static inline int check_deadlock_condition(znode * node) +{ + assert_spin_locked(&(node->lock.guard)); + return node->lock.nr_hipri_requests > 0 + && node->lock.nr_hipri_owners == 0; +} + +static int check_livelock_condition(znode * node, znode_lock_mode mode) +{ + zlock * lock = &node->lock; + + return mode == ZNODE_READ_LOCK && + lock->nr_readers >= 0 && lock->nr_hipri_write_requests > 0; +} + +/* checks lock/request compatibility */ +static int can_lock_object(lock_stack * owner) +{ + znode *node = owner->request.node; + + assert_spin_locked(&(node->lock.guard)); + + /* See if the node is disconnected. */ + if (unlikely(ZF_ISSET(node, JNODE_IS_DYING))) + return RETERR(-EINVAL); + + /* Do not ever try to take a lock if we are going in low priority + direction and a node have a high priority request without high + priority owners. */ + if (unlikely(!owner->curpri && check_deadlock_condition(node))) + return RETERR(-E_REPEAT); + if (unlikely(owner->curpri && + check_livelock_condition(node, owner->request.mode))) + return RETERR(-E_REPEAT); + if (unlikely(!is_lock_compatible(node, owner->request.mode))) + return RETERR(-E_REPEAT); + return 0; +} + +/* Setting of a high priority to the process. It clears "signaled" flags + because znode locked by high-priority process can't satisfy our "deadlock + condition". */ +static void set_high_priority(lock_stack * owner) +{ + assert("nikita-1846", owner == get_current_lock_stack()); + /* Do nothing if current priority is already high */ + if (!owner->curpri) { + /* We don't need locking for owner->locks list, because, this + * function is only called with the lock stack of the current + * thread, and no other thread can play with owner->locks list + * and/or change ->node pointers of lock handles in this list. + * + * (Interrupts also are not involved.) + */ + lock_handle *item = list_entry(owner->locks.next, lock_handle, + locks_link); + while (&owner->locks != &item->locks_link) { + znode *node = item->node; + + spin_lock_zlock(&node->lock); + + node->lock.nr_hipri_owners++; + + /* we can safely set signaled to zero, because + previous statement (nr_hipri_owners ++) guarantees + that signaled will be never set again. */ + item->signaled = 0; + spin_unlock_zlock(&node->lock); + + item = list_entry(item->locks_link.next, lock_handle, + locks_link); + } + owner->curpri = 1; + atomic_set(&owner->nr_signaled, 0); + } +} + +/* Sets a low priority to the process. */ +static void set_low_priority(lock_stack * owner) +{ + assert("nikita-3075", owner == get_current_lock_stack()); + /* Do nothing if current priority is already low */ + if (owner->curpri) { + /* scan all locks (lock handles) held by @owner, which is + actually current thread, and check whether we are reaching + deadlock possibility anywhere. + */ + lock_handle *handle = list_entry(owner->locks.next, lock_handle, + locks_link); + while (&owner->locks != &handle->locks_link) { + znode *node = handle->node; + spin_lock_zlock(&node->lock); + /* this thread just was hipri owner of @node, so + nr_hipri_owners has to be greater than zero. */ + assert("nikita-1835", node->lock.nr_hipri_owners > 0); + node->lock.nr_hipri_owners--; + /* If we have deadlock condition, adjust a nr_signaled + field. It is enough to set "signaled" flag only for + current process, other low-pri owners will be + signaled and waken up after current process unlocks + this object and any high-priority requestor takes + control. */ + if (check_deadlock_condition(node) + && !handle->signaled) { + handle->signaled = 1; + atomic_inc(&owner->nr_signaled); + } + spin_unlock_zlock(&node->lock); + handle = list_entry(handle->locks_link.next, + lock_handle, locks_link); + } + owner->curpri = 0; + } +} + +static void remove_lock_request(lock_stack * requestor) +{ + zlock * lock = &requestor->request.node->lock; + + if (requestor->curpri) { + assert("nikita-1838", lock->nr_hipri_requests > 0); + lock->nr_hipri_requests--; + if (requestor->request.mode == ZNODE_WRITE_LOCK) + lock->nr_hipri_write_requests--; + } + list_del(&requestor->requestors_link); +} + +static void invalidate_all_lock_requests(znode * node) +{ + lock_stack *requestor, *tmp; + + assert_spin_locked(&(node->lock.guard)); + + list_for_each_entry_safe(requestor, tmp, &node->lock.requestors, + requestors_link) { + remove_lock_request(requestor); + requestor->request.ret_code = -EINVAL; + reiser4_wake_up(requestor); + requestor->request.mode = ZNODE_NO_LOCK; + } +} + +static void dispatch_lock_requests(znode * node) +{ + lock_stack *requestor, *tmp; + + assert_spin_locked(&(node->lock.guard)); + + list_for_each_entry_safe(requestor, tmp, &node->lock.requestors, + requestors_link) { + if (znode_is_write_locked(node)) + break; + if (!can_lock_object(requestor)) { + lock_object(requestor); + remove_lock_request(requestor); + requestor->request.ret_code = 0; + reiser4_wake_up(requestor); + requestor->request.mode = ZNODE_NO_LOCK; + } + } +} + +/* release long-term lock, acquired by longterm_lock_znode() */ +void longterm_unlock_znode(lock_handle * handle) +{ + znode *node = handle->node; + lock_stack *oldowner = handle->owner; + int hipri; + int readers; + int rdelta; + int youdie; + + /* + * this is time-critical and highly optimized code. Modify carefully. + */ + + assert("jmacd-1021", handle != NULL); + assert("jmacd-1022", handle->owner != NULL); + assert("nikita-1392", LOCK_CNT_GTZ(long_term_locked_znode)); + + assert("zam-130", oldowner == get_current_lock_stack()); + + LOCK_CNT_DEC(long_term_locked_znode); + + /* + * to minimize amount of operations performed under lock, pre-compute + * all variables used within critical section. This makes code + * obscure. + */ + + /* was this lock of hi or lo priority */ + hipri = oldowner->curpri ? 1 : 0; + /* number of readers */ + readers = node->lock.nr_readers; + /* +1 if write lock, -1 if read lock */ + rdelta = (readers > 0) ? -1 : +1; + /* true if node is to die and write lock is released */ + youdie = ZF_ISSET(node, JNODE_HEARD_BANSHEE) && (readers < 0); + + spin_lock_zlock(&node->lock); + + assert("zam-101", znode_is_locked(node)); + + /* Adjust a number of high priority owners of this lock */ + assert("nikita-1836", node->lock.nr_hipri_owners >= hipri); + node->lock.nr_hipri_owners -= hipri; + + /* Handle znode deallocation on last write-lock release. */ + if (znode_is_wlocked_once(node)) { + if (youdie) { + forget_znode(handle); + assert("nikita-2191", znode_invariant(node)); + zput(node); + return; + } + } + + if (handle->signaled) + atomic_dec(&oldowner->nr_signaled); + + /* Unlocking means owner<->object link deletion */ + unlink_object(handle); + + /* This is enough to be sure whether an object is completely + unlocked. */ + node->lock.nr_readers += rdelta; + + /* If the node is locked it must have an owners list. Likewise, if + the node is unlocked it must have an empty owners list. */ + assert("zam-319", equi(znode_is_locked(node), + !list_empty_careful(&node->lock.owners))); + +#if REISER4_DEBUG + if (!znode_is_locked(node)) + ++node->times_locked; +#endif + + /* If there are pending lock requests we wake up a requestor */ + if (!znode_is_wlocked(node)) + dispatch_lock_requests(node); + if (check_deadlock_condition(node)) + wake_up_all_lopri_owners(node); + spin_unlock_zlock(&node->lock); + + /* minus one reference from handle->node */ + assert("nikita-2190", znode_invariant(node)); + ON_DEBUG(check_lock_data()); + ON_DEBUG(check_lock_node_data(node)); + zput(node); +} + +/* final portion of longterm-lock */ +static int +lock_tail(lock_stack * owner, int ok, znode_lock_mode mode) +{ + znode *node = owner->request.node; + + assert_spin_locked(&(node->lock.guard)); + + /* If we broke with (ok == 0) it means we can_lock, now do it. */ + if (ok == 0) { + lock_object(owner); + owner->request.mode = 0; + /* count a reference from lockhandle->node + + znode was already referenced at the entry to this function, + hence taking spin-lock here is not necessary (see comment + in the zref()). + */ + zref(node); + + LOCK_CNT_INC(long_term_locked_znode); + } + spin_unlock_zlock(&node->lock); + ON_DEBUG(check_lock_data()); + ON_DEBUG(check_lock_node_data(node)); + return ok; +} + +/* + * version of longterm_znode_lock() optimized for the most common case: read + * lock without any special flags. This is the kind of lock that any tree + * traversal takes on the root node of the tree, which is very frequent. + */ +static int longterm_lock_tryfast(lock_stack * owner) +{ + int result; + znode *node; + zlock *lock; + + node = owner->request.node; + lock = &node->lock; + + assert("nikita-3340", reiser4_schedulable()); + assert("nikita-3341", request_is_deadlock_safe(node, + ZNODE_READ_LOCK, + ZNODE_LOCK_LOPRI)); + spin_lock_zlock(lock); + result = can_lock_object(owner); + spin_unlock_zlock(lock); + + if (likely(result != -EINVAL)) { + spin_lock_znode(node); + result = reiser4_try_capture(ZJNODE(node), ZNODE_READ_LOCK, 0); + spin_unlock_znode(node); + spin_lock_zlock(lock); + if (unlikely(result != 0)) { + owner->request.mode = 0; + } else { + result = can_lock_object(owner); + if (unlikely(result == -E_REPEAT)) { + /* fall back to longterm_lock_znode() */ + spin_unlock_zlock(lock); + return 1; + } + } + return lock_tail(owner, result, ZNODE_READ_LOCK); + } else + return 1; +} + +/* locks given lock object */ +int longterm_lock_znode( + /* local link object (allocated by lock owner + * thread, usually on its own stack) */ + lock_handle * handle, + /* znode we want to lock. */ + znode * node, + /* {ZNODE_READ_LOCK, ZNODE_WRITE_LOCK}; */ + znode_lock_mode mode, + /* {0, -EINVAL, -E_DEADLOCK}, see return codes + description. */ + znode_lock_request request) { + int ret; + int hipri = (request & ZNODE_LOCK_HIPRI) != 0; + int non_blocking = 0; + int has_atom; + txn_capture cap_flags; + zlock *lock; + txn_handle *txnh; + tree_level level; + + /* Get current process context */ + lock_stack *owner = get_current_lock_stack(); + + /* Check that the lock handle is initialized and isn't already being + * used. */ + assert("jmacd-808", handle->owner == NULL); + assert("nikita-3026", reiser4_schedulable()); + assert("nikita-3219", request_is_deadlock_safe(node, mode, request)); + assert("zam-1056", atomic_read(&ZJNODE(node)->x_count) > 0); + /* long term locks are not allowed in the VM contexts (->writepage(), + * prune_{d,i}cache()). + * + * FIXME this doesn't work due to unused-dentry-with-unlinked-inode + * bug caused by d_splice_alias() only working for directories. + */ + assert("nikita-3547", 1 || ((current->flags & PF_MEMALLOC) == 0)); + assert("zam-1055", mode != ZNODE_NO_LOCK); + + cap_flags = 0; + if (request & ZNODE_LOCK_NONBLOCK) { + cap_flags |= TXN_CAPTURE_NONBLOCKING; + non_blocking = 1; + } + + if (request & ZNODE_LOCK_DONT_FUSE) + cap_flags |= TXN_CAPTURE_DONT_FUSE; + + /* If we are changing our process priority we must adjust a number + of high priority owners for each znode that we already lock */ + if (hipri) { + set_high_priority(owner); + } else { + set_low_priority(owner); + } + + level = znode_get_level(node); + + /* Fill request structure with our values. */ + owner->request.mode = mode; + owner->request.handle = handle; + owner->request.node = node; + + txnh = get_current_context()->trans; + lock = &node->lock; + + if (mode == ZNODE_READ_LOCK && request == 0) { + ret = longterm_lock_tryfast(owner); + if (ret <= 0) + return ret; + } + + has_atom = (txnh->atom != NULL); + + /* Synchronize on node's zlock guard lock. */ + spin_lock_zlock(lock); + + if (znode_is_locked(node) && + mode == ZNODE_WRITE_LOCK && recursive(owner)) + return lock_tail(owner, 0, mode); + + for (;;) { + /* Check the lock's availability: if it is unavaiable we get + E_REPEAT, 0 indicates "can_lock", otherwise the node is + invalid. */ + ret = can_lock_object(owner); + + if (unlikely(ret == -EINVAL)) { + /* @node is dying. Leave it alone. */ + break; + } + + if (unlikely(ret == -E_REPEAT && non_blocking)) { + /* either locking of @node by the current thread will + * lead to the deadlock, or lock modes are + * incompatible. */ + break; + } + + assert("nikita-1844", (ret == 0) + || ((ret == -E_REPEAT) && !non_blocking)); + /* If we can get the lock... Try to capture first before + taking the lock. */ + + /* first handle commonest case where node and txnh are already + * in the same atom. */ + /* safe to do without taking locks, because: + * + * 1. read of aligned word is atomic with respect to writes to + * this word + * + * 2. false negatives are handled in reiser4_try_capture(). + * + * 3. false positives are impossible. + * + * PROOF: left as an exercise to the curious reader. + * + * Just kidding. Here is one: + * + * At the time T0 txnh->atom is stored in txnh_atom. + * + * At the time T1 node->atom is stored in node_atom. + * + * At the time T2 we observe that + * + * txnh_atom != NULL && node_atom == txnh_atom. + * + * Imagine that at this moment we acquire node and txnh spin + * lock in this order. Suppose that under spin lock we have + * + * node->atom != txnh->atom, (S1) + * + * at the time T3. + * + * txnh->atom != NULL still, because txnh is open by the + * current thread. + * + * Suppose node->atom == NULL, that is, node was un-captured + * between T1, and T3. But un-capturing of formatted node is + * always preceded by the call to reiser4_invalidate_lock(), + * which marks znode as JNODE_IS_DYING under zlock spin + * lock. Contradiction, because can_lock_object() above checks + * for JNODE_IS_DYING. Hence, node->atom != NULL at T3. + * + * Suppose that node->atom != node_atom, that is, atom, node + * belongs to was fused into another atom: node_atom was fused + * into node->atom. Atom of txnh was equal to node_atom at T2, + * which means that under spin lock, txnh->atom == node->atom, + * because txnh->atom can only follow fusion + * chain. Contradicts S1. + * + * The same for hypothesis txnh->atom != txnh_atom. Hence, + * node->atom == node_atom == txnh_atom == txnh->atom. Again + * contradicts S1. Hence S1 is false. QED. + * + */ + + if (likely(has_atom && ZJNODE(node)->atom == txnh->atom)) { + ; + } else { + /* + * unlock zlock spin lock here. It is possible for + * longterm_unlock_znode() to sneak in here, but there + * is no harm: reiser4_invalidate_lock() will mark znode + * as JNODE_IS_DYING and this will be noted by + * can_lock_object() below. + */ + spin_unlock_zlock(lock); + spin_lock_znode(node); + ret = reiser4_try_capture(ZJNODE(node), mode, + cap_flags); + spin_unlock_znode(node); + spin_lock_zlock(lock); + if (unlikely(ret != 0)) { + /* In the failure case, the txnmgr releases + the znode's lock (or in some cases, it was + released a while ago). There's no need to + reacquire it so we should return here, + avoid releasing the lock. */ + owner->request.mode = 0; + break; + } + + /* Check the lock's availability again -- this is + because under some circumstances the capture code + has to release and reacquire the znode spinlock. */ + ret = can_lock_object(owner); + } + + /* This time, a return of (ret == 0) means we can lock, so we + should break out of the loop. */ + if (likely(ret != -E_REPEAT || non_blocking)) + break; + + /* Lock is unavailable, we have to wait. */ + ret = reiser4_prepare_to_sleep(owner); + if (unlikely(ret != 0)) + break; + + assert_spin_locked(&(node->lock.guard)); + if (hipri) { + /* If we are going in high priority direction then + increase high priority requests counter for the + node */ + lock->nr_hipri_requests++; + if (mode == ZNODE_WRITE_LOCK) + lock->nr_hipri_write_requests++; + /* If there are no high priority owners for a node, + then immediately wake up low priority owners, so + they can detect possible deadlock */ + if (lock->nr_hipri_owners == 0) + wake_up_all_lopri_owners(node); + } + list_add_tail(&owner->requestors_link, &lock->requestors); + + /* Ok, here we have prepared a lock request, so unlock + a znode ... */ + spin_unlock_zlock(lock); + /* ... and sleep */ + reiser4_go_to_sleep(owner); + if (owner->request.mode == ZNODE_NO_LOCK) + goto request_is_done; + spin_lock_zlock(lock); + if (owner->request.mode == ZNODE_NO_LOCK) { + spin_unlock_zlock(lock); +request_is_done: + if (owner->request.ret_code == 0) { + LOCK_CNT_INC(long_term_locked_znode); + zref(node); + } + return owner->request.ret_code; + } + remove_lock_request(owner); + } + + return lock_tail(owner, ret, mode); +} + +/* lock object invalidation means changing of lock object state to `INVALID' + and waiting for all other processes to cancel theirs lock requests. */ +void reiser4_invalidate_lock(lock_handle * handle /* path to lock + * owner and lock + * object is being + * invalidated. */ ) +{ + znode *node = handle->node; + lock_stack *owner = handle->owner; + + assert("zam-325", owner == get_current_lock_stack()); + assert("zam-103", znode_is_write_locked(node)); + assert("nikita-1393", !ZF_ISSET(node, JNODE_LEFT_CONNECTED)); + assert("nikita-1793", !ZF_ISSET(node, JNODE_RIGHT_CONNECTED)); + assert("nikita-1394", ZF_ISSET(node, JNODE_HEARD_BANSHEE)); + assert("nikita-3097", znode_is_wlocked_once(node)); + assert_spin_locked(&(node->lock.guard)); + + if (handle->signaled) + atomic_dec(&owner->nr_signaled); + + ZF_SET(node, JNODE_IS_DYING); + unlink_object(handle); + node->lock.nr_readers = 0; + + invalidate_all_lock_requests(node); + spin_unlock_zlock(&node->lock); +} + +/* Initializes lock_stack. */ +void init_lock_stack(lock_stack * owner /* pointer to + * allocated + * structure. */ ) +{ + INIT_LIST_HEAD(&owner->locks); + INIT_LIST_HEAD(&owner->requestors_link); + spin_lock_init(&owner->sguard); + owner->curpri = 1; + init_waitqueue_head(&owner->wait); +} + +/* Initializes lock object. */ +void reiser4_init_lock(zlock * lock /* pointer on allocated + * uninitialized lock object + * structure. */ ) +{ + memset(lock, 0, sizeof(zlock)); + spin_lock_init(&lock->guard); + INIT_LIST_HEAD(&lock->requestors); + INIT_LIST_HEAD(&lock->owners); +} + +/* Transfer a lock handle (presumably so that variables can be moved between + stack and heap locations). */ +static void +move_lh_internal(lock_handle * new, lock_handle * old, int unlink_old) +{ + znode *node = old->node; + lock_stack *owner = old->owner; + int signaled; + + /* locks_list, modified by link_object() is not protected by + anything. This is valid because only current thread ever modifies + locks_list of its lock_stack. + */ + assert("nikita-1827", owner == get_current_lock_stack()); + assert("nikita-1831", new->owner == NULL); + + spin_lock_zlock(&node->lock); + + signaled = old->signaled; + if (unlink_old) { + unlink_object(old); + } else { + if (node->lock.nr_readers > 0) { + node->lock.nr_readers += 1; + } else { + node->lock.nr_readers -= 1; + } + if (signaled) + atomic_inc(&owner->nr_signaled); + if (owner->curpri) + node->lock.nr_hipri_owners += 1; + LOCK_CNT_INC(long_term_locked_znode); + + zref(node); + } + link_object(new, owner, node); + new->signaled = signaled; + + spin_unlock_zlock(&node->lock); +} + +void move_lh(lock_handle * new, lock_handle * old) +{ + move_lh_internal(new, old, /*unlink_old */ 1); +} + +void copy_lh(lock_handle * new, lock_handle * old) +{ + move_lh_internal(new, old, /*unlink_old */ 0); +} + +/* after getting -E_DEADLOCK we unlock znodes until this function returns false + */ +int reiser4_check_deadlock(void) +{ + lock_stack *owner = get_current_lock_stack(); + return atomic_read(&owner->nr_signaled) != 0; +} + +/* Before going to sleep we re-check "release lock" requests which might come + from threads with hi-pri lock priorities. */ +int reiser4_prepare_to_sleep(lock_stack * owner) +{ + assert("nikita-1847", owner == get_current_lock_stack()); + + /* We return -E_DEADLOCK if one or more "give me the lock" messages are + * counted in nr_signaled */ + if (unlikely(atomic_read(&owner->nr_signaled) != 0)) { + assert("zam-959", !owner->curpri); + return RETERR(-E_DEADLOCK); + } + return 0; +} + +/* Wakes up a single thread */ +void __reiser4_wake_up(lock_stack * owner) +{ + atomic_set(&owner->wakeup, 1); + wake_up(&owner->wait); +} + +/* Puts a thread to sleep */ +void reiser4_go_to_sleep(lock_stack * owner) +{ + /* Well, we might sleep here, so holding of any spinlocks is no-no */ + assert("nikita-3027", reiser4_schedulable()); + + wait_event(owner->wait, atomic_read(&owner->wakeup)); + atomic_set(&owner->wakeup, 0); +} + +int lock_stack_isclean(lock_stack * owner) +{ + if (list_empty_careful(&owner->locks)) { + assert("zam-353", atomic_read(&owner->nr_signaled) == 0); + return 1; + } + + return 0; +} + +#if REISER4_DEBUG + +/* + * debugging functions + */ + +static void list_check(struct list_head *head) +{ + struct list_head *pos; + + list_for_each(pos, head) + assert("", (pos->prev != NULL && pos->next != NULL && + pos->prev->next == pos && pos->next->prev == pos)); +} + +/* check consistency of locking data-structures hanging of the @stack */ +static void check_lock_stack(lock_stack * stack) +{ + spin_lock_stack(stack); + /* check that stack->locks is not corrupted */ + list_check(&stack->locks); + spin_unlock_stack(stack); +} + +/* check consistency of locking data structures */ +void check_lock_data(void) +{ + check_lock_stack(&get_current_context()->stack); +} + +/* check consistency of locking data structures for @node */ +void check_lock_node_data(znode * node) +{ + spin_lock_zlock(&node->lock); + list_check(&node->lock.owners); + list_check(&node->lock.requestors); + spin_unlock_zlock(&node->lock); +} + +/* check that given lock request is dead lock safe. This check is, of course, + * not exhaustive. */ +static int +request_is_deadlock_safe(znode * node, znode_lock_mode mode, + znode_lock_request request) +{ + lock_stack *owner; + + owner = get_current_lock_stack(); + /* + * check that hipri lock request is not issued when there are locked + * nodes at the higher levels. + */ + if (request & ZNODE_LOCK_HIPRI && !(request & ZNODE_LOCK_NONBLOCK) && + znode_get_level(node) != 0) { + lock_handle *item; + + list_for_each_entry(item, &owner->locks, locks_link) { + znode *other; + + other = item->node; + + if (znode_get_level(other) == 0) + continue; + if (znode_get_level(other) > znode_get_level(node)) + return 0; + } + } + return 1; +} + +#endif + +/* return pointer to static storage with name of lock_mode. For + debugging */ +const char *lock_mode_name(znode_lock_mode lock/* lock mode to get name of */) +{ + if (lock == ZNODE_READ_LOCK) + return "read"; + else if (lock == ZNODE_WRITE_LOCK) + return "write"; + else { + static char buf[30]; + + sprintf(buf, "unknown: %i", lock); + return buf; + } +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 79 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/lock.h linux-2.6.30/fs/reiser4/lock.h --- linux-2.6.30.orig/fs/reiser4/lock.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/lock.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,250 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Long term locking data structures. See lock.c for details. */ + +#ifndef __LOCK_H__ +#define __LOCK_H__ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "plugin/node/node.h" +#include "txnmgr.h" +#include "readahead.h" + +#include +#include +#include /* for PAGE_CACHE_SIZE */ +#include +#include + +/* Per-znode lock object */ +struct zlock { + spinlock_t guard; + /* The number of readers if positive; the number of recursively taken + write locks if negative. Protected by zlock spin lock. */ + int nr_readers; + /* A number of processes (lock_stacks) that have this object + locked with high priority */ + unsigned nr_hipri_owners; + /* A number of attempts to lock znode in high priority direction */ + unsigned nr_hipri_requests; + /* A linked list of lock_handle objects that contains pointers + for all lock_stacks which have this lock object locked */ + unsigned nr_hipri_write_requests; + struct list_head owners; + /* A linked list of lock_stacks that wait for this lock */ + struct list_head requestors; +}; + +static inline void spin_lock_zlock(zlock *lock) +{ + /* check that zlock is not locked */ + assert("", LOCK_CNT_NIL(spin_locked_zlock)); + /* check that spinlocks of lower priorities are not held */ + assert("", LOCK_CNT_NIL(spin_locked_stack)); + + spin_lock(&lock->guard); + + LOCK_CNT_INC(spin_locked_zlock); + LOCK_CNT_INC(spin_locked); +} + +static inline void spin_unlock_zlock(zlock *lock) +{ + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_zlock)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(spin_locked_zlock); + LOCK_CNT_DEC(spin_locked); + + spin_unlock(&lock->guard); +} + +#define lock_is_locked(lock) ((lock)->nr_readers != 0) +#define lock_is_rlocked(lock) ((lock)->nr_readers > 0) +#define lock_is_wlocked(lock) ((lock)->nr_readers < 0) +#define lock_is_wlocked_once(lock) ((lock)->nr_readers == -1) +#define lock_can_be_rlocked(lock) ((lock)->nr_readers >= 0) +#define lock_mode_compatible(lock, mode) \ + (((mode) == ZNODE_WRITE_LOCK && !lock_is_locked(lock)) || \ + ((mode) == ZNODE_READ_LOCK && lock_can_be_rlocked(lock))) + +/* Since we have R/W znode locks we need additional bidirectional `link' + objects to implement n<->m relationship between lock owners and lock + objects. We call them `lock handles'. + + Locking: see lock.c/"SHORT-TERM LOCKING" +*/ +struct lock_handle { + /* This flag indicates that a signal to yield a lock was passed to + lock owner and counted in owner->nr_signalled + + Locking: this is accessed under spin lock on ->node. + */ + int signaled; + /* A link to owner of a lock */ + lock_stack *owner; + /* A link to znode locked */ + znode *node; + /* A list of all locks for a process */ + struct list_head locks_link; + /* A list of all owners for a znode */ + struct list_head owners_link; +}; + +struct lock_request { + /* A pointer to uninitialized link object */ + lock_handle *handle; + /* A pointer to the object we want to lock */ + znode *node; + /* Lock mode (ZNODE_READ_LOCK or ZNODE_WRITE_LOCK) */ + znode_lock_mode mode; + /* how dispatch_lock_requests() returns lock request result code */ + int ret_code; +}; + +/* A lock stack structure for accumulating locks owned by a process */ +struct lock_stack { + /* A guard lock protecting a lock stack */ + spinlock_t sguard; + /* number of znodes which were requested by high priority processes */ + atomic_t nr_signaled; + /* Current priority of a process + + This is only accessed by the current thread and thus requires no + locking. + */ + int curpri; + /* A list of all locks owned by this process. Elements can be added to + * this list only by the current thread. ->node pointers in this list + * can be only changed by the current thread. */ + struct list_head locks; + /* When lock_stack waits for the lock, it puts itself on double-linked + requestors list of that lock */ + struct list_head requestors_link; + /* Current lock request info. + + This is only accessed by the current thread and thus requires no + locking. + */ + struct lock_request request; + /* the following two fields are the lock stack's + * synchronization object to use with the standard linux/wait.h + * interface. See reiser4_go_to_sleep and __reiser4_wake_up for + * usage details. */ + wait_queue_head_t wait; + atomic_t wakeup; +#if REISER4_DEBUG + int nr_locks; /* number of lock handles in the above list */ +#endif +}; + +/* + User-visible znode locking functions +*/ + +extern int longterm_lock_znode(lock_handle * handle, + znode * node, + znode_lock_mode mode, + znode_lock_request request); + +extern void longterm_unlock_znode(lock_handle * handle); + +extern int reiser4_check_deadlock(void); + +extern lock_stack *get_current_lock_stack(void); + +extern void init_lock_stack(lock_stack * owner); +extern void reiser4_init_lock(zlock * lock); + +static inline void init_lh(lock_handle *lh) +{ +#if REISER4_DEBUG + memset(lh, 0, sizeof *lh); + INIT_LIST_HEAD(&lh->locks_link); + INIT_LIST_HEAD(&lh->owners_link); +#else + lh->node = NULL; +#endif +} + +static inline void done_lh(lock_handle *lh) +{ + assert("zam-342", lh != NULL); + if (lh->node != NULL) + longterm_unlock_znode(lh); +} + +extern void move_lh(lock_handle * new, lock_handle * old); +extern void copy_lh(lock_handle * new, lock_handle * old); + +extern int reiser4_prepare_to_sleep(lock_stack * owner); +extern void reiser4_go_to_sleep(lock_stack * owner); +extern void __reiser4_wake_up(lock_stack * owner); + +extern int lock_stack_isclean(lock_stack * owner); + +/* zlock object state check macros: only used in assertions. Both forms imply + that the lock is held by the current thread. */ +extern int znode_is_write_locked(const znode *); +extern void reiser4_invalidate_lock(lock_handle *); + +/* lock ordering is: first take zlock spin lock, then lock stack spin lock */ +#define spin_ordering_pred_stack(stack) \ + (LOCK_CNT_NIL(spin_locked_stack) && \ + LOCK_CNT_NIL(spin_locked_txnmgr) && \ + LOCK_CNT_NIL(spin_locked_inode) && \ + LOCK_CNT_NIL(rw_locked_cbk_cache) && \ + LOCK_CNT_NIL(spin_locked_super_eflush)) + +static inline void spin_lock_stack(lock_stack *stack) +{ + assert("", spin_ordering_pred_stack(stack)); + spin_lock(&(stack->sguard)); + LOCK_CNT_INC(spin_locked_stack); + LOCK_CNT_INC(spin_locked); +} + +static inline void spin_unlock_stack(lock_stack *stack) +{ + assert_spin_locked(&(stack->sguard)); + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_stack)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + LOCK_CNT_DEC(spin_locked_stack); + LOCK_CNT_DEC(spin_locked); + spin_unlock(&(stack->sguard)); +} + +static inline void reiser4_wake_up(lock_stack * owner) +{ + spin_lock_stack(owner); + __reiser4_wake_up(owner); + spin_unlock_stack(owner); +} + +const char *lock_mode_name(znode_lock_mode lock); + +#if REISER4_DEBUG +extern void check_lock_data(void); +extern void check_lock_node_data(znode * node); +#else +#define check_lock_data() noop +#define check_lock_node_data() noop +#endif + +/* __LOCK_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/Makefile linux-2.6.30/fs/reiser4/Makefile --- linux-2.6.30.orig/fs/reiser4/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,98 @@ +# +# reiser4/Makefile +# + +obj-$(CONFIG_REISER4_FS) += reiser4.o + +reiser4-y := \ + debug.o \ + jnode.o \ + znode.o \ + key.o \ + pool.o \ + tree_mod.o \ + estimate.o \ + carry.o \ + carry_ops.o \ + lock.o \ + tree.o \ + context.o \ + tap.o \ + coord.o \ + block_alloc.o \ + txnmgr.o \ + kassign.o \ + flush.o \ + wander.o \ + eottl.o \ + search.o \ + page_cache.o \ + seal.o \ + dscale.o \ + flush_queue.o \ + ktxnmgrd.o \ + blocknrset.o \ + super.o \ + super_ops.o \ + fsdata.o \ + export_ops.o \ + oid.o \ + tree_walk.o \ + inode.o \ + vfs_ops.o \ + as_ops.o \ + entd.o\ + readahead.o \ + status_flags.o \ + init_super.o \ + safe_link.o \ + \ + plugin/plugin.o \ + plugin/plugin_set.o \ + plugin/node/node.o \ + plugin/object.o \ + plugin/cluster.o \ + plugin/inode_ops.o \ + plugin/inode_ops_rename.o \ + plugin/file_ops.o \ + plugin/file_ops_readdir.o \ + plugin/file_plugin_common.o \ + plugin/file/file.o \ + plugin/file/tail_conversion.o \ + plugin/file/file_conversion.o \ + plugin/file/symlink.o \ + plugin/file/cryptcompress.o \ + plugin/dir_plugin_common.o \ + plugin/dir/hashed_dir.o \ + plugin/dir/seekable_dir.o \ + plugin/node/node40.o \ + \ + plugin/crypto/cipher.o \ + plugin/crypto/digest.o \ + \ + plugin/compress/compress.o \ + plugin/compress/compress_mode.o \ + \ + plugin/item/static_stat.o \ + plugin/item/sde.o \ + plugin/item/cde.o \ + plugin/item/blackbox.o \ + plugin/item/internal.o \ + plugin/item/tail.o \ + plugin/item/ctail.o \ + plugin/item/extent.o \ + plugin/item/extent_item_ops.o \ + plugin/item/extent_file_ops.o \ + plugin/item/extent_flush_ops.o \ + \ + plugin/hash.o \ + plugin/fibration.o \ + plugin/tail_policy.o \ + plugin/item/item.o \ + \ + plugin/security/perm.o \ + plugin/space/bitmap.o \ + \ + plugin/disk_format/disk_format40.o \ + plugin/disk_format/disk_format.o + diff -urN linux-2.6.30.orig/fs/reiser4/oid.c linux-2.6.30/fs/reiser4/oid.c --- linux-2.6.30.orig/fs/reiser4/oid.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/oid.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,141 @@ +/* Copyright 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "debug.h" +#include "super.h" +#include "txnmgr.h" + +/* we used to have oid allocation plugin. It was removed because it + was recognized as providing unneeded level of abstraction. If one + ever will find it useful - look at yet_unneeded_abstractions/oid +*/ + +/* + * initialize in-memory data for oid allocator at @super. @nr_files and @next + * are provided by disk format plugin that reads them from the disk during + * mount. + */ +int oid_init_allocator(struct super_block *super, oid_t nr_files, oid_t next) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(super); + + sbinfo->next_to_use = next; + sbinfo->oids_in_use = nr_files; + return 0; +} + +/* + * allocate oid and return it. ABSOLUTE_MAX_OID is returned when allocator + * runs out of oids. + */ +oid_t oid_allocate(struct super_block *super) +{ + reiser4_super_info_data *sbinfo; + oid_t oid; + + sbinfo = get_super_private(super); + + spin_lock_reiser4_super(sbinfo); + if (sbinfo->next_to_use != ABSOLUTE_MAX_OID) { + oid = sbinfo->next_to_use++; + sbinfo->oids_in_use++; + } else + oid = ABSOLUTE_MAX_OID; + spin_unlock_reiser4_super(sbinfo); + return oid; +} + +/* + * Tell oid allocator that @oid is now free. + */ +int oid_release(struct super_block *super, oid_t oid UNUSED_ARG) +{ + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(super); + + spin_lock_reiser4_super(sbinfo); + sbinfo->oids_in_use--; + spin_unlock_reiser4_super(sbinfo); + return 0; +} + +/* + * return next @oid that would be allocated (i.e., returned by oid_allocate()) + * without actually allocating it. This is used by disk format plugin to save + * oid allocator state on the disk. + */ +oid_t oid_next(const struct super_block *super) +{ + reiser4_super_info_data *sbinfo; + oid_t oid; + + sbinfo = get_super_private(super); + + spin_lock_reiser4_super(sbinfo); + oid = sbinfo->next_to_use; + spin_unlock_reiser4_super(sbinfo); + return oid; +} + +/* + * returns number of currently used oids. This is used by statfs(2) to report + * number of "inodes" and by disk format plugin to save oid allocator state on + * the disk. + */ +long oids_used(const struct super_block *super) +{ + reiser4_super_info_data *sbinfo; + oid_t used; + + sbinfo = get_super_private(super); + + spin_lock_reiser4_super(sbinfo); + used = sbinfo->oids_in_use; + spin_unlock_reiser4_super(sbinfo); + if (used < (__u64) ((long)~0) >> 1) + return (long)used; + else + return (long)-1; +} + +/* + * Count oid as allocated in atom. This is done after call to oid_allocate() + * at the point when we are irrevocably committed to creation of the new file + * (i.e., when oid allocation cannot be any longer rolled back due to some + * error). + */ +void oid_count_allocated(void) +{ + txn_atom *atom; + + atom = get_current_atom_locked(); + atom->nr_objects_created++; + spin_unlock_atom(atom); +} + +/* + * Count oid as free in atom. This is done after call to oid_release() at the + * point when we are irrevocably committed to the deletion of the file (i.e., + * when oid release cannot be any longer rolled back due to some error). + */ +void oid_count_released(void) +{ + txn_atom *atom; + + atom = get_current_atom_locked(); + atom->nr_objects_deleted++; + spin_unlock_atom(atom); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/page_cache.c linux-2.6.30/fs/reiser4/page_cache.c --- linux-2.6.30.orig/fs/reiser4/page_cache.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/page_cache.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,693 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Memory pressure hooks. Fake inodes handling. */ + +/* GLOSSARY + + . Formatted and unformatted nodes. + Elements of reiser4 balanced tree to store data and metadata. + Unformatted nodes are pointed to by extent pointers. Such nodes + are used to store data of large objects. Unlike unformatted nodes, + formatted ones have associated format described by node4X plugin. + + . Jnode (or journal node) + The in-memory header which is used to track formatted and unformatted + nodes, bitmap nodes, etc. In particular, jnodes are used to track + transactional information associated with each block(see reiser4/jnode.c + for details). + + . Znode + The in-memory header which is used to track formatted nodes. Contains + embedded jnode (see reiser4/znode.c for details). +*/ + +/* We store all file system meta data (and data, of course) in the page cache. + + What does this mean? In stead of using bread/brelse we create special + "fake" inode (one per super block) and store content of formatted nodes + into pages bound to this inode in the page cache. In newer kernels bread() + already uses inode attached to block device (bd_inode). Advantage of having + our own fake inode is that we can install appropriate methods in its + address_space operations. Such methods are called by VM on memory pressure + (or during background page flushing) and we can use them to react + appropriately. + + In initial version we only support one block per page. Support for multiple + blocks per page is complicated by relocation. + + To each page, used by reiser4, jnode is attached. jnode is analogous to + buffer head. Difference is that jnode is bound to the page permanently: + jnode cannot be removed from memory until its backing page is. + + jnode contain pointer to page (->pg field) and page contain pointer to + jnode in ->private field. Pointer from jnode to page is protected to by + jnode's spinlock and pointer from page to jnode is protected by page lock + (PG_locked bit). Lock ordering is: first take page lock, then jnode spin + lock. To go into reverse direction use jnode_lock_page() function that uses + standard try-lock-and-release device. + + Properties: + + 1. when jnode-to-page mapping is established (by jnode_attach_page()), page + reference counter is increased. + + 2. when jnode-to-page mapping is destroyed (by page_clear_jnode(), page + reference counter is decreased. + + 3. on jload() reference counter on jnode page is increased, page is + kmapped and `referenced'. + + 4. on jrelse() inverse operations are performed. + + 5. kmapping/kunmapping of unformatted pages is done by read/write methods. + + DEADLOCKS RELATED TO MEMORY PRESSURE. [OUTDATED. Only interesting + historically.] + + [In the following discussion, `lock' invariably means long term lock on + znode.] (What about page locks?) + + There is some special class of deadlock possibilities related to memory + pressure. Locks acquired by other reiser4 threads are accounted for in + deadlock prevention mechanism (lock.c), but when ->vm_writeback() is + invoked additional hidden arc is added to the locking graph: thread that + tries to allocate memory waits for ->vm_writeback() to finish. If this + thread keeps lock and ->vm_writeback() tries to acquire this lock, deadlock + prevention is useless. + + Another related problem is possibility for ->vm_writeback() to run out of + memory itself. This is not a problem for ext2 and friends, because their + ->vm_writeback() don't allocate much memory, but reiser4 flush is + definitely able to allocate huge amounts of memory. + + It seems that there is no reliable way to cope with the problems above. In + stead it was decided that ->vm_writeback() (as invoked in the kswapd + context) wouldn't perform any flushing itself, but rather should just wake + up some auxiliary thread dedicated for this purpose (or, the same thread + that does periodic commit of old atoms (ktxnmgrd.c)). + + Details: + + 1. Page is called `reclaimable' against particular reiser4 mount F if this + page can be ultimately released by try_to_free_pages() under presumptions + that: + + a. ->vm_writeback() for F is no-op, and + + b. none of the threads accessing F are making any progress, and + + c. other reiser4 mounts obey the same memory reservation protocol as F + (described below). + + For example, clean un-pinned page, or page occupied by ext2 data are + reclaimable against any reiser4 mount. + + When there is more than one reiser4 mount in a system, condition (c) makes + reclaim-ability not easily verifiable beyond trivial cases mentioned above. + + THIS COMMENT IS VALID FOR "MANY BLOCKS ON PAGE" CASE + + Fake inode is used to bound formatted nodes and each node is indexed within + fake inode by its block number. If block size of smaller than page size, it + may so happen that block mapped to the page with formatted node is occupied + by unformatted node or is unallocated. This lead to some complications, + because flushing whole page can lead to an incorrect overwrite of + unformatted node that is moreover, can be cached in some other place as + part of the file body. To avoid this, buffers for unformatted nodes are + never marked dirty. Also pages in the fake are never marked dirty. This + rules out usage of ->writepage() as memory pressure hook. In stead + ->releasepage() is used. + + Josh is concerned that page->buffer is going to die. This should not pose + significant problem though, because we need to add some data structures to + the page anyway (jnode) and all necessary book keeping can be put there. + +*/ + +/* Life cycle of pages/nodes. + + jnode contains reference to page and page contains reference back to + jnode. This reference is counted in page ->count. Thus, page bound to jnode + cannot be released back into free pool. + + 1. Formatted nodes. + + 1. formatted node is represented by znode. When new znode is created its + ->pg pointer is NULL initially. + + 2. when node content is loaded into znode (by call to zload()) for the + first time following happens (in call to ->read_node() or + ->allocate_node()): + + 1. new page is added to the page cache. + + 2. this page is attached to znode and its ->count is increased. + + 3. page is kmapped. + + 3. if more calls to zload() follow (without corresponding zrelses), page + counter is left intact and in its stead ->d_count is increased in znode. + + 4. each call to zrelse decreases ->d_count. When ->d_count drops to zero + ->release_node() is called and page is kunmapped as result. + + 5. at some moment node can be captured by a transaction. Its ->x_count + is then increased by transaction manager. + + 6. if node is removed from the tree (empty node with JNODE_HEARD_BANSHEE + bit set) following will happen (also see comment at the top of znode.c): + + 1. when last lock is released, node will be uncaptured from + transaction. This released reference that transaction manager acquired + at the step 5. + + 2. when last reference is released, zput() detects that node is + actually deleted and calls ->delete_node() + operation. page_cache_delete_node() implementation detaches jnode from + page and releases page. + + 7. otherwise (node wasn't removed from the tree), last reference to + znode will be released after transaction manager committed transaction + node was in. This implies squallocing of this node (see + flush.c). Nothing special happens at this point. Znode is still in the + hash table and page is still attached to it. + + 8. znode is actually removed from the memory because of the memory + pressure, or during umount (znodes_tree_done()). Anyway, znode is + removed by the call to zdrop(). At this moment, page is detached from + znode and removed from the inode address space. + +*/ + +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree.h" +#include "vfs_ops.h" +#include "inode.h" +#include "super.h" +#include "entd.h" +#include "page_cache.h" +#include "ktxnmgrd.h" + +#include +#include +#include /* for struct page */ +#include /* for struct page */ +#include +#include +#include +#include + +static struct bio *page_bio(struct page *, jnode * , int rw, gfp_t gfp); + +static struct address_space_operations formatted_fake_as_ops; + +static const oid_t fake_ino = 0x1; +static const oid_t bitmap_ino = 0x2; +static const oid_t cc_ino = 0x3; + +static void +init_fake_inode(struct super_block *super, struct inode *fake, + struct inode **pfake) +{ + assert("nikita-2168", fake->i_state & I_NEW); + fake->i_mapping->a_ops = &formatted_fake_as_ops; + *pfake = fake; + /* NOTE-NIKITA something else? */ + unlock_new_inode(fake); +} + +/** + * reiser4_init_formatted_fake - iget inodes for formatted nodes and bitmaps + * @super: super block to init fake inode for + * + * Initializes fake inode to which formatted nodes are bound in the page cache + * and inode for bitmaps. + */ +int reiser4_init_formatted_fake(struct super_block *super) +{ + struct inode *fake; + struct inode *bitmap; + struct inode *cc; + reiser4_super_info_data *sinfo; + + assert("nikita-1703", super != NULL); + + sinfo = get_super_private_nocheck(super); + fake = iget_locked(super, oid_to_ino(fake_ino)); + + if (fake != NULL) { + init_fake_inode(super, fake, &sinfo->fake); + + bitmap = iget_locked(super, oid_to_ino(bitmap_ino)); + if (bitmap != NULL) { + init_fake_inode(super, bitmap, &sinfo->bitmap); + + cc = iget_locked(super, oid_to_ino(cc_ino)); + if (cc != NULL) { + init_fake_inode(super, cc, &sinfo->cc); + return 0; + } else { + iput(sinfo->fake); + iput(sinfo->bitmap); + sinfo->fake = NULL; + sinfo->bitmap = NULL; + } + } else { + iput(sinfo->fake); + sinfo->fake = NULL; + } + } + return RETERR(-ENOMEM); +} + +/** + * reiser4_done_formatted_fake - release inode used by formatted nodes and bitmaps + * @super: super block to init fake inode for + * + * Releases inodes which were used as address spaces of bitmap and formatted + * nodes. + */ +void reiser4_done_formatted_fake(struct super_block *super) +{ + reiser4_super_info_data *sinfo; + + sinfo = get_super_private_nocheck(super); + + if (sinfo->fake != NULL) { + iput(sinfo->fake); + sinfo->fake = NULL; + } + + if (sinfo->bitmap != NULL) { + iput(sinfo->bitmap); + sinfo->bitmap = NULL; + } + + if (sinfo->cc != NULL) { + iput(sinfo->cc); + sinfo->cc = NULL; + } + return; +} + +void reiser4_wait_page_writeback(struct page *page) +{ + assert("zam-783", PageLocked(page)); + + do { + unlock_page(page); + wait_on_page_writeback(page); + lock_page(page); + } while (PageWriteback(page)); +} + +/* return tree @page is in */ +reiser4_tree *reiser4_tree_by_page(const struct page *page/* page to query */) +{ + assert("nikita-2461", page != NULL); + return &get_super_private(page->mapping->host->i_sb)->tree; +} + +/* completion handler for single page bio-based read. + + mpage_end_io_read() would also do. But it's static. + +*/ +static void +end_bio_single_page_read(struct bio *bio, int err UNUSED_ARG) +{ + struct page *page; + + page = bio->bi_io_vec[0].bv_page; + + if (test_bit(BIO_UPTODATE, &bio->bi_flags)) { + SetPageUptodate(page); + } else { + ClearPageUptodate(page); + SetPageError(page); + } + unlock_page(page); + bio_put(bio); +} + +/* completion handler for single page bio-based write. + + mpage_end_io_write() would also do. But it's static. + +*/ +static void +end_bio_single_page_write(struct bio *bio, int err UNUSED_ARG) +{ + struct page *page; + + page = bio->bi_io_vec[0].bv_page; + + if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) + SetPageError(page); + end_page_writeback(page); + bio_put(bio); +} + +/* ->readpage() method for formatted nodes */ +static int formatted_readpage(struct file *f UNUSED_ARG, + struct page *page/* page to read */) +{ + assert("nikita-2412", PagePrivate(page) && jprivate(page)); + return reiser4_page_io(page, jprivate(page), READ, + reiser4_ctx_gfp_mask_get()); +} + +/** + * reiser4_page_io - submit single-page bio request + * @page: page to perform io for + * @node: jnode of page + * @rw: read or write + * @gfp: gfp mask for bio allocation + * + * Submits single page read or write. + */ +int reiser4_page_io(struct page *page, jnode *node, int rw, gfp_t gfp) +{ + struct bio *bio; + int result; + + assert("nikita-2094", page != NULL); + assert("nikita-2226", PageLocked(page)); + assert("nikita-2634", node != NULL); + assert("nikita-2893", rw == READ || rw == WRITE); + + if (rw) { + if (unlikely(page->mapping->host->i_sb->s_flags & MS_RDONLY)) { + unlock_page(page); + return 0; + } + } + + bio = page_bio(page, node, rw, gfp); + if (!IS_ERR(bio)) { + if (rw == WRITE) { + set_page_writeback(page); + unlock_page(page); + } + reiser4_submit_bio(rw, bio); + result = 0; + } else { + unlock_page(page); + result = PTR_ERR(bio); + } + + return result; +} + +/* helper function to construct bio for page */ +static struct bio *page_bio(struct page *page, jnode * node, int rw, gfp_t gfp) +{ + struct bio *bio; + assert("nikita-2092", page != NULL); + assert("nikita-2633", node != NULL); + + /* Simple implementation in the assumption that blocksize == pagesize. + + We only have to submit one block, but submit_bh() will allocate bio + anyway, so lets use all the bells-and-whistles of bio code. + */ + + bio = bio_alloc(gfp, 1); + if (bio != NULL) { + int blksz; + struct super_block *super; + reiser4_block_nr blocknr; + + super = page->mapping->host->i_sb; + assert("nikita-2029", super != NULL); + blksz = super->s_blocksize; + assert("nikita-2028", blksz == (int)PAGE_CACHE_SIZE); + + spin_lock_jnode(node); + blocknr = *jnode_get_io_block(node); + spin_unlock_jnode(node); + + assert("nikita-2275", blocknr != (reiser4_block_nr) 0); + assert("nikita-2276", !reiser4_blocknr_is_fake(&blocknr)); + + bio->bi_bdev = super->s_bdev; + /* fill bio->bi_sector before calling bio_add_page(), because + * q->merge_bvec_fn may want to inspect it (see + * drivers/md/linear.c:linear_mergeable_bvec() for example. */ + bio->bi_sector = blocknr * (blksz >> 9); + + if (!bio_add_page(bio, page, blksz, 0)) { + warning("nikita-3452", + "Single page bio cannot be constructed"); + return ERR_PTR(RETERR(-EINVAL)); + } + + /* bio -> bi_idx is filled by bio_init() */ + bio->bi_end_io = (rw == READ) ? + end_bio_single_page_read : end_bio_single_page_write; + + return bio; + } else + return ERR_PTR(RETERR(-ENOMEM)); +} + +#if 0 +static int can_hit_entd(reiser4_context *ctx, struct super_block *s) +{ + if (ctx == NULL || ((unsigned long)ctx->magic) != context_magic) + return 1; + if (ctx->super != s) + return 1; + if (get_super_private(s)->entd.tsk == current) + return 0; + if (!lock_stack_isclean(&ctx->stack)) + return 0; + if (ctx->trans->atom != NULL) + return 0; + return 1; +} +#endif + +/** + * reiser4_writepage - writepage of struct address_space_operations + * @page: page to write + * @wbc: + * + * + */ +/* Common memory pressure notification. */ +int reiser4_writepage(struct page *page, + struct writeback_control *wbc) +{ + struct super_block *s; + reiser4_context *ctx; + + assert("vs-828", PageLocked(page)); + + s = page->mapping->host->i_sb; + ctx = get_current_context_check(); + + /* assert("", can_hit_entd(ctx, s)); */ + return write_page_by_ent(page, wbc); +} + +/* ->set_page_dirty() method of formatted address_space */ +static int formatted_set_page_dirty(struct page *page) +{ + assert("nikita-2173", page != NULL); + BUG(); + return __set_page_dirty_nobuffers(page); +} + +/* writepages method of address space operations in reiser4 is used to involve + into transactions pages which are dirtied via mmap. Only regular files can + have such pages. Fake inode is used to access formatted nodes via page + cache. As formatted nodes can never be mmaped, fake inode's writepages has + nothing to do */ +static int +writepages_fake(struct address_space *mapping, struct writeback_control *wbc) +{ + return 0; +} + +/* address space operations for the fake inode */ +static struct address_space_operations formatted_fake_as_ops = { + /* Perform a writeback of a single page as a memory-freeing + * operation. */ + .writepage = reiser4_writepage, + /* this is called to read formatted node */ + .readpage = formatted_readpage, + /* ->sync_page() method of fake inode address space operations. Called + from wait_on_page() and lock_page(). + + This is most annoyingly misnomered method. Actually it is called + from wait_on_page_bit() and lock_page() and its purpose is to + actually start io by jabbing device drivers. + */ + .sync_page = block_sync_page, + /* Write back some dirty pages from this mapping. Called from sync. + called during sync (pdflush) */ + .writepages = writepages_fake, + /* Set a page dirty */ + .set_page_dirty = formatted_set_page_dirty, + /* used for read-ahead. Not applicable */ + .readpages = NULL, + .write_begin = NULL, + .write_end = NULL, + .bmap = NULL, + /* called just before page is being detached from inode mapping and + removed from memory. Called on truncate, cut/squeeze, and + umount. */ + .invalidatepage = reiser4_invalidatepage, + /* this is called by shrink_cache() so that file system can try to + release objects (jnodes, buffers, journal heads) attached to page + and, may be made page itself free-able. + */ + .releasepage = reiser4_releasepage, + .direct_IO = NULL +}; + +/* called just before page is released (no longer used by reiser4). Callers: + jdelete() and extent2tail(). */ +void reiser4_drop_page(struct page *page) +{ + assert("nikita-2181", PageLocked(page)); + clear_page_dirty_for_io(page); + ClearPageUptodate(page); +#if defined(PG_skipped) + ClearPageSkipped(page); +#endif + unlock_page(page); +} + +#define JNODE_GANG_SIZE (16) + +/* find all jnodes from range specified and invalidate them */ +static int +truncate_jnodes_range(struct inode *inode, pgoff_t from, pgoff_t count) +{ + reiser4_inode *info; + int truncated_jnodes; + reiser4_tree *tree; + unsigned long index; + unsigned long end; + + if (inode_file_plugin(inode) == + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)) + /* + * No need to get rid of jnodes here: if the single jnode of + * page cluster did not have page, then it was found and killed + * before in + * truncate_complete_page_cluster()->jput()->jput_final(), + * otherwise it will be dropped by reiser4_invalidatepage() + */ + return 0; + truncated_jnodes = 0; + + info = reiser4_inode_data(inode); + tree = reiser4_tree_by_inode(inode); + + index = from; + end = from + count; + + while (1) { + jnode *gang[JNODE_GANG_SIZE]; + int taken; + int i; + jnode *node; + + assert("nikita-3466", index <= end); + + read_lock_tree(tree); + taken = + radix_tree_gang_lookup(jnode_tree_by_reiser4_inode(info), + (void **)gang, index, + JNODE_GANG_SIZE); + for (i = 0; i < taken; ++i) { + node = gang[i]; + if (index_jnode(node) < end) + jref(node); + else + gang[i] = NULL; + } + read_unlock_tree(tree); + + for (i = 0; i < taken; ++i) { + node = gang[i]; + if (node != NULL) { + index = max(index, index_jnode(node)); + spin_lock_jnode(node); + assert("edward-1457", node->pg == NULL); + /* this is always called after + truncate_inode_pages_range(). Therefore, here + jnode can not have page. New pages can not be + created because truncate_jnodes_range goes + under exclusive access on file obtained, + where as new page creation requires + non-exclusive access obtained */ + JF_SET(node, JNODE_HEARD_BANSHEE); + reiser4_uncapture_jnode(node); + unhash_unformatted_jnode(node); + truncated_jnodes++; + jput(node); + } else + break; + } + if (i != taken || taken == 0) + break; + } + return truncated_jnodes; +} + +/* Truncating files in reiser4: problems and solutions. + + VFS calls fs's truncate after it has called truncate_inode_pages() + to get rid of pages corresponding to part of file being truncated. + In reiser4 it may cause existence of unallocated extents which do + not have jnodes. Flush code does not expect that. Solution of this + problem is straightforward. As vfs's truncate is implemented using + setattr operation, it seems reasonable to have ->setattr() that + will cut file body. However, flush code also does not expect dirty + pages without parent items, so it is impossible to cut all items, + then truncate all pages in two steps. We resolve this problem by + cutting items one-by-one. Each such fine-grained step performed + under longterm znode lock calls at the end ->kill_hook() method of + a killed item to remove its binded pages and jnodes. + + The following function is a common part of mentioned kill hooks. + Also, this is called before tail-to-extent conversion (to not manage + few copies of the data). +*/ +void reiser4_invalidate_pages(struct address_space *mapping, pgoff_t from, + unsigned long count, int even_cows) +{ + loff_t from_bytes, count_bytes; + + if (count == 0) + return; + from_bytes = ((loff_t) from) << PAGE_CACHE_SHIFT; + count_bytes = ((loff_t) count) << PAGE_CACHE_SHIFT; + + unmap_mapping_range(mapping, from_bytes, count_bytes, even_cows); + truncate_inode_pages_range(mapping, from_bytes, + from_bytes + count_bytes - 1); + truncate_jnodes_range(mapping->host, from, count); +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/page_cache.h linux-2.6.30/fs/reiser4/page_cache.h --- linux-2.6.30.orig/fs/reiser4/page_cache.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/page_cache.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,66 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ +/* Memory pressure hooks. Fake inodes handling. See page_cache.c. */ + +#if !defined(__REISER4_PAGE_CACHE_H__) +#define __REISER4_PAGE_CACHE_H__ + +#include "forward.h" +#include "context.h" /* for reiser4_ctx_gfp_mask_get() */ + +#include /* for struct super_block, address_space */ +#include /* for struct page */ +#include /* for lock_page() */ +#include /* for __vmalloc() */ + +extern int reiser4_init_formatted_fake(struct super_block *); +extern void reiser4_done_formatted_fake(struct super_block *); + +extern reiser4_tree *reiser4_tree_by_page(const struct page *); + +#define reiser4_submit_bio(rw, bio) submit_bio((rw), (bio)) + +extern void reiser4_wait_page_writeback(struct page *); +static inline void lock_and_wait_page_writeback(struct page *page) +{ + lock_page(page); + if (unlikely(PageWriteback(page))) + reiser4_wait_page_writeback(page); +} + +#define jprivate(page) ((jnode *)page_private(page)) + +extern int reiser4_page_io(struct page *, jnode *, int rw, gfp_t); +extern void reiser4_drop_page(struct page *); +extern void reiser4_invalidate_pages(struct address_space *, pgoff_t from, + unsigned long count, int even_cows); +extern void capture_reiser4_inodes(struct super_block *, + struct writeback_control *); +static inline void *reiser4_vmalloc(unsigned long size) +{ + return __vmalloc(size, + reiser4_ctx_gfp_mask_get() | __GFP_HIGHMEM, + PAGE_KERNEL); +} + +#define PAGECACHE_TAG_REISER4_MOVED PAGECACHE_TAG_DIRTY + +#if REISER4_DEBUG +extern void print_page(const char *prefix, struct page *page); +#else +#define print_page(prf, p) noop +#endif + +/* __REISER4_PAGE_CACHE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/cluster.c linux-2.6.30/fs/reiser4/plugin/cluster.c --- linux-2.6.30.orig/fs/reiser4/plugin/cluster.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/cluster.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,72 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Contains reiser4 cluster plugins (see + http://www.namesys.com/cryptcompress_design.html + "Concepts of clustering" for details). */ + +#include "plugin_header.h" +#include "plugin.h" +#include "../inode.h" + +static int change_cluster(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + assert("edward-1324", inode != NULL); + assert("edward-1325", plugin != NULL); + assert("edward-1326", is_reiser4_inode(inode)); + assert("edward-1327", plugin->h.type_id == REISER4_CLUSTER_PLUGIN_TYPE); + + /* Can't change the cluster plugin for already existent regular files */ + if (!plugin_of_group(inode_file_plugin(inode), REISER4_DIRECTORY_FILE)) + return RETERR(-EINVAL); + + /* If matches, nothing to change. */ + if (inode_hash_plugin(inode) != NULL && + inode_hash_plugin(inode)->h.id == plugin->h.id) + return 0; + + return aset_set_unsafe(&reiser4_inode_data(inode)->pset, + PSET_CLUSTER, plugin); +} + +static reiser4_plugin_ops cluster_plugin_ops = { + .init = NULL, + .load = NULL, + .save_len = NULL, + .save = NULL, + .change = &change_cluster +}; + +#define SUPPORT_CLUSTER(SHIFT, ID, LABEL, DESC) \ + [CLUSTER_ ## ID ## _ID] = { \ + .h = { \ + .type_id = REISER4_CLUSTER_PLUGIN_TYPE, \ + .id = CLUSTER_ ## ID ## _ID, \ + .pops = &cluster_plugin_ops, \ + .label = LABEL, \ + .desc = DESC, \ + .linkage = {NULL, NULL} \ + }, \ + .shift = SHIFT \ + } + +cluster_plugin cluster_plugins[LAST_CLUSTER_ID] = { + SUPPORT_CLUSTER(16, 64K, "64K", "Large"), + SUPPORT_CLUSTER(15, 32K, "32K", "Big"), + SUPPORT_CLUSTER(14, 16K, "16K", "Average"), + SUPPORT_CLUSTER(13, 8K, "8K", "Small"), + SUPPORT_CLUSTER(12, 4K, "4K", "Minimal") +}; + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/cluster.h linux-2.6.30/fs/reiser4/plugin/cluster.h --- linux-2.6.30.orig/fs/reiser4/plugin/cluster.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/cluster.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,410 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* This file contains size/offset translators, modulators + and other helper functions. */ + +#if !defined(__FS_REISER4_CLUSTER_H__) +#define __FS_REISER4_CLUSTER_H__ + +#include "../inode.h" + +static inline int inode_cluster_shift(struct inode *inode) +{ + assert("edward-92", inode != NULL); + assert("edward-93", reiser4_inode_data(inode) != NULL); + + return inode_cluster_plugin(inode)->shift; +} + +static inline unsigned cluster_nrpages_shift(struct inode *inode) +{ + return inode_cluster_shift(inode) - PAGE_CACHE_SHIFT; +} + +/* cluster size in page units */ +static inline unsigned cluster_nrpages(struct inode *inode) +{ + return 1U << cluster_nrpages_shift(inode); +} + +static inline size_t inode_cluster_size(struct inode *inode) +{ + assert("edward-96", inode != NULL); + + return 1U << inode_cluster_shift(inode); +} + +static inline cloff_t pg_to_clust(pgoff_t idx, struct inode *inode) +{ + return idx >> cluster_nrpages_shift(inode); +} + +static inline pgoff_t clust_to_pg(cloff_t idx, struct inode *inode) +{ + return idx << cluster_nrpages_shift(inode); +} + +static inline pgoff_t pg_to_clust_to_pg(pgoff_t idx, struct inode *inode) +{ + return clust_to_pg(pg_to_clust(idx, inode), inode); +} + +static inline pgoff_t off_to_pg(loff_t off) +{ + return (off >> PAGE_CACHE_SHIFT); +} + +static inline loff_t pg_to_off(pgoff_t idx) +{ + return ((loff_t) (idx) << PAGE_CACHE_SHIFT); +} + +static inline cloff_t off_to_clust(loff_t off, struct inode *inode) +{ + return off >> inode_cluster_shift(inode); +} + +static inline loff_t clust_to_off(cloff_t idx, struct inode *inode) +{ + return (loff_t) idx << inode_cluster_shift(inode); +} + +static inline loff_t off_to_clust_to_off(loff_t off, struct inode *inode) +{ + return clust_to_off(off_to_clust(off, inode), inode); +} + +static inline pgoff_t off_to_clust_to_pg(loff_t off, struct inode *inode) +{ + return clust_to_pg(off_to_clust(off, inode), inode); +} + +static inline unsigned off_to_pgoff(loff_t off) +{ + return off & (PAGE_CACHE_SIZE - 1); +} + +static inline unsigned off_to_cloff(loff_t off, struct inode *inode) +{ + return off & ((loff_t) (inode_cluster_size(inode)) - 1); +} + +static inline pgoff_t offset_in_clust(struct page *page) +{ + assert("edward-1488", page != NULL); + assert("edward-1489", page->mapping != NULL); + + return page_index(page) & ((cluster_nrpages(page->mapping->host)) - 1); +} + +static inline int first_page_in_cluster(struct page *page) +{ + return offset_in_clust(page) == 0; +} + +static inline int last_page_in_cluster(struct page *page) +{ + return offset_in_clust(page) == + cluster_nrpages(page->mapping->host) - 1; +} + +static inline unsigned +pg_to_off_to_cloff(unsigned long idx, struct inode *inode) +{ + return off_to_cloff(pg_to_off(idx), inode); +} + +/*********************** Size translators **************************/ + +/* Translate linear size. + * New units are (1 << @blk_shift) times larger, then old ones. + * In other words, calculate number of logical blocks, occupied + * by @count elements + */ +static inline unsigned long size_in_blocks(loff_t count, unsigned blkbits) +{ + return (count + (1UL << blkbits) - 1) >> blkbits; +} + +/* size in pages */ +static inline pgoff_t size_in_pages(loff_t size) +{ + return size_in_blocks(size, PAGE_CACHE_SHIFT); +} + +/* size in logical clusters */ +static inline cloff_t size_in_lc(loff_t size, struct inode *inode) +{ + return size_in_blocks(size, inode_cluster_shift(inode)); +} + +/* size in pages to the size in page clusters */ +static inline cloff_t sp_to_spcl(pgoff_t size, struct inode *inode) +{ + return size_in_blocks(size, cluster_nrpages_shift(inode)); +} + +/*********************** Size modulators ***************************/ + +/* + Modulate linear size by nominated block size and offset. + + The "finite" function (which is zero almost everywhere). + How much is a height of the figure at a position @pos, + when trying to construct rectangle of height (1 << @blkbits), + and square @size. + + ****** + ******* + ******* + ******* + ----------> pos +*/ +static inline unsigned __mbb(loff_t size, unsigned long pos, int blkbits) +{ + unsigned end = size >> blkbits; + if (pos < end) + return 1U << blkbits; + if (unlikely(pos > end)) + return 0; + return size & ~(~0ull << blkbits); +} + +/* the same as above, but block size is page size */ +static inline unsigned __mbp(loff_t size, pgoff_t pos) +{ + return __mbb(size, pos, PAGE_CACHE_SHIFT); +} + +/* number of file's bytes in the nominated logical cluster */ +static inline unsigned lbytes(cloff_t index, struct inode *inode) +{ + return __mbb(i_size_read(inode), index, inode_cluster_shift(inode)); +} + +/* number of file's bytes in the nominated page */ +static inline unsigned pbytes(pgoff_t index, struct inode *inode) +{ + return __mbp(i_size_read(inode), index); +} + +/** + * number of pages occuped by @win->count bytes starting from + * @win->off at logical cluster defined by @win. This is exactly + * a number of pages to be modified and dirtied in any cluster operation. + */ +static inline pgoff_t win_count_to_nrpages(struct reiser4_slide * win) +{ + return ((win->off + win->count + + (1UL << PAGE_CACHE_SHIFT) - 1) >> PAGE_CACHE_SHIFT) - + off_to_pg(win->off); +} + +/* return true, if logical cluster is not occupied by the file */ +static inline int new_logical_cluster(struct cluster_handle *clust, + struct inode *inode) +{ + return clust_to_off(clust->index, inode) >= i_size_read(inode); +} + +/* return true, if pages @p1 and @p2 are of the same page cluster */ +static inline int same_page_cluster(struct page *p1, struct page *p2) +{ + assert("edward-1490", p1 != NULL); + assert("edward-1491", p2 != NULL); + assert("edward-1492", p1->mapping != NULL); + assert("edward-1493", p2->mapping != NULL); + + return (pg_to_clust(page_index(p1), p1->mapping->host) == + pg_to_clust(page_index(p2), p2->mapping->host)); +} + +static inline int cluster_is_complete(struct cluster_handle *clust, + struct inode *inode) +{ + return clust->tc.lsize == inode_cluster_size(inode); +} + +static inline void reiser4_slide_init(struct reiser4_slide *win) +{ + assert("edward-1084", win != NULL); + memset(win, 0, sizeof *win); +} + +static inline tfm_action +cluster_get_tfm_act(struct tfm_cluster *tc) +{ + assert("edward-1356", tc != NULL); + return tc->act; +} + +static inline void +cluster_set_tfm_act(struct tfm_cluster *tc, tfm_action act) +{ + assert("edward-1356", tc != NULL); + tc->act = act; +} + +static inline void cluster_init_act(struct cluster_handle *clust, + tfm_action act, + struct reiser4_slide *window) +{ + assert("edward-84", clust != NULL); + memset(clust, 0, sizeof *clust); + cluster_set_tfm_act(&clust->tc, act); + clust->dstat = INVAL_DISK_CLUSTER; + clust->win = window; +} + +static inline void cluster_init_read(struct cluster_handle *clust, + struct reiser4_slide *window) +{ + cluster_init_act(clust, TFMA_READ, window); +} + +static inline void cluster_init_write(struct cluster_handle *clust, + struct reiser4_slide *window) +{ + cluster_init_act(clust, TFMA_WRITE, window); +} + +/* true if @p1 and @p2 are items of the same disk cluster */ +static inline int same_disk_cluster(const coord_t *p1, const coord_t *p2) +{ + /* drop this if you have other items to aggregate */ + assert("edward-1494", item_id_by_coord(p1) == CTAIL_ID); + + return item_plugin_by_coord(p1)->b.mergeable(p1, p2); +} + +static inline int dclust_get_extension_dsize(hint_t *hint) +{ + return hint->ext_coord.extension.ctail.dsize; +} + +static inline void dclust_set_extension_dsize(hint_t *hint, int dsize) +{ + hint->ext_coord.extension.ctail.dsize = dsize; +} + +static inline int dclust_get_extension_shift(hint_t *hint) +{ + return hint->ext_coord.extension.ctail.shift; +} + +static inline int dclust_get_extension_ncount(hint_t *hint) +{ + return hint->ext_coord.extension.ctail.ncount; +} + +static inline void dclust_inc_extension_ncount(hint_t *hint) +{ + hint->ext_coord.extension.ctail.ncount++; +} + +static inline void dclust_init_extension(hint_t *hint) +{ + memset(&hint->ext_coord.extension.ctail, 0, + sizeof(hint->ext_coord.extension.ctail)); +} + +static inline int hint_is_unprepped_dclust(hint_t *hint) +{ + assert("edward-1451", hint_is_valid(hint)); + return dclust_get_extension_shift(hint) == (int)UCTAIL_SHIFT; +} + +static inline void coord_set_between_clusters(coord_t *coord) +{ +#if REISER4_DEBUG + int result; + result = zload(coord->node); + assert("edward-1296", !result); +#endif + if (!coord_is_between_items(coord)) { + coord->between = AFTER_ITEM; + coord->unit_pos = 0; + } +#if REISER4_DEBUG + zrelse(coord->node); +#endif +} + +int reiser4_inflate_cluster(struct cluster_handle *, struct inode *); +int find_disk_cluster(struct cluster_handle *, struct inode *, int read, + znode_lock_mode mode); +int checkout_logical_cluster(struct cluster_handle *, jnode * , struct inode *); +int reiser4_deflate_cluster(struct cluster_handle *, struct inode *); +void truncate_complete_page_cluster(struct inode *inode, cloff_t start, + int even_cows); +void invalidate_hint_cluster(struct cluster_handle *clust); +int get_disk_cluster_locked(struct cluster_handle *clust, struct inode *inode, + znode_lock_mode lock_mode); +void reset_cluster_params(struct cluster_handle *clust); +int set_cluster_by_page(struct cluster_handle *clust, struct page *page, + int count); +int prepare_page_cluster(struct inode *inode, struct cluster_handle *clust, + rw_op rw); +void __put_page_cluster(int from, int count, struct page **pages, + struct inode *inode); +void put_page_cluster(struct cluster_handle *clust, + struct inode *inode, rw_op rw); +void put_cluster_handle(struct cluster_handle *clust); +int grab_tfm_stream(struct inode *inode, struct tfm_cluster *tc, + tfm_stream_id id); +int tfm_cluster_is_uptodate(struct tfm_cluster *tc); +void tfm_cluster_set_uptodate(struct tfm_cluster *tc); +void tfm_cluster_clr_uptodate(struct tfm_cluster *tc); + +/* move cluster handle to the target position + specified by the page of index @pgidx */ +static inline void move_cluster_forward(struct cluster_handle *clust, + struct inode *inode, + pgoff_t pgidx) +{ + assert("edward-1297", clust != NULL); + assert("edward-1298", inode != NULL); + + reset_cluster_params(clust); + if (clust->index_valid && + /* Hole in the indices. Hint became invalid and can not be + used by find_cluster_item() even if seal/node versions + will coincide */ + pg_to_clust(pgidx, inode) != clust->index + 1) { + reiser4_unset_hint(clust->hint); + invalidate_hint_cluster(clust); + } + clust->index = pg_to_clust(pgidx, inode); + clust->index_valid = 1; +} + +static inline int alloc_clust_pages(struct cluster_handle *clust, + struct inode *inode) +{ + assert("edward-791", clust != NULL); + assert("edward-792", inode != NULL); + clust->pages = + kmalloc(sizeof(*clust->pages) << inode_cluster_shift(inode), + reiser4_ctx_gfp_mask_get()); + if (!clust->pages) + return -ENOMEM; + return 0; +} + +static inline void free_clust_pages(struct cluster_handle *clust) +{ + kfree(clust->pages); +} + +#endif /* __FS_REISER4_CLUSTER_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/compress/compress.c linux-2.6.30/fs/reiser4/plugin/compress/compress.c --- linux-2.6.30.orig/fs/reiser4/plugin/compress/compress.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/compress/compress.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,355 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* reiser4 compression transform plugins */ + +#include "../../debug.h" +#include "../../inode.h" +#include "../plugin.h" + +#include +#include +#include +#include + +static int change_compression(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + assert("edward-1316", inode != NULL); + assert("edward-1317", plugin != NULL); + assert("edward-1318", is_reiser4_inode(inode)); + assert("edward-1319", + plugin->h.type_id == REISER4_COMPRESSION_PLUGIN_TYPE); + + /* cannot change compression plugin of already existing regular object */ + if (!plugin_of_group(inode_file_plugin(inode), REISER4_DIRECTORY_FILE)) + return RETERR(-EINVAL); + + /* If matches, nothing to change. */ + if (inode_hash_plugin(inode) != NULL && + inode_hash_plugin(inode)->h.id == plugin->h.id) + return 0; + + return aset_set_unsafe(&reiser4_inode_data(inode)->pset, + PSET_COMPRESSION, plugin); +} + +static reiser4_plugin_ops compression_plugin_ops = { + .init = NULL, + .load = NULL, + .save_len = NULL, + .save = NULL, + .change = &change_compression +}; + +/******************************************************************************/ +/* gzip1 compression */ +/******************************************************************************/ + +#define GZIP1_DEF_LEVEL Z_BEST_SPEED +#define GZIP1_DEF_WINBITS 15 +#define GZIP1_DEF_MEMLEVEL MAX_MEM_LEVEL + +static int gzip1_init(void) +{ + return 0; +} + +static int gzip1_overrun(unsigned src_len UNUSED_ARG) +{ + return 0; +} + +static coa_t gzip1_alloc(tfm_action act) +{ + coa_t coa = NULL; + int ret = 0; + switch (act) { + case TFMA_WRITE: /* compress */ + coa = reiser4_vmalloc(zlib_deflate_workspacesize()); + if (!coa) { + ret = -ENOMEM; + break; + } + break; + case TFMA_READ: /* decompress */ + coa = reiser4_vmalloc(zlib_inflate_workspacesize()); + if (!coa) { + ret = -ENOMEM; + break; + } + break; + default: + impossible("edward-767", + "trying to alloc workspace for unknown tfm action"); + } + if (ret) { + warning("edward-768", + "alloc workspace for gzip1 (tfm action = %d) failed\n", + act); + return ERR_PTR(ret); + } + return coa; +} + +static void gzip1_free(coa_t coa, tfm_action act) +{ + assert("edward-769", coa != NULL); + + switch (act) { + case TFMA_WRITE: /* compress */ + vfree(coa); + break; + case TFMA_READ: /* decompress */ + vfree(coa); + break; + default: + impossible("edward-770", "unknown tfm action"); + } + return; +} + +static int gzip1_min_size_deflate(void) +{ + return 64; +} + +static void +gzip1_compress(coa_t coa, __u8 * src_first, unsigned src_len, + __u8 * dst_first, unsigned *dst_len) +{ + int ret = 0; + struct z_stream_s stream; + + assert("edward-842", coa != NULL); + assert("edward-875", src_len != 0); + + stream.workspace = coa; + ret = zlib_deflateInit2(&stream, GZIP1_DEF_LEVEL, Z_DEFLATED, + -GZIP1_DEF_WINBITS, GZIP1_DEF_MEMLEVEL, + Z_DEFAULT_STRATEGY); + if (ret != Z_OK) { + warning("edward-771", "zlib_deflateInit2 returned %d\n", ret); + goto rollback; + } + ret = zlib_deflateReset(&stream); + if (ret != Z_OK) { + warning("edward-772", "zlib_deflateReset returned %d\n", ret); + goto rollback; + } + stream.next_in = src_first; + stream.avail_in = src_len; + stream.next_out = dst_first; + stream.avail_out = *dst_len; + + ret = zlib_deflate(&stream, Z_FINISH); + if (ret != Z_STREAM_END) { + if (ret != Z_OK) + warning("edward-773", + "zlib_deflate returned %d\n", ret); + goto rollback; + } + *dst_len = stream.total_out; + return; + rollback: + *dst_len = src_len; + return; +} + +static void +gzip1_decompress(coa_t coa, __u8 * src_first, unsigned src_len, + __u8 * dst_first, unsigned *dst_len) +{ + int ret = 0; + struct z_stream_s stream; + + assert("edward-843", coa != NULL); + assert("edward-876", src_len != 0); + + stream.workspace = coa; + ret = zlib_inflateInit2(&stream, -GZIP1_DEF_WINBITS); + if (ret != Z_OK) { + warning("edward-774", "zlib_inflateInit2 returned %d\n", ret); + return; + } + ret = zlib_inflateReset(&stream); + if (ret != Z_OK) { + warning("edward-775", "zlib_inflateReset returned %d\n", ret); + return; + } + + stream.next_in = src_first; + stream.avail_in = src_len; + stream.next_out = dst_first; + stream.avail_out = *dst_len; + + ret = zlib_inflate(&stream, Z_SYNC_FLUSH); + /* + * Work around a bug in zlib, which sometimes wants to taste an extra + * byte when being used in the (undocumented) raw deflate mode. + * (From USAGI). + */ + if (ret == Z_OK && !stream.avail_in && stream.avail_out) { + u8 zerostuff = 0; + stream.next_in = &zerostuff; + stream.avail_in = 1; + ret = zlib_inflate(&stream, Z_FINISH); + } + if (ret != Z_STREAM_END) { + warning("edward-776", "zlib_inflate returned %d\n", ret); + return; + } + *dst_len = stream.total_out; + return; +} + +/******************************************************************************/ +/* lzo1 compression */ +/******************************************************************************/ + +static int lzo1_init(void) +{ + return 0; +} + +static int lzo1_overrun(unsigned in_len) +{ + return in_len / 64 + 16 + 3; +} + +static coa_t lzo1_alloc(tfm_action act) +{ + int ret = 0; + coa_t coa = NULL; + + switch (act) { + case TFMA_WRITE: /* compress */ + coa = reiser4_vmalloc(LZO1X_1_MEM_COMPRESS); + if (!coa) { + ret = -ENOMEM; + break; + } + case TFMA_READ: /* decompress */ + break; + default: + impossible("edward-877", + "trying to alloc workspace for unknown tfm action"); + } + if (ret) { + warning("edward-878", + "alloc workspace for lzo1 (tfm action = %d) failed\n", + act); + return ERR_PTR(ret); + } + return coa; +} + +static void lzo1_free(coa_t coa, tfm_action act) +{ + assert("edward-879", coa != NULL); + + switch (act) { + case TFMA_WRITE: /* compress */ + vfree(coa); + break; + case TFMA_READ: /* decompress */ + impossible("edward-1304", + "trying to free non-allocated workspace"); + default: + impossible("edward-880", "unknown tfm action"); + } + return; +} + +static int lzo1_min_size_deflate(void) +{ + return 256; +} + +static void +lzo1_compress(coa_t coa, __u8 * src_first, unsigned src_len, + __u8 * dst_first, unsigned *dst_len) +{ + int result; + + assert("edward-846", coa != NULL); + assert("edward-847", src_len != 0); + + result = lzo1x_1_compress(src_first, src_len, dst_first, dst_len, coa); + if (unlikely(result != LZO_E_OK)) { + warning("edward-849", "lzo1x_1_compress failed\n"); + goto out; + } + if (*dst_len >= src_len) { + //warning("edward-850", "lzo1x_1_compress: incompressible data\n"); + goto out; + } + return; + out: + *dst_len = src_len; + return; +} + +static void +lzo1_decompress(coa_t coa, __u8 * src_first, unsigned src_len, + __u8 * dst_first, unsigned *dst_len) +{ + int result; + + assert("edward-851", coa == NULL); + assert("edward-852", src_len != 0); + + result = lzo1x_decompress_safe(src_first, src_len, dst_first, dst_len); + if (result != LZO_E_OK) + warning("edward-853", "lzo1x_1_decompress failed\n"); + return; +} + +compression_plugin compression_plugins[LAST_COMPRESSION_ID] = { + [LZO1_COMPRESSION_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_PLUGIN_TYPE, + .id = LZO1_COMPRESSION_ID, + .pops = &compression_plugin_ops, + .label = "lzo1", + .desc = "lzo1 compression transform", + .linkage = {NULL, NULL} + }, + .init = lzo1_init, + .overrun = lzo1_overrun, + .alloc = lzo1_alloc, + .free = lzo1_free, + .min_size_deflate = lzo1_min_size_deflate, + .checksum = reiser4_adler32, + .compress = lzo1_compress, + .decompress = lzo1_decompress + }, + [GZIP1_COMPRESSION_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_PLUGIN_TYPE, + .id = GZIP1_COMPRESSION_ID, + .pops = &compression_plugin_ops, + .label = "gzip1", + .desc = "gzip1 compression transform", + .linkage = {NULL, NULL} + }, + .init = gzip1_init, + .overrun = gzip1_overrun, + .alloc = gzip1_alloc, + .free = gzip1_free, + .min_size_deflate = gzip1_min_size_deflate, + .checksum = reiser4_adler32, + .compress = gzip1_compress, + .decompress = gzip1_decompress + } +}; + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/compress/compress.h linux-2.6.30/fs/reiser4/plugin/compress/compress.h --- linux-2.6.30.orig/fs/reiser4/plugin/compress/compress.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/compress/compress.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,43 @@ +#if !defined( __FS_REISER4_COMPRESS_H__ ) +#define __FS_REISER4_COMPRESS_H__ + +#include +#include + +/* transform direction */ +typedef enum { + TFMA_READ, /* decrypt, decompress */ + TFMA_WRITE, /* encrypt, compress */ + TFMA_LAST +} tfm_action; + +/* supported compression algorithms */ +typedef enum { + LZO1_COMPRESSION_ID, + GZIP1_COMPRESSION_ID, + LAST_COMPRESSION_ID, +} reiser4_compression_id; + +/* the same as pgoff, but units are page clusters */ +typedef unsigned long cloff_t; + +/* working data of a (de)compression algorithm */ +typedef void *coa_t; + +/* table for all supported (de)compression algorithms */ +typedef coa_t coa_set[LAST_COMPRESSION_ID][TFMA_LAST]; + +__u32 reiser4_adler32(char *data, __u32 len); + +#endif /* __FS_REISER4_COMPRESS_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/compress/compress_mode.c linux-2.6.30/fs/reiser4/plugin/compress/compress_mode.c --- linux-2.6.30.orig/fs/reiser4/plugin/compress/compress_mode.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/compress/compress_mode.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,162 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* This file contains Reiser4 compression mode plugins. + + Compression mode plugin is a set of handlers called by compressor + at flush time and represent some heuristics including the ones + which are to avoid compression of incompressible data, see + http://www.namesys.com/cryptcompress_design.html for more details. +*/ +#include "../../inode.h" +#include "../plugin.h" + +static int should_deflate_none(struct inode * inode, cloff_t index) +{ + return 0; +} + +static int should_deflate_common(struct inode * inode, cloff_t index) +{ + return compression_is_on(cryptcompress_inode_data(inode)); +} + +static int discard_hook_ultim(struct inode *inode, cloff_t index) +{ + turn_off_compression(cryptcompress_inode_data(inode)); + return 0; +} + +static int discard_hook_lattd(struct inode *inode, cloff_t index) +{ + struct cryptcompress_info * info = cryptcompress_inode_data(inode); + + assert("edward-1462", + get_lattice_factor(info) >= MIN_LATTICE_FACTOR && + get_lattice_factor(info) <= MAX_LATTICE_FACTOR); + + turn_off_compression(info); + if (get_lattice_factor(info) < MAX_LATTICE_FACTOR) + set_lattice_factor(info, get_lattice_factor(info) << 1); + return 0; +} + +static int accept_hook_lattd(struct inode *inode, cloff_t index) +{ + turn_on_compression(cryptcompress_inode_data(inode)); + set_lattice_factor(cryptcompress_inode_data(inode), MIN_LATTICE_FACTOR); + return 0; +} + +/* Check on dynamic lattice, the adaptive compression modes which + defines the following behavior: + + Compression is on: try to compress everything and turn + it off, whenever cluster is incompressible. + + Compression is off: try to compress clusters of indexes + k * FACTOR (k = 0, 1, 2, ...) and turn it on, if some of + them is compressible. If incompressible, then increase FACTOR */ + +/* check if @index belongs to one-dimensional lattice + of sparce factor @factor */ +static int is_on_lattice(cloff_t index, int factor) +{ + return (factor ? index % factor == 0: index == 0); +} + +static int should_deflate_lattd(struct inode * inode, cloff_t index) +{ + return should_deflate_common(inode, index) || + is_on_lattice(index, + get_lattice_factor + (cryptcompress_inode_data(inode))); +} + +/* compression mode_plugins */ +compression_mode_plugin compression_mode_plugins[LAST_COMPRESSION_MODE_ID] = { + [NONE_COMPRESSION_MODE_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .id = NONE_COMPRESSION_MODE_ID, + .pops = NULL, + .label = "none", + .desc = "Compress nothing", + .linkage = {NULL, NULL} + }, + .should_deflate = should_deflate_none, + .accept_hook = NULL, + .discard_hook = NULL + }, + /* Check-on-dynamic-lattice adaptive compression mode */ + [LATTD_COMPRESSION_MODE_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .id = LATTD_COMPRESSION_MODE_ID, + .pops = NULL, + .label = "lattd", + .desc = "Check on dynamic lattice", + .linkage = {NULL, NULL} + }, + .should_deflate = should_deflate_lattd, + .accept_hook = accept_hook_lattd, + .discard_hook = discard_hook_lattd + }, + /* Check-ultimately compression mode: + Turn off compression forever as soon as we meet + incompressible data */ + [ULTIM_COMPRESSION_MODE_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .id = ULTIM_COMPRESSION_MODE_ID, + .pops = NULL, + .label = "ultim", + .desc = "Check ultimately", + .linkage = {NULL, NULL} + }, + .should_deflate = should_deflate_common, + .accept_hook = NULL, + .discard_hook = discard_hook_ultim + }, + /* Force-to-compress-everything compression mode */ + [FORCE_COMPRESSION_MODE_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .id = FORCE_COMPRESSION_MODE_ID, + .pops = NULL, + .label = "force", + .desc = "Force to compress everything", + .linkage = {NULL, NULL} + }, + .should_deflate = NULL, + .accept_hook = NULL, + .discard_hook = NULL + }, + /* Convert-to-extent compression mode. + In this mode items will be converted to extents and management + will be passed to (classic) unix file plugin as soon as ->write() + detects that the first complete logical cluster (of index #0) is + incompressible. */ + [CONVX_COMPRESSION_MODE_ID] = { + .h = { + .type_id = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .id = CONVX_COMPRESSION_MODE_ID, + .pops = NULL, + .label = "conv", + .desc = "Convert to extent", + .linkage = {NULL, NULL} + }, + .should_deflate = should_deflate_common, + .accept_hook = NULL, + .discard_hook = NULL + } +}; + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/compress/Makefile linux-2.6.30/fs/reiser4/plugin/compress/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/compress/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/compress/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,5 @@ +obj-$(CONFIG_REISER4_FS) += compress_plugins.o + +compress_plugins-objs := \ + compress.o \ + compress_mode.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/crypto/cipher.c linux-2.6.30/fs/reiser4/plugin/crypto/cipher.c --- linux-2.6.30.orig/fs/reiser4/plugin/crypto/cipher.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/crypto/cipher.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,37 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, + licensing governed by reiser4/README */ +/* Reiser4 cipher transform plugins */ + +#include "../../debug.h" +#include "../plugin.h" + +cipher_plugin cipher_plugins[LAST_CIPHER_ID] = { + [NONE_CIPHER_ID] = { + .h = { + .type_id = REISER4_CIPHER_PLUGIN_TYPE, + .id = NONE_CIPHER_ID, + .pops = NULL, + .label = "none", + .desc = "no cipher transform", + .linkage = {NULL, NULL} + }, + .alloc = NULL, + .free = NULL, + .scale = NULL, + .align_stream = NULL, + .setkey = NULL, + .encrypt = NULL, + .decrypt = NULL + } +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/crypto/cipher.h linux-2.6.30/fs/reiser4/plugin/crypto/cipher.h --- linux-2.6.30.orig/fs/reiser4/plugin/crypto/cipher.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/crypto/cipher.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,55 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* This file contains definitions for the objects operated + by reiser4 key manager, which is something like keyring + wrapped by appropriate reiser4 plugin */ + +#if !defined( __FS_REISER4_CRYPT_H__ ) +#define __FS_REISER4_CRYPT_H__ + +#include + +/* key info imported from user space */ +struct reiser4_crypto_data { + int keysize; /* uninstantiated key size */ + __u8 * key; /* uninstantiated key */ + int keyid_size; /* size of passphrase */ + __u8 * keyid; /* passphrase */ +}; + +/* This object contains all needed infrastructure to implement + cipher transform. This is operated (allocating, inheriting, + validating, binding to host inode, etc..) by reiser4 key manager. + + This info can be allocated in two cases: + 1. importing a key from user space. + 2. reading inode from disk */ +struct reiser4_crypto_info { + struct inode * host; + struct crypto_hash * digest; + struct crypto_blkcipher * cipher; +#if 0 + cipher_key_plugin * kplug; /* key manager */ +#endif + __u8 * keyid; /* key fingerprint, created by digest plugin, + using uninstantiated key and passphrase. + supposed to be stored in disk stat-data */ + int inst; /* this indicates if the cipher key is + instantiated (case 1 above) */ + int keysize; /* uninstantiated key size (bytes), supposed + to be stored in disk stat-data */ + int keyload_count; /* number of the objects which has this + crypto-stat attached */ +}; + +#endif /* __FS_REISER4_CRYPT_H__ */ + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/crypto/digest.c linux-2.6.30/fs/reiser4/plugin/crypto/digest.c --- linux-2.6.30.orig/fs/reiser4/plugin/crypto/digest.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/crypto/digest.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,58 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* reiser4 digest transform plugin (is used by cryptcompress object plugin) */ +/* EDWARD-FIXME-HANS: and it does what? a digest is a what? */ +#include "../../debug.h" +#include "../plugin_header.h" +#include "../plugin.h" +#include "../file/cryptcompress.h" + +#include + +extern digest_plugin digest_plugins[LAST_DIGEST_ID]; + +static struct crypto_hash * alloc_sha256 (void) +{ +#if REISER4_SHA256 + return crypto_alloc_hash ("sha256", 0, CRYPTO_ALG_ASYNC); +#else + warning("edward-1418", "sha256 unsupported"); + return ERR_PTR(-EINVAL); +#endif +} + +static void free_sha256 (struct crypto_hash * tfm) +{ +#if REISER4_SHA256 + crypto_free_hash(tfm); +#endif + return; +} + +/* digest plugins */ +digest_plugin digest_plugins[LAST_DIGEST_ID] = { + [SHA256_32_DIGEST_ID] = { + .h = { + .type_id = REISER4_DIGEST_PLUGIN_TYPE, + .id = SHA256_32_DIGEST_ID, + .pops = NULL, + .label = "sha256_32", + .desc = "sha256_32 digest transform", + .linkage = {NULL, NULL} + }, + .fipsize = sizeof(__u32), + .alloc = alloc_sha256, + .free = free_sha256 + } +}; + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/dir/dir.h linux-2.6.30/fs/reiser4/plugin/dir/dir.h --- linux-2.6.30.orig/fs/reiser4/plugin/dir/dir.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/dir/dir.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,36 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* this file contains declarations of methods implementing directory plugins */ + +#if !defined( __REISER4_DIR_H__ ) +#define __REISER4_DIR_H__ + +/*#include "../../key.h" + +#include */ + +/* declarations of functions implementing HASHED_DIR_PLUGIN_ID dir plugin */ + +/* "hashed" directory methods of dir plugin */ +void build_entry_key_hashed(const struct inode *, const struct qstr *, + reiser4_key *); + +/* declarations of functions implementing SEEKABLE_HASHED_DIR_PLUGIN_ID dir plugin */ + +/* "seekable" directory methods of dir plugin */ +void build_entry_key_seekable(const struct inode *, const struct qstr *, + reiser4_key *); + +/* __REISER4_DIR_H__ */ +#endif + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/dir/hashed_dir.c linux-2.6.30/fs/reiser4/plugin/dir/hashed_dir.c --- linux-2.6.30.orig/fs/reiser4/plugin/dir/hashed_dir.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/dir/hashed_dir.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,81 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Directory plugin using hashes (see fs/reiser4/plugin/hash.c) to map file + names to the files. */ + +/* + * Hashed directory logically consists of persistent directory + * entries. Directory entry is a pair of a file name and a key of stat-data of + * a file that has this name in the given directory. + * + * Directory entries are stored in the tree in the form of directory + * items. Directory item should implement dir_entry_ops portion of item plugin + * interface (see plugin/item/item.h). Hashed directory interacts with + * directory item plugin exclusively through dir_entry_ops operations. + * + * Currently there are two implementations of directory items: "simple + * directory item" (plugin/item/sde.[ch]), and "compound directory item" + * (plugin/item/cde.[ch]) with the latter being the default. + * + * There is, however some delicate way through which directory code interferes + * with item plugin: key assignment policy. A key for a directory item is + * chosen by directory code, and as described in kassign.c, this key contains + * a portion of file name. Directory item uses this knowledge to avoid storing + * this portion of file name twice: in the key and in the directory item body. + * + */ + +#include "../../inode.h" + +void complete_entry_key(const struct inode *, const char *name, + int len, reiser4_key * result); + +/* this is implementation of build_entry_key method of dir + plugin for HASHED_DIR_PLUGIN_ID + */ +void build_entry_key_hashed(const struct inode *dir, /* directory where entry is + * (or will be) in.*/ + const struct qstr *qname, /* name of file referenced + * by this entry */ + reiser4_key * result /* resulting key of directory + * entry */ ) +{ + const char *name; + int len; + + assert("nikita-1139", dir != NULL); + assert("nikita-1140", qname != NULL); + assert("nikita-1141", qname->name != NULL); + assert("nikita-1142", result != NULL); + + name = qname->name; + len = qname->len; + + assert("nikita-2867", strlen(name) == len); + + reiser4_key_init(result); + /* locality of directory entry's key is objectid of parent + directory */ + set_key_locality(result, get_inode_oid(dir)); + /* minor packing locality is constant */ + set_key_type(result, KEY_FILE_NAME_MINOR); + /* dot is special case---we always want it to be first entry in + a directory. Actually, we just want to have smallest + directory entry. + */ + if (len == 1 && name[0] == '.') + return; + + /* initialize part of entry key which depends on file name */ + complete_entry_key(dir, name, len, result); +} + +/* Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/dir/Makefile linux-2.6.30/fs/reiser4/plugin/dir/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/dir/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/dir/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,5 @@ +obj-$(CONFIG_REISER4_FS) += dir_plugins.o + +dir_plugins-objs := \ + hashed_dir.o \ + seekable_dir.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/dir/seekable_dir.c linux-2.6.30/fs/reiser4/plugin/dir/seekable_dir.c --- linux-2.6.30.orig/fs/reiser4/plugin/dir/seekable_dir.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/dir/seekable_dir.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,46 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "../../inode.h" + +/* this is implementation of build_entry_key method of dir + plugin for SEEKABLE_HASHED_DIR_PLUGIN_ID + This is for directories where we want repeatable and restartable readdir() + even in case 32bit user level struct dirent (readdir(3)). +*/ +void +build_entry_key_seekable(const struct inode *dir, const struct qstr *name, + reiser4_key * result) +{ + oid_t objectid; + + assert("nikita-2283", dir != NULL); + assert("nikita-2284", name != NULL); + assert("nikita-2285", name->name != NULL); + assert("nikita-2286", result != NULL); + + reiser4_key_init(result); + /* locality of directory entry's key is objectid of parent + directory */ + set_key_locality(result, get_inode_oid(dir)); + /* minor packing locality is constant */ + set_key_type(result, KEY_FILE_NAME_MINOR); + /* dot is special case---we always want it to be first entry in + a directory. Actually, we just want to have smallest + directory entry. + */ + if ((name->len == 1) && (name->name[0] == '.')) + return; + + /* objectid of key is 31 lowest bits of hash. */ + objectid = + inode_hash_plugin(dir)->hash(name->name, + (int)name->len) & 0x7fffffff; + + assert("nikita-2303", !(objectid & ~KEY_OBJECTID_MASK)); + set_key_objectid(result, objectid); + + /* offset is always 0. */ + set_key_offset(result, (__u64) 0); + return; +} diff -urN linux-2.6.30.orig/fs/reiser4/plugin/dir_plugin_common.c linux-2.6.30/fs/reiser4/plugin/dir_plugin_common.c --- linux-2.6.30.orig/fs/reiser4/plugin/dir_plugin_common.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/dir_plugin_common.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,865 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + reiser4/README */ + +/* this file contains typical implementations for most of methods of + directory plugin +*/ + +#include "../inode.h" + +int reiser4_find_entry(struct inode *dir, struct dentry *name, + lock_handle * , znode_lock_mode, reiser4_dir_entry_desc *); +int reiser4_lookup_name(struct inode *parent, struct dentry *dentry, + reiser4_key * key); +void check_light_weight(struct inode *inode, struct inode *parent); + +/* this is common implementation of get_parent method of dir plugin + this is used by NFS kernel server to "climb" up directory tree to + check permissions + */ +struct dentry *get_parent_common(struct inode *child) +{ + struct super_block *s; + struct inode *parent; + struct dentry dotdot; + struct dentry *dentry; + reiser4_key key; + int result; + + /* + * lookup dotdot entry. + */ + + s = child->i_sb; + memset(&dotdot, 0, sizeof(dotdot)); + dotdot.d_name.name = ".."; + dotdot.d_name.len = 2; + dotdot.d_op = &get_super_private(s)->ops.dentry; + + result = reiser4_lookup_name(child, &dotdot, &key); + if (result != 0) + return ERR_PTR(result); + + parent = reiser4_iget(s, &key, 1); + if (!IS_ERR(parent)) { + /* + * FIXME-NIKITA dubious: attributes are inherited from @child + * to @parent. But: + * + * (*) this is the only this we can do + * + * (*) attributes of light-weight object are inherited + * from a parent through which object was looked up first, + * so it is ambiguous anyway. + * + */ + check_light_weight(parent, child); + reiser4_iget_complete(parent); + dentry = d_obtain_alias(parent); + if (!IS_ERR(dentry)) + dentry->d_op = &get_super_private(s)->ops.dentry; + } else if (PTR_ERR(parent) == -ENOENT) + dentry = ERR_PTR(RETERR(-ESTALE)); + else + dentry = (void *)parent; + return dentry; +} + +/* this is common implementation of is_name_acceptable method of dir + plugin + */ +int is_name_acceptable_common(const struct inode *inode, /* directory to check*/ + const char *name UNUSED_ARG, /* name to check */ + int len/* @name's length */) +{ + assert("nikita-733", inode != NULL); + assert("nikita-734", name != NULL); + assert("nikita-735", len > 0); + + return len <= reiser4_max_filename_len(inode); +} + +/* there is no common implementation of build_entry_key method of dir + plugin. See plugin/dir/hashed_dir.c:build_entry_key_hashed() or + plugin/dir/seekable.c:build_entry_key_seekable() for example +*/ + +/* this is common implementation of build_readdir_key method of dir + plugin + see reiser4_readdir_common for more details +*/ +int build_readdir_key_common(struct file *dir /* directory being read */ , + reiser4_key * result/* where to store key */) +{ + reiser4_file_fsdata *fdata; + struct inode *inode; + + assert("nikita-1361", dir != NULL); + assert("nikita-1362", result != NULL); + assert("nikita-1363", dir->f_dentry != NULL); + inode = dir->f_dentry->d_inode; + assert("nikita-1373", inode != NULL); + + fdata = reiser4_get_file_fsdata(dir); + if (IS_ERR(fdata)) + return PTR_ERR(fdata); + assert("nikita-1364", fdata != NULL); + return extract_key_from_de_id(get_inode_oid(inode), + &fdata->dir.readdir.position. + dir_entry_key, result); + +} + +void reiser4_adjust_dir_file(struct inode *, const struct dentry *, int offset, + int adj); + +/* this is common implementation of add_entry method of dir plugin +*/ +int reiser4_add_entry_common(struct inode *object, /* directory to add new name + * in */ + struct dentry *where, /* new name */ + reiser4_object_create_data * data, /* parameters of + * new object */ + reiser4_dir_entry_desc * entry /* parameters of + * new directory + * entry */) +{ + int result; + coord_t *coord; + lock_handle lh; + struct reiser4_dentry_fsdata *fsdata; + reiser4_block_nr reserve; + + assert("nikita-1114", object != NULL); + assert("nikita-1250", where != NULL); + + fsdata = reiser4_get_dentry_fsdata(where); + if (unlikely(IS_ERR(fsdata))) + return PTR_ERR(fsdata); + + reserve = inode_dir_plugin(object)->estimate.add_entry(object); + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) + return RETERR(-ENOSPC); + + init_lh(&lh); + coord = &fsdata->dec.entry_coord; + coord_clear_iplug(coord); + + /* check for this entry in a directory. This is plugin method. */ + result = reiser4_find_entry(object, where, &lh, ZNODE_WRITE_LOCK, + entry); + if (likely(result == -ENOENT)) { + /* add new entry. Just pass control to the directory + item plugin. */ + assert("nikita-1709", inode_dir_item_plugin(object)); + assert("nikita-2230", coord->node == lh.node); + reiser4_seal_done(&fsdata->dec.entry_seal); + result = + inode_dir_item_plugin(object)->s.dir.add_entry(object, + coord, &lh, + where, + entry); + if (result == 0) { + reiser4_adjust_dir_file(object, where, + fsdata->dec.pos + 1, +1); + INODE_INC_FIELD(object, i_size); + } + } else if (result == 0) { + assert("nikita-2232", coord->node == lh.node); + result = RETERR(-EEXIST); + } + done_lh(&lh); + + return result; +} + +/** + * rem_entry - remove entry from directory item + * @dir: + * @dentry: + * @entry: + * @coord: + * @lh: + * + * Checks that coordinate @coord is set properly and calls item plugin + * method to cut entry. + */ +static int +rem_entry(struct inode *dir, struct dentry *dentry, + reiser4_dir_entry_desc * entry, coord_t *coord, lock_handle * lh) +{ + item_plugin *iplug; + struct inode *child; + + iplug = inode_dir_item_plugin(dir); + child = dentry->d_inode; + assert("nikita-3399", child != NULL); + + /* check that we are really destroying an entry for @child */ + if (REISER4_DEBUG) { + int result; + reiser4_key key; + + result = iplug->s.dir.extract_key(coord, &key); + if (result != 0) + return result; + if (get_key_objectid(&key) != get_inode_oid(child)) { + warning("nikita-3397", + "rem_entry: %#llx != %#llx\n", + get_key_objectid(&key), + (unsigned long long)get_inode_oid(child)); + return RETERR(-EIO); + } + } + return iplug->s.dir.rem_entry(dir, &dentry->d_name, coord, lh, entry); +} + +/** + * reiser4_rem_entry_common - remove entry from a directory + * @dir: directory to remove entry from + * @where: name that is being removed + * @entry: description of entry being removed + * + * This is common implementation of rem_entry method of dir plugin. + */ +int reiser4_rem_entry_common(struct inode *dir, + struct dentry *dentry, + reiser4_dir_entry_desc * entry) +{ + int result; + coord_t *coord; + lock_handle lh; + struct reiser4_dentry_fsdata *fsdata; + __u64 tograb; + + assert("nikita-1124", dir != NULL); + assert("nikita-1125", dentry != NULL); + + tograb = inode_dir_plugin(dir)->estimate.rem_entry(dir); + result = reiser4_grab_space(tograb, BA_CAN_COMMIT | BA_RESERVED); + if (result != 0) + return RETERR(-ENOSPC); + + init_lh(&lh); + + /* check for this entry in a directory. This is plugin method. */ + result = reiser4_find_entry(dir, dentry, &lh, ZNODE_WRITE_LOCK, entry); + fsdata = reiser4_get_dentry_fsdata(dentry); + if (IS_ERR(fsdata)) { + done_lh(&lh); + return PTR_ERR(fsdata); + } + + coord = &fsdata->dec.entry_coord; + + assert("nikita-3404", + get_inode_oid(dentry->d_inode) != get_inode_oid(dir) || + dir->i_size <= 1); + + coord_clear_iplug(coord); + if (result == 0) { + /* remove entry. Just pass control to the directory item + plugin. */ + assert("vs-542", inode_dir_item_plugin(dir)); + reiser4_seal_done(&fsdata->dec.entry_seal); + reiser4_adjust_dir_file(dir, dentry, fsdata->dec.pos, -1); + result = + WITH_COORD(coord, + rem_entry(dir, dentry, entry, coord, &lh)); + if (result == 0) { + if (dir->i_size >= 1) + INODE_DEC_FIELD(dir, i_size); + else { + warning("nikita-2509", "Dir %llu is runt", + (unsigned long long) + get_inode_oid(dir)); + result = RETERR(-EIO); + } + + assert("nikita-3405", dentry->d_inode->i_nlink != 1 || + dentry->d_inode->i_size != 2 || + inode_dir_plugin(dentry->d_inode) == NULL); + } + } + done_lh(&lh); + + return result; +} + +static reiser4_block_nr estimate_init(struct inode *parent, + struct inode *object); +static int create_dot_dotdot(struct inode *object, struct inode *parent); + +/* this is common implementation of init method of dir plugin + create "." and ".." entries +*/ +int reiser4_dir_init_common(struct inode *object, /* new directory */ + struct inode *parent, /* parent directory */ + reiser4_object_create_data * data /* info passed + * to us, this + * is filled by + * reiser4() + * syscall in + * particular */) +{ + reiser4_block_nr reserve; + + assert("nikita-680", object != NULL); + assert("nikita-681", S_ISDIR(object->i_mode)); + assert("nikita-682", parent != NULL); + assert("nikita-684", data != NULL); + assert("nikita-686", data->id == DIRECTORY_FILE_PLUGIN_ID); + assert("nikita-687", object->i_mode & S_IFDIR); + + reserve = estimate_init(parent, object); + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) + return RETERR(-ENOSPC); + + return create_dot_dotdot(object, parent); +} + +/* this is common implementation of done method of dir plugin + remove "." entry +*/ +int reiser4_dir_done_common(struct inode *object/* object being deleted */) +{ + int result; + reiser4_block_nr reserve; + struct dentry goodby_dots; + reiser4_dir_entry_desc entry; + + assert("nikita-1449", object != NULL); + + if (reiser4_inode_get_flag(object, REISER4_NO_SD)) + return 0; + + /* of course, this can be rewritten to sweep everything in one + reiser4_cut_tree(). */ + memset(&entry, 0, sizeof entry); + + /* FIXME: this done method is called from reiser4_delete_dir_common + * which reserved space already */ + reserve = inode_dir_plugin(object)->estimate.rem_entry(object); + if (reiser4_grab_space(reserve, BA_CAN_COMMIT | BA_RESERVED)) + return RETERR(-ENOSPC); + + memset(&goodby_dots, 0, sizeof goodby_dots); + entry.obj = goodby_dots.d_inode = object; + goodby_dots.d_name.name = "."; + goodby_dots.d_name.len = 1; + result = reiser4_rem_entry_common(object, &goodby_dots, &entry); + reiser4_free_dentry_fsdata(&goodby_dots); + if (unlikely(result != 0 && result != -ENOMEM && result != -ENOENT)) + warning("nikita-2252", "Cannot remove dot of %lli: %i", + (unsigned long long)get_inode_oid(object), result); + return 0; +} + +/* this is common implementation of attach method of dir plugin +*/ +int reiser4_attach_common(struct inode *child UNUSED_ARG, + struct inode *parent UNUSED_ARG) +{ + assert("nikita-2647", child != NULL); + assert("nikita-2648", parent != NULL); + + return 0; +} + +/* this is common implementation of detach method of dir plugin + remove "..", decrease nlink on parent +*/ +int reiser4_detach_common(struct inode *object, struct inode *parent) +{ + int result; + struct dentry goodby_dots; + reiser4_dir_entry_desc entry; + + assert("nikita-2885", object != NULL); + assert("nikita-2886", !reiser4_inode_get_flag(object, REISER4_NO_SD)); + + memset(&entry, 0, sizeof entry); + + /* NOTE-NIKITA this only works if @parent is -the- parent of + @object, viz. object whose key is stored in dotdot + entry. Wouldn't work with hard-links on directories. */ + memset(&goodby_dots, 0, sizeof goodby_dots); + entry.obj = goodby_dots.d_inode = parent; + goodby_dots.d_name.name = ".."; + goodby_dots.d_name.len = 2; + result = reiser4_rem_entry_common(object, &goodby_dots, &entry); + reiser4_free_dentry_fsdata(&goodby_dots); + if (result == 0) { + /* the dot should be the only entry remaining at this time... */ + assert("nikita-3400", + object->i_size == 1 && object->i_nlink <= 2); +#if 0 + /* and, together with the only name directory can have, they + * provides for the last 2 remaining references. If we get + * here as part of error handling during mkdir, @object + * possibly has no name yet, so its nlink == 1. If we get here + * from rename (targeting empty directory), it has no name + * already, so its nlink == 1. */ + assert("nikita-3401", + object->i_nlink == 2 || object->i_nlink == 1); +#endif + + /* decrement nlink of directory removed ".." pointed + to */ + reiser4_del_nlink(parent, NULL, 0); + } + return result; +} + +/* this is common implementation of estimate.add_entry method of + dir plugin + estimation of adding entry which supposes that entry is inserting a + unit into item +*/ +reiser4_block_nr estimate_add_entry_common(const struct inode *inode) +{ + return estimate_one_insert_into_item(reiser4_tree_by_inode(inode)); +} + +/* this is common implementation of estimate.rem_entry method of dir + plugin +*/ +reiser4_block_nr estimate_rem_entry_common(const struct inode *inode) +{ + return estimate_one_item_removal(reiser4_tree_by_inode(inode)); +} + +/* this is common implementation of estimate.unlink method of dir + plugin +*/ +reiser4_block_nr +dir_estimate_unlink_common(const struct inode *parent, + const struct inode *object) +{ + reiser4_block_nr res; + + /* hashed_rem_entry(object) */ + res = inode_dir_plugin(object)->estimate.rem_entry(object); + /* del_nlink(parent) */ + res += 2 * inode_file_plugin(parent)->estimate.update(parent); + + return res; +} + +/* + * helper for inode_ops ->lookup() and dir plugin's ->get_parent() + * methods: if @inode is a light-weight file, setup its credentials + * that are not stored in the stat-data in this case + */ +void check_light_weight(struct inode *inode, struct inode *parent) +{ + if (reiser4_inode_get_flag(inode, REISER4_LIGHT_WEIGHT)) { + inode->i_uid = parent->i_uid; + inode->i_gid = parent->i_gid; + /* clear light-weight flag. If inode would be read by any + other name, [ug]id wouldn't change. */ + reiser4_inode_clr_flag(inode, REISER4_LIGHT_WEIGHT); + } +} + +/* looks for name specified in @dentry in directory @parent and if name is + found - key of object found entry points to is stored in @entry->key */ +int reiser4_lookup_name(struct inode *parent, /* inode of directory to lookup + * for name in */ + struct dentry *dentry, /* name to look for */ + reiser4_key * key/* place to store key */) +{ + int result; + coord_t *coord; + lock_handle lh; + const char *name; + int len; + reiser4_dir_entry_desc entry; + struct reiser4_dentry_fsdata *fsdata; + + assert("nikita-1247", parent != NULL); + assert("nikita-1248", dentry != NULL); + assert("nikita-1123", dentry->d_name.name != NULL); + assert("vs-1486", + dentry->d_op == &get_super_private(parent->i_sb)->ops.dentry); + + name = dentry->d_name.name; + len = dentry->d_name.len; + + if (!inode_dir_plugin(parent)->is_name_acceptable(parent, name, len)) + /* some arbitrary error code to return */ + return RETERR(-ENAMETOOLONG); + + fsdata = reiser4_get_dentry_fsdata(dentry); + if (IS_ERR(fsdata)) + return PTR_ERR(fsdata); + + coord = &fsdata->dec.entry_coord; + coord_clear_iplug(coord); + init_lh(&lh); + + /* find entry in a directory. This is plugin method. */ + result = reiser4_find_entry(parent, dentry, &lh, ZNODE_READ_LOCK, + &entry); + if (result == 0) { + /* entry was found, extract object key from it. */ + result = + WITH_COORD(coord, + item_plugin_by_coord(coord)->s.dir. + extract_key(coord, key)); + } + done_lh(&lh); + return result; + +} + +/* helper for reiser4_dir_init_common(): estimate number of blocks to reserve */ +static reiser4_block_nr +estimate_init(struct inode *parent, struct inode *object) +{ + reiser4_block_nr res = 0; + + assert("vpf-321", parent != NULL); + assert("vpf-322", object != NULL); + + /* hashed_add_entry(object) */ + res += inode_dir_plugin(object)->estimate.add_entry(object); + /* reiser4_add_nlink(object) */ + res += inode_file_plugin(object)->estimate.update(object); + /* hashed_add_entry(object) */ + res += inode_dir_plugin(object)->estimate.add_entry(object); + /* reiser4_add_nlink(parent) */ + res += inode_file_plugin(parent)->estimate.update(parent); + + return 0; +} + +/* helper function for reiser4_dir_init_common(). Create "." and ".." */ +static int create_dot_dotdot(struct inode *object/* object to create dot and + * dotdot for */ , + struct inode *parent/* parent of @object */) +{ + int result; + struct dentry dots_entry; + reiser4_dir_entry_desc entry; + + assert("nikita-688", object != NULL); + assert("nikita-689", S_ISDIR(object->i_mode)); + assert("nikita-691", parent != NULL); + + /* We store dot and dotdot as normal directory entries. This is + not necessary, because almost all information stored in them + is already in the stat-data of directory, the only thing + being missed is objectid of grand-parent directory that can + easily be added there as extension. + + But it is done the way it is done, because not storing dot + and dotdot will lead to the following complications: + + . special case handling in ->lookup(). + . addition of another extension to the sd. + . dependency on key allocation policy for stat data. + + */ + + memset(&entry, 0, sizeof entry); + memset(&dots_entry, 0, sizeof dots_entry); + entry.obj = dots_entry.d_inode = object; + dots_entry.d_name.name = "."; + dots_entry.d_name.len = 1; + result = reiser4_add_entry_common(object, &dots_entry, NULL, &entry); + reiser4_free_dentry_fsdata(&dots_entry); + + if (result == 0) { + result = reiser4_add_nlink(object, object, 0); + if (result == 0) { + entry.obj = dots_entry.d_inode = parent; + dots_entry.d_name.name = ".."; + dots_entry.d_name.len = 2; + result = reiser4_add_entry_common(object, + &dots_entry, NULL, &entry); + reiser4_free_dentry_fsdata(&dots_entry); + /* if creation of ".." failed, iput() will delete + object with ".". */ + if (result == 0) { + result = reiser4_add_nlink(parent, object, 0); + if (result != 0) + /* + * if we failed to bump i_nlink, try + * to remove ".." + */ + reiser4_detach_common(object, parent); + } + } + } + + if (result != 0) { + /* + * in the case of error, at least update stat-data so that, + * ->i_nlink updates are not lingering. + */ + reiser4_update_sd(object); + reiser4_update_sd(parent); + } + + return result; +} + +/* + * return 0 iff @coord contains a directory entry for the file with the name + * @name. + */ +static int +check_item(const struct inode *dir, const coord_t *coord, const char *name) +{ + item_plugin *iplug; + char buf[DE_NAME_BUF_LEN]; + + iplug = item_plugin_by_coord(coord); + if (iplug == NULL) { + warning("nikita-1135", "Cannot get item plugin"); + print_coord("coord", coord, 1); + return RETERR(-EIO); + } else if (item_id_by_coord(coord) != + item_id_by_plugin(inode_dir_item_plugin(dir))) { + /* item id of current item does not match to id of items a + directory is built of */ + warning("nikita-1136", "Wrong item plugin"); + print_coord("coord", coord, 1); + return RETERR(-EIO); + } + assert("nikita-1137", iplug->s.dir.extract_name); + + /* Compare name stored in this entry with name we are looking for. + + NOTE-NIKITA Here should go code for support of something like + unicode, code tables, etc. + */ + return !!strcmp(name, iplug->s.dir.extract_name(coord, buf)); +} + +static int +check_entry(const struct inode *dir, coord_t *coord, const struct qstr *name) +{ + return WITH_COORD(coord, check_item(dir, coord, name->name)); +} + +/* + * argument package used by entry_actor to scan entries with identical keys. + */ +struct entry_actor_args { + /* name we are looking for */ + const char *name; + /* key of directory entry. entry_actor() scans through sequence of + * items/units having the same key */ + reiser4_key *key; + /* how many entries with duplicate key was scanned so far. */ + int non_uniq; +#if REISER4_USE_COLLISION_LIMIT + /* scan limit */ + int max_non_uniq; +#endif + /* return parameter: set to true, if ->name wasn't found */ + int not_found; + /* what type of lock to take when moving to the next node during + * scan */ + znode_lock_mode mode; + + /* last coord that was visited during scan */ + coord_t last_coord; + /* last node locked during scan */ + lock_handle last_lh; + /* inode of directory */ + const struct inode *inode; +}; + +/* Function called by reiser4_find_entry() to look for given name + in the directory. */ +static int entry_actor(reiser4_tree * tree UNUSED_ARG /* tree being scanned */ , + coord_t *coord /* current coord */ , + lock_handle * lh /* current lock handle */ , + void *entry_actor_arg/* argument to scan */) +{ + reiser4_key unit_key; + struct entry_actor_args *args; + + assert("nikita-1131", tree != NULL); + assert("nikita-1132", coord != NULL); + assert("nikita-1133", entry_actor_arg != NULL); + + args = entry_actor_arg; + ++args->non_uniq; +#if REISER4_USE_COLLISION_LIMIT + if (args->non_uniq > args->max_non_uniq) { + args->not_found = 1; + /* hash collision overflow. */ + return RETERR(-EBUSY); + } +#endif + + /* + * did we just reach the end of the sequence of items/units with + * identical keys? + */ + if (!keyeq(args->key, unit_key_by_coord(coord, &unit_key))) { + assert("nikita-1791", + keylt(args->key, unit_key_by_coord(coord, &unit_key))); + args->not_found = 1; + args->last_coord.between = AFTER_UNIT; + return 0; + } + + coord_dup(&args->last_coord, coord); + /* + * did scan just moved to the next node? + */ + if (args->last_lh.node != lh->node) { + int lock_result; + + /* + * if so, lock new node with the mode requested by the caller + */ + done_lh(&args->last_lh); + assert("nikita-1896", znode_is_any_locked(lh->node)); + lock_result = longterm_lock_znode(&args->last_lh, lh->node, + args->mode, ZNODE_LOCK_HIPRI); + if (lock_result != 0) + return lock_result; + } + return check_item(args->inode, coord, args->name); +} + +/* Look for given @name within directory @dir. + + This is called during lookup, creation and removal of directory + entries and on reiser4_rename_common + + First calculate key that directory entry for @name would have. Search + for this key in the tree. If such key is found, scan all items with + the same key, checking name in each directory entry along the way. +*/ +int reiser4_find_entry(struct inode *dir, /* directory to scan */ + struct dentry *de, /* name to search for */ + lock_handle * lh, /* resulting lock handle */ + znode_lock_mode mode, /* required lock mode */ + reiser4_dir_entry_desc * entry /* parameters of found + directory entry */) +{ + const struct qstr *name; + seal_t *seal; + coord_t *coord; + int result; + __u32 flags; + struct de_location *dec; + struct reiser4_dentry_fsdata *fsdata; + + assert("nikita-1130", lh != NULL); + assert("nikita-1128", dir != NULL); + + name = &de->d_name; + assert("nikita-1129", name != NULL); + + /* dentry private data don't require lock, because dentry + manipulations are protected by i_mutex on parent. + + This is not so for inodes, because there is no -the- parent in + inode case. + */ + fsdata = reiser4_get_dentry_fsdata(de); + if (IS_ERR(fsdata)) + return PTR_ERR(fsdata); + dec = &fsdata->dec; + + coord = &dec->entry_coord; + coord_clear_iplug(coord); + seal = &dec->entry_seal; + /* compose key of directory entry for @name */ + inode_dir_plugin(dir)->build_entry_key(dir, name, &entry->key); + + if (reiser4_seal_is_set(seal)) { + /* check seal */ + result = reiser4_seal_validate(seal, coord, &entry->key, + lh, mode, ZNODE_LOCK_LOPRI); + if (result == 0) { + /* key was found. Check that it is really item we are + looking for. */ + result = check_entry(dir, coord, name); + if (result == 0) + return 0; + } + } + flags = (mode == ZNODE_WRITE_LOCK) ? CBK_FOR_INSERT : 0; + /* + * find place in the tree where directory item should be located. + */ + result = reiser4_object_lookup(dir, &entry->key, coord, lh, mode, + FIND_EXACT, LEAF_LEVEL, LEAF_LEVEL, + flags, NULL/*ra_info */); + if (result == CBK_COORD_FOUND) { + struct entry_actor_args arg; + + /* fast path: no hash collisions */ + result = check_entry(dir, coord, name); + if (result == 0) { + reiser4_seal_init(seal, coord, &entry->key); + dec->pos = 0; + } else if (result > 0) { + /* Iterate through all units with the same keys. */ + arg.name = name->name; + arg.key = &entry->key; + arg.not_found = 0; + arg.non_uniq = 0; +#if REISER4_USE_COLLISION_LIMIT + arg.max_non_uniq = max_hash_collisions(dir); + assert("nikita-2851", arg.max_non_uniq > 1); +#endif + arg.mode = mode; + arg.inode = dir; + coord_init_zero(&arg.last_coord); + init_lh(&arg.last_lh); + + result = reiser4_iterate_tree + (reiser4_tree_by_inode(dir), + coord, lh, + entry_actor, &arg, mode, 1); + /* if end of the tree or extent was reached during + scanning. */ + if (arg.not_found || (result == -E_NO_NEIGHBOR)) { + /* step back */ + done_lh(lh); + + result = zload(arg.last_coord.node); + if (result == 0) { + coord_clear_iplug(&arg.last_coord); + coord_dup(coord, &arg.last_coord); + move_lh(lh, &arg.last_lh); + result = RETERR(-ENOENT); + zrelse(arg.last_coord.node); + --arg.non_uniq; + } + } + + done_lh(&arg.last_lh); + if (result == 0) + reiser4_seal_init(seal, coord, &entry->key); + + if (result == 0 || result == -ENOENT) { + assert("nikita-2580", arg.non_uniq > 0); + dec->pos = arg.non_uniq - 1; + } + } + } else + dec->pos = -1; + return result; +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format40.c linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format40.c --- linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format40.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format40.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,655 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "../../debug.h" +#include "../../dformat.h" +#include "../../key.h" +#include "../node/node.h" +#include "../space/space_allocator.h" +#include "disk_format40.h" +#include "../plugin.h" +#include "../../txnmgr.h" +#include "../../jnode.h" +#include "../../tree.h" +#include "../../super.h" +#include "../../wander.h" +#include "../../inode.h" +#include "../../ktxnmgrd.h" +#include "../../status_flags.h" + +#include /* for __u?? */ +#include /* for struct super_block */ +#include + +/* reiser 4.0 default disk layout */ + +/* Amount of free blocks needed to perform release_format40 when fs gets + mounted RW: 1 for SB, 1 for non-leaves in overwrite set, 2 for tx header + & tx record. */ +#define RELEASE_RESERVED 4 + +/* The greatest supported format40 version number */ +#define FORMAT40_VERSION PLUGIN_LIBRARY_VERSION + +/* This flag indicates that backup should be updated + (the update is performed by fsck) */ +#define FORMAT40_UPDATE_BACKUP (1 << 31) + +/* functions to access fields of format40_disk_super_block */ +static __u64 get_format40_block_count(const format40_disk_super_block * sb) +{ + return le64_to_cpu(get_unaligned(&sb->block_count)); +} + +static __u64 get_format40_free_blocks(const format40_disk_super_block * sb) +{ + return le64_to_cpu(get_unaligned(&sb->free_blocks)); +} + +static __u64 get_format40_root_block(const format40_disk_super_block * sb) +{ + return le64_to_cpu(get_unaligned(&sb->root_block)); +} + +static __u16 get_format40_tree_height(const format40_disk_super_block * sb) +{ + return le16_to_cpu(get_unaligned(&sb->tree_height)); +} + +static __u64 get_format40_file_count(const format40_disk_super_block * sb) +{ + return le64_to_cpu(get_unaligned(&sb->file_count)); +} + +static __u64 get_format40_oid(const format40_disk_super_block * sb) +{ + return le64_to_cpu(get_unaligned(&sb->oid)); +} + +static __u32 get_format40_mkfs_id(const format40_disk_super_block * sb) +{ + return le32_to_cpu(get_unaligned(&sb->mkfs_id)); +} + +static __u64 get_format40_flags(const format40_disk_super_block * sb) +{ + return le64_to_cpu(get_unaligned(&sb->flags)); +} + +static __u32 get_format40_version(const format40_disk_super_block * sb) +{ + return le32_to_cpu(get_unaligned(&sb->version)) & + ~FORMAT40_UPDATE_BACKUP; +} + +static int update_backup_version(const format40_disk_super_block * sb) +{ + return (le32_to_cpu(get_unaligned(&sb->version)) & + FORMAT40_UPDATE_BACKUP); +} + +static int update_disk_version(const format40_disk_super_block * sb) +{ + return (get_format40_version(sb) < FORMAT40_VERSION); +} + +static int incomplete_compatibility(const format40_disk_super_block * sb) +{ + return (get_format40_version(sb) > FORMAT40_VERSION); +} + +static format40_super_info *get_sb_info(struct super_block *super) +{ + return &get_super_private(super)->u.format40; +} + +static int consult_diskmap(struct super_block *s) +{ + format40_super_info *info; + journal_location *jloc; + + info = get_sb_info(s); + jloc = &get_super_private(s)->jloc; + /* Default format-specific locations, if there is nothing in + * diskmap */ + jloc->footer = FORMAT40_JOURNAL_FOOTER_BLOCKNR; + jloc->header = FORMAT40_JOURNAL_HEADER_BLOCKNR; + info->loc.super = FORMAT40_OFFSET / s->s_blocksize; +#ifdef CONFIG_REISER4_BADBLOCKS + reiser4_get_diskmap_value(FORMAT40_PLUGIN_DISKMAP_ID, FORMAT40_JF, + &jloc->footer); + reiser4_get_diskmap_value(FORMAT40_PLUGIN_DISKMAP_ID, FORMAT40_JH, + &jloc->header); + reiser4_get_diskmap_value(FORMAT40_PLUGIN_DISKMAP_ID, FORMAT40_SUPER, + &info->loc.super); +#endif + return 0; +} + +/* find any valid super block of disk_format40 (even if the first + super block is destroyed), will change block numbers of actual journal header/footer (jf/jh) + if needed */ +static struct buffer_head *find_a_disk_format40_super_block(struct super_block + *s) +{ + struct buffer_head *super_bh; + format40_disk_super_block *disk_sb; + format40_super_info *info; + + assert("umka-487", s != NULL); + + info = get_sb_info(s); + + super_bh = sb_bread(s, info->loc.super); + if (super_bh == NULL) + return ERR_PTR(RETERR(-EIO)); + + disk_sb = (format40_disk_super_block *) super_bh->b_data; + if (strncmp(disk_sb->magic, FORMAT40_MAGIC, sizeof(FORMAT40_MAGIC))) { + brelse(super_bh); + return ERR_PTR(RETERR(-EINVAL)); + } + + reiser4_set_block_count(s, le64_to_cpu(get_unaligned(&disk_sb->block_count))); + reiser4_set_data_blocks(s, le64_to_cpu(get_unaligned(&disk_sb->block_count)) - + le64_to_cpu(get_unaligned(&disk_sb->free_blocks))); + reiser4_set_free_blocks(s, le64_to_cpu(get_unaligned(&disk_sb->free_blocks))); + + return super_bh; +} + +/* find the most recent version of super block. This is called after journal is + replayed */ +static struct buffer_head *read_super_block(struct super_block *s UNUSED_ARG) +{ + /* Here the most recent superblock copy has to be read. However, as + journal replay isn't complete, we are using + find_a_disk_format40_super_block() function. */ + return find_a_disk_format40_super_block(s); +} + +static int get_super_jnode(struct super_block *s) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + jnode *sb_jnode; + int ret; + + sb_jnode = reiser4_alloc_io_head(&get_sb_info(s)->loc.super); + + ret = jload(sb_jnode); + + if (ret) { + reiser4_drop_io_head(sb_jnode); + return ret; + } + + pin_jnode_data(sb_jnode); + jrelse(sb_jnode); + + sbinfo->u.format40.sb_jnode = sb_jnode; + + return 0; +} + +static void done_super_jnode(struct super_block *s) +{ + jnode *sb_jnode = get_super_private(s)->u.format40.sb_jnode; + + if (sb_jnode) { + unpin_jnode_data(sb_jnode); + reiser4_drop_io_head(sb_jnode); + } +} + +typedef enum format40_init_stage { + NONE_DONE = 0, + CONSULT_DISKMAP, + FIND_A_SUPER, + INIT_JOURNAL_INFO, + INIT_STATUS, + JOURNAL_REPLAY, + READ_SUPER, + KEY_CHECK, + INIT_OID, + INIT_TREE, + JOURNAL_RECOVER, + INIT_SA, + INIT_JNODE, + ALL_DONE +} format40_init_stage; + +static format40_disk_super_block *copy_sb(const struct buffer_head *super_bh) +{ + format40_disk_super_block *sb_copy; + + sb_copy = kmalloc(sizeof(format40_disk_super_block), + reiser4_ctx_gfp_mask_get()); + if (sb_copy == NULL) + return ERR_PTR(RETERR(-ENOMEM)); + memcpy(sb_copy, ((format40_disk_super_block *) super_bh->b_data), + sizeof(format40_disk_super_block)); + return sb_copy; +} + +static int check_key_format(const format40_disk_super_block *sb_copy) +{ + if (!equi(REISER4_LARGE_KEY, + get_format40_flags(sb_copy) & (1 << FORMAT40_LARGE_KEYS))) { + warning("nikita-3228", "Key format mismatch. " + "Only %s keys are supported.", + REISER4_LARGE_KEY ? "large" : "small"); + return RETERR(-EINVAL); + } + return 0; +} + +/** + * try_init_format40 + * @super: + * @stage: + * + */ +static int try_init_format40(struct super_block *super, + format40_init_stage *stage) +{ + int result; + struct buffer_head *super_bh; + reiser4_super_info_data *sbinfo; + format40_disk_super_block *sb_copy; + tree_level height; + reiser4_block_nr root_block; + node_plugin *nplug; + + assert("vs-475", super != NULL); + assert("vs-474", get_super_private(super)); + + *stage = NONE_DONE; + + result = consult_diskmap(super); + if (result) + return result; + *stage = CONSULT_DISKMAP; + + super_bh = find_a_disk_format40_super_block(super); + if (IS_ERR(super_bh)) + return PTR_ERR(super_bh); + brelse(super_bh); + *stage = FIND_A_SUPER; + + /* ok, we are sure that filesystem format is a format40 format */ + + /* map jnodes for journal control blocks (header, footer) to disk */ + result = reiser4_init_journal_info(super); + if (result) + return result; + *stage = INIT_JOURNAL_INFO; + + /* ok, we are sure that filesystem format is a format40 format */ + /* Now check it's state */ + result = reiser4_status_init(FORMAT40_STATUS_BLOCKNR); + if (result != 0 && result != -EINVAL) + /* -EINVAL means there is no magic, so probably just old + * fs. */ + return result; + *stage = INIT_STATUS; + + result = reiser4_status_query(NULL, NULL); + if (result == REISER4_STATUS_MOUNT_WARN) + notice("vpf-1363", "Warning: mounting %s with errors.", + super->s_id); + if (result == REISER4_STATUS_MOUNT_RO) + notice("vpf-1364", "Warning: mounting %s with fatal errors," + " forcing read-only mount.", super->s_id); + result = reiser4_journal_replay(super); + if (result) + return result; + *stage = JOURNAL_REPLAY; + + super_bh = read_super_block(super); + if (IS_ERR(super_bh)) + return PTR_ERR(super_bh); + *stage = READ_SUPER; + + /* allocate and make a copy of format40_disk_super_block */ + sb_copy = copy_sb(super_bh); + brelse(super_bh); + + if (IS_ERR(sb_copy)) + return PTR_ERR(sb_copy); + printk("reiser4: %s: found disk format 4.0.%u.\n", + super->s_id, + get_format40_version(sb_copy)); + if (incomplete_compatibility(sb_copy)) + printk("reiser4: Warning: The last completely supported " + "version of disk format40 is %u. Some objects of " + "the semantic tree can be unaccessible.\n", + FORMAT40_VERSION); + /* make sure that key format of kernel and filesystem match */ + result = check_key_format(sb_copy); + if (result) { + kfree(sb_copy); + return result; + } + *stage = KEY_CHECK; + + result = oid_init_allocator(super, get_format40_file_count(sb_copy), + get_format40_oid(sb_copy)); + if (result) { + kfree(sb_copy); + return result; + } + *stage = INIT_OID; + + /* get things necessary to init reiser4_tree */ + root_block = get_format40_root_block(sb_copy); + height = get_format40_tree_height(sb_copy); + nplug = node_plugin_by_id(NODE40_ID); + + /* initialize reiser4_super_info_data */ + sbinfo = get_super_private(super); + assert("", sbinfo->tree.super == super); + /* init reiser4_tree for the filesystem */ + result = reiser4_init_tree(&sbinfo->tree, &root_block, height, nplug); + if (result) { + kfree(sb_copy); + return result; + } + *stage = INIT_TREE; + + /* + * initialize reiser4_super_info_data with data from format40 super + * block + */ + sbinfo->default_uid = 0; + sbinfo->default_gid = 0; + sbinfo->mkfs_id = get_format40_mkfs_id(sb_copy); + /* number of blocks in filesystem and reserved space */ + reiser4_set_block_count(super, get_format40_block_count(sb_copy)); + sbinfo->blocks_free = get_format40_free_blocks(sb_copy); + sbinfo->version = get_format40_version(sb_copy); + kfree(sb_copy); + + if (update_backup_version(sb_copy)) + printk("reiser4: Warning: metadata backup is not updated. " + "Please run 'fsck.reiser4 --fix' on %s.\n", + super->s_id); + + sbinfo->fsuid = 0; + sbinfo->fs_flags |= (1 << REISER4_ADG); /* hard links for directories + * are not supported */ + sbinfo->fs_flags |= (1 << REISER4_ONE_NODE_PLUGIN); /* all nodes in + * layout 40 are + * of one + * plugin */ + /* sbinfo->tmgr is initialized already */ + + /* recover sb data which were logged separately from sb block */ + + /* NOTE-NIKITA: reiser4_journal_recover_sb_data() calls + * oid_init_allocator() and reiser4_set_free_blocks() with new + * data. What's the reason to call them above? */ + result = reiser4_journal_recover_sb_data(super); + if (result != 0) + return result; + *stage = JOURNAL_RECOVER; + + /* + * Set number of used blocks. The number of used blocks is not stored + * neither in on-disk super block nor in the journal footer blocks. At + * this moment actual values of total blocks and free block counters + * are set in the reiser4 super block (in-memory structure) and we can + * calculate number of used blocks from them. + */ + reiser4_set_data_blocks(super, + reiser4_block_count(super) - + reiser4_free_blocks(super)); + +#if REISER4_DEBUG + sbinfo->min_blocks_used = 16 /* reserved area */ + + 2 /* super blocks */ + + 2 /* journal footer and header */ ; +#endif + + /* init disk space allocator */ + result = sa_init_allocator(reiser4_get_space_allocator(super), + super, NULL); + if (result) + return result; + *stage = INIT_SA; + + result = get_super_jnode(super); + if (result == 0) + *stage = ALL_DONE; + return result; +} + +/* plugin->u.format.get_ready */ +int init_format_format40(struct super_block *s, void *data UNUSED_ARG) +{ + int result; + format40_init_stage stage; + + result = try_init_format40(s, &stage); + switch (stage) { + case ALL_DONE: + assert("nikita-3458", result == 0); + break; + case INIT_JNODE: + done_super_jnode(s); + case INIT_SA: + sa_destroy_allocator(reiser4_get_space_allocator(s), s); + case JOURNAL_RECOVER: + case INIT_TREE: + reiser4_done_tree(&get_super_private(s)->tree); + case INIT_OID: + case KEY_CHECK: + case READ_SUPER: + case JOURNAL_REPLAY: + case INIT_STATUS: + reiser4_status_finish(); + case INIT_JOURNAL_INFO: + reiser4_done_journal_info(s); + case FIND_A_SUPER: + case CONSULT_DISKMAP: + case NONE_DONE: + break; + default: + impossible("nikita-3457", "init stage: %i", stage); + } + + if (!rofs_super(s) && reiser4_free_blocks(s) < RELEASE_RESERVED) + return RETERR(-ENOSPC); + + return result; +} + +static void pack_format40_super(const struct super_block *s, char *data) +{ + format40_disk_super_block *super_data = + (format40_disk_super_block *) data; + + reiser4_super_info_data *sbinfo = get_super_private(s); + + assert("zam-591", data != NULL); + + put_unaligned(cpu_to_le64(reiser4_free_committed_blocks(s)), + &super_data->free_blocks); + + put_unaligned(cpu_to_le64(sbinfo->tree.root_block), + &super_data->root_block); + + put_unaligned(cpu_to_le64(oid_next(s)), + &super_data->oid); + + put_unaligned(cpu_to_le64(oids_used(s)), + &super_data->file_count); + + put_unaligned(cpu_to_le16(sbinfo->tree.height), + &super_data->tree_height); + + if (update_disk_version(super_data)) { + __u32 version = FORMAT40_VERSION | FORMAT40_UPDATE_BACKUP; + + put_unaligned(cpu_to_le32(version), &super_data->version); + } +} + +/* plugin->u.format.log_super + return a jnode which should be added to transaction when the super block + gets logged */ +jnode *log_super_format40(struct super_block *s) +{ + jnode *sb_jnode; + + sb_jnode = get_super_private(s)->u.format40.sb_jnode; + + jload(sb_jnode); + + pack_format40_super(s, jdata(sb_jnode)); + + jrelse(sb_jnode); + + return sb_jnode; +} + +/* plugin->u.format.release */ +int release_format40(struct super_block *s) +{ + int ret; + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(s); + assert("zam-579", sbinfo != NULL); + + if (!rofs_super(s)) { + ret = reiser4_capture_super_block(s); + if (ret != 0) + warning("vs-898", + "reiser4_capture_super_block failed: %d", + ret); + + ret = txnmgr_force_commit_all(s, 1); + if (ret != 0) + warning("jmacd-74438", "txn_force failed: %d", ret); + + all_grabbed2free(); + } + + sa_destroy_allocator(&sbinfo->space_allocator, s); + reiser4_done_journal_info(s); + done_super_jnode(s); + + rcu_barrier(); + reiser4_done_tree(&sbinfo->tree); + /* call finish_rcu(), because some znode were "released" in + * reiser4_done_tree(). */ + rcu_barrier(); + + return 0; +} + +#define FORMAT40_ROOT_LOCALITY 41 +#define FORMAT40_ROOT_OBJECTID 42 + +/* plugin->u.format.root_dir_key */ +const reiser4_key *root_dir_key_format40(const struct super_block *super + UNUSED_ARG) +{ + static const reiser4_key FORMAT40_ROOT_DIR_KEY = { + .el = { + __constant_cpu_to_le64((FORMAT40_ROOT_LOCALITY << 4) | KEY_SD_MINOR), +#if REISER4_LARGE_KEY + ON_LARGE_KEY(0ull,) +#endif + __constant_cpu_to_le64(FORMAT40_ROOT_OBJECTID), + 0ull + } + }; + + return &FORMAT40_ROOT_DIR_KEY; +} + +/* plugin->u.format.check_open. + Check the opened object for validness. For now it checks for the valid oid & + locality only, can be improved later and it its work may depend on the mount + options. */ +int check_open_format40(const struct inode *object) +{ + oid_t max, oid; + + max = oid_next(object->i_sb) - 1; + + /* Check the oid. */ + oid = get_inode_oid(object); + if (oid > max) { + warning("vpf-1360", "The object with the oid %llu " + "greater then the max used oid %llu found.", + (unsigned long long)oid, (unsigned long long)max); + + return RETERR(-EIO); + } + + /* Check the locality. */ + oid = reiser4_inode_data(object)->locality_id; + if (oid > max) { + warning("vpf-1361", "The object with the locality %llu " + "greater then the max used oid %llu found.", + (unsigned long long)oid, (unsigned long long)max); + + return RETERR(-EIO); + } + + return 0; +} + +/* plugin->u.format.version_update. + Perform all version update operations from the on-disk + format40_disk_super_block.version on disk to FORMAT40_VERSION. + */ +int version_update_format40(struct super_block *super) { + txn_handle * trans; + lock_handle lh; + txn_atom *atom; + int ret; + + /* Nothing to do if RO mount or the on-disk version is not less. */ + if (super->s_flags & MS_RDONLY) + return 0; + + if (get_super_private(super)->version >= FORMAT40_VERSION) + return 0; + + printk("reiser4: Updating disk format to 4.0.%u. The reiser4 metadata " + "backup is left unchanged. Please run 'fsck.reiser4 --fix' " + "on %s to update it too.\n", FORMAT40_VERSION, super->s_id); + + /* Mark the uber znode dirty to call log_super on write_logs. */ + init_lh(&lh); + ret = get_uber_znode(reiser4_get_tree(super), ZNODE_WRITE_LOCK, + ZNODE_LOCK_HIPRI, &lh); + if (ret != 0) + return ret; + + znode_make_dirty(lh.node); + done_lh(&lh); + + /* Update the backup blocks. */ + + /* Force write_logs immediately. */ + trans = get_current_context()->trans; + atom = get_current_atom_locked(); + assert("vpf-1906", atom != NULL); + + spin_lock_txnh(trans); + return force_commit_atom(trans); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format40.h linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format40.h --- linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format40.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format40.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,109 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* this file contains: + - definition of ondisk super block of standart disk layout for + reiser 4.0 (layout 40) + - definition of layout 40 specific portion of in-core super block + - declarations of functions implementing methods of layout plugin + for layout 40 + - declarations of functions used to get/set fields in layout 40 super block +*/ + +#ifndef __DISK_FORMAT40_H__ +#define __DISK_FORMAT40_H__ + +/* magic for default reiser4 layout */ +#define FORMAT40_MAGIC "ReIsEr40FoRmAt" +#define FORMAT40_OFFSET (REISER4_MASTER_OFFSET + PAGE_CACHE_SIZE) + +#include "../../dformat.h" + +#include /* for struct super_block */ + +typedef enum { + FORMAT40_LARGE_KEYS +} format40_flags; + +/* ondisk super block for format 40. It is 512 bytes long */ +typedef struct format40_disk_super_block { + /* 0 */ d64 block_count; + /* number of block in a filesystem */ + /* 8 */ d64 free_blocks; + /* number of free blocks */ + /* 16 */ d64 root_block; + /* filesystem tree root block */ + /* 24 */ d64 oid; + /* smallest free objectid */ + /* 32 */ d64 file_count; + /* number of files in a filesystem */ + /* 40 */ d64 flushes; + /* number of times super block was + flushed. Needed if format 40 + will have few super blocks */ + /* 48 */ d32 mkfs_id; + /* unique identifier of fs */ + /* 52 */ char magic[16]; + /* magic string ReIsEr40FoRmAt */ + /* 68 */ d16 tree_height; + /* height of filesystem tree */ + /* 70 */ d16 formatting_policy; + /* not used anymore */ + /* 72 */ d64 flags; + /* 80 */ d32 version; + /* on-disk format version number + initially assigned by mkfs as the greatest format40 + version number supported by reiser4progs and updated + in mount time in accordance with the greatest format40 + version number supported by kernel. + Is used by fsck to catch possible corruption and + for various compatibility issues */ + /* 84 */ char not_used[428]; +} format40_disk_super_block; + +/* format 40 specific part of reiser4_super_info_data */ +typedef struct format40_super_info { +/* format40_disk_super_block actual_sb; */ + jnode *sb_jnode; + struct { + reiser4_block_nr super; + } loc; +} format40_super_info; + +/* Defines for journal header and footer respectively. */ +#define FORMAT40_JOURNAL_HEADER_BLOCKNR \ + ((REISER4_MASTER_OFFSET / PAGE_CACHE_SIZE) + 3) + +#define FORMAT40_JOURNAL_FOOTER_BLOCKNR \ + ((REISER4_MASTER_OFFSET / PAGE_CACHE_SIZE) + 4) + +#define FORMAT40_STATUS_BLOCKNR \ + ((REISER4_MASTER_OFFSET / PAGE_CACHE_SIZE) + 5) + +/* Diskmap declarations */ +#define FORMAT40_PLUGIN_DISKMAP_ID ((REISER4_FORMAT_PLUGIN_TYPE<<16) | (FORMAT40_ID)) +#define FORMAT40_SUPER 1 +#define FORMAT40_JH 2 +#define FORMAT40_JF 3 + +/* declarations of functions implementing methods of layout plugin for + format 40. The functions theirself are in disk_format40.c */ +extern int init_format_format40(struct super_block *, void *data); +extern const reiser4_key *root_dir_key_format40(const struct super_block *); +extern int release_format40(struct super_block *s); +extern jnode *log_super_format40(struct super_block *s); +extern int check_open_format40(const struct inode *object); +extern int version_update_format40(struct super_block *super); + +/* __DISK_FORMAT40_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format.c linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format.c --- linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,38 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "../../debug.h" +#include "../plugin_header.h" +#include "disk_format40.h" +#include "disk_format.h" +#include "../plugin.h" + +/* initialization of disk layout plugins */ +disk_format_plugin format_plugins[LAST_FORMAT_ID] = { + [FORMAT40_ID] = { + .h = { + .type_id = REISER4_FORMAT_PLUGIN_TYPE, + .id = FORMAT40_ID, + .pops = NULL, + .label = "reiser40", + .desc = "standard disk layout for reiser40", + .linkage = {NULL, NULL} + }, + .init_format = init_format_format40, + .root_dir_key = root_dir_key_format40, + .release = release_format40, + .log_super = log_super_format40, + .check_open = check_open_format40, + .version_update = version_update_format40 + } +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format.h linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format.h --- linux-2.6.30.orig/fs/reiser4/plugin/disk_format/disk_format.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/disk_format/disk_format.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,27 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* identifiers for disk layouts, they are also used as indexes in array of disk + plugins */ + +#if !defined( __REISER4_DISK_FORMAT_H__ ) +#define __REISER4_DISK_FORMAT_H__ + +typedef enum { + /* standard reiser4 disk layout plugin id */ + FORMAT40_ID, + LAST_FORMAT_ID +} disk_format_id; + +/* __REISER4_DISK_FORMAT_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/disk_format/Makefile linux-2.6.30/fs/reiser4/plugin/disk_format/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/disk_format/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/disk_format/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,5 @@ +obj-$(CONFIG_REISER4_FS) += df_plugins.o + +df_plugins-objs := \ + disk_format40.o \ + disk_format.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/fibration.c linux-2.6.30/fs/reiser4/plugin/fibration.c --- linux-2.6.30.orig/fs/reiser4/plugin/fibration.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/fibration.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,175 @@ +/* Copyright 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Directory fibrations */ + +/* + * Suppose we have a directory tree with sources of some project. During + * compilation .o files are created within this tree. This makes access + * to the original source files less efficient, because source files are + * now "diluted" by object files: default directory plugin uses prefix + * of a file name as a part of the key for directory entry (and this + * part is also inherited by the key of file body). This means that + * foo.o will be located close to foo.c and foo.h in the tree. + * + * To avoid this effect directory plugin fill highest 7 (unused + * originally) bits of the second component of the directory entry key + * by bit-pattern depending on the file name (see + * fs/reiser4/kassign.c:build_entry_key_common()). These bits are called + * "fibre". Fibre of the file name key is inherited by key of stat data + * and keys of file body (in the case of REISER4_LARGE_KEY). + * + * Fibre for a given file is chosen by per-directory fibration + * plugin. Names within given fibre are ordered lexicographically. + */ + +#include "../debug.h" +#include "plugin_header.h" +#include "plugin.h" +#include "../super.h" +#include "../inode.h" + +#include + +static const int fibre_shift = 57; + +#define FIBRE_NO(n) (((__u64)(n)) << fibre_shift) + +/* + * Trivial fibration: all files of directory are just ordered + * lexicographically. + */ +static __u64 fibre_trivial(const struct inode *dir, const char *name, int len) +{ + return FIBRE_NO(0); +} + +/* + * dot-o fibration: place .o files after all others. + */ +static __u64 fibre_dot_o(const struct inode *dir, const char *name, int len) +{ + /* special treatment for .*\.o */ + if (len > 2 && name[len - 1] == 'o' && name[len - 2] == '.') + return FIBRE_NO(1); + else + return FIBRE_NO(0); +} + +/* + * ext.1 fibration: subdivide directory into 128 fibrations one for each + * 7bit extension character (file "foo.h" goes into fibre "h"), plus + * default fibre for the rest. + */ +static __u64 fibre_ext_1(const struct inode *dir, const char *name, int len) +{ + if (len > 2 && name[len - 2] == '.') + return FIBRE_NO(name[len - 1]); + else + return FIBRE_NO(0); +} + +/* + * ext.3 fibration: try to separate files with different 3-character + * extensions from each other. + */ +static __u64 fibre_ext_3(const struct inode *dir, const char *name, int len) +{ + if (len > 4 && name[len - 4] == '.') + return FIBRE_NO(name[len - 3] + name[len - 2] + name[len - 1]); + else + return FIBRE_NO(0); +} + +static int change_fibration(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + int result; + + assert("nikita-3503", inode != NULL); + assert("nikita-3504", plugin != NULL); + + assert("nikita-3505", is_reiser4_inode(inode)); + assert("nikita-3506", inode_dir_plugin(inode) != NULL); + assert("nikita-3507", + plugin->h.type_id == REISER4_FIBRATION_PLUGIN_TYPE); + + result = 0; + if (inode_fibration_plugin(inode) == NULL || + inode_fibration_plugin(inode)->h.id != plugin->h.id) { + if (is_dir_empty(inode) == 0) + result = aset_set_unsafe(&reiser4_inode_data(inode)->pset, + PSET_FIBRATION, plugin); + else + result = RETERR(-ENOTEMPTY); + + } + return result; +} + +static reiser4_plugin_ops fibration_plugin_ops = { + .init = NULL, + .load = NULL, + .save_len = NULL, + .save = NULL, + .change = change_fibration +}; + +/* fibration plugins */ +fibration_plugin fibration_plugins[LAST_FIBRATION_ID] = { + [FIBRATION_LEXICOGRAPHIC] = { + .h = { + .type_id = REISER4_FIBRATION_PLUGIN_TYPE, + .id = FIBRATION_LEXICOGRAPHIC, + .pops = &fibration_plugin_ops, + .label = "lexicographic", + .desc = "no fibration", + .linkage = {NULL, NULL} + }, + .fibre = fibre_trivial + }, + [FIBRATION_DOT_O] = { + .h = { + .type_id = REISER4_FIBRATION_PLUGIN_TYPE, + .id = FIBRATION_DOT_O, + .pops = &fibration_plugin_ops, + .label = "dot-o", + .desc = "fibrate .o files separately", + .linkage = {NULL, NULL} + }, + .fibre = fibre_dot_o + }, + [FIBRATION_EXT_1] = { + .h = { + .type_id = REISER4_FIBRATION_PLUGIN_TYPE, + .id = FIBRATION_EXT_1, + .pops = &fibration_plugin_ops, + .label = "ext-1", + .desc = "fibrate file by single character extension", + .linkage = {NULL, NULL} + }, + .fibre = fibre_ext_1 + }, + [FIBRATION_EXT_3] = { + .h = { + .type_id = REISER4_FIBRATION_PLUGIN_TYPE, + .id = FIBRATION_EXT_3, + .pops = &fibration_plugin_ops, + .label = "ext-3", + .desc = "fibrate file by three character extension", + .linkage = {NULL, NULL} + }, + .fibre = fibre_ext_3 + } +}; + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/fibration.h linux-2.6.30/fs/reiser4/plugin/fibration.h --- linux-2.6.30.orig/fs/reiser4/plugin/fibration.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/fibration.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,37 @@ +/* Copyright 2004 by Hans Reiser, licensing governed by reiser4/README */ + +/* Fibration plugin used by hashed directory plugin to segment content + * of directory. See fs/reiser4/plugin/fibration.c for more on this. */ + +#if !defined(__FS_REISER4_PLUGIN_FIBRATION_H__) +#define __FS_REISER4_PLUGIN_FIBRATION_H__ + +#include "plugin_header.h" + +typedef struct fibration_plugin { + /* generic fields */ + plugin_header h; + + __u64(*fibre) (const struct inode *dir, const char *name, int len); +} fibration_plugin; + +typedef enum { + FIBRATION_LEXICOGRAPHIC, + FIBRATION_DOT_O, + FIBRATION_EXT_1, + FIBRATION_EXT_3, + LAST_FIBRATION_ID +} reiser4_fibration_id; + +/* __FS_REISER4_PLUGIN_FIBRATION_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/cryptcompress.c linux-2.6.30/fs/reiser4/plugin/file/cryptcompress.c --- linux-2.6.30.orig/fs/reiser4/plugin/file/cryptcompress.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/cryptcompress.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,3807 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + reiser4/README */ +/* + * Written by Edward Shishkin. + * + * Implementations of inode/file/address_space operations + * specific for cryptcompress file plugin which manages + * regular files built of compressed and(or) encrypted bodies. + * See http://dev.namesys.com/CryptcompressPlugin for details. + */ + +#include "../../inode.h" +#include "../cluster.h" +#include "../object.h" +#include "../../tree_walk.h" +#include "cryptcompress.h" + +#include +#include +#include +#include +#include +#include + +/* + Managing primary and secondary caches by Reiser4 + cryptcompress file plugin. Synchronization scheme. + + + +------------------+ + +------------------->| tfm stream | + | | (compressed data)| + flush | +------------------+ + +-----------------+ | + |(->)longterm lock| V +--+ writepages() | | +-***-+ reiser4 +---+ + | | +--+ | *** | storage tree | | + | | | +-***-+ (primary cache)| | +u | write() (secondary| cache) V / | \ | | +s | ----> +----+ +----+ +----+ +----+ +-***** ******* **----+ ----> | d | +e | | | |page cluster | | | **disk cluster** | | i | +r | <---- +----+ +----+ +----+ +----+ +-***** **********----+ <---- | s | + | read() ^ ^ | | k | + | | (->)longterm lock| | page_io()| | + | | +------+ | | +--+ readpages() | | +---+ + | V + | +------------------+ + +--------------------| tfm stream | + | (plain text) | + +------------------+ +*/ + +/* get cryptcompress specific portion of inode */ +struct cryptcompress_info *cryptcompress_inode_data(const struct inode *inode) +{ + return &reiser4_inode_data(inode)->file_plugin_data.cryptcompress_info; +} + +/* plugin->u.file.init_inode_data */ +void init_inode_data_cryptcompress(struct inode *inode, + reiser4_object_create_data * crd, + int create) +{ + struct cryptcompress_info *data; + + data = cryptcompress_inode_data(inode); + assert("edward-685", data != NULL); + + memset(data, 0, sizeof(*data)); + + mutex_init(&data->checkin_mutex); + data->trunc_index = ULONG_MAX; + turn_on_compression(data); + set_lattice_factor(data, MIN_LATTICE_FACTOR); + init_inode_ordering(inode, crd, create); +} + +/* The following is a part of reiser4 cipher key manager + which is called when opening/creating a cryptcompress file */ + +/* get/set cipher key info */ +struct reiser4_crypto_info * inode_crypto_info (struct inode * inode) +{ + assert("edward-90", inode != NULL); + assert("edward-91", reiser4_inode_data(inode) != NULL); + return cryptcompress_inode_data(inode)->crypt; +} + +static void set_inode_crypto_info (struct inode * inode, + struct reiser4_crypto_info * info) +{ + cryptcompress_inode_data(inode)->crypt = info; +} + +/* allocate a cipher key info */ +struct reiser4_crypto_info * reiser4_alloc_crypto_info (struct inode * inode) +{ + struct reiser4_crypto_info *info; + int fipsize; + + info = kzalloc(sizeof(*info), reiser4_ctx_gfp_mask_get()); + if (!info) + return ERR_PTR(-ENOMEM); + + fipsize = inode_digest_plugin(inode)->fipsize; + info->keyid = kmalloc(fipsize, reiser4_ctx_gfp_mask_get()); + if (!info->keyid) { + kfree(info); + return ERR_PTR(-ENOMEM); + } + info->host = inode; + return info; +} + +#if 0 +/* allocate/free low-level info for cipher and digest + transforms */ +static int alloc_crypto_tfms(struct reiser4_crypto_info * info) +{ + struct crypto_blkcipher * ctfm = NULL; + struct crypto_hash * dtfm = NULL; + cipher_plugin * cplug = inode_cipher_plugin(info->host); + digest_plugin * dplug = inode_digest_plugin(info->host); + + if (cplug->alloc) { + ctfm = cplug->alloc(); + if (IS_ERR(ctfm)) { + warning("edward-1364", + "Can not allocate info for %s\n", + cplug->h.desc); + return RETERR(PTR_ERR(ctfm)); + } + } + info_set_cipher(info, ctfm); + if (dplug->alloc) { + dtfm = dplug->alloc(); + if (IS_ERR(dtfm)) { + warning("edward-1365", + "Can not allocate info for %s\n", + dplug->h.desc); + goto unhappy_with_digest; + } + } + info_set_digest(info, dtfm); + return 0; + unhappy_with_digest: + if (cplug->free) { + cplug->free(ctfm); + info_set_cipher(info, NULL); + } + return RETERR(PTR_ERR(dtfm)); +} +#endif + +static void +free_crypto_tfms(struct reiser4_crypto_info * info) +{ + assert("edward-1366", info != NULL); + if (!info_get_cipher(info)) { + assert("edward-1601", !info_get_digest(info)); + return; + } + inode_cipher_plugin(info->host)->free(info_get_cipher(info)); + info_set_cipher(info, NULL); + inode_digest_plugin(info->host)->free(info_get_digest(info)); + info_set_digest(info, NULL); + return; +} + +#if 0 +/* create a key fingerprint for disk stat-data */ +static int create_keyid (struct reiser4_crypto_info * info, + struct reiser4_crypto_data * data) +{ + int ret = -ENOMEM; + size_t blk, pad; + __u8 * dmem; + __u8 * cmem; + struct hash_desc ddesc; + struct blkcipher_desc cdesc; + struct scatterlist sg; + + assert("edward-1367", info != NULL); + assert("edward-1368", info->keyid != NULL); + + ddesc.tfm = info_get_digest(info); + ddesc.flags = 0; + cdesc.tfm = info_get_cipher(info); + cdesc.flags = 0; + + dmem = kmalloc((size_t)crypto_hash_digestsize(ddesc.tfm), + reiser4_ctx_gfp_mask_get()); + if (!dmem) + goto exit1; + + blk = crypto_blkcipher_blocksize(cdesc.tfm); + + pad = data->keyid_size % blk; + pad = (pad ? blk - pad : 0); + + cmem = kmalloc((size_t)data->keyid_size + pad, + reiser4_ctx_gfp_mask_get()); + if (!cmem) + goto exit2; + memcpy(cmem, data->keyid, data->keyid_size); + memset(cmem + data->keyid_size, 0, pad); + + sg_init_one(&sg, cmem, data->keyid_size + pad); + + ret = crypto_blkcipher_encrypt(&cdesc, &sg, &sg, + data->keyid_size + pad); + if (ret) { + warning("edward-1369", + "encryption failed flags=%x\n", cdesc.flags); + goto exit3; + } + ret = crypto_hash_digest(&ddesc, &sg, sg.length, dmem); + if (ret) { + warning("edward-1602", + "digest failed flags=%x\n", ddesc.flags); + goto exit3; + } + memcpy(info->keyid, dmem, inode_digest_plugin(info->host)->fipsize); + exit3: + kfree(cmem); + exit2: + kfree(dmem); + exit1: + return ret; +} +#endif + +static void destroy_keyid(struct reiser4_crypto_info * info) +{ + assert("edward-1370", info != NULL); + assert("edward-1371", info->keyid != NULL); + kfree(info->keyid); + return; +} + +static void __free_crypto_info (struct inode * inode) +{ + struct reiser4_crypto_info * info = inode_crypto_info(inode); + assert("edward-1372", info != NULL); + + free_crypto_tfms(info); + destroy_keyid(info); + kfree(info); +} + +#if 0 +static void instantiate_crypto_info(struct reiser4_crypto_info * info) +{ + assert("edward-1373", info != NULL); + assert("edward-1374", info->inst == 0); + info->inst = 1; +} +#endif + +static void uninstantiate_crypto_info(struct reiser4_crypto_info * info) +{ + assert("edward-1375", info != NULL); + info->inst = 0; +} + +#if 0 +static int is_crypto_info_instantiated(struct reiser4_crypto_info * info) +{ + return info->inst; +} + +static int inode_has_cipher_key(struct inode * inode) +{ + assert("edward-1376", inode != NULL); + return inode_crypto_info(inode) && + is_crypto_info_instantiated(inode_crypto_info(inode)); +} +#endif + +static void free_crypto_info (struct inode * inode) +{ + uninstantiate_crypto_info(inode_crypto_info(inode)); + __free_crypto_info(inode); +} + +static int need_cipher(struct inode * inode) +{ + return inode_cipher_plugin(inode) != + cipher_plugin_by_id(NONE_CIPHER_ID); +} + +/* Parse @data which contains a (uninstantiated) cipher key imported + from user space, create a low-level cipher info and attach it to + the @object. If success, then info contains an instantiated key */ +#if 0 +struct reiser4_crypto_info * create_crypto_info(struct inode * object, + struct reiser4_crypto_data * data) +{ + int ret; + struct reiser4_crypto_info * info; + + assert("edward-1377", data != NULL); + assert("edward-1378", need_cipher(object)); + + if (inode_file_plugin(object) != + file_plugin_by_id(DIRECTORY_FILE_PLUGIN_ID)) + return ERR_PTR(-EINVAL); + + info = reiser4_alloc_crypto_info(object); + if (IS_ERR(info)) + return info; + ret = alloc_crypto_tfms(info); + if (ret) + goto err; + /* instantiating a key */ + ret = crypto_blkcipher_setkey(info_get_cipher(info), + data->key, + data->keysize); + if (ret) { + warning("edward-1379", + "setkey failed flags=%x", + crypto_blkcipher_get_flags(info_get_cipher(info))); + goto err; + } + info->keysize = data->keysize; + ret = create_keyid(info, data); + if (ret) + goto err; + instantiate_crypto_info(info); + return info; + err: + __free_crypto_info(object); + return ERR_PTR(ret); +} +#endif + +/* increment/decrement a load counter when + attaching/detaching the crypto-stat to any object */ +static void load_crypto_info(struct reiser4_crypto_info * info) +{ + assert("edward-1380", info != NULL); + inc_keyload_count(info); +} + +static void unload_crypto_info(struct inode * inode) +{ + struct reiser4_crypto_info * info = inode_crypto_info(inode); + assert("edward-1381", info->keyload_count > 0); + + dec_keyload_count(inode_crypto_info(inode)); + if (info->keyload_count == 0) + /* final release */ + free_crypto_info(inode); +} + +/* attach/detach an existing crypto-stat */ +void reiser4_attach_crypto_info(struct inode * inode, + struct reiser4_crypto_info * info) +{ + assert("edward-1382", inode != NULL); + assert("edward-1383", info != NULL); + assert("edward-1384", inode_crypto_info(inode) == NULL); + + set_inode_crypto_info(inode, info); + load_crypto_info(info); +} + +/* returns true, if crypto stat can be attached to the @host */ +#if REISER4_DEBUG +static int host_allows_crypto_info(struct inode * host) +{ + int ret; + file_plugin * fplug = inode_file_plugin(host); + + switch (fplug->h.id) { + case CRYPTCOMPRESS_FILE_PLUGIN_ID: + ret = 1; + break; + default: + ret = 0; + } + return ret; +} +#endif /* REISER4_DEBUG */ + +static void reiser4_detach_crypto_info(struct inode * inode) +{ + assert("edward-1385", inode != NULL); + assert("edward-1386", host_allows_crypto_info(inode)); + + if (inode_crypto_info(inode)) + unload_crypto_info(inode); + set_inode_crypto_info(inode, NULL); +} + +#if 0 + +/* compare fingerprints of @child and @parent */ +static int keyid_eq(struct reiser4_crypto_info * child, + struct reiser4_crypto_info * parent) +{ + return !memcmp(child->keyid, + parent->keyid, + info_digest_plugin(parent)->fipsize); +} + +/* check if a crypto-stat (which is bound to @parent) can be inherited */ +int can_inherit_crypto_cryptcompress(struct inode *child, struct inode *parent) +{ + if (!need_cipher(child)) + return 0; + /* the child is created */ + if (!inode_crypto_info(child)) + return 1; + /* the child is looked up */ + if (!inode_crypto_info(parent)) + return 0; + return (inode_cipher_plugin(child) == inode_cipher_plugin(parent) && + inode_digest_plugin(child) == inode_digest_plugin(parent) && + inode_crypto_info(child)->keysize == + inode_crypto_info(parent)->keysize && + keyid_eq(inode_crypto_info(child), inode_crypto_info(parent))); +} +#endif + +/* helper functions for ->create() method of the cryptcompress plugin */ +static int inode_set_crypto(struct inode * object) +{ + reiser4_inode * info; + if (!inode_crypto_info(object)) { + if (need_cipher(object)) + return RETERR(-EINVAL); + /* the file is not to be encrypted */ + return 0; + } + info = reiser4_inode_data(object); + info->extmask |= (1 << CRYPTO_STAT); + return 0; +} + +static int inode_init_compression(struct inode * object) +{ + int result = 0; + assert("edward-1461", object != NULL); + if (inode_compression_plugin(object)->init) + result = inode_compression_plugin(object)->init(); + return result; +} + +static int inode_check_cluster(struct inode * object) +{ + assert("edward-696", object != NULL); + + if (unlikely(inode_cluster_size(object) < PAGE_CACHE_SIZE)) { + warning("edward-1320", "Can not support '%s' " + "logical clusters (less then page size)", + inode_cluster_plugin(object)->h.label); + return RETERR(-EINVAL); + } + if (unlikely(inode_cluster_shift(object)) >= BITS_PER_BYTE*sizeof(int)){ + warning("edward-1463", "Can not support '%s' " + "logical clusters (too big for transform)", + inode_cluster_plugin(object)->h.label); + return RETERR(-EINVAL); + } + return 0; +} + +/* plugin->destroy_inode() */ +void destroy_inode_cryptcompress(struct inode * inode) +{ + assert("edward-1464", INODE_PGCOUNT(inode) == 0); + reiser4_detach_crypto_info(inode); + return; +} + +/* plugin->create_object(): +. install plugins +. attach crypto info if specified +. attach compression info if specified +. attach cluster info +*/ +int create_object_cryptcompress(struct inode *object, struct inode *parent, + reiser4_object_create_data * data) +{ + int result; + reiser4_inode *info; + + assert("edward-23", object != NULL); + assert("edward-24", parent != NULL); + assert("edward-30", data != NULL); + assert("edward-26", reiser4_inode_get_flag(object, REISER4_NO_SD)); + assert("edward-27", data->id == CRYPTCOMPRESS_FILE_PLUGIN_ID); + + info = reiser4_inode_data(object); + + assert("edward-29", info != NULL); + + /* set file bit */ + info->plugin_mask |= (1 << PSET_FILE); + + /* set crypto */ + result = inode_set_crypto(object); + if (result) + goto error; + /* set compression */ + result = inode_init_compression(object); + if (result) + goto error; + /* set cluster */ + result = inode_check_cluster(object); + if (result) + goto error; + + /* save everything in disk stat-data */ + result = write_sd_by_inode_common(object); + if (!result) + return 0; + error: + reiser4_detach_crypto_info(object); + return result; +} + +/* plugin->open() */ +int open_cryptcompress(struct inode * inode, struct file * file) +{ + return 0; +} + +/* returns a blocksize, the attribute of a cipher algorithm */ +static unsigned int +cipher_blocksize(struct inode * inode) +{ + assert("edward-758", need_cipher(inode)); + assert("edward-1400", inode_crypto_info(inode) != NULL); + return crypto_blkcipher_blocksize + (info_get_cipher(inode_crypto_info(inode))); +} + +/* returns offset translated by scale factor of the crypto-algorithm */ +static loff_t inode_scaled_offset (struct inode * inode, + const loff_t src_off /* input offset */) +{ + assert("edward-97", inode != NULL); + + if (!need_cipher(inode) || + src_off == get_key_offset(reiser4_min_key()) || + src_off == get_key_offset(reiser4_max_key())) + return src_off; + + return inode_cipher_plugin(inode)->scale(inode, + cipher_blocksize(inode), + src_off); +} + +/* returns disk cluster size */ +size_t inode_scaled_cluster_size(struct inode * inode) +{ + assert("edward-110", inode != NULL); + + return inode_scaled_offset(inode, inode_cluster_size(inode)); +} + +/* set number of cluster pages */ +static void set_cluster_nrpages(struct cluster_handle * clust, + struct inode *inode) +{ + struct reiser4_slide * win; + + assert("edward-180", clust != NULL); + assert("edward-1040", inode != NULL); + + clust->old_nrpages = size_in_pages(lbytes(clust->index, inode)); + win = clust->win; + if (!win) { + clust->nr_pages = size_in_pages(lbytes(clust->index, inode)); + return; + } + assert("edward-1176", clust->op != LC_INVAL); + assert("edward-1064", win->off + win->count + win->delta != 0); + + if (win->stat == HOLE_WINDOW && + win->off == 0 && win->count == inode_cluster_size(inode)) { + /* special case: writing a "fake" logical cluster */ + clust->nr_pages = 0; + return; + } + clust->nr_pages = size_in_pages(max(win->off + win->count + win->delta, + lbytes(clust->index, inode))); + return; +} + +/* plugin->key_by_inode() + build key of a disk cluster */ +int key_by_inode_cryptcompress(struct inode *inode, loff_t off, + reiser4_key * key) +{ + assert("edward-64", inode != 0); + + if (likely(off != get_key_offset(reiser4_max_key()))) + off = off_to_clust_to_off(off, inode); + if (inode_crypto_info(inode)) + off = inode_scaled_offset(inode, off); + + key_by_inode_and_offset_common(inode, 0, key); + set_key_offset(key, (__u64)off); + return 0; +} + +/* plugin->flow_by_inode() */ +/* flow is used to read/write disk clusters */ +int flow_by_inode_cryptcompress(struct inode *inode, const char __user * buf, + int user, /* 1: @buf is of user space, + 0: kernel space */ + loff_t size, /* @buf size */ + loff_t off, /* offset to start io from */ + rw_op op, /* READ or WRITE */ + flow_t * f /* resulting flow */) +{ + assert("edward-436", f != NULL); + assert("edward-149", inode != NULL); + assert("edward-150", inode_file_plugin(inode) != NULL); + assert("edward-1465", user == 0); /* we use flow to read/write + disk clusters located in + kernel space */ + f->length = size; + memcpy(&f->data, &buf, sizeof(buf)); + f->user = user; + f->op = op; + + return key_by_inode_cryptcompress(inode, off, &f->key); +} + +static int +cryptcompress_hint_validate(hint_t * hint, const reiser4_key * key, + znode_lock_mode lock_mode) +{ + coord_t *coord; + + assert("edward-704", hint != NULL); + assert("edward-1089", !hint_is_valid(hint)); + assert("edward-706", hint->lh.owner == NULL); + + coord = &hint->ext_coord.coord; + + if (!hint || !hint_is_set(hint) || hint->mode != lock_mode) + /* hint either not set or set by different operation */ + return RETERR(-E_REPEAT); + + if (get_key_offset(key) != hint->offset) + /* hint is set for different key */ + return RETERR(-E_REPEAT); + + assert("edward-707", reiser4_schedulable()); + + return reiser4_seal_validate(&hint->seal, &hint->ext_coord.coord, + key, &hint->lh, lock_mode, + ZNODE_LOCK_LOPRI); +} + +/* reserve disk space when writing a logical cluster */ +static int reserve4cluster(struct inode *inode, struct cluster_handle *clust) +{ + int result = 0; + + assert("edward-965", reiser4_schedulable()); + assert("edward-439", inode != NULL); + assert("edward-440", clust != NULL); + assert("edward-441", clust->pages != NULL); + + if (clust->nr_pages == 0) { + assert("edward-1152", clust->win != NULL); + assert("edward-1153", clust->win->stat == HOLE_WINDOW); + /* don't reserve disk space for fake logical cluster */ + return 0; + } + assert("edward-442", jprivate(clust->pages[0]) != NULL); + + result = reiser4_grab_space_force(estimate_insert_cluster(inode) + + estimate_update_cluster(inode), + BA_CAN_COMMIT); + if (result) + return result; + clust->reserved = 1; + grabbed2cluster_reserved(estimate_insert_cluster(inode) + + estimate_update_cluster(inode)); +#if REISER4_DEBUG + clust->reserved_prepped = estimate_update_cluster(inode); + clust->reserved_unprepped = estimate_insert_cluster(inode); +#endif + /* there can be space grabbed by txnmgr_force_commit_all */ + return 0; +} + +/* free reserved disk space if writing a logical cluster fails */ +static void free_reserved4cluster(struct inode *inode, + struct cluster_handle *ch, int count) +{ + assert("edward-967", ch->reserved == 1); + + cluster_reserved2free(count); + ch->reserved = 0; +} + +/* The core search procedure of the cryptcompress plugin. + If returned value is not cbk_errored, then current znode is locked */ +static int find_cluster_item(hint_t * hint, + const reiser4_key * key, /* key of the item we are + looking for */ + znode_lock_mode lock_mode /* which lock */ , + ra_info_t * ra_info, lookup_bias bias, __u32 flags) +{ + int result; + reiser4_key ikey; + int went_right = 0; + coord_t *coord = &hint->ext_coord.coord; + coord_t orig = *coord; + + assert("edward-152", hint != NULL); + + if (!hint_is_valid(hint)) { + result = cryptcompress_hint_validate(hint, key, lock_mode); + if (result == -E_REPEAT) + goto traverse_tree; + else if (result) { + assert("edward-1216", 0); + return result; + } + hint_set_valid(hint); + } + assert("edward-709", znode_is_any_locked(coord->node)); + + /* In-place lookup is going here, it means we just need to + check if next item of the @coord match to the @keyhint) */ + + if (equal_to_rdk(coord->node, key)) { + result = goto_right_neighbor(coord, &hint->lh); + if (result == -E_NO_NEIGHBOR) { + assert("edward-1217", 0); + return RETERR(-EIO); + } + if (result) + return result; + assert("edward-1218", equal_to_ldk(coord->node, key)); + went_right = 1; + } else { + coord->item_pos++; + coord->unit_pos = 0; + coord->between = AT_UNIT; + } + result = zload(coord->node); + if (result) + return result; + assert("edward-1219", !node_is_empty(coord->node)); + + if (!coord_is_existing_item(coord)) { + zrelse(coord->node); + goto not_found; + } + item_key_by_coord(coord, &ikey); + zrelse(coord->node); + if (!keyeq(key, &ikey)) + goto not_found; + /* Ok, item is found, update node counts */ + if (went_right) + dclust_inc_extension_ncount(hint); + return CBK_COORD_FOUND; + + not_found: + assert("edward-1220", coord->item_pos > 0); + //coord->item_pos--; + /* roll back */ + *coord = orig; + ON_DEBUG(coord_update_v(coord)); + return CBK_COORD_NOTFOUND; + + traverse_tree: + assert("edward-713", hint->lh.owner == NULL); + assert("edward-714", reiser4_schedulable()); + + reiser4_unset_hint(hint); + dclust_init_extension(hint); + coord_init_zero(coord); + result = coord_by_key(current_tree, key, coord, &hint->lh, + lock_mode, bias, LEAF_LEVEL, LEAF_LEVEL, + CBK_UNIQUE | flags, ra_info); + if (cbk_errored(result)) + return result; + if(result == CBK_COORD_FOUND) + dclust_inc_extension_ncount(hint); + hint_set_valid(hint); + return result; +} + +/* This function is called by deflate[inflate] manager when + creating a transformed/plain stream to check if we should + create/cut some overhead. If this returns true, then @oh + contains the size of this overhead. + */ +static int need_cut_or_align(struct inode * inode, + struct cluster_handle * ch, rw_op rw, int * oh) +{ + struct tfm_cluster * tc = &ch->tc; + switch (rw) { + case WRITE_OP: /* estimate align */ + *oh = tc->len % cipher_blocksize(inode); + if (*oh != 0) + return 1; + break; + case READ_OP: /* estimate cut */ + *oh = *(tfm_output_data(ch) + tc->len - 1); + break; + default: + impossible("edward-1401", "bad option"); + } + return (tc->len != tc->lsize); +} + +/* create/cut an overhead of transformed/plain stream */ +static void align_or_cut_overhead(struct inode * inode, + struct cluster_handle * ch, rw_op rw) +{ + unsigned int oh; + cipher_plugin * cplug = inode_cipher_plugin(inode); + + assert("edward-1402", need_cipher(inode)); + + if (!need_cut_or_align(inode, ch, rw, &oh)) + return; + switch (rw) { + case WRITE_OP: /* do align */ + ch->tc.len += + cplug->align_stream(tfm_input_data(ch) + + ch->tc.len, ch->tc.len, + cipher_blocksize(inode)); + *(tfm_input_data(ch) + ch->tc.len - 1) = + cipher_blocksize(inode) - oh; + break; + case READ_OP: /* do cut */ + assert("edward-1403", oh <= cipher_blocksize(inode)); + ch->tc.len -= oh; + break; + default: + impossible("edward-1404", "bad option"); + } + return; +} + +static unsigned max_cipher_overhead(struct inode * inode) +{ + if (!need_cipher(inode) || !inode_cipher_plugin(inode)->align_stream) + return 0; + return cipher_blocksize(inode); +} + +static int deflate_overhead(struct inode *inode) +{ + return (inode_compression_plugin(inode)-> + checksum ? DC_CHECKSUM_SIZE : 0); +} + +static unsigned deflate_overrun(struct inode * inode, int ilen) +{ + return coa_overrun(inode_compression_plugin(inode), ilen); +} + +/* Estimating compressibility of a logical cluster by various + policies represented by compression mode plugin. + If this returns false, then compressor won't be called for + the cluster of index @index. +*/ +static int should_compress(struct tfm_cluster * tc, cloff_t index, + struct inode *inode) +{ + compression_plugin *cplug = inode_compression_plugin(inode); + compression_mode_plugin *mplug = inode_compression_mode_plugin(inode); + + assert("edward-1321", tc->len != 0); + assert("edward-1322", cplug != NULL); + assert("edward-1323", mplug != NULL); + + return /* estimate by size */ + (cplug->min_size_deflate ? + tc->len >= cplug->min_size_deflate() : + 1) && + /* estimate by compression mode plugin */ + (mplug->should_deflate ? + mplug->should_deflate(inode, index) : + 1); +} + +/* Evaluating results of compression transform. + Returns true, if we need to accept this results */ +static int save_compressed(int size_before, int size_after, struct inode *inode) +{ + return (size_after + deflate_overhead(inode) + + max_cipher_overhead(inode) < size_before); +} + +/* Guess result of the evaluation above */ +static int need_inflate(struct cluster_handle * ch, struct inode * inode, + int encrypted /* is cluster encrypted */ ) +{ + struct tfm_cluster * tc = &ch->tc; + + assert("edward-142", tc != 0); + assert("edward-143", inode != NULL); + + return tc->len < + (encrypted ? + inode_scaled_offset(inode, tc->lsize) : + tc->lsize); +} + +/* If results of compression were accepted, then we add + a checksum to catch possible disk cluster corruption. + The following is a format of the data stored in disk clusters: + + data This is (transformed) logical cluster. + cipher_overhead This is created by ->align() method + of cipher plugin. May be absent. + checksum (4) This is created by ->checksum method + of compression plugin to check + integrity. May be absent. + + Crypto overhead format: + + data + control_byte (1) contains aligned overhead size: + 1 <= overhead <= cipher_blksize +*/ +/* Append a checksum at the end of a transformed stream */ +static void dc_set_checksum(compression_plugin * cplug, struct tfm_cluster * tc) +{ + __u32 checksum; + + assert("edward-1309", tc != NULL); + assert("edward-1310", tc->len > 0); + assert("edward-1311", cplug->checksum != NULL); + + checksum = cplug->checksum(tfm_stream_data(tc, OUTPUT_STREAM), tc->len); + put_unaligned(cpu_to_le32(checksum), + (d32 *)(tfm_stream_data(tc, OUTPUT_STREAM) + tc->len)); + tc->len += (int)DC_CHECKSUM_SIZE; +} + +/* Check a disk cluster checksum. + Returns 0 if checksum is correct, otherwise returns 1 */ +static int dc_check_checksum(compression_plugin * cplug, struct tfm_cluster * tc) +{ + assert("edward-1312", tc != NULL); + assert("edward-1313", tc->len > (int)DC_CHECKSUM_SIZE); + assert("edward-1314", cplug->checksum != NULL); + + if (cplug->checksum(tfm_stream_data(tc, INPUT_STREAM), + tc->len - (int)DC_CHECKSUM_SIZE) != + le32_to_cpu(get_unaligned((d32 *) + (tfm_stream_data(tc, INPUT_STREAM) + + tc->len - (int)DC_CHECKSUM_SIZE)))) { + warning("edward-156", + "Bad disk cluster checksum %d, (should be %d) Fsck?\n", + (int)le32_to_cpu + (get_unaligned((d32 *) + (tfm_stream_data(tc, INPUT_STREAM) + + tc->len - (int)DC_CHECKSUM_SIZE))), + (int)cplug->checksum + (tfm_stream_data(tc, INPUT_STREAM), + tc->len - (int)DC_CHECKSUM_SIZE)); + return 1; + } + tc->len -= (int)DC_CHECKSUM_SIZE; + return 0; +} + +/* get input/output stream for some transform action */ +int grab_tfm_stream(struct inode * inode, struct tfm_cluster * tc, + tfm_stream_id id) +{ + size_t size = inode_scaled_cluster_size(inode); + + assert("edward-901", tc != NULL); + assert("edward-1027", inode_compression_plugin(inode) != NULL); + + if (cluster_get_tfm_act(tc) == TFMA_WRITE) + size += deflate_overrun(inode, inode_cluster_size(inode)); + + if (!get_tfm_stream(tc, id) && id == INPUT_STREAM) + alternate_streams(tc); + if (!get_tfm_stream(tc, id)) + return alloc_tfm_stream(tc, size, id); + + assert("edward-902", tfm_stream_is_set(tc, id)); + + if (tfm_stream_size(tc, id) < size) + return realloc_tfm_stream(tc, size, id); + return 0; +} + +/* Common deflate manager */ +int reiser4_deflate_cluster(struct cluster_handle * clust, struct inode * inode) +{ + int result = 0; + int compressed = 0; + int encrypted = 0; + struct tfm_cluster * tc = &clust->tc; + compression_plugin * coplug; + + assert("edward-401", inode != NULL); + assert("edward-903", tfm_stream_is_set(tc, INPUT_STREAM)); + assert("edward-1348", cluster_get_tfm_act(tc) == TFMA_WRITE); + assert("edward-498", !tfm_cluster_is_uptodate(tc)); + + coplug = inode_compression_plugin(inode); + if (should_compress(tc, clust->index, inode)) { + /* try to compress, discard bad results */ + __u32 dst_len; + compression_mode_plugin * mplug = + inode_compression_mode_plugin(inode); + assert("edward-602", coplug != NULL); + assert("edward-1423", coplug->compress != NULL); + + result = grab_coa(tc, coplug); + if (result) { + warning("edward-1424", + "alloc_coa failed with ret=%d, skipped compression", + result); + goto cipher; + } + result = grab_tfm_stream(inode, tc, OUTPUT_STREAM); + if (result) { + warning("edward-1425", + "alloc stream failed with ret=%d, skipped compression", + result); + goto cipher; + } + dst_len = tfm_stream_size(tc, OUTPUT_STREAM); + coplug->compress(get_coa(tc, coplug->h.id, tc->act), + tfm_input_data(clust), tc->len, + tfm_output_data(clust), &dst_len); + /* make sure we didn't overwrite extra bytes */ + assert("edward-603", + dst_len <= tfm_stream_size(tc, OUTPUT_STREAM)); + + /* evaluate results of compression transform */ + if (save_compressed(tc->len, dst_len, inode)) { + /* good result, accept */ + tc->len = dst_len; + if (mplug->accept_hook != NULL) { + result = mplug->accept_hook(inode, clust->index); + if (result) + warning("edward-1426", + "accept_hook failed with ret=%d", + result); + } + compressed = 1; + } + else { + /* bad result, discard */ +#if 0 + if (cluster_is_complete(clust, inode)) + warning("edward-1496", + "incompressible cluster %lu (inode %llu)", + clust->index, + (unsigned long long)get_inode_oid(inode)); +#endif + if (mplug->discard_hook != NULL && + cluster_is_complete(clust, inode)) { + result = mplug->discard_hook(inode, + clust->index); + if (result) + warning("edward-1427", + "discard_hook failed with ret=%d", + result); + } + } + } + cipher: + if (need_cipher(inode)) { + cipher_plugin * ciplug; + struct blkcipher_desc desc; + struct scatterlist src; + struct scatterlist dst; + + ciplug = inode_cipher_plugin(inode); + desc.tfm = info_get_cipher(inode_crypto_info(inode)); + desc.flags = 0; + if (compressed) + alternate_streams(tc); + result = grab_tfm_stream(inode, tc, OUTPUT_STREAM); + if (result) + return result; + + align_or_cut_overhead(inode, clust, WRITE_OP); + sg_init_one(&src, tfm_input_data(clust), tc->len); + sg_init_one(&dst, tfm_output_data(clust), tc->len); + + result = crypto_blkcipher_encrypt(&desc, &dst, &src, tc->len); + if (result) { + warning("edward-1405", + "encryption failed flags=%x\n", desc.flags); + return result; + } + encrypted = 1; + } + if (compressed && coplug->checksum != NULL) + dc_set_checksum(coplug, tc); + if (!compressed && !encrypted) + alternate_streams(tc); + return result; +} + +/* Common inflate manager. */ +int reiser4_inflate_cluster(struct cluster_handle * clust, struct inode * inode) +{ + int result = 0; + int transformed = 0; + struct tfm_cluster * tc = &clust->tc; + compression_plugin * coplug; + + assert("edward-905", inode != NULL); + assert("edward-1178", clust->dstat == PREP_DISK_CLUSTER); + assert("edward-906", tfm_stream_is_set(&clust->tc, INPUT_STREAM)); + assert("edward-1349", tc->act == TFMA_READ); + assert("edward-907", !tfm_cluster_is_uptodate(tc)); + + /* Handle a checksum (if any) */ + coplug = inode_compression_plugin(inode); + if (need_inflate(clust, inode, need_cipher(inode)) && + coplug->checksum != NULL) { + result = dc_check_checksum(coplug, tc); + if (unlikely(result)) { + warning("edward-1460", + "Inode %llu: disk cluster %lu looks corrupted", + (unsigned long long)get_inode_oid(inode), + clust->index); + return RETERR(-EIO); + } + } + if (need_cipher(inode)) { + cipher_plugin * ciplug; + struct blkcipher_desc desc; + struct scatterlist src; + struct scatterlist dst; + + ciplug = inode_cipher_plugin(inode); + desc.tfm = info_get_cipher(inode_crypto_info(inode)); + desc.flags = 0; + result = grab_tfm_stream(inode, tc, OUTPUT_STREAM); + if (result) + return result; + assert("edward-909", tfm_cluster_is_set(tc)); + + sg_init_one(&src, tfm_input_data(clust), tc->len); + sg_init_one(&dst, tfm_output_data(clust), tc->len); + + result = crypto_blkcipher_decrypt(&desc, &dst, &src, tc->len); + if (result) { + warning("edward-1600", "decrypt failed flags=%x\n", + desc.flags); + return result; + } + align_or_cut_overhead(inode, clust, READ_OP); + transformed = 1; + } + if (need_inflate(clust, inode, 0)) { + unsigned dst_len = inode_cluster_size(inode); + if(transformed) + alternate_streams(tc); + + result = grab_tfm_stream(inode, tc, OUTPUT_STREAM); + if (result) + return result; + assert("edward-1305", coplug->decompress != NULL); + assert("edward-910", tfm_cluster_is_set(tc)); + + coplug->decompress(get_coa(tc, coplug->h.id, tc->act), + tfm_input_data(clust), tc->len, + tfm_output_data(clust), &dst_len); + /* check length */ + tc->len = dst_len; + assert("edward-157", dst_len == tc->lsize); + transformed = 1; + } + if (!transformed) + alternate_streams(tc); + return result; +} + +/* This is implementation of readpage method of struct + address_space_operations for cryptcompress plugin. */ +int readpage_cryptcompress(struct file *file, struct page *page) +{ + reiser4_context *ctx; + struct cluster_handle clust; + item_plugin *iplug; + int result; + + assert("edward-88", PageLocked(page)); + assert("vs-976", !PageUptodate(page)); + assert("edward-89", page->mapping && page->mapping->host); + + ctx = reiser4_init_context(page->mapping->host->i_sb); + if (IS_ERR(ctx)) { + unlock_page(page); + return PTR_ERR(ctx); + } + assert("edward-113", + ergo(file != NULL, + page->mapping == file->f_dentry->d_inode->i_mapping)); + + if (PageUptodate(page)) { + warning("edward-1338", "page is already uptodate\n"); + unlock_page(page); + reiser4_exit_context(ctx); + return 0; + } + cluster_init_read(&clust, NULL); + clust.file = file; + iplug = item_plugin_by_id(CTAIL_ID); + if (!iplug->s.file.readpage) { + unlock_page(page); + put_cluster_handle(&clust); + reiser4_exit_context(ctx); + return -EINVAL; + } + result = iplug->s.file.readpage(&clust, page); + + put_cluster_handle(&clust); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return result; +} + +/* number of pages to check in */ +static int get_new_nrpages(struct cluster_handle * clust) +{ + switch (clust->op) { + case LC_APPOV: + return clust->nr_pages; + case LC_TRUNC: + assert("edward-1179", clust->win != NULL); + return size_in_pages(clust->win->off + clust->win->count); + default: + impossible("edward-1180", "bad page cluster option"); + return 0; + } +} + +static void set_cluster_pages_dirty(struct cluster_handle * clust, + struct inode * inode) +{ + int i; + struct page *pg; + int nrpages = get_new_nrpages(clust); + + for (i = 0; i < nrpages; i++) { + + pg = clust->pages[i]; + assert("edward-968", pg != NULL); + lock_page(pg); + assert("edward-1065", PageUptodate(pg)); + set_page_dirty_notag(pg); + unlock_page(pg); + mark_page_accessed(pg); + } +} + +/* Grab a page cluster for read/write operations. + Attach a jnode for write operations (when preparing for modifications, which + are supposed to be committed). + + We allocate only one jnode per page cluster; this jnode is binded to the + first page of this cluster, so we have an extra-reference that will be put + as soon as jnode is evicted from memory), other references will be cleaned + up in flush time (assume that check in page cluster was successful). +*/ +int grab_page_cluster(struct inode * inode, + struct cluster_handle * clust, rw_op rw) +{ + int i; + int result = 0; + jnode *node = NULL; + + assert("edward-182", clust != NULL); + assert("edward-183", clust->pages != NULL); + assert("edward-1466", clust->node == NULL); + assert("edward-1428", inode != NULL); + assert("edward-1429", inode->i_mapping != NULL); + assert("edward-184", clust->nr_pages <= cluster_nrpages(inode)); + + if (clust->nr_pages == 0) + return 0; + + for (i = 0; i < clust->nr_pages; i++) { + + assert("edward-1044", clust->pages[i] == NULL); + + clust->pages[i] = + find_or_create_page(inode->i_mapping, + clust_to_pg(clust->index, inode) + i, + reiser4_ctx_gfp_mask_get()); + if (!clust->pages[i]) { + result = RETERR(-ENOMEM); + break; + } + if (i == 0 && rw == WRITE_OP) { + node = jnode_of_page(clust->pages[i]); + if (IS_ERR(node)) { + result = PTR_ERR(node); + unlock_page(clust->pages[i]); + break; + } + JF_SET(node, JNODE_CLUSTER_PAGE); + assert("edward-920", jprivate(clust->pages[0])); + } + INODE_PGCOUNT_INC(inode); + unlock_page(clust->pages[i]); + } + if (unlikely(result)) { + while (i) { + put_cluster_page(clust->pages[--i]); + INODE_PGCOUNT_DEC(inode); + } + if (node && !IS_ERR(node)) + jput(node); + return result; + } + clust->node = node; + return 0; +} + +static void truncate_page_cluster_range(struct inode * inode, + struct page ** pages, + cloff_t index, + int from, int count, + int even_cows) +{ + assert("edward-1467", count > 0); + reiser4_invalidate_pages(inode->i_mapping, + clust_to_pg(index, inode) + from, + count, even_cows); +} + +/* Put @count pages starting from @from offset */ +void __put_page_cluster(int from, int count, + struct page ** pages, struct inode * inode) +{ + int i; + assert("edward-1468", pages != NULL); + assert("edward-1469", inode != NULL); + assert("edward-1470", from >= 0 && count >= 0); + + for (i = 0; i < count; i++) { + assert("edward-1471", pages[from + i] != NULL); + assert("edward-1472", + pages[from + i]->index == pages[from]->index + i); + + put_cluster_page(pages[from + i]); + INODE_PGCOUNT_DEC(inode); + } +} + +/* + * This is dual to grab_page_cluster, + * however if @rw == WRITE_OP, then we call this function + * only if something is failed before checkin page cluster. + */ +void put_page_cluster(struct cluster_handle * clust, + struct inode * inode, rw_op rw) +{ + assert("edward-445", clust != NULL); + assert("edward-922", clust->pages != NULL); + assert("edward-446", + ergo(clust->nr_pages != 0, clust->pages[0] != NULL)); + + __put_page_cluster(0, clust->nr_pages, clust->pages, inode); + if (rw == WRITE_OP) { + if (unlikely(clust->node)) { + assert("edward-447", + clust->node == jprivate(clust->pages[0])); + jput(clust->node); + clust->node = NULL; + } + } +} + +#if REISER4_DEBUG +int cryptcompress_inode_ok(struct inode *inode) +{ + if (!(reiser4_inode_data(inode)->plugin_mask & (1 << PSET_FILE))) + return 0; + if (!cluster_shift_ok(inode_cluster_shift(inode))) + return 0; + return 1; +} + +static int window_ok(struct reiser4_slide * win, struct inode *inode) +{ + assert("edward-1115", win != NULL); + assert("edward-1116", ergo(win->delta, win->stat == HOLE_WINDOW)); + + return (win->off != inode_cluster_size(inode)) && + (win->off + win->count + win->delta <= inode_cluster_size(inode)); +} + +static int cluster_ok(struct cluster_handle * clust, struct inode *inode) +{ + assert("edward-279", clust != NULL); + + if (!clust->pages) + return 0; + return (clust->win ? window_ok(clust->win, inode) : 1); +} +#if 0 +static int pages_truncate_ok(struct inode *inode, pgoff_t start) +{ + int found; + struct page * page; + + found = find_get_pages(inode->i_mapping, start, 1, &page); + if (found) + put_cluster_page(page); + return !found; +} +#else +#define pages_truncate_ok(inode, start) 1 +#endif + +static int jnode_truncate_ok(struct inode *inode, cloff_t index) +{ + jnode *node; + node = jlookup(current_tree, get_inode_oid(inode), + clust_to_pg(index, inode)); + if (likely(!node)) + return 1; + jput(node); + return 0; +} + +static int find_fake_appended(struct inode *inode, cloff_t * index); + +static int body_truncate_ok(struct inode *inode, cloff_t aidx) +{ + int result; + cloff_t raidx; + + result = find_fake_appended(inode, &raidx); + return !result && (aidx == raidx); +} +#endif + +/* guess next window stat */ +static inline window_stat next_window_stat(struct reiser4_slide * win) +{ + assert("edward-1130", win != NULL); + return ((win->stat == HOLE_WINDOW && win->delta == 0) ? + HOLE_WINDOW : DATA_WINDOW); +} + +/* guess and set next cluster index and window params */ +static void move_update_window(struct inode * inode, + struct cluster_handle * clust, + loff_t file_off, loff_t to_file) +{ + struct reiser4_slide * win; + + assert("edward-185", clust != NULL); + assert("edward-438", clust->pages != NULL); + assert("edward-281", cluster_ok(clust, inode)); + + win = clust->win; + if (!win) + return; + + switch (win->stat) { + case DATA_WINDOW: + /* increment */ + clust->index++; + win->stat = DATA_WINDOW; + win->off = 0; + win->count = min((loff_t)inode_cluster_size(inode), to_file); + break; + case HOLE_WINDOW: + switch (next_window_stat(win)) { + case HOLE_WINDOW: + /* skip */ + clust->index = off_to_clust(file_off, inode); + win->stat = HOLE_WINDOW; + win->off = 0; + win->count = off_to_cloff(file_off, inode); + win->delta = min((loff_t)(inode_cluster_size(inode) - + win->count), to_file); + break; + case DATA_WINDOW: + /* stay */ + win->stat = DATA_WINDOW; + /* off+count+delta=inv */ + win->off = win->off + win->count; + win->count = win->delta; + win->delta = 0; + break; + default: + impossible("edward-282", "wrong next window state"); + } + break; + default: + impossible("edward-283", "wrong current window state"); + } + assert("edward-1068", cluster_ok(clust, inode)); +} + +static int update_sd_cryptcompress(struct inode *inode) +{ + int result = 0; + + assert("edward-978", reiser4_schedulable()); + + result = reiser4_grab_space_force(/* one for stat data update */ + estimate_update_common(inode), + BA_CAN_COMMIT); + if (result) + return result; + inode->i_ctime = inode->i_mtime = CURRENT_TIME; + result = reiser4_update_sd(inode); + + return result; +} + +static void uncapture_cluster_jnode(jnode * node) +{ + txn_atom *atom; + + assert_spin_locked(&(node->guard)); + + atom = jnode_get_atom(node); + if (atom == NULL) { + assert("jmacd-7111", !JF_ISSET(node, JNODE_DIRTY)); + spin_unlock_jnode(node); + return; + } + reiser4_uncapture_block(node); + spin_unlock_atom(atom); + jput(node); +} + +static void put_found_pages(struct page **pages, int nr) +{ + int i; + for (i = 0; i < nr; i++) { + assert("edward-1045", pages[i] != NULL); + put_cluster_page(pages[i]); + } +} + +/* Lifecycle of a logical cluster in the system. + * + * + * Logical cluster of a cryptcompress file is represented in the system by + * . page cluster (in memory, primary cache, contains plain text); + * . disk cluster (in memory, secondary cache, contains transformed text). + * Primary cache is to reduce number of transform operations (compression, + * encryption), i.e. to implement transform-caching strategy. + * Secondary cache is to reduce number of I/O operations, i.e. for usual + * write-caching strategy. Page cluster is a set of pages, i.e. mapping of + * a logical cluster to the primary cache. Disk cluster is a set of items + * of the same type defined by some reiser4 item plugin id. + * + * 1. Performing modifications + * + * Every modification of a cryptcompress file is considered as a set of + * operations performed on file's logical clusters. Every such "atomic" + * modification is truncate, append and(or) overwrite some bytes of a + * logical cluster performed in the primary cache with the following + * synchronization with the secondary cache (in flush time). Disk clusters, + * which live in the secondary cache, are supposed to be synchronized with + * disk. The mechanism of synchronization of primary and secondary caches + * includes so-called checkin/checkout technique described below. + * + * 2. Submitting modifications + * + * Each page cluster has associated jnode (a special in-memory header to + * keep a track of transactions in reiser4), which is attached to its first + * page when grabbing page cluster for modifications (see grab_page_cluster). + * Submitting modifications (see checkin_logical_cluster) is going per logical + * cluster and includes: + * . checkin_cluster_size; + * . checkin_page_cluster. + * checkin_cluster_size() is resolved to file size update (which completely + * defines new size of logical cluster (number of file's bytes in a logical + * cluster). + * checkin_page_cluster() captures jnode of a page cluster and installs + * jnode's dirty flag (if needed) to indicate that modifications are + * successfully checked in. + * + * 3. Checking out modifications + * + * Is going per logical cluster in flush time (see checkout_logical_cluster). + * This is the time of synchronizing primary and secondary caches. + * checkout_logical_cluster() includes: + * . checkout_page_cluster (retrieving checked in pages). + * . uncapture jnode (including clear dirty flag and unlock) + * + * 4. Committing modifications + * + * Proceeding a synchronization of primary and secondary caches. When checking + * out page cluster (the phase above) pages are locked/flushed/unlocked + * one-by-one in ascending order of their indexes to contiguous stream, which + * is supposed to be transformed (compressed, encrypted), chopped up into items + * and committed to disk as a disk cluster. + * + * 5. Managing page references + * + * Every checked in page have a special additional "control" reference, + * which is dropped at checkout. We need this to avoid unexpected evicting + * pages from memory before checkout. Control references are managed so + * they are not accumulated with every checkin: + * + * 0 + * checkin -> 1 + * 0 -> checkout + * checkin -> 1 + * checkin -> 1 + * checkin -> 1 + * 0 -> checkout + * ... + * + * Every page cluster has its own unique "cluster lock". Update/drop + * references are serialized via this lock. Number of checked in cluster + * pages is calculated by i_size under cluster lock. File size is updated + * at every checkin action also under cluster lock (except cases of + * appending/truncating fake logical clusters). + * + * Proof of correctness: + * + * Since we update file size under cluster lock, in the case of non-fake + * logical cluster with its lock held we do have expected number of checked + * in pages. On the other hand, append/truncate of fake logical clusters + * doesn't change number of checked in pages of any cluster. + * + * NOTE-EDWARD: As cluster lock we use guard (spinlock_t) of its jnode. + * Currently, I don't see any reason to create a special lock for those + * needs. + */ + +static inline void lock_cluster(jnode * node) +{ + spin_lock_jnode(node); +} + +static inline void unlock_cluster(jnode * node) +{ + spin_unlock_jnode(node); +} + +static inline void unlock_cluster_uncapture(jnode * node) +{ + uncapture_cluster_jnode(node); +} + +/* Set new file size by window. Cluster lock is required. */ +static void checkin_file_size(struct cluster_handle * clust, + struct inode * inode) +{ + loff_t new_size; + struct reiser4_slide * win; + + assert("edward-1181", clust != NULL); + assert("edward-1182", inode != NULL); + assert("edward-1473", clust->pages != NULL); + assert("edward-1474", clust->pages[0] != NULL); + assert("edward-1475", jprivate(clust->pages[0]) != NULL); + assert_spin_locked(&(jprivate(clust->pages[0])->guard)); + + + win = clust->win; + assert("edward-1183", win != NULL); + + new_size = clust_to_off(clust->index, inode) + win->off; + + switch (clust->op) { + case LC_APPOV: + if (new_size + win->count <= i_size_read(inode)) + /* overwrite only */ + return; + new_size += win->count; + break; + case LC_TRUNC: + break; + default: + impossible("edward-1184", "bad page cluster option"); + break; + } + inode_check_scale_nolock(inode, i_size_read(inode), new_size); + i_size_write(inode, new_size); + return; +} + +static inline void checkin_cluster_size(struct cluster_handle * clust, + struct inode * inode) +{ + if (clust->win) + checkin_file_size(clust, inode); +} + +static int checkin_page_cluster(struct cluster_handle * clust, + struct inode * inode) +{ + int result; + jnode * node; + int old_nrpages = clust->old_nrpages; + int new_nrpages = get_new_nrpages(clust); + + node = clust->node; + + assert("edward-221", node != NULL); + assert("edward-971", clust->reserved == 1); + assert("edward-1263", + clust->reserved_prepped == estimate_update_cluster(inode)); + assert("edward-1264", clust->reserved_unprepped == 0); + + if (JF_ISSET(node, JNODE_DIRTY)) { + /* + * page cluster was checked in, but not yet + * checked out, so release related resources + */ + free_reserved4cluster(inode, clust, + estimate_update_cluster(inode)); + __put_page_cluster(0, clust->old_nrpages, + clust->pages, inode); + } else { + result = capture_cluster_jnode(node); + if (unlikely(result)) { + unlock_cluster(node); + return result; + } + jnode_make_dirty_locked(node); + clust->reserved = 0; + } + unlock_cluster(node); + + if (new_nrpages < old_nrpages) { + /* truncate >= 1 complete pages */ + __put_page_cluster(new_nrpages, + old_nrpages - new_nrpages, + clust->pages, inode); + truncate_page_cluster_range(inode, + clust->pages, clust->index, + new_nrpages, + old_nrpages - new_nrpages, + 0); + } +#if REISER4_DEBUG + clust->reserved_prepped -= estimate_update_cluster(inode); +#endif + return 0; +} + +/* Submit modifications of a logical cluster */ +static int checkin_logical_cluster(struct cluster_handle * clust, + struct inode *inode) +{ + int result = 0; + jnode * node; + + node = clust->node; + + assert("edward-1035", node != NULL); + assert("edward-1029", clust != NULL); + assert("edward-1030", clust->reserved == 1); + assert("edward-1031", clust->nr_pages != 0); + assert("edward-1032", clust->pages != NULL); + assert("edward-1033", clust->pages[0] != NULL); + assert("edward-1446", jnode_is_cluster_page(node)); + assert("edward-1476", node == jprivate(clust->pages[0])); + + lock_cluster(node); + checkin_cluster_size(clust, inode); + /* this will unlock cluster */ + result = checkin_page_cluster(clust, inode); + jput(node); + clust->node = NULL; + return result; +} + +/* + * Retrieve size of logical cluster that was checked in at + * the latest modifying session (cluster lock is required) + */ +static inline void checkout_cluster_size(struct cluster_handle * clust, + struct inode * inode) +{ + struct tfm_cluster *tc = &clust->tc; + + tc->len = lbytes(clust->index, inode); + assert("edward-1478", tc->len != 0); +} + +/* + * Retrieve a page cluster with the latest submitted modifications + * and flush its pages to previously allocated contiguous stream. + */ +static void checkout_page_cluster(struct cluster_handle * clust, + jnode * node, struct inode * inode) +{ + int i; + int found; + int to_put; + struct tfm_cluster *tc = &clust->tc; + + /* find and put checked in pages: cluster is locked, + * so we must get expected number (to_put) of pages + */ + to_put = size_in_pages(lbytes(clust->index, inode)); + found = find_get_pages(inode->i_mapping, + clust_to_pg(clust->index, inode), + to_put, clust->pages); + BUG_ON(found != to_put); + + __put_page_cluster(0, to_put, clust->pages, inode); + unlock_cluster_uncapture(node); + + /* Flush found pages. + * + * Note, that we don't disable modifications while flushing, + * moreover, some found pages can be truncated, as we have + * released cluster lock. + */ + for (i = 0; i < found; i++) { + int in_page; + char * data; + assert("edward-1479", + clust->pages[i]->index == clust->pages[0]->index + i); + + lock_page(clust->pages[i]); + if (!PageUptodate(clust->pages[i])) { + /* page was truncated */ + assert("edward-1480", + i_size_read(inode) <= page_offset(clust->pages[i])); + assert("edward-1481", + clust->pages[i]->mapping != inode->i_mapping); + unlock_page(clust->pages[i]); + break; + } + /* Update the number of bytes in the logical cluster, + * as it could be partially truncated. Note, that only + * partial truncate is possible (complete truncate can + * not go here, as it is performed via ->kill_hook() + * called by cut_file_items(), and the last one must + * wait for znode locked with parent coord). + */ + checkout_cluster_size(clust, inode); + + /* this can be zero, as new file size is + checked in before truncating pages */ + in_page = __mbp(tc->len, i); + + data = kmap(clust->pages[i]); + memcpy(tfm_stream_data(tc, INPUT_STREAM) + pg_to_off(i), + data, in_page); + kunmap(clust->pages[i]); + + if (PageDirty(clust->pages[i])) + cancel_dirty_page(clust->pages[i], PAGE_CACHE_SIZE); + + unlock_page(clust->pages[i]); + + if (in_page < PAGE_CACHE_SIZE) + /* end of the file */ + break; + } + put_found_pages(clust->pages, found); /* find_get_pages */ + tc->lsize = tc->len; + return; +} + +/* Check out modifications of a logical cluster */ +int checkout_logical_cluster(struct cluster_handle * clust, + jnode * node, struct inode *inode) +{ + int result; + struct tfm_cluster *tc = &clust->tc; + + assert("edward-980", node != NULL); + assert("edward-236", inode != NULL); + assert("edward-237", clust != NULL); + assert("edward-240", !clust->win); + assert("edward-241", reiser4_schedulable()); + assert("edward-718", cryptcompress_inode_ok(inode)); + + result = grab_tfm_stream(inode, tc, INPUT_STREAM); + if (result) { + warning("edward-1430", "alloc stream failed with ret=%d", + result); + return RETERR(-E_REPEAT); + } + lock_cluster(node); + + if (unlikely(!JF_ISSET(node, JNODE_DIRTY))) { + /* race with another flush */ + warning("edward-982", + "checking out logical cluster %lu of inode %llu: " + "jnode is not dirty", clust->index, + (unsigned long long)get_inode_oid(inode)); + unlock_cluster(node); + return RETERR(-E_REPEAT); + } + cluster_reserved2grabbed(estimate_update_cluster(inode)); + + /* this will unlock cluster */ + checkout_page_cluster(clust, node, inode); + return 0; +} + +/* set hint for the cluster of the index @index */ +static void set_hint_cluster(struct inode *inode, hint_t * hint, + cloff_t index, znode_lock_mode mode) +{ + reiser4_key key; + assert("edward-722", cryptcompress_inode_ok(inode)); + assert("edward-723", + inode_file_plugin(inode) == + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)); + + inode_file_plugin(inode)->key_by_inode(inode, + clust_to_off(index, inode), + &key); + + reiser4_seal_init(&hint->seal, &hint->ext_coord.coord, &key); + hint->offset = get_key_offset(&key); + hint->mode = mode; +} + +void invalidate_hint_cluster(struct cluster_handle * clust) +{ + assert("edward-1291", clust != NULL); + assert("edward-1292", clust->hint != NULL); + + done_lh(&clust->hint->lh); + hint_clr_valid(clust->hint); +} + +static void put_hint_cluster(struct cluster_handle * clust, + struct inode *inode, znode_lock_mode mode) +{ + assert("edward-1286", clust != NULL); + assert("edward-1287", clust->hint != NULL); + + set_hint_cluster(inode, clust->hint, clust->index + 1, mode); + invalidate_hint_cluster(clust); +} + +static int balance_dirty_page_cluster(struct cluster_handle * clust, + struct inode *inode, loff_t off, + loff_t to_file, + int nr_dirtied) +{ + int result; + struct cryptcompress_info * info; + + assert("edward-724", inode != NULL); + assert("edward-725", cryptcompress_inode_ok(inode)); + assert("edward-1547", + nr_dirtied != 0 && nr_dirtied <= cluster_nrpages(inode)); + + /* set next window params */ + move_update_window(inode, clust, off, to_file); + + result = update_sd_cryptcompress(inode); + if (result) + return result; + assert("edward-726", clust->hint->lh.owner == NULL); + info = cryptcompress_inode_data(inode); + + mutex_unlock(&info->checkin_mutex); + reiser4_txn_restart_current(); + balance_dirty_pages_ratelimited_nr(inode->i_mapping, nr_dirtied); + mutex_lock(&info->checkin_mutex); + return 0; +} + +/* set zeroes to the page cluster, proceed it, and maybe, try to capture + its pages */ +static int write_hole(struct inode *inode, struct cluster_handle * clust, + loff_t file_off, loff_t to_file) +{ + int result = 0; + unsigned cl_off, cl_count = 0; + unsigned to_pg, pg_off; + struct reiser4_slide * win; + + assert("edward-190", clust != NULL); + assert("edward-1069", clust->win != NULL); + assert("edward-191", inode != NULL); + assert("edward-727", cryptcompress_inode_ok(inode)); + assert("edward-1171", clust->dstat != INVAL_DISK_CLUSTER); + assert("edward-1154", + ergo(clust->dstat != FAKE_DISK_CLUSTER, clust->reserved == 1)); + + win = clust->win; + + assert("edward-1070", win != NULL); + assert("edward-201", win->stat == HOLE_WINDOW); + assert("edward-192", cluster_ok(clust, inode)); + + if (win->off == 0 && win->count == inode_cluster_size(inode)) { + /* This part of the hole will be represented by "fake" + * logical cluster, i.e. which doesn't have appropriate + * disk cluster until someone modify this logical cluster + * and make it dirty. + * So go forward here.. + */ + move_update_window(inode, clust, file_off, to_file); + return 0; + } + cl_count = win->count; /* number of zeroes to write */ + cl_off = win->off; + pg_off = off_to_pgoff(win->off); + + while (cl_count) { + struct page *page; + page = clust->pages[off_to_pg(cl_off)]; + + assert("edward-284", page != NULL); + + to_pg = min((typeof(pg_off))PAGE_CACHE_SIZE - pg_off, cl_count); + lock_page(page); + zero_user(page, pg_off, to_pg); + SetPageUptodate(page); + set_page_dirty_notag(page); + mark_page_accessed(page); + unlock_page(page); + + cl_off += to_pg; + cl_count -= to_pg; + pg_off = 0; + } + if (!win->delta) { + /* only zeroes in this window, try to capture + */ + result = checkin_logical_cluster(clust, inode); + if (result) + return result; + put_hint_cluster(clust, inode, ZNODE_WRITE_LOCK); + result = balance_dirty_page_cluster(clust, + inode, file_off, to_file, + win_count_to_nrpages(win)); + } else + move_update_window(inode, clust, file_off, to_file); + return result; +} + +/* + The main disk search procedure for cryptcompress plugin, which + . scans all items of disk cluster with the lock mode @mode + . maybe reads each one (if @read) + . maybe makes its znode dirty (if write lock mode was specified) + + NOTE-EDWARD: Callers should handle the case when disk cluster + is incomplete (-EIO) +*/ +int find_disk_cluster(struct cluster_handle * clust, + struct inode *inode, int read, znode_lock_mode mode) +{ + flow_t f; + hint_t *hint; + int result = 0; + int was_grabbed; + ra_info_t ra_info; + file_plugin *fplug; + item_plugin *iplug; + struct tfm_cluster *tc; + struct cryptcompress_info * info; + + assert("edward-138", clust != NULL); + assert("edward-728", clust->hint != NULL); + assert("edward-226", reiser4_schedulable()); + assert("edward-137", inode != NULL); + assert("edward-729", cryptcompress_inode_ok(inode)); + + hint = clust->hint; + fplug = inode_file_plugin(inode); + was_grabbed = get_current_context()->grabbed_blocks; + info = cryptcompress_inode_data(inode); + tc = &clust->tc; + + assert("edward-462", !tfm_cluster_is_uptodate(tc)); + assert("edward-461", ergo(read, tfm_stream_is_set(tc, INPUT_STREAM))); + + dclust_init_extension(hint); + + /* set key of the first disk cluster item */ + fplug->flow_by_inode(inode, + (read ? (char __user *)tfm_stream_data(tc, INPUT_STREAM) : NULL), + 0 /* kernel space */ , + inode_scaled_cluster_size(inode), + clust_to_off(clust->index, inode), READ_OP, &f); + if (mode == ZNODE_WRITE_LOCK) { + /* reserve for flush to make dirty all the leaf nodes + which contain disk cluster */ + result = + reiser4_grab_space_force(estimate_dirty_cluster(inode), + BA_CAN_COMMIT); + if (result) + goto out; + } + + ra_info.key_to_stop = f.key; + set_key_offset(&ra_info.key_to_stop, get_key_offset(reiser4_max_key())); + + while (f.length) { + result = find_cluster_item(hint, &f.key, mode, + NULL, FIND_EXACT, + (mode == ZNODE_WRITE_LOCK ? + CBK_FOR_INSERT : 0)); + switch (result) { + case CBK_COORD_NOTFOUND: + result = 0; + if (inode_scaled_offset + (inode, clust_to_off(clust->index, inode)) == + get_key_offset(&f.key)) { + /* first item not found, this is treated + as disk cluster is absent */ + clust->dstat = FAKE_DISK_CLUSTER; + goto out; + } + /* we are outside the cluster, stop search here */ + assert("edward-146", + f.length != inode_scaled_cluster_size(inode)); + goto ok; + case CBK_COORD_FOUND: + assert("edward-148", + hint->ext_coord.coord.between == AT_UNIT); + assert("edward-460", + hint->ext_coord.coord.unit_pos == 0); + + coord_clear_iplug(&hint->ext_coord.coord); + result = zload_ra(hint->ext_coord.coord.node, &ra_info); + if (unlikely(result)) + goto out; + iplug = item_plugin_by_coord(&hint->ext_coord.coord); + assert("edward-147", + item_id_by_coord(&hint->ext_coord.coord) == + CTAIL_ID); + + result = iplug->s.file.read(NULL, &f, hint); + if (result) { + zrelse(hint->ext_coord.coord.node); + goto out; + } + if (mode == ZNODE_WRITE_LOCK) { + /* Don't make dirty more nodes then it was + estimated (see comments before + estimate_dirty_cluster). Missed nodes will be + read up in flush time if they are evicted from + memory */ + if (dclust_get_extension_ncount(hint) <= + estimate_dirty_cluster(inode)) + znode_make_dirty(hint->ext_coord.coord.node); + + znode_set_convertible(hint->ext_coord.coord. + node); + } + zrelse(hint->ext_coord.coord.node); + break; + default: + goto out; + } + } + ok: + /* at least one item was found */ + /* NOTE-EDWARD: Callers should handle the case + when disk cluster is incomplete (-EIO) */ + tc->len = inode_scaled_cluster_size(inode) - f.length; + tc->lsize = lbytes(clust->index, inode); + assert("edward-1196", tc->len > 0); + assert("edward-1406", tc->lsize > 0); + + if (hint_is_unprepped_dclust(clust->hint)) { + clust->dstat = UNPR_DISK_CLUSTER; + } else if (clust->index == info->trunc_index) { + clust->dstat = TRNC_DISK_CLUSTER; + } else { + clust->dstat = PREP_DISK_CLUSTER; + dclust_set_extension_dsize(clust->hint, tc->len); + } + out: + assert("edward-1339", + get_current_context()->grabbed_blocks >= was_grabbed); + grabbed2free(get_current_context(), + get_current_super_private(), + get_current_context()->grabbed_blocks - was_grabbed); + return result; +} + +int get_disk_cluster_locked(struct cluster_handle * clust, struct inode *inode, + znode_lock_mode lock_mode) +{ + reiser4_key key; + ra_info_t ra_info; + + assert("edward-730", reiser4_schedulable()); + assert("edward-731", clust != NULL); + assert("edward-732", inode != NULL); + + if (hint_is_valid(clust->hint)) { + assert("edward-1293", clust->dstat != INVAL_DISK_CLUSTER); + assert("edward-1294", + znode_is_write_locked(clust->hint->lh.node)); + /* already have a valid locked position */ + return (clust->dstat == + FAKE_DISK_CLUSTER ? CBK_COORD_NOTFOUND : + CBK_COORD_FOUND); + } + key_by_inode_cryptcompress(inode, clust_to_off(clust->index, inode), + &key); + ra_info.key_to_stop = key; + set_key_offset(&ra_info.key_to_stop, get_key_offset(reiser4_max_key())); + + return find_cluster_item(clust->hint, &key, lock_mode, NULL, FIND_EXACT, + CBK_FOR_INSERT); +} + +/* Read needed cluster pages before modifying. + If success, @clust->hint contains locked position in the tree. + Also: + . find and set disk cluster state + . make disk cluster dirty if its state is not FAKE_DISK_CLUSTER. +*/ +static int read_some_cluster_pages(struct inode * inode, + struct cluster_handle * clust) +{ + int i; + int result = 0; + item_plugin *iplug; + struct reiser4_slide * win = clust->win; + znode_lock_mode mode = ZNODE_WRITE_LOCK; + + iplug = item_plugin_by_id(CTAIL_ID); + + assert("edward-924", !tfm_cluster_is_uptodate(&clust->tc)); + +#if REISER4_DEBUG + if (clust->nr_pages == 0) { + /* start write hole from fake disk cluster */ + assert("edward-1117", win != NULL); + assert("edward-1118", win->stat == HOLE_WINDOW); + assert("edward-1119", new_logical_cluster(clust, inode)); + } +#endif + if (new_logical_cluster(clust, inode)) { + /* + new page cluster is about to be written, nothing to read, + */ + assert("edward-734", reiser4_schedulable()); + assert("edward-735", clust->hint->lh.owner == NULL); + + if (clust->nr_pages) { + int off; + struct page * pg; + assert("edward-1419", clust->pages != NULL); + pg = clust->pages[clust->nr_pages - 1]; + assert("edward-1420", pg != NULL); + off = off_to_pgoff(win->off+win->count+win->delta); + if (off) { + lock_page(pg); + zero_user_segment(pg, off, PAGE_CACHE_SIZE); + unlock_page(pg); + } + } + clust->dstat = FAKE_DISK_CLUSTER; + return 0; + } + /* + Here we should search for disk cluster to figure out its real state. + Also there is one more important reason to do disk search: we need + to make disk cluster _dirty_ if it exists + */ + + /* if windows is specified, read the only pages + that will be modified partially */ + + for (i = 0; i < clust->nr_pages; i++) { + struct page *pg = clust->pages[i]; + + lock_page(pg); + if (PageUptodate(pg)) { + unlock_page(pg); + continue; + } + unlock_page(pg); + + if (win && + i >= size_in_pages(win->off) && + i < off_to_pg(win->off + win->count + win->delta)) + /* page will be completely overwritten */ + continue; + + if (win && (i == clust->nr_pages - 1) && + /* the last page is + partially modified, + not uptodate .. */ + (size_in_pages(i_size_read(inode)) <= pg->index)) { + /* .. and appended, + so set zeroes to the rest */ + int offset; + lock_page(pg); + assert("edward-1260", + size_in_pages(win->off + win->count + + win->delta) - 1 == i); + + offset = + off_to_pgoff(win->off + win->count + win->delta); + zero_user_segment(pg, offset, PAGE_CACHE_SIZE); + unlock_page(pg); + /* still not uptodate */ + break; + } + lock_page(pg); + result = do_readpage_ctail(inode, clust, pg, mode); + + assert("edward-1526", ergo(!result, PageUptodate(pg))); + unlock_page(pg); + if (result) { + warning("edward-219", "do_readpage_ctail failed"); + goto out; + } + } + if (!tfm_cluster_is_uptodate(&clust->tc)) { + /* disk cluster unclaimed, but we need to make its znodes dirty + * to make flush update convert its content + */ + result = find_disk_cluster(clust, inode, + 0 /* do not read items */, + mode); + } + out: + tfm_cluster_clr_uptodate(&clust->tc); + return result; +} + +static int should_create_unprepped_cluster(struct cluster_handle * clust, + struct inode * inode) +{ + assert("edward-737", clust != NULL); + + switch (clust->dstat) { + case PREP_DISK_CLUSTER: + case UNPR_DISK_CLUSTER: + return 0; + case FAKE_DISK_CLUSTER: + if (clust->win && + clust->win->stat == HOLE_WINDOW && clust->nr_pages == 0) { + assert("edward-1172", + new_logical_cluster(clust, inode)); + return 0; + } + return 1; + default: + impossible("edward-1173", "bad disk cluster state"); + return 0; + } +} + +static int cryptcompress_make_unprepped_cluster(struct cluster_handle * clust, + struct inode *inode) +{ + int result; + + assert("edward-1123", reiser4_schedulable()); + assert("edward-737", clust != NULL); + assert("edward-738", inode != NULL); + assert("edward-739", cryptcompress_inode_ok(inode)); + assert("edward-1053", clust->hint != NULL); + + if (!should_create_unprepped_cluster(clust, inode)) { + if (clust->reserved) { + cluster_reserved2free(estimate_insert_cluster(inode)); +#if REISER4_DEBUG + assert("edward-1267", + clust->reserved_unprepped == + estimate_insert_cluster(inode)); + clust->reserved_unprepped -= + estimate_insert_cluster(inode); +#endif + } + return 0; + } + assert("edward-1268", clust->reserved); + cluster_reserved2grabbed(estimate_insert_cluster(inode)); +#if REISER4_DEBUG + assert("edward-1441", + clust->reserved_unprepped == estimate_insert_cluster(inode)); + clust->reserved_unprepped -= estimate_insert_cluster(inode); +#endif + result = ctail_insert_unprepped_cluster(clust, inode); + if (result) + return result; + + inode_add_bytes(inode, inode_cluster_size(inode)); + + assert("edward-743", cryptcompress_inode_ok(inode)); + assert("edward-744", znode_is_write_locked(clust->hint->lh.node)); + + clust->dstat = UNPR_DISK_CLUSTER; + return 0; +} + +/* . Grab page cluster for read, write, setattr, etc. operations; + * . Truncate its complete pages, if needed; + */ +int prepare_page_cluster(struct inode * inode, struct cluster_handle * clust, + rw_op rw) +{ + assert("edward-177", inode != NULL); + assert("edward-741", cryptcompress_inode_ok(inode)); + assert("edward-740", clust->pages != NULL); + + set_cluster_nrpages(clust, inode); + reset_cluster_pgset(clust, cluster_nrpages(inode)); + return grab_page_cluster(inode, clust, rw); +} + +/* Truncate complete page cluster of index @index. + * This is called by ->kill_hook() method of item + * plugin when deleting a disk cluster of such index. + */ +void truncate_complete_page_cluster(struct inode *inode, cloff_t index, + int even_cows) +{ + int found; + int nr_pages; + jnode *node; + struct page *pages[MAX_CLUSTER_NRPAGES]; + + node = jlookup(current_tree, get_inode_oid(inode), + clust_to_pg(index, inode)); + nr_pages = size_in_pages(lbytes(index, inode)); + assert("edward-1483", nr_pages != 0); + if (!node) + goto truncate; + found = find_get_pages(inode->i_mapping, + clust_to_pg(index, inode), + cluster_nrpages(inode), pages); + if (!found) { + assert("edward-1484", jnode_truncate_ok(inode, index)); + return; + } + lock_cluster(node); + + if (reiser4_inode_get_flag(inode, REISER4_FILE_CONV_IN_PROGRESS) + && index == 0) + /* converting to unix_file is in progress */ + JF_CLR(node, JNODE_CLUSTER_PAGE); + if (JF_ISSET(node, JNODE_DIRTY)) { + /* + * @nr_pages were checked in, but not yet checked out - + * we need to release them. (also there can be pages + * attached to page cache by read(), etc. - don't take + * them into account). + */ + assert("edward-1198", found >= nr_pages); + + /* free disk space grabbed for disk cluster converting */ + cluster_reserved2grabbed(estimate_update_cluster(inode)); + grabbed2free(get_current_context(), + get_current_super_private(), + estimate_update_cluster(inode)); + __put_page_cluster(0, nr_pages, pages, inode); + + /* This will clear dirty bit, uncapture and unlock jnode */ + unlock_cluster_uncapture(node); + } else + unlock_cluster(node); + jput(node); /* jlookup */ + put_found_pages(pages, found); /* find_get_pages */ + truncate: + if (reiser4_inode_get_flag(inode, REISER4_FILE_CONV_IN_PROGRESS) && + index == 0) + return; + truncate_page_cluster_range(inode, pages, index, 0, + cluster_nrpages(inode), + even_cows); + assert("edward-1201", + ergo(!reiser4_inode_get_flag(inode, + REISER4_FILE_CONV_IN_PROGRESS), + jnode_truncate_ok(inode, index))); + return; +} + +/* + * Set cluster handle @clust of a logical cluster before + * modifications which are supposed to be committed. + * + * . grab cluster pages; + * . reserve disk space; + * . maybe read pages from disk and set the disk cluster dirty; + * . maybe write hole and check in (partially zeroed) logical cluster; + * . create 'unprepped' disk cluster for new or fake logical one. + */ +static int prepare_logical_cluster(struct inode *inode, + loff_t file_off, /* write position + in the file */ + loff_t to_file, /* bytes of users data + to write to the file */ + struct cluster_handle * clust, + logical_cluster_op op) +{ + int result = 0; + struct reiser4_slide * win = clust->win; + + reset_cluster_params(clust); + cluster_set_tfm_act(&clust->tc, TFMA_READ); +#if REISER4_DEBUG + clust->ctx = get_current_context(); +#endif + assert("edward-1190", op != LC_INVAL); + + clust->op = op; + + result = prepare_page_cluster(inode, clust, WRITE_OP); + if (result) + return result; + assert("edward-1447", + ergo(clust->nr_pages != 0, jprivate(clust->pages[0]))); + assert("edward-1448", + ergo(clust->nr_pages != 0, + jnode_is_cluster_page(jprivate(clust->pages[0])))); + + result = reserve4cluster(inode, clust); + if (result) + goto err1; + result = read_some_cluster_pages(inode, clust); + if (result) { + free_reserved4cluster(inode, + clust, + estimate_update_cluster(inode) + + estimate_insert_cluster(inode)); + goto err1; + } + assert("edward-1124", clust->dstat != INVAL_DISK_CLUSTER); + + result = cryptcompress_make_unprepped_cluster(clust, inode); + if (result) + goto err2; + if (win && win->stat == HOLE_WINDOW) { + result = write_hole(inode, clust, file_off, to_file); + if (result) + goto err2; + } + return 0; + err2: + free_reserved4cluster(inode, clust, + estimate_update_cluster(inode)); + err1: + put_page_cluster(clust, inode, WRITE_OP); + assert("edward-1125", result == -ENOSPC); + return result; +} + +/* set window by two offsets */ +static void set_window(struct cluster_handle * clust, + struct reiser4_slide * win, struct inode *inode, + loff_t o1, loff_t o2) +{ + assert("edward-295", clust != NULL); + assert("edward-296", inode != NULL); + assert("edward-1071", win != NULL); + assert("edward-297", o1 <= o2); + + clust->index = off_to_clust(o1, inode); + + win->off = off_to_cloff(o1, inode); + win->count = min((loff_t)(inode_cluster_size(inode) - win->off), + o2 - o1); + win->delta = 0; + + clust->win = win; +} + +static int set_cluster_by_window(struct inode *inode, + struct cluster_handle * clust, + struct reiser4_slide * win, size_t length, + loff_t file_off) +{ + int result; + + assert("edward-197", clust != NULL); + assert("edward-1072", win != NULL); + assert("edward-198", inode != NULL); + + result = alloc_cluster_pgset(clust, cluster_nrpages(inode)); + if (result) + return result; + + if (file_off > i_size_read(inode)) { + /* Uhmm, hole in cryptcompress file... */ + loff_t hole_size; + hole_size = file_off - inode->i_size; + + set_window(clust, win, inode, inode->i_size, file_off); + win->stat = HOLE_WINDOW; + if (win->off + hole_size < inode_cluster_size(inode)) + /* there is also user's data to append to the hole */ + win->delta = min(inode_cluster_size(inode) - + (win->off + win->count), length); + return 0; + } + set_window(clust, win, inode, file_off, file_off + length); + win->stat = DATA_WINDOW; + return 0; +} + +int set_cluster_by_page(struct cluster_handle * clust, struct page * page, + int count) +{ + int result = 0; + int (*setting_actor)(struct cluster_handle * clust, int count); + + assert("edward-1358", clust != NULL); + assert("edward-1359", page != NULL); + assert("edward-1360", page->mapping != NULL); + assert("edward-1361", page->mapping->host != NULL); + + setting_actor = + (clust->pages ? reset_cluster_pgset : alloc_cluster_pgset); + result = setting_actor(clust, count); + clust->index = pg_to_clust(page->index, page->mapping->host); + return result; +} + +/* reset all the params that not get updated */ +void reset_cluster_params(struct cluster_handle * clust) +{ + assert("edward-197", clust != NULL); + + clust->dstat = INVAL_DISK_CLUSTER; + clust->tc.uptodate = 0; + clust->tc.len = 0; +} + +/* the heart of write_cryptcompress */ +static loff_t do_write_cryptcompress(struct file *file, struct inode *inode, + const char __user *buf, size_t to_write, + loff_t pos, struct psched_context *cont) +{ + int i; + hint_t *hint; + int result = 0; + size_t count; + struct reiser4_slide win; + struct cluster_handle clust; + struct cryptcompress_info * info; + + assert("edward-154", buf != NULL); + assert("edward-161", reiser4_schedulable()); + assert("edward-748", cryptcompress_inode_ok(inode)); + assert("edward-159", current_blocksize == PAGE_CACHE_SIZE); + assert("edward-1274", get_current_context()->grabbed_blocks == 0); + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + + result = load_file_hint(file, hint); + if (result) { + kfree(hint); + return result; + } + count = to_write; + + reiser4_slide_init(&win); + cluster_init_read(&clust, &win); + clust.hint = hint; + info = cryptcompress_inode_data(inode); + + mutex_lock(&info->checkin_mutex); + + result = set_cluster_by_window(inode, &clust, &win, to_write, pos); + if (result) + goto out; + + if (next_window_stat(&win) == HOLE_WINDOW) { + /* write hole in this iteration + separated from the loop below */ + result = write_pschedule_hook(file, inode, + pos, + &clust, + cont); + if (result) + goto out; + result = prepare_logical_cluster(inode, pos, count, &clust, + LC_APPOV); + if (result) + goto out; + } + do { + const char __user * src; + unsigned page_off, to_page; + + assert("edward-750", reiser4_schedulable()); + + result = write_pschedule_hook(file, inode, + pos + to_write - count, + &clust, + cont); + if (result) + goto out; + if (cont->state == PSCHED_ASSIGNED_NEW) + /* done_lh was called in write_pschedule_hook */ + goto out_no_longterm_lock; + + result = prepare_logical_cluster(inode, pos, count, &clust, + LC_APPOV); + if (result) + goto out; + + assert("edward-751", cryptcompress_inode_ok(inode)); + assert("edward-204", win.stat == DATA_WINDOW); + assert("edward-1288", hint_is_valid(clust.hint)); + assert("edward-752", + znode_is_write_locked(hint->ext_coord.coord.node)); + put_hint_cluster(&clust, inode, ZNODE_WRITE_LOCK); + + /* set write position in page */ + page_off = off_to_pgoff(win.off); + + /* copy user's data to cluster pages */ + for (i = off_to_pg(win.off), src = buf; + i < size_in_pages(win.off + win.count); + i++, src += to_page) { + to_page = __mbp(win.off + win.count, i) - page_off; + assert("edward-1039", + page_off + to_page <= PAGE_CACHE_SIZE); + assert("edward-287", clust.pages[i] != NULL); + + fault_in_pages_readable(src, to_page); + + lock_page(clust.pages[i]); + result = + __copy_from_user((char *)kmap(clust.pages[i]) + + page_off, src, to_page); + kunmap(clust.pages[i]); + if (unlikely(result)) { + unlock_page(clust.pages[i]); + result = -EFAULT; + goto err2; + } + SetPageUptodate(clust.pages[i]); + set_page_dirty_notag(clust.pages[i]); + flush_dcache_page(clust.pages[i]); + mark_page_accessed(clust.pages[i]); + unlock_page(clust.pages[i]); + page_off = 0; + } + assert("edward-753", cryptcompress_inode_ok(inode)); + + result = checkin_logical_cluster(&clust, inode); + if (result) + goto err2; + + buf += win.count; + count -= win.count; + + result = balance_dirty_page_cluster(&clust, inode, 0, count, + win_count_to_nrpages(&win)); + if (result) + goto err1; + assert("edward-755", hint->lh.owner == NULL); + reset_cluster_params(&clust); + continue; + err2: + put_page_cluster(&clust, inode, WRITE_OP); + err1: + if (clust.reserved) + free_reserved4cluster(inode, + &clust, + estimate_update_cluster(inode)); + break; + } while (count); + out: + done_lh(&hint->lh); + save_file_hint(file, hint); + out_no_longterm_lock: + mutex_unlock(&info->checkin_mutex); + kfree(hint); + put_cluster_handle(&clust); + assert("edward-195", + ergo((to_write == count), + (result < 0 || cont->state == PSCHED_ASSIGNED_NEW))); + return (to_write - count) ? (to_write - count) : result; +} + +/** + * plugin->write() + * @file: file to write to + * @buf: address of user-space buffer + * @read_amount: number of bytes to write + * @off: position in file to write to + */ +ssize_t write_cryptcompress(struct file *file, const char __user *buf, + size_t count, loff_t *off, + struct psched_context *cont) +{ + ssize_t result; + struct inode *inode; + reiser4_context *ctx; + loff_t pos = *off; + struct cryptcompress_info *info; + + assert("edward-1449", cont->state == PSCHED_INVAL_STATE); + + inode = file->f_dentry->d_inode; + assert("edward-196", cryptcompress_inode_ok(inode)); + + info = cryptcompress_inode_data(inode); + ctx = get_current_context(); + + result = generic_write_checks(file, &pos, &count, 0); + if (unlikely(result != 0)) { + context_set_commit_async(ctx); + return result; + } + if (unlikely(count == 0)) + return 0; + result = file_remove_suid(file); + if (unlikely(result != 0)) { + context_set_commit_async(ctx); + return result; + } + /* remove_suid might create a transaction */ + reiser4_txn_restart(ctx); + + result = do_write_cryptcompress(file, inode, buf, count, pos, cont); + + if (unlikely(result < 0)) { + context_set_commit_async(ctx); + return result; + } + /* update position in a file */ + *off = pos + result; + return result; +} + +/* plugin->readpages */ +int readpages_cryptcompress(struct file *file, struct address_space *mapping, + struct list_head *pages, unsigned nr_pages) +{ + reiser4_context * ctx; + int ret; + + ctx = reiser4_init_context(mapping->host->i_sb); + if (IS_ERR(ctx)) { + ret = PTR_ERR(ctx); + goto err; + } + /* cryptcompress file can be built of ctail items only */ + ret = readpages_ctail(file, mapping, pages); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + if (ret) { +err: + put_pages_list(pages); + } + return ret; +} + +static reiser4_block_nr cryptcompress_estimate_read(struct inode *inode) +{ + /* reserve one block to update stat data item */ + assert("edward-1193", + inode_file_plugin(inode)->estimate.update == + estimate_update_common); + return estimate_update_common(inode); +} + +/** + * plugin->read + * @file: file to read from + * @buf: address of user-space buffer + * @read_amount: number of bytes to read + * @off: position in file to read from + */ +ssize_t read_cryptcompress(struct file * file, char __user *buf, size_t size, + loff_t * off) +{ + ssize_t result; + struct inode *inode; + reiser4_context *ctx; + struct cryptcompress_info *info; + reiser4_block_nr needed; + + inode = file->f_dentry->d_inode; + assert("edward-1194", !reiser4_inode_get_flag(inode, REISER4_NO_SD)); + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + info = cryptcompress_inode_data(inode); + needed = cryptcompress_estimate_read(inode); + + result = reiser4_grab_space(needed, BA_CAN_COMMIT); + if (result != 0) { + reiser4_exit_context(ctx); + return result; + } + result = do_sync_read(file, buf, size, off); + + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + + return result; +} + +/* Look for a disk cluster and keep lookup result in @found. + * If @index > 0, then find disk cluster of the index (@index - 1); + * If @index == 0, then find the rightmost disk cluster. + * Keep incremented index of the found disk cluster in @found. + * @found == 0 means that disk cluster was not found (in the last + * case (@index == 0) it means that file doesn't have disk clusters). + */ +static int lookup_disk_cluster(struct inode *inode, cloff_t * found, + cloff_t index) +{ + int result; + reiser4_key key; + loff_t offset; + hint_t *hint; + lock_handle *lh; + lookup_bias bias; + coord_t *coord; + item_plugin *iplug; + + assert("edward-1131", inode != NULL); + assert("edward-95", cryptcompress_inode_ok(inode)); + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + hint_init_zero(hint); + lh = &hint->lh; + + bias = (index ? FIND_EXACT : FIND_MAX_NOT_MORE_THAN); + offset = + (index ? clust_to_off(index, inode) - + 1 : get_key_offset(reiser4_max_key())); + + key_by_inode_cryptcompress(inode, offset, &key); + + /* find the last item of this object */ + result = + find_cluster_item(hint, &key, ZNODE_READ_LOCK, NULL /* ra_info */, + bias, 0); + if (cbk_errored(result)) { + done_lh(lh); + kfree(hint); + return result; + } + if (result == CBK_COORD_NOTFOUND) { + /* no real disk clusters */ + done_lh(lh); + kfree(hint); + *found = 0; + return 0; + } + /* disk cluster is found */ + coord = &hint->ext_coord.coord; + coord_clear_iplug(coord); + result = zload(coord->node); + if (unlikely(result)) { + done_lh(lh); + kfree(hint); + return result; + } + iplug = item_plugin_by_coord(coord); + assert("edward-277", iplug == item_plugin_by_id(CTAIL_ID)); + assert("edward-1202", ctail_ok(coord)); + + item_key_by_coord(coord, &key); + *found = off_to_clust(get_key_offset(&key), inode) + 1; + + assert("edward-1132", ergo(index, index == *found)); + + zrelse(coord->node); + done_lh(lh); + kfree(hint); + return 0; +} + +static int find_fake_appended(struct inode *inode, cloff_t * index) +{ + return lookup_disk_cluster(inode, index, + 0 /* find last real one */ ); +} + +/* Set left coord when unit is not found after node_lookup() + This takes into account that there can be holes in a sequence + of disk clusters */ + +static void adjust_left_coord(coord_t * left_coord) +{ + switch (left_coord->between) { + case AFTER_UNIT: + left_coord->between = AFTER_ITEM; + case AFTER_ITEM: + case BEFORE_UNIT: + break; + default: + impossible("edward-1204", "bad left coord to cut"); + } + return; +} + +#define CRC_CUT_TREE_MIN_ITERATIONS 64 + +/* plugin->cut_tree_worker */ +int cut_tree_worker_cryptcompress(tap_t * tap, const reiser4_key * from_key, + const reiser4_key * to_key, + reiser4_key * smallest_removed, + struct inode *object, int truncate, + int *progress) +{ + lock_handle next_node_lock; + coord_t left_coord; + int result; + + assert("edward-1158", tap->coord->node != NULL); + assert("edward-1159", znode_is_write_locked(tap->coord->node)); + assert("edward-1160", znode_get_level(tap->coord->node) == LEAF_LEVEL); + + *progress = 0; + init_lh(&next_node_lock); + + while (1) { + znode *node; /* node from which items are cut */ + node_plugin *nplug; /* node plugin for @node */ + + node = tap->coord->node; + + /* Move next_node_lock to the next node on the left. */ + result = + reiser4_get_left_neighbor(&next_node_lock, node, + ZNODE_WRITE_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (result != 0 && result != -E_NO_NEIGHBOR) + break; + /* FIXME-EDWARD: Check can we delete the node as a whole. */ + result = reiser4_tap_load(tap); + if (result) + return result; + + /* Prepare the second (right) point for cut_node() */ + if (*progress) + coord_init_last_unit(tap->coord, node); + + else if (item_plugin_by_coord(tap->coord)->b.lookup == NULL) + /* set rightmost unit for the items without lookup method */ + tap->coord->unit_pos = coord_last_unit_pos(tap->coord); + + nplug = node->nplug; + + assert("edward-1161", nplug); + assert("edward-1162", nplug->lookup); + + /* left_coord is leftmost unit cut from @node */ + result = nplug->lookup(node, from_key, FIND_EXACT, &left_coord); + + if (IS_CBKERR(result)) + break; + + if (result == CBK_COORD_NOTFOUND) + adjust_left_coord(&left_coord); + + /* adjust coordinates so that they are set to existing units */ + if (coord_set_to_right(&left_coord) + || coord_set_to_left(tap->coord)) { + result = 0; + break; + } + + if (coord_compare(&left_coord, tap->coord) == + COORD_CMP_ON_RIGHT) { + /* keys from @from_key to @to_key are not in the tree */ + result = 0; + break; + } + + /* cut data from one node */ + *smallest_removed = *reiser4_min_key(); + result = kill_node_content(&left_coord, + tap->coord, + from_key, + to_key, + smallest_removed, + next_node_lock.node, + object, truncate); + reiser4_tap_relse(tap); + + if (result) + break; + + ++(*progress); + + /* Check whether all items with keys >= from_key were removed + * from the tree. */ + if (keyle(smallest_removed, from_key)) + /* result = 0; */ + break; + + if (next_node_lock.node == NULL) + break; + + result = reiser4_tap_move(tap, &next_node_lock); + done_lh(&next_node_lock); + if (result) + break; + + /* Break long cut_tree operation (deletion of a large file) if + * atom requires commit. */ + if (*progress > CRC_CUT_TREE_MIN_ITERATIONS + && current_atom_should_commit()) { + result = -E_REPEAT; + break; + } + } + done_lh(&next_node_lock); + return result; +} + +/* Append or expand hole in two steps: + * 1) set zeroes to the rightmost page of the rightmost non-fake + * logical cluster; + * 2) expand hole via fake logical clusters (just increase i_size) + */ +static int cryptcompress_append_hole(struct inode *inode /* with old size */, + loff_t new_size) +{ + int result = 0; + hint_t *hint; + lock_handle *lh; + loff_t hole_size; + int nr_zeroes; + struct reiser4_slide win; + struct cluster_handle clust; + + assert("edward-1133", inode->i_size < new_size); + assert("edward-1134", reiser4_schedulable()); + assert("edward-1135", cryptcompress_inode_ok(inode)); + assert("edward-1136", current_blocksize == PAGE_CACHE_SIZE); + assert("edward-1333", off_to_cloff(inode->i_size, inode) != 0); + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + hint_init_zero(hint); + lh = &hint->lh; + + reiser4_slide_init(&win); + cluster_init_read(&clust, &win); + clust.hint = hint; + + result = alloc_cluster_pgset(&clust, cluster_nrpages(inode)); + if (result) + goto out; + if (off_to_cloff(inode->i_size, inode) == 0) + goto append_fake; + hole_size = new_size - inode->i_size; + nr_zeroes = + inode_cluster_size(inode) - off_to_cloff(inode->i_size, inode); + if (hole_size < nr_zeroes) + nr_zeroes = hole_size; + set_window(&clust, &win, inode, inode->i_size, + inode->i_size + nr_zeroes); + win.stat = HOLE_WINDOW; + + assert("edward-1137", + clust.index == off_to_clust(inode->i_size, inode)); + + result = prepare_logical_cluster(inode, 0, 0, &clust, LC_APPOV); + + assert("edward-1271", !result || result == -ENOSPC); + if (result) + goto out; + assert("edward-1139", + clust.dstat == PREP_DISK_CLUSTER || + clust.dstat == UNPR_DISK_CLUSTER); + + assert("edward-1431", hole_size >= nr_zeroes); + if (hole_size == nr_zeroes) + /* nothing to append anymore */ + goto out; + append_fake: + INODE_SET_SIZE(inode, new_size); + out: + done_lh(lh); + kfree(hint); + put_cluster_handle(&clust); + return result; +} + +static int update_cryptcompress_size(struct inode *inode, loff_t new_size, + int update_sd) +{ + return (new_size & ((loff_t) (inode_cluster_size(inode)) - 1) + ? 0 : reiser4_update_file_size(inode, new_size, update_sd)); +} + +/* Prune cryptcompress file in two steps: + * 1) cut all nominated logical clusters except the leftmost one which + * is to be partially truncated. Note, that there can be "holes" + * represented by fake logical clusters. + * 2) set zeroes and capture leftmost partially truncated logical + * cluster, if it is not fake; otherwise prune fake logical cluster + * (just decrease i_size). + */ +static int prune_cryptcompress(struct inode *inode, loff_t new_size, + int update_sd, cloff_t aidx) +{ + int result = 0; + unsigned nr_zeroes; + loff_t to_prune; + loff_t old_size; + cloff_t ridx; + + hint_t *hint; + lock_handle *lh; + struct reiser4_slide win; + struct cluster_handle clust; + + assert("edward-1140", inode->i_size >= new_size); + assert("edward-1141", reiser4_schedulable()); + assert("edward-1142", cryptcompress_inode_ok(inode)); + assert("edward-1143", current_blocksize == PAGE_CACHE_SIZE); + + old_size = inode->i_size; + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + hint_init_zero(hint); + lh = &hint->lh; + + reiser4_slide_init(&win); + cluster_init_read(&clust, &win); + clust.hint = hint; + + /* calculate index of the rightmost logical cluster + that will be completely truncated */ + ridx = size_in_lc(new_size, inode); + + /* truncate all disk clusters starting from @ridx */ + assert("edward-1174", ridx <= aidx); + old_size = inode->i_size; + if (ridx != aidx) { + struct cryptcompress_info * info; + info = cryptcompress_inode_data(inode); + result = cut_file_items(inode, + clust_to_off(ridx, inode), + update_sd, + clust_to_off(aidx, inode), + update_cryptcompress_size); + info->trunc_index = ULONG_MAX; + if (result) + goto out; + } + /* + * there can be pages of fake logical clusters, truncate them + */ + truncate_inode_pages(inode->i_mapping, clust_to_off(ridx, inode)); + assert("edward-1524", + pages_truncate_ok(inode, clust_to_pg(ridx, inode))); + /* + * now perform partial truncate of last logical cluster + */ + if (!off_to_cloff(new_size, inode)) { + /* no partial truncate is needed */ + assert("edward-1145", inode->i_size == new_size); + goto truncate_fake; + } + assert("edward-1146", new_size < inode->i_size); + + to_prune = inode->i_size - new_size; + + /* check if the last logical cluster is fake */ + result = lookup_disk_cluster(inode, &aidx, ridx); + if (result) + goto out; + if (!aidx) + /* yup, this is fake one */ + goto truncate_fake; + + assert("edward-1148", aidx == ridx); + + /* do partial truncate of the last page cluster, + and try to capture this one */ + result = alloc_cluster_pgset(&clust, cluster_nrpages(inode)); + if (result) + goto out; + nr_zeroes = (off_to_pgoff(new_size) ? + PAGE_CACHE_SIZE - off_to_pgoff(new_size) : 0); + set_window(&clust, &win, inode, new_size, new_size + nr_zeroes); + win.stat = HOLE_WINDOW; + + assert("edward-1149", clust.index == ridx - 1); + + result = prepare_logical_cluster(inode, 0, 0, &clust, LC_TRUNC); + if (result) + goto out; + assert("edward-1151", + clust.dstat == PREP_DISK_CLUSTER || + clust.dstat == UNPR_DISK_CLUSTER); + + assert("edward-1191", inode->i_size == new_size); + assert("edward-1206", body_truncate_ok(inode, ridx)); + truncate_fake: + /* drop all the pages that don't have jnodes (i.e. pages + which can not be truncated by cut_file_items() because + of holes represented by fake disk clusters) including + the pages of partially truncated cluster which was + released by prepare_logical_cluster() */ + INODE_SET_SIZE(inode, new_size); + truncate_inode_pages(inode->i_mapping, new_size); + out: + assert("edward-1334", !result || result == -ENOSPC); + assert("edward-1497", + pages_truncate_ok(inode, size_in_pages(new_size))); + + done_lh(lh); + kfree(hint); + put_cluster_handle(&clust); + return result; +} + +/* Prepare cryptcompress file for truncate: + * prune or append rightmost fake logical clusters (if any) + */ +static int start_truncate_fake(struct inode *inode, cloff_t aidx, + loff_t new_size, int update_sd) +{ + int result = 0; + int bytes; + + if (new_size > inode->i_size) { + /* append */ + if (inode->i_size < clust_to_off(aidx, inode)) + /* no fake bytes */ + return 0; + bytes = new_size - inode->i_size; + INODE_SET_SIZE(inode, inode->i_size + bytes); + } else { + /* prune */ + if (inode->i_size <= clust_to_off(aidx, inode)) + /* no fake bytes */ + return 0; + bytes = inode->i_size - + max(new_size, clust_to_off(aidx, inode)); + if (!bytes) + return 0; + INODE_SET_SIZE(inode, inode->i_size - bytes); + /* In the case of fake prune we need to drop page cluster. + There are only 2 cases for partially truncated page: + 1. If is is dirty, therefore it is anonymous + (was dirtied via mmap), and will be captured + later via ->capture(). + 2. If is clean, therefore it is filled by zeroes. + In both cases we don't need to make it dirty and + capture here. + */ + truncate_inode_pages(inode->i_mapping, inode->i_size); + } + if (update_sd) + result = update_sd_cryptcompress(inode); + return result; +} + +/** + * This is called in setattr_cryptcompress when it is used to truncate, + * and in delete_object_cryptcompress + */ +static int cryptcompress_truncate(struct inode *inode, /* old size */ + loff_t new_size, /* new size */ + int update_sd) +{ + int result; + cloff_t aidx; + + result = find_fake_appended(inode, &aidx); + if (result) + return result; + assert("edward-1208", + ergo(aidx > 0, inode->i_size > clust_to_off(aidx - 1, inode))); + + result = start_truncate_fake(inode, aidx, new_size, update_sd); + if (result) + return result; + if (inode->i_size == new_size) + /* nothing to truncate anymore */ + return 0; + result = (inode->i_size < new_size ? + cryptcompress_append_hole(inode, new_size) : + prune_cryptcompress(inode, new_size, update_sd, aidx)); + if (!result && update_sd) + result = update_sd_cryptcompress(inode); + return result; +} + +/** + * Capture a pager cluster. + * @clust must be set up by a caller. + */ +static int capture_page_cluster(struct cluster_handle * clust, + struct inode * inode) +{ + int result; + + assert("edward-1073", clust != NULL); + assert("edward-1074", inode != NULL); + assert("edward-1075", clust->dstat == INVAL_DISK_CLUSTER); + + result = prepare_logical_cluster(inode, 0, 0, clust, LC_APPOV); + if (result) + return result; + + set_cluster_pages_dirty(clust, inode); + result = checkin_logical_cluster(clust, inode); + put_hint_cluster(clust, inode, ZNODE_WRITE_LOCK); + if (unlikely(result)) + put_page_cluster(clust, inode, WRITE_OP); + return result; +} + +/* Starting from @index find tagged pages of the same page cluster. + * Clear the tag for each of them. Return number of found pages. + */ +static int find_anon_page_cluster(struct address_space * mapping, + pgoff_t * index, struct page ** pages) +{ + int i = 0; + int found; + spin_lock_irq(&mapping->tree_lock); + do { + /* looking for one page */ + found = radix_tree_gang_lookup_tag(&mapping->page_tree, + (void **)&pages[i], + *index, 1, + PAGECACHE_TAG_REISER4_MOVED); + if (!found) + break; + if (!same_page_cluster(pages[0], pages[i])) + break; + + /* found */ + page_cache_get(pages[i]); + *index = pages[i]->index + 1; + + radix_tree_tag_clear(&mapping->page_tree, + pages[i]->index, + PAGECACHE_TAG_REISER4_MOVED); + if (last_page_in_cluster(pages[i++])) + break; + } while (1); + spin_unlock_irq(&mapping->tree_lock); + return i; +} + +#define MAX_PAGES_TO_CAPTURE (1024) + +/* Capture anonymous page clusters */ +static int capture_anon_pages(struct address_space * mapping, pgoff_t * index, + int to_capture) +{ + int count = 0; + int found = 0; + int result = 0; + hint_t *hint; + lock_handle *lh; + struct inode * inode; + struct cluster_handle clust; + struct page * pages[MAX_CLUSTER_NRPAGES]; + + assert("edward-1127", mapping != NULL); + assert("edward-1128", mapping->host != NULL); + assert("edward-1440", mapping->host->i_mapping == mapping); + + inode = mapping->host; + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + hint_init_zero(hint); + lh = &hint->lh; + + cluster_init_read(&clust, NULL); + clust.hint = hint; + + result = alloc_cluster_pgset(&clust, cluster_nrpages(inode)); + if (result) + goto out; + + while (to_capture > 0) { + found = find_anon_page_cluster(mapping, index, pages); + if (!found) { + *index = (pgoff_t) - 1; + break; + } + move_cluster_forward(&clust, inode, pages[0]->index); + result = capture_page_cluster(&clust, inode); + + put_found_pages(pages, found); /* find_anon_page_cluster */ + if (result) + break; + to_capture -= clust.nr_pages; + count += clust.nr_pages; + } + if (result) { + warning("edward-1077", + "Capture failed (inode %llu, result=%i, captured=%d)\n", + (unsigned long long)get_inode_oid(inode), result, count); + } else { + assert("edward-1078", ergo(found > 0, count > 0)); + if (to_capture <= 0) + /* there may be left more pages */ + __mark_inode_dirty(inode, I_DIRTY_PAGES); + result = count; + } + out: + done_lh(lh); + kfree(hint); + put_cluster_handle(&clust); + return result; +} + +/* Returns true if inode's mapping has dirty pages + which do not belong to any atom */ +static int cryptcompress_inode_has_anon_pages(struct inode *inode) +{ + int result; + spin_lock_irq(&inode->i_mapping->tree_lock); + result = radix_tree_tagged(&inode->i_mapping->page_tree, + PAGECACHE_TAG_REISER4_MOVED); + spin_unlock_irq(&inode->i_mapping->tree_lock); + return result; +} + +/* plugin->writepages */ +int writepages_cryptcompress(struct address_space *mapping, + struct writeback_control *wbc) +{ + int result = 0; + long to_capture; + pgoff_t nrpages; + pgoff_t index = 0; + struct inode *inode; + struct cryptcompress_info *info; + + inode = mapping->host; + if (!cryptcompress_inode_has_anon_pages(inode)) + goto end; + info = cryptcompress_inode_data(inode); + nrpages = size_in_pages(i_size_read(inode)); + + if (wbc->sync_mode != WB_SYNC_ALL) + to_capture = min(wbc->nr_to_write, (long)MAX_PAGES_TO_CAPTURE); + else + to_capture = MAX_PAGES_TO_CAPTURE; + do { + reiser4_context *ctx; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) { + result = PTR_ERR(ctx); + break; + } + /* avoid recursive calls to ->sync_inodes */ + ctx->nobalance = 1; + + assert("edward-1079", + lock_stack_isclean(get_current_lock_stack())); + + reiser4_txn_restart_current(); + + if (get_current_context()->entd) { + if (mutex_trylock(&info->checkin_mutex) == 0) { + /* the mutex might be occupied by + entd caller */ + result = RETERR(-EBUSY); + reiser4_exit_context(ctx); + break; + } + } else + mutex_lock(&info->checkin_mutex); + + result = capture_anon_pages(inode->i_mapping, &index, + to_capture); + mutex_unlock(&info->checkin_mutex); + + if (result < 0) { + reiser4_exit_context(ctx); + break; + } + wbc->nr_to_write -= result; + if (wbc->sync_mode != WB_SYNC_ALL) { + reiser4_exit_context(ctx); + break; + } + result = txnmgr_force_commit_all(inode->i_sb, 0); + reiser4_exit_context(ctx); + } while (result >= 0 && index < nrpages); + + end: + if (is_in_reiser4_context()) { + if (get_current_context()->nr_captured >= CAPTURE_APAGE_BURST) { + /* there are already pages to flush, flush them out, + do not delay until end of reiser4_sync_inodes */ + reiser4_writeout(inode->i_sb, wbc); + get_current_context()->nr_captured = 0; + } + } + return result; +} + +/* plugin->ioctl */ +int ioctl_cryptcompress(struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg) +{ + return RETERR(-ENOSYS); +} + +/* plugin->mmap */ +int mmap_cryptcompress(struct file *file, struct vm_area_struct *vma) +{ + int result; + struct inode *inode; + reiser4_context *ctx; + + inode = file->f_dentry->d_inode; + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + /* + * generic_file_mmap will do update_atime. Grab space for stat data + * update. + */ + result = reiser4_grab_space_force + (inode_file_plugin(inode)->estimate.update(inode), + BA_CAN_COMMIT); + if (result) { + reiser4_exit_context(ctx); + return result; + } + result = generic_file_mmap(file, vma); + reiser4_exit_context(ctx); + return result; +} + +/* plugin->delete_object */ +int delete_object_cryptcompress(struct inode *inode) +{ + int result; + struct cryptcompress_info * info; + + assert("edward-429", inode->i_nlink == 0); + + reiser4_txn_restart_current(); + info = cryptcompress_inode_data(inode); + + mutex_lock(&info->checkin_mutex); + result = cryptcompress_truncate(inode, 0, 0); + mutex_unlock(&info->checkin_mutex); + + if (result) { + warning("edward-430", + "cannot truncate cryptcompress file %lli: %i", + (unsigned long long)get_inode_oid(inode), + result); + } + truncate_inode_pages(inode->i_mapping, 0); + assert("edward-1487", pages_truncate_ok(inode, 0)); + /* and remove stat data */ + return reiser4_delete_object_common(inode); +} + +/* + * plugin->setattr + * This implements actual truncate (see comments in reiser4/page_cache.c) + */ +int setattr_cryptcompress(struct dentry *dentry, struct iattr *attr) +{ + int result; + struct inode *inode; + struct cryptcompress_info * info; + + inode = dentry->d_inode; + info = cryptcompress_inode_data(inode); + + if (attr->ia_valid & ATTR_SIZE) { + if (i_size_read(inode) != attr->ia_size) { + reiser4_context *ctx; + loff_t old_size; + + ctx = reiser4_init_context(dentry->d_inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + result = setattr_pschedule_hook(inode); + if (result) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + old_size = i_size_read(inode); + inode_check_scale(inode, old_size, attr->ia_size); + + mutex_lock(&info->checkin_mutex); + result = cryptcompress_truncate(inode, + attr->ia_size, + 1/* update sd */); + mutex_unlock(&info->checkin_mutex); + if (result) { + warning("edward-1192", + "truncate_cryptcompress failed: oid %lli, " + "old size %lld, new size %lld, retval %d", + (unsigned long long) + get_inode_oid(inode), old_size, + attr->ia_size, result); + } + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + } else + result = 0; + } else + result = reiser4_setattr_common(dentry, attr); + return result; +} + +/* plugin->release */ +int release_cryptcompress(struct inode *inode, struct file *file) +{ + reiser4_context *ctx = reiser4_init_context(inode->i_sb); + + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + reiser4_free_file_fsdata(file); + reiser4_exit_context(ctx); + return 0; +} + +/* plugin->prepare_write */ +int write_begin_cryptcompress(struct file *file, struct page *page, + unsigned from, unsigned to) +{ + return do_prepare_write(file, page, from, to); +} + +/* plugin->commit_write */ +int write_end_cryptcompress(struct file *file, struct page *page, + unsigned from, unsigned to) +{ + int ret; + hint_t *hint; + lock_handle *lh; + struct inode * inode; + struct cluster_handle clust; + + unlock_page(page); + + inode = page->mapping->host; + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + hint_init_zero(hint); + lh = &hint->lh; + + cluster_init_read(&clust, NULL); + clust.hint = hint; + + ret = alloc_cluster_pgset(&clust, cluster_nrpages(inode)); + if (ret) + goto out; + clust.index = pg_to_clust(page->index, inode); + ret = capture_page_cluster(&clust, inode); + if (ret) + warning("edward-1557", + "Capture failed (inode %llu, result=%i)", + (unsigned long long)get_inode_oid(inode), ret); + out: + done_lh(lh); + kfree(hint); + put_cluster_handle(&clust); + return ret; +} + +/* plugin->bmap */ +sector_t bmap_cryptcompress(struct address_space *mapping, sector_t lblock) +{ + return -EINVAL; +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/cryptcompress.h linux-2.6.30/fs/reiser4/plugin/file/cryptcompress.h --- linux-2.6.30.orig/fs/reiser4/plugin/file/cryptcompress.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/cryptcompress.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,616 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* See http://www.namesys.com/cryptcompress_design.html */ + +#if !defined( __FS_REISER4_CRYPTCOMPRESS_H__ ) +#define __FS_REISER4_CRYPTCOMPRESS_H__ + +#include "../../page_cache.h" +#include "../compress/compress.h" +#include "../crypto/cipher.h" + +#include + +#define MIN_CLUSTER_SHIFT PAGE_CACHE_SHIFT +#define MAX_CLUSTER_SHIFT 16 +#define MAX_CLUSTER_NRPAGES (1U << MAX_CLUSTER_SHIFT >> PAGE_CACHE_SHIFT) +#define DC_CHECKSUM_SIZE 4 + +#define MIN_LATTICE_FACTOR 1 +#define MAX_LATTICE_FACTOR 32 + +/* this mask contains all non-standard plugins that might + be present in reiser4-specific part of inode managed by + cryptcompress file plugin */ +#define cryptcompress_mask \ + ((1 << PSET_FILE) | \ + (1 << PSET_CLUSTER) | \ + (1 << PSET_CIPHER) | \ + (1 << PSET_DIGEST) | \ + (1 << PSET_COMPRESSION) | \ + (1 << PSET_COMPRESSION_MODE)) + +#if REISER4_DEBUG +static inline int cluster_shift_ok(int shift) +{ + return (shift >= MIN_CLUSTER_SHIFT) && (shift <= MAX_CLUSTER_SHIFT); +} +#endif + +#if REISER4_DEBUG +#define INODE_PGCOUNT(inode) \ +({ \ + assert("edward-1530", inode_file_plugin(inode) == \ + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)); \ + atomic_read(&cryptcompress_inode_data(inode)->pgcount); \ + }) +#define INODE_PGCOUNT_INC(inode) \ +do { \ + assert("edward-1531", inode_file_plugin(inode) == \ + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)); \ + atomic_inc(&cryptcompress_inode_data(inode)->pgcount); \ +} while (0) +#define INODE_PGCOUNT_DEC(inode) \ +do { \ + if (inode_file_plugin(inode) == \ + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)) \ + atomic_dec(&cryptcompress_inode_data(inode)->pgcount); \ +} while (0) +#else +#define INODE_PGCOUNT(inode) (0) +#define INODE_PGCOUNT_INC(inode) +#define INODE_PGCOUNT_DEC(inode) +#endif /* REISER4_DEBUG */ + +struct tfm_stream { + __u8 *data; + size_t size; +}; + +typedef enum { + INPUT_STREAM, + OUTPUT_STREAM, + LAST_STREAM +} tfm_stream_id; + +typedef struct tfm_stream * tfm_unit[LAST_STREAM]; + +static inline __u8 *ts_data(struct tfm_stream * stm) +{ + assert("edward-928", stm != NULL); + return stm->data; +} + +static inline size_t ts_size(struct tfm_stream * stm) +{ + assert("edward-929", stm != NULL); + return stm->size; +} + +static inline void set_ts_size(struct tfm_stream * stm, size_t size) +{ + assert("edward-930", stm != NULL); + + stm->size = size; +} + +static inline int alloc_ts(struct tfm_stream ** stm) +{ + assert("edward-931", stm); + assert("edward-932", *stm == NULL); + + *stm = kzalloc(sizeof(**stm), reiser4_ctx_gfp_mask_get()); + if (!*stm) + return -ENOMEM; + return 0; +} + +static inline void free_ts(struct tfm_stream * stm) +{ + assert("edward-933", !ts_data(stm)); + assert("edward-934", !ts_size(stm)); + + kfree(stm); +} + +static inline int alloc_ts_data(struct tfm_stream * stm, size_t size) +{ + assert("edward-935", !ts_data(stm)); + assert("edward-936", !ts_size(stm)); + assert("edward-937", size != 0); + + stm->data = reiser4_vmalloc(size); + if (!stm->data) + return -ENOMEM; + set_ts_size(stm, size); + return 0; +} + +static inline void free_ts_data(struct tfm_stream * stm) +{ + assert("edward-938", equi(ts_data(stm), ts_size(stm))); + + if (ts_data(stm)) + vfree(ts_data(stm)); + memset(stm, 0, sizeof *stm); +} + +/* Write modes for item conversion in flush convert phase */ +typedef enum { + CRC_APPEND_ITEM = 1, + CRC_OVERWRITE_ITEM = 2, + CRC_CUT_ITEM = 3 +} cryptcompress_write_mode_t; + +typedef enum { + LC_INVAL = 0, /* invalid value */ + LC_APPOV = 1, /* append and/or overwrite */ + LC_TRUNC = 2 /* truncate */ +} logical_cluster_op; + +/* Transform cluster. + * Intermediate state between page cluster and disk cluster + * Is used for data transform (compression/encryption) + */ +struct tfm_cluster { + coa_set coa; /* compression algorithms info */ + tfm_unit tun; /* plain and transformed streams */ + tfm_action act; + int uptodate; + int lsize; /* number of bytes in logical cluster */ + int len; /* length of the transform stream */ +}; + +static inline coa_t get_coa(struct tfm_cluster * tc, reiser4_compression_id id, + tfm_action act) +{ + return tc->coa[id][act]; +} + +static inline void set_coa(struct tfm_cluster * tc, reiser4_compression_id id, + tfm_action act, coa_t coa) +{ + tc->coa[id][act] = coa; +} + +static inline int alloc_coa(struct tfm_cluster * tc, compression_plugin * cplug) +{ + coa_t coa; + + coa = cplug->alloc(tc->act); + if (IS_ERR(coa)) + return PTR_ERR(coa); + set_coa(tc, cplug->h.id, tc->act, coa); + return 0; +} + +static inline int +grab_coa(struct tfm_cluster * tc, compression_plugin * cplug) +{ + return (cplug->alloc && !get_coa(tc, cplug->h.id, tc->act) ? + alloc_coa(tc, cplug) : 0); +} + +static inline void free_coa_set(struct tfm_cluster * tc) +{ + tfm_action j; + reiser4_compression_id i; + compression_plugin *cplug; + + assert("edward-810", tc != NULL); + + for (j = 0; j < TFMA_LAST; j++) + for (i = 0; i < LAST_COMPRESSION_ID; i++) { + if (!get_coa(tc, i, j)) + continue; + cplug = compression_plugin_by_id(i); + assert("edward-812", cplug->free != NULL); + cplug->free(get_coa(tc, i, j), j); + set_coa(tc, i, j, 0); + } + return; +} + +static inline struct tfm_stream * get_tfm_stream(struct tfm_cluster * tc, + tfm_stream_id id) +{ + return tc->tun[id]; +} + +static inline void set_tfm_stream(struct tfm_cluster * tc, + tfm_stream_id id, struct tfm_stream * ts) +{ + tc->tun[id] = ts; +} + +static inline __u8 *tfm_stream_data(struct tfm_cluster * tc, tfm_stream_id id) +{ + return ts_data(get_tfm_stream(tc, id)); +} + +static inline void set_tfm_stream_data(struct tfm_cluster * tc, + tfm_stream_id id, __u8 * data) +{ + get_tfm_stream(tc, id)->data = data; +} + +static inline size_t tfm_stream_size(struct tfm_cluster * tc, tfm_stream_id id) +{ + return ts_size(get_tfm_stream(tc, id)); +} + +static inline void +set_tfm_stream_size(struct tfm_cluster * tc, tfm_stream_id id, size_t size) +{ + get_tfm_stream(tc, id)->size = size; +} + +static inline int +alloc_tfm_stream(struct tfm_cluster * tc, size_t size, tfm_stream_id id) +{ + assert("edward-939", tc != NULL); + assert("edward-940", !get_tfm_stream(tc, id)); + + tc->tun[id] = kzalloc(sizeof(struct tfm_stream), + reiser4_ctx_gfp_mask_get()); + if (!tc->tun[id]) + return -ENOMEM; + return alloc_ts_data(get_tfm_stream(tc, id), size); +} + +static inline int +realloc_tfm_stream(struct tfm_cluster * tc, size_t size, tfm_stream_id id) +{ + assert("edward-941", tfm_stream_size(tc, id) < size); + free_ts_data(get_tfm_stream(tc, id)); + return alloc_ts_data(get_tfm_stream(tc, id), size); +} + +static inline void free_tfm_stream(struct tfm_cluster * tc, tfm_stream_id id) +{ + free_ts_data(get_tfm_stream(tc, id)); + free_ts(get_tfm_stream(tc, id)); + set_tfm_stream(tc, id, 0); +} + +static inline unsigned coa_overrun(compression_plugin * cplug, int ilen) +{ + return (cplug->overrun != NULL ? cplug->overrun(ilen) : 0); +} + +static inline void free_tfm_unit(struct tfm_cluster * tc) +{ + tfm_stream_id id; + for (id = 0; id < LAST_STREAM; id++) { + if (!get_tfm_stream(tc, id)) + continue; + free_tfm_stream(tc, id); + } +} + +static inline void put_tfm_cluster(struct tfm_cluster * tc) +{ + assert("edward-942", tc != NULL); + free_coa_set(tc); + free_tfm_unit(tc); +} + +static inline int tfm_cluster_is_uptodate(struct tfm_cluster * tc) +{ + assert("edward-943", tc != NULL); + assert("edward-944", tc->uptodate == 0 || tc->uptodate == 1); + return (tc->uptodate == 1); +} + +static inline void tfm_cluster_set_uptodate(struct tfm_cluster * tc) +{ + assert("edward-945", tc != NULL); + assert("edward-946", tc->uptodate == 0 || tc->uptodate == 1); + tc->uptodate = 1; + return; +} + +static inline void tfm_cluster_clr_uptodate(struct tfm_cluster * tc) +{ + assert("edward-947", tc != NULL); + assert("edward-948", tc->uptodate == 0 || tc->uptodate == 1); + tc->uptodate = 0; + return; +} + +static inline int tfm_stream_is_set(struct tfm_cluster * tc, tfm_stream_id id) +{ + return (get_tfm_stream(tc, id) && + tfm_stream_data(tc, id) && tfm_stream_size(tc, id)); +} + +static inline int tfm_cluster_is_set(struct tfm_cluster * tc) +{ + int i; + for (i = 0; i < LAST_STREAM; i++) + if (!tfm_stream_is_set(tc, i)) + return 0; + return 1; +} + +static inline void alternate_streams(struct tfm_cluster * tc) +{ + struct tfm_stream *tmp = get_tfm_stream(tc, INPUT_STREAM); + + set_tfm_stream(tc, INPUT_STREAM, get_tfm_stream(tc, OUTPUT_STREAM)); + set_tfm_stream(tc, OUTPUT_STREAM, tmp); +} + +/* Set of states to indicate a kind of data + * that will be written to the window */ +typedef enum { + DATA_WINDOW, /* user's data */ + HOLE_WINDOW /* zeroes (such kind of data can be written + * if we start to write from offset > i_size) */ +} window_stat; + +/* Window (of logical cluster size) discretely sliding along a file. + * Is used to locate hole region in a logical cluster to be properly + * represented on disk. + * We split a write to cryptcompress file into writes to its logical + * clusters. Before writing to a logical cluster we set a window, i.e. + * calculate values of the following fields: + */ +struct reiser4_slide { + unsigned off; /* offset to write from */ + unsigned count; /* number of bytes to write */ + unsigned delta; /* number of bytes to append to the hole */ + window_stat stat; /* what kind of data will be written starting + from @off */ +}; + +/* Possible states of a disk cluster */ +typedef enum { + INVAL_DISK_CLUSTER, /* unknown state */ + PREP_DISK_CLUSTER, /* disk cluster got converted by flush + * at least 1 time */ + UNPR_DISK_CLUSTER, /* disk cluster just created and should be + * converted by flush */ + FAKE_DISK_CLUSTER, /* disk cluster doesn't exist neither in memory + * nor on disk */ + TRNC_DISK_CLUSTER /* disk cluster is partially truncated */ +} disk_cluster_stat; + +/* The following structure represents various stages of the same logical + * cluster of index @index: + * . fixed slide + * . page cluster (stage in primary cache) + * . transform cluster (transition stage) + * . disk cluster (stage in secondary cache) + * This structure is used in transition and synchronizing operations, e.g. + * transform cluster is a transition state when synchronizing page cluster + * and disk cluster. + * FIXME: Encapsulate page cluster, disk cluster. + */ +struct cluster_handle { + cloff_t index; /* offset in a file (unit is a cluster size) */ + int index_valid; /* for validating the index above, if needed */ + struct file *file; /* host file */ + + /* logical cluster */ + struct reiser4_slide *win; /* sliding window to locate holes */ + logical_cluster_op op; /* logical cluster operation (truncate or + append/overwrite) */ + /* transform cluster */ + struct tfm_cluster tc; /* contains all needed info to synchronize + page cluster and disk cluster) */ + /* page cluster */ + int nr_pages; /* number of pages of current checkin action */ + int old_nrpages; /* number of pages of last checkin action */ + struct page **pages; /* attached pages */ + jnode * node; /* jnode for capture */ + + /* disk cluster */ + hint_t *hint; /* current position in the tree */ + disk_cluster_stat dstat; /* state of the current disk cluster */ + int reserved; /* is space for disk cluster reserved */ +#if REISER4_DEBUG + reiser4_context *ctx; + int reserved_prepped; + int reserved_unprepped; +#endif + +}; + +static inline __u8 * tfm_input_data (struct cluster_handle * clust) +{ + return tfm_stream_data(&clust->tc, INPUT_STREAM); +} + +static inline __u8 * tfm_output_data (struct cluster_handle * clust) +{ + return tfm_stream_data(&clust->tc, OUTPUT_STREAM); +} + +static inline int reset_cluster_pgset(struct cluster_handle * clust, + int nrpages) +{ + assert("edward-1057", clust->pages != NULL); + memset(clust->pages, 0, sizeof(*clust->pages) * nrpages); + return 0; +} + +static inline int alloc_cluster_pgset(struct cluster_handle * clust, + int nrpages) +{ + assert("edward-949", clust != NULL); + assert("edward-1362", clust->pages == NULL); + assert("edward-950", nrpages != 0 && nrpages <= MAX_CLUSTER_NRPAGES); + + clust->pages = kzalloc(sizeof(*clust->pages) * nrpages, + reiser4_ctx_gfp_mask_get()); + if (!clust->pages) + return RETERR(-ENOMEM); + return 0; +} + +static inline void move_cluster_pgset(struct cluster_handle *clust, + struct page ***pages, int * nr_pages) +{ + assert("edward-1545", clust != NULL && clust->pages != NULL); + assert("edward-1546", pages != NULL && *pages == NULL); + *pages = clust->pages; + *nr_pages = clust->nr_pages; + clust->pages = NULL; +} + +static inline void free_cluster_pgset(struct cluster_handle * clust) +{ + assert("edward-951", clust->pages != NULL); + kfree(clust->pages); + clust->pages = NULL; +} + +static inline void put_cluster_handle(struct cluster_handle * clust) +{ + assert("edward-435", clust != NULL); + + put_tfm_cluster(&clust->tc); + if (clust->pages) + free_cluster_pgset(clust); + memset(clust, 0, sizeof *clust); +} + +static inline void inc_keyload_count(struct reiser4_crypto_info * data) +{ + assert("edward-1410", data != NULL); + data->keyload_count++; +} + +static inline void dec_keyload_count(struct reiser4_crypto_info * data) +{ + assert("edward-1411", data != NULL); + assert("edward-1412", data->keyload_count > 0); + data->keyload_count--; +} + +static inline int capture_cluster_jnode(jnode * node) +{ + return reiser4_try_capture(node, ZNODE_WRITE_LOCK, 0); +} + +/* cryptcompress specific part of reiser4_inode */ +struct cryptcompress_info { + struct mutex checkin_mutex; /* This is to serialize + * checkin_logical_cluster operations */ + cloff_t trunc_index; /* Index of the leftmost truncated disk + * cluster (to resolve races with read) */ + struct reiser4_crypto_info *crypt; + /* + * the following 2 fields are controlled by compression mode plugin + */ + int compress_toggle; /* Current status of compressibility */ + int lattice_factor; /* Factor of dynamic lattice. FIXME: Have + * a compression_toggle to keep the factor + */ +#if REISER4_DEBUG + atomic_t pgcount; /* number of grabbed pages */ +#endif +}; + +static inline void set_compression_toggle (struct cryptcompress_info * info, int val) +{ + info->compress_toggle = val; +} + +static inline int get_compression_toggle (struct cryptcompress_info * info) +{ + return info->compress_toggle; +} + +static inline int compression_is_on(struct cryptcompress_info * info) +{ + return get_compression_toggle(info) == 1; +} + +static inline void turn_on_compression(struct cryptcompress_info * info) +{ + set_compression_toggle(info, 1); +} + +static inline void turn_off_compression(struct cryptcompress_info * info) +{ + set_compression_toggle(info, 0); +} + +static inline void set_lattice_factor(struct cryptcompress_info * info, int val) +{ + info->lattice_factor = val; +} + +static inline int get_lattice_factor(struct cryptcompress_info * info) +{ + return info->lattice_factor; +} + +struct cryptcompress_info *cryptcompress_inode_data(const struct inode *); +int equal_to_rdk(znode *, const reiser4_key *); +int goto_right_neighbor(coord_t *, lock_handle *); +int cryptcompress_inode_ok(struct inode *inode); +int coord_is_unprepped_ctail(const coord_t * coord); +extern int do_readpage_ctail(struct inode *, struct cluster_handle *, + struct page * page, znode_lock_mode mode); +extern int ctail_insert_unprepped_cluster(struct cluster_handle * clust, + struct inode * inode); +extern int readpages_cryptcompress(struct file*, struct address_space*, + struct list_head*, unsigned); +int bind_cryptcompress(struct inode *child, struct inode *parent); +void destroy_inode_cryptcompress(struct inode * inode); +int grab_page_cluster(struct inode *inode, struct cluster_handle * clust, + rw_op rw); +int write_pschedule_hook(struct file *file, struct inode * inode, + loff_t pos, struct cluster_handle * clust, + struct psched_context * cont); +int setattr_pschedule_hook(struct inode * inode); +struct reiser4_crypto_info * inode_crypto_info(struct inode * inode); +void inherit_crypto_info_common(struct inode * parent, struct inode * object, + int (*can_inherit)(struct inode * child, + struct inode * parent)); +void reiser4_attach_crypto_info(struct inode * inode, + struct reiser4_crypto_info * info); +void change_crypto_info(struct inode * inode, struct reiser4_crypto_info * new); +struct reiser4_crypto_info * reiser4_alloc_crypto_info (struct inode * inode); + +static inline struct crypto_blkcipher * info_get_cipher(struct reiser4_crypto_info * info) +{ + return info->cipher; +} + +static inline void info_set_cipher(struct reiser4_crypto_info * info, + struct crypto_blkcipher * tfm) +{ + info->cipher = tfm; +} + +static inline struct crypto_hash * info_get_digest(struct reiser4_crypto_info * info) +{ + return info->digest; +} + +static inline void info_set_digest(struct reiser4_crypto_info * info, + struct crypto_hash * tfm) +{ + info->digest = tfm; +} + +static inline void put_cluster_page(struct page * page) +{ + page_cache_release(page); +} + +#endif /* __FS_REISER4_CRYPTCOMPRESS_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/file.c linux-2.6.30/fs/reiser4/plugin/file/file.c --- linux-2.6.30.orig/fs/reiser4/plugin/file/file.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/file.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,2687 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* + * this file contains implementations of inode/file/address_space/file plugin + * operations specific for "unix file plugin" (plugin id is + * UNIX_FILE_PLUGIN_ID). "Unix file" is either built of tail items only + * (FORMATTING_ID) or of extent items only (EXTENT_POINTER_ID) or empty (have + * no items but stat data) + */ + +#include "../../inode.h" +#include "../../super.h" +#include "../../tree_walk.h" +#include "../../carry.h" +#include "../../page_cache.h" +#include "../../ioctl.h" +#include "../object.h" +#include "../cluster.h" +#include "../../safe_link.h" + +#include +#include +#include + + +static int unpack(struct file *file, struct inode *inode, int forever); +static void drop_access(struct unix_file_info *); +static int hint_validate(hint_t * hint, const reiser4_key * key, int check_key, + znode_lock_mode lock_mode); + +/* Get exclusive access and make sure that file is not partially + * converted (It may happen that another process is doing tail + * conversion. If so, wait until it completes) + */ +static inline void get_exclusive_access_careful(struct unix_file_info * uf_info, + struct inode *inode) +{ + do { + get_exclusive_access(uf_info); + if (!reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV)) + break; + drop_exclusive_access(uf_info); + schedule(); + } while (1); +} + +/* get unix file plugin specific portion of inode */ +struct unix_file_info *unix_file_inode_data(const struct inode *inode) +{ + return &reiser4_inode_data(inode)->file_plugin_data.unix_file_info; +} + +/** + * equal_to_rdk - compare key and znode's right delimiting key + * @node: node whose right delimiting key to compare with @key + * @key: key to compare with @node's right delimiting key + * + * Returns true if @key is equal to right delimiting key of @node. + */ +int equal_to_rdk(znode *node, const reiser4_key *key) +{ + int result; + + read_lock_dk(znode_get_tree(node)); + result = keyeq(key, znode_get_rd_key(node)); + read_unlock_dk(znode_get_tree(node)); + return result; +} + +#if REISER4_DEBUG + +/** + * equal_to_ldk - compare key and znode's left delimiting key + * @node: node whose left delimiting key to compare with @key + * @key: key to compare with @node's left delimiting key + * + * Returns true if @key is equal to left delimiting key of @node. + */ +int equal_to_ldk(znode *node, const reiser4_key *key) +{ + int result; + + read_lock_dk(znode_get_tree(node)); + result = keyeq(key, znode_get_ld_key(node)); + read_unlock_dk(znode_get_tree(node)); + return result; +} + +/** + * check_coord - check whether coord corresponds to key + * @coord: coord to check + * @key: key @coord has to correspond to + * + * Returns true if @coord is set as if it was set as result of lookup with @key + * in coord->node. + */ +static int check_coord(const coord_t *coord, const reiser4_key *key) +{ + coord_t twin; + + node_plugin_by_node(coord->node)->lookup(coord->node, key, + FIND_MAX_NOT_MORE_THAN, &twin); + return coords_equal(coord, &twin); +} + +#endif /* REISER4_DEBUG */ + +/** + * init_uf_coord - initialize extended coord + * @uf_coord: + * @lh: + * + * + */ +void init_uf_coord(uf_coord_t *uf_coord, lock_handle *lh) +{ + coord_init_zero(&uf_coord->coord); + coord_clear_iplug(&uf_coord->coord); + uf_coord->lh = lh; + init_lh(lh); + memset(&uf_coord->extension, 0, sizeof(uf_coord->extension)); + uf_coord->valid = 0; +} + +static void validate_extended_coord(uf_coord_t *uf_coord, loff_t offset) +{ + assert("vs-1333", uf_coord->valid == 0); + + if (coord_is_between_items(&uf_coord->coord)) + return; + + assert("vs-1348", + item_plugin_by_coord(&uf_coord->coord)->s.file. + init_coord_extension); + + item_body_by_coord(&uf_coord->coord); + item_plugin_by_coord(&uf_coord->coord)->s.file. + init_coord_extension(uf_coord, offset); +} + +/** + * goto_right_neighbor - lock right neighbor, drop current node lock + * @coord: + * @lh: + * + * Obtain lock on right neighbor and drop lock on current node. + */ +int goto_right_neighbor(coord_t *coord, lock_handle *lh) +{ + int result; + lock_handle lh_right; + + assert("vs-1100", znode_is_locked(coord->node)); + + init_lh(&lh_right); + result = reiser4_get_right_neighbor(&lh_right, coord->node, + znode_is_wlocked(coord->node) ? + ZNODE_WRITE_LOCK : ZNODE_READ_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (result) { + done_lh(&lh_right); + return result; + } + + /* + * we hold two longterm locks on neighboring nodes. Unlock left of + * them + */ + done_lh(lh); + + coord_init_first_unit_nocheck(coord, lh_right.node); + move_lh(lh, &lh_right); + + return 0; + +} + +/** + * set_file_state + * @uf_info: + * @cbk_result: + * @level: + * + * This is to be used by find_file_item and in find_file_state to + * determine real state of file + */ +static void set_file_state(struct unix_file_info *uf_info, int cbk_result, + tree_level level) +{ + if (cbk_errored(cbk_result)) + /* error happened in find_file_item */ + return; + + assert("vs-1164", level == LEAF_LEVEL || level == TWIG_LEVEL); + + if (uf_info->container == UF_CONTAINER_UNKNOWN) { + if (cbk_result == CBK_COORD_NOTFOUND) + uf_info->container = UF_CONTAINER_EMPTY; + else if (level == LEAF_LEVEL) + uf_info->container = UF_CONTAINER_TAILS; + else + uf_info->container = UF_CONTAINER_EXTENTS; + } else { + /* + * file state is known, check whether it is set correctly if + * file is not being tail converted + */ + if (!reiser4_inode_get_flag(unix_file_info_to_inode(uf_info), + REISER4_PART_IN_CONV)) { + assert("vs-1162", + ergo(level == LEAF_LEVEL && + cbk_result == CBK_COORD_FOUND, + uf_info->container == UF_CONTAINER_TAILS)); + assert("vs-1165", + ergo(level == TWIG_LEVEL && + cbk_result == CBK_COORD_FOUND, + uf_info->container == UF_CONTAINER_EXTENTS)); + } + } +} + +int find_file_item_nohint(coord_t *coord, lock_handle *lh, + const reiser4_key *key, znode_lock_mode lock_mode, + struct inode *inode) +{ + return reiser4_object_lookup(inode, key, coord, lh, lock_mode, + FIND_MAX_NOT_MORE_THAN, + TWIG_LEVEL, LEAF_LEVEL, + (lock_mode == ZNODE_READ_LOCK) ? CBK_UNIQUE : + (CBK_UNIQUE | CBK_FOR_INSERT), + NULL /* ra_info */ ); +} + +/** + * find_file_item - look for file item in the tree + * @hint: provides coordinate, lock handle, seal + * @key: key for search + * @mode: mode of lock to put on returned node + * @ra_info: + * @inode: + * + * This finds position in the tree corresponding to @key. It first tries to use + * @hint's seal if it is set. + */ +int find_file_item(hint_t *hint, const reiser4_key *key, + znode_lock_mode lock_mode, + struct inode *inode) +{ + int result; + coord_t *coord; + lock_handle *lh; + + assert("nikita-3030", reiser4_schedulable()); + assert("vs-1707", hint != NULL); + assert("vs-47", inode != NULL); + + coord = &hint->ext_coord.coord; + lh = hint->ext_coord.lh; + init_lh(lh); + + result = hint_validate(hint, key, 1 /* check key */, lock_mode); + if (!result) { + if (coord->between == AFTER_UNIT && + equal_to_rdk(coord->node, key)) { + result = goto_right_neighbor(coord, lh); + if (result == -E_NO_NEIGHBOR) + return RETERR(-EIO); + if (result) + return result; + assert("vs-1152", equal_to_ldk(coord->node, key)); + /* + * we moved to different node. Invalidate coord + * extension, zload is necessary to init it again + */ + hint->ext_coord.valid = 0; + } + + set_file_state(unix_file_inode_data(inode), CBK_COORD_FOUND, + znode_get_level(coord->node)); + + return CBK_COORD_FOUND; + } + + coord_init_zero(coord); + result = find_file_item_nohint(coord, lh, key, lock_mode, inode); + set_file_state(unix_file_inode_data(inode), result, + znode_get_level(coord->node)); + + /* FIXME: we might already have coord extension initialized */ + hint->ext_coord.valid = 0; + return result; +} + +/* plugin->u.file.write_flowom = NULL + plugin->u.file.read_flow = NULL */ + +void hint_init_zero(hint_t * hint) +{ + memset(hint, 0, sizeof(*hint)); + init_lh(&hint->lh); + hint->ext_coord.lh = &hint->lh; +} + +static int find_file_state(struct inode *inode, struct unix_file_info *uf_info) +{ + int result; + reiser4_key key; + coord_t coord; + lock_handle lh; + + assert("vs-1628", ea_obtained(uf_info)); + + if (uf_info->container == UF_CONTAINER_UNKNOWN) { + key_by_inode_and_offset_common(inode, 0, &key); + init_lh(&lh); + result = find_file_item_nohint(&coord, &lh, &key, + ZNODE_READ_LOCK, inode); + set_file_state(uf_info, result, znode_get_level(coord.node)); + done_lh(&lh); + if (!cbk_errored(result)) + result = 0; + } else + result = 0; + assert("vs-1074", + ergo(result == 0, uf_info->container != UF_CONTAINER_UNKNOWN)); + reiser4_txn_restart_current(); + return result; +} + +/** + * Estimate and reserve space needed to truncate page + * which gets partially truncated: one block for page + * itself, stat-data update (estimate_one_insert_into_item) + * and one item insertion (estimate_one_insert_into_item) + * which may happen if page corresponds to hole extent and + * unallocated one will have to be created + */ +static int reserve_partial_page(reiser4_tree * tree) +{ + grab_space_enable(); + return reiser4_grab_reserved(reiser4_get_current_sb(), + 1 + + 2 * estimate_one_insert_into_item(tree), + BA_CAN_COMMIT); +} + +/* estimate and reserve space needed to cut one item and update one stat data */ +static int reserve_cut_iteration(reiser4_tree * tree) +{ + __u64 estimate = estimate_one_item_removal(tree) + + estimate_one_insert_into_item(tree); + + assert("nikita-3172", lock_stack_isclean(get_current_lock_stack())); + + grab_space_enable(); + /* We need to double our estimate now that we can delete more than one + node. */ + return reiser4_grab_reserved(reiser4_get_current_sb(), estimate * 2, + BA_CAN_COMMIT); +} + +int reiser4_update_file_size(struct inode *inode, loff_t new_size, + int update_sd) +{ + int result = 0; + + INODE_SET_SIZE(inode, new_size); + if (update_sd) { + inode->i_ctime = inode->i_mtime = CURRENT_TIME; + result = reiser4_update_sd(inode); + } + return result; +} + +/** + * Cut file items one by one starting from the last one until + * new file size (inode->i_size) is reached. Reserve space + * and update file stat data on every single cut from the tree + */ +int cut_file_items(struct inode *inode, loff_t new_size, + int update_sd, loff_t cur_size, + int (*update_actor) (struct inode *, loff_t, int)) +{ + reiser4_key from_key, to_key; + reiser4_key smallest_removed; + file_plugin *fplug = inode_file_plugin(inode); + int result; + int progress = 0; + + assert("vs-1248", + fplug == file_plugin_by_id(UNIX_FILE_PLUGIN_ID) || + fplug == file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)); + + fplug->key_by_inode(inode, new_size, &from_key); + to_key = from_key; + set_key_offset(&to_key, cur_size - 1 /*get_key_offset(reiser4_max_key()) */ ); + /* this loop normally runs just once */ + while (1) { + result = reserve_cut_iteration(reiser4_tree_by_inode(inode)); + if (result) + break; + + result = reiser4_cut_tree_object(current_tree, &from_key, &to_key, + &smallest_removed, inode, 1, + &progress); + if (result == -E_REPEAT) { + /** + * -E_REPEAT is a signal to interrupt a long + * file truncation process + */ + if (progress) { + result = update_actor(inode, + get_key_offset(&smallest_removed), + update_sd); + if (result) + break; + } + /* the below does up(sbinfo->delete_mutex). + * Do not get folled */ + reiser4_release_reserved(inode->i_sb); + /** + * reiser4_cut_tree_object() was interrupted probably + * because current atom requires commit, we have to + * release transaction handle to allow atom commit. + */ + reiser4_txn_restart_current(); + continue; + } + if (result + && !(result == CBK_COORD_NOTFOUND && new_size == 0 + && inode->i_size == 0)) + break; + + set_key_offset(&smallest_removed, new_size); + /* Final sd update after the file gets its correct size */ + result = update_actor(inode, get_key_offset(&smallest_removed), + update_sd); + break; + } + + /* the below does up(sbinfo->delete_mutex). Do not get folled */ + reiser4_release_reserved(inode->i_sb); + + return result; +} + +int find_or_create_extent(struct page *page); + +/* part of truncate_file_body: it is called when truncate is used to make file + shorter */ +static int shorten_file(struct inode *inode, loff_t new_size) +{ + int result; + struct page *page; + int padd_from; + unsigned long index; + struct unix_file_info *uf_info; + + /* + * all items of ordinary reiser4 file are grouped together. That is why + * we can use reiser4_cut_tree. Plan B files (for instance) can not be + * truncated that simply + */ + result = cut_file_items(inode, new_size, 1 /*update_sd */ , + get_key_offset(reiser4_max_key()), + reiser4_update_file_size); + if (result) + return result; + + uf_info = unix_file_inode_data(inode); + assert("vs-1105", new_size == inode->i_size); + if (new_size == 0) { + uf_info->container = UF_CONTAINER_EMPTY; + return 0; + } + + result = find_file_state(inode, uf_info); + if (result) + return result; + if (uf_info->container == UF_CONTAINER_TAILS) + /* + * No need to worry about zeroing last page after new file + * end + */ + return 0; + + padd_from = inode->i_size & (PAGE_CACHE_SIZE - 1); + if (!padd_from) + /* file is truncated to page boundary */ + return 0; + + result = reserve_partial_page(reiser4_tree_by_inode(inode)); + if (result) { + reiser4_release_reserved(inode->i_sb); + return result; + } + + /* last page is partially truncated - zero its content */ + index = (inode->i_size >> PAGE_CACHE_SHIFT); + page = read_mapping_page(inode->i_mapping, index, NULL); + if (IS_ERR(page)) { + /* + * the below does up(sbinfo->delete_mutex). Do not get + * confused + */ + reiser4_release_reserved(inode->i_sb); + if (likely(PTR_ERR(page) == -EINVAL)) { + /* looks like file is built of tail items */ + return 0; + } + return PTR_ERR(page); + } + wait_on_page_locked(page); + if (!PageUptodate(page)) { + page_cache_release(page); + /* + * the below does up(sbinfo->delete_mutex). Do not get + * confused + */ + reiser4_release_reserved(inode->i_sb); + return RETERR(-EIO); + } + + /* + * if page correspons to hole extent unit - unallocated one will be + * created here. This is not necessary + */ + result = find_or_create_extent(page); + + /* + * FIXME: cut_file_items has already updated inode. Probably it would + * be better to update it here when file is really truncated + */ + if (result) { + page_cache_release(page); + /* + * the below does up(sbinfo->delete_mutex). Do not get + * confused + */ + reiser4_release_reserved(inode->i_sb); + return result; + } + + lock_page(page); + assert("vs-1066", PageLocked(page)); + zero_user_segment(page, padd_from, PAGE_CACHE_SIZE); + unlock_page(page); + page_cache_release(page); + /* the below does up(sbinfo->delete_mutex). Do not get confused */ + reiser4_release_reserved(inode->i_sb); + return 0; +} + +/** + * should_have_notail + * @uf_info: + * @new_size: + * + * Calls formatting plugin to see whether file of size @new_size has to be + * stored in unformatted nodes or in tail items. 0 is returned for later case. + */ +static int should_have_notail(const struct unix_file_info *uf_info, loff_t new_size) +{ + if (!uf_info->tplug) + return 1; + return !uf_info->tplug->have_tail(unix_file_info_to_inode(uf_info), + new_size); + +} + +/** + * truncate_file_body - change length of file + * @inode: inode of file + * @new_size: new file length + * + * Adjusts items file @inode is built of to match @new_size. It may either cut + * items or add them to represent a hole at the end of file. The caller has to + * obtain exclusive access to the file. + */ +static int truncate_file_body(struct inode *inode, struct iattr *attr) +{ + int result; + loff_t new_size = attr->ia_size; + + if (inode->i_size < new_size) { + /* expanding truncate */ + struct unix_file_info *uf_info = unix_file_inode_data(inode); + + result = find_file_state(inode, uf_info); + if (result) + return result; + + if (should_have_notail(uf_info, new_size)) { + /* + * file of size @new_size has to be built of + * extents. If it is built of tails - convert to + * extents + */ + if (uf_info->container == UF_CONTAINER_TAILS) { + /* + * if file is being convered by another process + * - wait until it completes + */ + while (1) { + if (reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)) { + drop_exclusive_access(uf_info); + schedule(); + get_exclusive_access(uf_info); + continue; + } + break; + } + + if (uf_info->container == UF_CONTAINER_TAILS) { + result = tail2extent(uf_info); + if (result) + return result; + } + } + result = reiser4_write_extent(NULL, inode, NULL, + 0, &new_size); + if (result) + return result; + uf_info->container = UF_CONTAINER_EXTENTS; + } else { + if (uf_info->container == UF_CONTAINER_EXTENTS) { + result = reiser4_write_extent(NULL, inode, NULL, + 0, &new_size); + if (result) + return result; + } else { + result = reiser4_write_tail(NULL, inode, NULL, + 0, &new_size); + if (result) + return result; + uf_info->container = UF_CONTAINER_TAILS; + } + } + BUG_ON(result > 0); + result = reiser4_update_file_size(inode, new_size, 1); + BUG_ON(result != 0); + } else + result = shorten_file(inode, new_size); + return result; +} + +/* plugin->u.write_sd_by_inode = write_sd_by_inode_common */ + +/** + * load_file_hint - copy hint from struct file to local variable + * @file: file to get hint from + * @hint: structure to fill + * + * Reiser4 specific portion of struct file may contain information (hint) + * stored on exiting from previous read or write. That information includes + * seal of znode and coord within that znode where previous read or write + * stopped. This function copies that information to @hint if it was stored or + * initializes @hint by 0s otherwise. + */ +int load_file_hint(struct file *file, hint_t *hint) +{ + reiser4_file_fsdata *fsdata; + + if (file) { + fsdata = reiser4_get_file_fsdata(file); + if (IS_ERR(fsdata)) + return PTR_ERR(fsdata); + + spin_lock_inode(file->f_dentry->d_inode); + if (reiser4_seal_is_set(&fsdata->reg.hint.seal)) { + *hint = fsdata->reg.hint; + init_lh(&hint->lh); + hint->ext_coord.lh = &hint->lh; + spin_unlock_inode(file->f_dentry->d_inode); + /* + * force re-validation of the coord on the first + * iteration of the read/write loop. + */ + hint->ext_coord.valid = 0; + assert("nikita-19892", coords_equal(&hint->seal.coord1, + &hint->ext_coord. + coord)); + return 0; + } + memset(&fsdata->reg.hint, 0, sizeof(hint_t)); + spin_unlock_inode(file->f_dentry->d_inode); + } + hint_init_zero(hint); + return 0; +} + +/** + * save_file_hint - copy hint to reiser4 private struct file's part + * @file: file to save hint in + * @hint: hint to save + * + * This copies @hint to reiser4 private part of struct file. It can help + * speedup future accesses to the file. + */ +void save_file_hint(struct file *file, const hint_t *hint) +{ + reiser4_file_fsdata *fsdata; + + assert("edward-1337", hint != NULL); + + if (!file || !reiser4_seal_is_set(&hint->seal)) + return; + fsdata = reiser4_get_file_fsdata(file); + assert("vs-965", !IS_ERR(fsdata)); + assert("nikita-19891", + coords_equal(&hint->seal.coord1, &hint->ext_coord.coord)); + assert("vs-30", hint->lh.owner == NULL); + spin_lock_inode(file->f_dentry->d_inode); + fsdata->reg.hint = *hint; + spin_unlock_inode(file->f_dentry->d_inode); + return; +} + +void reiser4_unset_hint(hint_t * hint) +{ + assert("vs-1315", hint); + hint->ext_coord.valid = 0; + reiser4_seal_done(&hint->seal); + done_lh(&hint->lh); +} + +/* coord must be set properly. So, that reiser4_set_hint + has nothing to do */ +void reiser4_set_hint(hint_t * hint, const reiser4_key * key, + znode_lock_mode mode) +{ + ON_DEBUG(coord_t * coord = &hint->ext_coord.coord); + assert("vs-1207", WITH_DATA(coord->node, check_coord(coord, key))); + + reiser4_seal_init(&hint->seal, &hint->ext_coord.coord, key); + hint->offset = get_key_offset(key); + hint->mode = mode; + done_lh(&hint->lh); +} + +int hint_is_set(const hint_t * hint) +{ + return reiser4_seal_is_set(&hint->seal); +} + +#if REISER4_DEBUG +static int all_but_offset_key_eq(const reiser4_key * k1, const reiser4_key * k2) +{ + return (get_key_locality(k1) == get_key_locality(k2) && + get_key_type(k1) == get_key_type(k2) && + get_key_band(k1) == get_key_band(k2) && + get_key_ordering(k1) == get_key_ordering(k2) && + get_key_objectid(k1) == get_key_objectid(k2)); +} +#endif + +static int +hint_validate(hint_t * hint, const reiser4_key * key, int check_key, + znode_lock_mode lock_mode) +{ + if (!hint || !hint_is_set(hint) || hint->mode != lock_mode) + /* hint either not set or set by different operation */ + return RETERR(-E_REPEAT); + + assert("vs-1277", all_but_offset_key_eq(key, &hint->seal.key)); + + if (check_key && get_key_offset(key) != hint->offset) + /* hint is set for different key */ + return RETERR(-E_REPEAT); + + assert("vs-31", hint->ext_coord.lh == &hint->lh); + return reiser4_seal_validate(&hint->seal, &hint->ext_coord.coord, key, + hint->ext_coord.lh, lock_mode, + ZNODE_LOCK_LOPRI); +} + +/** + * Look for place at twig level for extent corresponding to page, + * call extent's writepage method to create unallocated extent if + * it does not exist yet, initialize jnode, capture page + */ +int find_or_create_extent(struct page *page) +{ + int result; + struct inode *inode; + int plugged_hole; + + jnode *node; + + assert("vs-1065", page->mapping && page->mapping->host); + inode = page->mapping->host; + + lock_page(page); + node = jnode_of_page(page); + if (IS_ERR(node)) { + unlock_page(page); + return PTR_ERR(node); + } + JF_SET(node, JNODE_WRITE_PREPARED); + unlock_page(page); + + if (node->blocknr == 0) { + plugged_hole = 0; + result = reiser4_update_extent(inode, node, page_offset(page), + &plugged_hole); + if (result) { + JF_CLR(node, JNODE_WRITE_PREPARED); + jput(node); + warning("edward-1549", + "reiser4_update_extent failed: %d", result); + return result; + } + if (plugged_hole) + reiser4_update_sd(inode); + } else { + spin_lock_jnode(node); + result = reiser4_try_capture(node, ZNODE_WRITE_LOCK, 0); + BUG_ON(result != 0); + jnode_make_dirty_locked(node); + spin_unlock_jnode(node); + } + + BUG_ON(node->atom == NULL); + JF_CLR(node, JNODE_WRITE_PREPARED); + jput(node); + + if (get_current_context()->entd) { + entd_context *ent = get_entd_context(node->tree->super); + + if (ent->cur_request->page == page) + ent->cur_request->node = node; + } + return 0; +} + +/** + * has_anonymous_pages - check whether inode has pages dirtied via mmap + * @inode: inode to check + * + * Returns true if inode's mapping has dirty pages which do not belong to any + * atom. Those are either tagged PAGECACHE_TAG_REISER4_MOVED in mapping's page + * tree or were eflushed and can be found via jnodes tagged + * EFLUSH_TAG_ANONYMOUS in radix tree of jnodes. + */ +static int has_anonymous_pages(struct inode *inode) +{ + int result; + + spin_lock_irq(&inode->i_mapping->tree_lock); + result = radix_tree_tagged(&inode->i_mapping->page_tree, PAGECACHE_TAG_REISER4_MOVED); + spin_unlock_irq(&inode->i_mapping->tree_lock); + return result; +} + +/** + * capture_page_and_create_extent - + * @page: page to be captured + * + * Grabs space for extent creation and stat data update and calls function to + * do actual work. + */ +static int capture_page_and_create_extent(struct page *page) +{ + int result; + struct inode *inode; + + assert("vs-1084", page->mapping && page->mapping->host); + inode = page->mapping->host; + assert("vs-1139", + unix_file_inode_data(inode)->container == UF_CONTAINER_EXTENTS); + /* page belongs to file */ + assert("vs-1393", + inode->i_size > page_offset(page)); + + /* page capture may require extent creation (if it does not exist yet) + and stat data's update (number of blocks changes on extent + creation) */ + grab_space_enable(); + result = reiser4_grab_space(2 * estimate_one_insert_into_item + (reiser4_tree_by_inode(inode)), + BA_CAN_COMMIT); + if (likely(!result)) + result = find_or_create_extent(page); + + if (result != 0) + SetPageError(page); + return result; +} + +/* plugin->write_end() */ +int write_end_unix_file(struct file *file, struct page *page, + unsigned from, unsigned to) +{ + unlock_page(page); + return capture_page_and_create_extent(page); +} + +/* + * Support for "anonymous" pages and jnodes. + * + * When file is write-accessed through mmap pages can be dirtied from the user + * level. In this case kernel is not notified until one of following happens: + * + * (1) msync() + * + * (2) truncate() (either explicit or through unlink) + * + * (3) VM scanner starts reclaiming mapped pages, dirtying them before + * starting write-back. + * + * As a result of (3) ->writepage may be called on a dirty page without + * jnode. Such page is called "anonymous" in reiser4. Certain work-loads + * (iozone) generate huge number of anonymous pages. + * + * reiser4_sync_sb() method tries to insert anonymous pages into + * tree. This is done by capture_anonymous_*() functions below. + */ + +/** + * capture_anonymous_page - involve page into transaction + * @pg: page to deal with + * + * Takes care that @page has corresponding metadata in the tree, creates jnode + * for @page and captures it. On success 1 is returned. + */ +static int capture_anonymous_page(struct page *page) +{ + int result; + + if (PageWriteback(page)) + /* FIXME: do nothing? */ + return 0; + + result = capture_page_and_create_extent(page); + if (result == 0) { + result = 1; + } else + warning("nikita-3329", + "Cannot capture anon page: %i", result); + + return result; +} + +/** + * capture_anonymous_pages - find and capture pages dirtied via mmap + * @mapping: address space where to look for pages + * @index: start index + * @to_capture: maximum number of pages to capture + * + * Looks for pages tagged REISER4_MOVED starting from the *@index-th page, + * captures (involves into atom) them, returns number of captured pages, + * updates @index to next page after the last captured one. + */ +static int +capture_anonymous_pages(struct address_space *mapping, pgoff_t *index, + unsigned int to_capture) +{ + int result; + struct pagevec pvec; + unsigned int i, count; + int nr; + + pagevec_init(&pvec, 0); + count = min(pagevec_space(&pvec), to_capture); + nr = 0; + + /* find pages tagged MOVED */ + spin_lock_irq(&mapping->tree_lock); + pvec.nr = radix_tree_gang_lookup_tag(&mapping->page_tree, + (void **)pvec.pages, *index, count, + PAGECACHE_TAG_REISER4_MOVED); + if (pagevec_count(&pvec) == 0) { + /* + * there are no pages tagged MOVED in mapping->page_tree + * starting from *index + */ + spin_unlock_irq(&mapping->tree_lock); + *index = (pgoff_t)-1; + return 0; + } + + /* clear MOVED tag for all found pages */ + for (i = 0; i < pagevec_count(&pvec); i++) { + page_cache_get(pvec.pages[i]); + radix_tree_tag_clear(&mapping->page_tree, pvec.pages[i]->index, + PAGECACHE_TAG_REISER4_MOVED); + } + spin_unlock_irq(&mapping->tree_lock); + + + *index = pvec.pages[i - 1]->index + 1; + + for (i = 0; i < pagevec_count(&pvec); i++) { + result = capture_anonymous_page(pvec.pages[i]); + if (result == 1) + nr++; + else { + if (result < 0) { + warning("vs-1454", + "failed to capture page: " + "result=%d, captured=%d)\n", + result, i); + + /* + * set MOVED tag to all pages which left not + * captured + */ + spin_lock_irq(&mapping->tree_lock); + for (; i < pagevec_count(&pvec); i ++) { + radix_tree_tag_set(&mapping->page_tree, + pvec.pages[i]->index, + PAGECACHE_TAG_REISER4_MOVED); + } + spin_unlock_irq(&mapping->tree_lock); + + pagevec_release(&pvec); + return result; + } else { + /* + * result == 0. capture_anonymous_page returns + * 0 for Writeback-ed page. Set MOVED tag on + * that page + */ + spin_lock_irq(&mapping->tree_lock); + radix_tree_tag_set(&mapping->page_tree, + pvec.pages[i]->index, + PAGECACHE_TAG_REISER4_MOVED); + spin_unlock_irq(&mapping->tree_lock); + if (i == 0) + *index = pvec.pages[0]->index; + else + *index = pvec.pages[i - 1]->index + 1; + } + } + } + pagevec_release(&pvec); + return nr; +} + +/** + * capture_anonymous_jnodes - find and capture anonymous jnodes + * @mapping: address space where to look for jnodes + * @from: start index + * @to: end index + * @to_capture: maximum number of jnodes to capture + * + * Looks for jnodes tagged EFLUSH_TAG_ANONYMOUS in inode's tree of jnodes in + * the range of indexes @from-@to and captures them, returns number of captured + * jnodes, updates @from to next jnode after the last captured one. + */ +static int +capture_anonymous_jnodes(struct address_space *mapping, + pgoff_t *from, pgoff_t to, int to_capture) +{ + *from = to; + return 0; +} + +/* + * Commit atom of the jnode of a page. + */ +static int sync_page(struct page *page) +{ + int result; + do { + jnode *node; + txn_atom *atom; + + lock_page(page); + node = jprivate(page); + if (node != NULL) { + spin_lock_jnode(node); + atom = jnode_get_atom(node); + spin_unlock_jnode(node); + } else + atom = NULL; + unlock_page(page); + result = reiser4_sync_atom(atom); + } while (result == -E_REPEAT); + /* + * ZAM-FIXME-HANS: document the logic of this loop, is it just to + * handle the case where more pages get added to the atom while we are + * syncing it? + */ + assert("nikita-3485", ergo(result == 0, + get_current_context()->trans->atom == NULL)); + return result; +} + +/* + * Commit atoms of pages on @pages list. + * call sync_page for each page from mapping's page tree + */ +static int sync_page_list(struct inode *inode) +{ + int result; + struct address_space *mapping; + unsigned long from; /* start index for radix_tree_gang_lookup */ + unsigned int found; /* return value for radix_tree_gang_lookup */ + + mapping = inode->i_mapping; + from = 0; + result = 0; + spin_lock_irq(&mapping->tree_lock); + while (result == 0) { + struct page *page; + + found = + radix_tree_gang_lookup(&mapping->page_tree, (void **)&page, + from, 1); + assert("edward-1550", found < 2); + if (found == 0) + break; + /** + * page may not leave radix tree because it is protected from + * truncating by inode->i_mutex locked by sys_fsync + */ + page_cache_get(page); + spin_unlock_irq(&mapping->tree_lock); + + from = page->index + 1; + + result = sync_page(page); + + page_cache_release(page); + spin_lock_irq(&mapping->tree_lock); + } + + spin_unlock_irq(&mapping->tree_lock); + return result; +} + +static int commit_file_atoms(struct inode *inode) +{ + int result; + struct unix_file_info *uf_info; + + uf_info = unix_file_inode_data(inode); + + get_exclusive_access(uf_info); + /* + * find what items file is made from + */ + result = find_file_state(inode, uf_info); + drop_exclusive_access(uf_info); + if (result != 0) + return result; + + /* + * file state cannot change because we are under ->i_mutex + */ + switch (uf_info->container) { + case UF_CONTAINER_EXTENTS: + /* find_file_state might open join an atom */ + reiser4_txn_restart_current(); + result = + /* + * when we are called by + * filemap_fdatawrite-> + * do_writepages()-> + * reiser4_writepages() + * + * inode->i_mapping->dirty_pages are spices into + * ->io_pages, leaving ->dirty_pages dirty. + * + * When we are called from + * reiser4_fsync()->sync_unix_file(), we have to + * commit atoms of all pages on the ->dirty_list. + * + * So for simplicity we just commit ->io_pages and + * ->dirty_pages. + */ + sync_page_list(inode); + break; + case UF_CONTAINER_TAILS: + /* + * NOTE-NIKITA probably we can be smarter for tails. For now + * just commit all existing atoms. + */ + result = txnmgr_force_commit_all(inode->i_sb, 0); + break; + case UF_CONTAINER_EMPTY: + result = 0; + break; + case UF_CONTAINER_UNKNOWN: + default: + result = -EIO; + break; + } + + /* + * commit current transaction: there can be captured nodes from + * find_file_state() and finish_conversion(). + */ + reiser4_txn_restart_current(); + return result; +} + +/** + * writepages_unix_file - writepages of struct address_space_operations + * @mapping: + * @wbc: + * + * This captures anonymous pages and anonymous jnodes. Anonymous pages are + * pages which are dirtied via mmapping. Anonymous jnodes are ones which were + * created by reiser4_writepage. + */ +int writepages_unix_file(struct address_space *mapping, + struct writeback_control *wbc) +{ + int result; + struct unix_file_info *uf_info; + pgoff_t pindex, jindex, nr_pages; + long to_capture; + struct inode *inode; + + inode = mapping->host; + if (!has_anonymous_pages(inode)) { + result = 0; + goto end; + } + jindex = pindex = wbc->range_start >> PAGE_CACHE_SHIFT; + result = 0; + nr_pages = size_in_pages(i_size_read(inode)); + + uf_info = unix_file_inode_data(inode); + + do { + reiser4_context *ctx; + + if (wbc->sync_mode != WB_SYNC_ALL) + to_capture = min(wbc->nr_to_write, CAPTURE_APAGE_BURST); + else + to_capture = CAPTURE_APAGE_BURST; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) { + result = PTR_ERR(ctx); + break; + } + /* avoid recursive calls to ->sync_inodes */ + ctx->nobalance = 1; + assert("zam-760", lock_stack_isclean(get_current_lock_stack())); + assert("edward-1551", LOCK_CNT_NIL(inode_sem_w)); + assert("edward-1552", LOCK_CNT_NIL(inode_sem_r)); + + reiser4_txn_restart_current(); + + /* we have to get nonexclusive access to the file */ + if (get_current_context()->entd) { + /* + * use nonblocking version of nonexclusive_access to + * avoid deadlock which might look like the following: + * process P1 holds NEA on file F1 and called entd to + * reclaim some memory. Entd works for P1 and is going + * to capture pages of file F2. To do that entd has to + * get NEA to F2. F2 is held by process P2 which also + * called entd. But entd is serving P1 at the moment + * and P2 has to wait. Process P3 trying to get EA to + * file F2. Existence of pending EA request to file F2 + * makes impossible for entd to get NEA to file + * F2. Neither of these process can continue. Using + * nonblocking version of gettign NEA is supposed to + * avoid this deadlock. + */ + if (try_to_get_nonexclusive_access(uf_info) == 0) { + result = RETERR(-EBUSY); + reiser4_exit_context(ctx); + break; + } + } else + get_nonexclusive_access(uf_info); + + while (to_capture > 0) { + pgoff_t start; + + assert("vs-1727", jindex <= pindex); + if (pindex == jindex) { + start = pindex; + result = + capture_anonymous_pages(inode->i_mapping, + &pindex, + to_capture); + if (result <= 0) + break; + to_capture -= result; + wbc->nr_to_write -= result; + if (start + result == pindex) { + jindex = pindex; + continue; + } + if (to_capture <= 0) + break; + } + /* deal with anonymous jnodes between jindex and pindex */ + result = + capture_anonymous_jnodes(inode->i_mapping, &jindex, + pindex, to_capture); + if (result < 0) + break; + to_capture -= result; + get_current_context()->nr_captured += result; + + if (jindex == (pgoff_t) - 1) { + assert("vs-1728", pindex == (pgoff_t) - 1); + break; + } + } + if (to_capture <= 0) + /* there may be left more pages */ + __mark_inode_dirty(inode, I_DIRTY_PAGES); + + drop_nonexclusive_access(uf_info); + if (result < 0) { + /* error happened */ + reiser4_exit_context(ctx); + return result; + } + if (wbc->sync_mode != WB_SYNC_ALL) { + reiser4_exit_context(ctx); + return 0; + } + result = commit_file_atoms(inode); + reiser4_exit_context(ctx); + if (pindex >= nr_pages && jindex == pindex) + break; + } while (1); + + end: + if (is_in_reiser4_context()) { + if (get_current_context()->nr_captured >= CAPTURE_APAGE_BURST) { + /* + * there are already pages to flush, flush them out, do + * not delay until end of reiser4_sync_inodes + */ + reiser4_writeout(inode->i_sb, wbc); + get_current_context()->nr_captured = 0; + } + } + return result; +} + +/** + * readpage_unix_file_nolock - readpage of struct address_space_operations + * @file: + * @page: + * + * Compose a key and search for item containing information about @page + * data. If item is found - its readpage method is called. + */ +int readpage_unix_file(struct file *file, struct page *page) +{ + reiser4_context *ctx; + int result; + struct inode *inode; + reiser4_key key; + item_plugin *iplug; + hint_t *hint; + lock_handle *lh; + coord_t *coord; + + assert("vs-1062", PageLocked(page)); + assert("vs-976", !PageUptodate(page)); + assert("vs-1061", page->mapping && page->mapping->host); + + if (page->mapping->host->i_size <= page_offset(page)) { + /* page is out of file */ + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + unlock_page(page); + return 0; + } + + inode = page->mapping->host; + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) { + unlock_page(page); + return PTR_ERR(ctx); + } + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) { + unlock_page(page); + reiser4_exit_context(ctx); + return RETERR(-ENOMEM); + } + + result = load_file_hint(file, hint); + if (result) { + kfree(hint); + unlock_page(page); + reiser4_exit_context(ctx); + return result; + } + lh = &hint->lh; + + /* get key of first byte of the page */ + key_by_inode_and_offset_common(inode, page_offset(page), &key); + + /* look for file metadata corresponding to first byte of page */ + page_cache_get(page); + unlock_page(page); + result = find_file_item(hint, &key, ZNODE_READ_LOCK, inode); + lock_page(page); + page_cache_release(page); + + if (page->mapping == NULL) { + /* + * readpage allows truncate to run concurrently. Page was + * truncated while it was not locked + */ + done_lh(lh); + kfree(hint); + unlock_page(page); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return -EINVAL; + } + + if (result != CBK_COORD_FOUND || hint->ext_coord.coord.between != AT_UNIT) { + if (result == CBK_COORD_FOUND && + hint->ext_coord.coord.between != AT_UNIT) + /* file is truncated */ + result = -EINVAL; + done_lh(lh); + kfree(hint); + unlock_page(page); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return result; + } + + /* + * item corresponding to page is found. It can not be removed because + * znode lock is held + */ + if (PageUptodate(page)) { + done_lh(lh); + kfree(hint); + unlock_page(page); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return 0; + } + + coord = &hint->ext_coord.coord; + result = zload(coord->node); + if (result) { + done_lh(lh); + kfree(hint); + unlock_page(page); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return result; + } + + validate_extended_coord(&hint->ext_coord, page_offset(page)); + + if (!coord_is_existing_unit(coord)) { + /* this indicates corruption */ + warning("vs-280", + "Looking for page %lu of file %llu (size %lli). " + "No file items found (%d). File is corrupted?\n", + page->index, (unsigned long long)get_inode_oid(inode), + inode->i_size, result); + zrelse(coord->node); + done_lh(lh); + kfree(hint); + unlock_page(page); + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return RETERR(-EIO); + } + + /* + * get plugin of found item or use plugin if extent if there are no + * one + */ + iplug = item_plugin_by_coord(coord); + if (iplug->s.file.readpage) + result = iplug->s.file.readpage(coord, page); + else + result = RETERR(-EINVAL); + + if (!result) { + set_key_offset(&key, + (loff_t) (page->index + 1) << PAGE_CACHE_SHIFT); + /* FIXME should call reiser4_set_hint() */ + reiser4_unset_hint(hint); + } else { + unlock_page(page); + reiser4_unset_hint(hint); + } + assert("vs-979", + ergo(result == 0, (PageLocked(page) || PageUptodate(page)))); + assert("vs-9791", ergo(result != 0, !PageLocked(page))); + + zrelse(coord->node); + done_lh(lh); + + save_file_hint(file, hint); + kfree(hint); + + /* + * FIXME: explain why it is needed. HINT: page allocation in write can + * not be done when atom is not NULL because reiser4_writepage can not + * kick entd and have to eflush + */ + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return result; +} + +struct uf_readpages_context { + lock_handle lh; + coord_t coord; +}; + +/* A callback function for readpages_unix_file/read_cache_pages. + * If the file is build of tails, then return error (-ENOENT). + * + * @data -- a pointer to reiser4_readpages_context object, + * to save the twig lock and the coord between + * read_cache_page iterations. + * @page -- page to start read. + */ +static int uf_readpages_filler(void * data, struct page * page) +{ + struct uf_readpages_context *rc = data; + jnode * node; + int ret = 0; + reiser4_extent *ext; + __u64 ext_index; + int cbk_done = 0; + struct address_space * mapping = page->mapping; + + if (PageUptodate(page)) { + unlock_page(page); + return 0; + } + page_cache_get(page); + + if (rc->lh.node == 0) { + /* no twig lock - have to do tree search. */ + reiser4_key key; + repeat: + unlock_page(page); + key_by_inode_and_offset_common( + mapping->host, page_offset(page), &key); + ret = coord_by_key( + &get_super_private(mapping->host->i_sb)->tree, + &key, &rc->coord, &rc->lh, + ZNODE_READ_LOCK, FIND_EXACT, + TWIG_LEVEL, TWIG_LEVEL, CBK_UNIQUE, NULL); + if (unlikely(ret)) + goto exit; + lock_page(page); + if (PageUptodate(page)) + goto unlock; + cbk_done = 1; + } + ret = zload(rc->coord.node); + if (unlikely(ret)) + goto unlock; + if (!coord_is_existing_item(&rc->coord) || + !item_is_extent(&rc->coord)) { + zrelse(rc->coord.node); + ret = RETERR(-EIO); + goto unlock; + } + ext = extent_by_coord(&rc->coord); + ext_index = extent_unit_index(&rc->coord); + if (page->index < ext_index || + page->index >= ext_index + extent_get_width(ext)) { + /* the page index doesn't belong to the extent unit + which the coord points to - release the lock and + repeat with tree search. */ + zrelse(rc->coord.node); + done_lh(&rc->lh); + /* we can be here after a CBK call only in case of + corruption of the tree or the tree lookup algorithm bug. */ + if (unlikely(cbk_done)) { + ret = RETERR(-EIO); + goto unlock; + } + goto repeat; + } + node = jnode_of_page(page); + if (unlikely(IS_ERR(node))) { + zrelse(rc->coord.node); + ret = PTR_ERR(node); + goto unlock; + } + ret = reiser4_do_readpage_extent(ext, page->index - ext_index, page); + jput(node); + zrelse(rc->coord.node); + if (likely(!ret)) + goto exit; + unlock: + unlock_page(page); + exit: + page_cache_release(page); + return ret; +} + +/** + * readpages_unix_file - called by the readahead code, starts reading for each + * page of given list of pages + */ +int readpages_unix_file( + struct file *file, struct address_space *mapping, + struct list_head *pages, unsigned nr_pages) +{ + reiser4_context *ctx; + struct uf_readpages_context rc; + int ret; + + ctx = reiser4_init_context(mapping->host->i_sb); + if (IS_ERR(ctx)) { + put_pages_list(pages); + return PTR_ERR(ctx); + } + init_lh(&rc.lh); + ret = read_cache_pages(mapping, pages, uf_readpages_filler, &rc); + done_lh(&rc.lh); + context_set_commit_async(ctx); + /* close the transaction to protect further page allocation from deadlocks */ + reiser4_txn_restart(ctx); + reiser4_exit_context(ctx); + return ret; +} + +static reiser4_block_nr unix_file_estimate_read(struct inode *inode, + loff_t count UNUSED_ARG) +{ + /* We should reserve one block, because of updating of the stat data + item */ + assert("vs-1249", + inode_file_plugin(inode)->estimate.update == + estimate_update_common); + return estimate_update_common(inode); +} + +/* this is called with nonexclusive access obtained, file's container can not change */ +static ssize_t read_file(hint_t *hint, struct file *file, /* file to read from to */ + char __user *buf, /* address of user-space buffer */ + size_t count, /* number of bytes to read */ + loff_t *off) +{ + int result; + struct inode *inode; + flow_t flow; + int (*read_f) (struct file *, flow_t *, hint_t *); + coord_t *coord; + znode *loaded; + + inode = file->f_dentry->d_inode; + + /* build flow */ + assert("vs-1250", + inode_file_plugin(inode)->flow_by_inode == + flow_by_inode_unix_file); + result = + flow_by_inode_unix_file(inode, buf, 1 /* user space */ , count, + *off, READ_OP, &flow); + if (unlikely(result)) + return result; + + /* get seal and coord sealed with it from reiser4 private data + of struct file. The coord will tell us where our last read + of this file finished, and the seal will help to determine + if that location is still valid. + */ + coord = &hint->ext_coord.coord; + while (flow.length && result == 0) { + result = + find_file_item(hint, &flow.key, ZNODE_READ_LOCK, inode); + if (cbk_errored(result)) + /* error happened */ + break; + + if (coord->between != AT_UNIT) { + /* there were no items corresponding to given offset */ + done_lh(hint->ext_coord.lh); + break; + } + + loaded = coord->node; + result = zload(loaded); + if (unlikely(result)) { + done_lh(hint->ext_coord.lh); + break; + } + + if (hint->ext_coord.valid == 0) + validate_extended_coord(&hint->ext_coord, + get_key_offset(&flow.key)); + + assert("vs-4", hint->ext_coord.valid == 1); + assert("vs-33", hint->ext_coord.lh == &hint->lh); + /* call item's read method */ + read_f = item_plugin_by_coord(coord)->s.file.read; + result = read_f(file, &flow, hint); + zrelse(loaded); + done_lh(hint->ext_coord.lh); + } + + return (count - flow.length) ? (count - flow.length) : result; +} + +static ssize_t read_unix_file_container_tails(struct file*, char __user*, size_t, loff_t*); + +/** + * read_unix_file - read of struct file_operations + * @file: file to read from + * @buf: address of user-space buffer + * @read_amount: number of bytes to read + * @off: position in file to read from + * + * This is implementation of vfs's read method of struct file_operations for + * unix file plugin. + */ +ssize_t read_unix_file(struct file *file, char __user *buf, size_t read_amount, + loff_t *off) +{ + reiser4_context *ctx; + ssize_t result; + struct inode *inode; + struct unix_file_info *uf_info; + + if (unlikely(read_amount == 0)) + return 0; + + assert("umka-072", file != NULL); + assert("umka-074", off != NULL); + inode = file->f_dentry->d_inode; + assert("vs-972", !reiser4_inode_get_flag(inode, REISER4_NO_SD)); + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + uf_info = unix_file_inode_data(inode); + if (uf_info->container == UF_CONTAINER_UNKNOWN) { + get_exclusive_access(uf_info); + result = find_file_state(inode, uf_info); + if (unlikely(result != 0)) + goto out; + } else + get_nonexclusive_access(uf_info); + result = reiser4_grab_space_force(unix_file_estimate_read(inode, read_amount), + BA_CAN_COMMIT); + if (unlikely(result != 0)) + goto out; + if (uf_info->container == UF_CONTAINER_EXTENTS){ + result = do_sync_read(file, buf, read_amount, off); + } else if (uf_info->container == UF_CONTAINER_TAILS || + reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV) || + reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) { + result = read_unix_file_container_tails(file, buf, read_amount, off); + } else { + assert("zam-1085", uf_info->container == UF_CONTAINER_EMPTY); + result = 0; + } +out: + drop_access(uf_info); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} + +static ssize_t read_unix_file_container_tails( + struct file *file, char __user *buf, size_t read_amount, loff_t *off) +{ + int result; + struct inode *inode; + hint_t *hint; + struct unix_file_info *uf_info; + size_t count, read, left; + loff_t size; + + assert("umka-072", file != NULL); + assert("umka-074", off != NULL); + inode = file->f_dentry->d_inode; + assert("vs-972", !reiser4_inode_get_flag(inode, REISER4_NO_SD)); + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) + return RETERR(-ENOMEM); + + result = load_file_hint(file, hint); + if (result) { + kfree(hint); + return result; + } + + left = read_amount; + count = 0; + uf_info = unix_file_inode_data(inode); + while (left > 0) { + reiser4_txn_restart_current(); + size = i_size_read(inode); + if (*off >= size) + /* position to read from is past the end of file */ + break; + if (*off + left > size) + left = size - *off; + /* faultin user page */ + result = fault_in_pages_writeable(buf, left > PAGE_CACHE_SIZE ? PAGE_CACHE_SIZE : left); + if (result) + return RETERR(-EFAULT); + + read = read_file(hint, file, buf, + left > PAGE_CACHE_SIZE ? PAGE_CACHE_SIZE : left, + off); + if (read < 0) { + result = read; + break; + } + left -= read; + buf += read; + + /* update position in a file */ + *off += read; + /* total number of read bytes */ + count += read; + } + done_lh(&hint->lh); + save_file_hint(file, hint); + kfree(hint); + if (count) + file_accessed(file); + /* return number of read bytes or error code if nothing is read */ + return count ? count : result; +} + +/* This function takes care about @file's pages. First of all it checks if + filesystems readonly and if so gets out. Otherwise, it throws out all + pages of file if it was mapped for read and going to be mapped for write + and consists of tails. This is done in order to not manage few copies + of the data (first in page cache and second one in tails them selves) + for the case of mapping files consisting tails. + + Here also tail2extent conversion is performed if it is allowed and file + is going to be written or mapped for write. This functions may be called + from write_unix_file() or mmap_unix_file(). */ +static int check_pages_unix_file(struct file *file, struct inode *inode) +{ + reiser4_invalidate_pages(inode->i_mapping, 0, + (inode->i_size + PAGE_CACHE_SIZE - + 1) >> PAGE_CACHE_SHIFT, 0); + return unpack(file, inode, 0 /* not forever */ ); +} + +/** + * mmap_unix_file - mmap of struct file_operations + * @file: file to mmap + * @vma: + * + * This is implementation of vfs's mmap method of struct file_operations for + * unix file plugin. It converts file to extent if necessary. Sets + * reiser4_inode's flag - REISER4_HAS_MMAP. + */ +int mmap_unix_file(struct file *file, struct vm_area_struct *vma) +{ + reiser4_context *ctx; + int result; + struct inode *inode; + struct unix_file_info *uf_info; + reiser4_block_nr needed; + + inode = file->f_dentry->d_inode; + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + uf_info = unix_file_inode_data(inode); + + get_exclusive_access_careful(uf_info, inode); + + if (!IS_RDONLY(inode) && (vma->vm_flags & (VM_MAYWRITE | VM_SHARED))) { + /* + * we need file built of extent items. If it is still built of + * tail items we have to convert it. Find what items the file + * is built of + */ + result = find_file_state(inode, uf_info); + if (result != 0) { + drop_exclusive_access(uf_info); + reiser4_exit_context(ctx); + return result; + } + + assert("vs-1648", (uf_info->container == UF_CONTAINER_TAILS || + uf_info->container == UF_CONTAINER_EXTENTS || + uf_info->container == UF_CONTAINER_EMPTY)); + if (uf_info->container == UF_CONTAINER_TAILS) { + /* + * invalidate all pages and convert file from tails to + * extents + */ + result = check_pages_unix_file(file, inode); + if (result) { + drop_exclusive_access(uf_info); + reiser4_exit_context(ctx); + return result; + } + } + } + + /* + * generic_file_mmap will do update_atime. Grab space for stat data + * update. + */ + needed = inode_file_plugin(inode)->estimate.update(inode); + result = reiser4_grab_space_force(needed, BA_CAN_COMMIT); + if (result) { + drop_exclusive_access(uf_info); + reiser4_exit_context(ctx); + return result; + } + + result = generic_file_mmap(file, vma); + if (result == 0) { + /* mark file as having mapping. */ + reiser4_inode_set_flag(inode, REISER4_HAS_MMAP); + } + + drop_exclusive_access(uf_info); + reiser4_exit_context(ctx); + return result; +} + +/** + * find_first_item + * @inode: + * + * Finds file item which is responsible for first byte in the file. + */ +static int find_first_item(struct inode *inode) +{ + coord_t coord; + lock_handle lh; + reiser4_key key; + int result; + + coord_init_zero(&coord); + init_lh(&lh); + inode_file_plugin(inode)->key_by_inode(inode, 0, &key); + result = find_file_item_nohint(&coord, &lh, &key, ZNODE_READ_LOCK, + inode); + if (result == CBK_COORD_FOUND) { + if (coord.between == AT_UNIT) { + result = zload(coord.node); + if (result == 0) { + result = item_id_by_coord(&coord); + zrelse(coord.node); + if (result != EXTENT_POINTER_ID && + result != FORMATTING_ID) + result = RETERR(-EIO); + } + } else + result = RETERR(-EIO); + } + done_lh(&lh); + return result; +} + +/** + * open_unix_file + * @inode: + * @file: + * + * If filesystem is not readonly - complete uncompleted tail conversion if + * there was one + */ +int open_unix_file(struct inode *inode, struct file *file) +{ + int result; + reiser4_context *ctx; + struct unix_file_info *uf_info; + + if (IS_RDONLY(inode)) + return 0; + + if (!reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) + return 0; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + uf_info = unix_file_inode_data(inode); + + get_exclusive_access_careful(uf_info, inode); + + if (!reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) { + /* + * other process completed the conversion + */ + drop_exclusive_access(uf_info); + reiser4_exit_context(ctx); + return 0; + } + + /* + * file left in semi converted state after unclean shutdown or another + * thread is doing conversion and dropped exclusive access which doing + * balance dirty pages. Complete the conversion + */ + result = find_first_item(inode); + if (result == EXTENT_POINTER_ID) + /* + * first item is extent, therefore there was incomplete + * tail2extent conversion. Complete it + */ + result = tail2extent(unix_file_inode_data(inode)); + else if (result == FORMATTING_ID) + /* + * first item is formatting item, therefore there was + * incomplete extent2tail conversion. Complete it + */ + result = extent2tail(file, unix_file_inode_data(inode)); + else + result = -EIO; + + assert("vs-1712", + ergo(result == 0, + (!reiser4_inode_get_flag(inode, REISER4_PART_MIXED) && + !reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV)))); + drop_exclusive_access(uf_info); + reiser4_exit_context(ctx); + return result; +} + +#define NEITHER_OBTAINED 0 +#define EA_OBTAINED 1 +#define NEA_OBTAINED 2 + +static void drop_access(struct unix_file_info *uf_info) +{ + if (uf_info->exclusive_use) + drop_exclusive_access(uf_info); + else + drop_nonexclusive_access(uf_info); +} + +#define debug_wuf(format, ...) printk("%s: %d: %s: " format "\n", \ + __FILE__, __LINE__, __FUNCTION__, ## __VA_ARGS__) + +/** + * write_unix_file - private ->write() method of unix_file plugin. + * + * @file: file to write to + * @buf: address of user-space buffer + * @count: number of bytes to write + * @pos: position in file to write to + * @cont: unused argument, as we don't perform plugin conversion when being + * managed by unix_file plugin. + */ +ssize_t write_unix_file(struct file *file, const char __user *buf, + size_t count, loff_t *pos, struct psched_context *cont) +{ + int result; + reiser4_context *ctx; + struct inode *inode; + struct unix_file_info *uf_info; + ssize_t written; + int try_free_space; + int to_write = PAGE_CACHE_SIZE * WRITE_GRANULARITY; + size_t left; + ssize_t (*write_op)(struct file *, struct inode *, + const char __user *, size_t, + loff_t *pos); + int ea; + loff_t new_size; + + ctx = get_current_context(); + inode = file->f_dentry->d_inode; + + assert("vs-947", !reiser4_inode_get_flag(inode, REISER4_NO_SD)); + assert("vs-9471", (!reiser4_inode_get_flag(inode, REISER4_PART_MIXED))); + + /* check amount of bytes to write and writing position */ + result = generic_write_checks(file, pos, &count, 0); + if (result) { + context_set_commit_async(ctx); + return result; + } + + result = file_remove_suid(file); + if (result) { + context_set_commit_async(ctx); + return result; + } + /* remove_suid might create a transaction */ + reiser4_txn_restart(ctx); + + uf_info = unix_file_inode_data(inode); + + current->backing_dev_info = inode->i_mapping->backing_dev_info; + written = 0; + try_free_space = 0; + left = count; + ea = NEITHER_OBTAINED; + + new_size = i_size_read(inode); + if (*pos + count > new_size) + new_size = *pos + count; + + while (left) { + if (left < to_write) + to_write = left; + + if (uf_info->container == UF_CONTAINER_EMPTY) { + get_exclusive_access(uf_info); + ea = EA_OBTAINED; + if (uf_info->container != UF_CONTAINER_EMPTY) { + /* file is made not empty by another process */ + drop_exclusive_access(uf_info); + ea = NEITHER_OBTAINED; + continue; + } + } else if (uf_info->container == UF_CONTAINER_UNKNOWN) { + /* + * get exclusive access directly just to not have to + * re-obtain it if file will appear empty + */ + get_exclusive_access(uf_info); + ea = EA_OBTAINED; + result = find_file_state(inode, uf_info); + if (result) { + drop_exclusive_access(uf_info); + ea = NEITHER_OBTAINED; + break; + } + } else { + get_nonexclusive_access(uf_info); + ea = NEA_OBTAINED; + } + + /* either EA or NEA is obtained. Choose item write method */ + if (uf_info->container == UF_CONTAINER_EXTENTS) { + /* file is built of extent items */ + write_op = reiser4_write_extent; + } else if (uf_info->container == UF_CONTAINER_EMPTY) { + /* file is empty */ + if (should_have_notail(uf_info, new_size)) + write_op = reiser4_write_extent; + else + write_op = reiser4_write_tail; + } else { + /* file is built of tail items */ + if (should_have_notail(uf_info, new_size)) { + if (ea == NEA_OBTAINED) { + drop_nonexclusive_access(uf_info); + get_exclusive_access(uf_info); + ea = EA_OBTAINED; + } + if (uf_info->container == UF_CONTAINER_TAILS) { + /* + * if file is being convered by another + * process - wait until it completes + */ + while (1) { + if (reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)) { + drop_exclusive_access(uf_info); + schedule(); + get_exclusive_access(uf_info); + continue; + } + break; + } + if (uf_info->container == UF_CONTAINER_TAILS) { + result = tail2extent(uf_info); + if (result) { + drop_exclusive_access(uf_info); + context_set_commit_async(ctx); + break; + } + } + } + drop_exclusive_access(uf_info); + ea = NEITHER_OBTAINED; + continue; + } + write_op = reiser4_write_tail; + } + + written = write_op(file, inode, buf, to_write, pos); + if (written == -ENOSPC && try_free_space) { + drop_access(uf_info); + txnmgr_force_commit_all(inode->i_sb, 0); + try_free_space = 0; + continue; + } + if (written < 0) { + drop_access(uf_info); + result = written; + break; + } + /* something is written. */ + if (uf_info->container == UF_CONTAINER_EMPTY) { + assert("edward-1553", ea == EA_OBTAINED); + uf_info->container = + (write_op == reiser4_write_extent) ? + UF_CONTAINER_EXTENTS : UF_CONTAINER_TAILS; + } else { + assert("edward-1554", ergo(uf_info->container == UF_CONTAINER_EXTENTS, + write_op == reiser4_write_extent)); + assert("edward-1555", ergo(uf_info->container == UF_CONTAINER_TAILS, + write_op == reiser4_write_tail)); + } + if (*pos + written > inode->i_size) + INODE_SET_FIELD(inode, i_size, *pos + written); + file_update_time(file); + result = reiser4_update_sd(inode); + if (result) { + current->backing_dev_info = NULL; + drop_access(uf_info); + context_set_commit_async(ctx); + break; + } + drop_access(uf_info); + ea = NEITHER_OBTAINED; + reiser4_txn_restart(ctx); + current->journal_info = NULL; + /* + * tell VM how many pages were dirtied. Maybe number of pages + * which were dirty already should not be counted + */ + balance_dirty_pages_ratelimited_nr(inode->i_mapping, + (written + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE); + current->journal_info = ctx; + + left -= written; + buf += written; + *pos += written; + } + if (result == 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { + reiser4_txn_restart_current(); + grab_space_enable(); + result = reiser4_sync_file_common(file, file->f_dentry, + 0 /* data and stat data */); + if (result) + warning("reiser4-7", "failed to sync file %llu", + (unsigned long long)get_inode_oid(inode)); + } + + current->backing_dev_info = NULL; + + /* + * return number of written bytes or error code if nothing is + * written. Note, that it does not work correctly in case when + * sync_unix_file returns error + */ + return (count - left) ? (count - left) : result; +} + +/** + * release_unix_file - release of struct file_operations + * @inode: inode of released file + * @file: file to release + * + * Implementation of release method of struct file_operations for unix file + * plugin. If last reference to indode is released - convert all extent items + * into tail items if necessary. Frees reiser4 specific file data. + */ +int release_unix_file(struct inode *inode, struct file *file) +{ + reiser4_context *ctx; + struct unix_file_info *uf_info; + int result; + int in_reiser4; + + in_reiser4 = is_in_reiser4_context(); + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + result = 0; + if (in_reiser4 == 0) { + uf_info = unix_file_inode_data(inode); + + get_exclusive_access_careful(uf_info, inode); + if (atomic_read(&file->f_dentry->d_count) == 1 && + uf_info->container == UF_CONTAINER_EXTENTS && + !should_have_notail(uf_info, inode->i_size) && + !rofs_inode(inode)) { + result = extent2tail(file, uf_info); + if (result != 0) { + context_set_commit_async(ctx); + warning("nikita-3233", + "Failed (%d) to convert in %s (%llu)", + result, __FUNCTION__, + (unsigned long long) + get_inode_oid(inode)); + } + } + drop_exclusive_access(uf_info); + } else { + /* + we are within reiser4 context already. How latter is + possible? Simple: + + (gdb) bt + #0 get_exclusive_access () + #2 0xc01e56d3 in release_unix_file () + #3 0xc01c3643 in reiser4_release () + #4 0xc014cae0 in __fput () + #5 0xc013ffc3 in remove_vm_struct () + #6 0xc0141786 in exit_mmap () + #7 0xc0118480 in mmput () + #8 0xc0133205 in oom_kill () + #9 0xc01332d1 in out_of_memory () + #10 0xc013bc1d in try_to_free_pages () + #11 0xc013427b in __alloc_pages () + #12 0xc013f058 in do_anonymous_page () + #13 0xc013f19d in do_no_page () + #14 0xc013f60e in handle_mm_fault () + #15 0xc01131e5 in do_page_fault () + #16 0xc0104935 in error_code () + #17 0xc025c0c6 in __copy_to_user_ll () + #18 0xc01d496f in reiser4_read_tail () + #19 0xc01e4def in read_unix_file () + #20 0xc01c3504 in reiser4_read () + #21 0xc014bd4f in vfs_read () + #22 0xc014bf66 in sys_read () + */ + warning("vs-44", "out of memory?"); + } + + reiser4_free_file_fsdata(file); + + reiser4_exit_context(ctx); + return result; +} + +static void set_file_notail(struct inode *inode) +{ + reiser4_inode *state; + formatting_plugin *tplug; + + state = reiser4_inode_data(inode); + tplug = formatting_plugin_by_id(NEVER_TAILS_FORMATTING_ID); + force_plugin_pset(inode, PSET_FORMATTING, (reiser4_plugin *)tplug); +} + +/* if file is built of tails - convert it to extents */ +static int unpack(struct file *filp, struct inode *inode, int forever) +{ + int result = 0; + struct unix_file_info *uf_info; + + uf_info = unix_file_inode_data(inode); + assert("vs-1628", ea_obtained(uf_info)); + + result = find_file_state(inode, uf_info); + if (result) + return result; + assert("vs-1074", uf_info->container != UF_CONTAINER_UNKNOWN); + + if (uf_info->container == UF_CONTAINER_TAILS) { + /* + * if file is being convered by another process - wait until it + * completes + */ + while (1) { + if (reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)) { + drop_exclusive_access(uf_info); + schedule(); + get_exclusive_access(uf_info); + continue; + } + break; + } + if (uf_info->container == UF_CONTAINER_TAILS) { + result = tail2extent(uf_info); + if (result) + return result; + } + } + if (forever) { + /* safe new formatting plugin in stat data */ + __u64 tograb; + + set_file_notail(inode); + + grab_space_enable(); + tograb = inode_file_plugin(inode)->estimate.update(inode); + result = reiser4_grab_space(tograb, BA_CAN_COMMIT); + result = reiser4_update_sd(inode); + } + + return result; +} + +/* implentation of vfs' ioctl method of struct file_operations for unix file + plugin +*/ +int +ioctl_unix_file(struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg UNUSED_ARG) +{ + reiser4_context *ctx; + int result; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + switch (cmd) { + case REISER4_IOC_UNPACK: + get_exclusive_access(unix_file_inode_data(inode)); + result = unpack(filp, inode, 1 /* forever */ ); + drop_exclusive_access(unix_file_inode_data(inode)); + break; + + default: + result = RETERR(-ENOSYS); + break; + } + reiser4_exit_context(ctx); + return result; +} + +/* implentation of vfs' bmap method of struct address_space_operations for unix + file plugin +*/ +sector_t bmap_unix_file(struct address_space * mapping, sector_t lblock) +{ + reiser4_context *ctx; + sector_t result; + reiser4_key key; + coord_t coord; + lock_handle lh; + struct inode *inode; + item_plugin *iplug; + sector_t block; + + inode = mapping->host; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + key_by_inode_and_offset_common(inode, + (loff_t) lblock * current_blocksize, + &key); + + init_lh(&lh); + result = + find_file_item_nohint(&coord, &lh, &key, ZNODE_READ_LOCK, inode); + if (cbk_errored(result)) { + done_lh(&lh); + reiser4_exit_context(ctx); + return result; + } + + result = zload(coord.node); + if (result) { + done_lh(&lh); + reiser4_exit_context(ctx); + return result; + } + + iplug = item_plugin_by_coord(&coord); + if (iplug->s.file.get_block) { + result = iplug->s.file.get_block(&coord, lblock, &block); + if (result == 0) + result = block; + } else + result = RETERR(-EINVAL); + + zrelse(coord.node); + done_lh(&lh); + reiser4_exit_context(ctx); + return result; +} + +/** + * flow_by_inode_unix_file - initizlize structure flow + * @inode: inode of file for which read or write is abou + * @buf: buffer to perform read to or write from + * @user: flag showing whether @buf is user space or kernel space + * @size: size of buffer @buf + * @off: start offset fro read or write + * @op: READ or WRITE + * @flow: + * + * Initializes fields of @flow: key, size of data, i/o mode (read or write). + */ +int flow_by_inode_unix_file(struct inode *inode, + const char __user *buf, int user, + loff_t size, loff_t off, + rw_op op, flow_t *flow) +{ + assert("nikita-1100", inode != NULL); + + flow->length = size; + memcpy(&flow->data, &buf, sizeof(buf)); + flow->user = user; + flow->op = op; + assert("nikita-1931", inode_file_plugin(inode) != NULL); + assert("nikita-1932", + inode_file_plugin(inode)->key_by_inode == + key_by_inode_and_offset_common); + /* calculate key of write position and insert it into flow->key */ + return key_by_inode_and_offset_common(inode, off, &flow->key); +} + +/* plugin->u.file.set_plug_in_sd = NULL + plugin->u.file.set_plug_in_inode = NULL + plugin->u.file.create_blank_sd = NULL */ +/* plugin->u.file.delete */ +/* + plugin->u.file.add_link = reiser4_add_link_common + plugin->u.file.rem_link = NULL */ + +/* plugin->u.file.owns_item + this is common_file_owns_item with assertion */ +/* Audited by: green(2002.06.15) */ +int +owns_item_unix_file(const struct inode *inode /* object to check against */ , + const coord_t * coord /* coord to check */ ) +{ + int result; + + result = owns_item_common(inode, coord); + if (!result) + return 0; + if (!plugin_of_group(item_plugin_by_coord(coord), + UNIX_FILE_METADATA_ITEM_TYPE)) + return 0; + assert("vs-547", + item_id_by_coord(coord) == EXTENT_POINTER_ID || + item_id_by_coord(coord) == FORMATTING_ID); + return 1; +} + +static int setattr_truncate(struct inode *inode, struct iattr *attr) +{ + int result; + int s_result; + loff_t old_size; + reiser4_tree *tree; + + inode_check_scale(inode, inode->i_size, attr->ia_size); + + old_size = inode->i_size; + tree = reiser4_tree_by_inode(inode); + + result = safe_link_grab(tree, BA_CAN_COMMIT); + if (result == 0) + result = safe_link_add(inode, SAFE_TRUNCATE); + if (result == 0) + result = truncate_file_body(inode, attr); + if (result) + warning("vs-1588", "truncate_file failed: oid %lli, " + "old size %lld, new size %lld, retval %d", + (unsigned long long)get_inode_oid(inode), + old_size, attr->ia_size, result); + + s_result = safe_link_grab(tree, BA_CAN_COMMIT); + if (s_result == 0) + s_result = + safe_link_del(tree, get_inode_oid(inode), SAFE_TRUNCATE); + if (s_result != 0) { + warning("nikita-3417", "Cannot kill safelink %lli: %i", + (unsigned long long)get_inode_oid(inode), s_result); + } + safe_link_release(tree); + return result; +} + +/* plugin->u.file.setattr method */ +/* This calls inode_setattr and if truncate is in effect it also takes + exclusive inode access to avoid races */ +int setattr_unix_file(struct dentry *dentry, /* Object to change attributes */ + struct iattr *attr /* change description */ ) +{ + int result; + + if (attr->ia_valid & ATTR_SIZE) { + reiser4_context *ctx; + struct unix_file_info *uf_info; + + /* truncate does reservation itself and requires exclusive + access obtained */ + ctx = reiser4_init_context(dentry->d_inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + uf_info = unix_file_inode_data(dentry->d_inode); + get_exclusive_access_careful(uf_info, dentry->d_inode); + result = setattr_truncate(dentry->d_inode, attr); + drop_exclusive_access(uf_info); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + } else + result = reiser4_setattr_common(dentry, attr); + + return result; +} + +/* plugin->u.file.init_inode_data */ +void +init_inode_data_unix_file(struct inode *inode, + reiser4_object_create_data * crd, int create) +{ + struct unix_file_info *data; + + data = unix_file_inode_data(inode); + data->container = create ? UF_CONTAINER_EMPTY : UF_CONTAINER_UNKNOWN; + init_rwsem(&data->latch); + data->tplug = inode_formatting_plugin(inode); + data->exclusive_use = 0; + +#if REISER4_DEBUG + data->ea_owner = NULL; + atomic_set(&data->nr_neas, 0); +#endif + init_inode_ordering(inode, crd, create); +} + +/** + * delete_unix_file - delete_object of file_plugin + * @inode: inode to be deleted + * + * Truncates file to length 0, removes stat data and safe link. + */ +int delete_object_unix_file(struct inode *inode) +{ + struct unix_file_info *uf_info; + int result; + + if (reiser4_inode_get_flag(inode, REISER4_NO_SD)) + return 0; + + /* truncate file bogy first */ + uf_info = unix_file_inode_data(inode); + get_exclusive_access(uf_info); + result = shorten_file(inode, 0 /* size */ ); + drop_exclusive_access(uf_info); + + if (result) + warning("edward-1556", + "failed to truncate file (%llu) on removal: %d", + get_inode_oid(inode), result); + + /* remove stat data and safe link */ + return reiser4_delete_object_common(inode); +} + +/* plugin->write_begin() */ +int write_begin_unix_file(struct file *file, struct page *page, + unsigned from, unsigned to) +{ + int ret; + struct unix_file_info *info; + + info = unix_file_inode_data(file->f_dentry->d_inode); + get_exclusive_access(info); + ret = find_file_state(file->f_dentry->d_inode, info); + if (likely(ret == 0)) { + if (info->container == UF_CONTAINER_TAILS) + ret = -EINVAL; + else + ret = do_prepare_write(file, page, from, to); + } + drop_exclusive_access(info); + return ret; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/file_conversion.c linux-2.6.30/fs/reiser4/plugin/file/file_conversion.c --- linux-2.6.30.orig/fs/reiser4/plugin/file/file_conversion.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/file_conversion.c 2009-06-22 16:44:21.000000000 +0200 @@ -0,0 +1,779 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, + licensing governed by reiser4/README */ + +/** + * This file contains plugin schedule hooks, and plugin conversion methods. + * + * Plugin schedule hook makes a decision (at plugin schedule point) about the + * most reasonable plugins for managing a regular file. Usually such decisions + * is made by some O(1)-heuristic. + * + * By default we assign a unix_file plugin id when writing incompressible file + * managed by cryptcompress plugin id. Currently used heuristic for estimating + * compressibility is very simple: if first complete logical cluster (64K by + * default) of a file is incompressible, then we make a decision, that the whole + * file is incompressible (*). + * + * To enable a conversion we install a special "magic" compression mode plugin + * (CONVX_COMPRESSION_MODE_ID, see plugin/compress/compress_mode.c for details) + * at file creation time (**). + * + * Note, that we don't perform back conversion (unix_file->cryptcompress) + * because of compatibility reasons (see http://dev.namesys.com/Version4.X.Y + * for details). + * + * The conversion is accompanied by rebuilding disk structures of a file, so it + * is important to protect them from being interacted with other plugins which + * don't expect them to be in such inconsistent state. For this to be protected + * we serialize readers and writers of a file's conversion set (FCS). + * + * We define FCS as a file plugin installed in inode's pset plus file's data + * and metadata that this file plugin manipulates with (items, etc). + * Note, that FCS is defined per file. + * FCS reader is defined as a set of instruction of the following type: + * {inode_file_plugin(inode)->method()} (I.e. retrieving a file plugin id + * conjoined with all method's instructions should be atomic). + * FCS writer is a set of instructions that perform file plugin conversion + * (convert items, update pset, etc). + * Example: + * reiser4_write_careful() supplied to VFS as a ->write() file operation is + * composed of the following (optional) instructions: + * 1 2 3 + * *********************** ####### --------------------------------------------> + * + * 1) "****" are instructions performed on behalf of cryptcompress file plugin; + * 2) "####" is a FCS writer (performing a conversion cryptcompress->unix_file); + * 3) "----" are instructions performed on behalf of unix_file plugin; + * Here (1) and (3) are FCS readers. + * + * In this example FCS readers and writers are already serialized (by design), + * however there can be readers and writers executing at the same time in + * different contexts, so we need a common mechanism of serialization. + * + * Currently serialization of FCS readers and writers is performed via acquiring + * a special per-inode rw-semaphore (conv_sem). And yes, {down, up}_read is for + * FCS readers, and {down, up}_write is for FCS writers, see the macros below + * for passive/active protection. + * + * --- + * (*) This heuristic can be changed to a better one (benchmarking is needed). + * (**) Such technique allows to keep enable/disable state on disk. + */ + +#include "../../inode.h" +#include "../cluster.h" +#include "file.h" + +#define conversion_enabled(inode) \ + (inode_compression_mode_plugin(inode) == \ + compression_mode_plugin_by_id(CONVX_COMPRESSION_MODE_ID)) + +/** + * Located sections (readers and writers of @pset) are not permanently + * critical: cryptcompress file can be converted only if the conversion + * is enabled (see the macrio above). Also we don't perform back + * conversion. The following helper macro is a sanity check to decide + * if we need the protection (locks are always additional overheads). + */ +#define should_protect(inode) \ + (inode_file_plugin(inode) == \ + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID) && \ + conversion_enabled(inode)) +/** + * To avoid confusion with read/write file operations, we'll speak about + * "passive" protection for FCS readers and "active" protection for FCS + * writers. All methods with active or passive protection have suffix + * "careful". + */ +/** + * Macros for passive protection. + * + * Construct invariant operation to be supplied to VFS. + * The macro accepts the following lexemes: + * @type - type of the value represented by the compound statement; + * @method - name of an operation to be supplied to VFS (reiser4 file + * plugin also should contain a method with such name). + */ +#define PROT_PASSIVE(type, method, args) \ +({ \ + type _result; \ + struct rw_semaphore * guard = \ + &reiser4_inode_data(inode)->conv_sem; \ + \ + if (should_protect(inode)) { \ + down_read(guard); \ + if (!should_protect(inode)) \ + up_read(guard); \ + } \ + _result = inode_file_plugin(inode)->method args; \ + if (should_protect(inode)) \ + up_read(guard); \ + _result; \ +}) + +#define PROT_PASSIVE_VOID(method, args) \ +({ \ + struct rw_semaphore * guard = \ + &reiser4_inode_data(inode)->conv_sem; \ + \ + if (should_protect(inode)) { \ + down_read(guard); \ + if (!should_protect(inode)) \ + up_read(guard); \ + } \ + inode_file_plugin(inode)->method args; \ + \ + if (should_protect(inode)) \ + up_read(guard); \ +}) + +/* Pass management to the unix-file plugin with "notail" policy */ +static int __cryptcompress2unixfile(struct file *file, struct inode * inode) +{ + int result; + reiser4_inode *info; + struct unix_file_info * uf; + info = reiser4_inode_data(inode); + + result = aset_set_unsafe(&info->pset, + PSET_FILE, + (reiser4_plugin *) + file_plugin_by_id(UNIX_FILE_PLUGIN_ID)); + if (result) + return result; + result = aset_set_unsafe(&info->pset, + PSET_FORMATTING, + (reiser4_plugin *) + formatting_plugin_by_id(NEVER_TAILS_FORMATTING_ID)); + if (result) + return result; + /* get rid of non-standard plugins */ + info->plugin_mask &= ~cryptcompress_mask; + /* get rid of plugin stat-data extension */ + info->extmask &= ~(1 << PLUGIN_STAT); + + reiser4_inode_clr_flag(inode, REISER4_SDLEN_KNOWN); + + /* FIXME use init_inode_data_unix_file() instead, + but aviod init_inode_ordering() */ + /* Init unix-file specific part of inode */ + uf = unix_file_inode_data(inode); + uf->container = UF_CONTAINER_UNKNOWN; + init_rwsem(&uf->latch); + uf->tplug = inode_formatting_plugin(inode); + uf->exclusive_use = 0; +#if REISER4_DEBUG + uf->ea_owner = NULL; + atomic_set(&uf->nr_neas, 0); +#endif + /** + * we was carefull for file_ops, inode_ops and as_ops + * to be invariant for plugin conversion, so there is + * no need to update ones already installed in the + * vfs's residence. + */ + return 0; +} + +#if REISER4_DEBUG +static int disabled_conversion_inode_ok(struct inode * inode) +{ + __u64 extmask = reiser4_inode_data(inode)->extmask; + __u16 plugin_mask = reiser4_inode_data(inode)->plugin_mask; + + return ((extmask & (1 << LIGHT_WEIGHT_STAT)) && + (extmask & (1 << UNIX_STAT)) && + (extmask & (1 << LARGE_TIMES_STAT)) && + (extmask & (1 << PLUGIN_STAT)) && + (plugin_mask & (1 << PSET_COMPRESSION_MODE))); +} +#endif + +/** + * Disable future attempts to schedule/convert file plugin. + * This function is called by plugin schedule hooks. + * + * To disable conversion we assign any compression mode plugin id + * different from CONVX_COMPRESSION_MODE_ID. + */ +static int disable_conversion(struct inode * inode) +{ + int result; + result = + force_plugin_pset(inode, + PSET_COMPRESSION_MODE, + (reiser4_plugin *)compression_mode_plugin_by_id + (LATTD_COMPRESSION_MODE_ID)); + assert("edward-1500", + ergo(!result, disabled_conversion_inode_ok(inode))); + return result; +} + +/** + * Check if we really have achieved plugin scheduling point + */ +static int check_psched_point(struct inode * inode, + loff_t pos /* position in the + file to write from */, + struct cluster_handle * clust, + struct psched_context * cont) +{ + assert("edward-1505", conversion_enabled(inode)); + /* + * if file size is more then cluster size, then compressible + * status must be figured out (i.e. compression was disabled, + * or file plugin was converted to unix_file) + */ + assert("edward-1506", inode->i_size <= inode_cluster_size(inode)); + + if (pos > inode->i_size) + /* first logical cluster will contain a (partial) hole */ + return disable_conversion(inode); + if (pos < inode_cluster_size(inode)) + /* writing to the first logical cluster */ + return 0; + /* + * here we have: + * cluster_size <= pos <= i_size <= cluster_size, + * and, hence, pos == i_size == cluster_size + */ + assert("edward-1498", + pos == inode->i_size && + pos == inode_cluster_size(inode)); + assert("edward-1539", cont != NULL); + assert("edward-1540", cont->state == PSCHED_INVAL_STATE); + + cont->state = PSCHED_SCHED_POINT; + return 0; +} + +static void start_check_compressibility(struct inode * inode, + struct cluster_handle * clust, + hint_t * hint) +{ + assert("edward-1507", clust->index == 1); + assert("edward-1508", !tfm_cluster_is_uptodate(&clust->tc)); + assert("edward-1509", cluster_get_tfm_act(&clust->tc) == TFMA_READ); + + hint_init_zero(hint); + clust->hint = hint; + clust->index --; + clust->nr_pages = size_in_pages(lbytes(clust->index, inode)); + + /* first logical cluster (of index #0) must be complete */ + assert("edward-1510", lbytes(clust->index, inode) == + inode_cluster_size(inode)); +} + +static void finish_check_compressibility(struct inode * inode, + struct cluster_handle * clust, + hint_t * hint) +{ + reiser4_unset_hint(clust->hint); + clust->hint = hint; + clust->index ++; +} + +#if REISER4_DEBUG +static int prepped_dclust_ok(hint_t * hint) +{ + reiser4_key key; + coord_t * coord = &hint->ext_coord.coord; + + item_key_by_coord(coord, &key); + return (item_id_by_coord(coord) == CTAIL_ID && + !coord_is_unprepped_ctail(coord) && + (get_key_offset(&key) + nr_units_ctail(coord) == + dclust_get_extension_dsize(hint))); +} +#endif + +#define fifty_persent(size) (size >> 1) +/* evaluation of data compressibility */ +#define data_is_compressible(osize, isize) \ + (osize < fifty_persent(isize)) + +/** + * A simple O(1)-heuristic for compressibility. + * This is called not more then one time per file's life. + * Read first logical cluster (of index #0) and estimate its compressibility. + * Save estimation result in @cont. + */ +static int read_check_compressibility(struct inode * inode, + struct cluster_handle * clust, + struct psched_context * cont) +{ + int i; + int result; + __u32 dst_len; + hint_t tmp_hint; + hint_t * cur_hint = clust->hint; + assert("edward-1541", cont->state == PSCHED_SCHED_POINT); + + start_check_compressibility(inode, clust, &tmp_hint); + + reset_cluster_pgset(clust, cluster_nrpages(inode)); + result = grab_page_cluster(inode, clust, READ_OP); + if (result) + return result; + /* Read page cluster here */ + for (i = 0; i < clust->nr_pages; i++) { + struct page *page = clust->pages[i]; + lock_page(page); + result = do_readpage_ctail(inode, clust, page, + ZNODE_READ_LOCK); + unlock_page(page); + if (result) + goto error; + } + tfm_cluster_clr_uptodate(&clust->tc); + + cluster_set_tfm_act(&clust->tc, TFMA_WRITE); + + if (hint_is_valid(&tmp_hint) && !hint_is_unprepped_dclust(&tmp_hint)) { + /* lenght of compressed data is known, no need to compress */ + assert("edward-1511", + znode_is_any_locked(tmp_hint.lh.node)); + assert("edward-1512", + WITH_DATA(tmp_hint.ext_coord.coord.node, + prepped_dclust_ok(&tmp_hint))); + dst_len = dclust_get_extension_dsize(&tmp_hint); + } + else { + struct tfm_cluster * tc = &clust->tc; + compression_plugin * cplug = inode_compression_plugin(inode); + result = grab_tfm_stream(inode, tc, INPUT_STREAM); + if (result) + goto error; + for (i = 0; i < clust->nr_pages; i++) { + char *data; + lock_page(clust->pages[i]); + BUG_ON(!PageUptodate(clust->pages[i])); + data = kmap(clust->pages[i]); + memcpy(tfm_stream_data(tc, INPUT_STREAM) + pg_to_off(i), + data, PAGE_CACHE_SIZE); + kunmap(clust->pages[i]); + unlock_page(clust->pages[i]); + } + result = grab_tfm_stream(inode, tc, OUTPUT_STREAM); + if (result) + goto error; + result = grab_coa(tc, cplug); + if (result) + goto error; + tc->len = tc->lsize = lbytes(clust->index, inode); + assert("edward-1513", tc->len == inode_cluster_size(inode)); + dst_len = tfm_stream_size(tc, OUTPUT_STREAM); + cplug->compress(get_coa(tc, cplug->h.id, tc->act), + tfm_input_data(clust), tc->len, + tfm_output_data(clust), &dst_len); + assert("edward-1514", + dst_len <= tfm_stream_size(tc, OUTPUT_STREAM)); + } + finish_check_compressibility(inode, clust, cur_hint); + cont->state = + (data_is_compressible(dst_len, inode_cluster_size(inode)) ? + PSCHED_REMAINS_OLD : + PSCHED_ASSIGNED_NEW); + return 0; + error: + put_page_cluster(clust, inode, READ_OP); + return result; +} + +/* Cut disk cluster of index @idx */ +static int cut_disk_cluster(struct inode * inode, cloff_t idx) +{ + reiser4_key from, to; + assert("edward-1515", inode_file_plugin(inode) == + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)); + key_by_inode_cryptcompress(inode, clust_to_off(idx, inode), &from); + to = from; + set_key_offset(&to, + get_key_offset(&from) + inode_cluster_size(inode) - 1); + return reiser4_cut_tree(reiser4_tree_by_inode(inode), + &from, &to, inode, 0); +} + +static int reserve_cryptcompress2unixfile(struct inode *inode) +{ + reiser4_block_nr unformatted_nodes; + reiser4_tree *tree; + + tree = reiser4_tree_by_inode(inode); + + /* number of unformatted nodes which will be created */ + unformatted_nodes = cluster_nrpages(inode); /* N */ + + /* + * space required for one iteration of extent->tail conversion: + * + * 1. kill ctail items + * + * 2. insert N unformatted nodes + * + * 3. insert N (worst-case single-block + * extents) extent units. + * + * 4. drilling to the leaf level by coord_by_key() + * + * 5. possible update of stat-data + * + */ + grab_space_enable(); + return reiser4_grab_space + (2 * tree->height + + unformatted_nodes + + unformatted_nodes * estimate_one_insert_into_item(tree) + + 1 + estimate_one_insert_item(tree) + + inode_file_plugin(inode)->estimate.update(inode), + BA_CAN_COMMIT); +} + +/** + * Convert cryptcompress file plugin to unix_file plugin. + */ +static int cryptcompress2unixfile(struct file * file, struct inode * inode, + struct psched_context * cont) +{ + int i; + int result = 0; + struct cryptcompress_info *cr_info; + struct unix_file_info *uf_info; + assert("edward-1516", cont->pages[0]->index == 0); + + /* release all cryptcompress-specific resources */ + cr_info = cryptcompress_inode_data(inode); + result = reserve_cryptcompress2unixfile(inode); + if (result) + goto out; + /* tell kill_hook to not truncate pages */ + reiser4_inode_set_flag(inode, REISER4_FILE_CONV_IN_PROGRESS); + result = cut_disk_cluster(inode, 0); + if (result) + goto out; + /* captured jnode of cluster and assotiated resources (pages, + reserved disk space) were released by ->kill_hook() method + of the item plugin */ + + result = __cryptcompress2unixfile(file, inode); + if (result) + goto out; + /* At this point file is managed by unix file plugin */ + + uf_info = unix_file_inode_data(inode); + + assert("edward-1518", + ergo(jprivate(cont->pages[0]), + !jnode_is_cluster_page(jprivate(cont->pages[0])))); + for(i = 0; i < cont->nr_pages; i++) { + assert("edward-1519", cont->pages[i]); + assert("edward-1520", PageUptodate(cont->pages[i])); + + result = find_or_create_extent(cont->pages[i]); + if (result) + break; + } + if (unlikely(result)) + goto out; + uf_info->container = UF_CONTAINER_EXTENTS; + result = reiser4_update_sd(inode); + out: + all_grabbed2free(); + return result; +} + +#define convert_file_plugin cryptcompress2unixfile + +/** + * This is called by ->write() method of a cryptcompress file plugin. + * Make a decision about the most reasonable file plugin id to manage + * the file. + */ +int write_pschedule_hook(struct file * file, struct inode * inode, + loff_t pos, struct cluster_handle * clust, + struct psched_context * cont) +{ + int result; + if (!conversion_enabled(inode)) + return 0; + result = check_psched_point(inode, pos, clust, cont); + if (result || cont->state != PSCHED_SCHED_POINT) + return result; + result = read_check_compressibility(inode, clust, cont); + if (result) + return result; + if (cont->state == PSCHED_REMAINS_OLD) { + put_page_cluster(clust, inode, READ_OP); + return disable_conversion(inode); + } + assert("edward-1543", cont->state == PSCHED_ASSIGNED_NEW); + /* + * page cluster is grabbed and uptodate. It will be + * released with a pgset after plugin conversion is + * finished, see put_psched_context(). + */ + reiser4_unset_hint(clust->hint); + move_cluster_pgset(clust, &cont->pages, &cont->nr_pages); + return 0; +} + +/** + * This is called by ->setattr() method of cryptcompress file plugin. + */ +int setattr_pschedule_hook(struct inode * inode) +{ + if (conversion_enabled(inode)) + return disable_conversion(inode); + return 0; +} + +static inline void init_psched_context(struct psched_context * cont) +{ + memset(cont, 0, sizeof(*cont)); +} + +static inline void done_psched_context(struct psched_context * cont, + struct inode * inode) +{ + if (cont->pages) { + __put_page_cluster(0, cont->nr_pages, cont->pages, inode); + kfree(cont->pages); + } +} +/** + * Here are wrappers with "protection", aka Reiser4 "careful" methods. + * They are used by vfs (as methods of file_ops, inode_ops or as_ops), + * which is not aware of plugin conversion performed by Reiser4. + */ + +/* + * Wrappers with active protection for: + * + * ->write(); + */ + +/* + * ->write() file operation supplied to VFS. + * Write a file in 3 steps (some of them can be optional). + */ +ssize_t reiser4_write_careful(struct file *file, const char __user *buf, + size_t count, loff_t *off) +{ + int result; + reiser4_context *ctx; + ssize_t written_old = 0; /* bytes written with initial plugin */ + ssize_t written_new = 0; /* bytes written with new plugin */ + struct psched_context cont; + struct inode * inode = file->f_dentry->d_inode; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + init_psched_context(&cont); + mutex_lock(&inode->i_mutex); + /** + * First step. + * Start write with initial file plugin. + * Keep a plugin schedule status at @cont (if any). + */ + written_old = inode_file_plugin(inode)->write(file, + buf, + count, + off, + &cont); + if (cont.state != PSCHED_ASSIGNED_NEW || written_old < 0) + goto exit; + /** + * Second step. + * New file plugin has been scheduled. + * Perform conversion to the new plugin. + */ + down_read(&reiser4_inode_data(inode)->conv_sem); + result = convert_file_plugin(file, inode, &cont); + up_read(&reiser4_inode_data(inode)->conv_sem); + if (result) { + warning("edward-1544", + "Inode %llu: file plugin conversion failed (%d)", + (unsigned long long)get_inode_oid(inode), + result); + context_set_commit_async(ctx); + goto exit; + } + reiser4_txn_restart(ctx); + /** + * Third step: + * Finish write with the new file plugin. + */ + assert("edward-1536", + inode_file_plugin(inode) == + file_plugin_by_id(UNIX_FILE_PLUGIN_ID)); + + written_new = inode_file_plugin(inode)->write(file, + buf + written_old, + count - written_old, + off, + NULL); + exit: + mutex_unlock(&inode->i_mutex); + done_psched_context(&cont, inode); + reiser4_exit_context(ctx); + + return written_old + (written_new < 0 ? 0 : written_new); +} + +/* Wrappers with passive protection for: + * + * ->open(); + * ->read(); + * ->ioctl(); + * ->mmap(); + * ->release(); + * ->bmap(). + */ + +int reiser4_open_careful(struct inode *inode, struct file *file) +{ + return PROT_PASSIVE(int, open, (inode, file)); +} + +ssize_t reiser4_read_careful(struct file * file, char __user * buf, + size_t size, loff_t * off) +{ + struct inode * inode = file->f_dentry->d_inode; + return PROT_PASSIVE(ssize_t, read, (file, buf, size, off)); +} + +int reiser4_ioctl_careful(struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg) +{ + return PROT_PASSIVE(int, ioctl, (inode, filp, cmd, arg)); +} + +int reiser4_mmap_careful(struct file *file, struct vm_area_struct *vma) +{ + struct inode *inode = file->f_dentry->d_inode; + return PROT_PASSIVE(int, mmap, (file, vma)); +} + +int reiser4_release_careful(struct inode *inode, struct file *file) +{ + return PROT_PASSIVE(int, release, (inode, file)); +} + +sector_t reiser4_bmap_careful(struct address_space * mapping, sector_t lblock) +{ + struct inode *inode = mapping->host; + return PROT_PASSIVE(sector_t, bmap, (mapping, lblock)); +} + +/** + * NOTE: The following two methods are + * used only for loopback functionality. + * reiser4_write_end() can not cope with + * short writes for now. + */ +int reiser4_write_begin_careful(struct file *file, + struct address_space *mapping, + loff_t pos, + unsigned len, + unsigned flags, + struct page **pagep, + void **fsdata) +{ + int ret = 0; + unsigned start, end; + struct page *page; + pgoff_t index; + reiser4_context *ctx; + struct inode * inode = file->f_dentry->d_inode; + + index = pos >> PAGE_CACHE_SHIFT; + start = pos & (PAGE_CACHE_SIZE - 1); + end = start + len; + + page = grab_cache_page_write_begin(mapping, index, + flags & AOP_FLAG_NOFS); + *pagep = page; + if (!page) + return -ENOMEM; + + ctx = reiser4_init_context(file->f_dentry->d_inode->i_sb); + if (IS_ERR(ctx)) { + ret = PTR_ERR(ctx); + goto out; + } + ret = PROT_PASSIVE(int, write_begin, (file, page, start, end)); + + /* don't commit transaction under inode semaphore */ + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + out: + if (unlikely(ret)) { + unlock_page(page); + page_cache_release(page); + } + return ret; +} + +int reiser4_write_end_careful(struct file *file, + struct address_space *mapping, + loff_t pos, + unsigned len, + unsigned copied, + struct page *page, + void *fsdata) +{ + int ret; + reiser4_context *ctx; + unsigned start, end; + struct inode *inode = page->mapping->host; + + assert("umka-3101", file != NULL); + assert("umka-3102", page != NULL); + assert("umka-3093", PageLocked(page)); + + start = pos & (PAGE_CACHE_SIZE - 1); + end = start + len; + + flush_dcache_page(page); + SetPageUptodate(page); + + ctx = reiser4_init_context(page->mapping->host->i_sb); + if (IS_ERR(ctx)){ + unlock_page(page); + ret = PTR_ERR(ctx); + goto out; + } + ret = PROT_PASSIVE(int, write_end, (file, page, start, end)); + + /* don't commit transaction under inode semaphore */ + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + out: + page_cache_release(page); + if (!ret) + ret = copied; + return ret; +} + +/* + * Wrappers without protection for: + * + * ->setattr() + */ +int reiser4_setattr(struct dentry *dentry, struct iattr *attr) +{ + return inode_file_plugin(dentry->d_inode)->setattr(dentry, attr); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/file.h linux-2.6.30/fs/reiser4/plugin/file/file.h --- linux-2.6.30.orig/fs/reiser4/plugin/file/file.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/file.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,336 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* this file contains declarations of methods implementing + file plugins (UNIX_FILE_PLUGIN_ID, CRYPTCOMPRESS_FILE_PLUGIN_ID + and SYMLINK_FILE_PLUGIN_ID) */ + +#if !defined( __REISER4_FILE_H__ ) +#define __REISER4_FILE_H__ + +/* possible states when scheduling a new file plugin */ +typedef enum { + PSCHED_INVAL_STATE, /* invalid state */ + PSCHED_SCHED_POINT, /* scheduling point has been achieved */ + PSCHED_REMAINS_OLD, /* made a decision to be managed by old plugin */ + PSCHED_ASSIGNED_NEW /* new plugin has been scheduled */ +} psched_state; + +struct psched_context { + int nr_pages; + struct page **pages; + psched_state state; +}; + +/** + * Declarations of common/careful/generic methods. + * Suppose ->foo() is a vs method (of f_ops, i_ops, or a_ops); + * Then common reiser4 method for foo looks like reiser4_foo_common; + * careful method looks like reiser4_foo_careful; + * generic method looks like reiser4_foo. + * + * Common method is a simple instruction set eligible for more + * then one plugin id. + * + * Generic method looks at the plugin installed in inode's + * plugin set and calls its appropriate method. + * + * Careful method looks like generic method with protected pset + * (see plugin/file/file_conversion.c for details). + */ + +/* inode operations */ +int reiser4_setattr(struct dentry *, struct iattr *); + +/* file operations */ +ssize_t reiser4_read_careful(struct file *, char __user *buf, + size_t count, loff_t *off); +ssize_t reiser4_write_careful(struct file *, const char __user *buf, + size_t count, loff_t * off); +int reiser4_ioctl_careful(struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg); +int reiser4_mmap_careful(struct file *, struct vm_area_struct *); +int reiser4_open_careful(struct inode *inode, struct file *file); +int reiser4_release_careful(struct inode *, struct file *); +int reiser4_sync_file_common(struct file *, struct dentry *, int datasync); + +/* address space operations */ +int reiser4_readpage(struct file *, struct page *); +int reiser4_readpages(struct file*, struct address_space*, struct list_head*, + unsigned); +int reiser4_writepages(struct address_space *, struct writeback_control *); +int reiser4_write_begin_careful(struct file *file, + struct address_space *mapping, + loff_t pos, unsigned len, unsigned flags, + struct page **pagep, void **fsdata); +int reiser4_write_end_careful(struct file *file, + struct address_space *mapping, + loff_t pos, unsigned len, unsigned copied, + struct page *page, void *fsdata); +sector_t reiser4_bmap_careful(struct address_space *, sector_t lblock); + +/* + * Private methods of unix-file plugin + * (UNIX_FILE_PLUGIN_ID) + */ + +/* private inode operations */ +int setattr_unix_file(struct dentry *, struct iattr *); + +/* private file operations */ + +ssize_t read_unix_file(struct file *, char __user *buf, size_t read_amount, + loff_t *off); +ssize_t write_unix_file(struct file *, const char __user *buf, size_t write_amount, + loff_t * off, struct psched_context * cont); +int ioctl_unix_file(struct inode *, struct file *, unsigned int cmd, + unsigned long arg); +int mmap_unix_file(struct file *, struct vm_area_struct *); +int open_unix_file(struct inode *, struct file *); +int release_unix_file(struct inode *, struct file *); + +/* private address space operations */ +int readpage_unix_file(struct file *, struct page *); +int readpages_unix_file(struct file*, struct address_space*, struct list_head*, + unsigned); +int writepages_unix_file(struct address_space *, struct writeback_control *); +int write_begin_unix_file(struct file *file, struct page *page, + unsigned from, unsigned to); +int write_end_unix_file(struct file *file, struct page *page, + unsigned from, unsigned to); +sector_t bmap_unix_file(struct address_space *, sector_t lblock); + +/* other private methods */ +int delete_object_unix_file(struct inode *); +int flow_by_inode_unix_file(struct inode *, const char __user *buf, + int user, loff_t, loff_t, rw_op, flow_t *); +int owns_item_unix_file(const struct inode *, const coord_t *); +void init_inode_data_unix_file(struct inode *, reiser4_object_create_data *, + int create); + +/* + * Private methods of cryptcompress file plugin + * (CRYPTCOMPRESS_FILE_PLUGIN_ID) + */ + +/* private inode operations */ +int setattr_cryptcompress(struct dentry *, struct iattr *); + +/* private file operations */ +ssize_t read_cryptcompress(struct file *, char __user *buf, + size_t count, loff_t *off); +ssize_t write_cryptcompress(struct file *, const char __user *buf, + size_t count, loff_t * off, + struct psched_context *cont); +int ioctl_cryptcompress(struct inode *, struct file *, unsigned int cmd, + unsigned long arg); +int mmap_cryptcompress(struct file *, struct vm_area_struct *); +int open_cryptcompress(struct inode *, struct file *); +int release_cryptcompress(struct inode *, struct file *); + +/* private address space operations */ +int readpage_cryptcompress(struct file *, struct page *); +int readpages_cryptcompress(struct file*, struct address_space*, + struct list_head*, unsigned); +int writepages_cryptcompress(struct address_space *, + struct writeback_control *); +int write_begin_cryptcompress(struct file *file, struct page *page, + unsigned from, unsigned to); +int write_end_cryptcompress(struct file *file, struct page *page, + unsigned from, unsigned to); +sector_t bmap_cryptcompress(struct address_space *, sector_t lblock); + +/* other private methods */ +int flow_by_inode_cryptcompress(struct inode *, const char __user *buf, + int user, loff_t, loff_t, rw_op, flow_t *); +int key_by_inode_cryptcompress(struct inode *, loff_t off, reiser4_key *); +int create_object_cryptcompress(struct inode *, struct inode *, + reiser4_object_create_data *); +int delete_object_cryptcompress(struct inode *); +void init_inode_data_cryptcompress(struct inode *, reiser4_object_create_data *, + int create); +int cut_tree_worker_cryptcompress(tap_t *, const reiser4_key * from_key, + const reiser4_key * to_key, + reiser4_key * smallest_removed, + struct inode *object, int truncate, + int *progress); +void destroy_inode_cryptcompress(struct inode *); + +/* + * Private methods of symlink file plugin + * (SYMLINK_FILE_PLUGIN_ID) + */ +int reiser4_create_symlink(struct inode *symlink, struct inode *dir, + reiser4_object_create_data *); +void destroy_inode_symlink(struct inode *); + +/* + * all the write into unix file is performed by item write method. Write method + * of unix file plugin only decides which item plugin (extent or tail) and in + * which mode (one from the enum below) to call + */ +typedef enum { + FIRST_ITEM = 1, + APPEND_ITEM = 2, + OVERWRITE_ITEM = 3 +} write_mode_t; + +/* unix file may be in one the following states */ +typedef enum { + UF_CONTAINER_UNKNOWN = 0, + UF_CONTAINER_TAILS = 1, + UF_CONTAINER_EXTENTS = 2, + UF_CONTAINER_EMPTY = 3 +} file_container_t; + +struct formatting_plugin; +struct inode; + +/* unix file plugin specific part of reiser4 inode */ +struct unix_file_info { + /* + * this read-write lock protects file containerization change. Accesses + * which do not change file containerization (see file_container_t) + * (read, readpage, writepage, write (until tail conversion is + * involved)) take read-lock. Accesses which modify file + * containerization (truncate, conversion from tail to extent and back) + * take write-lock. + */ + struct rw_semaphore latch; + /* this enum specifies which items are used to build the file */ + file_container_t container; + /* + * plugin which controls when file is to be converted to extents and + * back to tail + */ + struct formatting_plugin *tplug; + /* if this is set, file is in exclusive use */ + int exclusive_use; +#if REISER4_DEBUG + /* pointer to task struct of thread owning exclusive access to file */ + void *ea_owner; + atomic_t nr_neas; + void *last_reader; +#endif +}; + +struct unix_file_info *unix_file_inode_data(const struct inode *inode); +void get_exclusive_access(struct unix_file_info *); +void drop_exclusive_access(struct unix_file_info *); +void get_nonexclusive_access(struct unix_file_info *); +void drop_nonexclusive_access(struct unix_file_info *); +int try_to_get_nonexclusive_access(struct unix_file_info *); +int find_file_item(hint_t *, const reiser4_key *, znode_lock_mode, + struct inode *); +int find_file_item_nohint(coord_t *, lock_handle *, + const reiser4_key *, znode_lock_mode, + struct inode *); + +int load_file_hint(struct file *, hint_t *); +void save_file_hint(struct file *, const hint_t *); + +#include "../item/extent.h" +#include "../item/tail.h" +#include "../item/ctail.h" + +struct uf_coord { + coord_t coord; + lock_handle *lh; + int valid; + union { + struct extent_coord_extension extent; + struct tail_coord_extension tail; + struct ctail_coord_extension ctail; + } extension; +}; + +#include "../../forward.h" +#include "../../seal.h" +#include "../../lock.h" + +/* + * This structure is used to speed up file operations (reads and writes). A + * hint is a suggestion about where a key resolved to last time. A seal + * indicates whether a node has been modified since a hint was last recorded. + * You check the seal, and if the seal is still valid, you can use the hint + * without traversing the tree again. + */ +struct hint { + seal_t seal; /* a seal over last file item accessed */ + uf_coord_t ext_coord; + loff_t offset; + znode_lock_mode mode; + lock_handle lh; +}; + +static inline int hint_is_valid(hint_t * hint) +{ + return hint->ext_coord.valid; +} + +static inline void hint_set_valid(hint_t * hint) +{ + hint->ext_coord.valid = 1; +} + +static inline void hint_clr_valid(hint_t * hint) +{ + hint->ext_coord.valid = 0; +} + +int load_file_hint(struct file *, hint_t *); +void save_file_hint(struct file *, const hint_t *); +void hint_init_zero(hint_t *); +void reiser4_set_hint(hint_t *, const reiser4_key *, znode_lock_mode); +int hint_is_set(const hint_t *); +void reiser4_unset_hint(hint_t *); + +int reiser4_update_file_size(struct inode *, loff_t, int update_sd); +int cut_file_items(struct inode *, loff_t new_size, + int update_sd, loff_t cur_size, + int (*update_actor) (struct inode *, loff_t, int)); +#if REISER4_DEBUG + +/* return 1 is exclusive access is obtained, 0 - otherwise */ +static inline int ea_obtained(struct unix_file_info * uf_info) +{ + int ret; + + ret = down_read_trylock(&uf_info->latch); + if (ret) + up_read(&uf_info->latch); + return !ret; +} + +#endif + +#define WRITE_GRANULARITY 32 + +int tail2extent(struct unix_file_info *); +int extent2tail(struct file *, struct unix_file_info *); + +int goto_right_neighbor(coord_t *, lock_handle *); +int find_or_create_extent(struct page *); +int equal_to_ldk(znode *, const reiser4_key *); + +void init_uf_coord(uf_coord_t *uf_coord, lock_handle *lh); + +static inline int cbk_errored(int cbk_result) +{ + return (cbk_result != CBK_COORD_NOTFOUND + && cbk_result != CBK_COORD_FOUND); +} + +/* __REISER4_FILE_H__ */ +#endif + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/Makefile linux-2.6.30/fs/reiser4/plugin/file/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/file/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,7 @@ +obj-$(CONFIG_REISER4_FS) += file_plugins.o + +file_plugins-objs := \ + file.o \ + tail_conversion.o \ + symlink.o \ + cryptcompress.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/symfile.c linux-2.6.30/fs/reiser4/plugin/file/symfile.c --- linux-2.6.30.orig/fs/reiser4/plugin/file/symfile.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/symfile.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,87 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Symfiles are a generalization of Unix symlinks. + + A symfile when read behaves as though you took its contents and + substituted them into the reiser4 naming system as the right hand side + of an assignment, and then read that which you had assigned to it. + + A key issue for symfiles is how to implement writes through to + subfiles. In general, one must have some method of determining what + of that which is written to the symfile is written to what subfile. + This can be done by use of custom plugin methods written by users, or + by using a few general methods we provide for those willing to endure + the insertion of delimiters into what is read. + + Writing to symfiles without delimiters to denote what is written to + what subfile is not supported by any plugins we provide in this + release. Our most sophisticated support for writes is that embodied + by the invert plugin (see invert.c). + + A read only version of the /etc/passwd file might be + constructed as a symfile whose contents are as follows: + + /etc/passwd/userlines/* + + or + + /etc/passwd/userlines/demidov+/etc/passwd/userlines/edward+/etc/passwd/userlines/reiser+/etc/passwd/userlines/root + + or + + /etc/passwd/userlines/(demidov+edward+reiser+root) + + A symfile with contents + + /filenameA+"(some text stored in the uninvertable symfile)+/filenameB + + will return when read + + The contents of filenameAsome text stored in the uninvertable symfileThe contents of filenameB + + and write of what has been read will not be possible to implement as + an identity operation because there are no delimiters denoting the + boundaries of what is to be written to what subfile. + + Note that one could make this a read/write symfile if one specified + delimiters, and the write method understood those delimiters delimited + what was written to subfiles. + + So, specifying the symfile in a manner that allows writes: + + /etc/passwd/userlines/demidov+"( + )+/etc/passwd/userlines/edward+"( + )+/etc/passwd/userlines/reiser+"( + )+/etc/passwd/userlines/root+"( + ) + + or + + /etc/passwd/userlines/(demidov+"( + )+edward+"( + )+reiser+"( + )+root+"( + )) + + and the file demidov might be specified as: + + /etc/passwd/userlines/demidov/username+"(:)+/etc/passwd/userlines/demidov/password+"(:)+/etc/passwd/userlines/demidov/userid+"(:)+/etc/passwd/userlines/demidov/groupid+"(:)+/etc/passwd/userlines/demidov/gecos+"(:)+/etc/passwd/userlines/demidov/home+"(:)+/etc/passwd/userlines/demidov/shell + + or + + /etc/passwd/userlines/demidov/(username+"(:)+password+"(:)+userid+"(:)+groupid+"(:)+gecos+"(:)+home+"(:)+shell) + + Notice that if the file demidov has a carriage return in it, the + parsing fails, but then if you put carriage returns in the wrong place + in a normal /etc/passwd file it breaks things also. + + Note that it is forbidden to have no text between two interpolations + if one wants to be able to define what parts of a write go to what + subfiles referenced in an interpolation. + + If one wants to be able to add new lines by writing to the file, one + must either write a custom plugin for /etc/passwd that knows how to + name an added line, or one must use an invert, or one must use a more + sophisticated symfile syntax that we are not planning to write for + version 4.0. +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/symlink.c linux-2.6.30/fs/reiser4/plugin/file/symlink.c --- linux-2.6.30.orig/fs/reiser4/plugin/file/symlink.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/symlink.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,95 @@ +/* Copyright 2002, 2003, 2005 by Hans Reiser, licensing governed by reiser4/README */ + +#include "../../inode.h" + +#include +#include + +/* file plugin methods specific for symlink files + (SYMLINK_FILE_PLUGIN_ID) */ + +/* this is implementation of create_object method of file plugin for + SYMLINK_FILE_PLUGIN_ID + */ + +/** + * reiser4_create_symlink - create_object of file plugin for SYMLINK_FILE_PLUGIN_ID + * @symlink: inode of symlink object + * @dir: inode of parent directory + * @info: parameters of new object + * + * Inserts stat data with symlink extension where into the tree. + */ +int reiser4_create_symlink(struct inode *symlink, + struct inode *dir UNUSED_ARG, + reiser4_object_create_data *data /* info passed to us + * this is filled by + * reiser4() syscall + * in particular */) +{ + int result; + + assert("nikita-680", symlink != NULL); + assert("nikita-681", S_ISLNK(symlink->i_mode)); + assert("nikita-685", reiser4_inode_get_flag(symlink, REISER4_NO_SD)); + assert("nikita-682", dir != NULL); + assert("nikita-684", data != NULL); + assert("nikita-686", data->id == SYMLINK_FILE_PLUGIN_ID); + + /* + * stat data of symlink has symlink extension in which we store + * symlink content, that is, path symlink is pointing to. + */ + reiser4_inode_data(symlink)->extmask |= (1 << SYMLINK_STAT); + + assert("vs-838", symlink->i_private == NULL); + symlink->i_private = (void *)data->name; + + assert("vs-843", symlink->i_size == 0); + INODE_SET_FIELD(symlink, i_size, strlen(data->name)); + + /* insert stat data appended with data->name */ + result = inode_file_plugin(symlink)->write_sd_by_inode(symlink); + if (result) { + /* FIXME-VS: Make sure that symlink->i_private is not attached + to kmalloced data */ + INODE_SET_FIELD(symlink, i_size, 0); + } else { + assert("vs-849", symlink->i_private + && reiser4_inode_get_flag(symlink, + REISER4_GENERIC_PTR_USED)); + assert("vs-850", + !memcmp((char *)symlink->i_private, data->name, + (size_t) symlink->i_size + 1)); + } + return result; +} + +/* this is implementation of destroy_inode method of file plugin for + SYMLINK_FILE_PLUGIN_ID + */ +void destroy_inode_symlink(struct inode *inode) +{ + assert("edward-799", + inode_file_plugin(inode) == + file_plugin_by_id(SYMLINK_FILE_PLUGIN_ID)); + assert("edward-800", !is_bad_inode(inode) && is_inode_loaded(inode)); + assert("edward-801", reiser4_inode_get_flag(inode, + REISER4_GENERIC_PTR_USED)); + assert("vs-839", S_ISLNK(inode->i_mode)); + + kfree(inode->i_private); + inode->i_private = NULL; + reiser4_inode_clr_flag(inode, REISER4_GENERIC_PTR_USED); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file/tail_conversion.c linux-2.6.30/fs/reiser4/plugin/file/tail_conversion.c --- linux-2.6.30.orig/fs/reiser4/plugin/file/tail_conversion.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file/tail_conversion.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,737 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "../../inode.h" +#include "../../super.h" +#include "../../page_cache.h" +#include "../../carry.h" +#include "../../safe_link.h" +#include "../../vfs_ops.h" + +#include + +/* this file contains: + tail2extent and extent2tail */ + +/* exclusive access to a file is acquired when file state changes: tail2extent, empty2tail, extent2tail, etc */ +void get_exclusive_access(struct unix_file_info * uf_info) +{ + assert("nikita-3028", reiser4_schedulable()); + assert("nikita-3047", LOCK_CNT_NIL(inode_sem_w)); + assert("nikita-3048", LOCK_CNT_NIL(inode_sem_r)); + /* + * "deadlock avoidance": sometimes we commit a transaction under + * rw-semaphore on a file. Such commit can deadlock with another + * thread that captured some block (hence preventing atom from being + * committed) and waits on rw-semaphore. + */ + reiser4_txn_restart_current(); + LOCK_CNT_INC(inode_sem_w); + down_write(&uf_info->latch); + uf_info->exclusive_use = 1; + assert("vs-1713", uf_info->ea_owner == NULL); + assert("vs-1713", atomic_read(&uf_info->nr_neas) == 0); + ON_DEBUG(uf_info->ea_owner = current); +} + +void drop_exclusive_access(struct unix_file_info * uf_info) +{ + assert("vs-1714", uf_info->ea_owner == current); + assert("vs-1715", atomic_read(&uf_info->nr_neas) == 0); + ON_DEBUG(uf_info->ea_owner = NULL); + uf_info->exclusive_use = 0; + up_write(&uf_info->latch); + assert("nikita-3049", LOCK_CNT_NIL(inode_sem_r)); + assert("nikita-3049", LOCK_CNT_GTZ(inode_sem_w)); + LOCK_CNT_DEC(inode_sem_w); + reiser4_txn_restart_current(); +} + +/** + * nea_grabbed - do something when file semaphore is down_read-ed + * @uf_info: + * + * This is called when nonexclisive access is obtained on file. All it does is + * for debugging purposes. + */ +static void nea_grabbed(struct unix_file_info *uf_info) +{ +#if REISER4_DEBUG + LOCK_CNT_INC(inode_sem_r); + assert("vs-1716", uf_info->ea_owner == NULL); + atomic_inc(&uf_info->nr_neas); + uf_info->last_reader = current; +#endif +} + +/** + * get_nonexclusive_access - get nonexclusive access to a file + * @uf_info: unix file specific part of inode to obtain access to + * + * Nonexclusive access is obtained on a file before read, write, readpage. + */ +void get_nonexclusive_access(struct unix_file_info *uf_info) +{ + assert("nikita-3029", reiser4_schedulable()); + assert("nikita-3361", get_current_context()->trans->atom == NULL); + + down_read(&uf_info->latch); + nea_grabbed(uf_info); +} + +/** + * try_to_get_nonexclusive_access - try to get nonexclusive access to a file + * @uf_info: unix file specific part of inode to obtain access to + * + * Non-blocking version of nonexclusive access obtaining. + */ +int try_to_get_nonexclusive_access(struct unix_file_info *uf_info) +{ + int result; + + result = down_read_trylock(&uf_info->latch); + if (result) + nea_grabbed(uf_info); + return result; +} + +void drop_nonexclusive_access(struct unix_file_info * uf_info) +{ + assert("vs-1718", uf_info->ea_owner == NULL); + assert("vs-1719", atomic_read(&uf_info->nr_neas) > 0); + ON_DEBUG(atomic_dec(&uf_info->nr_neas)); + + up_read(&uf_info->latch); + + LOCK_CNT_DEC(inode_sem_r); + reiser4_txn_restart_current(); +} + +/* part of tail2extent. Cut all items covering @count bytes starting from + @offset */ +/* Audited by: green(2002.06.15) */ +static int cut_formatting_items(struct inode *inode, loff_t offset, int count) +{ + reiser4_key from, to; + + /* AUDIT: How about putting an assertion here, what would check + all provided range is covered by tail items only? */ + /* key of first byte in the range to be cut */ + inode_file_plugin(inode)->key_by_inode(inode, offset, &from); + + /* key of last byte in that range */ + to = from; + set_key_offset(&to, (__u64) (offset + count - 1)); + + /* cut everything between those keys */ + return reiser4_cut_tree(reiser4_tree_by_inode(inode), &from, &to, + inode, 0); +} + +static void release_all_pages(struct page **pages, unsigned nr_pages) +{ + unsigned i; + + for (i = 0; i < nr_pages; i++) { + if (pages[i] == NULL) { +#if REISER4_DEBUG + unsigned j; + for (j = i + 1; j < nr_pages; j++) + assert("vs-1620", pages[j] == NULL); +#endif + break; + } + page_cache_release(pages[i]); + pages[i] = NULL; + } +} + +/* part of tail2extent. replace tail items with extent one. Content of tail + items (@count bytes) being cut are copied already into + pages. extent_writepage method is called to create extents corresponding to + those pages */ +static int replace(struct inode *inode, struct page **pages, unsigned nr_pages, int count) +{ + int result; + unsigned i; + STORE_COUNTERS; + + if (nr_pages == 0) + return 0; + + assert("vs-596", pages[0]); + + /* cut copied items */ + result = cut_formatting_items(inode, page_offset(pages[0]), count); + if (result) + return result; + + CHECK_COUNTERS; + + /* put into tree replacement for just removed items: extent item, namely */ + for (i = 0; i < nr_pages; i++) { + result = add_to_page_cache_lru(pages[i], inode->i_mapping, + pages[i]->index, + mapping_gfp_mask(inode-> + i_mapping)); + if (result) + break; + unlock_page(pages[i]); + result = find_or_create_extent(pages[i]); + if (result) + break; + SetPageUptodate(pages[i]); + } + return result; +} + +#define TAIL2EXTENT_PAGE_NUM 3 /* number of pages to fill before cutting tail + * items */ + +static int reserve_tail2extent_iteration(struct inode *inode) +{ + reiser4_block_nr unformatted_nodes; + reiser4_tree *tree; + + tree = reiser4_tree_by_inode(inode); + + /* number of unformatted nodes which will be created */ + unformatted_nodes = TAIL2EXTENT_PAGE_NUM; + + /* + * space required for one iteration of extent->tail conversion: + * + * 1. kill N tail items + * + * 2. insert TAIL2EXTENT_PAGE_NUM unformatted nodes + * + * 3. insert TAIL2EXTENT_PAGE_NUM (worst-case single-block + * extents) extent units. + * + * 4. drilling to the leaf level by coord_by_key() + * + * 5. possible update of stat-data + * + */ + grab_space_enable(); + return reiser4_grab_space + (2 * tree->height + + TAIL2EXTENT_PAGE_NUM + + TAIL2EXTENT_PAGE_NUM * estimate_one_insert_into_item(tree) + + 1 + estimate_one_insert_item(tree) + + inode_file_plugin(inode)->estimate.update(inode), BA_CAN_COMMIT); +} + +/* clear stat data's flag indicating that conversion is being converted */ +static int complete_conversion(struct inode *inode) +{ + int result; + + grab_space_enable(); + result = + reiser4_grab_space(inode_file_plugin(inode)->estimate.update(inode), + BA_CAN_COMMIT); + if (result == 0) { + reiser4_inode_clr_flag(inode, REISER4_PART_MIXED); + result = reiser4_update_sd(inode); + } + if (result) + warning("vs-1696", "Failed to clear converting bit of %llu: %i", + (unsigned long long)get_inode_oid(inode), result); + return 0; +} + +/** + * find_start + * @inode: + * @id: + * @offset: + * + * this is used by tail2extent and extent2tail to detect where previous + * uncompleted conversion stopped + */ +static int find_start(struct inode *inode, reiser4_plugin_id id, __u64 *offset) +{ + int result; + lock_handle lh; + coord_t coord; + struct unix_file_info *ufo; + int found; + reiser4_key key; + + ufo = unix_file_inode_data(inode); + init_lh(&lh); + result = 0; + found = 0; + inode_file_plugin(inode)->key_by_inode(inode, *offset, &key); + do { + init_lh(&lh); + result = find_file_item_nohint(&coord, &lh, &key, + ZNODE_READ_LOCK, inode); + + if (result == CBK_COORD_FOUND) { + if (coord.between == AT_UNIT) { + /*coord_clear_iplug(&coord); */ + result = zload(coord.node); + if (result == 0) { + if (item_id_by_coord(&coord) == id) + found = 1; + else + item_plugin_by_coord(&coord)->s. + file.append_key(&coord, + &key); + zrelse(coord.node); + } + } else + result = RETERR(-ENOENT); + } + done_lh(&lh); + } while (result == 0 && !found); + *offset = get_key_offset(&key); + return result; +} + +/** + * tail2extent + * @uf_info: + * + * + */ +int tail2extent(struct unix_file_info *uf_info) +{ + int result; + reiser4_key key; /* key of next byte to be moved to page */ + char *p_data; /* data of page */ + unsigned page_off = 0, /* offset within the page where to copy data */ + count; /* number of bytes of item which can be + * copied to page */ + struct page *pages[TAIL2EXTENT_PAGE_NUM]; + struct page *page; + int done; /* set to 1 when all file is read */ + char *item; + int i; + struct inode *inode; + int first_iteration; + int bytes; + __u64 offset; + + assert("nikita-3362", ea_obtained(uf_info)); + inode = unix_file_info_to_inode(uf_info); + assert("nikita-3412", !IS_RDONLY(inode)); + assert("vs-1649", uf_info->container != UF_CONTAINER_EXTENTS); + assert("", !reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV)); + + offset = 0; + first_iteration = 1; + result = 0; + if (reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) { + /* + * file is marked on disk as there was a conversion which did + * not complete due to either crash or some error. Find which + * offset tail conversion stopped at + */ + result = find_start(inode, FORMATTING_ID, &offset); + if (result == -ENOENT) { + /* no tail items found, everything is converted */ + uf_info->container = UF_CONTAINER_EXTENTS; + complete_conversion(inode); + return 0; + } else if (result != 0) + /* some other error */ + return result; + first_iteration = 0; + } + + reiser4_inode_set_flag(inode, REISER4_PART_IN_CONV); + + /* get key of first byte of a file */ + inode_file_plugin(inode)->key_by_inode(inode, offset, &key); + + done = 0; + while (done == 0) { + memset(pages, 0, sizeof(pages)); + result = reserve_tail2extent_iteration(inode); + if (result != 0) { + reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV); + goto out; + } + if (first_iteration) { + reiser4_inode_set_flag(inode, REISER4_PART_MIXED); + reiser4_update_sd(inode); + first_iteration = 0; + } + bytes = 0; + for (i = 0; i < sizeof_array(pages) && done == 0; i++) { + assert("vs-598", + (get_key_offset(&key) & ~PAGE_CACHE_MASK) == 0); + page = alloc_page(reiser4_ctx_gfp_mask_get()); + if (!page) { + result = RETERR(-ENOMEM); + goto error; + } + + page->index = + (unsigned long)(get_key_offset(&key) >> + PAGE_CACHE_SHIFT); + /* + * usually when one is going to longterm lock znode (as + * find_file_item does, for instance) he must not hold + * locked pages. However, there is an exception for + * case tail2extent. Pages appearing here are not + * reachable to everyone else, they are clean, they do + * not have jnodes attached so keeping them locked do + * not risk deadlock appearance + */ + assert("vs-983", !PagePrivate(page)); + reiser4_invalidate_pages(inode->i_mapping, page->index, + 1, 0); + + for (page_off = 0; page_off < PAGE_CACHE_SIZE;) { + coord_t coord; + lock_handle lh; + + /* get next item */ + /* FIXME: we might want to readahead here */ + init_lh(&lh); + result = + find_file_item_nohint(&coord, &lh, &key, + ZNODE_READ_LOCK, + inode); + if (result != CBK_COORD_FOUND) { + /* + * error happened of not items of file + * were found + */ + done_lh(&lh); + page_cache_release(page); + goto error; + } + + if (coord.between == AFTER_UNIT) { + /* + * end of file is reached. Padd page + * with zeros + */ + done_lh(&lh); + done = 1; + p_data = kmap_atomic(page, KM_USER0); + memset(p_data + page_off, 0, + PAGE_CACHE_SIZE - page_off); + kunmap_atomic(p_data, KM_USER0); + break; + } + + result = zload(coord.node); + if (result) { + page_cache_release(page); + done_lh(&lh); + goto error; + } + assert("vs-856", coord.between == AT_UNIT); + item = ((char *)item_body_by_coord(&coord)) + + coord.unit_pos; + + /* how many bytes to copy */ + count = + item_length_by_coord(&coord) - + coord.unit_pos; + /* limit length of copy to end of page */ + if (count > PAGE_CACHE_SIZE - page_off) + count = PAGE_CACHE_SIZE - page_off; + + /* + * copy item (as much as will fit starting from + * the beginning of the item) into the page + */ + p_data = kmap_atomic(page, KM_USER0); + memcpy(p_data + page_off, item, count); + kunmap_atomic(p_data, KM_USER0); + + page_off += count; + bytes += count; + set_key_offset(&key, + get_key_offset(&key) + count); + + zrelse(coord.node); + done_lh(&lh); + } /* end of loop which fills one page by content of + * formatting items */ + + if (page_off) { + /* something was copied into page */ + pages[i] = page; + } else { + page_cache_release(page); + assert("vs-1648", done == 1); + break; + } + } /* end of loop through pages of one conversion iteration */ + + if (i > 0) { + result = replace(inode, pages, i, bytes); + release_all_pages(pages, sizeof_array(pages)); + if (result) + goto error; + /* + * We have to drop exclusive access to avoid deadlock + * which may happen because called by reiser4_writepages + * capture_unix_file requires to get non-exclusive + * access to a file. It is safe to drop EA in the middle + * of tail2extent conversion because write_unix_file, + * setattr_unix_file(truncate), mmap_unix_file, + * release_unix_file(extent2tail) checks if conversion + * is not in progress (see comments before + * get_exclusive_access_careful(). + * Other processes that acquire non-exclusive access + * (read_unix_file, reiser4_writepages, etc) should work + * on partially converted files. + */ + drop_exclusive_access(uf_info); + /* throttle the conversion */ + reiser4_throttle_write(inode); + get_exclusive_access(uf_info); + + /* + * nobody is allowed to complete conversion but a + * process which started it + */ + assert("", reiser4_inode_get_flag(inode, + REISER4_PART_MIXED)); + } + } + if (result == 0) { + /* file is converted to extent items */ + reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV); + assert("vs-1697", reiser4_inode_get_flag(inode, + REISER4_PART_MIXED)); + + uf_info->container = UF_CONTAINER_EXTENTS; + complete_conversion(inode); + } else { + /* + * conversion is not complete. Inode was already marked as + * REISER4_PART_MIXED and stat-data were updated at the first + * iteration of the loop above. + */ + error: + release_all_pages(pages, sizeof_array(pages)); + reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV); + warning("edward-1548", "Partial conversion of %llu: %i", + (unsigned long long)get_inode_oid(inode), result); + } + + out: + /* this flag should be cleared, otherwise get_exclusive_access_careful() + will fall into infinite loop */ + assert("edward-1549", !reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)); + return result; +} + +static int reserve_extent2tail_iteration(struct inode *inode) +{ + reiser4_tree *tree; + + tree = reiser4_tree_by_inode(inode); + /* + * reserve blocks for (in this order): + * + * 1. removal of extent item + * + * 2. insertion of tail by insert_flow() + * + * 3. drilling to the leaf level by coord_by_key() + * + * 4. possible update of stat-data + */ + grab_space_enable(); + return reiser4_grab_space + (estimate_one_item_removal(tree) + + estimate_insert_flow(tree->height) + + 1 + estimate_one_insert_item(tree) + + inode_file_plugin(inode)->estimate.update(inode), BA_CAN_COMMIT); +} + +/* for every page of file: read page, cut part of extent pointing to this page, + put data of page tree by tail item */ +int extent2tail(struct file * file, struct unix_file_info *uf_info) +{ + int result; + struct inode *inode; + struct page *page; + unsigned long num_pages, i; + unsigned long start_page; + reiser4_key from; + reiser4_key to; + unsigned count; + __u64 offset; + + assert("nikita-3362", ea_obtained(uf_info)); + inode = unix_file_info_to_inode(uf_info); + assert("nikita-3412", !IS_RDONLY(inode)); + assert("vs-1649", uf_info->container != UF_CONTAINER_TAILS); + assert("", !reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV)); + + offset = 0; + if (reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) { + /* + * file is marked on disk as there was a conversion which did + * not complete due to either crash or some error. Find which + * offset tail conversion stopped at + */ + result = find_start(inode, EXTENT_POINTER_ID, &offset); + if (result == -ENOENT) { + /* no extent found, everything is converted */ + uf_info->container = UF_CONTAINER_TAILS; + complete_conversion(inode); + return 0; + } else if (result != 0) + /* some other error */ + return result; + } + + reiser4_inode_set_flag(inode, REISER4_PART_IN_CONV); + + /* number of pages in the file */ + num_pages = + (inode->i_size + - offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + start_page = offset >> PAGE_CACHE_SHIFT; + + inode_file_plugin(inode)->key_by_inode(inode, offset, &from); + to = from; + + result = 0; + for (i = 0; i < num_pages; i++) { + __u64 start_byte; + + result = reserve_extent2tail_iteration(inode); + if (result != 0) + break; + if (i == 0 && offset == 0) { + reiser4_inode_set_flag(inode, REISER4_PART_MIXED); + reiser4_update_sd(inode); + } + + page = read_mapping_page(inode->i_mapping, + (unsigned)(i + start_page), NULL); + if (IS_ERR(page)) { + result = PTR_ERR(page); + break; + } + + wait_on_page_locked(page); + + if (!PageUptodate(page)) { + page_cache_release(page); + result = RETERR(-EIO); + break; + } + + /* cut part of file we have read */ + start_byte = (__u64) ((i + start_page) << PAGE_CACHE_SHIFT); + set_key_offset(&from, start_byte); + set_key_offset(&to, start_byte + PAGE_CACHE_SIZE - 1); + /* + * reiser4_cut_tree_object() returns -E_REPEAT to allow atom + * commits during over-long truncates. But + * extent->tail conversion should be performed in one + * transaction. + */ + result = reiser4_cut_tree(reiser4_tree_by_inode(inode), &from, + &to, inode, 0); + + if (result) { + page_cache_release(page); + break; + } + + /* put page data into tree via tail_write */ + count = PAGE_CACHE_SIZE; + if ((i == (num_pages - 1)) && + (inode->i_size & ~PAGE_CACHE_MASK)) + /* last page can be incompleted */ + count = (inode->i_size & ~PAGE_CACHE_MASK); + while (count) { + loff_t pos = start_byte; + + assert("edward-1537", + file != NULL && file->f_dentry != NULL); + assert("edward-1538", + file->f_dentry->d_inode == inode); + + result = reiser4_write_tail(file, inode, + (char __user *)kmap(page), + count, &pos); + reiser4_free_file_fsdata(file); + if (result <= 0) { + warning("", "reiser4_write_tail failed"); + page_cache_release(page); + reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV); + return result; + } + count -= result; + } + + /* release page */ + lock_page(page); + /* page is already detached from jnode and mapping. */ + assert("vs-1086", page->mapping == NULL); + assert("nikita-2690", + (!PagePrivate(page) && jprivate(page) == 0)); + /* waiting for writeback completion with page lock held is + * perfectly valid. */ + wait_on_page_writeback(page); + reiser4_drop_page(page); + /* release reference taken by read_cache_page() above */ + page_cache_release(page); + + drop_exclusive_access(uf_info); + /* throttle the conversion */ + reiser4_throttle_write(inode); + get_exclusive_access(uf_info); + /* + * nobody is allowed to complete conversion but a process which + * started it + */ + assert("", reiser4_inode_get_flag(inode, REISER4_PART_MIXED)); + } + + reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV); + + if (i == num_pages) { + /* file is converted to formatted items */ + assert("vs-1698", reiser4_inode_get_flag(inode, + REISER4_PART_MIXED)); + assert("vs-1260", + inode_has_no_jnodes(reiser4_inode_data(inode))); + + uf_info->container = UF_CONTAINER_TAILS; + complete_conversion(inode); + return 0; + } + /* + * conversion is not complete. Inode was already marked as + * REISER4_PART_MIXED and stat-data were updated at the first + * iteration of the loop above. + */ + warning("nikita-2282", + "Partial conversion of %llu: %lu of %lu: %i", + (unsigned long long)get_inode_oid(inode), i, + num_pages, result); + + /* this flag should be cleared, otherwise get_exclusive_access_careful() + will fall into infinite loop */ + assert("edward-1550", !reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)); + return result; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file_ops.c linux-2.6.30/fs/reiser4/plugin/file_ops.c --- linux-2.6.30.orig/fs/reiser4/plugin/file_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,162 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + reiser4/README */ + +/* this file contains typical implementations for some of methods of + struct file_operations and of struct address_space_operations +*/ + +#include "../inode.h" +#include "object.h" + +/* file operations */ + +/* implementation of vfs's llseek method of struct file_operations for + typical directory can be found in readdir_common.c +*/ +loff_t reiser4_llseek_dir_common(struct file *, loff_t, int origin); + +/* implementation of vfs's readdir method of struct file_operations for + typical directory can be found in readdir_common.c +*/ +int reiser4_readdir_common(struct file *, void *dirent, filldir_t); + +/** + * reiser4_release_dir_common - release of struct file_operations + * @inode: inode of released file + * @file: file to release + * + * Implementation of release method of struct file_operations for typical + * directory. All it does is freeing of reiser4 specific file data. +*/ +int reiser4_release_dir_common(struct inode *inode, struct file *file) +{ + reiser4_context *ctx; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + reiser4_free_file_fsdata(file); + reiser4_exit_context(ctx); + return 0; +} + +/* this is common implementation of vfs's fsync method of struct + file_operations +*/ +int reiser4_sync_common(struct file *file, struct dentry *dentry, int datasync) +{ + reiser4_context *ctx; + int result; + + ctx = reiser4_init_context(dentry->d_inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + result = txnmgr_force_commit_all(dentry->d_inode->i_sb, 0); + + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} + +/* + * common sync method for regular files. + * + * We are trying to be smart here. Instead of committing all atoms (original + * solution), we scan dirty pages of this file and commit all atoms they are + * part of. + * + * Situation is complicated by anonymous pages: i.e., extent-less pages + * dirtied through mmap. Fortunately sys_fsync() first calls + * filemap_fdatawrite() that will ultimately call reiser4_writepages(), insert + * all missing extents and capture anonymous pages. + */ +int reiser4_sync_file_common(struct file *file, + struct dentry *dentry, int datasync) +{ + reiser4_context *ctx; + txn_atom *atom; + reiser4_block_nr reserve; + + ctx = reiser4_init_context(dentry->d_inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + reserve = estimate_update_common(dentry->d_inode); + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) { + reiser4_exit_context(ctx); + return RETERR(-ENOSPC); + } + write_sd_by_inode_common(dentry->d_inode); + + atom = get_current_atom_locked(); + spin_lock_txnh(ctx->trans); + force_commit_atom(ctx->trans); + reiser4_exit_context(ctx); + return 0; +} + + +/* address space operations */ + + +/* this is helper for plugin->write_begin() */ +int do_prepare_write(struct file *file, struct page *page, unsigned from, + unsigned to) +{ + int result; + file_plugin *fplug; + struct inode *inode; + + assert("umka-3099", file != NULL); + assert("umka-3100", page != NULL); + assert("umka-3095", PageLocked(page)); + + if (to - from == PAGE_CACHE_SIZE || PageUptodate(page)) + return 0; + + inode = page->mapping->host; + fplug = inode_file_plugin(inode); + + if (page->mapping->a_ops->readpage == NULL) + return RETERR(-EINVAL); + + result = page->mapping->a_ops->readpage(file, page); + if (result != 0) { + SetPageError(page); + ClearPageUptodate(page); + /* All reiser4 readpage() implementations should return the + * page locked in case of error. */ + assert("nikita-3472", PageLocked(page)); + } else { + /* + * ->readpage() either: + * + * 1. starts IO against @page. @page is locked for IO in + * this case. + * + * 2. doesn't start IO. @page is unlocked. + * + * In either case, page should be locked. + */ + lock_page(page); + /* + * IO (if any) is completed at this point. Check for IO + * errors. + */ + if (!PageUptodate(page)) + result = RETERR(-EIO); + } + assert("umka-3098", PageLocked(page)); + return result; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file_ops_readdir.c linux-2.6.30/fs/reiser4/plugin/file_ops_readdir.c --- linux-2.6.30.orig/fs/reiser4/plugin/file_ops_readdir.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file_ops_readdir.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,658 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "../inode.h" + +/* return true, iff @coord points to the valid directory item that is part of + * @inode directory. */ +static int is_valid_dir_coord(struct inode *inode, coord_t *coord) +{ + return plugin_of_group(item_plugin_by_coord(coord), + DIR_ENTRY_ITEM_TYPE) && + inode_file_plugin(inode)->owns_item(inode, coord); +} + +/* compare two logical positions within the same directory */ +static cmp_t dir_pos_cmp(const struct dir_pos *p1, const struct dir_pos *p2) +{ + cmp_t result; + + assert("nikita-2534", p1 != NULL); + assert("nikita-2535", p2 != NULL); + + result = de_id_cmp(&p1->dir_entry_key, &p2->dir_entry_key); + if (result == EQUAL_TO) { + int diff; + + diff = p1->pos - p2->pos; + result = + (diff < 0) ? LESS_THAN : (diff ? GREATER_THAN : EQUAL_TO); + } + return result; +} + +/* see comment before reiser4_readdir_common() for overview of why "adjustment" + * is necessary. */ +static void +adjust_dir_pos(struct file *dir, struct readdir_pos *readdir_spot, + const struct dir_pos *mod_point, int adj) +{ + struct dir_pos *pos; + + /* + * new directory entry was added (adj == +1) or removed (adj == -1) at + * the @mod_point. Directory file descriptor @dir is doing readdir and + * is currently positioned at @readdir_spot. Latter has to be updated + * to maintain stable readdir. + */ + /* directory is positioned to the beginning. */ + if (readdir_spot->entry_no == 0) + return; + + pos = &readdir_spot->position; + switch (dir_pos_cmp(mod_point, pos)) { + case LESS_THAN: + /* @mod_pos is _before_ @readdir_spot, that is, entry was + * added/removed on the left (in key order) of current + * position. */ + /* logical number of directory entry readdir is "looking" at + * changes */ + readdir_spot->entry_no += adj; + assert("nikita-2577", + ergo(dir != NULL, reiser4_get_dir_fpos(dir) + adj >= 0)); + if (de_id_cmp(&pos->dir_entry_key, + &mod_point->dir_entry_key) == EQUAL_TO) { + assert("nikita-2575", mod_point->pos < pos->pos); + /* + * if entry added/removed has the same key as current + * for readdir, update counter of duplicate keys in + * @readdir_spot. + */ + pos->pos += adj; + } + break; + case GREATER_THAN: + /* directory is modified after @pos: nothing to do. */ + break; + case EQUAL_TO: + /* cannot insert an entry readdir is looking at, because it + already exists. */ + assert("nikita-2576", adj < 0); + /* directory entry to which @pos points to is being + removed. + + NOTE-NIKITA: Right thing to do is to update @pos to point + to the next entry. This is complex (we are under spin-lock + for one thing). Just rewind it to the beginning. Next + readdir will have to scan the beginning of + directory. Proper solution is to use semaphore in + spin lock's stead and use rewind_right() here. + + NOTE-NIKITA: now, semaphore is used, so... + */ + memset(readdir_spot, 0, sizeof *readdir_spot); + } +} + +/* scan all file-descriptors for this directory and adjust their + positions respectively. Should be used by implementations of + add_entry and rem_entry of dir plugin */ +void reiser4_adjust_dir_file(struct inode *dir, const struct dentry *de, + int offset, int adj) +{ + reiser4_file_fsdata *scan; + struct dir_pos mod_point; + + assert("nikita-2536", dir != NULL); + assert("nikita-2538", de != NULL); + assert("nikita-2539", adj != 0); + + build_de_id(dir, &de->d_name, &mod_point.dir_entry_key); + mod_point.pos = offset; + + spin_lock_inode(dir); + + /* + * new entry was added/removed in directory @dir. Scan all file + * descriptors for @dir that are currently involved into @readdir and + * update them. + */ + + list_for_each_entry(scan, get_readdir_list(dir), dir.linkage) + adjust_dir_pos(scan->back, &scan->dir.readdir, &mod_point, adj); + + spin_unlock_inode(dir); +} + +/* + * traverse tree to start/continue readdir from the readdir position @pos. + */ +static int dir_go_to(struct file *dir, struct readdir_pos *pos, tap_t *tap) +{ + reiser4_key key; + int result; + struct inode *inode; + + assert("nikita-2554", pos != NULL); + + inode = dir->f_dentry->d_inode; + result = inode_dir_plugin(inode)->build_readdir_key(dir, &key); + if (result != 0) + return result; + result = reiser4_object_lookup(inode, + &key, + tap->coord, + tap->lh, + tap->mode, + FIND_EXACT, + LEAF_LEVEL, LEAF_LEVEL, + 0, &tap->ra_info); + if (result == CBK_COORD_FOUND) + result = rewind_right(tap, (int)pos->position.pos); + else { + tap->coord->node = NULL; + done_lh(tap->lh); + result = RETERR(-EIO); + } + return result; +} + +/* + * handling of non-unique keys: calculate at what ordinal position within + * sequence of directory items with identical keys @pos is. + */ +static int set_pos(struct inode *inode, struct readdir_pos *pos, tap_t *tap) +{ + int result; + coord_t coord; + lock_handle lh; + tap_t scan; + de_id *did; + reiser4_key de_key; + + coord_init_zero(&coord); + init_lh(&lh); + reiser4_tap_init(&scan, &coord, &lh, ZNODE_READ_LOCK); + reiser4_tap_copy(&scan, tap); + reiser4_tap_load(&scan); + pos->position.pos = 0; + + did = &pos->position.dir_entry_key; + + if (is_valid_dir_coord(inode, scan.coord)) { + + build_de_id_by_key(unit_key_by_coord(scan.coord, &de_key), did); + + while (1) { + + result = go_prev_unit(&scan); + if (result != 0) + break; + + if (!is_valid_dir_coord(inode, scan.coord)) { + result = -EINVAL; + break; + } + + /* get key of directory entry */ + unit_key_by_coord(scan.coord, &de_key); + if (de_id_key_cmp(did, &de_key) != EQUAL_TO) { + /* duplicate-sequence is over */ + break; + } + pos->position.pos++; + } + } else + result = RETERR(-ENOENT); + reiser4_tap_relse(&scan); + reiser4_tap_done(&scan); + return result; +} + +/* + * "rewind" directory to @offset, i.e., set @pos and @tap correspondingly. + */ +static int dir_rewind(struct file *dir, struct readdir_pos *pos, tap_t *tap) +{ + __u64 destination; + __s64 shift; + int result; + struct inode *inode; + loff_t dirpos; + + assert("nikita-2553", dir != NULL); + assert("nikita-2548", pos != NULL); + assert("nikita-2551", tap->coord != NULL); + assert("nikita-2552", tap->lh != NULL); + + dirpos = reiser4_get_dir_fpos(dir); + shift = dirpos - pos->fpos; + /* this is logical directory entry within @dir which we are rewinding + * to */ + destination = pos->entry_no + shift; + + inode = dir->f_dentry->d_inode; + if (dirpos < 0) + return RETERR(-EINVAL); + else if (destination == 0ll || dirpos == 0) { + /* rewind to the beginning of directory */ + memset(pos, 0, sizeof *pos); + return dir_go_to(dir, pos, tap); + } else if (destination >= inode->i_size) + return RETERR(-ENOENT); + + if (shift < 0) { + /* I am afraid of negative numbers */ + shift = -shift; + /* rewinding to the left */ + if (shift <= (int)pos->position.pos) { + /* destination is within sequence of entries with + duplicate keys. */ + result = dir_go_to(dir, pos, tap); + } else { + shift -= pos->position.pos; + while (1) { + /* repetitions: deadlock is possible when + going to the left. */ + result = dir_go_to(dir, pos, tap); + if (result == 0) { + result = rewind_left(tap, shift); + if (result == -E_DEADLOCK) { + reiser4_tap_done(tap); + continue; + } + } + break; + } + } + } else { + /* rewinding to the right */ + result = dir_go_to(dir, pos, tap); + if (result == 0) + result = rewind_right(tap, shift); + } + if (result == 0) { + result = set_pos(inode, pos, tap); + if (result == 0) { + /* update pos->position.pos */ + pos->entry_no = destination; + pos->fpos = dirpos; + } + } + return result; +} + +/* + * Function that is called by common_readdir() on each directory entry while + * doing readdir. ->filldir callback may block, so we had to release long term + * lock while calling it. To avoid repeating tree traversal, seal is used. If + * seal is broken, we return -E_REPEAT. Node is unlocked in this case. + * + * Whether node is unlocked in case of any other error is undefined. It is + * guaranteed to be still locked if success (0) is returned. + * + * When ->filldir() wants no more, feed_entry() returns 1, and node is + * unlocked. + */ +static int +feed_entry(struct file *f, struct readdir_pos *pos, tap_t *tap, + filldir_t filldir, void *dirent) +{ + item_plugin *iplug; + char *name; + reiser4_key sd_key; + int result; + char buf[DE_NAME_BUF_LEN]; + char name_buf[32]; + char *local_name; + unsigned file_type; + seal_t seal; + coord_t *coord; + reiser4_key entry_key; + + coord = tap->coord; + iplug = item_plugin_by_coord(coord); + + /* pointer to name within the node */ + name = iplug->s.dir.extract_name(coord, buf); + assert("nikita-1371", name != NULL); + + /* key of object the entry points to */ + if (iplug->s.dir.extract_key(coord, &sd_key) != 0) + return RETERR(-EIO); + + /* we must release longterm znode lock before calling filldir to avoid + deadlock which may happen if filldir causes page fault. So, copy + name to intermediate buffer */ + if (strlen(name) + 1 > sizeof(name_buf)) { + local_name = kmalloc(strlen(name) + 1, + reiser4_ctx_gfp_mask_get()); + if (local_name == NULL) + return RETERR(-ENOMEM); + } else + local_name = name_buf; + + strcpy(local_name, name); + file_type = iplug->s.dir.extract_file_type(coord); + + unit_key_by_coord(coord, &entry_key); + reiser4_seal_init(&seal, coord, &entry_key); + + longterm_unlock_znode(tap->lh); + + /* + * send information about directory entry to the ->filldir() filler + * supplied to us by caller (VFS). + * + * ->filldir is entitled to do weird things. For example, ->filldir + * supplied by knfsd re-enters file system. Make sure no locks are + * held. + */ + assert("nikita-3436", lock_stack_isclean(get_current_lock_stack())); + + reiser4_txn_restart_current(); + result = filldir(dirent, name, (int)strlen(name), + /* offset of this entry */ + f->f_pos, + /* inode number of object bounden by this entry */ + oid_to_uino(get_key_objectid(&sd_key)), file_type); + if (local_name != name_buf) + kfree(local_name); + if (result < 0) + /* ->filldir() is satisfied. (no space in buffer, IOW) */ + result = 1; + else + result = reiser4_seal_validate(&seal, coord, &entry_key, + tap->lh, tap->mode, + ZNODE_LOCK_HIPRI); + return result; +} + +static void move_entry(struct readdir_pos *pos, coord_t *coord) +{ + reiser4_key de_key; + de_id *did; + + /* update @pos */ + ++pos->entry_no; + did = &pos->position.dir_entry_key; + + /* get key of directory entry */ + unit_key_by_coord(coord, &de_key); + + if (de_id_key_cmp(did, &de_key) == EQUAL_TO) + /* we are within sequence of directory entries + with duplicate keys. */ + ++pos->position.pos; + else { + pos->position.pos = 0; + build_de_id_by_key(&de_key, did); + } + ++pos->fpos; +} + +/* + * STATELESS READDIR + * + * readdir support in reiser4 relies on ability to update readdir_pos embedded + * into reiser4_file_fsdata on each directory modification (name insertion and + * removal), see reiser4_readdir_common() function below. This obviously doesn't + * work when reiser4 is accessed over NFS, because NFS doesn't keep any state + * across client READDIR requests for the same directory. + * + * To address this we maintain a "pool" of detached reiser4_file_fsdata + * (d_cursor). Whenever NFS readdir request comes, we detect this, and try to + * find detached reiser4_file_fsdata corresponding to previous readdir + * request. In other words, additional state is maintained on the + * server. (This is somewhat contrary to the design goals of NFS protocol.) + * + * To efficiently detect when our ->readdir() method is called by NFS server, + * dentry is marked as "stateless" in reiser4_decode_fh() (this is checked by + * file_is_stateless() function). + * + * To find out d_cursor in the pool, we encode client id (cid) in the highest + * bits of NFS readdir cookie: when first readdir request comes to the given + * directory from the given client, cookie is set to 0. This situation is + * detected, global cid_counter is incremented, and stored in highest bits of + * all direntry offsets returned to the client, including last one. As the + * only valid readdir cookie is one obtained as direntry->offset, we are + * guaranteed that next readdir request (continuing current one) will have + * current cid in the highest bits of starting readdir cookie. All d_cursors + * are hashed into per-super-block hash table by (oid, cid) key. + * + * In addition d_cursors are placed into per-super-block radix tree where they + * are keyed by oid alone. This is necessary to efficiently remove them during + * rmdir. + * + * At last, currently unused d_cursors are linked into special list. This list + * is used d_cursor_shrink to reclaim d_cursors on memory pressure. + * + */ + +/* + * prepare for readdir. + */ +static int dir_readdir_init(struct file *f, tap_t *tap, + struct readdir_pos **pos) +{ + struct inode *inode; + reiser4_file_fsdata *fsdata; + int result; + + assert("nikita-1359", f != NULL); + inode = f->f_dentry->d_inode; + assert("nikita-1360", inode != NULL); + + if (!S_ISDIR(inode->i_mode)) + return RETERR(-ENOTDIR); + + /* try to find detached readdir state */ + result = reiser4_attach_fsdata(f, inode); + if (result != 0) + return result; + + fsdata = reiser4_get_file_fsdata(f); + assert("nikita-2571", fsdata != NULL); + if (IS_ERR(fsdata)) + return PTR_ERR(fsdata); + + /* add file descriptor to the readdir list hanging of directory + * inode. This list is used to scan "readdirs-in-progress" while + * inserting or removing names in the directory. */ + spin_lock_inode(inode); + if (list_empty_careful(&fsdata->dir.linkage)) + list_add(&fsdata->dir.linkage, get_readdir_list(inode)); + *pos = &fsdata->dir.readdir; + spin_unlock_inode(inode); + + /* move @tap to the current position */ + return dir_rewind(f, *pos, tap); +} + +/* this is implementation of vfs's llseek method of struct file_operations for + typical directory + See comment before reiser4_readdir_common() for explanation. +*/ +loff_t reiser4_llseek_dir_common(struct file *file, loff_t off, int origin) +{ + reiser4_context *ctx; + loff_t result; + struct inode *inode; + + inode = file->f_dentry->d_inode; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + mutex_lock(&inode->i_mutex); + + /* update ->f_pos */ + result = default_llseek(file, off, origin); + if (result >= 0) { + int ff; + coord_t coord; + lock_handle lh; + tap_t tap; + struct readdir_pos *pos; + + coord_init_zero(&coord); + init_lh(&lh); + reiser4_tap_init(&tap, &coord, &lh, ZNODE_READ_LOCK); + + ff = dir_readdir_init(file, &tap, &pos); + reiser4_detach_fsdata(file); + if (ff != 0) + result = (loff_t) ff; + reiser4_tap_done(&tap); + } + reiser4_detach_fsdata(file); + mutex_unlock(&inode->i_mutex); + + reiser4_exit_context(ctx); + return result; +} + +/* this is common implementation of vfs's readdir method of struct + file_operations + + readdir problems: + + readdir(2)/getdents(2) interface is based on implicit assumption that + readdir can be restarted from any particular point by supplying file system + with off_t-full of data. That is, file system fills ->d_off field in struct + dirent and later user passes ->d_off to the seekdir(3), which is, actually, + implemented by glibc as lseek(2) on directory. + + Reiser4 cannot restart readdir from 64 bits of data, because two last + components of the key of directory entry are unknown, which given 128 bits: + locality and type fields in the key of directory entry are always known, to + start readdir() from given point objectid and offset fields have to be + filled. + + Traditional UNIX API for scanning through directory + (readdir/seekdir/telldir/opendir/closedir/rewindir/getdents) is based on the + assumption that directory is structured very much like regular file, in + particular, it is implied that each name within given directory (directory + entry) can be uniquely identified by scalar offset and that such offset is + stable across the life-time of the name is identifies. + + This is manifestly not so for reiser4. In reiser4 the only stable unique + identifies for the directory entry is its key that doesn't fit into + seekdir/telldir API. + + solution: + + Within each file descriptor participating in readdir-ing of directory + plugin/dir/dir.h:readdir_pos is maintained. This structure keeps track of + the "current" directory entry that file descriptor looks at. It contains a + key of directory entry (plus some additional info to deal with non-unique + keys that we wouldn't dwell onto here) and a logical position of this + directory entry starting from the beginning of the directory, that is + ordinal number of this entry in the readdir order. + + Obviously this logical position is not stable in the face of directory + modifications. To work around this, on each addition or removal of directory + entry all file descriptors for directory inode are scanned and their + readdir_pos are updated accordingly (adjust_dir_pos()). +*/ +int reiser4_readdir_common(struct file *f /* directory file being read */, + void *dirent /* opaque data passed to us by VFS */, + filldir_t filld /* filler function passed to us + * by VFS */) +{ + reiser4_context *ctx; + int result; + struct inode *inode; + coord_t coord; + lock_handle lh; + tap_t tap; + struct readdir_pos *pos; + + assert("nikita-1359", f != NULL); + inode = f->f_dentry->d_inode; + assert("nikita-1360", inode != NULL); + + if (!S_ISDIR(inode->i_mode)) + return RETERR(-ENOTDIR); + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + coord_init_zero(&coord); + init_lh(&lh); + reiser4_tap_init(&tap, &coord, &lh, ZNODE_READ_LOCK); + + reiser4_readdir_readahead_init(inode, &tap); + +repeat: + result = dir_readdir_init(f, &tap, &pos); + if (result == 0) { + result = reiser4_tap_load(&tap); + /* scan entries one by one feeding them to @filld */ + while (result == 0) { + coord_t *coord; + + coord = tap.coord; + assert("nikita-2572", coord_is_existing_unit(coord)); + assert("nikita-3227", is_valid_dir_coord(inode, coord)); + + result = feed_entry(f, pos, &tap, filld, dirent); + if (result > 0) { + break; + } else if (result == 0) { + ++f->f_pos; + result = go_next_unit(&tap); + if (result == -E_NO_NEIGHBOR || + result == -ENOENT) { + result = 0; + break; + } else if (result == 0) { + if (is_valid_dir_coord(inode, coord)) + move_entry(pos, coord); + else + break; + } + } else if (result == -E_REPEAT) { + /* feed_entry() had to restart. */ + ++f->f_pos; + reiser4_tap_relse(&tap); + goto repeat; + } else + warning("vs-1617", + "reiser4_readdir_common: unexpected error %d", + result); + } + reiser4_tap_relse(&tap); + + if (result >= 0) + f->f_version = inode->i_version; + } else if (result == -E_NO_NEIGHBOR || result == -ENOENT) + result = 0; + reiser4_tap_done(&tap); + reiser4_detach_fsdata(f); + + /* try to update directory's atime */ + if (reiser4_grab_space_force(inode_file_plugin(inode)->estimate.update(inode), + BA_CAN_COMMIT) != 0) + warning("", "failed to update atime on readdir: %llu", + get_inode_oid(inode)); + else + file_accessed(f); + + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + + return (result <= 0) ? result : 0; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/file_plugin_common.c linux-2.6.30/fs/reiser4/plugin/file_plugin_common.c --- linux-2.6.30.orig/fs/reiser4/plugin/file_plugin_common.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/file_plugin_common.c 2009-06-22 17:27:31.000000000 +0200 @@ -0,0 +1,1008 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + reiser4/README */ + +/* this file contains typical implementations for most of methods of + file plugin +*/ + +#include "../inode.h" +#include "object.h" +#include "../safe_link.h" + +#include + +static int insert_new_sd(struct inode *inode); +static int update_sd(struct inode *inode); + +/* this is common implementation of write_sd_by_inode method of file plugin + either insert stat data or update it + */ +int write_sd_by_inode_common(struct inode *inode/* object to save */) +{ + int result; + + assert("nikita-730", inode != NULL); + + if (reiser4_inode_get_flag(inode, REISER4_NO_SD)) + /* object doesn't have stat-data yet */ + result = insert_new_sd(inode); + else + result = update_sd(inode); + if (result != 0 && result != -ENAMETOOLONG && result != -ENOMEM) + /* Don't issue warnings about "name is too long" */ + warning("nikita-2221", "Failed to save sd for %llu: %i", + (unsigned long long)get_inode_oid(inode), result); + return result; +} + +/* this is common implementation of key_by_inode method of file plugin + */ +int +key_by_inode_and_offset_common(struct inode *inode, loff_t off, + reiser4_key * key) +{ + reiser4_key_init(key); + set_key_locality(key, reiser4_inode_data(inode)->locality_id); + set_key_ordering(key, get_inode_ordering(inode)); + set_key_objectid(key, get_inode_oid(inode)); /*FIXME: inode->i_ino */ + set_key_type(key, KEY_BODY_MINOR); + set_key_offset(key, (__u64) off); + return 0; +} + +/* this is common implementation of set_plug_in_inode method of file plugin + */ +int set_plug_in_inode_common(struct inode *object /* inode to set plugin on */ , + struct inode *parent /* parent object */ , + reiser4_object_create_data * data /* creational + * data */ ) +{ + __u64 mask; + + object->i_mode = data->mode; + /* this should be plugin decision */ + object->i_uid = current->cred->fsuid; + object->i_mtime = object->i_atime = object->i_ctime = CURRENT_TIME; + + /* support for BSD style group-id assignment. See mount's manual page + description of bsdgroups ext2 mount options for more details */ + if (reiser4_is_set(object->i_sb, REISER4_BSD_GID)) + object->i_gid = parent->i_gid; + else if (parent->i_mode & S_ISGID) { + /* parent directory has sguid bit */ + object->i_gid = parent->i_gid; + if (S_ISDIR(object->i_mode)) + /* sguid is inherited by sub-directories */ + object->i_mode |= S_ISGID; + } else + object->i_gid = current->cred->fsgid; + + /* this object doesn't have stat-data yet */ + reiser4_inode_set_flag(object, REISER4_NO_SD); +#if 0 + /* this is now called after all inode plugins are initialized: + do_create_vfs_child after adjust_to_parent */ + /* setup inode and file-operations for this inode */ + setup_inode_ops(object, data); +#endif + object->i_nlink = 0; + reiser4_seal_init(&reiser4_inode_data(object)->sd_seal, NULL, NULL); + mask = (1 << UNIX_STAT) | (1 << LIGHT_WEIGHT_STAT); + if (!reiser4_is_set(object->i_sb, REISER4_32_BIT_TIMES)) + mask |= (1 << LARGE_TIMES_STAT); + + reiser4_inode_data(object)->extmask = mask; + return 0; +} + +/* this is common implementation of adjust_to_parent method of file plugin for + regular files + */ +int adjust_to_parent_common(struct inode *object /* new object */ , + struct inode *parent /* parent directory */ , + struct inode *root/* root directory */) +{ + assert("nikita-2165", object != NULL); + if (parent == NULL) + parent = root; + assert("nikita-2069", parent != NULL); + + /* + * inherit missing plugins from parent + */ + + grab_plugin_pset(object, parent, PSET_FILE); + grab_plugin_pset(object, parent, PSET_SD); + grab_plugin_pset(object, parent, PSET_FORMATTING); + grab_plugin_pset(object, parent, PSET_PERM); + return 0; +} + +/* this is common implementation of adjust_to_parent method of file plugin for + typical directories + */ +int adjust_to_parent_common_dir(struct inode *object /* new object */ , + struct inode *parent /* parent directory */ , + struct inode *root/* root directory */) +{ + int result = 0; + pset_member memb; + + assert("nikita-2166", object != NULL); + if (parent == NULL) + parent = root; + assert("nikita-2167", parent != NULL); + + /* + * inherit missing plugins from parent + */ + for (memb = 0; memb < PSET_LAST; ++memb) { + result = grab_plugin_pset(object, parent, memb); + if (result != 0) + break; + } + return result; +} + +int adjust_to_parent_cryptcompress(struct inode *object /* new object */ , + struct inode *parent /* parent directory */, + struct inode *root/* root directory */) +{ + int result; + result = adjust_to_parent_common(object, parent, root); + if (result) + return result; + assert("edward-1416", parent != NULL); + + grab_plugin_pset(object, parent, PSET_CLUSTER); + grab_plugin_pset(object, parent, PSET_CIPHER); + grab_plugin_pset(object, parent, PSET_DIGEST); + grab_plugin_pset(object, parent, PSET_COMPRESSION); + grab_plugin_pset(object, parent, PSET_COMPRESSION_MODE); + + return 0; +} + +/* this is common implementation of create_object method of file plugin + */ +int reiser4_create_object_common(struct inode *object, struct inode *parent, + reiser4_object_create_data * data) +{ + reiser4_block_nr reserve; + assert("nikita-744", object != NULL); + assert("nikita-745", parent != NULL); + assert("nikita-747", data != NULL); + assert("nikita-748", reiser4_inode_get_flag(object, REISER4_NO_SD)); + + reserve = estimate_create_common(object); + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) + return RETERR(-ENOSPC); + return write_sd_by_inode_common(object); +} + +static int common_object_delete_no_reserve(struct inode *inode); + +/** + * reiser4_delete_object_common - delete_object of file_plugin + * @inode: inode to be deleted + * + * This is common implementation of delete_object method of file_plugin. It + * applies to object its deletion consists of removing two items - stat data + * and safe-link. + */ +int reiser4_delete_object_common(struct inode *inode) +{ + int result; + + assert("nikita-1477", inode != NULL); + /* FIXME: if file body deletion failed (i/o error, for instance), + inode->i_size can be != 0 here */ + assert("nikita-3420", inode->i_size == 0 || S_ISLNK(inode->i_mode)); + assert("nikita-3421", inode->i_nlink == 0); + + if (!reiser4_inode_get_flag(inode, REISER4_NO_SD)) { + reiser4_block_nr reserve; + + /* grab space which is needed to remove 2 items from the tree: + stat data and safe-link */ + reserve = 2 * + estimate_one_item_removal(reiser4_tree_by_inode(inode)); + if (reiser4_grab_space_force(reserve, + BA_RESERVED | BA_CAN_COMMIT)) + return RETERR(-ENOSPC); + result = common_object_delete_no_reserve(inode); + } else + result = 0; + return result; +} + +/** + * reiser4_delete_dir_common - delete_object of file_plugin + * @inode: inode to be deleted + * + * This is common implementation of delete_object method of file_plugin for + * typical directory. It calls done method of dir_plugin to remove "." and + * removes stat data and safe-link. + */ +int reiser4_delete_dir_common(struct inode *inode) +{ + int result; + dir_plugin *dplug; + + assert("", (get_current_context() && + get_current_context()->trans->atom == NULL)); + + dplug = inode_dir_plugin(inode); + assert("vs-1101", dplug && dplug->done); + + /* kill cursors which might be attached to inode */ + reiser4_kill_cursors(inode); + + /* grab space enough for removing two items */ + if (reiser4_grab_space + (2 * estimate_one_item_removal(reiser4_tree_by_inode(inode)), + BA_RESERVED | BA_CAN_COMMIT)) + return RETERR(-ENOSPC); + + result = dplug->done(inode); + if (!result) + result = common_object_delete_no_reserve(inode); + return result; +} + +/* this is common implementation of add_link method of file plugin + */ +int reiser4_add_link_common(struct inode *object, struct inode *parent) +{ + /* + * increment ->i_nlink and update ->i_ctime + */ + + INODE_INC_FIELD(object, i_nlink); + object->i_ctime = CURRENT_TIME; + return 0; +} + +/* this is common implementation of rem_link method of file plugin + */ +int reiser4_rem_link_common(struct inode *object, struct inode *parent) +{ + assert("nikita-2021", object != NULL); + assert("nikita-2163", object->i_nlink > 0); + + /* + * decrement ->i_nlink and update ->i_ctime + */ + + INODE_DEC_FIELD(object, i_nlink); + object->i_ctime = CURRENT_TIME; + return 0; +} + +/* this is common implementation of rem_link method of file plugin for typical + directory +*/ +int rem_link_common_dir(struct inode *object, struct inode *parent UNUSED_ARG) +{ + assert("nikita-20211", object != NULL); + assert("nikita-21631", object->i_nlink > 0); + + /* + * decrement ->i_nlink and update ->i_ctime + */ + INODE_DEC_FIELD(object, i_nlink); + if (object->i_nlink == 1) + INODE_DEC_FIELD(object, i_nlink); + object->i_ctime = CURRENT_TIME; + return 0; +} + +/* this is common implementation of owns_item method of file plugin + compare objectids of keys in inode and coord */ +int owns_item_common(const struct inode *inode, /* object to check + * against */ + const coord_t *coord/* coord to check */) +{ + reiser4_key item_key; + reiser4_key file_key; + + assert("nikita-760", inode != NULL); + assert("nikita-761", coord != NULL); + + return coord_is_existing_item(coord) && + (get_key_objectid(build_sd_key(inode, &file_key)) == + get_key_objectid(item_key_by_coord(coord, &item_key))); +} + +/* this is common implementation of owns_item method of file plugin + for typical directory +*/ +int owns_item_common_dir(const struct inode *inode,/* object to check against */ + const coord_t *coord/* coord of item to check */) +{ + reiser4_key item_key; + + assert("nikita-1335", inode != NULL); + assert("nikita-1334", coord != NULL); + + if (plugin_of_group(item_plugin_by_coord(coord), DIR_ENTRY_ITEM_TYPE)) + return get_key_locality(item_key_by_coord(coord, &item_key)) == + get_inode_oid(inode); + else + return owns_item_common(inode, coord); +} + +/* this is common implementation of can_add_link method of file plugin + checks whether yet another hard links to this object can be added +*/ +int can_add_link_common(const struct inode *object/* object to check */) +{ + assert("nikita-732", object != NULL); + + /* inode->i_nlink is unsigned int, so just check for integer + overflow */ + return object->i_nlink + 1 != 0; +} + +/* this is common implementation of can_rem_link method of file plugin for + typical directory +*/ +int can_rem_link_common_dir(const struct inode *inode) +{ + /* is_dir_empty() returns 0 is dir is empty */ + return !is_dir_empty(inode); +} + +/* this is common implementation of detach method of file plugin for typical + directory +*/ +int reiser4_detach_common_dir(struct inode *child, struct inode *parent) +{ + dir_plugin *dplug; + + dplug = inode_dir_plugin(child); + assert("nikita-2883", dplug != NULL); + assert("nikita-2884", dplug->detach != NULL); + return dplug->detach(child, parent); +} + +/* this is common implementation of bind method of file plugin for typical + directory +*/ +int reiser4_bind_common_dir(struct inode *child, struct inode *parent) +{ + dir_plugin *dplug; + + dplug = inode_dir_plugin(child); + assert("nikita-2646", dplug != NULL); + return dplug->attach(child, parent); +} + +static int process_truncate(struct inode *, __u64 size); + +/* this is common implementation of safelink method of file plugin + */ +int safelink_common(struct inode *object, reiser4_safe_link_t link, __u64 value) +{ + int result; + + assert("vs-1705", get_current_context()->trans->atom == NULL); + if (link == SAFE_UNLINK) + /* nothing to do. iput() in the caller (process_safelink) will + * finish with file */ + result = 0; + else if (link == SAFE_TRUNCATE) + result = process_truncate(object, value); + else { + warning("nikita-3438", "Unrecognized safe-link type: %i", link); + result = RETERR(-EIO); + } + return result; +} + +/* this is common implementation of estimate.create method of file plugin + can be used when object creation involves insertion of one item (usually stat + data) into tree +*/ +reiser4_block_nr estimate_create_common(const struct inode *object) +{ + return estimate_one_insert_item(reiser4_tree_by_inode(object)); +} + +/* this is common implementation of estimate.create method of file plugin for + typical directory + can be used when directory creation involves insertion of two items (usually + stat data and item containing "." and "..") into tree +*/ +reiser4_block_nr estimate_create_common_dir(const struct inode *object) +{ + return 2 * estimate_one_insert_item(reiser4_tree_by_inode(object)); +} + +/* this is common implementation of estimate.update method of file plugin + can be used when stat data update does not do more than inserting a unit + into a stat data item which is probably true for most cases +*/ +reiser4_block_nr estimate_update_common(const struct inode *inode) +{ + return estimate_one_insert_into_item(reiser4_tree_by_inode(inode)); +} + +/* this is common implementation of estimate.unlink method of file plugin + */ +reiser4_block_nr +estimate_unlink_common(const struct inode *object UNUSED_ARG, + const struct inode *parent UNUSED_ARG) +{ + return 0; +} + +/* this is common implementation of estimate.unlink method of file plugin for + typical directory +*/ +reiser4_block_nr +estimate_unlink_common_dir(const struct inode *object, + const struct inode *parent) +{ + dir_plugin *dplug; + + dplug = inode_dir_plugin(object); + assert("nikita-2888", dplug != NULL); + assert("nikita-2887", dplug->estimate.unlink != NULL); + return dplug->estimate.unlink(object, parent); +} + +char *wire_write_common(struct inode *inode, char *start) +{ + return build_inode_onwire(inode, start); +} + +char *wire_read_common(char *addr, reiser4_object_on_wire * obj) +{ + if (!obj) + return locate_obj_key_id_onwire(addr); + return extract_obj_key_id_from_onwire(addr, &obj->u.std.key_id); +} + +struct dentry *wire_get_common(struct super_block *sb, + reiser4_object_on_wire * obj) +{ + struct inode *inode; + struct dentry *dentry; + reiser4_key key; + + extract_key_from_id(&obj->u.std.key_id, &key); + inode = reiser4_iget(sb, &key, 1); + if (!IS_ERR(inode)) { + reiser4_iget_complete(inode); + dentry = d_obtain_alias(inode); + if (!IS_ERR(dentry)) + dentry->d_op = &get_super_private(sb)->ops.dentry; + } else if (PTR_ERR(inode) == -ENOENT) + /* + * inode wasn't found at the key encoded in the file + * handle. Hence, file handle is stale. + */ + dentry = ERR_PTR(RETERR(-ESTALE)); + else + dentry = (void *)inode; + return dentry; +} + +int wire_size_common(struct inode *inode) +{ + return inode_onwire_size(inode); +} + +void wire_done_common(reiser4_object_on_wire * obj) +{ + /* nothing to do */ +} + +/* helper function to print errors */ +static void key_warning(const reiser4_key * key /* key to print */ , + const struct inode *inode, + int code/* error code to print */) +{ + assert("nikita-716", key != NULL); + + if (code != -ENOMEM) { + warning("nikita-717", "Error for inode %llu (%i)", + (unsigned long long)get_key_objectid(key), code); + reiser4_print_key("for key", key); + } +} + +/* NIKITA-FIXME-HANS: perhaps this function belongs in another file? */ +#if REISER4_DEBUG +static void +check_inode_seal(const struct inode *inode, + const coord_t *coord, const reiser4_key * key) +{ + reiser4_key unit_key; + + unit_key_by_coord(coord, &unit_key); + assert("nikita-2752", + WITH_DATA_RET(coord->node, 1, keyeq(key, &unit_key))); + assert("nikita-2753", get_inode_oid(inode) == get_key_objectid(key)); +} + +static void check_sd_coord(coord_t *coord, const reiser4_key * key) +{ + reiser4_key ukey; + + coord_clear_iplug(coord); + if (zload(coord->node)) + return; + + if (!coord_is_existing_unit(coord) || + !item_plugin_by_coord(coord) || + !keyeq(unit_key_by_coord(coord, &ukey), key) || + (znode_get_level(coord->node) != LEAF_LEVEL) || + !item_is_statdata(coord)) { + warning("nikita-1901", "Conspicuous seal"); + reiser4_print_key("key", key); + print_coord("coord", coord, 1); + impossible("nikita-2877", "no way"); + } + zrelse(coord->node); +} + +#else +#define check_inode_seal(inode, coord, key) noop +#define check_sd_coord(coord, key) noop +#endif + +/* insert new stat-data into tree. Called with inode state + locked. Return inode state locked. */ +static int insert_new_sd(struct inode *inode/* inode to create sd for */) +{ + int result; + reiser4_key key; + coord_t coord; + reiser4_item_data data; + char *area; + reiser4_inode *ref; + lock_handle lh; + oid_t oid; + + assert("nikita-723", inode != NULL); + assert("nikita-3406", reiser4_inode_get_flag(inode, REISER4_NO_SD)); + + ref = reiser4_inode_data(inode); + spin_lock_inode(inode); + + if (ref->plugin_mask != 0) + /* inode has non-standard plugins */ + inode_set_extension(inode, PLUGIN_STAT); + /* + * prepare specification of new item to be inserted + */ + + data.iplug = inode_sd_plugin(inode); + data.length = data.iplug->s.sd.save_len(inode); + spin_unlock_inode(inode); + + data.data = NULL; + data.user = 0; +/* could be optimized for case where there is only one node format in + * use in the filesystem, probably there are lots of such + * places we could optimize for only one node layout.... -Hans */ + if (data.length > reiser4_tree_by_inode(inode)->nplug->max_item_size()) { + /* This is silly check, but we don't know actual node where + insertion will go into. */ + return RETERR(-ENAMETOOLONG); + } + oid = oid_allocate(inode->i_sb); +/* NIKITA-FIXME-HANS: what is your opinion on whether this error check should be + * encapsulated into oid_allocate? */ + if (oid == ABSOLUTE_MAX_OID) + return RETERR(-EOVERFLOW); + + set_inode_oid(inode, oid); + + coord_init_zero(&coord); + init_lh(&lh); + + result = insert_by_key(reiser4_tree_by_inode(inode), + build_sd_key(inode, &key), &data, &coord, &lh, + /* stat data lives on a leaf level */ + LEAF_LEVEL, CBK_UNIQUE); + + /* we don't want to re-check that somebody didn't insert + stat-data while we were doing io, because if it did, + insert_by_key() returned error. */ + /* but what _is_ possible is that plugin for inode's stat-data, + list of non-standard plugins or their state would change + during io, so that stat-data wouldn't fit into sd. To avoid + this race we keep inode_state lock. This lock has to be + taken each time you access inode in a way that would cause + changes in sd size: changing plugins etc. + */ + + if (result == IBK_INSERT_OK) { + coord_clear_iplug(&coord); + result = zload(coord.node); + if (result == 0) { + /* have we really inserted stat data? */ + assert("nikita-725", item_is_statdata(&coord)); + + /* inode was just created. It is inserted into hash + table, but no directory entry was yet inserted into + parent. So, inode is inaccessible through + ->lookup(). All places that directly grab inode + from hash-table (like old knfsd), should check + IMMUTABLE flag that is set by common_create_child. + */ + assert("nikita-3240", data.iplug != NULL); + assert("nikita-3241", data.iplug->s.sd.save != NULL); + area = item_body_by_coord(&coord); + result = data.iplug->s.sd.save(inode, &area); + znode_make_dirty(coord.node); + if (result == 0) { + /* object has stat-data now */ + reiser4_inode_clr_flag(inode, REISER4_NO_SD); + reiser4_inode_set_flag(inode, + REISER4_SDLEN_KNOWN); + /* initialise stat-data seal */ + reiser4_seal_init(&ref->sd_seal, &coord, &key); + ref->sd_coord = coord; + check_inode_seal(inode, &coord, &key); + } else if (result != -ENOMEM) + /* + * convert any other error code to -EIO to + * avoid confusing user level with unexpected + * errors. + */ + result = RETERR(-EIO); + zrelse(coord.node); + } + } + done_lh(&lh); + + if (result != 0) + key_warning(&key, inode, result); + else + oid_count_allocated(); + + return result; +} + +/* find sd of inode in a tree, deal with errors */ +int lookup_sd(struct inode *inode /* inode to look sd for */ , + znode_lock_mode lock_mode /* lock mode */ , + coord_t *coord /* resulting coord */ , + lock_handle * lh /* resulting lock handle */ , + const reiser4_key * key /* resulting key */ , + int silent) +{ + int result; + __u32 flags; + + assert("nikita-1692", inode != NULL); + assert("nikita-1693", coord != NULL); + assert("nikita-1694", key != NULL); + + /* look for the object's stat data in a tree. + This returns in "node" pointer to a locked znode and in "pos" + position of an item found in node. Both are only valid if + coord_found is returned. */ + flags = (lock_mode == ZNODE_WRITE_LOCK) ? CBK_FOR_INSERT : 0; + flags |= CBK_UNIQUE; + /* + * traverse tree to find stat data. We cannot use vroot here, because + * it only covers _body_ of the file, and stat data don't belong + * there. + */ + result = coord_by_key(reiser4_tree_by_inode(inode), + key, + coord, + lh, + lock_mode, + FIND_EXACT, LEAF_LEVEL, LEAF_LEVEL, flags, NULL); + if (REISER4_DEBUG && result == 0) + check_sd_coord(coord, key); + + if (result != 0 && !silent) + key_warning(key, inode, result); + return result; +} + +static int +locate_inode_sd(struct inode *inode, + reiser4_key * key, coord_t *coord, lock_handle * lh) +{ + reiser4_inode *state; + seal_t seal; + int result; + + assert("nikita-3483", inode != NULL); + + state = reiser4_inode_data(inode); + spin_lock_inode(inode); + *coord = state->sd_coord; + coord_clear_iplug(coord); + seal = state->sd_seal; + spin_unlock_inode(inode); + + build_sd_key(inode, key); + if (reiser4_seal_is_set(&seal)) { + /* first, try to use seal */ + result = reiser4_seal_validate(&seal, + coord, + key, + lh, ZNODE_WRITE_LOCK, + ZNODE_LOCK_LOPRI); + if (result == 0) + check_sd_coord(coord, key); + } else + result = -E_REPEAT; + + if (result != 0) { + coord_init_zero(coord); + result = lookup_sd(inode, ZNODE_WRITE_LOCK, coord, lh, key, 0); + } + return result; +} + +#if REISER4_DEBUG +static int all_but_offset_key_eq(const reiser4_key * k1, const reiser4_key * k2) +{ + return (get_key_locality(k1) == get_key_locality(k2) && + get_key_type(k1) == get_key_type(k2) && + get_key_band(k1) == get_key_band(k2) && + get_key_ordering(k1) == get_key_ordering(k2) && + get_key_objectid(k1) == get_key_objectid(k2)); +} + +#include "../tree_walk.h" + +/* make some checks before and after stat-data resize operation */ +static int check_sd_resize(struct inode *inode, coord_t *coord, + int length, int progress/* 1 means after resize */) +{ + int ret = 0; + lock_handle left_lock; + coord_t left_coord; + reiser4_key left_key; + reiser4_key key; + + if (inode_file_plugin(inode) != + file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)) + return 0; + if (!length) + return 0; + if (coord->item_pos != 0) + return 0; + + init_lh(&left_lock); + ret = reiser4_get_left_neighbor(&left_lock, + coord->node, + ZNODE_WRITE_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (ret == -E_REPEAT || ret == -E_NO_NEIGHBOR || + ret == -ENOENT || ret == -EINVAL + || ret == -E_DEADLOCK) { + ret = 0; + goto exit; + } + ret = zload(left_lock.node); + if (ret) + goto exit; + coord_init_last_unit(&left_coord, left_lock.node); + item_key_by_coord(&left_coord, &left_key); + item_key_by_coord(coord, &key); + + if (all_but_offset_key_eq(&key, &left_key)) + /* corruption occured */ + ret = 1; + zrelse(left_lock.node); + exit: + done_lh(&left_lock); + return ret; +} +#endif + +/* update stat-data at @coord */ +static int +update_sd_at(struct inode *inode, coord_t *coord, reiser4_key * key, + lock_handle * lh) +{ + int result; + reiser4_item_data data; + char *area; + reiser4_inode *state; + znode *loaded; + + state = reiser4_inode_data(inode); + + coord_clear_iplug(coord); + result = zload(coord->node); + if (result != 0) + return result; + loaded = coord->node; + + spin_lock_inode(inode); + assert("nikita-728", inode_sd_plugin(inode) != NULL); + data.iplug = inode_sd_plugin(inode); + + /* if inode has non-standard plugins, add appropriate stat data + * extension */ + if (state->extmask & (1 << PLUGIN_STAT)) { + if (state->plugin_mask == 0) + inode_clr_extension(inode, PLUGIN_STAT); + } else if (state->plugin_mask != 0) + inode_set_extension(inode, PLUGIN_STAT); + + if (state->extmask & (1 << HEIR_STAT)) { + if (state->heir_mask == 0) + inode_clr_extension(inode, HEIR_STAT); + } else if (state->heir_mask != 0) + inode_set_extension(inode, HEIR_STAT); + + /* data.length is how much space to add to (or remove + from if negative) sd */ + if (!reiser4_inode_get_flag(inode, REISER4_SDLEN_KNOWN)) { + /* recalculate stat-data length */ + data.length = + data.iplug->s.sd.save_len(inode) - + item_length_by_coord(coord); + reiser4_inode_set_flag(inode, REISER4_SDLEN_KNOWN); + } else + data.length = 0; + spin_unlock_inode(inode); + + /* if on-disk stat data is of different length than required + for this inode, resize it */ + + if (data.length != 0) { + data.data = NULL; + data.user = 0; + + assert("edward-1441", + !check_sd_resize(inode, coord, + data.length, 0/* before resize */)); + + /* insertion code requires that insertion point (coord) was + * between units. */ + coord->between = AFTER_UNIT; + result = reiser4_resize_item(coord, &data, key, lh, + COPI_DONT_SHIFT_LEFT); + if (result != 0) { + key_warning(key, inode, result); + zrelse(loaded); + return result; + } + if (loaded != coord->node) { + /* reiser4_resize_item moved coord to another node. + Zload it */ + zrelse(loaded); + coord_clear_iplug(coord); + result = zload(coord->node); + if (result != 0) + return result; + loaded = coord->node; + } + assert("edward-1442", + !check_sd_resize(inode, coord, + data.length, 1/* after resize */)); + } + area = item_body_by_coord(coord); + spin_lock_inode(inode); + result = data.iplug->s.sd.save(inode, &area); + znode_make_dirty(coord->node); + + /* re-initialise stat-data seal */ + + /* + * coord.between was possibly skewed from AT_UNIT when stat-data size + * was changed and new extensions were pasted into item. + */ + coord->between = AT_UNIT; + reiser4_seal_init(&state->sd_seal, coord, key); + state->sd_coord = *coord; + spin_unlock_inode(inode); + check_inode_seal(inode, coord, key); + zrelse(loaded); + return result; +} + +/* Update existing stat-data in a tree. Called with inode state locked. Return + inode state locked. */ +static int update_sd(struct inode *inode/* inode to update sd for */) +{ + int result; + reiser4_key key; + coord_t coord; + lock_handle lh; + + assert("nikita-726", inode != NULL); + + /* no stat-data, nothing to update?! */ + assert("nikita-3482", !reiser4_inode_get_flag(inode, REISER4_NO_SD)); + + init_lh(&lh); + + result = locate_inode_sd(inode, &key, &coord, &lh); + if (result == 0) + result = update_sd_at(inode, &coord, &key, &lh); + done_lh(&lh); + + return result; +} + +/* helper for reiser4_delete_object_common and reiser4_delete_dir_common. + Remove object stat data. Space for that must be reserved by caller before +*/ +static int +common_object_delete_no_reserve(struct inode *inode/* object to remove */) +{ + int result; + + assert("nikita-1477", inode != NULL); + + if (!reiser4_inode_get_flag(inode, REISER4_NO_SD)) { + reiser4_key sd_key; + + vfs_dq_free_inode(inode); + vfs_dq_drop(inode); + + build_sd_key(inode, &sd_key); + result = + reiser4_cut_tree(reiser4_tree_by_inode(inode), + &sd_key, &sd_key, NULL, 0); + if (result == 0) { + reiser4_inode_set_flag(inode, REISER4_NO_SD); + result = oid_release(inode->i_sb, get_inode_oid(inode)); + if (result == 0) { + oid_count_released(); + + result = safe_link_del(reiser4_tree_by_inode(inode), + get_inode_oid(inode), + SAFE_UNLINK); + } + } + } else + result = 0; + return result; +} + +/* helper for safelink_common */ +static int process_truncate(struct inode *inode, __u64 size) +{ + int result; + struct iattr attr; + file_plugin *fplug; + reiser4_context *ctx; + struct dentry dentry; + + assert("vs-21", is_in_reiser4_context()); + ctx = reiser4_init_context(inode->i_sb); + assert("vs-22", !IS_ERR(ctx)); + + attr.ia_size = size; + attr.ia_valid = ATTR_SIZE | ATTR_CTIME; + fplug = inode_file_plugin(inode); + + mutex_lock(&inode->i_mutex); + assert("vs-1704", get_current_context()->trans->atom == NULL); + dentry.d_inode = inode; + result = inode->i_op->setattr(&dentry, &attr); + mutex_unlock(&inode->i_mutex); + + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + + return result; +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/hash.c linux-2.6.30/fs/reiser4/plugin/hash.c --- linux-2.6.30.orig/fs/reiser4/plugin/hash.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/hash.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,352 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Hash functions */ + +#include "../debug.h" +#include "plugin_header.h" +#include "plugin.h" +#include "../super.h" +#include "../inode.h" + +#include + +/* old rupasov (yura) hash */ +static __u64 hash_rupasov(const unsigned char *name /* name to hash */ , + int len/* @name's length */) +{ + int i; + int j; + int pow; + __u64 a; + __u64 c; + + assert("nikita-672", name != NULL); + assert("nikita-673", len >= 0); + + for (pow = 1, i = 1; i < len; ++i) + pow = pow * 10; + + if (len == 1) + a = name[0] - 48; + else + a = (name[0] - 48) * pow; + + for (i = 1; i < len; ++i) { + c = name[i] - 48; + for (pow = 1, j = i; j < len - 1; ++j) + pow = pow * 10; + a = a + c * pow; + } + for (; i < 40; ++i) { + c = '0' - 48; + for (pow = 1, j = i; j < len - 1; ++j) + pow = pow * 10; + a = a + c * pow; + } + + for (; i < 256; ++i) { + c = i; + for (pow = 1, j = i; j < len - 1; ++j) + pow = pow * 10; + a = a + c * pow; + } + + a = a << 7; + return a; +} + +/* r5 hash */ +static __u64 hash_r5(const unsigned char *name /* name to hash */ , + int len UNUSED_ARG/* @name's length */) +{ + __u64 a = 0; + + assert("nikita-674", name != NULL); + assert("nikita-675", len >= 0); + + while (*name) { + a += *name << 4; + a += *name >> 4; + a *= 11; + name++; + } + return a; +} + +/* Keyed 32-bit hash function using TEA in a Davis-Meyer function + H0 = Key + Hi = E Mi(Hi-1) + Hi-1 + + (see Applied Cryptography, 2nd edition, p448). + + Jeremy Fitzhardinge 1998 + + Jeremy has agreed to the contents of reiserfs/README. -Hans + + This code was blindly upgraded to __u64 by s/__u32/__u64/g. +*/ +static __u64 hash_tea(const unsigned char *name /* name to hash */ , + int len/* @name's length */) +{ + __u64 k[] = { 0x9464a485u, 0x542e1a94u, 0x3e846bffu, 0xb75bcfc3u }; + + __u64 h0 = k[0], h1 = k[1]; + __u64 a, b, c, d; + __u64 pad; + int i; + + assert("nikita-676", name != NULL); + assert("nikita-677", len >= 0); + +#define DELTA 0x9E3779B9u +#define FULLROUNDS 10 /* 32 is overkill, 16 is strong crypto */ +#define PARTROUNDS 6 /* 6 gets complete mixing */ + +/* a, b, c, d - data; h0, h1 - accumulated hash */ +#define TEACORE(rounds) \ + do { \ + __u64 sum = 0; \ + int n = rounds; \ + __u64 b0, b1; \ + \ + b0 = h0; \ + b1 = h1; \ + \ + do { \ + sum += DELTA; \ + b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b); \ + b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d); \ + } while (--n); \ + \ + h0 += b0; \ + h1 += b1; \ + } while (0) + + pad = (__u64) len | ((__u64) len << 8); + pad |= pad << 16; + + while (len >= 16) { + a = (__u64) name[0] | (__u64) name[1] << 8 | (__u64) name[2] << + 16 | (__u64) name[3] << 24; + b = (__u64) name[4] | (__u64) name[5] << 8 | (__u64) name[6] << + 16 | (__u64) name[7] << 24; + c = (__u64) name[8] | (__u64) name[9] << 8 | (__u64) name[10] << + 16 | (__u64) name[11] << 24; + d = (__u64) name[12] | (__u64) name[13] << 8 | (__u64) name[14] + << 16 | (__u64) name[15] << 24; + + TEACORE(PARTROUNDS); + + len -= 16; + name += 16; + } + + if (len >= 12) { + /* assert(len < 16); */ + if (len >= 16) + *(int *)0 = 0; + + a = (__u64) name[0] | (__u64) name[1] << 8 | (__u64) name[2] << + 16 | (__u64) name[3] << 24; + b = (__u64) name[4] | (__u64) name[5] << 8 | (__u64) name[6] << + 16 | (__u64) name[7] << 24; + c = (__u64) name[8] | (__u64) name[9] << 8 | (__u64) name[10] << + 16 | (__u64) name[11] << 24; + + d = pad; + for (i = 12; i < len; i++) { + d <<= 8; + d |= name[i]; + } + } else if (len >= 8) { + /* assert(len < 12); */ + if (len >= 12) + *(int *)0 = 0; + a = (__u64) name[0] | (__u64) name[1] << 8 | (__u64) name[2] << + 16 | (__u64) name[3] << 24; + b = (__u64) name[4] | (__u64) name[5] << 8 | (__u64) name[6] << + 16 | (__u64) name[7] << 24; + + c = d = pad; + for (i = 8; i < len; i++) { + c <<= 8; + c |= name[i]; + } + } else if (len >= 4) { + /* assert(len < 8); */ + if (len >= 8) + *(int *)0 = 0; + a = (__u64) name[0] | (__u64) name[1] << 8 | (__u64) name[2] << + 16 | (__u64) name[3] << 24; + + b = c = d = pad; + for (i = 4; i < len; i++) { + b <<= 8; + b |= name[i]; + } + } else { + /* assert(len < 4); */ + if (len >= 4) + *(int *)0 = 0; + a = b = c = d = pad; + for (i = 0; i < len; i++) { + a <<= 8; + a |= name[i]; + } + } + + TEACORE(FULLROUNDS); + +/* return 0;*/ + return h0 ^ h1; + +} + +/* classical 64 bit Fowler/Noll/Vo-1 (FNV-1) hash. + + See http://www.isthe.com/chongo/tech/comp/fnv/ for details. + + Excerpts: + + FNV hashes are designed to be fast while maintaining a low collision + rate. + + [This version also seems to preserve lexicographical order locally.] + + FNV hash algorithms and source code have been released into the public + domain. + +*/ +static __u64 hash_fnv1(const unsigned char *name /* name to hash */ , + int len UNUSED_ARG/* @name's length */) +{ + unsigned long long a = 0xcbf29ce484222325ull; + const unsigned long long fnv_64_prime = 0x100000001b3ull; + + assert("nikita-678", name != NULL); + assert("nikita-679", len >= 0); + + /* FNV-1 hash each octet in the buffer */ + for (; *name; ++name) { + /* multiply by the 32 bit FNV magic prime mod 2^64 */ + a *= fnv_64_prime; + /* xor the bottom with the current octet */ + a ^= (unsigned long long)(*name); + } + /* return our new hash value */ + return a; +} + +/* degenerate hash function used to simplify testing of non-unique key + handling */ +static __u64 hash_deg(const unsigned char *name UNUSED_ARG /* name to hash */ , + int len UNUSED_ARG/* @name's length */) +{ + return 0xc0c0c0c010101010ull; +} + +static int change_hash(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + int result; + + assert("nikita-3503", inode != NULL); + assert("nikita-3504", plugin != NULL); + + assert("nikita-3505", is_reiser4_inode(inode)); + assert("nikita-3507", plugin->h.type_id == REISER4_HASH_PLUGIN_TYPE); + + if (!plugin_of_group(inode_file_plugin(inode), REISER4_DIRECTORY_FILE)) + return RETERR(-EINVAL); + + result = 0; + if (inode_hash_plugin(inode) == NULL || + inode_hash_plugin(inode)->h.id != plugin->h.id) { + if (is_dir_empty(inode) == 0) + result = aset_set_unsafe(&reiser4_inode_data(inode)->pset, + PSET_HASH, plugin); + else + result = RETERR(-ENOTEMPTY); + + } + return result; +} + +static reiser4_plugin_ops hash_plugin_ops = { + .init = NULL, + .load = NULL, + .save_len = NULL, + .save = NULL, + .change = change_hash +}; + +/* hash plugins */ +hash_plugin hash_plugins[LAST_HASH_ID] = { + [RUPASOV_HASH_ID] = { + .h = { + .type_id = REISER4_HASH_PLUGIN_TYPE, + .id = RUPASOV_HASH_ID, + .pops = &hash_plugin_ops, + .label = "rupasov", + .desc = "Original Yura's hash", + .linkage = {NULL, NULL} + }, + .hash = hash_rupasov + }, + [R5_HASH_ID] = { + .h = { + .type_id = REISER4_HASH_PLUGIN_TYPE, + .id = R5_HASH_ID, + .pops = &hash_plugin_ops, + .label = "r5", + .desc = "r5 hash", + .linkage = {NULL, NULL} + }, + .hash = hash_r5 + }, + [TEA_HASH_ID] = { + .h = { + .type_id = REISER4_HASH_PLUGIN_TYPE, + .id = TEA_HASH_ID, + .pops = &hash_plugin_ops, + .label = "tea", + .desc = "tea hash", + .linkage = {NULL, NULL} + }, + .hash = hash_tea + }, + [FNV1_HASH_ID] = { + .h = { + .type_id = REISER4_HASH_PLUGIN_TYPE, + .id = FNV1_HASH_ID, + .pops = &hash_plugin_ops, + .label = "fnv1", + .desc = "fnv1 hash", + .linkage = {NULL, NULL} + }, + .hash = hash_fnv1 + }, + [DEGENERATE_HASH_ID] = { + .h = { + .type_id = REISER4_HASH_PLUGIN_TYPE, + .id = DEGENERATE_HASH_ID, + .pops = &hash_plugin_ops, + .label = "degenerate hash", + .desc = "Degenerate hash: only for testing", + .linkage = {NULL, NULL} + }, + .hash = hash_deg + } +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/inode_ops.c linux-2.6.30/fs/reiser4/plugin/inode_ops.c --- linux-2.6.30.orig/fs/reiser4/plugin/inode_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/inode_ops.c 2009-06-22 17:27:31.000000000 +0200 @@ -0,0 +1,906 @@ +/* + * Copyright 2005 by Hans Reiser, licensing governed by reiser4/README + */ + +/* + * this file contains typical implementations for most of methods of struct + * inode_operations + */ + +#include "../inode.h" +#include "../safe_link.h" + +#include +#include + +static int create_vfs_object(struct inode *parent, struct dentry *dentry, + reiser4_object_create_data *data); + +/** + * reiser4_create_common - create of inode operations + * @parent: inode of parent directory + * @dentry: dentry of new object to create + * @mode: the permissions to use + * @nameidata: + * + * This is common implementation of vfs's create method of struct + * inode_operations. + * Creates regular file using file plugin from parent directory plugin set. + */ +int reiser4_create_common(struct inode *parent, struct dentry *dentry, + int mode, struct nameidata *nameidata) +{ + reiser4_object_create_data data; + file_plugin *fplug; + + memset(&data, 0, sizeof data); + data.mode = S_IFREG | mode; + fplug = child_create_plugin(parent) ? : inode_create_plugin(parent); + if (!plugin_of_group(fplug, REISER4_REGULAR_FILE)) { + warning("vpf-1900", "'%s' is not a regular file plugin.", + fplug->h.label); + return RETERR(-EIO); + } + data.id = fplug->h.id; + return create_vfs_object(parent, dentry, &data); +} + +int reiser4_lookup_name(struct inode *dir, struct dentry *, reiser4_key *); +void check_light_weight(struct inode *inode, struct inode *parent); + +/** + * reiser4_lookup_common - lookup of inode operations + * @parent: inode of directory to lookup into + * @dentry: name to look for + * @nameidata: + * + * This is common implementation of vfs's lookup method of struct + * inode_operations. + */ +struct dentry *reiser4_lookup_common(struct inode *parent, + struct dentry *dentry, + struct nameidata *nameidata) +{ + reiser4_context *ctx; + int result; + struct dentry *new; + struct inode *inode; + reiser4_dir_entry_desc entry; + + ctx = reiser4_init_context(parent->i_sb); + if (IS_ERR(ctx)) + return (struct dentry *)ctx; + + /* set up operations on dentry. */ + dentry->d_op = &get_super_private(parent->i_sb)->ops.dentry; + + result = reiser4_lookup_name(parent, dentry, &entry.key); + if (result) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + if (result == -ENOENT) { + /* object not found */ + if (!IS_DEADDIR(parent)) + d_add(dentry, NULL); + return NULL; + } + return ERR_PTR(result); + } + + inode = reiser4_iget(parent->i_sb, &entry.key, 0); + if (IS_ERR(inode)) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return ERR_PTR(PTR_ERR(inode)); + } + + /* success */ + check_light_weight(inode, parent); + new = d_splice_alias(inode, dentry); + reiser4_iget_complete(inode); + + /* prevent balance_dirty_pages() from being called: we don't want to + * do this under directory i_mutex. */ + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return new; +} + +static reiser4_block_nr common_estimate_link(struct inode *parent, + struct inode *object); +int reiser4_update_dir(struct inode *); + +/** + * reiser4_link_common - link of inode operations + * @existing: dentry of object which is to get new name + * @parent: directory where new name is to be created + * @newname: new name + * + * This is common implementation of vfs's link method of struct + * inode_operations. + */ +int reiser4_link_common(struct dentry *existing, struct inode *parent, + struct dentry *newname) +{ + reiser4_context *ctx; + int result; + struct inode *object; + dir_plugin *parent_dplug; + reiser4_dir_entry_desc entry; + reiser4_object_create_data data; + reiser4_block_nr reserve; + + ctx = reiser4_init_context(parent->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + assert("nikita-1431", existing != NULL); + assert("nikita-1432", parent != NULL); + assert("nikita-1433", newname != NULL); + + object = existing->d_inode; + assert("nikita-1434", object != NULL); + + /* check for race with create_object() */ + if (reiser4_inode_get_flag(object, REISER4_IMMUTABLE)) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return RETERR(-E_REPEAT); + } + + parent_dplug = inode_dir_plugin(parent); + + memset(&entry, 0, sizeof entry); + entry.obj = object; + + data.mode = object->i_mode; + data.id = inode_file_plugin(object)->h.id; + + reserve = common_estimate_link(parent, existing->d_inode); + if ((__s64) reserve < 0) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return reserve; + } + + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return RETERR(-ENOSPC); + } + + /* + * Subtle race handling: sys_link() doesn't take i_mutex on @parent. It + * means that link(2) can race against unlink(2) or rename(2), and + * inode is dead (->i_nlink == 0) when reiser4_link() is entered. + * + * For such inode we have to undo special processing done in + * reiser4_unlink() viz. creation of safe-link. + */ + if (unlikely(object->i_nlink == 0)) { + result = safe_link_del(reiser4_tree_by_inode(object), + get_inode_oid(object), SAFE_UNLINK); + if (result != 0) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + } + + /* increment nlink of @existing and update its stat data */ + result = reiser4_add_nlink(object, parent, 1); + if (result == 0) { + /* add entry to the parent */ + result = + parent_dplug->add_entry(parent, newname, &data, &entry); + if (result != 0) { + /* failed to add entry to the parent, decrement nlink + of @existing */ + reiser4_del_nlink(object, parent, 1); + /* + * now, if that failed, we have a file with too big + * nlink---space leak, much better than directory + * entry pointing to nowhere + */ + } + } + if (result == 0) { + atomic_inc(&object->i_count); + /* + * Upon successful completion, link() shall mark for update + * the st_ctime field of the file. Also, the st_ctime and + * st_mtime fields of the directory that contains the new + * entry shall be marked for update. --SUS + */ + result = reiser4_update_dir(parent); + } + if (result == 0) + d_instantiate(newname, existing->d_inode); + + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} + +static int unlink_check_and_grab(struct inode *parent, struct dentry *victim); + +/** + * reiser4_unlink_common - unlink of inode operations + * @parent: inode of directory to remove name from + * @victim: name to be removed + * + * This is common implementation of vfs's unlink method of struct + * inode_operations. + */ +int reiser4_unlink_common(struct inode *parent, struct dentry *victim) +{ + reiser4_context *ctx; + int result; + struct inode *object; + file_plugin *fplug; + + ctx = reiser4_init_context(parent->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + object = victim->d_inode; + fplug = inode_file_plugin(object); + assert("nikita-2882", fplug->detach != NULL); + + result = unlink_check_and_grab(parent, victim); + if (result != 0) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + + result = fplug->detach(object, parent); + if (result == 0) { + dir_plugin *parent_dplug; + reiser4_dir_entry_desc entry; + + parent_dplug = inode_dir_plugin(parent); + memset(&entry, 0, sizeof entry); + + /* first, delete directory entry */ + result = parent_dplug->rem_entry(parent, victim, &entry); + if (result == 0) { + /* + * if name was removed successfully, we _have_ to + * return 0 from this function, because upper level + * caller (vfs_{rmdir,unlink}) expect this. + * + * now that directory entry is removed, update + * stat-data + */ + reiser4_del_nlink(object, parent, 1); + /* + * Upon successful completion, unlink() shall mark for + * update the st_ctime and st_mtime fields of the + * parent directory. Also, if the file's link count is + * not 0, the st_ctime field of the file shall be + * marked for update. --SUS + */ + reiser4_update_dir(parent); + /* add safe-link for this file */ + if (object->i_nlink == 0) + safe_link_add(object, SAFE_UNLINK); + } + } + + if (unlikely(result != 0)) { + if (result != -ENOMEM) + warning("nikita-3398", "Cannot unlink %llu (%i)", + (unsigned long long)get_inode_oid(object), + result); + /* if operation failed commit pending inode modifications to + * the stat-data */ + reiser4_update_sd(object); + reiser4_update_sd(parent); + } + + reiser4_release_reserved(object->i_sb); + + /* @object's i_ctime was updated by ->rem_link() method(). */ + + /* @victim can be already removed from the disk by this time. Inode is + then marked so that iput() wouldn't try to remove stat data. But + inode itself is still there. + */ + + /* + * we cannot release directory semaphore here, because name has + * already been deleted, but dentry (@victim) still exists. Prevent + * balance_dirty_pages() from being called on exiting this context: we + * don't want to do this under directory i_mutex. + */ + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} + +/** + * reiser4_symlink_common - symlink of inode operations + * @parent: inode of parent directory + * @dentry: dentry of object to be created + * @linkname: string symlink is to contain + * + * This is common implementation of vfs's symlink method of struct + * inode_operations. + * Creates object using file plugin SYMLINK_FILE_PLUGIN_ID. + */ +int reiser4_symlink_common(struct inode *parent, struct dentry *dentry, + const char *linkname) +{ + reiser4_object_create_data data; + + memset(&data, 0, sizeof data); + data.name = linkname; + data.id = SYMLINK_FILE_PLUGIN_ID; + data.mode = S_IFLNK | S_IRWXUGO; + return create_vfs_object(parent, dentry, &data); +} + +/** + * reiser4_mkdir_common - mkdir of inode operations + * @parent: inode of parent directory + * @dentry: dentry of object to be created + * @mode: the permissions to use + * + * This is common implementation of vfs's mkdir method of struct + * inode_operations. + * Creates object using file plugin DIRECTORY_FILE_PLUGIN_ID. + */ +int reiser4_mkdir_common(struct inode *parent, struct dentry *dentry, int mode) +{ + reiser4_object_create_data data; + + memset(&data, 0, sizeof data); + data.mode = S_IFDIR | mode; + data.id = DIRECTORY_FILE_PLUGIN_ID; + return create_vfs_object(parent, dentry, &data); +} + +/** + * reiser4_mknod_common - mknod of inode operations + * @parent: inode of parent directory + * @dentry: dentry of object to be created + * @mode: the permissions to use and file type + * @rdev: minor and major of new device file + * + * This is common implementation of vfs's mknod method of struct + * inode_operations. + * Creates object using file plugin SPECIAL_FILE_PLUGIN_ID. + */ +int reiser4_mknod_common(struct inode *parent, struct dentry *dentry, + int mode, dev_t rdev) +{ + reiser4_object_create_data data; + + memset(&data, 0, sizeof data); + data.mode = mode; + data.rdev = rdev; + data.id = SPECIAL_FILE_PLUGIN_ID; + return create_vfs_object(parent, dentry, &data); +} + +/* + * implementation of vfs's rename method of struct inode_operations for typical + * directory is in inode_ops_rename.c + */ + +/** + * reiser4_follow_link_common - follow_link of inode operations + * @dentry: dentry of symlink + * @data: + * + * This is common implementation of vfs's followlink method of struct + * inode_operations. + * Assumes that inode's i_private points to the content of symbolic link. + */ +void *reiser4_follow_link_common(struct dentry *dentry, struct nameidata *nd) +{ + assert("vs-851", S_ISLNK(dentry->d_inode->i_mode)); + + if (!dentry->d_inode->i_private + || !reiser4_inode_get_flag(dentry->d_inode, + REISER4_GENERIC_PTR_USED)) + return ERR_PTR(RETERR(-EINVAL)); + nd_set_link(nd, dentry->d_inode->i_private); + return NULL; +} + +/** + * reiser4_permission_common - permission of inode operations + * @inode: inode to check permissions for + * @mask: mode bits to check permissions for + * @nameidata: + * + * Uses generic function to check for rwx permissions. + */ +int reiser4_permission_common(struct inode *inode, int mask) +{ + return generic_permission(inode, mask, NULL); +} + +static int setattr_reserve(reiser4_tree *); + +/* this is common implementation of vfs's setattr method of struct + inode_operations +*/ +int reiser4_setattr_common(struct dentry *dentry, struct iattr *attr) +{ + reiser4_context *ctx; + struct inode *inode; + int result; + + inode = dentry->d_inode; + result = inode_change_ok(inode, attr); + if (result) + return result; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + assert("nikita-3119", !(attr->ia_valid & ATTR_SIZE)); + + /* + * grab disk space and call standard inode_setattr(). + */ + result = setattr_reserve(reiser4_tree_by_inode(inode)); + if (!result) { + if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) + || (attr->ia_valid & ATTR_GID + && attr->ia_gid != inode->i_gid)) { + result = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; + if (result) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + } + result = inode_setattr(inode, attr); + if (!result) + reiser4_update_sd(inode); + } + + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} + +/* this is common implementation of vfs's getattr method of struct + inode_operations +*/ +int reiser4_getattr_common(struct vfsmount *mnt UNUSED_ARG, + struct dentry *dentry, struct kstat *stat) +{ + struct inode *obj; + + assert("nikita-2298", dentry != NULL); + assert("nikita-2299", stat != NULL); + assert("nikita-2300", dentry->d_inode != NULL); + + obj = dentry->d_inode; + + stat->dev = obj->i_sb->s_dev; + stat->ino = oid_to_uino(get_inode_oid(obj)); + stat->mode = obj->i_mode; + /* don't confuse userland with huge nlink. This is not entirely + * correct, because nlink_t is not necessary 16 bit signed. */ + stat->nlink = min(obj->i_nlink, (typeof(obj->i_nlink)) 0x7fff); + stat->uid = obj->i_uid; + stat->gid = obj->i_gid; + stat->rdev = obj->i_rdev; + stat->atime = obj->i_atime; + stat->mtime = obj->i_mtime; + stat->ctime = obj->i_ctime; + stat->size = obj->i_size; + stat->blocks = + (inode_get_bytes(obj) + VFS_BLKSIZE - 1) >> VFS_BLKSIZE_BITS; + /* "preferred" blocksize for efficient file system I/O */ + stat->blksize = get_super_private(obj->i_sb)->optimal_io_size; + + return 0; +} + +/* Estimate the maximum amount of nodes which might be allocated or changed on + typical new object creation. Typical creation consists of calling create + method of file plugin, adding directory entry to parent and update parent + directory's stat data. +*/ +static reiser4_block_nr estimate_create_vfs_object(struct inode *parent, + /* parent object */ + struct inode *object + /* object */) +{ + assert("vpf-309", parent != NULL); + assert("vpf-307", object != NULL); + + return + /* object creation estimation */ + inode_file_plugin(object)->estimate.create(object) + + /* stat data of parent directory estimation */ + inode_file_plugin(parent)->estimate.update(parent) + + /* adding entry estimation */ + inode_dir_plugin(parent)->estimate.add_entry(parent) + + /* to undo in the case of failure */ + inode_dir_plugin(parent)->estimate.rem_entry(parent); +} + +/* Create child in directory. + + . get object's plugin + . get fresh inode + . initialize inode + . add object's stat-data + . initialize object's directory + . add entry to the parent + . instantiate dentry + +*/ +static int do_create_vfs_child(reiser4_object_create_data * data,/* parameters + of new + object */ + struct inode **retobj) +{ + int result; + + struct dentry *dentry; /* parent object */ + struct inode *parent; /* new name */ + + dir_plugin *par_dir; /* directory plugin on the parent */ + dir_plugin *obj_dir; /* directory plugin on the new object */ + file_plugin *obj_plug; /* object plugin on the new object */ + struct inode *object; /* new object */ + reiser4_block_nr reserve; + + reiser4_dir_entry_desc entry; /* new directory entry */ + + assert("nikita-1420", data != NULL); + parent = data->parent; + dentry = data->dentry; + + assert("nikita-1418", parent != NULL); + assert("nikita-1419", dentry != NULL); + + /* check, that name is acceptable for parent */ + par_dir = inode_dir_plugin(parent); + if (par_dir->is_name_acceptable && + !par_dir->is_name_acceptable(parent, + dentry->d_name.name, + (int)dentry->d_name.len)) + return RETERR(-ENAMETOOLONG); + + result = 0; + obj_plug = file_plugin_by_id((int)data->id); + if (obj_plug == NULL) { + warning("nikita-430", "Cannot find plugin %i", data->id); + return RETERR(-ENOENT); + } + object = new_inode(parent->i_sb); + if (object == NULL) + return RETERR(-ENOMEM); + /* we'll update i_nlink below */ + object->i_nlink = 0; + /* new_inode() initializes i_ino to "arbitrary" value. Reset it to 0, + * to simplify error handling: if some error occurs before i_ino is + * initialized with oid, i_ino should already be set to some + * distinguished value. */ + object->i_ino = 0; + + /* So that on error iput will be called. */ + *retobj = object; + + if (vfs_dq_alloc_inode(object)) { + vfs_dq_drop(object); + object->i_flags |= S_NOQUOTA; + return RETERR(-EDQUOT); + } + + memset(&entry, 0, sizeof entry); + entry.obj = object; + + set_plugin(&reiser4_inode_data(object)->pset, PSET_FILE, + file_plugin_to_plugin(obj_plug)); + result = obj_plug->set_plug_in_inode(object, parent, data); + if (result) { + warning("nikita-431", "Cannot install plugin %i on %llx", + data->id, (unsigned long long)get_inode_oid(object)); + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + return result; + } + + /* reget plugin after installation */ + obj_plug = inode_file_plugin(object); + + if (obj_plug->create_object == NULL) { + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + return RETERR(-EPERM); + } + + /* if any of hash, tail, sd or permission plugins for newly created + object are not set yet set them here inheriting them from parent + directory + */ + assert("nikita-2070", obj_plug->adjust_to_parent != NULL); + result = obj_plug->adjust_to_parent(object, + parent, + object->i_sb->s_root->d_inode); + if (result == 0) + result = finish_pset(object); + if (result != 0) { + warning("nikita-432", "Cannot inherit from %llx to %llx", + (unsigned long long)get_inode_oid(parent), + (unsigned long long)get_inode_oid(object)); + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + return result; + } + + /* setup inode and file-operations for this inode */ + setup_inode_ops(object, data); + + /* call file plugin's method to initialize plugin specific part of + * inode */ + if (obj_plug->init_inode_data) + obj_plug->init_inode_data(object, data, 1/*create */); + + /* obtain directory plugin (if any) for new object. */ + obj_dir = inode_dir_plugin(object); + if (obj_dir != NULL && obj_dir->init == NULL) { + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + return RETERR(-EPERM); + } + + reiser4_inode_data(object)->locality_id = get_inode_oid(parent); + + reserve = estimate_create_vfs_object(parent, object); + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) { + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + return RETERR(-ENOSPC); + } + + /* mark inode `immutable'. We disable changes to the file being + created until valid directory entry for it is inserted. Otherwise, + if file were expanded and insertion of directory entry fails, we + have to remove file, but we only alloted enough space in + transaction to remove _empty_ file. 3.x code used to remove stat + data in different transaction thus possibly leaking disk space on + crash. This all only matters if it's possible to access file + without name, for example, by inode number + */ + reiser4_inode_set_flag(object, REISER4_IMMUTABLE); + + /* create empty object, this includes allocation of new objectid. For + directories this implies creation of dot and dotdot */ + assert("nikita-2265", reiser4_inode_get_flag(object, REISER4_NO_SD)); + + /* mark inode as `loaded'. From this point onward + reiser4_delete_inode() will try to remove its stat-data. */ + reiser4_inode_set_flag(object, REISER4_LOADED); + + result = obj_plug->create_object(object, parent, data); + if (result != 0) { + reiser4_inode_clr_flag(object, REISER4_IMMUTABLE); + if (result != -ENAMETOOLONG && result != -ENOMEM) + warning("nikita-2219", + "Failed to create sd for %llu", + (unsigned long long)get_inode_oid(object)); + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + return result; + } + + if (obj_dir != NULL) + result = obj_dir->init(object, parent, data); + if (result == 0) { + assert("nikita-434", !reiser4_inode_get_flag(object, + REISER4_NO_SD)); + /* insert inode into VFS hash table */ + insert_inode_hash(object); + /* create entry */ + result = par_dir->add_entry(parent, dentry, data, &entry); + if (result == 0) { + result = reiser4_add_nlink(object, parent, 0); + /* If O_CREAT is set and the file did not previously + exist, upon successful completion, open() shall + mark for update the st_atime, st_ctime, and + st_mtime fields of the file and the st_ctime and + st_mtime fields of the parent directory. --SUS + */ + /* @object times are already updated by + reiser4_add_nlink() */ + if (result == 0) + reiser4_update_dir(parent); + if (result != 0) + /* cleanup failure to add nlink */ + par_dir->rem_entry(parent, dentry, &entry); + } + if (result != 0) + /* cleanup failure to add entry */ + obj_plug->detach(object, parent); + } else if (result != -ENOMEM) + warning("nikita-2219", "Failed to initialize dir for %llu: %i", + (unsigned long long)get_inode_oid(object), result); + + /* + * update stat-data, committing all pending modifications to the inode + * fields. + */ + reiser4_update_sd(object); + if (result != 0) { + vfs_dq_free_inode(object); + object->i_flags |= S_NOQUOTA; + /* if everything was ok (result == 0), parent stat-data is + * already updated above (update_parent_dir()) */ + reiser4_update_sd(parent); + /* failure to create entry, remove object */ + obj_plug->delete_object(object); + } + + /* file has name now, clear immutable flag */ + reiser4_inode_clr_flag(object, REISER4_IMMUTABLE); + + /* on error, iput() will call ->delete_inode(). We should keep track + of the existence of stat-data for this inode and avoid attempt to + remove it in reiser4_delete_inode(). This is accomplished through + REISER4_NO_SD bit in inode.u.reiser4_i.plugin.flags + */ + return result; +} + +/* this is helper for common implementations of reiser4_mkdir, reiser4_create, + reiser4_mknod and reiser4_symlink +*/ +static int +create_vfs_object(struct inode *parent, + struct dentry *dentry, reiser4_object_create_data * data) +{ + reiser4_context *ctx; + int result; + struct inode *child; + + ctx = reiser4_init_context(parent->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + context_set_commit_async(ctx); + + data->parent = parent; + data->dentry = dentry; + child = NULL; + result = do_create_vfs_child(data, &child); + if (unlikely(result != 0)) { + if (child != NULL) { + reiser4_make_bad_inode(child); + iput(child); + } + } else + d_instantiate(dentry, child); + + reiser4_exit_context(ctx); + return result; +} + +/** + * helper for link_common. Estimate disk space necessary to add a link + * from @parent to @object + */ +static reiser4_block_nr common_estimate_link(struct inode *parent /* parent + * directory + */, + struct inode *object /* object to + * which new + * link is + * being + * created */) +{ + reiser4_block_nr res = 0; + file_plugin *fplug; + dir_plugin *dplug; + + assert("vpf-317", object != NULL); + assert("vpf-318", parent != NULL); + + fplug = inode_file_plugin(object); + dplug = inode_dir_plugin(parent); + /* VS-FIXME-HANS: why do we do fplug->estimate.update(object) twice + * instead of multiplying by 2? */ + /* reiser4_add_nlink(object) */ + res += fplug->estimate.update(object); + /* add_entry(parent) */ + res += dplug->estimate.add_entry(parent); + /* reiser4_del_nlink(object) */ + res += fplug->estimate.update(object); + /* update_dir(parent) */ + res += inode_file_plugin(parent)->estimate.update(parent); + /* safe-link */ + res += estimate_one_item_removal(reiser4_tree_by_inode(object)); + + return res; +} + +/* Estimate disk space necessary to remove a link between @parent and + @object. +*/ +static reiser4_block_nr estimate_unlink(struct inode *parent /* parent + * directory */, + struct inode *object /* object to which + * new link is + * being created + */) +{ + reiser4_block_nr res = 0; + file_plugin *fplug; + dir_plugin *dplug; + + assert("vpf-317", object != NULL); + assert("vpf-318", parent != NULL); + + fplug = inode_file_plugin(object); + dplug = inode_dir_plugin(parent); + + /* rem_entry(parent) */ + res += dplug->estimate.rem_entry(parent); + /* reiser4_del_nlink(object) */ + res += fplug->estimate.update(object); + /* update_dir(parent) */ + res += inode_file_plugin(parent)->estimate.update(parent); + /* fplug->unlink */ + res += fplug->estimate.unlink(object, parent); + /* safe-link */ + res += estimate_one_insert_item(reiser4_tree_by_inode(object)); + + return res; +} + +/* helper for reiser4_unlink_common. Estimate and grab space for unlink. */ +static int unlink_check_and_grab(struct inode *parent, struct dentry *victim) +{ + file_plugin *fplug; + struct inode *child; + int result; + + result = 0; + child = victim->d_inode; + fplug = inode_file_plugin(child); + + /* check for race with create_object() */ + if (reiser4_inode_get_flag(child, REISER4_IMMUTABLE)) + return RETERR(-E_REPEAT); + /* object being deleted should have stat data */ + assert("vs-949", !reiser4_inode_get_flag(child, REISER4_NO_SD)); + + /* ask object plugin */ + if (fplug->can_rem_link != NULL && !fplug->can_rem_link(child)) + return RETERR(-ENOTEMPTY); + + result = (int)estimate_unlink(parent, child); + if (result < 0) + return result; + + return reiser4_grab_reserved(child->i_sb, result, BA_CAN_COMMIT); +} + +/* helper for reiser4_setattr_common */ +static int setattr_reserve(reiser4_tree * tree) +{ + assert("vs-1096", is_grab_enabled(get_current_context())); + return reiser4_grab_space(estimate_one_insert_into_item(tree), + BA_CAN_COMMIT); +} + +/* helper function. Standards require that for many file-system operations + on success ctime and mtime of parent directory is to be updated. */ +int reiser4_update_dir(struct inode *dir) +{ + assert("nikita-2525", dir != NULL); + + dir->i_ctime = dir->i_mtime = CURRENT_TIME; + return reiser4_update_sd(dir); +} diff -urN linux-2.6.30.orig/fs/reiser4/plugin/inode_ops_rename.c linux-2.6.30/fs/reiser4/plugin/inode_ops_rename.c --- linux-2.6.30.orig/fs/reiser4/plugin/inode_ops_rename.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/inode_ops_rename.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,925 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "../inode.h" +#include "../safe_link.h" + +static const char *possible_leak = "Possible disk space leak."; + +/* re-bind existing name at @from_coord in @from_dir to point to @to_inode. + + Helper function called from hashed_rename() */ +static int replace_name(struct inode *to_inode, /* inode where @from_coord is + * to be re-targeted at */ + struct inode *from_dir, /* directory where @from_coord + * lives */ + struct inode *from_inode, /* inode @from_coord + * originally point to */ + coord_t *from_coord, /* where directory entry is in + * the tree */ + lock_handle * from_lh/* lock handle on @from_coord */) +{ + item_plugin *from_item; + int result; + znode *node; + + coord_clear_iplug(from_coord); + node = from_coord->node; + result = zload(node); + if (result != 0) + return result; + from_item = item_plugin_by_coord(from_coord); + if (plugin_of_group(item_plugin_by_coord(from_coord), + DIR_ENTRY_ITEM_TYPE)) { + reiser4_key to_key; + + build_sd_key(to_inode, &to_key); + + /* everything is found and prepared to change directory entry + at @from_coord to point to @to_inode. + + @to_inode is just about to get new name, so bump its link + counter. + + */ + result = reiser4_add_nlink(to_inode, from_dir, 0); + if (result != 0) { + /* Don't issue warning: this may be plain -EMLINK */ + zrelse(node); + return result; + } + + result = + from_item->s.dir.update_key(from_coord, &to_key, from_lh); + if (result != 0) { + reiser4_del_nlink(to_inode, from_dir, 0); + zrelse(node); + return result; + } + + /* @from_inode just lost its name, he-he. + + If @from_inode was directory, it contained dotdot pointing + to @from_dir. @from_dir i_nlink will be decreased when + iput() will be called on @from_inode. + + If file-system is not ADG (hard-links are + supported on directories), iput(from_inode) will not remove + @from_inode, and thus above is incorrect, but hard-links on + directories are problematic in many other respects. + */ + result = reiser4_del_nlink(from_inode, from_dir, 0); + if (result != 0) { + warning("nikita-2330", + "Cannot remove link from source: %i. %s", + result, possible_leak); + } + /* Has to return success, because entry is already + * modified. */ + result = 0; + + /* NOTE-NIKITA consider calling plugin method in stead of + accessing inode fields directly. */ + from_dir->i_mtime = CURRENT_TIME; + } else { + warning("nikita-2326", "Unexpected item type"); + result = RETERR(-EIO); + } + zrelse(node); + return result; +} + +/* add new entry pointing to @inode into @dir at @coord, locked by @lh + + Helper function used by hashed_rename(). */ +static int add_name(struct inode *inode, /* inode where @coord is to be + * re-targeted at */ + struct inode *dir, /* directory where @coord lives */ + struct dentry *name, /* new name */ + coord_t *coord, /* where directory entry is in the tree + */ + lock_handle * lh, /* lock handle on @coord */ + int is_dir/* true, if @inode is directory */) +{ + int result; + reiser4_dir_entry_desc entry; + + assert("nikita-2333", lh->node == coord->node); + assert("nikita-2334", is_dir == S_ISDIR(inode->i_mode)); + + memset(&entry, 0, sizeof entry); + entry.obj = inode; + /* build key of directory entry description */ + inode_dir_plugin(dir)->build_entry_key(dir, &name->d_name, &entry.key); + + /* ext2 does this in different order: first inserts new entry, + then increases directory nlink. We don't want do this, + because reiser4_add_nlink() calls ->add_link() plugin + method that can fail for whatever reason, leaving as with + cleanup problems. + */ + /* @inode is getting new name */ + reiser4_add_nlink(inode, dir, 0); + /* create @new_name in @new_dir pointing to + @old_inode */ + result = WITH_COORD(coord, + inode_dir_item_plugin(dir)->s.dir.add_entry(dir, + coord, + lh, + name, + &entry)); + if (result != 0) { + int result2; + result2 = reiser4_del_nlink(inode, dir, 0); + if (result2 != 0) { + warning("nikita-2327", + "Cannot drop link on %lli %i. %s", + (unsigned long long)get_inode_oid(inode), + result2, possible_leak); + } + } else + INODE_INC_FIELD(dir, i_size); + return result; +} + +static reiser4_block_nr estimate_rename(struct inode *old_dir, /* directory + * where @old is + * located */ + struct dentry *old_name,/* old name */ + struct inode *new_dir, /* directory + * where @new is + * located */ + struct dentry *new_name /* new name */) +{ + reiser4_block_nr res1, res2; + dir_plugin * p_parent_old, *p_parent_new; + file_plugin * p_child_old, *p_child_new; + + assert("vpf-311", old_dir != NULL); + assert("vpf-312", new_dir != NULL); + assert("vpf-313", old_name != NULL); + assert("vpf-314", new_name != NULL); + + p_parent_old = inode_dir_plugin(old_dir); + p_parent_new = inode_dir_plugin(new_dir); + p_child_old = inode_file_plugin(old_name->d_inode); + if (new_name->d_inode) + p_child_new = inode_file_plugin(new_name->d_inode); + else + p_child_new = NULL; + + /* find_entry - can insert one leaf. */ + res1 = res2 = 1; + + /* replace_name */ + { + /* reiser4_add_nlink(p_child_old) and + * reiser4_del_nlink(p_child_old) */ + res1 += 2 * p_child_old->estimate.update(old_name->d_inode); + /* update key */ + res1 += 1; + /* reiser4_del_nlink(p_child_new) */ + if (p_child_new) + res1 += p_child_new->estimate.update(new_name->d_inode); + } + + /* else add_name */ + { + /* reiser4_add_nlink(p_parent_new) and + * reiser4_del_nlink(p_parent_new) */ + res2 += + 2 * inode_file_plugin(new_dir)->estimate.update(new_dir); + /* reiser4_add_nlink(p_parent_old) */ + res2 += p_child_old->estimate.update(old_name->d_inode); + /* add_entry(p_parent_new) */ + res2 += p_parent_new->estimate.add_entry(new_dir); + /* reiser4_del_nlink(p_parent_old) */ + res2 += p_child_old->estimate.update(old_name->d_inode); + } + + res1 = res1 < res2 ? res2 : res1; + + /* reiser4_write_sd(p_parent_new) */ + res1 += inode_file_plugin(new_dir)->estimate.update(new_dir); + + /* reiser4_write_sd(p_child_new) */ + if (p_child_new) + res1 += p_child_new->estimate.update(new_name->d_inode); + + /* hashed_rem_entry(p_parent_old) */ + res1 += p_parent_old->estimate.rem_entry(old_dir); + + /* reiser4_del_nlink(p_child_old) */ + res1 += p_child_old->estimate.update(old_name->d_inode); + + /* replace_name */ + { + /* reiser4_add_nlink(p_parent_dir_new) */ + res1 += inode_file_plugin(new_dir)->estimate.update(new_dir); + /* update_key */ + res1 += 1; + /* reiser4_del_nlink(p_parent_new) */ + res1 += inode_file_plugin(new_dir)->estimate.update(new_dir); + /* reiser4_del_nlink(p_parent_old) */ + res1 += inode_file_plugin(old_dir)->estimate.update(old_dir); + } + + /* reiser4_write_sd(p_parent_old) */ + res1 += inode_file_plugin(old_dir)->estimate.update(old_dir); + + /* reiser4_write_sd(p_child_old) */ + res1 += p_child_old->estimate.update(old_name->d_inode); + + return res1; +} + +static int hashed_rename_estimate_and_grab(struct inode *old_dir, /* directory + * where @old + * is located + */ + struct dentry *old_name,/* old name + */ + struct inode *new_dir, /* directory + * where @new + * is located + */ + struct dentry *new_name /* new name + */) +{ + reiser4_block_nr reserve; + + reserve = estimate_rename(old_dir, old_name, new_dir, new_name); + + if (reiser4_grab_space(reserve, BA_CAN_COMMIT)) + return RETERR(-ENOSPC); + + return 0; +} + +/* check whether @old_inode and @new_inode can be moved within file system + * tree. This singles out attempts to rename pseudo-files, for example. */ +static int can_rename(struct inode *old_dir, struct inode *old_inode, + struct inode *new_dir, struct inode *new_inode) +{ + file_plugin *fplug; + dir_plugin *dplug; + + assert("nikita-3370", old_inode != NULL); + + dplug = inode_dir_plugin(new_dir); + fplug = inode_file_plugin(old_inode); + + if (dplug == NULL) + return RETERR(-ENOTDIR); + else if (new_dir->i_op->create == NULL) + return RETERR(-EPERM); + else if (!fplug->can_add_link(old_inode)) + return RETERR(-EMLINK); + else if (new_inode != NULL) { + fplug = inode_file_plugin(new_inode); + if (fplug->can_rem_link != NULL && + !fplug->can_rem_link(new_inode)) + return RETERR(-EBUSY); + } + return 0; +} + +int reiser4_find_entry(struct inode *, struct dentry *, lock_handle * , + znode_lock_mode, reiser4_dir_entry_desc *); +int reiser4_update_dir(struct inode *); + +/* this is common implementation of vfs's rename method of struct + inode_operations + See comments in the body. + + It is arguable that this function can be made generic so, that it + will be applicable to any kind of directory plugin that deals with + directories composed out of directory entries. The only obstacle + here is that we don't have any data-type to represent directory + entry. This should be re-considered when more than one different + directory plugin will be implemented. +*/ +int reiser4_rename_common(struct inode *old_dir /* directory where @old + * is located */ , + struct dentry *old_name /* old name */ , + struct inode *new_dir /* directory where @new + * is located */ , + struct dentry *new_name/* new name */) +{ + /* From `The Open Group Base Specifications Issue 6' + + If either the old or new argument names a symbolic link, rename() + shall operate on the symbolic link itself, and shall not resolve + the last component of the argument. If the old argument and the new + argument resolve to the same existing file, rename() shall return + successfully and perform no other action. + + [this is done by VFS: vfs_rename()] + + If the old argument points to the pathname of a file that is not a + directory, the new argument shall not point to the pathname of a + directory. + + [checked by VFS: vfs_rename->may_delete()] + + If the link named by the new argument exists, it shall + be removed and old renamed to new. In this case, a link named new + shall remain visible to other processes throughout the renaming + operation and refer either to the file referred to by new or old + before the operation began. + + [we should assure this] + + Write access permission is required for + both the directory containing old and the directory containing new. + + [checked by VFS: vfs_rename->may_delete(), may_create()] + + If the old argument points to the pathname of a directory, the new + argument shall not point to the pathname of a file that is not a + directory. + + [checked by VFS: vfs_rename->may_delete()] + + If the directory named by the new argument exists, it + shall be removed and old renamed to new. In this case, a link named + new shall exist throughout the renaming operation and shall refer + either to the directory referred to by new or old before the + operation began. + + [we should assure this] + + If new names an existing directory, it shall be + required to be an empty directory. + + [we should check this] + + If the old argument points to a pathname of a symbolic link, the + symbolic link shall be renamed. If the new argument points to a + pathname of a symbolic link, the symbolic link shall be removed. + + The new pathname shall not contain a path prefix that names + old. Write access permission is required for the directory + containing old and the directory containing new. If the old + argument points to the pathname of a directory, write access + permission may be required for the directory named by old, and, if + it exists, the directory named by new. + + [checked by VFS: vfs_rename(), vfs_rename_dir()] + + If the link named by the new argument exists and the file's link + count becomes 0 when it is removed and no process has the file + open, the space occupied by the file shall be freed and the file + shall no longer be accessible. If one or more processes have the + file open when the last link is removed, the link shall be removed + before rename() returns, but the removal of the file contents shall + be postponed until all references to the file are closed. + + [iput() handles this, but we can do this manually, a la + reiser4_unlink()] + + Upon successful completion, rename() shall mark for update the + st_ctime and st_mtime fields of the parent directory of each file. + + [N/A] + + */ + reiser4_context *ctx; + int result; + int is_dir; /* is @old_name directory */ + + struct inode *old_inode; + struct inode *new_inode; + coord_t *new_coord; + + struct reiser4_dentry_fsdata *new_fsdata; + dir_plugin *dplug; + file_plugin *fplug; + + reiser4_dir_entry_desc *old_entry, *new_entry, *dotdot_entry; + lock_handle * new_lh, *dotdot_lh; + struct dentry *dotdot_name; + struct reiser4_dentry_fsdata *dataonstack; + + ctx = reiser4_init_context(old_dir->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + old_entry = kzalloc(3 * sizeof(*old_entry) + 2 * sizeof(*new_lh) + + sizeof(*dotdot_name) + sizeof(*dataonstack), + reiser4_ctx_gfp_mask_get()); + if (!old_entry) { + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return RETERR(-ENOMEM); + } + + new_entry = old_entry + 1; + dotdot_entry = old_entry + 2; + new_lh = (lock_handle *)(old_entry + 3); + dotdot_lh = new_lh + 1; + dotdot_name = (struct dentry *)(new_lh + 2); + dataonstack = (struct reiser4_dentry_fsdata *)(dotdot_name + 1); + + assert("nikita-2318", old_dir != NULL); + assert("nikita-2319", new_dir != NULL); + assert("nikita-2320", old_name != NULL); + assert("nikita-2321", new_name != NULL); + + old_inode = old_name->d_inode; + new_inode = new_name->d_inode; + + dplug = inode_dir_plugin(old_dir); + fplug = NULL; + + new_fsdata = reiser4_get_dentry_fsdata(new_name); + if (IS_ERR(new_fsdata)) { + kfree(old_entry); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return PTR_ERR(new_fsdata); + } + + new_coord = &new_fsdata->dec.entry_coord; + coord_clear_iplug(new_coord); + + is_dir = S_ISDIR(old_inode->i_mode); + + assert("nikita-3461", old_inode->i_nlink >= 1 + !!is_dir); + + /* if target is existing directory and it's not empty---return error. + + This check is done specifically, because is_dir_empty() requires + tree traversal and have to be done before locks are taken. + */ + if (is_dir && new_inode != NULL && is_dir_empty(new_inode) != 0) { + kfree(old_entry); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return RETERR(-ENOTEMPTY); + } + + result = can_rename(old_dir, old_inode, new_dir, new_inode); + if (result != 0) { + kfree(old_entry); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + + result = hashed_rename_estimate_and_grab(old_dir, old_name, + new_dir, new_name); + if (result != 0) { + kfree(old_entry); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + + init_lh(new_lh); + + /* find entry for @new_name */ + result = reiser4_find_entry(new_dir, new_name, new_lh, ZNODE_WRITE_LOCK, + new_entry); + + if (IS_CBKERR(result)) { + done_lh(new_lh); + kfree(old_entry); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; + } + + reiser4_seal_done(&new_fsdata->dec.entry_seal); + + /* add or replace name for @old_inode as @new_name */ + if (new_inode != NULL) { + /* target (@new_name) exists. */ + /* Not clear what to do with objects that are + both directories and files at the same time. */ + if (result == CBK_COORD_FOUND) { + result = replace_name(old_inode, + new_dir, + new_inode, new_coord, new_lh); + if (result == 0) + fplug = inode_file_plugin(new_inode); + } else if (result == CBK_COORD_NOTFOUND) { + /* VFS told us that @new_name is bound to existing + inode, but we failed to find directory entry. */ + warning("nikita-2324", "Target not found"); + result = RETERR(-ENOENT); + } + } else { + /* target (@new_name) doesn't exists. */ + if (result == CBK_COORD_NOTFOUND) + result = add_name(old_inode, + new_dir, + new_name, new_coord, new_lh, is_dir); + else if (result == CBK_COORD_FOUND) { + /* VFS told us that @new_name is "negative" dentry, + but we found directory entry. */ + warning("nikita-2331", "Target found unexpectedly"); + result = RETERR(-EIO); + } + } + + assert("nikita-3462", ergo(result == 0, + old_inode->i_nlink >= 2 + !!is_dir)); + + /* We are done with all modifications to the @new_dir, release lock on + node. */ + done_lh(new_lh); + + if (fplug != NULL) { + /* detach @new_inode from name-space */ + result = fplug->detach(new_inode, new_dir); + if (result != 0) + warning("nikita-2330", "Cannot detach %lli: %i. %s", + (unsigned long long)get_inode_oid(new_inode), + result, possible_leak); + } + + if (new_inode != NULL) + reiser4_update_sd(new_inode); + + if (result == 0) { + old_entry->obj = old_inode; + + dplug->build_entry_key(old_dir, + &old_name->d_name, &old_entry->key); + + /* At this stage new name was introduced for + @old_inode. @old_inode, @new_dir, and @new_inode i_nlink + counters were updated. + + We want to remove @old_name now. If @old_inode wasn't + directory this is simple. + */ + result = dplug->rem_entry(old_dir, old_name, old_entry); + if (result != 0 && result != -ENOMEM) { + warning("nikita-2335", + "Cannot remove old name: %i", result); + } else { + result = reiser4_del_nlink(old_inode, old_dir, 0); + if (result != 0 && result != -ENOMEM) { + warning("nikita-2337", + "Cannot drop link on old: %i", result); + } + } + + if (result == 0 && is_dir) { + /* @old_inode is directory. We also have to update + dotdot entry. */ + coord_t *dotdot_coord; + + memset(dataonstack, 0, sizeof dataonstack); + memset(dotdot_entry, 0, sizeof dotdot_entry); + dotdot_entry->obj = old_dir; + memset(dotdot_name, 0, sizeof dotdot_name); + dotdot_name->d_name.name = ".."; + dotdot_name->d_name.len = 2; + /* + * allocate ->d_fsdata on the stack to avoid using + * reiser4_get_dentry_fsdata(). Locking is not needed, + * because dentry is private to the current thread. + */ + dotdot_name->d_fsdata = dataonstack; + init_lh(dotdot_lh); + + dotdot_coord = &dataonstack->dec.entry_coord; + coord_clear_iplug(dotdot_coord); + + result = reiser4_find_entry(old_inode, dotdot_name, + dotdot_lh, ZNODE_WRITE_LOCK, + dotdot_entry); + if (result == 0) { + /* replace_name() decreases i_nlink on + * @old_dir */ + result = replace_name(new_dir, + old_inode, + old_dir, + dotdot_coord, dotdot_lh); + } else + result = RETERR(-EIO); + done_lh(dotdot_lh); + } + } + reiser4_update_dir(new_dir); + reiser4_update_dir(old_dir); + reiser4_update_sd(old_inode); + if (result == 0) { + file_plugin *fplug; + + if (new_inode != NULL) { + /* add safe-link for target file (in case we removed + * last reference to the poor fellow */ + fplug = inode_file_plugin(new_inode); + if (new_inode->i_nlink == 0) + result = safe_link_add(new_inode, SAFE_UNLINK); + } + } + kfree(old_entry); + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} + +#if 0 +int reiser4_rename_common(struct inode *old_dir /* directory where @old + * is located */ , + struct dentry *old_name /* old name */ , + struct inode *new_dir /* directory where @new + * is located */ , + struct dentry *new_name/* new name */) +{ + /* From `The Open Group Base Specifications Issue 6' + + If either the old or new argument names a symbolic link, rename() + shall operate on the symbolic link itself, and shall not resolve + the last component of the argument. If the old argument and the new + argument resolve to the same existing file, rename() shall return + successfully and perform no other action. + + [this is done by VFS: vfs_rename()] + + If the old argument points to the pathname of a file that is not a + directory, the new argument shall not point to the pathname of a + directory. + + [checked by VFS: vfs_rename->may_delete()] + + If the link named by the new argument exists, it shall + be removed and old renamed to new. In this case, a link named new + shall remain visible to other processes throughout the renaming + operation and refer either to the file referred to by new or old + before the operation began. + + [we should assure this] + + Write access permission is required for + both the directory containing old and the directory containing new. + + [checked by VFS: vfs_rename->may_delete(), may_create()] + + If the old argument points to the pathname of a directory, the new + argument shall not point to the pathname of a file that is not a + directory. + + [checked by VFS: vfs_rename->may_delete()] + + If the directory named by the new argument exists, it + shall be removed and old renamed to new. In this case, a link named + new shall exist throughout the renaming operation and shall refer + either to the directory referred to by new or old before the + operation began. + + [we should assure this] + + If new names an existing directory, it shall be + required to be an empty directory. + + [we should check this] + + If the old argument points to a pathname of a symbolic link, the + symbolic link shall be renamed. If the new argument points to a + pathname of a symbolic link, the symbolic link shall be removed. + + The new pathname shall not contain a path prefix that names + old. Write access permission is required for the directory + containing old and the directory containing new. If the old + argument points to the pathname of a directory, write access + permission may be required for the directory named by old, and, if + it exists, the directory named by new. + + [checked by VFS: vfs_rename(), vfs_rename_dir()] + + If the link named by the new argument exists and the file's link + count becomes 0 when it is removed and no process has the file + open, the space occupied by the file shall be freed and the file + shall no longer be accessible. If one or more processes have the + file open when the last link is removed, the link shall be removed + before rename() returns, but the removal of the file contents shall + be postponed until all references to the file are closed. + + [iput() handles this, but we can do this manually, a la + reiser4_unlink()] + + Upon successful completion, rename() shall mark for update the + st_ctime and st_mtime fields of the parent directory of each file. + + [N/A] + + */ + reiser4_context *ctx; + int result; + int is_dir; /* is @old_name directory */ + struct inode *old_inode; + struct inode *new_inode; + reiser4_dir_entry_desc old_entry; + reiser4_dir_entry_desc new_entry; + coord_t *new_coord; + struct reiser4_dentry_fsdata *new_fsdata; + lock_handle new_lh; + dir_plugin *dplug; + file_plugin *fplug; + + ctx = reiser4_init_context(old_dir->i_sb); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + assert("nikita-2318", old_dir != NULL); + assert("nikita-2319", new_dir != NULL); + assert("nikita-2320", old_name != NULL); + assert("nikita-2321", new_name != NULL); + + old_inode = old_name->d_inode; + new_inode = new_name->d_inode; + + dplug = inode_dir_plugin(old_dir); + fplug = NULL; + + new_fsdata = reiser4_get_dentry_fsdata(new_name); + if (IS_ERR(new_fsdata)) { + result = PTR_ERR(new_fsdata); + goto exit; + } + + new_coord = &new_fsdata->dec.entry_coord; + coord_clear_iplug(new_coord); + + is_dir = S_ISDIR(old_inode->i_mode); + + assert("nikita-3461", old_inode->i_nlink >= 1 + !!is_dir); + + /* if target is existing directory and it's not empty---return error. + + This check is done specifically, because is_dir_empty() requires + tree traversal and have to be done before locks are taken. + */ + if (is_dir && new_inode != NULL && is_dir_empty(new_inode) != 0) + return RETERR(-ENOTEMPTY); + + result = can_rename(old_dir, old_inode, new_dir, new_inode); + if (result != 0) + goto exit; + + result = hashed_rename_estimate_and_grab(old_dir, old_name, + new_dir, new_name); + if (result != 0) + goto exit; + + init_lh(&new_lh); + + /* find entry for @new_name */ + result = reiser4_find_entry(new_dir, new_name, &new_lh, + ZNODE_WRITE_LOCK, &new_entry); + + if (IS_CBKERR(result)) { + done_lh(&new_lh); + goto exit; + } + + reiser4_seal_done(&new_fsdata->dec.entry_seal); + + /* add or replace name for @old_inode as @new_name */ + if (new_inode != NULL) { + /* target (@new_name) exists. */ + /* Not clear what to do with objects that are + both directories and files at the same time. */ + if (result == CBK_COORD_FOUND) { + result = replace_name(old_inode, + new_dir, + new_inode, new_coord, &new_lh); + if (result == 0) + fplug = inode_file_plugin(new_inode); + } else if (result == CBK_COORD_NOTFOUND) { + /* VFS told us that @new_name is bound to existing + inode, but we failed to find directory entry. */ + warning("nikita-2324", "Target not found"); + result = RETERR(-ENOENT); + } + } else { + /* target (@new_name) doesn't exists. */ + if (result == CBK_COORD_NOTFOUND) + result = add_name(old_inode, + new_dir, + new_name, new_coord, &new_lh, is_dir); + else if (result == CBK_COORD_FOUND) { + /* VFS told us that @new_name is "negative" dentry, + but we found directory entry. */ + warning("nikita-2331", "Target found unexpectedly"); + result = RETERR(-EIO); + } + } + + assert("nikita-3462", ergo(result == 0, + old_inode->i_nlink >= 2 + !!is_dir)); + + /* We are done with all modifications to the @new_dir, release lock on + node. */ + done_lh(&new_lh); + + if (fplug != NULL) { + /* detach @new_inode from name-space */ + result = fplug->detach(new_inode, new_dir); + if (result != 0) + warning("nikita-2330", "Cannot detach %lli: %i. %s", + (unsigned long long)get_inode_oid(new_inode), + result, possible_leak); + } + + if (new_inode != NULL) + reiser4_update_sd(new_inode); + + if (result == 0) { + memset(&old_entry, 0, sizeof old_entry); + old_entry.obj = old_inode; + + dplug->build_entry_key(old_dir, + &old_name->d_name, &old_entry.key); + + /* At this stage new name was introduced for + @old_inode. @old_inode, @new_dir, and @new_inode i_nlink + counters were updated. + + We want to remove @old_name now. If @old_inode wasn't + directory this is simple. + */ + result = dplug->rem_entry(old_dir, old_name, &old_entry); + /*result = rem_entry_hashed(old_dir, old_name, &old_entry); */ + if (result != 0 && result != -ENOMEM) { + warning("nikita-2335", + "Cannot remove old name: %i", result); + } else { + result = reiser4_del_nlink(old_inode, old_dir, 0); + if (result != 0 && result != -ENOMEM) { + warning("nikita-2337", + "Cannot drop link on old: %i", result); + } + } + + if (result == 0 && is_dir) { + /* @old_inode is directory. We also have to update + dotdot entry. */ + coord_t *dotdot_coord; + lock_handle dotdot_lh; + struct dentry dotdot_name; + reiser4_dir_entry_desc dotdot_entry; + struct reiser4_dentry_fsdata dataonstack; + struct reiser4_dentry_fsdata *fsdata; + + memset(&dataonstack, 0, sizeof dataonstack); + memset(&dotdot_entry, 0, sizeof dotdot_entry); + dotdot_entry.obj = old_dir; + memset(&dotdot_name, 0, sizeof dotdot_name); + dotdot_name.d_name.name = ".."; + dotdot_name.d_name.len = 2; + /* + * allocate ->d_fsdata on the stack to avoid using + * reiser4_get_dentry_fsdata(). Locking is not needed, + * because dentry is private to the current thread. + */ + dotdot_name.d_fsdata = &dataonstack; + init_lh(&dotdot_lh); + + fsdata = &dataonstack; + dotdot_coord = &fsdata->dec.entry_coord; + coord_clear_iplug(dotdot_coord); + + result = reiser4_find_entry(old_inode, + &dotdot_name, + &dotdot_lh, + ZNODE_WRITE_LOCK, + &dotdot_entry); + if (result == 0) { + /* replace_name() decreases i_nlink on + * @old_dir */ + result = replace_name(new_dir, + old_inode, + old_dir, + dotdot_coord, &dotdot_lh); + } else + result = RETERR(-EIO); + done_lh(&dotdot_lh); + } + } + reiser4_update_dir(new_dir); + reiser4_update_dir(old_dir); + reiser4_update_sd(old_inode); + if (result == 0) { + file_plugin *fplug; + + if (new_inode != NULL) { + /* add safe-link for target file (in case we removed + * last reference to the poor fellow */ + fplug = inode_file_plugin(new_inode); + if (new_inode->i_nlink == 0) + result = safe_link_add(new_inode, SAFE_UNLINK); + } + } +exit: + context_set_commit_async(ctx); + reiser4_exit_context(ctx); + return result; +} +#endif diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/acl.h linux-2.6.30/fs/reiser4/plugin/item/acl.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/acl.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/acl.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,66 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Directory entry. */ + +#if !defined( __FS_REISER4_PLUGIN_DIRECTORY_ENTRY_H__ ) +#define __FS_REISER4_PLUGIN_DIRECTORY_ENTRY_H__ + +#include "../../forward.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../key.h" + +#include +#include /* for struct dentry */ + +typedef struct directory_entry_format { + /* key of object stat-data. It's not necessary to store whole + key here, because it's always key of stat-data, so minor + packing locality and offset can be omitted here. But this + relies on particular key allocation scheme for stat-data, so, + for extensibility sake, whole key can be stored here. + + We store key as array of bytes, because we don't want 8-byte + alignment of dir entries. + */ + obj_key_id id; + /* file name. Null terminated string. */ + d8 name[0]; +} directory_entry_format; + +void print_de(const char *prefix, coord_t * coord); +int extract_key_de(const coord_t * coord, reiser4_key * key); +int update_key_de(const coord_t * coord, const reiser4_key * key, + lock_handle * lh); +char *extract_name_de(const coord_t * coord, char *buf); +unsigned extract_file_type_de(const coord_t * coord); +int add_entry_de(struct inode *dir, coord_t * coord, + lock_handle * lh, const struct dentry *name, + reiser4_dir_entry_desc * entry); +int rem_entry_de(struct inode *dir, const struct qstr *name, coord_t * coord, + lock_handle * lh, reiser4_dir_entry_desc * entry); +int max_name_len_de(const struct inode *dir); + +int de_rem_and_shrink(struct inode *dir, coord_t * coord, int length); + +char *extract_dent_name(const coord_t * coord, + directory_entry_format * dent, char *buf); + +#if REISER4_LARGE_KEY +#define DE_NAME_BUF_LEN (24) +#else +#define DE_NAME_BUF_LEN (16) +#endif + +/* __FS_REISER4_PLUGIN_DIRECTORY_ENTRY_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/blackbox.c linux-2.6.30/fs/reiser4/plugin/item/blackbox.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/blackbox.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/blackbox.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,142 @@ +/* Copyright 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Black box item implementation */ + +#include "../../forward.h" +#include "../../debug.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../coord.h" +#include "../../tree.h" +#include "../../lock.h" + +#include "blackbox.h" +#include "item.h" +#include "../plugin.h" + +int +store_black_box(reiser4_tree * tree, + const reiser4_key * key, void *data, int length) +{ + int result; + reiser4_item_data idata; + coord_t coord; + lock_handle lh; + + memset(&idata, 0, sizeof idata); + + idata.data = data; + idata.user = 0; + idata.length = length; + idata.iplug = item_plugin_by_id(BLACK_BOX_ID); + + init_lh(&lh); + result = insert_by_key(tree, key, + &idata, &coord, &lh, LEAF_LEVEL, CBK_UNIQUE); + + assert("nikita-3413", + ergo(result == 0, + WITH_COORD(&coord, + item_length_by_coord(&coord) == length))); + + done_lh(&lh); + return result; +} + +int +load_black_box(reiser4_tree * tree, + reiser4_key * key, void *data, int length, int exact) +{ + int result; + coord_t coord; + lock_handle lh; + + init_lh(&lh); + result = coord_by_key(tree, key, + &coord, &lh, ZNODE_READ_LOCK, + exact ? FIND_EXACT : FIND_MAX_NOT_MORE_THAN, + LEAF_LEVEL, LEAF_LEVEL, CBK_UNIQUE, NULL); + + if (result == 0) { + int ilen; + + result = zload(coord.node); + if (result == 0) { + ilen = item_length_by_coord(&coord); + if (ilen <= length) { + memcpy(data, item_body_by_coord(&coord), ilen); + unit_key_by_coord(&coord, key); + } else if (exact) { + /* + * item is larger than buffer provided by the + * user. Only issue a warning if @exact is + * set. If @exact is false, we are iterating + * over all safe-links and here we are reaching + * the end of the iteration. + */ + warning("nikita-3415", + "Wrong black box length: %i > %i", + ilen, length); + result = RETERR(-EIO); + } + zrelse(coord.node); + } + } + + done_lh(&lh); + return result; + +} + +int +update_black_box(reiser4_tree * tree, + const reiser4_key * key, void *data, int length) +{ + int result; + coord_t coord; + lock_handle lh; + + init_lh(&lh); + result = coord_by_key(tree, key, + &coord, &lh, ZNODE_READ_LOCK, + FIND_EXACT, + LEAF_LEVEL, LEAF_LEVEL, CBK_UNIQUE, NULL); + if (result == 0) { + int ilen; + + result = zload(coord.node); + if (result == 0) { + ilen = item_length_by_coord(&coord); + if (length <= ilen) { + memcpy(item_body_by_coord(&coord), data, + length); + } else { + warning("nikita-3437", + "Wrong black box length: %i < %i", + ilen, length); + result = RETERR(-EIO); + } + zrelse(coord.node); + } + } + + done_lh(&lh); + return result; + +} + +int kill_black_box(reiser4_tree * tree, const reiser4_key * key) +{ + return reiser4_cut_tree(tree, key, key, NULL, 1); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/blackbox.h linux-2.6.30/fs/reiser4/plugin/item/blackbox.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/blackbox.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/blackbox.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,33 @@ +/* Copyright 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* "Black box" entry to fixed-width contain user supplied data */ + +#if !defined( __FS_REISER4_BLACK_BOX_H__ ) +#define __FS_REISER4_BLACK_BOX_H__ + +#include "../../forward.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../key.h" + +extern int store_black_box(reiser4_tree * tree, + const reiser4_key * key, void *data, int length); +extern int load_black_box(reiser4_tree * tree, + reiser4_key * key, void *data, int length, int exact); +extern int kill_black_box(reiser4_tree * tree, const reiser4_key * key); +extern int update_black_box(reiser4_tree * tree, + const reiser4_key * key, void *data, int length); + +/* __FS_REISER4_BLACK_BOX_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/cde.c linux-2.6.30/fs/reiser4/plugin/item/cde.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/cde.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/cde.c 2009-06-22 17:27:31.000000000 +0200 @@ -0,0 +1,1008 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Directory entry implementation */ + +/* DESCRIPTION: + + This is "compound" directory item plugin implementation. This directory + item type is compound (as opposed to the "simple directory item" in + fs/reiser4/plugin/item/sde.[ch]), because it consists of several directory + entries. + + The reason behind this decision is disk space efficiency: all directory + entries inside the same directory have identical fragment in their + keys. This, of course, depends on key assignment policy. In our default key + assignment policy, all directory entries have the same locality which is + equal to the object id of their directory. + + Composing directory item out of several directory entries for the same + directory allows us to store said key fragment only once. That is, this is + some ad hoc form of key compression (stem compression) that is implemented + here, because general key compression is not supposed to be implemented in + v4.0. + + Another decision that was made regarding all directory item plugins, is + that they will store entry keys unaligned. This is for that sake of disk + space efficiency again. + + In should be noted, that storing keys unaligned increases CPU consumption, + at least on some architectures. + + Internal on-disk structure of the compound directory item is the following: + + HEADER cde_item_format. Here number of entries is stored. + ENTRY_HEADER_0 cde_unit_header. Here part of entry key and + ENTRY_HEADER_1 offset of entry body are stored. + ENTRY_HEADER_2 (basically two last parts of key) + ... + ENTRY_HEADER_N + ENTRY_BODY_0 directory_entry_format. Here part of stat data key and + ENTRY_BODY_1 NUL-terminated name are stored. + ENTRY_BODY_2 (part of statadta key in the + sence that since all SDs have + zero offset, this offset is not + stored on disk). + ... + ENTRY_BODY_N + + When it comes to the balancing, each directory entry in compound directory + item is unit, that is, something that can be cut from one item and pasted + into another item of the same type. Handling of unit cut and paste is major + reason for the complexity of code below. + +*/ + +#include "../../forward.h" +#include "../../debug.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../key.h" +#include "../../coord.h" +#include "sde.h" +#include "cde.h" +#include "item.h" +#include "../node/node.h" +#include "../plugin.h" +#include "../../znode.h" +#include "../../carry.h" +#include "../../tree.h" +#include "../../inode.h" + +#include /* for struct inode */ +#include /* for struct dentry */ +#include + +#if 0 +#define CHECKME(coord) \ +({ \ + const char *message; \ + coord_t dup; \ + \ + coord_dup_nocheck(&dup, (coord)); \ + dup.unit_pos = 0; \ + assert("nikita-2871", cde_check(&dup, &message) == 0); \ +}) +#else +#define CHECKME(coord) noop +#endif + +/* return body of compound directory item at @coord */ +static inline cde_item_format *formatted_at(const coord_t * coord) +{ + assert("nikita-1282", coord != NULL); + return item_body_by_coord(coord); +} + +/* return entry header at @coord */ +static inline cde_unit_header *header_at(const coord_t * + coord /* coord of item */ , + int idx /* index of unit */ ) +{ + assert("nikita-1283", coord != NULL); + return &formatted_at(coord)->entry[idx]; +} + +/* return number of units in compound directory item at @coord */ +static int units(const coord_t * coord /* coord of item */ ) +{ + return le16_to_cpu(get_unaligned(&formatted_at(coord)->num_of_entries)); +} + +/* return offset of the body of @idx-th entry in @coord */ +static unsigned int offset_of(const coord_t * coord /* coord of item */ , + int idx /* index of unit */ ) +{ + if (idx < units(coord)) + return le16_to_cpu(get_unaligned(&header_at(coord, idx)->offset)); + else if (idx == units(coord)) + return item_length_by_coord(coord); + else + impossible("nikita-1308", "Wrong idx"); + return 0; +} + +/* set offset of the body of @idx-th entry in @coord */ +static void set_offset(const coord_t * coord /* coord of item */ , + int idx /* index of unit */ , + unsigned int offset /* new offset */ ) +{ + put_unaligned(cpu_to_le16((__u16) offset), &header_at(coord, idx)->offset); +} + +static void adj_offset(const coord_t * coord /* coord of item */ , + int idx /* index of unit */ , + int delta /* offset change */ ) +{ + d16 *doffset; + __u16 offset; + + doffset = &header_at(coord, idx)->offset; + offset = le16_to_cpu(get_unaligned(doffset)); + offset += delta; + put_unaligned(cpu_to_le16((__u16) offset), doffset); +} + +/* return pointer to @offset-th byte from the beginning of @coord */ +static char *address(const coord_t * coord /* coord of item */ , + int offset) +{ + return ((char *)item_body_by_coord(coord)) + offset; +} + +/* return pointer to the body of @idx-th entry in @coord */ +static directory_entry_format *entry_at(const coord_t * coord /* coord of + * item */ , + int idx /* index of unit */ ) +{ + return (directory_entry_format *) address(coord, + (int)offset_of(coord, idx)); +} + +/* return number of unit referenced by @coord */ +static int idx_of(const coord_t * coord /* coord of item */ ) +{ + assert("nikita-1285", coord != NULL); + return coord->unit_pos; +} + +/* find position where entry with @entry_key would be inserted into @coord */ +static int find(const coord_t * coord /* coord of item */ , + const reiser4_key * entry_key /* key to look for */ , + cmp_t * last /* result of last comparison */ ) +{ + int entries; + + int left; + int right; + + cde_unit_header *header; + + assert("nikita-1295", coord != NULL); + assert("nikita-1296", entry_key != NULL); + assert("nikita-1297", last != NULL); + + entries = units(coord); + left = 0; + right = entries - 1; + while (right - left >= REISER4_SEQ_SEARCH_BREAK) { + int median; + + median = (left + right) >> 1; + + header = header_at(coord, median); + *last = de_id_key_cmp(&header->hash, entry_key); + switch (*last) { + case LESS_THAN: + left = median; + break; + case GREATER_THAN: + right = median; + break; + case EQUAL_TO:{ + do { + median--; + header--; + } while (median >= 0 && + de_id_key_cmp(&header->hash, + entry_key) == EQUAL_TO); + return median + 1; + } + } + } + header = header_at(coord, left); + for (; left < entries; ++left, ++header) { + prefetch(header + 1); + *last = de_id_key_cmp(&header->hash, entry_key); + if (*last != LESS_THAN) + break; + } + if (left < entries) + return left; + else + return RETERR(-ENOENT); + +} + +/* expand @coord as to accommodate for insertion of @no new entries starting + from @pos, with total bodies size @size. */ +static int expand_item(const coord_t * coord /* coord of item */ , + int pos /* unit position */ , int no /* number of new + * units*/ , + int size /* total size of new units' data */ , + unsigned int data_size /* free space already reserved + * in the item for insertion */ ) +{ + int entries; + cde_unit_header *header; + char *dent; + int i; + + assert("nikita-1310", coord != NULL); + assert("nikita-1311", pos >= 0); + assert("nikita-1312", no > 0); + assert("nikita-1313", data_size >= no * sizeof(directory_entry_format)); + assert("nikita-1343", + item_length_by_coord(coord) >= + (int)(size + data_size + no * sizeof *header)); + + entries = units(coord); + + if (pos == entries) + dent = address(coord, size); + else + dent = (char *)entry_at(coord, pos); + /* place where new header will be in */ + header = header_at(coord, pos); + /* free space for new entry headers */ + memmove(header + no, header, + (unsigned)(address(coord, size) - (char *)header)); + /* if adding to the end initialise first new header */ + if (pos == entries) { + set_offset(coord, pos, (unsigned)size); + } + + /* adjust entry pointer and size */ + dent = dent + no * sizeof *header; + size += no * sizeof *header; + /* free space for new entries */ + memmove(dent + data_size, dent, + (unsigned)(address(coord, size) - dent)); + + /* increase counter */ + entries += no; + put_unaligned(cpu_to_le16((__u16) entries), &formatted_at(coord)->num_of_entries); + + /* [ 0 ... pos ] entries were shifted by no * ( sizeof *header ) + bytes. */ + for (i = 0; i <= pos; ++i) + adj_offset(coord, i, no * sizeof *header); + /* [ pos + no ... +\infty ) entries were shifted by ( no * + sizeof *header + data_size ) bytes */ + for (i = pos + no; i < entries; ++i) + adj_offset(coord, i, no * sizeof *header + data_size); + return 0; +} + +/* insert new @entry into item */ +static int expand(const coord_t * coord /* coord of item */ , + struct cde_entry * entry /* entry to insert */ , + int len /* length of @entry data */ , + int *pos /* position to insert */ , + reiser4_dir_entry_desc * dir_entry /* parameters for new + * entry */ ) +{ + cmp_t cmp_res; + int datasize; + + *pos = find(coord, &dir_entry->key, &cmp_res); + if (*pos < 0) + *pos = units(coord); + + datasize = sizeof(directory_entry_format); + if (is_longname(entry->name->name, entry->name->len)) + datasize += entry->name->len + 1; + + expand_item(coord, *pos, 1, item_length_by_coord(coord) - len, + datasize); + return 0; +} + +/* paste body of @entry into item */ +static int paste_entry(const coord_t * coord /* coord of item */ , + struct cde_entry * entry /* new entry */ , + int pos /* position to insert */ , + reiser4_dir_entry_desc * dir_entry /* parameters for + * new entry */ ) +{ + cde_unit_header *header; + directory_entry_format *dent; + const char *name; + int len; + + header = header_at(coord, pos); + dent = entry_at(coord, pos); + + build_de_id_by_key(&dir_entry->key, &header->hash); + build_inode_key_id(entry->obj, &dent->id); + /* AUDIT unsafe strcpy() operation! It should be replaced with + much less CPU hungry + memcpy( ( char * ) dent -> name, entry -> name -> name , entry -> name -> len ); + + Also a more major thing is that there should be a way to figure out + amount of space in dent -> name and be able to check that we are + not going to overwrite more than we supposed to */ + name = entry->name->name; + len = entry->name->len; + if (is_longname(name, len)) { + strcpy((unsigned char *)dent->name, name); + put_unaligned(0, &dent->name[len]); + } + return 0; +} + +/* estimate how much space is necessary in item to insert/paste set of entries + described in @data. */ +int estimate_cde(const coord_t * coord /* coord of item */ , + const reiser4_item_data * data /* parameters for new item */ ) +{ + struct cde_entry_data *e; + int result; + int i; + + e = (struct cde_entry_data *) data->data; + + assert("nikita-1288", e != NULL); + assert("nikita-1289", e->num_of_entries >= 0); + + if (coord == NULL) + /* insert */ + result = sizeof(cde_item_format); + else + /* paste */ + result = 0; + + result += e->num_of_entries * + (sizeof(cde_unit_header) + sizeof(directory_entry_format)); + for (i = 0; i < e->num_of_entries; ++i) { + const char *name; + int len; + + name = e->entry[i].name->name; + len = e->entry[i].name->len; + assert("nikita-2054", strlen(name) == len); + if (is_longname(name, len)) + result += len + 1; + } + ((reiser4_item_data *) data)->length = result; + return result; +} + +/* ->nr_units() method for this item plugin. */ +pos_in_node_t nr_units_cde(const coord_t * coord /* coord of item */ ) +{ + return units(coord); +} + +/* ->unit_key() method for this item plugin. */ +reiser4_key *unit_key_cde(const coord_t * coord /* coord of item */ , + reiser4_key * key /* resulting key */ ) +{ + assert("nikita-1452", coord != NULL); + assert("nikita-1345", idx_of(coord) < units(coord)); + assert("nikita-1346", key != NULL); + + item_key_by_coord(coord, key); + extract_key_from_de_id(extract_dir_id_from_key(key), + &header_at(coord, idx_of(coord))->hash, key); + return key; +} + +/* mergeable_cde(): implementation of ->mergeable() item method. + + Two directory items are mergeable iff they are from the same + directory. That simple. + +*/ +int mergeable_cde(const coord_t * p1 /* coord of first item */ , + const coord_t * p2 /* coord of second item */ ) +{ + reiser4_key k1; + reiser4_key k2; + + assert("nikita-1339", p1 != NULL); + assert("nikita-1340", p2 != NULL); + + return + (item_plugin_by_coord(p1) == item_plugin_by_coord(p2)) && + (extract_dir_id_from_key(item_key_by_coord(p1, &k1)) == + extract_dir_id_from_key(item_key_by_coord(p2, &k2))); + +} + +/* ->max_key_inside() method for this item plugin. */ +reiser4_key *max_key_inside_cde(const coord_t * coord /* coord of item */ , + reiser4_key * result /* resulting key */ ) +{ + assert("nikita-1342", coord != NULL); + + item_key_by_coord(coord, result); + set_key_ordering(result, get_key_ordering(reiser4_max_key())); + set_key_fulloid(result, get_key_fulloid(reiser4_max_key())); + set_key_offset(result, get_key_offset(reiser4_max_key())); + return result; +} + +/* @data contains data which are to be put into tree */ +int can_contain_key_cde(const coord_t * coord /* coord of item */ , + const reiser4_key * key /* key to check */ , + const reiser4_item_data * data /* parameters of new + * item/unit being + * created */ ) +{ + reiser4_key item_key; + + /* FIXME-VS: do not rely on anything but iplug field of @data. Only + data->iplug is initialized */ + assert("vs-457", data && data->iplug); +/* assert( "vs-553", data -> user == 0 );*/ + item_key_by_coord(coord, &item_key); + + return (item_plugin_by_coord(coord) == data->iplug) && + (extract_dir_id_from_key(&item_key) == + extract_dir_id_from_key(key)); +} + +#if REISER4_DEBUG +/* cde_check ->check() method for compressed directory items + + used for debugging, every item should have here the most complete + possible check of the consistency of the item that the inventor can + construct +*/ +int reiser4_check_cde(const coord_t * coord /* coord of item to check */, + const char **error /* where to store error message */) +{ + int i; + int result; + char *item_start; + char *item_end; + reiser4_key key; + + coord_t c; + + assert("nikita-1357", coord != NULL); + assert("nikita-1358", error != NULL); + + if (!ergo(coord->item_pos != 0, + is_dot_key(item_key_by_coord(coord, &key)))) { + *error = "CDE doesn't start with dot"; + return -1; + } + item_start = item_body_by_coord(coord); + item_end = item_start + item_length_by_coord(coord); + + coord_dup(&c, coord); + result = 0; + for (i = 0; i < units(coord); ++i) { + directory_entry_format *entry; + + if ((char *)(header_at(coord, i) + 1) > + item_end - units(coord) * sizeof *entry) { + *error = "CDE header is out of bounds"; + result = -1; + break; + } + entry = entry_at(coord, i); + if ((char *)entry < item_start + sizeof(cde_item_format)) { + *error = "CDE header is too low"; + result = -1; + break; + } + if ((char *)(entry + 1) > item_end) { + *error = "CDE header is too high"; + result = -1; + break; + } + } + + return result; +} +#endif + +/* ->init() method for this item plugin. */ +int init_cde(coord_t * coord /* coord of item */ , + coord_t * from UNUSED_ARG, reiser4_item_data * data /* structure used for insertion */ + UNUSED_ARG) +{ + put_unaligned(cpu_to_le16(0), &formatted_at(coord)->num_of_entries); + return 0; +} + +/* ->lookup() method for this item plugin. */ +lookup_result lookup_cde(const reiser4_key * key /* key to search for */ , + lookup_bias bias /* search bias */ , + coord_t * coord /* coord of item to lookup in */ ) +{ + cmp_t last_comp; + int pos; + + reiser4_key utmost_key; + + assert("nikita-1293", coord != NULL); + assert("nikita-1294", key != NULL); + + CHECKME(coord); + + if (keygt(item_key_by_coord(coord, &utmost_key), key)) { + coord->unit_pos = 0; + coord->between = BEFORE_UNIT; + return CBK_COORD_NOTFOUND; + } + pos = find(coord, key, &last_comp); + if (pos >= 0) { + coord->unit_pos = (int)pos; + switch (last_comp) { + case EQUAL_TO: + coord->between = AT_UNIT; + return CBK_COORD_FOUND; + case GREATER_THAN: + coord->between = BEFORE_UNIT; + return RETERR(-ENOENT); + case LESS_THAN: + default: + impossible("nikita-1298", "Broken find"); + return RETERR(-EIO); + } + } else { + coord->unit_pos = units(coord) - 1; + coord->between = AFTER_UNIT; + return (bias == + FIND_MAX_NOT_MORE_THAN) ? CBK_COORD_FOUND : + CBK_COORD_NOTFOUND; + } +} + +/* ->paste() method for this item plugin. */ +int paste_cde(coord_t * coord /* coord of item */ , + reiser4_item_data * data /* parameters of new unit being + * inserted */ , + carry_plugin_info * info UNUSED_ARG /* todo carry queue */ ) +{ + struct cde_entry_data *e; + int result; + int i; + + CHECKME(coord); + e = (struct cde_entry_data *) data->data; + + result = 0; + for (i = 0; i < e->num_of_entries; ++i) { + int pos; + int phantom_size; + + phantom_size = data->length; + if (units(coord) == 0) + phantom_size -= sizeof(cde_item_format); + + result = + expand(coord, e->entry + i, phantom_size, &pos, data->arg); + if (result != 0) + break; + result = paste_entry(coord, e->entry + i, pos, data->arg); + if (result != 0) + break; + } + CHECKME(coord); + return result; +} + +/* amount of space occupied by all entries starting from @idx both headers and + bodies. */ +static unsigned int part_size(const coord_t * coord /* coord of item */ , + int idx /* index of unit */ ) +{ + assert("nikita-1299", coord != NULL); + assert("nikita-1300", idx < (int)units(coord)); + + return sizeof(cde_item_format) + + (idx + 1) * sizeof(cde_unit_header) + offset_of(coord, + idx + 1) - + offset_of(coord, 0); +} + +/* how many but not more than @want units of @source can be merged with + item in @target node. If pend == append - we try to append last item + of @target by first units of @source. If pend == prepend - we try to + "prepend" first item in @target by last units of @source. @target + node has @free_space bytes of free space. Total size of those units + are returned via @size */ +int can_shift_cde(unsigned free_space /* free space in item */ , + coord_t * coord /* coord of source item */ , + znode * target /* target node */ , + shift_direction pend /* shift direction */ , + unsigned *size /* resulting number of shifted bytes */ , + unsigned want /* maximal number of bytes to shift */ ) +{ + int shift; + + CHECKME(coord); + if (want == 0) { + *size = 0; + return 0; + } + + /* pend == SHIFT_LEFT <==> shifting to the left */ + if (pend == SHIFT_LEFT) { + for (shift = min((int)want - 1, units(coord)); shift >= 0; + --shift) { + *size = part_size(coord, shift); + if (target != NULL) + *size -= sizeof(cde_item_format); + if (*size <= free_space) + break; + } + shift = shift + 1; + } else { + int total_size; + + assert("nikita-1301", pend == SHIFT_RIGHT); + + total_size = item_length_by_coord(coord); + for (shift = units(coord) - want - 1; shift < units(coord) - 1; + ++shift) { + *size = total_size - part_size(coord, shift); + if (target == NULL) + *size += sizeof(cde_item_format); + if (*size <= free_space) + break; + } + shift = units(coord) - shift - 1; + } + if (shift == 0) + *size = 0; + CHECKME(coord); + return shift; +} + +/* ->copy_units() method for this item plugin. */ +void copy_units_cde(coord_t * target /* coord of target item */ , + coord_t * source /* coord of source item */ , + unsigned from /* starting unit */ , + unsigned count /* how many units to copy */ , + shift_direction where_is_free_space /* shift direction */ , + unsigned free_space /* free space in item */ ) +{ + char *header_from; + char *header_to; + + char *entry_from; + char *entry_to; + + int pos_in_target; + int data_size; + int data_delta; + int i; + + assert("nikita-1303", target != NULL); + assert("nikita-1304", source != NULL); + assert("nikita-1305", (int)from < units(source)); + assert("nikita-1307", (int)(from + count) <= units(source)); + + if (where_is_free_space == SHIFT_LEFT) { + assert("nikita-1453", from == 0); + pos_in_target = units(target); + } else { + assert("nikita-1309", (int)(from + count) == units(source)); + pos_in_target = 0; + memmove(item_body_by_coord(target), + (char *)item_body_by_coord(target) + free_space, + item_length_by_coord(target) - free_space); + } + + CHECKME(target); + CHECKME(source); + + /* expand @target */ + data_size = + offset_of(source, (int)(from + count)) - offset_of(source, + (int)from); + + if (units(target) == 0) + free_space -= sizeof(cde_item_format); + + expand_item(target, pos_in_target, (int)count, + (int)(item_length_by_coord(target) - free_space), + (unsigned)data_size); + + /* copy first @count units of @source into @target */ + data_delta = + offset_of(target, pos_in_target) - offset_of(source, (int)from); + + /* copy entries */ + entry_from = (char *)entry_at(source, (int)from); + entry_to = (char *)entry_at(source, (int)(from + count)); + memmove(entry_at(target, pos_in_target), entry_from, + (unsigned)(entry_to - entry_from)); + + /* copy headers */ + header_from = (char *)header_at(source, (int)from); + header_to = (char *)header_at(source, (int)(from + count)); + memmove(header_at(target, pos_in_target), header_from, + (unsigned)(header_to - header_from)); + + /* update offsets */ + for (i = pos_in_target; i < (int)(pos_in_target + count); ++i) + adj_offset(target, i, data_delta); + CHECKME(target); + CHECKME(source); +} + +/* ->cut_units() method for this item plugin. */ +int cut_units_cde(coord_t * coord /* coord of item */ , + pos_in_node_t from /* start unit pos */ , + pos_in_node_t to /* stop unit pos */ , + struct carry_cut_data *cdata UNUSED_ARG, + reiser4_key * smallest_removed, reiser4_key * new_first) +{ + char *header_from; + char *header_to; + + char *entry_from; + char *entry_to; + + int size; + int entry_delta; + int header_delta; + int i; + + unsigned count; + + CHECKME(coord); + + count = to - from + 1; + + assert("nikita-1454", coord != NULL); + assert("nikita-1455", (int)(from + count) <= units(coord)); + + if (smallest_removed) + unit_key_by_coord(coord, smallest_removed); + + if (new_first) { + coord_t next; + + /* not everything is cut from item head */ + assert("vs-1527", from == 0); + assert("vs-1528", to < units(coord) - 1); + + coord_dup(&next, coord); + next.unit_pos++; + unit_key_by_coord(&next, new_first); + } + + size = item_length_by_coord(coord); + if (count == (unsigned)units(coord)) { + return size; + } + + header_from = (char *)header_at(coord, (int)from); + header_to = (char *)header_at(coord, (int)(from + count)); + + entry_from = (char *)entry_at(coord, (int)from); + entry_to = (char *)entry_at(coord, (int)(from + count)); + + /* move headers */ + memmove(header_from, header_to, + (unsigned)(address(coord, size) - header_to)); + + header_delta = header_to - header_from; + + entry_from -= header_delta; + entry_to -= header_delta; + size -= header_delta; + + /* copy entries */ + memmove(entry_from, entry_to, + (unsigned)(address(coord, size) - entry_to)); + + entry_delta = entry_to - entry_from; + size -= entry_delta; + + /* update offsets */ + + for (i = 0; i < (int)from; ++i) + adj_offset(coord, i, -header_delta); + + for (i = from; i < units(coord) - (int)count; ++i) + adj_offset(coord, i, -header_delta - entry_delta); + + put_unaligned(cpu_to_le16((__u16) units(coord) - count), + &formatted_at(coord)->num_of_entries); + + if (from == 0) { + /* entries from head was removed - move remaining to right */ + memmove((char *)item_body_by_coord(coord) + + header_delta + entry_delta, item_body_by_coord(coord), + (unsigned)size); + if (REISER4_DEBUG) + memset(item_body_by_coord(coord), 0, + (unsigned)header_delta + entry_delta); + } else { + /* freed space is already at the end of item */ + if (REISER4_DEBUG) + memset((char *)item_body_by_coord(coord) + size, 0, + (unsigned)header_delta + entry_delta); + } + + return header_delta + entry_delta; +} + +int kill_units_cde(coord_t * coord /* coord of item */ , + pos_in_node_t from /* start unit pos */ , + pos_in_node_t to /* stop unit pos */ , + struct carry_kill_data *kdata UNUSED_ARG, + reiser4_key * smallest_removed, reiser4_key * new_first) +{ + return cut_units_cde(coord, from, to, NULL, smallest_removed, new_first); +} + +/* ->s.dir.extract_key() method for this item plugin. */ +int extract_key_cde(const coord_t * coord /* coord of item */ , + reiser4_key * key /* resulting key */ ) +{ + directory_entry_format *dent; + + assert("nikita-1155", coord != NULL); + assert("nikita-1156", key != NULL); + + dent = entry_at(coord, idx_of(coord)); + return extract_key_from_id(&dent->id, key); +} + +int +update_key_cde(const coord_t * coord, const reiser4_key * key, + lock_handle * lh UNUSED_ARG) +{ + directory_entry_format *dent; + obj_key_id obj_id; + int result; + + assert("nikita-2344", coord != NULL); + assert("nikita-2345", key != NULL); + + dent = entry_at(coord, idx_of(coord)); + result = build_obj_key_id(key, &obj_id); + if (result == 0) { + dent->id = obj_id; + znode_make_dirty(coord->node); + } + return 0; +} + +/* ->s.dir.extract_name() method for this item plugin. */ +char *extract_name_cde(const coord_t * coord /* coord of item */ , char *buf) +{ + directory_entry_format *dent; + + assert("nikita-1157", coord != NULL); + + dent = entry_at(coord, idx_of(coord)); + return extract_dent_name(coord, dent, buf); +} + +static int cde_bytes(int pasting, const reiser4_item_data * data) +{ + int result; + + result = data->length; + if (!pasting) + result -= sizeof(cde_item_format); + return result; +} + +/* ->s.dir.add_entry() method for this item plugin */ +int add_entry_cde(struct inode *dir /* directory object */ , + coord_t * coord /* coord of item */ , + lock_handle * lh /* lock handle for insertion */ , + const struct dentry *name /* name to insert */ , + reiser4_dir_entry_desc * dir_entry /* parameters of new + * directory entry */ ) +{ + reiser4_item_data data; + struct cde_entry entry; + struct cde_entry_data edata; + int result; + + assert("nikita-1656", coord->node == lh->node); + assert("nikita-1657", znode_is_write_locked(coord->node)); + + edata.num_of_entries = 1; + edata.entry = &entry; + + entry.dir = dir; + entry.obj = dir_entry->obj; + entry.name = &name->d_name; + + data.data = (char *)&edata; + data.user = 0; /* &edata is not user space */ + data.iplug = item_plugin_by_id(COMPOUND_DIR_ID); + data.arg = dir_entry; + assert("nikita-1302", data.iplug != NULL); + + result = is_dot_key(&dir_entry->key); + data.length = estimate_cde(result ? coord : NULL, &data); + + /* NOTE-NIKITA quota plugin? */ + if (vfs_dq_alloc_space_nodirty(dir, cde_bytes(result, &data))) + return RETERR(-EDQUOT); + + if (result) + result = insert_by_coord(coord, &data, &dir_entry->key, lh, 0); + else + result = reiser4_resize_item(coord, &data, &dir_entry->key, + lh, 0); + return result; +} + +/* ->s.dir.rem_entry() */ +int rem_entry_cde(struct inode *dir /* directory of item */ , + const struct qstr *name, coord_t * coord /* coord of item */ , + lock_handle * lh UNUSED_ARG /* lock handle for + * removal */ , + reiser4_dir_entry_desc * entry UNUSED_ARG /* parameters of + * directory entry + * being removed */ ) +{ + coord_t shadow; + int result; + int length; + ON_DEBUG(char buf[DE_NAME_BUF_LEN]); + + assert("nikita-2870", strlen(name->name) == name->len); + assert("nikita-2869", + !strcmp(name->name, extract_name_cde(coord, buf))); + + length = sizeof(directory_entry_format) + sizeof(cde_unit_header); + if (is_longname(name->name, name->len)) + length += name->len + 1; + + if (inode_get_bytes(dir) < length) { + warning("nikita-2628", "Dir is broke: %llu: %llu", + (unsigned long long)get_inode_oid(dir), + inode_get_bytes(dir)); + + return RETERR(-EIO); + } + + /* cut_node() is supposed to take pointers to _different_ + coords, because it will modify them without respect to + possible aliasing. To work around this, create temporary copy + of @coord. + */ + coord_dup(&shadow, coord); + result = + kill_node_content(coord, &shadow, NULL, NULL, NULL, NULL, NULL, 0); + if (result == 0) { + /* NOTE-NIKITA quota plugin? */ + vfs_dq_free_space_nodirty(dir, length); + } + return result; +} + +/* ->s.dir.max_name_len() method for this item plugin */ +int max_name_len_cde(const struct inode *dir /* directory */ ) +{ + return + reiser4_tree_by_inode(dir)->nplug->max_item_size() - + sizeof(directory_entry_format) - sizeof(cde_item_format) - + sizeof(cde_unit_header) - 2; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/cde.h linux-2.6.30/fs/reiser4/plugin/item/cde.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/cde.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/cde.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,87 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Compound directory item. See cde.c for description. */ + +#if !defined( __FS_REISER4_PLUGIN_COMPRESSED_DE_H__ ) +#define __FS_REISER4_PLUGIN_COMPRESSED_DE_H__ + +#include "../../forward.h" +#include "../../kassign.h" +#include "../../dformat.h" + +#include /* for struct inode */ +#include /* for struct dentry, etc */ + +typedef struct cde_unit_header { + de_id hash; + d16 offset; +} cde_unit_header; + +typedef struct cde_item_format { + d16 num_of_entries; + cde_unit_header entry[0]; +} cde_item_format; + +struct cde_entry { + const struct inode *dir; + const struct inode *obj; + const struct qstr *name; +}; + +struct cde_entry_data { + int num_of_entries; + struct cde_entry *entry; +}; + +/* plugin->item.b.* */ +reiser4_key *max_key_inside_cde(const coord_t * coord, reiser4_key * result); +int can_contain_key_cde(const coord_t * coord, const reiser4_key * key, + const reiser4_item_data *); +int mergeable_cde(const coord_t * p1, const coord_t * p2); +pos_in_node_t nr_units_cde(const coord_t * coord); +reiser4_key *unit_key_cde(const coord_t * coord, reiser4_key * key); +int estimate_cde(const coord_t * coord, const reiser4_item_data * data); +void print_cde(const char *prefix, coord_t * coord); +int init_cde(coord_t * coord, coord_t * from, reiser4_item_data * data); +lookup_result lookup_cde(const reiser4_key * key, lookup_bias bias, + coord_t * coord); +int paste_cde(coord_t * coord, reiser4_item_data * data, + carry_plugin_info * info UNUSED_ARG); +int can_shift_cde(unsigned free_space, coord_t * coord, znode * target, + shift_direction pend, unsigned *size, unsigned want); +void copy_units_cde(coord_t * target, coord_t * source, unsigned from, + unsigned count, shift_direction where_is_free_space, + unsigned free_space); +int cut_units_cde(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_cut_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +int kill_units_cde(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +void print_cde(const char *prefix, coord_t * coord); +int reiser4_check_cde(const coord_t * coord, const char **error); + +/* plugin->u.item.s.dir.* */ +int extract_key_cde(const coord_t * coord, reiser4_key * key); +int update_key_cde(const coord_t * coord, const reiser4_key * key, + lock_handle * lh); +char *extract_name_cde(const coord_t * coord, char *buf); +int add_entry_cde(struct inode *dir, coord_t * coord, + lock_handle * lh, const struct dentry *name, + reiser4_dir_entry_desc * entry); +int rem_entry_cde(struct inode *dir, const struct qstr *name, coord_t * coord, + lock_handle * lh, reiser4_dir_entry_desc * entry); +int max_name_len_cde(const struct inode *dir); + +/* __FS_REISER4_PLUGIN_COMPRESSED_DE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/ctail.c linux-2.6.30/fs/reiser4/plugin/item/ctail.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/ctail.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/ctail.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1613 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* ctails (aka "clustered tails") are items for cryptcompress objects */ + +/* DESCRIPTION: + +Each cryptcompress object is stored on disk as a set of clusters sliced +into ctails. + +Internal on-disk structure: + + HEADER (1) Here stored disk cluster shift + BODY +*/ + +#include "../../forward.h" +#include "../../debug.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../key.h" +#include "../../coord.h" +#include "item.h" +#include "../node/node.h" +#include "../plugin.h" +#include "../object.h" +#include "../../znode.h" +#include "../../carry.h" +#include "../../tree.h" +#include "../../inode.h" +#include "../../super.h" +#include "../../context.h" +#include "../../page_cache.h" +#include "../cluster.h" +#include "../../flush.h" +#include "../../tree_walk.h" + +#include +#include +#include + +/* return body of ctail item at @coord */ +static ctail_item_format *ctail_formatted_at(const coord_t * coord) +{ + assert("edward-60", coord != NULL); + return item_body_by_coord(coord); +} + +static int cluster_shift_by_coord(const coord_t * coord) +{ + return get_unaligned(&ctail_formatted_at(coord)->cluster_shift); +} + +static inline void dclust_set_extension_shift(hint_t * hint) +{ + assert("edward-1270", + item_id_by_coord(&hint->ext_coord.coord) == CTAIL_ID); + hint->ext_coord.extension.ctail.shift = + cluster_shift_by_coord(&hint->ext_coord.coord); +} + +static loff_t off_by_coord(const coord_t * coord) +{ + reiser4_key key; + return get_key_offset(item_key_by_coord(coord, &key)); +} + +int coord_is_unprepped_ctail(const coord_t * coord) +{ + assert("edward-1233", coord != NULL); + assert("edward-1234", item_id_by_coord(coord) == CTAIL_ID); + assert("edward-1235", + ergo((int)cluster_shift_by_coord(coord) == (int)UCTAIL_SHIFT, + nr_units_ctail(coord) == (pos_in_node_t) UCTAIL_NR_UNITS)); + + return (int)cluster_shift_by_coord(coord) == (int)UCTAIL_SHIFT; +} + +static cloff_t clust_by_coord(const coord_t * coord, struct inode *inode) +{ + int shift; + + if (inode != NULL) { + shift = inode_cluster_shift(inode); + assert("edward-1236", + ergo(!coord_is_unprepped_ctail(coord), + shift == cluster_shift_by_coord(coord))); + } else { + assert("edward-1237", !coord_is_unprepped_ctail(coord)); + shift = cluster_shift_by_coord(coord); + } + return off_by_coord(coord) >> shift; +} + +static int disk_cluster_size(const coord_t * coord) +{ + assert("edward-1156", + item_plugin_by_coord(coord) == item_plugin_by_id(CTAIL_ID)); + /* calculation of disk cluster size + is meaninless if ctail is unprepped */ + assert("edward-1238", !coord_is_unprepped_ctail(coord)); + + return 1 << cluster_shift_by_coord(coord); +} + +/* true if the key is of first disk cluster item */ +static int is_disk_cluster_key(const reiser4_key * key, const coord_t * coord) +{ + assert("edward-1239", item_id_by_coord(coord) == CTAIL_ID); + + return coord_is_unprepped_ctail(coord) || + ((get_key_offset(key) & + ((loff_t) disk_cluster_size(coord) - 1)) == 0); +} + +static char *first_unit(coord_t * coord) +{ + /* FIXME: warning: pointer of type `void *' used in arithmetic */ + return (char *)item_body_by_coord(coord) + sizeof(ctail_item_format); +} + +/* plugin->u.item.b.max_key_inside : + tail_max_key_inside */ + +/* plugin->u.item.b.can_contain_key */ +int +can_contain_key_ctail(const coord_t * coord, const reiser4_key * key, + const reiser4_item_data * data) +{ + reiser4_key item_key; + + if (item_plugin_by_coord(coord) != data->iplug) + return 0; + + item_key_by_coord(coord, &item_key); + if (get_key_locality(key) != get_key_locality(&item_key) || + get_key_objectid(key) != get_key_objectid(&item_key)) + return 0; + if (get_key_offset(&item_key) + nr_units_ctail(coord) != + get_key_offset(key)) + return 0; + if (is_disk_cluster_key(key, coord)) + return 0; + return 1; +} + +/* plugin->u.item.b.mergeable */ +int mergeable_ctail(const coord_t * p1, const coord_t * p2) +{ + reiser4_key key1, key2; + + assert("edward-62", item_id_by_coord(p1) == CTAIL_ID); + assert("edward-61", plugin_of_group(item_plugin_by_coord(p1), + UNIX_FILE_METADATA_ITEM_TYPE)); + + if (item_id_by_coord(p2) != CTAIL_ID) { + /* second item is of another type */ + return 0; + } + + item_key_by_coord(p1, &key1); + item_key_by_coord(p2, &key2); + if (get_key_locality(&key1) != get_key_locality(&key2) || + get_key_objectid(&key1) != get_key_objectid(&key2) || + get_key_type(&key1) != get_key_type(&key2)) { + /* items of different objects */ + return 0; + } + if (get_key_offset(&key1) + nr_units_ctail(p1) != get_key_offset(&key2)) + /* not adjacent items */ + return 0; + if (is_disk_cluster_key(&key2, p2)) + return 0; + return 1; +} + +/* plugin->u.item.b.nr_units */ +pos_in_node_t nr_units_ctail(const coord_t * coord) +{ + return (item_length_by_coord(coord) - + sizeof(ctail_formatted_at(coord)->cluster_shift)); +} + +/* plugin->u.item.b.estimate: + estimate how much space is needed to insert/paste @data->length bytes + into ctail at @coord */ +int estimate_ctail(const coord_t * coord /* coord of item */ , + const reiser4_item_data * + data /* parameters for new item */ ) +{ + if (coord == NULL) + /* insert */ + return (sizeof(ctail_item_format) + data->length); + else + /* paste */ + return data->length; +} + +/* ->init() method for this item plugin. */ +int init_ctail(coord_t * to /* coord of item */ , + coord_t * from /* old_item */ , + reiser4_item_data * data /* structure used for insertion */ ) +{ + int cluster_shift; /* cpu value to convert */ + + if (data) { + assert("edward-463", data->length > sizeof(ctail_item_format)); + cluster_shift = *((int *)(data->arg)); + data->length -= sizeof(ctail_item_format); + } else { + assert("edward-464", from != NULL); + assert("edward-855", ctail_ok(from)); + cluster_shift = (int)(cluster_shift_by_coord(from)); + } + put_unaligned((d8)cluster_shift, &ctail_formatted_at(to)->cluster_shift); + assert("edward-856", ctail_ok(to)); + return 0; +} + +/* plugin->u.item.b.lookup: + NULL: We are looking for item keys only */ + +#if REISER4_DEBUG +int ctail_ok(const coord_t * coord) +{ + return coord_is_unprepped_ctail(coord) || + cluster_shift_ok(cluster_shift_by_coord(coord)); +} + +/* plugin->u.item.b.check */ +int check_ctail(const coord_t * coord, const char **error) +{ + if (!ctail_ok(coord)) { + if (error) + *error = "bad cluster shift in ctail"; + return 1; + } + return 0; +} +#endif + +/* plugin->u.item.b.paste */ +int +paste_ctail(coord_t * coord, reiser4_item_data * data, + carry_plugin_info * info UNUSED_ARG) +{ + unsigned old_nr_units; + + assert("edward-268", data->data != NULL); + /* copy only from kernel space */ + assert("edward-66", data->user == 0); + + old_nr_units = + item_length_by_coord(coord) - sizeof(ctail_item_format) - + data->length; + + /* ctail items never get pasted in the middle */ + + if (coord->unit_pos == 0 && coord->between == AT_UNIT) { + + /* paste at the beginning when create new item */ + assert("edward-450", + item_length_by_coord(coord) == + data->length + sizeof(ctail_item_format)); + assert("edward-451", old_nr_units == 0); + } else if (coord->unit_pos == old_nr_units - 1 + && coord->between == AFTER_UNIT) { + + /* paste at the end */ + coord->unit_pos++; + } else + impossible("edward-453", "bad paste position"); + + memcpy(first_unit(coord) + coord->unit_pos, data->data, data->length); + + assert("edward-857", ctail_ok(coord)); + + return 0; +} + +/* plugin->u.item.b.fast_paste */ + +/* plugin->u.item.b.can_shift + number of units is returned via return value, number of bytes via @size. For + ctail items they coincide */ +int +can_shift_ctail(unsigned free_space, coord_t * source, + znode * target, shift_direction direction UNUSED_ARG, + unsigned *size /* number of bytes */ , unsigned want) +{ + /* make sure that that we do not want to shift more than we have */ + assert("edward-68", want > 0 && want <= nr_units_ctail(source)); + + *size = min(want, free_space); + + if (!target) { + /* new item will be created */ + if (*size <= sizeof(ctail_item_format)) { + *size = 0; + return 0; + } + return *size - sizeof(ctail_item_format); + } + return *size; +} + +/* plugin->u.item.b.copy_units + cooperates with ->can_shift() */ +void +copy_units_ctail(coord_t * target, coord_t * source, + unsigned from, unsigned count /* units */ , + shift_direction where_is_free_space, + unsigned free_space /* bytes */ ) +{ + /* make sure that item @target is expanded already */ + assert("edward-69", (unsigned)item_length_by_coord(target) >= count); + assert("edward-70", free_space == count || free_space == count + 1); + + assert("edward-858", ctail_ok(source)); + + if (where_is_free_space == SHIFT_LEFT) { + /* append item @target with @count first bytes of @source: + this restriction came from ordinary tails */ + assert("edward-71", from == 0); + assert("edward-860", ctail_ok(target)); + + memcpy(first_unit(target) + nr_units_ctail(target) - count, + first_unit(source), count); + } else { + /* target item is moved to right already */ + reiser4_key key; + + assert("edward-72", nr_units_ctail(source) == from + count); + + if (free_space == count) { + init_ctail(target, source, NULL); + } else { + /* new item has been created */ + assert("edward-862", ctail_ok(target)); + } + memcpy(first_unit(target), first_unit(source) + from, count); + + assert("edward-863", ctail_ok(target)); + + /* new units are inserted before first unit in an item, + therefore, we have to update item key */ + item_key_by_coord(source, &key); + set_key_offset(&key, get_key_offset(&key) + from); + + node_plugin_by_node(target->node)->update_item_key(target, &key, + NULL /*info */); + } +} + +/* plugin->u.item.b.create_hook */ +int create_hook_ctail(const coord_t * coord, void *arg) +{ + assert("edward-864", znode_is_loaded(coord->node)); + + znode_set_convertible(coord->node); + return 0; +} + +/* plugin->u.item.b.kill_hook */ +int kill_hook_ctail(const coord_t * coord, pos_in_node_t from, + pos_in_node_t count, carry_kill_data * kdata) +{ + struct inode *inode; + + assert("edward-1157", item_id_by_coord(coord) == CTAIL_ID); + assert("edward-291", znode_is_write_locked(coord->node)); + + inode = kdata->inode; + if (inode) { + reiser4_key key; + struct cryptcompress_info * info; + cloff_t index; + + item_key_by_coord(coord, &key); + info = cryptcompress_inode_data(inode); + index = off_to_clust(get_key_offset(&key), inode); + + if (from == 0) { + info->trunc_index = index; + if (is_disk_cluster_key(&key, coord)) { + /* + * first item of disk cluster is to be killed + */ + truncate_complete_page_cluster( + inode, index, kdata->params.truncate); + inode_sub_bytes(inode, + inode_cluster_size(inode)); + } + } + } + return 0; +} + +/* for shift_hook_ctail(), + return true if the first disk cluster item has dirty child +*/ +static int ctail_convertible(const coord_t * coord) +{ + int result; + reiser4_key key; + jnode *child = NULL; + + assert("edward-477", coord != NULL); + assert("edward-478", item_id_by_coord(coord) == CTAIL_ID); + + if (coord_is_unprepped_ctail(coord)) + /* unprepped ctail should be converted */ + return 1; + + item_key_by_coord(coord, &key); + child = jlookup(current_tree, + get_key_objectid(&key), + off_to_pg(off_by_coord(coord))); + if (!child) + return 0; + result = JF_ISSET(child, JNODE_DIRTY); + jput(child); + return result; +} + +/* FIXME-EDWARD */ +/* plugin->u.item.b.shift_hook */ +int shift_hook_ctail(const coord_t * item /* coord of item */ , + unsigned from UNUSED_ARG /* start unit */ , + unsigned count UNUSED_ARG /* stop unit */ , + znode * old_node /* old parent */ ) +{ + assert("edward-479", item != NULL); + assert("edward-480", item->node != old_node); + + if (!znode_convertible(old_node) || znode_convertible(item->node)) + return 0; + if (ctail_convertible(item)) + znode_set_convertible(item->node); + return 0; +} + +static int +cut_or_kill_ctail_units(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + int cut, void *p, reiser4_key * smallest_removed, + reiser4_key * new_first) +{ + pos_in_node_t count; /* number of units to cut */ + char *item; + + count = to - from + 1; + item = item_body_by_coord(coord); + + assert("edward-74", ergo(from != 0, to == coord_last_unit_pos(coord))); + + if (smallest_removed) { + /* store smallest key removed */ + item_key_by_coord(coord, smallest_removed); + set_key_offset(smallest_removed, + get_key_offset(smallest_removed) + from); + } + + if (new_first) { + assert("vs-1531", from == 0); + + item_key_by_coord(coord, new_first); + set_key_offset(new_first, + get_key_offset(new_first) + from + count); + } + + if (!cut) + kill_hook_ctail(coord, from, 0, (struct carry_kill_data *)p); + + if (from == 0) { + if (count != nr_units_ctail(coord)) { + /* part of item is removed, so move free space at the beginning + of the item and update item key */ + reiser4_key key; + memcpy(item + to + 1, item, sizeof(ctail_item_format)); + item_key_by_coord(coord, &key); + set_key_offset(&key, get_key_offset(&key) + count); + node_plugin_by_node(coord->node)->update_item_key(coord, + &key, + NULL); + } else { + /* cut_units should not be called to cut evrything */ + assert("vs-1532", ergo(cut, 0)); + /* whole item is cut, so more then amount of space occupied + by units got freed */ + count += sizeof(ctail_item_format); + } + if (REISER4_DEBUG) + memset(item, 0, count); + } else if (REISER4_DEBUG) + memset(item + sizeof(ctail_item_format) + from, 0, count); + return count; +} + +/* plugin->u.item.b.cut_units */ +int +cut_units_ctail(coord_t * item, pos_in_node_t from, pos_in_node_t to, + carry_cut_data * cdata, reiser4_key * smallest_removed, + reiser4_key * new_first) +{ + return cut_or_kill_ctail_units(item, from, to, 1, NULL, + smallest_removed, new_first); +} + +/* plugin->u.item.b.kill_units */ +int +kill_units_ctail(coord_t * item, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *kdata, reiser4_key * smallest_removed, + reiser4_key * new_first) +{ + return cut_or_kill_ctail_units(item, from, to, 0, kdata, + smallest_removed, new_first); +} + +/* plugin->u.item.s.file.read */ +int read_ctail(struct file *file UNUSED_ARG, flow_t * f, hint_t * hint) +{ + uf_coord_t *uf_coord; + coord_t *coord; + + uf_coord = &hint->ext_coord; + coord = &uf_coord->coord; + assert("edward-127", f->user == 0); + assert("edward-129", coord && coord->node); + assert("edward-130", coord_is_existing_unit(coord)); + assert("edward-132", znode_is_loaded(coord->node)); + + /* start read only from the beginning of ctail */ + assert("edward-133", coord->unit_pos == 0); + /* read only whole ctails */ + assert("edward-135", nr_units_ctail(coord) <= f->length); + + assert("edward-136", reiser4_schedulable()); + assert("edward-886", ctail_ok(coord)); + + if (f->data) + memcpy(f->data, (char *)first_unit(coord), + (size_t) nr_units_ctail(coord)); + + dclust_set_extension_shift(hint); + mark_page_accessed(znode_page(coord->node)); + move_flow_forward(f, nr_units_ctail(coord)); + + return 0; +} + +/** + * Prepare transform stream with plain text for page + * @page taking into account synchronization issues. + */ +static int ctail_read_disk_cluster(struct cluster_handle * clust, + struct inode * inode, struct page * page, + znode_lock_mode mode) +{ + int result; + + assert("edward-1450", mode == ZNODE_READ_LOCK || ZNODE_WRITE_LOCK); + assert("edward-671", clust->hint != NULL); + assert("edward-140", clust->dstat == INVAL_DISK_CLUSTER); + assert("edward-672", cryptcompress_inode_ok(inode)); + assert("edward-1527", PageLocked(page)); + + unlock_page(page); + + /* set input stream */ + result = grab_tfm_stream(inode, &clust->tc, INPUT_STREAM); + if (result) { + lock_page(page); + return result; + } + result = find_disk_cluster(clust, inode, 1 /* read items */, mode); + lock_page(page); + if (result) + return result; + /* + * at this point we have locked position in the tree + */ + assert("edward-1528", znode_is_any_locked(clust->hint->lh.node)); + + if (page->mapping != inode->i_mapping) { + /* page was truncated */ + reiser4_unset_hint(clust->hint); + reset_cluster_params(clust); + return AOP_TRUNCATED_PAGE; + } + if (PageUptodate(page)) { + /* disk cluster can be obsolete, don't use it! */ + reiser4_unset_hint(clust->hint); + reset_cluster_params(clust); + return 0; + } + if (clust->dstat == FAKE_DISK_CLUSTER || + clust->dstat == UNPR_DISK_CLUSTER || + clust->dstat == TRNC_DISK_CLUSTER) { + /* + * this information about disk cluster will be valid + * as long as we keep the position in the tree locked + */ + tfm_cluster_set_uptodate(&clust->tc); + return 0; + } + /* now prepare output stream.. */ + result = grab_coa(&clust->tc, inode_compression_plugin(inode)); + if (result) + return result; + /* ..and fill this with plain text */ + result = reiser4_inflate_cluster(clust, inode); + if (result) + return result; + /* + * The stream is ready! It won't be obsolete as + * long as we keep last disk cluster item locked. + */ + tfm_cluster_set_uptodate(&clust->tc); + return 0; +} + +/* + * fill one page with plain text. + */ +int do_readpage_ctail(struct inode * inode, struct cluster_handle * clust, + struct page *page, znode_lock_mode mode) +{ + int ret; + unsigned cloff; + char *data; + size_t to_page; + struct tfm_cluster * tc = &clust->tc; + + assert("edward-212", PageLocked(page)); + + if (unlikely(page->mapping != inode->i_mapping)) + return AOP_TRUNCATED_PAGE; + if (PageUptodate(page)) + goto exit; + to_page = pbytes(page_index(page), inode); + if (to_page == 0) { + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + goto exit; + } + if (!tfm_cluster_is_uptodate(&clust->tc)) { + clust->index = pg_to_clust(page->index, inode); + + /* this will unlock/lock the page */ + ret = ctail_read_disk_cluster(clust, inode, page, mode); + + assert("edward-212", PageLocked(page)); + if (ret) + return ret; + + /* refresh bytes */ + to_page = pbytes(page_index(page), inode); + if (to_page == 0) { + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + goto exit; + } + } + if (PageUptodate(page)) + /* somebody else fill it already */ + goto exit; + + assert("edward-119", tfm_cluster_is_uptodate(tc)); + assert("edward-1529", znode_is_any_locked(clust->hint->lh.node)); + + switch (clust->dstat) { + case UNPR_DISK_CLUSTER: + BUG_ON(1); + case TRNC_DISK_CLUSTER: + /* + * Race with truncate! + * We resolve it in favour of the last one (the only way, + * as in this case plain text is unrecoverable) + */ + case FAKE_DISK_CLUSTER: + /* fill the page by zeroes */ + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + break; + case PREP_DISK_CLUSTER: + /* fill page by transformed stream with plain text */ + assert("edward-1058", !PageUptodate(page)); + assert("edward-120", tc->len <= inode_cluster_size(inode)); + + /* page index in this logical cluster */ + cloff = pg_to_off_to_cloff(page->index, inode); + + data = kmap(page); + memcpy(data, tfm_stream_data(tc, OUTPUT_STREAM) + cloff, to_page); + memset(data + to_page, 0, (size_t) PAGE_CACHE_SIZE - to_page); + flush_dcache_page(page); + kunmap(page); + SetPageUptodate(page); + break; + default: + impossible("edward-1169", "bad disk cluster state"); + } + exit: + return 0; +} + +/* plugin->u.item.s.file.readpage */ +int readpage_ctail(void *vp, struct page *page) +{ + int result; + hint_t * hint; + struct cluster_handle * clust = vp; + + assert("edward-114", clust != NULL); + assert("edward-115", PageLocked(page)); + assert("edward-116", !PageUptodate(page)); + assert("edward-118", page->mapping && page->mapping->host); + assert("edward-867", !tfm_cluster_is_uptodate(&clust->tc)); + + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) { + unlock_page(page); + return RETERR(-ENOMEM); + } + clust->hint = hint; + result = load_file_hint(clust->file, hint); + if (result) { + kfree(hint); + unlock_page(page); + return result; + } + assert("vs-25", hint->ext_coord.lh == &hint->lh); + + result = do_readpage_ctail(page->mapping->host, clust, page, + ZNODE_READ_LOCK); + assert("edward-213", PageLocked(page)); + assert("edward-1163", ergo(!result, PageUptodate(page))); + + unlock_page(page); + done_lh(&hint->lh); + hint->ext_coord.valid = 0; + save_file_hint(clust->file, hint); + kfree(hint); + tfm_cluster_clr_uptodate(&clust->tc); + + return result; +} + +/* Helper function for ->readpages() */ +static int ctail_read_page_cluster(struct cluster_handle * clust, + struct inode *inode) +{ + int i; + int result; + assert("edward-779", clust != NULL); + assert("edward-1059", clust->win == NULL); + assert("edward-780", inode != NULL); + + result = prepare_page_cluster(inode, clust, READ_OP); + if (result) + return result; + + assert("edward-781", !tfm_cluster_is_uptodate(&clust->tc)); + + for (i = 0; i < clust->nr_pages; i++) { + struct page *page = clust->pages[i]; + lock_page(page); + result = do_readpage_ctail(inode, clust, page, ZNODE_READ_LOCK); + unlock_page(page); + if (result) + break; + } + tfm_cluster_clr_uptodate(&clust->tc); + put_page_cluster(clust, inode, READ_OP); + return result; +} + +/* filler for read_cache_pages() */ +static int ctail_readpages_filler(void * data, struct page * page) +{ + int ret = 0; + struct cluster_handle * clust = data; + struct inode * inode = clust->file->f_dentry->d_inode; + + assert("edward-1525", page->mapping == inode->i_mapping); + + if (PageUptodate(page)) { + unlock_page(page); + return 0; + } + if (pbytes(page_index(page), inode) == 0) { + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + unlock_page(page); + return 0; + } + move_cluster_forward(clust, inode, page->index); + unlock_page(page); + /* + * read the whole page cluster + */ + ret = ctail_read_page_cluster(clust, inode); + + assert("edward-869", !tfm_cluster_is_uptodate(&clust->tc)); + return ret; +} + +/* + * We populate a bit more then upper readahead suggests: + * with each nominated page we read the whole page cluster + * this page belongs to. + */ +int readpages_ctail(struct file *file, struct address_space *mapping, + struct list_head *pages) +{ + int ret = 0; + hint_t *hint; + struct cluster_handle clust; + struct inode *inode = mapping->host; + + assert("edward-1521", inode == file->f_dentry->d_inode); + + cluster_init_read(&clust, NULL); + clust.file = file; + hint = kmalloc(sizeof(*hint), reiser4_ctx_gfp_mask_get()); + if (hint == NULL) { + warning("vs-28", "failed to allocate hint"); + ret = RETERR(-ENOMEM); + goto exit1; + } + clust.hint = hint; + ret = load_file_hint(clust.file, hint); + if (ret) { + warning("edward-1522", "failed to load hint"); + goto exit2; + } + assert("vs-26", hint->ext_coord.lh == &hint->lh); + ret = alloc_cluster_pgset(&clust, cluster_nrpages(inode)); + if (ret) { + warning("edward-1523", "failed to alloc pgset"); + goto exit3; + } + ret = read_cache_pages(mapping, pages, ctail_readpages_filler, &clust); + + assert("edward-870", !tfm_cluster_is_uptodate(&clust.tc)); + exit3: + done_lh(&hint->lh); + save_file_hint(file, hint); + hint->ext_coord.valid = 0; + exit2: + kfree(hint); + exit1: + put_cluster_handle(&clust); + return ret; +} + +/* + plugin->u.item.s.file.append_key + key of the first item of the next disk cluster +*/ +reiser4_key *append_key_ctail(const coord_t * coord, reiser4_key * key) +{ + assert("edward-1241", item_id_by_coord(coord) == CTAIL_ID); + assert("edward-1242", cluster_shift_ok(cluster_shift_by_coord(coord))); + + item_key_by_coord(coord, key); + set_key_offset(key, ((__u64) (clust_by_coord(coord, NULL)) + 1) + << cluster_shift_by_coord(coord)); + return key; +} + +static int insert_unprepped_ctail(struct cluster_handle * clust, + struct inode *inode) +{ + int result; + char buf[UCTAIL_NR_UNITS]; + reiser4_item_data data; + reiser4_key key; + int shift = (int)UCTAIL_SHIFT; + + memset(buf, 0, (size_t) UCTAIL_NR_UNITS); + result = key_by_inode_cryptcompress(inode, + clust_to_off(clust->index, inode), + &key); + if (result) + return result; + data.user = 0; + data.iplug = item_plugin_by_id(CTAIL_ID); + data.arg = &shift; + data.length = sizeof(ctail_item_format) + (size_t) UCTAIL_NR_UNITS; + data.data = buf; + + result = insert_by_coord(&clust->hint->ext_coord.coord, + &data, &key, clust->hint->ext_coord.lh, 0); + return result; +} + +static int +insert_cryptcompress_flow(coord_t * coord, lock_handle * lh, flow_t * f, + int cluster_shift) +{ + int result; + carry_pool *pool; + carry_level *lowest_level; + reiser4_item_data *data; + carry_op *op; + + pool = + init_carry_pool(sizeof(*pool) + 3 * sizeof(*lowest_level) + + sizeof(*data)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + lowest_level = (carry_level *) (pool + 1); + init_carry_level(lowest_level, pool); + data = (reiser4_item_data *) (lowest_level + 3); + + assert("edward-466", coord->between == AFTER_ITEM + || coord->between == AFTER_UNIT || coord->between == BEFORE_ITEM + || coord->between == EMPTY_NODE + || coord->between == BEFORE_UNIT); + + if (coord->between == AFTER_UNIT) { + coord->unit_pos = 0; + coord->between = AFTER_ITEM; + } + op = reiser4_post_carry(lowest_level, COP_INSERT_FLOW, coord->node, + 0 /* operate directly on coord -> node */); + if (IS_ERR(op) || (op == NULL)) { + done_carry_pool(pool); + return RETERR(op ? PTR_ERR(op) : -EIO); + } + data->user = 0; + data->iplug = item_plugin_by_id(CTAIL_ID); + data->arg = &cluster_shift; + + data->length = 0; + data->data = NULL; + + op->u.insert_flow.flags = COPI_DONT_SHIFT_LEFT | COPI_DONT_SHIFT_RIGHT; + op->u.insert_flow.insert_point = coord; + op->u.insert_flow.flow = f; + op->u.insert_flow.data = data; + op->u.insert_flow.new_nodes = 0; + + lowest_level->track_type = CARRY_TRACK_CHANGE; + lowest_level->tracked = lh; + + result = reiser4_carry(lowest_level, NULL); + done_carry_pool(pool); + + return result; +} + +/* Implementation of CRC_APPEND_ITEM mode of ctail conversion */ +static int insert_cryptcompress_flow_in_place(coord_t * coord, + lock_handle * lh, flow_t * f, + int cluster_shift) +{ + int ret; + coord_t pos; + lock_handle lock; + + assert("edward-484", + coord->between == AT_UNIT || coord->between == AFTER_ITEM); + assert("edward-485", item_id_by_coord(coord) == CTAIL_ID); + + coord_dup(&pos, coord); + pos.unit_pos = 0; + pos.between = AFTER_ITEM; + + init_lh(&lock); + copy_lh(&lock, lh); + + ret = insert_cryptcompress_flow(&pos, &lock, f, cluster_shift); + done_lh(&lock); + assert("edward-1347", znode_is_write_locked(lh->node)); + assert("edward-1228", !ret); + return ret; +} + +/* Implementation of CRC_OVERWRITE_ITEM mode of ctail conversion */ +static int overwrite_ctail(coord_t * coord, flow_t * f) +{ + unsigned count; + + assert("edward-269", f->user == 0); + assert("edward-270", f->data != NULL); + assert("edward-271", f->length > 0); + assert("edward-272", coord_is_existing_unit(coord)); + assert("edward-273", coord->unit_pos == 0); + assert("edward-274", znode_is_write_locked(coord->node)); + assert("edward-275", reiser4_schedulable()); + assert("edward-467", item_id_by_coord(coord) == CTAIL_ID); + assert("edward-1243", ctail_ok(coord)); + + count = nr_units_ctail(coord); + + if (count > f->length) + count = f->length; + memcpy(first_unit(coord), f->data, count); + move_flow_forward(f, count); + coord->unit_pos += count; + return 0; +} + +/* Implementation of CRC_CUT_ITEM mode of ctail conversion: + cut ctail (part or whole) starting from next unit position */ +static int cut_ctail(coord_t * coord) +{ + coord_t stop; + + assert("edward-435", coord->between == AT_UNIT && + coord->item_pos < coord_num_items(coord) && + coord->unit_pos <= coord_num_units(coord)); + + if (coord->unit_pos == coord_num_units(coord)) + /* nothing to cut */ + return 0; + coord_dup(&stop, coord); + stop.unit_pos = coord_last_unit_pos(coord); + + return cut_node_content(coord, &stop, NULL, NULL, NULL); +} + +int ctail_insert_unprepped_cluster(struct cluster_handle * clust, + struct inode * inode) +{ + int result; + assert("edward-1244", inode != NULL); + assert("edward-1245", clust->hint != NULL); + assert("edward-1246", clust->dstat == FAKE_DISK_CLUSTER); + assert("edward-1247", clust->reserved == 1); + + result = get_disk_cluster_locked(clust, inode, ZNODE_WRITE_LOCK); + if (cbk_errored(result)) + return result; + assert("edward-1249", result == CBK_COORD_NOTFOUND); + assert("edward-1250", znode_is_write_locked(clust->hint->lh.node)); + + assert("edward-1295", + clust->hint->ext_coord.lh->node == + clust->hint->ext_coord.coord.node); + + coord_set_between_clusters(&clust->hint->ext_coord.coord); + + result = insert_unprepped_ctail(clust, inode); + all_grabbed2free(); + + assert("edward-1251", !result); + assert("edward-1252", cryptcompress_inode_ok(inode)); + assert("edward-1253", znode_is_write_locked(clust->hint->lh.node)); + assert("edward-1254", + reiser4_clustered_blocks(reiser4_get_current_sb())); + assert("edward-1255", + znode_convertible(clust->hint->ext_coord.coord.node)); + + return result; +} + +static int do_convert_ctail(flush_pos_t * pos, cryptcompress_write_mode_t mode) +{ + int result = 0; + struct convert_item_info * info; + + assert("edward-468", pos != NULL); + assert("edward-469", pos->sq != NULL); + assert("edward-845", item_convert_data(pos) != NULL); + + info = item_convert_data(pos); + assert("edward-679", info->flow.data != NULL); + + switch (mode) { + case CRC_APPEND_ITEM: + assert("edward-1229", info->flow.length != 0); + assert("edward-1256", + cluster_shift_ok(cluster_shift_by_coord(&pos->coord))); + result = + insert_cryptcompress_flow_in_place(&pos->coord, + &pos->lock, + &info->flow, + info->cluster_shift); + break; + case CRC_OVERWRITE_ITEM: + assert("edward-1230", info->flow.length != 0); + overwrite_ctail(&pos->coord, &info->flow); + if (info->flow.length != 0) + break; + case CRC_CUT_ITEM: + assert("edward-1231", info->flow.length == 0); + result = cut_ctail(&pos->coord); + break; + default: + result = RETERR(-EIO); + impossible("edward-244", "bad convert mode"); + } + return result; +} + +/* plugin->u.item.f.scan */ +int scan_ctail(flush_scan * scan) +{ + int result = 0; + struct page *page; + struct inode *inode; + jnode *node = scan->node; + + assert("edward-227", scan->node != NULL); + assert("edward-228", jnode_is_cluster_page(scan->node)); + assert("edward-639", znode_is_write_locked(scan->parent_lock.node)); + + page = jnode_page(node); + inode = page->mapping->host; + + if (!reiser4_scanning_left(scan)) + return result; + if (!ZF_ISSET(scan->parent_lock.node, JNODE_DIRTY)) + znode_make_dirty(scan->parent_lock.node); + + if (!znode_convertible(scan->parent_lock.node)) { + if (JF_ISSET(scan->node, JNODE_DIRTY)) + znode_set_convertible(scan->parent_lock.node); + else { + warning("edward-681", + "cluster page is already processed"); + return -EAGAIN; + } + } + return result; +} + +/* If true, this function attaches children */ +static int should_attach_convert_idata(flush_pos_t * pos) +{ + int result; + assert("edward-431", pos != NULL); + assert("edward-432", pos->child == NULL); + assert("edward-619", znode_is_write_locked(pos->coord.node)); + assert("edward-470", + item_plugin_by_coord(&pos->coord) == + item_plugin_by_id(CTAIL_ID)); + + /* check for leftmost child */ + utmost_child_ctail(&pos->coord, LEFT_SIDE, &pos->child); + + if (!pos->child) + return 0; + spin_lock_jnode(pos->child); + result = (JF_ISSET(pos->child, JNODE_DIRTY) && + pos->child->atom == ZJNODE(pos->coord.node)->atom); + spin_unlock_jnode(pos->child); + if (!result && pos->child) { + /* existing child isn't to attach, clear up this one */ + jput(pos->child); + pos->child = NULL; + } + return result; +} + +/** + * Collect all needed information about the object here, + * as in-memory inode can be evicted from memory before + * disk update completion. + */ +static int init_convert_data_ctail(struct convert_item_info * idata, + struct inode *inode) +{ + assert("edward-813", idata != NULL); + assert("edward-814", inode != NULL); + + idata->cluster_shift = inode_cluster_shift(inode); + idata->d_cur = DC_FIRST_ITEM; + idata->d_next = DC_INVALID_STATE; + + return 0; +} + +static int alloc_item_convert_data(struct convert_info * sq) +{ + assert("edward-816", sq != NULL); + assert("edward-817", sq->itm == NULL); + + sq->itm = kmalloc(sizeof(*sq->itm), reiser4_ctx_gfp_mask_get()); + if (sq->itm == NULL) + return RETERR(-ENOMEM); + return 0; +} + +static void free_item_convert_data(struct convert_info * sq) +{ + assert("edward-818", sq != NULL); + assert("edward-819", sq->itm != NULL); + assert("edward-820", sq->iplug != NULL); + + kfree(sq->itm); + sq->itm = NULL; + return; +} + +static int alloc_convert_data(flush_pos_t * pos) +{ + assert("edward-821", pos != NULL); + assert("edward-822", pos->sq == NULL); + + pos->sq = kmalloc(sizeof(*pos->sq), reiser4_ctx_gfp_mask_get()); + if (!pos->sq) + return RETERR(-ENOMEM); + memset(pos->sq, 0, sizeof(*pos->sq)); + cluster_init_write(&pos->sq->clust, NULL); + return 0; +} + +void free_convert_data(flush_pos_t * pos) +{ + struct convert_info *sq; + + assert("edward-823", pos != NULL); + assert("edward-824", pos->sq != NULL); + + sq = pos->sq; + if (sq->itm) + free_item_convert_data(sq); + put_cluster_handle(&sq->clust); + kfree(pos->sq); + pos->sq = NULL; + return; +} + +static int init_item_convert_data(flush_pos_t * pos, struct inode *inode) +{ + struct convert_info *sq; + + assert("edward-825", pos != NULL); + assert("edward-826", pos->sq != NULL); + assert("edward-827", item_convert_data(pos) != NULL); + assert("edward-828", inode != NULL); + + sq = pos->sq; + + memset(sq->itm, 0, sizeof(*sq->itm)); + + /* iplug->init_convert_data() */ + return init_convert_data_ctail(sq->itm, inode); +} + +/* create and attach disk cluster info used by 'convert' phase of the flush + squalloc() */ +static int attach_convert_idata(flush_pos_t * pos, struct inode *inode) +{ + int ret = 0; + struct convert_item_info *info; + struct cluster_handle *clust; + file_plugin *fplug = inode_file_plugin(inode); + compression_plugin *cplug = inode_compression_plugin(inode); + + assert("edward-248", pos != NULL); + assert("edward-249", pos->child != NULL); + assert("edward-251", inode != NULL); + assert("edward-682", cryptcompress_inode_ok(inode)); + assert("edward-252", + fplug == file_plugin_by_id(CRYPTCOMPRESS_FILE_PLUGIN_ID)); + assert("edward-473", + item_plugin_by_coord(&pos->coord) == + item_plugin_by_id(CTAIL_ID)); + + if (!pos->sq) { + ret = alloc_convert_data(pos); + if (ret) + return ret; + } + clust = &pos->sq->clust; + ret = grab_coa(&clust->tc, cplug); + if (ret) + goto err; + ret = set_cluster_by_page(clust, + jnode_page(pos->child), + MAX_CLUSTER_NRPAGES); + if (ret) + goto err; + + assert("edward-829", pos->sq != NULL); + assert("edward-250", item_convert_data(pos) == NULL); + + pos->sq->iplug = item_plugin_by_id(CTAIL_ID); + + ret = alloc_item_convert_data(pos->sq); + if (ret) + goto err; + ret = init_item_convert_data(pos, inode); + if (ret) + goto err; + info = item_convert_data(pos); + + ret = checkout_logical_cluster(clust, pos->child, inode); + if (ret) + goto err; + + reiser4_deflate_cluster(clust, inode); + inc_item_convert_count(pos); + + /* prepare flow for insertion */ + fplug->flow_by_inode(inode, + (const char __user *)tfm_stream_data(&clust->tc, OUTPUT_STREAM), + 0 /* kernel space */ , + clust->tc.len, + clust_to_off(clust->index, inode), + WRITE_OP, &info->flow); + jput(pos->child); + return 0; + err: + jput(pos->child); + free_convert_data(pos); + return ret; +} + +/* clear up disk cluster info */ +static void detach_convert_idata(struct convert_info * sq) +{ + struct convert_item_info *info; + + assert("edward-253", sq != NULL); + assert("edward-840", sq->itm != NULL); + + info = sq->itm; + assert("edward-1212", info->flow.length == 0); + + free_item_convert_data(sq); + return; +} + +/* plugin->u.item.f.utmost_child */ + +/* This function sets leftmost child for a first cluster item, + if the child exists, and NULL in other cases. + NOTE-EDWARD: Do not call this for RIGHT_SIDE */ + +int utmost_child_ctail(const coord_t * coord, sideof side, jnode ** child) +{ + reiser4_key key; + + item_key_by_coord(coord, &key); + + assert("edward-257", coord != NULL); + assert("edward-258", child != NULL); + assert("edward-259", side == LEFT_SIDE); + assert("edward-260", + item_plugin_by_coord(coord) == item_plugin_by_id(CTAIL_ID)); + + if (!is_disk_cluster_key(&key, coord)) + *child = NULL; + else + *child = jlookup(current_tree, + get_key_objectid(item_key_by_coord + (coord, &key)), + off_to_pg(get_key_offset(&key))); + return 0; +} + +/* Returns true if @p2 is the next item to @p1 + in the _same_ disk cluster. + Disk cluster is a set of items. If ->clustered() != NULL, + with each item the whole disk cluster should be read/modified +*/ + +/* Go rightward and check for next disk cluster item, set + * d_next to DC_CHAINED_ITEM, if the last one exists. + * If the current position is last item, go to right neighbor. + * Skip empty nodes. Note, that right neighbors may be not in + * the slum because of races. If so, make it dirty and + * convertible. + */ +static int next_item_dc_stat(flush_pos_t * pos) +{ + int ret = 0; + int stop = 0; + znode *cur; + coord_t coord; + lock_handle lh; + lock_handle right_lock; + + assert("edward-1232", !node_is_empty(pos->coord.node)); + assert("edward-1014", + pos->coord.item_pos < coord_num_items(&pos->coord)); + assert("edward-1015", chaining_data_present(pos)); + assert("edward-1017", + item_convert_data(pos)->d_next == DC_INVALID_STATE); + + item_convert_data(pos)->d_next = DC_AFTER_CLUSTER; + + if (item_convert_data(pos)->d_cur == DC_AFTER_CLUSTER) + return ret; + if (pos->coord.item_pos < coord_num_items(&pos->coord) - 1) + return ret; + + /* Check next slum item. + * Note, that it can not be killed by concurrent truncate, + * as the last one will want the lock held by us. + */ + init_lh(&right_lock); + cur = pos->coord.node; + + while (!stop) { + init_lh(&lh); + ret = reiser4_get_right_neighbor(&lh, + cur, + ZNODE_WRITE_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (ret) + break; + ret = zload(lh.node); + if (ret) { + done_lh(&lh); + break; + } + coord_init_before_first_item(&coord, lh.node); + + if (node_is_empty(lh.node)) { + znode_make_dirty(lh.node); + znode_set_convertible(lh.node); + stop = 0; + } else if (same_disk_cluster(&pos->coord, &coord)) { + + item_convert_data(pos)->d_next = DC_CHAINED_ITEM; + + if (!ZF_ISSET(lh.node, JNODE_DIRTY)) { + /* + warning("edward-1024", + "next slum item mergeable, " + "but znode %p isn't dirty\n", + lh.node); + */ + znode_make_dirty(lh.node); + } + if (!znode_convertible(lh.node)) { + /* + warning("edward-1272", + "next slum item mergeable, " + "but znode %p isn't convertible\n", + lh.node); + */ + znode_set_convertible(lh.node); + } + stop = 1; + } else + stop = 1; + zrelse(lh.node); + done_lh(&right_lock); + copy_lh(&right_lock, &lh); + done_lh(&lh); + cur = right_lock.node; + } + done_lh(&right_lock); + + if (ret == -E_NO_NEIGHBOR) + ret = 0; + return ret; +} + +static int +assign_convert_mode(struct convert_item_info * idata, + cryptcompress_write_mode_t * mode) +{ + int result = 0; + + assert("edward-1025", idata != NULL); + + if (idata->flow.length) { + /* append or overwrite */ + switch (idata->d_cur) { + case DC_FIRST_ITEM: + case DC_CHAINED_ITEM: + *mode = CRC_OVERWRITE_ITEM; + break; + case DC_AFTER_CLUSTER: + *mode = CRC_APPEND_ITEM; + break; + default: + impossible("edward-1018", "wrong current item state"); + } + } else { + /* cut or invalidate */ + switch (idata->d_cur) { + case DC_FIRST_ITEM: + case DC_CHAINED_ITEM: + *mode = CRC_CUT_ITEM; + break; + case DC_AFTER_CLUSTER: + result = 1; + break; + default: + impossible("edward-1019", "wrong current item state"); + } + } + return result; +} + +/* plugin->u.item.f.convert */ +/* write ctail in guessed mode */ +int convert_ctail(flush_pos_t * pos) +{ + int result; + int nr_items; + cryptcompress_write_mode_t mode = CRC_OVERWRITE_ITEM; + + assert("edward-1020", pos != NULL); + assert("edward-1213", coord_num_items(&pos->coord) != 0); + assert("edward-1257", item_id_by_coord(&pos->coord) == CTAIL_ID); + assert("edward-1258", ctail_ok(&pos->coord)); + assert("edward-261", pos->coord.node != NULL); + + nr_items = coord_num_items(&pos->coord); + if (!chaining_data_present(pos)) { + if (should_attach_convert_idata(pos)) { + /* attach convert item info */ + struct inode *inode; + + assert("edward-264", pos->child != NULL); + assert("edward-265", jnode_page(pos->child) != NULL); + assert("edward-266", + jnode_page(pos->child)->mapping != NULL); + + inode = jnode_page(pos->child)->mapping->host; + + assert("edward-267", inode != NULL); + + /* attach item convert info by child and put the last one */ + result = attach_convert_idata(pos, inode); + pos->child = NULL; + if (result == -E_REPEAT) { + /* jnode became clean, or there is no dirty + pages (nothing to update in disk cluster) */ + warning("edward-1021", + "convert_ctail: nothing to attach"); + return 0; + } + if (result != 0) + return result; + } else + /* unconvertible */ + return 0; + } else { + /* use old convert info */ + + struct convert_item_info *idata; + + idata = item_convert_data(pos); + + result = assign_convert_mode(idata, &mode); + if (result) { + /* disk cluster is over, + nothing to update anymore */ + detach_convert_idata(pos->sq); + return 0; + } + } + + assert("edward-433", chaining_data_present(pos)); + assert("edward-1022", + pos->coord.item_pos < coord_num_items(&pos->coord)); + + /* check if next item is of current disk cluster */ + result = next_item_dc_stat(pos); + if (result) { + detach_convert_idata(pos->sq); + return result; + } + result = do_convert_ctail(pos, mode); + if (result) { + detach_convert_idata(pos->sq); + return result; + } + switch (mode) { + case CRC_CUT_ITEM: + assert("edward-1214", item_convert_data(pos)->flow.length == 0); + assert("edward-1215", + coord_num_items(&pos->coord) == nr_items || + coord_num_items(&pos->coord) == nr_items - 1); + if (item_convert_data(pos)->d_next == DC_CHAINED_ITEM) + break; + if (coord_num_items(&pos->coord) != nr_items) { + /* the item was killed, no more chained items */ + detach_convert_idata(pos->sq); + if (!node_is_empty(pos->coord.node)) + /* make sure the next item will be scanned */ + coord_init_before_item(&pos->coord); + break; + } + case CRC_APPEND_ITEM: + assert("edward-434", item_convert_data(pos)->flow.length == 0); + detach_convert_idata(pos->sq); + break; + case CRC_OVERWRITE_ITEM: + if (coord_is_unprepped_ctail(&pos->coord)) { + /* convert unpprepped ctail to prepped one */ + assert("edward-1259", + cluster_shift_ok(item_convert_data(pos)-> + cluster_shift)); + put_unaligned((d8)item_convert_data(pos)->cluster_shift, + &ctail_formatted_at(&pos->coord)-> + cluster_shift); + } + break; + } + return result; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/ctail.h linux-2.6.30/fs/reiser4/plugin/item/ctail.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/ctail.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/ctail.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,102 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Ctail items are fragments (or bodies) of special tipe to provide + optimal storage of encrypted and(or) compressed files. */ + + +#if !defined( __FS_REISER4_CTAIL_H__ ) +#define __FS_REISER4_CTAIL_H__ + +/* Disk format of ctail item */ +typedef struct ctail_item_format { + /* packed shift; + if its value is different from UCTAIL_SHIFT (see below), then + size of disk cluster is calculated as (1 << cluster_shift) */ + d8 cluster_shift; + /* ctail body */ + d8 body[0]; +} __attribute__ ((packed)) ctail_item_format; + +/* "Unprepped" disk cluster is represented by a single ctail item + with the following "magic" attributes: */ +/* "magic" cluster_shift */ +#define UCTAIL_SHIFT 0xff +/* How many units unprepped ctail item has */ +#define UCTAIL_NR_UNITS 1 + +/* The following is a set of various item states in a disk cluster. + Disk cluster is a set of items whose keys belong to the interval + [dc_key , dc_key + disk_cluster_size - 1] */ +typedef enum { + DC_INVALID_STATE = 0, + DC_FIRST_ITEM = 1, + DC_CHAINED_ITEM = 2, + DC_AFTER_CLUSTER = 3 +} dc_item_stat; + +/* ctail-specific extension. + In particular this describes parameters of disk cluster an item belongs to */ +struct ctail_coord_extension { + int shift; /* this contains cluster_shift extracted from + ctail_item_format (above), or UCTAIL_SHIFT + (the last one is the "magic" of unprepped disk clusters)*/ + int dsize; /* size of a prepped disk cluster */ + int ncount; /* count of nodes occupied by a disk cluster */ +}; + +struct cut_list; + +/* plugin->item.b.* */ +int can_contain_key_ctail(const coord_t *, const reiser4_key *, + const reiser4_item_data *); +int mergeable_ctail(const coord_t * p1, const coord_t * p2); +pos_in_node_t nr_units_ctail(const coord_t * coord); +int estimate_ctail(const coord_t * coord, const reiser4_item_data * data); +void print_ctail(const char *prefix, coord_t * coord); +lookup_result lookup_ctail(const reiser4_key *, lookup_bias, coord_t *); + +int paste_ctail(coord_t * coord, reiser4_item_data * data, + carry_plugin_info * info UNUSED_ARG); +int init_ctail(coord_t *, coord_t *, reiser4_item_data *); +int can_shift_ctail(unsigned free_space, coord_t * coord, + znode * target, shift_direction pend, unsigned *size, + unsigned want); +void copy_units_ctail(coord_t * target, coord_t * source, unsigned from, + unsigned count, shift_direction where_is_free_space, + unsigned free_space); +int cut_units_ctail(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + carry_cut_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +int kill_units_ctail(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + carry_kill_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +int ctail_ok(const coord_t * coord); +int check_ctail(const coord_t * coord, const char **error); + +/* plugin->u.item.s.* */ +int read_ctail(struct file *, flow_t *, hint_t *); +int readpage_ctail(void *, struct page *); +int readpages_ctail(struct file *, struct address_space *, struct list_head *); +reiser4_key *append_key_ctail(const coord_t *, reiser4_key *); +int create_hook_ctail(const coord_t * coord, void *arg); +int kill_hook_ctail(const coord_t *, pos_in_node_t, pos_in_node_t, + carry_kill_data *); +int shift_hook_ctail(const coord_t *, unsigned, unsigned, znode *); + +/* plugin->u.item.f */ +int utmost_child_ctail(const coord_t *, sideof, jnode **); +int scan_ctail(flush_scan *); +int convert_ctail(flush_pos_t *); +size_t inode_scaled_cluster_size(struct inode *); + +#endif /* __FS_REISER4_CTAIL_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/extent.c linux-2.6.30/fs/reiser4/plugin/item/extent.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/extent.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/extent.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,197 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "item.h" +#include "../../key.h" +#include "../../super.h" +#include "../../carry.h" +#include "../../inode.h" +#include "../../page_cache.h" +#include "../../flush.h" +#include "../object.h" + +/* prepare structure reiser4_item_data. It is used to put one extent unit into tree */ +/* Audited by: green(2002.06.13) */ +reiser4_item_data *init_new_extent(reiser4_item_data * data, void *ext_unit, + int nr_extents) +{ + data->data = ext_unit; + /* data->data is kernel space */ + data->user = 0; + data->length = sizeof(reiser4_extent) * nr_extents; + data->arg = NULL; + data->iplug = item_plugin_by_id(EXTENT_POINTER_ID); + return data; +} + +/* how many bytes are addressed by @nr first extents of the extent item */ +reiser4_block_nr reiser4_extent_size(const coord_t * coord, pos_in_node_t nr) +{ + pos_in_node_t i; + reiser4_block_nr blocks; + reiser4_extent *ext; + + ext = item_body_by_coord(coord); + assert("vs-263", nr <= nr_units_extent(coord)); + + blocks = 0; + for (i = 0; i < nr; i++, ext++) { + blocks += extent_get_width(ext); + } + + return blocks * current_blocksize; +} + +extent_state state_of_extent(reiser4_extent * ext) +{ + switch ((int)extent_get_start(ext)) { + case 0: + return HOLE_EXTENT; + case 1: + return UNALLOCATED_EXTENT; + default: + break; + } + return ALLOCATED_EXTENT; +} + +int extent_is_unallocated(const coord_t * item) +{ + assert("jmacd-5133", item_is_extent(item)); + + return state_of_extent(extent_by_coord(item)) == UNALLOCATED_EXTENT; +} + +/* set extent's start and width */ +void reiser4_set_extent(reiser4_extent * ext, reiser4_block_nr start, + reiser4_block_nr width) +{ + extent_set_start(ext, start); + extent_set_width(ext, width); +} + +/** + * reiser4_replace_extent - replace extent and paste 1 or 2 after it + * @un_extent: coordinate of extent to be overwritten + * @lh: need better comment + * @key: need better comment + * @exts_to_add: data prepared for insertion into tree + * @replace: need better comment + * @flags: need better comment + * @return_insert_position: need better comment + * + * Overwrites one extent, pastes 1 or 2 more ones after overwritten one. If + * @return_inserted_position is 1 - @un_extent and @lh are returned set to + * first of newly inserted units, if it is 0 - @un_extent and @lh are returned + * set to extent which was overwritten. + */ +int reiser4_replace_extent(struct replace_handle *h, + int return_inserted_position) +{ + int result; + znode *orig_znode; + /*ON_DEBUG(reiser4_extent orig_ext);*/ /* this is for debugging */ + + assert("vs-990", coord_is_existing_unit(h->coord)); + assert("vs-1375", znode_is_write_locked(h->coord->node)); + assert("vs-1426", extent_get_width(&h->overwrite) != 0); + assert("vs-1427", extent_get_width(&h->new_extents[0]) != 0); + assert("vs-1427", ergo(h->nr_new_extents == 2, + extent_get_width(&h->new_extents[1]) != 0)); + + /* compose structure for paste */ + init_new_extent(&h->item, &h->new_extents[0], h->nr_new_extents); + + coord_dup(&h->coord_after, h->coord); + init_lh(&h->lh_after); + copy_lh(&h->lh_after, h->lh); + reiser4_tap_init(&h->watch, &h->coord_after, &h->lh_after, ZNODE_WRITE_LOCK); + reiser4_tap_monitor(&h->watch); + + ON_DEBUG(h->orig_ext = *extent_by_coord(h->coord)); + orig_znode = h->coord->node; + +#if REISER4_DEBUG + /* make sure that key is set properly */ + unit_key_by_coord(h->coord, &h->tmp); + set_key_offset(&h->tmp, + get_key_offset(&h->tmp) + + extent_get_width(&h->overwrite) * current_blocksize); + assert("vs-1080", keyeq(&h->tmp, &h->paste_key)); +#endif + + /* set insert point after unit to be replaced */ + h->coord->between = AFTER_UNIT; + + result = insert_into_item(h->coord, return_inserted_position ? h->lh : NULL, + &h->paste_key, &h->item, h->flags); + if (!result) { + /* now we have to replace the unit after which new units were + inserted. Its position is tracked by @watch */ + reiser4_extent *ext; + znode *node; + + node = h->coord_after.node; + if (node != orig_znode) { + coord_clear_iplug(&h->coord_after); + result = zload(node); + } + + if (likely(!result)) { + ext = extent_by_coord(&h->coord_after); + + assert("vs-987", znode_is_loaded(node)); + assert("vs-988", !memcmp(ext, &h->orig_ext, sizeof(*ext))); + + /* overwrite extent unit */ + memcpy(ext, &h->overwrite, sizeof(reiser4_extent)); + znode_make_dirty(node); + + if (node != orig_znode) + zrelse(node); + + if (return_inserted_position == 0) { + /* coord and lh are to be set to overwritten + extent */ + assert("vs-1662", + WITH_DATA(node, !memcmp(&h->overwrite, + extent_by_coord( + &h->coord_after), + sizeof(reiser4_extent)))); + + *h->coord = h->coord_after; + done_lh(h->lh); + copy_lh(h->lh, &h->lh_after); + } else { + /* h->coord and h->lh are to be set to first of + inserted units */ + assert("vs-1663", + WITH_DATA(h->coord->node, + !memcmp(&h->new_extents[0], + extent_by_coord(h->coord), + sizeof(reiser4_extent)))); + assert("vs-1664", h->lh->node == h->coord->node); + } + } + } + reiser4_tap_done(&h->watch); + + return result; +} + +lock_handle *znode_lh(znode *node) +{ + assert("vs-1371", znode_is_write_locked(node)); + assert("vs-1372", znode_is_wlocked_once(node)); + return list_entry(node->lock.owners.next, lock_handle, owners_link); +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/extent_file_ops.c linux-2.6.30/fs/reiser4/plugin/item/extent_file_ops.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/extent_file_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/extent_file_ops.c 2009-06-22 17:27:31.000000000 +0200 @@ -0,0 +1,1453 @@ +/* COPYRIGHT 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "item.h" +#include "../../inode.h" +#include "../../page_cache.h" +#include "../object.h" + +#include +#include + +static inline reiser4_extent *ext_by_offset(const znode *node, int offset) +{ + reiser4_extent *ext; + + ext = (reiser4_extent *) (zdata(node) + offset); + return ext; +} + +/** + * check_uf_coord - verify coord extension + * @uf_coord: + * @key: + * + * Makes sure that all fields of @uf_coord are set properly. If @key is + * specified - check whether @uf_coord is set correspondingly. + */ +static void check_uf_coord(const uf_coord_t *uf_coord, const reiser4_key *key) +{ +#if REISER4_DEBUG + const coord_t *coord; + const struct extent_coord_extension *ext_coord; + reiser4_extent *ext; + + coord = &uf_coord->coord; + ext_coord = &uf_coord->extension.extent; + ext = ext_by_offset(coord->node, uf_coord->extension.extent.ext_offset); + + assert("", + WITH_DATA(coord->node, + (uf_coord->valid == 1 && + coord_is_iplug_set(coord) && + item_is_extent(coord) && + ext_coord->nr_units == nr_units_extent(coord) && + ext == extent_by_coord(coord) && + ext_coord->width == extent_get_width(ext) && + coord->unit_pos < ext_coord->nr_units && + ext_coord->pos_in_unit < ext_coord->width && + memcmp(ext, &ext_coord->extent, + sizeof(reiser4_extent)) == 0))); + if (key) { + reiser4_key coord_key; + + unit_key_by_coord(&uf_coord->coord, &coord_key); + set_key_offset(&coord_key, + get_key_offset(&coord_key) + + (uf_coord->extension.extent. + pos_in_unit << PAGE_CACHE_SHIFT)); + assert("", keyeq(key, &coord_key)); + } +#endif +} + +static inline reiser4_extent *ext_by_ext_coord(const uf_coord_t *uf_coord) +{ + check_uf_coord(uf_coord, NULL); + + return ext_by_offset(uf_coord->coord.node, + uf_coord->extension.extent.ext_offset); +} + +#if REISER4_DEBUG + +/** + * offset_is_in_unit + * + * + * + */ +/* return 1 if offset @off is inside of extent unit pointed to by @coord. Set + pos_in_unit inside of unit correspondingly */ +static int offset_is_in_unit(const coord_t *coord, loff_t off) +{ + reiser4_key unit_key; + __u64 unit_off; + reiser4_extent *ext; + + ext = extent_by_coord(coord); + + unit_key_extent(coord, &unit_key); + unit_off = get_key_offset(&unit_key); + if (off < unit_off) + return 0; + if (off >= (unit_off + (current_blocksize * extent_get_width(ext)))) + return 0; + return 1; +} + +static int +coord_matches_key_extent(const coord_t * coord, const reiser4_key * key) +{ + reiser4_key item_key; + + assert("vs-771", coord_is_existing_unit(coord)); + assert("vs-1258", keylt(key, append_key_extent(coord, &item_key))); + assert("vs-1259", keyge(key, item_key_by_coord(coord, &item_key))); + + return offset_is_in_unit(coord, get_key_offset(key)); +} + +#endif + +/** + * can_append - + * @key: + * @coord: + * + * Returns 1 if @key is equal to an append key of item @coord is set to + */ +static int can_append(const reiser4_key *key, const coord_t *coord) +{ + reiser4_key append_key; + + return keyeq(key, append_key_extent(coord, &append_key)); +} + +/** + * append_hole + * @coord: + * @lh: + * @key: + * + */ +static int append_hole(coord_t *coord, lock_handle *lh, + const reiser4_key *key) +{ + reiser4_key append_key; + reiser4_block_nr hole_width; + reiser4_extent *ext, new_ext; + reiser4_item_data idata; + + /* last item of file may have to be appended with hole */ + assert("vs-708", znode_get_level(coord->node) == TWIG_LEVEL); + assert("vs-714", item_id_by_coord(coord) == EXTENT_POINTER_ID); + + /* key of first byte which is not addressed by this extent */ + append_key_extent(coord, &append_key); + + assert("", keyle(&append_key, key)); + + /* + * extent item has to be appended with hole. Calculate length of that + * hole + */ + hole_width = ((get_key_offset(key) - get_key_offset(&append_key) + + current_blocksize - 1) >> current_blocksize_bits); + assert("vs-954", hole_width > 0); + + /* set coord after last unit */ + coord_init_after_item_end(coord); + + /* get last extent in the item */ + ext = extent_by_coord(coord); + if (state_of_extent(ext) == HOLE_EXTENT) { + /* + * last extent of a file is hole extent. Widen that extent by + * @hole_width blocks. Note that we do not worry about + * overflowing - extent width is 64 bits + */ + reiser4_set_extent(ext, HOLE_EXTENT_START, + extent_get_width(ext) + hole_width); + znode_make_dirty(coord->node); + return 0; + } + + /* append last item of the file with hole extent unit */ + assert("vs-713", (state_of_extent(ext) == ALLOCATED_EXTENT || + state_of_extent(ext) == UNALLOCATED_EXTENT)); + + reiser4_set_extent(&new_ext, HOLE_EXTENT_START, hole_width); + init_new_extent(&idata, &new_ext, 1); + return insert_into_item(coord, lh, &append_key, &idata, 0); +} + +/** + * check_jnodes + * @twig: longterm locked twig node + * @key: + * + */ +static void check_jnodes(znode *twig, const reiser4_key *key, int count) +{ +#if REISER4_DEBUG + coord_t c; + reiser4_key node_key, jnode_key; + + jnode_key = *key; + + assert("", twig != NULL); + assert("", znode_get_level(twig) == TWIG_LEVEL); + assert("", znode_is_write_locked(twig)); + + zload(twig); + /* get the smallest key in twig node */ + coord_init_first_unit(&c, twig); + unit_key_by_coord(&c, &node_key); + assert("", keyle(&node_key, &jnode_key)); + + coord_init_last_unit(&c, twig); + unit_key_by_coord(&c, &node_key); + if (item_plugin_by_coord(&c)->s.file.append_key) + item_plugin_by_coord(&c)->s.file.append_key(&c, &node_key); + set_key_offset(&jnode_key, + get_key_offset(&jnode_key) + (loff_t)count * PAGE_CACHE_SIZE - 1); + assert("", keylt(&jnode_key, &node_key)); + zrelse(twig); +#endif +} + +/** + * append_last_extent - append last file item + * @uf_coord: coord to start insertion from + * @jnodes: array of jnodes + * @count: number of jnodes in the array + * + * There is already at least one extent item of file @inode in the tree. Append + * the last of them with unallocated extent unit of width @count. Assign + * fake block numbers to jnodes corresponding to the inserted extent. + */ +static int append_last_extent(uf_coord_t *uf_coord, const reiser4_key *key, + jnode **jnodes, int count) +{ + int result; + reiser4_extent new_ext; + reiser4_item_data idata; + coord_t *coord; + struct extent_coord_extension *ext_coord; + reiser4_extent *ext; + reiser4_block_nr block; + jnode *node; + int i; + + coord = &uf_coord->coord; + ext_coord = &uf_coord->extension.extent; + ext = ext_by_ext_coord(uf_coord); + + /* check correctness of position in the item */ + assert("vs-228", coord->unit_pos == coord_last_unit_pos(coord)); + assert("vs-1311", coord->between == AFTER_UNIT); + assert("vs-1302", ext_coord->pos_in_unit == ext_coord->width - 1); + + if (!can_append(key, coord)) { + /* hole extent has to be inserted */ + result = append_hole(coord, uf_coord->lh, key); + uf_coord->valid = 0; + return result; + } + + if (count == 0) + return 0; + + assert("", get_key_offset(key) == (loff_t)index_jnode(jnodes[0]) * PAGE_CACHE_SIZE); + + result = vfs_dq_alloc_block_nodirty(mapping_jnode(jnodes[0])->host, + count); + BUG_ON(result != 0); + + switch (state_of_extent(ext)) { + case UNALLOCATED_EXTENT: + /* + * last extent unit of the file is unallocated one. Increase + * its width by @count + */ + reiser4_set_extent(ext, UNALLOCATED_EXTENT_START, + extent_get_width(ext) + count); + znode_make_dirty(coord->node); + + /* update coord extension */ + ext_coord->width += count; + ON_DEBUG(extent_set_width + (&uf_coord->extension.extent.extent, + ext_coord->width)); + break; + + case HOLE_EXTENT: + case ALLOCATED_EXTENT: + /* + * last extent unit of the file is either hole or allocated + * one. Append one unallocated extent of width @count + */ + reiser4_set_extent(&new_ext, UNALLOCATED_EXTENT_START, count); + init_new_extent(&idata, &new_ext, 1); + result = insert_into_item(coord, uf_coord->lh, key, &idata, 0); + uf_coord->valid = 0; + if (result) + return result; + break; + + default: + return RETERR(-EIO); + } + + /* + * make sure that we hold long term locked twig node containing all + * jnodes we are about to capture + */ + check_jnodes(uf_coord->lh->node, key, count); + + /* + * assign fake block numbers to all jnodes. FIXME: make sure whether + * twig node containing inserted extent item is locked + */ + block = fake_blocknr_unformatted(count); + for (i = 0; i < count; i ++, block ++) { + node = jnodes[i]; + spin_lock_jnode(node); + JF_SET(node, JNODE_CREATED); + jnode_set_block(node, &block); + result = reiser4_try_capture(node, ZNODE_WRITE_LOCK, 0); + BUG_ON(result != 0); + jnode_make_dirty_locked(node); + spin_unlock_jnode(node); + } + return count; +} + +/** + * insert_first_hole - inser hole extent into tree + * @coord: + * @lh: + * @key: + * + * + */ +static int insert_first_hole(coord_t *coord, lock_handle *lh, + const reiser4_key *key) +{ + reiser4_extent new_ext; + reiser4_item_data idata; + reiser4_key item_key; + reiser4_block_nr hole_width; + + /* @coord must be set for inserting of new item */ + assert("vs-711", coord_is_between_items(coord)); + + item_key = *key; + set_key_offset(&item_key, 0ull); + + hole_width = ((get_key_offset(key) + current_blocksize - 1) >> + current_blocksize_bits); + assert("vs-710", hole_width > 0); + + /* compose body of hole extent and insert item into tree */ + reiser4_set_extent(&new_ext, HOLE_EXTENT_START, hole_width); + init_new_extent(&idata, &new_ext, 1); + return insert_extent_by_coord(coord, &idata, &item_key, lh); +} + + +/** + * insert_first_extent - insert first file item + * @inode: inode of file + * @uf_coord: coord to start insertion from + * @jnodes: array of jnodes + * @count: number of jnodes in the array + * @inode: + * + * There are no items of file @inode in the tree yet. Insert unallocated extent + * of width @count into tree or hole extent if writing not to the + * beginning. Assign fake block numbers to jnodes corresponding to the inserted + * unallocated extent. Returns number of jnodes or error code. + */ +static int insert_first_extent(uf_coord_t *uf_coord, const reiser4_key *key, + jnode **jnodes, int count, + struct inode *inode) +{ + int result; + int i; + reiser4_extent new_ext; + reiser4_item_data idata; + reiser4_block_nr block; + struct unix_file_info *uf_info; + jnode *node; + + /* first extent insertion starts at leaf level */ + assert("vs-719", znode_get_level(uf_coord->coord.node) == LEAF_LEVEL); + assert("vs-711", coord_is_between_items(&uf_coord->coord)); + + if (get_key_offset(key) != 0) { + result = insert_first_hole(&uf_coord->coord, uf_coord->lh, key); + uf_coord->valid = 0; + uf_info = unix_file_inode_data(inode); + + /* + * first item insertion is only possible when writing to empty + * file or performing tail conversion + */ + assert("", (uf_info->container == UF_CONTAINER_EMPTY || + (reiser4_inode_get_flag(inode, + REISER4_PART_MIXED) && + reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)))); + /* if file was empty - update its state */ + if (result == 0 && uf_info->container == UF_CONTAINER_EMPTY) + uf_info->container = UF_CONTAINER_EXTENTS; + return result; + } + + if (count == 0) + return 0; + + result = vfs_dq_alloc_block_nodirty(mapping_jnode(jnodes[0])->host, + count); + BUG_ON(result != 0); + + /* + * prepare for tree modification: compose body of item and item data + * structure needed for insertion + */ + reiser4_set_extent(&new_ext, UNALLOCATED_EXTENT_START, count); + init_new_extent(&idata, &new_ext, 1); + + /* insert extent item into the tree */ + result = insert_extent_by_coord(&uf_coord->coord, &idata, key, + uf_coord->lh); + if (result) + return result; + + /* + * make sure that we hold long term locked twig node containing all + * jnodes we are about to capture + */ + check_jnodes(uf_coord->lh->node, key, count); + /* + * assign fake block numbers to all jnodes, capture and mark them dirty + */ + block = fake_blocknr_unformatted(count); + for (i = 0; i < count; i ++, block ++) { + node = jnodes[i]; + spin_lock_jnode(node); + JF_SET(node, JNODE_CREATED); + jnode_set_block(node, &block); + result = reiser4_try_capture(node, ZNODE_WRITE_LOCK, 0); + BUG_ON(result != 0); + jnode_make_dirty_locked(node); + spin_unlock_jnode(node); + } + + /* + * invalidate coordinate, research must be performed to continue + * because write will continue on twig level + */ + uf_coord->valid = 0; + return count; +} + +/** + * plug_hole - replace hole extent with unallocated and holes + * @uf_coord: + * @key: + * @node: + * @h: structure containing coordinate, lock handle, key, etc + * + * Creates an unallocated extent of width 1 within a hole. In worst case two + * additional extents can be created. + */ +static int plug_hole(uf_coord_t *uf_coord, const reiser4_key *key, int *how) +{ + struct replace_handle rh; + reiser4_extent *ext; + reiser4_block_nr width, pos_in_unit; + coord_t *coord; + struct extent_coord_extension *ext_coord; + int return_inserted_position; + + check_uf_coord(uf_coord, key); + + rh.coord = coord_by_uf_coord(uf_coord); + rh.lh = uf_coord->lh; + rh.flags = 0; + + coord = coord_by_uf_coord(uf_coord); + ext_coord = ext_coord_by_uf_coord(uf_coord); + ext = ext_by_ext_coord(uf_coord); + + width = ext_coord->width; + pos_in_unit = ext_coord->pos_in_unit; + + *how = 0; + if (width == 1) { + reiser4_set_extent(ext, UNALLOCATED_EXTENT_START, 1); + znode_make_dirty(coord->node); + /* update uf_coord */ + ON_DEBUG(ext_coord->extent = *ext); + *how = 1; + return 0; + } else if (pos_in_unit == 0) { + /* we deal with first element of extent */ + if (coord->unit_pos) { + /* there is an extent to the left */ + if (state_of_extent(ext - 1) == UNALLOCATED_EXTENT) { + /* + * left neighboring unit is an unallocated + * extent. Increase its width and decrease + * width of hole + */ + extent_set_width(ext - 1, + extent_get_width(ext - 1) + 1); + extent_set_width(ext, width - 1); + znode_make_dirty(coord->node); + + /* update coord extension */ + coord->unit_pos--; + ext_coord->width = extent_get_width(ext - 1); + ext_coord->pos_in_unit = ext_coord->width - 1; + ext_coord->ext_offset -= sizeof(reiser4_extent); + ON_DEBUG(ext_coord->extent = + *extent_by_coord(coord)); + *how = 2; + return 0; + } + } + /* extent for replace */ + reiser4_set_extent(&rh.overwrite, UNALLOCATED_EXTENT_START, 1); + /* extent to be inserted */ + reiser4_set_extent(&rh.new_extents[0], HOLE_EXTENT_START, + width - 1); + rh.nr_new_extents = 1; + + /* have reiser4_replace_extent to return with @coord and + @uf_coord->lh set to unit which was replaced */ + return_inserted_position = 0; + *how = 3; + } else if (pos_in_unit == width - 1) { + /* we deal with last element of extent */ + if (coord->unit_pos < nr_units_extent(coord) - 1) { + /* there is an extent unit to the right */ + if (state_of_extent(ext + 1) == UNALLOCATED_EXTENT) { + /* + * right neighboring unit is an unallocated + * extent. Increase its width and decrease + * width of hole + */ + extent_set_width(ext + 1, + extent_get_width(ext + 1) + 1); + extent_set_width(ext, width - 1); + znode_make_dirty(coord->node); + + /* update coord extension */ + coord->unit_pos++; + ext_coord->width = extent_get_width(ext + 1); + ext_coord->pos_in_unit = 0; + ext_coord->ext_offset += sizeof(reiser4_extent); + ON_DEBUG(ext_coord->extent = + *extent_by_coord(coord)); + *how = 4; + return 0; + } + } + /* extent for replace */ + reiser4_set_extent(&rh.overwrite, HOLE_EXTENT_START, width - 1); + /* extent to be inserted */ + reiser4_set_extent(&rh.new_extents[0], UNALLOCATED_EXTENT_START, + 1); + rh.nr_new_extents = 1; + + /* have reiser4_replace_extent to return with @coord and + @uf_coord->lh set to unit which was inserted */ + return_inserted_position = 1; + *how = 5; + } else { + /* extent for replace */ + reiser4_set_extent(&rh.overwrite, HOLE_EXTENT_START, + pos_in_unit); + /* extents to be inserted */ + reiser4_set_extent(&rh.new_extents[0], UNALLOCATED_EXTENT_START, + 1); + reiser4_set_extent(&rh.new_extents[1], HOLE_EXTENT_START, + width - pos_in_unit - 1); + rh.nr_new_extents = 2; + + /* have reiser4_replace_extent to return with @coord and + @uf_coord->lh set to first of units which were inserted */ + return_inserted_position = 1; + *how = 6; + } + unit_key_by_coord(coord, &rh.paste_key); + set_key_offset(&rh.paste_key, get_key_offset(&rh.paste_key) + + extent_get_width(&rh.overwrite) * current_blocksize); + + uf_coord->valid = 0; + return reiser4_replace_extent(&rh, return_inserted_position); +} + +/** + * overwrite_one_block - + * @uf_coord: + * @key: + * @node: + * + * If @node corresponds to hole extent - create unallocated extent for it and + * assign fake block number. If @node corresponds to allocated extent - assign + * block number of jnode + */ +static int overwrite_one_block(uf_coord_t *uf_coord, const reiser4_key *key, + jnode *node, int *hole_plugged) +{ + int result; + struct extent_coord_extension *ext_coord; + reiser4_extent *ext; + reiser4_block_nr block; + int how; + + assert("vs-1312", uf_coord->coord.between == AT_UNIT); + + result = 0; + ext_coord = ext_coord_by_uf_coord(uf_coord); + ext = ext_by_ext_coord(uf_coord); + assert("", state_of_extent(ext) != UNALLOCATED_EXTENT); + + switch (state_of_extent(ext)) { + case ALLOCATED_EXTENT: + block = extent_get_start(ext) + ext_coord->pos_in_unit; + break; + + case HOLE_EXTENT: + result = vfs_dq_alloc_block_nodirty(mapping_jnode(node)->host, + 1); + BUG_ON(result != 0); + result = plug_hole(uf_coord, key, &how); + if (result) + return result; + block = fake_blocknr_unformatted(1); + if (hole_plugged) + *hole_plugged = 1; + JF_SET(node, JNODE_CREATED); + break; + + default: + return RETERR(-EIO); + } + + jnode_set_block(node, &block); + return 0; +} + +/** + * move_coord - move coordinate forward + * @uf_coord: + * + * Move coordinate one data block pointer forward. Return 1 if coord is set to + * the last one already or is invalid. + */ +static int move_coord(uf_coord_t *uf_coord) +{ + struct extent_coord_extension *ext_coord; + + if (uf_coord->valid == 0) + return 1; + ext_coord = &uf_coord->extension.extent; + ext_coord->pos_in_unit ++; + if (ext_coord->pos_in_unit < ext_coord->width) + /* coordinate moved within the unit */ + return 0; + + /* end of unit is reached. Try to move to next unit */ + ext_coord->pos_in_unit = 0; + uf_coord->coord.unit_pos ++; + if (uf_coord->coord.unit_pos < ext_coord->nr_units) { + /* coordinate moved to next unit */ + ext_coord->ext_offset += sizeof(reiser4_extent); + ext_coord->width = + extent_get_width(ext_by_offset + (uf_coord->coord.node, + ext_coord->ext_offset)); + ON_DEBUG(ext_coord->extent = + *ext_by_offset(uf_coord->coord.node, + ext_coord->ext_offset)); + return 0; + } + /* end of item is reached */ + uf_coord->valid = 0; + return 1; +} + +/** + * overwrite_extent - + * @inode: + * + * Returns number of handled jnodes. + */ +static int overwrite_extent(uf_coord_t *uf_coord, const reiser4_key *key, + jnode **jnodes, int count, int *plugged_hole) +{ + int result; + reiser4_key k; + int i; + jnode *node; + + k = *key; + for (i = 0; i < count; i ++) { + node = jnodes[i]; + if (*jnode_get_block(node) == 0) { + result = overwrite_one_block(uf_coord, &k, node, plugged_hole); + if (result) + return result; + } + /* + * make sure that we hold long term locked twig node containing + * all jnodes we are about to capture + */ + check_jnodes(uf_coord->lh->node, &k, 1); + /* + * assign fake block numbers to all jnodes, capture and mark + * them dirty + */ + spin_lock_jnode(node); + result = reiser4_try_capture(node, ZNODE_WRITE_LOCK, 0); + BUG_ON(result != 0); + jnode_make_dirty_locked(node); + spin_unlock_jnode(node); + + if (uf_coord->valid == 0) + return i + 1; + + check_uf_coord(uf_coord, &k); + + if (move_coord(uf_coord)) { + /* + * failed to move to the next node pointer. Either end + * of file or end of twig node is reached. In the later + * case we might go to the right neighbor. + */ + uf_coord->valid = 0; + return i + 1; + } + set_key_offset(&k, get_key_offset(&k) + PAGE_CACHE_SIZE); + } + + return count; +} + +/** + * reiser4_update_extent + * @file: + * @jnodes: + * @count: + * @off: + * + */ +int reiser4_update_extent(struct inode *inode, jnode *node, loff_t pos, + int *plugged_hole) +{ + int result; + znode *loaded; + uf_coord_t uf_coord; + coord_t *coord; + lock_handle lh; + reiser4_key key; + + assert("", reiser4_lock_counters()->d_refs == 0); + + key_by_inode_and_offset_common(inode, pos, &key); + + init_uf_coord(&uf_coord, &lh); + coord = &uf_coord.coord; + result = find_file_item_nohint(coord, &lh, &key, + ZNODE_WRITE_LOCK, inode); + if (IS_CBKERR(result)) { + assert("", reiser4_lock_counters()->d_refs == 0); + return result; + } + + result = zload(coord->node); + BUG_ON(result != 0); + loaded = coord->node; + + if (coord->between == AFTER_UNIT) { + /* + * append existing extent item with unallocated extent of width + * nr_jnodes + */ + init_coord_extension_extent(&uf_coord, + get_key_offset(&key)); + result = append_last_extent(&uf_coord, &key, + &node, 1); + } else if (coord->between == AT_UNIT) { + /* + * overwrite + * not optimal yet. Will be optimized if new write will show + * performance win. + */ + init_coord_extension_extent(&uf_coord, + get_key_offset(&key)); + result = overwrite_extent(&uf_coord, &key, + &node, 1, plugged_hole); + } else { + /* + * there are no items of this file in the tree yet. Create + * first item of the file inserting one unallocated extent of + * width nr_jnodes + */ + result = insert_first_extent(&uf_coord, &key, &node, 1, inode); + } + assert("", result == 1 || result < 0); + zrelse(loaded); + done_lh(&lh); + assert("", reiser4_lock_counters()->d_refs == 0); + return (result == 1) ? 0 : result; +} + +/** + * update_extents + * @file: + * @jnodes: + * @count: + * @off: + * + */ +static int update_extents(struct file *file, struct inode *inode, + jnode **jnodes, int count, loff_t pos) +{ + struct hint hint; + reiser4_key key; + int result; + znode *loaded; + + result = load_file_hint(file, &hint); + BUG_ON(result != 0); + + if (count != 0) + /* + * count == 0 is special case: expanding truncate + */ + pos = (loff_t)index_jnode(jnodes[0]) << PAGE_CACHE_SHIFT; + key_by_inode_and_offset_common(inode, pos, &key); + + assert("", reiser4_lock_counters()->d_refs == 0); + + do { + result = find_file_item(&hint, &key, ZNODE_WRITE_LOCK, inode); + if (IS_CBKERR(result)) { + assert("", reiser4_lock_counters()->d_refs == 0); + return result; + } + + result = zload(hint.ext_coord.coord.node); + BUG_ON(result != 0); + loaded = hint.ext_coord.coord.node; + + if (hint.ext_coord.coord.between == AFTER_UNIT) { + /* + * append existing extent item with unallocated extent + * of width nr_jnodes + */ + if (hint.ext_coord.valid == 0) + /* NOTE: get statistics on this */ + init_coord_extension_extent(&hint.ext_coord, + get_key_offset(&key)); + result = append_last_extent(&hint.ext_coord, &key, + jnodes, count); + } else if (hint.ext_coord.coord.between == AT_UNIT) { + /* + * overwrite + * not optimal yet. Will be optimized if new write will + * show performance win. + */ + if (hint.ext_coord.valid == 0) + /* NOTE: get statistics on this */ + init_coord_extension_extent(&hint.ext_coord, + get_key_offset(&key)); + result = overwrite_extent(&hint.ext_coord, &key, + jnodes, count, NULL); + } else { + /* + * there are no items of this file in the tree + * yet. Create first item of the file inserting one + * unallocated extent of * width nr_jnodes + */ + result = insert_first_extent(&hint.ext_coord, &key, + jnodes, count, inode); + } + zrelse(loaded); + if (result < 0) { + done_lh(hint.ext_coord.lh); + break; + } + + jnodes += result; + count -= result; + set_key_offset(&key, get_key_offset(&key) + result * PAGE_CACHE_SIZE); + + /* seal and unlock znode */ + if (hint.ext_coord.valid) + reiser4_set_hint(&hint, &key, ZNODE_WRITE_LOCK); + else + reiser4_unset_hint(&hint); + + } while (count > 0); + + save_file_hint(file, &hint); + assert("", reiser4_lock_counters()->d_refs == 0); + return result; +} + +/** + * write_extent_reserve_space - reserve space for extent write operation + * @inode: + * + * Estimates and reserves space which may be required for writing + * WRITE_GRANULARITY pages of file. + */ +static int write_extent_reserve_space(struct inode *inode) +{ + __u64 count; + reiser4_tree *tree; + + /* + * to write WRITE_GRANULARITY pages to a file by extents we have to + * reserve disk space for: + + * 1. find_file_item may have to insert empty node to the tree (empty + * leaf node between two extent items). This requires 1 block and + * number of blocks which are necessary to perform insertion of an + * internal item into twig level. + + * 2. for each of written pages there might be needed 1 block and + * number of blocks which might be necessary to perform insertion of or + * paste to an extent item. + + * 3. stat data update + */ + tree = reiser4_tree_by_inode(inode); + count = estimate_one_insert_item(tree) + + WRITE_GRANULARITY * (1 + estimate_one_insert_into_item(tree)) + + estimate_one_insert_item(tree); + grab_space_enable(); + return reiser4_grab_space(count, 0 /* flags */); +} + +/* + * filemap_copy_from_user no longer exists in generic code, because it + * is deadlocky (copying from user while holding the page lock is bad). + * As a temporary fix for reiser4, just define it here. + */ +static inline size_t +filemap_copy_from_user(struct page *page, unsigned long offset, + const char __user *buf, unsigned bytes) +{ + char *kaddr; + int left; + + kaddr = kmap_atomic(page, KM_USER0); + left = __copy_from_user_inatomic_nocache(kaddr + offset, buf, bytes); + kunmap_atomic(kaddr, KM_USER0); + + if (left != 0) { + /* Do it the slow way */ + kaddr = kmap(page); + left = __copy_from_user_nocache(kaddr + offset, buf, bytes); + kunmap(page); + } + return bytes - left; +} + +/** + * reiser4_write_extent - write method of extent item plugin + * @file: file to write to + * @buf: address of user-space buffer + * @count: number of bytes to write + * @pos: position in file to write to + * + */ +ssize_t reiser4_write_extent(struct file *file, struct inode * inode, + const char __user *buf, size_t count, loff_t *pos) +{ + int have_to_update_extent; + int nr_pages, nr_dirty; + struct page *page; + jnode *jnodes[WRITE_GRANULARITY + 1]; + unsigned long index; + unsigned long end; + int i; + int to_page, page_off; + size_t left, written; + int result = 0; + + if (write_extent_reserve_space(inode)) + return RETERR(-ENOSPC); + + if (count == 0) { + /* truncate case */ + update_extents(file, inode, jnodes, 0, *pos); + return 0; + } + + BUG_ON(get_current_context()->trans->atom != NULL); + + left = count; + index = *pos >> PAGE_CACHE_SHIFT; + /* calculate number of pages which are to be written */ + end = ((*pos + count - 1) >> PAGE_CACHE_SHIFT); + nr_pages = end - index + 1; + nr_dirty = 0; + assert("", nr_pages <= WRITE_GRANULARITY + 1); + + /* get pages and jnodes */ + for (i = 0; i < nr_pages; i ++) { + page = find_or_create_page(inode->i_mapping, index + i, + reiser4_ctx_gfp_mask_get()); + if (page == NULL) { + nr_pages = i; + result = RETERR(-ENOMEM); + goto out; + } + + jnodes[i] = jnode_of_page(page); + if (IS_ERR(jnodes[i])) { + unlock_page(page); + page_cache_release(page); + nr_pages = i; + result = RETERR(-ENOMEM); + goto out; + } + /* prevent jnode and page from disconnecting */ + JF_SET(jnodes[i], JNODE_WRITE_PREPARED); + unlock_page(page); + } + + BUG_ON(get_current_context()->trans->atom != NULL); + + have_to_update_extent = 0; + + page_off = (*pos & (PAGE_CACHE_SIZE - 1)); + for (i = 0; i < nr_pages; i ++) { + to_page = PAGE_CACHE_SIZE - page_off; + if (to_page > left) + to_page = left; + page = jnode_page(jnodes[i]); + if (page_offset(page) < inode->i_size && + !PageUptodate(page) && to_page != PAGE_CACHE_SIZE) { + /* + * the above is not optimal for partial write to last + * page of file when file size is not at boundary of + * page + */ + lock_page(page); + if (!PageUptodate(page)) { + result = readpage_unix_file(NULL, page); + BUG_ON(result != 0); + /* wait for read completion */ + lock_page(page); + BUG_ON(!PageUptodate(page)); + } else + result = 0; + unlock_page(page); + } + + BUG_ON(get_current_context()->trans->atom != NULL); + fault_in_pages_readable(buf, to_page); + BUG_ON(get_current_context()->trans->atom != NULL); + + lock_page(page); + if (!PageUptodate(page) && to_page != PAGE_CACHE_SIZE) + zero_user_segments(page, 0, page_off, + page_off + to_page, + PAGE_CACHE_SIZE); + + written = filemap_copy_from_user(page, page_off, buf, to_page); + if (unlikely(written != to_page)) { + unlock_page(page); + result = RETERR(-EFAULT); + break; + } + + flush_dcache_page(page); + set_page_dirty_notag(page); + unlock_page(page); + nr_dirty++; + + mark_page_accessed(page); + SetPageUptodate(page); + + if (jnodes[i]->blocknr == 0) + have_to_update_extent ++; + + page_off = 0; + buf += to_page; + left -= to_page; + BUG_ON(get_current_context()->trans->atom != NULL); + } + + if (have_to_update_extent) { + update_extents(file, inode, jnodes, nr_dirty, *pos); + } else { + for (i = 0; i < nr_dirty; i ++) { + int ret; + spin_lock_jnode(jnodes[i]); + ret = reiser4_try_capture(jnodes[i], + ZNODE_WRITE_LOCK, 0); + BUG_ON(ret != 0); + jnode_make_dirty_locked(jnodes[i]); + spin_unlock_jnode(jnodes[i]); + } + } +out: + for (i = 0; i < nr_pages; i ++) { + page_cache_release(jnode_page(jnodes[i])); + JF_CLR(jnodes[i], JNODE_WRITE_PREPARED); + jput(jnodes[i]); + } + + /* the only errors handled so far is ENOMEM and + EFAULT on copy_from_user */ + + return (count - left) ? (count - left) : result; +} + +int reiser4_do_readpage_extent(reiser4_extent * ext, reiser4_block_nr pos, + struct page *page) +{ + jnode *j; + struct address_space *mapping; + unsigned long index; + oid_t oid; + reiser4_block_nr block; + + mapping = page->mapping; + oid = get_inode_oid(mapping->host); + index = page->index; + + switch (state_of_extent(ext)) { + case HOLE_EXTENT: + /* + * it is possible to have hole page with jnode, if page was + * eflushed previously. + */ + j = jfind(mapping, index); + if (j == NULL) { + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + unlock_page(page); + return 0; + } + spin_lock_jnode(j); + if (!jnode_page(j)) { + jnode_attach_page(j, page); + } else { + BUG_ON(jnode_page(j) != page); + assert("vs-1504", jnode_page(j) == page); + } + block = *jnode_get_io_block(j); + spin_unlock_jnode(j); + if (block == 0) { + zero_user(page, 0, PAGE_CACHE_SIZE); + SetPageUptodate(page); + unlock_page(page); + jput(j); + return 0; + } + break; + + case ALLOCATED_EXTENT: + j = jnode_of_page(page); + if (IS_ERR(j)) + return PTR_ERR(j); + if (*jnode_get_block(j) == 0) { + reiser4_block_nr blocknr; + + blocknr = extent_get_start(ext) + pos; + jnode_set_block(j, &blocknr); + } else + assert("vs-1403", + j->blocknr == extent_get_start(ext) + pos); + break; + + case UNALLOCATED_EXTENT: + j = jfind(mapping, index); + assert("nikita-2688", j); + assert("vs-1426", jnode_page(j) == NULL); + + spin_lock_jnode(j); + jnode_attach_page(j, page); + spin_unlock_jnode(j); + break; + + default: + warning("vs-957", "wrong extent\n"); + return RETERR(-EIO); + } + + BUG_ON(j == 0); + reiser4_page_io(page, j, READ, reiser4_ctx_gfp_mask_get()); + jput(j); + return 0; +} + +/* Implements plugin->u.item.s.file.read operation for extent items. */ +int reiser4_read_extent(struct file *file, flow_t *flow, hint_t *hint) +{ + int result; + struct page *page; + unsigned long cur_page, next_page; + unsigned long page_off, count; + struct address_space *mapping; + loff_t file_off; + uf_coord_t *uf_coord; + coord_t *coord; + struct extent_coord_extension *ext_coord; + unsigned long nr_pages; + char *kaddr; + + assert("vs-1353", current_blocksize == PAGE_CACHE_SIZE); + assert("vs-572", flow->user == 1); + assert("vs-1351", flow->length > 0); + + uf_coord = &hint->ext_coord; + + check_uf_coord(uf_coord, NULL); + assert("vs-33", uf_coord->lh == &hint->lh); + + coord = &uf_coord->coord; + assert("vs-1119", znode_is_rlocked(coord->node)); + assert("vs-1120", znode_is_loaded(coord->node)); + assert("vs-1256", coord_matches_key_extent(coord, &flow->key)); + + mapping = file->f_dentry->d_inode->i_mapping; + ext_coord = &uf_coord->extension.extent; + + /* offset in a file to start read from */ + file_off = get_key_offset(&flow->key); + /* offset within the page to start read from */ + page_off = (unsigned long)(file_off & (PAGE_CACHE_SIZE - 1)); + /* bytes which can be read from the page which contains file_off */ + count = PAGE_CACHE_SIZE - page_off; + + /* index of page containing offset read is to start from */ + cur_page = (unsigned long)(file_off >> PAGE_CACHE_SHIFT); + next_page = cur_page; + /* number of pages flow spans over */ + nr_pages = + ((file_off + flow->length + PAGE_CACHE_SIZE - + 1) >> PAGE_CACHE_SHIFT) - cur_page; + + /* we start having twig node read locked. However, we do not want to + keep that lock all the time readahead works. So, set a sel and + release twig node. */ + reiser4_set_hint(hint, &flow->key, ZNODE_READ_LOCK); + /* &hint->lh is done-ed */ + + do { + reiser4_txn_restart_current(); + page = read_mapping_page(mapping, cur_page, file); + if (IS_ERR(page)) + return PTR_ERR(page); + lock_page(page); + if (!PageUptodate(page)) { + unlock_page(page); + page_cache_release(page); + warning("jmacd-97178", "extent_read: page is not up to date"); + return RETERR(-EIO); + } + mark_page_accessed(page); + unlock_page(page); + + /* If users can be writing to this page using arbitrary virtual + addresses, take care about potential aliasing before reading + the page on the kernel side. + */ + if (mapping_writably_mapped(mapping)) + flush_dcache_page(page); + + assert("nikita-3034", reiser4_schedulable()); + + /* number of bytes which are to be read from the page */ + if (count > flow->length) + count = flow->length; + + result = fault_in_pages_writeable(flow->data, count); + if (result) { + page_cache_release(page); + return RETERR(-EFAULT); + } + + kaddr = kmap_atomic(page, KM_USER0); + result = __copy_to_user_inatomic(flow->data, + kaddr + page_off, count); + kunmap_atomic(kaddr, KM_USER0); + if (result != 0) { + kaddr = kmap(page); + result = __copy_to_user(flow->data, kaddr + page_off, count); + kunmap(page); + if (unlikely(result)) + return RETERR(-EFAULT); + } + + page_cache_release(page); + + /* increase key (flow->key), update user area pointer (flow->data) */ + move_flow_forward(flow, count); + + page_off = 0; + cur_page ++; + count = PAGE_CACHE_SIZE; + nr_pages--; + } while (flow->length); + + return 0; +} + +/* + plugin->s.file.readpage + reiser4_read->unix_file_read->page_cache_readahead->reiser4_readpage->unix_file_readpage->extent_readpage + or + filemap_fault->reiser4_readpage->readpage_unix_file->->readpage_extent + + At the beginning: coord->node is read locked, zloaded, page is + locked, coord is set to existing unit inside of extent item (it is not necessary that coord matches to page->index) +*/ +int reiser4_readpage_extent(void *vp, struct page *page) +{ + uf_coord_t *uf_coord = vp; + ON_DEBUG(coord_t * coord = &uf_coord->coord); + ON_DEBUG(reiser4_key key); + + assert("vs-1040", PageLocked(page)); + assert("vs-1050", !PageUptodate(page)); + assert("vs-1039", page->mapping && page->mapping->host); + + assert("vs-1044", znode_is_loaded(coord->node)); + assert("vs-758", item_is_extent(coord)); + assert("vs-1046", coord_is_existing_unit(coord)); + assert("vs-1045", znode_is_rlocked(coord->node)); + assert("vs-1047", + page->mapping->host->i_ino == + get_key_objectid(item_key_by_coord(coord, &key))); + check_uf_coord(uf_coord, NULL); + + return reiser4_do_readpage_extent( + ext_by_ext_coord(uf_coord), + uf_coord->extension.extent.pos_in_unit, page); +} + +/** + * get_block_address_extent + * @coord: + * @block: + * @result: + * + * + */ +int get_block_address_extent(const coord_t *coord, sector_t block, + sector_t *result) +{ + reiser4_extent *ext; + + if (!coord_is_existing_unit(coord)) + return RETERR(-EINVAL); + + ext = extent_by_coord(coord); + + if (state_of_extent(ext) != ALLOCATED_EXTENT) + /* FIXME: bad things may happen if it is unallocated extent */ + *result = 0; + else { + reiser4_key key; + + unit_key_by_coord(coord, &key); + assert("vs-1645", + block >= get_key_offset(&key) >> current_blocksize_bits); + assert("vs-1646", + block < + (get_key_offset(&key) >> current_blocksize_bits) + + extent_get_width(ext)); + *result = + extent_get_start(ext) + (block - + (get_key_offset(&key) >> + current_blocksize_bits)); + } + return 0; +} + +/* + plugin->u.item.s.file.append_key + key of first byte which is the next to last byte by addressed by this extent +*/ +reiser4_key *append_key_extent(const coord_t * coord, reiser4_key * key) +{ + item_key_by_coord(coord, key); + set_key_offset(key, + get_key_offset(key) + reiser4_extent_size(coord, + nr_units_extent + (coord))); + + assert("vs-610", get_key_offset(key) + && (get_key_offset(key) & (current_blocksize - 1)) == 0); + return key; +} + +/* plugin->u.item.s.file.init_coord_extension */ +void init_coord_extension_extent(uf_coord_t * uf_coord, loff_t lookuped) +{ + coord_t *coord; + struct extent_coord_extension *ext_coord; + reiser4_key key; + loff_t offset; + + assert("vs-1295", uf_coord->valid == 0); + + coord = &uf_coord->coord; + assert("vs-1288", coord_is_iplug_set(coord)); + assert("vs-1327", znode_is_loaded(coord->node)); + + if (coord->between != AFTER_UNIT && coord->between != AT_UNIT) + return; + + ext_coord = &uf_coord->extension.extent; + ext_coord->nr_units = nr_units_extent(coord); + ext_coord->ext_offset = + (char *)extent_by_coord(coord) - zdata(coord->node); + ext_coord->width = extent_get_width(extent_by_coord(coord)); + ON_DEBUG(ext_coord->extent = *extent_by_coord(coord)); + uf_coord->valid = 1; + + /* pos_in_unit is the only uninitialized field in extended coord */ + if (coord->between == AFTER_UNIT) { + assert("vs-1330", + coord->unit_pos == nr_units_extent(coord) - 1); + + ext_coord->pos_in_unit = ext_coord->width - 1; + } else { + /* AT_UNIT */ + unit_key_by_coord(coord, &key); + offset = get_key_offset(&key); + + assert("vs-1328", offset <= lookuped); + assert("vs-1329", + lookuped < + offset + ext_coord->width * current_blocksize); + ext_coord->pos_in_unit = + ((lookuped - offset) >> current_blocksize_bits); + } +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/extent_flush_ops.c linux-2.6.30/fs/reiser4/plugin/item/extent_flush_ops.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/extent_flush_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/extent_flush_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1028 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "item.h" +#include "../../tree.h" +#include "../../jnode.h" +#include "../../super.h" +#include "../../flush.h" +#include "../../carry.h" +#include "../object.h" + +#include + +static reiser4_block_nr extent_unit_start(const coord_t * item); + +/* Return either first or last extent (depending on @side) of the item + @coord is set to. Set @pos_in_unit either to first or to last block + of extent. */ +static reiser4_extent *extent_utmost_ext(const coord_t * coord, sideof side, + reiser4_block_nr * pos_in_unit) +{ + reiser4_extent *ext; + + if (side == LEFT_SIDE) { + /* get first extent of item */ + ext = extent_item(coord); + *pos_in_unit = 0; + } else { + /* get last extent of item and last position within it */ + assert("vs-363", side == RIGHT_SIDE); + ext = extent_item(coord) + coord_last_unit_pos(coord); + *pos_in_unit = extent_get_width(ext) - 1; + } + + return ext; +} + +/* item_plugin->f.utmost_child */ +/* Return the child. Coord is set to extent item. Find jnode corresponding + either to first or to last unformatted node pointed by the item */ +int utmost_child_extent(const coord_t * coord, sideof side, jnode ** childp) +{ + reiser4_extent *ext; + reiser4_block_nr pos_in_unit; + + ext = extent_utmost_ext(coord, side, &pos_in_unit); + + switch (state_of_extent(ext)) { + case HOLE_EXTENT: + *childp = NULL; + return 0; + case ALLOCATED_EXTENT: + case UNALLOCATED_EXTENT: + break; + default: + /* this should never happen */ + assert("vs-1417", 0); + } + + { + reiser4_key key; + reiser4_tree *tree; + unsigned long index; + + if (side == LEFT_SIDE) { + /* get key of first byte addressed by the extent */ + item_key_by_coord(coord, &key); + } else { + /* get key of byte which next after last byte addressed by the extent */ + append_key_extent(coord, &key); + } + + assert("vs-544", + (get_key_offset(&key) >> PAGE_CACHE_SHIFT) < ~0ul); + /* index of first or last (depending on @side) page addressed + by the extent */ + index = + (unsigned long)(get_key_offset(&key) >> PAGE_CACHE_SHIFT); + if (side == RIGHT_SIDE) + index--; + + tree = coord->node->zjnode.tree; + *childp = jlookup(tree, get_key_objectid(&key), index); + } + + return 0; +} + +/* item_plugin->f.utmost_child_real_block */ +/* Return the child's block, if allocated. */ +int +utmost_child_real_block_extent(const coord_t * coord, sideof side, + reiser4_block_nr * block) +{ + reiser4_extent *ext; + + ext = extent_by_coord(coord); + + switch (state_of_extent(ext)) { + case ALLOCATED_EXTENT: + *block = extent_get_start(ext); + if (side == RIGHT_SIDE) + *block += extent_get_width(ext) - 1; + break; + case HOLE_EXTENT: + case UNALLOCATED_EXTENT: + *block = 0; + break; + default: + /* this should never happen */ + assert("vs-1418", 0); + } + + return 0; +} + +/* item_plugin->f.scan */ +/* Performs leftward scanning starting from an unformatted node and its parent coordinate. + This scan continues, advancing the parent coordinate, until either it encounters a + formatted child or it finishes scanning this node. + + If unallocated, the entire extent must be dirty and in the same atom. (Actually, I'm + not sure this is last property (same atom) is enforced, but it should be the case since + one atom must write the parent and the others must read the parent, thus fusing?). In + any case, the code below asserts this case for unallocated extents. Unallocated + extents are thus optimized because we can skip to the endpoint when scanning. + + It returns control to reiser4_scan_extent, handles these terminating conditions, + e.g., by loading the next twig. +*/ +int reiser4_scan_extent(flush_scan * scan) +{ + coord_t coord; + jnode *neighbor; + unsigned long scan_index, unit_index, unit_width, scan_max, scan_dist; + reiser4_block_nr unit_start; + __u64 oid; + reiser4_key key; + int ret = 0, allocated, incr; + reiser4_tree *tree; + + if (!JF_ISSET(scan->node, JNODE_DIRTY)) { + scan->stop = 1; + return 0; /* Race with truncate, this node is already + * truncated. */ + } + + coord_dup(&coord, &scan->parent_coord); + + assert("jmacd-1404", !reiser4_scan_finished(scan)); + assert("jmacd-1405", jnode_get_level(scan->node) == LEAF_LEVEL); + assert("jmacd-1406", jnode_is_unformatted(scan->node)); + + /* The scan_index variable corresponds to the current page index of the + unformatted block scan position. */ + scan_index = index_jnode(scan->node); + + assert("jmacd-7889", item_is_extent(&coord)); + + repeat: + /* objectid of file */ + oid = get_key_objectid(item_key_by_coord(&coord, &key)); + + allocated = !extent_is_unallocated(&coord); + /* Get the values of this extent unit: */ + unit_index = extent_unit_index(&coord); + unit_width = extent_unit_width(&coord); + unit_start = extent_unit_start(&coord); + + assert("jmacd-7187", unit_width > 0); + assert("jmacd-7188", scan_index >= unit_index); + assert("jmacd-7189", scan_index <= unit_index + unit_width - 1); + + /* Depending on the scan direction, we set different maximum values for scan_index + (scan_max) and the number of nodes that would be passed if the scan goes the + entire way (scan_dist). Incr is an integer reflecting the incremental + direction of scan_index. */ + if (reiser4_scanning_left(scan)) { + scan_max = unit_index; + scan_dist = scan_index - unit_index; + incr = -1; + } else { + scan_max = unit_index + unit_width - 1; + scan_dist = scan_max - unit_index; + incr = +1; + } + + tree = coord.node->zjnode.tree; + + /* If the extent is allocated we have to check each of its blocks. If the extent + is unallocated we can skip to the scan_max. */ + if (allocated) { + do { + neighbor = jlookup(tree, oid, scan_index); + if (neighbor == NULL) + goto stop_same_parent; + + if (scan->node != neighbor + && !reiser4_scan_goto(scan, neighbor)) { + /* @neighbor was jput() by reiser4_scan_goto */ + goto stop_same_parent; + } + + ret = scan_set_current(scan, neighbor, 1, &coord); + if (ret != 0) { + goto exit; + } + + /* reference to @neighbor is stored in @scan, no need + to jput(). */ + scan_index += incr; + + } while (incr + scan_max != scan_index); + + } else { + /* Optimized case for unallocated extents, skip to the end. */ + neighbor = jlookup(tree, oid, scan_max /*index */ ); + if (neighbor == NULL) { + /* Race with truncate */ + scan->stop = 1; + ret = 0; + goto exit; + } + + assert("zam-1043", + reiser4_blocknr_is_fake(jnode_get_block(neighbor))); + + ret = scan_set_current(scan, neighbor, scan_dist, &coord); + if (ret != 0) { + goto exit; + } + } + + if (coord_sideof_unit(&coord, scan->direction) == 0 + && item_is_extent(&coord)) { + /* Continue as long as there are more extent units. */ + + scan_index = + extent_unit_index(&coord) + + (reiser4_scanning_left(scan) ? + extent_unit_width(&coord) - 1 : 0); + goto repeat; + } + + if (0) { + stop_same_parent: + + /* If we are scanning left and we stop in the middle of an allocated + extent, we know the preceder immediately.. */ + /* middle of extent is (scan_index - unit_index) != 0. */ + if (reiser4_scanning_left(scan) && + (scan_index - unit_index) != 0) { + /* FIXME(B): Someone should step-through and verify that this preceder + calculation is indeed correct. */ + /* @unit_start is starting block (number) of extent + unit. Flush stopped at the @scan_index block from + the beginning of the file, which is (scan_index - + unit_index) block within extent. + */ + if (unit_start) { + /* skip preceder update when we are at hole */ + scan->preceder_blk = + unit_start + scan_index - unit_index; + check_preceder(scan->preceder_blk); + } + } + + /* In this case, we leave coord set to the parent of scan->node. */ + scan->stop = 1; + + } else { + /* In this case, we are still scanning, coord is set to the next item which is + either off-the-end of the node or not an extent. */ + assert("jmacd-8912", scan->stop == 0); + assert("jmacd-7812", + (coord_is_after_sideof_unit(&coord, scan->direction) + || !item_is_extent(&coord))); + } + + ret = 0; + exit: + return ret; +} + +/* ask block allocator for some blocks */ +static void extent_allocate_blocks(reiser4_blocknr_hint *preceder, + reiser4_block_nr wanted_count, + reiser4_block_nr *first_allocated, + reiser4_block_nr *allocated, + block_stage_t block_stage) +{ + *allocated = wanted_count; + preceder->max_dist = 0; /* scan whole disk, if needed */ + + /* that number of blocks (wanted_count) is either in UNALLOCATED or in GRABBED */ + preceder->block_stage = block_stage; + + /* FIXME: we do not handle errors here now */ + check_me("vs-420", + reiser4_alloc_blocks(preceder, first_allocated, allocated, + BA_PERMANENT) == 0); + /* update flush_pos's preceder to last allocated block number */ + preceder->blk = *first_allocated + *allocated - 1; +} + +/* when on flush time unallocated extent is to be replaced with allocated one it may happen that one unallocated extent + will have to be replaced with set of allocated extents. In this case insert_into_item will be called which may have + to add new nodes into tree. Space for that is taken from inviolable reserve (5%). */ +static reiser4_block_nr reserve_replace(void) +{ + reiser4_block_nr grabbed, needed; + + grabbed = get_current_context()->grabbed_blocks; + needed = estimate_one_insert_into_item(current_tree); + check_me("vpf-340", !reiser4_grab_space_force(needed, BA_RESERVED)); + return grabbed; +} + +static void free_replace_reserved(reiser4_block_nr grabbed) +{ + reiser4_context *ctx; + + ctx = get_current_context(); + grabbed2free(ctx, get_super_private(ctx->super), + ctx->grabbed_blocks - grabbed); +} + +/* Block offset of first block addressed by unit */ +__u64 extent_unit_index(const coord_t * item) +{ + reiser4_key key; + + assert("vs-648", coord_is_existing_unit(item)); + unit_key_by_coord(item, &key); + return get_key_offset(&key) >> current_blocksize_bits; +} + +/* AUDIT shouldn't return value be of reiser4_block_nr type? + Josh's answer: who knows? Is a "number of blocks" the same type as "block offset"? */ +__u64 extent_unit_width(const coord_t * item) +{ + assert("vs-649", coord_is_existing_unit(item)); + return width_by_coord(item); +} + +/* Starting block location of this unit */ +static reiser4_block_nr extent_unit_start(const coord_t * item) +{ + return extent_get_start(extent_by_coord(item)); +} + +/** + * split_allocated_extent - + * @coord: + * @pos_in_unit: + * + * replace allocated extent with two allocated extents + */ +static int split_allocated_extent(coord_t *coord, reiser4_block_nr pos_in_unit) +{ + int result; + struct replace_handle *h; + reiser4_extent *ext; + reiser4_block_nr grabbed; + + ext = extent_by_coord(coord); + assert("vs-1410", state_of_extent(ext) == ALLOCATED_EXTENT); + assert("vs-1411", extent_get_width(ext) > pos_in_unit); + + h = kmalloc(sizeof(*h), reiser4_ctx_gfp_mask_get()); + if (h == NULL) + return RETERR(-ENOMEM); + h->coord = coord; + h->lh = znode_lh(coord->node); + h->pkey = &h->key; + unit_key_by_coord(coord, h->pkey); + set_key_offset(h->pkey, + (get_key_offset(h->pkey) + + pos_in_unit * current_blocksize)); + reiser4_set_extent(&h->overwrite, extent_get_start(ext), + pos_in_unit); + reiser4_set_extent(&h->new_extents[0], + extent_get_start(ext) + pos_in_unit, + extent_get_width(ext) - pos_in_unit); + h->nr_new_extents = 1; + h->flags = COPI_DONT_SHIFT_LEFT; + h->paste_key = h->key; + + /* reserve space for extent unit paste, @grabbed is reserved before */ + grabbed = reserve_replace(); + result = reiser4_replace_extent(h, 0 /* leave @coord set to overwritten + extent */); + /* restore reserved */ + free_replace_reserved(grabbed); + kfree(h); + return result; +} + +/* replace extent @ext by extent @replace. Try to merge @replace with previous extent of the item (if there is + one). Return 1 if it succeeded, 0 - otherwise */ +static int try_to_merge_with_left(coord_t *coord, reiser4_extent *ext, + reiser4_extent *replace) +{ + assert("vs-1415", extent_by_coord(coord) == ext); + + if (coord->unit_pos == 0 + || state_of_extent(ext - 1) != ALLOCATED_EXTENT) + /* @ext either does not exist or is not allocated extent */ + return 0; + if (extent_get_start(ext - 1) + extent_get_width(ext - 1) != + extent_get_start(replace)) + return 0; + + /* we can glue, widen previous unit */ + extent_set_width(ext - 1, + extent_get_width(ext - 1) + extent_get_width(replace)); + + if (extent_get_width(ext) != extent_get_width(replace)) { + /* make current extent narrower */ + if (state_of_extent(ext) == ALLOCATED_EXTENT) + extent_set_start(ext, + extent_get_start(ext) + + extent_get_width(replace)); + extent_set_width(ext, + extent_get_width(ext) - + extent_get_width(replace)); + } else { + /* current extent completely glued with its left neighbor, remove it */ + coord_t from, to; + + coord_dup(&from, coord); + from.unit_pos = nr_units_extent(coord) - 1; + coord_dup(&to, &from); + + /* currently cut from extent can cut either from the beginning or from the end. Move place which got + freed after unit removal to end of item */ + memmove(ext, ext + 1, + (from.unit_pos - + coord->unit_pos) * sizeof(reiser4_extent)); + /* wipe part of item which is going to be cut, so that node_check will not be confused */ + cut_node_content(&from, &to, NULL, NULL, NULL); + } + znode_make_dirty(coord->node); + /* move coord back */ + coord->unit_pos--; + return 1; +} + +/** + * conv_extent - replace extent with 2 ones + * @coord: coordinate of extent to be replaced + * @replace: extent to overwrite the one @coord is set to + * + * Overwrites extent @coord is set to and paste one extent unit after + * overwritten one if @replace is shorter than initial extent + */ +static int conv_extent(coord_t *coord, reiser4_extent *replace) +{ + int result; + struct replace_handle *h; + reiser4_extent *ext; + reiser4_block_nr start, width, new_width; + reiser4_block_nr grabbed; + extent_state state; + + ext = extent_by_coord(coord); + state = state_of_extent(ext); + start = extent_get_start(ext); + width = extent_get_width(ext); + new_width = extent_get_width(replace); + + assert("vs-1458", (state == UNALLOCATED_EXTENT || + state == ALLOCATED_EXTENT)); + assert("vs-1459", width >= new_width); + + if (try_to_merge_with_left(coord, ext, replace)) { + /* merged @replace with left neighbor. Current unit is either + removed or narrowed */ + return 0; + } + + if (width == new_width) { + /* replace current extent with @replace */ + *ext = *replace; + znode_make_dirty(coord->node); + return 0; + } + + h = kmalloc(sizeof(*h), reiser4_ctx_gfp_mask_get()); + if (h == NULL) + return RETERR(-ENOMEM); + h->coord = coord; + h->lh = znode_lh(coord->node); + h->pkey = &h->key; + unit_key_by_coord(coord, h->pkey); + set_key_offset(h->pkey, + (get_key_offset(h->pkey) + new_width * current_blocksize)); + h->overwrite = *replace; + + /* replace @ext with @replace and padding extent */ + reiser4_set_extent(&h->new_extents[0], + (state == ALLOCATED_EXTENT) ? + (start + new_width) : + UNALLOCATED_EXTENT_START, + width - new_width); + h->nr_new_extents = 1; + h->flags = COPI_DONT_SHIFT_LEFT; + h->paste_key = h->key; + + /* reserve space for extent unit paste, @grabbed is reserved before */ + grabbed = reserve_replace(); + result = reiser4_replace_extent(h, 0 /* leave @coord set to overwritten + extent */); + + /* restore reserved */ + free_replace_reserved(grabbed); + kfree(h); + return result; +} + +/** + * assign_real_blocknrs + * @flush_pos: + * @oid: objectid of file jnodes to assign block number to belongs to + * @index: first jnode on the range + * @count: number of jnodes to assign block numbers to + * @first: start of allocated block range + * + * Assigns block numbers to each of @count jnodes. Index of first jnode is + * @index. Jnodes get lookuped with jlookup. + */ +static void assign_real_blocknrs(flush_pos_t *flush_pos, oid_t oid, + unsigned long index, reiser4_block_nr count, + reiser4_block_nr first) +{ + unsigned long i; + reiser4_tree *tree; + txn_atom *atom; + int nr; + + atom = atom_locked_by_fq(flush_pos->fq); + assert("vs-1468", atom); + BUG_ON(atom == NULL); + + nr = 0; + tree = current_tree; + for (i = 0; i < count; ++i, ++index) { + jnode *node; + + node = jlookup(tree, oid, index); + assert("", node != NULL); + BUG_ON(node == NULL); + + spin_lock_jnode(node); + assert("", !jnode_is_flushprepped(node)); + assert("vs-1475", node->atom == atom); + assert("vs-1476", atomic_read(&node->x_count) > 0); + + JF_CLR(node, JNODE_FLUSH_RESERVED); + jnode_set_block(node, &first); + unformatted_make_reloc(node, flush_pos->fq); + ON_DEBUG(count_jnode(node->atom, node, NODE_LIST(node), + FQ_LIST, 0)); + spin_unlock_jnode(node); + first++; + + atomic_dec(&node->x_count); + nr ++; + } + + spin_unlock_atom(atom); + return; +} + +/** + * make_node_ovrwr - assign node to overwrite set + * @jnodes: overwrite set list head + * @node: jnode to belong to overwrite set + * + * Sets OVRWR jnode state bit and puts @node to the end of list head @jnodes + * which is an accumulator for nodes before they get to overwrite set list of + * atom. + */ +static void make_node_ovrwr(struct list_head *jnodes, jnode *node) +{ + spin_lock_jnode(node); + + assert("zam-917", !JF_ISSET(node, JNODE_RELOC)); + assert("zam-918", !JF_ISSET(node, JNODE_OVRWR)); + + JF_SET(node, JNODE_OVRWR); + list_move_tail(&node->capture_link, jnodes); + ON_DEBUG(count_jnode(node->atom, node, DIRTY_LIST, OVRWR_LIST, 0)); + + spin_unlock_jnode(node); +} + +/** + * mark_jnodes_overwrite - put bunch of jnodes to overwrite set + * @flush_pos: flush position + * @oid: objectid of file jnodes belong to + * @index: starting index + * @width: extent width + * + * Puts nodes of one extent (file objectid @oid, extent width @width) to atom's + * overwrite set. Starting from the one with index @index. If end of slum is + * detected (node is not found or flushprepped) - stop iterating and set flush + * position's state to POS_INVALID. + */ +static void mark_jnodes_overwrite(flush_pos_t *flush_pos, oid_t oid, + unsigned long index, reiser4_block_nr width) +{ + unsigned long i; + reiser4_tree *tree; + jnode *node; + txn_atom *atom; + LIST_HEAD(jnodes); + + tree = current_tree; + + atom = atom_locked_by_fq(reiser4_pos_fq(flush_pos)); + assert("vs-1478", atom); + + for (i = flush_pos->pos_in_unit; i < width; i++, index++) { + node = jlookup(tree, oid, index); + if (!node) { + flush_pos->state = POS_INVALID; + break; + } + if (jnode_check_flushprepped(node)) { + flush_pos->state = POS_INVALID; + atomic_dec(&node->x_count); + break; + } + if (node->atom != atom) { + flush_pos->state = POS_INVALID; + atomic_dec(&node->x_count); + break; + } + make_node_ovrwr(&jnodes, node); + atomic_dec(&node->x_count); + } + + list_splice_init(&jnodes, ATOM_OVRWR_LIST(atom)->prev); + spin_unlock_atom(atom); +} + +/** + * allocated_extent_slum_size + * @flush_pos: + * @oid: + * @index: + * @count: + * + * + */ +static int allocated_extent_slum_size(flush_pos_t *flush_pos, oid_t oid, + unsigned long index, unsigned long count) +{ + unsigned long i; + reiser4_tree *tree; + txn_atom *atom; + int nr; + + atom = atom_locked_by_fq(reiser4_pos_fq(flush_pos)); + assert("vs-1468", atom); + + nr = 0; + tree = current_tree; + for (i = 0; i < count; ++i, ++index) { + jnode *node; + + node = jlookup(tree, oid, index); + if (!node) + break; + + if (jnode_check_flushprepped(node)) { + atomic_dec(&node->x_count); + break; + } + + if (node->atom != atom) { + /* + * this is possible on overwrite: extent_write may + * capture several unformatted nodes without capturing + * any formatted nodes. + */ + atomic_dec(&node->x_count); + break; + } + + assert("vs-1476", atomic_read(&node->x_count) > 1); + atomic_dec(&node->x_count); + nr ++; + } + + spin_unlock_atom(atom); + return nr; +} + +/** + * alloc_extent + * @flush_pos: + * + * + * this is called by handle_pos_on_twig to proceed extent unit flush_pos->coord + * is set to. It is to prepare for flushing sequence of not flushprepped nodes + * (slum). It supposes that slum starts at flush_pos->pos_in_unit position + * within the extent. Slum gets to relocate set if flush_pos->leaf_relocate is + * set to 1 and to overwrite set otherwise + */ +int reiser4_alloc_extent(flush_pos_t *flush_pos) +{ + coord_t *coord; + reiser4_extent *ext; + reiser4_extent replace_ext; + oid_t oid; + reiser4_block_nr protected; + reiser4_block_nr start; + __u64 index; + __u64 width; + extent_state state; + int result; + reiser4_block_nr first_allocated; + __u64 allocated; + reiser4_key key; + block_stage_t block_stage; + + assert("vs-1468", flush_pos->state == POS_ON_EPOINT); + assert("vs-1469", coord_is_existing_unit(&flush_pos->coord) + && item_is_extent(&flush_pos->coord)); + + coord = &flush_pos->coord; + + ext = extent_by_coord(coord); + state = state_of_extent(ext); + if (state == HOLE_EXTENT) { + flush_pos->state = POS_INVALID; + return 0; + } + + item_key_by_coord(coord, &key); + oid = get_key_objectid(&key); + index = extent_unit_index(coord) + flush_pos->pos_in_unit; + start = extent_get_start(ext); + width = extent_get_width(ext); + + assert("vs-1457", width > flush_pos->pos_in_unit); + + if (flush_pos->leaf_relocate || state == UNALLOCATED_EXTENT) { + /* relocate */ + if (flush_pos->pos_in_unit) { + /* split extent unit into two */ + result = + split_allocated_extent(coord, + flush_pos->pos_in_unit); + flush_pos->pos_in_unit = 0; + return result; + } + + /* limit number of nodes to allocate */ + if (flush_pos->nr_to_write < width) + width = flush_pos->nr_to_write; + + if (state == ALLOCATED_EXTENT) { + /* + * all protected nodes are not flushprepped, therefore + * they are counted as flush_reserved + */ + block_stage = BLOCK_FLUSH_RESERVED; + protected = allocated_extent_slum_size(flush_pos, oid, + index, width); + if (protected == 0) { + flush_pos->state = POS_INVALID; + flush_pos->pos_in_unit = 0; + return 0; + } + } else { + block_stage = BLOCK_UNALLOCATED; + protected = width; + } + + /* + * look at previous unit if possible. If it is allocated, make + * preceder more precise + */ + if (coord->unit_pos && + (state_of_extent(ext - 1) == ALLOCATED_EXTENT)) + reiser4_pos_hint(flush_pos)->blk = + extent_get_start(ext - 1) + + extent_get_width(ext - 1); + + /* allocate new block numbers for protected nodes */ + extent_allocate_blocks(reiser4_pos_hint(flush_pos), + protected, + &first_allocated, &allocated, + block_stage); + + if (state == ALLOCATED_EXTENT) + /* + * on relocating - free nodes which are going to be + * relocated + */ + reiser4_dealloc_blocks(&start, &allocated, + BLOCK_ALLOCATED, BA_DEFER); + + /* assign new block numbers to protected nodes */ + assign_real_blocknrs(flush_pos, oid, index, allocated, first_allocated); + + /* prepare extent which will replace current one */ + reiser4_set_extent(&replace_ext, first_allocated, allocated); + + /* adjust extent item */ + result = conv_extent(coord, &replace_ext); + if (result != 0 && result != -ENOMEM) { + warning("vs-1461", + "Failed to allocate extent. Should not happen\n"); + return result; + } + + /* + * break flush: we prepared for flushing as many blocks as we + * were asked for + */ + if (flush_pos->nr_to_write == allocated) + flush_pos->state = POS_INVALID; + } else { + /* overwrite */ + mark_jnodes_overwrite(flush_pos, oid, index, width); + } + flush_pos->pos_in_unit = 0; + return 0; +} + +/* if @key is glueable to the item @coord is set to */ +static int must_insert(const coord_t *coord, const reiser4_key *key) +{ + reiser4_key last; + + if (item_id_by_coord(coord) == EXTENT_POINTER_ID + && keyeq(append_key_extent(coord, &last), key)) + return 0; + return 1; +} + +/* copy extent @copy to the end of @node. It may have to either insert new item after the last one, or append last item, + or modify last unit of last item to have greater width */ +static int put_unit_to_end(znode *node, const reiser4_key *key, + reiser4_extent *copy_ext) +{ + int result; + coord_t coord; + cop_insert_flag flags; + reiser4_extent *last_ext; + reiser4_item_data data; + + /* set coord after last unit in an item */ + coord_init_last_unit(&coord, node); + coord.between = AFTER_UNIT; + + flags = + COPI_DONT_SHIFT_LEFT | COPI_DONT_SHIFT_RIGHT | COPI_DONT_ALLOCATE; + if (must_insert(&coord, key)) { + result = + insert_by_coord(&coord, init_new_extent(&data, copy_ext, 1), + key, NULL /*lh */ , flags); + + } else { + /* try to glue with last unit */ + last_ext = extent_by_coord(&coord); + if (state_of_extent(last_ext) && + extent_get_start(last_ext) + extent_get_width(last_ext) == + extent_get_start(copy_ext)) { + /* widen last unit of node */ + extent_set_width(last_ext, + extent_get_width(last_ext) + + extent_get_width(copy_ext)); + znode_make_dirty(node); + return 0; + } + + /* FIXME: put an assertion here that we can not merge last unit in @node and new unit */ + result = + insert_into_item(&coord, NULL /*lh */ , key, + init_new_extent(&data, copy_ext, 1), + flags); + } + + assert("vs-438", result == 0 || result == -E_NODE_FULL); + return result; +} + +/* @coord is set to extent unit */ +squeeze_result squalloc_extent(znode *left, const coord_t *coord, + flush_pos_t *flush_pos, + reiser4_key *stop_key) +{ + reiser4_extent *ext; + __u64 index; + __u64 width; + reiser4_block_nr start; + extent_state state; + oid_t oid; + reiser4_block_nr first_allocated; + __u64 allocated; + __u64 protected; + reiser4_extent copy_extent; + reiser4_key key; + int result; + block_stage_t block_stage; + + assert("vs-1457", flush_pos->pos_in_unit == 0); + assert("vs-1467", coord_is_leftmost_unit(coord)); + assert("vs-1467", item_is_extent(coord)); + + ext = extent_by_coord(coord); + index = extent_unit_index(coord); + start = extent_get_start(ext); + width = extent_get_width(ext); + state = state_of_extent(ext); + unit_key_by_coord(coord, &key); + oid = get_key_objectid(&key); + + if ((flush_pos->leaf_relocate && state == ALLOCATED_EXTENT) || + (state == UNALLOCATED_EXTENT)) { + /* relocate */ + if (state == ALLOCATED_EXTENT) { + /* all protected nodes are not flushprepped, therefore + * they are counted as flush_reserved */ + block_stage = BLOCK_FLUSH_RESERVED; + protected = allocated_extent_slum_size(flush_pos, oid, + index, width); + if (protected == 0) { + flush_pos->state = POS_INVALID; + flush_pos->pos_in_unit = 0; + return 0; + } + } else { + block_stage = BLOCK_UNALLOCATED; + protected = width; + } + + /* + * look at previous unit if possible. If it is allocated, make + * preceder more precise + */ + if (coord->unit_pos && + (state_of_extent(ext - 1) == ALLOCATED_EXTENT)) + reiser4_pos_hint(flush_pos)->blk = + extent_get_start(ext - 1) + + extent_get_width(ext - 1); + + /* allocate new block numbers for protected nodes */ + extent_allocate_blocks(reiser4_pos_hint(flush_pos), + protected, + &first_allocated, &allocated, + block_stage); + + /* prepare extent which will be copied to left */ + reiser4_set_extent(©_extent, first_allocated, allocated); + + result = put_unit_to_end(left, &key, ©_extent); + if (result == -E_NODE_FULL) { + int target_block_stage; + + /* free blocks which were just allocated */ + target_block_stage = + (state == + ALLOCATED_EXTENT) ? BLOCK_FLUSH_RESERVED : + BLOCK_UNALLOCATED; + reiser4_dealloc_blocks(&first_allocated, &allocated, + target_block_stage, + BA_PERMANENT); + + /* rewind the preceder. */ + flush_pos->preceder.blk = first_allocated; + check_preceder(flush_pos->preceder.blk); + + return SQUEEZE_TARGET_FULL; + } + + if (state == ALLOCATED_EXTENT) { + /* free nodes which were relocated */ + reiser4_dealloc_blocks(&start, &allocated, + BLOCK_ALLOCATED, BA_DEFER); + } + + /* assign new block numbers to protected nodes */ + assign_real_blocknrs(flush_pos, oid, index, allocated, + first_allocated); + + set_key_offset(&key, + get_key_offset(&key) + + (allocated << current_blocksize_bits)); + } else { + /* + * overwrite: try to copy unit as it is to left neighbor and + * make all first not flushprepped nodes overwrite nodes + */ + reiser4_set_extent(©_extent, start, width); + result = put_unit_to_end(left, &key, ©_extent); + if (result == -E_NODE_FULL) + return SQUEEZE_TARGET_FULL; + + if (state != HOLE_EXTENT) + mark_jnodes_overwrite(flush_pos, oid, index, width); + set_key_offset(&key, + get_key_offset(&key) + + (width << current_blocksize_bits)); + } + *stop_key = key; + return SQUEEZE_CONTINUE; +} + +int key_by_offset_extent(struct inode *inode, loff_t off, reiser4_key * key) +{ + return key_by_inode_and_offset_common(inode, off, key); +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/extent.h linux-2.6.30/fs/reiser4/plugin/item/extent.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/extent.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/extent.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,231 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#ifndef __REISER4_EXTENT_H__ +#define __REISER4_EXTENT_H__ + +/* on disk extent */ +typedef struct { + reiser4_dblock_nr start; + reiser4_dblock_nr width; +} reiser4_extent; + +struct extent_stat { + int unallocated_units; + int unallocated_blocks; + int allocated_units; + int allocated_blocks; + int hole_units; + int hole_blocks; +}; + +/* extents in an extent item can be either holes, or unallocated or allocated + extents */ +typedef enum { + HOLE_EXTENT, + UNALLOCATED_EXTENT, + ALLOCATED_EXTENT +} extent_state; + +#define HOLE_EXTENT_START 0 +#define UNALLOCATED_EXTENT_START 1 +#define UNALLOCATED_EXTENT_START2 2 + +struct extent_coord_extension { + reiser4_block_nr pos_in_unit; + reiser4_block_nr width; /* width of current unit */ + pos_in_node_t nr_units; /* number of units */ + int ext_offset; /* offset from the beginning of zdata() */ + unsigned long expected_page; +#if REISER4_DEBUG + reiser4_extent extent; +#endif +}; + +/* macros to set/get fields of on-disk extent */ +static inline reiser4_block_nr extent_get_start(const reiser4_extent * ext) +{ + return le64_to_cpu(ext->start); +} + +static inline reiser4_block_nr extent_get_width(const reiser4_extent * ext) +{ + return le64_to_cpu(ext->width); +} + +extern __u64 reiser4_current_block_count(void); + +static inline void +extent_set_start(reiser4_extent * ext, reiser4_block_nr start) +{ + cassert(sizeof(ext->start) == 8); + assert("nikita-2510", + ergo(start > 1, start < reiser4_current_block_count())); + put_unaligned(cpu_to_le64(start), &ext->start); +} + +static inline void +extent_set_width(reiser4_extent * ext, reiser4_block_nr width) +{ + cassert(sizeof(ext->width) == 8); + assert("", width > 0); + put_unaligned(cpu_to_le64(width), &ext->width); + assert("nikita-2511", + ergo(extent_get_start(ext) > 1, + extent_get_start(ext) + width <= + reiser4_current_block_count())); +} + +#define extent_item(coord) \ +({ \ + assert("nikita-3143", item_is_extent(coord)); \ + ((reiser4_extent *)item_body_by_coord (coord)); \ +}) + +#define extent_by_coord(coord) \ +({ \ + assert("nikita-3144", item_is_extent(coord)); \ + (extent_item (coord) + (coord)->unit_pos); \ +}) + +#define width_by_coord(coord) \ +({ \ + assert("nikita-3145", item_is_extent(coord)); \ + extent_get_width (extent_by_coord(coord)); \ +}) + +struct carry_cut_data; +struct carry_kill_data; + +/* plugin->u.item.b.* */ +reiser4_key *max_key_inside_extent(const coord_t *, reiser4_key *); +int can_contain_key_extent(const coord_t * coord, const reiser4_key * key, + const reiser4_item_data *); +int mergeable_extent(const coord_t * p1, const coord_t * p2); +pos_in_node_t nr_units_extent(const coord_t *); +lookup_result lookup_extent(const reiser4_key *, lookup_bias, coord_t *); +void init_coord_extent(coord_t *); +int init_extent(coord_t *, reiser4_item_data *); +int paste_extent(coord_t *, reiser4_item_data *, carry_plugin_info *); +int can_shift_extent(unsigned free_space, + coord_t * source, znode * target, shift_direction, + unsigned *size, unsigned want); +void copy_units_extent(coord_t * target, coord_t * source, unsigned from, + unsigned count, shift_direction where_is_free_space, + unsigned free_space); +int kill_hook_extent(const coord_t *, pos_in_node_t from, pos_in_node_t count, + struct carry_kill_data *); +int create_hook_extent(const coord_t * coord, void *arg); +int cut_units_extent(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_cut_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +int kill_units_extent(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +reiser4_key *unit_key_extent(const coord_t *, reiser4_key *); +reiser4_key *max_unit_key_extent(const coord_t *, reiser4_key *); +void print_extent(const char *, coord_t *); +int utmost_child_extent(const coord_t * coord, sideof side, jnode ** child); +int utmost_child_real_block_extent(const coord_t * coord, sideof side, + reiser4_block_nr * block); +void item_stat_extent(const coord_t * coord, void *vp); +int reiser4_check_extent(const coord_t * coord, const char **error); + +/* plugin->u.item.s.file.* */ +ssize_t reiser4_write_extent(struct file *, struct inode * inode, + const char __user *, size_t, loff_t *); +int reiser4_read_extent(struct file *, flow_t *, hint_t *); +int reiser4_readpage_extent(void *, struct page *); +int reiser4_do_readpage_extent(reiser4_extent*, reiser4_block_nr, struct page*); +reiser4_key *append_key_extent(const coord_t *, reiser4_key *); +void init_coord_extension_extent(uf_coord_t *, loff_t offset); +int get_block_address_extent(const coord_t *, sector_t block, + sector_t * result); + +/* these are used in flush.c + FIXME-VS: should they be somewhere in item_plugin? */ +int allocate_extent_item_in_place(coord_t *, lock_handle *, flush_pos_t * pos); +int allocate_and_copy_extent(znode * left, coord_t * right, flush_pos_t * pos, + reiser4_key * stop_key); + +int extent_is_unallocated(const coord_t * item); /* True if this extent is unallocated (i.e., not a hole, not allocated). */ +__u64 extent_unit_index(const coord_t * item); /* Block offset of this unit. */ +__u64 extent_unit_width(const coord_t * item); /* Number of blocks in this unit. */ + +/* plugin->u.item.f. */ +int reiser4_scan_extent(flush_scan * scan); +extern int key_by_offset_extent(struct inode *, loff_t, reiser4_key *); + +reiser4_item_data *init_new_extent(reiser4_item_data * data, void *ext_unit, + int nr_extents); +reiser4_block_nr reiser4_extent_size(const coord_t * coord, pos_in_node_t nr); +extent_state state_of_extent(reiser4_extent * ext); +void reiser4_set_extent(reiser4_extent *, reiser4_block_nr start, + reiser4_block_nr width); +int reiser4_update_extent(struct inode *, jnode *, loff_t pos, + int *plugged_hole); + +#include "../../coord.h" +#include "../../lock.h" +#include "../../tap.h" + +struct replace_handle { + /* these are to be set before calling reiser4_replace_extent */ + coord_t *coord; + lock_handle *lh; + reiser4_key key; + reiser4_key *pkey; + reiser4_extent overwrite; + reiser4_extent new_extents[2]; + int nr_new_extents; + unsigned flags; + + /* these are used by reiser4_replace_extent */ + reiser4_item_data item; + coord_t coord_after; + lock_handle lh_after; + tap_t watch; + reiser4_key paste_key; +#if REISER4_DEBUG + reiser4_extent orig_ext; + reiser4_key tmp; +#endif +}; + +/* this structure is kmalloced before calling make_extent to avoid excessive + stack consumption on plug_hole->reiser4_replace_extent */ +struct make_extent_handle { + uf_coord_t *uf_coord; + reiser4_block_nr blocknr; + int created; + struct inode *inode; + union { + struct { + } append; + struct replace_handle replace; + } u; +}; + +int reiser4_replace_extent(struct replace_handle *, + int return_inserted_position); +lock_handle *znode_lh(znode *); + +/* the reiser4 repacker support */ +struct repacker_cursor; +extern int process_extent_backward_for_repacking(tap_t *, + struct repacker_cursor *); +extern int mark_extent_for_repacking(tap_t *, int); + +#define coord_by_uf_coord(uf_coord) (&((uf_coord)->coord)) +#define ext_coord_by_uf_coord(uf_coord) (&((uf_coord)->extension.extent)) + +/* __REISER4_EXTENT_H__ */ +#endif +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/extent_item_ops.c linux-2.6.30/fs/reiser4/plugin/item/extent_item_ops.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/extent_item_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/extent_item_ops.c 2009-06-22 17:27:31.000000000 +0200 @@ -0,0 +1,889 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "item.h" +#include "../../inode.h" +#include "../../tree_walk.h" /* check_sibling_list() */ +#include "../../page_cache.h" +#include "../../carry.h" + +#include + +/* item_plugin->b.max_key_inside */ +reiser4_key *max_key_inside_extent(const coord_t * coord, reiser4_key * key) +{ + item_key_by_coord(coord, key); + set_key_offset(key, get_key_offset(reiser4_max_key())); + return key; +} + +/* item_plugin->b.can_contain_key + this checks whether @key of @data is matching to position set by @coord */ +int +can_contain_key_extent(const coord_t * coord, const reiser4_key * key, + const reiser4_item_data * data) +{ + reiser4_key item_key; + + if (item_plugin_by_coord(coord) != data->iplug) + return 0; + + item_key_by_coord(coord, &item_key); + if (get_key_locality(key) != get_key_locality(&item_key) || + get_key_objectid(key) != get_key_objectid(&item_key) || + get_key_ordering(key) != get_key_ordering(&item_key)) + return 0; + + return 1; +} + +/* item_plugin->b.mergeable + first item is of extent type */ +/* Audited by: green(2002.06.13) */ +int mergeable_extent(const coord_t * p1, const coord_t * p2) +{ + reiser4_key key1, key2; + + assert("vs-299", item_id_by_coord(p1) == EXTENT_POINTER_ID); + /* FIXME-VS: Which is it? Assert or return 0 */ + if (item_id_by_coord(p2) != EXTENT_POINTER_ID) { + return 0; + } + + item_key_by_coord(p1, &key1); + item_key_by_coord(p2, &key2); + if (get_key_locality(&key1) != get_key_locality(&key2) || + get_key_objectid(&key1) != get_key_objectid(&key2) || + get_key_ordering(&key1) != get_key_ordering(&key2) || + get_key_type(&key1) != get_key_type(&key2)) + return 0; + if (get_key_offset(&key1) + + reiser4_extent_size(p1, nr_units_extent(p1)) != + get_key_offset(&key2)) + return 0; + return 1; +} + +/* item_plugin->b.nr_units */ +pos_in_node_t nr_units_extent(const coord_t * coord) +{ + /* length of extent item has to be multiple of extent size */ + assert("vs-1424", + (item_length_by_coord(coord) % sizeof(reiser4_extent)) == 0); + return item_length_by_coord(coord) / sizeof(reiser4_extent); +} + +/* item_plugin->b.lookup */ +lookup_result +lookup_extent(const reiser4_key * key, lookup_bias bias UNUSED_ARG, + coord_t * coord) +{ /* znode and item_pos are + set to an extent item to + look through */ + reiser4_key item_key; + reiser4_block_nr lookuped, offset; + unsigned i, nr_units; + reiser4_extent *ext; + unsigned blocksize; + unsigned char blocksize_bits; + + item_key_by_coord(coord, &item_key); + offset = get_key_offset(&item_key); + + /* key we are looking for must be greater than key of item @coord */ + assert("vs-414", keygt(key, &item_key)); + + assert("umka-99945", + !keygt(key, max_key_inside_extent(coord, &item_key))); + + ext = extent_item(coord); + assert("vs-1350", (char *)ext == (zdata(coord->node) + coord->offset)); + + blocksize = current_blocksize; + blocksize_bits = current_blocksize_bits; + + /* offset we are looking for */ + lookuped = get_key_offset(key); + + nr_units = nr_units_extent(coord); + /* go through all extents until the one which address given offset */ + for (i = 0; i < nr_units; i++, ext++) { + offset += (extent_get_width(ext) << blocksize_bits); + if (offset > lookuped) { + /* desired byte is somewhere in this extent */ + coord->unit_pos = i; + coord->between = AT_UNIT; + return CBK_COORD_FOUND; + } + } + + /* set coord after last unit */ + coord->unit_pos = nr_units - 1; + coord->between = AFTER_UNIT; + return CBK_COORD_FOUND; +} + +/* item_plugin->b.paste + item @coord is set to has been appended with @data->length of free + space. data->data contains data to be pasted into the item in position + @coord->in_item.unit_pos. It must fit into that free space. + @coord must be set between units. +*/ +int +paste_extent(coord_t * coord, reiser4_item_data * data, + carry_plugin_info * info UNUSED_ARG) +{ + unsigned old_nr_units; + reiser4_extent *ext; + int item_length; + + ext = extent_item(coord); + item_length = item_length_by_coord(coord); + old_nr_units = (item_length - data->length) / sizeof(reiser4_extent); + + /* this is also used to copy extent into newly created item, so + old_nr_units could be 0 */ + assert("vs-260", item_length >= data->length); + + /* make sure that coord is set properly */ + assert("vs-35", + ((!coord_is_existing_unit(coord)) + || (!old_nr_units && !coord->unit_pos))); + + /* first unit to be moved */ + switch (coord->between) { + case AFTER_UNIT: + coord->unit_pos++; + case BEFORE_UNIT: + coord->between = AT_UNIT; + break; + case AT_UNIT: + assert("vs-331", !old_nr_units && !coord->unit_pos); + break; + default: + impossible("vs-330", "coord is set improperly"); + } + + /* prepare space for new units */ + memmove(ext + coord->unit_pos + data->length / sizeof(reiser4_extent), + ext + coord->unit_pos, + (old_nr_units - coord->unit_pos) * sizeof(reiser4_extent)); + + /* copy new data from kernel space */ + assert("vs-556", data->user == 0); + memcpy(ext + coord->unit_pos, data->data, (unsigned)data->length); + + /* after paste @coord is set to first of pasted units */ + assert("vs-332", coord_is_existing_unit(coord)); + assert("vs-333", + !memcmp(data->data, extent_by_coord(coord), + (unsigned)data->length)); + return 0; +} + +/* item_plugin->b.can_shift */ +int +can_shift_extent(unsigned free_space, coord_t * source, + znode * target UNUSED_ARG, shift_direction pend UNUSED_ARG, + unsigned *size, unsigned want) +{ + *size = item_length_by_coord(source); + if (*size > free_space) + /* never split a unit of extent item */ + *size = free_space - free_space % sizeof(reiser4_extent); + + /* we can shift *size bytes, calculate how many do we want to shift */ + if (*size > want * sizeof(reiser4_extent)) + *size = want * sizeof(reiser4_extent); + + if (*size % sizeof(reiser4_extent) != 0) + impossible("vs-119", "Wrong extent size: %i %zd", *size, + sizeof(reiser4_extent)); + return *size / sizeof(reiser4_extent); + +} + +/* item_plugin->b.copy_units */ +void +copy_units_extent(coord_t * target, coord_t * source, + unsigned from, unsigned count, + shift_direction where_is_free_space, unsigned free_space) +{ + char *from_ext, *to_ext; + + assert("vs-217", free_space == count * sizeof(reiser4_extent)); + + from_ext = item_body_by_coord(source); + to_ext = item_body_by_coord(target); + + if (where_is_free_space == SHIFT_LEFT) { + assert("vs-215", from == 0); + + /* At this moment, item length was already updated in the item + header by shifting code, hence nr_units_extent() will + return "new" number of units---one we obtain after copying + units. + */ + to_ext += + (nr_units_extent(target) - count) * sizeof(reiser4_extent); + } else { + reiser4_key key; + coord_t coord; + + assert("vs-216", + from + count == coord_last_unit_pos(source) + 1); + + from_ext += item_length_by_coord(source) - free_space; + + /* new units are inserted before first unit in an item, + therefore, we have to update item key */ + coord = *source; + coord.unit_pos = from; + unit_key_extent(&coord, &key); + + node_plugin_by_node(target->node)->update_item_key(target, &key, + NULL /*info */); + } + + memcpy(to_ext, from_ext, free_space); +} + +/* item_plugin->b.create_hook + @arg is znode of leaf node for which we need to update right delimiting key */ +int create_hook_extent(const coord_t * coord, void *arg) +{ + coord_t *child_coord; + znode *node; + reiser4_key key; + reiser4_tree *tree; + + if (!arg) + return 0; + + child_coord = arg; + tree = znode_get_tree(coord->node); + + assert("nikita-3246", znode_get_level(child_coord->node) == LEAF_LEVEL); + + write_lock_tree(tree); + write_lock_dk(tree); + /* find a node on the left level for which right delimiting key has to + be updated */ + if (coord_wrt(child_coord) == COORD_ON_THE_LEFT) { + assert("vs-411", znode_is_left_connected(child_coord->node)); + node = child_coord->node->left; + } else { + assert("vs-412", coord_wrt(child_coord) == COORD_ON_THE_RIGHT); + node = child_coord->node; + assert("nikita-3314", node != NULL); + } + + if (node != NULL) { + znode_set_rd_key(node, item_key_by_coord(coord, &key)); + + assert("nikita-3282", check_sibling_list(node)); + /* break sibling links */ + if (ZF_ISSET(node, JNODE_RIGHT_CONNECTED) && node->right) { + ON_DEBUG(node->right->left_version = + atomic_inc_return(&delim_key_version); + node->right_version = + atomic_inc_return(&delim_key_version);); + + node->right->left = NULL; + node->right = NULL; + } + } + write_unlock_dk(tree); + write_unlock_tree(tree); + return 0; +} + +#define ITEM_TAIL_KILLED 0 +#define ITEM_HEAD_KILLED 1 +#define ITEM_KILLED 2 + +/* item_plugin->b.kill_hook + this is called when @count units starting from @from-th one are going to be removed + */ +int +kill_hook_extent(const coord_t * coord, pos_in_node_t from, pos_in_node_t count, + struct carry_kill_data *kdata) +{ + reiser4_extent *ext; + reiser4_block_nr start, length; + const reiser4_key *pfrom_key, *pto_key; + struct inode *inode; + reiser4_tree *tree; + pgoff_t from_off, to_off, offset, skip; + int retval; + + /* these are located in memory kmalloc-ed by kill_node_content */ + reiser4_key *min_item_key, *max_item_key, *from_key, *to_key, *key; + coord_t *dup, *next; + + assert("zam-811", znode_is_write_locked(coord->node)); + assert("nikita-3315", kdata != NULL); + assert("vs-34", kdata->buf != NULL); + + /* map structures to kdata->buf */ + min_item_key = (reiser4_key *) (kdata->buf); + max_item_key = min_item_key + 1; + from_key = max_item_key + 1; + to_key = from_key + 1; + key = to_key + 1; + dup = (coord_t *) (key + 1); + next = dup + 1; + + item_key_by_coord(coord, min_item_key); + max_item_key_by_coord(coord, max_item_key); + + if (kdata->params.from_key) { + pfrom_key = kdata->params.from_key; + pto_key = kdata->params.to_key; + } else { + assert("vs-1549", from == coord->unit_pos); + unit_key_by_coord(coord, from_key); + pfrom_key = from_key; + + coord_dup(dup, coord); + dup->unit_pos = from + count - 1; + max_unit_key_by_coord(dup, to_key); + pto_key = to_key; + } + + if (!keylt(pto_key, max_item_key)) { + if (!keygt(pfrom_key, min_item_key)) { + znode *left, *right; + + /* item is to be removed completely */ + assert("nikita-3316", kdata->left != NULL + && kdata->right != NULL); + + left = kdata->left->node; + right = kdata->right->node; + + tree = current_tree; + /* we have to do two things: + * + * 1. link left and right formatted neighbors of + * extent being removed, and + * + * 2. update their delimiting keys. + * + * atomicity of these operations is protected by + * taking dk-lock and tree-lock. + */ + /* if neighbors of item being removed are znodes - + * link them */ + write_lock_tree(tree); + write_lock_dk(tree); + link_left_and_right(left, right); + if (left) { + /* update right delimiting key of left + * neighbor of extent item */ + /*coord_t next; + reiser4_key key; */ + + coord_dup(next, coord); + + if (coord_next_item(next)) + *key = *znode_get_rd_key(coord->node); + else + item_key_by_coord(next, key); + znode_set_rd_key(left, key); + } + write_unlock_dk(tree); + write_unlock_tree(tree); + + from_off = + get_key_offset(min_item_key) >> PAGE_CACHE_SHIFT; + to_off = + (get_key_offset(max_item_key) + + 1) >> PAGE_CACHE_SHIFT; + retval = ITEM_KILLED; + } else { + /* tail of item is to be removed */ + from_off = + (get_key_offset(pfrom_key) + PAGE_CACHE_SIZE - + 1) >> PAGE_CACHE_SHIFT; + to_off = + (get_key_offset(max_item_key) + + 1) >> PAGE_CACHE_SHIFT; + retval = ITEM_TAIL_KILLED; + } + } else { + /* head of item is to be removed */ + assert("vs-1571", keyeq(pfrom_key, min_item_key)); + assert("vs-1572", + (get_key_offset(pfrom_key) & (PAGE_CACHE_SIZE - 1)) == + 0); + assert("vs-1573", + ((get_key_offset(pto_key) + 1) & (PAGE_CACHE_SIZE - + 1)) == 0); + + if (kdata->left->node) { + /* update right delimiting key of left neighbor of extent item */ + /*reiser4_key key; */ + + *key = *pto_key; + set_key_offset(key, get_key_offset(pto_key) + 1); + + write_lock_dk(current_tree); + znode_set_rd_key(kdata->left->node, key); + write_unlock_dk(current_tree); + } + + from_off = get_key_offset(pfrom_key) >> PAGE_CACHE_SHIFT; + to_off = (get_key_offset(pto_key) + 1) >> PAGE_CACHE_SHIFT; + retval = ITEM_HEAD_KILLED; + } + + inode = kdata->inode; + assert("vs-1545", inode != NULL); + if (inode != NULL) + /* take care of pages and jnodes corresponding to part of item being killed */ + reiser4_invalidate_pages(inode->i_mapping, from_off, + to_off - from_off, + kdata->params.truncate); + + ext = extent_item(coord) + from; + offset = + (get_key_offset(min_item_key) + + reiser4_extent_size(coord, from)) >> PAGE_CACHE_SHIFT; + + assert("vs-1551", from_off >= offset); + assert("vs-1552", from_off - offset <= extent_get_width(ext)); + skip = from_off - offset; + offset = from_off; + + while (offset < to_off) { + length = extent_get_width(ext) - skip; + if (state_of_extent(ext) == HOLE_EXTENT) { + skip = 0; + offset += length; + ext++; + continue; + } + + if (offset + length > to_off) { + length = to_off - offset; + } + + vfs_dq_free_block_nodirty(inode, length); + + if (state_of_extent(ext) == UNALLOCATED_EXTENT) { + /* some jnodes corresponding to this unallocated extent */ + fake_allocated2free(length, 0 /* unformatted */ ); + + skip = 0; + offset += length; + ext++; + continue; + } + + assert("vs-1218", state_of_extent(ext) == ALLOCATED_EXTENT); + + if (length != 0) { + start = extent_get_start(ext) + skip; + + /* BA_DEFER bit parameter is turned on because blocks which get freed are not safe to be freed + immediately */ + reiser4_dealloc_blocks(&start, &length, + 0 /* not used */ , + BA_DEFER + /* unformatted with defer */ ); + } + skip = 0; + offset += length; + ext++; + } + return retval; +} + +/* item_plugin->b.kill_units */ +int +kill_units_extent(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *kdata, reiser4_key * smallest_removed, + reiser4_key * new_first) +{ + reiser4_extent *ext; + reiser4_key item_key; + pos_in_node_t count; + reiser4_key from_key, to_key; + const reiser4_key *pfrom_key, *pto_key; + loff_t off; + int result; + + assert("vs-1541", + ((kdata->params.from_key == NULL && kdata->params.to_key == NULL) + || (kdata->params.from_key != NULL + && kdata->params.to_key != NULL))); + + if (kdata->params.from_key) { + pfrom_key = kdata->params.from_key; + pto_key = kdata->params.to_key; + } else { + coord_t dup; + + /* calculate key range of kill */ + assert("vs-1549", from == coord->unit_pos); + unit_key_by_coord(coord, &from_key); + pfrom_key = &from_key; + + coord_dup(&dup, coord); + dup.unit_pos = to; + max_unit_key_by_coord(&dup, &to_key); + pto_key = &to_key; + } + + item_key_by_coord(coord, &item_key); + +#if REISER4_DEBUG + { + reiser4_key max_item_key; + + max_item_key_by_coord(coord, &max_item_key); + + if (new_first) { + /* head of item is to be cut */ + assert("vs-1542", keyeq(pfrom_key, &item_key)); + assert("vs-1538", keylt(pto_key, &max_item_key)); + } else { + /* tail of item is to be cut */ + assert("vs-1540", keygt(pfrom_key, &item_key)); + assert("vs-1543", !keylt(pto_key, &max_item_key)); + } + } +#endif + + if (smallest_removed) + *smallest_removed = *pfrom_key; + + if (new_first) { + /* item head is cut. Item key will change. This new key is calculated here */ + assert("vs-1556", + (get_key_offset(pto_key) & (PAGE_CACHE_SIZE - 1)) == + (PAGE_CACHE_SIZE - 1)); + *new_first = *pto_key; + set_key_offset(new_first, get_key_offset(new_first) + 1); + } + + count = to - from + 1; + result = kill_hook_extent(coord, from, count, kdata); + if (result == ITEM_TAIL_KILLED) { + assert("vs-1553", + get_key_offset(pfrom_key) >= + get_key_offset(&item_key) + + reiser4_extent_size(coord, from)); + off = + get_key_offset(pfrom_key) - + (get_key_offset(&item_key) + + reiser4_extent_size(coord, from)); + if (off) { + /* unit @from is to be cut partially. Its width decreases */ + ext = extent_item(coord) + from; + extent_set_width(ext, + (off + PAGE_CACHE_SIZE - + 1) >> PAGE_CACHE_SHIFT); + count--; + } + } else { + __u64 max_to_offset; + __u64 rest; + + assert("vs-1575", result == ITEM_HEAD_KILLED); + assert("", from == 0); + assert("", + ((get_key_offset(pto_key) + 1) & (PAGE_CACHE_SIZE - + 1)) == 0); + assert("", + get_key_offset(pto_key) + 1 > + get_key_offset(&item_key) + + reiser4_extent_size(coord, to)); + max_to_offset = + get_key_offset(&item_key) + + reiser4_extent_size(coord, to + 1) - 1; + assert("", get_key_offset(pto_key) <= max_to_offset); + + rest = + (max_to_offset - + get_key_offset(pto_key)) >> PAGE_CACHE_SHIFT; + if (rest) { + /* unit @to is to be cut partially */ + ext = extent_item(coord) + to; + + assert("", extent_get_width(ext) > rest); + + if (state_of_extent(ext) == ALLOCATED_EXTENT) + extent_set_start(ext, + extent_get_start(ext) + + (extent_get_width(ext) - + rest)); + + extent_set_width(ext, rest); + count--; + } + } + return count * sizeof(reiser4_extent); +} + +/* item_plugin->b.cut_units + this is too similar to kill_units_extent */ +int +cut_units_extent(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_cut_data *cdata, reiser4_key * smallest_removed, + reiser4_key * new_first) +{ + reiser4_extent *ext; + reiser4_key item_key; + pos_in_node_t count; + reiser4_key from_key, to_key; + const reiser4_key *pfrom_key, *pto_key; + loff_t off; + + assert("vs-1541", + ((cdata->params.from_key == NULL && cdata->params.to_key == NULL) + || (cdata->params.from_key != NULL + && cdata->params.to_key != NULL))); + + if (cdata->params.from_key) { + pfrom_key = cdata->params.from_key; + pto_key = cdata->params.to_key; + } else { + coord_t dup; + + /* calculate key range of kill */ + coord_dup(&dup, coord); + dup.unit_pos = from; + unit_key_by_coord(&dup, &from_key); + + dup.unit_pos = to; + max_unit_key_by_coord(&dup, &to_key); + + pfrom_key = &from_key; + pto_key = &to_key; + } + + assert("vs-1555", + (get_key_offset(pfrom_key) & (PAGE_CACHE_SIZE - 1)) == 0); + assert("vs-1556", + (get_key_offset(pto_key) & (PAGE_CACHE_SIZE - 1)) == + (PAGE_CACHE_SIZE - 1)); + + item_key_by_coord(coord, &item_key); + +#if REISER4_DEBUG + { + reiser4_key max_item_key; + + assert("vs-1584", + get_key_locality(pfrom_key) == + get_key_locality(&item_key)); + assert("vs-1585", + get_key_type(pfrom_key) == get_key_type(&item_key)); + assert("vs-1586", + get_key_objectid(pfrom_key) == + get_key_objectid(&item_key)); + assert("vs-1587", + get_key_ordering(pfrom_key) == + get_key_ordering(&item_key)); + + max_item_key_by_coord(coord, &max_item_key); + + if (new_first != NULL) { + /* head of item is to be cut */ + assert("vs-1542", keyeq(pfrom_key, &item_key)); + assert("vs-1538", keylt(pto_key, &max_item_key)); + } else { + /* tail of item is to be cut */ + assert("vs-1540", keygt(pfrom_key, &item_key)); + assert("vs-1543", keyeq(pto_key, &max_item_key)); + } + } +#endif + + if (smallest_removed) + *smallest_removed = *pfrom_key; + + if (new_first) { + /* item head is cut. Item key will change. This new key is calculated here */ + *new_first = *pto_key; + set_key_offset(new_first, get_key_offset(new_first) + 1); + } + + count = to - from + 1; + + assert("vs-1553", + get_key_offset(pfrom_key) >= + get_key_offset(&item_key) + reiser4_extent_size(coord, from)); + off = + get_key_offset(pfrom_key) - (get_key_offset(&item_key) + + reiser4_extent_size(coord, from)); + if (off) { + /* tail of unit @from is to be cut partially. Its width decreases */ + assert("vs-1582", new_first == NULL); + ext = extent_item(coord) + from; + extent_set_width(ext, off >> PAGE_CACHE_SHIFT); + count--; + } + + assert("vs-1554", + get_key_offset(pto_key) <= + get_key_offset(&item_key) + + reiser4_extent_size(coord, to + 1) - 1); + off = + (get_key_offset(&item_key) + + reiser4_extent_size(coord, to + 1) - 1) - + get_key_offset(pto_key); + if (off) { + /* @to_key is smaller than max key of unit @to. Unit @to will not be removed. It gets start increased + and width decreased. */ + assert("vs-1583", (off & (PAGE_CACHE_SIZE - 1)) == 0); + ext = extent_item(coord) + to; + if (state_of_extent(ext) == ALLOCATED_EXTENT) + extent_set_start(ext, + extent_get_start(ext) + + (extent_get_width(ext) - + (off >> PAGE_CACHE_SHIFT))); + + extent_set_width(ext, (off >> PAGE_CACHE_SHIFT)); + count--; + } + return count * sizeof(reiser4_extent); +} + +/* item_plugin->b.unit_key */ +reiser4_key *unit_key_extent(const coord_t * coord, reiser4_key * key) +{ + assert("vs-300", coord_is_existing_unit(coord)); + + item_key_by_coord(coord, key); + set_key_offset(key, + (get_key_offset(key) + + reiser4_extent_size(coord, coord->unit_pos))); + + return key; +} + +/* item_plugin->b.max_unit_key */ +reiser4_key *max_unit_key_extent(const coord_t * coord, reiser4_key * key) +{ + assert("vs-300", coord_is_existing_unit(coord)); + + item_key_by_coord(coord, key); + set_key_offset(key, + (get_key_offset(key) + + reiser4_extent_size(coord, coord->unit_pos + 1) - 1)); + return key; +} + +/* item_plugin->b.estimate + item_plugin->b.item_data_by_flow */ + +#if REISER4_DEBUG + +/* item_plugin->b.check + used for debugging, every item should have here the most complete + possible check of the consistency of the item that the inventor can + construct +*/ +int reiser4_check_extent(const coord_t * coord /* coord of item to check */, + const char **error /* where to store error message */) +{ + reiser4_extent *ext, *first; + unsigned i, j; + reiser4_block_nr start, width, blk_cnt; + unsigned num_units; + reiser4_tree *tree; + oid_t oid; + reiser4_key key; + coord_t scan; + + assert("vs-933", REISER4_DEBUG); + + if (znode_get_level(coord->node) != TWIG_LEVEL) { + *error = "Extent on the wrong level"; + return -1; + } + if (item_length_by_coord(coord) % sizeof(reiser4_extent) != 0) { + *error = "Wrong item size"; + return -1; + } + ext = first = extent_item(coord); + blk_cnt = reiser4_block_count(reiser4_get_current_sb()); + num_units = coord_num_units(coord); + tree = znode_get_tree(coord->node); + item_key_by_coord(coord, &key); + oid = get_key_objectid(&key); + coord_dup(&scan, coord); + + for (i = 0; i < num_units; ++i, ++ext) { + __u64 index; + + scan.unit_pos = i; + index = extent_unit_index(&scan); + +#if 0 + /* check that all jnodes are present for the unallocated + * extent */ + if (state_of_extent(ext) == UNALLOCATED_EXTENT) { + for (j = 0; j < extent_get_width(ext); j++) { + jnode *node; + + node = jlookup(tree, oid, index + j); + if (node == NULL) { + print_coord("scan", &scan, 0); + *error = "Jnode missing"; + return -1; + } + jput(node); + } + } +#endif + + start = extent_get_start(ext); + if (start < 2) + continue; + /* extent is allocated one */ + width = extent_get_width(ext); + if (start >= blk_cnt) { + *error = "Start too large"; + return -1; + } + if (start + width > blk_cnt) { + *error = "End too large"; + return -1; + } + /* make sure that this extent does not overlap with other + allocated extents extents */ + for (j = 0; j < i; j++) { + if (state_of_extent(first + j) != ALLOCATED_EXTENT) + continue; + if (! + ((extent_get_start(ext) >= + extent_get_start(first + j) + + extent_get_width(first + j)) + || (extent_get_start(ext) + + extent_get_width(ext) <= + extent_get_start(first + j)))) { + *error = "Extent overlaps with others"; + return -1; + } + } + + } + + return 0; +} + +#endif /* REISER4_DEBUG */ + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/internal.c linux-2.6.30/fs/reiser4/plugin/item/internal.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/internal.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/internal.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,404 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Implementation of internal-item plugin methods. */ + +#include "../../forward.h" +#include "../../debug.h" +#include "../../dformat.h" +#include "../../key.h" +#include "../../coord.h" +#include "internal.h" +#include "item.h" +#include "../node/node.h" +#include "../plugin.h" +#include "../../jnode.h" +#include "../../znode.h" +#include "../../tree_walk.h" +#include "../../tree_mod.h" +#include "../../tree.h" +#include "../../super.h" +#include "../../block_alloc.h" + +/* see internal.h for explanation */ + +/* plugin->u.item.b.mergeable */ +int mergeable_internal(const coord_t * p1 UNUSED_ARG /* first item */ , + const coord_t * p2 UNUSED_ARG /* second item */ ) +{ + /* internal items are not mergeable */ + return 0; +} + +/* ->lookup() method for internal items */ +lookup_result lookup_internal(const reiser4_key * key /* key to look up */ , + lookup_bias bias UNUSED_ARG /* lookup bias */ , + coord_t * coord /* coord of item */ ) +{ + reiser4_key ukey; + + switch (keycmp(unit_key_by_coord(coord, &ukey), key)) { + default: + impossible("", "keycmp()?!"); + case LESS_THAN: + /* FIXME-VS: AFTER_ITEM used to be here. But with new coord + item plugin can not be taken using coord set this way */ + assert("vs-681", coord->unit_pos == 0); + coord->between = AFTER_UNIT; + case EQUAL_TO: + return CBK_COORD_FOUND; + case GREATER_THAN: + return CBK_COORD_NOTFOUND; + } +} + +/* return body of internal item at @coord */ +static internal_item_layout *internal_at(const coord_t * coord /* coord of + * item */ ) +{ + assert("nikita-607", coord != NULL); + assert("nikita-1650", + item_plugin_by_coord(coord) == + item_plugin_by_id(NODE_POINTER_ID)); + return (internal_item_layout *) item_body_by_coord(coord); +} + +void reiser4_update_internal(const coord_t * coord, + const reiser4_block_nr * blocknr) +{ + internal_item_layout *item = internal_at(coord); + assert("nikita-2959", reiser4_blocknr_is_sane(blocknr)); + + put_unaligned(cpu_to_le64(*blocknr), &item->pointer); +} + +/* return child block number stored in the internal item at @coord */ +static reiser4_block_nr pointer_at(const coord_t * coord /* coord of item */ ) +{ + assert("nikita-608", coord != NULL); + return le64_to_cpu(get_unaligned(&internal_at(coord)->pointer)); +} + +/* get znode pointed to by internal @item */ +static znode *znode_at(const coord_t * item /* coord of item */ , + znode * parent /* parent node */ ) +{ + return child_znode(item, parent, 1, 0); +} + +/* store pointer from internal item into "block". Implementation of + ->down_link() method */ +void down_link_internal(const coord_t * coord /* coord of item */ , + const reiser4_key * key UNUSED_ARG /* key to get + * pointer for */ , + reiser4_block_nr * block /* resulting block number */ ) +{ + ON_DEBUG(reiser4_key item_key); + + assert("nikita-609", coord != NULL); + assert("nikita-611", block != NULL); + assert("nikita-612", (key == NULL) || + /* twig horrors */ + (znode_get_level(coord->node) == TWIG_LEVEL) + || keyle(item_key_by_coord(coord, &item_key), key)); + + *block = pointer_at(coord); + assert("nikita-2960", reiser4_blocknr_is_sane(block)); +} + +/* Get the child's block number, or 0 if the block is unallocated. */ +int +utmost_child_real_block_internal(const coord_t * coord, sideof side UNUSED_ARG, + reiser4_block_nr * block) +{ + assert("jmacd-2059", coord != NULL); + + *block = pointer_at(coord); + assert("nikita-2961", reiser4_blocknr_is_sane(block)); + + if (reiser4_blocknr_is_fake(block)) { + *block = 0; + } + + return 0; +} + +/* Return the child. */ +int +utmost_child_internal(const coord_t * coord, sideof side UNUSED_ARG, + jnode ** childp) +{ + reiser4_block_nr block = pointer_at(coord); + znode *child; + + assert("jmacd-2059", childp != NULL); + assert("nikita-2962", reiser4_blocknr_is_sane(&block)); + + child = zlook(znode_get_tree(coord->node), &block); + + if (IS_ERR(child)) { + return PTR_ERR(child); + } + + *childp = ZJNODE(child); + + return 0; +} + +#if REISER4_DEBUG + +static void check_link(znode * left, znode * right) +{ + znode *scan; + + for (scan = left; scan != right; scan = scan->right) { + if (ZF_ISSET(scan, JNODE_RIP)) + break; + if (znode_is_right_connected(scan) && scan->right != NULL) { + if (ZF_ISSET(scan->right, JNODE_RIP)) + break; + assert("nikita-3285", + znode_is_left_connected(scan->right)); + assert("nikita-3265", + ergo(scan != left, + ZF_ISSET(scan, JNODE_HEARD_BANSHEE))); + assert("nikita-3284", scan->right->left == scan); + } else + break; + } +} + +int check__internal(const coord_t * coord, const char **error) +{ + reiser4_block_nr blk; + znode *child; + coord_t cpy; + + blk = pointer_at(coord); + if (!reiser4_blocknr_is_sane(&blk)) { + *error = "Invalid pointer"; + return -1; + } + coord_dup(&cpy, coord); + child = znode_at(&cpy, cpy.node); + if (child != NULL) { + znode *left_child; + znode *right_child; + + left_child = right_child = NULL; + + assert("nikita-3256", znode_invariant(child)); + if (coord_prev_item(&cpy) == 0 && item_is_internal(&cpy)) { + left_child = znode_at(&cpy, cpy.node); + if (left_child != NULL) { + read_lock_tree(znode_get_tree(child)); + check_link(left_child, child); + read_unlock_tree(znode_get_tree(child)); + zput(left_child); + } + } + coord_dup(&cpy, coord); + if (coord_next_item(&cpy) == 0 && item_is_internal(&cpy)) { + right_child = znode_at(&cpy, cpy.node); + if (right_child != NULL) { + read_lock_tree(znode_get_tree(child)); + check_link(child, right_child); + read_unlock_tree(znode_get_tree(child)); + zput(right_child); + } + } + zput(child); + } + return 0; +} + +#endif /* REISER4_DEBUG */ + +/* return true only if this item really points to "block" */ +/* Audited by: green(2002.06.14) */ +int has_pointer_to_internal(const coord_t * coord /* coord of item */ , + const reiser4_block_nr * block /* block number to + * check */ ) +{ + assert("nikita-613", coord != NULL); + assert("nikita-614", block != NULL); + + return pointer_at(coord) == *block; +} + +/* hook called by ->create_item() method of node plugin after new internal + item was just created. + + This is point where pointer to new node is inserted into tree. Initialize + parent pointer in child znode, insert child into sibling list and slum. + +*/ +int create_hook_internal(const coord_t * item /* coord of item */ , + void *arg /* child's left neighbor, if any */ ) +{ + znode *child; + __u64 child_ptr; + + assert("nikita-1252", item != NULL); + assert("nikita-1253", item->node != NULL); + assert("nikita-1181", znode_get_level(item->node) > LEAF_LEVEL); + assert("nikita-1450", item->unit_pos == 0); + + /* + * preparing to item insertion build_child_ptr_data sets pointer to + * data to be inserted to jnode's blocknr which is in cpu byte + * order. Node's create_item simply copied those data. As result we + * have child pointer in cpu's byte order. Convert content of internal + * item to little endian byte order. + */ + child_ptr = get_unaligned((__u64 *)item_body_by_coord(item)); + reiser4_update_internal(item, &child_ptr); + + child = znode_at(item, item->node); + if (child != NULL && !IS_ERR(child)) { + znode *left; + int result = 0; + reiser4_tree *tree; + + left = arg; + tree = znode_get_tree(item->node); + write_lock_tree(tree); + write_lock_dk(tree); + assert("nikita-1400", (child->in_parent.node == NULL) + || (znode_above_root(child->in_parent.node))); + ++item->node->c_count; + coord_to_parent_coord(item, &child->in_parent); + sibling_list_insert_nolock(child, left); + + assert("nikita-3297", ZF_ISSET(child, JNODE_ORPHAN)); + ZF_CLR(child, JNODE_ORPHAN); + + if ((left != NULL) && !keyeq(znode_get_rd_key(left), + znode_get_rd_key(child))) { + znode_set_rd_key(child, znode_get_rd_key(left)); + } + write_unlock_dk(tree); + write_unlock_tree(tree); + zput(child); + return result; + } else { + if (child == NULL) + child = ERR_PTR(-EIO); + return PTR_ERR(child); + } +} + +/* hook called by ->cut_and_kill() method of node plugin just before internal + item is removed. + + This is point where empty node is removed from the tree. Clear parent + pointer in child, and mark node for pending deletion. + + Node will be actually deleted later and in several installations: + + . when last lock on this node will be released, node will be removed from + the sibling list and its lock will be invalidated + + . when last reference to this node will be dropped, bitmap will be updated + and node will be actually removed from the memory. + +*/ +int kill_hook_internal(const coord_t * item /* coord of item */ , + pos_in_node_t from UNUSED_ARG /* start unit */ , + pos_in_node_t count UNUSED_ARG /* stop unit */ , + struct carry_kill_data *p UNUSED_ARG) +{ + znode *child; + int result = 0; + + assert("nikita-1222", item != NULL); + assert("nikita-1224", from == 0); + assert("nikita-1225", count == 1); + + child = znode_at(item, item->node); + if (child == NULL) + return 0; + if (IS_ERR(child)) + return PTR_ERR(child); + result = zload(child); + if (result) { + zput(child); + return result; + } + if (node_is_empty(child)) { + reiser4_tree *tree; + + assert("nikita-1397", znode_is_write_locked(child)); + assert("nikita-1398", child->c_count == 0); + assert("nikita-2546", ZF_ISSET(child, JNODE_HEARD_BANSHEE)); + + tree = znode_get_tree(item->node); + write_lock_tree(tree); + init_parent_coord(&child->in_parent, NULL); + --item->node->c_count; + write_unlock_tree(tree); + } else { + warning("nikita-1223", + "Cowardly refuse to remove link to non-empty node"); + result = RETERR(-EIO); + } + zrelse(child); + zput(child); + return result; +} + +/* hook called by ->shift() node plugin method when iternal item was just + moved from one node to another. + + Update parent pointer in child and c_counts in old and new parent + +*/ +int shift_hook_internal(const coord_t * item /* coord of item */ , + unsigned from UNUSED_ARG /* start unit */ , + unsigned count UNUSED_ARG /* stop unit */ , + znode * old_node /* old parent */ ) +{ + znode *child; + znode *new_node; + reiser4_tree *tree; + + assert("nikita-1276", item != NULL); + assert("nikita-1277", from == 0); + assert("nikita-1278", count == 1); + assert("nikita-1451", item->unit_pos == 0); + + new_node = item->node; + assert("nikita-2132", new_node != old_node); + tree = znode_get_tree(item->node); + child = child_znode(item, old_node, 1, 0); + if (child == NULL) + return 0; + if (!IS_ERR(child)) { + write_lock_tree(tree); + ++new_node->c_count; + assert("nikita-1395", znode_parent(child) == old_node); + assert("nikita-1396", old_node->c_count > 0); + coord_to_parent_coord(item, &child->in_parent); + assert("nikita-1781", znode_parent(child) == new_node); + assert("nikita-1782", + check_tree_pointer(item, child) == NS_FOUND); + --old_node->c_count; + write_unlock_tree(tree); + zput(child); + return 0; + } else + return PTR_ERR(child); +} + +/* plugin->u.item.b.max_key_inside - not defined */ + +/* plugin->u.item.b.nr_units - item.c:single_unit */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/internal.h linux-2.6.30/fs/reiser4/plugin/item/internal.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/internal.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/internal.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,57 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* Internal item contains down-link to the child of the internal/twig + node in a tree. It is internal items that are actually used during + tree traversal. */ + +#if !defined( __FS_REISER4_PLUGIN_ITEM_INTERNAL_H__ ) +#define __FS_REISER4_PLUGIN_ITEM_INTERNAL_H__ + +#include "../../forward.h" +#include "../../dformat.h" + +/* on-disk layout of internal item */ +typedef struct internal_item_layout { + /* 0 */ reiser4_dblock_nr pointer; + /* 4 */ +} internal_item_layout; + +struct cut_list; + +int mergeable_internal(const coord_t * p1, const coord_t * p2); +lookup_result lookup_internal(const reiser4_key * key, lookup_bias bias, + coord_t * coord); +/* store pointer from internal item into "block". Implementation of + ->down_link() method */ +extern void down_link_internal(const coord_t * coord, const reiser4_key * key, + reiser4_block_nr * block); +extern int has_pointer_to_internal(const coord_t * coord, + const reiser4_block_nr * block); +extern int create_hook_internal(const coord_t * item, void *arg); +extern int kill_hook_internal(const coord_t * item, pos_in_node_t from, + pos_in_node_t count, struct carry_kill_data *); +extern int shift_hook_internal(const coord_t * item, unsigned from, + unsigned count, znode * old_node); +extern void reiser4_print_internal(const char *prefix, coord_t * coord); + +extern int utmost_child_internal(const coord_t * coord, sideof side, + jnode ** child); +int utmost_child_real_block_internal(const coord_t * coord, sideof side, + reiser4_block_nr * block); + +extern void reiser4_update_internal(const coord_t * coord, + const reiser4_block_nr * blocknr); +/* FIXME: reiserfs has check_internal */ +extern int check__internal(const coord_t * coord, const char **error); + +/* __FS_REISER4_PLUGIN_ITEM_INTERNAL_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/item.c linux-2.6.30/fs/reiser4/plugin/item/item.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/item.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/item.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,719 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* definition of item plugins. */ + +#include "../../forward.h" +#include "../../debug.h" +#include "../../key.h" +#include "../../coord.h" +#include "../plugin_header.h" +#include "sde.h" +#include "internal.h" +#include "item.h" +#include "static_stat.h" +#include "../plugin.h" +#include "../../znode.h" +#include "../../tree.h" +#include "../../context.h" +#include "ctail.h" + +/* return pointer to item body */ +void item_body_by_coord_hard(coord_t * coord /* coord to query */ ) +{ + assert("nikita-324", coord != NULL); + assert("nikita-325", coord->node != NULL); + assert("nikita-326", znode_is_loaded(coord->node)); + assert("nikita-3200", coord->offset == INVALID_OFFSET); + + coord->offset = + node_plugin_by_node(coord->node)->item_by_coord(coord) - + zdata(coord->node); + ON_DEBUG(coord->body_v = coord->node->times_locked); +} + +void *item_body_by_coord_easy(const coord_t * coord /* coord to query */ ) +{ + return zdata(coord->node) + coord->offset; +} + +#if REISER4_DEBUG + +int item_body_is_valid(const coord_t * coord) +{ + return + coord->offset == + node_plugin_by_node(coord->node)->item_by_coord(coord) - + zdata(coord->node); +} + +#endif + +/* return length of item at @coord */ +pos_in_node_t item_length_by_coord(const coord_t * coord /* coord to query */ ) +{ + int len; + + assert("nikita-327", coord != NULL); + assert("nikita-328", coord->node != NULL); + assert("nikita-329", znode_is_loaded(coord->node)); + + len = node_plugin_by_node(coord->node)->length_by_coord(coord); + return len; +} + +void obtain_item_plugin(const coord_t * coord) +{ + assert("nikita-330", coord != NULL); + assert("nikita-331", coord->node != NULL); + assert("nikita-332", znode_is_loaded(coord->node)); + + coord_set_iplug((coord_t *) coord, + node_plugin_by_node(coord->node)-> + plugin_by_coord(coord)); + assert("nikita-2479", + coord_iplug(coord) == + node_plugin_by_node(coord->node)->plugin_by_coord(coord)); +} + +/* return id of item */ +/* Audited by: green(2002.06.15) */ +item_id item_id_by_coord(const coord_t * coord /* coord to query */ ) +{ + assert("vs-539", coord != NULL); + assert("vs-538", coord->node != NULL); + assert("vs-537", znode_is_loaded(coord->node)); + assert("vs-536", item_plugin_by_coord(coord) != NULL); + assert("vs-540", + item_id_by_plugin(item_plugin_by_coord(coord)) < LAST_ITEM_ID); + + return item_id_by_plugin(item_plugin_by_coord(coord)); +} + +/* return key of item at @coord */ +/* Audited by: green(2002.06.15) */ +reiser4_key *item_key_by_coord(const coord_t * coord /* coord to query */ , + reiser4_key * key /* result */ ) +{ + assert("nikita-338", coord != NULL); + assert("nikita-339", coord->node != NULL); + assert("nikita-340", znode_is_loaded(coord->node)); + + return node_plugin_by_node(coord->node)->key_at(coord, key); +} + +/* this returns max key in the item */ +reiser4_key *max_item_key_by_coord(const coord_t * coord /* coord to query */ , + reiser4_key * key /* result */ ) +{ + coord_t last; + + assert("nikita-338", coord != NULL); + assert("nikita-339", coord->node != NULL); + assert("nikita-340", znode_is_loaded(coord->node)); + + /* make coord pointing to last item's unit */ + coord_dup(&last, coord); + last.unit_pos = coord_num_units(&last) - 1; + assert("vs-1560", coord_is_existing_unit(&last)); + + max_unit_key_by_coord(&last, key); + return key; +} + +/* return key of unit at @coord */ +reiser4_key *unit_key_by_coord(const coord_t * coord /* coord to query */ , + reiser4_key * key /* result */ ) +{ + assert("nikita-772", coord != NULL); + assert("nikita-774", coord->node != NULL); + assert("nikita-775", znode_is_loaded(coord->node)); + + if (item_plugin_by_coord(coord)->b.unit_key != NULL) + return item_plugin_by_coord(coord)->b.unit_key(coord, key); + else + return item_key_by_coord(coord, key); +} + +/* return the biggest key contained the unit @coord */ +reiser4_key *max_unit_key_by_coord(const coord_t * coord /* coord to query */ , + reiser4_key * key /* result */ ) +{ + assert("nikita-772", coord != NULL); + assert("nikita-774", coord->node != NULL); + assert("nikita-775", znode_is_loaded(coord->node)); + + if (item_plugin_by_coord(coord)->b.max_unit_key != NULL) + return item_plugin_by_coord(coord)->b.max_unit_key(coord, key); + else + return unit_key_by_coord(coord, key); +} + +/* ->max_key_inside() method for items consisting of exactly one key (like + stat-data) */ +static reiser4_key *max_key_inside_single_key(const coord_t * + coord /* coord of item */ , + reiser4_key * + result /* resulting key */ ) +{ + assert("nikita-604", coord != NULL); + + /* coord -> key is starting key of this item and it has to be already + filled in */ + return unit_key_by_coord(coord, result); +} + +/* ->nr_units() method for items consisting of exactly one unit always */ +pos_in_node_t +nr_units_single_unit(const coord_t * coord UNUSED_ARG /* coord of item */ ) +{ + return 1; +} + +static int +paste_no_paste(coord_t * coord UNUSED_ARG, + reiser4_item_data * data UNUSED_ARG, + carry_plugin_info * info UNUSED_ARG) +{ + return 0; +} + +/* default ->fast_paste() method */ +static int +agree_to_fast_op(const coord_t * coord UNUSED_ARG /* coord of item */ ) +{ + return 1; +} + +int item_can_contain_key(const coord_t * item /* coord of item */ , + const reiser4_key * key /* key to check */ , + const reiser4_item_data * data /* parameters of item + * being created */ ) +{ + item_plugin *iplug; + reiser4_key min_key_in_item; + reiser4_key max_key_in_item; + + assert("nikita-1658", item != NULL); + assert("nikita-1659", key != NULL); + + iplug = item_plugin_by_coord(item); + if (iplug->b.can_contain_key != NULL) + return iplug->b.can_contain_key(item, key, data); + else { + assert("nikita-1681", iplug->b.max_key_inside != NULL); + item_key_by_coord(item, &min_key_in_item); + iplug->b.max_key_inside(item, &max_key_in_item); + + /* can contain key if + min_key_in_item <= key && + key <= max_key_in_item + */ + return keyle(&min_key_in_item, key) + && keyle(key, &max_key_in_item); + } +} + +/* mergeable method for non mergeable items */ +static int +not_mergeable(const coord_t * i1 UNUSED_ARG, const coord_t * i2 UNUSED_ARG) +{ + return 0; +} + +/* return 0 if @item1 and @item2 are not mergeable, !0 - otherwise */ +int are_items_mergeable(const coord_t * i1 /* coord of first item */ , + const coord_t * i2 /* coord of second item */ ) +{ + item_plugin *iplug; + reiser4_key k1; + reiser4_key k2; + + assert("nikita-1336", i1 != NULL); + assert("nikita-1337", i2 != NULL); + + iplug = item_plugin_by_coord(i1); + assert("nikita-1338", iplug != NULL); + + /* NOTE-NIKITA are_items_mergeable() is also called by assertions in + shifting code when nodes are in "suspended" state. */ + assert("nikita-1663", + keyle(item_key_by_coord(i1, &k1), item_key_by_coord(i2, &k2))); + + if (iplug->b.mergeable != NULL) { + return iplug->b.mergeable(i1, i2); + } else if (iplug->b.max_key_inside != NULL) { + iplug->b.max_key_inside(i1, &k1); + item_key_by_coord(i2, &k2); + + /* mergeable if ->max_key_inside() >= key of i2; */ + return keyge(iplug->b.max_key_inside(i1, &k1), + item_key_by_coord(i2, &k2)); + } else { + item_key_by_coord(i1, &k1); + item_key_by_coord(i2, &k2); + + return + (get_key_locality(&k1) == get_key_locality(&k2)) && + (get_key_objectid(&k1) == get_key_objectid(&k2)) + && (iplug == item_plugin_by_coord(i2)); + } +} + +int item_is_extent(const coord_t * item) +{ + assert("vs-482", coord_is_existing_item(item)); + return item_id_by_coord(item) == EXTENT_POINTER_ID; +} + +int item_is_tail(const coord_t * item) +{ + assert("vs-482", coord_is_existing_item(item)); + return item_id_by_coord(item) == FORMATTING_ID; +} + +#if REISER4_DEBUG + +int item_is_statdata(const coord_t * item) +{ + assert("vs-516", coord_is_existing_item(item)); + return plugin_of_group(item_plugin_by_coord(item), STAT_DATA_ITEM_TYPE); +} + +int item_is_ctail(const coord_t * item) +{ + assert("edward-xx", coord_is_existing_item(item)); + return item_id_by_coord(item) == CTAIL_ID; +} + +#endif /* REISER4_DEBUG */ + +static int change_item(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + /* cannot change constituent item (sd, or dir_item) */ + return RETERR(-EINVAL); +} + +static reiser4_plugin_ops item_plugin_ops = { + .init = NULL, + .load = NULL, + .save_len = NULL, + .save = NULL, + .change = change_item +}; + +item_plugin item_plugins[LAST_ITEM_ID] = { + [STATIC_STAT_DATA_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = STATIC_STAT_DATA_ID, + .groups = (1 << STAT_DATA_ITEM_TYPE), + .pops = &item_plugin_ops, + .label = "sd", + .desc = "stat-data", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = max_key_inside_single_key, + .can_contain_key = NULL, + .mergeable = not_mergeable, + .nr_units = nr_units_single_unit, + .lookup = NULL, + .init = NULL, + .paste = paste_no_paste, + .fast_paste = NULL, + .can_shift = NULL, + .copy_units = NULL, + .create_hook = NULL, + .kill_hook = NULL, + .shift_hook = NULL, + .cut_units = NULL, + .kill_units = NULL, + .unit_key = NULL, + .max_unit_key = NULL, + .estimate = NULL, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = NULL +#endif + }, + .f = { + .utmost_child = NULL, + .utmost_child_real_block = NULL, + .update = NULL, + .scan = NULL, + .convert = NULL + }, + .s = { + .sd = { + .init_inode = init_inode_static_sd, + .save_len = save_len_static_sd, + .save = save_static_sd + } + } + }, + [SIMPLE_DIR_ENTRY_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = SIMPLE_DIR_ENTRY_ID, + .groups = (1 << DIR_ENTRY_ITEM_TYPE), + .pops = &item_plugin_ops, + .label = "de", + .desc = "directory entry", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = max_key_inside_single_key, + .can_contain_key = NULL, + .mergeable = NULL, + .nr_units = nr_units_single_unit, + .lookup = NULL, + .init = NULL, + .paste = NULL, + .fast_paste = NULL, + .can_shift = NULL, + .copy_units = NULL, + .create_hook = NULL, + .kill_hook = NULL, + .shift_hook = NULL, + .cut_units = NULL, + .kill_units = NULL, + .unit_key = NULL, + .max_unit_key = NULL, + .estimate = NULL, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = NULL +#endif + }, + .f = { + .utmost_child = NULL, + .utmost_child_real_block = NULL, + .update = NULL, + .scan = NULL, + .convert = NULL + }, + .s = { + .dir = { + .extract_key = extract_key_de, + .update_key = update_key_de, + .extract_name = extract_name_de, + .extract_file_type = extract_file_type_de, + .add_entry = add_entry_de, + .rem_entry = rem_entry_de, + .max_name_len = max_name_len_de + } + } + }, + [COMPOUND_DIR_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = COMPOUND_DIR_ID, + .groups = (1 << DIR_ENTRY_ITEM_TYPE), + .pops = &item_plugin_ops, + .label = "cde", + .desc = "compressed directory entry", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = max_key_inside_cde, + .can_contain_key = can_contain_key_cde, + .mergeable = mergeable_cde, + .nr_units = nr_units_cde, + .lookup = lookup_cde, + .init = init_cde, + .paste = paste_cde, + .fast_paste = agree_to_fast_op, + .can_shift = can_shift_cde, + .copy_units = copy_units_cde, + .create_hook = NULL, + .kill_hook = NULL, + .shift_hook = NULL, + .cut_units = cut_units_cde, + .kill_units = kill_units_cde, + .unit_key = unit_key_cde, + .max_unit_key = unit_key_cde, + .estimate = estimate_cde, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = reiser4_check_cde +#endif + }, + .f = { + .utmost_child = NULL, + .utmost_child_real_block = NULL, + .update = NULL, + .scan = NULL, + .convert = NULL + }, + .s = { + .dir = { + .extract_key = extract_key_cde, + .update_key = update_key_cde, + .extract_name = extract_name_cde, + .extract_file_type = extract_file_type_de, + .add_entry = add_entry_cde, + .rem_entry = rem_entry_cde, + .max_name_len = max_name_len_cde + } + } + }, + [NODE_POINTER_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = NODE_POINTER_ID, + .groups = (1 << INTERNAL_ITEM_TYPE), + .pops = NULL, + .label = "internal", + .desc = "internal item", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = NULL, + .can_contain_key = NULL, + .mergeable = mergeable_internal, + .nr_units = nr_units_single_unit, + .lookup = lookup_internal, + .init = NULL, + .paste = NULL, + .fast_paste = NULL, + .can_shift = NULL, + .copy_units = NULL, + .create_hook = create_hook_internal, + .kill_hook = kill_hook_internal, + .shift_hook = shift_hook_internal, + .cut_units = NULL, + .kill_units = NULL, + .unit_key = NULL, + .max_unit_key = NULL, + .estimate = NULL, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = check__internal +#endif + }, + .f = { + .utmost_child = utmost_child_internal, + .utmost_child_real_block = + utmost_child_real_block_internal, + .update = reiser4_update_internal, + .scan = NULL, + .convert = NULL + }, + .s = { + .internal = { + .down_link = down_link_internal, + .has_pointer_to = has_pointer_to_internal + } + } + }, + [EXTENT_POINTER_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = EXTENT_POINTER_ID, + .groups = (1 << UNIX_FILE_METADATA_ITEM_TYPE), + .pops = NULL, + .label = "extent", + .desc = "extent item", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = max_key_inside_extent, + .can_contain_key = can_contain_key_extent, + .mergeable = mergeable_extent, + .nr_units = nr_units_extent, + .lookup = lookup_extent, + .init = NULL, + .paste = paste_extent, + .fast_paste = agree_to_fast_op, + .can_shift = can_shift_extent, + .create_hook = create_hook_extent, + .copy_units = copy_units_extent, + .kill_hook = kill_hook_extent, + .shift_hook = NULL, + .cut_units = cut_units_extent, + .kill_units = kill_units_extent, + .unit_key = unit_key_extent, + .max_unit_key = max_unit_key_extent, + .estimate = NULL, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = reiser4_check_extent +#endif + }, + .f = { + .utmost_child = utmost_child_extent, + .utmost_child_real_block = + utmost_child_real_block_extent, + .update = NULL, + .scan = reiser4_scan_extent, + .convert = NULL, + .key_by_offset = key_by_offset_extent + }, + .s = { + .file = { + .write = reiser4_write_extent, + .read = reiser4_read_extent, + .readpage = reiser4_readpage_extent, + .get_block = get_block_address_extent, + .append_key = append_key_extent, + .init_coord_extension = + init_coord_extension_extent + } + } + }, + [FORMATTING_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = FORMATTING_ID, + .groups = (1 << UNIX_FILE_METADATA_ITEM_TYPE), + .pops = NULL, + .label = "body", + .desc = "body (or tail?) item", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = max_key_inside_tail, + .can_contain_key = can_contain_key_tail, + .mergeable = mergeable_tail, + .nr_units = nr_units_tail, + .lookup = lookup_tail, + .init = NULL, + .paste = paste_tail, + .fast_paste = agree_to_fast_op, + .can_shift = can_shift_tail, + .create_hook = NULL, + .copy_units = copy_units_tail, + .kill_hook = kill_hook_tail, + .shift_hook = NULL, + .cut_units = cut_units_tail, + .kill_units = kill_units_tail, + .unit_key = unit_key_tail, + .max_unit_key = unit_key_tail, + .estimate = NULL, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = NULL +#endif + }, + .f = { + .utmost_child = NULL, + .utmost_child_real_block = NULL, + .update = NULL, + .scan = NULL, + .convert = NULL + }, + .s = { + .file = { + .write = reiser4_write_tail, + .read = reiser4_read_tail, + .readpage = readpage_tail, + .get_block = get_block_address_tail, + .append_key = append_key_tail, + .init_coord_extension = + init_coord_extension_tail + } + } + }, + [CTAIL_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = CTAIL_ID, + .groups = (1 << UNIX_FILE_METADATA_ITEM_TYPE), + .pops = NULL, + .label = "ctail", + .desc = "cryptcompress tail item", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = max_key_inside_tail, + .can_contain_key = can_contain_key_ctail, + .mergeable = mergeable_ctail, + .nr_units = nr_units_ctail, + .lookup = NULL, + .init = init_ctail, + .paste = paste_ctail, + .fast_paste = agree_to_fast_op, + .can_shift = can_shift_ctail, + .create_hook = create_hook_ctail, + .copy_units = copy_units_ctail, + .kill_hook = kill_hook_ctail, + .shift_hook = shift_hook_ctail, + .cut_units = cut_units_ctail, + .kill_units = kill_units_ctail, + .unit_key = unit_key_tail, + .max_unit_key = unit_key_tail, + .estimate = estimate_ctail, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = check_ctail +#endif + }, + .f = { + .utmost_child = utmost_child_ctail, + /* FIXME-EDWARD: write this */ + .utmost_child_real_block = NULL, + .update = NULL, + .scan = scan_ctail, + .convert = convert_ctail + }, + .s = { + .file = { + .write = NULL, + .read = read_ctail, + .readpage = readpage_ctail, + .get_block = get_block_address_tail, + .append_key = append_key_ctail, + .init_coord_extension = + init_coord_extension_tail + } + } + }, + [BLACK_BOX_ID] = { + .h = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .id = BLACK_BOX_ID, + .groups = (1 << OTHER_ITEM_TYPE), + .pops = NULL, + .label = "blackbox", + .desc = "black box item", + .linkage = {NULL, NULL} + }, + .b = { + .max_key_inside = NULL, + .can_contain_key = NULL, + .mergeable = not_mergeable, + .nr_units = nr_units_single_unit, + /* to need for ->lookup method */ + .lookup = NULL, + .init = NULL, + .paste = NULL, + .fast_paste = NULL, + .can_shift = NULL, + .copy_units = NULL, + .create_hook = NULL, + .kill_hook = NULL, + .shift_hook = NULL, + .cut_units = NULL, + .kill_units = NULL, + .unit_key = NULL, + .max_unit_key = NULL, + .estimate = NULL, + .item_data_by_flow = NULL, +#if REISER4_DEBUG + .check = NULL +#endif + } + } +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/item.h linux-2.6.30/fs/reiser4/plugin/item/item.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/item.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/item.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,398 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* first read balance.c comments before reading this */ + +/* An item_plugin implements all of the operations required for + balancing that are item specific. */ + +/* an item plugin also implements other operations that are specific to that + item. These go into the item specific operations portion of the item + handler, and all of the item specific portions of the item handler are put + into a union. */ + +#if !defined( __REISER4_ITEM_H__ ) +#define __REISER4_ITEM_H__ + +#include "../../forward.h" +#include "../plugin_header.h" +#include "../../dformat.h" +#include "../../seal.h" +#include "../../plugin/file/file.h" + +#include /* for struct file, struct inode */ +#include /* for struct page */ +#include /* for struct dentry */ + +typedef enum { + STAT_DATA_ITEM_TYPE, + DIR_ENTRY_ITEM_TYPE, + INTERNAL_ITEM_TYPE, + UNIX_FILE_METADATA_ITEM_TYPE, + OTHER_ITEM_TYPE +} item_type_id; + +/* this is the part of each item plugin that all items are expected to + support or at least explicitly fail to support by setting the + pointer to null. */ +struct balance_ops { + /* operations called by balancing + + It is interesting to consider that some of these item + operations could be given sources or targets that are not + really items in nodes. This could be ok/useful. + + */ + /* maximal key that can _possibly_ be occupied by this item + + When inserting, and node ->lookup() method (called by + coord_by_key()) reaches an item after binary search, + the ->max_key_inside() item plugin method is used to determine + whether new item should pasted into existing item + (new_key<=max_key_inside()) or new item has to be created + (new_key>max_key_inside()). + + For items that occupy exactly one key (like stat-data) + this method should return this key. For items that can + grow indefinitely (extent, directory item) this should + return reiser4_max_key(). + + For example extent with the key + + (LOCALITY,4,OBJID,STARTING-OFFSET), and length BLK blocks, + + ->max_key_inside is (LOCALITY,4,OBJID,0xffffffffffffffff), and + */ + reiser4_key *(*max_key_inside) (const coord_t *, reiser4_key *); + + /* true if item @coord can merge data at @key. */ + int (*can_contain_key) (const coord_t *, const reiser4_key *, + const reiser4_item_data *); + /* mergeable() - check items for mergeability + + Optional method. Returns true if two items can be merged. + + */ + int (*mergeable) (const coord_t *, const coord_t *); + + /* number of atomic things in an item. + NOTE FOR CONTRIBUTORS: use a generic method + nr_units_single_unit() for solid (atomic) items, as + tree operations use it as a criterion of solidness + (see is_solid_item macro) */ + pos_in_node_t(*nr_units) (const coord_t *); + + /* search within item for a unit within the item, and return a + pointer to it. This can be used to calculate how many + bytes to shrink an item if you use pointer arithmetic and + compare to the start of the item body if the item's data + are continuous in the node, if the item's data are not + continuous in the node, all sorts of other things are maybe + going to break as well. */ + lookup_result(*lookup) (const reiser4_key *, lookup_bias, coord_t *); + /* method called by ode_plugin->create_item() to initialise new + item */ + int (*init) (coord_t * target, coord_t * from, + reiser4_item_data * data); + /* method called (e.g., by reiser4_resize_item()) to place new data + into item when it grows */ + int (*paste) (coord_t *, reiser4_item_data *, carry_plugin_info *); + /* return true if paste into @coord is allowed to skip + carry. That is, if such paste would require any changes + at the parent level + */ + int (*fast_paste) (const coord_t *); + /* how many but not more than @want units of @source can be + shifted into @target node. If pend == append - we try to + append last item of @target by first units of @source. If + pend == prepend - we try to "prepend" first item in @target + by last units of @source. @target node has @free_space + bytes of free space. Total size of those units are returned + via @size. + + @target is not NULL if shifting to the mergeable item and + NULL is new item will be created during shifting. + */ + int (*can_shift) (unsigned free_space, coord_t *, + znode *, shift_direction, unsigned *size, + unsigned want); + + /* starting off @from-th unit of item @source append or + prepend @count units to @target. @target has been already + expanded by @free_space bytes. That must be exactly what is + needed for those items in @target. If @where_is_free_space + == SHIFT_LEFT - free space is at the end of @target item, + othersize - it is in the beginning of it. */ + void (*copy_units) (coord_t *, coord_t *, + unsigned from, unsigned count, + shift_direction where_is_free_space, + unsigned free_space); + + int (*create_hook) (const coord_t *, void *); + /* do whatever is necessary to do when @count units starting + from @from-th one are removed from the tree */ + /* FIXME-VS: this is used to be here for, in particular, + extents and items of internal type to free blocks they point + to at the same time with removing items from a + tree. Problems start, however, when dealloc_block fails due + to some reason. Item gets removed, but blocks it pointed to + are not freed. It is not clear how to fix this for items of + internal type because a need to remove internal item may + appear in the middle of balancing, and there is no way to + undo changes made. OTOH, if space allocator involves + balancing to perform dealloc_block - this will probably + break balancing due to deadlock issues + */ + int (*kill_hook) (const coord_t *, pos_in_node_t from, + pos_in_node_t count, struct carry_kill_data *); + int (*shift_hook) (const coord_t *, unsigned from, unsigned count, + znode * _node); + + /* unit @*from contains @from_key. unit @*to contains @to_key. Cut all keys between @from_key and @to_key + including boundaries. When units are cut from item beginning - move space which gets freed to head of + item. When units are cut from item end - move freed space to item end. When units are cut from the middle of + item - move freed space to item head. Return amount of space which got freed. Save smallest removed key in + @smallest_removed if it is not 0. Save new first item key in @new_first_key if it is not 0 + */ + int (*cut_units) (coord_t *, pos_in_node_t from, pos_in_node_t to, + struct carry_cut_data *, + reiser4_key * smallest_removed, + reiser4_key * new_first_key); + + /* like cut_units, except that these units are removed from the + tree, not only from a node */ + int (*kill_units) (coord_t *, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *, + reiser4_key * smallest_removed, + reiser4_key * new_first); + + /* if @key_of_coord == 1 - returned key of coord, otherwise - + key of unit is returned. If @coord is not set to certain + unit - ERR_PTR(-ENOENT) is returned */ + reiser4_key *(*unit_key) (const coord_t *, reiser4_key *); + reiser4_key *(*max_unit_key) (const coord_t *, reiser4_key *); + /* estimate how much space is needed for paste @data into item at + @coord. if @coord==0 - estimate insertion, otherwise - estimate + pasting + */ + int (*estimate) (const coord_t *, const reiser4_item_data *); + + /* converts flow @f to item data. @coord == 0 on insert */ + int (*item_data_by_flow) (const coord_t *, const flow_t *, + reiser4_item_data *); + + /*void (*show) (struct seq_file *, coord_t *); */ + +#if REISER4_DEBUG + /* used for debugging, every item should have here the most + complete possible check of the consistency of the item that + the inventor can construct */ + int (*check) (const coord_t *, const char **error); +#endif + +}; + +struct flush_ops { + /* return the right or left child of @coord, only if it is in memory */ + int (*utmost_child) (const coord_t *, sideof side, jnode ** child); + + /* return whether the right or left child of @coord has a non-fake + block number. */ + int (*utmost_child_real_block) (const coord_t *, sideof side, + reiser4_block_nr *); + /* relocate child at @coord to the @block */ + void (*update) (const coord_t *, const reiser4_block_nr *); + /* count unformatted nodes per item for leave relocation policy, etc.. */ + int (*scan) (flush_scan * scan); + /* convert item by flush */ + int (*convert) (flush_pos_t * pos); + /* backward mapping from jnode offset to a key. */ + int (*key_by_offset) (struct inode *, loff_t, reiser4_key *); +}; + +/* operations specific to the directory item */ +struct dir_entry_iops { + /* extract stat-data key from directory entry at @coord and place it + into @key. */ + int (*extract_key) (const coord_t *, reiser4_key * key); + /* update object key in item. */ + int (*update_key) (const coord_t *, const reiser4_key *, lock_handle *); + /* extract name from directory entry at @coord and return it */ + char *(*extract_name) (const coord_t *, char *buf); + /* extract file type (DT_* stuff) from directory entry at @coord and + return it */ + unsigned (*extract_file_type) (const coord_t *); + int (*add_entry) (struct inode * dir, + coord_t *, lock_handle *, + const struct dentry * name, + reiser4_dir_entry_desc * entry); + int (*rem_entry) (struct inode * dir, const struct qstr * name, + coord_t *, lock_handle *, + reiser4_dir_entry_desc * entry); + int (*max_name_len) (const struct inode * dir); +}; + +/* operations specific to items regular (unix) file metadata are built of */ +struct file_iops{ + int (*write) (struct file *, struct inode *, + const char __user *, size_t, loff_t *pos); + int (*read) (struct file *, flow_t *, hint_t *); + int (*readpage) (void *, struct page *); + int (*get_block) (const coord_t *, sector_t, sector_t *); + /* + * key of first byte which is not addressed by the item @coord is set + * to. + * For example, for extent item with the key + * + * (LOCALITY,4,OBJID,STARTING-OFFSET), and length BLK blocks, + * + * ->append_key is + * + * (LOCALITY,4,OBJID,STARTING-OFFSET + BLK * block_size) + */ + reiser4_key *(*append_key) (const coord_t *, reiser4_key *); + + void (*init_coord_extension) (uf_coord_t *, loff_t); +}; + +/* operations specific to items of stat data type */ +struct sd_iops { + int (*init_inode) (struct inode * inode, char *sd, int len); + int (*save_len) (struct inode * inode); + int (*save) (struct inode * inode, char **area); +}; + +/* operations specific to internal item */ +struct internal_iops{ + /* all tree traversal want to know from internal item is where + to go next. */ + void (*down_link) (const coord_t * coord, + const reiser4_key * key, reiser4_block_nr * block); + /* check that given internal item contains given pointer. */ + int (*has_pointer_to) (const coord_t * coord, + const reiser4_block_nr * block); +}; + +struct item_plugin { + /* generic fields */ + plugin_header h; + /* methods common for all item types */ + struct balance_ops b; /* balance operations */ + struct flush_ops f; /* flush operates with items via this methods */ + + /* methods specific to particular type of item */ + union { + struct dir_entry_iops dir; + struct file_iops file; + struct sd_iops sd; + struct internal_iops internal; + } s; +}; + +#define is_solid_item(iplug) ((iplug)->b.nr_units == nr_units_single_unit) + +static inline item_id item_id_by_plugin(item_plugin * plugin) +{ + return plugin->h.id; +} + +static inline char get_iplugid(item_plugin * iplug) +{ + assert("nikita-2838", iplug != NULL); + assert("nikita-2839", iplug->h.id < 0xff); + return (char)item_id_by_plugin(iplug); +} + +extern unsigned long znode_times_locked(const znode * z); + +static inline void coord_set_iplug(coord_t * coord, item_plugin * iplug) +{ + assert("nikita-2837", coord != NULL); + assert("nikita-2838", iplug != NULL); + coord->iplugid = get_iplugid(iplug); + ON_DEBUG(coord->plug_v = znode_times_locked(coord->node)); +} + +static inline item_plugin *coord_iplug(const coord_t * coord) +{ + assert("nikita-2833", coord != NULL); + assert("nikita-2834", coord->iplugid != INVALID_PLUGID); + assert("nikita-3549", coord->plug_v == znode_times_locked(coord->node)); + return (item_plugin *) plugin_by_id(REISER4_ITEM_PLUGIN_TYPE, + coord->iplugid); +} + +extern int item_can_contain_key(const coord_t * item, const reiser4_key * key, + const reiser4_item_data *); +extern int are_items_mergeable(const coord_t * i1, const coord_t * i2); +extern int item_is_extent(const coord_t *); +extern int item_is_tail(const coord_t *); +extern int item_is_statdata(const coord_t * item); +extern int item_is_ctail(const coord_t *); + +extern pos_in_node_t item_length_by_coord(const coord_t * coord); +extern pos_in_node_t nr_units_single_unit(const coord_t * coord); +extern item_id item_id_by_coord(const coord_t * coord /* coord to query */ ); +extern reiser4_key *item_key_by_coord(const coord_t * coord, reiser4_key * key); +extern reiser4_key *max_item_key_by_coord(const coord_t *, reiser4_key *); +extern reiser4_key *unit_key_by_coord(const coord_t * coord, reiser4_key * key); +extern reiser4_key *max_unit_key_by_coord(const coord_t * coord, + reiser4_key * key); +extern void obtain_item_plugin(const coord_t * coord); + +#if defined(REISER4_DEBUG) +extern int znode_is_loaded(const znode * node); +#endif + +/* return plugin of item at @coord */ +static inline item_plugin *item_plugin_by_coord(const coord_t * + coord /* coord to query */ ) +{ + assert("nikita-330", coord != NULL); + assert("nikita-331", coord->node != NULL); + assert("nikita-332", znode_is_loaded(coord->node)); + + if (unlikely(!coord_is_iplug_set(coord))) + obtain_item_plugin(coord); + return coord_iplug(coord); +} + +/* this returns true if item is of internal type */ +static inline int item_is_internal(const coord_t * item) +{ + assert("vs-483", coord_is_existing_item(item)); + return plugin_of_group(item_plugin_by_coord(item), INTERNAL_ITEM_TYPE); +} + +extern void item_body_by_coord_hard(coord_t * coord); +extern void *item_body_by_coord_easy(const coord_t * coord); +#if REISER4_DEBUG +extern int item_body_is_valid(const coord_t * coord); +#endif + +/* return pointer to item body */ +static inline void *item_body_by_coord(const coord_t * + coord /* coord to query */ ) +{ + assert("nikita-324", coord != NULL); + assert("nikita-325", coord->node != NULL); + assert("nikita-326", znode_is_loaded(coord->node)); + + if (coord->offset == INVALID_OFFSET) + item_body_by_coord_hard((coord_t *) coord); + assert("nikita-3201", item_body_is_valid(coord)); + assert("nikita-3550", coord->body_v == znode_times_locked(coord->node)); + return item_body_by_coord_easy(coord); +} + +/* __REISER4_ITEM_H__ */ +#endif +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/Makefile linux-2.6.30/fs/reiser4/plugin/item/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/item/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,18 @@ +obj-$(CONFIG_REISER4_FS) += item_plugins.o + +item_plugins-objs := \ + item.o \ + static_stat.o \ + sde.o \ + cde.o \ + blackbox.o \ + internal.o \ + tail.o \ + ctail.o \ + extent.o \ + extent_item_ops.o \ + extent_file_ops.o \ + extent_flush_ops.o + + + diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/sde.c linux-2.6.30/fs/reiser4/plugin/item/sde.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/sde.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/sde.c 2009-06-22 17:27:31.000000000 +0200 @@ -0,0 +1,190 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Directory entry implementation */ +#include "../../forward.h" +#include "../../debug.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../coord.h" +#include "sde.h" +#include "item.h" +#include "../plugin.h" +#include "../../znode.h" +#include "../../carry.h" +#include "../../tree.h" +#include "../../inode.h" + +#include /* for struct inode */ +#include /* for struct dentry */ +#include + +/* ->extract_key() method of simple directory item plugin. */ +int extract_key_de(const coord_t * coord /* coord of item */ , + reiser4_key * key /* resulting key */ ) +{ + directory_entry_format *dent; + + assert("nikita-1458", coord != NULL); + assert("nikita-1459", key != NULL); + + dent = (directory_entry_format *) item_body_by_coord(coord); + assert("nikita-1158", item_length_by_coord(coord) >= (int)sizeof *dent); + return extract_key_from_id(&dent->id, key); +} + +int +update_key_de(const coord_t * coord, const reiser4_key * key, + lock_handle * lh UNUSED_ARG) +{ + directory_entry_format *dent; + obj_key_id obj_id; + int result; + + assert("nikita-2342", coord != NULL); + assert("nikita-2343", key != NULL); + + dent = (directory_entry_format *) item_body_by_coord(coord); + result = build_obj_key_id(key, &obj_id); + if (result == 0) { + dent->id = obj_id; + znode_make_dirty(coord->node); + } + return 0; +} + +char *extract_dent_name(const coord_t * coord, directory_entry_format * dent, + char *buf) +{ + reiser4_key key; + + unit_key_by_coord(coord, &key); + if (get_key_type(&key) != KEY_FILE_NAME_MINOR) + reiser4_print_address("oops", znode_get_block(coord->node)); + if (!is_longname_key(&key)) { + if (is_dot_key(&key)) + return (char *)"."; + else + return extract_name_from_key(&key, buf); + } else + return (char *)dent->name; +} + +/* ->extract_name() method of simple directory item plugin. */ +char *extract_name_de(const coord_t * coord /* coord of item */ , char *buf) +{ + directory_entry_format *dent; + + assert("nikita-1460", coord != NULL); + + dent = (directory_entry_format *) item_body_by_coord(coord); + return extract_dent_name(coord, dent, buf); +} + +/* ->extract_file_type() method of simple directory item plugin. */ +unsigned extract_file_type_de(const coord_t * coord UNUSED_ARG /* coord of + * item */ ) +{ + assert("nikita-1764", coord != NULL); + /* we don't store file type in the directory entry yet. + + But see comments at kassign.h:obj_key_id + */ + return DT_UNKNOWN; +} + +int add_entry_de(struct inode *dir /* directory of item */ , + coord_t * coord /* coord of item */ , + lock_handle * lh /* insertion lock handle */ , + const struct dentry *de /* name to add */ , + reiser4_dir_entry_desc * entry /* parameters of new directory + * entry */ ) +{ + reiser4_item_data data; + directory_entry_format *dent; + int result; + const char *name; + int len; + int longname; + + name = de->d_name.name; + len = de->d_name.len; + assert("nikita-1163", strlen(name) == len); + + longname = is_longname(name, len); + + data.length = sizeof *dent; + if (longname) + data.length += len + 1; + data.data = NULL; + data.user = 0; + data.iplug = item_plugin_by_id(SIMPLE_DIR_ENTRY_ID); + + /* NOTE-NIKITA quota plugin */ + if (vfs_dq_alloc_space_nodirty(dir, data.length)) + return -EDQUOT; + + result = insert_by_coord(coord, &data, &entry->key, lh, 0 /*flags */ ); + if (result != 0) + return result; + + dent = (directory_entry_format *) item_body_by_coord(coord); + build_inode_key_id(entry->obj, &dent->id); + if (longname) { + memcpy(dent->name, name, len); + put_unaligned(0, &dent->name[len]); + } + return 0; +} + +int rem_entry_de(struct inode *dir /* directory of item */ , + const struct qstr *name UNUSED_ARG, + coord_t * coord /* coord of item */ , + lock_handle * lh UNUSED_ARG /* lock handle for + * removal */ , + reiser4_dir_entry_desc * entry UNUSED_ARG /* parameters of + * directory entry + * being removed */ ) +{ + coord_t shadow; + int result; + int length; + + length = item_length_by_coord(coord); + if (inode_get_bytes(dir) < length) { + warning("nikita-2627", "Dir is broke: %llu: %llu", + (unsigned long long)get_inode_oid(dir), + inode_get_bytes(dir)); + + return RETERR(-EIO); + } + + /* cut_node() is supposed to take pointers to _different_ + coords, because it will modify them without respect to + possible aliasing. To work around this, create temporary copy + of @coord. + */ + coord_dup(&shadow, coord); + result = + kill_node_content(coord, &shadow, NULL, NULL, NULL, NULL, NULL, 0); + if (result == 0) { + /* NOTE-NIKITA quota plugin */ + vfs_dq_free_space_nodirty(dir, length); + } + return result; +} + +int max_name_len_de(const struct inode *dir) +{ + return reiser4_tree_by_inode(dir)->nplug->max_item_size() - + sizeof(directory_entry_format) - 2; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/sde.h linux-2.6.30/fs/reiser4/plugin/item/sde.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/sde.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/sde.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,66 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Directory entry. */ + +#if !defined( __FS_REISER4_PLUGIN_DIRECTORY_ENTRY_H__ ) +#define __FS_REISER4_PLUGIN_DIRECTORY_ENTRY_H__ + +#include "../../forward.h" +#include "../../dformat.h" +#include "../../kassign.h" +#include "../../key.h" + +#include +#include /* for struct dentry */ + +typedef struct directory_entry_format { + /* key of object stat-data. It's not necessary to store whole + key here, because it's always key of stat-data, so minor + packing locality and offset can be omitted here. But this + relies on particular key allocation scheme for stat-data, so, + for extensibility sake, whole key can be stored here. + + We store key as array of bytes, because we don't want 8-byte + alignment of dir entries. + */ + obj_key_id id; + /* file name. Null terminated string. */ + d8 name[0]; +} directory_entry_format; + +void print_de(const char *prefix, coord_t * coord); +int extract_key_de(const coord_t * coord, reiser4_key * key); +int update_key_de(const coord_t * coord, const reiser4_key * key, + lock_handle * lh); +char *extract_name_de(const coord_t * coord, char *buf); +unsigned extract_file_type_de(const coord_t * coord); +int add_entry_de(struct inode *dir, coord_t * coord, + lock_handle * lh, const struct dentry *name, + reiser4_dir_entry_desc * entry); +int rem_entry_de(struct inode *dir, const struct qstr *name, coord_t * coord, + lock_handle * lh, reiser4_dir_entry_desc * entry); +int max_name_len_de(const struct inode *dir); + +int de_rem_and_shrink(struct inode *dir, coord_t * coord, int length); + +char *extract_dent_name(const coord_t * coord, + directory_entry_format * dent, char *buf); + +#if REISER4_LARGE_KEY +#define DE_NAME_BUF_LEN (24) +#else +#define DE_NAME_BUF_LEN (16) +#endif + +/* __FS_REISER4_PLUGIN_DIRECTORY_ENTRY_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/static_stat.c linux-2.6.30/fs/reiser4/plugin/item/static_stat.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/static_stat.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/static_stat.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1107 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* stat data manipulation. */ + +#include "../../forward.h" +#include "../../super.h" +#include "../../vfs_ops.h" +#include "../../inode.h" +#include "../../debug.h" +#include "../../dformat.h" +#include "../object.h" +#include "../plugin.h" +#include "../plugin_header.h" +#include "static_stat.h" +#include "item.h" + +#include +#include + +/* see static_stat.h for explanation */ + +/* helper function used while we are dumping/loading inode/plugin state + to/from the stat-data. */ + +static void move_on(int *length /* space remaining in stat-data */ , + char **area /* current coord in stat data */ , + int size_of /* how many bytes to move forward */ ) +{ + assert("nikita-615", length != NULL); + assert("nikita-616", area != NULL); + + *length -= size_of; + *area += size_of; + + assert("nikita-617", *length >= 0); +} + +/* helper function used while loading inode/plugin state from stat-data. + Complain if there is less space in stat-data than was expected. + Can only happen on disk corruption. */ +static int not_enough_space(struct inode *inode /* object being processed */ , + const char *where /* error message */ ) +{ + assert("nikita-618", inode != NULL); + + warning("nikita-619", "Not enough space in %llu while loading %s", + (unsigned long long)get_inode_oid(inode), where); + + return RETERR(-EINVAL); +} + +/* helper function used while loading inode/plugin state from + stat-data. Call it if invalid plugin id was found. */ +static int unknown_plugin(reiser4_plugin_id id /* invalid id */ , + struct inode *inode /* object being processed */ ) +{ + warning("nikita-620", "Unknown plugin %i in %llu", + id, (unsigned long long)get_inode_oid(inode)); + + return RETERR(-EINVAL); +} + +/* this is installed as ->init_inode() method of + item_plugins[ STATIC_STAT_DATA_IT ] (fs/reiser4/plugin/item/item.c). + Copies data from on-disk stat-data format into inode. + Handles stat-data extensions. */ +/* was sd_load */ +int init_inode_static_sd(struct inode *inode /* object being processed */ , + char *sd /* stat-data body */ , + int len /* length of stat-data */ ) +{ + int result; + int bit; + int chunk; + __u16 mask; + __u64 bigmask; + reiser4_stat_data_base *sd_base; + reiser4_inode *state; + + assert("nikita-625", inode != NULL); + assert("nikita-626", sd != NULL); + + result = 0; + sd_base = (reiser4_stat_data_base *) sd; + state = reiser4_inode_data(inode); + mask = le16_to_cpu(get_unaligned(&sd_base->extmask)); + bigmask = mask; + reiser4_inode_set_flag(inode, REISER4_SDLEN_KNOWN); + + move_on(&len, &sd, sizeof *sd_base); + for (bit = 0, chunk = 0; + mask != 0 || bit <= LAST_IMPORTANT_SD_EXTENSION; + ++bit, mask >>= 1) { + if (((bit + 1) % 16) != 0) { + /* handle extension */ + sd_ext_plugin *sdplug; + + if (bit >= LAST_SD_EXTENSION) { + warning("vpf-1904", + "No such extension %i in inode %llu", + bit, + (unsigned long long) + get_inode_oid(inode)); + + result = RETERR(-EINVAL); + break; + } + + sdplug = sd_ext_plugin_by_id(bit); + if (sdplug == NULL) { + warning("nikita-627", + "No such extension %i in inode %llu", + bit, + (unsigned long long) + get_inode_oid(inode)); + + result = RETERR(-EINVAL); + break; + } + if (mask & 1) { + assert("nikita-628", sdplug->present); + /* alignment is not supported in node layout + plugin yet. + result = align( inode, &len, &sd, + sdplug -> alignment ); + if( result != 0 ) + return result; */ + result = sdplug->present(inode, &sd, &len); + } else if (sdplug->absent != NULL) + result = sdplug->absent(inode); + if (result) + break; + /* else, we are looking at the last bit in 16-bit + portion of bitmask */ + } else if (mask & 1) { + /* next portion of bitmask */ + if (len < (int)sizeof(d16)) { + warning("nikita-629", + "No space for bitmap in inode %llu", + (unsigned long long) + get_inode_oid(inode)); + + result = RETERR(-EINVAL); + break; + } + mask = le16_to_cpu(get_unaligned((d16 *)sd)); + bigmask <<= 16; + bigmask |= mask; + move_on(&len, &sd, sizeof(d16)); + ++chunk; + if (chunk == 3) { + if (!(mask & 0x8000)) { + /* clear last bit */ + mask &= ~0x8000; + continue; + } + /* too much */ + warning("nikita-630", + "Too many extensions in %llu", + (unsigned long long) + get_inode_oid(inode)); + + result = RETERR(-EINVAL); + break; + } + } else + /* bitmask exhausted */ + break; + } + state->extmask = bigmask; + /* common initialisations */ + if (len - (bit / 16 * sizeof(d16)) > 0) { + /* alignment in save_len_static_sd() is taken into account + -edward */ + warning("nikita-631", "unused space in inode %llu", + (unsigned long long)get_inode_oid(inode)); + } + + return result; +} + +/* estimates size of stat-data required to store inode. + Installed as ->save_len() method of + item_plugins[ STATIC_STAT_DATA_IT ] (fs/reiser4/plugin/item/item.c). */ +/* was sd_len */ +int save_len_static_sd(struct inode *inode /* object being processed */ ) +{ + unsigned int result; + __u64 mask; + int bit; + + assert("nikita-632", inode != NULL); + + result = sizeof(reiser4_stat_data_base); + mask = reiser4_inode_data(inode)->extmask; + for (bit = 0; mask != 0; ++bit, mask >>= 1) { + if (mask & 1) { + sd_ext_plugin *sdplug; + + sdplug = sd_ext_plugin_by_id(bit); + assert("nikita-633", sdplug != NULL); + /* no aligment support + result += + round_up( result, sdplug -> alignment ) - result; */ + result += sdplug->save_len(inode); + } + } + result += bit / 16 * sizeof(d16); + return result; +} + +/* saves inode into stat-data. + Installed as ->save() method of + item_plugins[ STATIC_STAT_DATA_IT ] (fs/reiser4/plugin/item/item.c). */ +/* was sd_save */ +int save_static_sd(struct inode *inode /* object being processed */ , + char **area /* where to save stat-data */ ) +{ + int result; + __u64 emask; + int bit; + unsigned int len; + reiser4_stat_data_base *sd_base; + + assert("nikita-634", inode != NULL); + assert("nikita-635", area != NULL); + + result = 0; + emask = reiser4_inode_data(inode)->extmask; + sd_base = (reiser4_stat_data_base *) * area; + put_unaligned(cpu_to_le16((__u16)(emask & 0xffff)), &sd_base->extmask); + /*cputod16((unsigned)(emask & 0xffff), &sd_base->extmask);*/ + + *area += sizeof *sd_base; + len = 0xffffffffu; + for (bit = 0; emask != 0; ++bit, emask >>= 1) { + if (emask & 1) { + if ((bit + 1) % 16 != 0) { + sd_ext_plugin *sdplug; + sdplug = sd_ext_plugin_by_id(bit); + assert("nikita-636", sdplug != NULL); + /* no alignment support yet + align( inode, &len, area, + sdplug -> alignment ); */ + result = sdplug->save(inode, area); + if (result) + break; + } else { + put_unaligned(cpu_to_le16((__u16)(emask & 0xffff)), + (d16 *)(*area)); + /*cputod16((unsigned)(emask & 0xffff), + (d16 *) * area);*/ + *area += sizeof(d16); + } + } + } + return result; +} + +/* stat-data extension handling functions. */ + +static int present_lw_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ , + int *len /* remaining length */ ) +{ + if (*len >= (int)sizeof(reiser4_light_weight_stat)) { + reiser4_light_weight_stat *sd_lw; + + sd_lw = (reiser4_light_weight_stat *) * area; + + inode->i_mode = le16_to_cpu(get_unaligned(&sd_lw->mode)); + inode->i_nlink = le32_to_cpu(get_unaligned(&sd_lw->nlink)); + inode->i_size = le64_to_cpu(get_unaligned(&sd_lw->size)); + if ((inode->i_mode & S_IFMT) == (S_IFREG | S_IFIFO)) { + inode->i_mode &= ~S_IFIFO; + warning("", "partially converted file is encountered"); + reiser4_inode_set_flag(inode, REISER4_PART_MIXED); + } + move_on(len, area, sizeof *sd_lw); + return 0; + } else + return not_enough_space(inode, "lw sd"); +} + +static int save_len_lw_sd(struct inode *inode UNUSED_ARG /* object being + * processed */ ) +{ + return sizeof(reiser4_light_weight_stat); +} + +static int save_lw_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ ) +{ + reiser4_light_weight_stat *sd; + mode_t delta; + + assert("nikita-2705", inode != NULL); + assert("nikita-2706", area != NULL); + assert("nikita-2707", *area != NULL); + + sd = (reiser4_light_weight_stat *) * area; + + delta = (reiser4_inode_get_flag(inode, + REISER4_PART_MIXED) ? S_IFIFO : 0); + put_unaligned(cpu_to_le16(inode->i_mode | delta), &sd->mode); + put_unaligned(cpu_to_le32(inode->i_nlink), &sd->nlink); + put_unaligned(cpu_to_le64((__u64) inode->i_size), &sd->size); + *area += sizeof *sd; + return 0; +} + +static int present_unix_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ , + int *len /* remaining length */ ) +{ + assert("nikita-637", inode != NULL); + assert("nikita-638", area != NULL); + assert("nikita-639", *area != NULL); + assert("nikita-640", len != NULL); + assert("nikita-641", *len > 0); + + if (*len >= (int)sizeof(reiser4_unix_stat)) { + reiser4_unix_stat *sd; + + sd = (reiser4_unix_stat *) * area; + + inode->i_uid = le32_to_cpu(get_unaligned(&sd->uid)); + inode->i_gid = le32_to_cpu(get_unaligned(&sd->gid)); + inode->i_atime.tv_sec = le32_to_cpu(get_unaligned(&sd->atime)); + inode->i_mtime.tv_sec = le32_to_cpu(get_unaligned(&sd->mtime)); + inode->i_ctime.tv_sec = le32_to_cpu(get_unaligned(&sd->ctime)); + if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) + inode->i_rdev = le64_to_cpu(get_unaligned(&sd->u.rdev)); + else + inode_set_bytes(inode, (loff_t) le64_to_cpu(get_unaligned(&sd->u.bytes))); + move_on(len, area, sizeof *sd); + return 0; + } else + return not_enough_space(inode, "unix sd"); +} + +static int absent_unix_sd(struct inode *inode /* object being processed */ ) +{ + inode->i_uid = get_super_private(inode->i_sb)->default_uid; + inode->i_gid = get_super_private(inode->i_sb)->default_gid; + inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; + inode_set_bytes(inode, inode->i_size); + /* mark inode as lightweight, so that caller (lookup_common) will + complete initialisation by copying [ug]id from a parent. */ + reiser4_inode_set_flag(inode, REISER4_LIGHT_WEIGHT); + return 0; +} + +/* Audited by: green(2002.06.14) */ +static int save_len_unix_sd(struct inode *inode UNUSED_ARG /* object being + * processed */ ) +{ + return sizeof(reiser4_unix_stat); +} + +static int save_unix_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ ) +{ + reiser4_unix_stat *sd; + + assert("nikita-642", inode != NULL); + assert("nikita-643", area != NULL); + assert("nikita-644", *area != NULL); + + sd = (reiser4_unix_stat *) * area; + put_unaligned(cpu_to_le32(inode->i_uid), &sd->uid); + put_unaligned(cpu_to_le32(inode->i_gid), &sd->gid); + put_unaligned(cpu_to_le32((__u32) inode->i_atime.tv_sec), &sd->atime); + put_unaligned(cpu_to_le32((__u32) inode->i_ctime.tv_sec), &sd->ctime); + put_unaligned(cpu_to_le32((__u32) inode->i_mtime.tv_sec), &sd->mtime); + if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) + put_unaligned(cpu_to_le64(inode->i_rdev), &sd->u.rdev); + else + put_unaligned(cpu_to_le64((__u64) inode_get_bytes(inode)), &sd->u.bytes); + *area += sizeof *sd; + return 0; +} + +static int +present_large_times_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ , + int *len /* remaining length */ ) +{ + if (*len >= (int)sizeof(reiser4_large_times_stat)) { + reiser4_large_times_stat *sd_lt; + + sd_lt = (reiser4_large_times_stat *) * area; + + inode->i_atime.tv_nsec = le32_to_cpu(get_unaligned(&sd_lt->atime)); + inode->i_mtime.tv_nsec = le32_to_cpu(get_unaligned(&sd_lt->mtime)); + inode->i_ctime.tv_nsec = le32_to_cpu(get_unaligned(&sd_lt->ctime)); + + move_on(len, area, sizeof *sd_lt); + return 0; + } else + return not_enough_space(inode, "large times sd"); +} + +static int +save_len_large_times_sd(struct inode *inode UNUSED_ARG + /* object being processed */ ) +{ + return sizeof(reiser4_large_times_stat); +} + +static int +save_large_times_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ ) +{ + reiser4_large_times_stat *sd; + + assert("nikita-2817", inode != NULL); + assert("nikita-2818", area != NULL); + assert("nikita-2819", *area != NULL); + + sd = (reiser4_large_times_stat *) * area; + + put_unaligned(cpu_to_le32((__u32) inode->i_atime.tv_nsec), &sd->atime); + put_unaligned(cpu_to_le32((__u32) inode->i_ctime.tv_nsec), &sd->ctime); + put_unaligned(cpu_to_le32((__u32) inode->i_mtime.tv_nsec), &sd->mtime); + + *area += sizeof *sd; + return 0; +} + +/* symlink stat data extension */ + +/* allocate memory for symlink target and attach it to inode->i_private */ +static int +symlink_target_to_inode(struct inode *inode, const char *target, int len) +{ + assert("vs-845", inode->i_private == NULL); + assert("vs-846", !reiser4_inode_get_flag(inode, + REISER4_GENERIC_PTR_USED)); + /* FIXME-VS: this is prone to deadlock. Not more than other similar + places, though */ + inode->i_private = kmalloc((size_t) len + 1, + reiser4_ctx_gfp_mask_get()); + if (!inode->i_private) + return RETERR(-ENOMEM); + + memcpy((char *)(inode->i_private), target, (size_t) len); + ((char *)(inode->i_private))[len] = 0; + reiser4_inode_set_flag(inode, REISER4_GENERIC_PTR_USED); + return 0; +} + +/* this is called on read_inode. There is nothing to do actually, but some + sanity checks */ +static int present_symlink_sd(struct inode *inode, char **area, int *len) +{ + int result; + int length; + reiser4_symlink_stat *sd; + + length = (int)inode->i_size; + /* + * *len is number of bytes in stat data item from *area to the end of + * item. It must be not less than size of symlink + 1 for ending 0 + */ + if (length > *len) + return not_enough_space(inode, "symlink"); + + if (*(*area + length) != 0) { + warning("vs-840", "Symlink is not zero terminated"); + return RETERR(-EIO); + } + + sd = (reiser4_symlink_stat *) * area; + result = symlink_target_to_inode(inode, sd->body, length); + + move_on(len, area, length + 1); + return result; +} + +static int save_len_symlink_sd(struct inode *inode) +{ + return inode->i_size + 1; +} + +/* this is called on create and update stat data. Do nothing on update but + update @area */ +static int save_symlink_sd(struct inode *inode, char **area) +{ + int result; + int length; + reiser4_symlink_stat *sd; + + length = (int)inode->i_size; + /* inode->i_size must be set already */ + assert("vs-841", length); + + result = 0; + sd = (reiser4_symlink_stat *) * area; + if (!reiser4_inode_get_flag(inode, REISER4_GENERIC_PTR_USED)) { + const char *target; + + target = (const char *)(inode->i_private); + inode->i_private = NULL; + + result = symlink_target_to_inode(inode, target, length); + + /* copy symlink to stat data */ + memcpy(sd->body, target, (size_t) length); + (*area)[length] = 0; + } else { + /* there is nothing to do in update but move area */ + assert("vs-844", + !memcmp(inode->i_private, sd->body, + (size_t) length + 1)); + } + + *area += (length + 1); + return result; +} + +static int present_flags_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ , + int *len /* remaining length */ ) +{ + assert("nikita-645", inode != NULL); + assert("nikita-646", area != NULL); + assert("nikita-647", *area != NULL); + assert("nikita-648", len != NULL); + assert("nikita-649", *len > 0); + + if (*len >= (int)sizeof(reiser4_flags_stat)) { + reiser4_flags_stat *sd; + + sd = (reiser4_flags_stat *) * area; + inode->i_flags = le32_to_cpu(get_unaligned(&sd->flags)); + move_on(len, area, sizeof *sd); + return 0; + } else + return not_enough_space(inode, "generation and attrs"); +} + +/* Audited by: green(2002.06.14) */ +static int save_len_flags_sd(struct inode *inode UNUSED_ARG /* object being + * processed */ ) +{ + return sizeof(reiser4_flags_stat); +} + +static int save_flags_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ ) +{ + reiser4_flags_stat *sd; + + assert("nikita-650", inode != NULL); + assert("nikita-651", area != NULL); + assert("nikita-652", *area != NULL); + + sd = (reiser4_flags_stat *) * area; + put_unaligned(cpu_to_le32(inode->i_flags), &sd->flags); + *area += sizeof *sd; + return 0; +} + +static int absent_plugin_sd(struct inode *inode); +static int present_plugin_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */ , + int *len /* remaining length */, + int is_pset /* 1 if plugin set, 0 if heir set. */) +{ + reiser4_plugin_stat *sd; + reiser4_plugin *plugin; + reiser4_inode *info; + int i; + __u16 mask; + int result; + int num_of_plugins; + + assert("nikita-653", inode != NULL); + assert("nikita-654", area != NULL); + assert("nikita-655", *area != NULL); + assert("nikita-656", len != NULL); + assert("nikita-657", *len > 0); + + if (*len < (int)sizeof(reiser4_plugin_stat)) + return not_enough_space(inode, "plugin"); + + sd = (reiser4_plugin_stat *) * area; + info = reiser4_inode_data(inode); + + mask = 0; + num_of_plugins = le16_to_cpu(get_unaligned(&sd->plugins_no)); + move_on(len, area, sizeof *sd); + result = 0; + for (i = 0; i < num_of_plugins; ++i) { + reiser4_plugin_slot *slot; + reiser4_plugin_type type; + pset_member memb; + + slot = (reiser4_plugin_slot *) * area; + if (*len < (int)sizeof *slot) + return not_enough_space(inode, "additional plugin"); + + memb = le16_to_cpu(get_unaligned(&slot->pset_memb)); + type = aset_member_to_type_unsafe(memb); + + if (type == REISER4_PLUGIN_TYPES) { + warning("nikita-3502", + "wrong %s member (%i) for %llu", is_pset ? + "pset" : "hset", memb, + (unsigned long long)get_inode_oid(inode)); + return RETERR(-EINVAL); + } + plugin = plugin_by_disk_id(reiser4_tree_by_inode(inode), + type, &slot->id); + if (plugin == NULL) + return unknown_plugin(le16_to_cpu(get_unaligned(&slot->id)), inode); + + /* plugin is loaded into inode, mark this into inode's + bitmask of loaded non-standard plugins */ + if (!(mask & (1 << memb))) { + mask |= (1 << memb); + } else { + warning("nikita-658", "duplicate plugin for %llu", + (unsigned long long)get_inode_oid(inode)); + return RETERR(-EINVAL); + } + move_on(len, area, sizeof *slot); + /* load plugin data, if any */ + if (plugin->h.pops != NULL && plugin->h.pops->load) + result = plugin->h.pops->load(inode, plugin, area, len); + else + result = aset_set_unsafe(is_pset ? &info->pset : + &info->hset, memb, plugin); + if (result) + return result; + } + if (is_pset) { + /* if object plugin wasn't loaded from stat-data, guess it by + mode bits */ + plugin = file_plugin_to_plugin(inode_file_plugin(inode)); + if (plugin == NULL) + result = absent_plugin_sd(inode); + info->plugin_mask = mask; + } else + info->heir_mask = mask; + + return result; +} + +static int present_pset_sd(struct inode *inode, char **area, int *len) { + return present_plugin_sd(inode, area, len, 1 /* pset */); +} + +/* Determine object plugin for @inode based on i_mode. + + Many objects in reiser4 file system are controlled by standard object + plugins that emulate traditional unix objects: unix file, directory, symlink, fifo, and so on. + + For such files we don't explicitly store plugin id in object stat + data. Rather required plugin is guessed from mode bits, where file "type" + is encoded (see stat(2)). +*/ +static int +guess_plugin_by_mode(struct inode *inode /* object to guess plugins for */ ) +{ + int fplug_id; + int dplug_id; + reiser4_inode *info; + + assert("nikita-736", inode != NULL); + + dplug_id = fplug_id = -1; + + switch (inode->i_mode & S_IFMT) { + case S_IFSOCK: + case S_IFBLK: + case S_IFCHR: + case S_IFIFO: + fplug_id = SPECIAL_FILE_PLUGIN_ID; + break; + case S_IFLNK: + fplug_id = SYMLINK_FILE_PLUGIN_ID; + break; + case S_IFDIR: + fplug_id = DIRECTORY_FILE_PLUGIN_ID; + dplug_id = HASHED_DIR_PLUGIN_ID; + break; + default: + warning("nikita-737", "wrong file mode: %o", inode->i_mode); + return RETERR(-EIO); + case S_IFREG: + fplug_id = UNIX_FILE_PLUGIN_ID; + break; + } + info = reiser4_inode_data(inode); + set_plugin(&info->pset, PSET_FILE, (fplug_id >= 0) ? + plugin_by_id(REISER4_FILE_PLUGIN_TYPE, fplug_id) : NULL); + set_plugin(&info->pset, PSET_DIR, (dplug_id >= 0) ? + plugin_by_id(REISER4_DIR_PLUGIN_TYPE, dplug_id) : NULL); + return 0; +} + +/* Audited by: green(2002.06.14) */ +static int absent_plugin_sd(struct inode *inode /* object being processed */ ) +{ + int result; + + assert("nikita-659", inode != NULL); + + result = guess_plugin_by_mode(inode); + /* if mode was wrong, guess_plugin_by_mode() returns "regular file", + but setup_inode_ops() will call make_bad_inode(). + Another, more logical but bit more complex solution is to add + "bad-file plugin". */ + /* FIXME-VS: activate was called here */ + return result; +} + +/* helper function for plugin_sd_save_len(): calculate how much space + required to save state of given plugin */ +/* Audited by: green(2002.06.14) */ +static int len_for(reiser4_plugin * plugin /* plugin to save */ , + struct inode *inode /* object being processed */ , + pset_member memb, + int len, int is_pset) +{ + reiser4_inode *info; + assert("nikita-661", inode != NULL); + + if (plugin == NULL) + return len; + + info = reiser4_inode_data(inode); + if (is_pset ? + info->plugin_mask & (1 << memb) : + info->heir_mask & (1 << memb)) { + len += sizeof(reiser4_plugin_slot); + if (plugin->h.pops && plugin->h.pops->save_len != NULL) { + /* non-standard plugin, call method */ + /* commented as it is incompatible with alignment + * policy in save_plug() -edward */ + /* len = round_up(len, plugin->h.pops->alignment); */ + len += plugin->h.pops->save_len(inode, plugin); + } + } + return len; +} + +/* calculate how much space is required to save state of all plugins, + associated with inode */ +static int save_len_plugin_sd(struct inode *inode /* object being processed */, + int is_pset) +{ + int len; + int last; + reiser4_inode *state; + pset_member memb; + + assert("nikita-663", inode != NULL); + + state = reiser4_inode_data(inode); + + /* common case: no non-standard plugins */ + if (is_pset ? state->plugin_mask == 0 : state->heir_mask == 0) + return 0; + len = sizeof(reiser4_plugin_stat); + last = PSET_LAST; + + for (memb = 0; memb < last; ++memb) { + len = len_for(aset_get(is_pset ? state->pset : state->hset, memb), + inode, memb, len, is_pset); + } + assert("nikita-664", len > (int)sizeof(reiser4_plugin_stat)); + return len; +} + +static int save_len_pset_sd(struct inode *inode) { + return save_len_plugin_sd(inode, 1 /* pset */); +} + +/* helper function for plugin_sd_save(): save plugin, associated with + inode. */ +static int save_plug(reiser4_plugin * plugin /* plugin to save */ , + struct inode *inode /* object being processed */ , + int memb /* what element of pset is saved */ , + char **area /* position in stat-data */ , + int *count /* incremented if plugin were actually saved. */, + int is_pset /* 1 for plugin set, 0 for heir set */) +{ + reiser4_plugin_slot *slot; + int fake_len; + int result; + + assert("nikita-665", inode != NULL); + assert("nikita-666", area != NULL); + assert("nikita-667", *area != NULL); + + if (plugin == NULL) + return 0; + + if (is_pset ? + !(reiser4_inode_data(inode)->plugin_mask & (1 << memb)) : + !(reiser4_inode_data(inode)->heir_mask & (1 << memb))) + return 0; + slot = (reiser4_plugin_slot *) * area; + put_unaligned(cpu_to_le16(memb), &slot->pset_memb); + put_unaligned(cpu_to_le16(plugin->h.id), &slot->id); + fake_len = (int)0xffff; + move_on(&fake_len, area, sizeof *slot); + ++*count; + result = 0; + if (plugin->h.pops != NULL) { + if (plugin->h.pops->save != NULL) + result = plugin->h.pops->save(inode, plugin, area); + } + return result; +} + +/* save state of all non-standard plugins associated with inode */ +static int save_plugin_sd(struct inode *inode /* object being processed */ , + char **area /* position in stat-data */, + int is_pset /* 1 for pset, 0 for hset */) +{ + int fake_len; + int result = 0; + int num_of_plugins; + reiser4_plugin_stat *sd; + reiser4_inode *state; + pset_member memb; + + assert("nikita-669", inode != NULL); + assert("nikita-670", area != NULL); + assert("nikita-671", *area != NULL); + + state = reiser4_inode_data(inode); + if (is_pset ? state->plugin_mask == 0 : state->heir_mask == 0) + return 0; + sd = (reiser4_plugin_stat *) * area; + fake_len = (int)0xffff; + move_on(&fake_len, area, sizeof *sd); + + num_of_plugins = 0; + for (memb = 0; memb < PSET_LAST; ++memb) { + result = save_plug(aset_get(is_pset ? state->pset : state->hset, + memb), + inode, memb, area, &num_of_plugins, is_pset); + if (result != 0) + break; + } + + put_unaligned(cpu_to_le16((__u16)num_of_plugins), &sd->plugins_no); + return result; +} + +static int save_pset_sd(struct inode *inode, char **area) { + return save_plugin_sd(inode, area, 1 /* pset */); +} + +static int present_hset_sd(struct inode *inode, char **area, int *len) { + return present_plugin_sd(inode, area, len, 0 /* hset */); +} + +static int save_len_hset_sd(struct inode *inode) { + return save_len_plugin_sd(inode, 0 /* pset */); +} + +static int save_hset_sd(struct inode *inode, char **area) { + return save_plugin_sd(inode, area, 0 /* hset */); +} + +/* helper function for crypto_sd_present(), crypto_sd_save. + Extract crypto info from stat-data and attach it to inode */ +static int extract_crypto_info (struct inode * inode, + reiser4_crypto_stat * sd) +{ + struct reiser4_crypto_info * info; + assert("edward-11", !inode_crypto_info(inode)); + assert("edward-1413", + !reiser4_inode_get_flag(inode, REISER4_CRYPTO_STAT_LOADED)); + /* create and attach a crypto-stat without secret key loaded */ + info = reiser4_alloc_crypto_info(inode); + if (IS_ERR(info)) + return PTR_ERR(info); + info->keysize = le16_to_cpu(get_unaligned(&sd->keysize)); + memcpy(info->keyid, sd->keyid, inode_digest_plugin(inode)->fipsize); + reiser4_attach_crypto_info(inode, info); + reiser4_inode_set_flag(inode, REISER4_CRYPTO_STAT_LOADED); + return 0; +} + +/* crypto stat-data extension */ + +static int present_crypto_sd(struct inode *inode, char **area, int *len) +{ + int result; + reiser4_crypto_stat *sd; + digest_plugin *dplug = inode_digest_plugin(inode); + + assert("edward-06", dplug != NULL); + assert("edward-684", dplug->fipsize); + assert("edward-07", area != NULL); + assert("edward-08", *area != NULL); + assert("edward-09", len != NULL); + assert("edward-10", *len > 0); + + if (*len < (int)sizeof(reiser4_crypto_stat)) { + return not_enough_space(inode, "crypto-sd"); + } + /* *len is number of bytes in stat data item from *area to the end of + item. It must be not less than size of this extension */ + assert("edward-75", sizeof(*sd) + dplug->fipsize <= *len); + + sd = (reiser4_crypto_stat *) * area; + result = extract_crypto_info(inode, sd); + move_on(len, area, sizeof(*sd) + dplug->fipsize); + + return result; +} + +static int save_len_crypto_sd(struct inode *inode) +{ + return sizeof(reiser4_crypto_stat) + + inode_digest_plugin(inode)->fipsize; +} + +static int save_crypto_sd(struct inode *inode, char **area) +{ + int result = 0; + reiser4_crypto_stat *sd; + struct reiser4_crypto_info * info = inode_crypto_info(inode); + digest_plugin *dplug = inode_digest_plugin(inode); + + assert("edward-12", dplug != NULL); + assert("edward-13", area != NULL); + assert("edward-14", *area != NULL); + assert("edward-15", info != NULL); + assert("edward-1414", info->keyid != NULL); + assert("edward-1415", info->keysize != 0); + assert("edward-76", reiser4_inode_data(inode) != NULL); + + if (!reiser4_inode_get_flag(inode, REISER4_CRYPTO_STAT_LOADED)) { + /* file is just created */ + sd = (reiser4_crypto_stat *) *area; + /* copy everything but private key to the disk stat-data */ + put_unaligned(cpu_to_le16(info->keysize), &sd->keysize); + memcpy(sd->keyid, info->keyid, (size_t) dplug->fipsize); + reiser4_inode_set_flag(inode, REISER4_CRYPTO_STAT_LOADED); + } + *area += (sizeof(*sd) + dplug->fipsize); + return result; +} + +static int eio(struct inode *inode, char **area, int *len) +{ + return RETERR(-EIO); +} + +sd_ext_plugin sd_ext_plugins[LAST_SD_EXTENSION] = { + [LIGHT_WEIGHT_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = LIGHT_WEIGHT_STAT, + .pops = NULL, + .label = "light-weight sd", + .desc = "sd for light-weight files", + .linkage = {NULL,NULL} + }, + .present = present_lw_sd, + .absent = NULL, + .save_len = save_len_lw_sd, + .save = save_lw_sd, + .alignment = 8 + }, + [UNIX_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = UNIX_STAT, + .pops = NULL, + .label = "unix-sd", + .desc = "unix stat-data fields", + .linkage = {NULL,NULL} + }, + .present = present_unix_sd, + .absent = absent_unix_sd, + .save_len = save_len_unix_sd, + .save = save_unix_sd, + .alignment = 8 + }, + [LARGE_TIMES_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = LARGE_TIMES_STAT, + .pops = NULL, + .label = "64time-sd", + .desc = "nanosecond resolution for times", + .linkage = {NULL,NULL} + }, + .present = present_large_times_sd, + .absent = NULL, + .save_len = save_len_large_times_sd, + .save = save_large_times_sd, + .alignment = 8 + }, + [SYMLINK_STAT] = { + /* stat data of symlink has this extension */ + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = SYMLINK_STAT, + .pops = NULL, + .label = "symlink-sd", + .desc = + "stat data is appended with symlink name", + .linkage = {NULL,NULL} + }, + .present = present_symlink_sd, + .absent = NULL, + .save_len = save_len_symlink_sd, + .save = save_symlink_sd, + .alignment = 8 + }, + [PLUGIN_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = PLUGIN_STAT, + .pops = NULL, + .label = "plugin-sd", + .desc = "plugin stat-data fields", + .linkage = {NULL,NULL} + }, + .present = present_pset_sd, + .absent = absent_plugin_sd, + .save_len = save_len_pset_sd, + .save = save_pset_sd, + .alignment = 8 + }, + [HEIR_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = HEIR_STAT, + .pops = NULL, + .label = "heir-plugin-sd", + .desc = "heir plugin stat-data fields", + .linkage = {NULL,NULL} + }, + .present = present_hset_sd, + .absent = NULL, + .save_len = save_len_hset_sd, + .save = save_hset_sd, + .alignment = 8 + }, + [FLAGS_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = FLAGS_STAT, + .pops = NULL, + .label = "flags-sd", + .desc = "inode bit flags", + .linkage = {NULL, NULL} + }, + .present = present_flags_sd, + .absent = NULL, + .save_len = save_len_flags_sd, + .save = save_flags_sd, + .alignment = 8 + }, + [CAPABILITIES_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = CAPABILITIES_STAT, + .pops = NULL, + .label = "capabilities-sd", + .desc = "capabilities", + .linkage = {NULL, NULL} + }, + .present = eio, + .absent = NULL, + .save_len = save_len_flags_sd, + .save = save_flags_sd, + .alignment = 8 + }, + [CRYPTO_STAT] = { + .h = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .id = CRYPTO_STAT, + .pops = NULL, + .label = "crypto-sd", + .desc = "secret key size and id", + .linkage = {NULL, NULL} + }, + .present = present_crypto_sd, + .absent = NULL, + .save_len = save_len_crypto_sd, + .save = save_crypto_sd, + .alignment = 8 + } +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/static_stat.h linux-2.6.30/fs/reiser4/plugin/item/static_stat.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/static_stat.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/static_stat.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,224 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* This describes the static_stat item, used to hold all information needed by the stat() syscall. + +In the case where each file has not less than the fields needed by the +stat() syscall, it is more compact to store those fields in this +struct. + +If this item does not exist, then all stats are dynamically resolved. +At the moment, we either resolve all stats dynamically or all of them +statically. If you think this is not fully optimal, and the rest of +reiser4 is working, then fix it...:-) + +*/ + +#if !defined( __FS_REISER4_PLUGIN_ITEM_STATIC_STAT_H__ ) +#define __FS_REISER4_PLUGIN_ITEM_STATIC_STAT_H__ + +#include "../../forward.h" +#include "../../dformat.h" + +#include /* for struct inode */ + +/* Stat data layout: goals and implementation. + + We want to be able to have lightweight files which have complete flexibility in what semantic metadata is attached to + them, including not having semantic metadata attached to them. + + There is one problem with doing that, which is that if in fact you have exactly the same metadata for most files you + want to store, then it takes more space to store that metadata in a dynamically sized structure than in a statically + sized structure because the statically sized structure knows without recording it what the names and lengths of the + attributes are. + + This leads to a natural compromise, which is to special case those files which have simply the standard unix file + attributes, and only employ the full dynamic stat data mechanism for those files that differ from the standard unix + file in their use of file attributes. + + Yet this compromise deserves to be compromised a little. + + We accommodate the case where you have no more than the standard unix file attributes by using an "extension + bitmask": each bit in it indicates presence or absence of or particular stat data extension (see sd_ext_bits enum). + + If the first bit of the extension bitmask bit is 0, we have light-weight file whose attributes are either inherited + from parent directory (as uid, gid) or initialised to some sane values. + + To capitalize on existing code infrastructure, extensions are + implemented as plugins of type REISER4_SD_EXT_PLUGIN_TYPE. + Each stat-data extension plugin implements four methods: + + ->present() called by sd_load() when this extension is found in stat-data + ->absent() called by sd_load() when this extension is not found in stat-data + ->save_len() called by sd_len() to calculate total length of stat-data + ->save() called by sd_save() to store extension data into stat-data + + Implementation is in fs/reiser4/plugin/item/static_stat.c +*/ + +/* stat-data extension. Please order this by presumed frequency of use */ +typedef enum { + /* support for light-weight files */ + LIGHT_WEIGHT_STAT, + /* data required to implement unix stat(2) call. Layout is in + reiser4_unix_stat. If this is not present, file is light-weight */ + UNIX_STAT, + /* this contains additional set of 32bit [anc]time fields to implement + nanosecond resolution. Layout is in reiser4_large_times_stat. Usage + if this extension is governed by 32bittimes mount option. */ + LARGE_TIMES_STAT, + /* stat data has link name included */ + SYMLINK_STAT, + /* on-disk slots of non-standard plugins for main plugin table + (@reiser4_inode->pset), that is, plugins that cannot be deduced + from file mode bits), for example, aggregation, interpolation etc. */ + PLUGIN_STAT, + /* this extension contains persistent inode flags. These flags are + single bits: immutable, append, only, etc. Layout is in + reiser4_flags_stat. */ + FLAGS_STAT, + /* this extension contains capabilities sets, associated with this + file. Layout is in reiser4_capabilities_stat */ + CAPABILITIES_STAT, + /* this extension contains size and public id of the secret key. + Layout is in reiser4_crypto_stat */ + CRYPTO_STAT, + /* on-disk slots of non-default plugins for inheritance, which + are extracted to special plugin table (@reiser4_inode->hset). + By default, children of the object will inherit plugins from + its main plugin table (pset). */ + HEIR_STAT, + LAST_SD_EXTENSION, + /* + * init_inode_static_sd() iterates over extension mask until all + * non-zero bits are processed. This means, that neither ->present(), + * nor ->absent() methods will be called for stat-data extensions that + * go after last present extension. But some basic extensions, we want + * either ->absent() or ->present() method to be called, because these + * extensions set up something in inode even when they are not + * present. This is what LAST_IMPORTANT_SD_EXTENSION is for: for all + * extensions before and including LAST_IMPORTANT_SD_EXTENSION either + * ->present(), or ->absent() method will be called, independently of + * what other extensions are present. + */ + LAST_IMPORTANT_SD_EXTENSION = PLUGIN_STAT +} sd_ext_bits; + +/* minimal stat-data. This allows to support light-weight files. */ +typedef struct reiser4_stat_data_base { + /* 0 */ __le16 extmask; + /* 2 */ +} PACKED reiser4_stat_data_base; + +typedef struct reiser4_light_weight_stat { + /* 0 */ __le16 mode; + /* 2 */ __le32 nlink; + /* 6 */ __le64 size; + /* size in bytes */ + /* 14 */ +} PACKED reiser4_light_weight_stat; + +typedef struct reiser4_unix_stat { + /* owner id */ + /* 0 */ __le32 uid; + /* group id */ + /* 4 */ __le32 gid; + /* access time */ + /* 8 */ __le32 atime; + /* modification time */ + /* 12 */ __le32 mtime; + /* change time */ + /* 16 */ __le32 ctime; + union { + /* minor:major for device files */ + /* 20 */ __le64 rdev; + /* bytes used by file */ + /* 20 */ __le64 bytes; + } u; + /* 28 */ +} PACKED reiser4_unix_stat; + +/* symlink stored as part of inode */ +typedef struct reiser4_symlink_stat { + char body[0]; +} PACKED reiser4_symlink_stat; + +typedef struct reiser4_plugin_slot { + /* 0 */ __le16 pset_memb; + /* 2 */ __le16 id; + /* 4 *//* here plugin stores its persistent state */ +} PACKED reiser4_plugin_slot; + +/* stat-data extension for files with non-standard plugin. */ +typedef struct reiser4_plugin_stat { + /* number of additional plugins, associated with this object */ + /* 0 */ __le16 plugins_no; + /* 2 */ reiser4_plugin_slot slot[0]; + /* 2 */ +} PACKED reiser4_plugin_stat; + +/* stat-data extension for inode flags. Currently it is just fixed-width 32 + * bit mask. If need arise, this can be replaced with variable width + * bitmask. */ +typedef struct reiser4_flags_stat { + /* 0 */ __le32 flags; + /* 4 */ +} PACKED reiser4_flags_stat; + +typedef struct reiser4_capabilities_stat { + /* 0 */ __le32 effective; + /* 8 */ __le32 permitted; + /* 16 */ +} PACKED reiser4_capabilities_stat; + +typedef struct reiser4_cluster_stat { +/* this defines cluster size (an attribute of cryptcompress objects) as PAGE_SIZE << cluster shift */ + /* 0 */ d8 cluster_shift; + /* 1 */ +} PACKED reiser4_cluster_stat; + +typedef struct reiser4_crypto_stat { + /* secret key size, bits */ + /* 0 */ d16 keysize; + /* secret key id */ + /* 2 */ d8 keyid[0]; + /* 2 */ +} PACKED reiser4_crypto_stat; + +typedef struct reiser4_large_times_stat { + /* access time */ + /* 0 */ d32 atime; + /* modification time */ + /* 4 */ d32 mtime; + /* change time */ + /* 8 */ d32 ctime; + /* 12 */ +} PACKED reiser4_large_times_stat; + +/* this structure is filled by sd_item_stat */ +typedef struct sd_stat { + int dirs; + int files; + int others; +} sd_stat; + +/* plugin->item.common.* */ +extern void print_sd(const char *prefix, coord_t * coord); +extern void item_stat_static_sd(const coord_t * coord, void *vp); + +/* plugin->item.s.sd.* */ +extern int init_inode_static_sd(struct inode *inode, char *sd, int len); +extern int save_len_static_sd(struct inode *inode); +extern int save_static_sd(struct inode *inode, char **area); + +/* __FS_REISER4_PLUGIN_ITEM_STATIC_STAT_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/tail.c linux-2.6.30/fs/reiser4/plugin/item/tail.c --- linux-2.6.30.orig/fs/reiser4/plugin/item/tail.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/tail.c 2009-06-22 17:36:20.000000000 +0200 @@ -0,0 +1,807 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "item.h" +#include "../../inode.h" +#include "../../page_cache.h" +#include "../../carry.h" +#include "../../vfs_ops.h" + +#include +#include +#include +#include + +/* plugin->u.item.b.max_key_inside */ +reiser4_key *max_key_inside_tail(const coord_t *coord, reiser4_key *key) +{ + item_key_by_coord(coord, key); + set_key_offset(key, get_key_offset(reiser4_max_key())); + return key; +} + +/* plugin->u.item.b.can_contain_key */ +int can_contain_key_tail(const coord_t *coord, const reiser4_key *key, + const reiser4_item_data *data) +{ + reiser4_key item_key; + + if (item_plugin_by_coord(coord) != data->iplug) + return 0; + + item_key_by_coord(coord, &item_key); + if (get_key_locality(key) != get_key_locality(&item_key) || + get_key_objectid(key) != get_key_objectid(&item_key)) + return 0; + + return 1; +} + +/* plugin->u.item.b.mergeable + first item is of tail type */ +/* Audited by: green(2002.06.14) */ +int mergeable_tail(const coord_t *p1, const coord_t *p2) +{ + reiser4_key key1, key2; + + assert("vs-535", plugin_of_group(item_plugin_by_coord(p1), + UNIX_FILE_METADATA_ITEM_TYPE)); + assert("vs-365", item_id_by_coord(p1) == FORMATTING_ID); + + if (item_id_by_coord(p2) != FORMATTING_ID) { + /* second item is of another type */ + return 0; + } + + item_key_by_coord(p1, &key1); + item_key_by_coord(p2, &key2); + if (get_key_locality(&key1) != get_key_locality(&key2) || + get_key_objectid(&key1) != get_key_objectid(&key2) + || get_key_type(&key1) != get_key_type(&key2)) { + /* items of different objects */ + return 0; + } + if (get_key_offset(&key1) + nr_units_tail(p1) != get_key_offset(&key2)) { + /* not adjacent items */ + return 0; + } + return 1; +} + +/* plugin->u.item.b.print + plugin->u.item.b.check */ + +/* plugin->u.item.b.nr_units */ +pos_in_node_t nr_units_tail(const coord_t * coord) +{ + return item_length_by_coord(coord); +} + +/* plugin->u.item.b.lookup */ +lookup_result +lookup_tail(const reiser4_key * key, lookup_bias bias, coord_t * coord) +{ + reiser4_key item_key; + __u64 lookuped, offset; + unsigned nr_units; + + item_key_by_coord(coord, &item_key); + offset = get_key_offset(item_key_by_coord(coord, &item_key)); + nr_units = nr_units_tail(coord); + + /* key we are looking for must be greater than key of item @coord */ + assert("vs-416", keygt(key, &item_key)); + + /* offset we are looking for */ + lookuped = get_key_offset(key); + + if (lookuped >= offset && lookuped < offset + nr_units) { + /* byte we are looking for is in this item */ + coord->unit_pos = lookuped - offset; + coord->between = AT_UNIT; + return CBK_COORD_FOUND; + } + + /* set coord after last unit */ + coord->unit_pos = nr_units - 1; + coord->between = AFTER_UNIT; + return bias == + FIND_MAX_NOT_MORE_THAN ? CBK_COORD_FOUND : CBK_COORD_NOTFOUND; +} + +/* plugin->u.item.b.paste */ +int +paste_tail(coord_t *coord, reiser4_item_data *data, + carry_plugin_info *info UNUSED_ARG) +{ + unsigned old_item_length; + char *item; + + /* length the item had before resizing has been performed */ + old_item_length = item_length_by_coord(coord) - data->length; + + /* tail items never get pasted in the middle */ + assert("vs-363", + (coord->unit_pos == 0 && coord->between == BEFORE_UNIT) || + (coord->unit_pos == old_item_length - 1 && + coord->between == AFTER_UNIT) || + (coord->unit_pos == 0 && old_item_length == 0 + && coord->between == AT_UNIT)); + + item = item_body_by_coord(coord); + if (coord->unit_pos == 0) + /* make space for pasted data when pasting at the beginning of + the item */ + memmove(item + data->length, item, old_item_length); + + if (coord->between == AFTER_UNIT) + coord->unit_pos++; + + if (data->data) { + assert("vs-554", data->user == 0 || data->user == 1); + if (data->user) { + assert("nikita-3035", reiser4_schedulable()); + /* copy from user space */ + if (__copy_from_user(item + coord->unit_pos, + (const char __user *)data->data, + (unsigned)data->length)) + return RETERR(-EFAULT); + } else + /* copy from kernel space */ + memcpy(item + coord->unit_pos, data->data, + (unsigned)data->length); + } else { + memset(item + coord->unit_pos, 0, (unsigned)data->length); + } + return 0; +} + +/* plugin->u.item.b.fast_paste */ + +/* plugin->u.item.b.can_shift + number of units is returned via return value, number of bytes via @size. For + tail items they coincide */ +int +can_shift_tail(unsigned free_space, coord_t * source UNUSED_ARG, + znode * target UNUSED_ARG, shift_direction direction UNUSED_ARG, + unsigned *size, unsigned want) +{ + /* make sure that that we do not want to shift more than we have */ + assert("vs-364", want > 0 + && want <= (unsigned)item_length_by_coord(source)); + + *size = min(want, free_space); + return *size; +} + +/* plugin->u.item.b.copy_units */ +void +copy_units_tail(coord_t * target, coord_t * source, + unsigned from, unsigned count, + shift_direction where_is_free_space, + unsigned free_space UNUSED_ARG) +{ + /* make sure that item @target is expanded already */ + assert("vs-366", (unsigned)item_length_by_coord(target) >= count); + assert("vs-370", free_space >= count); + + if (where_is_free_space == SHIFT_LEFT) { + /* append item @target with @count first bytes of @source */ + assert("vs-365", from == 0); + + memcpy((char *)item_body_by_coord(target) + + item_length_by_coord(target) - count, + (char *)item_body_by_coord(source), count); + } else { + /* target item is moved to right already */ + reiser4_key key; + + assert("vs-367", + (unsigned)item_length_by_coord(source) == from + count); + + memcpy((char *)item_body_by_coord(target), + (char *)item_body_by_coord(source) + from, count); + + /* new units are inserted before first unit in an item, + therefore, we have to update item key */ + item_key_by_coord(source, &key); + set_key_offset(&key, get_key_offset(&key) + from); + + node_plugin_by_node(target->node)->update_item_key(target, &key, + NULL /*info */); + } +} + +/* plugin->u.item.b.create_hook */ + +/* item_plugin->b.kill_hook + this is called when @count units starting from @from-th one are going to be removed + */ +int +kill_hook_tail(const coord_t * coord, pos_in_node_t from, + pos_in_node_t count, struct carry_kill_data *kdata) +{ + reiser4_key key; + loff_t start, end; + + assert("vs-1577", kdata); + assert("vs-1579", kdata->inode); + + item_key_by_coord(coord, &key); + start = get_key_offset(&key) + from; + end = start + count; + fake_kill_hook_tail(kdata->inode, start, end, kdata->params.truncate); + return 0; +} + +/* plugin->u.item.b.shift_hook */ + +/* helper for kill_units_tail and cut_units_tail */ +static int +do_cut_or_kill(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + reiser4_key * smallest_removed, reiser4_key * new_first) +{ + pos_in_node_t count; + + /* this method is only called to remove part of item */ + assert("vs-374", (to - from + 1) < item_length_by_coord(coord)); + /* tails items are never cut from the middle of an item */ + assert("vs-396", ergo(from != 0, to == coord_last_unit_pos(coord))); + assert("vs-1558", ergo(from == 0, to < coord_last_unit_pos(coord))); + + count = to - from + 1; + + if (smallest_removed) { + /* store smallest key removed */ + item_key_by_coord(coord, smallest_removed); + set_key_offset(smallest_removed, + get_key_offset(smallest_removed) + from); + } + if (new_first) { + /* head of item is cut */ + assert("vs-1529", from == 0); + + item_key_by_coord(coord, new_first); + set_key_offset(new_first, + get_key_offset(new_first) + from + count); + } + + if (REISER4_DEBUG) + memset((char *)item_body_by_coord(coord) + from, 0, count); + return count; +} + +/* plugin->u.item.b.cut_units */ +int +cut_units_tail(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_cut_data *cdata UNUSED_ARG, + reiser4_key * smallest_removed, reiser4_key * new_first) +{ + return do_cut_or_kill(coord, from, to, smallest_removed, new_first); +} + +/* plugin->u.item.b.kill_units */ +int +kill_units_tail(coord_t * coord, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *kdata, reiser4_key * smallest_removed, + reiser4_key * new_first) +{ + kill_hook_tail(coord, from, to - from + 1, kdata); + return do_cut_or_kill(coord, from, to, smallest_removed, new_first); +} + +/* plugin->u.item.b.unit_key */ +reiser4_key *unit_key_tail(const coord_t * coord, reiser4_key * key) +{ + assert("vs-375", coord_is_existing_unit(coord)); + + item_key_by_coord(coord, key); + set_key_offset(key, (get_key_offset(key) + coord->unit_pos)); + + return key; +} + +/* plugin->u.item.b.estimate + plugin->u.item.b.item_data_by_flow */ + +/* tail redpage function. It is called from readpage_tail(). */ +static int do_readpage_tail(uf_coord_t *uf_coord, struct page *page) +{ + tap_t tap; + int result; + coord_t coord; + lock_handle lh; + int count, mapped; + struct inode *inode; + char *pagedata; + + /* saving passed coord in order to do not move it by tap. */ + init_lh(&lh); + copy_lh(&lh, uf_coord->lh); + inode = page->mapping->host; + coord_dup(&coord, &uf_coord->coord); + + reiser4_tap_init(&tap, &coord, &lh, ZNODE_READ_LOCK); + + if ((result = reiser4_tap_load(&tap))) + goto out_tap_done; + + /* lookup until page is filled up. */ + for (mapped = 0; mapped < PAGE_CACHE_SIZE; ) { + /* number of bytes to be copied to page */ + count = item_length_by_coord(&coord) - coord.unit_pos; + if (count > PAGE_CACHE_SIZE - mapped) + count = PAGE_CACHE_SIZE - mapped; + + /* attach @page to address space and get data address */ + pagedata = kmap_atomic(page, KM_USER0); + + /* copy tail item to page */ + memcpy(pagedata + mapped, + ((char *)item_body_by_coord(&coord) + coord.unit_pos), + count); + mapped += count; + + flush_dcache_page(page); + + /* dettach page from address space */ + kunmap_atomic(pagedata, KM_USER0); + + /* Getting next tail item. */ + if (mapped < PAGE_CACHE_SIZE) { + /* + * unlock page in order to avoid keep it locked + * during tree lookup, which takes long term locks + */ + unlock_page(page); + + /* getting right neighbour. */ + result = go_dir_el(&tap, RIGHT_SIDE, 0); + + /* lock page back */ + lock_page(page); + if (PageUptodate(page)) { + /* + * another thread read the page, we have + * nothing to do + */ + result = 0; + goto out_unlock_page; + } + + if (result) { + if (result == -E_NO_NEIGHBOR) { + /* + * rigth neighbor is not a formatted + * node + */ + result = 0; + goto done; + } else { + goto out_tap_relse; + } + } else { + if (!inode_file_plugin(inode)-> + owns_item(inode, &coord)) { + /* item of another file is found */ + result = 0; + goto done; + } + } + } + } + + done: + if (mapped != PAGE_CACHE_SIZE) + zero_user_segment(page, mapped, PAGE_CACHE_SIZE); + SetPageUptodate(page); + out_unlock_page: + unlock_page(page); + out_tap_relse: + reiser4_tap_relse(&tap); + out_tap_done: + reiser4_tap_done(&tap); + return result; +} + +/* + plugin->s.file.readpage + reiser4_read->unix_file_read->page_cache_readahead->reiser4_readpage->unix_file_readpage->readpage_tail + or + filemap_fault->reiser4_readpage->readpage_unix_file->->readpage_tail + + At the beginning: coord->node is read locked, zloaded, page is locked, coord is set to existing unit inside of tail + item. */ +int readpage_tail(void *vp, struct page *page) +{ + uf_coord_t *uf_coord = vp; + ON_DEBUG(coord_t * coord = &uf_coord->coord); + ON_DEBUG(reiser4_key key); + + assert("umka-2515", PageLocked(page)); + assert("umka-2516", !PageUptodate(page)); + assert("umka-2517", !jprivate(page) && !PagePrivate(page)); + assert("umka-2518", page->mapping && page->mapping->host); + + assert("umka-2519", znode_is_loaded(coord->node)); + assert("umka-2520", item_is_tail(coord)); + assert("umka-2521", coord_is_existing_unit(coord)); + assert("umka-2522", znode_is_rlocked(coord->node)); + assert("umka-2523", + page->mapping->host->i_ino == + get_key_objectid(item_key_by_coord(coord, &key))); + + return do_readpage_tail(uf_coord, page); +} + +/** + * overwrite_tail + * @flow: + * @coord: + * + * Overwrites tail item or its part by user data. Returns number of bytes + * written or error code. + */ +static int overwrite_tail(flow_t *flow, coord_t *coord) +{ + unsigned count; + + assert("vs-570", flow->user == 1); + assert("vs-946", flow->data); + assert("vs-947", coord_is_existing_unit(coord)); + assert("vs-948", znode_is_write_locked(coord->node)); + assert("nikita-3036", reiser4_schedulable()); + + count = item_length_by_coord(coord) - coord->unit_pos; + if (count > flow->length) + count = flow->length; + + if (__copy_from_user((char *)item_body_by_coord(coord) + coord->unit_pos, + (const char __user *)flow->data, count)) + return RETERR(-EFAULT); + + znode_make_dirty(coord->node); + return count; +} + +/** + * insert_first_tail + * @inode: + * @flow: + * @coord: + * @lh: + * + * Returns number of bytes written or error code. + */ +static ssize_t insert_first_tail(struct inode *inode, flow_t *flow, + coord_t *coord, lock_handle *lh) +{ + int result; + loff_t to_write; + struct unix_file_info *uf_info; + + if (get_key_offset(&flow->key) != 0) { + /* + * file is empty and we have to write not to the beginning of + * file. Create a hole at the beginning of file. On success + * insert_flow returns 0 as number of written bytes which is + * what we have to return on padding a file with holes + */ + flow->data = NULL; + flow->length = get_key_offset(&flow->key); + set_key_offset(&flow->key, 0); + /* + * holes in files built of tails are stored just like if there + * were real data which are all zeros. Therefore we have to + * allocate quota here as well + */ + if (vfs_dq_alloc_space_nodirty(inode, flow->length)) + return RETERR(-EDQUOT); + result = reiser4_insert_flow(coord, lh, flow); + if (flow->length) + vfs_dq_free_space_nodirty(inode, flow->length); + + uf_info = unix_file_inode_data(inode); + + /* + * first item insertion is only possible when writing to empty + * file or performing tail conversion + */ + assert("", (uf_info->container == UF_CONTAINER_EMPTY || + (reiser4_inode_get_flag(inode, + REISER4_PART_MIXED) && + reiser4_inode_get_flag(inode, + REISER4_PART_IN_CONV)))); + /* if file was empty - update its state */ + if (result == 0 && uf_info->container == UF_CONTAINER_EMPTY) + uf_info->container = UF_CONTAINER_TAILS; + return result; + } + + /* check quota before appending data */ + if (vfs_dq_alloc_space_nodirty(inode, flow->length)) + return RETERR(-EDQUOT); + + to_write = flow->length; + result = reiser4_insert_flow(coord, lh, flow); + if (flow->length) + vfs_dq_free_space_nodirty(inode, flow->length); + return (to_write - flow->length) ? (to_write - flow->length) : result; +} + +/** + * append_tail + * @inode: + * @flow: + * @coord: + * @lh: + * + * Returns number of bytes written or error code. + */ +static ssize_t append_tail(struct inode *inode, + flow_t *flow, coord_t *coord, lock_handle *lh) +{ + int result; + reiser4_key append_key; + loff_t to_write; + + if (!keyeq(&flow->key, append_key_tail(coord, &append_key))) { + flow->data = NULL; + flow->length = get_key_offset(&flow->key) - get_key_offset(&append_key); + set_key_offset(&flow->key, get_key_offset(&append_key)); + /* + * holes in files built of tails are stored just like if there + * were real data which are all zeros. Therefore we have to + * allocate quota here as well + */ + if (vfs_dq_alloc_space_nodirty(inode, flow->length)) + return RETERR(-EDQUOT); + result = reiser4_insert_flow(coord, lh, flow); + if (flow->length) + vfs_dq_free_space_nodirty(inode, flow->length); + return result; + } + + /* check quota before appending data */ + if (vfs_dq_alloc_space_nodirty(inode, flow->length)) + return RETERR(-EDQUOT); + + to_write = flow->length; + result = reiser4_insert_flow(coord, lh, flow); + if (flow->length) + vfs_dq_free_space_nodirty(inode, flow->length); + return (to_write - flow->length) ? (to_write - flow->length) : result; +} + +/** + * write_tail_reserve_space - reserve space for tail write operation + * @inode: + * + * Estimates and reserves space which may be required for writing one flow to a + * file + */ +static int write_extent_reserve_space(struct inode *inode) +{ + __u64 count; + reiser4_tree *tree; + + /* + * to write one flow to a file by tails we have to reserve disk space for: + + * 1. find_file_item may have to insert empty node to the tree (empty + * leaf node between two extent items). This requires 1 block and + * number of blocks which are necessary to perform insertion of an + * internal item into twig level. + * + * 2. flow insertion + * + * 3. stat data update + */ + tree = reiser4_tree_by_inode(inode); + count = estimate_one_insert_item(tree) + + estimate_insert_flow(tree->height) + + estimate_one_insert_item(tree); + grab_space_enable(); + return reiser4_grab_space(count, 0 /* flags */); +} + +#define PAGE_PER_FLOW 4 + +static loff_t faultin_user_pages(const char __user *buf, size_t count) +{ + loff_t faulted; + int to_fault; + + if (count > PAGE_PER_FLOW * PAGE_CACHE_SIZE) + count = PAGE_PER_FLOW * PAGE_CACHE_SIZE; + faulted = 0; + while (count > 0) { + to_fault = PAGE_CACHE_SIZE; + if (count < to_fault) + to_fault = count; + fault_in_pages_readable(buf + faulted, to_fault); + count -= to_fault; + faulted += to_fault; + } + return faulted; +} + +/** + * reiser4_write_tail - write method of tail item plugin + * @file: file to write to + * @buf: address of user-space buffer + * @count: number of bytes to write + * @pos: position in file to write to + * + * Returns number of written bytes or error code. + */ +ssize_t reiser4_write_tail(struct file *file, struct inode * inode, + const char __user *buf, size_t count, loff_t *pos) +{ + struct hint hint; + int result; + flow_t flow; + coord_t *coord; + lock_handle *lh; + znode *loaded; + + assert("edward-1548", inode != NULL); + + if (write_extent_reserve_space(inode)) + return RETERR(-ENOSPC); + + result = load_file_hint(file, &hint); + BUG_ON(result != 0); + + flow.length = faultin_user_pages(buf, count); + flow.user = 1; + memcpy(&flow.data, &buf, sizeof(buf)); + flow.op = WRITE_OP; + key_by_inode_and_offset_common(inode, *pos, &flow.key); + + result = find_file_item(&hint, &flow.key, ZNODE_WRITE_LOCK, inode); + if (IS_CBKERR(result)) + return result; + + coord = &hint.ext_coord.coord; + lh = hint.ext_coord.lh; + + result = zload(coord->node); + BUG_ON(result != 0); + loaded = coord->node; + + if (coord->between == AFTER_UNIT) { + /* append with data or hole */ + result = append_tail(inode, &flow, coord, lh); + } else if (coord->between == AT_UNIT) { + /* overwrite */ + result = overwrite_tail(&flow, coord); + } else { + /* no items of this file yet. insert data or hole */ + result = insert_first_tail(inode, &flow, coord, lh); + } + zrelse(loaded); + if (result < 0) { + done_lh(lh); + return result; + } + + /* seal and unlock znode */ + hint.ext_coord.valid = 0; + if (hint.ext_coord.valid) + reiser4_set_hint(&hint, &flow.key, ZNODE_WRITE_LOCK); + else + reiser4_unset_hint(&hint); + + save_file_hint(file, &hint); + return result; +} + +#if REISER4_DEBUG + +static int +coord_matches_key_tail(const coord_t * coord, const reiser4_key * key) +{ + reiser4_key item_key; + + assert("vs-1356", coord_is_existing_unit(coord)); + assert("vs-1354", keylt(key, append_key_tail(coord, &item_key))); + assert("vs-1355", keyge(key, item_key_by_coord(coord, &item_key))); + return get_key_offset(key) == + get_key_offset(&item_key) + coord->unit_pos; + +} + +#endif + +/* plugin->u.item.s.file.read */ +int reiser4_read_tail(struct file *file UNUSED_ARG, flow_t *f, hint_t *hint) +{ + unsigned count; + int item_length; + coord_t *coord; + uf_coord_t *uf_coord; + + uf_coord = &hint->ext_coord; + coord = &uf_coord->coord; + + assert("vs-571", f->user == 1); + assert("vs-571", f->data); + assert("vs-967", coord && coord->node); + assert("vs-1117", znode_is_rlocked(coord->node)); + assert("vs-1118", znode_is_loaded(coord->node)); + + assert("nikita-3037", reiser4_schedulable()); + assert("vs-1357", coord_matches_key_tail(coord, &f->key)); + + /* calculate number of bytes to read off the item */ + item_length = item_length_by_coord(coord); + count = item_length_by_coord(coord) - coord->unit_pos; + if (count > f->length) + count = f->length; + + /* user page has to be brought in so that major page fault does not + * occur here when longtem lock is held */ + if (__copy_to_user((char __user *)f->data, + ((char *)item_body_by_coord(coord) + coord->unit_pos), + count)) + return RETERR(-EFAULT); + + /* probably mark_page_accessed() should only be called if + * coord->unit_pos is zero. */ + mark_page_accessed(znode_page(coord->node)); + move_flow_forward(f, count); + + coord->unit_pos += count; + if (item_length == coord->unit_pos) { + coord->unit_pos--; + coord->between = AFTER_UNIT; + } + reiser4_set_hint(hint, &f->key, ZNODE_READ_LOCK); + return 0; +} + +/* + plugin->u.item.s.file.append_key + key of first byte which is the next to last byte by addressed by this item +*/ +reiser4_key *append_key_tail(const coord_t * coord, reiser4_key * key) +{ + item_key_by_coord(coord, key); + set_key_offset(key, get_key_offset(key) + item_length_by_coord(coord)); + return key; +} + +/* plugin->u.item.s.file.init_coord_extension */ +void init_coord_extension_tail(uf_coord_t * uf_coord, loff_t lookuped) +{ + uf_coord->valid = 1; +} + +/* + plugin->u.item.s.file.get_block +*/ +int +get_block_address_tail(const coord_t * coord, sector_t lblock, sector_t * block) +{ + assert("nikita-3252", znode_get_level(coord->node) == LEAF_LEVEL); + + if (reiser4_blocknr_is_fake(znode_get_block(coord->node))) + /* if node has'nt obtainet its block number yet, return 0. + * Lets avoid upsetting users with some cosmic numbers beyond + * the device capacity.*/ + *block = 0; + else + *block = *znode_get_block(coord->node); + return 0; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/item/tail.h linux-2.6.30/fs/reiser4/plugin/item/tail.h --- linux-2.6.30.orig/fs/reiser4/plugin/item/tail.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/item/tail.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,58 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#if !defined( __REISER4_TAIL_H__ ) +#define __REISER4_TAIL_H__ + +struct tail_coord_extension { + int not_used; +}; + +struct cut_list; + +/* plugin->u.item.b.* */ +reiser4_key *max_key_inside_tail(const coord_t *, reiser4_key *); +int can_contain_key_tail(const coord_t * coord, const reiser4_key * key, + const reiser4_item_data *); +int mergeable_tail(const coord_t * p1, const coord_t * p2); +pos_in_node_t nr_units_tail(const coord_t *); +lookup_result lookup_tail(const reiser4_key *, lookup_bias, coord_t *); +int paste_tail(coord_t *, reiser4_item_data *, carry_plugin_info *); +int can_shift_tail(unsigned free_space, coord_t * source, + znode * target, shift_direction, unsigned *size, + unsigned want); +void copy_units_tail(coord_t * target, coord_t * source, unsigned from, + unsigned count, shift_direction, unsigned free_space); +int kill_hook_tail(const coord_t *, pos_in_node_t from, pos_in_node_t count, + struct carry_kill_data *); +int cut_units_tail(coord_t *, pos_in_node_t from, pos_in_node_t to, + struct carry_cut_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +int kill_units_tail(coord_t *, pos_in_node_t from, pos_in_node_t to, + struct carry_kill_data *, reiser4_key * smallest_removed, + reiser4_key * new_first); +reiser4_key *unit_key_tail(const coord_t *, reiser4_key *); + +/* plugin->u.item.s.* */ +ssize_t reiser4_write_tail(struct file *file, struct inode * inode, + const char __user *buf, size_t count, loff_t *pos); +int reiser4_read_tail(struct file *, flow_t *, hint_t *); +int readpage_tail(void *vp, struct page *page); +reiser4_key *append_key_tail(const coord_t *, reiser4_key *); +void init_coord_extension_tail(uf_coord_t *, loff_t offset); +int get_block_address_tail(const coord_t *, sector_t, sector_t *); +int item_balance_dirty_pages(struct address_space *, const flow_t *, + hint_t *, int back_to_dirty, int set_hint); + +/* __REISER4_TAIL_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/Makefile linux-2.6.30/fs/reiser4/plugin/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,26 @@ +obj-$(CONFIG_REISER4_FS) += plugins.o + +plugins-objs := \ + plugin.o \ + plugin_set.o \ + object.o \ + inode_ops.o \ + inode_ops_rename.o \ + file_ops.o \ + file_ops_readdir.o \ + file_plugin_common.o \ + dir_plugin_common.o \ + digest.o \ + hash.o \ + fibration.o \ + tail_policy.o \ + regular.o + +obj-$(CONFIG_REISER4_FS) += item/ +obj-$(CONFIG_REISER4_FS) += file/ +obj-$(CONFIG_REISER4_FS) += dir/ +obj-$(CONFIG_REISER4_FS) += node/ +obj-$(CONFIG_REISER4_FS) += compress/ +obj-$(CONFIG_REISER4_FS) += space/ +obj-$(CONFIG_REISER4_FS) += disk_format/ +obj-$(CONFIG_REISER4_FS) += security/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/node/Makefile linux-2.6.30/fs/reiser4/plugin/node/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/node/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/node/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,5 @@ +obj-$(CONFIG_REISER4_FS) += node_plugins.o + +node_plugins-objs := \ + node.o \ + node40.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/node/node40.c linux-2.6.30/fs/reiser4/plugin/node/node40.c --- linux-2.6.30.orig/fs/reiser4/plugin/node/node40.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/node/node40.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,2924 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "../../debug.h" +#include "../../key.h" +#include "../../coord.h" +#include "../plugin_header.h" +#include "../item/item.h" +#include "node.h" +#include "node40.h" +#include "../plugin.h" +#include "../../jnode.h" +#include "../../znode.h" +#include "../../pool.h" +#include "../../carry.h" +#include "../../tap.h" +#include "../../tree.h" +#include "../../super.h" +#include "../../reiser4.h" + +#include +#include +#include + +/* leaf 40 format: + + [node header | item 0, item 1, .., item N-1 | free space | item_head N-1, .. item_head 1, item head 0 ] + plugin_id (16) key + free_space (16) pluginid (16) + free_space_start (16) offset (16) + level (8) + num_items (16) + magic (32) + flush_time (32) +*/ +/* NIKITA-FIXME-HANS: I told you guys not less than 10 times to not call it r4fs. Change to "ReIs". */ +/* magic number that is stored in ->magic field of node header */ +static const __u32 REISER4_NODE_MAGIC = 0x52344653; /* (*(__u32 *)"R4FS"); */ + +static int prepare_for_update(znode * left, znode * right, + carry_plugin_info * info); + +/* header of node of reiser40 format is at the beginning of node */ +static inline node40_header *node40_node_header(const znode * node /* node to + * query */ ) +{ + assert("nikita-567", node != NULL); + assert("nikita-568", znode_page(node) != NULL); + assert("nikita-569", zdata(node) != NULL); + return (node40_header *) zdata(node); +} + +/* functions to get/set fields of node40_header */ +#define nh40_get_magic(nh) le32_to_cpu(get_unaligned(&(nh)->magic)) +#define nh40_get_free_space(nh) le16_to_cpu(get_unaligned(&(nh)->free_space)) +#define nh40_get_free_space_start(nh) le16_to_cpu(get_unaligned(&(nh)->free_space_start)) +#define nh40_get_level(nh) get_unaligned(&(nh)->level) +#define nh40_get_num_items(nh) le16_to_cpu(get_unaligned(&(nh)->nr_items)) +#define nh40_get_flush_id(nh) le64_to_cpu(get_unaligned(&(nh)->flush_id)) + +#define nh40_set_magic(nh, value) put_unaligned(cpu_to_le32(value), &(nh)->magic) +#define nh40_set_free_space(nh, value) put_unaligned(cpu_to_le16(value), &(nh)->free_space) +#define nh40_set_free_space_start(nh, value) put_unaligned(cpu_to_le16(value), &(nh)->free_space_start) +#define nh40_set_level(nh, value) put_unaligned(value, &(nh)->level) +#define nh40_set_num_items(nh, value) put_unaligned(cpu_to_le16(value), &(nh)->nr_items) +#define nh40_set_mkfs_id(nh, value) put_unaligned(cpu_to_le32(value), &(nh)->mkfs_id) + +/* plugin field of node header should be read/set by + plugin_by_disk_id/save_disk_plugin */ + +/* array of item headers is at the end of node */ +static inline item_header40 *node40_ih_at(const znode * node, unsigned pos) +{ + return (item_header40 *) (zdata(node) + znode_size(node)) - pos - 1; +} + +/* ( page_address( node -> pg ) + PAGE_CACHE_SIZE ) - pos - 1 + */ +static inline item_header40 *node40_ih_at_coord(const coord_t * coord) +{ + return (item_header40 *) (zdata(coord->node) + + znode_size(coord->node)) - (coord->item_pos) - + 1; +} + +/* functions to get/set fields of item_header40 */ +#define ih40_get_offset(ih) le16_to_cpu(get_unaligned(&(ih)->offset)) + +#define ih40_set_offset(ih, value) put_unaligned(cpu_to_le16(value), &(ih)->offset) + +/* plugin field of item header should be read/set by + plugin_by_disk_id/save_disk_plugin */ + +/* plugin methods */ + +/* plugin->u.node.item_overhead + look for description of this method in plugin/node/node.h */ +size_t +item_overhead_node40(const znode * node UNUSED_ARG, flow_t * f UNUSED_ARG) +{ + return sizeof(item_header40); +} + +/* plugin->u.node.free_space + look for description of this method in plugin/node/node.h */ +size_t free_space_node40(znode * node) +{ + assert("nikita-577", node != NULL); + assert("nikita-578", znode_is_loaded(node)); + assert("nikita-579", zdata(node) != NULL); + + return nh40_get_free_space(node40_node_header(node)); +} + +/* private inline version of node40_num_of_items() for use in this file. This + is necessary, because address of node40_num_of_items() is taken and it is + never inlined as a result. */ +static inline short node40_num_of_items_internal(const znode * node) +{ + return nh40_get_num_items(node40_node_header(node)); +} + +#if REISER4_DEBUG +static inline void check_num_items(const znode * node) +{ + assert("nikita-2749", + node40_num_of_items_internal(node) == node->nr_items); + assert("nikita-2746", znode_is_write_locked(node)); +} +#else +#define check_num_items(node) noop +#endif + +/* plugin->u.node.num_of_items + look for description of this method in plugin/node/node.h */ +int num_of_items_node40(const znode * node) +{ + return node40_num_of_items_internal(node); +} + +static void +node40_set_num_items(znode * node, node40_header * nh, unsigned value) +{ + assert("nikita-2751", node != NULL); + assert("nikita-2750", nh == node40_node_header(node)); + + check_num_items(node); + nh40_set_num_items(nh, value); + node->nr_items = value; + check_num_items(node); +} + +/* plugin->u.node.item_by_coord + look for description of this method in plugin/node/node.h */ +char *item_by_coord_node40(const coord_t * coord) +{ + item_header40 *ih; + char *p; + + /* @coord is set to existing item */ + assert("nikita-596", coord != NULL); + assert("vs-255", coord_is_existing_item(coord)); + + ih = node40_ih_at_coord(coord); + p = zdata(coord->node) + ih40_get_offset(ih); + return p; +} + +/* plugin->u.node.length_by_coord + look for description of this method in plugin/node/node.h */ +int length_by_coord_node40(const coord_t * coord) +{ + item_header40 *ih; + int result; + + /* @coord is set to existing item */ + assert("vs-256", coord != NULL); + assert("vs-257", coord_is_existing_item(coord)); + + ih = node40_ih_at_coord(coord); + if ((int)coord->item_pos == + node40_num_of_items_internal(coord->node) - 1) + result = + nh40_get_free_space_start(node40_node_header(coord->node)) - + ih40_get_offset(ih); + else + result = ih40_get_offset(ih - 1) - ih40_get_offset(ih); + + return result; +} + +static pos_in_node_t +node40_item_length(const znode * node, pos_in_node_t item_pos) +{ + item_header40 *ih; + pos_in_node_t result; + + /* @coord is set to existing item */ + assert("vs-256", node != NULL); + assert("vs-257", node40_num_of_items_internal(node) > item_pos); + + ih = node40_ih_at(node, item_pos); + if (item_pos == node40_num_of_items_internal(node) - 1) + result = + nh40_get_free_space_start(node40_node_header(node)) - + ih40_get_offset(ih); + else + result = ih40_get_offset(ih - 1) - ih40_get_offset(ih); + + return result; +} + +/* plugin->u.node.plugin_by_coord + look for description of this method in plugin/node/node.h */ +item_plugin *plugin_by_coord_node40(const coord_t * coord) +{ + item_header40 *ih; + item_plugin *result; + + /* @coord is set to existing item */ + assert("vs-258", coord != NULL); + assert("vs-259", coord_is_existing_item(coord)); + + ih = node40_ih_at_coord(coord); + /* pass NULL in stead of current tree. This is time critical call. */ + result = item_plugin_by_disk_id(NULL, &ih->plugin_id); + return result; +} + +/* plugin->u.node.key_at + look for description of this method in plugin/node/node.h */ +reiser4_key *key_at_node40(const coord_t * coord, reiser4_key * key) +{ + item_header40 *ih; + + assert("nikita-1765", coord_is_existing_item(coord)); + + /* @coord is set to existing item */ + ih = node40_ih_at_coord(coord); + memcpy(key, &ih->key, sizeof(reiser4_key)); + return key; +} + +/* VS-FIXME-HANS: please review whether the below are properly disabled when debugging is disabled */ + +#define NODE_INCSTAT(n, counter) \ + reiser4_stat_inc_at_level(znode_get_level(n), node.lookup.counter) + +#define NODE_ADDSTAT(n, counter, val) \ + reiser4_stat_add_at_level(znode_get_level(n), node.lookup.counter, val) + +/* plugin->u.node.lookup + look for description of this method in plugin/node/node.h */ +node_search_result lookup_node40(znode * node /* node to query */ , + const reiser4_key * key /* key to look for */ , + lookup_bias bias /* search bias */ , + coord_t * coord /* resulting coord */ ) +{ + int left; + int right; + int found; + int items; + + item_header40 *lefth; + item_header40 *righth; + + item_plugin *iplug; + item_header40 *bstop; + item_header40 *ih; + cmp_t order; + + assert("nikita-583", node != NULL); + assert("nikita-584", key != NULL); + assert("nikita-585", coord != NULL); + assert("nikita-2693", znode_is_any_locked(node)); + cassert(REISER4_SEQ_SEARCH_BREAK > 2); + + items = node_num_items(node); + + if (unlikely(items == 0)) { + coord_init_first_unit(coord, node); + return NS_NOT_FOUND; + } + + /* binary search for item that can contain given key */ + left = 0; + right = items - 1; + coord->node = node; + coord_clear_iplug(coord); + found = 0; + + lefth = node40_ih_at(node, left); + righth = node40_ih_at(node, right); + + /* It is known that for small arrays sequential search is on average + more efficient than binary. This is because sequential search is + coded as tight loop that can be better optimized by compilers and + for small array size gain from this optimization makes sequential + search the winner. Another, maybe more important, reason for this, + is that sequential array is more CPU cache friendly, whereas binary + search effectively destroys CPU caching. + + Critical here is the notion of "smallness". Reasonable value of + REISER4_SEQ_SEARCH_BREAK can be found by playing with code in + fs/reiser4/ulevel/ulevel.c:test_search(). + + Don't try to further optimize sequential search by scanning from + right to left in attempt to use more efficient loop termination + condition (comparison with 0). This doesn't work. + + */ + + while (right - left >= REISER4_SEQ_SEARCH_BREAK) { + int median; + item_header40 *medianh; + + median = (left + right) / 2; + medianh = node40_ih_at(node, median); + + assert("nikita-1084", median >= 0); + assert("nikita-1085", median < items); + switch (keycmp(key, &medianh->key)) { + case LESS_THAN: + right = median; + righth = medianh; + break; + default: + wrong_return_value("nikita-586", "keycmp"); + case GREATER_THAN: + left = median; + lefth = medianh; + break; + case EQUAL_TO: + do { + --median; + /* headers are ordered from right to left */ + ++medianh; + } while (median >= 0 && keyeq(key, &medianh->key)); + right = left = median + 1; + ih = lefth = righth = medianh - 1; + found = 1; + break; + } + } + /* sequential scan. Item headers, and, therefore, keys are stored at + the rightmost part of a node from right to left. We are trying to + access memory from left to right, and hence, scan in _descending_ + order of item numbers. + */ + if (!found) { + for (left = right, ih = righth; left >= 0; ++ih, --left) { + cmp_t comparison; + + prefetchkey(&(ih + 1)->key); + comparison = keycmp(&ih->key, key); + if (comparison == GREATER_THAN) + continue; + if (comparison == EQUAL_TO) { + found = 1; + do { + --left; + ++ih; + } while (left >= 0 && keyeq(&ih->key, key)); + ++left; + --ih; + } else { + assert("nikita-1256", comparison == LESS_THAN); + } + break; + } + if (unlikely(left < 0)) + left = 0; + } + + assert("nikita-3212", right >= left); + assert("nikita-3214", + equi(found, keyeq(&node40_ih_at(node, left)->key, key))); + + coord_set_item_pos(coord, left); + coord->unit_pos = 0; + coord->between = AT_UNIT; + + /* key < leftmost key in a mode or node is corrupted and keys + are not sorted */ + bstop = node40_ih_at(node, (unsigned)left); + order = keycmp(&bstop->key, key); + if (unlikely(order == GREATER_THAN)) { + if (unlikely(left != 0)) { + /* screw up */ + warning("nikita-587", "Key less than %i key in a node", + left); + reiser4_print_key("key", key); + reiser4_print_key("min", &bstop->key); + print_coord_content("coord", coord); + return RETERR(-EIO); + } else { + coord->between = BEFORE_UNIT; + return NS_NOT_FOUND; + } + } + /* left <= key, ok */ + iplug = item_plugin_by_disk_id(znode_get_tree(node), &bstop->plugin_id); + + if (unlikely(iplug == NULL)) { + warning("nikita-588", "Unknown plugin %i", + le16_to_cpu(get_unaligned(&bstop->plugin_id))); + reiser4_print_key("key", key); + print_coord_content("coord", coord); + return RETERR(-EIO); + } + + coord_set_iplug(coord, iplug); + + /* if exact key from item header was found by binary search, no + further checks are necessary. */ + if (found) { + assert("nikita-1259", order == EQUAL_TO); + return NS_FOUND; + } + if (iplug->b.max_key_inside != NULL) { + reiser4_key max_item_key; + + /* key > max_item_key --- outside of an item */ + if (keygt(key, iplug->b.max_key_inside(coord, &max_item_key))) { + coord->unit_pos = 0; + coord->between = AFTER_ITEM; + /* FIXME-VS: key we are looking for does not fit into + found item. Return NS_NOT_FOUND then. Without that + the following case does not work: there is extent of + file 10000, 10001. File 10000, 10002 has been just + created. When writing to position 0 in that file - + traverse_tree will stop here on twig level. When we + want it to go down to leaf level + */ + return NS_NOT_FOUND; + } + } + + if (iplug->b.lookup != NULL) { + return iplug->b.lookup(key, bias, coord); + } else { + assert("nikita-1260", order == LESS_THAN); + coord->between = AFTER_UNIT; + return (bias == FIND_EXACT) ? NS_NOT_FOUND : NS_FOUND; + } +} + +#undef NODE_ADDSTAT +#undef NODE_INCSTAT + +/* plugin->u.node.estimate + look for description of this method in plugin/node/node.h */ +size_t estimate_node40(znode * node) +{ + size_t result; + + assert("nikita-597", node != NULL); + + result = free_space_node40(node) - sizeof(item_header40); + + return (result > 0) ? result : 0; +} + +/* plugin->u.node.check + look for description of this method in plugin/node/node.h */ +int check_node40(const znode * node /* node to check */ , + __u32 flags /* check flags */ , + const char **error /* where to store error message */ ) +{ + int nr_items; + int i; + reiser4_key prev; + unsigned old_offset; + tree_level level; + coord_t coord; + int result; + + assert("nikita-580", node != NULL); + assert("nikita-581", error != NULL); + assert("nikita-2948", znode_is_loaded(node)); + + if (ZF_ISSET(node, JNODE_HEARD_BANSHEE)) + return 0; + + assert("nikita-582", zdata(node) != NULL); + + nr_items = node40_num_of_items_internal(node); + if (nr_items < 0) { + *error = "Negative number of items"; + return -1; + } + + if (flags & REISER4_NODE_DKEYS) + prev = *znode_get_ld_key((znode *) node); + else + prev = *reiser4_min_key(); + + old_offset = 0; + coord_init_zero(&coord); + coord.node = (znode *) node; + coord.unit_pos = 0; + coord.between = AT_UNIT; + level = znode_get_level(node); + for (i = 0; i < nr_items; i++) { + item_header40 *ih; + reiser4_key unit_key; + unsigned j; + + ih = node40_ih_at(node, (unsigned)i); + coord_set_item_pos(&coord, i); + if ((ih40_get_offset(ih) >= + znode_size(node) - nr_items * sizeof(item_header40)) || + (ih40_get_offset(ih) < sizeof(node40_header))) { + *error = "Offset is out of bounds"; + return -1; + } + if (ih40_get_offset(ih) <= old_offset) { + *error = "Offsets are in wrong order"; + return -1; + } + if ((i == 0) && (ih40_get_offset(ih) != sizeof(node40_header))) { + *error = "Wrong offset of first item"; + return -1; + } + old_offset = ih40_get_offset(ih); + + if (keygt(&prev, &ih->key)) { + *error = "Keys are in wrong order"; + return -1; + } + if (!keyeq(&ih->key, unit_key_by_coord(&coord, &unit_key))) { + *error = "Wrong key of first unit"; + return -1; + } + prev = ih->key; + for (j = 0; j < coord_num_units(&coord); ++j) { + coord.unit_pos = j; + unit_key_by_coord(&coord, &unit_key); + if (keygt(&prev, &unit_key)) { + *error = "Unit keys are in wrong order"; + return -1; + } + prev = unit_key; + } + coord.unit_pos = 0; + if (level != TWIG_LEVEL && item_is_extent(&coord)) { + *error = "extent on the wrong level"; + return -1; + } + if (level == LEAF_LEVEL && item_is_internal(&coord)) { + *error = "internal item on the wrong level"; + return -1; + } + if (level != LEAF_LEVEL && + !item_is_internal(&coord) && !item_is_extent(&coord)) { + *error = "wrong item on the internal level"; + return -1; + } + if (level > TWIG_LEVEL && !item_is_internal(&coord)) { + *error = "non-internal item on the internal level"; + return -1; + } +#if REISER4_DEBUG + if (item_plugin_by_coord(&coord)->b.check + && item_plugin_by_coord(&coord)->b.check(&coord, error)) + return -1; +#endif + if (i) { + coord_t prev_coord; + /* two neighboring items can not be mergeable */ + coord_dup(&prev_coord, &coord); + coord_prev_item(&prev_coord); + if (are_items_mergeable(&prev_coord, &coord)) { + *error = "mergeable items in one node"; + return -1; + } + + } + } + + if ((flags & REISER4_NODE_DKEYS) && !node_is_empty(node)) { + coord_t coord; + item_plugin *iplug; + + coord_init_last_unit(&coord, node); + iplug = item_plugin_by_coord(&coord); + if ((item_is_extent(&coord) || item_is_tail(&coord)) && + iplug->s.file.append_key != NULL) { + reiser4_key mkey; + + iplug->s.file.append_key(&coord, &mkey); + set_key_offset(&mkey, get_key_offset(&mkey) - 1); + read_lock_dk(current_tree); + result = keygt(&mkey, znode_get_rd_key((znode *) node)); + read_unlock_dk(current_tree); + if (result) { + *error = "key of rightmost item is too large"; + return -1; + } + } + } + if (flags & REISER4_NODE_DKEYS) { + read_lock_tree(current_tree); + read_lock_dk(current_tree); + + flags |= REISER4_NODE_TREE_STABLE; + + if (keygt(&prev, znode_get_rd_key((znode *) node))) { + if (flags & REISER4_NODE_TREE_STABLE) { + *error = "Last key is greater than rdkey"; + read_unlock_dk(current_tree); + read_unlock_tree(current_tree); + return -1; + } + } + if (keygt + (znode_get_ld_key((znode *) node), + znode_get_rd_key((znode *) node))) { + *error = "ldkey is greater than rdkey"; + read_unlock_dk(current_tree); + read_unlock_tree(current_tree); + return -1; + } + if (ZF_ISSET(node, JNODE_LEFT_CONNECTED) && + (node->left != NULL) && + !ZF_ISSET(node->left, JNODE_HEARD_BANSHEE) && + ergo(flags & REISER4_NODE_TREE_STABLE, + !keyeq(znode_get_rd_key(node->left), + znode_get_ld_key((znode *) node))) + && ergo(!(flags & REISER4_NODE_TREE_STABLE), + keygt(znode_get_rd_key(node->left), + znode_get_ld_key((znode *) node)))) { + *error = "left rdkey or ldkey is wrong"; + read_unlock_dk(current_tree); + read_unlock_tree(current_tree); + return -1; + } + if (ZF_ISSET(node, JNODE_RIGHT_CONNECTED) && + (node->right != NULL) && + !ZF_ISSET(node->right, JNODE_HEARD_BANSHEE) && + ergo(flags & REISER4_NODE_TREE_STABLE, + !keyeq(znode_get_rd_key((znode *) node), + znode_get_ld_key(node->right))) + && ergo(!(flags & REISER4_NODE_TREE_STABLE), + keygt(znode_get_rd_key((znode *) node), + znode_get_ld_key(node->right)))) { + *error = "rdkey or right ldkey is wrong"; + read_unlock_dk(current_tree); + read_unlock_tree(current_tree); + return -1; + } + + read_unlock_dk(current_tree); + read_unlock_tree(current_tree); + } + + return 0; +} + +/* plugin->u.node.parse + look for description of this method in plugin/node/node.h */ +int parse_node40(znode * node /* node to parse */ ) +{ + node40_header *header; + int result; + d8 level; + + header = node40_node_header((znode *) node); + result = -EIO; + level = nh40_get_level(header); + if (unlikely(((__u8) znode_get_level(node)) != level)) + warning("nikita-494", "Wrong level found in node: %i != %i", + znode_get_level(node), level); + else if (unlikely(nh40_get_magic(header) != REISER4_NODE_MAGIC)) + warning("nikita-495", + "Wrong magic in tree node: want %x, got %x", + REISER4_NODE_MAGIC, nh40_get_magic(header)); + else { + node->nr_items = node40_num_of_items_internal(node); + result = 0; + } + return RETERR(result); +} + +/* plugin->u.node.init + look for description of this method in plugin/node/node.h */ +int init_node40(znode * node /* node to initialise */ ) +{ + node40_header *header; + + assert("nikita-570", node != NULL); + assert("nikita-572", zdata(node) != NULL); + + header = node40_node_header(node); + memset(header, 0, sizeof(node40_header)); + nh40_set_free_space(header, znode_size(node) - sizeof(node40_header)); + nh40_set_free_space_start(header, sizeof(node40_header)); + /* sane hypothesis: 0 in CPU format is 0 in disk format */ + /* items: 0 */ + save_plugin_id(node_plugin_to_plugin(node->nplug), + &header->common_header.plugin_id); + nh40_set_level(header, znode_get_level(node)); + nh40_set_magic(header, REISER4_NODE_MAGIC); + node->nr_items = 0; + nh40_set_mkfs_id(header, reiser4_mkfs_id(reiser4_get_current_sb())); + + /* flags: 0 */ + return 0; +} + +#ifdef GUESS_EXISTS +int guess_node40(const znode * node /* node to guess plugin of */ ) +{ + node40_header *nethack; + + assert("nikita-1058", node != NULL); + nethack = node40_node_header(node); + return + (nh40_get_magic(nethack) == REISER4_NODE_MAGIC) && + (plugin_by_disk_id(znode_get_tree(node), + REISER4_NODE_PLUGIN_TYPE, + &nethack->common_header.plugin_id)->h.id == + NODE40_ID); +} +#endif + +/* plugin->u.node.chage_item_size + look for description of this method in plugin/node/node.h */ +void change_item_size_node40(coord_t * coord, int by) +{ + node40_header *nh; + item_header40 *ih; + char *item_data; + int item_length; + unsigned i; + + /* make sure that @item is coord of existing item */ + assert("vs-210", coord_is_existing_item(coord)); + + nh = node40_node_header(coord->node); + + item_data = item_by_coord_node40(coord); + item_length = length_by_coord_node40(coord); + + /* move item bodies */ + ih = node40_ih_at_coord(coord); + memmove(item_data + item_length + by, item_data + item_length, + nh40_get_free_space_start(node40_node_header(coord->node)) - + (ih40_get_offset(ih) + item_length)); + + /* update offsets of moved items */ + for (i = coord->item_pos + 1; i < nh40_get_num_items(nh); i++) { + ih = node40_ih_at(coord->node, i); + ih40_set_offset(ih, ih40_get_offset(ih) + by); + } + + /* update node header */ + nh40_set_free_space(nh, nh40_get_free_space(nh) - by); + nh40_set_free_space_start(nh, nh40_get_free_space_start(nh) + by); +} + +static int should_notify_parent(const znode * node) +{ + /* FIXME_JMACD This looks equivalent to znode_is_root(), right? -josh */ + return !disk_addr_eq(znode_get_block(node), + &znode_get_tree(node)->root_block); +} + +/* plugin->u.node.create_item + look for description of this method in plugin/node/node.h */ +int +create_item_node40(coord_t *target, const reiser4_key *key, + reiser4_item_data *data, carry_plugin_info *info) +{ + node40_header *nh; + item_header40 *ih; + unsigned offset; + unsigned i; + + nh = node40_node_header(target->node); + + assert("vs-212", coord_is_between_items(target)); + /* node must have enough free space */ + assert("vs-254", + free_space_node40(target->node) >= + data->length + sizeof(item_header40)); + assert("vs-1410", data->length >= 0); + + if (coord_set_to_right(target)) + /* there are not items to the right of @target, so, new item + will be inserted after last one */ + coord_set_item_pos(target, nh40_get_num_items(nh)); + + if (target->item_pos < nh40_get_num_items(nh)) { + /* there are items to be moved to prepare space for new + item */ + ih = node40_ih_at_coord(target); + /* new item will start at this offset */ + offset = ih40_get_offset(ih); + + memmove(zdata(target->node) + offset + data->length, + zdata(target->node) + offset, + nh40_get_free_space_start(nh) - offset); + /* update headers of moved items */ + for (i = target->item_pos; i < nh40_get_num_items(nh); i++) { + ih = node40_ih_at(target->node, i); + ih40_set_offset(ih, ih40_get_offset(ih) + data->length); + } + + /* @ih is set to item header of the last item, move item headers */ + memmove(ih - 1, ih, + sizeof(item_header40) * (nh40_get_num_items(nh) - + target->item_pos)); + } else { + /* new item will start at this offset */ + offset = nh40_get_free_space_start(nh); + } + + /* make item header for the new item */ + ih = node40_ih_at_coord(target); + memcpy(&ih->key, key, sizeof(reiser4_key)); + ih40_set_offset(ih, offset); + save_plugin_id(item_plugin_to_plugin(data->iplug), &ih->plugin_id); + + /* update node header */ + nh40_set_free_space(nh, + nh40_get_free_space(nh) - data->length - + sizeof(item_header40)); + nh40_set_free_space_start(nh, + nh40_get_free_space_start(nh) + data->length); + node40_set_num_items(target->node, nh, nh40_get_num_items(nh) + 1); + + /* FIXME: check how does create_item work when between is set to BEFORE_UNIT */ + target->unit_pos = 0; + target->between = AT_UNIT; + coord_clear_iplug(target); + + /* initialize item */ + if (data->iplug->b.init != NULL) { + data->iplug->b.init(target, NULL, data); + } + /* copy item body */ + if (data->iplug->b.paste != NULL) { + data->iplug->b.paste(target, data, info); + } else if (data->data != NULL) { + if (data->user) { + /* AUDIT: Are we really should not check that pointer + from userspace was valid and data bytes were + available? How will we return -EFAULT of some kind + without this check? */ + assert("nikita-3038", reiser4_schedulable()); + /* copy data from user space */ + __copy_from_user(zdata(target->node) + offset, + (const char __user *)data->data, + (unsigned)data->length); + } else + /* copy from kernel space */ + memcpy(zdata(target->node) + offset, data->data, + (unsigned)data->length); + } + + if (target->item_pos == 0) { + /* left delimiting key has to be updated */ + prepare_for_update(NULL, target->node, info); + } + + if (item_plugin_by_coord(target)->b.create_hook != NULL) { + item_plugin_by_coord(target)->b.create_hook(target, data->arg); + } + + return 0; +} + +/* plugin->u.node.update_item_key + look for description of this method in plugin/node/node.h */ +void +update_item_key_node40(coord_t * target, const reiser4_key * key, + carry_plugin_info * info) +{ + item_header40 *ih; + + ih = node40_ih_at_coord(target); + memcpy(&ih->key, key, sizeof(reiser4_key)); + + if (target->item_pos == 0) { + prepare_for_update(NULL, target->node, info); + } +} + +/* this bits encode cut mode */ +#define CMODE_TAIL 1 +#define CMODE_WHOLE 2 +#define CMODE_HEAD 4 + +struct cut40_info { + int mode; + pos_in_node_t tail_removed; /* position of item which gets tail removed */ + pos_in_node_t first_removed; /* position of first the leftmost item among items removed completely */ + pos_in_node_t removed_count; /* number of items removed completely */ + pos_in_node_t head_removed; /* position of item which gets head removed */ + + pos_in_node_t freed_space_start; + pos_in_node_t freed_space_end; + pos_in_node_t first_moved; + pos_in_node_t head_removed_location; +}; + +static void init_cinfo(struct cut40_info *cinfo) +{ + cinfo->mode = 0; + cinfo->tail_removed = MAX_POS_IN_NODE; + cinfo->first_removed = MAX_POS_IN_NODE; + cinfo->removed_count = MAX_POS_IN_NODE; + cinfo->head_removed = MAX_POS_IN_NODE; + cinfo->freed_space_start = MAX_POS_IN_NODE; + cinfo->freed_space_end = MAX_POS_IN_NODE; + cinfo->first_moved = MAX_POS_IN_NODE; + cinfo->head_removed_location = MAX_POS_IN_NODE; +} + +/* complete cut_node40/kill_node40 content by removing the gap created by */ +static void compact(znode * node, struct cut40_info *cinfo) +{ + node40_header *nh; + item_header40 *ih; + pos_in_node_t freed; + pos_in_node_t pos, nr_items; + + assert("vs-1526", (cinfo->freed_space_start != MAX_POS_IN_NODE && + cinfo->freed_space_end != MAX_POS_IN_NODE && + cinfo->first_moved != MAX_POS_IN_NODE)); + assert("vs-1523", cinfo->freed_space_end >= cinfo->freed_space_start); + + nh = node40_node_header(node); + nr_items = nh40_get_num_items(nh); + + /* remove gap made up by removal */ + memmove(zdata(node) + cinfo->freed_space_start, + zdata(node) + cinfo->freed_space_end, + nh40_get_free_space_start(nh) - cinfo->freed_space_end); + + /* update item headers of moved items - change their locations */ + pos = cinfo->first_moved; + ih = node40_ih_at(node, pos); + if (cinfo->head_removed_location != MAX_POS_IN_NODE) { + assert("vs-1580", pos == cinfo->head_removed); + ih40_set_offset(ih, cinfo->head_removed_location); + pos++; + ih--; + } + + freed = cinfo->freed_space_end - cinfo->freed_space_start; + for (; pos < nr_items; pos++, ih--) { + assert("vs-1581", ih == node40_ih_at(node, pos)); + ih40_set_offset(ih, ih40_get_offset(ih) - freed); + } + + /* free space start moved to right */ + nh40_set_free_space_start(nh, nh40_get_free_space_start(nh) - freed); + + if (cinfo->removed_count != MAX_POS_IN_NODE) { + /* number of items changed. Remove item headers of those items */ + ih = node40_ih_at(node, nr_items - 1); + memmove(ih + cinfo->removed_count, ih, + sizeof(item_header40) * (nr_items - + cinfo->removed_count - + cinfo->first_removed)); + freed += sizeof(item_header40) * cinfo->removed_count; + node40_set_num_items(node, nh, nr_items - cinfo->removed_count); + } + + /* total amount of free space increased */ + nh40_set_free_space(nh, nh40_get_free_space(nh) + freed); +} + +int shrink_item_node40(coord_t * coord, int delta) +{ + node40_header *nh; + item_header40 *ih; + pos_in_node_t pos; + pos_in_node_t nr_items; + char *end; + znode *node; + int off; + + assert("nikita-3487", coord != NULL); + assert("nikita-3488", delta >= 0); + + node = coord->node; + nh = node40_node_header(node); + nr_items = nh40_get_num_items(nh); + + ih = node40_ih_at_coord(coord); + assert("nikita-3489", delta <= length_by_coord_node40(coord)); + off = ih40_get_offset(ih) + length_by_coord_node40(coord); + end = zdata(node) + off; + + /* remove gap made up by removal */ + memmove(end - delta, end, nh40_get_free_space_start(nh) - off); + + /* update item headers of moved items - change their locations */ + pos = coord->item_pos + 1; + ih = node40_ih_at(node, pos); + for (; pos < nr_items; pos++, ih--) { + assert("nikita-3490", ih == node40_ih_at(node, pos)); + ih40_set_offset(ih, ih40_get_offset(ih) - delta); + } + + /* free space start moved to left */ + nh40_set_free_space_start(nh, nh40_get_free_space_start(nh) - delta); + /* total amount of free space increased */ + nh40_set_free_space(nh, nh40_get_free_space(nh) + delta); + /* + * This method does _not_ changes number of items. Hence, it cannot + * make node empty. Also it doesn't remove items at all, which means + * that no keys have to be updated either. + */ + return 0; +} + +/* this is used by cut_node40 and kill_node40. It analyses input parameters and calculates cut mode. There are 2 types + of cut. First is when a unit is removed from the middle of an item. In this case this function returns 1. All the + rest fits into second case: 0 or 1 of items getting tail cut, 0 or more items removed completely and 0 or 1 item + getting head cut. Function returns 0 in this case */ +static int +parse_cut(struct cut40_info *cinfo, const struct cut_kill_params *params) +{ + reiser4_key left_key, right_key; + reiser4_key min_from_key, max_to_key; + const reiser4_key *from_key, *to_key; + + init_cinfo(cinfo); + + /* calculate minimal key stored in first item of items to be cut (params->from) */ + item_key_by_coord(params->from, &min_from_key); + /* and max key stored in last item of items to be cut (params->to) */ + max_item_key_by_coord(params->to, &max_to_key); + + /* if cut key range is not defined in input parameters - define it using cut coord range */ + if (params->from_key == NULL) { + assert("vs-1513", params->to_key == NULL); + unit_key_by_coord(params->from, &left_key); + from_key = &left_key; + max_unit_key_by_coord(params->to, &right_key); + to_key = &right_key; + } else { + from_key = params->from_key; + to_key = params->to_key; + } + + if (params->from->item_pos == params->to->item_pos) { + if (keylt(&min_from_key, from_key) + && keylt(to_key, &max_to_key)) + return 1; + + if (keygt(from_key, &min_from_key)) { + /* tail of item is to be cut cut */ + cinfo->tail_removed = params->from->item_pos; + cinfo->mode |= CMODE_TAIL; + } else if (keylt(to_key, &max_to_key)) { + /* head of item is to be cut */ + cinfo->head_removed = params->from->item_pos; + cinfo->mode |= CMODE_HEAD; + } else { + /* item is removed completely */ + cinfo->first_removed = params->from->item_pos; + cinfo->removed_count = 1; + cinfo->mode |= CMODE_WHOLE; + } + } else { + cinfo->first_removed = params->from->item_pos + 1; + cinfo->removed_count = + params->to->item_pos - params->from->item_pos - 1; + + if (keygt(from_key, &min_from_key)) { + /* first item is not cut completely */ + cinfo->tail_removed = params->from->item_pos; + cinfo->mode |= CMODE_TAIL; + } else { + cinfo->first_removed--; + cinfo->removed_count++; + } + if (keylt(to_key, &max_to_key)) { + /* last item is not cut completely */ + cinfo->head_removed = params->to->item_pos; + cinfo->mode |= CMODE_HEAD; + } else { + cinfo->removed_count++; + } + if (cinfo->removed_count) + cinfo->mode |= CMODE_WHOLE; + } + + return 0; +} + +static void +call_kill_hooks(znode * node, pos_in_node_t from, pos_in_node_t count, + carry_kill_data * kdata) +{ + coord_t coord; + item_plugin *iplug; + pos_in_node_t pos; + + coord.node = node; + coord.unit_pos = 0; + coord.between = AT_UNIT; + for (pos = 0; pos < count; pos++) { + coord_set_item_pos(&coord, from + pos); + coord.unit_pos = 0; + coord.between = AT_UNIT; + iplug = item_plugin_by_coord(&coord); + if (iplug->b.kill_hook) { + iplug->b.kill_hook(&coord, 0, coord_num_units(&coord), + kdata); + } + } +} + +/* this is used to kill item partially */ +static pos_in_node_t +kill_units(coord_t * coord, pos_in_node_t from, pos_in_node_t to, void *data, + reiser4_key * smallest_removed, reiser4_key * new_first_key) +{ + struct carry_kill_data *kdata; + item_plugin *iplug; + + kdata = data; + iplug = item_plugin_by_coord(coord); + + assert("vs-1524", iplug->b.kill_units); + return iplug->b.kill_units(coord, from, to, kdata, smallest_removed, + new_first_key); +} + +/* call item plugin to cut tail of file */ +static pos_in_node_t +kill_tail(coord_t * coord, void *data, reiser4_key * smallest_removed) +{ + struct carry_kill_data *kdata; + pos_in_node_t to; + + kdata = data; + to = coord_last_unit_pos(coord); + return kill_units(coord, coord->unit_pos, to, kdata, smallest_removed, + NULL); +} + +/* call item plugin to cut head of item */ +static pos_in_node_t +kill_head(coord_t * coord, void *data, reiser4_key * smallest_removed, + reiser4_key * new_first_key) +{ + return kill_units(coord, 0, coord->unit_pos, data, smallest_removed, + new_first_key); +} + +/* this is used to cut item partially */ +static pos_in_node_t +cut_units(coord_t * coord, pos_in_node_t from, pos_in_node_t to, void *data, + reiser4_key * smallest_removed, reiser4_key * new_first_key) +{ + carry_cut_data *cdata; + item_plugin *iplug; + + cdata = data; + iplug = item_plugin_by_coord(coord); + assert("vs-302", iplug->b.cut_units); + return iplug->b.cut_units(coord, from, to, cdata, smallest_removed, + new_first_key); +} + +/* call item plugin to cut tail of file */ +static pos_in_node_t +cut_tail(coord_t * coord, void *data, reiser4_key * smallest_removed) +{ + carry_cut_data *cdata; + pos_in_node_t to; + + cdata = data; + to = coord_last_unit_pos(cdata->params.from); + return cut_units(coord, coord->unit_pos, to, data, smallest_removed, NULL); +} + +/* call item plugin to cut head of item */ +static pos_in_node_t +cut_head(coord_t * coord, void *data, reiser4_key * smallest_removed, + reiser4_key * new_first_key) +{ + return cut_units(coord, 0, coord->unit_pos, data, smallest_removed, + new_first_key); +} + +/* this returns 1 of key of first item changed, 0 - if it did not */ +static int +prepare_for_compact(struct cut40_info *cinfo, + const struct cut_kill_params *params, int is_cut, + void *data, carry_plugin_info * info) +{ + znode *node; + item_header40 *ih; + pos_in_node_t freed; + pos_in_node_t item_pos; + coord_t coord; + reiser4_key new_first_key; + pos_in_node_t(*kill_units_f) (coord_t *, pos_in_node_t, pos_in_node_t, + void *, reiser4_key *, reiser4_key *); + pos_in_node_t(*kill_tail_f) (coord_t *, void *, reiser4_key *); + pos_in_node_t(*kill_head_f) (coord_t *, void *, reiser4_key *, + reiser4_key *); + int retval; + + retval = 0; + + node = params->from->node; + + assert("vs-184", node == params->to->node); + assert("vs-312", !node_is_empty(node)); + assert("vs-297", + coord_compare(params->from, params->to) != COORD_CMP_ON_RIGHT); + + if (is_cut) { + kill_units_f = cut_units; + kill_tail_f = cut_tail; + kill_head_f = cut_head; + } else { + kill_units_f = kill_units; + kill_tail_f = kill_tail; + kill_head_f = kill_head; + } + + if (parse_cut(cinfo, params) == 1) { + /* cut from the middle of item */ + freed = + kill_units_f(params->from, params->from->unit_pos, + params->to->unit_pos, data, + params->smallest_removed, NULL); + + item_pos = params->from->item_pos; + ih = node40_ih_at(node, item_pos); + cinfo->freed_space_start = + ih40_get_offset(ih) + node40_item_length(node, + item_pos) - freed; + cinfo->freed_space_end = cinfo->freed_space_start + freed; + cinfo->first_moved = item_pos + 1; + } else { + assert("vs-1521", (cinfo->tail_removed != MAX_POS_IN_NODE || + cinfo->first_removed != MAX_POS_IN_NODE || + cinfo->head_removed != MAX_POS_IN_NODE)); + + switch (cinfo->mode) { + case CMODE_TAIL: + /* one item gets cut partially from its end */ + assert("vs-1562", + cinfo->tail_removed == params->from->item_pos); + + freed = + kill_tail_f(params->from, data, + params->smallest_removed); + + item_pos = cinfo->tail_removed; + ih = node40_ih_at(node, item_pos); + cinfo->freed_space_start = + ih40_get_offset(ih) + node40_item_length(node, + item_pos) - + freed; + cinfo->freed_space_end = + cinfo->freed_space_start + freed; + cinfo->first_moved = cinfo->tail_removed + 1; + break; + + case CMODE_WHOLE: + /* one or more items get removed completely */ + assert("vs-1563", + cinfo->first_removed == params->from->item_pos); + assert("vs-1564", cinfo->removed_count > 0 + && cinfo->removed_count != MAX_POS_IN_NODE); + + /* call kill hook for all items removed completely */ + if (is_cut == 0) + call_kill_hooks(node, cinfo->first_removed, + cinfo->removed_count, data); + + item_pos = cinfo->first_removed; + ih = node40_ih_at(node, item_pos); + + if (params->smallest_removed) + memcpy(params->smallest_removed, &ih->key, + sizeof(reiser4_key)); + + cinfo->freed_space_start = ih40_get_offset(ih); + + item_pos += (cinfo->removed_count - 1); + ih -= (cinfo->removed_count - 1); + cinfo->freed_space_end = + ih40_get_offset(ih) + node40_item_length(node, + item_pos); + cinfo->first_moved = item_pos + 1; + if (cinfo->first_removed == 0) + /* key of first item of the node changes */ + retval = 1; + break; + + case CMODE_HEAD: + /* one item gets cut partially from its head */ + assert("vs-1565", + cinfo->head_removed == params->from->item_pos); + + freed = + kill_head_f(params->to, data, + params->smallest_removed, + &new_first_key); + + item_pos = cinfo->head_removed; + ih = node40_ih_at(node, item_pos); + cinfo->freed_space_start = ih40_get_offset(ih); + cinfo->freed_space_end = ih40_get_offset(ih) + freed; + cinfo->first_moved = cinfo->head_removed + 1; + + /* item head is removed, therefore, item key changed */ + coord.node = node; + coord_set_item_pos(&coord, item_pos); + coord.unit_pos = 0; + coord.between = AT_UNIT; + update_item_key_node40(&coord, &new_first_key, NULL); + if (item_pos == 0) + /* key of first item of the node changes */ + retval = 1; + break; + + case CMODE_TAIL | CMODE_WHOLE: + /* one item gets cut from its end and one or more items get removed completely */ + assert("vs-1566", + cinfo->tail_removed == params->from->item_pos); + assert("vs-1567", + cinfo->first_removed == cinfo->tail_removed + 1); + assert("vs-1564", cinfo->removed_count > 0 + && cinfo->removed_count != MAX_POS_IN_NODE); + + freed = + kill_tail_f(params->from, data, + params->smallest_removed); + + item_pos = cinfo->tail_removed; + ih = node40_ih_at(node, item_pos); + cinfo->freed_space_start = + ih40_get_offset(ih) + node40_item_length(node, + item_pos) - + freed; + + /* call kill hook for all items removed completely */ + if (is_cut == 0) + call_kill_hooks(node, cinfo->first_removed, + cinfo->removed_count, data); + + item_pos += cinfo->removed_count; + ih -= cinfo->removed_count; + cinfo->freed_space_end = + ih40_get_offset(ih) + node40_item_length(node, + item_pos); + cinfo->first_moved = item_pos + 1; + break; + + case CMODE_WHOLE | CMODE_HEAD: + /* one or more items get removed completely and one item gets cut partially from its head */ + assert("vs-1568", + cinfo->first_removed == params->from->item_pos); + assert("vs-1564", cinfo->removed_count > 0 + && cinfo->removed_count != MAX_POS_IN_NODE); + assert("vs-1569", + cinfo->head_removed == + cinfo->first_removed + cinfo->removed_count); + + /* call kill hook for all items removed completely */ + if (is_cut == 0) + call_kill_hooks(node, cinfo->first_removed, + cinfo->removed_count, data); + + item_pos = cinfo->first_removed; + ih = node40_ih_at(node, item_pos); + + if (params->smallest_removed) + memcpy(params->smallest_removed, &ih->key, + sizeof(reiser4_key)); + + freed = + kill_head_f(params->to, data, NULL, &new_first_key); + + cinfo->freed_space_start = ih40_get_offset(ih); + + ih = node40_ih_at(node, cinfo->head_removed); + /* this is the most complex case. Item which got head removed and items which are to be moved + intact change their location differently. */ + cinfo->freed_space_end = ih40_get_offset(ih) + freed; + cinfo->first_moved = cinfo->head_removed; + cinfo->head_removed_location = cinfo->freed_space_start; + + /* item head is removed, therefore, item key changed */ + coord.node = node; + coord_set_item_pos(&coord, cinfo->head_removed); + coord.unit_pos = 0; + coord.between = AT_UNIT; + update_item_key_node40(&coord, &new_first_key, NULL); + + assert("vs-1579", cinfo->first_removed == 0); + /* key of first item of the node changes */ + retval = 1; + break; + + case CMODE_TAIL | CMODE_HEAD: + /* one item get cut from its end and its neighbor gets cut from its tail */ + impossible("vs-1576", "this can not happen currently"); + break; + + case CMODE_TAIL | CMODE_WHOLE | CMODE_HEAD: + impossible("vs-1577", "this can not happen currently"); + break; + default: + impossible("vs-1578", "unexpected cut mode"); + break; + } + } + return retval; +} + +/* plugin->u.node.kill + return value is number of items removed completely */ +int kill_node40(struct carry_kill_data *kdata, carry_plugin_info * info) +{ + znode *node; + struct cut40_info cinfo; + int first_key_changed; + + node = kdata->params.from->node; + + first_key_changed = + prepare_for_compact(&cinfo, &kdata->params, 0 /* not cut */ , kdata, + info); + compact(node, &cinfo); + + if (info) { + /* it is not called by node40_shift, so we have to take care + of changes on upper levels */ + if (node_is_empty(node) + && !(kdata->flags & DELETE_RETAIN_EMPTY)) + /* all contents of node is deleted */ + prepare_removal_node40(node, info); + else if (first_key_changed) { + prepare_for_update(NULL, node, info); + } + } + + coord_clear_iplug(kdata->params.from); + coord_clear_iplug(kdata->params.to); + + znode_make_dirty(node); + return cinfo.removed_count == MAX_POS_IN_NODE ? 0 : cinfo.removed_count; +} + +/* plugin->u.node.cut + return value is number of items removed completely */ +int cut_node40(struct carry_cut_data *cdata, carry_plugin_info * info) +{ + znode *node; + struct cut40_info cinfo; + int first_key_changed; + + node = cdata->params.from->node; + + first_key_changed = + prepare_for_compact(&cinfo, &cdata->params, 1 /* not cut */ , cdata, + info); + compact(node, &cinfo); + + if (info) { + /* it is not called by node40_shift, so we have to take care + of changes on upper levels */ + if (node_is_empty(node)) + /* all contents of node is deleted */ + prepare_removal_node40(node, info); + else if (first_key_changed) { + prepare_for_update(NULL, node, info); + } + } + + coord_clear_iplug(cdata->params.from); + coord_clear_iplug(cdata->params.to); + + znode_make_dirty(node); + return cinfo.removed_count == MAX_POS_IN_NODE ? 0 : cinfo.removed_count; +} + +/* this structure is used by shift method of node40 plugin */ +struct shift_params { + shift_direction pend; /* when @pend == append - we are shifting to + left, when @pend == prepend - to right */ + coord_t wish_stop; /* when shifting to left this is last unit we + want shifted, when shifting to right - this + is set to unit we want to start shifting + from */ + znode *target; + int everything; /* it is set to 1 if everything we have to shift is + shifted, 0 - otherwise */ + + /* FIXME-VS: get rid of read_stop */ + + /* these are set by estimate_shift */ + coord_t real_stop; /* this will be set to last unit which will be + really shifted */ + + /* coordinate in source node before operation of unit which becomes + first after shift to left of last after shift to right */ + union { + coord_t future_first; + coord_t future_last; + } u; + + unsigned merging_units; /* number of units of first item which have to + be merged with last item of target node */ + unsigned merging_bytes; /* number of bytes in those units */ + + unsigned entire; /* items shifted in their entirety */ + unsigned entire_bytes; /* number of bytes in those items */ + + unsigned part_units; /* number of units of partially copied item */ + unsigned part_bytes; /* number of bytes in those units */ + + unsigned shift_bytes; /* total number of bytes in items shifted (item + headers not included) */ + +}; + +static int item_creation_overhead(coord_t *item) +{ + return node_plugin_by_coord(item)->item_overhead(item->node, NULL); +} + +/* how many units are there in @source starting from source->unit_pos + but not further than @stop_coord */ +static int +wanted_units(coord_t *source, coord_t *stop_coord, shift_direction pend) +{ + if (pend == SHIFT_LEFT) { + assert("vs-181", source->unit_pos == 0); + } else { + assert("vs-182", + source->unit_pos == coord_last_unit_pos(source)); + } + + if (source->item_pos != stop_coord->item_pos) { + /* @source and @stop_coord are different items */ + return coord_last_unit_pos(source) + 1; + } + + if (pend == SHIFT_LEFT) { + return stop_coord->unit_pos + 1; + } else { + return source->unit_pos - stop_coord->unit_pos + 1; + } +} + +/* this calculates what can be copied from @shift->wish_stop.node to + @shift->target */ +static void +estimate_shift(struct shift_params *shift, const reiser4_context * ctx) +{ + unsigned target_free_space, size; + pos_in_node_t stop_item; /* item which estimating should not consider */ + unsigned want; /* number of units of item we want shifted */ + coord_t source; /* item being estimated */ + item_plugin *iplug; + + /* shifting to left/right starts from first/last units of + @shift->wish_stop.node */ + if (shift->pend == SHIFT_LEFT) { + coord_init_first_unit(&source, shift->wish_stop.node); + } else { + coord_init_last_unit(&source, shift->wish_stop.node); + } + shift->real_stop = source; + + /* free space in target node and number of items in source */ + target_free_space = znode_free_space(shift->target); + + shift->everything = 0; + if (!node_is_empty(shift->target)) { + /* target node is not empty, check for boundary items + mergeability */ + coord_t to; + + /* item we try to merge @source with */ + if (shift->pend == SHIFT_LEFT) { + coord_init_last_unit(&to, shift->target); + } else { + coord_init_first_unit(&to, shift->target); + } + + if ((shift->pend == SHIFT_LEFT) ? are_items_mergeable(&to, + &source) : + are_items_mergeable(&source, &to)) { + /* how many units of @source do we want to merge to + item @to */ + want = + wanted_units(&source, &shift->wish_stop, + shift->pend); + + /* how many units of @source we can merge to item + @to */ + iplug = item_plugin_by_coord(&source); + if (iplug->b.can_shift != NULL) + shift->merging_units = + iplug->b.can_shift(target_free_space, + &source, shift->target, + shift->pend, &size, + want); + else { + shift->merging_units = 0; + size = 0; + } + shift->merging_bytes = size; + shift->shift_bytes += size; + /* update stop coord to be set to last unit of @source + we can merge to @target */ + if (shift->merging_units) + /* at least one unit can be shifted */ + shift->real_stop.unit_pos = + (shift->merging_units - source.unit_pos - + 1) * shift->pend; + else { + /* nothing can be shifted */ + if (shift->pend == SHIFT_LEFT) + coord_init_before_first_item(&shift-> + real_stop, + source. + node); + else + coord_init_after_last_item(&shift-> + real_stop, + source.node); + } + assert("nikita-2081", shift->real_stop.unit_pos + 1); + + if (shift->merging_units != want) { + /* we could not copy as many as we want, so, + there is no reason for estimating any + longer */ + return; + } + + target_free_space -= size; + coord_add_item_pos(&source, shift->pend); + } + } + + /* number of item nothing of which we want to shift */ + stop_item = shift->wish_stop.item_pos + shift->pend; + + /* calculate how many items can be copied into given free + space as whole */ + for (; source.item_pos != stop_item; + coord_add_item_pos(&source, shift->pend)) { + if (shift->pend == SHIFT_RIGHT) + source.unit_pos = coord_last_unit_pos(&source); + + /* how many units of @source do we want to copy */ + want = wanted_units(&source, &shift->wish_stop, shift->pend); + + if (want == coord_last_unit_pos(&source) + 1) { + /* we want this item to be copied entirely */ + size = + item_length_by_coord(&source) + + item_creation_overhead(&source); + if (size <= target_free_space) { + /* item fits into target node as whole */ + target_free_space -= size; + shift->shift_bytes += + size - item_creation_overhead(&source); + shift->entire_bytes += + size - item_creation_overhead(&source); + shift->entire++; + + /* update shift->real_stop coord to be set to + last unit of @source we can merge to + @target */ + shift->real_stop = source; + if (shift->pend == SHIFT_LEFT) + shift->real_stop.unit_pos = + coord_last_unit_pos(&shift-> + real_stop); + else + shift->real_stop.unit_pos = 0; + continue; + } + } + + /* we reach here only for an item which does not fit into + target node in its entirety. This item may be either + partially shifted, or not shifted at all. We will have to + create new item in target node, so decrease amout of free + space by an item creation overhead. We can reach here also + if stop coord is in this item */ + if (target_free_space >= + (unsigned)item_creation_overhead(&source)) { + target_free_space -= item_creation_overhead(&source); + iplug = item_plugin_by_coord(&source); + if (iplug->b.can_shift) { + shift->part_units = iplug->b.can_shift(target_free_space, + &source, + NULL, /* target */ + shift->pend, + &size, + want); + } else { + target_free_space = 0; + shift->part_units = 0; + size = 0; + } + } else { + target_free_space = 0; + shift->part_units = 0; + size = 0; + } + shift->part_bytes = size; + shift->shift_bytes += size; + + /* set @shift->real_stop to last unit of @source we can merge + to @shift->target */ + if (shift->part_units) { + shift->real_stop = source; + shift->real_stop.unit_pos = + (shift->part_units - source.unit_pos - + 1) * shift->pend; + assert("nikita-2082", shift->real_stop.unit_pos + 1); + } + + if (want != shift->part_units) + /* not everything wanted were shifted */ + return; + break; + } + + shift->everything = 1; +} + +static void +copy_units(coord_t * target, coord_t * source, unsigned from, unsigned count, + shift_direction dir, unsigned free_space) +{ + item_plugin *iplug; + + assert("nikita-1463", target != NULL); + assert("nikita-1464", source != NULL); + assert("nikita-1465", from + count <= coord_num_units(source)); + + iplug = item_plugin_by_coord(source); + assert("nikita-1468", iplug == item_plugin_by_coord(target)); + iplug->b.copy_units(target, source, from, count, dir, free_space); + + if (dir == SHIFT_RIGHT) { + /* FIXME-VS: this looks not necessary. update_item_key was + called already by copy_units method */ + reiser4_key split_key; + + assert("nikita-1469", target->unit_pos == 0); + + unit_key_by_coord(target, &split_key); + node_plugin_by_coord(target)->update_item_key(target, + &split_key, NULL); + } +} + +/* copy part of @shift->real_stop.node starting either from its beginning or + from its end and ending at @shift->real_stop to either the end or the + beginning of @shift->target */ +static void copy(struct shift_params *shift) +{ + node40_header *nh; + coord_t from; + coord_t to; + item_header40 *from_ih, *to_ih; + int free_space_start; + int new_items; + unsigned old_items; + int old_offset; + unsigned i; + + nh = node40_node_header(shift->target); + free_space_start = nh40_get_free_space_start(nh); + old_items = nh40_get_num_items(nh); + new_items = shift->entire + (shift->part_units ? 1 : 0); + assert("vs-185", + shift->shift_bytes == + shift->merging_bytes + shift->entire_bytes + shift->part_bytes); + + from = shift->wish_stop; + + coord_init_first_unit(&to, shift->target); + + /* NOTE:NIKITA->VS not sure what I am doing: shift->target is empty, + hence to.between is set to EMPTY_NODE above. Looks like we want it + to be AT_UNIT. + + Oh, wonders of ->betweeness... + + */ + to.between = AT_UNIT; + + if (shift->pend == SHIFT_LEFT) { + /* copying to left */ + + coord_set_item_pos(&from, 0); + from_ih = node40_ih_at(from.node, 0); + + coord_set_item_pos(&to, + node40_num_of_items_internal(to.node) - 1); + if (shift->merging_units) { + /* expand last item, so that plugin methods will see + correct data */ + free_space_start += shift->merging_bytes; + nh40_set_free_space_start(nh, + (unsigned)free_space_start); + nh40_set_free_space(nh, + nh40_get_free_space(nh) - + shift->merging_bytes); + + /* appending last item of @target */ + copy_units(&to, &from, 0, /* starting from 0-th unit */ + shift->merging_units, SHIFT_LEFT, + shift->merging_bytes); + coord_inc_item_pos(&from); + from_ih--; + coord_inc_item_pos(&to); + } + + to_ih = node40_ih_at(shift->target, old_items); + if (shift->entire) { + /* copy @entire items entirely */ + + /* copy item headers */ + memcpy(to_ih - shift->entire + 1, + from_ih - shift->entire + 1, + shift->entire * sizeof(item_header40)); + /* update item header offset */ + old_offset = ih40_get_offset(from_ih); + /* AUDIT: Looks like if we calculate old_offset + free_space_start here instead of just old_offset, we can perform one "add" operation less per each iteration */ + for (i = 0; i < shift->entire; i++, to_ih--, from_ih--) + ih40_set_offset(to_ih, + ih40_get_offset(from_ih) - + old_offset + free_space_start); + + /* copy item bodies */ + memcpy(zdata(shift->target) + free_space_start, zdata(from.node) + old_offset, /*ih40_get_offset (from_ih), */ + shift->entire_bytes); + + coord_add_item_pos(&from, (int)shift->entire); + coord_add_item_pos(&to, (int)shift->entire); + } + + nh40_set_free_space_start(nh, + free_space_start + + shift->shift_bytes - + shift->merging_bytes); + nh40_set_free_space(nh, + nh40_get_free_space(nh) - + (shift->shift_bytes - shift->merging_bytes + + sizeof(item_header40) * new_items)); + + /* update node header */ + node40_set_num_items(shift->target, nh, old_items + new_items); + assert("vs-170", + nh40_get_free_space(nh) < znode_size(shift->target)); + + if (shift->part_units) { + /* copy heading part (@part units) of @source item as + a new item into @target->node */ + + /* copy item header of partially copied item */ + coord_set_item_pos(&to, + node40_num_of_items_internal(to.node) + - 1); + memcpy(to_ih, from_ih, sizeof(item_header40)); + ih40_set_offset(to_ih, + nh40_get_free_space_start(nh) - + shift->part_bytes); + if (item_plugin_by_coord(&to)->b.init) + item_plugin_by_coord(&to)->b.init(&to, &from, + NULL); + copy_units(&to, &from, 0, shift->part_units, SHIFT_LEFT, + shift->part_bytes); + } + + } else { + /* copying to right */ + + coord_set_item_pos(&from, + node40_num_of_items_internal(from.node) - 1); + from_ih = node40_ih_at_coord(&from); + + coord_set_item_pos(&to, 0); + + /* prepare space for new items */ + memmove(zdata(to.node) + sizeof(node40_header) + + shift->shift_bytes, + zdata(to.node) + sizeof(node40_header), + free_space_start - sizeof(node40_header)); + /* update item headers of moved items */ + to_ih = node40_ih_at(to.node, 0); + /* first item gets @merging_bytes longer. free space appears + at its beginning */ + if (!node_is_empty(to.node)) + ih40_set_offset(to_ih, + ih40_get_offset(to_ih) + + shift->shift_bytes - + shift->merging_bytes); + + for (i = 1; i < old_items; i++) + ih40_set_offset(to_ih - i, + ih40_get_offset(to_ih - i) + + shift->shift_bytes); + + /* move item headers to make space for new items */ + memmove(to_ih - old_items + 1 - new_items, + to_ih - old_items + 1, + sizeof(item_header40) * old_items); + to_ih -= (new_items - 1); + + nh40_set_free_space_start(nh, + free_space_start + + shift->shift_bytes); + nh40_set_free_space(nh, + nh40_get_free_space(nh) - + (shift->shift_bytes + + sizeof(item_header40) * new_items)); + + /* update node header */ + node40_set_num_items(shift->target, nh, old_items + new_items); + assert("vs-170", + nh40_get_free_space(nh) < znode_size(shift->target)); + + if (shift->merging_units) { + coord_add_item_pos(&to, new_items); + to.unit_pos = 0; + to.between = AT_UNIT; + /* prepend first item of @to */ + copy_units(&to, &from, + coord_last_unit_pos(&from) - + shift->merging_units + 1, + shift->merging_units, SHIFT_RIGHT, + shift->merging_bytes); + coord_dec_item_pos(&from); + from_ih++; + } + + if (shift->entire) { + /* copy @entire items entirely */ + + /* copy item headers */ + memcpy(to_ih, from_ih, + shift->entire * sizeof(item_header40)); + + /* update item header offset */ + old_offset = + ih40_get_offset(from_ih + shift->entire - 1); + /* AUDIT: old_offset + sizeof (node40_header) + shift->part_bytes calculation can be taken off the loop. */ + for (i = 0; i < shift->entire; i++, to_ih++, from_ih++) + ih40_set_offset(to_ih, + ih40_get_offset(from_ih) - + old_offset + + sizeof(node40_header) + + shift->part_bytes); + /* copy item bodies */ + coord_add_item_pos(&from, -(int)(shift->entire - 1)); + memcpy(zdata(to.node) + sizeof(node40_header) + + shift->part_bytes, item_by_coord_node40(&from), + shift->entire_bytes); + coord_dec_item_pos(&from); + } + + if (shift->part_units) { + coord_set_item_pos(&to, 0); + to.unit_pos = 0; + to.between = AT_UNIT; + /* copy heading part (@part units) of @source item as + a new item into @target->node */ + + /* copy item header of partially copied item */ + memcpy(to_ih, from_ih, sizeof(item_header40)); + ih40_set_offset(to_ih, sizeof(node40_header)); + if (item_plugin_by_coord(&to)->b.init) + item_plugin_by_coord(&to)->b.init(&to, &from, + NULL); + copy_units(&to, &from, + coord_last_unit_pos(&from) - + shift->part_units + 1, shift->part_units, + SHIFT_RIGHT, shift->part_bytes); + } + } +} + +/* remove everything either before or after @fact_stop. Number of items + removed completely is returned */ +static int delete_copied(struct shift_params *shift) +{ + coord_t from; + coord_t to; + struct carry_cut_data cdata; + + if (shift->pend == SHIFT_LEFT) { + /* we were shifting to left, remove everything from the + beginning of @shift->wish_stop->node upto + @shift->wish_stop */ + coord_init_first_unit(&from, shift->real_stop.node); + to = shift->real_stop; + + /* store old coordinate of unit which will be first after + shift to left */ + shift->u.future_first = to; + coord_next_unit(&shift->u.future_first); + } else { + /* we were shifting to right, remove everything from + @shift->stop_coord upto to end of + @shift->stop_coord->node */ + from = shift->real_stop; + coord_init_last_unit(&to, from.node); + + /* store old coordinate of unit which will be last after + shift to right */ + shift->u.future_last = from; + coord_prev_unit(&shift->u.future_last); + } + + cdata.params.from = &from; + cdata.params.to = &to; + cdata.params.from_key = NULL; + cdata.params.to_key = NULL; + cdata.params.smallest_removed = NULL; + return cut_node40(&cdata, NULL); +} + +/* something was moved between @left and @right. Add carry operation to @info + list to have carry to update delimiting key between them */ +static int +prepare_for_update(znode * left, znode * right, carry_plugin_info * info) +{ + carry_op *op; + carry_node *cn; + + if (info == NULL) + /* nowhere to send operation to. */ + return 0; + + if (!should_notify_parent(right)) + return 0; + + op = node_post_carry(info, COP_UPDATE, right, 1); + if (IS_ERR(op) || op == NULL) + return op ? PTR_ERR(op) : -EIO; + + if (left != NULL) { + carry_node *reference; + + if (info->doing) + reference = insert_carry_node(info->doing, + info->todo, left); + else + reference = op->node; + assert("nikita-2992", reference != NULL); + cn = reiser4_add_carry(info->todo, POOLO_BEFORE, reference); + if (IS_ERR(cn)) + return PTR_ERR(cn); + cn->parent = 1; + cn->node = left; + if (ZF_ISSET(left, JNODE_ORPHAN)) + cn->left_before = 1; + op->u.update.left = cn; + } else + op->u.update.left = NULL; + return 0; +} + +/* plugin->u.node.prepare_removal + to delete a pointer to @empty from the tree add corresponding carry + operation (delete) to @info list */ +int prepare_removal_node40(znode * empty, carry_plugin_info * info) +{ + carry_op *op; + reiser4_tree *tree; + + if (!should_notify_parent(empty)) + return 0; + /* already on a road to Styx */ + if (ZF_ISSET(empty, JNODE_HEARD_BANSHEE)) + return 0; + op = node_post_carry(info, COP_DELETE, empty, 1); + if (IS_ERR(op) || op == NULL) + return RETERR(op ? PTR_ERR(op) : -EIO); + + op->u.delete.child = NULL; + op->u.delete.flags = 0; + + /* fare thee well */ + tree = znode_get_tree(empty); + read_lock_tree(tree); + write_lock_dk(tree); + znode_set_ld_key(empty, znode_get_rd_key(empty)); + if (znode_is_left_connected(empty) && empty->left) + znode_set_rd_key(empty->left, znode_get_rd_key(empty)); + write_unlock_dk(tree); + read_unlock_tree(tree); + + ZF_SET(empty, JNODE_HEARD_BANSHEE); + return 0; +} + +/* something were shifted from @insert_coord->node to @shift->target, update + @insert_coord correspondingly */ +static void +adjust_coord(coord_t * insert_coord, struct shift_params *shift, int removed, + int including_insert_coord) +{ + /* item plugin was invalidated by shifting */ + coord_clear_iplug(insert_coord); + + if (node_is_empty(shift->wish_stop.node)) { + assert("vs-242", shift->everything); + if (including_insert_coord) { + if (shift->pend == SHIFT_RIGHT) { + /* set @insert_coord before first unit of + @shift->target node */ + coord_init_before_first_item(insert_coord, + shift->target); + } else { + /* set @insert_coord after last in target node */ + coord_init_after_last_item(insert_coord, + shift->target); + } + } else { + /* set @insert_coord inside of empty node. There is + only one possible coord within an empty + node. init_first_unit will set that coord */ + coord_init_first_unit(insert_coord, + shift->wish_stop.node); + } + return; + } + + if (shift->pend == SHIFT_RIGHT) { + /* there was shifting to right */ + if (shift->everything) { + /* everything wanted was shifted */ + if (including_insert_coord) { + /* @insert_coord is set before first unit of + @to node */ + coord_init_before_first_item(insert_coord, + shift->target); + insert_coord->between = BEFORE_UNIT; + } else { + /* @insert_coord is set after last unit of + @insert->node */ + coord_init_last_unit(insert_coord, + shift->wish_stop.node); + insert_coord->between = AFTER_UNIT; + } + } + return; + } + + /* there was shifting to left */ + if (shift->everything) { + /* everything wanted was shifted */ + if (including_insert_coord) { + /* @insert_coord is set after last unit in @to node */ + coord_init_after_last_item(insert_coord, shift->target); + } else { + /* @insert_coord is set before first unit in the same + node */ + coord_init_before_first_item(insert_coord, + shift->wish_stop.node); + } + return; + } + + /* FIXME-VS: the code below is complicated because with between == + AFTER_ITEM unit_pos is set to 0 */ + + if (!removed) { + /* no items were shifted entirely */ + assert("vs-195", shift->merging_units == 0 + || shift->part_units == 0); + + if (shift->real_stop.item_pos == insert_coord->item_pos) { + if (shift->merging_units) { + if (insert_coord->between == AFTER_UNIT) { + assert("nikita-1441", + insert_coord->unit_pos >= + shift->merging_units); + insert_coord->unit_pos -= + shift->merging_units; + } else if (insert_coord->between == BEFORE_UNIT) { + assert("nikita-2090", + insert_coord->unit_pos > + shift->merging_units); + insert_coord->unit_pos -= + shift->merging_units; + } + + assert("nikita-2083", + insert_coord->unit_pos + 1); + } else { + if (insert_coord->between == AFTER_UNIT) { + assert("nikita-1442", + insert_coord->unit_pos >= + shift->part_units); + insert_coord->unit_pos -= + shift->part_units; + } else if (insert_coord->between == BEFORE_UNIT) { + assert("nikita-2089", + insert_coord->unit_pos > + shift->part_units); + insert_coord->unit_pos -= + shift->part_units; + } + + assert("nikita-2084", + insert_coord->unit_pos + 1); + } + } + return; + } + + /* we shifted to left and there was no enough space for everything */ + switch (insert_coord->between) { + case AFTER_UNIT: + case BEFORE_UNIT: + if (shift->real_stop.item_pos == insert_coord->item_pos) + insert_coord->unit_pos -= shift->part_units; + case AFTER_ITEM: + coord_add_item_pos(insert_coord, -removed); + break; + default: + impossible("nikita-2087", "not ready"); + } + assert("nikita-2085", insert_coord->unit_pos + 1); +} + +static int call_shift_hooks(struct shift_params *shift) +{ + unsigned i, shifted; + coord_t coord; + item_plugin *iplug; + + assert("vs-275", !node_is_empty(shift->target)); + + /* number of items shift touches */ + shifted = + shift->entire + (shift->merging_units ? 1 : 0) + + (shift->part_units ? 1 : 0); + + if (shift->pend == SHIFT_LEFT) { + /* moved items are at the end */ + coord_init_last_unit(&coord, shift->target); + coord.unit_pos = 0; + + assert("vs-279", shift->pend == 1); + for (i = 0; i < shifted; i++) { + unsigned from, count; + + iplug = item_plugin_by_coord(&coord); + if (i == 0 && shift->part_units) { + assert("vs-277", + coord_num_units(&coord) == + shift->part_units); + count = shift->part_units; + from = 0; + } else if (i == shifted - 1 && shift->merging_units) { + count = shift->merging_units; + from = coord_num_units(&coord) - count; + } else { + count = coord_num_units(&coord); + from = 0; + } + + if (iplug->b.shift_hook) { + iplug->b.shift_hook(&coord, from, count, + shift->wish_stop.node); + } + coord_add_item_pos(&coord, -shift->pend); + } + } else { + /* moved items are at the beginning */ + coord_init_first_unit(&coord, shift->target); + + assert("vs-278", shift->pend == -1); + for (i = 0; i < shifted; i++) { + unsigned from, count; + + iplug = item_plugin_by_coord(&coord); + if (i == 0 && shift->part_units) { + assert("vs-277", + coord_num_units(&coord) == + shift->part_units); + count = coord_num_units(&coord); + from = 0; + } else if (i == shifted - 1 && shift->merging_units) { + count = shift->merging_units; + from = 0; + } else { + count = coord_num_units(&coord); + from = 0; + } + + if (iplug->b.shift_hook) { + iplug->b.shift_hook(&coord, from, count, + shift->wish_stop.node); + } + coord_add_item_pos(&coord, -shift->pend); + } + } + + return 0; +} + +/* shift to left is completed. Return 1 if unit @old was moved to left neighbor */ +static int +unit_moved_left(const struct shift_params *shift, const coord_t * old) +{ + assert("vs-944", shift->real_stop.node == old->node); + + if (shift->real_stop.item_pos < old->item_pos) + return 0; + if (shift->real_stop.item_pos == old->item_pos) { + if (shift->real_stop.unit_pos < old->unit_pos) + return 0; + } + return 1; +} + +/* shift to right is completed. Return 1 if unit @old was moved to right + neighbor */ +static int +unit_moved_right(const struct shift_params *shift, const coord_t * old) +{ + assert("vs-944", shift->real_stop.node == old->node); + + if (shift->real_stop.item_pos > old->item_pos) + return 0; + if (shift->real_stop.item_pos == old->item_pos) { + if (shift->real_stop.unit_pos > old->unit_pos) + return 0; + } + return 1; +} + +/* coord @old was set in node from which shift was performed. What was shifted + is stored in @shift. Update @old correspondingly to performed shift */ +static coord_t *adjust_coord2(const struct shift_params *shift, + const coord_t * old, coord_t * new) +{ + coord_clear_iplug(new); + new->between = old->between; + + coord_clear_iplug(new); + if (old->node == shift->target) { + if (shift->pend == SHIFT_LEFT) { + /* coord which is set inside of left neighbor does not + change during shift to left */ + coord_dup(new, old); + return new; + } + new->node = old->node; + coord_set_item_pos(new, + old->item_pos + shift->entire + + (shift->part_units ? 1 : 0)); + new->unit_pos = old->unit_pos; + if (old->item_pos == 0 && shift->merging_units) + new->unit_pos += shift->merging_units; + return new; + } + + assert("vs-977", old->node == shift->wish_stop.node); + if (shift->pend == SHIFT_LEFT) { + if (unit_moved_left(shift, old)) { + /* unit @old moved to left neighbor. Calculate its + coordinate there */ + new->node = shift->target; + coord_set_item_pos(new, + node_num_items(shift->target) - + shift->entire - + (shift->part_units ? 1 : 0) + + old->item_pos); + + new->unit_pos = old->unit_pos; + if (shift->merging_units) { + coord_dec_item_pos(new); + if (old->item_pos == 0) { + /* unit_pos only changes if item got + merged */ + new->unit_pos = + coord_num_units(new) - + (shift->merging_units - + old->unit_pos); + } + } + } else { + /* unit @old did not move to left neighbor. + + Use _nocheck, because @old is outside of its node. + */ + coord_dup_nocheck(new, old); + coord_add_item_pos(new, + -shift->u.future_first.item_pos); + if (new->item_pos == 0) + new->unit_pos -= shift->u.future_first.unit_pos; + } + } else { + if (unit_moved_right(shift, old)) { + /* unit @old moved to right neighbor */ + new->node = shift->target; + coord_set_item_pos(new, + old->item_pos - + shift->real_stop.item_pos); + if (new->item_pos == 0) { + /* unit @old might change unit pos */ + coord_set_item_pos(new, + old->unit_pos - + shift->real_stop.unit_pos); + } + } else { + /* unit @old did not move to right neighbor, therefore + it did not change */ + coord_dup(new, old); + } + } + coord_set_iplug(new, item_plugin_by_coord(new)); + return new; +} + +/* this is called when shift is completed (something of source node is copied + to target and deleted in source) to update all taps set in current + context */ +static void update_taps(const struct shift_params *shift) +{ + tap_t *tap; + coord_t new; + + for_all_taps(tap) { + /* update only taps set to nodes participating in shift */ + if (tap->coord->node == shift->wish_stop.node + || tap->coord->node == shift->target) + tap_to_coord(tap, + adjust_coord2(shift, tap->coord, &new)); + } +} + +#if REISER4_DEBUG + +struct shift_check { + reiser4_key key; + __u16 plugin_id; + union { + __u64 bytes; + __u64 entries; + void *unused; + } u; +}; + +void *shift_check_prepare(const znode * left, const znode * right) +{ + pos_in_node_t i, nr_items; + int mergeable; + struct shift_check *data; + item_header40 *ih; + + if (node_is_empty(left) || node_is_empty(right)) + mergeable = 0; + else { + coord_t l, r; + + coord_init_last_unit(&l, left); + coord_init_first_unit(&r, right); + mergeable = are_items_mergeable(&l, &r); + } + nr_items = + node40_num_of_items_internal(left) + + node40_num_of_items_internal(right) - (mergeable ? 1 : 0); + data = + kmalloc(sizeof(struct shift_check) * nr_items, + reiser4_ctx_gfp_mask_get()); + if (data != NULL) { + coord_t coord; + pos_in_node_t item_pos; + + coord_init_first_unit(&coord, left); + i = 0; + + for (item_pos = 0; + item_pos < node40_num_of_items_internal(left); + item_pos++) { + + coord_set_item_pos(&coord, item_pos); + ih = node40_ih_at_coord(&coord); + + data[i].key = ih->key; + data[i].plugin_id = le16_to_cpu(get_unaligned(&ih->plugin_id)); + switch (data[i].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + data[i].u.bytes = coord_num_units(&coord); + break; + case EXTENT_POINTER_ID: + data[i].u.bytes = + reiser4_extent_size(&coord, + coord_num_units(&coord)); + break; + case COMPOUND_DIR_ID: + data[i].u.entries = coord_num_units(&coord); + break; + default: + data[i].u.unused = NULL; + break; + } + i++; + } + + coord_init_first_unit(&coord, right); + + if (mergeable) { + assert("vs-1609", i != 0); + + ih = node40_ih_at_coord(&coord); + + assert("vs-1589", + data[i - 1].plugin_id == + le16_to_cpu(get_unaligned(&ih->plugin_id))); + switch (data[i - 1].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + data[i - 1].u.bytes += coord_num_units(&coord); + break; + case EXTENT_POINTER_ID: + data[i - 1].u.bytes += + reiser4_extent_size(&coord, + coord_num_units(&coord)); + break; + case COMPOUND_DIR_ID: + data[i - 1].u.entries += + coord_num_units(&coord); + break; + default: + impossible("vs-1605", "wrong mergeable item"); + break; + } + item_pos = 1; + } else + item_pos = 0; + for (; item_pos < node40_num_of_items_internal(right); + item_pos++) { + + assert("vs-1604", i < nr_items); + coord_set_item_pos(&coord, item_pos); + ih = node40_ih_at_coord(&coord); + + data[i].key = ih->key; + data[i].plugin_id = le16_to_cpu(get_unaligned(&ih->plugin_id)); + switch (data[i].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + data[i].u.bytes = coord_num_units(&coord); + break; + case EXTENT_POINTER_ID: + data[i].u.bytes = + reiser4_extent_size(&coord, + coord_num_units(&coord)); + break; + case COMPOUND_DIR_ID: + data[i].u.entries = coord_num_units(&coord); + break; + default: + data[i].u.unused = NULL; + break; + } + i++; + } + assert("vs-1606", i == nr_items); + } + return data; +} + +void shift_check(void *vp, const znode * left, const znode * right) +{ + pos_in_node_t i, nr_items; + coord_t coord; + __u64 last_bytes; + int mergeable; + item_header40 *ih; + pos_in_node_t item_pos; + struct shift_check *data; + + data = (struct shift_check *)vp; + + if (data == NULL) + return; + + if (node_is_empty(left) || node_is_empty(right)) + mergeable = 0; + else { + coord_t l, r; + + coord_init_last_unit(&l, left); + coord_init_first_unit(&r, right); + mergeable = are_items_mergeable(&l, &r); + } + + nr_items = + node40_num_of_items_internal(left) + + node40_num_of_items_internal(right) - (mergeable ? 1 : 0); + + i = 0; + last_bytes = 0; + + coord_init_first_unit(&coord, left); + + for (item_pos = 0; item_pos < node40_num_of_items_internal(left); + item_pos++) { + + coord_set_item_pos(&coord, item_pos); + ih = node40_ih_at_coord(&coord); + + assert("vs-1611", i == item_pos); + assert("vs-1590", keyeq(&ih->key, &data[i].key)); + assert("vs-1591", + le16_to_cpu(get_unaligned(&ih->plugin_id)) == data[i].plugin_id); + if ((i < (node40_num_of_items_internal(left) - 1)) + || !mergeable) { + switch (data[i].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + assert("vs-1592", + data[i].u.bytes == + coord_num_units(&coord)); + break; + case EXTENT_POINTER_ID: + assert("vs-1593", + data[i].u.bytes == + reiser4_extent_size(&coord, + coord_num_units + (&coord))); + break; + case COMPOUND_DIR_ID: + assert("vs-1594", + data[i].u.entries == + coord_num_units(&coord)); + break; + default: + break; + } + } + if (item_pos == (node40_num_of_items_internal(left) - 1) + && mergeable) { + switch (data[i].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + last_bytes = coord_num_units(&coord); + break; + case EXTENT_POINTER_ID: + last_bytes = + reiser4_extent_size(&coord, + coord_num_units(&coord)); + break; + case COMPOUND_DIR_ID: + last_bytes = coord_num_units(&coord); + break; + default: + impossible("vs-1595", "wrong mergeable item"); + break; + } + } + i++; + } + + coord_init_first_unit(&coord, right); + if (mergeable) { + ih = node40_ih_at_coord(&coord); + + assert("vs-1589", + data[i - 1].plugin_id == le16_to_cpu(get_unaligned(&ih->plugin_id))); + assert("vs-1608", last_bytes != 0); + switch (data[i - 1].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + assert("vs-1596", + data[i - 1].u.bytes == + last_bytes + coord_num_units(&coord)); + break; + + case EXTENT_POINTER_ID: + assert("vs-1597", + data[i - 1].u.bytes == + last_bytes + reiser4_extent_size(&coord, + coord_num_units + (&coord))); + break; + + case COMPOUND_DIR_ID: + assert("vs-1598", + data[i - 1].u.bytes == + last_bytes + coord_num_units(&coord)); + break; + default: + impossible("vs-1599", "wrong mergeable item"); + break; + } + item_pos = 1; + } else + item_pos = 0; + + for (; item_pos < node40_num_of_items_internal(right); item_pos++) { + + coord_set_item_pos(&coord, item_pos); + ih = node40_ih_at_coord(&coord); + + assert("vs-1612", keyeq(&ih->key, &data[i].key)); + assert("vs-1613", + le16_to_cpu(get_unaligned(&ih->plugin_id)) == data[i].plugin_id); + switch (data[i].plugin_id) { + case CTAIL_ID: + case FORMATTING_ID: + assert("vs-1600", + data[i].u.bytes == coord_num_units(&coord)); + break; + case EXTENT_POINTER_ID: + assert("vs-1601", + data[i].u.bytes == + reiser4_extent_size(&coord, + coord_num_units + (&coord))); + break; + case COMPOUND_DIR_ID: + assert("vs-1602", + data[i].u.entries == coord_num_units(&coord)); + break; + default: + break; + } + i++; + } + + assert("vs-1603", i == nr_items); + kfree(data); +} + +#endif + +/* plugin->u.node.shift + look for description of this method in plugin/node/node.h */ +int shift_node40(coord_t * from, znode * to, shift_direction pend, int delete_child, /* if @from->node becomes empty - it will be + deleted from the tree if this is set to 1 */ + int including_stop_coord, carry_plugin_info * info) +{ + struct shift_params shift; + int result; + znode *left, *right; + znode *source; + int target_empty; + + assert("nikita-2161", coord_check(from)); + + memset(&shift, 0, sizeof(shift)); + shift.pend = pend; + shift.wish_stop = *from; + shift.target = to; + + assert("nikita-1473", znode_is_write_locked(from->node)); + assert("nikita-1474", znode_is_write_locked(to)); + + source = from->node; + + /* set @shift.wish_stop to rightmost/leftmost unit among units we want + shifted */ + if (pend == SHIFT_LEFT) { + result = coord_set_to_left(&shift.wish_stop); + left = to; + right = from->node; + } else { + result = coord_set_to_right(&shift.wish_stop); + left = from->node; + right = to; + } + + if (result) { + /* move insertion coord even if there is nothing to move */ + if (including_stop_coord) { + /* move insertion coord (@from) */ + if (pend == SHIFT_LEFT) { + /* after last item in target node */ + coord_init_after_last_item(from, to); + } else { + /* before first item in target node */ + coord_init_before_first_item(from, to); + } + } + + if (delete_child && node_is_empty(shift.wish_stop.node)) + result = + prepare_removal_node40(shift.wish_stop.node, info); + else + result = 0; + /* there is nothing to shift */ + assert("nikita-2078", coord_check(from)); + return result; + } + + target_empty = node_is_empty(to); + + /* when first node plugin with item body compression is implemented, + this must be changed to call node specific plugin */ + + /* shift->stop_coord is updated to last unit which really will be + shifted */ + estimate_shift(&shift, get_current_context()); + if (!shift.shift_bytes) { + /* we could not shift anything */ + assert("nikita-2079", coord_check(from)); + return 0; + } + + copy(&shift); + + /* result value of this is important. It is used by adjust_coord below */ + result = delete_copied(&shift); + + assert("vs-1610", result >= 0); + assert("vs-1471", + ((reiser4_context *) current->journal_info)->magic == + context_magic); + + /* item which has been moved from one node to another might want to do + something on that event. This can be done by item's shift_hook + method, which will be now called for every moved items */ + call_shift_hooks(&shift); + + assert("vs-1472", + ((reiser4_context *) current->journal_info)->magic == + context_magic); + + update_taps(&shift); + + assert("vs-1473", + ((reiser4_context *) current->journal_info)->magic == + context_magic); + + /* adjust @from pointer in accordance with @including_stop_coord flag + and amount of data which was really shifted */ + adjust_coord(from, &shift, result, including_stop_coord); + + if (target_empty) + /* + * items were shifted into empty node. Update delimiting key. + */ + result = prepare_for_update(NULL, left, info); + + /* add update operation to @info, which is the list of operations to + be performed on a higher level */ + result = prepare_for_update(left, right, info); + if (!result && node_is_empty(source) && delete_child) { + /* all contents of @from->node is moved to @to and @from->node + has to be removed from the tree, so, on higher level we + will be removing the pointer to node @from->node */ + result = prepare_removal_node40(source, info); + } + assert("nikita-2080", coord_check(from)); + return result ? result : (int)shift.shift_bytes; +} + +/* plugin->u.node.fast_insert() + look for description of this method in plugin/node/node.h */ +int fast_insert_node40(const coord_t * coord UNUSED_ARG /* node to query */ ) +{ + return 1; +} + +/* plugin->u.node.fast_paste() + look for description of this method in plugin/node/node.h */ +int fast_paste_node40(const coord_t * coord UNUSED_ARG /* node to query */ ) +{ + return 1; +} + +/* plugin->u.node.fast_cut() + look for description of this method in plugin/node/node.h */ +int fast_cut_node40(const coord_t * coord UNUSED_ARG /* node to query */ ) +{ + return 1; +} + +/* plugin->u.node.modify - not defined */ + +/* plugin->u.node.max_item_size */ +int max_item_size_node40(void) +{ + return reiser4_get_current_sb()->s_blocksize - sizeof(node40_header) - + sizeof(item_header40); +} + +/* plugin->u.node.set_item_plugin */ +int set_item_plugin_node40(coord_t *coord, item_id id) +{ + item_header40 *ih; + + ih = node40_ih_at_coord(coord); + put_unaligned(cpu_to_le16(id), &ih->plugin_id); + coord->iplugid = id; + return 0; +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/node/node40.h linux-2.6.30/fs/reiser4/plugin/node/node40.h --- linux-2.6.30.orig/fs/reiser4/plugin/node/node40.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/node/node40.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,125 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#if !defined( __REISER4_NODE40_H__ ) +#define __REISER4_NODE40_H__ + +#include "../../forward.h" +#include "../../dformat.h" +#include "node.h" + +#include + +/* format of node header for 40 node layouts. Keep bloat out of this struct. */ +typedef struct node40_header { + /* identifier of node plugin. Must be located at the very beginning + of a node. */ + common_node_header common_header; /* this is 16 bits */ + /* number of items. Should be first element in the node header, + because we haven't yet finally decided whether it shouldn't go into + common_header. + */ +/* NIKITA-FIXME-HANS: Create a macro such that if there is only one + * node format at compile time, and it is this one, accesses do not function dereference when + * accessing these fields (and otherwise they do). Probably 80% of users will only have one node format at a time throughout the life of reiser4. */ + d16 nr_items; + /* free space in node measured in bytes */ + d16 free_space; + /* offset to start of free space in node */ + d16 free_space_start; + /* for reiser4_fsck. When information about what is a free + block is corrupted, and we try to recover everything even + if marked as freed, then old versions of data may + duplicate newer versions, and this field allows us to + restore the newer version. Also useful for when users + who don't have the new trashcan installed on their linux distro + delete the wrong files and send us desperate emails + offering $25 for them back. */ + + /* magic field we need to tell formatted nodes NIKITA-FIXME-HANS: improve this comment */ + d32 magic; + /* flushstamp is made of mk_id and write_counter. mk_id is an + id generated randomly at mkreiserfs time. So we can just + skip all nodes with different mk_id. write_counter is d64 + incrementing counter of writes on disk. It is used for + choosing the newest data at fsck time. NIKITA-FIXME-HANS: why was field name changed but not comment? */ + + d32 mkfs_id; + d64 flush_id; + /* node flags to be used by fsck (reiser4ck or reiser4fsck?) + and repacker NIKITA-FIXME-HANS: say more or reference elsewhere that says more */ + d16 flags; + + /* 1 is leaf level, 2 is twig level, root is the numerically + largest level */ + d8 level; + + d8 pad; +} PACKED node40_header; + +/* item headers are not standard across all node layouts, pass + pos_in_node to functions instead */ +typedef struct item_header40 { + /* key of item */ + /* 0 */ reiser4_key key; + /* offset from start of a node measured in 8-byte chunks */ + /* 24 */ d16 offset; + /* 26 */ d16 flags; + /* 28 */ d16 plugin_id; +} PACKED item_header40; + +size_t item_overhead_node40(const znode * node, flow_t * aflow); +size_t free_space_node40(znode * node); +node_search_result lookup_node40(znode * node, const reiser4_key * key, + lookup_bias bias, coord_t * coord); +int num_of_items_node40(const znode * node); +char *item_by_coord_node40(const coord_t * coord); +int length_by_coord_node40(const coord_t * coord); +item_plugin *plugin_by_coord_node40(const coord_t * coord); +reiser4_key *key_at_node40(const coord_t * coord, reiser4_key * key); +size_t estimate_node40(znode * node); +int check_node40(const znode * node, __u32 flags, const char **error); +int parse_node40(znode * node); +int init_node40(znode * node); +#ifdef GUESS_EXISTS +int guess_node40(const znode * node); +#endif +void change_item_size_node40(coord_t * coord, int by); +int create_item_node40(coord_t * target, const reiser4_key * key, + reiser4_item_data * data, carry_plugin_info * info); +void update_item_key_node40(coord_t * target, const reiser4_key * key, + carry_plugin_info * info); +int kill_node40(struct carry_kill_data *, carry_plugin_info *); +int cut_node40(struct carry_cut_data *, carry_plugin_info *); +int shift_node40(coord_t * from, znode * to, shift_direction pend, + /* if @from->node becomes + empty - it will be deleted from + the tree if this is set to 1 + */ + int delete_child, int including_stop_coord, + carry_plugin_info * info); + +int fast_insert_node40(const coord_t * coord); +int fast_paste_node40(const coord_t * coord); +int fast_cut_node40(const coord_t * coord); +int max_item_size_node40(void); +int prepare_removal_node40(znode * empty, carry_plugin_info * info); +int set_item_plugin_node40(coord_t * coord, item_id id); +int shrink_item_node40(coord_t * coord, int delta); + +#if REISER4_DEBUG +void *shift_check_prepare(const znode *left, const znode *right); +void shift_check(void *vp, const znode *left, const znode *right); +#endif + +/* __REISER4_NODE40_H__ */ +#endif +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/node/node.c linux-2.6.30/fs/reiser4/plugin/node/node.c --- linux-2.6.30.orig/fs/reiser4/plugin/node/node.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/node/node.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,131 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Node plugin interface. + + Description: The tree provides the abstraction of flows, which it + internally fragments into items which it stores in nodes. + + A key_atom is a piece of data bound to a single key. + + For reasonable space efficiency to be achieved it is often + necessary to store key_atoms in the nodes in the form of items, where + an item is a sequence of key_atoms of the same or similar type. It is + more space-efficient, because the item can implement (very) + efficient compression of key_atom's bodies using internal knowledge + about their semantics, and it can often avoid having a key for each + key_atom. Each type of item has specific operations implemented by its + item handler (see balance.c). + + Rationale: the rest of the code (specifically balancing routines) + accesses leaf level nodes through this interface. This way we can + implement various block layouts and even combine various layouts + within the same tree. Balancing/allocating algorithms should not + care about peculiarities of splitting/merging specific item types, + but rather should leave that to the item's item handler. + + Items, including those that provide the abstraction of flows, have + the property that if you move them in part or in whole to another + node, the balancing code invokes their is_left_mergeable() + item_operation to determine if they are mergeable with their new + neighbor in the node you have moved them to. For some items the + is_left_mergeable() function always returns null. + + When moving the bodies of items from one node to another: + + if a partial item is shifted to another node the balancing code invokes + an item handler method to handle the item splitting. + + if the balancing code needs to merge with an item in the node it + is shifting to, it will invoke an item handler method to handle + the item merging. + + if it needs to move whole item bodies unchanged, the balancing code uses xmemcpy() + adjusting the item headers after the move is done using the node handler. +*/ + +#include "../../forward.h" +#include "../../debug.h" +#include "../../key.h" +#include "../../coord.h" +#include "../plugin_header.h" +#include "../item/item.h" +#include "node.h" +#include "../plugin.h" +#include "../../znode.h" +#include "../../tree.h" +#include "../../super.h" +#include "../../reiser4.h" + +/** + * leftmost_key_in_node - get the smallest key in node + * @node: + * @key: store result here + * + * Stores the leftmost key of @node in @key. + */ +reiser4_key *leftmost_key_in_node(const znode *node, reiser4_key *key) +{ + assert("nikita-1634", node != NULL); + assert("nikita-1635", key != NULL); + + if (!node_is_empty(node)) { + coord_t first_item; + + coord_init_first_unit(&first_item, (znode *) node); + item_key_by_coord(&first_item, key); + } else + *key = *reiser4_max_key(); + return key; +} + +node_plugin node_plugins[LAST_NODE_ID] = { + [NODE40_ID] = { + .h = { + .type_id = REISER4_NODE_PLUGIN_TYPE, + .id = NODE40_ID, + .pops = NULL, + .label = "unified", + .desc = "unified node layout", + .linkage = {NULL, NULL} + }, + .item_overhead = item_overhead_node40, + .free_space = free_space_node40, + .lookup = lookup_node40, + .num_of_items = num_of_items_node40, + .item_by_coord = item_by_coord_node40, + .length_by_coord = length_by_coord_node40, + .plugin_by_coord = plugin_by_coord_node40, + .key_at = key_at_node40, + .estimate = estimate_node40, + .check = check_node40, + .parse = parse_node40, + .init = init_node40, +#ifdef GUESS_EXISTS + .guess = guess_node40, +#endif + .change_item_size = change_item_size_node40, + .create_item = create_item_node40, + .update_item_key = update_item_key_node40, + .cut_and_kill = kill_node40, + .cut = cut_node40, + .shift = shift_node40, + .shrink_item = shrink_item_node40, + .fast_insert = fast_insert_node40, + .fast_paste = fast_paste_node40, + .fast_cut = fast_cut_node40, + .max_item_size = max_item_size_node40, + .prepare_removal = prepare_removal_node40, + .set_item_plugin = set_item_plugin_node40 + } +}; + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/node/node.h linux-2.6.30/fs/reiser4/plugin/node/node.h --- linux-2.6.30.orig/fs/reiser4/plugin/node/node.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/node/node.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,272 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* We need a definition of the default node layout here. */ + +/* Generally speaking, it is best to have free space in the middle of the + node so that two sets of things can grow towards it, and to have the + item bodies on the left so that the last one of them grows into free + space. We optimize for the case where we append new items to the end + of the node, or grow the last item, because it hurts nothing to so + optimize and it is a common special case to do massive insertions in + increasing key order (and one of cases more likely to have a real user + notice the delay time for). + + formatted leaf default layout: (leaf1) + + |node header:item bodies:free space:key + pluginid + item offset| + + We grow towards the middle, optimizing layout for the case where we + append new items to the end of the node. The node header is fixed + length. Keys, and item offsets plus pluginids for the items + corresponding to them are in increasing key order, and are fixed + length. Item offsets are relative to start of node (16 bits creating + a node size limit of 64k, 12 bits might be a better choice....). Item + bodies are in decreasing key order. Item bodies have a variable size. + There is a one to one to one mapping of keys to item offsets to item + bodies. Item offsets consist of pointers to the zeroth byte of the + item body. Item length equals the start of the next item minus the + start of this item, except the zeroth item whose length equals the end + of the node minus the start of that item (plus a byte). In other + words, the item length is not recorded anywhere, and it does not need + to be since it is computable. + + Leaf variable length items and keys layout : (lvar) + + |node header:key offset + item offset + pluginid triplets:free space:key bodies:item bodies| + + We grow towards the middle, optimizing layout for the case where we + append new items to the end of the node. The node header is fixed + length. Keys and item offsets for the items corresponding to them are + in increasing key order, and keys are variable length. Item offsets + are relative to start of node (16 bits). Item bodies are in + decreasing key order. Item bodies have a variable size. There is a + one to one to one mapping of keys to item offsets to item bodies. + Item offsets consist of pointers to the zeroth byte of the item body. + Item length equals the start of the next item's key minus the start of + this item, except the zeroth item whose length equals the end of the + node minus the start of that item (plus a byte). + + leaf compressed keys layout: (lcomp) + + |node header:key offset + key inherit + item offset pairs:free space:key bodies:item bodies| + + We grow towards the middle, optimizing layout for the case where we + append new items to the end of the node. The node header is fixed + length. Keys and item offsets for the items corresponding to them are + in increasing key order, and keys are variable length. The "key + inherit" field indicates how much of the key prefix is identical to + the previous key (stem compression as described in "Managing + Gigabytes" is used). key_inherit is a one byte integer. The + intra-node searches performed through this layout are linear searches, + and this is theorized to not hurt performance much due to the high + cost of processor stalls on modern CPUs, and the small number of keys + in a single node. Item offsets are relative to start of node (16 + bits). Item bodies are in decreasing key order. Item bodies have a + variable size. There is a one to one to one mapping of keys to item + offsets to item bodies. Item offsets consist of pointers to the + zeroth byte of the item body. Item length equals the start of the + next item minus the start of this item, except the zeroth item whose + length equals the end of the node minus the start of that item (plus a + byte). In other words, item length and key length is not recorded + anywhere, and it does not need to be since it is computable. + + internal node default layout: (idef1) + + just like ldef1 except that item bodies are either blocknrs of + children or extents, and moving them may require updating parent + pointers in the nodes that they point to. +*/ + +/* There is an inherent 3-way tradeoff between optimizing and + exchanging disks between different architectures and code + complexity. This is optimal and simple and inexchangeable. + Someone else can do the code for exchanging disks and make it + complex. It would not be that hard. Using other than the PAGE_SIZE + might be suboptimal. +*/ + +#if !defined( __REISER4_NODE_H__ ) +#define __REISER4_NODE_H__ + +#define LEAF40_NODE_SIZE PAGE_CACHE_SIZE + +#include "../../dformat.h" +#include "../plugin_header.h" + +#include + +typedef enum { + NS_FOUND = 0, + NS_NOT_FOUND = -ENOENT +} node_search_result; + +/* Maximal possible space overhead for creation of new item in a node */ +#define REISER4_NODE_MAX_OVERHEAD ( sizeof( reiser4_key ) + 32 ) + +typedef enum { + REISER4_NODE_DKEYS = (1 << 0), + REISER4_NODE_TREE_STABLE = (1 << 1) +} reiser4_node_check_flag; + +/* cut and cut_and_kill have too long list of parameters. This structure is just to safe some space on stack */ +struct cut_list { + coord_t *from; + coord_t *to; + const reiser4_key *from_key; + const reiser4_key *to_key; + reiser4_key *smallest_removed; + carry_plugin_info *info; + __u32 flags; + struct inode *inode; /* this is to pass list of eflushed jnodes down to extent_kill_hook */ + lock_handle *left; + lock_handle *right; +}; + +struct carry_cut_data; +struct carry_kill_data; + +/* The responsibility of the node plugin is to store and give access + to the sequence of items within the node. */ +typedef struct node_plugin { + /* generic plugin fields */ + plugin_header h; + + /* calculates the amount of space that will be required to store an + item which is in addition to the space consumed by the item body. + (the space consumed by the item body can be gotten by calling + item->estimate) */ + size_t(*item_overhead) (const znode * node, flow_t * f); + + /* returns free space by looking into node (i.e., without using + znode->free_space). */ + size_t(*free_space) (znode * node); + /* search within the node for the one item which might + contain the key, invoking item->search_within to search within + that item to see if it is in there */ + node_search_result(*lookup) (znode * node, const reiser4_key * key, + lookup_bias bias, coord_t * coord); + /* number of items in node */ + int (*num_of_items) (const znode * node); + + /* store information about item in @coord in @data */ + /* break into several node ops, don't add any more uses of this before doing so */ + /*int ( *item_at )( const coord_t *coord, reiser4_item_data *data ); */ + char *(*item_by_coord) (const coord_t * coord); + int (*length_by_coord) (const coord_t * coord); + item_plugin *(*plugin_by_coord) (const coord_t * coord); + + /* store item key in @key */ + reiser4_key *(*key_at) (const coord_t * coord, reiser4_key * key); + /* conservatively estimate whether unit of what size can fit + into node. This estimation should be performed without + actually looking into the node's content (free space is saved in + znode). */ + size_t(*estimate) (znode * node); + + /* performs every consistency check the node plugin author could + imagine. Optional. */ + int (*check) (const znode * node, __u32 flags, const char **error); + + /* Called when node is read into memory and node plugin is + already detected. This should read some data into znode (like free + space counter) and, optionally, check data consistency. + */ + int (*parse) (znode * node); + /* This method is called on a new node to initialise plugin specific + data (header, etc.) */ + int (*init) (znode * node); + /* Check whether @node content conforms to this plugin format. + Probably only useful after support for old V3.x formats is added. + Uncomment after 4.0 only. + */ + /* int ( *guess )( const znode *node ); */ +#if REISER4_DEBUG + void (*print) (const char *prefix, const znode * node, __u32 flags); +#endif + /* change size of @item by @by bytes. @item->node has enough free + space. When @by > 0 - free space is appended to end of item. When + @by < 0 - item is truncated - it is assumed that last @by bytes if + the item are freed already */ + void (*change_item_size) (coord_t * item, int by); + + /* create new item @length bytes long in coord @target */ + int (*create_item) (coord_t * target, const reiser4_key * key, + reiser4_item_data * data, carry_plugin_info * info); + + /* update key of item. */ + void (*update_item_key) (coord_t * target, const reiser4_key * key, + carry_plugin_info * info); + + int (*cut_and_kill) (struct carry_kill_data *, carry_plugin_info *); + int (*cut) (struct carry_cut_data *, carry_plugin_info *); + + /* + * shrink item pointed to by @coord by @delta bytes. + */ + int (*shrink_item) (coord_t * coord, int delta); + + /* copy as much as possible but not more than up to @stop from + @stop->node to @target. If (pend == append) then data from beginning of + @stop->node are copied to the end of @target. If (pend == prepend) then + data from the end of @stop->node are copied to the beginning of + @target. Copied data are removed from @stop->node. Information + about what to do on upper level is stored in @todo */ + int (*shift) (coord_t * stop, znode * target, shift_direction pend, + int delete_node, int including_insert_coord, + carry_plugin_info * info); + /* return true if this node allows skip carry() in some situations + (see fs/reiser4/tree.c:insert_by_coord()). Reiser3.x format + emulation doesn't. + + This will speedup insertions that doesn't require updates to the + parent, by bypassing initialisation of carry() structures. It's + believed that majority of insertions will fit there. + + */ + int (*fast_insert) (const coord_t * coord); + int (*fast_paste) (const coord_t * coord); + int (*fast_cut) (const coord_t * coord); + /* this limits max size of item which can be inserted into a node and + number of bytes item in a node may be appended with */ + int (*max_item_size) (void); + int (*prepare_removal) (znode * empty, carry_plugin_info * info); + /* change plugin id of items which are in a node already. Currently it is Used in tail conversion for regular + * files */ + int (*set_item_plugin) (coord_t * coord, item_id); +} node_plugin; + +typedef enum { + /* standard unified node layout used for both leaf and internal + nodes */ + NODE40_ID, + LAST_NODE_ID +} reiser4_node_id; + +extern reiser4_key *leftmost_key_in_node(const znode * node, reiser4_key * key); +#if REISER4_DEBUG +extern void print_node_content(const char *prefix, const znode * node, + __u32 flags); +#endif + +extern void indent_znode(const znode * node); + +typedef struct common_node_header { + /* + * identifier of node plugin. Must be located at the very beginning of + * a node. + */ + __le16 plugin_id; +} common_node_header; + +/* __REISER4_NODE_H__ */ +#endif +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/object.c linux-2.6.30/fs/reiser4/plugin/object.c --- linux-2.6.30.orig/fs/reiser4/plugin/object.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/object.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,531 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* + * Examples of object plugins: file, directory, symlink, special file. + * + * Plugins associated with inode: + * + * Plugin of inode is plugin referenced by plugin-id field of on-disk + * stat-data. How we store this plugin in in-core inode is not + * important. Currently pointers are used, another variant is to store offsets + * and do array lookup on each access. + * + * Now, each inode has one selected plugin: object plugin that + * determines what type of file this object is: directory, regular etc. + * + * This main plugin can use other plugins that are thus subordinated to + * it. Directory instance of object plugin uses hash; regular file + * instance uses tail policy plugin. + * + * Object plugin is either taken from id in stat-data or guessed from + * i_mode bits. Once it is established we ask it to install its + * subordinate plugins, by looking again in stat-data or inheriting them + * from parent. + * + * How new inode is initialized during ->read_inode(): + * 1 read stat-data and initialize inode fields: i_size, i_mode, + * i_generation, capabilities etc. + * 2 read plugin id from stat data or try to guess plugin id + * from inode->i_mode bits if plugin id is missing. + * 3 Call ->init_inode() method of stat-data plugin to initialise inode fields. + * + * NIKITA-FIXME-HANS: can you say a little about 1 being done before 3? What + * if stat data does contain i_size, etc., due to it being an unusual plugin? + * + * 4 Call ->activate() method of object's plugin. Plugin is either read from + * from stat-data or guessed from mode bits + * 5 Call ->inherit() method of object plugin to inherit as yet un initialized + * plugins from parent. + * + * Easy induction proves that on last step all plugins of inode would be + * initialized. + * + * When creating new object: + * 1 obtain object plugin id (see next period) + * NIKITA-FIXME-HANS: period? + * 2 ->install() this plugin + * 3 ->inherit() the rest from the parent + * + * We need some examples of creating an object with default and non-default + * plugin ids. Nikita, please create them. + */ + +#include "../inode.h" + +static int _bugop(void) +{ + BUG_ON(1); + return 0; +} + +#define bugop ((void *)_bugop) + +static int _dummyop(void) +{ + return 0; +} + +#define dummyop ((void *)_dummyop) + +static int change_file(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + /* cannot change object plugin of already existing object */ + if (memb == PSET_FILE) + return RETERR(-EINVAL); + + /* Change PSET_CREATE */ + return aset_set_unsafe(&reiser4_inode_data(inode)->pset, memb, plugin); +} + +static reiser4_plugin_ops file_plugin_ops = { + .change = change_file +}; + +static struct inode_operations null_i_ops = {.create = NULL}; +static struct file_operations null_f_ops = {.owner = NULL}; +static struct address_space_operations null_a_ops = {.writepage = NULL}; + +/* VFS methods for regular files */ +static struct inode_operations regular_file_i_ops = { + .permission = reiser4_permission_common, + .setattr = reiser4_setattr, + .getattr = reiser4_getattr_common +}; +static struct file_operations regular_file_f_ops = { + .llseek = generic_file_llseek, + .read = reiser4_read_careful, + .write = reiser4_write_careful, + .aio_read = generic_file_aio_read, + .ioctl = reiser4_ioctl_careful, + .mmap = reiser4_mmap_careful, + .open = reiser4_open_careful, + .release = reiser4_release_careful, + .fsync = reiser4_sync_file_common, + .splice_read = generic_file_splice_read, + .splice_write = generic_file_splice_write +}; +static struct address_space_operations regular_file_a_ops = { + .writepage = reiser4_writepage, + .readpage = reiser4_readpage, + .sync_page = block_sync_page, + .writepages = reiser4_writepages, + .set_page_dirty = reiser4_set_page_dirty, + .readpages = reiser4_readpages, + .write_begin = reiser4_write_begin_careful, + .write_end = reiser4_write_end_careful, + .bmap = reiser4_bmap_careful, + .invalidatepage = reiser4_invalidatepage, + .releasepage = reiser4_releasepage +}; + +/* VFS methods for symlink files */ +static struct inode_operations symlink_file_i_ops = { + .readlink = generic_readlink, + .follow_link = reiser4_follow_link_common, + .permission = reiser4_permission_common, + .setattr = reiser4_setattr_common, + .getattr = reiser4_getattr_common +}; + +/* VFS methods for special files */ +static struct inode_operations special_file_i_ops = { + .permission = reiser4_permission_common, + .setattr = reiser4_setattr_common, + .getattr = reiser4_getattr_common +}; + +/* VFS methods for directories */ +static struct inode_operations directory_i_ops = { + .create = reiser4_create_common, + .lookup = reiser4_lookup_common, + .link = reiser4_link_common, + .unlink = reiser4_unlink_common, + .symlink = reiser4_symlink_common, + .mkdir = reiser4_mkdir_common, + .rmdir = reiser4_unlink_common, + .mknod = reiser4_mknod_common, + .rename = reiser4_rename_common, + .permission = reiser4_permission_common, + .setattr = reiser4_setattr_common, + .getattr = reiser4_getattr_common +}; +static struct file_operations directory_f_ops = { + .llseek = reiser4_llseek_dir_common, + .read = generic_read_dir, + .readdir = reiser4_readdir_common, + .release = reiser4_release_dir_common, + .fsync = reiser4_sync_common +}; +static struct address_space_operations directory_a_ops = { + .writepage = bugop, + .sync_page = bugop, + .writepages = dummyop, + .set_page_dirty = bugop, + .readpages = bugop, + .write_begin = bugop, + .write_end = bugop, + .bmap = bugop, + .invalidatepage = bugop, + .releasepage = bugop +}; + +/* + * Definitions of object plugins. + */ + +file_plugin file_plugins[LAST_FILE_PLUGIN_ID] = { + [UNIX_FILE_PLUGIN_ID] = { + .h = { + .type_id = REISER4_FILE_PLUGIN_TYPE, + .id = UNIX_FILE_PLUGIN_ID, + .groups = (1 << REISER4_REGULAR_FILE), + .pops = &file_plugin_ops, + .label = "reg", + .desc = "regular file", + .linkage = {NULL, NULL}, + }, + /* + * invariant vfs ops + */ + .inode_ops = ®ular_file_i_ops, + .file_ops = ®ular_file_f_ops, + .as_ops = ®ular_file_a_ops, + /* + * private i_ops + */ + .setattr = setattr_unix_file, + .open = open_unix_file, + .read = read_unix_file, + .write = write_unix_file, + .ioctl = ioctl_unix_file, + .mmap = mmap_unix_file, + .release = release_unix_file, + /* + * private f_ops + */ + .readpage = readpage_unix_file, + .readpages = readpages_unix_file, + .writepages = writepages_unix_file, + .write_begin = write_begin_unix_file, + .write_end = write_end_unix_file, + /* + * private a_ops + */ + .bmap = bmap_unix_file, + /* + * other private methods + */ + .write_sd_by_inode = write_sd_by_inode_common, + .flow_by_inode = flow_by_inode_unix_file, + .key_by_inode = key_by_inode_and_offset_common, + .set_plug_in_inode = set_plug_in_inode_common, + .adjust_to_parent = adjust_to_parent_common, + .create_object = reiser4_create_object_common, + .delete_object = delete_object_unix_file, + .add_link = reiser4_add_link_common, + .rem_link = reiser4_rem_link_common, + .owns_item = owns_item_unix_file, + .can_add_link = can_add_link_common, + .detach = dummyop, + .bind = dummyop, + .safelink = safelink_common, + .estimate = { + .create = estimate_create_common, + .update = estimate_update_common, + .unlink = estimate_unlink_common + }, + .init_inode_data = init_inode_data_unix_file, + .cut_tree_worker = cut_tree_worker_common, + .wire = { + .write = wire_write_common, + .read = wire_read_common, + .get = wire_get_common, + .size = wire_size_common, + .done = wire_done_common + } + }, + [DIRECTORY_FILE_PLUGIN_ID] = { + .h = { + .type_id = REISER4_FILE_PLUGIN_TYPE, + .id = DIRECTORY_FILE_PLUGIN_ID, + .groups = (1 << REISER4_DIRECTORY_FILE), + .pops = &file_plugin_ops, + .label = "dir", + .desc = "directory", + .linkage = {NULL, NULL} + }, + .inode_ops = &null_i_ops, + .file_ops = &null_f_ops, + .as_ops = &null_a_ops, + + .write_sd_by_inode = write_sd_by_inode_common, + .flow_by_inode = bugop, + .key_by_inode = bugop, + .set_plug_in_inode = set_plug_in_inode_common, + .adjust_to_parent = adjust_to_parent_common_dir, + .create_object = reiser4_create_object_common, + .delete_object = reiser4_delete_dir_common, + .add_link = reiser4_add_link_common, + .rem_link = rem_link_common_dir, + .owns_item = owns_item_common_dir, + .can_add_link = can_add_link_common, + .can_rem_link = can_rem_link_common_dir, + .detach = reiser4_detach_common_dir, + .bind = reiser4_bind_common_dir, + .safelink = safelink_common, + .estimate = { + .create = estimate_create_common_dir, + .update = estimate_update_common, + .unlink = estimate_unlink_common_dir + }, + .wire = { + .write = wire_write_common, + .read = wire_read_common, + .get = wire_get_common, + .size = wire_size_common, + .done = wire_done_common + }, + .init_inode_data = init_inode_ordering, + .cut_tree_worker = cut_tree_worker_common, + }, + [SYMLINK_FILE_PLUGIN_ID] = { + .h = { + .type_id = REISER4_FILE_PLUGIN_TYPE, + .id = SYMLINK_FILE_PLUGIN_ID, + .groups = (1 << REISER4_SYMLINK_FILE), + .pops = &file_plugin_ops, + .label = "symlink", + .desc = "symbolic link", + .linkage = {NULL,NULL} + }, + .inode_ops = &symlink_file_i_ops, + /* inode->i_fop of symlink is initialized + by NULL in setup_inode_ops */ + .file_ops = &null_f_ops, + .as_ops = &null_a_ops, + + .write_sd_by_inode = write_sd_by_inode_common, + .set_plug_in_inode = set_plug_in_inode_common, + .adjust_to_parent = adjust_to_parent_common, + .create_object = reiser4_create_symlink, + .delete_object = reiser4_delete_object_common, + .add_link = reiser4_add_link_common, + .rem_link = reiser4_rem_link_common, + .can_add_link = can_add_link_common, + .detach = dummyop, + .bind = dummyop, + .safelink = safelink_common, + .estimate = { + .create = estimate_create_common, + .update = estimate_update_common, + .unlink = estimate_unlink_common + }, + .init_inode_data = init_inode_ordering, + .cut_tree_worker = cut_tree_worker_common, + .destroy_inode = destroy_inode_symlink, + .wire = { + .write = wire_write_common, + .read = wire_read_common, + .get = wire_get_common, + .size = wire_size_common, + .done = wire_done_common + } + }, + [SPECIAL_FILE_PLUGIN_ID] = { + .h = { + .type_id = REISER4_FILE_PLUGIN_TYPE, + .id = SPECIAL_FILE_PLUGIN_ID, + .groups = (1 << REISER4_SPECIAL_FILE), + .pops = &file_plugin_ops, + .label = "special", + .desc = + "special: fifo, device or socket", + .linkage = {NULL, NULL} + }, + .inode_ops = &special_file_i_ops, + /* file_ops of special files (sockets, block, char, fifo) are + initialized by init_special_inode. */ + .file_ops = &null_f_ops, + .as_ops = &null_a_ops, + + .write_sd_by_inode = write_sd_by_inode_common, + .set_plug_in_inode = set_plug_in_inode_common, + .adjust_to_parent = adjust_to_parent_common, + .create_object = reiser4_create_object_common, + .delete_object = reiser4_delete_object_common, + .add_link = reiser4_add_link_common, + .rem_link = reiser4_rem_link_common, + .owns_item = owns_item_common, + .can_add_link = can_add_link_common, + .detach = dummyop, + .bind = dummyop, + .safelink = safelink_common, + .estimate = { + .create = estimate_create_common, + .update = estimate_update_common, + .unlink = estimate_unlink_common + }, + .init_inode_data = init_inode_ordering, + .cut_tree_worker = cut_tree_worker_common, + .wire = { + .write = wire_write_common, + .read = wire_read_common, + .get = wire_get_common, + .size = wire_size_common, + .done = wire_done_common + } + }, + [CRYPTCOMPRESS_FILE_PLUGIN_ID] = { + .h = { + .type_id = REISER4_FILE_PLUGIN_TYPE, + .id = CRYPTCOMPRESS_FILE_PLUGIN_ID, + .groups = (1 << REISER4_REGULAR_FILE), + .pops = &file_plugin_ops, + .label = "cryptcompress", + .desc = "cryptcompress file", + .linkage = {NULL, NULL} + }, + .inode_ops = ®ular_file_i_ops, + .file_ops = ®ular_file_f_ops, + .as_ops = ®ular_file_a_ops, + + .setattr = setattr_cryptcompress, + .open = open_cryptcompress, + .read = read_cryptcompress, + .write = write_cryptcompress, + .ioctl = ioctl_cryptcompress, + .mmap = mmap_cryptcompress, + .release = release_cryptcompress, + + .readpage = readpage_cryptcompress, + .readpages = readpages_cryptcompress, + .writepages = writepages_cryptcompress, + .write_begin = write_begin_cryptcompress, + .write_end = write_end_cryptcompress, + + .bmap = bmap_cryptcompress, + + .write_sd_by_inode = write_sd_by_inode_common, + .flow_by_inode = flow_by_inode_cryptcompress, + .key_by_inode = key_by_inode_cryptcompress, + .set_plug_in_inode = set_plug_in_inode_common, + .adjust_to_parent = adjust_to_parent_cryptcompress, + .create_object = create_object_cryptcompress, + .delete_object = delete_object_cryptcompress, + .add_link = reiser4_add_link_common, + .rem_link = reiser4_rem_link_common, + .owns_item = owns_item_common, + .can_add_link = can_add_link_common, + .detach = dummyop, + .bind = dummyop, + .safelink = safelink_common, + .estimate = { + .create = estimate_create_common, + .update = estimate_update_common, + .unlink = estimate_unlink_common + }, + .init_inode_data = init_inode_data_cryptcompress, + .cut_tree_worker = cut_tree_worker_cryptcompress, + .destroy_inode = destroy_inode_cryptcompress, + .wire = { + .write = wire_write_common, + .read = wire_read_common, + .get = wire_get_common, + .size = wire_size_common, + .done = wire_done_common + } + } +}; + +static int change_dir(struct inode *inode, + reiser4_plugin * plugin, + pset_member memb) +{ + /* cannot change dir plugin of already existing object */ + return RETERR(-EINVAL); +} + +static reiser4_plugin_ops dir_plugin_ops = { + .change = change_dir +}; + +/* + * definition of directory plugins + */ + +dir_plugin dir_plugins[LAST_DIR_ID] = { + /* standard hashed directory plugin */ + [HASHED_DIR_PLUGIN_ID] = { + .h = { + .type_id = REISER4_DIR_PLUGIN_TYPE, + .id = HASHED_DIR_PLUGIN_ID, + .pops = &dir_plugin_ops, + .label = "dir", + .desc = "hashed directory", + .linkage = {NULL, NULL} + }, + .inode_ops = &directory_i_ops, + .file_ops = &directory_f_ops, + .as_ops = &directory_a_ops, + + .get_parent = get_parent_common, + .is_name_acceptable = is_name_acceptable_common, + .build_entry_key = build_entry_key_hashed, + .build_readdir_key = build_readdir_key_common, + .add_entry = reiser4_add_entry_common, + .rem_entry = reiser4_rem_entry_common, + .init = reiser4_dir_init_common, + .done = reiser4_dir_done_common, + .attach = reiser4_attach_common, + .detach = reiser4_detach_common, + .estimate = { + .add_entry = estimate_add_entry_common, + .rem_entry = estimate_rem_entry_common, + .unlink = dir_estimate_unlink_common + } + }, + /* hashed directory for which seekdir/telldir are guaranteed to + * work. Brain-damage. */ + [SEEKABLE_HASHED_DIR_PLUGIN_ID] = { + .h = { + .type_id = REISER4_DIR_PLUGIN_TYPE, + .id = SEEKABLE_HASHED_DIR_PLUGIN_ID, + .pops = &dir_plugin_ops, + .label = "dir32", + .desc = "directory hashed with 31 bit hash", + .linkage = {NULL, NULL} + }, + .inode_ops = &directory_i_ops, + .file_ops = &directory_f_ops, + .as_ops = &directory_a_ops, + + .get_parent = get_parent_common, + .is_name_acceptable = is_name_acceptable_common, + .build_entry_key = build_entry_key_seekable, + .build_readdir_key = build_readdir_key_common, + .add_entry = reiser4_add_entry_common, + .rem_entry = reiser4_rem_entry_common, + .init = reiser4_dir_init_common, + .done = reiser4_dir_done_common, + .attach = reiser4_attach_common, + .detach = reiser4_detach_common, + .estimate = { + .add_entry = estimate_add_entry_common, + .rem_entry = estimate_rem_entry_common, + .unlink = dir_estimate_unlink_common + } + } +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/object.h linux-2.6.30/fs/reiser4/plugin/object.h --- linux-2.6.30.orig/fs/reiser4/plugin/object.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/object.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,117 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Declaration of object plugin functions. */ + +#if !defined(__FS_REISER4_PLUGIN_OBJECT_H__) +#define __FS_REISER4_PLUGIN_OBJECT_H__ + +#include "../type_safe_hash.h" + +/* common implementations of inode operations */ +int reiser4_create_common(struct inode *parent, struct dentry *dentry, + int mode, struct nameidata *); +struct dentry *reiser4_lookup_common(struct inode *parent, + struct dentry *dentry, + struct nameidata *nameidata); +int reiser4_link_common(struct dentry *existing, struct inode *parent, + struct dentry *newname); +int reiser4_unlink_common(struct inode *parent, struct dentry *victim); +int reiser4_mkdir_common(struct inode *parent, struct dentry *dentry, int mode); +int reiser4_symlink_common(struct inode *parent, struct dentry *dentry, + const char *linkname); +int reiser4_mknod_common(struct inode *parent, struct dentry *dentry, + int mode, dev_t rdev); +int reiser4_rename_common(struct inode *old_dir, struct dentry *old_name, + struct inode *new_dir, struct dentry *new_name); +void *reiser4_follow_link_common(struct dentry *, struct nameidata *data); +int reiser4_permission_common(struct inode *, int mask); +int reiser4_setattr_common(struct dentry *, struct iattr *); +int reiser4_getattr_common(struct vfsmount *mnt, struct dentry *, + struct kstat *); + +/* common implementations of file operations */ +loff_t reiser4_llseek_dir_common(struct file *, loff_t off, int origin); +int reiser4_readdir_common(struct file *, void *dirent, filldir_t); +int reiser4_release_dir_common(struct inode *, struct file *); +int reiser4_sync_common(struct file *, struct dentry *, int datasync); + + +/* file plugin operations: common implementations */ +int write_sd_by_inode_common(struct inode *); +int key_by_inode_and_offset_common(struct inode *, loff_t, reiser4_key *); +int set_plug_in_inode_common(struct inode *object, struct inode *parent, + reiser4_object_create_data *); +int adjust_to_parent_common(struct inode *object, struct inode *parent, + struct inode *root); +int adjust_to_parent_common_dir(struct inode *object, struct inode *parent, + struct inode *root); +int adjust_to_parent_cryptcompress(struct inode *object, struct inode *parent, + struct inode *root); +int reiser4_create_object_common(struct inode *object, struct inode *parent, + reiser4_object_create_data *); +int reiser4_delete_object_common(struct inode *); +int reiser4_delete_dir_common(struct inode *); +int reiser4_add_link_common(struct inode *object, struct inode *parent); +int reiser4_rem_link_common(struct inode *object, struct inode *parent); +int rem_link_common_dir(struct inode *object, struct inode *parent); +int owns_item_common(const struct inode *, const coord_t *); +int owns_item_common_dir(const struct inode *, const coord_t *); +int can_add_link_common(const struct inode *); +int can_rem_link_common_dir(const struct inode *); +int reiser4_detach_common_dir(struct inode *child, struct inode *parent); +int reiser4_bind_common_dir(struct inode *child, struct inode *parent); +int safelink_common(struct inode *, reiser4_safe_link_t, __u64 value); +reiser4_block_nr estimate_create_common(const struct inode *); +reiser4_block_nr estimate_create_common_dir(const struct inode *); +reiser4_block_nr estimate_update_common(const struct inode *); +reiser4_block_nr estimate_unlink_common(const struct inode *, + const struct inode *); +reiser4_block_nr estimate_unlink_common_dir(const struct inode *, + const struct inode *); +char *wire_write_common(struct inode *, char *start); +char *wire_read_common(char *addr, reiser4_object_on_wire *); +struct dentry *wire_get_common(struct super_block *, reiser4_object_on_wire *); +int wire_size_common(struct inode *); +void wire_done_common(reiser4_object_on_wire *); + +/* dir plugin operations: common implementations */ +struct dentry *get_parent_common(struct inode *child); +int is_name_acceptable_common(const struct inode *, const char *name, int len); +void build_entry_key_common(const struct inode *, + const struct qstr *qname, reiser4_key *); +int build_readdir_key_common(struct file *dir, reiser4_key *); +int reiser4_add_entry_common(struct inode *object, struct dentry *where, + reiser4_object_create_data * , reiser4_dir_entry_desc *); +int reiser4_rem_entry_common(struct inode *object, struct dentry *where, + reiser4_dir_entry_desc *); +int reiser4_dir_init_common(struct inode *object, struct inode *parent, + reiser4_object_create_data *); +int reiser4_dir_done_common(struct inode *); +int reiser4_attach_common(struct inode *child, struct inode *parent); +int reiser4_detach_common(struct inode *object, struct inode *parent); +reiser4_block_nr estimate_add_entry_common(const struct inode *); +reiser4_block_nr estimate_rem_entry_common(const struct inode *); +reiser4_block_nr dir_estimate_unlink_common(const struct inode *, + const struct inode *); + +/* these are essential parts of common implementations, they are to make + customized implementations easier */ +int do_prepare_write(struct file *, struct page *, unsigned from, unsigned to); + +/* merely useful functions */ +int lookup_sd(struct inode *, znode_lock_mode, coord_t *, lock_handle * , + const reiser4_key * , int silent); + +/* __FS_REISER4_PLUGIN_OBJECT_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/plugin.c linux-2.6.30/fs/reiser4/plugin/plugin.c --- linux-2.6.30.orig/fs/reiser4/plugin/plugin.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/plugin.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,560 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Basic plugin infrastructure, lookup etc. */ + +/* PLUGINS: + + Plugins are internal Reiser4 "modules" or "objects" used to increase + extensibility and allow external users to easily adapt reiser4 to + their needs. + + Plugins are classified into several disjoint "types". Plugins + belonging to the particular plugin type are termed "instances" of + this type. Existing types are listed by enum reiser4_plugin_type + (see plugin/plugin_header.h) + +NIKITA-FIXME-HANS: update this list, and review this entire comment for currency + + Object (file) plugin determines how given file-system object serves + standard VFS requests for read, write, seek, mmap etc. Instances of + file plugins are: regular file, directory, symlink. Another example + of file plugin is audit plugin, that optionally records accesses to + underlying object and forwards requests to it. + + Hash plugins compute hashes used by reiser4 to store and locate + files within directories. Instances of hash plugin type are: r5, + tea, rupasov. + + Tail plugins (or, more precisely, tail policy plugins) determine + when last part of the file should be stored in a formatted item. + + Scope and lookup: + + label such that pair ( type_label, plugin_label ) is unique. This + pair is a globally persistent and user-visible plugin + identifier. Internally kernel maintains plugins and plugin types in + arrays using an index into those arrays as plugin and plugin type + identifiers. File-system in turn, also maintains persistent + "dictionary" which is mapping from plugin label to numerical + identifier which is stored in file-system objects. That is, we + store the offset into the plugin array for that plugin type as the + plugin id in the stat data of the filesystem object. + + Internal kernel plugin type identifier (index in plugins[] array) is + of type reiser4_plugin_type. Set of available plugin types is + currently static, but dynamic loading doesn't seem to pose + insurmountable problems. + + Within each type plugins are addressed by the identifiers of type + reiser4_plugin_id (indices in reiser4_plugin_type_data.builtin[]). + Such identifiers are only required to be unique within one type, + not globally. + + Thus, plugin in memory is uniquely identified by the pair (type_id, + id). + + Usage: + + There exists only one instance of each plugin instance, but this + single instance can be associated with many entities (file-system + objects, items, nodes, transactions, file-descriptors etc.). Entity + to which plugin of given type is termed (due to the lack of + imagination) "subject" of this plugin type and, by abuse of + terminology, subject of particular instance of this type to which + it's attached currently. For example, inode is subject of object + plugin type. Inode representing directory is subject of directory + plugin, hash plugin type and some particular instance of hash plugin + type. Inode, representing regular file is subject of "regular file" + plugin, tail-policy plugin type etc. + + With each subject the plugin possibly stores some state. For example, + the state of a directory plugin (instance of object plugin type) is pointer + to hash plugin (if directories always use hashing that is). + + Interface: + + In addition to a scalar identifier, each plugin type and plugin + proper has a "label": short string and a "description"---longer + descriptive string. Labels and descriptions of plugin types are + hard-coded into plugins[] array, declared and defined in + plugin.c. Label and description of plugin are stored in .label and + .desc fields of reiser4_plugin_header respectively. It's possible to + locate plugin by the pair of labels. + + Features (not implemented): + + . user-level plugin manipulations: + + reiser4("filename/..file_plugin<='audit'"); + + write(open("filename/..file_plugin"), "audit", 8); + + . user level utilities lsplug and chplug to manipulate plugins. + Utilities are not of primary priority. Possibly they will be not + working on v4.0 + + NIKITA-FIXME-HANS: this should be a mkreiserfs option not a mount + option, do you agree? I don't think that specifying it at mount time, + and then changing it with each mount, is a good model for usage. + + . mount option "plug" to set-up plugins of root-directory. + "plug=foo:bar" will set "bar" as default plugin of type "foo". + + Limitations: + + . each plugin type has to provide at least one builtin + plugin. This is technical limitation and it can be lifted in the + future. + + TODO: + + New plugin types/plugings: + Things we should be able to separately choose to inherit: + + security plugins + + stat data + + file bodies + + file plugins + + dir plugins + + . perm:acl + + . audi---audit plugin intercepting and possibly logging all + accesses to object. Requires to put stub functions in file_operations + in stead of generic_file_*. + +NIKITA-FIXME-HANS: why make overflows a plugin? + . over---handle hash overflows + + . sqnt---handle different access patterns and instruments read-ahead + +NIKITA-FIXME-HANS: describe the line below in more detail. + + . hier---handle inheritance of plugins along file-system hierarchy + + Different kinds of inheritance: on creation vs. on access. + Compatible/incompatible plugins. + Inheritance for multi-linked files. + Layered plugins. + Notion of plugin context is abandoned. + +Each file is associated + with one plugin and dependant plugins (hash, etc.) are stored as + main plugin state. Now, if we have plugins used for regular files + but not for directories, how such plugins would be inherited? + . always store them with directories also + +NIKTIA-FIXME-HANS: Do the line above. It is not exclusive of doing +the line below which is also useful. + + . use inheritance hierarchy, independent of file-system namespace +*/ + +#include "../debug.h" +#include "../dformat.h" +#include "plugin_header.h" +#include "item/static_stat.h" +#include "node/node.h" +#include "security/perm.h" +#include "space/space_allocator.h" +#include "disk_format/disk_format.h" +#include "plugin.h" +#include "../reiser4.h" +#include "../jnode.h" +#include "../inode.h" + +#include /* for struct super_block */ + +/* + * init_plugins - initialize plugin sub-system. + * Just call this once on reiser4 startup. + * + * Initializes plugin sub-system. It is part of reiser4 module + * initialization. For each plugin of each type init method is called and each + * plugin is put into list of plugins. + */ +int init_plugins(void) +{ + reiser4_plugin_type type_id; + + for (type_id = 0; type_id < REISER4_PLUGIN_TYPES; ++type_id) { + struct reiser4_plugin_type_data *ptype; + int i; + + ptype = &plugins[type_id]; + assert("nikita-3508", ptype->label != NULL); + assert("nikita-3509", ptype->type_id == type_id); + + INIT_LIST_HEAD(&ptype->plugins_list); +/* NIKITA-FIXME-HANS: change builtin_num to some other name lacking the term + * builtin. */ + for (i = 0; i < ptype->builtin_num; ++i) { + reiser4_plugin *plugin; + + plugin = plugin_at(ptype, i); + + if (plugin->h.label == NULL) + /* uninitialized slot encountered */ + continue; + assert("nikita-3445", plugin->h.type_id == type_id); + plugin->h.id = i; + if (plugin->h.pops != NULL && + plugin->h.pops->init != NULL) { + int result; + + result = plugin->h.pops->init(plugin); + if (result != 0) + return result; + } + INIT_LIST_HEAD(&plugin->h.linkage); + list_add_tail(&plugin->h.linkage, &ptype->plugins_list); + } + } + return 0; +} + +/* true if plugin type id is valid */ +int is_plugin_type_valid(reiser4_plugin_type type) +{ + /* "type" is unsigned, so no comparison with 0 is + necessary */ + return (type < REISER4_PLUGIN_TYPES); +} + +/* true if plugin id is valid */ +int is_plugin_id_valid(reiser4_plugin_type type, reiser4_plugin_id id) +{ + assert("nikita-1653", is_plugin_type_valid(type)); + return id < plugins[type].builtin_num; +} + +/* return plugin by its @type and @id. + + Both arguments are checked for validness: this is supposed to be called + from user-level. + +NIKITA-FIXME-HANS: Do you instead mean that this checks ids created in +user space, and passed to the filesystem by use of method files? Your +comment really confused me on the first reading.... + +*/ +reiser4_plugin *plugin_by_unsafe_id(reiser4_plugin_type type /* plugin type + * unchecked */, + reiser4_plugin_id id /* plugin id, + * unchecked */) +{ + if (is_plugin_type_valid(type)) { + if (is_plugin_id_valid(type, id)) + return plugin_at(&plugins[type], id); + else + /* id out of bounds */ + warning("nikita-2913", + "Invalid plugin id: [%i:%i]", type, id); + } else + /* type_id out of bounds */ + warning("nikita-2914", "Invalid type_id: %i", type); + return NULL; +} + +/** + * save_plugin_id - store plugin id in disk format + * @plugin: plugin to convert + * @area: where to store result + * + * Puts id of @plugin in little endian format to address @area. + */ +int save_plugin_id(reiser4_plugin *plugin /* plugin to convert */ , + d16 * area/* where to store result */) +{ + assert("nikita-1261", plugin != NULL); + assert("nikita-1262", area != NULL); + + put_unaligned(cpu_to_le16(plugin->h.id), area); + return 0; +} + +/* list of all plugins of given type */ +struct list_head *get_plugin_list(reiser4_plugin_type type) +{ + assert("nikita-1056", is_plugin_type_valid(type)); + return &plugins[type].plugins_list; +} + +static void update_pset_mask(reiser4_inode * info, pset_member memb) +{ + struct dentry *rootdir; + reiser4_inode *root; + + assert("edward-1443", memb != PSET_FILE); + + rootdir = inode_by_reiser4_inode(info)->i_sb->s_root; + if (rootdir != NULL) { + root = reiser4_inode_data(rootdir->d_inode); + /* + * if inode is different from the default one, or we are + * changing plugin of root directory, update plugin_mask + */ + if (aset_get(info->pset, memb) != + aset_get(root->pset, memb) || + info == root) + info->plugin_mask |= (1 << memb); + else + info->plugin_mask &= ~(1 << memb); + } +} + +/* Get specified plugin set member from parent, + or from fs-defaults (if no parent is given) and + install the result to pset of @self */ +int grab_plugin_pset(struct inode *self, + struct inode *ancestor, + pset_member memb) +{ + reiser4_plugin *plug; + reiser4_inode *info; + int result = 0; + + /* Do not grab if initialised already. */ + info = reiser4_inode_data(self); + if (aset_get(info->pset, memb) != NULL) + return 0; + if (ancestor) { + reiser4_inode *parent; + + parent = reiser4_inode_data(ancestor); + plug = aset_get(parent->hset, memb) ? : + aset_get(parent->pset, memb); + } else + plug = get_default_plugin(memb); + + result = set_plugin(&info->pset, memb, plug); + if (result == 0) { + if (!ancestor || self->i_sb->s_root->d_inode != self) + update_pset_mask(info, memb); + } + return result; +} + +/* Take missing pset members from root inode */ +int finish_pset(struct inode *inode) +{ + reiser4_plugin *plug; + reiser4_inode *root; + reiser4_inode *info; + pset_member memb; + int result = 0; + + root = reiser4_inode_data(inode->i_sb->s_root->d_inode); + info = reiser4_inode_data(inode); + + assert("edward-1455", root != NULL); + assert("edward-1456", info != NULL); + + /* file and directory plugins are already initialized. */ + for (memb = PSET_DIR + 1; memb < PSET_LAST; ++memb) { + + /* Do not grab if initialised already. */ + if (aset_get(info->pset, memb) != NULL) + continue; + + plug = aset_get(root->pset, memb); + result = set_plugin(&info->pset, memb, plug); + if (result != 0) + break; + } + if (result != 0) { + warning("nikita-3447", + "Cannot set up plugins for %lli", + (unsigned long long) + get_inode_oid(inode)); + } + return result; +} + +int force_plugin_pset(struct inode *self, pset_member memb, + reiser4_plugin * plug) +{ + reiser4_inode *info; + int result = 0; + + if (!self->i_sb->s_root || self->i_sb->s_root->d_inode == self) { + /* Changing pset in the root object. */ + return RETERR(-EINVAL); + } + + info = reiser4_inode_data(self); + if (plug->h.pops != NULL && plug->h.pops->change != NULL) + result = plug->h.pops->change(self, plug, memb); + else + result = aset_set_unsafe(&info->pset, memb, plug); + if (result == 0) { + __u16 oldmask = info->plugin_mask; + + update_pset_mask(info, memb); + if (oldmask != info->plugin_mask) + reiser4_inode_clr_flag(self, REISER4_SDLEN_KNOWN); + } + return result; +} + +struct reiser4_plugin_type_data plugins[REISER4_PLUGIN_TYPES] = { + /* C90 initializers */ + [REISER4_FILE_PLUGIN_TYPE] = { + .type_id = REISER4_FILE_PLUGIN_TYPE, + .label = "file", + .desc = "Object plugins", + .builtin_num = sizeof_array(file_plugins), + .builtin = file_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(file_plugin) + }, + [REISER4_DIR_PLUGIN_TYPE] = { + .type_id = REISER4_DIR_PLUGIN_TYPE, + .label = "dir", + .desc = "Directory plugins", + .builtin_num = sizeof_array(dir_plugins), + .builtin = dir_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(dir_plugin) + }, + [REISER4_HASH_PLUGIN_TYPE] = { + .type_id = REISER4_HASH_PLUGIN_TYPE, + .label = "hash", + .desc = "Directory hashes", + .builtin_num = sizeof_array(hash_plugins), + .builtin = hash_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(hash_plugin) + }, + [REISER4_FIBRATION_PLUGIN_TYPE] = { + .type_id = + REISER4_FIBRATION_PLUGIN_TYPE, + .label = "fibration", + .desc = "Directory fibrations", + .builtin_num = sizeof_array(fibration_plugins), + .builtin = fibration_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(fibration_plugin) + }, + [REISER4_CIPHER_PLUGIN_TYPE] = { + .type_id = REISER4_CIPHER_PLUGIN_TYPE, + .label = "cipher", + .desc = "Cipher plugins", + .builtin_num = sizeof_array(cipher_plugins), + .builtin = cipher_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(cipher_plugin) + }, + [REISER4_DIGEST_PLUGIN_TYPE] = { + .type_id = REISER4_DIGEST_PLUGIN_TYPE, + .label = "digest", + .desc = "Digest plugins", + .builtin_num = sizeof_array(digest_plugins), + .builtin = digest_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(digest_plugin) + }, + [REISER4_COMPRESSION_PLUGIN_TYPE] = { + .type_id = REISER4_COMPRESSION_PLUGIN_TYPE, + .label = "compression", + .desc = "Compression plugins", + .builtin_num = sizeof_array(compression_plugins), + .builtin = compression_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(compression_plugin) + }, + [REISER4_FORMATTING_PLUGIN_TYPE] = { + .type_id = REISER4_FORMATTING_PLUGIN_TYPE, + .label = "formatting", + .desc = "Tail inlining policies", + .builtin_num = sizeof_array(formatting_plugins), + .builtin = formatting_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(formatting_plugin) + }, + [REISER4_PERM_PLUGIN_TYPE] = { + .type_id = REISER4_PERM_PLUGIN_TYPE, + .label = "perm", + .desc = "Permission checks", + .builtin_num = sizeof_array(perm_plugins), + .builtin = perm_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(perm_plugin) + }, + [REISER4_ITEM_PLUGIN_TYPE] = { + .type_id = REISER4_ITEM_PLUGIN_TYPE, + .label = "item", + .desc = "Item handlers", + .builtin_num = sizeof_array(item_plugins), + .builtin = item_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(item_plugin) + }, + [REISER4_NODE_PLUGIN_TYPE] = { + .type_id = REISER4_NODE_PLUGIN_TYPE, + .label = "node", + .desc = "node layout handlers", + .builtin_num = sizeof_array(node_plugins), + .builtin = node_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(node_plugin) + }, + [REISER4_SD_EXT_PLUGIN_TYPE] = { + .type_id = REISER4_SD_EXT_PLUGIN_TYPE, + .label = "sd_ext", + .desc = "Parts of stat-data", + .builtin_num = sizeof_array(sd_ext_plugins), + .builtin = sd_ext_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(sd_ext_plugin) + }, + [REISER4_FORMAT_PLUGIN_TYPE] = { + .type_id = REISER4_FORMAT_PLUGIN_TYPE, + .label = "disk_layout", + .desc = "defines filesystem on disk layout", + .builtin_num = sizeof_array(format_plugins), + .builtin = format_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(disk_format_plugin) + }, + [REISER4_JNODE_PLUGIN_TYPE] = { + .type_id = REISER4_JNODE_PLUGIN_TYPE, + .label = "jnode", + .desc = "defines kind of jnode", + .builtin_num = sizeof_array(jnode_plugins), + .builtin = jnode_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(jnode_plugin) + }, + [REISER4_COMPRESSION_MODE_PLUGIN_TYPE] = { + .type_id = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .label = "compression_mode", + .desc = "Defines compression mode", + .builtin_num = sizeof_array(compression_mode_plugins), + .builtin = compression_mode_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(compression_mode_plugin) + }, + [REISER4_CLUSTER_PLUGIN_TYPE] = { + .type_id = REISER4_CLUSTER_PLUGIN_TYPE, + .label = "cluster", + .desc = "Defines cluster size", + .builtin_num = sizeof_array(cluster_plugins), + .builtin = cluster_plugins, + .plugins_list = {NULL, NULL}, + .size = sizeof(cluster_plugin) + } +}; + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/plugin.h linux-2.6.30/fs/reiser4/plugin/plugin.h --- linux-2.6.30.orig/fs/reiser4/plugin/plugin.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/plugin.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,942 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Basic plugin data-types. + see fs/reiser4/plugin/plugin.c for details */ + +#if !defined(__FS_REISER4_PLUGIN_TYPES_H__) +#define __FS_REISER4_PLUGIN_TYPES_H__ + +#include "../forward.h" +#include "../debug.h" +#include "../dformat.h" +#include "../key.h" +#include "compress/compress.h" +#include "crypto/cipher.h" +#include "plugin_header.h" +#include "item/static_stat.h" +#include "item/internal.h" +#include "item/sde.h" +#include "item/cde.h" +#include "item/item.h" +#include "node/node.h" +#include "node/node40.h" +#include "security/perm.h" +#include "fibration.h" + +#include "space/bitmap.h" +#include "space/space_allocator.h" + +#include "disk_format/disk_format40.h" +#include "disk_format/disk_format.h" + +#include /* for struct super_block, address_space */ +#include /* for struct page */ +#include /* for struct buffer_head */ +#include /* for struct dentry */ +#include +#include + +typedef struct reiser4_object_on_wire reiser4_object_on_wire; + +/* + * File plugin. Defines the set of methods that file plugins implement, some + * of which are optional. + * + * A file plugin offers to the caller an interface for IO ( writing to and/or + * reading from) to what the caller sees as one sequence of bytes. An IO to it + * may affect more than one physical sequence of bytes, or no physical sequence + * of bytes, it may affect sequences of bytes offered by other file plugins to + * the semantic layer, and the file plugin may invoke other plugins and + * delegate work to them, but its interface is structured for offering the + * caller the ability to read and/or write what the caller sees as being a + * single sequence of bytes. + * + * The file plugin must present a sequence of bytes to the caller, but it does + * not necessarily have to store a sequence of bytes, it does not necessarily + * have to support efficient tree traversal to any offset in the sequence of + * bytes (tail and extent items, whose keys contain offsets, do however provide + * efficient non-sequential lookup of any offset in the sequence of bytes). + * + * Directory plugins provide methods for selecting file plugins by resolving a + * name for them. + * + * The functionality other filesystems call an attribute, and rigidly tie + * together, we decompose into orthogonal selectable features of files. Using + * the terminology we will define next, an attribute is a perhaps constrained, + * perhaps static length, file whose parent has a uni-count-intra-link to it, + * which might be grandparent-major-packed, and whose parent has a deletion + * method that deletes it. + * + * File plugins can implement constraints. + * + * Files can be of variable length (e.g. regular unix files), or of static + * length (e.g. static sized attributes). + * + * An object may have many sequences of bytes, and many file plugins, but, it + * has exactly one objectid. It is usually desirable that an object has a + * deletion method which deletes every item with that objectid. Items cannot + * in general be found by just their objectids. This means that an object must + * have either a method built into its deletion plugin method for knowing what + * items need to be deleted, or links stored with the object that provide the + * plugin with a method for finding those items. Deleting a file within an + * object may or may not have the effect of deleting the entire object, + * depending on the file plugin's deletion method. + * + * LINK TAXONOMY: + * + * Many objects have a reference count, and when the reference count reaches 0 + * the object's deletion method is invoked. Some links embody a reference + * count increase ("countlinks"), and others do not ("nocountlinks"). + * + * Some links are bi-directional links ("bilinks"), and some are + * uni-directional("unilinks"). + * + * Some links are between parts of the same object ("intralinks"), and some are + * between different objects ("interlinks"). + * + * PACKING TAXONOMY: + * + * Some items of an object are stored with a major packing locality based on + * their object's objectid (e.g. unix directory items in plan A), and these are + * called "self-major-packed". + * + * Some items of an object are stored with a major packing locality based on + * their semantic parent object's objectid (e.g. unix file bodies in plan A), + * and these are called "parent-major-packed". + * + * Some items of an object are stored with a major packing locality based on + * their semantic grandparent, and these are called "grandparent-major-packed". + * Now carefully notice that we run into trouble with key length if we have to + * store a 8 byte major+minor grandparent based packing locality, an 8 byte + * parent objectid, an 8 byte attribute objectid, and an 8 byte offset, all in + * a 24 byte key. One of these fields must be sacrificed if an item is to be + * grandparent-major-packed, and which to sacrifice is left to the item author + * choosing to make the item grandparent-major-packed. You cannot make tail + * items and extent items grandparent-major-packed, though you could make them + * self-major-packed (usually they are parent-major-packed). + * + * In the case of ACLs (which are composed of fixed length ACEs which consist + * of {subject-type, subject, and permission bitmask} triples), it makes sense + * to not have an offset field in the ACE item key, and to allow duplicate keys + * for ACEs. Thus, the set of ACES for a given file is found by looking for a + * key consisting of the objectid of the grandparent (thus grouping all ACLs in + * a directory together), the minor packing locality of ACE, the objectid of + * the file, and 0. + * + * IO involves moving data from one location to another, which means that two + * locations must be specified, source and destination. + * + * This source and destination can be in the filesystem, or they can be a + * pointer in the user process address space plus a byte count. + * + * If both source and destination are in the filesystem, then at least one of + * them must be representable as a pure stream of bytes (which we call a flow, + * and define as a struct containing a key, a data pointer, and a length). + * This may mean converting one of them into a flow. We provide a generic + * cast_into_flow() method, which will work for any plugin supporting + * read_flow(), though it is inefficiently implemented in that it temporarily + * stores the flow in a buffer (Question: what to do with huge flows that + * cannot fit into memory? Answer: we must not convert them all at once. ) + * + * Performing a write requires resolving the write request into a flow defining + * the source, and a method that performs the write, and a key that defines + * where in the tree the write is to go. + * + * Performing a read requires resolving the read request into a flow defining + * the target, and a method that performs the read, and a key that defines + * where in the tree the read is to come from. + * + * There will exist file plugins which have no pluginid stored on the disk for + * them, and which are only invoked by other plugins. + */ + +/* This should be incremented with each new contributed + pair (plugin type, plugin id). + NOTE: Make sure there is a release of reiser4progs + with the corresponding version number */ +#define PLUGIN_LIBRARY_VERSION 0 + + /* enumeration of fields within plugin_set */ +typedef enum { + PSET_FILE, + PSET_DIR, /* PSET_FILE and PSET_DIR should be first + * elements: inode.c:read_inode() depends on + * this. */ + PSET_PERM, + PSET_FORMATTING, + PSET_HASH, + PSET_FIBRATION, + PSET_SD, + PSET_DIR_ITEM, + PSET_CIPHER, + PSET_DIGEST, + PSET_COMPRESSION, + PSET_COMPRESSION_MODE, + PSET_CLUSTER, + PSET_CREATE, + PSET_LAST +} pset_member; + +/* builtin file-plugins */ +typedef enum { + /* regular file */ + UNIX_FILE_PLUGIN_ID, + /* directory */ + DIRECTORY_FILE_PLUGIN_ID, + /* symlink */ + SYMLINK_FILE_PLUGIN_ID, + /* for objects completely handled by the VFS: fifos, devices, + sockets */ + SPECIAL_FILE_PLUGIN_ID, + /* regular cryptcompress file */ + CRYPTCOMPRESS_FILE_PLUGIN_ID, + /* number of file plugins. Used as size of arrays to hold + file plugins. */ + LAST_FILE_PLUGIN_ID +} reiser4_file_id; + +typedef struct file_plugin { + + /* generic fields */ + plugin_header h; + + /* VFS methods. + * Must be invariant with respect to plugin conversion. + * It can be achieved by using "common" methods, which + * are the same for all plugins that take participation in + * conversion, or by using "generic" or "careful" methods, + * which provide automatic redirection to proper private + * plugin methods ("careful" are the same as "generic", + * but with protection of pset and other disk structures + * from being rebuilt during conversion. + */ + struct inode_operations * inode_ops; + struct file_operations * file_ops; + struct address_space_operations * as_ops; + /** + * Private methods. These are optional. If used they will allow you + * to minimize the amount of code needed to implement a deviation + * from some other method that also uses them. + */ + /* + * private inode_ops + */ + int (*setattr)(struct dentry *, struct iattr *); + /* + * private file_ops + */ + /* do whatever is necessary to do when object is opened */ + int (*open) (struct inode *inode, struct file *file); + ssize_t (*read) (struct file *, char __user *buf, size_t read_amount, + loff_t *off); + /* write as much as possible bytes from nominated @write_amount + * before plugin scheduling is occurred. Save scheduling state + * in @cont */ + ssize_t (*write) (struct file *, const char __user *buf, + size_t write_amount, loff_t * off, + struct psched_context * cont); + int (*ioctl) (struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg); + int (*mmap) (struct file *, struct vm_area_struct *); + int (*release) (struct inode *, struct file *); + /* + * private a_ops + */ + int (*readpage) (struct file *file, struct page *page); + int (*readpages)(struct file *file, struct address_space *mapping, + struct list_head *pages, unsigned nr_pages); + int (*writepages)(struct address_space *mapping, + struct writeback_control *wbc); + int (*write_begin)(struct file *file, struct page *page, + unsigned from, unsigned to); + int (*write_end)(struct file *file, struct page *page, + unsigned from, unsigned to); + sector_t (*bmap) (struct address_space * mapping, sector_t lblock); + /* other private methods */ + /* save inode cached stat-data onto disk. It was called + reiserfs_update_sd() in 3.x */ + int (*write_sd_by_inode) (struct inode *); + /* + * Construct flow into @flow according to user-supplied data. + * + * This is used by read/write methods to construct a flow to + * write/read. ->flow_by_inode() is plugin method, rather than single + * global implementation, because key in a flow used by plugin may + * depend on data in a @buf. + * + * NIKITA-FIXME-HANS: please create statistics on what functions are + * dereferenced how often for the mongo benchmark. You can supervise + * Elena doing this for you if that helps. Email me the list of the + * top 10, with their counts, and an estimate of the total number of + * CPU cycles spent dereferencing as a percentage of CPU cycles spent + * processing (non-idle processing). If the total percent is, say, + * less than 1%, it will make our coding discussions much easier, and + * keep me from questioning whether functions like the below are too + * frequently called to be dereferenced. If the total percent is more + * than 1%, perhaps private methods should be listed in a "required" + * comment at the top of each plugin (with stern language about how if + * the comment is missing it will not be accepted by the maintainer), + * and implemented using macros not dereferenced functions. How about + * replacing this whole private methods part of the struct with a + * thorough documentation of what the standard helper functions are for + * use in constructing plugins? I think users have been asking for + * that, though not in so many words. + */ + int (*flow_by_inode) (struct inode *, const char __user *buf, + int user, loff_t size, + loff_t off, rw_op op, flow_t *); + /* + * Return the key used to retrieve an offset of a file. It is used by + * default implementation of ->flow_by_inode() method + * (common_build_flow()) and, among other things, to get to the extent + * from jnode of unformatted node. + */ + int (*key_by_inode) (struct inode *, loff_t off, reiser4_key *); + + /* NIKITA-FIXME-HANS: this comment is not as clear to others as you + * think.... */ + /* + * set the plugin for a file. Called during file creation in creat() + * but not reiser4() unless an inode already exists for the file. + */ + int (*set_plug_in_inode) (struct inode *inode, struct inode *parent, + reiser4_object_create_data *); + + /* NIKITA-FIXME-HANS: comment and name seem to say different things, + * are you setting up the object itself also or just adjusting the + * parent?.... */ + /* set up plugins for new @object created in @parent. @root is root + directory. */ + int (*adjust_to_parent) (struct inode *object, struct inode *parent, + struct inode *root); + /* + * this does whatever is necessary to do when object is created. For + * instance, for unix files stat data is inserted. It is supposed to be + * called by create of struct inode_operations. + */ + int (*create_object) (struct inode *object, struct inode *parent, + reiser4_object_create_data *); + /* + * this method should check REISER4_NO_SD and set REISER4_NO_SD on + * success. Deletion of an object usually includes removal of items + * building file body (for directories this is removal of "." and "..") + * and removal of stat-data item. + */ + int (*delete_object) (struct inode *); + + /* add link from @parent to @object */ + int (*add_link) (struct inode *object, struct inode *parent); + + /* remove link from @parent to @object */ + int (*rem_link) (struct inode *object, struct inode *parent); + + /* + * return true if item addressed by @coord belongs to @inode. This is + * used by read/write to properly slice flow into items in presence of + * multiple key assignment policies, because items of a file are not + * necessarily contiguous in a key space, for example, in a plan-b. + */ + int (*owns_item) (const struct inode *, const coord_t *); + + /* checks whether yet another hard links to this object can be + added */ + int (*can_add_link) (const struct inode *); + + /* checks whether hard links to this object can be removed */ + int (*can_rem_link) (const struct inode *); + + /* not empty for DIRECTORY_FILE_PLUGIN_ID only currently. It calls + detach of directory plugin to remove ".." */ + int (*detach) (struct inode *child, struct inode *parent); + + /* called when @child was just looked up in the @parent. It is not + empty for DIRECTORY_FILE_PLUGIN_ID only where it calls attach of + directory plugin */ + int (*bind) (struct inode *child, struct inode *parent); + + /* process safe-link during mount */ + int (*safelink) (struct inode *object, reiser4_safe_link_t link, + __u64 value); + + /* The couple of estimate methods for all file operations */ + struct { + reiser4_block_nr(*create) (const struct inode *); + reiser4_block_nr(*update) (const struct inode *); + reiser4_block_nr(*unlink) (const struct inode *, + const struct inode *); + } estimate; + + /* + * reiser4 specific part of inode has a union of structures which are + * specific to a plugin. This method is called when inode is read + * (read_inode) and when file is created (common_create_child) so that + * file plugin could initialize its inode data + */ + void (*init_inode_data) (struct inode *, reiser4_object_create_data * , + int); + + /* + * This method performs progressive deletion of items and whole nodes + * from right to left. + * + * @tap: the point deletion process begins from, + * @from_key: the beginning of the deleted key range, + * @to_key: the end of the deleted key range, + * @smallest_removed: the smallest removed key, + * + * @return: 0 if success, error code otherwise, -E_REPEAT means that + * long cut_tree operation was interrupted for allowing atom commit . + */ + int (*cut_tree_worker) (tap_t *, const reiser4_key * from_key, + const reiser4_key * to_key, + reiser4_key * smallest_removed, struct inode *, + int, int *); + + /* called from ->destroy_inode() */ + void (*destroy_inode) (struct inode *); + + /* + * methods to serialize object identify. This is used, for example, by + * reiser4_{en,de}code_fh(). + */ + struct { + /* store object's identity at @area */ + char *(*write) (struct inode *inode, char *area); + /* parse object from wire to the @obj */ + char *(*read) (char *area, reiser4_object_on_wire * obj); + /* given object identity in @obj, find or create its dentry */ + struct dentry *(*get) (struct super_block *s, + reiser4_object_on_wire * obj); + /* how many bytes ->wire.write() consumes */ + int (*size) (struct inode *inode); + /* finish with object identify */ + void (*done) (reiser4_object_on_wire * obj); + } wire; +} file_plugin; + +extern file_plugin file_plugins[LAST_FILE_PLUGIN_ID]; + +struct reiser4_object_on_wire { + file_plugin *plugin; + union { + struct { + obj_key_id key_id; + } std; + void *generic; + } u; +}; + +/* builtin dir-plugins */ +typedef enum { + HASHED_DIR_PLUGIN_ID, + SEEKABLE_HASHED_DIR_PLUGIN_ID, + LAST_DIR_ID +} reiser4_dir_id; + +typedef struct dir_plugin { + /* generic fields */ + plugin_header h; + + struct inode_operations * inode_ops; + struct file_operations * file_ops; + struct address_space_operations * as_ops; + + /* + * private methods: These are optional. If used they will allow you to + * minimize the amount of code needed to implement a deviation from + * some other method that uses them. You could logically argue that + * they should be a separate type of plugin. + */ + + struct dentry *(*get_parent) (struct inode *childdir); + + /* + * check whether "name" is acceptable name to be inserted into this + * object. Optionally implemented by directory-like objects. Can check + * for maximal length, reserved symbols etc + */ + int (*is_name_acceptable) (const struct inode *inode, const char *name, + int len); + + void (*build_entry_key) (const struct inode *dir /* directory where + * entry is (or will + * be) in.*/ , + const struct qstr *name /* name of file + * referenced by this + * entry */ , + reiser4_key * result /* resulting key of + * directory entry */ ); + int (*build_readdir_key) (struct file *dir, reiser4_key * result); + int (*add_entry) (struct inode *object, struct dentry *where, + reiser4_object_create_data * data, + reiser4_dir_entry_desc * entry); + int (*rem_entry) (struct inode *object, struct dentry *where, + reiser4_dir_entry_desc * entry); + + /* + * initialize directory structure for newly created object. For normal + * unix directories, insert dot and dotdot. + */ + int (*init) (struct inode *object, struct inode *parent, + reiser4_object_create_data * data); + + /* destroy directory */ + int (*done) (struct inode *child); + + /* called when @subdir was just looked up in the @dir */ + int (*attach) (struct inode *subdir, struct inode *dir); + int (*detach) (struct inode *subdir, struct inode *dir); + + struct { + reiser4_block_nr(*add_entry) (const struct inode *); + reiser4_block_nr(*rem_entry) (const struct inode *); + reiser4_block_nr(*unlink) (const struct inode *, + const struct inode *); + } estimate; +} dir_plugin; + +extern dir_plugin dir_plugins[LAST_DIR_ID]; + +typedef struct formatting_plugin { + /* generic fields */ + plugin_header h; + /* returns non-zero iff file's tail has to be stored + in a direct item. */ + int (*have_tail) (const struct inode *inode, loff_t size); +} formatting_plugin; + +typedef struct hash_plugin { + /* generic fields */ + plugin_header h; + /* computes hash of the given name */ + __u64(*hash) (const unsigned char *name, int len); +} hash_plugin; + +typedef struct cipher_plugin { + /* generic fields */ + plugin_header h; + struct crypto_blkcipher * (*alloc) (void); + void (*free) (struct crypto_blkcipher *tfm); + /* Offset translator. For each offset this returns (k * offset), where + k (k >= 1) is an expansion factor of the cipher algorithm. + For all symmetric algorithms k == 1. For asymmetric algorithms (which + inflate data) offset translation guarantees that all disk cluster's + units will have keys smaller then next cluster's one. + */ + loff_t(*scale) (struct inode *inode, size_t blocksize, loff_t src); + /* Cipher algorithms can accept data only by chunks of cipher block + size. This method is to align any flow up to cipher block size when + we pass it to cipher algorithm. To align means to append padding of + special format specific to the cipher algorithm */ + int (*align_stream) (__u8 *tail, int clust_size, int blocksize); + /* low-level key manager (check, install, etc..) */ + int (*setkey) (struct crypto_tfm *tfm, const __u8 *key, + unsigned int keylen); + /* main text processing procedures */ + void (*encrypt) (__u32 *expkey, __u8 *dst, const __u8 *src); + void (*decrypt) (__u32 *expkey, __u8 *dst, const __u8 *src); +} cipher_plugin; + +typedef struct digest_plugin { + /* generic fields */ + plugin_header h; + /* fingerprint size in bytes */ + int fipsize; + struct crypto_hash * (*alloc) (void); + void (*free) (struct crypto_hash *tfm); +} digest_plugin; + +typedef struct compression_plugin { + /* generic fields */ + plugin_header h; + int (*init) (void); + /* the maximum number of bytes the size of the "compressed" data can + * exceed the uncompressed data. */ + int (*overrun) (unsigned src_len); + coa_t(*alloc) (tfm_action act); + void (*free) (coa_t coa, tfm_action act); + /* minimal size of the flow we still try to compress */ + int (*min_size_deflate) (void); + __u32(*checksum) (char *data, __u32 length); + /* main transform procedures */ + void (*compress) (coa_t coa, __u8 *src_first, unsigned src_len, + __u8 *dst_first, unsigned *dst_len); + void (*decompress) (coa_t coa, __u8 *src_first, unsigned src_len, + __u8 *dst_first, unsigned *dst_len); +} compression_plugin; + +typedef struct compression_mode_plugin { + /* generic fields */ + plugin_header h; + /* this is called when estimating compressibility + of a logical cluster by its content */ + int (*should_deflate) (struct inode *inode, cloff_t index); + /* this is called when results of compression should be saved */ + int (*accept_hook) (struct inode *inode, cloff_t index); + /* this is called when results of compression should be discarded */ + int (*discard_hook) (struct inode *inode, cloff_t index); +} compression_mode_plugin; + +typedef struct cluster_plugin { + /* generic fields */ + plugin_header h; + int shift; +} cluster_plugin; + +typedef struct sd_ext_plugin { + /* generic fields */ + plugin_header h; + int (*present) (struct inode *inode, char **area, int *len); + int (*absent) (struct inode *inode); + int (*save_len) (struct inode *inode); + int (*save) (struct inode *inode, char **area); + /* alignment requirement for this stat-data part */ + int alignment; +} sd_ext_plugin; + +/* this plugin contains methods to allocate objectid for newly created files, + to deallocate objectid when file gets removed, to report number of used and + free objectids */ +typedef struct oid_allocator_plugin { + /* generic fields */ + plugin_header h; + int (*init_oid_allocator) (reiser4_oid_allocator * map, __u64 nr_files, + __u64 oids); + /* used to report statfs->f_files */ + __u64(*oids_used) (reiser4_oid_allocator * map); + /* get next oid to use */ + __u64(*next_oid) (reiser4_oid_allocator * map); + /* used to report statfs->f_ffree */ + __u64(*oids_free) (reiser4_oid_allocator * map); + /* allocate new objectid */ + int (*allocate_oid) (reiser4_oid_allocator * map, oid_t *); + /* release objectid */ + int (*release_oid) (reiser4_oid_allocator * map, oid_t); + /* how many pages to reserve in transaction for allocation of new + objectid */ + int (*oid_reserve_allocate) (reiser4_oid_allocator * map); + /* how many pages to reserve in transaction for freeing of an + objectid */ + int (*oid_reserve_release) (reiser4_oid_allocator * map); + void (*print_info) (const char *, reiser4_oid_allocator *); +} oid_allocator_plugin; + +/* disk layout plugin: this specifies super block, journal, bitmap (if there + are any) locations, etc */ +typedef struct disk_format_plugin { + /* generic fields */ + plugin_header h; + /* replay journal, initialize super_info_data, etc */ + int (*init_format) (struct super_block *, void *data); + + /* key of root directory stat data */ + const reiser4_key * (*root_dir_key) (const struct super_block *); + + int (*release) (struct super_block *); + jnode * (*log_super) (struct super_block *); + int (*check_open) (const struct inode *object); + int (*version_update) (struct super_block *); +} disk_format_plugin; + +struct jnode_plugin { + /* generic fields */ + plugin_header h; + int (*init) (jnode * node); + int (*parse) (jnode * node); + struct address_space *(*mapping) (const jnode * node); + unsigned long (*index) (const jnode * node); + jnode * (*clone) (jnode * node); +}; + +/* plugin instance. */ +/* */ +/* This is "wrapper" union for all types of plugins. Most of the code uses */ +/* plugins of particular type (file_plugin, dir_plugin, etc.) rather than */ +/* operates with pointers to reiser4_plugin. This union is only used in */ +/* some generic code in plugin/plugin.c that operates on all */ +/* plugins. Technically speaking purpose of this union is to add type */ +/* safety to said generic code: each plugin type (file_plugin, for */ +/* example), contains plugin_header as its first memeber. This first member */ +/* is located at the same place in memory as .h member of */ +/* reiser4_plugin. Generic code, obtains pointer to reiser4_plugin and */ +/* looks in the .h which is header of plugin type located in union. This */ +/* allows to avoid type-casts. */ +union reiser4_plugin { + /* generic fields */ + plugin_header h; + /* file plugin */ + file_plugin file; + /* directory plugin */ + dir_plugin dir; + /* hash plugin, used by directory plugin */ + hash_plugin hash; + /* fibration plugin used by directory plugin */ + fibration_plugin fibration; + /* cipher transform plugin, used by file plugin */ + cipher_plugin cipher; + /* digest transform plugin, used by file plugin */ + digest_plugin digest; + /* compression transform plugin, used by file plugin */ + compression_plugin compression; + /* tail plugin, used by file plugin */ + formatting_plugin formatting; + /* permission plugin */ + perm_plugin perm; + /* node plugin */ + node_plugin node; + /* item plugin */ + item_plugin item; + /* stat-data extension plugin */ + sd_ext_plugin sd_ext; + /* disk layout plugin */ + disk_format_plugin format; + /* object id allocator plugin */ + oid_allocator_plugin oid_allocator; + /* plugin for different jnode types */ + jnode_plugin jnode; + /* compression mode plugin, used by object plugin */ + compression_mode_plugin compression_mode; + /* cluster plugin, used by object plugin */ + cluster_plugin clust; + /* place-holder for new plugin types that can be registered + dynamically, and used by other dynamically loaded plugins. */ + void *generic; +}; + +struct reiser4_plugin_ops { + /* called when plugin is initialized */ + int (*init) (reiser4_plugin * plugin); + /* called when plugin is unloaded */ + int (*done) (reiser4_plugin * plugin); + /* load given plugin from disk */ + int (*load) (struct inode *inode, + reiser4_plugin * plugin, char **area, int *len); + /* how many space is required to store this plugin's state + in stat-data */ + int (*save_len) (struct inode *inode, reiser4_plugin * plugin); + /* save persistent plugin-data to disk */ + int (*save) (struct inode *inode, reiser4_plugin * plugin, + char **area); + /* alignment requirement for on-disk state of this plugin + in number of bytes */ + int alignment; + /* install itself into given inode. This can return error + (e.g., you cannot change hash of non-empty directory). */ + int (*change) (struct inode *inode, reiser4_plugin * plugin, + pset_member memb); + /* install itself into given inode. This can return error + (e.g., you cannot change hash of non-empty directory). */ + int (*inherit) (struct inode *inode, struct inode *parent, + reiser4_plugin * plugin); +}; + +/* functions implemented in fs/reiser4/plugin/plugin.c */ + +/* stores plugin reference in reiser4-specific part of inode */ +extern int set_object_plugin(struct inode *inode, reiser4_plugin_id id); +extern int init_plugins(void); + +/* builtin plugins */ + +/* builtin hash-plugins */ + +typedef enum { + RUPASOV_HASH_ID, + R5_HASH_ID, + TEA_HASH_ID, + FNV1_HASH_ID, + DEGENERATE_HASH_ID, + LAST_HASH_ID +} reiser4_hash_id; + +/* builtin cipher plugins */ + +typedef enum { + NONE_CIPHER_ID, + LAST_CIPHER_ID +} reiser4_cipher_id; + +/* builtin digest plugins */ + +typedef enum { + SHA256_32_DIGEST_ID, + LAST_DIGEST_ID +} reiser4_digest_id; + +/* builtin compression mode plugins */ +typedef enum { + NONE_COMPRESSION_MODE_ID, + LATTD_COMPRESSION_MODE_ID, + ULTIM_COMPRESSION_MODE_ID, + FORCE_COMPRESSION_MODE_ID, + CONVX_COMPRESSION_MODE_ID, + LAST_COMPRESSION_MODE_ID +} reiser4_compression_mode_id; + +/* builtin cluster plugins */ +typedef enum { + CLUSTER_64K_ID, + CLUSTER_32K_ID, + CLUSTER_16K_ID, + CLUSTER_8K_ID, + CLUSTER_4K_ID, + LAST_CLUSTER_ID +} reiser4_cluster_id; + +/* builtin tail-plugins */ + +typedef enum { + NEVER_TAILS_FORMATTING_ID, + ALWAYS_TAILS_FORMATTING_ID, + SMALL_FILE_FORMATTING_ID, + LAST_TAIL_FORMATTING_ID +} reiser4_formatting_id; + +/* data type used to pack parameters that we pass to vfs object creation + function create_object() */ +struct reiser4_object_create_data { + /* plugin to control created object */ + reiser4_file_id id; + /* mode of regular file, directory or special file */ +/* what happens if some other sort of perm plugin is in use? */ + int mode; + /* rdev of special file */ + dev_t rdev; + /* symlink target */ + const char *name; + /* add here something for non-standard objects you invent, like + query for interpolation file etc. */ + + struct reiser4_crypto_info *crypto; + + struct inode *parent; + struct dentry *dentry; +}; + +/* description of directory entry being created/destroyed/sought for + + It is passed down to the directory plugin and farther to the + directory item plugin methods. Creation of new directory is done in + several stages: first we search for an entry with the same name, then + create new one. reiser4_dir_entry_desc is used to store some information + collected at some stage of this process and required later: key of + item that we want to insert/delete and pointer to an object that will + be bound by the new directory entry. Probably some more fields will + be added there. + +*/ +struct reiser4_dir_entry_desc { + /* key of directory entry */ + reiser4_key key; + /* object bound by this entry. */ + struct inode *obj; +}; + +#define MAX_PLUGIN_TYPE_LABEL_LEN 32 +#define MAX_PLUGIN_PLUG_LABEL_LEN 32 + +#define PLUGIN_BY_ID(TYPE, ID, FIELD) \ +static inline TYPE *TYPE ## _by_id(reiser4_plugin_id id) \ +{ \ + reiser4_plugin *plugin = plugin_by_id(ID, id); \ + return plugin ? &plugin->FIELD : NULL; \ +} \ +static inline TYPE *TYPE ## _by_disk_id(reiser4_tree * tree, d16 *id) \ +{ \ + reiser4_plugin *plugin = plugin_by_disk_id(tree, ID, id); \ + return plugin ? &plugin->FIELD : NULL; \ +} \ +static inline TYPE *TYPE ## _by_unsafe_id(reiser4_plugin_id id) \ +{ \ + reiser4_plugin *plugin = plugin_by_unsafe_id(ID, id); \ + return plugin ? &plugin->FIELD : NULL; \ +} \ +static inline reiser4_plugin* TYPE ## _to_plugin(TYPE* plugin) \ +{ \ + return (reiser4_plugin *) plugin; \ +} \ +static inline reiser4_plugin_id TYPE ## _id(TYPE* plugin) \ +{ \ + return TYPE ## _to_plugin(plugin)->h.id; \ +} \ +typedef struct { int foo; } TYPE ## _plugin_dummy + +PLUGIN_BY_ID(item_plugin, REISER4_ITEM_PLUGIN_TYPE, item); +PLUGIN_BY_ID(file_plugin, REISER4_FILE_PLUGIN_TYPE, file); +PLUGIN_BY_ID(dir_plugin, REISER4_DIR_PLUGIN_TYPE, dir); +PLUGIN_BY_ID(node_plugin, REISER4_NODE_PLUGIN_TYPE, node); +PLUGIN_BY_ID(sd_ext_plugin, REISER4_SD_EXT_PLUGIN_TYPE, sd_ext); +PLUGIN_BY_ID(perm_plugin, REISER4_PERM_PLUGIN_TYPE, perm); +PLUGIN_BY_ID(hash_plugin, REISER4_HASH_PLUGIN_TYPE, hash); +PLUGIN_BY_ID(fibration_plugin, REISER4_FIBRATION_PLUGIN_TYPE, fibration); +PLUGIN_BY_ID(cipher_plugin, REISER4_CIPHER_PLUGIN_TYPE, cipher); +PLUGIN_BY_ID(digest_plugin, REISER4_DIGEST_PLUGIN_TYPE, digest); +PLUGIN_BY_ID(compression_plugin, REISER4_COMPRESSION_PLUGIN_TYPE, compression); +PLUGIN_BY_ID(formatting_plugin, REISER4_FORMATTING_PLUGIN_TYPE, formatting); +PLUGIN_BY_ID(disk_format_plugin, REISER4_FORMAT_PLUGIN_TYPE, format); +PLUGIN_BY_ID(jnode_plugin, REISER4_JNODE_PLUGIN_TYPE, jnode); +PLUGIN_BY_ID(compression_mode_plugin, REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + compression_mode); +PLUGIN_BY_ID(cluster_plugin, REISER4_CLUSTER_PLUGIN_TYPE, clust); + +extern int save_plugin_id(reiser4_plugin * plugin, d16 * area); + +extern struct list_head *get_plugin_list(reiser4_plugin_type type_id); + +#define for_all_plugins(ptype, plugin) \ +for (plugin = list_entry(get_plugin_list(ptype)->next, reiser4_plugin, h.linkage); \ + get_plugin_list(ptype) != &plugin->h.linkage; \ + plugin = list_entry(plugin->h.linkage.next, reiser4_plugin, h.linkage)) + + +extern int grab_plugin_pset(struct inode *self, struct inode *ancestor, + pset_member memb); +extern int force_plugin_pset(struct inode *self, pset_member memb, + reiser4_plugin *plug); +extern int finish_pset(struct inode *inode); + +/* defined in fs/reiser4/plugin/object.c */ +extern file_plugin file_plugins[LAST_FILE_PLUGIN_ID]; +/* defined in fs/reiser4/plugin/object.c */ +extern dir_plugin dir_plugins[LAST_DIR_ID]; +/* defined in fs/reiser4/plugin/item/static_stat.c */ +extern sd_ext_plugin sd_ext_plugins[LAST_SD_EXTENSION]; +/* defined in fs/reiser4/plugin/hash.c */ +extern hash_plugin hash_plugins[LAST_HASH_ID]; +/* defined in fs/reiser4/plugin/fibration.c */ +extern fibration_plugin fibration_plugins[LAST_FIBRATION_ID]; +/* defined in fs/reiser4/plugin/crypt.c */ +extern cipher_plugin cipher_plugins[LAST_CIPHER_ID]; +/* defined in fs/reiser4/plugin/digest.c */ +extern digest_plugin digest_plugins[LAST_DIGEST_ID]; +/* defined in fs/reiser4/plugin/compress/compress.c */ +extern compression_plugin compression_plugins[LAST_COMPRESSION_ID]; +/* defined in fs/reiser4/plugin/compress/compression_mode.c */ +extern compression_mode_plugin +compression_mode_plugins[LAST_COMPRESSION_MODE_ID]; +/* defined in fs/reiser4/plugin/cluster.c */ +extern cluster_plugin cluster_plugins[LAST_CLUSTER_ID]; +/* defined in fs/reiser4/plugin/tail.c */ +extern formatting_plugin formatting_plugins[LAST_TAIL_FORMATTING_ID]; +/* defined in fs/reiser4/plugin/security/security.c */ +extern perm_plugin perm_plugins[LAST_PERM_ID]; +/* defined in fs/reiser4/plugin/item/item.c */ +extern item_plugin item_plugins[LAST_ITEM_ID]; +/* defined in fs/reiser4/plugin/node/node.c */ +extern node_plugin node_plugins[LAST_NODE_ID]; +/* defined in fs/reiser4/plugin/disk_format/disk_format.c */ +extern disk_format_plugin format_plugins[LAST_FORMAT_ID]; + +/* __FS_REISER4_PLUGIN_TYPES_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/plugin_header.h linux-2.6.30/fs/reiser4/plugin/plugin_header.h --- linux-2.6.30.orig/fs/reiser4/plugin/plugin_header.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/plugin_header.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,157 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* plugin header. Data structures required by all plugin types. */ + +#if !defined(__PLUGIN_HEADER_H__) +#define __PLUGIN_HEADER_H__ + +/* plugin data-types and constants */ + +#include "../debug.h" +#include "../dformat.h" + +/* Every plugin type can be considered as a class of virtual objects + {(type, i) | i = 0, 1, ...}, which has one the following categories + of virtualization: + A - no virtualization; + F - per-file virtualization; + S - per-superblock virtualization; + FIXME-EDWARD: Define every such category */ + +/* Supported plugin types: (id, (virtualization category), short description) */ +typedef enum { + REISER4_FILE_PLUGIN_TYPE, /* (F) service VFS enry-points */ + REISER4_DIR_PLUGIN_TYPE, /* (F) service VFS enry-points */ + REISER4_ITEM_PLUGIN_TYPE, /* (F) manage items */ + REISER4_NODE_PLUGIN_TYPE, /* (S) manage formatted nodes */ + REISER4_HASH_PLUGIN_TYPE, /* (F) compute hash */ + REISER4_FIBRATION_PLUGIN_TYPE, /* (F) directory fibrations */ + REISER4_FORMATTING_PLUGIN_TYPE, /* (F) tail-packing policy */ + REISER4_PERM_PLUGIN_TYPE, /* stub (vacancy) */ + REISER4_SD_EXT_PLUGIN_TYPE, /* (A) stat-data extensions */ + REISER4_FORMAT_PLUGIN_TYPE, /* (S) specify disk format */ + REISER4_JNODE_PLUGIN_TYPE, /* (A) in-memory node headers */ + REISER4_CIPHER_PLUGIN_TYPE, /* (F) cipher transform algs */ + REISER4_DIGEST_PLUGIN_TYPE, /* (F) digest transform algs */ + REISER4_COMPRESSION_PLUGIN_TYPE, /* (F) compression tfm algs */ + REISER4_COMPRESSION_MODE_PLUGIN_TYPE, /* (F) compression heuristic */ + REISER4_CLUSTER_PLUGIN_TYPE, /* (F) size of logical cluster */ + REISER4_PLUGIN_TYPES +} reiser4_plugin_type; + +/* Supported plugin groups */ +typedef enum { + REISER4_DIRECTORY_FILE, + REISER4_REGULAR_FILE, + REISER4_SYMLINK_FILE, + REISER4_SPECIAL_FILE, +} file_plugin_group; + +struct reiser4_plugin_ops; +/* generic plugin operations, supported by each + plugin type. */ +typedef struct reiser4_plugin_ops reiser4_plugin_ops; + +/* the common part of all plugin instances. */ +typedef struct plugin_header { + /* plugin type */ + reiser4_plugin_type type_id; + /* id of this plugin */ + reiser4_plugin_id id; + /* bitmask of groups the plugin belongs to. */ + reiser4_plugin_groups groups; + /* plugin operations */ + reiser4_plugin_ops *pops; +/* NIKITA-FIXME-HANS: usage of and access to label and desc is not commented and + * defined. */ + /* short label of this plugin */ + const char *label; + /* descriptive string.. */ + const char *desc; + /* list linkage */ + struct list_head linkage; +} plugin_header; + +#define plugin_of_group(plug, group) (plug->h.groups & (1 << group)) + +/* PRIVATE INTERFACES */ +/* NIKITA-FIXME-HANS: what is this for and why does it duplicate what is in + * plugin_header? */ +/* plugin type representation. */ +struct reiser4_plugin_type_data { + /* internal plugin type identifier. Should coincide with + index of this item in plugins[] array. */ + reiser4_plugin_type type_id; + /* short symbolic label of this plugin type. Should be no longer + than MAX_PLUGIN_TYPE_LABEL_LEN characters including '\0'. */ + const char *label; + /* plugin type description longer than .label */ + const char *desc; + +/* NIKITA-FIXME-HANS: define built-in */ + /* number of built-in plugin instances of this type */ + int builtin_num; + /* array of built-in plugins */ + void *builtin; + struct list_head plugins_list; + size_t size; +}; + +extern struct reiser4_plugin_type_data plugins[REISER4_PLUGIN_TYPES]; + +int is_plugin_type_valid(reiser4_plugin_type type); +int is_plugin_id_valid(reiser4_plugin_type type, reiser4_plugin_id id); + +static inline reiser4_plugin *plugin_at(struct reiser4_plugin_type_data *ptype, + int i) +{ + char *builtin; + + builtin = ptype->builtin; + return (reiser4_plugin *) (builtin + i * ptype->size); +} + +/* return plugin by its @type_id and @id */ +static inline reiser4_plugin *plugin_by_id(reiser4_plugin_type type, + reiser4_plugin_id id) +{ + assert("nikita-1651", is_plugin_type_valid(type)); + assert("nikita-1652", is_plugin_id_valid(type, id)); + return plugin_at(&plugins[type], id); +} + +extern reiser4_plugin *plugin_by_unsafe_id(reiser4_plugin_type type_id, + reiser4_plugin_id id); + +/** + * plugin_by_disk_id - get reiser4_plugin + * @type_id: plugin type id + * @did: plugin id in disk format + * + * Returns reiser4_plugin by plugin type id an dplugin_id. + */ +static inline reiser4_plugin *plugin_by_disk_id(reiser4_tree * tree UNUSED_ARG, + reiser4_plugin_type type_id, + __le16 *plugin_id) +{ + /* + * what we should do properly is to maintain within each file-system a + * dictionary that maps on-disk plugin ids to "universal" ids. This + * dictionary will be resolved on mount time, so that this function + * will perform just one additional array lookup. + */ + return plugin_by_unsafe_id(type_id, le16_to_cpu(*plugin_id)); +} + +/* __PLUGIN_HEADER_H__ */ +#endif + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/plugin_set.c linux-2.6.30/fs/reiser4/plugin/plugin_set.c --- linux-2.6.30.orig/fs/reiser4/plugin/plugin_set.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/plugin_set.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,380 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ +/* This file contains Reiser4 plugin set operations */ + +/* plugin sets + * + * Each file in reiser4 is controlled by a whole set of plugins (file plugin, + * directory plugin, hash plugin, tail policy plugin, security plugin, etc.) + * assigned (inherited, deduced from mode bits, etc.) at creation time. This + * set of plugins (so called pset) is described by structure plugin_set (see + * plugin/plugin_set.h), which contains pointers to all required plugins. + * + * Children can inherit some pset members from their parent, however sometimes + * it is useful to specify members different from parent ones. Since object's + * pset can not be easily changed without fatal consequences, we use for this + * purpose another special plugin table (so called hset, or heir set) described + * by the same structure. + * + * Inode only stores a pointers to pset and hset. Different inodes with the + * same set of pset (hset) members point to the same pset (hset). This is + * archived by storing psets and hsets in global hash table. Races are avoided + * by simple (and efficient so far) solution of never recycling psets, even + * when last inode pointing to it is destroyed. + */ + +#include "../debug.h" +#include "../super.h" +#include "plugin_set.h" + +#include +#include + +/* slab for plugin sets */ +static struct kmem_cache *plugin_set_slab; + +static spinlock_t plugin_set_lock[8] __cacheline_aligned_in_smp = { + [0 ... 7] = SPIN_LOCK_UNLOCKED +}; + +/* hash table support */ + +#define PS_TABLE_SIZE (32) + +static inline plugin_set *cast_to(const unsigned long *a) +{ + return container_of(a, plugin_set, hashval); +} + +static inline int pseq(const unsigned long *a1, const unsigned long *a2) +{ + plugin_set *set1; + plugin_set *set2; + + /* make sure fields are not missed in the code below */ + cassert(sizeof *set1 == + sizeof set1->hashval + + sizeof set1->link + + sizeof set1->file + + sizeof set1->dir + + sizeof set1->perm + + sizeof set1->formatting + + sizeof set1->hash + + sizeof set1->fibration + + sizeof set1->sd + + sizeof set1->dir_item + + sizeof set1->cipher + + sizeof set1->digest + + sizeof set1->compression + + sizeof set1->compression_mode + + sizeof set1->cluster + + sizeof set1->create); + + set1 = cast_to(a1); + set2 = cast_to(a2); + return + set1->hashval == set2->hashval && + set1->file == set2->file && + set1->dir == set2->dir && + set1->perm == set2->perm && + set1->formatting == set2->formatting && + set1->hash == set2->hash && + set1->fibration == set2->fibration && + set1->sd == set2->sd && + set1->dir_item == set2->dir_item && + set1->cipher == set2->cipher && + set1->digest == set2->digest && + set1->compression == set2->compression && + set1->compression_mode == set2->compression_mode && + set1->cluster == set2->cluster && + set1->create == set2->create; +} + +#define HASH_FIELD(hash, set, field) \ +({ \ + (hash) += (unsigned long)(set)->field >> 2; \ +}) + +static inline unsigned long calculate_hash(const plugin_set * set) +{ + unsigned long result; + + result = 0; + HASH_FIELD(result, set, file); + HASH_FIELD(result, set, dir); + HASH_FIELD(result, set, perm); + HASH_FIELD(result, set, formatting); + HASH_FIELD(result, set, hash); + HASH_FIELD(result, set, fibration); + HASH_FIELD(result, set, sd); + HASH_FIELD(result, set, dir_item); + HASH_FIELD(result, set, cipher); + HASH_FIELD(result, set, digest); + HASH_FIELD(result, set, compression); + HASH_FIELD(result, set, compression_mode); + HASH_FIELD(result, set, cluster); + HASH_FIELD(result, set, create); + return result & (PS_TABLE_SIZE - 1); +} + +static inline unsigned long +pshash(ps_hash_table * table, const unsigned long *a) +{ + return *a; +} + +/* The hash table definition */ +#define KMALLOC(size) kmalloc((size), reiser4_ctx_gfp_mask_get()) +#define KFREE(ptr, size) kfree(ptr) +TYPE_SAFE_HASH_DEFINE(ps, plugin_set, unsigned long, hashval, link, pshash, + pseq); +#undef KFREE +#undef KMALLOC + +static ps_hash_table ps_table; +static plugin_set empty_set = { + .hashval = 0, + .file = NULL, + .dir = NULL, + .perm = NULL, + .formatting = NULL, + .hash = NULL, + .fibration = NULL, + .sd = NULL, + .dir_item = NULL, + .cipher = NULL, + .digest = NULL, + .compression = NULL, + .compression_mode = NULL, + .cluster = NULL, + .create = NULL, + .link = {NULL} +}; + +plugin_set *plugin_set_get_empty(void) +{ + return &empty_set; +} + +void plugin_set_put(plugin_set * set) +{ +} + +static inline unsigned long *pset_field(plugin_set * set, int offset) +{ + return (unsigned long *)(((char *)set) + offset); +} + +static int plugin_set_field(plugin_set ** set, const unsigned long val, + const int offset) +{ + unsigned long *spot; + spinlock_t *lock; + plugin_set replica; + plugin_set *twin; + plugin_set *psal; + plugin_set *orig; + + assert("nikita-2902", set != NULL); + assert("nikita-2904", *set != NULL); + + spot = pset_field(*set, offset); + if (unlikely(*spot == val)) + return 0; + + replica = *(orig = *set); + *pset_field(&replica, offset) = val; + replica.hashval = calculate_hash(&replica); + rcu_read_lock(); + twin = ps_hash_find(&ps_table, &replica.hashval); + if (unlikely(twin == NULL)) { + rcu_read_unlock(); + psal = kmem_cache_alloc(plugin_set_slab, + reiser4_ctx_gfp_mask_get()); + if (psal == NULL) + return RETERR(-ENOMEM); + *psal = replica; + lock = &plugin_set_lock[replica.hashval & 7]; + spin_lock(lock); + twin = ps_hash_find(&ps_table, &replica.hashval); + if (likely(twin == NULL)) { + *set = psal; + ps_hash_insert_rcu(&ps_table, psal); + } else { + *set = twin; + kmem_cache_free(plugin_set_slab, psal); + } + spin_unlock(lock); + } else { + rcu_read_unlock(); + *set = twin; + } + return 0; +} + +static struct { + int offset; + reiser4_plugin_groups groups; + reiser4_plugin_type type; +} pset_descr[PSET_LAST] = { + [PSET_FILE] = { + .offset = offsetof(plugin_set, file), + .type = REISER4_FILE_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_DIR] = { + .offset = offsetof(plugin_set, dir), + .type = REISER4_DIR_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_PERM] = { + .offset = offsetof(plugin_set, perm), + .type = REISER4_PERM_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_FORMATTING] = { + .offset = offsetof(plugin_set, formatting), + .type = REISER4_FORMATTING_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_HASH] = { + .offset = offsetof(plugin_set, hash), + .type = REISER4_HASH_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_FIBRATION] = { + .offset = offsetof(plugin_set, fibration), + .type = REISER4_FIBRATION_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_SD] = { + .offset = offsetof(plugin_set, sd), + .type = REISER4_ITEM_PLUGIN_TYPE, + .groups = (1 << STAT_DATA_ITEM_TYPE) + }, + [PSET_DIR_ITEM] = { + .offset = offsetof(plugin_set, dir_item), + .type = REISER4_ITEM_PLUGIN_TYPE, + .groups = (1 << DIR_ENTRY_ITEM_TYPE) + }, + [PSET_CIPHER] = { + .offset = offsetof(plugin_set, cipher), + .type = REISER4_CIPHER_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_DIGEST] = { + .offset = offsetof(plugin_set, digest), + .type = REISER4_DIGEST_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_COMPRESSION] = { + .offset = offsetof(plugin_set, compression), + .type = REISER4_COMPRESSION_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_COMPRESSION_MODE] = { + .offset = offsetof(plugin_set, compression_mode), + .type = REISER4_COMPRESSION_MODE_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_CLUSTER] = { + .offset = offsetof(plugin_set, cluster), + .type = REISER4_CLUSTER_PLUGIN_TYPE, + .groups = 0 + }, + [PSET_CREATE] = { + .offset = offsetof(plugin_set, create), + .type = REISER4_FILE_PLUGIN_TYPE, + .groups = (1 << REISER4_REGULAR_FILE) + } +}; + +#define DEFINE_PSET_OPS(PREFIX) \ + reiser4_plugin_type PREFIX##_member_to_type_unsafe(pset_member memb) \ +{ \ + if (memb > PSET_LAST) \ + return REISER4_PLUGIN_TYPES; \ + return pset_descr[memb].type; \ +} \ + \ +int PREFIX##_set_unsafe(plugin_set ** set, pset_member memb, \ + reiser4_plugin * plugin) \ +{ \ + assert("nikita-3492", set != NULL); \ + assert("nikita-3493", *set != NULL); \ + assert("nikita-3494", plugin != NULL); \ + assert("nikita-3495", 0 <= memb && memb < PSET_LAST); \ + assert("nikita-3496", plugin->h.type_id == pset_descr[memb].type); \ + \ + if (pset_descr[memb].groups) \ + if (!(pset_descr[memb].groups & plugin->h.groups)) \ + return -EINVAL; \ + \ + return plugin_set_field(set, \ + (unsigned long)plugin, pset_descr[memb].offset); \ +} \ + \ +reiser4_plugin *PREFIX##_get(plugin_set * set, pset_member memb) \ +{ \ + assert("nikita-3497", set != NULL); \ + assert("nikita-3498", 0 <= memb && memb < PSET_LAST); \ + \ + return *(reiser4_plugin **) (((char *)set) + pset_descr[memb].offset); \ +} + +DEFINE_PSET_OPS(aset); + +int set_plugin(plugin_set ** set, pset_member memb, reiser4_plugin * plugin) +{ + return plugin_set_field(set, + (unsigned long)plugin, pset_descr[memb].offset); +} + +/** + * init_plugin_set - create plugin set cache and hash table + * + * Initializes slab cache of plugin_set-s and their hash table. It is part of + * reiser4 module initialization. + */ +int init_plugin_set(void) +{ + int result; + + result = ps_hash_init(&ps_table, PS_TABLE_SIZE); + if (result == 0) { + plugin_set_slab = kmem_cache_create("plugin_set", + sizeof(plugin_set), 0, + SLAB_HWCACHE_ALIGN, + NULL); + if (plugin_set_slab == NULL) + result = RETERR(-ENOMEM); + } + return result; +} + +/** + * done_plugin_set - delete plugin_set cache and plugin_set hash table + * + * This is called on reiser4 module unloading or system shutdown. + */ +void done_plugin_set(void) +{ + plugin_set *cur, *next; + + for_all_in_htable(&ps_table, ps, cur, next) { + ps_hash_remove(&ps_table, cur); + kmem_cache_free(plugin_set_slab, cur); + } + destroy_reiser4_cache(&plugin_set_slab); + ps_hash_done(&ps_table); +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/plugin_set.h linux-2.6.30/fs/reiser4/plugin/plugin_set.h --- linux-2.6.30.orig/fs/reiser4/plugin/plugin_set.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/plugin_set.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,78 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Reiser4 plugin set definition. + See fs/reiser4/plugin/plugin_set.c for details */ + +#if !defined(__PLUGIN_SET_H__) +#define __PLUGIN_SET_H__ + +#include "../type_safe_hash.h" +#include "plugin.h" + +#include + +struct plugin_set; +typedef struct plugin_set plugin_set; + +TYPE_SAFE_HASH_DECLARE(ps, plugin_set); + +struct plugin_set { + unsigned long hashval; + /* plugin of file */ + file_plugin *file; + /* plugin of dir */ + dir_plugin *dir; + /* perm plugin for this file */ + perm_plugin *perm; + /* tail policy plugin. Only meaningful for regular files */ + formatting_plugin *formatting; + /* hash plugin. Only meaningful for directories. */ + hash_plugin *hash; + /* fibration plugin. Only meaningful for directories. */ + fibration_plugin *fibration; + /* plugin of stat-data */ + item_plugin *sd; + /* plugin of items a directory is built of */ + item_plugin *dir_item; + /* cipher plugin */ + cipher_plugin *cipher; + /* digest plugin */ + digest_plugin *digest; + /* compression plugin */ + compression_plugin *compression; + /* compression mode plugin */ + compression_mode_plugin *compression_mode; + /* cluster plugin */ + cluster_plugin *cluster; + /* this specifies file plugin of regular children. + only meaningful for directories */ + file_plugin *create; + ps_hash_link link; +}; + +extern plugin_set *plugin_set_get_empty(void); +extern void plugin_set_put(plugin_set * set); + +extern int init_plugin_set(void); +extern void done_plugin_set(void); + +extern reiser4_plugin *aset_get(plugin_set * set, pset_member memb); +extern int set_plugin(plugin_set ** set, pset_member memb, + reiser4_plugin * plugin); +extern int aset_set_unsafe(plugin_set ** set, pset_member memb, + reiser4_plugin * plugin); +extern reiser4_plugin_type aset_member_to_type_unsafe(pset_member memb); + +/* __PLUGIN_SET_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/security/Makefile linux-2.6.30/fs/reiser4/plugin/security/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/security/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/security/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,4 @@ +obj-$(CONFIG_REISER4_FS) += security_plugins.o + +security_plugins-objs := \ + perm.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/security/perm.c linux-2.6.30/fs/reiser4/plugin/security/perm.c --- linux-2.6.30.orig/fs/reiser4/plugin/security/perm.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/security/perm.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,33 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* + * This file contains implementation of permission plugins. + * See the comments in perm.h + */ + +#include "../plugin.h" +#include "../plugin_header.h" +#include "../../debug.h" + +perm_plugin perm_plugins[LAST_PERM_ID] = { + [NULL_PERM_ID] = { + .h = { + .type_id = REISER4_PERM_PLUGIN_TYPE, + .id = NULL_PERM_ID, + .pops = NULL, + .label = "null", + .desc = "stub permission plugin", + .linkage = {NULL, NULL} + } + } +}; + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/security/perm.h linux-2.6.30/fs/reiser4/plugin/security/perm.h --- linux-2.6.30.orig/fs/reiser4/plugin/security/perm.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/security/perm.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,38 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Perm (short for "permissions") plugins common stuff. */ + +#if !defined( __REISER4_PERM_H__ ) +#define __REISER4_PERM_H__ + +#include "../../forward.h" +#include "../plugin_header.h" + +#include + +/* Definition of permission plugin */ +/* NIKITA-FIXME-HANS: define what this is targeted for. + It does not seem to be intended for use with sys_reiser4. Explain. */ + +/* NOTE-EDWARD: This seems to be intended for deprecated sys_reiser4. + Consider it like a temporary "seam" and reserved pset member. + If you have something usefull to add, then rename this plugin and add here */ +typedef struct perm_plugin { + /* generic plugin fields */ + plugin_header h; +} perm_plugin; + +typedef enum { NULL_PERM_ID, LAST_PERM_ID } reiser4_perm_id; + +/* __REISER4_PERM_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/space/bitmap.c linux-2.6.30/fs/reiser4/plugin/space/bitmap.c --- linux-2.6.30.orig/fs/reiser4/plugin/space/bitmap.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/space/bitmap.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1585 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#include "../../debug.h" +#include "../../dformat.h" +#include "../../txnmgr.h" +#include "../../jnode.h" +#include "../../block_alloc.h" +#include "../../tree.h" +#include "../../super.h" +#include "../plugin.h" +#include "space_allocator.h" +#include "bitmap.h" + +#include +#include /* for struct super_block */ +#include +#include + +/* Proposed (but discarded) optimization: dynamic loading/unloading of bitmap + * blocks + + A useful optimization of reiser4 bitmap handling would be dynamic bitmap + blocks loading/unloading which is different from v3.x where all bitmap + blocks are loaded at mount time. + + To implement bitmap blocks unloading we need to count bitmap block usage + and detect currently unused blocks allowing them to be unloaded. It is not + a simple task since we allow several threads to modify one bitmap block + simultaneously. + + Briefly speaking, the following schema is proposed: we count in special + variable associated with each bitmap block. That is for counting of block + alloc/dealloc operations on that bitmap block. With a deferred block + deallocation feature of reiser4 all those operation will be represented in + atom dirty/deleted lists as jnodes for freshly allocated or deleted + nodes. + + So, we increment usage counter for each new node allocated or deleted, and + decrement it at atom commit one time for each node from the dirty/deleted + atom's list. Of course, freshly allocated node deletion and node reusing + from atom deleted (if we do so) list should decrement bitmap usage counter + also. + + This schema seems to be working but that reference counting is + not easy to debug. I think we should agree with Hans and do not implement + it in v4.0. Current code implements "on-demand" bitmap blocks loading only. + + For simplicity all bitmap nodes (both commit and working bitmap blocks) are + loaded into memory on fs mount time or each bitmap nodes are loaded at the + first access to it, the "dont_load_bitmap" mount option controls whether + bimtap nodes should be loaded at mount time. Dynamic unloading of bitmap + nodes currently is not supported. */ + +#define CHECKSUM_SIZE 4 + +#define BYTES_PER_LONG (sizeof(long)) + +#if BITS_PER_LONG == 64 +# define LONG_INT_SHIFT (6) +#else +# define LONG_INT_SHIFT (5) +#endif + +#define LONG_INT_MASK (BITS_PER_LONG - 1UL) + +typedef unsigned long ulong_t; + +#define bmap_size(blocksize) ((blocksize) - CHECKSUM_SIZE) +#define bmap_bit_count(blocksize) (bmap_size(blocksize) << 3) + +/* Block allocation/deallocation are done through special bitmap objects which + are allocated in an array at fs mount. */ +struct bitmap_node { + struct mutex mutex; /* long term lock object */ + + jnode *wjnode; /* j-nodes for WORKING ... */ + jnode *cjnode; /* ... and COMMIT bitmap blocks */ + + bmap_off_t first_zero_bit; /* for skip_busy option implementation */ + + atomic_t loaded; /* a flag which shows that bnode is loaded + * already */ +}; + +static inline char *bnode_working_data(struct bitmap_node *bnode) +{ + char *data; + + data = jdata(bnode->wjnode); + assert("zam-429", data != NULL); + + return data + CHECKSUM_SIZE; +} + +static inline char *bnode_commit_data(const struct bitmap_node *bnode) +{ + char *data; + + data = jdata(bnode->cjnode); + assert("zam-430", data != NULL); + + return data + CHECKSUM_SIZE; +} + +static inline __u32 bnode_commit_crc(const struct bitmap_node *bnode) +{ + char *data; + + data = jdata(bnode->cjnode); + assert("vpf-261", data != NULL); + + return le32_to_cpu(get_unaligned((d32 *)data)); +} + +static inline void bnode_set_commit_crc(struct bitmap_node *bnode, __u32 crc) +{ + char *data; + + data = jdata(bnode->cjnode); + assert("vpf-261", data != NULL); + + put_unaligned(cpu_to_le32(crc), (d32 *)data); +} + +/* ZAM-FIXME-HANS: is the idea that this might be a union someday? having + * written the code, does this added abstraction still have */ +/* ANSWER(Zam): No, the abstractions is in the level above (exact place is the + * reiser4_space_allocator structure) */ +/* ZAM-FIXME-HANS: I don't understand your english in comment above. */ +/* FIXME-HANS(Zam): I don't understand the questions like "might be a union + * someday?". What they about? If there is a reason to have a union, it should + * be a union, if not, it should not be a union. "..might be someday" means no + * reason. */ +struct bitmap_allocator_data { + /* an array for bitmap blocks direct access */ + struct bitmap_node *bitmap; +}; + +#define get_barray(super) \ +(((struct bitmap_allocator_data *)(get_super_private(super)->space_allocator.u.generic)) -> bitmap) + +#define get_bnode(super, i) (get_barray(super) + i) + +/* allocate and initialize jnode with JNODE_BITMAP type */ +static jnode *bnew(void) +{ + jnode *jal = jalloc(); + + if (jal) + jnode_init(jal, current_tree, JNODE_BITMAP); + + return jal; +} + +/* this file contains: + - bitmap based implementation of space allocation plugin + - all the helper functions like set bit, find_first_zero_bit, etc */ + +/* Audited by: green(2002.06.12) */ +static int find_next_zero_bit_in_word(ulong_t word, int start_bit) +{ + ulong_t mask = 1UL << start_bit; + int i = start_bit; + + while ((word & mask) != 0) { + mask <<= 1; + if (++i >= BITS_PER_LONG) + break; + } + + return i; +} + +#include + +#if BITS_PER_LONG == 64 + +#define OFF(addr) (((ulong_t)(addr) & (BYTES_PER_LONG - 1)) << 3) +#define BASE(addr) ((ulong_t*) ((ulong_t)(addr) & ~(BYTES_PER_LONG - 1))) + +static inline void reiser4_set_bit(int nr, void *addr) +{ + ext2_set_bit(nr + OFF(addr), BASE(addr)); +} + +static inline void reiser4_clear_bit(int nr, void *addr) +{ + ext2_clear_bit(nr + OFF(addr), BASE(addr)); +} + +static inline int reiser4_test_bit(int nr, void *addr) +{ + return ext2_test_bit(nr + OFF(addr), BASE(addr)); +} +static inline int reiser4_find_next_zero_bit(void *addr, int maxoffset, + int offset) +{ + int off = OFF(addr); + + return ext2_find_next_zero_bit(BASE(addr), maxoffset + off, + offset + off) - off; +} + +#else + +#define reiser4_set_bit(nr, addr) ext2_set_bit(nr, addr) +#define reiser4_clear_bit(nr, addr) ext2_clear_bit(nr, addr) +#define reiser4_test_bit(nr, addr) ext2_test_bit(nr, addr) + +#define reiser4_find_next_zero_bit(addr, maxoffset, offset) \ +ext2_find_next_zero_bit(addr, maxoffset, offset) +#endif + +/* Search for a set bit in the bit array [@start_offset, @max_offset[, offsets + * are counted from @addr, return the offset of the first bit if it is found, + * @maxoffset otherwise. */ +static bmap_off_t __reiser4_find_next_set_bit(void *addr, bmap_off_t max_offset, + bmap_off_t start_offset) +{ + ulong_t *base = addr; + /* start_offset is in bits, convert it to byte offset within bitmap. */ + int word_nr = start_offset >> LONG_INT_SHIFT; + /* bit number within the byte. */ + int bit_nr = start_offset & LONG_INT_MASK; + int max_word_nr = (max_offset - 1) >> LONG_INT_SHIFT; + + assert("zam-387", max_offset != 0); + + /* Unaligned @start_offset case. */ + if (bit_nr != 0) { + bmap_nr_t nr; + + nr = find_next_zero_bit_in_word(~(base[word_nr]), bit_nr); + + if (nr < BITS_PER_LONG) + return (word_nr << LONG_INT_SHIFT) + nr; + + ++word_nr; + } + + /* Fast scan trough aligned words. */ + while (word_nr <= max_word_nr) { + if (base[word_nr] != 0) { + return (word_nr << LONG_INT_SHIFT) + + find_next_zero_bit_in_word(~(base[word_nr]), 0); + } + + ++word_nr; + } + + return max_offset; +} + +#if BITS_PER_LONG == 64 + +static bmap_off_t reiser4_find_next_set_bit(void *addr, bmap_off_t max_offset, + bmap_off_t start_offset) +{ + bmap_off_t off = OFF(addr); + + return __reiser4_find_next_set_bit(BASE(addr), max_offset + off, + start_offset + off) - off; +} + +#else +#define reiser4_find_next_set_bit(addr, max_offset, start_offset) \ + __reiser4_find_next_set_bit(addr, max_offset, start_offset) +#endif + +/* search for the first set bit in single word. */ +static int find_last_set_bit_in_word(ulong_t word, int start_bit) +{ + ulong_t bit_mask; + int nr = start_bit; + + assert("zam-965", start_bit < BITS_PER_LONG); + assert("zam-966", start_bit >= 0); + + bit_mask = (1UL << nr); + + while (bit_mask != 0) { + if (bit_mask & word) + return nr; + bit_mask >>= 1; + nr--; + } + return BITS_PER_LONG; +} + +/* Search bitmap for a set bit in backward direction from the end to the + * beginning of given region + * + * @result: result offset of the last set bit + * @addr: base memory address, + * @low_off: low end of the search region, edge bit included into the region, + * @high_off: high end of the search region, edge bit included into the region, + * + * @return: 0 - set bit was found, -1 otherwise. + */ +static int +reiser4_find_last_set_bit(bmap_off_t * result, void *addr, bmap_off_t low_off, + bmap_off_t high_off) +{ + ulong_t *base = addr; + int last_word; + int first_word; + int last_bit; + int nr; + + assert("zam-962", high_off >= low_off); + + last_word = high_off >> LONG_INT_SHIFT; + last_bit = high_off & LONG_INT_MASK; + first_word = low_off >> LONG_INT_SHIFT; + + if (last_bit < BITS_PER_LONG) { + nr = find_last_set_bit_in_word(base[last_word], last_bit); + if (nr < BITS_PER_LONG) { + *result = (last_word << LONG_INT_SHIFT) + nr; + return 0; + } + --last_word; + } + while (last_word >= first_word) { + if (base[last_word] != 0x0) { + last_bit = + find_last_set_bit_in_word(base[last_word], + BITS_PER_LONG - 1); + assert("zam-972", last_bit < BITS_PER_LONG); + *result = (last_word << LONG_INT_SHIFT) + last_bit; + return 0; + } + --last_word; + } + + return -1; /* set bit not found */ +} + +/* Search bitmap for a clear bit in backward direction from the end to the + * beginning of given region */ +static int +reiser4_find_last_zero_bit(bmap_off_t * result, void *addr, bmap_off_t low_off, + bmap_off_t high_off) +{ + ulong_t *base = addr; + int last_word; + int first_word; + int last_bit; + int nr; + + last_word = high_off >> LONG_INT_SHIFT; + last_bit = high_off & LONG_INT_MASK; + first_word = low_off >> LONG_INT_SHIFT; + + if (last_bit < BITS_PER_LONG) { + nr = find_last_set_bit_in_word(~base[last_word], last_bit); + if (nr < BITS_PER_LONG) { + *result = (last_word << LONG_INT_SHIFT) + nr; + return 0; + } + --last_word; + } + while (last_word >= first_word) { + if (base[last_word] != (ulong_t) (-1)) { + *result = (last_word << LONG_INT_SHIFT) + + find_last_set_bit_in_word(~base[last_word], + BITS_PER_LONG - 1); + return 0; + } + --last_word; + } + + return -1; /* zero bit not found */ +} + +/* Audited by: green(2002.06.12) */ +static void reiser4_clear_bits(char *addr, bmap_off_t start, bmap_off_t end) +{ + int first_byte; + int last_byte; + + unsigned char first_byte_mask = 0xFF; + unsigned char last_byte_mask = 0xFF; + + assert("zam-410", start < end); + + first_byte = start >> 3; + last_byte = (end - 1) >> 3; + + if (last_byte > first_byte + 1) + memset(addr + first_byte + 1, 0, + (size_t) (last_byte - first_byte - 1)); + + first_byte_mask >>= 8 - (start & 0x7); + last_byte_mask <<= ((end - 1) & 0x7) + 1; + + if (first_byte == last_byte) { + addr[first_byte] &= (first_byte_mask | last_byte_mask); + } else { + addr[first_byte] &= first_byte_mask; + addr[last_byte] &= last_byte_mask; + } +} + +/* Audited by: green(2002.06.12) */ +/* ZAM-FIXME-HANS: comment this */ +static void reiser4_set_bits(char *addr, bmap_off_t start, bmap_off_t end) +{ + int first_byte; + int last_byte; + + unsigned char first_byte_mask = 0xFF; + unsigned char last_byte_mask = 0xFF; + + assert("zam-386", start < end); + + first_byte = start >> 3; + last_byte = (end - 1) >> 3; + + if (last_byte > first_byte + 1) + memset(addr + first_byte + 1, 0xFF, + (size_t) (last_byte - first_byte - 1)); + + first_byte_mask <<= start & 0x7; + last_byte_mask >>= 7 - ((end - 1) & 0x7); + + if (first_byte == last_byte) { + addr[first_byte] |= (first_byte_mask & last_byte_mask); + } else { + addr[first_byte] |= first_byte_mask; + addr[last_byte] |= last_byte_mask; + } +} + +#define ADLER_BASE 65521 +#define ADLER_NMAX 5552 + +/* Calculates the adler32 checksum for the data pointed by `data` of the + length `len`. This function was originally taken from zlib, version 1.1.3, + July 9th, 1998. + + Copyright (C) 1995-1998 Jean-loup Gailly and Mark Adler + + This software is provided 'as-is', without any express or implied + warranty. In no event will the authors be held liable for any damages + arising from the use of this software. + + Permission is granted to anyone to use this software for any purpose, + including commercial applications, and to alter it and redistribute it + freely, subject to the following restrictions: + + 1. The origin of this software must not be misrepresented; you must not + claim that you wrote the original software. If you use this software + in a product, an acknowledgment in the product documentation would be + appreciated but is not required. + 2. Altered source versions must be plainly marked as such, and must not be + misrepresented as being the original software. + 3. This notice may not be removed or altered from any source distribution. + + Jean-loup Gailly Mark Adler + jloup@gzip.org madler@alumni.caltech.edu + + The above comment applies only to the reiser4_adler32 function. +*/ + +__u32 reiser4_adler32(char *data, __u32 len) +{ + unsigned char *t = data; + __u32 s1 = 1; + __u32 s2 = 0; + int k; + + while (len > 0) { + k = len < ADLER_NMAX ? len : ADLER_NMAX; + len -= k; + + while (k--) { + s1 += *t++; + s2 += s1; + } + + s1 %= ADLER_BASE; + s2 %= ADLER_BASE; + } + return (s2 << 16) | s1; +} + +#define sb_by_bnode(bnode) \ + ((struct super_block *)jnode_get_tree(bnode->wjnode)->super) + +static __u32 bnode_calc_crc(const struct bitmap_node *bnode, unsigned long size) +{ + return reiser4_adler32(bnode_commit_data(bnode), bmap_size(size)); +} + +static int +bnode_check_adler32(const struct bitmap_node *bnode, unsigned long size) +{ + if (bnode_calc_crc(bnode, size) != bnode_commit_crc(bnode)) { + bmap_nr_t bmap; + + bmap = bnode - get_bnode(sb_by_bnode(bnode), 0); + + warning("vpf-263", + "Checksum for the bitmap block %llu is incorrect", + bmap); + + return RETERR(-EIO); + } + + return 0; +} + +#define REISER4_CHECK_BMAP_CRC (0) + +#if REISER4_CHECK_BMAP_CRC +static int bnode_check_crc(const struct bitmap_node *bnode) +{ + return bnode_check_adler32(bnode, + bmap_size(sb_by_bnode(bnode)->s_blocksize)); +} + +/* REISER4_CHECK_BMAP_CRC */ +#else + +#define bnode_check_crc(bnode) (0) + +/* REISER4_CHECK_BMAP_CRC */ +#endif + +/* Recalculates the adler32 checksum for only 1 byte change. + adler - previous adler checksum + old_data, data - old, new byte values. + tail == (chunk - offset) : length, checksum was calculated for, - offset of + the changed byte within this chunk. + This function can be used for checksum calculation optimisation. +*/ + +static __u32 +adler32_recalc(__u32 adler, unsigned char old_data, unsigned char data, + __u32 tail) +{ + __u32 delta = data - old_data + 2 * ADLER_BASE; + __u32 s1 = adler & 0xffff; + __u32 s2 = (adler >> 16) & 0xffff; + + s1 = (delta + s1) % ADLER_BASE; + s2 = (delta * tail + s2) % ADLER_BASE; + + return (s2 << 16) | s1; +} + +#define LIMIT(val, boundary) ((val) > (boundary) ? (boundary) : (val)) + +/** + * get_nr_bitmap - calculate number of bitmap blocks + * @super: super block with initialized blocksize and block count + * + * Calculates number of bitmap blocks of a filesystem which uses bitmaps to + * maintain free disk space. It assumes that each bitmap addresses the same + * number of blocks which is calculated by bmap_block_count macro defined in + * above. Number of blocks in the filesystem has to be initialized in reiser4 + * private data of super block already so that it can be obtained via + * reiser4_block_count(). Unfortunately, number of blocks addressed by a bitmap + * is not power of 2 because 4 bytes are used for checksum. Therefore, we have + * to use special function to divide and modulo 64bits filesystem block + * counters. + * + * Example: suppose filesystem have 32768 blocks. Blocksize is 4096. Each bitmap + * block addresses (4096 - 4) * 8 = 32736 blocks. Number of bitmaps to address + * all 32768 blocks is calculated as (32768 - 1) / 32736 + 1 = 2. + */ +static bmap_nr_t get_nr_bmap(const struct super_block *super) +{ + u64 quotient; + + assert("zam-393", reiser4_block_count(super) != 0); + + quotient = reiser4_block_count(super) - 1; + do_div(quotient, bmap_bit_count(super->s_blocksize)); + return quotient + 1; +} + +/** + * parse_blocknr - calculate bitmap number and offset in it by block number + * @block: pointer to block number to calculate location in bitmap of + * @bmap: pointer where to store bitmap block number + * @offset: pointer where to store offset within bitmap block + * + * Calculates location of bit which is responsible for allocation/freeing of + * block @*block. That location is represented by bitmap block number and offset + * within that bitmap block. + */ +static void +parse_blocknr(const reiser4_block_nr *block, bmap_nr_t *bmap, + bmap_off_t *offset) +{ + struct super_block *super = get_current_context()->super; + u64 quotient = *block; + + *offset = do_div(quotient, bmap_bit_count(super->s_blocksize)); + *bmap = quotient; + + assert("zam-433", *bmap < get_nr_bmap(super)); + assert("", *offset < bmap_bit_count(super->s_blocksize)); +} + +#if REISER4_DEBUG +/* Audited by: green(2002.06.12) */ +static void +check_block_range(const reiser4_block_nr * start, const reiser4_block_nr * len) +{ + struct super_block *sb = reiser4_get_current_sb(); + + assert("zam-436", sb != NULL); + + assert("zam-455", start != NULL); + assert("zam-437", *start != 0); + assert("zam-541", !reiser4_blocknr_is_fake(start)); + assert("zam-441", *start < reiser4_block_count(sb)); + + if (len != NULL) { + assert("zam-438", *len != 0); + assert("zam-442", *start + *len <= reiser4_block_count(sb)); + } +} + +static void check_bnode_loaded(const struct bitmap_node *bnode) +{ + assert("zam-485", bnode != NULL); + assert("zam-483", jnode_page(bnode->wjnode) != NULL); + assert("zam-484", jnode_page(bnode->cjnode) != NULL); + assert("nikita-2820", jnode_is_loaded(bnode->wjnode)); + assert("nikita-2821", jnode_is_loaded(bnode->cjnode)); +} + +#else + +# define check_block_range(start, len) do { /* nothing */} while(0) +# define check_bnode_loaded(bnode) do { /* nothing */} while(0) + +#endif + +/* modify bnode->first_zero_bit (if we free bits before); bnode should be + spin-locked */ +static inline void +adjust_first_zero_bit(struct bitmap_node *bnode, bmap_off_t offset) +{ + if (offset < bnode->first_zero_bit) + bnode->first_zero_bit = offset; +} + +/* return a physical disk address for logical bitmap number @bmap */ +/* FIXME-VS: this is somehow related to disk layout? */ +/* ZAM-FIXME-HANS: your answer is? Use not more than one function dereference + * per block allocation so that performance is not affected. Probably this + * whole file should be considered part of the disk layout plugin, and other + * disk layouts can use other defines and efficiency will not be significantly + * affected. */ + +#define REISER4_FIRST_BITMAP_BLOCK \ + ((REISER4_MASTER_OFFSET / PAGE_CACHE_SIZE) + 2) + +/* Audited by: green(2002.06.12) */ +static void +get_bitmap_blocknr(struct super_block *super, bmap_nr_t bmap, + reiser4_block_nr * bnr) +{ + + assert("zam-390", bmap < get_nr_bmap(super)); + +#ifdef CONFIG_REISER4_BADBLOCKS +#define BITMAP_PLUGIN_DISKMAP_ID ((0xc0e1<<16) | (0xe0ff)) + /* Check if the diskmap have this already, first. */ + if (reiser4_get_diskmap_value(BITMAP_PLUGIN_DISKMAP_ID, bmap, bnr) == 0) + return; /* Found it in diskmap */ +#endif + /* FIXME_ZAM: before discussing of disk layouts and disk format + plugins I implement bitmap location scheme which is close to scheme + used in reiser 3.6 */ + if (bmap == 0) { + *bnr = REISER4_FIRST_BITMAP_BLOCK; + } else { + *bnr = bmap * bmap_bit_count(super->s_blocksize); + } +} + +/* construct a fake block number for shadow bitmap (WORKING BITMAP) block */ +/* Audited by: green(2002.06.12) */ +static void get_working_bitmap_blocknr(bmap_nr_t bmap, reiser4_block_nr * bnr) +{ + *bnr = + (reiser4_block_nr) ((bmap & ~REISER4_BLOCKNR_STATUS_BIT_MASK) | + REISER4_BITMAP_BLOCKS_STATUS_VALUE); +} + +/* bnode structure initialization */ +static void +init_bnode(struct bitmap_node *bnode, + struct super_block *super UNUSED_ARG, bmap_nr_t bmap UNUSED_ARG) +{ + memset(bnode, 0, sizeof(struct bitmap_node)); + + mutex_init(&bnode->mutex); + atomic_set(&bnode->loaded, 0); +} + +static void release(jnode * node) +{ + jrelse(node); + JF_SET(node, JNODE_HEARD_BANSHEE); + jput(node); +} + +/* This function is for internal bitmap.c use because it assumes that jnode is + in under full control of this thread */ +static void done_bnode(struct bitmap_node *bnode) +{ + if (bnode) { + atomic_set(&bnode->loaded, 0); + if (bnode->wjnode != NULL) + release(bnode->wjnode); + if (bnode->cjnode != NULL) + release(bnode->cjnode); + bnode->wjnode = bnode->cjnode = NULL; + } +} + +/* ZAM-FIXME-HANS: comment this. Called only by load_and_lock_bnode()*/ +static int prepare_bnode(struct bitmap_node *bnode, jnode **cjnode_ret, + jnode **wjnode_ret) +{ + struct super_block *super; + jnode *cjnode; + jnode *wjnode; + bmap_nr_t bmap; + int ret; + + super = reiser4_get_current_sb(); + + *wjnode_ret = wjnode = bnew(); + if (wjnode == NULL) { + *cjnode_ret = NULL; + return RETERR(-ENOMEM); + } + + *cjnode_ret = cjnode = bnew(); + if (cjnode == NULL) + return RETERR(-ENOMEM); + + bmap = bnode - get_bnode(super, 0); + + get_working_bitmap_blocknr(bmap, &wjnode->blocknr); + get_bitmap_blocknr(super, bmap, &cjnode->blocknr); + + jref(cjnode); + jref(wjnode); + + /* load commit bitmap */ + ret = jload_gfp(cjnode, GFP_NOFS, 1); + + if (ret) + goto error; + + /* allocate memory for working bitmap block. Note that for + * bitmaps jinit_new() doesn't actually modifies node content, + * so parallel calls to this are ok. */ + ret = jinit_new(wjnode, GFP_NOFS); + + if (ret != 0) { + jrelse(cjnode); + goto error; + } + + return 0; + + error: + jput(cjnode); + jput(wjnode); + *wjnode_ret = *cjnode_ret = NULL; + return ret; + +} + +/* Check the bnode data on read. */ +static int check_struct_bnode(struct bitmap_node *bnode, __u32 blksize) +{ + void *data; + int ret; + + /* Check CRC */ + ret = bnode_check_adler32(bnode, blksize); + + if (ret) { + return ret; + } + + data = jdata(bnode->cjnode) + CHECKSUM_SIZE; + + /* Check the very first bit -- it must be busy. */ + if (!reiser4_test_bit(0, data)) { + warning("vpf-1362", "The allocator block %llu is not marked " + "as used.", (unsigned long long)bnode->cjnode->blocknr); + + return -EINVAL; + } + + return 0; +} + +/* load bitmap blocks "on-demand" */ +static int load_and_lock_bnode(struct bitmap_node *bnode) +{ + int ret; + + jnode *cjnode; + jnode *wjnode; + + assert("nikita-3040", reiser4_schedulable()); + +/* ZAM-FIXME-HANS: since bitmaps are never unloaded, this does not + * need to be atomic, right? Just leave a comment that if bitmaps were + * unloadable, this would need to be atomic. */ + if (atomic_read(&bnode->loaded)) { + /* bitmap is already loaded, nothing to do */ + check_bnode_loaded(bnode); + mutex_lock(&bnode->mutex); + assert("nikita-2827", atomic_read(&bnode->loaded)); + return 0; + } + + ret = prepare_bnode(bnode, &cjnode, &wjnode); + if (ret == 0) { + mutex_lock(&bnode->mutex); + + if (!atomic_read(&bnode->loaded)) { + assert("nikita-2822", cjnode != NULL); + assert("nikita-2823", wjnode != NULL); + assert("nikita-2824", jnode_is_loaded(cjnode)); + assert("nikita-2825", jnode_is_loaded(wjnode)); + + bnode->wjnode = wjnode; + bnode->cjnode = cjnode; + + ret = check_struct_bnode(bnode, current_blocksize); + if (!ret) { + cjnode = wjnode = NULL; + atomic_set(&bnode->loaded, 1); + /* working bitmap is initialized by on-disk + * commit bitmap. This should be performed + * under mutex. */ + memcpy(bnode_working_data(bnode), + bnode_commit_data(bnode), + bmap_size(current_blocksize)); + } else + mutex_unlock(&bnode->mutex); + } else + /* race: someone already loaded bitmap while we were + * busy initializing data. */ + check_bnode_loaded(bnode); + } + + if (wjnode != NULL) { + release(wjnode); + bnode->wjnode = NULL; + } + if (cjnode != NULL) { + release(cjnode); + bnode->cjnode = NULL; + } + + return ret; +} + +static void release_and_unlock_bnode(struct bitmap_node *bnode) +{ + check_bnode_loaded(bnode); + mutex_unlock(&bnode->mutex); +} + +/* This function does all block allocation work but only for one bitmap + block.*/ +/* FIXME_ZAM: It does not allow us to allocate block ranges across bitmap + block responsibility zone boundaries. This had no sense in v3.6 but may + have it in v4.x */ +/* ZAM-FIXME-HANS: do you mean search one bitmap block forward? */ +static int +search_one_bitmap_forward(bmap_nr_t bmap, bmap_off_t * offset, + bmap_off_t max_offset, int min_len, int max_len) +{ + struct super_block *super = get_current_context()->super; + struct bitmap_node *bnode = get_bnode(super, bmap); + + char *data; + + bmap_off_t search_end; + bmap_off_t start; + bmap_off_t end; + + int set_first_zero_bit = 0; + + int ret; + + assert("zam-364", min_len > 0); + assert("zam-365", max_len >= min_len); + assert("zam-366", *offset <= max_offset); + + ret = load_and_lock_bnode(bnode); + + if (ret) + return ret; + + data = bnode_working_data(bnode); + + start = *offset; + + if (bnode->first_zero_bit >= start) { + start = bnode->first_zero_bit; + set_first_zero_bit = 1; + } + + while (start + min_len < max_offset) { + + start = + reiser4_find_next_zero_bit((long *)data, max_offset, start); + if (set_first_zero_bit) { + bnode->first_zero_bit = start; + set_first_zero_bit = 0; + } + if (start >= max_offset) + break; + + search_end = LIMIT(start + max_len, max_offset); + end = + reiser4_find_next_set_bit((long *)data, search_end, start); + if (end >= start + min_len) { + /* we can't trust find_next_set_bit result if set bit + was not fount, result may be bigger than + max_offset */ + if (end > search_end) + end = search_end; + + ret = end - start; + *offset = start; + + reiser4_set_bits(data, start, end); + + /* FIXME: we may advance first_zero_bit if [start, + end] region overlaps the first_zero_bit point */ + + break; + } + + start = end + 1; + } + + release_and_unlock_bnode(bnode); + + return ret; +} + +static int +search_one_bitmap_backward(bmap_nr_t bmap, bmap_off_t * start_offset, + bmap_off_t end_offset, int min_len, int max_len) +{ + struct super_block *super = get_current_context()->super; + struct bitmap_node *bnode = get_bnode(super, bmap); + char *data; + bmap_off_t start; + int ret; + + assert("zam-958", min_len > 0); + assert("zam-959", max_len >= min_len); + assert("zam-960", *start_offset >= end_offset); + + ret = load_and_lock_bnode(bnode); + if (ret) + return ret; + + data = bnode_working_data(bnode); + start = *start_offset; + + while (1) { + bmap_off_t end, search_end; + + /* Find the beginning of the zero filled region */ + if (reiser4_find_last_zero_bit(&start, data, end_offset, start)) + break; + /* Is there more than `min_len' bits from `start' to + * `end_offset'? */ + if (start < end_offset + min_len - 1) + break; + + /* Do not search to `end_offset' if we need to find less than + * `max_len' zero bits. */ + if (end_offset + max_len - 1 < start) + search_end = start - max_len + 1; + else + search_end = end_offset; + + if (reiser4_find_last_set_bit(&end, data, search_end, start)) + end = search_end; + else + end++; + + if (end + min_len <= start + 1) { + if (end < search_end) + end = search_end; + ret = start - end + 1; + *start_offset = end; /* `end' is lowest offset */ + assert("zam-987", + reiser4_find_next_set_bit(data, start + 1, + end) >= start + 1); + reiser4_set_bits(data, end, start + 1); + break; + } + + if (end <= end_offset) + /* left search boundary reached. */ + break; + start = end - 1; + } + + release_and_unlock_bnode(bnode); + return ret; +} + +/* allocate contiguous range of blocks in bitmap */ +static int bitmap_alloc_forward(reiser4_block_nr * start, + const reiser4_block_nr * end, int min_len, + int max_len) +{ + bmap_nr_t bmap, end_bmap; + bmap_off_t offset, end_offset; + int len; + + reiser4_block_nr tmp; + + struct super_block *super = get_current_context()->super; + const bmap_off_t max_offset = bmap_bit_count(super->s_blocksize); + + parse_blocknr(start, &bmap, &offset); + + tmp = *end - 1; + parse_blocknr(&tmp, &end_bmap, &end_offset); + ++end_offset; + + assert("zam-358", end_bmap >= bmap); + assert("zam-359", ergo(end_bmap == bmap, end_offset >= offset)); + + for (; bmap < end_bmap; bmap++, offset = 0) { + len = + search_one_bitmap_forward(bmap, &offset, max_offset, + min_len, max_len); + if (len != 0) + goto out; + } + + len = + search_one_bitmap_forward(bmap, &offset, end_offset, min_len, + max_len); + out: + *start = bmap * max_offset + offset; + return len; +} + +/* allocate contiguous range of blocks in bitmap (from @start to @end in + * backward direction) */ +static int bitmap_alloc_backward(reiser4_block_nr * start, + const reiser4_block_nr * end, int min_len, + int max_len) +{ + bmap_nr_t bmap, end_bmap; + bmap_off_t offset, end_offset; + int len; + struct super_block *super = get_current_context()->super; + const bmap_off_t max_offset = bmap_bit_count(super->s_blocksize); + + parse_blocknr(start, &bmap, &offset); + parse_blocknr(end, &end_bmap, &end_offset); + + assert("zam-961", end_bmap <= bmap); + assert("zam-962", ergo(end_bmap == bmap, end_offset <= offset)); + + for (; bmap > end_bmap; bmap--, offset = max_offset - 1) { + len = + search_one_bitmap_backward(bmap, &offset, 0, min_len, + max_len); + if (len != 0) + goto out; + } + + len = + search_one_bitmap_backward(bmap, &offset, end_offset, min_len, + max_len); + out: + *start = bmap * max_offset + offset; + return len; +} + +/* plugin->u.space_allocator.alloc_blocks() */ +static int alloc_blocks_forward(reiser4_blocknr_hint *hint, int needed, + reiser4_block_nr *start, reiser4_block_nr *len) +{ + struct super_block *super = get_current_context()->super; + int actual_len; + + reiser4_block_nr search_start; + reiser4_block_nr search_end; + + assert("zam-398", super != NULL); + assert("zam-412", hint != NULL); + assert("zam-397", hint->blk <= reiser4_block_count(super)); + + if (hint->max_dist == 0) + search_end = reiser4_block_count(super); + else + search_end = + LIMIT(hint->blk + hint->max_dist, + reiser4_block_count(super)); + + /* We use @hint -> blk as a search start and search from it to the end + of the disk or in given region if @hint -> max_dist is not zero */ + search_start = hint->blk; + + actual_len = + bitmap_alloc_forward(&search_start, &search_end, 1, needed); + + /* There is only one bitmap search if max_dist was specified or first + pass was from the beginning of the bitmap. We also do one pass for + scanning bitmap in backward direction. */ + if (!(actual_len != 0 || hint->max_dist != 0 || search_start == 0)) { + /* next step is a scanning from 0 to search_start */ + search_end = search_start; + search_start = 0; + actual_len = + bitmap_alloc_forward(&search_start, &search_end, 1, needed); + } + if (actual_len == 0) + return RETERR(-ENOSPC); + if (actual_len < 0) + return RETERR(actual_len); + *len = actual_len; + *start = search_start; + return 0; +} + +static int alloc_blocks_backward(reiser4_blocknr_hint * hint, int needed, + reiser4_block_nr * start, + reiser4_block_nr * len) +{ + reiser4_block_nr search_start; + reiser4_block_nr search_end; + int actual_len; + + ON_DEBUG(struct super_block *super = reiser4_get_current_sb()); + + assert("zam-969", super != NULL); + assert("zam-970", hint != NULL); + assert("zam-971", hint->blk <= reiser4_block_count(super)); + + search_start = hint->blk; + if (hint->max_dist == 0 || search_start <= hint->max_dist) + search_end = 0; + else + search_end = search_start - hint->max_dist; + + actual_len = + bitmap_alloc_backward(&search_start, &search_end, 1, needed); + if (actual_len == 0) + return RETERR(-ENOSPC); + if (actual_len < 0) + return RETERR(actual_len); + *len = actual_len; + *start = search_start; + return 0; +} + +/* plugin->u.space_allocator.alloc_blocks() */ +int reiser4_alloc_blocks_bitmap(reiser4_space_allocator * allocator, + reiser4_blocknr_hint * hint, int needed, + reiser4_block_nr * start, reiser4_block_nr * len) +{ + if (hint->backward) + return alloc_blocks_backward(hint, needed, start, len); + return alloc_blocks_forward(hint, needed, start, len); +} + +/* plugin->u.space_allocator.dealloc_blocks(). */ +/* It just frees blocks in WORKING BITMAP. Usually formatted an unformatted + nodes deletion is deferred until transaction commit. However, deallocation + of temporary objects like wandered blocks and transaction commit records + requires immediate node deletion from WORKING BITMAP.*/ +void reiser4_dealloc_blocks_bitmap(reiser4_space_allocator * allocator, + reiser4_block_nr start, reiser4_block_nr len) +{ + struct super_block *super = reiser4_get_current_sb(); + + bmap_nr_t bmap; + bmap_off_t offset; + + struct bitmap_node *bnode; + int ret; + + assert("zam-468", len != 0); + check_block_range(&start, &len); + + parse_blocknr(&start, &bmap, &offset); + + assert("zam-469", offset + len <= bmap_bit_count(super->s_blocksize)); + + bnode = get_bnode(super, bmap); + + assert("zam-470", bnode != NULL); + + ret = load_and_lock_bnode(bnode); + assert("zam-481", ret == 0); + + reiser4_clear_bits(bnode_working_data(bnode), offset, + (bmap_off_t) (offset + len)); + + adjust_first_zero_bit(bnode, offset); + + release_and_unlock_bnode(bnode); +} + +/* plugin->u.space_allocator.check_blocks(). */ +void reiser4_check_blocks_bitmap(const reiser4_block_nr * start, + const reiser4_block_nr * len, int desired) +{ +#if REISER4_DEBUG + struct super_block *super = reiser4_get_current_sb(); + + bmap_nr_t bmap; + bmap_off_t start_offset; + bmap_off_t end_offset; + + struct bitmap_node *bnode; + int ret; + + assert("zam-622", len != NULL); + check_block_range(start, len); + parse_blocknr(start, &bmap, &start_offset); + + end_offset = start_offset + *len; + assert("nikita-2214", end_offset <= bmap_bit_count(super->s_blocksize)); + + bnode = get_bnode(super, bmap); + + assert("nikita-2215", bnode != NULL); + + ret = load_and_lock_bnode(bnode); + assert("zam-626", ret == 0); + + assert("nikita-2216", jnode_is_loaded(bnode->wjnode)); + + if (desired) { + assert("zam-623", + reiser4_find_next_zero_bit(bnode_working_data(bnode), + end_offset, start_offset) + >= end_offset); + } else { + assert("zam-624", + reiser4_find_next_set_bit(bnode_working_data(bnode), + end_offset, start_offset) + >= end_offset); + } + + release_and_unlock_bnode(bnode); +#endif +} + +/* conditional insertion of @node into atom's overwrite set if it was not there */ +static void cond_add_to_overwrite_set(txn_atom * atom, jnode * node) +{ + assert("zam-546", atom != NULL); + assert("zam-547", atom->stage == ASTAGE_PRE_COMMIT); + assert("zam-548", node != NULL); + + spin_lock_atom(atom); + spin_lock_jnode(node); + + if (node->atom == NULL) { + JF_SET(node, JNODE_OVRWR); + insert_into_atom_ovrwr_list(atom, node); + } else { + assert("zam-549", node->atom == atom); + } + + spin_unlock_jnode(node); + spin_unlock_atom(atom); +} + +/* an actor which applies delete set to COMMIT bitmap pages and link modified + pages in a single-linked list */ +static int +apply_dset_to_commit_bmap(txn_atom * atom, const reiser4_block_nr * start, + const reiser4_block_nr * len, void *data) +{ + + bmap_nr_t bmap; + bmap_off_t offset; + int ret; + + long long *blocks_freed_p = data; + + struct bitmap_node *bnode; + + struct super_block *sb = reiser4_get_current_sb(); + + check_block_range(start, len); + + parse_blocknr(start, &bmap, &offset); + + /* FIXME-ZAM: we assume that all block ranges are allocated by this + bitmap-based allocator and each block range can't go over a zone of + responsibility of one bitmap block; same assumption is used in + other journal hooks in bitmap code. */ + bnode = get_bnode(sb, bmap); + assert("zam-448", bnode != NULL); + + /* it is safe to unlock atom with is in ASTAGE_PRE_COMMIT */ + assert("zam-767", atom->stage == ASTAGE_PRE_COMMIT); + ret = load_and_lock_bnode(bnode); + if (ret) + return ret; + + /* put bnode into atom's overwrite set */ + cond_add_to_overwrite_set(atom, bnode->cjnode); + + data = bnode_commit_data(bnode); + + ret = bnode_check_crc(bnode); + if (ret != 0) + return ret; + + if (len != NULL) { + /* FIXME-ZAM: a check that all bits are set should be there */ + assert("zam-443", + offset + *len <= bmap_bit_count(sb->s_blocksize)); + reiser4_clear_bits(data, offset, (bmap_off_t) (offset + *len)); + + (*blocks_freed_p) += *len; + } else { + reiser4_clear_bit(offset, data); + (*blocks_freed_p)++; + } + + bnode_set_commit_crc(bnode, bnode_calc_crc(bnode, sb->s_blocksize)); + + release_and_unlock_bnode(bnode); + + return 0; +} + +/* plugin->u.space_allocator.pre_commit_hook(). */ +/* It just applies transaction changes to fs-wide COMMIT BITMAP, hoping the + rest is done by transaction manager (allocate wandered locations for COMMIT + BITMAP blocks, copy COMMIT BITMAP blocks data). */ +/* Only one instance of this function can be running at one given time, because + only one transaction can be committed a time, therefore it is safe to access + some global variables without any locking */ + +int reiser4_pre_commit_hook_bitmap(void) +{ + struct super_block *super = reiser4_get_current_sb(); + txn_atom *atom; + + long long blocks_freed = 0; + + atom = get_current_atom_locked(); + assert("zam-876", atom->stage == ASTAGE_PRE_COMMIT); + spin_unlock_atom(atom); + + { /* scan atom's captured list and find all freshly allocated nodes, + * mark corresponded bits in COMMIT BITMAP as used */ + struct list_head *head = ATOM_CLEAN_LIST(atom); + jnode *node = list_entry(head->next, jnode, capture_link); + + while (head != &node->capture_link) { + /* we detect freshly allocated jnodes */ + if (JF_ISSET(node, JNODE_RELOC)) { + int ret; + bmap_nr_t bmap; + + bmap_off_t offset; + bmap_off_t index; + struct bitmap_node *bn; + __u32 size = bmap_size(super->s_blocksize); + __u32 crc; + char byte; + + assert("zam-559", !JF_ISSET(node, JNODE_OVRWR)); + assert("zam-460", + !reiser4_blocknr_is_fake(&node->blocknr)); + + parse_blocknr(&node->blocknr, &bmap, &offset); + bn = get_bnode(super, bmap); + + index = offset >> 3; + assert("vpf-276", index < size); + + ret = bnode_check_crc(bnode); + if (ret != 0) + return ret; + + check_bnode_loaded(bn); + load_and_lock_bnode(bn); + + byte = *(bnode_commit_data(bn) + index); + reiser4_set_bit(offset, bnode_commit_data(bn)); + + crc = adler32_recalc(bnode_commit_crc(bn), byte, + *(bnode_commit_data(bn) + + index), + size - index), + bnode_set_commit_crc(bn, crc); + + release_and_unlock_bnode(bn); + + ret = bnode_check_crc(bn); + if (ret != 0) + return ret; + + /* working of this depends on how it inserts + new j-node into clean list, because we are + scanning the same list now. It is OK, if + insertion is done to the list front */ + cond_add_to_overwrite_set(atom, bn->cjnode); + } + + node = list_entry(node->capture_link.next, jnode, capture_link); + } + } + + blocknr_set_iterator(atom, &atom->delete_set, apply_dset_to_commit_bmap, + &blocks_freed, 0); + + blocks_freed -= atom->nr_blocks_allocated; + + { + reiser4_super_info_data *sbinfo; + + sbinfo = get_super_private(super); + + spin_lock_reiser4_super(sbinfo); + sbinfo->blocks_free_committed += blocks_freed; + spin_unlock_reiser4_super(sbinfo); + } + + return 0; +} + +/* plugin->u.space_allocator.init_allocator + constructor of reiser4_space_allocator object. It is called on fs mount */ +int reiser4_init_allocator_bitmap(reiser4_space_allocator * allocator, + struct super_block *super, void *arg) +{ + struct bitmap_allocator_data *data = NULL; + bmap_nr_t bitmap_blocks_nr; + bmap_nr_t i; + + assert("nikita-3039", reiser4_schedulable()); + + /* getting memory for bitmap allocator private data holder */ + data = + kmalloc(sizeof(struct bitmap_allocator_data), + reiser4_ctx_gfp_mask_get()); + + if (data == NULL) + return RETERR(-ENOMEM); + + /* allocation and initialization for the array of bnodes */ + bitmap_blocks_nr = get_nr_bmap(super); + + /* FIXME-ZAM: it is not clear what to do with huge number of bitmaps + which is bigger than 2^32 (= 8 * 4096 * 4096 * 2^32 bytes = 5.76e+17, + may I never meet someone who still uses the ia32 architecture when + storage devices of that size enter the market, and wants to use ia32 + with that storage device, much less reiser4. ;-) -Hans). Kmalloc is not possible and, + probably, another dynamic data structure should replace a static + array of bnodes. */ + /*data->bitmap = reiser4_kmalloc((size_t) (sizeof (struct bitmap_node) * bitmap_blocks_nr), GFP_KERNEL); */ + data->bitmap = reiser4_vmalloc(sizeof(struct bitmap_node) * bitmap_blocks_nr); + if (data->bitmap == NULL) { + kfree(data); + return RETERR(-ENOMEM); + } + + for (i = 0; i < bitmap_blocks_nr; i++) + init_bnode(data->bitmap + i, super, i); + + allocator->u.generic = data; + +#if REISER4_DEBUG + get_super_private(super)->min_blocks_used += bitmap_blocks_nr; +#endif + + /* Load all bitmap blocks at mount time. */ + if (!test_bit + (REISER4_DONT_LOAD_BITMAP, &get_super_private(super)->fs_flags)) { + __u64 start_time, elapsed_time; + struct bitmap_node *bnode; + int ret; + + if (REISER4_DEBUG) + printk(KERN_INFO "loading reiser4 bitmap..."); + start_time = jiffies; + + for (i = 0; i < bitmap_blocks_nr; i++) { + bnode = data->bitmap + i; + ret = load_and_lock_bnode(bnode); + if (ret) { + reiser4_destroy_allocator_bitmap(allocator, + super); + return ret; + } + release_and_unlock_bnode(bnode); + } + + elapsed_time = jiffies - start_time; + if (REISER4_DEBUG) + printk("...done (%llu jiffies)\n", + (unsigned long long)elapsed_time); + } + + return 0; +} + +/* plugin->u.space_allocator.destroy_allocator + destructor. It is called on fs unmount */ +int reiser4_destroy_allocator_bitmap(reiser4_space_allocator * allocator, + struct super_block *super) +{ + bmap_nr_t bitmap_blocks_nr; + bmap_nr_t i; + + struct bitmap_allocator_data *data = allocator->u.generic; + + assert("zam-414", data != NULL); + assert("zam-376", data->bitmap != NULL); + + bitmap_blocks_nr = get_nr_bmap(super); + + for (i = 0; i < bitmap_blocks_nr; i++) { + struct bitmap_node *bnode = data->bitmap + i; + + mutex_lock(&bnode->mutex); + +#if REISER4_DEBUG + if (atomic_read(&bnode->loaded)) { + jnode *wj = bnode->wjnode; + jnode *cj = bnode->cjnode; + + assert("zam-480", jnode_page(cj) != NULL); + assert("zam-633", jnode_page(wj) != NULL); + + assert("zam-634", + memcmp(jdata(wj), jdata(wj), + bmap_size(super->s_blocksize)) == 0); + + } +#endif + done_bnode(bnode); + mutex_unlock(&bnode->mutex); + } + + vfree(data->bitmap); + kfree(data); + + allocator->u.generic = NULL; + + return 0; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * scroll-step: 1 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/space/bitmap.h linux-2.6.30/fs/reiser4/plugin/space/bitmap.h --- linux-2.6.30.orig/fs/reiser4/plugin/space/bitmap.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/space/bitmap.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,47 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#if !defined (__REISER4_PLUGIN_SPACE_BITMAP_H__) +#define __REISER4_PLUGIN_SPACE_BITMAP_H__ + +#include "../../dformat.h" +#include "../../block_alloc.h" + +#include /* for __u?? */ +#include /* for struct super_block */ +/* EDWARD-FIXME-HANS: write something as informative as the below for every .h file lacking it. */ +/* declarations of functions implementing methods of space allocator plugin for + bitmap based allocator. The functions themselves are in bitmap.c */ +extern int reiser4_init_allocator_bitmap(reiser4_space_allocator *, + struct super_block *, void *); +extern int reiser4_destroy_allocator_bitmap(reiser4_space_allocator *, + struct super_block *); +extern int reiser4_alloc_blocks_bitmap(reiser4_space_allocator *, + reiser4_blocknr_hint *, int needed, + reiser4_block_nr * start, + reiser4_block_nr * len); +extern void reiser4_check_blocks_bitmap(const reiser4_block_nr *, + const reiser4_block_nr *, int); +extern void reiser4_dealloc_blocks_bitmap(reiser4_space_allocator *, + reiser4_block_nr, + reiser4_block_nr); +extern int reiser4_pre_commit_hook_bitmap(void); + +#define reiser4_post_commit_hook_bitmap() do{}while(0) +#define reiser4_post_write_back_hook_bitmap() do{}while(0) +#define reiser4_print_info_bitmap(pref, al) do{}while(0) + +typedef __u64 bmap_nr_t; +typedef __u32 bmap_off_t; + +#endif /* __REISER4_PLUGIN_SPACE_BITMAP_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/space/Makefile linux-2.6.30/fs/reiser4/plugin/space/Makefile --- linux-2.6.30.orig/fs/reiser4/plugin/space/Makefile 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/space/Makefile 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,4 @@ +obj-$(CONFIG_REISER4_FS) += space_plugins.o + +space_plugins-objs := \ + bitmap.o diff -urN linux-2.6.30.orig/fs/reiser4/plugin/space/space_allocator.h linux-2.6.30/fs/reiser4/plugin/space/space_allocator.h --- linux-2.6.30.orig/fs/reiser4/plugin/space/space_allocator.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/space/space_allocator.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,80 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#ifndef __SPACE_ALLOCATOR_H__ +#define __SPACE_ALLOCATOR_H__ + +#include "../../forward.h" +#include "bitmap.h" +/* NIKITA-FIXME-HANS: surely this could use a comment. Something about how bitmap is the only space allocator for now, + * but... */ +#define DEF_SPACE_ALLOCATOR(allocator) \ + \ +static inline int sa_init_allocator (reiser4_space_allocator * al, struct super_block *s, void * opaque) \ +{ \ + return reiser4_init_allocator_##allocator (al, s, opaque); \ +} \ + \ +static inline void sa_destroy_allocator (reiser4_space_allocator *al, struct super_block *s) \ +{ \ + reiser4_destroy_allocator_##allocator (al, s); \ +} \ + \ +static inline int sa_alloc_blocks (reiser4_space_allocator *al, reiser4_blocknr_hint * hint, \ + int needed, reiser4_block_nr * start, reiser4_block_nr * len) \ +{ \ + return reiser4_alloc_blocks_##allocator (al, hint, needed, start, len); \ +} \ +static inline void sa_dealloc_blocks (reiser4_space_allocator * al, reiser4_block_nr start, reiser4_block_nr len) \ +{ \ + reiser4_dealloc_blocks_##allocator (al, start, len); \ +} \ + \ +static inline void sa_check_blocks (const reiser4_block_nr * start, const reiser4_block_nr * end, int desired) \ +{ \ + reiser4_check_blocks_##allocator (start, end, desired); \ +} \ + \ +static inline void sa_pre_commit_hook (void) \ +{ \ + reiser4_pre_commit_hook_##allocator (); \ +} \ + \ +static inline void sa_post_commit_hook (void) \ +{ \ + reiser4_post_commit_hook_##allocator (); \ +} \ + \ +static inline void sa_post_write_back_hook (void) \ +{ \ + reiser4_post_write_back_hook_##allocator(); \ +} \ + \ +static inline void sa_print_info(const char * prefix, reiser4_space_allocator * al) \ +{ \ + reiser4_print_info_##allocator (prefix, al); \ +} + +DEF_SPACE_ALLOCATOR(bitmap) + +/* this object is part of reiser4 private in-core super block */ +struct reiser4_space_allocator { + union { + /* space allocators might use this pointer to reference their + * data. */ + void *generic; + } u; +}; + +/* __SPACE_ALLOCATOR_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/plugin/tail_policy.c linux-2.6.30/fs/reiser4/plugin/tail_policy.c --- linux-2.6.30.orig/fs/reiser4/plugin/tail_policy.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/plugin/tail_policy.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,113 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Formatting policy plugins */ + +/* + * Formatting policy plugin is used by object plugin (of regular file) to + * convert file between two representations. + * + * Currently following policies are implemented: + * never store file in formatted nodes + * always store file in formatted nodes + * store file in formatted nodes if file is smaller than 4 blocks (default) + */ + +#include "../tree.h" +#include "../inode.h" +#include "../super.h" +#include "object.h" +#include "plugin.h" +#include "node/node.h" +#include "plugin_header.h" + +#include +#include /* For struct inode */ + +/** + * have_formatting_never - + * @inode: + * @size: + * + * + */ +/* Never store file's tail as direct item */ +/* Audited by: green(2002.06.12) */ +static int have_formatting_never(const struct inode *inode UNUSED_ARG + /* inode to operate on */ , + loff_t size UNUSED_ARG/* new object size */) +{ + return 0; +} + +/* Always store file's tail as direct item */ +/* Audited by: green(2002.06.12) */ +static int +have_formatting_always(const struct inode *inode UNUSED_ARG + /* inode to operate on */ , + loff_t size UNUSED_ARG/* new object size */) +{ + return 1; +} + +/* This function makes test if we should store file denoted @inode as tails only + or as extents only. */ +static int +have_formatting_default(const struct inode *inode UNUSED_ARG + /* inode to operate on */ , + loff_t size/* new object size */) +{ + assert("umka-1253", inode != NULL); + + if (size > inode->i_sb->s_blocksize * 4) + return 0; + + return 1; +} + +/* tail plugins */ +formatting_plugin formatting_plugins[LAST_TAIL_FORMATTING_ID] = { + [NEVER_TAILS_FORMATTING_ID] = { + .h = { + .type_id = REISER4_FORMATTING_PLUGIN_TYPE, + .id = NEVER_TAILS_FORMATTING_ID, + .pops = NULL, + .label = "never", + .desc = "Never store file's tail", + .linkage = {NULL, NULL} + }, + .have_tail = have_formatting_never + }, + [ALWAYS_TAILS_FORMATTING_ID] = { + .h = { + .type_id = REISER4_FORMATTING_PLUGIN_TYPE, + .id = ALWAYS_TAILS_FORMATTING_ID, + .pops = NULL, + .label = "always", + .desc = "Always store file's tail", + .linkage = {NULL, NULL} + }, + .have_tail = have_formatting_always + }, + [SMALL_FILE_FORMATTING_ID] = { + .h = { + .type_id = REISER4_FORMATTING_PLUGIN_TYPE, + .id = SMALL_FILE_FORMATTING_ID, + .pops = NULL, + .label = "4blocks", + .desc = "store files shorter than 4 blocks in tail items", + .linkage = {NULL, NULL} + }, + .have_tail = have_formatting_default + } +}; + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/pool.c linux-2.6.30/fs/reiser4/pool.c --- linux-2.6.30.orig/fs/reiser4/pool.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/pool.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,231 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Fast pool allocation. + + There are situations when some sub-system normally asks memory allocator + for only few objects, but under some circumstances could require much + more. Typical and actually motivating example is tree balancing. It needs + to keep track of nodes that were involved into it, and it is well-known + that in reasonable packed balanced tree most (92.938121%) percent of all + balancings end up after working with only few nodes (3.141592 on + average). But in rare cases balancing can involve much more nodes + (3*tree_height+1 in extremal situation). + + On the one hand, we don't want to resort to dynamic allocation (slab, + malloc(), etc.) to allocate data structures required to keep track of + nodes during balancing. On the other hand, we cannot statically allocate + required amount of space on the stack, because first: it is useless wastage + of precious resource, and second: this amount is unknown in advance (tree + height can change). + + Pools, implemented in this file are solution for this problem: + + - some configurable amount of objects is statically preallocated on the + stack + + - if this preallocated pool is exhausted and more objects is requested + they are allocated dynamically. + + Pools encapsulate distinction between statically and dynamically allocated + objects. Both allocation and recycling look exactly the same. + + To keep track of dynamically allocated objects, pool adds its own linkage + to each object. + + NOTE-NIKITA This linkage also contains some balancing-specific data. This + is not perfect. On the other hand, balancing is currently the only client + of pool code. + + NOTE-NIKITA Another desirable feature is to rewrite all pool manipulation + functions in the style of tslist/tshash, i.e., make them unreadable, but + type-safe. + +*/ + +#include "debug.h" +#include "pool.h" +#include "super.h" + +#include +#include + +/* initialize new pool object @h */ +static void reiser4_init_pool_obj(struct reiser4_pool_header *h) +{ + INIT_LIST_HEAD(&h->usage_linkage); + INIT_LIST_HEAD(&h->level_linkage); + INIT_LIST_HEAD(&h->extra_linkage); +} + +/* initialize new pool */ +void reiser4_init_pool(struct reiser4_pool *pool /* pool to initialize */ , + size_t obj_size /* size of objects in @pool */ , + int num_of_objs /* number of preallocated objects */ , + char *data/* area for preallocated objects */) +{ + struct reiser4_pool_header *h; + int i; + + assert("nikita-955", pool != NULL); + assert("nikita-1044", obj_size > 0); + assert("nikita-956", num_of_objs >= 0); + assert("nikita-957", data != NULL); + + memset(pool, 0, sizeof *pool); + pool->obj_size = obj_size; + pool->data = data; + INIT_LIST_HEAD(&pool->free); + INIT_LIST_HEAD(&pool->used); + INIT_LIST_HEAD(&pool->extra); + memset(data, 0, obj_size * num_of_objs); + for (i = 0; i < num_of_objs; ++i) { + h = (struct reiser4_pool_header *) (data + i * obj_size); + reiser4_init_pool_obj(h); + /* add pool header to the end of pool's free list */ + list_add_tail(&h->usage_linkage, &pool->free); + } +} + +/* release pool resources + + Release all resources acquired by this pool, specifically, dynamically + allocated objects. + +*/ +void reiser4_done_pool(struct reiser4_pool *pool UNUSED_ARG) +{ +} + +/* allocate carry object from @pool + + First, try to get preallocated object. If this fails, resort to dynamic + allocation. + +*/ +static void *reiser4_pool_alloc(struct reiser4_pool *pool) +{ + struct reiser4_pool_header *result; + + assert("nikita-959", pool != NULL); + + if (!list_empty(&pool->free)) { + struct list_head *linkage; + + linkage = pool->free.next; + list_del(linkage); + INIT_LIST_HEAD(linkage); + result = list_entry(linkage, struct reiser4_pool_header, + usage_linkage); + BUG_ON(!list_empty(&result->level_linkage) || + !list_empty(&result->extra_linkage)); + } else { + /* pool is empty. Extra allocations don't deserve dedicated + slab to be served from, as they are expected to be rare. */ + result = kmalloc(pool->obj_size, reiser4_ctx_gfp_mask_get()); + if (result != 0) { + reiser4_init_pool_obj(result); + list_add(&result->extra_linkage, &pool->extra); + } else + return ERR_PTR(RETERR(-ENOMEM)); + BUG_ON(!list_empty(&result->usage_linkage) || + !list_empty(&result->level_linkage)); + } + ++pool->objs; + list_add(&result->usage_linkage, &pool->used); + memset(result + 1, 0, pool->obj_size - sizeof *result); + return result; +} + +/* return object back to the pool */ +void reiser4_pool_free(struct reiser4_pool *pool, + struct reiser4_pool_header *h) +{ + assert("nikita-961", h != NULL); + assert("nikita-962", pool != NULL); + + --pool->objs; + assert("nikita-963", pool->objs >= 0); + + list_del_init(&h->usage_linkage); + list_del_init(&h->level_linkage); + + if (list_empty(&h->extra_linkage)) + /* + * pool header is not an extra one. Push it onto free list + * using usage_linkage + */ + list_add(&h->usage_linkage, &pool->free); + else { + /* remove pool header from pool's extra list and kfree it */ + list_del(&h->extra_linkage); + kfree(h); + } +} + +/* add new object to the carry level list + + Carry level is FIFO most of the time, but not always. Complications arise + when make_space() function tries to go to the left neighbor and thus adds + carry node before existing nodes, and also, when updating delimiting keys + after moving data between two nodes, we want left node to be locked before + right node. + + Latter case is confusing at the first glance. Problem is that COP_UPDATE + opration that updates delimiting keys is sometimes called with two nodes + (when data are moved between two nodes) and sometimes with only one node + (when leftmost item is deleted in a node). In any case operation is + supplied with at least node whose left delimiting key is to be updated + (that is "right" node). + + @pool - from which to allocate new object; + @list - where to add object; + @reference - after (or before) which existing object to add +*/ +struct reiser4_pool_header *reiser4_add_obj(struct reiser4_pool *pool, + struct list_head *list, + pool_ordering order, + struct reiser4_pool_header *reference) +{ + struct reiser4_pool_header *result; + + assert("nikita-972", pool != NULL); + + result = reiser4_pool_alloc(pool); + if (IS_ERR(result)) + return result; + + assert("nikita-973", result != NULL); + + switch (order) { + case POOLO_BEFORE: + __list_add(&result->level_linkage, + reference->level_linkage.prev, + &reference->level_linkage); + break; + case POOLO_AFTER: + __list_add(&result->level_linkage, + &reference->level_linkage, + reference->level_linkage.next); + break; + case POOLO_LAST: + list_add_tail(&result->level_linkage, list); + break; + case POOLO_FIRST: + list_add(&result->level_linkage, list); + break; + default: + wrong_return_value("nikita-927", "order"); + } + return result; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/pool.h linux-2.6.30/fs/reiser4/pool.h --- linux-2.6.30.orig/fs/reiser4/pool.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/pool.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,57 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Fast pool allocation */ + +#ifndef __REISER4_POOL_H__ +#define __REISER4_POOL_H__ + +#include + +struct reiser4_pool { + size_t obj_size; + int objs; + char *data; + struct list_head free; + struct list_head used; + struct list_head extra; +}; + +struct reiser4_pool_header { + /* object is either on free or "used" lists */ + struct list_head usage_linkage; + struct list_head level_linkage; + struct list_head extra_linkage; +}; + +typedef enum { + POOLO_BEFORE, + POOLO_AFTER, + POOLO_LAST, + POOLO_FIRST +} pool_ordering; + +/* pool manipulation functions */ + +extern void reiser4_init_pool(struct reiser4_pool *pool, size_t obj_size, + int num_of_objs, char *data); +extern void reiser4_done_pool(struct reiser4_pool *pool); +extern void reiser4_pool_free(struct reiser4_pool *pool, + struct reiser4_pool_header *h); +struct reiser4_pool_header *reiser4_add_obj(struct reiser4_pool *pool, + struct list_head *list, + pool_ordering order, + struct reiser4_pool_header *reference); + +/* __REISER4_POOL_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/readahead.c linux-2.6.30/fs/reiser4/readahead.c --- linux-2.6.30.orig/fs/reiser4/readahead.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/readahead.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,140 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "forward.h" +#include "tree.h" +#include "tree_walk.h" +#include "super.h" +#include "inode.h" +#include "key.h" +#include "znode.h" + +#include /* for totalram_pages */ + +void reiser4_init_ra_info(ra_info_t *rai) +{ + rai->key_to_stop = *reiser4_min_key(); +} + +/* global formatted node readahead parameter. It can be set by mount option + * -o readahead:NUM:1 */ +static inline int ra_adjacent_only(int flags) +{ + return flags & RA_ADJACENT_ONLY; +} + +/* this is used by formatted_readahead to decide whether read for right neighbor + * of node is to be issued. It returns 1 if right neighbor's first key is less + * or equal to readahead's stop key */ +static int should_readahead_neighbor(znode * node, ra_info_t *info) +{ + int result; + + read_lock_dk(znode_get_tree(node)); + result = keyle(znode_get_rd_key(node), &info->key_to_stop); + read_unlock_dk(znode_get_tree(node)); + return result; +} + +#define LOW_MEM_PERCENTAGE (5) + +static int low_on_memory(void) +{ + unsigned int freepages; + + freepages = nr_free_pages(); + return freepages < (totalram_pages * LOW_MEM_PERCENTAGE / 100); +} + +/* start read for @node and for a few of its right neighbors */ +void formatted_readahead(znode * node, ra_info_t *info) +{ + struct formatted_ra_params *ra_params; + znode *cur; + int i; + int grn_flags; + lock_handle next_lh; + + /* do nothing if node block number has not been assigned to node (which + * means it is still in cache). */ + if (reiser4_blocknr_is_fake(znode_get_block(node))) + return; + + ra_params = get_current_super_ra_params(); + + if (znode_page(node) == NULL) + jstartio(ZJNODE(node)); + + if (znode_get_level(node) != LEAF_LEVEL) + return; + + /* don't waste memory for read-ahead when low on memory */ + if (low_on_memory()) + return; + + /* We can have locked nodes on upper tree levels, in this situation lock + priorities do not help to resolve deadlocks, we have to use TRY_LOCK + here. */ + grn_flags = (GN_CAN_USE_UPPER_LEVELS | GN_TRY_LOCK); + + i = 0; + cur = zref(node); + init_lh(&next_lh); + while (i < ra_params->max) { + const reiser4_block_nr * nextblk; + + if (!should_readahead_neighbor(cur, info)) + break; + + if (reiser4_get_right_neighbor + (&next_lh, cur, ZNODE_READ_LOCK, grn_flags)) + break; + + nextblk = znode_get_block(next_lh.node); + if (reiser4_blocknr_is_fake(nextblk) || + (ra_adjacent_only(ra_params->flags) + && *nextblk != *znode_get_block(cur) + 1)) + break; + + zput(cur); + cur = zref(next_lh.node); + done_lh(&next_lh); + if (znode_page(cur) == NULL) + jstartio(ZJNODE(cur)); + else + /* Do not scan read-ahead window if pages already + * allocated (and i/o already started). */ + break; + + i++; + } + zput(cur); + done_lh(&next_lh); +} + +void reiser4_readdir_readahead_init(struct inode *dir, tap_t *tap) +{ + reiser4_key *stop_key; + + assert("nikita-3542", dir != NULL); + assert("nikita-3543", tap != NULL); + + stop_key = &tap->ra_info.key_to_stop; + /* initialize readdir readahead information: include into readahead + * stat data of all files of the directory */ + set_key_locality(stop_key, get_inode_oid(dir)); + set_key_type(stop_key, KEY_SD_MINOR); + set_key_ordering(stop_key, get_key_ordering(reiser4_max_key())); + set_key_objectid(stop_key, get_key_objectid(reiser4_max_key())); + set_key_offset(stop_key, get_key_offset(reiser4_max_key())); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/readahead.h linux-2.6.30/fs/reiser4/readahead.h --- linux-2.6.30.orig/fs/reiser4/readahead.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/readahead.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,52 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +#ifndef __READAHEAD_H__ +#define __READAHEAD_H__ + +#include "key.h" + +typedef enum { + RA_ADJACENT_ONLY = 1, /* only requests nodes which are adjacent. + Default is NO (not only adjacent) */ +} ra_global_flags; + +/* reiser4 super block has a field of this type. + It controls readahead during tree traversals */ +struct formatted_ra_params { + unsigned long max; /* request not more than this amount of nodes. + Default is totalram_pages / 4 */ + int flags; +}; + +typedef struct { + reiser4_key key_to_stop; +} ra_info_t; + +void formatted_readahead(znode * , ra_info_t *); +void reiser4_init_ra_info(ra_info_t *rai); + +struct reiser4_file_ra_state { + loff_t start; /* Current window */ + loff_t size; + loff_t next_size; /* Next window size */ + loff_t ahead_start; /* Ahead window */ + loff_t ahead_size; + loff_t max_window_size; /* Maximum readahead window */ + loff_t slow_start; /* enlarging r/a size algorithm. */ +}; + +extern void reiser4_readdir_readahead_init(struct inode *dir, tap_t *tap); + +/* __READAHEAD_H__ */ +#endif + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/README linux-2.6.30/fs/reiser4/README --- linux-2.6.30.orig/fs/reiser4/README 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/README 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,128 @@ +[LICENSING] + +Reiser4 is hereby licensed under the GNU General +Public License version 2. + +Source code files that contain the phrase "licensing governed by +reiser4/README" are "governed files" throughout this file. Governed +files are licensed under the GPL. The portions of them owned by Hans +Reiser, or authorized to be licensed by him, have been in the past, +and likely will be in the future, licensed to other parties under +other licenses. If you add your code to governed files, and don't +want it to be owned by Hans Reiser, put your copyright label on that +code so the poor blight and his customers can keep things straight. +All portions of governed files not labeled otherwise are owned by Hans +Reiser, and by adding your code to it, widely distributing it to +others or sending us a patch, and leaving the sentence in stating that +licensing is governed by the statement in this file, you accept this. +It will be a kindness if you identify whether Hans Reiser is allowed +to license code labeled as owned by you on your behalf other than +under the GPL, because he wants to know if it is okay to do so and put +a check in the mail to you (for non-trivial improvements) when he +makes his next sale. He makes no guarantees as to the amount if any, +though he feels motivated to motivate contributors, and you can surely +discuss this with him before or after contributing. You have the +right to decline to allow him to license your code contribution other +than under the GPL. + +Further licensing options are available for commercial and/or other +interests directly from Hans Reiser: reiser@namesys.com. If you interpret +the GPL as not allowing those additional licensing options, you read +it wrongly, and Richard Stallman agrees with me, when carefully read +you can see that those restrictions on additional terms do not apply +to the owner of the copyright, and my interpretation of this shall +govern for this license. + +[END LICENSING] + +Reiser4 is a file system based on dancing tree algorithms, and is +described at http://www.namesys.com + +mkfs.reiser4 and other utilities are on our webpage or wherever your +Linux provider put them. You really want to be running the latest +version off the website if you use fsck. + +Yes, if you update your reiser4 kernel module you do have to +recompile your kernel, most of the time. The errors you get will be +quite cryptic if your forget to do so. + +Hideous Commercial Pitch: Spread your development costs across other OS +vendors. Select from the best in the world, not the best in your +building, by buying from third party OS component suppliers. Leverage +the software component development power of the internet. Be the most +aggressive in taking advantage of the commercial possibilities of +decentralized internet development, and add value through your branded +integration that you sell as an operating system. Let your competitors +be the ones to compete against the entire internet by themselves. Be +hip, get with the new economic trend, before your competitors do. Send +email to reiser@namesys.com + +Hans Reiser was the primary architect of Reiser4, but a whole team +chipped their ideas in. He invested everything he had into Namesys +for 5.5 dark years of no money before Reiser3 finally started to work well +enough to bring in money. He owns the copyright. + +DARPA was the primary sponsor of Reiser4. DARPA does not endorse +Reiser4, it merely sponsors it. DARPA is, in solely Hans's personal +opinion, unique in its willingness to invest into things more +theoretical than the VC community can readily understand, and more +longterm than allows them to be sure that they will be the ones to +extract the economic benefits from. DARPA also integrated us into a +security community that transformed our security worldview. + +Vladimir Saveliev is our lead programmer, with us from the beginning, +and he worked long hours writing the cleanest code. This is why he is +now the lead programmer after years of commitment to our work. He +always made the effort to be the best he could be, and to make his +code the best that it could be. What resulted was quite remarkable. I +don't think that money can ever motivate someone to work the way he +did, he is one of the most selfless men I know. + +Alexander Lyamin was our sysadmin, and helped to educate us in +security issues. Moscow State University and IMT were very generous +in the internet access they provided us, and in lots of other little +ways that a generous institution can be. + +Alexander Zarochentcev (sometimes known as zam, or sasha), wrote the +locking code, the block allocator, and finished the flushing code. +His code is always crystal clean and well structured. + +Nikita Danilov wrote the core of the balancing code, the core of the +plugins code, and the directory code. He worked a steady pace of long +hours that produced a whole lot of well abstracted code. He is our +senior computer scientist. + +Vladimir Demidov wrote the parser. Writing an in kernel parser is +something very few persons have the skills for, and it is thanks to +him that we can say that the parser is really not so big compared to +various bits of our other code, and making a parser work in the kernel +was not so complicated as everyone would imagine mainly because it was +him doing it... + +Joshua McDonald wrote the transaction manager, and the flush code. +The flush code unexpectedly turned out be extremely hairy for reasons +you can read about on our web page, and he did a great job on an +extremely difficult task. + +Nina Reiser handled our accounting, government relations, and much +more. + +Ramon Reiser developed our website. + +Beverly Palmer drew our graphics. + +Vitaly Fertman developed librepair, userspace plugins repair code, fsck +and worked with Umka on developing libreiser4 and userspace plugins. + +Yury Umanets (aka Umka) developed libreiser4, userspace plugins and +userspace tools (reiser4progs). + +Oleg Drokin (aka Green) is the release manager who fixes everything. +It is so nice to have someone like that on the team. He (plus Chris +and Jeff) make it possible for the entire rest of the Namesys team to +focus on Reiser4, and he fixed a whole lot of Reiser4 bugs also. It +is just amazing to watch his talent for spotting bugs in action. + +Edward Shishkin wrote cryptcompress file plugin (which manages files +built of encrypted and(or) compressed bodies) and other plugins related +to transparent encryption and compression support. diff -urN linux-2.6.30.orig/fs/reiser4/reiser4.h linux-2.6.30/fs/reiser4/reiser4.h --- linux-2.6.30.orig/fs/reiser4/reiser4.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/reiser4.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,259 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* definitions of common constants used by reiser4 */ + +#if !defined( __REISER4_H__ ) +#define __REISER4_H__ + +#include /* for HZ */ +#include +#include +#include +#include +#include + +/* + * reiser4 compilation options. + */ + +#if defined(CONFIG_REISER4_DEBUG) +/* turn on assertion checks */ +#define REISER4_DEBUG (1) +#else +#define REISER4_DEBUG (0) +#endif + +#define REISER4_SHA256 (0) + +/* + * Turn on large keys mode. In his mode (which is default), reiser4 key has 4 + * 8-byte components. In the old "small key" mode, it's 3 8-byte + * components. Additional component, referred to as "ordering" is used to + * order items from which given object is composed of. As such, ordering is + * placed between locality and objectid. For directory item ordering contains + * initial prefix of the file name this item is for. This sorts all directory + * items within given directory lexicographically (but see + * fibration.[ch]). For file body and stat-data, ordering contains initial + * prefix of the name file was initially created with. In the common case + * (files with single name) this allows to order file bodies and stat-datas in + * the same order as their respective directory entries, thus speeding up + * readdir. + * + * Note, that kernel can only mount file system with the same key size as one + * it is compiled for, so flipping this option may render your data + * inaccessible. + */ +#define REISER4_LARGE_KEY (1) +/*#define REISER4_LARGE_KEY (0)*/ + +/*#define GUESS_EXISTS 1*/ + +/* + * PLEASE update fs/reiser4/kattr.c:show_options() when adding new compilation + * option + */ + +extern const char *REISER4_SUPER_MAGIC_STRING; +extern const int REISER4_MAGIC_OFFSET; /* offset to magic string from the + * beginning of device */ + +/* here go tunable parameters that are not worth special entry in kernel + configuration */ + +/* default number of slots in coord-by-key caches */ +#define CBK_CACHE_SLOTS (16) +/* how many elementary tree operation to carry on the next level */ +#define CARRIES_POOL_SIZE (5) +/* size of pool of preallocated nodes for carry process. */ +#define NODES_LOCKED_POOL_SIZE (5) + +#define REISER4_NEW_NODE_FLAGS (COPI_LOAD_LEFT | COPI_LOAD_RIGHT | COPI_GO_LEFT) +#define REISER4_NEW_EXTENT_FLAGS (COPI_LOAD_LEFT | COPI_LOAD_RIGHT | COPI_GO_LEFT) +#define REISER4_PASTE_FLAGS (COPI_GO_LEFT) +#define REISER4_INSERT_FLAGS (COPI_GO_LEFT) + +/* we are supporting reservation of disk space on uid basis */ +#define REISER4_SUPPORT_UID_SPACE_RESERVATION (0) +/* we are supporting reservation of disk space for groups */ +#define REISER4_SUPPORT_GID_SPACE_RESERVATION (0) +/* we are supporting reservation of disk space for root */ +#define REISER4_SUPPORT_ROOT_SPACE_RESERVATION (0) +/* we use rapid flush mode, see flush.c for comments. */ +#define REISER4_USE_RAPID_FLUSH (1) + +/* + * set this to 0 if you don't want to use wait-for-flush in ->writepage(). + */ +#define REISER4_USE_ENTD (1) + +/* key allocation is Plan-A */ +#define REISER4_PLANA_KEY_ALLOCATION (1) +/* key allocation follows good old 3.x scheme */ +#define REISER4_3_5_KEY_ALLOCATION (0) + +/* size of hash-table for znodes */ +#define REISER4_ZNODE_HASH_TABLE_SIZE (1 << 13) + +/* number of buckets in lnode hash-table */ +#define LNODE_HTABLE_BUCKETS (1024) + +/* some ridiculously high maximal limit on height of znode tree. This + is used in declaration of various per level arrays and + to allocate stattistics gathering array for per-level stats. */ +#define REISER4_MAX_ZTREE_HEIGHT (8) + +#define REISER4_PANIC_MSG_BUFFER_SIZE (1024) + +/* If array contains less than REISER4_SEQ_SEARCH_BREAK elements then, + sequential search is on average faster than binary. This is because + of better optimization and because sequential search is more CPU + cache friendly. This number (25) was found by experiments on dual AMD + Athlon(tm), 1400MHz. + + NOTE: testing in kernel has shown that binary search is more effective than + implied by results of the user level benchmarking. Probably because in the + node keys are separated by other data. So value was adjusted after few + tests. More thorough tuning is needed. +*/ +#define REISER4_SEQ_SEARCH_BREAK (3) + +/* don't allow tree to be lower than this */ +#define REISER4_MIN_TREE_HEIGHT (TWIG_LEVEL) + +/* NOTE NIKITA this is no longer used: maximal atom size is auto-adjusted to + * available memory. */ +/* Default value of maximal atom size. Can be ovewritten by + tmgr.atom_max_size mount option. By default infinity. */ +#define REISER4_ATOM_MAX_SIZE ((unsigned)(~0)) + +/* Default value of maximal atom age (in jiffies). After reaching this age + atom will be forced to commit, either synchronously or asynchronously. Can + be overwritten by tmgr.atom_max_age mount option. */ +#define REISER4_ATOM_MAX_AGE (600 * HZ) + +/* sleeping period for ktxnmrgd */ +#define REISER4_TXNMGR_TIMEOUT (5 * HZ) + +/* timeout to wait for ent thread in writepage. Default: 3 milliseconds. */ +#define REISER4_ENTD_TIMEOUT (3 * HZ / 1000) + +/* start complaining after that many restarts in coord_by_key(). + + This either means incredibly heavy contention for this part of a tree, or + some corruption or bug. +*/ +#define REISER4_CBK_ITERATIONS_LIMIT (100) + +/* return -EIO after that many iterations in coord_by_key(). + + I have witnessed more than 800 iterations (in 30 thread test) before cbk + finished. --nikita +*/ +#define REISER4_MAX_CBK_ITERATIONS 500000 + +/* put a per-inode limit on maximal number of directory entries with identical + keys in hashed directory. + + Disable this until inheritance interfaces stabilize: we need some way to + set per directory limit. +*/ +#define REISER4_USE_COLLISION_LIMIT (0) + +/* If flush finds more than FLUSH_RELOCATE_THRESHOLD adjacent dirty leaf-level + blocks it will force them to be relocated. */ +#define FLUSH_RELOCATE_THRESHOLD 64 +/* If flush finds can find a block allocation closer than at most + FLUSH_RELOCATE_DISTANCE from the preceder it will relocate to that position. + */ +#define FLUSH_RELOCATE_DISTANCE 64 + +/* If we have written this much or more blocks before encountering busy jnode + in flush list - abort flushing hoping that next time we get called + this jnode will be clean already, and we will save some seeks. */ +#define FLUSH_WRITTEN_THRESHOLD 50 + +/* The maximum number of nodes to scan left on a level during flush. */ +#define FLUSH_SCAN_MAXNODES 10000 + +/* per-atom limit of flushers */ +#define ATOM_MAX_FLUSHERS (1) + +/* default tracing buffer size */ +#define REISER4_TRACE_BUF_SIZE (1 << 15) + +/* what size units of IO we would like cp, etc., to use, in writing to + reiser4. In bytes. + + Can be overwritten by optimal_io_size mount option. +*/ +#define REISER4_OPTIMAL_IO_SIZE (64 * 1024) + +/* see comments in inode.c:oid_to_uino() */ +#define REISER4_UINO_SHIFT (1 << 30) + +/* Mark function argument as unused to avoid compiler warnings. */ +#define UNUSED_ARG __attribute__((unused)) + +#if ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 3)) || (__GNUC__ > 3) +#define NONNULL __attribute__((nonnull)) +#else +#define NONNULL +#endif + +/* master super block offset in bytes.*/ +#define REISER4_MASTER_OFFSET 65536 + +/* size of VFS block */ +#define VFS_BLKSIZE 512 +/* number of bits in size of VFS block (512==2^9) */ +#define VFS_BLKSIZE_BITS 9 + +#define REISER4_I reiser4_inode_data + +/* implication */ +#define ergo(antecedent, consequent) (!(antecedent) || (consequent)) +/* logical equivalence */ +#define equi(p1, p2) (ergo((p1), (p2)) && ergo((p2), (p1))) + +#define sizeof_array(x) ((int) (sizeof(x) / sizeof(x[0]))) + +#define NOT_YET (0) + +/** Reiser4 specific error codes **/ + +#define REISER4_ERROR_CODE_BASE 10000 + +/* Neighbor is not available (side neighbor or parent) */ +#define E_NO_NEIGHBOR (REISER4_ERROR_CODE_BASE) + +/* Node was not found in cache */ +#define E_NOT_IN_CACHE (REISER4_ERROR_CODE_BASE + 1) + +/* node has no free space enough for completion of balancing operation */ +#define E_NODE_FULL (REISER4_ERROR_CODE_BASE + 2) + +/* repeat operation */ +#define E_REPEAT (REISER4_ERROR_CODE_BASE + 3) + +/* deadlock happens */ +#define E_DEADLOCK (REISER4_ERROR_CODE_BASE + 4) + +/* operation cannot be performed, because it would block and non-blocking mode + * was requested. */ +#define E_BLOCK (REISER4_ERROR_CODE_BASE + 5) + +/* wait some event (depends on context), then repeat */ +#define E_WAIT (REISER4_ERROR_CODE_BASE + 6) + +#endif /* __REISER4_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/safe_link.c linux-2.6.30/fs/reiser4/safe_link.c --- linux-2.6.30.orig/fs/reiser4/safe_link.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/safe_link.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,354 @@ +/* Copyright 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Safe-links. */ + +/* + * Safe-links are used to maintain file system consistency during operations + * that spawns multiple transactions. For example: + * + * 1. Unlink. UNIX supports "open-but-unlinked" files, that is files + * without user-visible names in the file system, but still opened by some + * active process. What happens here is that unlink proper (i.e., removal + * of the last file name) and file deletion (truncate of file body to zero + * and deletion of stat-data, that happens when last file descriptor is + * closed), may belong to different transactions T1 and T2. If a crash + * happens after T1 commit, but before T2 commit, on-disk file system has + * a file without name, that is, disk space leak. + * + * 2. Truncate. Truncate of large file may spawn multiple transactions. If + * system crashes while truncate was in-progress, file is left partially + * truncated, which violates "atomicity guarantees" of reiser4, viz. that + * every system is atomic. + * + * Safe-links address both above cases. Basically, safe-link is a way post + * some operation to be executed during commit of some other transaction than + * current one. (Another way to look at the safe-link is to interpret it as a + * logical logging.) + * + * Specifically, at the beginning of unlink safe-link in inserted in the + * tree. This safe-link is normally removed by file deletion code (during + * transaction T2 in the above terms). Truncate also inserts safe-link that is + * normally removed when truncate operation is finished. + * + * This means, that in the case of "clean umount" there are no safe-links in + * the tree. If safe-links are observed during mount, it means that (a) system + * was terminated abnormally, and (b) safe-link correspond to the "pending" + * (i.e., not finished) operations that were in-progress during system + * termination. Each safe-link record enough information to complete + * corresponding operation, and mount simply "replays" them (hence, the + * analogy with the logical logging). + * + * Safe-links are implemented as blackbox items (see + * plugin/item/blackbox.[ch]). + * + * For the reference: ext3 also has similar mechanism, it's called "an orphan + * list" there. + */ + +#include "safe_link.h" +#include "debug.h" +#include "inode.h" + +#include "plugin/item/blackbox.h" + +#include + +/* + * On-disk format of safe-link. + */ +typedef struct safelink { + reiser4_key sdkey; /* key of stat-data for the file safe-link is + * for */ + d64 size; /* size to which file should be truncated */ +} safelink_t; + +/* + * locality where safe-link items are stored. Next to the objectid of root + * directory. + */ +static oid_t safe_link_locality(reiser4_tree * tree) +{ + return get_key_objectid(get_super_private(tree->super)->df_plug-> + root_dir_key(tree->super)) + 1; +} + +/* + Construct a key for the safe-link. Key has the following format: + +| 60 | 4 | 64 | 4 | 60 | 64 | ++---------------+---+------------------+---+---------------+------------------+ +| locality | 0 | 0 | 0 | objectid | link type | ++---------------+---+------------------+---+---------------+------------------+ +| | | | | +| 8 bytes | 8 bytes | 8 bytes | 8 bytes | + + This is in large keys format. In small keys format second 8 byte chunk is + out. Locality is a constant returned by safe_link_locality(). objectid is + an oid of a file on which operation protected by this safe-link is + performed. link-type is used to distinguish safe-links for different + operations. + + */ +static reiser4_key *build_link_key(reiser4_tree * tree, oid_t oid, + reiser4_safe_link_t link, reiser4_key * key) +{ + reiser4_key_init(key); + set_key_locality(key, safe_link_locality(tree)); + set_key_objectid(key, oid); + set_key_offset(key, link); + return key; +} + +/* + * how much disk space is necessary to insert and remove (in the + * error-handling path) safe-link. + */ +static __u64 safe_link_tograb(reiser4_tree * tree) +{ + return + /* insert safe link */ + estimate_one_insert_item(tree) + + /* remove safe link */ + estimate_one_item_removal(tree) + + /* drill to the leaf level during insertion */ + 1 + estimate_one_insert_item(tree) + + /* + * possible update of existing safe-link. Actually, if + * safe-link existed already (we failed to remove it), then no + * insertion is necessary, so this term is already "covered", + * but for simplicity let's left it. + */ + 1; +} + +/* + * grab enough disk space to insert and remove (in the error-handling path) + * safe-link. + */ +int safe_link_grab(reiser4_tree * tree, reiser4_ba_flags_t flags) +{ + int result; + + grab_space_enable(); + /* The sbinfo->delete_mutex can be taken here. + * safe_link_release() should be called before leaving reiser4 + * context. */ + result = + reiser4_grab_reserved(tree->super, safe_link_tograb(tree), flags); + grab_space_enable(); + return result; +} + +/* + * release unused disk space reserved by safe_link_grab(). + */ +void safe_link_release(reiser4_tree * tree) +{ + reiser4_release_reserved(tree->super); +} + +/* + * insert into tree safe-link for operation @link on inode @inode. + */ +int safe_link_add(struct inode *inode, reiser4_safe_link_t link) +{ + reiser4_key key; + safelink_t sl; + int length; + int result; + reiser4_tree *tree; + + build_sd_key(inode, &sl.sdkey); + length = sizeof sl.sdkey; + + if (link == SAFE_TRUNCATE) { + /* + * for truncate we have to store final file length also, + * expand item. + */ + length += sizeof(sl.size); + put_unaligned(cpu_to_le64(inode->i_size), &sl.size); + } + tree = reiser4_tree_by_inode(inode); + build_link_key(tree, get_inode_oid(inode), link, &key); + + result = store_black_box(tree, &key, &sl, length); + if (result == -EEXIST) + result = update_black_box(tree, &key, &sl, length); + return result; +} + +/* + * remove safe-link corresponding to the operation @link on inode @inode from + * the tree. + */ +int safe_link_del(reiser4_tree * tree, oid_t oid, reiser4_safe_link_t link) +{ + reiser4_key key; + + return kill_black_box(tree, build_link_key(tree, oid, link, &key)); +} + +/* + * in-memory structure to keep information extracted from safe-link. This is + * used to iterate over all safe-links. + */ +struct safe_link_context { + reiser4_tree *tree; /* internal tree */ + reiser4_key key; /* safe-link key */ + reiser4_key sdkey; /* key of object stat-data */ + reiser4_safe_link_t link; /* safe-link type */ + oid_t oid; /* object oid */ + __u64 size; /* final size for truncate */ +}; + +/* + * start iterating over all safe-links. + */ +static void safe_link_iter_begin(reiser4_tree * tree, + struct safe_link_context *ctx) +{ + ctx->tree = tree; + reiser4_key_init(&ctx->key); + set_key_locality(&ctx->key, safe_link_locality(tree)); + set_key_objectid(&ctx->key, get_key_objectid(reiser4_max_key())); + set_key_offset(&ctx->key, get_key_offset(reiser4_max_key())); +} + +/* + * return next safe-link. + */ +static int safe_link_iter_next(struct safe_link_context *ctx) +{ + int result; + safelink_t sl; + + result = load_black_box(ctx->tree, &ctx->key, &sl, sizeof sl, 0); + if (result == 0) { + ctx->oid = get_key_objectid(&ctx->key); + ctx->link = get_key_offset(&ctx->key); + ctx->sdkey = sl.sdkey; + if (ctx->link == SAFE_TRUNCATE) + ctx->size = le64_to_cpu(get_unaligned(&sl.size)); + } + return result; +} + +/* + * check are there any more safe-links left in the tree. + */ +static int safe_link_iter_finished(struct safe_link_context *ctx) +{ + return get_key_locality(&ctx->key) != safe_link_locality(ctx->tree); +} + +/* + * finish safe-link iteration. + */ +static void safe_link_iter_end(struct safe_link_context *ctx) +{ + /* nothing special */ +} + +/* + * process single safe-link. + */ +static int process_safelink(struct super_block *super, reiser4_safe_link_t link, + reiser4_key * sdkey, oid_t oid, __u64 size) +{ + struct inode *inode; + int result; + + /* + * obtain object inode by reiser4_iget(), then call object plugin + * ->safelink() method to do actual work, then delete safe-link on + * success. + */ + inode = reiser4_iget(super, sdkey, 1); + if (!IS_ERR(inode)) { + file_plugin *fplug; + + fplug = inode_file_plugin(inode); + assert("nikita-3428", fplug != NULL); + assert("", oid == get_inode_oid(inode)); + if (fplug->safelink != NULL) { + /* reiser4_txn_restart_current is not necessary because + * mounting is signle thread. However, without it + * deadlock detection code will complain (see + * nikita-3361). */ + reiser4_txn_restart_current(); + result = fplug->safelink(inode, link, size); + } else { + warning("nikita-3430", + "Cannot handle safelink for %lli", + (unsigned long long)oid); + reiser4_print_key("key", sdkey); + result = 0; + } + if (result != 0) { + warning("nikita-3431", + "Error processing safelink for %lli: %i", + (unsigned long long)oid, result); + } + reiser4_iget_complete(inode); + iput(inode); + if (result == 0) { + result = safe_link_grab(reiser4_get_tree(super), + BA_CAN_COMMIT); + if (result == 0) + result = + safe_link_del(reiser4_get_tree(super), oid, + link); + safe_link_release(reiser4_get_tree(super)); + /* + * restart transaction: if there was large number of + * safe-links, their processing may fail to fit into + * single transaction. + */ + if (result == 0) + reiser4_txn_restart_current(); + } + } else + result = PTR_ERR(inode); + return result; +} + +/* + * iterate over all safe-links in the file-system processing them one by one. + */ +int process_safelinks(struct super_block *super) +{ + struct safe_link_context ctx; + int result; + + if (rofs_super(super)) + /* do nothing on the read-only file system */ + return 0; + safe_link_iter_begin(&get_super_private(super)->tree, &ctx); + result = 0; + do { + result = safe_link_iter_next(&ctx); + if (safe_link_iter_finished(&ctx) || result == -ENOENT) { + result = 0; + break; + } + if (result == 0) + result = process_safelink(super, ctx.link, + &ctx.sdkey, ctx.oid, + ctx.size); + } while (result == 0); + safe_link_iter_end(&ctx); + return result; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/safe_link.h linux-2.6.30/fs/reiser4/safe_link.h --- linux-2.6.30.orig/fs/reiser4/safe_link.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/safe_link.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,29 @@ +/* Copyright 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Safe-links. See safe_link.c for details. */ + +#if !defined(__FS_SAFE_LINK_H__) +#define __FS_SAFE_LINK_H__ + +#include "tree.h" + +int safe_link_grab(reiser4_tree * tree, reiser4_ba_flags_t flags); +void safe_link_release(reiser4_tree * tree); +int safe_link_add(struct inode *inode, reiser4_safe_link_t link); +int safe_link_del(reiser4_tree *, oid_t oid, reiser4_safe_link_t link); + +int process_safelinks(struct super_block *super); + +/* __FS_SAFE_LINK_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/seal.c linux-2.6.30/fs/reiser4/seal.c --- linux-2.6.30.orig/fs/reiser4/seal.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/seal.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,218 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ +/* Seals implementation. */ +/* Seals are "weak" tree pointers. They are analogous to tree coords in + allowing to bypass tree traversal. But normal usage of coords implies that + node pointed to by coord is locked, whereas seals don't keep a lock (or + even a reference) to znode. In stead, each znode contains a version number, + increased on each znode modification. This version number is copied into a + seal when seal is created. Later, one can "validate" seal by calling + reiser4_seal_validate(). If znode is in cache and its version number is + still the same, seal is "pristine" and coord associated with it can be + re-used immediately. + + If, on the other hand, znode is out of cache, or it is obviously different + one from the znode seal was initially attached to (for example, it is on + the different level, or is being removed from the tree), seal is + irreparably invalid ("burned") and tree traversal has to be repeated. + + Otherwise, there is some hope, that while znode was modified (and seal was + "broken" as a result), key attached to the seal is still in the node. This + is checked by first comparing this key with delimiting keys of node and, if + key is ok, doing intra-node lookup. + + Znode version is maintained in the following way: + + there is reiser4_tree.znode_epoch counter. Whenever new znode is created, + znode_epoch is incremented and its new value is stored in ->version field + of new znode. Whenever znode is dirtied (which means it was probably + modified), znode_epoch is also incremented and its new value is stored in + znode->version. This is done so, because just incrementing znode->version + on each update is not enough: it may so happen, that znode get deleted, new + znode is allocated for the same disk block and gets the same version + counter, tricking seal code into false positive. +*/ + +#include "forward.h" +#include "debug.h" +#include "key.h" +#include "coord.h" +#include "seal.h" +#include "plugin/item/item.h" +#include "plugin/node/node.h" +#include "jnode.h" +#include "znode.h" +#include "super.h" + +static znode *seal_node(const seal_t *seal); +static int seal_matches(const seal_t *seal, znode * node); + +/* initialise seal. This can be called several times on the same seal. @coord + and @key can be NULL. */ +void reiser4_seal_init(seal_t *seal /* seal to initialise */ , + const coord_t *coord /* coord @seal will be + * attached to */ , + const reiser4_key * key UNUSED_ARG /* key @seal will be + * attached to */ ) +{ + assert("nikita-1886", seal != NULL); + memset(seal, 0, sizeof *seal); + if (coord != NULL) { + znode *node; + + node = coord->node; + assert("nikita-1987", node != NULL); + spin_lock_znode(node); + seal->version = node->version; + assert("nikita-1988", seal->version != 0); + seal->block = *znode_get_block(node); +#if REISER4_DEBUG + seal->coord1 = *coord; + if (key != NULL) + seal->key = *key; +#endif + spin_unlock_znode(node); + } +} + +/* finish with seal */ +void reiser4_seal_done(seal_t *seal/* seal to clear */) +{ + assert("nikita-1887", seal != NULL); + seal->version = 0; +} + +/* true if seal was initialised */ +int reiser4_seal_is_set(const seal_t *seal/* seal to query */) +{ + assert("nikita-1890", seal != NULL); + return seal->version != 0; +} + +#if REISER4_DEBUG +/* helper function for reiser4_seal_validate(). It checks that item at @coord + * has expected key. This is to detect cases where node was modified but wasn't + * marked dirty. */ +static inline int check_seal_match(const coord_t *coord /* coord to check */ , + const reiser4_key * k/* expected key */) +{ + reiser4_key ukey; + + return (coord->between != AT_UNIT) || + /* FIXME-VS: we only can compare keys for items whose units + represent exactly one key */ + ((coord_is_existing_unit(coord)) + && (item_is_extent(coord) + || keyeq(k, unit_key_by_coord(coord, &ukey)))) + || ((coord_is_existing_unit(coord)) && (item_is_ctail(coord)) + && keyge(k, unit_key_by_coord(coord, &ukey))); +} +#endif + +/* this is used by reiser4_seal_validate. It accepts return value of + * longterm_lock_znode and returns 1 if it can be interpreted as seal + * validation failure. For instance, when longterm_lock_znode returns -EINVAL, + * reiser4_seal_validate returns -E_REPEAT and caller will call tre search. + * We cannot do this in longterm_lock_znode(), because sometimes we want to + * distinguish between -EINVAL and -E_REPEAT. */ +static int should_repeat(int return_code) +{ + return return_code == -EINVAL; +} + +/* (re-)validate seal. + + Checks whether seal is pristine, and try to revalidate it if possible. + + If seal was burned, or broken irreparably, return -E_REPEAT. + + NOTE-NIKITA currently reiser4_seal_validate() returns -E_REPEAT if key we are + looking for is in range of keys covered by the sealed node, but item wasn't + found by node ->lookup() method. Alternative is to return -ENOENT in this + case, but this would complicate callers logic. + +*/ +int reiser4_seal_validate(seal_t *seal /* seal to validate */, + coord_t *coord /* coord to validate against */, + const reiser4_key * key /* key to validate against */, + lock_handle * lh /* resulting lock handle */, + znode_lock_mode mode /* lock node */, + znode_lock_request request/* locking priority */) +{ + znode *node; + int result; + + assert("nikita-1889", seal != NULL); + assert("nikita-1881", reiser4_seal_is_set(seal)); + assert("nikita-1882", key != NULL); + assert("nikita-1883", coord != NULL); + assert("nikita-1884", lh != NULL); + assert("nikita-1885", keyeq(&seal->key, key)); + assert("nikita-1989", coords_equal(&seal->coord1, coord)); + + /* obtain znode by block number */ + node = seal_node(seal); + if (node != NULL) { + /* znode was in cache, lock it */ + result = longterm_lock_znode(lh, node, mode, request); + zput(node); + if (result == 0) { + if (seal_matches(seal, node)) { + /* if seal version and znode version + coincide */ + ON_DEBUG(coord_update_v(coord)); + assert("nikita-1990", + node == seal->coord1.node); + assert("nikita-1898", + WITH_DATA_RET(coord->node, 1, + check_seal_match(coord, + key))); + } else + result = RETERR(-E_REPEAT); + } + if (result != 0) { + if (should_repeat(result)) + result = RETERR(-E_REPEAT); + /* unlock node on failure */ + done_lh(lh); + } + } else { + /* znode wasn't in cache */ + result = RETERR(-E_REPEAT); + } + return result; +} + +/* helpers functions */ + +/* obtain reference to znode seal points to, if in cache */ +static znode *seal_node(const seal_t *seal/* seal to query */) +{ + assert("nikita-1891", seal != NULL); + return zlook(current_tree, &seal->block); +} + +/* true if @seal version and @node version coincide */ +static int seal_matches(const seal_t *seal /* seal to check */ , + znode * node/* node to check */) +{ + int result; + + assert("nikita-1991", seal != NULL); + assert("nikita-1993", node != NULL); + + spin_lock_znode(node); + result = (seal->version == node->version); + spin_unlock_znode(node); + return result; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/seal.h linux-2.6.30/fs/reiser4/seal.h --- linux-2.6.30.orig/fs/reiser4/seal.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/seal.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,49 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Declaration of seals: "weak" tree pointers. See seal.c for comments. */ + +#ifndef __SEAL_H__ +#define __SEAL_H__ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" + +/* for __u?? types */ +/*#include */ + +/* seal. See comment at the top of seal.c */ +typedef struct seal_s { + /* version of znode recorder at the time of seal creation */ + __u64 version; + /* block number of znode attached to this seal */ + reiser4_block_nr block; +#if REISER4_DEBUG + /* coord this seal is attached to. For debugging. */ + coord_t coord1; + /* key this seal is attached to. For debugging. */ + reiser4_key key; +#endif +} seal_t; + +extern void reiser4_seal_init(seal_t *, const coord_t *, const reiser4_key *); +extern void reiser4_seal_done(seal_t *); +extern int reiser4_seal_is_set(const seal_t *); +extern int reiser4_seal_validate(seal_t *, coord_t *, + const reiser4_key *, lock_handle * , + znode_lock_mode mode, znode_lock_request request); + +/* __SEAL_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/search.c linux-2.6.30/fs/reiser4/search.c --- linux-2.6.30.orig/fs/reiser4/search.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/search.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1612 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "seal.h" +#include "plugin/item/item.h" +#include "plugin/node/node.h" +#include "plugin/plugin.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree_walk.h" +#include "tree.h" +#include "reiser4.h" +#include "super.h" +#include "inode.h" + +#include + +static const char *bias_name(lookup_bias bias); + +/* tree searching algorithm, intranode searching algorithms are in + plugin/node/ */ + +/* tree lookup cache + * + * The coord by key cache consists of small list of recently accessed nodes + * maintained according to the LRU discipline. Before doing real top-to-down + * tree traversal this cache is scanned for nodes that can contain key + * requested. + * + * The efficiency of coord cache depends heavily on locality of reference for + * tree accesses. Our user level simulations show reasonably good hit ratios + * for coord cache under most loads so far. + */ + +/* Initialise coord cache slot */ +static void cbk_cache_init_slot(cbk_cache_slot *slot) +{ + assert("nikita-345", slot != NULL); + + INIT_LIST_HEAD(&slot->lru); + slot->node = NULL; +} + +/* Initialize coord cache */ +int cbk_cache_init(cbk_cache * cache/* cache to init */) +{ + int i; + + assert("nikita-346", cache != NULL); + + cache->slot = + kmalloc(sizeof(cbk_cache_slot) * cache->nr_slots, + reiser4_ctx_gfp_mask_get()); + if (cache->slot == NULL) + return RETERR(-ENOMEM); + + INIT_LIST_HEAD(&cache->lru); + for (i = 0; i < cache->nr_slots; ++i) { + cbk_cache_init_slot(cache->slot + i); + list_add_tail(&((cache->slot + i)->lru), &cache->lru); + } + rwlock_init(&cache->guard); + return 0; +} + +/* free cbk cache data */ +void cbk_cache_done(cbk_cache * cache/* cache to release */) +{ + assert("nikita-2493", cache != NULL); + if (cache->slot != NULL) { + kfree(cache->slot); + cache->slot = NULL; + } +} + +/* macro to iterate over all cbk cache slots */ +#define for_all_slots(cache, slot) \ + for ((slot) = list_entry((cache)->lru.next, cbk_cache_slot, lru); \ + &(cache)->lru != &(slot)->lru; \ + (slot) = list_entry(slot->lru.next, cbk_cache_slot, lru)) + +#if REISER4_DEBUG +/* this function assures that [cbk-cache-invariant] invariant holds */ +static int cbk_cache_invariant(const cbk_cache * cache) +{ + cbk_cache_slot *slot; + int result; + int unused; + + if (cache->nr_slots == 0) + return 1; + + assert("nikita-2469", cache != NULL); + unused = 0; + result = 1; + read_lock(&((cbk_cache *)cache)->guard); + for_all_slots(cache, slot) { + /* in LRU first go all `used' slots followed by `unused' */ + if (unused && (slot->node != NULL)) + result = 0; + if (slot->node == NULL) + unused = 1; + else { + cbk_cache_slot *scan; + + /* all cached nodes are different */ + scan = slot; + while (result) { + scan = list_entry(scan->lru.next, + cbk_cache_slot, lru); + if (&cache->lru == &scan->lru) + break; + if (slot->node == scan->node) + result = 0; + } + } + if (!result) + break; + } + read_unlock(&((cbk_cache *)cache)->guard); + return result; +} + +#endif + +/* Remove references, if any, to @node from coord cache */ +void cbk_cache_invalidate(const znode * node /* node to remove from cache */ , + reiser4_tree * tree/* tree to remove node from */) +{ + cbk_cache_slot *slot; + cbk_cache *cache; + int i; + + assert("nikita-350", node != NULL); + assert("nikita-1479", LOCK_CNT_GTZ(rw_locked_tree)); + + cache = &tree->cbk_cache; + assert("nikita-2470", cbk_cache_invariant(cache)); + + write_lock(&(cache->guard)); + for (i = 0, slot = cache->slot; i < cache->nr_slots; ++i, ++slot) { + if (slot->node == node) { + list_move_tail(&slot->lru, &cache->lru); + slot->node = NULL; + break; + } + } + write_unlock(&(cache->guard)); + assert("nikita-2471", cbk_cache_invariant(cache)); +} + +/* add to the cbk-cache in the "tree" information about "node". This + can actually be update of existing slot in a cache. */ +static void cbk_cache_add(const znode * node/* node to add to the cache */) +{ + cbk_cache *cache; + + cbk_cache_slot *slot; + int i; + + assert("nikita-352", node != NULL); + + cache = &znode_get_tree(node)->cbk_cache; + assert("nikita-2472", cbk_cache_invariant(cache)); + + if (cache->nr_slots == 0) + return; + + write_lock(&(cache->guard)); + /* find slot to update/add */ + for (i = 0, slot = cache->slot; i < cache->nr_slots; ++i, ++slot) { + /* oops, this node is already in a cache */ + if (slot->node == node) + break; + } + /* if all slots are used, reuse least recently used one */ + if (i == cache->nr_slots) { + slot = list_entry(cache->lru.prev, cbk_cache_slot, lru); + slot->node = (znode *) node; + } + list_move(&slot->lru, &cache->lru); + write_unlock(&(cache->guard)); + assert("nikita-2473", cbk_cache_invariant(cache)); +} + +static int setup_delimiting_keys(cbk_handle * h); +static lookup_result coord_by_handle(cbk_handle * handle); +static lookup_result traverse_tree(cbk_handle * h); +static int cbk_cache_search(cbk_handle * h); + +static level_lookup_result cbk_level_lookup(cbk_handle * h); +static level_lookup_result cbk_node_lookup(cbk_handle * h); + +/* helper functions */ + +static void update_stale_dk(reiser4_tree * tree, znode * node); + +/* release parent node during traversal */ +static void put_parent(cbk_handle * h); +/* check consistency of fields */ +static int sanity_check(cbk_handle * h); +/* release resources in handle */ +static void hput(cbk_handle * h); + +static level_lookup_result search_to_left(cbk_handle * h); + +/* pack numerous (numberous I should say) arguments of coord_by_key() into + * cbk_handle */ +static cbk_handle *cbk_pack(cbk_handle * handle, + reiser4_tree * tree, + const reiser4_key * key, + coord_t *coord, + lock_handle * active_lh, + lock_handle * parent_lh, + znode_lock_mode lock_mode, + lookup_bias bias, + tree_level lock_level, + tree_level stop_level, + __u32 flags, ra_info_t *info) +{ + memset(handle, 0, sizeof *handle); + + handle->tree = tree; + handle->key = key; + handle->lock_mode = lock_mode; + handle->bias = bias; + handle->lock_level = lock_level; + handle->stop_level = stop_level; + handle->coord = coord; + /* set flags. See comment in tree.h:cbk_flags */ + handle->flags = flags | CBK_TRUST_DK | CBK_USE_CRABLOCK; + + handle->active_lh = active_lh; + handle->parent_lh = parent_lh; + handle->ra_info = info; + return handle; +} + +/* main tree lookup procedure + + Check coord cache. If key we are looking for is not found there, call cbk() + to do real tree traversal. + + As we have extents on the twig level, @lock_level and @stop_level can + be different from LEAF_LEVEL and each other. + + Thread cannot keep any reiser4 locks (tree, znode, dk spin-locks, or znode + long term locks) while calling this. +*/ +lookup_result coord_by_key(reiser4_tree * tree /* tree to perform search + * in. Usually this tree is + * part of file-system + * super-block */ , + const reiser4_key * key /* key to look for */ , + coord_t *coord /* where to store found + * position in a tree. Fields + * in "coord" are only valid if + * coord_by_key() returned + * "CBK_COORD_FOUND" */ , + lock_handle * lh, /* resulting lock handle */ + znode_lock_mode lock_mode /* type of lookup we + * want on node. Pass + * ZNODE_READ_LOCK here + * if you only want to + * read item found and + * ZNODE_WRITE_LOCK if + * you want to modify + * it */ , + lookup_bias bias /* what to return if coord + * with exactly the @key is + * not in the tree */ , + tree_level lock_level/* tree level where to start + * taking @lock type of + * locks */ , + tree_level stop_level/* tree level to stop. Pass + * LEAF_LEVEL or TWIG_LEVEL + * here Item being looked + * for has to be between + * @lock_level and + * @stop_level, inclusive */ , + __u32 flags /* search flags */ , + ra_info_t * + info + /* information about desired tree traversal + * readahead */ + ) +{ + cbk_handle handle; + lock_handle parent_lh; + lookup_result result; + + init_lh(lh); + init_lh(&parent_lh); + + assert("nikita-3023", reiser4_schedulable()); + + assert("nikita-353", tree != NULL); + assert("nikita-354", key != NULL); + assert("nikita-355", coord != NULL); + assert("nikita-356", (bias == FIND_EXACT) + || (bias == FIND_MAX_NOT_MORE_THAN)); + assert("nikita-357", stop_level >= LEAF_LEVEL); + /* no locks can be held during tree traversal */ + assert("nikita-2104", lock_stack_isclean(get_current_lock_stack())); + + cbk_pack(&handle, + tree, + key, + coord, + lh, + &parent_lh, + lock_mode, bias, lock_level, stop_level, flags, info); + + result = coord_by_handle(&handle); + assert("nikita-3247", + ergo(!IS_CBKERR(result), coord->node == lh->node)); + return result; +} + +/* like coord_by_key(), but starts traversal from vroot of @object rather than + * from tree root. */ +lookup_result reiser4_object_lookup(struct inode *object, + const reiser4_key * key, + coord_t *coord, + lock_handle * lh, + znode_lock_mode lock_mode, + lookup_bias bias, + tree_level lock_level, + tree_level stop_level, __u32 flags, + ra_info_t *info) +{ + cbk_handle handle; + lock_handle parent_lh; + lookup_result result; + + init_lh(lh); + init_lh(&parent_lh); + + assert("nikita-3023", reiser4_schedulable()); + + assert("nikita-354", key != NULL); + assert("nikita-355", coord != NULL); + assert("nikita-356", (bias == FIND_EXACT) + || (bias == FIND_MAX_NOT_MORE_THAN)); + assert("nikita-357", stop_level >= LEAF_LEVEL); + /* no locks can be held during tree search by key */ + assert("nikita-2104", lock_stack_isclean(get_current_lock_stack())); + + cbk_pack(&handle, + object != NULL ? reiser4_tree_by_inode(object) : current_tree, + key, + coord, + lh, + &parent_lh, + lock_mode, bias, lock_level, stop_level, flags, info); + handle.object = object; + + result = coord_by_handle(&handle); + assert("nikita-3247", + ergo(!IS_CBKERR(result), coord->node == lh->node)); + return result; +} + +/* lookup by cbk_handle. Common part of coord_by_key() and + reiser4_object_lookup(). */ +static lookup_result coord_by_handle(cbk_handle * handle) +{ + /* + * first check cbk_cache (which is look-aside cache for our tree) and + * of this fails, start traversal. + */ + /* first check whether "key" is in cache of recent lookups. */ + if (cbk_cache_search(handle) == 0) + return handle->result; + else + return traverse_tree(handle); +} + +/* Execute actor for each item (or unit, depending on @through_units_p), + starting from @coord, right-ward, until either: + + - end of the tree is reached + - unformatted node is met + - error occurred + - @actor returns 0 or less + + Error code, or last actor return value is returned. + + This is used by plugin/dir/hashe_dir.c:reiser4_find_entry() to move through + sequence of entries with identical keys and alikes. +*/ +int reiser4_iterate_tree(reiser4_tree * tree /* tree to scan */ , + coord_t *coord /* coord to start from */ , + lock_handle * lh /* lock handle to start with and to + * update along the way */ , + tree_iterate_actor_t actor /* function to call on each + * item/unit */ , + void *arg /* argument to pass to @actor */ , + znode_lock_mode mode /* lock mode on scanned nodes */ , + int through_units_p /* call @actor on each item or on + * each unit */ ) +{ + int result; + + assert("nikita-1143", tree != NULL); + assert("nikita-1145", coord != NULL); + assert("nikita-1146", lh != NULL); + assert("nikita-1147", actor != NULL); + + result = zload(coord->node); + coord_clear_iplug(coord); + if (result != 0) + return result; + if (!coord_is_existing_unit(coord)) { + zrelse(coord->node); + return -ENOENT; + } + while ((result = actor(tree, coord, lh, arg)) > 0) { + /* move further */ + if ((through_units_p && coord_next_unit(coord)) || + (!through_units_p && coord_next_item(coord))) { + do { + lock_handle couple; + + /* move to the next node */ + init_lh(&couple); + result = + reiser4_get_right_neighbor(&couple, + coord->node, + (int)mode, + GN_CAN_USE_UPPER_LEVELS); + zrelse(coord->node); + if (result == 0) { + + result = zload(couple.node); + if (result != 0) { + done_lh(&couple); + return result; + } + + coord_init_first_unit(coord, + couple.node); + done_lh(lh); + move_lh(lh, &couple); + } else + return result; + } while (node_is_empty(coord->node)); + } + + assert("nikita-1149", coord_is_existing_unit(coord)); + } + zrelse(coord->node); + return result; +} + +/* return locked uber znode for @tree */ +int get_uber_znode(reiser4_tree * tree, znode_lock_mode mode, + znode_lock_request pri, lock_handle * lh) +{ + int result; + + result = longterm_lock_znode(lh, tree->uber, mode, pri); + return result; +} + +/* true if @key is strictly within @node + + we are looking for possibly non-unique key and it is item is at the edge of + @node. May be it is in the neighbor. +*/ +static int znode_contains_key_strict(znode * node /* node to check key + * against */ , + const reiser4_key * + key /* key to check */ , + int isunique) +{ + int answer; + + assert("nikita-1760", node != NULL); + assert("nikita-1722", key != NULL); + + if (keyge(key, &node->rd_key)) + return 0; + + answer = keycmp(&node->ld_key, key); + + if (isunique) + return answer != GREATER_THAN; + else + return answer == LESS_THAN; +} + +/* + * Virtual Root (vroot) code. + * + * For given file system object (e.g., regular file or directory) let's + * define its "virtual root" as lowest in the tree (that is, furtherest + * from the tree root) node such that all body items of said object are + * located in a tree rooted at this node. + * + * Once vroot of object is found all tree lookups for items within body of + * this object ("object lookups") can be started from its vroot rather + * than from real root. This has following advantages: + * + * 1. amount of nodes traversed during lookup (and, hence, amount of + * key comparisons made) decreases, and + * + * 2. contention on tree root is decreased. This latter was actually + * motivating reason behind vroot, because spin lock of root node, + * which is taken when acquiring long-term lock on root node is the + * hottest lock in the reiser4. + * + * How to find vroot. + * + * When vroot of object F is not yet determined, all object lookups start + * from the root of the tree. At each tree level during traversal we have + * a node N such that a key we are looking for (which is the key inside + * object's body) is located within N. In function handle_vroot() called + * from cbk_level_lookup() we check whether N is possible vroot for + * F. Check is trivial---if neither leftmost nor rightmost item of N + * belongs to F (and we already have helpful ->owns_item() method of + * object plugin for this), then N is possible vroot of F. This, of + * course, relies on the assumption that each object occupies contiguous + * range of keys in the tree. + * + * Thus, traversing tree downward and checking each node as we go, we can + * find lowest such node, which, by definition, is vroot. + * + * How to track vroot. + * + * Nohow. If actual vroot changes, next object lookup will just restart + * from the actual tree root, refreshing object's vroot along the way. + * + */ + +/* + * Check whether @node is possible vroot of @object. + */ +static void handle_vroot(struct inode *object, znode * node) +{ + file_plugin *fplug; + coord_t coord; + + fplug = inode_file_plugin(object); + assert("nikita-3353", fplug != NULL); + assert("nikita-3354", fplug->owns_item != NULL); + + if (unlikely(node_is_empty(node))) + return; + + coord_init_first_unit(&coord, node); + /* + * if leftmost item of @node belongs to @object, we cannot be sure + * that @node is vroot of @object, because, some items of @object are + * probably in the sub-tree rooted at the left neighbor of @node. + */ + if (fplug->owns_item(object, &coord)) + return; + coord_init_last_unit(&coord, node); + /* mutatis mutandis for the rightmost item */ + if (fplug->owns_item(object, &coord)) + return; + /* otherwise, @node is possible vroot of @object */ + inode_set_vroot(object, node); +} + +/* + * helper function used by traverse tree to start tree traversal not from the + * tree root, but from @h->object's vroot, if possible. + */ +static int prepare_object_lookup(cbk_handle * h) +{ + znode *vroot; + int result; + + vroot = inode_get_vroot(h->object); + if (vroot == NULL) { + /* + * object doesn't have known vroot, start from real tree root. + */ + return LOOKUP_CONT; + } + + h->level = znode_get_level(vroot); + /* take a long-term lock on vroot */ + h->result = longterm_lock_znode(h->active_lh, vroot, + cbk_lock_mode(h->level, h), + ZNODE_LOCK_LOPRI); + result = LOOKUP_REST; + if (h->result == 0) { + int isunique; + int inside; + + isunique = h->flags & CBK_UNIQUE; + /* check that key is inside vroot */ + read_lock_dk(h->tree); + inside = (znode_contains_key_strict(vroot, h->key, isunique) && + !ZF_ISSET(vroot, JNODE_HEARD_BANSHEE)); + read_unlock_dk(h->tree); + if (inside) { + h->result = zload(vroot); + if (h->result == 0) { + /* search for key in vroot. */ + result = cbk_node_lookup(h); + zrelse(vroot); /*h->active_lh->node); */ + if (h->active_lh->node != vroot) { + result = LOOKUP_REST; + } else if (result == LOOKUP_CONT) { + move_lh(h->parent_lh, h->active_lh); + h->flags &= ~CBK_DKSET; + } + } + } + } + + zput(vroot); + + if (IS_CBKERR(h->result) || result == LOOKUP_REST) + hput(h); + return result; +} + +/* main function that handles common parts of tree traversal: starting + (fake znode handling), restarts, error handling, completion */ +static lookup_result traverse_tree(cbk_handle * h/* search handle */) +{ + int done; + int iterations; + int vroot_used; + + assert("nikita-365", h != NULL); + assert("nikita-366", h->tree != NULL); + assert("nikita-367", h->key != NULL); + assert("nikita-368", h->coord != NULL); + assert("nikita-369", (h->bias == FIND_EXACT) + || (h->bias == FIND_MAX_NOT_MORE_THAN)); + assert("nikita-370", h->stop_level >= LEAF_LEVEL); + assert("nikita-2949", !(h->flags & CBK_DKSET)); + assert("zam-355", lock_stack_isclean(get_current_lock_stack())); + + done = 0; + iterations = 0; + vroot_used = 0; + + /* loop for restarts */ +restart: + + assert("nikita-3024", reiser4_schedulable()); + + h->result = CBK_COORD_FOUND; + /* connect_znode() needs it */ + h->ld_key = *reiser4_min_key(); + h->rd_key = *reiser4_max_key(); + h->flags |= CBK_DKSET; + h->error = NULL; + + if (!vroot_used && h->object != NULL) { + vroot_used = 1; + done = prepare_object_lookup(h); + if (done == LOOKUP_REST) + goto restart; + else if (done == LOOKUP_DONE) + return h->result; + } + if (h->parent_lh->node == NULL) { + done = + get_uber_znode(h->tree, ZNODE_READ_LOCK, ZNODE_LOCK_LOPRI, + h->parent_lh); + + assert("nikita-1637", done != -E_DEADLOCK); + + h->block = h->tree->root_block; + h->level = h->tree->height; + h->coord->node = h->parent_lh->node; + + if (done != 0) + return done; + } + + /* loop descending a tree */ + while (!done) { + + if (unlikely((iterations > REISER4_CBK_ITERATIONS_LIMIT) && + IS_POW(iterations))) { + warning("nikita-1481", "Too many iterations: %i", + iterations); + reiser4_print_key("key", h->key); + ++iterations; + } else if (unlikely(iterations > REISER4_MAX_CBK_ITERATIONS)) { + h->error = + "reiser-2018: Too many iterations. Tree corrupted, or (less likely) starvation occurring."; + h->result = RETERR(-EIO); + break; + } + switch (cbk_level_lookup(h)) { + case LOOKUP_CONT: + move_lh(h->parent_lh, h->active_lh); + continue; + default: + wrong_return_value("nikita-372", "cbk_level"); + case LOOKUP_DONE: + done = 1; + break; + case LOOKUP_REST: + hput(h); + /* deadlock avoidance is normal case. */ + if (h->result != -E_DEADLOCK) + ++iterations; + reiser4_preempt_point(); + goto restart; + } + } + /* that's all. The rest is error handling */ + if (unlikely(h->error != NULL)) { + warning("nikita-373", "%s: level: %i, " + "lock_level: %i, stop_level: %i " + "lock_mode: %s, bias: %s", + h->error, h->level, h->lock_level, h->stop_level, + lock_mode_name(h->lock_mode), bias_name(h->bias)); + reiser4_print_address("block", &h->block); + reiser4_print_key("key", h->key); + print_coord_content("coord", h->coord); + } + /* `unlikely' error case */ + if (unlikely(IS_CBKERR(h->result))) { + /* failure. do cleanup */ + hput(h); + } else { + assert("nikita-1605", WITH_DATA_RET + (h->coord->node, 1, + ergo((h->result == CBK_COORD_FOUND) && + (h->bias == FIND_EXACT) && + (!node_is_empty(h->coord->node)), + coord_is_existing_item(h->coord)))); + } + return h->result; +} + +/* find delimiting keys of child + + Determine left and right delimiting keys for child pointed to by + @parent_coord. + +*/ +static void find_child_delimiting_keys(znode * parent /* parent znode, passed + * locked */ , + const coord_t *parent_coord + /* coord where pointer + * to child is stored + */ , + reiser4_key * ld /* where to store left + * delimiting key */ , + reiser4_key * rd /* where to store right + * delimiting key */ ) +{ + coord_t neighbor; + + assert("nikita-1484", parent != NULL); + assert_rw_locked(&(znode_get_tree(parent)->dk_lock)); + + coord_dup(&neighbor, parent_coord); + + if (neighbor.between == AT_UNIT) + /* imitate item ->lookup() behavior. */ + neighbor.between = AFTER_UNIT; + + if (coord_set_to_left(&neighbor) == 0) + unit_key_by_coord(&neighbor, ld); + else { + assert("nikita-14851", 0); + *ld = *znode_get_ld_key(parent); + } + + coord_dup(&neighbor, parent_coord); + if (neighbor.between == AT_UNIT) + neighbor.between = AFTER_UNIT; + if (coord_set_to_right(&neighbor) == 0) + unit_key_by_coord(&neighbor, rd); + else + *rd = *znode_get_rd_key(parent); +} + +/* + * setup delimiting keys for a child + * + * @parent parent node + * + * @coord location in @parent where pointer to @child is + * + * @child child node + */ +int +set_child_delimiting_keys(znode * parent, const coord_t *coord, znode * child) +{ + reiser4_tree *tree; + + assert("nikita-2952", + znode_get_level(parent) == znode_get_level(coord->node)); + + /* fast check without taking dk lock. This is safe, because + * JNODE_DKSET is never cleared once set. */ + if (!ZF_ISSET(child, JNODE_DKSET)) { + tree = znode_get_tree(parent); + write_lock_dk(tree); + if (likely(!ZF_ISSET(child, JNODE_DKSET))) { + find_child_delimiting_keys(parent, coord, + &child->ld_key, + &child->rd_key); + ON_DEBUG(child->ld_key_version = + atomic_inc_return(&delim_key_version); + child->rd_key_version = + atomic_inc_return(&delim_key_version);); + ZF_SET(child, JNODE_DKSET); + } + write_unlock_dk(tree); + return 1; + } + return 0; +} + +/* Perform tree lookup at one level. This is called from cbk_traverse() + function that drives lookup through tree and calls cbk_node_lookup() to + perform lookup within one node. + + See comments in a code. +*/ +static level_lookup_result cbk_level_lookup(cbk_handle * h/* search handle */) +{ + int ret; + int setdk; + int ldkeyset = 0; + reiser4_key ldkey; + reiser4_key key; + znode *active; + + assert("nikita-3025", reiser4_schedulable()); + + /* acquire reference to @active node */ + active = + zget(h->tree, &h->block, h->parent_lh->node, h->level, + reiser4_ctx_gfp_mask_get()); + + if (IS_ERR(active)) { + h->result = PTR_ERR(active); + return LOOKUP_DONE; + } + + /* lock @active */ + h->result = longterm_lock_znode(h->active_lh, + active, + cbk_lock_mode(h->level, h), + ZNODE_LOCK_LOPRI); + /* longterm_lock_znode() acquires additional reference to znode (which + will be later released by longterm_unlock_znode()). Release + reference acquired by zget(). + */ + zput(active); + if (unlikely(h->result != 0)) + goto fail_or_restart; + + setdk = 0; + /* if @active is accessed for the first time, setup delimiting keys on + it. Delimiting keys are taken from the parent node. See + setup_delimiting_keys() for details. + */ + if (h->flags & CBK_DKSET) { + setdk = setup_delimiting_keys(h); + h->flags &= ~CBK_DKSET; + } else { + znode *parent; + + parent = h->parent_lh->node; + h->result = zload(parent); + if (unlikely(h->result != 0)) + goto fail_or_restart; + + if (!ZF_ISSET(active, JNODE_DKSET)) + setdk = set_child_delimiting_keys(parent, + h->coord, active); + else { + read_lock_dk(h->tree); + find_child_delimiting_keys(parent, h->coord, &ldkey, + &key); + read_unlock_dk(h->tree); + ldkeyset = 1; + } + zrelse(parent); + } + + /* this is ugly kludge. Reminder: this is necessary, because + ->lookup() method returns coord with ->between field probably set + to something different from AT_UNIT. + */ + h->coord->between = AT_UNIT; + + if (znode_just_created(active) && (h->coord->node != NULL)) { + write_lock_tree(h->tree); + /* if we are going to load znode right now, setup + ->in_parent: coord where pointer to this node is stored in + parent. + */ + coord_to_parent_coord(h->coord, &active->in_parent); + write_unlock_tree(h->tree); + } + + /* check connectedness without holding tree lock---false negatives + * will be re-checked by connect_znode(), and false positives are + * impossible---@active cannot suddenly turn into unconnected + * state. */ + if (!znode_is_connected(active)) { + h->result = connect_znode(h->coord, active); + if (unlikely(h->result != 0)) { + put_parent(h); + goto fail_or_restart; + } + } + + jload_prefetch(ZJNODE(active)); + + if (setdk) + update_stale_dk(h->tree, active); + + /* put_parent() cannot be called earlier, because connect_znode() + assumes parent node is referenced; */ + put_parent(h); + + if ((!znode_contains_key_lock(active, h->key) && + (h->flags & CBK_TRUST_DK)) + || ZF_ISSET(active, JNODE_HEARD_BANSHEE)) { + /* 1. key was moved out of this node while this thread was + waiting for the lock. Restart. More elaborate solution is + to determine where key moved (to the left, or to the right) + and try to follow it through sibling pointers. + + 2. or, node itself is going to be removed from the + tree. Release lock and restart. + */ + h->result = -E_REPEAT; + } + if (h->result == -E_REPEAT) + return LOOKUP_REST; + + h->result = zload_ra(active, h->ra_info); + if (h->result) + return LOOKUP_DONE; + + /* sanity checks */ + if (sanity_check(h)) { + zrelse(active); + return LOOKUP_DONE; + } + + /* check that key of leftmost item in the @active is the same as in + * its parent */ + if (ldkeyset && !node_is_empty(active) && + !keyeq(leftmost_key_in_node(active, &key), &ldkey)) { + warning("vs-3533", "Keys are inconsistent. Fsck?"); + reiser4_print_key("inparent", &ldkey); + reiser4_print_key("inchild", &key); + h->result = RETERR(-EIO); + zrelse(active); + return LOOKUP_DONE; + } + + if (h->object != NULL) + handle_vroot(h->object, active); + + ret = cbk_node_lookup(h); + + /* h->active_lh->node might change, but active is yet to be zrelsed */ + zrelse(active); + + return ret; + +fail_or_restart: + if (h->result == -E_DEADLOCK) + return LOOKUP_REST; + return LOOKUP_DONE; +} + +#if REISER4_DEBUG +/* check left and right delimiting keys of a znode */ +void check_dkeys(znode * node) +{ + znode *left; + znode *right; + + read_lock_tree(current_tree); + read_lock_dk(current_tree); + + assert("vs-1710", znode_is_any_locked(node)); + assert("vs-1197", + !keygt(znode_get_ld_key(node), znode_get_rd_key(node))); + + left = node->left; + right = node->right; + + if (ZF_ISSET(node, JNODE_LEFT_CONNECTED) && ZF_ISSET(node, JNODE_DKSET) + && left != NULL && ZF_ISSET(left, JNODE_DKSET)) + /* check left neighbor. Note that left neighbor is not locked, + so it might get wrong delimiting keys therefore */ + assert("vs-1198", + (keyeq(znode_get_rd_key(left), znode_get_ld_key(node)) + || ZF_ISSET(left, JNODE_HEARD_BANSHEE))); + + if (ZF_ISSET(node, JNODE_RIGHT_CONNECTED) && ZF_ISSET(node, JNODE_DKSET) + && right != NULL && ZF_ISSET(right, JNODE_DKSET)) + /* check right neighbor. Note that right neighbor is not + locked, so it might get wrong delimiting keys therefore */ + assert("vs-1199", + (keyeq(znode_get_rd_key(node), znode_get_ld_key(right)) + || ZF_ISSET(right, JNODE_HEARD_BANSHEE))); + + read_unlock_dk(current_tree); + read_unlock_tree(current_tree); +} +#endif + +/* true if @key is left delimiting key of @node */ +static int key_is_ld(znode * node, const reiser4_key * key) +{ + int ld; + + assert("nikita-1716", node != NULL); + assert("nikita-1758", key != NULL); + + read_lock_dk(znode_get_tree(node)); + assert("nikita-1759", znode_contains_key(node, key)); + ld = keyeq(znode_get_ld_key(node), key); + read_unlock_dk(znode_get_tree(node)); + return ld; +} + +/* Process one node during tree traversal. + + This is called by cbk_level_lookup(). */ +static level_lookup_result cbk_node_lookup(cbk_handle * h/* search handle */) +{ + /* node plugin of @active */ + node_plugin *nplug; + /* item plugin of item that was found */ + item_plugin *iplug; + /* search bias */ + lookup_bias node_bias; + /* node we are operating upon */ + znode *active; + /* tree we are searching in */ + reiser4_tree *tree; + /* result */ + int result; + + assert("nikita-379", h != NULL); + + active = h->active_lh->node; + tree = h->tree; + + nplug = active->nplug; + assert("nikita-380", nplug != NULL); + + ON_DEBUG(check_dkeys(active)); + + /* return item from "active" node with maximal key not greater than + "key" */ + node_bias = h->bias; + result = nplug->lookup(active, h->key, node_bias, h->coord); + if (unlikely(result != NS_FOUND && result != NS_NOT_FOUND)) { + /* error occurred */ + h->result = result; + return LOOKUP_DONE; + } + if (h->level == h->stop_level) { + /* welcome to the stop level */ + assert("nikita-381", h->coord->node == active); + if (result == NS_FOUND) { + /* success of tree lookup */ + if (!(h->flags & CBK_UNIQUE) + && key_is_ld(active, h->key)) + return search_to_left(h); + else + h->result = CBK_COORD_FOUND; + } else { + h->result = CBK_COORD_NOTFOUND; + } + if (!(h->flags & CBK_IN_CACHE)) + cbk_cache_add(active); + return LOOKUP_DONE; + } + + if (h->level > TWIG_LEVEL && result == NS_NOT_FOUND) { + h->error = "not found on internal node"; + h->result = result; + return LOOKUP_DONE; + } + + assert("vs-361", h->level > h->stop_level); + + if (handle_eottl(h, &result)) { + assert("vs-1674", (result == LOOKUP_DONE || + result == LOOKUP_REST)); + return result; + } + + /* go down to next level */ + check_me("vs-12", zload(h->coord->node) == 0); + assert("nikita-2116", item_is_internal(h->coord)); + iplug = item_plugin_by_coord(h->coord); + iplug->s.internal.down_link(h->coord, h->key, &h->block); + zrelse(h->coord->node); + --h->level; + return LOOKUP_CONT; /* continue */ +} + +/* scan cbk_cache slots looking for a match for @h */ +static int cbk_cache_scan_slots(cbk_handle * h/* cbk handle */) +{ + level_lookup_result llr; + znode *node; + reiser4_tree *tree; + cbk_cache_slot *slot; + cbk_cache *cache; + tree_level level; + int isunique; + const reiser4_key *key; + int result; + + assert("nikita-1317", h != NULL); + assert("nikita-1315", h->tree != NULL); + assert("nikita-1316", h->key != NULL); + + tree = h->tree; + cache = &tree->cbk_cache; + if (cache->nr_slots == 0) + /* size of cbk cache was set to 0 by mount time option. */ + return RETERR(-ENOENT); + + assert("nikita-2474", cbk_cache_invariant(cache)); + node = NULL; /* to keep gcc happy */ + level = h->level; + key = h->key; + isunique = h->flags & CBK_UNIQUE; + result = RETERR(-ENOENT); + + /* + * this is time-critical function and dragons had, hence, been settled + * here. + * + * Loop below scans cbk cache slots trying to find matching node with + * suitable range of delimiting keys and located at the h->level. + * + * Scan is done under cbk cache spin lock that protects slot->node + * pointers. If suitable node is found we want to pin it in + * memory. But slot->node can point to the node with x_count 0 + * (unreferenced). Such node can be recycled at any moment, or can + * already be in the process of being recycled (within jput()). + * + * As we found node in the cbk cache, it means that jput() hasn't yet + * called cbk_cache_invalidate(). + * + * We acquire reference to the node without holding tree lock, and + * later, check node's RIP bit. This avoids races with jput(). + */ + + rcu_read_lock(); + read_lock(&((cbk_cache *)cache)->guard); + + slot = list_entry(cache->lru.next, cbk_cache_slot, lru); + slot = list_entry(slot->lru.prev, cbk_cache_slot, lru); + BUG_ON(&slot->lru != &cache->lru);/*????*/ + while (1) { + + slot = list_entry(slot->lru.next, cbk_cache_slot, lru); + + if (&cache->lru != &slot->lru) + node = slot->node; + else + node = NULL; + + if (unlikely(node == NULL)) + break; + + /* + * this is (hopefully) the only place in the code where we are + * working with delimiting keys without holding dk lock. This + * is fine here, because this is only "guess" anyway---keys + * are rechecked under dk lock below. + */ + if (znode_get_level(node) == level && + /* reiser4_min_key < key < reiser4_max_key */ + znode_contains_key_strict(node, key, isunique)) { + zref(node); + result = 0; + spin_lock_prefetch(&tree->tree_lock); + break; + } + } + read_unlock(&((cbk_cache *)cache)->guard); + + assert("nikita-2475", cbk_cache_invariant(cache)); + + if (unlikely(result == 0 && ZF_ISSET(node, JNODE_RIP))) + result = -ENOENT; + + rcu_read_unlock(); + + if (result != 0) { + h->result = CBK_COORD_NOTFOUND; + return RETERR(-ENOENT); + } + + result = + longterm_lock_znode(h->active_lh, node, cbk_lock_mode(level, h), + ZNODE_LOCK_LOPRI); + zput(node); + if (result != 0) + return result; + result = zload(node); + if (result != 0) + return result; + + /* recheck keys */ + read_lock_dk(tree); + result = (znode_contains_key_strict(node, key, isunique) && + !ZF_ISSET(node, JNODE_HEARD_BANSHEE)); + read_unlock_dk(tree); + if (result) { + /* do lookup inside node */ + llr = cbk_node_lookup(h); + /* if cbk_node_lookup() wandered to another node (due to eottl + or non-unique keys), adjust @node */ + /*node = h->active_lh->node; */ + + if (llr != LOOKUP_DONE) { + /* restart or continue on the next level */ + result = RETERR(-ENOENT); + } else if (IS_CBKERR(h->result)) + /* io or oom */ + result = RETERR(-ENOENT); + else { + /* good. Either item found or definitely not found. */ + result = 0; + + write_lock(&(cache->guard)); + if (slot->node == h->active_lh->node) { + /* if this node is still in cbk cache---move + its slot to the head of the LRU list. */ + list_move(&slot->lru, &cache->lru); + } + write_unlock(&(cache->guard)); + } + } else { + /* race. While this thread was waiting for the lock, node was + rebalanced and item we are looking for, shifted out of it + (if it ever was here). + + Continuing scanning is almost hopeless: node key range was + moved to, is almost certainly at the beginning of the LRU + list at this time, because it's hot, but restarting + scanning from the very beginning is complex. Just return, + so that cbk() will be performed. This is not that + important, because such races should be rare. Are they? + */ + result = RETERR(-ENOENT); /* -ERAUGHT */ + } + zrelse(node); + assert("nikita-2476", cbk_cache_invariant(cache)); + return result; +} + +/* look for item with given key in the coord cache + + This function, called by coord_by_key(), scans "coord cache" (&cbk_cache) + which is a small LRU list of znodes accessed lately. For each znode in + znode in this list, it checks whether key we are looking for fits into key + range covered by this node. If so, and in addition, node lies at allowed + level (this is to handle extents on a twig level), node is locked, and + lookup inside it is performed. + + we need a measurement of the cost of this cache search compared to the cost + of coord_by_key. + +*/ +static int cbk_cache_search(cbk_handle * h/* cbk handle */) +{ + int result = 0; + tree_level level; + + /* add CBK_IN_CACHE to the handle flags. This means that + * cbk_node_lookup() assumes that cbk_cache is scanned and would add + * found node to the cache. */ + h->flags |= CBK_IN_CACHE; + for (level = h->stop_level; level <= h->lock_level; ++level) { + h->level = level; + result = cbk_cache_scan_slots(h); + if (result != 0) { + done_lh(h->active_lh); + done_lh(h->parent_lh); + } else { + assert("nikita-1319", !IS_CBKERR(h->result)); + break; + } + } + h->flags &= ~CBK_IN_CACHE; + return result; +} + +/* type of lock we want to obtain during tree traversal. On stop level + we want type of lock user asked for, on upper levels: read lock. */ +znode_lock_mode cbk_lock_mode(tree_level level, cbk_handle * h) +{ + assert("nikita-382", h != NULL); + + return (level <= h->lock_level) ? h->lock_mode : ZNODE_READ_LOCK; +} + +/* update outdated delimiting keys */ +static void stale_dk(reiser4_tree * tree, znode * node) +{ + znode *right; + + read_lock_tree(tree); + write_lock_dk(tree); + right = node->right; + + if (ZF_ISSET(node, JNODE_RIGHT_CONNECTED) && + right && ZF_ISSET(right, JNODE_DKSET) && + !keyeq(znode_get_rd_key(node), znode_get_ld_key(right))) + znode_set_rd_key(node, znode_get_ld_key(right)); + + write_unlock_dk(tree); + read_unlock_tree(tree); +} + +/* check for possibly outdated delimiting keys, and update them if + * necessary. */ +static void update_stale_dk(reiser4_tree * tree, znode * node) +{ + znode *right; + reiser4_key rd; + + read_lock_tree(tree); + read_lock_dk(tree); + rd = *znode_get_rd_key(node); + right = node->right; + if (unlikely(ZF_ISSET(node, JNODE_RIGHT_CONNECTED) && + right && ZF_ISSET(right, JNODE_DKSET) && + !keyeq(&rd, znode_get_ld_key(right)))) { + assert("nikita-38211", ZF_ISSET(node, JNODE_DKSET)); + read_unlock_dk(tree); + read_unlock_tree(tree); + stale_dk(tree, node); + return; + } + read_unlock_dk(tree); + read_unlock_tree(tree); +} + +/* + * handle searches a the non-unique key. + * + * Suppose that we are looking for an item with possibly non-unique key 100. + * + * Root node contains two pointers: one to a node with left delimiting key 0, + * and another to a node with left delimiting key 100. Item we interested in + * may well happen in the sub-tree rooted at the first pointer. + * + * To handle this search_to_left() is called when search reaches stop + * level. This function checks it is _possible_ that item we are looking for + * is in the left neighbor (this can be done by comparing delimiting keys) and + * if so, tries to lock left neighbor (this is low priority lock, so it can + * deadlock, tree traversal is just restarted if it did) and then checks + * whether left neighbor actually contains items with our key. + * + * Note that this is done on the stop level only. It is possible to try such + * left-check on each level, but as duplicate keys are supposed to be rare + * (very unlikely that more than one node is completely filled with items with + * duplicate keys), it sis cheaper to scan to the left on the stop level once. + * + */ +static level_lookup_result search_to_left(cbk_handle * h/* search handle */) +{ + level_lookup_result result; + coord_t *coord; + znode *node; + znode *neighbor; + + lock_handle lh; + + assert("nikita-1761", h != NULL); + assert("nikita-1762", h->level == h->stop_level); + + init_lh(&lh); + coord = h->coord; + node = h->active_lh->node; + assert("nikita-1763", coord_is_leftmost_unit(coord)); + + h->result = + reiser4_get_left_neighbor(&lh, node, (int)h->lock_mode, + GN_CAN_USE_UPPER_LEVELS); + neighbor = NULL; + switch (h->result) { + case -E_DEADLOCK: + result = LOOKUP_REST; + break; + case 0:{ + node_plugin *nplug; + coord_t crd; + lookup_bias bias; + + neighbor = lh.node; + h->result = zload(neighbor); + if (h->result != 0) { + result = LOOKUP_DONE; + break; + } + + nplug = neighbor->nplug; + + coord_init_zero(&crd); + bias = h->bias; + h->bias = FIND_EXACT; + h->result = + nplug->lookup(neighbor, h->key, h->bias, &crd); + h->bias = bias; + + if (h->result == NS_NOT_FOUND) { + case -E_NO_NEIGHBOR: + h->result = CBK_COORD_FOUND; + if (!(h->flags & CBK_IN_CACHE)) + cbk_cache_add(node); + default: /* some other error */ + result = LOOKUP_DONE; + } else if (h->result == NS_FOUND) { + read_lock_dk(znode_get_tree(neighbor)); + h->rd_key = *znode_get_ld_key(node); + leftmost_key_in_node(neighbor, &h->ld_key); + read_unlock_dk(znode_get_tree(neighbor)); + h->flags |= CBK_DKSET; + + h->block = *znode_get_block(neighbor); + /* clear coord->node so that cbk_level_lookup() + wouldn't overwrite parent hint in neighbor. + + Parent hint was set up by + reiser4_get_left_neighbor() + */ + /* FIXME: why do we have to spinlock here? */ + write_lock_tree(znode_get_tree(neighbor)); + h->coord->node = NULL; + write_unlock_tree(znode_get_tree(neighbor)); + result = LOOKUP_CONT; + } else { + result = LOOKUP_DONE; + } + if (neighbor != NULL) + zrelse(neighbor); + } + } + done_lh(&lh); + return result; +} + +/* debugging aid: return symbolic name of search bias */ +static const char *bias_name(lookup_bias bias/* bias to get name of */) +{ + if (bias == FIND_EXACT) + return "exact"; + else if (bias == FIND_MAX_NOT_MORE_THAN) + return "left-slant"; +/* else if( bias == RIGHT_SLANT_BIAS ) */ +/* return "right-bias"; */ + else { + static char buf[30]; + + sprintf(buf, "unknown: %i", bias); + return buf; + } +} + +#if REISER4_DEBUG +/* debugging aid: print human readable information about @p */ +void print_coord_content(const char *prefix /* prefix to print */ , + coord_t *p/* coord to print */) +{ + reiser4_key key; + + if (p == NULL) { + printk("%s: null\n", prefix); + return; + } + if ((p->node != NULL) && znode_is_loaded(p->node) + && coord_is_existing_item(p)) + printk("%s: data: %p, length: %i\n", prefix, + item_body_by_coord(p), item_length_by_coord(p)); + if (znode_is_loaded(p->node)) { + item_key_by_coord(p, &key); + reiser4_print_key(prefix, &key); + } +} + +/* debugging aid: print human readable information about @block */ +void reiser4_print_address(const char *prefix /* prefix to print */ , + const reiser4_block_nr * block/* block number to print */) +{ + printk("%s: %s\n", prefix, sprint_address(block)); +} +#endif + +/* return string containing human readable representation of @block */ +char *sprint_address(const reiser4_block_nr * + block/* block number to print */) +{ + static char address[30]; + + if (block == NULL) + sprintf(address, "null"); + else if (reiser4_blocknr_is_fake(block)) + sprintf(address, "%llx", (unsigned long long)(*block)); + else + sprintf(address, "%llu", (unsigned long long)(*block)); + return address; +} + +/* release parent node during traversal */ +static void put_parent(cbk_handle * h/* search handle */) +{ + assert("nikita-383", h != NULL); + if (h->parent_lh->node != NULL) + longterm_unlock_znode(h->parent_lh); +} + +/* helper function used by coord_by_key(): release reference to parent znode + stored in handle before processing its child. */ +static void hput(cbk_handle * h/* search handle */) +{ + assert("nikita-385", h != NULL); + done_lh(h->parent_lh); + done_lh(h->active_lh); +} + +/* Helper function used by cbk(): update delimiting keys of child node (stored + in h->active_lh->node) using key taken from parent on the parent level. */ +static int setup_delimiting_keys(cbk_handle * h/* search handle */) +{ + znode *active; + reiser4_tree *tree; + + assert("nikita-1088", h != NULL); + + active = h->active_lh->node; + + /* fast check without taking dk lock. This is safe, because + * JNODE_DKSET is never cleared once set. */ + if (!ZF_ISSET(active, JNODE_DKSET)) { + tree = znode_get_tree(active); + write_lock_dk(tree); + if (!ZF_ISSET(active, JNODE_DKSET)) { + znode_set_ld_key(active, &h->ld_key); + znode_set_rd_key(active, &h->rd_key); + ZF_SET(active, JNODE_DKSET); + } + write_unlock_dk(tree); + return 1; + } + return 0; +} + +/* true if @block makes sense for the @tree. Used to detect corrupted node + * pointers */ +static int +block_nr_is_correct(reiser4_block_nr * block /* block number to check */ , + reiser4_tree * tree/* tree to check against */) +{ + assert("nikita-757", block != NULL); + assert("nikita-758", tree != NULL); + + /* check to see if it exceeds the size of the device. */ + return reiser4_blocknr_is_sane_for(tree->super, block); +} + +/* check consistency of fields */ +static int sanity_check(cbk_handle * h/* search handle */) +{ + assert("nikita-384", h != NULL); + + if (h->level < h->stop_level) { + h->error = "Buried under leaves"; + h->result = RETERR(-EIO); + return LOOKUP_DONE; + } else if (!block_nr_is_correct(&h->block, h->tree)) { + h->error = "bad block number"; + h->result = RETERR(-EIO); + return LOOKUP_DONE; + } else + return 0; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/status_flags.c linux-2.6.30/fs/reiser4/status_flags.c --- linux-2.6.30.orig/fs/reiser4/status_flags.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/status_flags.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,174 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Functions that deal with reiser4 status block, query status and update it, + * if needed */ + +#include +#include +#include +#include +#include "debug.h" +#include "dformat.h" +#include "status_flags.h" +#include "super.h" + +/* This is our end I/O handler that marks page uptodate if IO was successful. + It also unconditionally unlocks the page, so we can see that io was done. + We do not free bio, because we hope to reuse that. */ +static void reiser4_status_endio(struct bio *bio, int err) +{ + if (test_bit(BIO_UPTODATE, &bio->bi_flags)) { + SetPageUptodate(bio->bi_io_vec->bv_page); + } else { + ClearPageUptodate(bio->bi_io_vec->bv_page); + SetPageError(bio->bi_io_vec->bv_page); + } + unlock_page(bio->bi_io_vec->bv_page); +} + +/* Initialise status code. This is expected to be called from the disk format + code. block paremeter is where status block lives. */ +int reiser4_status_init(reiser4_block_nr block) +{ + struct super_block *sb = reiser4_get_current_sb(); + struct reiser4_status *statuspage; + struct bio *bio; + struct page *page; + + get_super_private(sb)->status_page = NULL; + get_super_private(sb)->status_bio = NULL; + + page = alloc_pages(reiser4_ctx_gfp_mask_get(), 0); + if (!page) + return -ENOMEM; + + bio = bio_alloc(reiser4_ctx_gfp_mask_get(), 1); + if (bio != NULL) { + bio->bi_sector = block * (sb->s_blocksize >> 9); + bio->bi_bdev = sb->s_bdev; + bio->bi_io_vec[0].bv_page = page; + bio->bi_io_vec[0].bv_len = sb->s_blocksize; + bio->bi_io_vec[0].bv_offset = 0; + bio->bi_vcnt = 1; + bio->bi_size = sb->s_blocksize; + bio->bi_end_io = reiser4_status_endio; + } else { + __free_pages(page, 0); + return -ENOMEM; + } + lock_page(page); + submit_bio(READ, bio); + blk_run_address_space(reiser4_get_super_fake(sb)->i_mapping); + wait_on_page_locked(page); + if (!PageUptodate(page)) { + warning("green-2007", + "I/O error while tried to read status page\n"); + return -EIO; + } + + statuspage = (struct reiser4_status *)kmap_atomic(page, KM_USER0); + if (memcmp + (statuspage->magic, REISER4_STATUS_MAGIC, + sizeof(REISER4_STATUS_MAGIC))) { + /* Magic does not match. */ + kunmap_atomic((char *)statuspage, KM_USER0); + warning("green-2008", "Wrong magic in status block\n"); + __free_pages(page, 0); + bio_put(bio); + return -EINVAL; + } + kunmap_atomic((char *)statuspage, KM_USER0); + + get_super_private(sb)->status_page = page; + get_super_private(sb)->status_bio = bio; + return 0; +} + +/* Query the status of fs. Returns if the FS can be safely mounted. + Also if "status" and "extended" parameters are given, it will fill + actual parts of status from disk there. */ +int reiser4_status_query(u64 *status, u64 *extended) +{ + struct super_block *sb = reiser4_get_current_sb(); + struct reiser4_status *statuspage; + int retval; + + if (!get_super_private(sb)->status_page) + /* No status page? */ + return REISER4_STATUS_MOUNT_UNKNOWN; + statuspage = (struct reiser4_status *) + kmap_atomic(get_super_private(sb)->status_page, KM_USER0); + switch ((long)le64_to_cpu(get_unaligned(&statuspage->status))) { + /* FIXME: this cast is a hack for 32 bit arches to work. */ + case REISER4_STATUS_OK: + retval = REISER4_STATUS_MOUNT_OK; + break; + case REISER4_STATUS_CORRUPTED: + retval = REISER4_STATUS_MOUNT_WARN; + break; + case REISER4_STATUS_DAMAGED: + case REISER4_STATUS_DESTROYED: + case REISER4_STATUS_IOERROR: + retval = REISER4_STATUS_MOUNT_RO; + break; + default: + retval = REISER4_STATUS_MOUNT_UNKNOWN; + break; + } + + if (status) + *status = le64_to_cpu(get_unaligned(&statuspage->status)); + if (extended) + *extended = le64_to_cpu(get_unaligned(&statuspage->extended_status)); + + kunmap_atomic((char *)statuspage, KM_USER0); + return retval; +} + +/* This function should be called when something bad happens (e.g. from + reiser4_panic). It fills the status structure and tries to push it to disk.*/ +int reiser4_status_write(__u64 status, __u64 extended_status, char *message) +{ + struct super_block *sb = reiser4_get_current_sb(); + struct reiser4_status *statuspage; + struct bio *bio = get_super_private(sb)->status_bio; + + if (!get_super_private(sb)->status_page) + /* No status page? */ + return -1; + statuspage = (struct reiser4_status *) + kmap_atomic(get_super_private(sb)->status_page, KM_USER0); + + put_unaligned(cpu_to_le64(status), &statuspage->status); + put_unaligned(cpu_to_le64(extended_status), &statuspage->extended_status); + strncpy(statuspage->texterror, message, REISER4_TEXTERROR_LEN); + + kunmap_atomic((char *)statuspage, KM_USER0); + bio->bi_bdev = sb->s_bdev; + bio->bi_io_vec[0].bv_page = get_super_private(sb)->status_page; + bio->bi_io_vec[0].bv_len = sb->s_blocksize; + bio->bi_io_vec[0].bv_offset = 0; + bio->bi_vcnt = 1; + bio->bi_size = sb->s_blocksize; + bio->bi_end_io = reiser4_status_endio; + lock_page(get_super_private(sb)->status_page); /* Safe as nobody should + * touch our page. */ + /* We can block now, but we have no other choice anyway */ + submit_bio(WRITE, bio); + blk_run_address_space(reiser4_get_super_fake(sb)->i_mapping); + return 0; /* We do not wait for io to finish. */ +} + +/* Frees the page with status and bio structure. Should be called by disk format + * at umount time */ +int reiser4_status_finish(void) +{ + struct super_block *sb = reiser4_get_current_sb(); + + __free_pages(get_super_private(sb)->status_page, 0); + get_super_private(sb)->status_page = NULL; + bio_put(get_super_private(sb)->status_bio); + get_super_private(sb)->status_bio = NULL; + return 0; +} diff -urN linux-2.6.30.orig/fs/reiser4/status_flags.h linux-2.6.30/fs/reiser4/status_flags.h --- linux-2.6.30.orig/fs/reiser4/status_flags.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/status_flags.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,47 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Here we declare structures and flags that store reiser4 status on disk. + The status that helps us to find out if the filesystem is valid or if it + contains some critical, or not so critical errors */ + +#if !defined(__REISER4_STATUS_FLAGS_H__) +#define __REISER4_STATUS_FLAGS_H__ + +#include "dformat.h" +/* These are major status flags */ +#define REISER4_STATUS_OK 0 +#define REISER4_STATUS_CORRUPTED 0x1 +#define REISER4_STATUS_DAMAGED 0x2 +#define REISER4_STATUS_DESTROYED 0x4 +#define REISER4_STATUS_IOERROR 0x8 + +/* Return values for reiser4_status_query() */ +#define REISER4_STATUS_MOUNT_OK 0 +#define REISER4_STATUS_MOUNT_WARN 1 +#define REISER4_STATUS_MOUNT_RO 2 +#define REISER4_STATUS_MOUNT_UNKNOWN -1 + +#define REISER4_TEXTERROR_LEN 256 + +#define REISER4_STATUS_MAGIC "ReiSeR4StATusBl" +/* We probably need to keep its size under sector size which is 512 bytes */ +struct reiser4_status { + char magic[16]; + d64 status; /* Current FS state */ + d64 extended_status; /* Any additional info that might have sense in + * addition to "status". E.g. last sector where + * io error happened if status is + * "io error encountered" */ + d64 stacktrace[10]; /* Last ten functional calls made (addresses) */ + char texterror[REISER4_TEXTERROR_LEN]; /* Any error message if + * appropriate, otherwise filled + * with zeroes */ +}; + +int reiser4_status_init(reiser4_block_nr block); +int reiser4_status_query(u64 *status, u64 *extended); +int reiser4_status_write(u64 status, u64 extended_status, char *message); +int reiser4_status_finish(void); + +#endif diff -urN linux-2.6.30.orig/fs/reiser4/super.c linux-2.6.30/fs/reiser4/super.c --- linux-2.6.30.orig/fs/reiser4/super.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/super.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,306 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Super-block manipulations. */ + +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "plugin/security/perm.h" +#include "plugin/space/space_allocator.h" +#include "plugin/plugin.h" +#include "tree.h" +#include "vfs_ops.h" +#include "super.h" +#include "reiser4.h" + +#include /* for __u?? */ +#include /* for struct super_block */ + +static __u64 reserved_for_gid(const struct super_block *super, gid_t gid); +static __u64 reserved_for_uid(const struct super_block *super, uid_t uid); +static __u64 reserved_for_root(const struct super_block *super); + +/* Return reiser4-specific part of super block */ +reiser4_super_info_data *get_super_private_nocheck(const struct super_block *super) +{ + return (reiser4_super_info_data *) super->s_fs_info; +} + +/* Return reiser4 fstype: value that is returned in ->f_type field by statfs() + */ +long reiser4_statfs_type(const struct super_block *super UNUSED_ARG) +{ + assert("nikita-448", super != NULL); + assert("nikita-449", is_reiser4_super(super)); + return (long)REISER4_SUPER_MAGIC; +} + +/* functions to read/modify fields of reiser4_super_info_data */ + +/* get number of blocks in file system */ +__u64 reiser4_block_count(const struct super_block *super /* super block + queried */ ) +{ + assert("vs-494", super != NULL); + assert("vs-495", is_reiser4_super(super)); + return get_super_private(super)->block_count; +} + +#if REISER4_DEBUG +/* + * number of blocks in the current file system + */ +__u64 reiser4_current_block_count(void) +{ + return get_current_super_private()->block_count; +} +#endif /* REISER4_DEBUG */ + +/* set number of block in filesystem */ +void reiser4_set_block_count(const struct super_block *super, __u64 nr) +{ + assert("vs-501", super != NULL); + assert("vs-502", is_reiser4_super(super)); + get_super_private(super)->block_count = nr; + /* + * The proper calculation of the reserved space counter (%5 of device + * block counter) we need a 64 bit division which is missing in Linux + * on i386 platform. Because we do not need a precise calculation here + * we can replace a div64 operation by this combination of + * multiplication and shift: 51. / (2^10) == .0498 . + * FIXME: this is a bug. It comes up only for very small filesystems + * which probably are never used. Nevertheless, it is a bug. Number of + * reserved blocks must be not less than maximal number of blocks which + * get grabbed with BA_RESERVED. + */ + get_super_private(super)->blocks_reserved = ((nr * 51) >> 10); +} + +/* amount of blocks used (allocated for data) in file system */ +__u64 reiser4_data_blocks(const struct super_block *super /* super block + queried */ ) +{ + assert("nikita-452", super != NULL); + assert("nikita-453", is_reiser4_super(super)); + return get_super_private(super)->blocks_used; +} + +/* set number of block used in filesystem */ +void reiser4_set_data_blocks(const struct super_block *super, __u64 nr) +{ + assert("vs-503", super != NULL); + assert("vs-504", is_reiser4_super(super)); + get_super_private(super)->blocks_used = nr; +} + +/* amount of free blocks in file system */ +__u64 reiser4_free_blocks(const struct super_block *super /* super block + queried */ ) +{ + assert("nikita-454", super != NULL); + assert("nikita-455", is_reiser4_super(super)); + return get_super_private(super)->blocks_free; +} + +/* set number of blocks free in filesystem */ +void reiser4_set_free_blocks(const struct super_block *super, __u64 nr) +{ + assert("vs-505", super != NULL); + assert("vs-506", is_reiser4_super(super)); + get_super_private(super)->blocks_free = nr; +} + +/* get mkfs unique identifier */ +__u32 reiser4_mkfs_id(const struct super_block *super /* super block + queried */ ) +{ + assert("vpf-221", super != NULL); + assert("vpf-222", is_reiser4_super(super)); + return get_super_private(super)->mkfs_id; +} + +/* amount of free blocks in file system */ +__u64 reiser4_free_committed_blocks(const struct super_block *super) +{ + assert("vs-497", super != NULL); + assert("vs-498", is_reiser4_super(super)); + return get_super_private(super)->blocks_free_committed; +} + +/* amount of blocks in the file system reserved for @uid and @gid */ +long reiser4_reserved_blocks(const struct super_block *super /* super block + queried */ , + uid_t uid /* user id */ , + gid_t gid/* group id */) +{ + long reserved; + + assert("nikita-456", super != NULL); + assert("nikita-457", is_reiser4_super(super)); + + reserved = 0; + if (REISER4_SUPPORT_GID_SPACE_RESERVATION) + reserved += reserved_for_gid(super, gid); + if (REISER4_SUPPORT_UID_SPACE_RESERVATION) + reserved += reserved_for_uid(super, uid); + if (REISER4_SUPPORT_ROOT_SPACE_RESERVATION && (uid == 0)) + reserved += reserved_for_root(super); + return reserved; +} + +/* get/set value of/to grabbed blocks counter */ +__u64 reiser4_grabbed_blocks(const struct super_block * super) +{ + assert("zam-512", super != NULL); + assert("zam-513", is_reiser4_super(super)); + + return get_super_private(super)->blocks_grabbed; +} + +__u64 reiser4_flush_reserved(const struct super_block *super) +{ + assert("vpf-285", super != NULL); + assert("vpf-286", is_reiser4_super(super)); + + return get_super_private(super)->blocks_flush_reserved; +} + +/* get/set value of/to counter of fake allocated formatted blocks */ +__u64 reiser4_fake_allocated(const struct super_block *super) +{ + assert("zam-516", super != NULL); + assert("zam-517", is_reiser4_super(super)); + + return get_super_private(super)->blocks_fake_allocated; +} + +/* get/set value of/to counter of fake allocated unformatted blocks */ +__u64 reiser4_fake_allocated_unformatted(const struct super_block *super) +{ + assert("zam-516", super != NULL); + assert("zam-517", is_reiser4_super(super)); + + return get_super_private(super)->blocks_fake_allocated_unformatted; +} + +/* get/set value of/to counter of clustered blocks */ +__u64 reiser4_clustered_blocks(const struct super_block *super) +{ + assert("edward-601", super != NULL); + assert("edward-602", is_reiser4_super(super)); + + return get_super_private(super)->blocks_clustered; +} + +/* space allocator used by this file system */ +reiser4_space_allocator * reiser4_get_space_allocator(const struct super_block + *super) +{ + assert("nikita-1965", super != NULL); + assert("nikita-1966", is_reiser4_super(super)); + return &get_super_private(super)->space_allocator; +} + +/* return fake inode used to bind formatted nodes in the page cache */ +struct inode *reiser4_get_super_fake(const struct super_block *super) +{ + assert("nikita-1757", super != NULL); + return get_super_private(super)->fake; +} + +/* return fake inode used to bind copied on capture nodes in the page cache */ +struct inode *reiser4_get_cc_fake(const struct super_block *super) +{ + assert("nikita-1757", super != NULL); + return get_super_private(super)->cc; +} + +/* return fake inode used to bind bitmaps and journlal heads */ +struct inode *reiser4_get_bitmap_fake(const struct super_block *super) +{ + assert("nikita-17571", super != NULL); + return get_super_private(super)->bitmap; +} + +/* tree used by this file system */ +reiser4_tree *reiser4_get_tree(const struct super_block *super) +{ + assert("nikita-460", super != NULL); + assert("nikita-461", is_reiser4_super(super)); + return &get_super_private(super)->tree; +} + +/* Check that @super is (looks like) reiser4 super block. This is mainly for + use in assertions. */ +int is_reiser4_super(const struct super_block *super) +{ + return + super != NULL && + get_super_private(super) != NULL && + super->s_op == &(get_super_private(super)->ops.super); +} + +int reiser4_is_set(const struct super_block *super, reiser4_fs_flag f) +{ + return test_bit((int)f, &get_super_private(super)->fs_flags); +} + +/* amount of blocks reserved for given group in file system */ +static __u64 reserved_for_gid(const struct super_block *super UNUSED_ARG, + gid_t gid UNUSED_ARG/* group id */) +{ + return 0; +} + +/* amount of blocks reserved for given user in file system */ +static __u64 reserved_for_uid(const struct super_block *super UNUSED_ARG, + uid_t uid UNUSED_ARG/* user id */) +{ + return 0; +} + +/* amount of blocks reserved for super user in file system */ +static __u64 reserved_for_root(const struct super_block *super UNUSED_ARG) +{ + return 0; +} + +/* + * true if block number @blk makes sense for the file system at @super. + */ +int +reiser4_blocknr_is_sane_for(const struct super_block *super, + const reiser4_block_nr * blk) +{ + reiser4_super_info_data *sbinfo; + + assert("nikita-2957", super != NULL); + assert("nikita-2958", blk != NULL); + + if (reiser4_blocknr_is_fake(blk)) + return 1; + + sbinfo = get_super_private(super); + return *blk < sbinfo->block_count; +} + +#if REISER4_DEBUG +/* + * true, if block number @blk makes sense for the current file system + */ +int reiser4_blocknr_is_sane(const reiser4_block_nr * blk) +{ + return reiser4_blocknr_is_sane_for(reiser4_get_current_sb(), blk); +} +#endif /* REISER4_DEBUG */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/super.h linux-2.6.30/fs/reiser4/super.h --- linux-2.6.30.orig/fs/reiser4/super.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/super.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,466 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Super-block functions. See super.c for details. */ + +#if !defined(__REISER4_SUPER_H__) +#define __REISER4_SUPER_H__ + +#include + +#include "tree.h" +#include "entd.h" +#include "wander.h" +#include "fsdata.h" +#include "plugin/object.h" +#include "plugin/space/space_allocator.h" + +/* + * Flush algorithms parameters. + */ +struct flush_params { + unsigned relocate_threshold; + unsigned relocate_distance; + unsigned written_threshold; + unsigned scan_maxnodes; +}; + +typedef enum { + /* + * True if this file system doesn't support hard-links (multiple names) + * for directories: this is default UNIX behavior. + * + * If hard-links on directoires are not allowed, file system is Acyclic + * Directed Graph (modulo dot, and dotdot, of course). + * + * This is used by reiser4_link(). + */ + REISER4_ADG = 0, + /* + * set if all nodes in internal tree have the same node layout plugin. + * If so, znode_guess_plugin() will return tree->node_plugin in stead + * of guessing plugin by plugin id stored in the node. + */ + REISER4_ONE_NODE_PLUGIN = 1, + /* if set, bsd gid assignment is supported. */ + REISER4_BSD_GID = 2, + /* [mac]_time are 32 bit in inode */ + REISER4_32_BIT_TIMES = 3, + /* load all bitmap blocks at mount time */ + REISER4_DONT_LOAD_BITMAP = 5, + /* enforce atomicity during write(2) */ + REISER4_ATOMIC_WRITE = 6, + /* don't use write barriers in the log writer code. */ + REISER4_NO_WRITE_BARRIER = 7 +} reiser4_fs_flag; + +/* + * VFS related operation vectors. + */ +struct object_ops { + struct super_operations super; + struct dentry_operations dentry; + struct export_operations export; +}; + +/* reiser4-specific part of super block + + Locking + + Fields immutable after mount: + + ->oid* + ->space* + ->default_[ug]id + ->mkfs_id + ->trace_flags + ->debug_flags + ->fs_flags + ->df_plug + ->optimal_io_size + ->plug + ->flush + ->u (bad name) + ->txnmgr + ->ra_params + ->fsuid + ->journal_header + ->journal_footer + + Fields protected by ->lnode_guard + + ->lnode_htable + + Fields protected by per-super block spin lock + + ->block_count + ->blocks_used + ->blocks_free + ->blocks_free_committed + ->blocks_grabbed + ->blocks_fake_allocated_unformatted + ->blocks_fake_allocated + ->blocks_flush_reserved + ->eflushed + ->blocknr_hint_default + + After journal replaying during mount, + + ->last_committed_tx + + is protected by ->tmgr.commit_mutex + + Invariants involving this data-type: + + [sb-block-counts] + [sb-grabbed] + [sb-fake-allocated] +*/ +struct reiser4_super_info_data { + /* + * guard spinlock which protects reiser4 super block fields (currently + * blocks_free, blocks_free_committed) + */ + spinlock_t guard; + + /* next oid that will be returned by oid_allocate() */ + oid_t next_to_use; + /* total number of used oids */ + oid_t oids_in_use; + + /* space manager plugin */ + reiser4_space_allocator space_allocator; + + /* reiser4 internal tree */ + reiser4_tree tree; + + /* + * default user id used for light-weight files without their own + * stat-data. + */ + uid_t default_uid; + + /* + * default group id used for light-weight files without their own + * stat-data. + */ + gid_t default_gid; + + /* mkfs identifier generated at mkfs time. */ + __u32 mkfs_id; + /* amount of blocks in a file system */ + __u64 block_count; + + /* inviolable reserve */ + __u64 blocks_reserved; + + /* amount of blocks used by file system data and meta-data. */ + __u64 blocks_used; + + /* + * amount of free blocks. This is "working" free blocks counter. It is + * like "working" bitmap, please see block_alloc.c for description. + */ + __u64 blocks_free; + + /* + * free block count for fs committed state. This is "commit" version of + * free block counter. + */ + __u64 blocks_free_committed; + + /* + * number of blocks reserved for further allocation, for all + * threads. + */ + __u64 blocks_grabbed; + + /* number of fake allocated unformatted blocks in tree. */ + __u64 blocks_fake_allocated_unformatted; + + /* number of fake allocated formatted blocks in tree. */ + __u64 blocks_fake_allocated; + + /* number of blocks reserved for flush operations. */ + __u64 blocks_flush_reserved; + + /* number of blocks reserved for cluster operations. */ + __u64 blocks_clustered; + + /* unique file-system identifier */ + __u32 fsuid; + + /* On-disk format version. If does not equal to the disk_format + plugin version, some format updates (e.g. enlarging plugin + set, etc) may have place on mount. */ + int version; + + /* file-system wide flags. See reiser4_fs_flag enum */ + unsigned long fs_flags; + + /* transaction manager */ + txn_mgr tmgr; + + /* ent thread */ + entd_context entd; + + /* fake inode used to bind formatted nodes */ + struct inode *fake; + /* inode used to bind bitmaps (and journal heads) */ + struct inode *bitmap; + /* inode used to bind copied on capture nodes */ + struct inode *cc; + + /* disk layout plugin */ + disk_format_plugin *df_plug; + + /* disk layout specific part of reiser4 super info data */ + union { + format40_super_info format40; + } u; + + /* value we return in st_blksize on stat(2) */ + unsigned long optimal_io_size; + + /* parameters for the flush algorithm */ + struct flush_params flush; + + /* pointers to jnodes for journal header and footer */ + jnode *journal_header; + jnode *journal_footer; + + journal_location jloc; + + /* head block number of last committed transaction */ + __u64 last_committed_tx; + + /* + * we remember last written location for using as a hint for new block + * allocation + */ + __u64 blocknr_hint_default; + + /* committed number of files (oid allocator state variable ) */ + __u64 nr_files_committed; + + struct formatted_ra_params ra_params; + + /* + * A mutex for serializing cut tree operation if out-of-free-space: + * the only one cut_tree thread is allowed to grab space from reserved + * area (it is 5% of disk space) + */ + struct mutex delete_mutex; + /* task owning ->delete_mutex */ + struct task_struct *delete_mutex_owner; + + /* Diskmap's blocknumber */ + __u64 diskmap_block; + + /* What to do in case of error */ + int onerror; + + /* operations for objects on this file system */ + struct object_ops ops; + + /* + * structure to maintain d_cursors. See plugin/file_ops_readdir.c for + * more details + */ + struct d_cursor_info d_info; + +#ifdef CONFIG_REISER4_BADBLOCKS + /* Alternative master superblock offset (in bytes) */ + unsigned long altsuper; +#endif + struct repacker *repacker; + struct page *status_page; + struct bio *status_bio; + +#if REISER4_DEBUG + /* + * minimum used blocks value (includes super blocks, bitmap blocks and + * other fs reserved areas), depends on fs format and fs size. + */ + __u64 min_blocks_used; + + /* + * when debugging is on, all jnodes (including znodes, bitmaps, etc.) + * are kept on a list anchored at sbinfo->all_jnodes. This list is + * protected by sbinfo->all_guard spin lock. This lock should be taken + * with _irq modifier, because it is also modified from interrupt + * contexts (by RCU). + */ + spinlock_t all_guard; + /* list of all jnodes */ + struct list_head all_jnodes; +#endif + struct dentry *debugfs_root; +}; + +extern reiser4_super_info_data *get_super_private_nocheck(const struct + super_block * super); + +/* Return reiser4-specific part of super block */ +static inline reiser4_super_info_data *get_super_private(const struct + super_block * super) +{ + assert("nikita-447", super != NULL); + + return (reiser4_super_info_data *) super->s_fs_info; +} + +/* get ent context for the @super */ +static inline entd_context *get_entd_context(struct super_block *super) +{ + return &get_super_private(super)->entd; +} + +/* "Current" super-block: main super block used during current system + call. Reference to this super block is stored in reiser4_context. */ +static inline struct super_block *reiser4_get_current_sb(void) +{ + return get_current_context()->super; +} + +/* Reiser4-specific part of "current" super-block: main super block used + during current system call. Reference to this super block is stored in + reiser4_context. */ +static inline reiser4_super_info_data *get_current_super_private(void) +{ + return get_super_private(reiser4_get_current_sb()); +} + +static inline struct formatted_ra_params *get_current_super_ra_params(void) +{ + return &(get_current_super_private()->ra_params); +} + +/* + * true, if file system on @super is read-only + */ +static inline int rofs_super(struct super_block *super) +{ + return super->s_flags & MS_RDONLY; +} + +/* + * true, if @tree represents read-only file system + */ +static inline int rofs_tree(reiser4_tree * tree) +{ + return rofs_super(tree->super); +} + +/* + * true, if file system where @inode lives on, is read-only + */ +static inline int rofs_inode(struct inode *inode) +{ + return rofs_super(inode->i_sb); +} + +/* + * true, if file system where @node lives on, is read-only + */ +static inline int rofs_jnode(jnode * node) +{ + return rofs_tree(jnode_get_tree(node)); +} + +extern __u64 reiser4_current_block_count(void); + +extern void build_object_ops(struct super_block *super, struct object_ops *ops); + +#define REISER4_SUPER_MAGIC 0x52345362 /* (*(__u32 *)"R4Sb"); */ + +static inline void spin_lock_reiser4_super(reiser4_super_info_data *sbinfo) +{ + spin_lock(&(sbinfo->guard)); +} + +static inline void spin_unlock_reiser4_super(reiser4_super_info_data *sbinfo) +{ + assert_spin_locked(&(sbinfo->guard)); + spin_unlock(&(sbinfo->guard)); +} + +extern __u64 reiser4_flush_reserved(const struct super_block *); +extern int reiser4_is_set(const struct super_block *super, reiser4_fs_flag f); +extern long reiser4_statfs_type(const struct super_block *super); +extern __u64 reiser4_block_count(const struct super_block *super); +extern void reiser4_set_block_count(const struct super_block *super, __u64 nr); +extern __u64 reiser4_data_blocks(const struct super_block *super); +extern void reiser4_set_data_blocks(const struct super_block *super, __u64 nr); +extern __u64 reiser4_free_blocks(const struct super_block *super); +extern void reiser4_set_free_blocks(const struct super_block *super, __u64 nr); +extern __u32 reiser4_mkfs_id(const struct super_block *super); + +extern __u64 reiser4_free_committed_blocks(const struct super_block *super); + +extern __u64 reiser4_grabbed_blocks(const struct super_block *); +extern __u64 reiser4_fake_allocated(const struct super_block *); +extern __u64 reiser4_fake_allocated_unformatted(const struct super_block *); +extern __u64 reiser4_clustered_blocks(const struct super_block *); + +extern long reiser4_reserved_blocks(const struct super_block *super, uid_t uid, + gid_t gid); + +extern reiser4_space_allocator * +reiser4_get_space_allocator(const struct super_block *super); +extern reiser4_oid_allocator * +reiser4_get_oid_allocator(const struct super_block *super); +extern struct inode *reiser4_get_super_fake(const struct super_block *super); +extern struct inode *reiser4_get_cc_fake(const struct super_block *super); +extern struct inode *reiser4_get_bitmap_fake(const struct super_block *super); +extern reiser4_tree *reiser4_get_tree(const struct super_block *super); +extern int is_reiser4_super(const struct super_block *super); + +extern int reiser4_blocknr_is_sane(const reiser4_block_nr * blk); +extern int reiser4_blocknr_is_sane_for(const struct super_block *super, + const reiser4_block_nr * blk); +extern int reiser4_fill_super(struct super_block *s, void *data, int silent); +extern int reiser4_done_super(struct super_block *s); + +/* step of fill super */ +extern int reiser4_init_fs_info(struct super_block *); +extern void reiser4_done_fs_info(struct super_block *); +extern int reiser4_init_super_data(struct super_block *, char *opt_string); +extern int reiser4_init_read_super(struct super_block *, int silent); +extern int reiser4_init_root_inode(struct super_block *); +extern reiser4_plugin *get_default_plugin(pset_member memb); + +/* Maximal possible object id. */ +#define ABSOLUTE_MAX_OID ((oid_t)~0) + +#define OIDS_RESERVED (1 << 16) +int oid_init_allocator(struct super_block *, oid_t nr_files, oid_t next); +oid_t oid_allocate(struct super_block *); +int oid_release(struct super_block *, oid_t); +oid_t oid_next(const struct super_block *); +void oid_count_allocated(void); +void oid_count_released(void); +long oids_used(const struct super_block *); + +#if REISER4_DEBUG +void print_fs_info(const char *prefix, const struct super_block *); +#endif + +extern void destroy_reiser4_cache(struct kmem_cache **); + +extern struct super_operations reiser4_super_operations; +extern struct export_operations reiser4_export_operations; +extern struct dentry_operations reiser4_dentry_operations; + +/* __REISER4_SUPER_H__ */ +#endif + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 120 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/super_ops.c linux-2.6.30/fs/reiser4/super_ops.c --- linux-2.6.30.orig/fs/reiser4/super_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/super_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,725 @@ +/* Copyright 2005 by Hans Reiser, licensing governed by + * reiser4/README */ + +#include "inode.h" +#include "page_cache.h" +#include "ktxnmgrd.h" +#include "flush.h" +#include "safe_link.h" + +#include +#include +#include +#include +#include + +/* slab cache for inodes */ +static struct kmem_cache *inode_cache; + +static struct dentry *reiser4_debugfs_root = NULL; + +/** + * init_once - constructor for reiser4 inodes + * @cache: cache @obj belongs to + * @obj: inode to be initialized + * + * Initialization function to be called when new page is allocated by reiser4 + * inode cache. It is set on inode cache creation. + */ +static void init_once(void *obj) +{ + struct reiser4_inode_object *info; + + info = obj; + + /* initialize vfs inode */ + inode_init_once(&info->vfs_inode); + + /* + * initialize reiser4 specific part fo inode. + * NOTE-NIKITA add here initializations for locks, list heads, + * etc. that will be added to our private inode part. + */ + INIT_LIST_HEAD(get_readdir_list(&info->vfs_inode)); + init_rwsem(&info->p.conv_sem); + /* init semaphore which is used during inode loading */ + loading_init_once(&info->p); + INIT_RADIX_TREE(jnode_tree_by_reiser4_inode(&info->p), + GFP_ATOMIC); +#if REISER4_DEBUG + info->p.nr_jnodes = 0; +#endif +} + +/** + * init_inodes - create znode cache + * + * Initializes slab cache of inodes. It is part of reiser4 module initialization + */ +static int init_inodes(void) +{ + inode_cache = kmem_cache_create("reiser4_inode", + sizeof(struct reiser4_inode_object), + 0, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT, init_once); + if (inode_cache == NULL) + return RETERR(-ENOMEM); + return 0; +} + +/** + * done_inodes - delete inode cache + * + * This is called on reiser4 module unloading or system shutdown. + */ +static void done_inodes(void) +{ + destroy_reiser4_cache(&inode_cache); +} + +/** + * reiser4_alloc_inode - alloc_inode of super operations + * @super: super block new inode is allocated for + * + * Allocates new inode, initializes reiser4 specific part of it. + */ +static struct inode *reiser4_alloc_inode(struct super_block *super) +{ + struct reiser4_inode_object *obj; + + assert("nikita-1696", super != NULL); + obj = kmem_cache_alloc(inode_cache, reiser4_ctx_gfp_mask_get()); + if (obj != NULL) { + reiser4_inode *info; + + info = &obj->p; + + info->pset = plugin_set_get_empty(); + info->hset = plugin_set_get_empty(); + info->extmask = 0; + info->locality_id = 0ull; + info->plugin_mask = 0; + info->heir_mask = 0; +#if !REISER4_INO_IS_OID + info->oid_hi = 0; +#endif + reiser4_seal_init(&info->sd_seal, NULL, NULL); + coord_init_invalid(&info->sd_coord, NULL); + info->flags = 0; + spin_lock_init(&info->guard); + /* this deals with info's loading semaphore */ + loading_alloc(info); + info->vroot = UBER_TREE_ADDR; + return &obj->vfs_inode; + } else + return NULL; +} + +/** + * reiser4_destroy_inode - destroy_inode of super operations + * @inode: inode being destroyed + * + * Puts reiser4 specific portion of inode, frees memory occupied by inode. + */ +static void reiser4_destroy_inode(struct inode *inode) +{ + reiser4_inode *info; + + info = reiser4_inode_data(inode); + + assert("vs-1220", inode_has_no_jnodes(info)); + + if (!is_bad_inode(inode) && is_inode_loaded(inode)) { + file_plugin *fplug = inode_file_plugin(inode); + if (fplug->destroy_inode != NULL) + fplug->destroy_inode(inode); + } + reiser4_dispose_cursors(inode); + if (info->pset) + plugin_set_put(info->pset); + if (info->hset) + plugin_set_put(info->hset); + + /* + * cannot add similar assertion about ->i_list as prune_icache return + * inode into slab with dangling ->list.{next,prev}. This is safe, + * because they are re-initialized in the new_inode(). + */ + assert("nikita-2895", list_empty(&inode->i_dentry)); + assert("nikita-2896", hlist_unhashed(&inode->i_hash)); + assert("nikita-2898", list_empty_careful(get_readdir_list(inode))); + + /* this deals with info's loading semaphore */ + loading_destroy(info); + + kmem_cache_free(inode_cache, + container_of(info, struct reiser4_inode_object, p)); +} + +/** + * reiser4_dirty_inode - dirty_inode of super operations + * @inode: inode being dirtied + * + * Updates stat data. + */ +static void reiser4_dirty_inode(struct inode *inode) +{ + int result; + + if (!is_in_reiser4_context()) + return; + assert("", !IS_RDONLY(inode)); + assert("", (inode_file_plugin(inode)->estimate.update(inode) <= + get_current_context()->grabbed_blocks)); + + result = reiser4_update_sd(inode); + if (result) + warning("", "failed to dirty inode for %llu: %d", + get_inode_oid(inode), result); +} + +/** + * reiser4_delete_inode - delete_inode of super operations + * @inode: inode to delete + * + * Calls file plugin's delete_object method to delete object items from + * filesystem tree and calls clear_inode. + */ +static void reiser4_delete_inode(struct inode *inode) +{ + reiser4_context *ctx; + file_plugin *fplug; + + ctx = reiser4_init_context(inode->i_sb); + if (IS_ERR(ctx)) { + warning("vs-15", "failed to init context"); + return; + } + + if (is_inode_loaded(inode)) { + fplug = inode_file_plugin(inode); + if (fplug != NULL && fplug->delete_object != NULL) + fplug->delete_object(inode); + } + + truncate_inode_pages(&inode->i_data, 0); + inode->i_blocks = 0; + clear_inode(inode); + reiser4_exit_context(ctx); +} + +/** + * reiser4_put_super - put_super of super operations + * @super: super block to free + * + * Stops daemons, release resources, umounts in short. + */ +static void reiser4_put_super(struct super_block *super) +{ + reiser4_super_info_data *sbinfo; + reiser4_context *ctx; + + sbinfo = get_super_private(super); + assert("vs-1699", sbinfo); + + debugfs_remove(sbinfo->tmgr.debugfs_atom_count); + debugfs_remove(sbinfo->tmgr.debugfs_id_count); + debugfs_remove(sbinfo->debugfs_root); + + ctx = reiser4_init_context(super); + if (IS_ERR(ctx)) { + warning("vs-17", "failed to init context"); + return; + } + + /* have disk format plugin to free its resources */ + if (get_super_private(super)->df_plug->release) + get_super_private(super)->df_plug->release(super); + + reiser4_done_formatted_fake(super); + + /* stop daemons: ktxnmgr and entd */ + reiser4_done_entd(super); + reiser4_done_ktxnmgrd(super); + reiser4_done_txnmgr(&sbinfo->tmgr); + + reiser4_done_fs_info(super); + reiser4_exit_context(ctx); +} + +/** + * reiser4_write_super - write_super of super operations + * @super: super block to write + * + * Captures znode associated with super block, comit all transactions. + */ +static void reiser4_write_super(struct super_block *super) +{ + int ret; + reiser4_context *ctx; + + assert("vs-1700", !rofs_super(super)); + + ctx = reiser4_init_context(super); + if (IS_ERR(ctx)) { + warning("vs-16", "failed to init context"); + return; + } + + ret = reiser4_capture_super_block(super); + if (ret != 0) + warning("vs-1701", + "reiser4_capture_super_block failed in write_super: %d", + ret); + ret = txnmgr_force_commit_all(super, 0); + if (ret != 0) + warning("jmacd-77113", + "txn_force failed in write_super: %d", ret); + + super->s_dirt = 0; + + reiser4_exit_context(ctx); +} + +/** + * reiser4_statfs - statfs of super operations + * @super: super block of file system in queried + * @stafs: buffer to fill with statistics + * + * Returns information about filesystem. + */ +static int reiser4_statfs(struct dentry *dentry, struct kstatfs *statfs) +{ + sector_t total; + sector_t reserved; + sector_t free; + sector_t forroot; + sector_t deleted; + reiser4_context *ctx; + struct super_block *super = dentry->d_sb; + + assert("nikita-408", super != NULL); + assert("nikita-409", statfs != NULL); + + ctx = reiser4_init_context(super); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + statfs->f_type = reiser4_statfs_type(super); + statfs->f_bsize = super->s_blocksize; + + /* + * 5% of total block space is reserved. This is needed for flush and + * for truncates (so that we are able to perform truncate/unlink even + * on the otherwise completely full file system). If this reservation + * is hidden from statfs(2), users will mistakenly guess that they + * have enough free space to complete some operation, which is + * frustrating. + * + * Another possible solution is to subtract ->blocks_reserved from + * ->f_bfree, but changing available space seems less intrusive than + * letting user to see 5% of disk space to be used directly after + * mkfs. + */ + total = reiser4_block_count(super); + reserved = get_super_private(super)->blocks_reserved; + deleted = txnmgr_count_deleted_blocks(); + free = reiser4_free_blocks(super) + deleted; + forroot = reiser4_reserved_blocks(super, 0, 0); + + /* + * These counters may be in inconsistent state because we take the + * values without keeping any global spinlock. Here we do a sanity + * check that free block counter does not exceed the number of all + * blocks. + */ + if (free > total) + free = total; + statfs->f_blocks = total - reserved; + /* make sure statfs->f_bfree is never larger than statfs->f_blocks */ + if (free > reserved) + free -= reserved; + else + free = 0; + statfs->f_bfree = free; + + if (free > forroot) + free -= forroot; + else + free = 0; + statfs->f_bavail = free; + + statfs->f_files = 0; + statfs->f_ffree = 0; + + /* maximal acceptable name length depends on directory plugin. */ + assert("nikita-3351", super->s_root->d_inode != NULL); + statfs->f_namelen = reiser4_max_filename_len(super->s_root->d_inode); + reiser4_exit_context(ctx); + return 0; +} + +/** + * reiser4_clear_inode - clear_inode of super operation + * @inode: inode about to destroy + * + * Does sanity checks: being destroyed should have all jnodes detached. + */ +static void reiser4_clear_inode(struct inode *inode) +{ +#if REISER4_DEBUG + reiser4_inode *r4_inode; + + r4_inode = reiser4_inode_data(inode); + if (!inode_has_no_jnodes(r4_inode)) + warning("vs-1732", "reiser4 inode has %ld jnodes\n", + r4_inode->nr_jnodes); +#endif +} + +/** + * reiser4_sync_inodes - sync_inodes of super operations + * @super: + * @wbc: + * + * This method is called by background and non-backgound writeback. Reiser4's + * implementation uses generic_sync_sb_inodes to call reiser4_writepages for + * each of dirty inodes. Reiser4_writepages handles pages dirtied via shared + * mapping - dirty pages get into atoms. Writeout is called to flush some + * atoms. + */ +static void reiser4_sync_inodes(struct super_block *super, + struct writeback_control *wbc) +{ + reiser4_context *ctx; + long to_write; + + if (wbc->for_kupdate) + /* reiser4 has its own means of periodical write-out */ + return; + + to_write = wbc->nr_to_write; + assert("vs-49", wbc->older_than_this == NULL); + + ctx = reiser4_init_context(super); + if (IS_ERR(ctx)) { + warning("vs-13", "failed to init context"); + return; + } + + /* + * call reiser4_writepages for each of dirty inodes to turn dirty pages + * into transactions if they were not yet. + */ + generic_sync_sb_inodes(super, wbc); + + /* flush goes here */ + wbc->nr_to_write = to_write; + reiser4_writeout(super, wbc); + + /* avoid recursive calls to ->sync_inodes */ + context_set_commit_async(ctx); + reiser4_exit_context(ctx); +} + +/** + * reiser4_show_options - show_options of super operations + * @m: file where to write information + * @mnt: mount structure + * + * Makes reiser4 mount options visible in /proc/mounts. + */ +static int reiser4_show_options(struct seq_file *m, struct vfsmount *mnt) +{ + struct super_block *super; + reiser4_super_info_data *sbinfo; + + super = mnt->mnt_sb; + sbinfo = get_super_private(super); + + seq_printf(m, ",atom_max_size=0x%x", sbinfo->tmgr.atom_max_size); + seq_printf(m, ",atom_max_age=0x%x", sbinfo->tmgr.atom_max_age); + seq_printf(m, ",atom_min_size=0x%x", sbinfo->tmgr.atom_min_size); + seq_printf(m, ",atom_max_flushers=0x%x", + sbinfo->tmgr.atom_max_flushers); + seq_printf(m, ",cbk_cache_slots=0x%x", + sbinfo->tree.cbk_cache.nr_slots); + + return 0; +} + +struct super_operations reiser4_super_operations = { + .alloc_inode = reiser4_alloc_inode, + .destroy_inode = reiser4_destroy_inode, + .dirty_inode = reiser4_dirty_inode, + .delete_inode = reiser4_delete_inode, + .put_super = reiser4_put_super, + .write_super = reiser4_write_super, + .statfs = reiser4_statfs, + .clear_inode = reiser4_clear_inode, + .sync_inodes = reiser4_sync_inodes, + .show_options = reiser4_show_options +}; + +/** + * fill_super - initialize super block on mount + * @super: super block to fill + * @data: reiser4 specific mount option + * @silent: + * + * This is to be called by reiser4_get_sb. Mounts filesystem. + */ +static int fill_super(struct super_block *super, void *data, int silent) +{ + reiser4_context ctx; + int result; + reiser4_super_info_data *sbinfo; + + assert("zam-989", super != NULL); + + super->s_op = NULL; + init_stack_context(&ctx, super); + + /* allocate reiser4 specific super block */ + if ((result = reiser4_init_fs_info(super)) != 0) + goto failed_init_sinfo; + + sbinfo = get_super_private(super); + /* initialize various reiser4 parameters, parse mount options */ + if ((result = reiser4_init_super_data(super, data)) != 0) + goto failed_init_super_data; + + /* read reiser4 master super block, initialize disk format plugin */ + if ((result = reiser4_init_read_super(super, silent)) != 0) + goto failed_init_read_super; + + /* initialize transaction manager */ + reiser4_init_txnmgr(&sbinfo->tmgr); + + /* initialize ktxnmgrd context and start kernel thread ktxnmrgd */ + if ((result = reiser4_init_ktxnmgrd(super)) != 0) + goto failed_init_ktxnmgrd; + + /* initialize entd context and start kernel thread entd */ + if ((result = reiser4_init_entd(super)) != 0) + goto failed_init_entd; + + /* initialize address spaces for formatted nodes and bitmaps */ + if ((result = reiser4_init_formatted_fake(super)) != 0) + goto failed_init_formatted_fake; + + /* initialize disk format plugin */ + if ((result = get_super_private(super)->df_plug->init_format(super, + data)) != 0) + goto failed_init_disk_format; + + /* + * There are some 'committed' versions of reiser4 super block counters, + * which correspond to reiser4 on-disk state. These counters are + * initialized here + */ + sbinfo->blocks_free_committed = sbinfo->blocks_free; + sbinfo->nr_files_committed = oids_used(super); + + /* get inode of root directory */ + if ((result = reiser4_init_root_inode(super)) != 0) + goto failed_init_root_inode; + + if ((result = get_super_private(super)->df_plug->version_update(super)) != 0) + goto failed_update_format_version; + + process_safelinks(super); + reiser4_exit_context(&ctx); + + sbinfo->debugfs_root = debugfs_create_dir(super->s_id, + reiser4_debugfs_root); + if (sbinfo->debugfs_root) { + sbinfo->tmgr.debugfs_atom_count = + debugfs_create_u32("atom_count", S_IFREG|S_IRUSR, + sbinfo->debugfs_root, + &sbinfo->tmgr.atom_count); + sbinfo->tmgr.debugfs_id_count = + debugfs_create_u32("id_count", S_IFREG|S_IRUSR, + sbinfo->debugfs_root, + &sbinfo->tmgr.id_count); + } + return 0; + + failed_update_format_version: + failed_init_root_inode: + if (sbinfo->df_plug->release) + sbinfo->df_plug->release(super); + failed_init_disk_format: + reiser4_done_formatted_fake(super); + failed_init_formatted_fake: + reiser4_done_entd(super); + failed_init_entd: + reiser4_done_ktxnmgrd(super); + failed_init_ktxnmgrd: + reiser4_done_txnmgr(&sbinfo->tmgr); + failed_init_read_super: + failed_init_super_data: + reiser4_done_fs_info(super); + failed_init_sinfo: + reiser4_exit_context(&ctx); + return result; +} + +/** + * reiser4_get_sb - get_sb of file_system_type operations + * @fs_type: + * @flags: mount flags MS_RDONLY, MS_VERBOSE, etc + * @dev_name: block device file name + * @data: specific mount options + * + * Reiser4 mount entry. + */ +static int reiser4_get_sb(struct file_system_type *fs_type, int flags, + const char *dev_name, void *data, struct vfsmount *mnt) +{ + return get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt); +} + +/* structure describing the reiser4 filesystem implementation */ +static struct file_system_type reiser4_fs_type = { + .owner = THIS_MODULE, + .name = "reiser4", + .fs_flags = FS_REQUIRES_DEV, + .get_sb = reiser4_get_sb, + .kill_sb = kill_block_super, + .next = NULL +}; + +void destroy_reiser4_cache(struct kmem_cache **cachep) +{ + BUG_ON(*cachep == NULL); + kmem_cache_destroy(*cachep); + *cachep = NULL; +} + +/** + * init_reiser4 - reiser4 initialization entry point + * + * Initializes reiser4 slabs, registers reiser4 filesystem type. It is called + * on kernel initialization or during reiser4 module load. + */ +static int __init init_reiser4(void) +{ + int result; + + printk(KERN_INFO + "Loading Reiser4. " + "See www.namesys.com for a description of Reiser4.\n"); + + /* initialize slab cache of inodes */ + if ((result = init_inodes()) != 0) + goto failed_inode_cache; + + /* initialize cache of znodes */ + if ((result = init_znodes()) != 0) + goto failed_init_znodes; + + /* initialize all plugins */ + if ((result = init_plugins()) != 0) + goto failed_init_plugins; + + /* initialize cache of plugin_set-s and plugin_set's hash table */ + if ((result = init_plugin_set()) != 0) + goto failed_init_plugin_set; + + /* initialize caches of txn_atom-s and txn_handle-s */ + if ((result = init_txnmgr_static()) != 0) + goto failed_init_txnmgr_static; + + /* initialize cache of jnodes */ + if ((result = init_jnodes()) != 0) + goto failed_init_jnodes; + + /* initialize cache of flush queues */ + if ((result = reiser4_init_fqs()) != 0) + goto failed_init_fqs; + + /* initialize cache of structures attached to dentry->d_fsdata */ + if ((result = reiser4_init_dentry_fsdata()) != 0) + goto failed_init_dentry_fsdata; + + /* initialize cache of structures attached to file->private_data */ + if ((result = reiser4_init_file_fsdata()) != 0) + goto failed_init_file_fsdata; + + /* + * initialize cache of d_cursors. See plugin/file_ops_readdir.c for + * more details + */ + if ((result = reiser4_init_d_cursor()) != 0) + goto failed_init_d_cursor; + + if ((result = register_filesystem(&reiser4_fs_type)) == 0) { + reiser4_debugfs_root = debugfs_create_dir("reiser4", NULL); + return 0; + } + + reiser4_done_d_cursor(); + failed_init_d_cursor: + reiser4_done_file_fsdata(); + failed_init_file_fsdata: + reiser4_done_dentry_fsdata(); + failed_init_dentry_fsdata: + reiser4_done_fqs(); + failed_init_fqs: + done_jnodes(); + failed_init_jnodes: + done_txnmgr_static(); + failed_init_txnmgr_static: + done_plugin_set(); + failed_init_plugin_set: + failed_init_plugins: + done_znodes(); + failed_init_znodes: + done_inodes(); + failed_inode_cache: + return result; +} + +/** + * done_reiser4 - reiser4 exit entry point + * + * Unregister reiser4 filesystem type, deletes caches. It is called on shutdown + * or at module unload. + */ +static void __exit done_reiser4(void) +{ + int result; + + debugfs_remove(reiser4_debugfs_root); + result = unregister_filesystem(&reiser4_fs_type); + BUG_ON(result != 0); + reiser4_done_d_cursor(); + reiser4_done_file_fsdata(); + reiser4_done_dentry_fsdata(); + reiser4_done_fqs(); + done_jnodes(); + done_txnmgr_static(); + done_plugin_set(); + done_znodes(); + destroy_reiser4_cache(&inode_cache); +} + +module_init(init_reiser4); +module_exit(done_reiser4); + +MODULE_DESCRIPTION("Reiser4 filesystem"); +MODULE_AUTHOR("Hans Reiser "); + +MODULE_LICENSE("GPL"); + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/tap.c linux-2.6.30/fs/reiser4/tap.c --- linux-2.6.30.orig/fs/reiser4/tap.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tap.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,376 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* + Tree Access Pointer (tap). + + tap is data structure combining coord and lock handle (mostly). It is + useful when one has to scan tree nodes (for example, in readdir, or flush), + for tap functions allow to move tap in either direction transparently + crossing unit/item/node borders. + + Tap doesn't provide automatic synchronization of its fields as it is + supposed to be per-thread object. +*/ + +#include "forward.h" +#include "debug.h" +#include "coord.h" +#include "tree.h" +#include "context.h" +#include "tap.h" +#include "znode.h" +#include "tree_walk.h" + +#if REISER4_DEBUG +static int tap_invariant(const tap_t *tap); +static void tap_check(const tap_t *tap); +#else +#define tap_check(tap) noop +#endif + +/** load node tap is pointing to, if not loaded already */ +int reiser4_tap_load(tap_t *tap) +{ + tap_check(tap); + if (tap->loaded == 0) { + int result; + + result = zload_ra(tap->coord->node, &tap->ra_info); + if (result != 0) + return result; + coord_clear_iplug(tap->coord); + } + ++tap->loaded; + tap_check(tap); + return 0; +} + +/** release node tap is pointing to. Dual to tap_load() */ +void reiser4_tap_relse(tap_t *tap) +{ + tap_check(tap); + if (tap->loaded > 0) { + --tap->loaded; + if (tap->loaded == 0) + zrelse(tap->coord->node); + } + tap_check(tap); +} + +/** + * init tap to consist of @coord and @lh. Locks on nodes will be acquired with + * @mode + */ +void reiser4_tap_init(tap_t *tap, coord_t *coord, lock_handle * lh, + znode_lock_mode mode) +{ + tap->coord = coord; + tap->lh = lh; + tap->mode = mode; + tap->loaded = 0; + INIT_LIST_HEAD(&tap->linkage); + reiser4_init_ra_info(&tap->ra_info); +} + +/** add @tap to the per-thread list of all taps */ +void reiser4_tap_monitor(tap_t *tap) +{ + assert("nikita-2623", tap != NULL); + tap_check(tap); + list_add(&tap->linkage, reiser4_taps_list()); + tap_check(tap); +} + +/* duplicate @src into @dst. Copy lock handle. @dst is not initially + * loaded. */ +void reiser4_tap_copy(tap_t *dst, tap_t *src) +{ + assert("nikita-3193", src != NULL); + assert("nikita-3194", dst != NULL); + + *dst->coord = *src->coord; + if (src->lh->node) + copy_lh(dst->lh, src->lh); + dst->mode = src->mode; + dst->loaded = 0; + INIT_LIST_HEAD(&dst->linkage); + dst->ra_info = src->ra_info; +} + +/** finish with @tap */ +void reiser4_tap_done(tap_t *tap) +{ + assert("nikita-2565", tap != NULL); + tap_check(tap); + if (tap->loaded > 0) + zrelse(tap->coord->node); + done_lh(tap->lh); + tap->loaded = 0; + list_del_init(&tap->linkage); + tap->coord->node = NULL; +} + +/** + * move @tap to the new node, locked with @target. Load @target, if @tap was + * already loaded. + */ +int reiser4_tap_move(tap_t *tap, lock_handle * target) +{ + int result = 0; + + assert("nikita-2567", tap != NULL); + assert("nikita-2568", target != NULL); + assert("nikita-2570", target->node != NULL); + assert("nikita-2569", tap->coord->node == tap->lh->node); + + tap_check(tap); + if (tap->loaded > 0) + result = zload_ra(target->node, &tap->ra_info); + + if (result == 0) { + if (tap->loaded > 0) + zrelse(tap->coord->node); + done_lh(tap->lh); + copy_lh(tap->lh, target); + tap->coord->node = target->node; + coord_clear_iplug(tap->coord); + } + tap_check(tap); + return result; +} + +/** + * move @tap to @target. Acquire lock on @target, if @tap was already + * loaded. + */ +static int tap_to(tap_t *tap, znode * target) +{ + int result; + + assert("nikita-2624", tap != NULL); + assert("nikita-2625", target != NULL); + + tap_check(tap); + result = 0; + if (tap->coord->node != target) { + lock_handle here; + + init_lh(&here); + result = longterm_lock_znode(&here, target, + tap->mode, ZNODE_LOCK_HIPRI); + if (result == 0) { + result = reiser4_tap_move(tap, &here); + done_lh(&here); + } + } + tap_check(tap); + return result; +} + +/** + * move @tap to given @target, loading and locking @target->node if + * necessary + */ +int tap_to_coord(tap_t *tap, coord_t *target) +{ + int result; + + tap_check(tap); + result = tap_to(tap, target->node); + if (result == 0) + coord_dup(tap->coord, target); + tap_check(tap); + return result; +} + +/** return list of all taps */ +struct list_head *reiser4_taps_list(void) +{ + return &get_current_context()->taps; +} + +/** helper function for go_{next,prev}_{item,unit,node}() */ +int go_dir_el(tap_t *tap, sideof dir, int units_p) +{ + coord_t dup; + coord_t *coord; + int result; + + int (*coord_dir) (coord_t *); + int (*get_dir_neighbor) (lock_handle *, znode *, int, int); + void (*coord_init) (coord_t *, const znode *); + ON_DEBUG(int (*coord_check) (const coord_t *)); + + assert("nikita-2556", tap != NULL); + assert("nikita-2557", tap->coord != NULL); + assert("nikita-2558", tap->lh != NULL); + assert("nikita-2559", tap->coord->node != NULL); + + tap_check(tap); + if (dir == LEFT_SIDE) { + coord_dir = units_p ? coord_prev_unit : coord_prev_item; + get_dir_neighbor = reiser4_get_left_neighbor; + coord_init = coord_init_last_unit; + } else { + coord_dir = units_p ? coord_next_unit : coord_next_item; + get_dir_neighbor = reiser4_get_right_neighbor; + coord_init = coord_init_first_unit; + } + ON_DEBUG(coord_check = + units_p ? coord_is_existing_unit : coord_is_existing_item); + assert("nikita-2560", coord_check(tap->coord)); + + coord = tap->coord; + coord_dup(&dup, coord); + if (coord_dir(&dup) != 0) { + do { + /* move to the left neighboring node */ + lock_handle dup; + + init_lh(&dup); + result = + get_dir_neighbor(&dup, coord->node, (int)tap->mode, + GN_CAN_USE_UPPER_LEVELS); + if (result == 0) { + result = reiser4_tap_move(tap, &dup); + if (result == 0) + coord_init(tap->coord, dup.node); + done_lh(&dup); + } + /* skip empty nodes */ + } while ((result == 0) && node_is_empty(coord->node)); + } else { + result = 0; + coord_dup(coord, &dup); + } + assert("nikita-2564", ergo(!result, coord_check(tap->coord))); + tap_check(tap); + return result; +} + +/** + * move @tap to the next unit, transparently crossing item and node + * boundaries + */ +int go_next_unit(tap_t *tap) +{ + return go_dir_el(tap, RIGHT_SIDE, 1); +} + +/** + * move @tap to the previous unit, transparently crossing item and node + * boundaries + */ +int go_prev_unit(tap_t *tap) +{ + return go_dir_el(tap, LEFT_SIDE, 1); +} + +/** + * @shift times apply @actor to the @tap. This is used to move @tap by + * @shift units (or items, or nodes) in either direction. + */ +static int rewind_to(tap_t *tap, go_actor_t actor, int shift) +{ + int result; + + assert("nikita-2555", shift >= 0); + assert("nikita-2562", tap->coord->node == tap->lh->node); + + tap_check(tap); + result = reiser4_tap_load(tap); + if (result != 0) + return result; + + for (; shift > 0; --shift) { + result = actor(tap); + assert("nikita-2563", tap->coord->node == tap->lh->node); + if (result != 0) + break; + } + reiser4_tap_relse(tap); + tap_check(tap); + return result; +} + +/** move @tap @shift units rightward */ +int rewind_right(tap_t *tap, int shift) +{ + return rewind_to(tap, go_next_unit, shift); +} + +/** move @tap @shift units leftward */ +int rewind_left(tap_t *tap, int shift) +{ + return rewind_to(tap, go_prev_unit, shift); +} + +#if REISER4_DEBUG +/** debugging function: print @tap content in human readable form */ +static void print_tap(const char *prefix, const tap_t *tap) +{ + if (tap == NULL) { + printk("%s: null tap\n", prefix); + return; + } + printk("%s: loaded: %i, in-list: %i, node: %p, mode: %s\n", prefix, + tap->loaded, (&tap->linkage == tap->linkage.next && + &tap->linkage == tap->linkage.prev), + tap->lh->node, + lock_mode_name(tap->mode)); + print_coord("\tcoord", tap->coord, 0); +} + +/** check [tap-sane] invariant */ +static int tap_invariant(const tap_t *tap) +{ + /* [tap-sane] invariant */ + + if (tap == NULL) + return 1; + /* tap->mode is one of + * + * {ZNODE_NO_LOCK, ZNODE_READ_LOCK, ZNODE_WRITE_LOCK}, and + */ + if (tap->mode != ZNODE_NO_LOCK && + tap->mode != ZNODE_READ_LOCK && tap->mode != ZNODE_WRITE_LOCK) + return 2; + /* tap->coord != NULL, and */ + if (tap->coord == NULL) + return 3; + /* tap->lh != NULL, and */ + if (tap->lh == NULL) + return 4; + /* tap->loaded > 0 => znode_is_loaded(tap->coord->node), and */ + if (!ergo(tap->loaded, znode_is_loaded(tap->coord->node))) + return 5; + /* tap->coord->node == tap->lh->node if tap->lh->node is not 0 */ + if (tap->lh->node != NULL && tap->coord->node != tap->lh->node) + return 6; + return 0; +} + +/** debugging function: check internal @tap consistency */ +static void tap_check(const tap_t *tap) +{ + int result; + + result = tap_invariant(tap); + if (result != 0) { + print_tap("broken", tap); + reiser4_panic("nikita-2831", "tap broken: %i\n", result); + } +} +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tap.h linux-2.6.30/fs/reiser4/tap.h --- linux-2.6.30.orig/fs/reiser4/tap.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tap.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,70 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* Tree Access Pointers. See tap.c for more details. */ + +#if !defined(__REISER4_TAP_H__) +#define __REISER4_TAP_H__ + +#include "forward.h" +#include "readahead.h" + +/** + tree_access_pointer aka tap. Data structure combining coord_t and lock + handle. + Invariants involving this data-type, see doc/lock-ordering for details: + + [tap-sane] + */ +struct tree_access_pointer { + /* coord tap is at */ + coord_t *coord; + /* lock handle on ->coord->node */ + lock_handle *lh; + /* mode of lock acquired by this tap */ + znode_lock_mode mode; + /* incremented by reiser4_tap_load(). + Decremented by reiser4_tap_relse(). */ + int loaded; + /* list of taps */ + struct list_head linkage; + /* read-ahead hint */ + ra_info_t ra_info; +}; + +typedef int (*go_actor_t) (tap_t *tap); + +extern int reiser4_tap_load(tap_t *tap); +extern void reiser4_tap_relse(tap_t *tap); +extern void reiser4_tap_init(tap_t *tap, coord_t *coord, lock_handle * lh, + znode_lock_mode mode); +extern void reiser4_tap_monitor(tap_t *tap); +extern void reiser4_tap_copy(tap_t *dst, tap_t *src); +extern void reiser4_tap_done(tap_t *tap); +extern int reiser4_tap_move(tap_t *tap, lock_handle * target); +extern int tap_to_coord(tap_t *tap, coord_t *target); + +extern int go_dir_el(tap_t *tap, sideof dir, int units_p); +extern int go_next_unit(tap_t *tap); +extern int go_prev_unit(tap_t *tap); +extern int rewind_right(tap_t *tap, int shift); +extern int rewind_left(tap_t *tap, int shift); + +extern struct list_head *reiser4_taps_list(void); + +#define for_all_taps(tap) \ + for (tap = list_entry(reiser4_taps_list()->next, tap_t, linkage); \ + reiser4_taps_list() != &tap->linkage; \ + tap = list_entry(tap->linkage.next, tap_t, linkage)) + +/* __REISER4_TAP_H__ */ +#endif +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tree.c linux-2.6.30/fs/reiser4/tree.c --- linux-2.6.30.orig/fs/reiser4/tree.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tree.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1878 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* + * KEYS IN A TREE. + * + * The tree consists of nodes located on the disk. Node in the tree is either + * formatted or unformatted. Formatted node is one that has structure + * understood by the tree balancing and traversal code. Formatted nodes are + * further classified into leaf and internal nodes. Latter distinctions is + * (almost) of only historical importance: general structure of leaves and + * internal nodes is the same in Reiser4. Unformatted nodes contain raw data + * that are part of bodies of ordinary files and attributes. + * + * Each node in the tree spawns some interval in the key space. Key ranges for + * all nodes in the tree are disjoint. Actually, this only holds in some weak + * sense, because of the non-unique keys: intersection of key ranges for + * different nodes is either empty, or consists of exactly one key. + * + * Formatted node consists of a sequence of items. Each item spawns some + * interval in key space. Key ranges for all items in a tree are disjoint, + * modulo non-unique keys again. Items within nodes are ordered in the key + * order of the smallest key in a item. + * + * Particular type of item can be further split into units. Unit is piece of + * item that can be cut from item and moved into another item of the same + * time. Units are used by balancing code to repack data during balancing. + * + * Unit can be further split into smaller entities (for example, extent unit + * represents several pages, and it is natural for extent code to operate on + * particular pages and even bytes within one unit), but this is of no + * relevance to the generic balancing and lookup code. + * + * Although item is said to "spawn" range or interval of keys, it is not + * necessary that item contains piece of data addressable by each and every + * key in this range. For example, compound directory item, consisting of + * units corresponding to directory entries and keyed by hashes of file names, + * looks more as having "discrete spectrum": only some disjoint keys inside + * range occupied by this item really address data. + * + * No than less, each item always has well-defined least (minimal) key, that + * is recorded in item header, stored in the node this item is in. Also, item + * plugin can optionally define method ->max_key_inside() returning maximal + * key that can _possibly_ be located within this item. This method is used + * (mainly) to determine when given piece of data should be merged into + * existing item, in stead of creating new one. Because of this, even though + * ->max_key_inside() can be larger that any key actually located in the item, + * intervals + * + * [ reiser4_min_key( item ), ->max_key_inside( item ) ] + * + * are still disjoint for all items within the _same_ node. + * + * In memory node is represented by znode. It plays several roles: + * + * . something locks are taken on + * + * . something tracked by transaction manager (this is going to change) + * + * . something used to access node data + * + * . something used to maintain tree structure in memory: sibling and + * parental linkage. + * + * . something used to organize nodes into "slums" + * + * More on znodes see in znode.[ch] + * + * DELIMITING KEYS + * + * To simplify balancing, allow some flexibility in locking and speed up + * important coord cache optimization, we keep delimiting keys of nodes in + * memory. Depending on disk format (implemented by appropriate node plugin) + * node on disk can record both left and right delimiting key, only one of + * them, or none. Still, our balancing and tree traversal code keep both + * delimiting keys for a node that is in memory stored in the znode. When + * node is first brought into memory during tree traversal, its left + * delimiting key is taken from its parent, and its right delimiting key is + * either next key in its parent, or is right delimiting key of parent if + * node is the rightmost child of parent. + * + * Physical consistency of delimiting key is protected by special dk + * read-write lock. That is, delimiting keys can only be inspected or + * modified under this lock. But dk lock is only sufficient for fast + * "pessimistic" check, because to simplify code and to decrease lock + * contention, balancing (carry) only updates delimiting keys right before + * unlocking all locked nodes on the given tree level. For example, + * coord-by-key cache scans LRU list of recently accessed znodes. For each + * node it first does fast check under dk spin lock. If key looked for is + * not between delimiting keys for this node, next node is inspected and so + * on. If key is inside of the key range, long term lock is taken on node + * and key range is rechecked. + * + * COORDINATES + * + * To find something in the tree, you supply a key, and the key is resolved + * by coord_by_key() into a coord (coordinate) that is valid as long as the + * node the coord points to remains locked. As mentioned above trees + * consist of nodes that consist of items that consist of units. A unit is + * the smallest and indivisible piece of tree as far as balancing and tree + * search are concerned. Each node, item, and unit can be addressed by + * giving its level in the tree and the key occupied by this entity. A node + * knows what the key ranges are of the items within it, and how to find its + * items and invoke their item handlers, but it does not know how to access + * individual units within its items except through the item handlers. + * coord is a structure containing a pointer to the node, the ordinal number + * of the item within this node (a sort of item offset), and the ordinal + * number of the unit within this item. + * + * TREE LOOKUP + * + * There are two types of access to the tree: lookup and modification. + * + * Lookup is a search for the key in the tree. Search can look for either + * exactly the key given to it, or for the largest key that is not greater + * than the key given to it. This distinction is determined by "bias" + * parameter of search routine (coord_by_key()). coord_by_key() either + * returns error (key is not in the tree, or some kind of external error + * occurred), or successfully resolves key into coord. + * + * This resolution is done by traversing tree top-to-bottom from root level + * to the desired level. On levels above twig level (level one above the + * leaf level) nodes consist exclusively of internal items. Internal item is + * nothing more than pointer to the tree node on the child level. On twig + * level nodes consist of internal items intermixed with extent + * items. Internal items form normal search tree structure used by traversal + * to descent through the tree. + * + * TREE LOOKUP OPTIMIZATIONS + * + * Tree lookup described above is expensive even if all nodes traversed are + * already in the memory: for each node binary search within it has to be + * performed and binary searches are CPU consuming and tend to destroy CPU + * caches. + * + * Several optimizations are used to work around this: + * + * . cbk_cache (look-aside cache for tree traversals, see search.c for + * details) + * + * . seals (see seal.[ch]) + * + * . vroot (see search.c) + * + * General search-by-key is layered thusly: + * + * [check seal, if any] --ok--> done + * | + * failed + * | + * V + * [vroot defined] --no--> node = tree_root + * | | + * yes | + * | | + * V | + * node = vroot | + * | | + * | | + * | | + * V V + * [check cbk_cache for key] --ok--> done + * | + * failed + * | + * V + * [start tree traversal from node] + * + */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "plugin/item/static_stat.h" +#include "plugin/item/item.h" +#include "plugin/node/node.h" +#include "plugin/plugin.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree_walk.h" +#include "carry.h" +#include "carry_ops.h" +#include "tap.h" +#include "tree.h" +#include "vfs_ops.h" +#include "page_cache.h" +#include "super.h" +#include "reiser4.h" +#include "inode.h" + +#include /* for struct super_block */ +#include + +/* Disk address (block number) never ever used for any real tree node. This is + used as block number of "uber" znode. + + Invalid block addresses are 0 by tradition. + +*/ +const reiser4_block_nr UBER_TREE_ADDR = 0ull; + +#define CUT_TREE_MIN_ITERATIONS 64 + +static int find_child_by_addr(znode * parent, znode * child, coord_t *result); + +/* return node plugin of coord->node */ +node_plugin *node_plugin_by_coord(const coord_t *coord) +{ + assert("vs-1", coord != NULL); + assert("vs-2", coord->node != NULL); + + return coord->node->nplug; +} + +/* insert item into tree. Fields of @coord are updated so that they can be + * used by consequent insert operation. */ +insert_result insert_by_key(reiser4_tree * tree /* tree to insert new item + * into */ , + const reiser4_key * key /* key of new item */ , + reiser4_item_data * data /* parameters for item + * creation */ , + coord_t *coord /* resulting insertion coord */ , + lock_handle * lh /* resulting lock + * handle */ , + tree_level stop_level /* level where to insert */ , + __u32 flags/* insertion flags */) +{ + int result; + + assert("nikita-358", tree != NULL); + assert("nikita-360", coord != NULL); + + result = coord_by_key(tree, key, coord, lh, ZNODE_WRITE_LOCK, + FIND_EXACT, stop_level, stop_level, + flags | CBK_FOR_INSERT, NULL/*ra_info */); + switch (result) { + default: + break; + case CBK_COORD_FOUND: + result = IBK_ALREADY_EXISTS; + break; + case CBK_COORD_NOTFOUND: + assert("nikita-2017", coord->node != NULL); + result = insert_by_coord(coord, data, key, lh, 0/*flags */); + break; + } + return result; +} + +/* insert item by calling carry. Helper function called if short-cut + insertion failed */ +static insert_result insert_with_carry_by_coord(coord_t *coord, + /* coord where to insert */ + lock_handle * lh, + /* lock handle of insertion node */ + reiser4_item_data * data, + /* parameters of new item */ + const reiser4_key * key, + /* key of new item */ + carry_opcode cop, + /* carry operation to perform */ + cop_insert_flag flags + /* carry flags */ ) +{ + int result; + carry_pool *pool; + carry_level *lowest_level; + carry_insert_data *cdata; + carry_op *op; + + assert("umka-314", coord != NULL); + + /* allocate carry_pool and 3 carry_level-s */ + pool = + init_carry_pool(sizeof(*pool) + 3 * sizeof(*lowest_level) + + sizeof(*cdata)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + lowest_level = (carry_level *) (pool + 1); + init_carry_level(lowest_level, pool); + + op = reiser4_post_carry(lowest_level, cop, coord->node, 0); + if (IS_ERR(op) || (op == NULL)) { + done_carry_pool(pool); + return RETERR(op ? PTR_ERR(op) : -EIO); + } + cdata = (carry_insert_data *) (lowest_level + 3); + cdata->coord = coord; + cdata->data = data; + cdata->key = key; + op->u.insert.d = cdata; + if (flags == 0) + flags = znode_get_tree(coord->node)->carry.insert_flags; + op->u.insert.flags = flags; + op->u.insert.type = COPT_ITEM_DATA; + op->u.insert.child = NULL; + if (lh != NULL) { + assert("nikita-3245", lh->node == coord->node); + lowest_level->track_type = CARRY_TRACK_CHANGE; + lowest_level->tracked = lh; + } + + result = reiser4_carry(lowest_level, NULL); + done_carry_pool(pool); + + return result; +} + +/* form carry queue to perform paste of @data with @key at @coord, and launch + its execution by calling carry(). + + Instruct carry to update @lh it after balancing insertion coord moves into + different block. + +*/ +static int paste_with_carry(coord_t *coord, /* coord of paste */ + lock_handle * lh, /* lock handle of node + * where item is + * pasted */ + reiser4_item_data * data, /* parameters of new + * item */ + const reiser4_key * key, /* key of new item */ + unsigned flags/* paste flags */) +{ + int result; + carry_pool *pool; + carry_level *lowest_level; + carry_insert_data *cdata; + carry_op *op; + + assert("umka-315", coord != NULL); + assert("umka-316", key != NULL); + + pool = + init_carry_pool(sizeof(*pool) + 3 * sizeof(*lowest_level) + + sizeof(*cdata)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + lowest_level = (carry_level *) (pool + 1); + init_carry_level(lowest_level, pool); + + op = reiser4_post_carry(lowest_level, COP_PASTE, coord->node, 0); + if (IS_ERR(op) || (op == NULL)) { + done_carry_pool(pool); + return RETERR(op ? PTR_ERR(op) : -EIO); + } + cdata = (carry_insert_data *) (lowest_level + 3); + cdata->coord = coord; + cdata->data = data; + cdata->key = key; + op->u.paste.d = cdata; + if (flags == 0) + flags = znode_get_tree(coord->node)->carry.paste_flags; + op->u.paste.flags = flags; + op->u.paste.type = COPT_ITEM_DATA; + if (lh != NULL) { + lowest_level->track_type = CARRY_TRACK_CHANGE; + lowest_level->tracked = lh; + } + + result = reiser4_carry(lowest_level, NULL); + done_carry_pool(pool); + + return result; +} + +/* insert item at the given coord. + + First try to skip carry by directly calling ->create_item() method of node + plugin. If this is impossible (there is not enough free space in the node, + or leftmost item in the node is created), call insert_with_carry_by_coord() + that will do full carry(). + +*/ +insert_result insert_by_coord(coord_t *coord /* coord where to + * insert. coord->node has + * to be write locked by + * caller */ , + reiser4_item_data * data /* data to be + * inserted */ , + const reiser4_key * key /* key of new item */ , + lock_handle * lh /* lock handle of write + * lock on node */ , + __u32 flags/* insertion flags */) +{ + unsigned item_size; + int result; + znode *node; + + assert("vs-247", coord != NULL); + assert("vs-248", data != NULL); + assert("vs-249", data->length >= 0); + assert("nikita-1191", znode_is_write_locked(coord->node)); + + node = coord->node; + coord_clear_iplug(coord); + result = zload(node); + if (result != 0) + return result; + + item_size = space_needed(node, NULL, data, 1); + if (item_size > znode_free_space(node) && + (flags & COPI_DONT_SHIFT_LEFT) && (flags & COPI_DONT_SHIFT_RIGHT) + && (flags & COPI_DONT_ALLOCATE)) { + /* we are forced to use free space of coord->node and new item + does not fit into it. + + Currently we get here only when we allocate and copy units + of extent item from a node to its left neighbor during + "squalloc"-ing. If @node (this is left neighbor) does not + have enough free space - we do not want to attempt any + shifting and allocations because we are in squeezing and + everything to the left of @node is tightly packed. + */ + result = -E_NODE_FULL; + } else if ((item_size <= znode_free_space(node)) && + !coord_is_before_leftmost(coord) && + (node_plugin_by_node(node)->fast_insert != NULL) + && node_plugin_by_node(node)->fast_insert(coord)) { + /* shortcut insertion without carry() overhead. + + Only possible if: + + - there is enough free space + + - insertion is not into the leftmost position in a node + (otherwise it would require updating of delimiting key in a + parent) + + - node plugin agrees with this + + */ + result = + node_plugin_by_node(node)->create_item(coord, key, data, + NULL); + znode_make_dirty(node); + } else { + /* otherwise do full-fledged carry(). */ + result = + insert_with_carry_by_coord(coord, lh, data, key, COP_INSERT, + flags); + } + zrelse(node); + return result; +} + +/* @coord is set to leaf level and @data is to be inserted to twig level */ +insert_result +insert_extent_by_coord(coord_t *coord, /* coord where to insert. + * coord->node has to be write + * locked by caller */ + reiser4_item_data *data,/* data to be inserted */ + const reiser4_key *key, /* key of new item */ + lock_handle *lh /* lock handle of write lock + on node */) +{ + assert("vs-405", coord != NULL); + assert("vs-406", data != NULL); + assert("vs-407", data->length > 0); + assert("vs-408", znode_is_write_locked(coord->node)); + assert("vs-409", znode_get_level(coord->node) == LEAF_LEVEL); + + return insert_with_carry_by_coord(coord, lh, data, key, COP_EXTENT, + 0 /*flags */ ); +} + +/* Insert into the item at the given coord. + + First try to skip carry by directly calling ->paste() method of item + plugin. If this is impossible (there is not enough free space in the node, + or we are pasting into leftmost position in the node), call + paste_with_carry() that will do full carry(). + +*/ +/* paste_into_item */ +int insert_into_item(coord_t * coord /* coord of pasting */ , + lock_handle * lh /* lock handle on node involved */ , + const reiser4_key * key /* key of unit being pasted */ , + reiser4_item_data * data /* parameters for new unit */ , + unsigned flags /* insert/paste flags */ ) +{ + int result; + int size_change; + node_plugin *nplug; + item_plugin *iplug; + + assert("umka-317", coord != NULL); + assert("umka-318", key != NULL); + + iplug = item_plugin_by_coord(coord); + nplug = node_plugin_by_coord(coord); + + assert("nikita-1480", iplug == data->iplug); + + size_change = space_needed(coord->node, coord, data, 0); + if (size_change > (int)znode_free_space(coord->node) && + (flags & COPI_DONT_SHIFT_LEFT) && (flags & COPI_DONT_SHIFT_RIGHT) + && (flags & COPI_DONT_ALLOCATE)) { + /* we are forced to use free space of coord->node and new data + does not fit into it. */ + return -E_NODE_FULL; + } + + /* shortcut paste without carry() overhead. + + Only possible if: + + - there is enough free space + + - paste is not into the leftmost unit in a node (otherwise + it would require updating of delimiting key in a parent) + + - node plugin agrees with this + + - item plugin agrees with us + */ + if (size_change <= (int)znode_free_space(coord->node) && + (coord->item_pos != 0 || + coord->unit_pos != 0 || coord->between == AFTER_UNIT) && + coord->unit_pos != 0 && nplug->fast_paste != NULL && + nplug->fast_paste(coord) && + iplug->b.fast_paste != NULL && iplug->b.fast_paste(coord)) { + if (size_change > 0) + nplug->change_item_size(coord, size_change); + /* NOTE-NIKITA: huh? where @key is used? */ + result = iplug->b.paste(coord, data, NULL); + if (size_change < 0) + nplug->change_item_size(coord, size_change); + znode_make_dirty(coord->node); + } else + /* otherwise do full-fledged carry(). */ + result = paste_with_carry(coord, lh, data, key, flags); + return result; +} + +/* this either appends or truncates item @coord */ +int reiser4_resize_item(coord_t * coord /* coord of item being resized */ , + reiser4_item_data * data /* parameters of resize */ , + reiser4_key * key /* key of new unit */ , + lock_handle * lh /* lock handle of node + * being modified */ , + cop_insert_flag flags /* carry flags */ ) +{ + int result; + znode *node; + + assert("nikita-362", coord != NULL); + assert("nikita-363", data != NULL); + assert("vs-245", data->length != 0); + + node = coord->node; + coord_clear_iplug(coord); + result = zload(node); + if (result != 0) + return result; + + if (data->length < 0) + result = node_plugin_by_coord(coord)->shrink_item(coord, + -data->length); + else + result = insert_into_item(coord, lh, key, data, flags); + + zrelse(node); + return result; +} + +/* insert flow @f */ +int reiser4_insert_flow(coord_t * coord, lock_handle * lh, flow_t * f) +{ + int result; + carry_pool *pool; + carry_level *lowest_level; + reiser4_item_data *data; + carry_op *op; + + pool = + init_carry_pool(sizeof(*pool) + 3 * sizeof(*lowest_level) + + sizeof(*data)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + lowest_level = (carry_level *) (pool + 1); + init_carry_level(lowest_level, pool); + + op = reiser4_post_carry(lowest_level, COP_INSERT_FLOW, coord->node, + 0 /* operate directly on coord -> node */ ); + if (IS_ERR(op) || (op == NULL)) { + done_carry_pool(pool); + return RETERR(op ? PTR_ERR(op) : -EIO); + } + + /* these are permanent during insert_flow */ + data = (reiser4_item_data *) (lowest_level + 3); + data->user = 1; + data->iplug = item_plugin_by_id(FORMATTING_ID); + data->arg = NULL; + /* data.length and data.data will be set before calling paste or + insert */ + data->length = 0; + data->data = NULL; + + op->u.insert_flow.flags = 0; + op->u.insert_flow.insert_point = coord; + op->u.insert_flow.flow = f; + op->u.insert_flow.data = data; + op->u.insert_flow.new_nodes = 0; + + lowest_level->track_type = CARRY_TRACK_CHANGE; + lowest_level->tracked = lh; + + result = reiser4_carry(lowest_level, NULL); + done_carry_pool(pool); + + return result; +} + +/* Given a coord in parent node, obtain a znode for the corresponding child */ +znode *child_znode(const coord_t * parent_coord /* coord of pointer to + * child */ , + znode * parent /* parent of child */ , + int incore_p /* if !0 only return child if already in + * memory */ , + int setup_dkeys_p /* if !0 update delimiting keys of + * child */ ) +{ + znode *child; + + assert("nikita-1374", parent_coord != NULL); + assert("nikita-1482", parent != NULL); +#if REISER4_DEBUG + if (setup_dkeys_p) + assert_rw_not_locked(&(znode_get_tree(parent)->dk_lock)); +#endif + assert("nikita-2947", znode_is_any_locked(parent)); + + if (znode_get_level(parent) <= LEAF_LEVEL) { + /* trying to get child of leaf node */ + warning("nikita-1217", "Child of maize?"); + return ERR_PTR(RETERR(-EIO)); + } + if (item_is_internal(parent_coord)) { + reiser4_block_nr addr; + item_plugin *iplug; + reiser4_tree *tree; + + iplug = item_plugin_by_coord(parent_coord); + assert("vs-512", iplug->s.internal.down_link); + iplug->s.internal.down_link(parent_coord, NULL, &addr); + + tree = znode_get_tree(parent); + if (incore_p) + child = zlook(tree, &addr); + else + child = + zget(tree, &addr, parent, + znode_get_level(parent) - 1, + reiser4_ctx_gfp_mask_get()); + if ((child != NULL) && !IS_ERR(child) && setup_dkeys_p) + set_child_delimiting_keys(parent, parent_coord, child); + } else { + warning("nikita-1483", "Internal item expected"); + child = ERR_PTR(RETERR(-EIO)); + } + return child; +} + +/* remove znode from transaction */ +static void uncapture_znode(znode * node) +{ + struct page *page; + + assert("zam-1001", ZF_ISSET(node, JNODE_HEARD_BANSHEE)); + + if (!reiser4_blocknr_is_fake(znode_get_block(node))) { + int ret; + + /* An already allocated block goes right to the atom's delete set. */ + ret = + reiser4_dealloc_block(znode_get_block(node), 0, + BA_DEFER | BA_FORMATTED); + if (ret) + warning("zam-942", + "can\'t add a block (%llu) number to atom's delete set\n", + (unsigned long long)(*znode_get_block(node))); + + spin_lock_znode(node); + /* Here we return flush reserved block which was reserved at the + * moment when this allocated node was marked dirty and still + * not used by flush in node relocation procedure. */ + if (ZF_ISSET(node, JNODE_FLUSH_RESERVED)) { + txn_atom *atom; + + atom = jnode_get_atom(ZJNODE(node)); + assert("zam-939", atom != NULL); + spin_unlock_znode(node); + flush_reserved2grabbed(atom, (__u64) 1); + spin_unlock_atom(atom); + } else + spin_unlock_znode(node); + } else { + /* znode has assigned block which is counted as "fake + allocated". Return it back to "free blocks") */ + fake_allocated2free((__u64) 1, BA_FORMATTED); + } + + /* + * uncapture page from transaction. There is a possibility of a race + * with ->releasepage(): reiser4_releasepage() detaches page from this + * jnode and we have nothing to uncapture. To avoid this, get + * reference of node->pg under jnode spin lock. reiser4_uncapture_page() + * will deal with released page itself. + */ + spin_lock_znode(node); + page = znode_page(node); + if (likely(page != NULL)) { + /* + * reiser4_uncapture_page() can only be called when we are sure + * that znode is pinned in memory, which we are, because + * forget_znode() is only called from longterm_unlock_znode(). + */ + page_cache_get(page); + spin_unlock_znode(node); + lock_page(page); + reiser4_uncapture_page(page); + unlock_page(page); + page_cache_release(page); + } else { + txn_atom *atom; + + /* handle "flush queued" znodes */ + while (1) { + atom = jnode_get_atom(ZJNODE(node)); + assert("zam-943", atom != NULL); + + if (!ZF_ISSET(node, JNODE_FLUSH_QUEUED) + || !atom->nr_running_queues) + break; + + spin_unlock_znode(node); + reiser4_atom_wait_event(atom); + spin_lock_znode(node); + } + + reiser4_uncapture_block(ZJNODE(node)); + spin_unlock_atom(atom); + zput(node); + } +} + +/* This is called from longterm_unlock_znode() when last lock is released from + the node that has been removed from the tree. At this point node is removed + from sibling list and its lock is invalidated. */ +void forget_znode(lock_handle * handle) +{ + znode *node; + reiser4_tree *tree; + + assert("umka-319", handle != NULL); + + node = handle->node; + tree = znode_get_tree(node); + + assert("vs-164", znode_is_write_locked(node)); + assert("nikita-1280", ZF_ISSET(node, JNODE_HEARD_BANSHEE)); + assert_rw_locked(&(node->lock.guard)); + + /* We assume that this node was detached from its parent before + * unlocking, it gives no way to reach this node from parent through a + * down link. The node should have no children and, thereby, can't be + * reached from them by their parent pointers. The only way to obtain a + * reference to the node is to use sibling pointers from its left and + * right neighbors. In the next several lines we remove the node from + * the sibling list. */ + + write_lock_tree(tree); + sibling_list_remove(node); + znode_remove(node, tree); + write_unlock_tree(tree); + + /* Here we set JNODE_DYING and cancel all pending lock requests. It + * forces all lock requestor threads to repeat iterations of getting + * lock on a child, neighbor or parent node. But, those threads can't + * come to this node again, because this node is no longer a child, + * neighbor or parent of any other node. This order of znode + * invalidation does not allow other threads to waste cpu time is a busy + * loop, trying to lock dying object. The exception is in the flush + * code when we take node directly from atom's capture list.*/ + reiser4_invalidate_lock(handle); + uncapture_znode(node); +} + +/* Check that internal item at @pointer really contains pointer to @child. */ +int check_tree_pointer(const coord_t * pointer /* would-be pointer to + * @child */ , + const znode * child /* child znode */ ) +{ + assert("nikita-1016", pointer != NULL); + assert("nikita-1017", child != NULL); + assert("nikita-1018", pointer->node != NULL); + + assert("nikita-1325", znode_is_any_locked(pointer->node)); + + assert("nikita-2985", + znode_get_level(pointer->node) == znode_get_level(child) + 1); + + coord_clear_iplug((coord_t *) pointer); + + if (coord_is_existing_unit(pointer)) { + item_plugin *iplug; + reiser4_block_nr addr; + + if (item_is_internal(pointer)) { + iplug = item_plugin_by_coord(pointer); + assert("vs-513", iplug->s.internal.down_link); + iplug->s.internal.down_link(pointer, NULL, &addr); + /* check that cached value is correct */ + if (disk_addr_eq(&addr, znode_get_block(child))) { + return NS_FOUND; + } + } + } + /* warning ("jmacd-1002", "tree pointer incorrect"); */ + return NS_NOT_FOUND; +} + +/* find coord of pointer to new @child in @parent. + + Find the &coord_t in the @parent where pointer to a given @child will + be in. + +*/ +int find_new_child_ptr(znode * parent /* parent znode, passed locked */ , + znode * + child UNUSED_ARG /* child znode, passed locked */ , + znode * left /* left brother of new node */ , + coord_t * result /* where result is stored in */ ) +{ + int ret; + + assert("nikita-1486", parent != NULL); + assert("nikita-1487", child != NULL); + assert("nikita-1488", result != NULL); + + ret = find_child_ptr(parent, left, result); + if (ret != NS_FOUND) { + warning("nikita-1489", "Cannot find brother position: %i", ret); + return RETERR(-EIO); + } else { + result->between = AFTER_UNIT; + return RETERR(NS_NOT_FOUND); + } +} + +/* find coord of pointer to @child in @parent. + + Find the &coord_t in the @parent where pointer to a given @child is in. + +*/ +int find_child_ptr(znode * parent /* parent znode, passed locked */ , + znode * child /* child znode, passed locked */ , + coord_t * result /* where result is stored in */ ) +{ + int lookup_res; + node_plugin *nplug; + /* left delimiting key of a child */ + reiser4_key ld; + reiser4_tree *tree; + + assert("nikita-934", parent != NULL); + assert("nikita-935", child != NULL); + assert("nikita-936", result != NULL); + assert("zam-356", znode_is_loaded(parent)); + + coord_init_zero(result); + result->node = parent; + + nplug = parent->nplug; + assert("nikita-939", nplug != NULL); + + tree = znode_get_tree(parent); + /* NOTE-NIKITA taking read-lock on tree here assumes that @result is + * not aliased to ->in_parent of some znode. Otherwise, + * parent_coord_to_coord() below would modify data protected by tree + * lock. */ + read_lock_tree(tree); + /* fast path. Try to use cached value. Lock tree to keep + node->pos_in_parent and pos->*_blocknr consistent. */ + if (child->in_parent.item_pos + 1 != 0) { + parent_coord_to_coord(&child->in_parent, result); + if (check_tree_pointer(result, child) == NS_FOUND) { + read_unlock_tree(tree); + return NS_FOUND; + } + + child->in_parent.item_pos = (unsigned short)~0; + } + read_unlock_tree(tree); + + /* is above failed, find some key from @child. We are looking for the + least key in a child. */ + read_lock_dk(tree); + ld = *znode_get_ld_key(child); + read_unlock_dk(tree); + /* + * now, lookup parent with key just found. Note, that left delimiting + * key doesn't identify node uniquely, because (in extremely rare + * case) two nodes can have equal left delimiting keys, if one of them + * is completely filled with directory entries that all happened to be + * hash collision. But, we check block number in check_tree_pointer() + * and, so, are safe. + */ + lookup_res = nplug->lookup(parent, &ld, FIND_EXACT, result); + /* update cached pos_in_node */ + if (lookup_res == NS_FOUND) { + write_lock_tree(tree); + coord_to_parent_coord(result, &child->in_parent); + write_unlock_tree(tree); + lookup_res = check_tree_pointer(result, child); + } + if (lookup_res == NS_NOT_FOUND) + lookup_res = find_child_by_addr(parent, child, result); + return lookup_res; +} + +/* find coord of pointer to @child in @parent by scanning + + Find the &coord_t in the @parent where pointer to a given @child + is in by scanning all internal items in @parent and comparing block + numbers in them with that of @child. + +*/ +static int find_child_by_addr(znode * parent /* parent znode, passed locked */ , + znode * child /* child znode, passed locked */ , + coord_t * result /* where result is stored in */ ) +{ + int ret; + + assert("nikita-1320", parent != NULL); + assert("nikita-1321", child != NULL); + assert("nikita-1322", result != NULL); + + ret = NS_NOT_FOUND; + + for_all_units(result, parent) { + if (check_tree_pointer(result, child) == NS_FOUND) { + write_lock_tree(znode_get_tree(parent)); + coord_to_parent_coord(result, &child->in_parent); + write_unlock_tree(znode_get_tree(parent)); + ret = NS_FOUND; + break; + } + } + return ret; +} + +/* true, if @addr is "unallocated block number", which is just address, with + highest bit set. */ +int is_disk_addr_unallocated(const reiser4_block_nr * addr /* address to + * check */ ) +{ + assert("nikita-1766", addr != NULL); + cassert(sizeof(reiser4_block_nr) == 8); + return (*addr & REISER4_BLOCKNR_STATUS_BIT_MASK) == + REISER4_UNALLOCATED_STATUS_VALUE; +} + +/* returns true if removing bytes of given range of key [from_key, to_key] + causes removing of whole item @from */ +static int +item_removed_completely(coord_t * from, const reiser4_key * from_key, + const reiser4_key * to_key) +{ + item_plugin *iplug; + reiser4_key key_in_item; + + assert("umka-325", from != NULL); + assert("", item_is_extent(from)); + + /* check first key just for case */ + item_key_by_coord(from, &key_in_item); + if (keygt(from_key, &key_in_item)) + return 0; + + /* check last key */ + iplug = item_plugin_by_coord(from); + assert("vs-611", iplug && iplug->s.file.append_key); + + iplug->s.file.append_key(from, &key_in_item); + set_key_offset(&key_in_item, get_key_offset(&key_in_item) - 1); + + if (keylt(to_key, &key_in_item)) + /* last byte is not removed */ + return 0; + return 1; +} + +/* helper function for prepare_twig_kill(): @left and @right are formatted + * neighbors of extent item being completely removed. Load and lock neighbors + * and store lock handles into @cdata for later use by kill_hook_extent() */ +static int +prepare_children(znode * left, znode * right, carry_kill_data * kdata) +{ + int result; + int left_loaded; + int right_loaded; + + result = 0; + left_loaded = right_loaded = 0; + + if (left != NULL) { + result = zload(left); + if (result == 0) { + left_loaded = 1; + result = longterm_lock_znode(kdata->left, left, + ZNODE_READ_LOCK, + ZNODE_LOCK_LOPRI); + } + } + if (result == 0 && right != NULL) { + result = zload(right); + if (result == 0) { + right_loaded = 1; + result = longterm_lock_znode(kdata->right, right, + ZNODE_READ_LOCK, + ZNODE_LOCK_HIPRI | + ZNODE_LOCK_NONBLOCK); + } + } + if (result != 0) { + done_lh(kdata->left); + done_lh(kdata->right); + if (left_loaded != 0) + zrelse(left); + if (right_loaded != 0) + zrelse(right); + } + return result; +} + +static void done_children(carry_kill_data * kdata) +{ + if (kdata->left != NULL && kdata->left->node != NULL) { + zrelse(kdata->left->node); + done_lh(kdata->left); + } + if (kdata->right != NULL && kdata->right->node != NULL) { + zrelse(kdata->right->node); + done_lh(kdata->right); + } +} + +/* part of cut_node. It is called when cut_node is called to remove or cut part + of extent item. When head of that item is removed - we have to update right + delimiting of left neighbor of extent. When item is removed completely - we + have to set sibling link between left and right neighbor of removed + extent. This may return -E_DEADLOCK because of trying to get left neighbor + locked. So, caller should repeat an attempt +*/ +/* Audited by: umka (2002.06.16) */ +static int +prepare_twig_kill(carry_kill_data * kdata, znode * locked_left_neighbor) +{ + int result; + reiser4_key key; + lock_handle left_lh; + lock_handle right_lh; + coord_t left_coord; + coord_t *from; + znode *left_child; + znode *right_child; + reiser4_tree *tree; + int left_zloaded_here, right_zloaded_here; + + from = kdata->params.from; + assert("umka-326", from != NULL); + assert("umka-327", kdata->params.to != NULL); + + /* for one extent item only yet */ + assert("vs-591", item_is_extent(from)); + assert("vs-592", from->item_pos == kdata->params.to->item_pos); + + if ((kdata->params.from_key + && keygt(kdata->params.from_key, item_key_by_coord(from, &key))) + || from->unit_pos != 0) { + /* head of item @from is not removed, there is nothing to + worry about */ + return 0; + } + + result = 0; + left_zloaded_here = 0; + right_zloaded_here = 0; + + left_child = right_child = NULL; + + coord_dup(&left_coord, from); + init_lh(&left_lh); + init_lh(&right_lh); + if (coord_prev_unit(&left_coord)) { + /* @from is leftmost item in its node */ + if (!locked_left_neighbor) { + result = + reiser4_get_left_neighbor(&left_lh, from->node, + ZNODE_READ_LOCK, + GN_CAN_USE_UPPER_LEVELS); + switch (result) { + case 0: + break; + case -E_NO_NEIGHBOR: + /* there is no formatted node to the left of + from->node */ + warning("vs-605", + "extent item has smallest key in " + "the tree and it is about to be removed"); + return 0; + case -E_DEADLOCK: + /* need to restart */ + default: + return result; + } + + /* we have acquired left neighbor of from->node */ + result = zload(left_lh.node); + if (result) + goto done; + + locked_left_neighbor = left_lh.node; + } else { + /* squalloc_right_twig_cut should have supplied locked + * left neighbor */ + assert("vs-834", + znode_is_write_locked(locked_left_neighbor)); + result = zload(locked_left_neighbor); + if (result) + return result; + } + + left_zloaded_here = 1; + coord_init_last_unit(&left_coord, locked_left_neighbor); + } + + if (!item_is_internal(&left_coord)) { + /* what else but extent can be on twig level */ + assert("vs-606", item_is_extent(&left_coord)); + + /* there is no left formatted child */ + if (left_zloaded_here) + zrelse(locked_left_neighbor); + done_lh(&left_lh); + return 0; + } + + tree = znode_get_tree(left_coord.node); + left_child = child_znode(&left_coord, left_coord.node, 1, 0); + + if (IS_ERR(left_child)) { + result = PTR_ERR(left_child); + goto done; + } + + /* left child is acquired, calculate new right delimiting key for it + and get right child if it is necessary */ + if (item_removed_completely + (from, kdata->params.from_key, kdata->params.to_key)) { + /* try to get right child of removed item */ + coord_t right_coord; + + assert("vs-607", + kdata->params.to->unit_pos == + coord_last_unit_pos(kdata->params.to)); + coord_dup(&right_coord, kdata->params.to); + if (coord_next_unit(&right_coord)) { + /* @to is rightmost unit in the node */ + result = + reiser4_get_right_neighbor(&right_lh, from->node, + ZNODE_READ_LOCK, + GN_CAN_USE_UPPER_LEVELS); + switch (result) { + case 0: + result = zload(right_lh.node); + if (result) + goto done; + + right_zloaded_here = 1; + coord_init_first_unit(&right_coord, + right_lh.node); + item_key_by_coord(&right_coord, &key); + break; + + case -E_NO_NEIGHBOR: + /* there is no formatted node to the right of + from->node */ + read_lock_dk(tree); + key = *znode_get_rd_key(from->node); + read_unlock_dk(tree); + right_coord.node = NULL; + result = 0; + break; + default: + /* real error */ + goto done; + } + } else { + /* there is an item to the right of @from - take its key */ + item_key_by_coord(&right_coord, &key); + } + + /* try to get right child of @from */ + if (right_coord.node && /* there is right neighbor of @from */ + item_is_internal(&right_coord)) { /* it is internal item */ + right_child = child_znode(&right_coord, + right_coord.node, 1, 0); + + if (IS_ERR(right_child)) { + result = PTR_ERR(right_child); + goto done; + } + + } + /* whole extent is removed between znodes left_child and right_child. Prepare them for linking and + update of right delimiting key of left_child */ + result = prepare_children(left_child, right_child, kdata); + } else { + /* head of item @to is removed. left_child has to get right delimting key update. Prepare it for that */ + result = prepare_children(left_child, NULL, kdata); + } + + done: + if (right_child) + zput(right_child); + if (right_zloaded_here) + zrelse(right_lh.node); + done_lh(&right_lh); + + if (left_child) + zput(left_child); + if (left_zloaded_here) + zrelse(locked_left_neighbor); + done_lh(&left_lh); + return result; +} + +/* this is used to remove part of node content between coordinates @from and @to. Units to which @from and @to are set + are to be cut completely */ +/* for try_to_merge_with_left, delete_copied, reiser4_delete_node */ +int cut_node_content(coord_t * from, coord_t * to, const reiser4_key * from_key, /* first key to be removed */ + const reiser4_key * to_key, /* last key to be removed */ + reiser4_key * + smallest_removed /* smallest key actually removed */ ) +{ + int result; + carry_pool *pool; + carry_level *lowest_level; + carry_cut_data *cut_data; + carry_op *op; + + assert("vs-1715", coord_compare(from, to) != COORD_CMP_ON_RIGHT); + + pool = + init_carry_pool(sizeof(*pool) + 3 * sizeof(*lowest_level) + + sizeof(*cut_data)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + lowest_level = (carry_level *) (pool + 1); + init_carry_level(lowest_level, pool); + + op = reiser4_post_carry(lowest_level, COP_CUT, from->node, 0); + assert("vs-1509", op != 0); + if (IS_ERR(op)) { + done_carry_pool(pool); + return PTR_ERR(op); + } + + cut_data = (carry_cut_data *) (lowest_level + 3); + cut_data->params.from = from; + cut_data->params.to = to; + cut_data->params.from_key = from_key; + cut_data->params.to_key = to_key; + cut_data->params.smallest_removed = smallest_removed; + + op->u.cut_or_kill.is_cut = 1; + op->u.cut_or_kill.u.cut = cut_data; + + result = reiser4_carry(lowest_level, NULL); + done_carry_pool(pool); + + return result; +} + +/* cut part of the node + + Cut part or whole content of node. + + cut data between @from and @to of @from->node and call carry() to make + corresponding changes in the tree. @from->node may become empty. If so - + pointer to it will be removed. Neighboring nodes are not changed. Smallest + removed key is stored in @smallest_removed + +*/ +int kill_node_content(coord_t * from, /* coord of the first unit/item that will be eliminated */ + coord_t * to, /* coord of the last unit/item that will be eliminated */ + const reiser4_key * from_key, /* first key to be removed */ + const reiser4_key * to_key, /* last key to be removed */ + reiser4_key * smallest_removed, /* smallest key actually removed */ + znode * locked_left_neighbor, /* this is set when kill_node_content is called with left neighbor + * locked (in squalloc_right_twig_cut, namely) */ + struct inode *inode, /* inode of file whose item (or its part) is to be killed. This is necessary to + invalidate pages together with item pointing to them */ + int truncate) +{ /* this call is made for file truncate) */ + int result; + carry_pool *pool; + carry_level *lowest_level; + carry_kill_data *kdata; + lock_handle *left_child; + lock_handle *right_child; + carry_op *op; + + assert("umka-328", from != NULL); + assert("vs-316", !node_is_empty(from->node)); + assert("nikita-1812", coord_is_existing_unit(from) + && coord_is_existing_unit(to)); + + /* allocate carry_pool, 3 carry_level-s, carry_kill_data and structures for kill_hook_extent */ + pool = init_carry_pool(sizeof(*pool) + 3 * sizeof(*lowest_level) + + sizeof(carry_kill_data) + + 2 * sizeof(lock_handle) + + 5 * sizeof(reiser4_key) + 2 * sizeof(coord_t)); + if (IS_ERR(pool)) + return PTR_ERR(pool); + + lowest_level = (carry_level *) (pool + 1); + init_carry_level(lowest_level, pool); + + kdata = (carry_kill_data *) (lowest_level + 3); + left_child = (lock_handle *) (kdata + 1); + right_child = left_child + 1; + + init_lh(left_child); + init_lh(right_child); + + kdata->params.from = from; + kdata->params.to = to; + kdata->params.from_key = from_key; + kdata->params.to_key = to_key; + kdata->params.smallest_removed = smallest_removed; + kdata->params.truncate = truncate; + kdata->flags = 0; + kdata->inode = inode; + kdata->left = left_child; + kdata->right = right_child; + /* memory for 5 reiser4_key and 2 coord_t will be used in kill_hook_extent */ + kdata->buf = (char *)(right_child + 1); + + if (znode_get_level(from->node) == TWIG_LEVEL && item_is_extent(from)) { + /* left child of extent item may have to get updated right + delimiting key and to get linked with right child of extent + @from if it will be removed completely */ + result = prepare_twig_kill(kdata, locked_left_neighbor); + if (result) { + done_children(kdata); + done_carry_pool(pool); + return result; + } + } + + op = reiser4_post_carry(lowest_level, COP_CUT, from->node, 0); + if (IS_ERR(op) || (op == NULL)) { + done_children(kdata); + done_carry_pool(pool); + return RETERR(op ? PTR_ERR(op) : -EIO); + } + + op->u.cut_or_kill.is_cut = 0; + op->u.cut_or_kill.u.kill = kdata; + + result = reiser4_carry(lowest_level, NULL); + + done_children(kdata); + done_carry_pool(pool); + return result; +} + +void +fake_kill_hook_tail(struct inode *inode, loff_t start, loff_t end, int truncate) +{ + if (reiser4_inode_get_flag(inode, REISER4_HAS_MMAP)) { + pgoff_t start_pg, end_pg; + + start_pg = start >> PAGE_CACHE_SHIFT; + end_pg = (end - 1) >> PAGE_CACHE_SHIFT; + + if ((start & (PAGE_CACHE_SIZE - 1)) == 0) { + /* + * kill up to the page boundary. + */ + assert("vs-123456", start_pg == end_pg); + reiser4_invalidate_pages(inode->i_mapping, start_pg, 1, + truncate); + } else if (start_pg != end_pg) { + /* + * page boundary is within killed portion of node. + */ + assert("vs-654321", end_pg - start_pg == 1); + reiser4_invalidate_pages(inode->i_mapping, end_pg, + end_pg - start_pg, 1); + } + } + inode_sub_bytes(inode, end - start); +} + +/** + * Delete whole @node from the reiser4 tree without loading it. + * + * @left: locked left neighbor, + * @node: node to be deleted, + * @smallest_removed: leftmost key of deleted node, + * @object: inode pointer, if we truncate a file body. + * @truncate: true if called for file truncate. + * + * @return: 0 if success, error code otherwise. + * + * NOTE: if @object!=NULL we assume that @smallest_removed != NULL and it + * contains the right value of the smallest removed key from the previous + * cut_worker() iteration. This is needed for proper accounting of + * "i_blocks" and "i_bytes" fields of the @object. + */ +int reiser4_delete_node(znode * node, reiser4_key * smallest_removed, + struct inode *object, int truncate) +{ + lock_handle parent_lock; + coord_t cut_from; + coord_t cut_to; + reiser4_tree *tree; + int ret; + + assert("zam-937", node != NULL); + assert("zam-933", znode_is_write_locked(node)); + assert("zam-999", smallest_removed != NULL); + + init_lh(&parent_lock); + + ret = reiser4_get_parent(&parent_lock, node, ZNODE_WRITE_LOCK); + if (ret) + return ret; + + assert("zam-934", !znode_above_root(parent_lock.node)); + + ret = zload(parent_lock.node); + if (ret) + goto failed_nozrelse; + + ret = find_child_ptr(parent_lock.node, node, &cut_from); + if (ret) + goto failed; + + /* decrement child counter and set parent pointer to NULL before + deleting the list from parent node because of checks in + internal_kill_item_hook (we can delete the last item from the parent + node, the parent node is going to be deleted and its c_count should + be zero). */ + + tree = znode_get_tree(node); + write_lock_tree(tree); + init_parent_coord(&node->in_parent, NULL); + --parent_lock.node->c_count; + write_unlock_tree(tree); + + assert("zam-989", item_is_internal(&cut_from)); + + /* @node should be deleted after unlocking. */ + ZF_SET(node, JNODE_HEARD_BANSHEE); + + /* remove a pointer from the parent node to the node being deleted. */ + coord_dup(&cut_to, &cut_from); + /* FIXME: shouldn't this be kill_node_content */ + ret = cut_node_content(&cut_from, &cut_to, NULL, NULL, NULL); + if (ret) + /* FIXME(Zam): Should we re-connect the node to its parent if + * cut_node fails? */ + goto failed; + + { + reiser4_tree *tree = current_tree; + __u64 start_offset = 0, end_offset = 0; + + read_lock_tree(tree); + write_lock_dk(tree); + if (object) { + /* We use @smallest_removed and the left delimiting of + * the current node for @object->i_blocks, i_bytes + * calculation. We assume that the items after the + * *@smallest_removed key have been deleted from the + * file body. */ + start_offset = get_key_offset(znode_get_ld_key(node)); + end_offset = get_key_offset(smallest_removed); + } + + assert("zam-1021", znode_is_connected(node)); + if (node->left) + znode_set_rd_key(node->left, znode_get_rd_key(node)); + + *smallest_removed = *znode_get_ld_key(node); + + write_unlock_dk(tree); + read_unlock_tree(tree); + + if (object) { + /* we used to perform actions which are to be performed on items on their removal from tree in + special item method - kill_hook. Here for optimization reasons we avoid reading node + containing item we remove and can not call item's kill hook. Instead we call function which + does exactly the same things as tail kill hook in assumption that node we avoid reading + contains only one item and that item is a tail one. */ + fake_kill_hook_tail(object, start_offset, end_offset, + truncate); + } + } + failed: + zrelse(parent_lock.node); + failed_nozrelse: + done_lh(&parent_lock); + + return ret; +} + +static int can_delete(const reiser4_key *key, znode *node) +{ + int result; + + read_lock_dk(current_tree); + result = keyle(key, znode_get_ld_key(node)); + read_unlock_dk(current_tree); + return result; +} + +/** + * This subroutine is not optimal but implementation seems to + * be easier). + * + * @tap: the point deletion process begins from, + * @from_key: the beginning of the deleted key range, + * @to_key: the end of the deleted key range, + * @smallest_removed: the smallest removed key, + * @truncate: true if called for file truncate. + * @progress: return true if a progress in file items deletions was made, + * @smallest_removed value is actual in that case. + * + * @return: 0 if success, error code otherwise, -E_REPEAT means that long + * reiser4_cut_tree operation was interrupted for allowing atom commit. + */ +int +cut_tree_worker_common(tap_t * tap, const reiser4_key * from_key, + const reiser4_key * to_key, + reiser4_key * smallest_removed, struct inode *object, + int truncate, int *progress) +{ + lock_handle next_node_lock; + coord_t left_coord; + int result; + + assert("zam-931", tap->coord->node != NULL); + assert("zam-932", znode_is_write_locked(tap->coord->node)); + + *progress = 0; + init_lh(&next_node_lock); + + while (1) { + znode *node; /* node from which items are cut */ + node_plugin *nplug; /* node plugin for @node */ + + node = tap->coord->node; + + /* Move next_node_lock to the next node on the left. */ + result = + reiser4_get_left_neighbor(&next_node_lock, node, + ZNODE_WRITE_LOCK, + GN_CAN_USE_UPPER_LEVELS); + if (result != 0 && result != -E_NO_NEIGHBOR) + break; + /* Check can we delete the node as a whole. */ + if (*progress && znode_get_level(node) == LEAF_LEVEL && + can_delete(from_key, node)) { + result = reiser4_delete_node(node, smallest_removed, + object, truncate); + } else { + result = reiser4_tap_load(tap); + if (result) + return result; + + /* Prepare the second (right) point for cut_node() */ + if (*progress) + coord_init_last_unit(tap->coord, node); + + else if (item_plugin_by_coord(tap->coord)->b.lookup == + NULL) + /* set rightmost unit for the items without lookup method */ + tap->coord->unit_pos = + coord_last_unit_pos(tap->coord); + + nplug = node->nplug; + + assert("vs-686", nplug); + assert("vs-687", nplug->lookup); + + /* left_coord is leftmost unit cut from @node */ + result = nplug->lookup(node, from_key, + FIND_MAX_NOT_MORE_THAN, + &left_coord); + + if (IS_CBKERR(result)) + break; + + /* adjust coordinates so that they are set to existing units */ + if (coord_set_to_right(&left_coord) + || coord_set_to_left(tap->coord)) { + result = 0; + break; + } + + if (coord_compare(&left_coord, tap->coord) == + COORD_CMP_ON_RIGHT) { + /* keys from @from_key to @to_key are not in the tree */ + result = 0; + break; + } + + if (left_coord.item_pos != tap->coord->item_pos) { + /* do not allow to cut more than one item. It is added to solve problem of truncating + partially converted files. If file is partially converted there may exist a twig node + containing both internal item or items pointing to leaf nodes with formatting items + and extent item. We do not want to kill internal items being at twig node here + because cut_tree_worker assumes killing them from level level */ + coord_dup(&left_coord, tap->coord); + assert("vs-1652", + coord_is_existing_unit(&left_coord)); + left_coord.unit_pos = 0; + } + + /* cut data from one node */ + /* *smallest_removed = *reiser4_min_key(); */ + result = + kill_node_content(&left_coord, tap->coord, from_key, + to_key, smallest_removed, + next_node_lock.node, object, + truncate); + reiser4_tap_relse(tap); + } + if (result) + break; + + ++(*progress); + + /* Check whether all items with keys >= from_key were removed + * from the tree. */ + if (keyle(smallest_removed, from_key)) + /* result = 0; */ + break; + + if (next_node_lock.node == NULL) + break; + + result = reiser4_tap_move(tap, &next_node_lock); + done_lh(&next_node_lock); + if (result) + break; + + /* Break long reiser4_cut_tree operation (deletion of a large + file) if atom requires commit. */ + if (*progress > CUT_TREE_MIN_ITERATIONS + && current_atom_should_commit()) { + result = -E_REPEAT; + break; + } + } + done_lh(&next_node_lock); + /* assert("vs-301", !keyeq(&smallest_removed, reiser4_min_key())); */ + return result; +} + +/* there is a fundamental problem with optimizing deletes: VFS does it + one file at a time. Another problem is that if an item can be + anything, then deleting items must be done one at a time. It just + seems clean to writes this to specify a from and a to key, and cut + everything between them though. */ + +/* use this function with care if deleting more than what is part of a single file. */ +/* do not use this when cutting a single item, it is suboptimal for that */ + +/* You are encouraged to write plugin specific versions of this. It + cannot be optimal for all plugins because it works item at a time, + and some plugins could sometimes work node at a time. Regular files + however are not optimizable to work node at a time because of + extents needing to free the blocks they point to. + + Optimizations compared to v3 code: + + It does not balance (that task is left to memory pressure code). + + Nodes are deleted only if empty. + + Uses extents. + + Performs read-ahead of formatted nodes whose contents are part of + the deletion. +*/ + +/** + * Delete everything from the reiser4 tree between two keys: @from_key and + * @to_key. + * + * @from_key: the beginning of the deleted key range, + * @to_key: the end of the deleted key range, + * @smallest_removed: the smallest removed key, + * @object: owner of cutting items. + * @truncate: true if called for file truncate. + * @progress: return true if a progress in file items deletions was made, + * @smallest_removed value is actual in that case. + * + * @return: 0 if success, error code otherwise, -E_REPEAT means that long cut_tree + * operation was interrupted for allowing atom commit . + */ + +int reiser4_cut_tree_object(reiser4_tree * tree, const reiser4_key * from_key, + const reiser4_key * to_key, + reiser4_key * smallest_removed_p, + struct inode *object, int truncate, int *progress) +{ + lock_handle lock; + int result; + tap_t tap; + coord_t right_coord; + reiser4_key smallest_removed; + int (*cut_tree_worker) (tap_t *, const reiser4_key *, + const reiser4_key *, reiser4_key *, + struct inode *, int, int *); + STORE_COUNTERS; + + assert("umka-329", tree != NULL); + assert("umka-330", from_key != NULL); + assert("umka-331", to_key != NULL); + assert("zam-936", keyle(from_key, to_key)); + + if (smallest_removed_p == NULL) + smallest_removed_p = &smallest_removed; + + init_lh(&lock); + + do { + /* Find rightmost item to cut away from the tree. */ + result = reiser4_object_lookup(object, to_key, &right_coord, + &lock, ZNODE_WRITE_LOCK, + FIND_MAX_NOT_MORE_THAN, + TWIG_LEVEL, LEAF_LEVEL, + CBK_UNIQUE, NULL /*ra_info */); + if (result != CBK_COORD_FOUND) + break; + if (object == NULL + || inode_file_plugin(object)->cut_tree_worker == NULL) + cut_tree_worker = cut_tree_worker_common; + else + cut_tree_worker = + inode_file_plugin(object)->cut_tree_worker; + reiser4_tap_init(&tap, &right_coord, &lock, ZNODE_WRITE_LOCK); + result = + cut_tree_worker(&tap, from_key, to_key, smallest_removed_p, + object, truncate, progress); + reiser4_tap_done(&tap); + + reiser4_preempt_point(); + + } while (0); + + done_lh(&lock); + + if (result) { + switch (result) { + case -E_NO_NEIGHBOR: + result = 0; + break; + case -E_DEADLOCK: + result = -E_REPEAT; + case -E_REPEAT: + case -ENOMEM: + case -ENOENT: + break; + default: + warning("nikita-2861", "failure: %i", result); + } + } + + CHECK_COUNTERS; + return result; +} + +/* repeat reiser4_cut_tree_object until everything is deleted. + * unlike cut_file_items, it does not end current transaction if -E_REPEAT + * is returned by cut_tree_object. */ +int reiser4_cut_tree(reiser4_tree * tree, const reiser4_key * from, + const reiser4_key * to, struct inode *inode, int truncate) +{ + int result; + int progress; + + do { + result = reiser4_cut_tree_object(tree, from, to, NULL, + inode, truncate, &progress); + } while (result == -E_REPEAT); + + return result; +} + +/* finishing reiser4 initialization */ +int reiser4_init_tree(reiser4_tree * tree /* pointer to structure being + * initialized */ , + const reiser4_block_nr * root_block /* address of a root block + * on a disk */ , + tree_level height /* height of a tree */ , + node_plugin * nplug /* default node plugin */ ) +{ + int result; + + assert("nikita-306", tree != NULL); + assert("nikita-307", root_block != NULL); + assert("nikita-308", height > 0); + assert("nikita-309", nplug != NULL); + assert("zam-587", tree->super != NULL); + + tree->root_block = *root_block; + tree->height = height; + tree->estimate_one_insert = calc_estimate_one_insert(height); + tree->nplug = nplug; + + tree->znode_epoch = 1ull; + + cbk_cache_init(&tree->cbk_cache); + + result = znodes_tree_init(tree); + if (result == 0) + result = jnodes_tree_init(tree); + if (result == 0) { + tree->uber = zget(tree, &UBER_TREE_ADDR, NULL, 0, + reiser4_ctx_gfp_mask_get()); + if (IS_ERR(tree->uber)) { + result = PTR_ERR(tree->uber); + tree->uber = NULL; + } + } + return result; +} + +/* release resources associated with @tree */ +void reiser4_done_tree(reiser4_tree * tree /* tree to release */ ) +{ + if (tree == NULL) + return; + + if (tree->uber != NULL) { + zput(tree->uber); + tree->uber = NULL; + } + znodes_tree_done(tree); + jnodes_tree_done(tree); + cbk_cache_done(&tree->cbk_cache); +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tree.h linux-2.6.30/fs/reiser4/tree.h --- linux-2.6.30.orig/fs/reiser4/tree.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tree.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,577 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Tree operations. See fs/reiser4/tree.c for comments */ + +#if !defined( __REISER4_TREE_H__ ) +#define __REISER4_TREE_H__ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "plugin/node/node.h" +#include "plugin/plugin.h" +#include "znode.h" +#include "tap.h" + +#include /* for __u?? */ +#include /* for struct super_block */ +#include +#include /* for struct task_struct */ + +/* fictive block number never actually used */ +extern const reiser4_block_nr UBER_TREE_ADDR; + +/* &cbk_cache_slot - entry in a coord cache. + + This is entry in a coord_by_key (cbk) cache, represented by + &cbk_cache. + +*/ +typedef struct cbk_cache_slot { + /* cached node */ + znode *node; + /* linkage to the next cbk cache slot in a LRU order */ + struct list_head lru; +} cbk_cache_slot; + +/* &cbk_cache - coord cache. This is part of reiser4_tree. + + cbk_cache is supposed to speed up tree lookups by caching results of recent + successful lookups (we don't cache negative results as dentry cache + does). Cache consists of relatively small number of entries kept in a LRU + order. Each entry (&cbk_cache_slot) contains a pointer to znode, from + which we can obtain a range of keys that covered by this znode. Before + embarking into real tree traversal we scan cbk_cache slot by slot and for + each slot check whether key we are looking for is between minimal and + maximal keys for node pointed to by this slot. If no match is found, real + tree traversal is performed and if result is successful, appropriate entry + is inserted into cache, possibly pulling least recently used entry out of + it. + + Tree spin lock is used to protect coord cache. If contention for this + lock proves to be too high, more finer grained locking can be added. + + Invariants involving parts of this data-type: + + [cbk-cache-invariant] +*/ +typedef struct cbk_cache { + /* serializator */ + rwlock_t guard; + int nr_slots; + /* head of LRU list of cache slots */ + struct list_head lru; + /* actual array of slots */ + cbk_cache_slot *slot; +} cbk_cache; + +/* level_lookup_result - possible outcome of looking up key at some level. + This is used by coord_by_key when traversing tree downward. */ +typedef enum { + /* continue to the next level */ + LOOKUP_CONT, + /* done. Either required item was found, or we can prove it + doesn't exist, or some error occurred. */ + LOOKUP_DONE, + /* restart traversal from the root. Infamous "repetition". */ + LOOKUP_REST +} level_lookup_result; + +/* This is representation of internal reiser4 tree where all file-system + data and meta-data are stored. This structure is passed to all tree + manipulation functions. It's different from the super block because: + we don't want to limit ourselves to strictly one to one mapping + between super blocks and trees, and, because they are logically + different: there are things in a super block that have no relation to + the tree (bitmaps, journalling area, mount options, etc.) and there + are things in a tree that bear no relation to the super block, like + tree of znodes. + + At this time, there is only one tree + per filesystem, and this struct is part of the super block. We only + call the super block the super block for historical reasons (most + other filesystems call the per filesystem metadata the super block). +*/ + +struct reiser4_tree { + /* block_nr == 0 is fake znode. Write lock it, while changing + tree height. */ + /* disk address of root node of a tree */ + reiser4_block_nr root_block; + + /* level of the root node. If this is 1, tree consists of root + node only */ + tree_level height; + + /* + * this is cached here avoid calling plugins through function + * dereference all the time. + */ + __u64 estimate_one_insert; + + /* cache of recent tree lookup results */ + cbk_cache cbk_cache; + + /* hash table to look up znodes by block number. */ + z_hash_table zhash_table; + z_hash_table zfake_table; + /* hash table to look up jnodes by inode and offset. */ + j_hash_table jhash_table; + + /* lock protecting: + - parent pointers, + - sibling pointers, + - znode hash table + - coord cache + */ + /* NOTE: The "giant" tree lock can be replaced by more spin locks, + hoping they will be less contented. We can use one spin lock per one + znode hash bucket. With adding of some code complexity, sibling + pointers can be protected by both znode spin locks. However it looks + more SMP scalable we should test this locking change on n-ways (n > + 4) SMP machines. Current 4-ways machine test does not show that tree + lock is contented and it is a bottleneck (2003.07.25). */ + + rwlock_t tree_lock; + + /* lock protecting delimiting keys */ + rwlock_t dk_lock; + + /* spin lock protecting znode_epoch */ + spinlock_t epoch_lock; + /* version stamp used to mark znode updates. See seal.[ch] for more + * information. */ + __u64 znode_epoch; + + znode *uber; + node_plugin *nplug; + struct super_block *super; + struct { + /* carry flags used for insertion of new nodes */ + __u32 new_node_flags; + /* carry flags used for insertion of new extents */ + __u32 new_extent_flags; + /* carry flags used for paste operations */ + __u32 paste_flags; + /* carry flags used for insert operations */ + __u32 insert_flags; + } carry; +}; + +extern int reiser4_init_tree(reiser4_tree * tree, + const reiser4_block_nr * root_block, + tree_level height, node_plugin * default_plugin); +extern void reiser4_done_tree(reiser4_tree * tree); + +/* cbk flags: options for coord_by_key() */ +typedef enum { + /* coord_by_key() is called for insertion. This is necessary because + of extents being located at the twig level. For explanation, see + comment just above is_next_item_internal(). + */ + CBK_FOR_INSERT = (1 << 0), + /* coord_by_key() is called with key that is known to be unique */ + CBK_UNIQUE = (1 << 1), + /* coord_by_key() can trust delimiting keys. This options is not user + accessible. coord_by_key() will set it automatically. It will be + only cleared by special-case in extents-on-the-twig-level handling + where it is necessary to insert item with a key smaller than + leftmost key in a node. This is necessary because of extents being + located at the twig level. For explanation, see comment just above + is_next_item_internal(). + */ + CBK_TRUST_DK = (1 << 2), + CBK_READA = (1 << 3), /* original: readahead leaves which contain items of certain file */ + CBK_READDIR_RA = (1 << 4), /* readdir: readahead whole directory and all its stat datas */ + CBK_DKSET = (1 << 5), + CBK_EXTENDED_COORD = (1 << 6), /* coord_t is actually */ + CBK_IN_CACHE = (1 << 7), /* node is already in cache */ + CBK_USE_CRABLOCK = (1 << 8) /* use crab_lock in stead of long term + * lock */ +} cbk_flags; + +/* insertion outcome. IBK = insert by key */ +typedef enum { + IBK_INSERT_OK = 0, + IBK_ALREADY_EXISTS = -EEXIST, + IBK_IO_ERROR = -EIO, + IBK_NO_SPACE = -E_NODE_FULL, + IBK_OOM = -ENOMEM +} insert_result; + +#define IS_CBKERR(err) ((err) != CBK_COORD_FOUND && (err) != CBK_COORD_NOTFOUND) + +typedef int (*tree_iterate_actor_t) (reiser4_tree * tree, coord_t * coord, + lock_handle * lh, void *arg); +extern int reiser4_iterate_tree(reiser4_tree * tree, coord_t * coord, + lock_handle * lh, + tree_iterate_actor_t actor, void *arg, + znode_lock_mode mode, int through_units_p); +extern int get_uber_znode(reiser4_tree * tree, znode_lock_mode mode, + znode_lock_request pri, lock_handle * lh); + +/* return node plugin of @node */ +static inline node_plugin *node_plugin_by_node(const znode * + node /* node to query */ ) +{ + assert("vs-213", node != NULL); + assert("vs-214", znode_is_loaded(node)); + + return node->nplug; +} + +/* number of items in @node */ +static inline pos_in_node_t node_num_items(const znode * node) +{ + assert("nikita-2754", znode_is_loaded(node)); + assert("nikita-2468", + node_plugin_by_node(node)->num_of_items(node) == node->nr_items); + + return node->nr_items; +} + +/* Return the number of items at the present node. Asserts coord->node != + NULL. */ +static inline unsigned coord_num_items(const coord_t * coord) +{ + assert("jmacd-9805", coord->node != NULL); + + return node_num_items(coord->node); +} + +/* true if @node is empty */ +static inline int node_is_empty(const znode * node) +{ + return node_num_items(node) == 0; +} + +typedef enum { + SHIFTED_SOMETHING = 0, + SHIFT_NO_SPACE = -E_NODE_FULL, + SHIFT_IO_ERROR = -EIO, + SHIFT_OOM = -ENOMEM, +} shift_result; + +extern node_plugin *node_plugin_by_coord(const coord_t * coord); +extern int is_coord_in_node(const coord_t * coord); +extern int key_in_node(const reiser4_key *, const coord_t *); +extern void coord_item_move_to(coord_t * coord, int items); +extern void coord_unit_move_to(coord_t * coord, int units); + +/* there are two types of repetitive accesses (ra): intra-syscall + (local) and inter-syscall (global). Local ra is used when + during single syscall we add/delete several items and units in the + same place in a tree. Note that plan-A fragments local ra by + separating stat-data and file body in key-space. Global ra is + used when user does repetitive modifications in the same place in a + tree. + + Our ra implementation serves following purposes: + 1 it affects balancing decisions so that next operation in a row + can be performed faster; + 2 it affects lower-level read-ahead in page-cache; + 3 it allows to avoid unnecessary lookups by maintaining some state + across several operations (this is only for local ra); + 4 it leaves room for lazy-micro-balancing: when we start a sequence of + operations they are performed without actually doing any intra-node + shifts, until we finish sequence or scope of sequence leaves + current node, only then we really pack node (local ra only). +*/ + +/* another thing that can be useful is to keep per-tree and/or + per-process cache of recent lookups. This cache can be organised as a + list of block numbers of formatted nodes sorted by starting key in + this node. Balancings should invalidate appropriate parts of this + cache. +*/ + +lookup_result coord_by_key(reiser4_tree * tree, const reiser4_key * key, + coord_t * coord, lock_handle * handle, + znode_lock_mode lock, lookup_bias bias, + tree_level lock_level, tree_level stop_level, + __u32 flags, ra_info_t *); + +lookup_result reiser4_object_lookup(struct inode *object, + const reiser4_key * key, + coord_t * coord, + lock_handle * lh, + znode_lock_mode lock_mode, + lookup_bias bias, + tree_level lock_level, + tree_level stop_level, + __u32 flags, ra_info_t * info); + +insert_result insert_by_key(reiser4_tree * tree, const reiser4_key * key, + reiser4_item_data * data, coord_t * coord, + lock_handle * lh, + tree_level stop_level, __u32 flags); +insert_result insert_by_coord(coord_t * coord, + reiser4_item_data * data, const reiser4_key * key, + lock_handle * lh, __u32); +insert_result insert_extent_by_coord(coord_t * coord, + reiser4_item_data * data, + const reiser4_key * key, lock_handle * lh); +int cut_node_content(coord_t * from, coord_t * to, const reiser4_key * from_key, + const reiser4_key * to_key, + reiser4_key * smallest_removed); +int kill_node_content(coord_t * from, coord_t * to, + const reiser4_key * from_key, const reiser4_key * to_key, + reiser4_key * smallest_removed, + znode * locked_left_neighbor, struct inode *inode, + int truncate); + +int reiser4_resize_item(coord_t * coord, reiser4_item_data * data, + reiser4_key * key, lock_handle * lh, cop_insert_flag); +int insert_into_item(coord_t * coord, lock_handle * lh, const reiser4_key * key, + reiser4_item_data * data, unsigned); +int reiser4_insert_flow(coord_t * coord, lock_handle * lh, flow_t * f); +int find_new_child_ptr(znode * parent, znode * child, znode * left, + coord_t * result); + +int shift_right_of_but_excluding_insert_coord(coord_t * insert_coord); +int shift_left_of_and_including_insert_coord(coord_t * insert_coord); + +void fake_kill_hook_tail(struct inode *, loff_t start, loff_t end, int); + +extern int cut_tree_worker_common(tap_t *, const reiser4_key *, + const reiser4_key *, reiser4_key *, + struct inode *, int, int *); +extern int reiser4_cut_tree_object(reiser4_tree *, const reiser4_key *, + const reiser4_key *, reiser4_key *, + struct inode *, int, int *); +extern int reiser4_cut_tree(reiser4_tree * tree, const reiser4_key * from, + const reiser4_key * to, struct inode *, int); + +extern int reiser4_delete_node(znode *, reiser4_key *, struct inode *, int); +extern int check_tree_pointer(const coord_t * pointer, const znode * child); +extern int find_new_child_ptr(znode * parent, znode * child UNUSED_ARG, + znode * left, coord_t * result); +extern int find_child_ptr(znode * parent, znode * child, coord_t * result); +extern int set_child_delimiting_keys(znode * parent, const coord_t * in_parent, + znode * child); +extern znode *child_znode(const coord_t * in_parent, znode * parent, + int incore_p, int setup_dkeys_p); + +extern int cbk_cache_init(cbk_cache * cache); +extern void cbk_cache_done(cbk_cache * cache); +extern void cbk_cache_invalidate(const znode * node, reiser4_tree * tree); + +extern char *sprint_address(const reiser4_block_nr * block); + +#if REISER4_DEBUG +extern void print_coord_content(const char *prefix, coord_t * p); +extern void reiser4_print_address(const char *prefix, + const reiser4_block_nr * block); +extern void print_tree_rec(const char *prefix, reiser4_tree * tree, + __u32 flags); +extern void check_dkeys(znode *node); +#else +#define print_coord_content(p, c) noop +#define reiser4_print_address(p, b) noop +#endif + +extern void forget_znode(lock_handle * handle); +extern int deallocate_znode(znode * node); + +extern int is_disk_addr_unallocated(const reiser4_block_nr * addr); + +/* struct used internally to pack all numerous arguments of tree lookup. + Used to avoid passing a lot of arguments to helper functions. */ +typedef struct cbk_handle { + /* tree we are in */ + reiser4_tree *tree; + /* key we are going after */ + const reiser4_key *key; + /* coord we will store result in */ + coord_t *coord; + /* type of lock to take on target node */ + znode_lock_mode lock_mode; + /* lookup bias. See comments at the declaration of lookup_bias */ + lookup_bias bias; + /* lock level: level starting from which tree traversal starts taking + * write locks. */ + tree_level lock_level; + /* level where search will stop. Either item will be found between + lock_level and stop_level, or CBK_COORD_NOTFOUND will be + returned. + */ + tree_level stop_level; + /* level we are currently at */ + tree_level level; + /* block number of @active node. Tree traversal operates on two + nodes: active and parent. */ + reiser4_block_nr block; + /* put here error message to be printed by caller */ + const char *error; + /* result passed back to caller */ + lookup_result result; + /* lock handles for active and parent */ + lock_handle *parent_lh; + lock_handle *active_lh; + reiser4_key ld_key; + reiser4_key rd_key; + /* flags, passed to the cbk routine. Bits of this bitmask are defined + in tree.h:cbk_flags enum. */ + __u32 flags; + ra_info_t *ra_info; + struct inode *object; +} cbk_handle; + +extern znode_lock_mode cbk_lock_mode(tree_level level, cbk_handle * h); + +/* eottl.c */ +extern int handle_eottl(cbk_handle *h, int *outcome); + +int lookup_multikey(cbk_handle * handle, int nr_keys); +int lookup_couple(reiser4_tree * tree, + const reiser4_key * key1, const reiser4_key * key2, + coord_t * coord1, coord_t * coord2, + lock_handle * lh1, lock_handle * lh2, + znode_lock_mode lock_mode, lookup_bias bias, + tree_level lock_level, tree_level stop_level, __u32 flags, + int *result1, int *result2); + +static inline void read_lock_tree(reiser4_tree *tree) +{ + /* check that tree is not locked */ + assert("", (LOCK_CNT_NIL(rw_locked_tree) && + LOCK_CNT_NIL(read_locked_tree) && + LOCK_CNT_NIL(write_locked_tree))); + /* check that spinlocks of lower priorities are not held */ + assert("", (LOCK_CNT_NIL(spin_locked_txnh) && + LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(spin_locked_stack))); + + read_lock(&(tree->tree_lock)); + + LOCK_CNT_INC(read_locked_tree); + LOCK_CNT_INC(rw_locked_tree); + LOCK_CNT_INC(spin_locked); +} + +static inline void read_unlock_tree(reiser4_tree *tree) +{ + assert("nikita-1375", LOCK_CNT_GTZ(read_locked_tree)); + assert("nikita-1376", LOCK_CNT_GTZ(rw_locked_tree)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(read_locked_tree); + LOCK_CNT_DEC(rw_locked_tree); + LOCK_CNT_DEC(spin_locked); + + read_unlock(&(tree->tree_lock)); +} + +static inline void write_lock_tree(reiser4_tree *tree) +{ + /* check that tree is not locked */ + assert("", (LOCK_CNT_NIL(rw_locked_tree) && + LOCK_CNT_NIL(read_locked_tree) && + LOCK_CNT_NIL(write_locked_tree))); + /* check that spinlocks of lower priorities are not held */ + assert("", (LOCK_CNT_NIL(spin_locked_txnh) && + LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(spin_locked_stack))); + + write_lock(&(tree->tree_lock)); + + LOCK_CNT_INC(write_locked_tree); + LOCK_CNT_INC(rw_locked_tree); + LOCK_CNT_INC(spin_locked); +} + +static inline void write_unlock_tree(reiser4_tree *tree) +{ + assert("nikita-1375", LOCK_CNT_GTZ(write_locked_tree)); + assert("nikita-1376", LOCK_CNT_GTZ(rw_locked_tree)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(write_locked_tree); + LOCK_CNT_DEC(rw_locked_tree); + LOCK_CNT_DEC(spin_locked); + + write_unlock(&(tree->tree_lock)); +} + +static inline void read_lock_dk(reiser4_tree *tree) +{ + /* check that dk is not locked */ + assert("", (LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(read_locked_dk) && + LOCK_CNT_NIL(write_locked_dk))); + /* check that spinlocks of lower priorities are not held */ + assert("", LOCK_CNT_NIL(spin_locked_stack)); + + read_lock(&((tree)->dk_lock)); + + LOCK_CNT_INC(read_locked_dk); + LOCK_CNT_INC(rw_locked_dk); + LOCK_CNT_INC(spin_locked); +} + +static inline void read_unlock_dk(reiser4_tree *tree) +{ + assert("nikita-1375", LOCK_CNT_GTZ(read_locked_dk)); + assert("nikita-1376", LOCK_CNT_GTZ(rw_locked_dk)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(read_locked_dk); + LOCK_CNT_DEC(rw_locked_dk); + LOCK_CNT_DEC(spin_locked); + + read_unlock(&(tree->dk_lock)); +} + +static inline void write_lock_dk(reiser4_tree *tree) +{ + /* check that dk is not locked */ + assert("", (LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(read_locked_dk) && + LOCK_CNT_NIL(write_locked_dk))); + /* check that spinlocks of lower priorities are not held */ + assert("", LOCK_CNT_NIL(spin_locked_stack)); + + write_lock(&((tree)->dk_lock)); + + LOCK_CNT_INC(write_locked_dk); + LOCK_CNT_INC(rw_locked_dk); + LOCK_CNT_INC(spin_locked); +} + +static inline void write_unlock_dk(reiser4_tree *tree) +{ + assert("nikita-1375", LOCK_CNT_GTZ(write_locked_dk)); + assert("nikita-1376", LOCK_CNT_GTZ(rw_locked_dk)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(write_locked_dk); + LOCK_CNT_DEC(rw_locked_dk); + LOCK_CNT_DEC(spin_locked); + + write_unlock(&(tree->dk_lock)); +} + +/* estimate api. Implementation is in estimate.c */ +reiser4_block_nr estimate_one_insert_item(reiser4_tree *); +reiser4_block_nr estimate_one_insert_into_item(reiser4_tree *); +reiser4_block_nr estimate_insert_flow(tree_level); +reiser4_block_nr estimate_one_item_removal(reiser4_tree *); +reiser4_block_nr calc_estimate_one_insert(tree_level); +reiser4_block_nr estimate_dirty_cluster(struct inode *); +reiser4_block_nr estimate_insert_cluster(struct inode *); +reiser4_block_nr estimate_update_cluster(struct inode *); + +/* __REISER4_TREE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tree_mod.c linux-2.6.30/fs/reiser4/tree_mod.c --- linux-2.6.30.orig/fs/reiser4/tree_mod.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tree_mod.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,386 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* + * Functions to add/delete new nodes to/from the tree. + * + * Functions from this file are used by carry (see carry*) to handle: + * + * . insertion of new formatted node into tree + * + * . addition of new tree root, increasing tree height + * + * . removing tree root, decreasing tree height + * + */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "plugin/plugin.h" +#include "jnode.h" +#include "znode.h" +#include "tree_mod.h" +#include "block_alloc.h" +#include "tree_walk.h" +#include "tree.h" +#include "super.h" + +#include + +static int add_child_ptr(znode * parent, znode * child); +/* warning only issued if error is not -E_REPEAT */ +#define ewarning( error, ... ) \ + if( ( error ) != -E_REPEAT ) \ + warning( __VA_ARGS__ ) + +/* allocate new node on the @level and immediately on the right of @brother. */ +znode * reiser4_new_node(znode * brother /* existing left neighbor + * of new node */, + tree_level level /* tree level at which new node is to + * be allocated */) +{ + znode *result; + int retcode; + reiser4_block_nr blocknr; + + assert("nikita-930", brother != NULL); + assert("umka-264", level < REAL_MAX_ZTREE_HEIGHT); + + retcode = assign_fake_blocknr_formatted(&blocknr); + if (retcode == 0) { + result = + zget(znode_get_tree(brother), &blocknr, NULL, level, + reiser4_ctx_gfp_mask_get()); + if (IS_ERR(result)) { + ewarning(PTR_ERR(result), "nikita-929", + "Cannot allocate znode for carry: %li", + PTR_ERR(result)); + return result; + } + /* cheap test, can be executed even when debugging is off */ + if (!znode_just_created(result)) { + warning("nikita-2213", + "Allocated already existing block: %llu", + (unsigned long long)blocknr); + zput(result); + return ERR_PTR(RETERR(-EIO)); + } + + assert("nikita-931", result != NULL); + result->nplug = znode_get_tree(brother)->nplug; + assert("nikita-933", result->nplug != NULL); + + retcode = zinit_new(result, reiser4_ctx_gfp_mask_get()); + if (retcode == 0) { + ZF_SET(result, JNODE_CREATED); + zrelse(result); + } else { + zput(result); + result = ERR_PTR(retcode); + } + } else { + /* failure to allocate new node during balancing. + This should never happen. Ever. Returning -E_REPEAT + is not viable solution, because "out of disk space" + is not transient error that will go away by itself. + */ + ewarning(retcode, "nikita-928", + "Cannot allocate block for carry: %i", retcode); + result = ERR_PTR(retcode); + } + assert("nikita-1071", result != NULL); + return result; +} + +/* allocate new root and add it to the tree + + This helper function is called by add_new_root(). + +*/ +znode *reiser4_add_tree_root(znode * old_root /* existing tree root */ , + znode * fake /* "fake" znode */ ) +{ + reiser4_tree *tree = znode_get_tree(old_root); + znode *new_root = NULL; /* to shut gcc up */ + int result; + + assert("nikita-1069", old_root != NULL); + assert("umka-262", fake != NULL); + assert("umka-263", tree != NULL); + + /* "fake" znode---one always hanging just above current root. This + node is locked when new root is created or existing root is + deleted. Downward tree traversal takes lock on it before taking + lock on a root node. This avoids race conditions with root + manipulations. + + */ + assert("nikita-1348", znode_above_root(fake)); + assert("nikita-1211", znode_is_root(old_root)); + + result = 0; + if (tree->height >= REAL_MAX_ZTREE_HEIGHT) { + warning("nikita-1344", "Tree is too tall: %i", tree->height); + /* ext2 returns -ENOSPC when it runs out of free inodes with a + following comment (fs/ext2/ialloc.c:441): Is it really + ENOSPC? + + -EXFULL? -EINVAL? + */ + result = RETERR(-ENOSPC); + } else { + /* Allocate block for new root. It's not that + important where it will be allocated, as root is + almost always in memory. Moreover, allocate on + flush can be going here. + */ + assert("nikita-1448", znode_is_root(old_root)); + new_root = reiser4_new_node(fake, tree->height + 1); + if (!IS_ERR(new_root) && (result = zload(new_root)) == 0) { + lock_handle rlh; + + init_lh(&rlh); + result = + longterm_lock_znode(&rlh, new_root, + ZNODE_WRITE_LOCK, + ZNODE_LOCK_LOPRI); + if (result == 0) { + parent_coord_t *in_parent; + + znode_make_dirty(fake); + + /* new root is a child of "fake" node */ + write_lock_tree(tree); + + ++tree->height; + + /* recalculate max balance overhead */ + tree->estimate_one_insert = + estimate_one_insert_item(tree); + + tree->root_block = *znode_get_block(new_root); + in_parent = &new_root->in_parent; + init_parent_coord(in_parent, fake); + /* manually insert new root into sibling + * list. With this all nodes involved into + * balancing are connected after balancing is + * done---useful invariant to check. */ + sibling_list_insert_nolock(new_root, NULL); + write_unlock_tree(tree); + + /* insert into new root pointer to the + @old_root. */ + assert("nikita-1110", + WITH_DATA(new_root, + node_is_empty(new_root))); + write_lock_dk(tree); + znode_set_ld_key(new_root, reiser4_min_key()); + znode_set_rd_key(new_root, reiser4_max_key()); + write_unlock_dk(tree); + if (REISER4_DEBUG) { + ZF_CLR(old_root, JNODE_LEFT_CONNECTED); + ZF_CLR(old_root, JNODE_RIGHT_CONNECTED); + ZF_SET(old_root, JNODE_ORPHAN); + } + result = add_child_ptr(new_root, old_root); + done_lh(&rlh); + } + zrelse(new_root); + } + } + if (result != 0) + new_root = ERR_PTR(result); + return new_root; +} + +/* build &reiser4_item_data for inserting child pointer + + Build &reiser4_item_data that can be later used to insert pointer to @child + in its parent. + +*/ +void build_child_ptr_data(znode * child /* node pointer to which will be + * inserted */ , + reiser4_item_data * data /* where to store result */ ) +{ + assert("nikita-1116", child != NULL); + assert("nikita-1117", data != NULL); + + /* + * NOTE: use address of child's blocknr as address of data to be + * inserted. As result of this data gets into on-disk structure in cpu + * byte order. internal's create_hook converts it to little endian byte + * order. + */ + data->data = (char *)znode_get_block(child); + /* data -> data is kernel space */ + data->user = 0; + data->length = sizeof(reiser4_block_nr); + /* FIXME-VS: hardcoded internal item? */ + + /* AUDIT: Is it possible that "item_plugin_by_id" may find nothing? */ + data->iplug = item_plugin_by_id(NODE_POINTER_ID); +} + +/* add pointer to @child into empty @parent. + + This is used when pointer to old root is inserted into new root which is + empty. +*/ +static int add_child_ptr(znode * parent, znode * child) +{ + coord_t coord; + reiser4_item_data data; + int result; + reiser4_key key; + + assert("nikita-1111", parent != NULL); + assert("nikita-1112", child != NULL); + assert("nikita-1115", + znode_get_level(parent) == znode_get_level(child) + 1); + + result = zload(parent); + if (result != 0) + return result; + assert("nikita-1113", node_is_empty(parent)); + coord_init_first_unit(&coord, parent); + + build_child_ptr_data(child, &data); + data.arg = NULL; + + read_lock_dk(znode_get_tree(parent)); + key = *znode_get_ld_key(child); + read_unlock_dk(znode_get_tree(parent)); + + result = node_plugin_by_node(parent)->create_item(&coord, &key, &data, + NULL); + znode_make_dirty(parent); + zrelse(parent); + return result; +} + +/* actually remove tree root */ +static int reiser4_kill_root(reiser4_tree * tree /* tree from which root is + * being removed */, + znode * old_root /* root node that is being + * removed */ , + znode * new_root /* new root---sole child of + * @old_root */, + const reiser4_block_nr * new_root_blk /* disk address of + * @new_root */) +{ + znode *uber; + int result; + lock_handle handle_for_uber; + + assert("umka-265", tree != NULL); + assert("nikita-1198", new_root != NULL); + assert("nikita-1199", + znode_get_level(new_root) + 1 == znode_get_level(old_root)); + + assert("nikita-1201", znode_is_write_locked(old_root)); + + assert("nikita-1203", + disk_addr_eq(new_root_blk, znode_get_block(new_root))); + + init_lh(&handle_for_uber); + /* obtain and lock "fake" znode protecting changes in tree height. */ + result = get_uber_znode(tree, ZNODE_WRITE_LOCK, ZNODE_LOCK_HIPRI, + &handle_for_uber); + if (result == 0) { + uber = handle_for_uber.node; + + znode_make_dirty(uber); + + /* don't take long term lock a @new_root. Take spinlock. */ + + write_lock_tree(tree); + + tree->root_block = *new_root_blk; + --tree->height; + + /* recalculate max balance overhead */ + tree->estimate_one_insert = estimate_one_insert_item(tree); + + assert("nikita-1202", + tree->height == znode_get_level(new_root)); + + /* new root is child on "fake" node */ + init_parent_coord(&new_root->in_parent, uber); + ++uber->c_count; + + /* sibling_list_insert_nolock(new_root, NULL); */ + write_unlock_tree(tree); + + /* reinitialise old root. */ + result = node_plugin_by_node(old_root)->init(old_root); + znode_make_dirty(old_root); + if (result == 0) { + assert("nikita-1279", node_is_empty(old_root)); + ZF_SET(old_root, JNODE_HEARD_BANSHEE); + old_root->c_count = 0; + } + } + done_lh(&handle_for_uber); + + return result; +} + +/* remove tree root + + This function removes tree root, decreasing tree height by one. Tree root + and its only child (that is going to become new tree root) are write locked + at the entry. + + To remove tree root we need to take lock on special "fake" znode that + protects changes of tree height. See comments in reiser4_add_tree_root() for + more on this. + + Also parent pointers have to be updated in + old and new root. To simplify code, function is split into two parts: outer + reiser4_kill_tree_root() collects all necessary arguments and calls + reiser4_kill_root() to do the actual job. + +*/ +int reiser4_kill_tree_root(znode * old_root /* tree root that we are + removing*/) +{ + int result; + coord_t down_link; + znode *new_root; + reiser4_tree *tree; + + assert("umka-266", current_tree != NULL); + assert("nikita-1194", old_root != NULL); + assert("nikita-1196", znode_is_root(old_root)); + assert("nikita-1200", node_num_items(old_root) == 1); + assert("nikita-1401", znode_is_write_locked(old_root)); + + coord_init_first_unit(&down_link, old_root); + + tree = znode_get_tree(old_root); + new_root = child_znode(&down_link, old_root, 0, 1); + if (!IS_ERR(new_root)) { + result = + reiser4_kill_root(tree, old_root, new_root, + znode_get_block(new_root)); + zput(new_root); + } else + result = PTR_ERR(new_root); + + return result; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tree_mod.h linux-2.6.30/fs/reiser4/tree_mod.h --- linux-2.6.30.orig/fs/reiser4/tree_mod.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tree_mod.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,29 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Functions to add/delete new nodes to/from the tree. See tree_mod.c for + * comments. */ + +#if !defined( __REISER4_TREE_MOD_H__ ) +#define __REISER4_TREE_MOD_H__ + +#include "forward.h" + +znode *reiser4_new_node(znode * brother, tree_level level); +znode *reiser4_add_tree_root(znode * old_root, znode * fake); +int reiser4_kill_tree_root(znode * old_root); +void build_child_ptr_data(znode * child, reiser4_item_data * data); + +/* __REISER4_TREE_MOD_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tree_walk.c linux-2.6.30/fs/reiser4/tree_walk.c --- linux-2.6.30.orig/fs/reiser4/tree_walk.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tree_walk.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,927 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Routines and macros to: + + get_left_neighbor() + + get_right_neighbor() + + get_parent() + + get_first_child() + + get_last_child() + + various routines to walk the whole tree and do things to it like + repack it, or move it to tertiary storage. Please make them as + generic as is reasonable. + +*/ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "jnode.h" +#include "znode.h" +#include "tree_walk.h" +#include "tree.h" +#include "super.h" + +/* These macros are used internally in tree_walk.c in attempt to make + lock_neighbor() code usable to build lock_parent(), lock_right_neighbor, + lock_left_neighbor */ +#define GET_NODE_BY_PTR_OFFSET(node, off) (*(znode**)(((unsigned long)(node)) + (off))) +#define FIELD_OFFSET(name) offsetof(znode, name) +#define PARENT_PTR_OFFSET FIELD_OFFSET(in_parent.node) +#define LEFT_PTR_OFFSET FIELD_OFFSET(left) +#define RIGHT_PTR_OFFSET FIELD_OFFSET(right) + +/* This is the generic procedure to get and lock `generic' neighbor (left or + right neighbor or parent). It implements common algorithm for all cases of + getting lock on neighbor node, only znode structure field is different in + each case. This is parameterized by ptr_offset argument, which is byte + offset for the pointer to the desired neighbor within the current node's + znode structure. This function should be called with the tree lock held */ +static int lock_neighbor( + /* resulting lock handle */ + lock_handle * result, + /* znode to lock */ + znode * node, + /* pointer to neighbor (or parent) znode field offset, in bytes from + the base address of znode structure */ + int ptr_offset, + /* lock mode for longterm_lock_znode call */ + znode_lock_mode mode, + /* lock request for longterm_lock_znode call */ + znode_lock_request req, + /* GN_* flags */ + int flags, int rlocked) +{ + reiser4_tree *tree = znode_get_tree(node); + znode *neighbor; + int ret; + + assert("umka-236", node != NULL); + assert("umka-237", tree != NULL); + assert_rw_locked(&(tree->tree_lock)); + + if (flags & GN_TRY_LOCK) + req |= ZNODE_LOCK_NONBLOCK; + if (flags & GN_SAME_ATOM) + req |= ZNODE_LOCK_DONT_FUSE; + + /* get neighbor's address by using of sibling link, quit while loop + (and return) if link is not available. */ + while (1) { + neighbor = GET_NODE_BY_PTR_OFFSET(node, ptr_offset); + + /* return -E_NO_NEIGHBOR if parent or side pointer is NULL or if + * node pointed by it is not connected. + * + * However, GN_ALLOW_NOT_CONNECTED option masks "connected" + * check and allows passing reference to not connected znode to + * subsequent longterm_lock_znode() call. This kills possible + * busy loop if we are trying to get longterm lock on locked but + * not yet connected parent node. */ + if (neighbor == NULL || !((flags & GN_ALLOW_NOT_CONNECTED) + || znode_is_connected(neighbor))) { + return RETERR(-E_NO_NEIGHBOR); + } + + /* protect it from deletion. */ + zref(neighbor); + + rlocked ? read_unlock_tree(tree) : write_unlock_tree(tree); + + ret = longterm_lock_znode(result, neighbor, mode, req); + + /* The lock handle obtains its own reference, release the one from above. */ + zput(neighbor); + + rlocked ? read_lock_tree(tree) : write_lock_tree(tree); + + /* restart if node we got reference to is being + invalidated. we should not get reference to this node + again. */ + if (ret == -EINVAL) + continue; + if (ret) + return ret; + + /* check if neighbor link still points to just locked znode; + the link could have been changed while the process slept. */ + if (neighbor == GET_NODE_BY_PTR_OFFSET(node, ptr_offset)) + return 0; + + /* znode was locked by mistake; unlock it and restart locking + process from beginning. */ + rlocked ? read_unlock_tree(tree) : write_unlock_tree(tree); + longterm_unlock_znode(result); + rlocked ? read_lock_tree(tree) : write_lock_tree(tree); + } +} + +/* get parent node with longterm lock, accepts GN* flags. */ +int reiser4_get_parent_flags(lock_handle * lh /* resulting lock handle */ , + znode * node /* child node */ , + znode_lock_mode mode + /* type of lock: read or write */ , + int flags /* GN_* flags */ ) +{ + int result; + + read_lock_tree(znode_get_tree(node)); + result = lock_neighbor(lh, node, PARENT_PTR_OFFSET, mode, + ZNODE_LOCK_HIPRI, flags, 1); + read_unlock_tree(znode_get_tree(node)); + return result; +} + +/* wrapper function to lock right or left neighbor depending on GN_GO_LEFT + bit in @flags parameter */ +/* Audited by: umka (2002.06.14) */ +static inline int +lock_side_neighbor(lock_handle * result, + znode * node, znode_lock_mode mode, int flags, int rlocked) +{ + int ret; + int ptr_offset; + znode_lock_request req; + + if (flags & GN_GO_LEFT) { + ptr_offset = LEFT_PTR_OFFSET; + req = ZNODE_LOCK_LOPRI; + } else { + ptr_offset = RIGHT_PTR_OFFSET; + req = ZNODE_LOCK_HIPRI; + } + + ret = + lock_neighbor(result, node, ptr_offset, mode, req, flags, rlocked); + + if (ret == -E_NO_NEIGHBOR) /* if we walk left or right -E_NO_NEIGHBOR does not + * guarantee that neighbor is absent in the + * tree; in this case we return -ENOENT -- + * means neighbor at least not found in + * cache */ + return RETERR(-ENOENT); + + return ret; +} + +#if REISER4_DEBUG + +int check_sibling_list(znode * node) +{ + znode *scan; + znode *next; + + assert("nikita-3283", LOCK_CNT_GTZ(write_locked_tree)); + + if (node == NULL) + return 1; + + if (ZF_ISSET(node, JNODE_RIP)) + return 1; + + assert("nikita-3270", node != NULL); + assert_rw_write_locked(&(znode_get_tree(node)->tree_lock)); + + for (scan = node; znode_is_left_connected(scan); scan = next) { + next = scan->left; + if (next != NULL && !ZF_ISSET(next, JNODE_RIP)) { + assert("nikita-3271", znode_is_right_connected(next)); + assert("nikita-3272", next->right == scan); + } else + break; + } + for (scan = node; znode_is_right_connected(scan); scan = next) { + next = scan->right; + if (next != NULL && !ZF_ISSET(next, JNODE_RIP)) { + assert("nikita-3273", znode_is_left_connected(next)); + assert("nikita-3274", next->left == scan); + } else + break; + } + return 1; +} + +#endif + +/* Znode sibling pointers maintenence. */ + +/* Znode sibling pointers are established between any neighbored nodes which are + in cache. There are two znode state bits (JNODE_LEFT_CONNECTED, + JNODE_RIGHT_CONNECTED), if left or right sibling pointer contains actual + value (even NULL), corresponded JNODE_*_CONNECTED bit is set. + + Reiser4 tree operations which may allocate new znodes (CBK, tree balancing) + take care about searching (hash table lookup may be required) of znode + neighbors, establishing sibling pointers between them and setting + JNODE_*_CONNECTED state bits. */ + +/* adjusting of sibling pointers and `connected' states for two + neighbors; works if one neighbor is NULL (was not found). */ + +/* FIXME-VS: this is unstatic-ed to use in tree.c in prepare_twig_cut */ +void link_left_and_right(znode * left, znode * right) +{ + assert("nikita-3275", check_sibling_list(left)); + assert("nikita-3275", check_sibling_list(right)); + + if (left != NULL) { + if (left->right == NULL) { + left->right = right; + ZF_SET(left, JNODE_RIGHT_CONNECTED); + + ON_DEBUG(left->right_version = + atomic_inc_return(&delim_key_version); + ); + + } else if (ZF_ISSET(left->right, JNODE_HEARD_BANSHEE) + && left->right != right) { + + ON_DEBUG(left->right->left_version = + atomic_inc_return(&delim_key_version); + left->right_version = + atomic_inc_return(&delim_key_version);); + + left->right->left = NULL; + left->right = right; + ZF_SET(left, JNODE_RIGHT_CONNECTED); + } else + /* + * there is a race condition in renew_sibling_link() + * and assertions below check that it is only one + * there. Thread T1 calls renew_sibling_link() without + * GN_NO_ALLOC flag. zlook() doesn't find neighbor + * node, but before T1 gets to the + * link_left_and_right(), another thread T2 creates + * neighbor node and connects it. check for + * left->right == NULL above protects T1 from + * overwriting correct left->right pointer installed + * by T2. + */ + assert("nikita-3302", + right == NULL || left->right == right); + } + if (right != NULL) { + if (right->left == NULL) { + right->left = left; + ZF_SET(right, JNODE_LEFT_CONNECTED); + + ON_DEBUG(right->left_version = + atomic_inc_return(&delim_key_version); + ); + + } else if (ZF_ISSET(right->left, JNODE_HEARD_BANSHEE) + && right->left != left) { + + ON_DEBUG(right->left->right_version = + atomic_inc_return(&delim_key_version); + right->left_version = + atomic_inc_return(&delim_key_version);); + + right->left->right = NULL; + right->left = left; + ZF_SET(right, JNODE_LEFT_CONNECTED); + + } else + assert("nikita-3303", + left == NULL || right->left == left); + } + assert("nikita-3275", check_sibling_list(left)); + assert("nikita-3275", check_sibling_list(right)); +} + +/* Audited by: umka (2002.06.14) */ +static void link_znodes(znode * first, znode * second, int to_left) +{ + if (to_left) + link_left_and_right(second, first); + else + link_left_and_right(first, second); +} + +/* getting of next (to left or to right, depend on gn_to_left bit in flags) + coord's unit position in horizontal direction, even across node + boundary. Should be called under tree lock, it protects nonexistence of + sibling link on parent level, if lock_side_neighbor() fails with + -ENOENT. */ +static int far_next_coord(coord_t * coord, lock_handle * handle, int flags) +{ + int ret; + znode *node; + reiser4_tree *tree; + + assert("umka-243", coord != NULL); + assert("umka-244", handle != NULL); + assert("zam-1069", handle->node == NULL); + + ret = + (flags & GN_GO_LEFT) ? coord_prev_unit(coord) : + coord_next_unit(coord); + if (!ret) + return 0; + + ret = + lock_side_neighbor(handle, coord->node, ZNODE_READ_LOCK, flags, 0); + if (ret) + return ret; + + node = handle->node; + tree = znode_get_tree(node); + write_unlock_tree(tree); + + coord_init_zero(coord); + + /* We avoid synchronous read here if it is specified by flag. */ + if ((flags & GN_ASYNC) && znode_page(handle->node) == NULL) { + ret = jstartio(ZJNODE(handle->node)); + if (!ret) + ret = -E_REPEAT; + goto error_locked; + } + + /* corresponded zrelse() should be called by the clients of + far_next_coord(), in place when this node gets unlocked. */ + ret = zload(handle->node); + if (ret) + goto error_locked; + + if (flags & GN_GO_LEFT) + coord_init_last_unit(coord, node); + else + coord_init_first_unit(coord, node); + + if (0) { + error_locked: + longterm_unlock_znode(handle); + } + write_lock_tree(tree); + return ret; +} + +/* Very significant function which performs a step in horizontal direction + when sibling pointer is not available. Actually, it is only function which + does it. + Note: this function does not restore locking status at exit, + caller should does care about proper unlocking and zrelsing */ +static int +renew_sibling_link(coord_t * coord, lock_handle * handle, znode * child, + tree_level level, int flags, int *nr_locked) +{ + int ret; + int to_left = flags & GN_GO_LEFT; + reiser4_block_nr da; + /* parent of the neighbor node; we set it to parent until not sharing + of one parent between child and neighbor node is detected */ + znode *side_parent = coord->node; + reiser4_tree *tree = znode_get_tree(child); + znode *neighbor = NULL; + + assert("umka-245", coord != NULL); + assert("umka-246", handle != NULL); + assert("umka-247", child != NULL); + assert("umka-303", tree != NULL); + + init_lh(handle); + write_lock_tree(tree); + ret = far_next_coord(coord, handle, flags); + + if (ret) { + if (ret != -ENOENT) { + write_unlock_tree(tree); + return ret; + } + } else { + item_plugin *iplug; + + if (handle->node != NULL) { + (*nr_locked)++; + side_parent = handle->node; + } + + /* does coord object points to internal item? We do not + support sibling pointers between znode for formatted and + unformatted nodes and return -E_NO_NEIGHBOR in that case. */ + iplug = item_plugin_by_coord(coord); + if (!item_is_internal(coord)) { + link_znodes(child, NULL, to_left); + write_unlock_tree(tree); + /* we know there can't be formatted neighbor */ + return RETERR(-E_NO_NEIGHBOR); + } + write_unlock_tree(tree); + + iplug->s.internal.down_link(coord, NULL, &da); + + if (flags & GN_NO_ALLOC) { + neighbor = zlook(tree, &da); + } else { + neighbor = + zget(tree, &da, side_parent, level, + reiser4_ctx_gfp_mask_get()); + } + + if (IS_ERR(neighbor)) { + ret = PTR_ERR(neighbor); + return ret; + } + + if (neighbor) + /* update delimiting keys */ + set_child_delimiting_keys(coord->node, coord, neighbor); + + write_lock_tree(tree); + } + + if (likely(neighbor == NULL || + (znode_get_level(child) == znode_get_level(neighbor) + && child != neighbor))) + link_znodes(child, neighbor, to_left); + else { + warning("nikita-3532", + "Sibling nodes on the different levels: %i != %i\n", + znode_get_level(child), znode_get_level(neighbor)); + ret = RETERR(-EIO); + } + + write_unlock_tree(tree); + + /* if GN_NO_ALLOC isn't set we keep reference to neighbor znode */ + if (neighbor != NULL && (flags & GN_NO_ALLOC)) + /* atomic_dec(&ZJNODE(neighbor)->x_count); */ + zput(neighbor); + + return ret; +} + +/* This function is for establishing of one side relation. */ +/* Audited by: umka (2002.06.14) */ +static int connect_one_side(coord_t * coord, znode * node, int flags) +{ + coord_t local; + lock_handle handle; + int nr_locked; + int ret; + + assert("umka-248", coord != NULL); + assert("umka-249", node != NULL); + + coord_dup_nocheck(&local, coord); + + init_lh(&handle); + + ret = + renew_sibling_link(&local, &handle, node, znode_get_level(node), + flags | GN_NO_ALLOC, &nr_locked); + + if (handle.node != NULL) { + /* complementary operations for zload() and lock() in far_next_coord() */ + zrelse(handle.node); + longterm_unlock_znode(&handle); + } + + /* we catch error codes which are not interesting for us because we + run renew_sibling_link() only for znode connection. */ + if (ret == -ENOENT || ret == -E_NO_NEIGHBOR) + return 0; + + return ret; +} + +/* if @child is not in `connected' state, performs hash searches for left and + right neighbor nodes and establishes horizontal sibling links */ +/* Audited by: umka (2002.06.14), umka (2002.06.15) */ +int connect_znode(coord_t * parent_coord, znode * child) +{ + reiser4_tree *tree = znode_get_tree(child); + int ret = 0; + + assert("zam-330", parent_coord != NULL); + assert("zam-331", child != NULL); + assert("zam-332", parent_coord->node != NULL); + assert("umka-305", tree != NULL); + + /* it is trivial to `connect' root znode because it can't have + neighbors */ + if (znode_above_root(parent_coord->node)) { + child->left = NULL; + child->right = NULL; + ZF_SET(child, JNODE_LEFT_CONNECTED); + ZF_SET(child, JNODE_RIGHT_CONNECTED); + + ON_DEBUG(child->left_version = + atomic_inc_return(&delim_key_version); + child->right_version = + atomic_inc_return(&delim_key_version);); + + return 0; + } + + /* load parent node */ + coord_clear_iplug(parent_coord); + ret = zload(parent_coord->node); + + if (ret != 0) + return ret; + + /* protect `connected' state check by tree_lock */ + read_lock_tree(tree); + + if (!znode_is_right_connected(child)) { + read_unlock_tree(tree); + /* connect right (default is right) */ + ret = connect_one_side(parent_coord, child, GN_NO_ALLOC); + if (ret) + goto zrelse_and_ret; + + read_lock_tree(tree); + } + + ret = znode_is_left_connected(child); + + read_unlock_tree(tree); + + if (!ret) { + ret = + connect_one_side(parent_coord, child, + GN_NO_ALLOC | GN_GO_LEFT); + } else + ret = 0; + + zrelse_and_ret: + zrelse(parent_coord->node); + + return ret; +} + +/* this function is like renew_sibling_link() but allocates neighbor node if + it doesn't exist and `connects' it. It may require making two steps in + horizontal direction, first one for neighbor node finding/allocation, + second one is for finding neighbor of neighbor to connect freshly allocated + znode. */ +/* Audited by: umka (2002.06.14), umka (2002.06.15) */ +static int +renew_neighbor(coord_t * coord, znode * node, tree_level level, int flags) +{ + coord_t local; + lock_handle empty[2]; + reiser4_tree *tree = znode_get_tree(node); + znode *neighbor = NULL; + int nr_locked = 0; + int ret; + + assert("umka-250", coord != NULL); + assert("umka-251", node != NULL); + assert("umka-307", tree != NULL); + assert("umka-308", level <= tree->height); + + /* umka (2002.06.14) + Here probably should be a check for given "level" validness. + Something like assert("xxx-yyy", level < REAL_MAX_ZTREE_HEIGHT); + */ + + coord_dup(&local, coord); + + ret = + renew_sibling_link(&local, &empty[0], node, level, + flags & ~GN_NO_ALLOC, &nr_locked); + if (ret) + goto out; + + /* tree lock is not needed here because we keep parent node(s) locked + and reference to neighbor znode incremented */ + neighbor = (flags & GN_GO_LEFT) ? node->left : node->right; + + read_lock_tree(tree); + ret = znode_is_connected(neighbor); + read_unlock_tree(tree); + if (ret) { + ret = 0; + goto out; + } + + ret = + renew_sibling_link(&local, &empty[nr_locked], neighbor, level, + flags | GN_NO_ALLOC, &nr_locked); + /* second renew_sibling_link() call is used for znode connection only, + so we can live with these errors */ + if (-ENOENT == ret || -E_NO_NEIGHBOR == ret) + ret = 0; + + out: + + for (--nr_locked; nr_locked >= 0; --nr_locked) { + zrelse(empty[nr_locked].node); + longterm_unlock_znode(&empty[nr_locked]); + } + + if (neighbor != NULL) + /* decrement znode reference counter without actually + releasing it. */ + atomic_dec(&ZJNODE(neighbor)->x_count); + + return ret; +} + +/* + reiser4_get_neighbor() -- lock node's neighbor. + + reiser4_get_neighbor() locks node's neighbor (left or right one, depends on + given parameter) using sibling link to it. If sibling link is not available + (i.e. neighbor znode is not in cache) and flags allow read blocks, we go one + level up for information about neighbor's disk address. We lock node's + parent, if it is common parent for both 'node' and its neighbor, neighbor's + disk address is in next (to left or to right) down link from link that points + to original node. If not, we need to lock parent's neighbor, read its content + and take first(last) downlink with neighbor's disk address. That locking + could be done by using sibling link and lock_neighbor() function, if sibling + link exists. In another case we have to go level up again until we find + common parent or valid sibling link. Then go down + allocating/connecting/locking/reading nodes until neighbor of first one is + locked. + + @neighbor: result lock handle, + @node: a node which we lock neighbor of, + @lock_mode: lock mode {LM_READ, LM_WRITE}, + @flags: logical OR of {GN_*} (see description above) subset. + + @return: 0 if success, negative value if lock was impossible due to an error + or lack of neighbor node. +*/ + +/* Audited by: umka (2002.06.14), umka (2002.06.15) */ +int +reiser4_get_neighbor(lock_handle * neighbor, znode * node, + znode_lock_mode lock_mode, int flags) +{ + reiser4_tree *tree = znode_get_tree(node); + lock_handle path[REAL_MAX_ZTREE_HEIGHT]; + + coord_t coord; + + tree_level base_level; + tree_level h = 0; + int ret; + + assert("umka-252", tree != NULL); + assert("umka-253", neighbor != NULL); + assert("umka-254", node != NULL); + + base_level = znode_get_level(node); + + assert("umka-310", base_level <= tree->height); + + coord_init_zero(&coord); + + again: + /* first, we try to use simple lock_neighbor() which requires sibling + link existence */ + read_lock_tree(tree); + ret = lock_side_neighbor(neighbor, node, lock_mode, flags, 1); + read_unlock_tree(tree); + if (!ret) { + /* load znode content if it was specified */ + if (flags & GN_LOAD_NEIGHBOR) { + ret = zload(node); + if (ret) + longterm_unlock_znode(neighbor); + } + return ret; + } + + /* only -ENOENT means we may look upward and try to connect + @node with its neighbor (if @flags allow us to do it) */ + if (ret != -ENOENT || !(flags & GN_CAN_USE_UPPER_LEVELS)) + return ret; + + /* before establishing of sibling link we lock parent node; it is + required by renew_neighbor() to work. */ + init_lh(&path[0]); + ret = reiser4_get_parent(&path[0], node, ZNODE_READ_LOCK); + if (ret) + return ret; + if (znode_above_root(path[0].node)) { + longterm_unlock_znode(&path[0]); + return RETERR(-E_NO_NEIGHBOR); + } + + while (1) { + znode *child = (h == 0) ? node : path[h - 1].node; + znode *parent = path[h].node; + + ret = zload(parent); + if (ret) + break; + + ret = find_child_ptr(parent, child, &coord); + + if (ret) { + zrelse(parent); + break; + } + + /* try to establish missing sibling link */ + ret = renew_neighbor(&coord, child, h + base_level, flags); + + zrelse(parent); + + switch (ret) { + case 0: + /* unlocking of parent znode prevents simple + deadlock situation */ + done_lh(&path[h]); + + /* depend on tree level we stay on we repeat first + locking attempt ... */ + if (h == 0) + goto again; + + /* ... or repeat establishing of sibling link at + one level below. */ + --h; + break; + + case -ENOENT: + /* sibling link is not available -- we go + upward. */ + init_lh(&path[h + 1]); + ret = + reiser4_get_parent(&path[h + 1], parent, + ZNODE_READ_LOCK); + if (ret) + goto fail; + ++h; + if (znode_above_root(path[h].node)) { + ret = RETERR(-E_NO_NEIGHBOR); + goto fail; + } + break; + + case -E_DEADLOCK: + /* there was lock request from hi-pri locker. if + it is possible we unlock last parent node and + re-lock it again. */ + for (; reiser4_check_deadlock(); h--) { + done_lh(&path[h]); + if (h == 0) + goto fail; + } + + break; + + default: /* other errors. */ + goto fail; + } + } + fail: + ON_DEBUG(check_lock_node_data(node)); + ON_DEBUG(check_lock_data()); + + /* unlock path */ + do { + /* FIXME-Zam: when we get here from case -E_DEADLOCK's goto + fail; path[0] is already done_lh-ed, therefore + longterm_unlock_znode(&path[h]); is not applicable */ + done_lh(&path[h]); + --h; + } while (h + 1 != 0); + + return ret; +} + +/* remove node from sibling list */ +/* Audited by: umka (2002.06.14) */ +void sibling_list_remove(znode * node) +{ + reiser4_tree *tree; + + tree = znode_get_tree(node); + assert("umka-255", node != NULL); + assert_rw_write_locked(&(tree->tree_lock)); + assert("nikita-3275", check_sibling_list(node)); + + write_lock_dk(tree); + if (znode_is_right_connected(node) && node->right != NULL && + znode_is_left_connected(node) && node->left != NULL) { + assert("zam-32245", + keyeq(znode_get_rd_key(node), + znode_get_ld_key(node->right))); + znode_set_rd_key(node->left, znode_get_ld_key(node->right)); + } + write_unlock_dk(tree); + + if (znode_is_right_connected(node) && node->right != NULL) { + assert("zam-322", znode_is_left_connected(node->right)); + node->right->left = node->left; + ON_DEBUG(node->right->left_version = + atomic_inc_return(&delim_key_version); + ); + } + if (znode_is_left_connected(node) && node->left != NULL) { + assert("zam-323", znode_is_right_connected(node->left)); + node->left->right = node->right; + ON_DEBUG(node->left->right_version = + atomic_inc_return(&delim_key_version); + ); + } + + ZF_CLR(node, JNODE_LEFT_CONNECTED); + ZF_CLR(node, JNODE_RIGHT_CONNECTED); + ON_DEBUG(node->left = node->right = NULL; + node->left_version = atomic_inc_return(&delim_key_version); + node->right_version = atomic_inc_return(&delim_key_version);); + assert("nikita-3276", check_sibling_list(node)); +} + +/* disconnect node from sibling list */ +void sibling_list_drop(znode * node) +{ + znode *right; + znode *left; + + assert("nikita-2464", node != NULL); + assert("nikita-3277", check_sibling_list(node)); + + right = node->right; + if (right != NULL) { + assert("nikita-2465", znode_is_left_connected(right)); + right->left = NULL; + ON_DEBUG(right->left_version = + atomic_inc_return(&delim_key_version); + ); + } + left = node->left; + if (left != NULL) { + assert("zam-323", znode_is_right_connected(left)); + left->right = NULL; + ON_DEBUG(left->right_version = + atomic_inc_return(&delim_key_version); + ); + } + ZF_CLR(node, JNODE_LEFT_CONNECTED); + ZF_CLR(node, JNODE_RIGHT_CONNECTED); + ON_DEBUG(node->left = node->right = NULL; + node->left_version = atomic_inc_return(&delim_key_version); + node->right_version = atomic_inc_return(&delim_key_version);); +} + +/* Insert new node into sibling list. Regular balancing inserts new node + after (at right side) existing and locked node (@before), except one case + of adding new tree root node. @before should be NULL in that case. */ +void sibling_list_insert_nolock(znode * new, znode * before) +{ + assert("zam-334", new != NULL); + assert("nikita-3298", !znode_is_left_connected(new)); + assert("nikita-3299", !znode_is_right_connected(new)); + assert("nikita-3300", new->left == NULL); + assert("nikita-3301", new->right == NULL); + assert("nikita-3278", check_sibling_list(new)); + assert("nikita-3279", check_sibling_list(before)); + + if (before != NULL) { + assert("zam-333", znode_is_connected(before)); + new->right = before->right; + new->left = before; + ON_DEBUG(new->right_version = + atomic_inc_return(&delim_key_version); + new->left_version = + atomic_inc_return(&delim_key_version);); + if (before->right != NULL) { + before->right->left = new; + ON_DEBUG(before->right->left_version = + atomic_inc_return(&delim_key_version); + ); + } + before->right = new; + ON_DEBUG(before->right_version = + atomic_inc_return(&delim_key_version); + ); + } else { + new->right = NULL; + new->left = NULL; + ON_DEBUG(new->right_version = + atomic_inc_return(&delim_key_version); + new->left_version = + atomic_inc_return(&delim_key_version);); + } + ZF_SET(new, JNODE_LEFT_CONNECTED); + ZF_SET(new, JNODE_RIGHT_CONNECTED); + assert("nikita-3280", check_sibling_list(new)); + assert("nikita-3281", check_sibling_list(before)); +} + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/tree_walk.h linux-2.6.30/fs/reiser4/tree_walk.h --- linux-2.6.30.orig/fs/reiser4/tree_walk.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/tree_walk.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,125 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +/* definitions of reiser4 tree walk functions */ + +#ifndef __FS_REISER4_TREE_WALK_H__ +#define __FS_REISER4_TREE_WALK_H__ + +#include "debug.h" +#include "forward.h" + +/* establishes horizontal links between cached znodes */ +int connect_znode(coord_t * coord, znode * node); + +/* tree traversal functions (reiser4_get_parent(), reiser4_get_neighbor()) + have the following common arguments: + + return codes: + + @return : 0 - OK, + +ZAM-FIXME-HANS: wrong return code name. Change them all. + -ENOENT - neighbor is not in cache, what is detected by sibling + link absence. + + -E_NO_NEIGHBOR - we are sure that neighbor (or parent) node cannot be + found (because we are left-/right- most node of the + tree, for example). Also, this return code is for + reiser4_get_parent() when we see no parent link -- it + means that our node is root node. + + -E_DEADLOCK - deadlock detected (request from high-priority process + received), other error codes are conformed to + /usr/include/asm/errno.h . +*/ + +int +reiser4_get_parent_flags(lock_handle * result, znode * node, + znode_lock_mode mode, int flags); + +/* bits definition for reiser4_get_neighbor function `flags' arg. */ +typedef enum { + /* If sibling pointer is NULL, this flag allows get_neighbor() to try to + * find not allocated not connected neigbor by going though upper + * levels */ + GN_CAN_USE_UPPER_LEVELS = 0x1, + /* locking left neighbor instead of right one */ + GN_GO_LEFT = 0x2, + /* automatically load neighbor node content */ + GN_LOAD_NEIGHBOR = 0x4, + /* return -E_REPEAT if can't lock */ + GN_TRY_LOCK = 0x8, + /* used internally in tree_walk.c, causes renew_sibling to not + allocate neighbor znode, but only search for it in znode cache */ + GN_NO_ALLOC = 0x10, + /* do not go across atom boundaries */ + GN_SAME_ATOM = 0x20, + /* allow to lock not connected nodes */ + GN_ALLOW_NOT_CONNECTED = 0x40, + /* Avoid synchronous jload, instead, call jstartio() and return -E_REPEAT. */ + GN_ASYNC = 0x80 +} znode_get_neigbor_flags; + +/* A commonly used wrapper for reiser4_get_parent_flags(). */ +static inline int reiser4_get_parent(lock_handle * result, znode * node, + znode_lock_mode mode) +{ + return reiser4_get_parent_flags(result, node, mode, + GN_ALLOW_NOT_CONNECTED); +} + +int reiser4_get_neighbor(lock_handle * neighbor, znode * node, + znode_lock_mode lock_mode, int flags); + +/* there are wrappers for most common usages of reiser4_get_neighbor() */ +static inline int +reiser4_get_left_neighbor(lock_handle * result, znode * node, int lock_mode, + int flags) +{ + return reiser4_get_neighbor(result, node, lock_mode, + flags | GN_GO_LEFT); +} + +static inline int +reiser4_get_right_neighbor(lock_handle * result, znode * node, int lock_mode, + int flags) +{ + ON_DEBUG(check_lock_node_data(node)); + ON_DEBUG(check_lock_data()); + return reiser4_get_neighbor(result, node, lock_mode, + flags & (~GN_GO_LEFT)); +} + +extern void sibling_list_remove(znode * node); +extern void sibling_list_drop(znode * node); +extern void sibling_list_insert_nolock(znode * new, znode * before); +extern void link_left_and_right(znode * left, znode * right); + +/* Functions called by tree_walk() when tree_walk() ... */ +struct tree_walk_actor { + /* ... meets a formatted node, */ + int (*process_znode) (tap_t *, void *); + /* ... meets an extent, */ + int (*process_extent) (tap_t *, void *); + /* ... begins tree traversal or repeats it after -E_REPEAT was returned by + * node or extent processing functions. */ + int (*before) (void *); +}; + +#if REISER4_DEBUG +int check_sibling_list(znode * node); +#else +#define check_sibling_list(n) (1) +#endif + +#endif /* __FS_REISER4_TREE_WALK_H__ */ + +/* + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/txnmgr.c linux-2.6.30/fs/reiser4/txnmgr.c --- linux-2.6.30.orig/fs/reiser4/txnmgr.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/txnmgr.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,3164 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Joshua MacDonald wrote the first draft of this code. */ + +/* ZAM-LONGTERM-FIXME-HANS: The locking in this file is badly designed, and a +filesystem scales only as well as its worst locking design. You need to +substantially restructure this code. Josh was not as experienced a programmer +as you. Particularly review how the locking style differs from what you did +for znodes usingt hi-lo priority locking, and present to me an opinion on +whether the differences are well founded. */ + +/* I cannot help but to disagree with the sentiment above. Locking of + * transaction manager is _not_ badly designed, and, at the very least, is not + * the scaling bottleneck. Scaling bottleneck is _exactly_ hi-lo priority + * locking on znodes, especially on the root node of the tree. --nikita, + * 2003.10.13 */ + +/* The txnmgr is a set of interfaces that keep track of atoms and transcrash handles. The + txnmgr processes capture_block requests and manages the relationship between jnodes and + atoms through the various stages of a transcrash, and it also oversees the fusion and + capture-on-copy processes. The main difficulty with this task is maintaining a + deadlock-free lock ordering between atoms and jnodes/handles. The reason for the + difficulty is that jnodes, handles, and atoms contain pointer circles, and the cycle + must be broken. The main requirement is that atom-fusion be deadlock free, so once you + hold the atom_lock you may then wait to acquire any jnode or handle lock. This implies + that any time you check the atom-pointer of a jnode or handle and then try to lock that + atom, you must use trylock() and possibly reverse the order. + + This code implements the design documented at: + + http://namesys.com/txn-doc.html + +ZAM-FIXME-HANS: update v4.html to contain all of the information present in the above (but updated), and then remove the +above document and reference the new. Be sure to provide some credit to Josh. I already have some writings on this +topic in v4.html, but they are lacking in details present in the above. Cure that. Remember to write for the bright 12 +year old --- define all technical terms used. + +*/ + +/* Thoughts on the external transaction interface: + + In the current code, a TRANSCRASH handle is created implicitly by reiser4_init_context() (which + creates state that lasts for the duration of a system call and is called at the start + of ReiserFS methods implementing VFS operations), and closed by reiser4_exit_context(), + occupying the scope of a single system call. We wish to give certain applications an + interface to begin and close (commit) transactions. Since our implementation of + transactions does not yet support isolation, allowing an application to open a + transaction implies trusting it to later close the transaction. Part of the + transaction interface will be aimed at enabling that trust, but the interface for + actually using transactions is fairly narrow. + + BEGIN_TRANSCRASH: Returns a transcrash identifier. It should be possible to translate + this identifier into a string that a shell-script could use, allowing you to start a + transaction by issuing a command. Once open, the transcrash should be set in the task + structure, and there should be options (I suppose) to allow it to be carried across + fork/exec. A transcrash has several options: + + - READ_FUSING or WRITE_FUSING: The default policy is for txn-capture to capture only + on writes (WRITE_FUSING) and allow "dirty reads". If the application wishes to + capture on reads as well, it should set READ_FUSING. + + - TIMEOUT: Since a non-isolated transcrash cannot be undone, every transcrash must + eventually close (or else the machine must crash). If the application dies an + unexpected death with an open transcrash, for example, or if it hangs for a long + duration, one solution (to avoid crashing the machine) is to simply close it anyway. + This is a dangerous option, but it is one way to solve the problem until isolated + transcrashes are available for untrusted applications. + + It seems to be what databases do, though it is unclear how one avoids a DoS attack + creating a vulnerability based on resource starvation. Guaranteeing that some + minimum amount of computational resources are made available would seem more correct + than guaranteeing some amount of time. When we again have someone to code the work, + this issue should be considered carefully. -Hans + + RESERVE_BLOCKS: A running transcrash should indicate to the transaction manager how + many dirty blocks it expects. The reserve_blocks interface should be called at a point + where it is safe for the application to fail, because the system may not be able to + grant the allocation and the application must be able to back-out. For this reason, + the number of reserve-blocks can also be passed as an argument to BEGIN_TRANSCRASH, but + the application may also wish to extend the allocation after beginning its transcrash. + + CLOSE_TRANSCRASH: The application closes the transcrash when it is finished making + modifications that require transaction protection. When isolated transactions are + supported the CLOSE operation is replaced by either COMMIT or ABORT. For example, if a + RESERVE_BLOCKS call fails for the application, it should "abort" by calling + CLOSE_TRANSCRASH, even though it really commits any changes that were made (which is + why, for safety, the application should call RESERVE_BLOCKS before making any changes). + + For actually implementing these out-of-system-call-scopped transcrashes, the + reiser4_context has a "txn_handle *trans" pointer that may be set to an open + transcrash. Currently there are no dynamically-allocated transcrashes, but there is a + "struct kmem_cache *_txnh_slab" created for that purpose in this file. +*/ + +/* Extending the other system call interfaces for future transaction features: + + Specialized applications may benefit from passing flags to the ordinary system call + interface such as read(), write(), or stat(). For example, the application specifies + WRITE_FUSING by default but wishes to add that a certain read() command should be + treated as READ_FUSING. But which read? Is it the directory-entry read, the stat-data + read, or the file-data read? These issues are straight-forward, but there are a lot of + them and adding the necessary flags-passing code will be tedious. + + When supporting isolated transactions, there is a corresponding READ_MODIFY_WRITE (RMW) + flag, which specifies that although it is a read operation being requested, a + write-lock should be taken. The reason is that read-locks are shared while write-locks + are exclusive, so taking a read-lock when a later-write is known in advance will often + leads to deadlock. If a reader knows it will write later, it should issue read + requests with the RMW flag set. +*/ + +/* + The znode/atom deadlock avoidance. + + FIXME(Zam): writing of this comment is in progress. + + The atom's special stage ASTAGE_CAPTURE_WAIT introduces a kind of atom's + long-term locking, which makes reiser4 locking scheme more complex. It had + deadlocks until we implement deadlock avoidance algorithms. That deadlocks + looked as the following: one stopped thread waits for a long-term lock on + znode, the thread who owns that lock waits when fusion with another atom will + be allowed. + + The source of the deadlocks is an optimization of not capturing index nodes + for read. Let's prove it. Suppose we have dumb node capturing scheme which + unconditionally captures each block before locking it. + + That scheme has no deadlocks. Let's begin with the thread which stage is + ASTAGE_CAPTURE_WAIT and it waits for a znode lock. The thread can't wait for + a capture because it's stage allows fusion with any atom except which are + being committed currently. A process of atom commit can't deadlock because + atom commit procedure does not acquire locks and does not fuse with other + atoms. Reiser4 does capturing right before going to sleep inside the + longtertm_lock_znode() function, it means the znode which we want to lock is + already captured and its atom is in ASTAGE_CAPTURE_WAIT stage. If we + continue the analysis we understand that no one process in the sequence may + waits atom fusion. Thereby there are no deadlocks of described kind. + + The capturing optimization makes the deadlocks possible. A thread can wait a + lock which owner did not captured that node. The lock owner's current atom + is not fused with the first atom and it does not get a ASTAGE_CAPTURE_WAIT + state. A deadlock is possible when that atom meets another one which is in + ASTAGE_CAPTURE_WAIT already. + + The deadlock avoidance scheme includes two algorithms: + + First algorithm is used when a thread captures a node which is locked but not + captured by another thread. Those nodes are marked MISSED_IN_CAPTURE at the + moment we skip their capturing. If such a node (marked MISSED_IN_CAPTURE) is + being captured by a thread with current atom is in ASTAGE_CAPTURE_WAIT, the + routine which forces all lock owners to join with current atom is executed. + + Second algorithm does not allow to skip capturing of already captured nodes. + + Both algorithms together prevent waiting a longterm lock without atom fusion + with atoms of all lock owners, which is a key thing for getting atom/znode + locking deadlocks. +*/ + +/* + * Transactions and mmap(2). + * + * 1. Transactions are not supported for accesses through mmap(2), because + * this would effectively amount to user-level transactions whose duration + * is beyond control of the kernel. + * + * 2. That said, we still want to preserve some decency with regard to + * mmap(2). During normal write(2) call, following sequence of events + * happens: + * + * 1. page is created; + * + * 2. jnode is created, dirtied and captured into current atom. + * + * 3. extent is inserted and modified. + * + * Steps (2) and (3) take place under long term lock on the twig node. + * + * When file is accessed through mmap(2) page is always created during + * page fault. + * After this (in reiser4_readpage()->reiser4_readpage_extent()): + * + * 1. if access is made to non-hole page new jnode is created, (if + * necessary) + * + * 2. if access is made to the hole page, jnode is not created (XXX + * not clear why). + * + * Also, even if page is created by write page fault it is not marked + * dirty immediately by handle_mm_fault(). Probably this is to avoid races + * with page write-out. + * + * Dirty bit installed by hardware is only transferred to the struct page + * later, when page is unmapped (in zap_pte_range(), or + * try_to_unmap_one()). + * + * So, with mmap(2) we have to handle following irksome situations: + * + * 1. there exists modified page (clean or dirty) without jnode + * + * 2. there exists modified page (clean or dirty) with clean jnode + * + * 3. clean page which is a part of atom can be transparently modified + * at any moment through mapping without becoming dirty. + * + * (1) and (2) can lead to the out-of-memory situation: ->writepage() + * doesn't know what to do with such pages and ->sync_sb()/->writepages() + * don't see them, because these methods operate on atoms. + * + * (3) can lead to the loss of data: suppose we have dirty page with dirty + * captured jnode captured by some atom. As part of early flush (for + * example) page was written out. Dirty bit was cleared on both page and + * jnode. After this page is modified through mapping, but kernel doesn't + * notice and just discards page and jnode as part of commit. (XXX + * actually it doesn't, because to reclaim page ->releasepage() has to be + * called and before this dirty bit will be transferred to the struct + * page). + * + */ + +#include "debug.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree.h" +#include "wander.h" +#include "ktxnmgrd.h" +#include "super.h" +#include "page_cache.h" +#include "reiser4.h" +#include "vfs_ops.h" +#include "inode.h" +#include "flush.h" + +#include +#include +#include +#include +#include +#include +#include +#include /* for totalram_pages */ + +static void atom_free(txn_atom * atom); + +static int commit_txnh(txn_handle * txnh); + +static void wakeup_atom_waitfor_list(txn_atom * atom); +static void wakeup_atom_waiting_list(txn_atom * atom); + +static void capture_assign_txnh_nolock(txn_atom * atom, txn_handle * txnh); + +static void capture_assign_block_nolock(txn_atom * atom, jnode * node); + +static void fuse_not_fused_lock_owners(txn_handle * txnh, znode * node); + +static int capture_init_fusion(jnode * node, txn_handle * txnh, + txn_capture mode); + +static int capture_fuse_wait(txn_handle *, txn_atom *, txn_atom *, txn_capture); + +static void capture_fuse_into(txn_atom * small, txn_atom * large); + +void reiser4_invalidate_list(struct list_head *); + +/* GENERIC STRUCTURES */ + +typedef struct _txn_wait_links txn_wait_links; + +struct _txn_wait_links { + lock_stack *_lock_stack; + struct list_head _fwaitfor_link; + struct list_head _fwaiting_link; + int (*waitfor_cb) (txn_atom * atom, struct _txn_wait_links * wlinks); + int (*waiting_cb) (txn_atom * atom, struct _txn_wait_links * wlinks); +}; + +/* FIXME: In theory, we should be using the slab cache init & destructor + methods instead of, e.g., jnode_init, etc. */ +static struct kmem_cache *_atom_slab = NULL; +/* this is for user-visible, cross system-call transactions. */ +static struct kmem_cache *_txnh_slab = NULL; + +/** + * init_txnmgr_static - create transaction manager slab caches + * + * Initializes caches of txn-atoms and txn_handle. It is part of reiser4 module + * initialization. + */ +int init_txnmgr_static(void) +{ + assert("jmacd-600", _atom_slab == NULL); + assert("jmacd-601", _txnh_slab == NULL); + + ON_DEBUG(atomic_set(&flush_cnt, 0)); + + _atom_slab = kmem_cache_create("txn_atom", sizeof(txn_atom), 0, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT, NULL); + if (_atom_slab == NULL) + return RETERR(-ENOMEM); + + _txnh_slab = kmem_cache_create("txn_handle", sizeof(txn_handle), 0, + SLAB_HWCACHE_ALIGN, NULL); + if (_txnh_slab == NULL) { + kmem_cache_destroy(_atom_slab); + _atom_slab = NULL; + return RETERR(-ENOMEM); + } + + return 0; +} + +/** + * done_txnmgr_static - delete txn_atom and txn_handle caches + * + * This is called on reiser4 module unloading or system shutdown. + */ +void done_txnmgr_static(void) +{ + destroy_reiser4_cache(&_atom_slab); + destroy_reiser4_cache(&_txnh_slab); +} + +/** + * init_txnmgr - initialize a new transaction manager + * @mgr: pointer to transaction manager embedded in reiser4 super block + * + * This is called on mount. Makes necessary initializations. + */ +void reiser4_init_txnmgr(txn_mgr *mgr) +{ + assert("umka-169", mgr != NULL); + + mgr->atom_count = 0; + mgr->id_count = 1; + INIT_LIST_HEAD(&mgr->atoms_list); + spin_lock_init(&mgr->tmgr_lock); + mutex_init(&mgr->commit_mutex); +} + +/** + * reiser4_done_txnmgr - stop transaction manager + * @mgr: pointer to transaction manager embedded in reiser4 super block + * + * This is called on umount. Does sanity checks. + */ +void reiser4_done_txnmgr(txn_mgr *mgr) +{ + assert("umka-170", mgr != NULL); + assert("umka-1701", list_empty_careful(&mgr->atoms_list)); + assert("umka-1702", mgr->atom_count == 0); +} + +/* Initialize a transaction handle. */ +/* Audited by: umka (2002.06.13) */ +static void txnh_init(txn_handle * txnh, txn_mode mode) +{ + assert("umka-171", txnh != NULL); + + txnh->mode = mode; + txnh->atom = NULL; + reiser4_ctx_gfp_mask_set(); + txnh->flags = 0; + spin_lock_init(&txnh->hlock); + INIT_LIST_HEAD(&txnh->txnh_link); +} + +#if REISER4_DEBUG +/* Check if a transaction handle is clean. */ +static int txnh_isclean(txn_handle * txnh) +{ + assert("umka-172", txnh != NULL); + return txnh->atom == NULL && + LOCK_CNT_NIL(spin_locked_txnh); +} +#endif + +/* Initialize an atom. */ +static void atom_init(txn_atom * atom) +{ + int level; + + assert("umka-173", atom != NULL); + + memset(atom, 0, sizeof(txn_atom)); + + atom->stage = ASTAGE_FREE; + atom->start_time = jiffies; + + for (level = 0; level < REAL_MAX_ZTREE_HEIGHT + 1; level += 1) + INIT_LIST_HEAD(ATOM_DIRTY_LIST(atom, level)); + + INIT_LIST_HEAD(ATOM_CLEAN_LIST(atom)); + INIT_LIST_HEAD(ATOM_OVRWR_LIST(atom)); + INIT_LIST_HEAD(ATOM_WB_LIST(atom)); + INIT_LIST_HEAD(&atom->inodes); + spin_lock_init(&(atom->alock)); + /* list of transaction handles */ + INIT_LIST_HEAD(&atom->txnh_list); + /* link to transaction manager's list of atoms */ + INIT_LIST_HEAD(&atom->atom_link); + INIT_LIST_HEAD(&atom->fwaitfor_list); + INIT_LIST_HEAD(&atom->fwaiting_list); + blocknr_set_init(&atom->delete_set); + blocknr_set_init(&atom->wandered_map); + + init_atom_fq_parts(atom); +} + +#if REISER4_DEBUG +/* Check if an atom is clean. */ +static int atom_isclean(txn_atom * atom) +{ + int level; + + assert("umka-174", atom != NULL); + + for (level = 0; level < REAL_MAX_ZTREE_HEIGHT + 1; level += 1) { + if (!list_empty_careful(ATOM_DIRTY_LIST(atom, level))) { + return 0; + } + } + + return atom->stage == ASTAGE_FREE && + atom->txnh_count == 0 && + atom->capture_count == 0 && + atomic_read(&atom->refcount) == 0 && + (&atom->atom_link == atom->atom_link.next && + &atom->atom_link == atom->atom_link.prev) && + list_empty_careful(&atom->txnh_list) && + list_empty_careful(ATOM_CLEAN_LIST(atom)) && + list_empty_careful(ATOM_OVRWR_LIST(atom)) && + list_empty_careful(ATOM_WB_LIST(atom)) && + list_empty_careful(&atom->fwaitfor_list) && + list_empty_careful(&atom->fwaiting_list) && + atom_fq_parts_are_clean(atom); +} +#endif + +/* Begin a transaction in this context. Currently this uses the reiser4_context's + trans_in_ctx, which means that transaction handles are stack-allocated. Eventually + this will be extended to allow transaction handles to span several contexts. */ +/* Audited by: umka (2002.06.13) */ +void reiser4_txn_begin(reiser4_context * context) +{ + assert("jmacd-544", context->trans == NULL); + + context->trans = &context->trans_in_ctx; + + /* FIXME_LATER_JMACD Currently there's no way to begin a TXN_READ_FUSING + transcrash. Default should be TXN_WRITE_FUSING. Also, the _trans variable is + stack allocated right now, but we would like to allow for dynamically allocated + transcrashes that span multiple system calls. + */ + txnh_init(context->trans, TXN_WRITE_FUSING); +} + +/* Finish a transaction handle context. */ +int reiser4_txn_end(reiser4_context * context) +{ + long ret = 0; + txn_handle *txnh; + + assert("umka-283", context != NULL); + assert("nikita-3012", reiser4_schedulable()); + assert("vs-24", context == get_current_context()); + assert("nikita-2967", lock_stack_isclean(get_current_lock_stack())); + + txnh = context->trans; + if (txnh != NULL) { + if (txnh->atom != NULL) + ret = commit_txnh(txnh); + assert("jmacd-633", txnh_isclean(txnh)); + context->trans = NULL; + } + return ret; +} + +void reiser4_txn_restart(reiser4_context * context) +{ + reiser4_txn_end(context); + reiser4_preempt_point(); + reiser4_txn_begin(context); +} + +void reiser4_txn_restart_current(void) +{ + reiser4_txn_restart(get_current_context()); +} + +/* TXN_ATOM */ + +/* Get the atom belonging to a txnh, which is not locked. Return txnh locked. Locks atom, if atom + is not NULL. This performs the necessary spin_trylock to break the lock-ordering cycle. May + return NULL. */ +static txn_atom *txnh_get_atom(txn_handle * txnh) +{ + txn_atom *atom; + + assert("umka-180", txnh != NULL); + assert_spin_not_locked(&(txnh->hlock)); + + while (1) { + spin_lock_txnh(txnh); + atom = txnh->atom; + + if (atom == NULL) + break; + + if (spin_trylock_atom(atom)) + break; + + atomic_inc(&atom->refcount); + + spin_unlock_txnh(txnh); + spin_lock_atom(atom); + spin_lock_txnh(txnh); + + if (txnh->atom == atom) { + atomic_dec(&atom->refcount); + break; + } + + spin_unlock_txnh(txnh); + atom_dec_and_unlock(atom); + } + + return atom; +} + +/* Get the current atom and spinlock it if current atom present. May return NULL */ +txn_atom *get_current_atom_locked_nocheck(void) +{ + reiser4_context *cx; + txn_atom *atom; + txn_handle *txnh; + + cx = get_current_context(); + assert("zam-437", cx != NULL); + + txnh = cx->trans; + assert("zam-435", txnh != NULL); + + atom = txnh_get_atom(txnh); + + spin_unlock_txnh(txnh); + return atom; +} + +/* Get the atom belonging to a jnode, which is initially locked. Return with + both jnode and atom locked. This performs the necessary spin_trylock to + break the lock-ordering cycle. Assumes the jnode is already locked, and + returns NULL if atom is not set. */ +txn_atom *jnode_get_atom(jnode * node) +{ + txn_atom *atom; + + assert("umka-181", node != NULL); + + while (1) { + assert_spin_locked(&(node->guard)); + + atom = node->atom; + /* node is not in any atom */ + if (atom == NULL) + break; + + /* If atom is not locked, grab the lock and return */ + if (spin_trylock_atom(atom)) + break; + + /* At least one jnode belongs to this atom it guarantees that + * atom->refcount > 0, we can safely increment refcount. */ + atomic_inc(&atom->refcount); + spin_unlock_jnode(node); + + /* re-acquire spin locks in the right order */ + spin_lock_atom(atom); + spin_lock_jnode(node); + + /* check if node still points to the same atom. */ + if (node->atom == atom) { + atomic_dec(&atom->refcount); + break; + } + + /* releasing of atom lock and reference requires not holding + * locks on jnodes. */ + spin_unlock_jnode(node); + + /* We do not sure that this atom has extra references except our + * one, so we should call proper function which may free atom if + * last reference is released. */ + atom_dec_and_unlock(atom); + + /* lock jnode again for getting valid node->atom pointer + * value. */ + spin_lock_jnode(node); + } + + return atom; +} + +/* Returns true if @node is dirty and part of the same atom as one of its neighbors. Used + by flush code to indicate whether the next node (in some direction) is suitable for + flushing. */ +int +same_slum_check(jnode * node, jnode * check, int alloc_check, int alloc_value) +{ + int compat; + txn_atom *atom; + + assert("umka-182", node != NULL); + assert("umka-183", check != NULL); + + /* Not sure what this function is supposed to do if supplied with @check that is + neither formatted nor unformatted (bitmap or so). */ + assert("nikita-2373", jnode_is_znode(check) + || jnode_is_unformatted(check)); + + /* Need a lock on CHECK to get its atom and to check various state bits. + Don't need a lock on NODE once we get the atom lock. */ + /* It is not enough to lock two nodes and check (node->atom == + check->atom) because atom could be locked and being fused at that + moment, jnodes of the atom of that state (being fused) can point to + different objects, but the atom is the same. */ + spin_lock_jnode(check); + + atom = jnode_get_atom(check); + + if (atom == NULL) { + compat = 0; + } else { + compat = (node->atom == atom && JF_ISSET(check, JNODE_DIRTY)); + + if (compat && jnode_is_znode(check)) { + compat &= znode_is_connected(JZNODE(check)); + } + + if (compat && alloc_check) { + compat &= (alloc_value == jnode_is_flushprepped(check)); + } + + spin_unlock_atom(atom); + } + + spin_unlock_jnode(check); + + return compat; +} + +/* Decrement the atom's reference count and if it falls to zero, free it. */ +void atom_dec_and_unlock(txn_atom * atom) +{ + txn_mgr *mgr = &get_super_private(reiser4_get_current_sb())->tmgr; + + assert("umka-186", atom != NULL); + assert_spin_locked(&(atom->alock)); + assert("zam-1039", atomic_read(&atom->refcount) > 0); + + if (atomic_dec_and_test(&atom->refcount)) { + /* take txnmgr lock and atom lock in proper order. */ + if (!spin_trylock_txnmgr(mgr)) { + /* This atom should exist after we re-acquire its + * spinlock, so we increment its reference counter. */ + atomic_inc(&atom->refcount); + spin_unlock_atom(atom); + spin_lock_txnmgr(mgr); + spin_lock_atom(atom); + + if (!atomic_dec_and_test(&atom->refcount)) { + spin_unlock_atom(atom); + spin_unlock_txnmgr(mgr); + return; + } + } + assert_spin_locked(&(mgr->tmgr_lock)); + atom_free(atom); + spin_unlock_txnmgr(mgr); + } else + spin_unlock_atom(atom); +} + +/* Create new atom and connect it to given transaction handle. This adds the + atom to the transaction manager's list and sets its reference count to 1, an + artificial reference which is kept until it commits. We play strange games + to avoid allocation under jnode & txnh spinlocks.*/ + +static int atom_begin_and_assign_to_txnh(txn_atom ** atom_alloc, txn_handle * txnh) +{ + txn_atom *atom; + txn_mgr *mgr; + + if (REISER4_DEBUG && rofs_tree(current_tree)) { + warning("nikita-3366", "Creating atom on rofs"); + dump_stack(); + } + + if (*atom_alloc == NULL) { + (*atom_alloc) = kmem_cache_alloc(_atom_slab, + reiser4_ctx_gfp_mask_get()); + + if (*atom_alloc == NULL) + return RETERR(-ENOMEM); + } + + /* and, also, txnmgr spin lock should be taken before jnode and txnh + locks. */ + mgr = &get_super_private(reiser4_get_current_sb())->tmgr; + spin_lock_txnmgr(mgr); + spin_lock_txnh(txnh); + + /* Check whether new atom still needed */ + if (txnh->atom != NULL) { + /* NOTE-NIKITA probably it is rather better to free + * atom_alloc here than thread it up to reiser4_try_capture() */ + + spin_unlock_txnh(txnh); + spin_unlock_txnmgr(mgr); + + return -E_REPEAT; + } + + atom = *atom_alloc; + *atom_alloc = NULL; + + atom_init(atom); + + assert("jmacd-17", atom_isclean(atom)); + + /* + * lock ordering is broken here. It is ok, as long as @atom is new + * and inaccessible for others. We can't use spin_lock_atom or + * spin_lock(&atom->alock) because they care about locking + * dependencies. spin_trylock_lock doesn't. + */ + check_me("", spin_trylock_atom(atom)); + + /* add atom to the end of transaction manager's list of atoms */ + list_add_tail(&atom->atom_link, &mgr->atoms_list); + atom->atom_id = mgr->id_count++; + mgr->atom_count += 1; + + /* Release txnmgr lock */ + spin_unlock_txnmgr(mgr); + + /* One reference until it commits. */ + atomic_inc(&atom->refcount); + atom->stage = ASTAGE_CAPTURE_FUSE; + atom->super = reiser4_get_current_sb(); + capture_assign_txnh_nolock(atom, txnh); + + spin_unlock_atom(atom); + spin_unlock_txnh(txnh); + + return -E_REPEAT; +} + +/* Return true if an atom is currently "open". */ +static int atom_isopen(const txn_atom * atom) +{ + assert("umka-185", atom != NULL); + + return atom->stage > 0 && atom->stage < ASTAGE_PRE_COMMIT; +} + +/* Return the number of pointers to this atom that must be updated during fusion. This + approximates the amount of work to be done. Fusion chooses the atom with fewer + pointers to fuse into the atom with more pointers. */ +static int atom_pointer_count(const txn_atom * atom) +{ + assert("umka-187", atom != NULL); + + /* This is a measure of the amount of work needed to fuse this atom + * into another. */ + return atom->txnh_count + atom->capture_count; +} + +/* Called holding the atom lock, this removes the atom from the transaction manager list + and frees it. */ +static void atom_free(txn_atom * atom) +{ + txn_mgr *mgr = &get_super_private(reiser4_get_current_sb())->tmgr; + + assert("umka-188", atom != NULL); + assert_spin_locked(&(atom->alock)); + + /* Remove from the txn_mgr's atom list */ + assert_spin_locked(&(mgr->tmgr_lock)); + mgr->atom_count -= 1; + list_del_init(&atom->atom_link); + + /* Clean the atom */ + assert("jmacd-16", + (atom->stage == ASTAGE_INVALID || atom->stage == ASTAGE_DONE)); + atom->stage = ASTAGE_FREE; + + blocknr_set_destroy(&atom->delete_set); + blocknr_set_destroy(&atom->wandered_map); + + assert("jmacd-16", atom_isclean(atom)); + + spin_unlock_atom(atom); + + kmem_cache_free(_atom_slab, atom); +} + +static int atom_is_dotard(const txn_atom * atom) +{ + return time_after(jiffies, atom->start_time + + get_current_super_private()->tmgr.atom_max_age); +} + +static int atom_can_be_committed(txn_atom * atom) +{ + assert_spin_locked(&(atom->alock)); + assert("zam-885", atom->txnh_count > atom->nr_waiters); + return atom->txnh_count == atom->nr_waiters + 1; +} + +/* Return true if an atom should commit now. This is determined by aging, atom + size or atom flags. */ +static int atom_should_commit(const txn_atom * atom) +{ + assert("umka-189", atom != NULL); + return + (atom->flags & ATOM_FORCE_COMMIT) || + ((unsigned)atom_pointer_count(atom) > + get_current_super_private()->tmgr.atom_max_size) + || atom_is_dotard(atom); +} + +/* return 1 if current atom exists and requires commit. */ +int current_atom_should_commit(void) +{ + txn_atom *atom; + int result = 0; + + atom = get_current_atom_locked_nocheck(); + if (atom) { + result = atom_should_commit(atom); + spin_unlock_atom(atom); + } + return result; +} + +static int atom_should_commit_asap(const txn_atom * atom) +{ + unsigned int captured; + unsigned int pinnedpages; + + assert("nikita-3309", atom != NULL); + + captured = (unsigned)atom->capture_count; + pinnedpages = (captured >> PAGE_CACHE_SHIFT) * sizeof(znode); + + return (pinnedpages > (totalram_pages >> 3)) || (atom->flushed > 100); +} + +static jnode *find_first_dirty_in_list(struct list_head *head, int flags) +{ + jnode *first_dirty; + + list_for_each_entry(first_dirty, head, capture_link) { + if (!(flags & JNODE_FLUSH_COMMIT)) { + /* + * skip jnodes which "heard banshee" or having active + * I/O + */ + if (JF_ISSET(first_dirty, JNODE_HEARD_BANSHEE) || + JF_ISSET(first_dirty, JNODE_WRITEBACK)) + continue; + } + return first_dirty; + } + return NULL; +} + +/* Get first dirty node from the atom's dirty_nodes[n] lists; return NULL if atom has no dirty + nodes on atom's lists */ +jnode *find_first_dirty_jnode(txn_atom * atom, int flags) +{ + jnode *first_dirty; + tree_level level; + + assert_spin_locked(&(atom->alock)); + + /* The flush starts from LEAF_LEVEL (=1). */ + for (level = 1; level < REAL_MAX_ZTREE_HEIGHT + 1; level += 1) { + if (list_empty_careful(ATOM_DIRTY_LIST(atom, level))) + continue; + + first_dirty = + find_first_dirty_in_list(ATOM_DIRTY_LIST(atom, level), + flags); + if (first_dirty) + return first_dirty; + } + + /* znode-above-root is on the list #0. */ + return find_first_dirty_in_list(ATOM_DIRTY_LIST(atom, 0), flags); +} + +static void dispatch_wb_list(txn_atom * atom, flush_queue_t * fq) +{ + jnode *cur; + + assert("zam-905", atom_is_protected(atom)); + + cur = list_entry(ATOM_WB_LIST(atom)->next, jnode, capture_link); + while (ATOM_WB_LIST(atom) != &cur->capture_link) { + jnode *next = list_entry(cur->capture_link.next, jnode, capture_link); + + spin_lock_jnode(cur); + if (!JF_ISSET(cur, JNODE_WRITEBACK)) { + if (JF_ISSET(cur, JNODE_DIRTY)) { + queue_jnode(fq, cur); + } else { + /* move jnode to atom's clean list */ + list_move_tail(&cur->capture_link, + ATOM_CLEAN_LIST(atom)); + } + } + spin_unlock_jnode(cur); + + cur = next; + } +} + +/* Scan current atom->writeback_nodes list, re-submit dirty and !writeback + * jnodes to disk. */ +static int submit_wb_list(void) +{ + int ret; + flush_queue_t *fq; + + fq = get_fq_for_current_atom(); + if (IS_ERR(fq)) + return PTR_ERR(fq); + + dispatch_wb_list(fq->atom, fq); + spin_unlock_atom(fq->atom); + + ret = reiser4_write_fq(fq, NULL, 1); + reiser4_fq_put(fq); + + return ret; +} + +/* Wait completion of all writes, re-submit atom writeback list if needed. */ +static int current_atom_complete_writes(void) +{ + int ret; + + /* Each jnode from that list was modified and dirtied when it had i/o + * request running already. After i/o completion we have to resubmit + * them to disk again.*/ + ret = submit_wb_list(); + if (ret < 0) + return ret; + + /* Wait all i/o completion */ + ret = current_atom_finish_all_fq(); + if (ret) + return ret; + + /* Scan wb list again; all i/o should be completed, we re-submit dirty + * nodes to disk */ + ret = submit_wb_list(); + if (ret < 0) + return ret; + + /* Wait all nodes we just submitted */ + return current_atom_finish_all_fq(); +} + +#if REISER4_DEBUG + +static void reiser4_info_atom(const char *prefix, const txn_atom * atom) +{ + if (atom == NULL) { + printk("%s: no atom\n", prefix); + return; + } + + printk("%s: refcount: %i id: %i flags: %x txnh_count: %i" + " capture_count: %i stage: %x start: %lu, flushed: %i\n", prefix, + atomic_read(&atom->refcount), atom->atom_id, atom->flags, + atom->txnh_count, atom->capture_count, atom->stage, + atom->start_time, atom->flushed); +} + +#else /* REISER4_DEBUG */ + +static inline void reiser4_info_atom(const char *prefix, const txn_atom * atom) {} + +#endif /* REISER4_DEBUG */ + +#define TOOMANYFLUSHES (1 << 13) + +/* Called with the atom locked and no open "active" transaction handlers except + ours, this function calls flush_current_atom() until all dirty nodes are + processed. Then it initiates commit processing. + + Called by the single remaining open "active" txnh, which is closing. Other + open txnhs belong to processes which wait atom commit in commit_txnh() + routine. They are counted as "waiters" in atom->nr_waiters. Therefore as + long as we hold the atom lock none of the jnodes can be captured and/or + locked. + + Return value is an error code if commit fails. +*/ +static int commit_current_atom(long *nr_submitted, txn_atom ** atom) +{ + reiser4_super_info_data *sbinfo = get_current_super_private(); + long ret = 0; + /* how many times jnode_flush() was called as a part of attempt to + * commit this atom. */ + int flushiters; + + assert("zam-888", atom != NULL && *atom != NULL); + assert_spin_locked(&((*atom)->alock)); + assert("zam-887", get_current_context()->trans->atom == *atom); + assert("jmacd-151", atom_isopen(*atom)); + + assert("nikita-3184", + get_current_super_private()->delete_mutex_owner != current); + + for (flushiters = 0;; ++flushiters) { + ret = + flush_current_atom(JNODE_FLUSH_WRITE_BLOCKS | + JNODE_FLUSH_COMMIT, + LONG_MAX /* nr_to_write */ , + nr_submitted, atom, NULL); + if (ret != -E_REPEAT) + break; + + /* if atom's dirty list contains one znode which is + HEARD_BANSHEE and is locked we have to allow lock owner to + continue and uncapture that znode */ + reiser4_preempt_point(); + + *atom = get_current_atom_locked(); + if (flushiters > TOOMANYFLUSHES && IS_POW(flushiters)) { + warning("nikita-3176", + "Flushing like mad: %i", flushiters); + reiser4_info_atom("atom", *atom); + DEBUGON(flushiters > (1 << 20)); + } + } + + if (ret) + return ret; + + assert_spin_locked(&((*atom)->alock)); + + if (!atom_can_be_committed(*atom)) { + spin_unlock_atom(*atom); + return RETERR(-E_REPEAT); + } + + if ((*atom)->capture_count == 0) + goto done; + + /* Up to this point we have been flushing and after flush is called we + return -E_REPEAT. Now we can commit. We cannot return -E_REPEAT + at this point, commit should be successful. */ + reiser4_atom_set_stage(*atom, ASTAGE_PRE_COMMIT); + ON_DEBUG(((*atom)->committer = current)); + spin_unlock_atom(*atom); + + ret = current_atom_complete_writes(); + if (ret) + return ret; + + assert("zam-906", list_empty(ATOM_WB_LIST(*atom))); + + /* isolate critical code path which should be executed by only one + * thread using tmgr mutex */ + mutex_lock(&sbinfo->tmgr.commit_mutex); + + ret = reiser4_write_logs(nr_submitted); + if (ret < 0) + reiser4_panic("zam-597", "write log failed (%ld)\n", ret); + + /* The atom->ovrwr_nodes list is processed under commit mutex held + because of bitmap nodes which are captured by special way in + reiser4_pre_commit_hook_bitmap(), that way does not include + capture_fuse_wait() as a capturing of other nodes does -- the commit + mutex is used for transaction isolation instead. */ + reiser4_invalidate_list(ATOM_OVRWR_LIST(*atom)); + mutex_unlock(&sbinfo->tmgr.commit_mutex); + + reiser4_invalidate_list(ATOM_CLEAN_LIST(*atom)); + reiser4_invalidate_list(ATOM_WB_LIST(*atom)); + assert("zam-927", list_empty(&(*atom)->inodes)); + + spin_lock_atom(*atom); + done: + reiser4_atom_set_stage(*atom, ASTAGE_DONE); + ON_DEBUG((*atom)->committer = NULL); + + /* Atom's state changes, so wake up everybody waiting for this + event. */ + wakeup_atom_waiting_list(*atom); + + /* Decrement the "until commit" reference, at least one txnh (the caller) is + still open. */ + atomic_dec(&(*atom)->refcount); + + assert("jmacd-1070", atomic_read(&(*atom)->refcount) > 0); + assert("jmacd-1062", (*atom)->capture_count == 0); + BUG_ON((*atom)->capture_count != 0); + assert_spin_locked(&((*atom)->alock)); + + return ret; +} + +/* TXN_TXNH */ + +/** + * force_commit_atom - commit current atom and wait commit completion + * @txnh: + * + * Commits current atom and wait commit completion; current atom and @txnh have + * to be spinlocked before call, this function unlocks them on exit. + */ +int force_commit_atom(txn_handle *txnh) +{ + txn_atom *atom; + + assert("zam-837", txnh != NULL); + assert_spin_locked(&(txnh->hlock)); + assert("nikita-2966", lock_stack_isclean(get_current_lock_stack())); + + atom = txnh->atom; + + assert("zam-834", atom != NULL); + assert_spin_locked(&(atom->alock)); + + /* + * Set flags for atom and txnh: forcing atom commit and waiting for + * commit completion + */ + txnh->flags |= TXNH_WAIT_COMMIT; + atom->flags |= ATOM_FORCE_COMMIT; + + spin_unlock_txnh(txnh); + spin_unlock_atom(atom); + + /* commit is here */ + reiser4_txn_restart_current(); + return 0; +} + +/* Called to force commit of any outstanding atoms. @commit_all_atoms controls + * should we commit all atoms including new ones which are created after this + * functions is called. */ +int txnmgr_force_commit_all(struct super_block *super, int commit_all_atoms) +{ + int ret; + txn_atom *atom; + txn_mgr *mgr; + txn_handle *txnh; + unsigned long start_time = jiffies; + reiser4_context *ctx = get_current_context(); + + assert("nikita-2965", lock_stack_isclean(get_current_lock_stack())); + assert("nikita-3058", reiser4_commit_check_locks()); + + reiser4_txn_restart_current(); + + mgr = &get_super_private(super)->tmgr; + + txnh = ctx->trans; + + again: + + spin_lock_txnmgr(mgr); + + list_for_each_entry(atom, &mgr->atoms_list, atom_link) { + spin_lock_atom(atom); + + /* Commit any atom which can be committed. If @commit_new_atoms + * is not set we commit only atoms which were created before + * this call is started. */ + if (commit_all_atoms + || time_before_eq(atom->start_time, start_time)) { + if (atom->stage <= ASTAGE_POST_COMMIT) { + spin_unlock_txnmgr(mgr); + + if (atom->stage < ASTAGE_PRE_COMMIT) { + spin_lock_txnh(txnh); + /* Add force-context txnh */ + capture_assign_txnh_nolock(atom, txnh); + ret = force_commit_atom(txnh); + if (ret) + return ret; + } else + /* wait atom commit */ + reiser4_atom_wait_event(atom); + + goto again; + } + } + + spin_unlock_atom(atom); + } + +#if REISER4_DEBUG + if (commit_all_atoms) { + reiser4_super_info_data *sbinfo = get_super_private(super); + spin_lock_reiser4_super(sbinfo); + assert("zam-813", + sbinfo->blocks_fake_allocated_unformatted == 0); + assert("zam-812", sbinfo->blocks_fake_allocated == 0); + spin_unlock_reiser4_super(sbinfo); + } +#endif + + spin_unlock_txnmgr(mgr); + + return 0; +} + +/* check whether commit_some_atoms() can commit @atom. Locking is up to the + * caller */ +static int atom_is_committable(txn_atom * atom) +{ + return + atom->stage < ASTAGE_PRE_COMMIT && + atom->txnh_count == atom->nr_waiters && atom_should_commit(atom); +} + +/* called periodically from ktxnmgrd to commit old atoms. Releases ktxnmgrd spin + * lock at exit */ +int commit_some_atoms(txn_mgr * mgr) +{ + int ret = 0; + txn_atom *atom; + txn_handle *txnh; + reiser4_context *ctx; + struct list_head *pos, *tmp; + + ctx = get_current_context(); + assert("nikita-2444", ctx != NULL); + + txnh = ctx->trans; + spin_lock_txnmgr(mgr); + + /* + * this is to avoid gcc complain that atom might be used + * uninitialized + */ + atom = NULL; + + /* look for atom to commit */ + list_for_each_safe(pos, tmp, &mgr->atoms_list) { + atom = list_entry(pos, txn_atom, atom_link); + /* + * first test without taking atom spin lock, whether it is + * eligible for committing at all + */ + if (atom_is_committable(atom)) { + /* now, take spin lock and re-check */ + spin_lock_atom(atom); + if (atom_is_committable(atom)) + break; + spin_unlock_atom(atom); + } + } + + ret = (&mgr->atoms_list == pos); + spin_unlock_txnmgr(mgr); + + if (ret) { + /* nothing found */ + spin_unlock(&mgr->daemon->guard); + return 0; + } + + spin_lock_txnh(txnh); + + BUG_ON(atom == NULL); + /* Set the atom to force committing */ + atom->flags |= ATOM_FORCE_COMMIT; + + /* Add force-context txnh */ + capture_assign_txnh_nolock(atom, txnh); + + spin_unlock_txnh(txnh); + spin_unlock_atom(atom); + + /* we are about to release daemon spin lock, notify daemon it + has to rescan atoms */ + mgr->daemon->rescan = 1; + spin_unlock(&mgr->daemon->guard); + reiser4_txn_restart_current(); + return 0; +} + +static int txn_try_to_fuse_small_atom(txn_mgr * tmgr, txn_atom * atom) +{ + int atom_stage; + txn_atom *atom_2; + int repeat; + + assert("zam-1051", atom->stage < ASTAGE_PRE_COMMIT); + + atom_stage = atom->stage; + repeat = 0; + + if (!spin_trylock_txnmgr(tmgr)) { + atomic_inc(&atom->refcount); + spin_unlock_atom(atom); + spin_lock_txnmgr(tmgr); + spin_lock_atom(atom); + repeat = 1; + if (atom->stage != atom_stage) { + spin_unlock_txnmgr(tmgr); + atom_dec_and_unlock(atom); + return -E_REPEAT; + } + atomic_dec(&atom->refcount); + } + + list_for_each_entry(atom_2, &tmgr->atoms_list, atom_link) { + if (atom == atom_2) + continue; + /* + * if trylock does not succeed we just do not fuse with that + * atom. + */ + if (spin_trylock_atom(atom_2)) { + if (atom_2->stage < ASTAGE_PRE_COMMIT) { + spin_unlock_txnmgr(tmgr); + capture_fuse_into(atom_2, atom); + /* all locks are lost we can only repeat here */ + return -E_REPEAT; + } + spin_unlock_atom(atom_2); + } + } + atom->flags |= ATOM_CANCEL_FUSION; + spin_unlock_txnmgr(tmgr); + if (repeat) { + spin_unlock_atom(atom); + return -E_REPEAT; + } + return 0; +} + +/* Calls jnode_flush for current atom if it exists; if not, just take another + atom and call jnode_flush() for him. If current transaction handle has + already assigned atom (current atom) we have to close current transaction + prior to switch to another atom or do something with current atom. This + code tries to flush current atom. + + flush_some_atom() is called as part of memory clearing process. It is + invoked from balance_dirty_pages(), pdflushd, and entd. + + If we can flush no nodes, atom is committed, because this frees memory. + + If atom is too large or too old it is committed also. +*/ +int +flush_some_atom(jnode * start, long *nr_submitted, const struct writeback_control *wbc, + int flags) +{ + reiser4_context *ctx = get_current_context(); + txn_mgr *tmgr = &get_super_private(ctx->super)->tmgr; + txn_handle *txnh = ctx->trans; + txn_atom *atom; + int ret; + + BUG_ON(wbc->nr_to_write == 0); + BUG_ON(*nr_submitted != 0); + assert("zam-1042", txnh != NULL); + repeat: + if (txnh->atom == NULL) { + /* current atom is not available, take first from txnmgr */ + spin_lock_txnmgr(tmgr); + + /* traverse the list of all atoms */ + list_for_each_entry(atom, &tmgr->atoms_list, atom_link) { + /* lock atom before checking its state */ + spin_lock_atom(atom); + + /* + * we need an atom which is not being committed and + * which has no flushers (jnode_flush() add one flusher + * at the beginning and subtract one at the end). + */ + if (atom->stage < ASTAGE_PRE_COMMIT && + atom->nr_flushers == 0) { + spin_lock_txnh(txnh); + capture_assign_txnh_nolock(atom, txnh); + spin_unlock_txnh(txnh); + + goto found; + } + + spin_unlock_atom(atom); + } + + /* + * Write throttling is case of no one atom can be + * flushed/committed. + */ + if (!current_is_pdflush() && !wbc->nonblocking) { + list_for_each_entry(atom, &tmgr->atoms_list, atom_link) { + spin_lock_atom(atom); + /* Repeat the check from the above. */ + if (atom->stage < ASTAGE_PRE_COMMIT + && atom->nr_flushers == 0) { + spin_lock_txnh(txnh); + capture_assign_txnh_nolock(atom, txnh); + spin_unlock_txnh(txnh); + + goto found; + } + if (atom->stage <= ASTAGE_POST_COMMIT) { + spin_unlock_txnmgr(tmgr); + /* + * we just wait until atom's flusher + * makes a progress in flushing or + * committing the atom + */ + reiser4_atom_wait_event(atom); + goto repeat; + } + spin_unlock_atom(atom); + } + } + spin_unlock_txnmgr(tmgr); + return 0; + found: + spin_unlock_txnmgr(tmgr); + } else + atom = get_current_atom_locked(); + + BUG_ON(atom->super != ctx->super); + assert("vs-35", atom->super == ctx->super); + if (start) { + spin_lock_jnode(start); + ret = (atom == start->atom) ? 1 : 0; + spin_unlock_jnode(start); + if (ret == 0) + start = NULL; + } + ret = flush_current_atom(flags, wbc->nr_to_write, nr_submitted, &atom, start); + if (ret == 0) { + /* flush_current_atom returns 0 only if it submitted for write + nothing */ + BUG_ON(*nr_submitted != 0); + if (*nr_submitted == 0 || atom_should_commit_asap(atom)) { + if (atom->capture_count < tmgr->atom_min_size && + !(atom->flags & ATOM_CANCEL_FUSION)) { + ret = txn_try_to_fuse_small_atom(tmgr, atom); + if (ret == -E_REPEAT) { + reiser4_preempt_point(); + goto repeat; + } + } + /* if early flushing could not make more nodes clean, + * or atom is too old/large, + * we force current atom to commit */ + /* wait for commit completion but only if this + * wouldn't stall pdflushd and ent thread. */ + if (!wbc->nonblocking && !ctx->entd) + txnh->flags |= TXNH_WAIT_COMMIT; + atom->flags |= ATOM_FORCE_COMMIT; + } + spin_unlock_atom(atom); + } else if (ret == -E_REPEAT) { + if (*nr_submitted == 0) { + /* let others who hampers flushing (hold longterm locks, + for instance) to free the way for flush */ + reiser4_preempt_point(); + goto repeat; + } + ret = 0; + } +/* + if (*nr_submitted > wbc->nr_to_write) + warning("", "asked for %ld, written %ld\n", wbc->nr_to_write, *nr_submitted); +*/ + reiser4_txn_restart(ctx); + + return ret; +} + +/* Remove processed nodes from atom's clean list (thereby remove them from transaction). */ +void reiser4_invalidate_list(struct list_head *head) +{ + while (!list_empty(head)) { + jnode *node; + + node = list_entry(head->next, jnode, capture_link); + spin_lock_jnode(node); + reiser4_uncapture_block(node); + jput(node); + } +} + +static void init_wlinks(txn_wait_links * wlinks) +{ + wlinks->_lock_stack = get_current_lock_stack(); + INIT_LIST_HEAD(&wlinks->_fwaitfor_link); + INIT_LIST_HEAD(&wlinks->_fwaiting_link); + wlinks->waitfor_cb = NULL; + wlinks->waiting_cb = NULL; +} + +/* Add atom to the atom's waitfor list and wait for somebody to wake us up; */ +void reiser4_atom_wait_event(txn_atom * atom) +{ + txn_wait_links _wlinks; + + assert_spin_locked(&(atom->alock)); + assert("nikita-3156", + lock_stack_isclean(get_current_lock_stack()) || + atom->nr_running_queues > 0); + + init_wlinks(&_wlinks); + list_add_tail(&_wlinks._fwaitfor_link, &atom->fwaitfor_list); + atomic_inc(&atom->refcount); + spin_unlock_atom(atom); + + reiser4_prepare_to_sleep(_wlinks._lock_stack); + reiser4_go_to_sleep(_wlinks._lock_stack); + + spin_lock_atom(atom); + list_del(&_wlinks._fwaitfor_link); + atom_dec_and_unlock(atom); +} + +void reiser4_atom_set_stage(txn_atom * atom, txn_stage stage) +{ + assert("nikita-3535", atom != NULL); + assert_spin_locked(&(atom->alock)); + assert("nikita-3536", stage <= ASTAGE_INVALID); + /* Excelsior! */ + assert("nikita-3537", stage >= atom->stage); + if (atom->stage != stage) { + atom->stage = stage; + reiser4_atom_send_event(atom); + } +} + +/* wake all threads which wait for an event */ +void reiser4_atom_send_event(txn_atom * atom) +{ + assert_spin_locked(&(atom->alock)); + wakeup_atom_waitfor_list(atom); +} + +/* Informs txn manager code that owner of this txn_handle should wait atom commit completion (for + example, because it does fsync(2)) */ +static int should_wait_commit(txn_handle * h) +{ + return h->flags & TXNH_WAIT_COMMIT; +} + +typedef struct commit_data { + txn_atom *atom; + txn_handle *txnh; + long nr_written; + /* as an optimization we start committing atom by first trying to + * flush it few times without switching into ASTAGE_CAPTURE_WAIT. This + * allows to reduce stalls due to other threads waiting for atom in + * ASTAGE_CAPTURE_WAIT stage. ->preflush is counter of these + * preliminary flushes. */ + int preflush; + /* have we waited on atom. */ + int wait; + int failed; + int wake_ktxnmgrd_up; +} commit_data; + +/* + * Called from commit_txnh() repeatedly, until either error happens, or atom + * commits successfully. + */ +static int try_commit_txnh(commit_data * cd) +{ + int result; + + assert("nikita-2968", lock_stack_isclean(get_current_lock_stack())); + + /* Get the atom and txnh locked. */ + cd->atom = txnh_get_atom(cd->txnh); + assert("jmacd-309", cd->atom != NULL); + spin_unlock_txnh(cd->txnh); + + if (cd->wait) { + cd->atom->nr_waiters--; + cd->wait = 0; + } + + if (cd->atom->stage == ASTAGE_DONE) + return 0; + + if (cd->failed) + return 0; + + if (atom_should_commit(cd->atom)) { + /* if atom is _very_ large schedule it for commit as soon as + * possible. */ + if (atom_should_commit_asap(cd->atom)) { + /* + * When atom is in PRE_COMMIT or later stage following + * invariant (encoded in atom_can_be_committed()) + * holds: there is exactly one non-waiter transaction + * handle opened on this atom. When thread wants to + * wait until atom commits (for example sync()) it + * waits on atom event after increasing + * atom->nr_waiters (see blow in this function). It + * cannot be guaranteed that atom is already committed + * after receiving event, so loop has to be + * re-started. But if atom switched into PRE_COMMIT + * stage and became too large, we cannot change its + * state back to CAPTURE_WAIT (atom stage can only + * increase monotonically), hence this check. + */ + if (cd->atom->stage < ASTAGE_CAPTURE_WAIT) + reiser4_atom_set_stage(cd->atom, + ASTAGE_CAPTURE_WAIT); + cd->atom->flags |= ATOM_FORCE_COMMIT; + } + if (cd->txnh->flags & TXNH_DONT_COMMIT) { + /* + * this thread (transaction handle that is) doesn't + * want to commit atom. Notify waiters that handle is + * closed. This can happen, for example, when we are + * under VFS directory lock and don't want to commit + * atom right now to avoid stalling other threads + * working in the same directory. + */ + + /* Wake the ktxnmgrd up if the ktxnmgrd is needed to + * commit this atom: no atom waiters and only one + * (our) open transaction handle. */ + cd->wake_ktxnmgrd_up = + cd->atom->txnh_count == 1 && + cd->atom->nr_waiters == 0; + reiser4_atom_send_event(cd->atom); + result = 0; + } else if (!atom_can_be_committed(cd->atom)) { + if (should_wait_commit(cd->txnh)) { + /* sync(): wait for commit */ + cd->atom->nr_waiters++; + cd->wait = 1; + reiser4_atom_wait_event(cd->atom); + result = RETERR(-E_REPEAT); + } else { + result = 0; + } + } else if (cd->preflush > 0 && !is_current_ktxnmgrd()) { + /* + * optimization: flush atom without switching it into + * ASTAGE_CAPTURE_WAIT. + * + * But don't do this for ktxnmgrd, because ktxnmgrd + * should never block on atom fusion. + */ + result = flush_current_atom(JNODE_FLUSH_WRITE_BLOCKS, + LONG_MAX, &cd->nr_written, + &cd->atom, NULL); + if (result == 0) { + spin_unlock_atom(cd->atom); + cd->preflush = 0; + result = RETERR(-E_REPEAT); + } else /* Atoms wasn't flushed + * completely. Rinse. Repeat. */ + --cd->preflush; + } else { + /* We change atom state to ASTAGE_CAPTURE_WAIT to + prevent atom fusion and count ourself as an active + flusher */ + reiser4_atom_set_stage(cd->atom, ASTAGE_CAPTURE_WAIT); + cd->atom->flags |= ATOM_FORCE_COMMIT; + + result = + commit_current_atom(&cd->nr_written, &cd->atom); + if (result != 0 && result != -E_REPEAT) + cd->failed = 1; + } + } else + result = 0; + +#if REISER4_DEBUG + if (result == 0) + assert_spin_locked(&(cd->atom->alock)); +#endif + + /* perfectly valid assertion, except that when atom/txnh is not locked + * fusion can take place, and cd->atom points nowhere. */ + /* + assert("jmacd-1028", ergo(result != 0, spin_atom_is_not_locked(cd->atom))); + */ + return result; +} + +/* Called to commit a transaction handle. This decrements the atom's number of open + handles and if it is the last handle to commit and the atom should commit, initiates + atom commit. if commit does not fail, return number of written blocks */ +static int commit_txnh(txn_handle * txnh) +{ + commit_data cd; + assert("umka-192", txnh != NULL); + + memset(&cd, 0, sizeof cd); + cd.txnh = txnh; + cd.preflush = 10; + + /* calls try_commit_txnh() until either atom commits, or error + * happens */ + while (try_commit_txnh(&cd) != 0) + reiser4_preempt_point(); + + spin_lock_txnh(txnh); + + cd.atom->txnh_count -= 1; + txnh->atom = NULL; + /* remove transaction handle from atom's list of transaction handles */ + list_del_init(&txnh->txnh_link); + + spin_unlock_txnh(txnh); + atom_dec_and_unlock(cd.atom); + /* if we don't want to do a commit (TXNH_DONT_COMMIT is set, probably + * because it takes time) by current thread, we do that work + * asynchronously by ktxnmgrd daemon. */ + if (cd.wake_ktxnmgrd_up) + ktxnmgrd_kick(&get_current_super_private()->tmgr); + + return 0; +} + +/* TRY_CAPTURE */ + +/* This routine attempts a single block-capture request. It may return -E_REPEAT if some + condition indicates that the request should be retried, and it may block if the + txn_capture mode does not include the TXN_CAPTURE_NONBLOCKING request flag. + + This routine encodes the basic logic of block capturing described by: + + http://namesys.com/v4/v4.html + + Our goal here is to ensure that any two blocks that contain dependent modifications + should commit at the same time. This function enforces this discipline by initiating + fusion whenever a transaction handle belonging to one atom requests to read or write a + block belonging to another atom (TXN_CAPTURE_WRITE or TXN_CAPTURE_READ_ATOMIC). + + In addition, this routine handles the initial assignment of atoms to blocks and + transaction handles. These are possible outcomes of this function: + + 1. The block and handle are already part of the same atom: return immediate success + + 2. The block is assigned but the handle is not: call capture_assign_txnh to assign + the handle to the block's atom. + + 3. The handle is assigned but the block is not: call capture_assign_block to assign + the block to the handle's atom. + + 4. Both handle and block are assigned, but to different atoms: call capture_init_fusion + to fuse atoms. + + 5. Neither block nor handle are assigned: create a new atom and assign them both. + + 6. A read request for a non-captured block: return immediate success. + + This function acquires and releases the handle's spinlock. This function is called + under the jnode lock and if the return value is 0, it returns with the jnode lock still + held. If the return is -E_REPEAT or some other error condition, the jnode lock is + released. The external interface (reiser4_try_capture) manages re-aquiring the jnode + lock in the failure case. +*/ +static int try_capture_block( + txn_handle * txnh, jnode * node, txn_capture mode, + txn_atom ** atom_alloc) +{ + txn_atom *block_atom; + txn_atom *txnh_atom; + + /* Should not call capture for READ_NONCOM requests, handled in reiser4_try_capture. */ + assert("jmacd-567", CAPTURE_TYPE(mode) != TXN_CAPTURE_READ_NONCOM); + + /* FIXME-ZAM-HANS: FIXME_LATER_JMACD Should assert that atom->tree == + * node->tree somewhere. */ + assert("umka-194", txnh != NULL); + assert("umka-195", node != NULL); + + /* The jnode is already locked! Being called from reiser4_try_capture(). */ + assert_spin_locked(&(node->guard)); + block_atom = node->atom; + + /* Get txnh spinlock, this allows us to compare txn_atom pointers but it doesn't + let us touch the atoms themselves. */ + spin_lock_txnh(txnh); + txnh_atom = txnh->atom; + /* Process of capturing continues into one of four branches depends on + which atoms from (block atom (node->atom), current atom (txnh->atom)) + exist. */ + if (txnh_atom == NULL) { + if (block_atom == NULL) { + spin_unlock_txnh(txnh); + spin_unlock_jnode(node); + /* assign empty atom to the txnh and repeat */ + return atom_begin_and_assign_to_txnh(atom_alloc, txnh); + } else { + atomic_inc(&block_atom->refcount); + /* node spin-lock isn't needed anymore */ + spin_unlock_jnode(node); + if (!spin_trylock_atom(block_atom)) { + spin_unlock_txnh(txnh); + spin_lock_atom(block_atom); + spin_lock_txnh(txnh); + } + /* re-check state after getting txnh and the node + * atom spin-locked */ + if (node->atom != block_atom || txnh->atom != NULL) { + spin_unlock_txnh(txnh); + atom_dec_and_unlock(block_atom); + return RETERR(-E_REPEAT); + } + atomic_dec(&block_atom->refcount); + if (block_atom->stage > ASTAGE_CAPTURE_WAIT || + (block_atom->stage == ASTAGE_CAPTURE_WAIT && + block_atom->txnh_count != 0)) + return capture_fuse_wait(txnh, block_atom, NULL, mode); + capture_assign_txnh_nolock(block_atom, txnh); + spin_unlock_txnh(txnh); + spin_unlock_atom(block_atom); + return RETERR(-E_REPEAT); + } + } else { + /* It is time to perform deadlock prevention check over the + node we want to capture. It is possible this node was locked + for read without capturing it. The optimization which allows + to do it helps us in keeping atoms independent as long as + possible but it may cause lock/fuse deadlock problems. + + A number of similar deadlock situations with locked but not + captured nodes were found. In each situation there are two + or more threads: one of them does flushing while another one + does routine balancing or tree lookup. The flushing thread + (F) sleeps in long term locking request for node (N), another + thread (A) sleeps in trying to capture some node already + belonging the atom F, F has a state which prevents + immediately fusion . + + Deadlocks of this kind cannot happen if node N was properly + captured by thread A. The F thread fuse atoms before locking + therefore current atom of thread F and current atom of thread + A became the same atom and thread A may proceed. This does + not work if node N was not captured because the fusion of + atom does not happens. + + The following scheme solves the deadlock: If + longterm_lock_znode locks and does not capture a znode, that + znode is marked as MISSED_IN_CAPTURE. A node marked this way + is processed by the code below which restores the missed + capture and fuses current atoms of all the node lock owners + by calling the fuse_not_fused_lock_owners() function. */ + if (JF_ISSET(node, JNODE_MISSED_IN_CAPTURE)) { + JF_CLR(node, JNODE_MISSED_IN_CAPTURE); + if (jnode_is_znode(node) && znode_is_locked(JZNODE(node))) { + spin_unlock_txnh(txnh); + spin_unlock_jnode(node); + fuse_not_fused_lock_owners(txnh, JZNODE(node)); + return RETERR(-E_REPEAT); + } + } + if (block_atom == NULL) { + atomic_inc(&txnh_atom->refcount); + spin_unlock_txnh(txnh); + if (!spin_trylock_atom(txnh_atom)) { + spin_unlock_jnode(node); + spin_lock_atom(txnh_atom); + spin_lock_jnode(node); + } + if (txnh->atom != txnh_atom || node->atom != NULL + || JF_ISSET(node, JNODE_IS_DYING)) { + spin_unlock_jnode(node); + atom_dec_and_unlock(txnh_atom); + return RETERR(-E_REPEAT); + } + atomic_dec(&txnh_atom->refcount); + capture_assign_block_nolock(txnh_atom, node); + spin_unlock_atom(txnh_atom); + } else { + if (txnh_atom != block_atom) { + if (mode & TXN_CAPTURE_DONT_FUSE) { + spin_unlock_txnh(txnh); + spin_unlock_jnode(node); + /* we are in a "no-fusion" mode and @node is + * already part of transaction. */ + return RETERR(-E_NO_NEIGHBOR); + } + return capture_init_fusion(node, txnh, mode); + } + spin_unlock_txnh(txnh); + } + } + return 0; +} + +static txn_capture +build_capture_mode(jnode * node, znode_lock_mode lock_mode, txn_capture flags) +{ + txn_capture cap_mode; + + assert_spin_locked(&(node->guard)); + + /* FIXME_JMACD No way to set TXN_CAPTURE_READ_MODIFY yet. */ + + if (lock_mode == ZNODE_WRITE_LOCK) { + cap_mode = TXN_CAPTURE_WRITE; + } else if (node->atom != NULL) { + cap_mode = TXN_CAPTURE_WRITE; + } else if (0 && /* txnh->mode == TXN_READ_FUSING && */ + jnode_get_level(node) == LEAF_LEVEL) { + /* NOTE-NIKITA TXN_READ_FUSING is not currently used */ + /* We only need a READ_FUSING capture at the leaf level. This + is because the internal levels of the tree (twigs included) + are redundant from the point of the user that asked for a + read-fusing transcrash. The user only wants to read-fuse + atoms due to reading uncommitted data that another user has + written. It is the file system that reads/writes the + internal tree levels, the user only reads/writes leaves. */ + cap_mode = TXN_CAPTURE_READ_ATOMIC; + } else { + /* In this case (read lock at a non-leaf) there's no reason to + * capture. */ + /* cap_mode = TXN_CAPTURE_READ_NONCOM; */ + return 0; + } + + cap_mode |= (flags & (TXN_CAPTURE_NONBLOCKING | TXN_CAPTURE_DONT_FUSE)); + assert("nikita-3186", cap_mode != 0); + return cap_mode; +} + +/* This is an external interface to try_capture_block(), it calls + try_capture_block() repeatedly as long as -E_REPEAT is returned. + + @node: node to capture, + @lock_mode: read or write lock is used in capture mode calculation, + @flags: see txn_capture flags enumeration, + @can_coc : can copy-on-capture + + @return: 0 - node was successfully captured, -E_REPEAT - capture request + cannot be processed immediately as it was requested in flags, + < 0 - other errors. +*/ +int reiser4_try_capture(jnode *node, znode_lock_mode lock_mode, + txn_capture flags) +{ + txn_atom *atom_alloc = NULL; + txn_capture cap_mode; + txn_handle *txnh = get_current_context()->trans; + int ret; + + assert_spin_locked(&(node->guard)); + + repeat: + if (JF_ISSET(node, JNODE_IS_DYING)) + return RETERR(-EINVAL); + if (node->atom != NULL && txnh->atom == node->atom) + return 0; + cap_mode = build_capture_mode(node, lock_mode, flags); + if (cap_mode == 0 || + (!(cap_mode & TXN_CAPTURE_WTYPES) && node->atom == NULL)) { + /* Mark this node as "MISSED". It helps in further deadlock + * analysis */ + if (jnode_is_znode(node)) + JF_SET(node, JNODE_MISSED_IN_CAPTURE); + return 0; + } + /* Repeat try_capture as long as -E_REPEAT is returned. */ + ret = try_capture_block(txnh, node, cap_mode, &atom_alloc); + /* Regardless of non_blocking: + + If ret == 0 then jnode is still locked. + If ret != 0 then jnode is unlocked. + */ +#if REISER4_DEBUG + if (ret == 0) + assert_spin_locked(&(node->guard)); + else + assert_spin_not_locked(&(node->guard)); +#endif + assert_spin_not_locked(&(txnh->guard)); + + if (ret == -E_REPEAT) { + /* E_REPEAT implies all locks were released, therefore we need + to take the jnode's lock again. */ + spin_lock_jnode(node); + + /* Although this may appear to be a busy loop, it is not. + There are several conditions that cause E_REPEAT to be + returned by the call to try_capture_block, all cases + indicating some kind of state change that means you should + retry the request and will get a different result. In some + cases this could be avoided with some extra code, but + generally it is done because the necessary locks were + released as a result of the operation and repeating is the + simplest thing to do (less bug potential). The cases are: + atom fusion returns E_REPEAT after it completes (jnode and + txnh were unlocked); race conditions in assign_block, + assign_txnh, and init_fusion return E_REPEAT (trylock + failure); after going to sleep in capture_fuse_wait + (request was blocked but may now succeed). I'm not quite + sure how capture_copy works yet, but it may also return + E_REPEAT. When the request is legitimately blocked, the + requestor goes to sleep in fuse_wait, so this is not a busy + loop. */ + /* NOTE-NIKITA: still don't understand: + + try_capture_block->capture_assign_txnh->spin_trylock_atom->E_REPEAT + + looks like busy loop? + */ + goto repeat; + } + + /* free extra atom object that was possibly allocated by + try_capture_block(). + + Do this before acquiring jnode spin lock to + minimize time spent under lock. --nikita */ + if (atom_alloc != NULL) { + kmem_cache_free(_atom_slab, atom_alloc); + } + + if (ret != 0) { + if (ret == -E_BLOCK) { + assert("nikita-3360", + cap_mode & TXN_CAPTURE_NONBLOCKING); + ret = -E_REPEAT; + } + + /* Failure means jnode is not locked. FIXME_LATER_JMACD May + want to fix the above code to avoid releasing the lock and + re-acquiring it, but there are cases were failure occurs + when the lock is not held, and those cases would need to be + modified to re-take the lock. */ + spin_lock_jnode(node); + } + + /* Jnode is still locked. */ + assert_spin_locked(&(node->guard)); + return ret; +} + +static void release_two_atoms(txn_atom *one, txn_atom *two) +{ + spin_unlock_atom(one); + atom_dec_and_unlock(two); + spin_lock_atom(one); + atom_dec_and_unlock(one); +} + +/* This function sets up a call to try_capture_block and repeats as long as -E_REPEAT is + returned by that routine. The txn_capture request mode is computed here depending on + the transaction handle's type and the lock request. This is called from the depths of + the lock manager with the jnode lock held and it always returns with the jnode lock + held. +*/ + +/* fuse all 'active' atoms of lock owners of given node. */ +static void fuse_not_fused_lock_owners(txn_handle * txnh, znode * node) +{ + lock_handle *lh; + int repeat; + txn_atom *atomh, *atomf; + reiser4_context *me = get_current_context(); + reiser4_context *ctx = NULL; + + assert_spin_not_locked(&(ZJNODE(node)->guard)); + assert_spin_not_locked(&(txnh->hlock)); + + repeat: + repeat = 0; + atomh = txnh_get_atom(txnh); + spin_unlock_txnh(txnh); + assert("zam-692", atomh != NULL); + + spin_lock_zlock(&node->lock); + /* inspect list of lock owners */ + list_for_each_entry(lh, &node->lock.owners, owners_link) { + ctx = get_context_by_lock_stack(lh->owner); + if (ctx == me) + continue; + /* below we use two assumptions to avoid addition spin-locks + for checking the condition : + + 1) if the lock stack has lock, the transaction should be + opened, i.e. ctx->trans != NULL; + + 2) reading of well-aligned ctx->trans->atom is atomic, if it + equals to the address of spin-locked atomh, we take that + the atoms are the same, nothing has to be captured. */ + if (atomh != ctx->trans->atom) { + reiser4_wake_up(lh->owner); + repeat = 1; + break; + } + } + if (repeat) { + if (!spin_trylock_txnh(ctx->trans)) { + spin_unlock_zlock(&node->lock); + spin_unlock_atom(atomh); + goto repeat; + } + atomf = ctx->trans->atom; + if (atomf == NULL) { + capture_assign_txnh_nolock(atomh, ctx->trans); + /* release zlock lock _after_ assigning the atom to the + * transaction handle, otherwise the lock owner thread + * may unlock all znodes, exit kernel context and here + * we would access an invalid transaction handle. */ + spin_unlock_zlock(&node->lock); + spin_unlock_atom(atomh); + spin_unlock_txnh(ctx->trans); + goto repeat; + } + assert("zam-1059", atomf != atomh); + spin_unlock_zlock(&node->lock); + atomic_inc(&atomh->refcount); + atomic_inc(&atomf->refcount); + spin_unlock_txnh(ctx->trans); + if (atomf > atomh) { + spin_lock_atom_nested(atomf); + } else { + spin_unlock_atom(atomh); + spin_lock_atom(atomf); + spin_lock_atom_nested(atomh); + } + if (atomh == atomf || !atom_isopen(atomh) || !atom_isopen(atomf)) { + release_two_atoms(atomf, atomh); + goto repeat; + } + atomic_dec(&atomh->refcount); + atomic_dec(&atomf->refcount); + capture_fuse_into(atomf, atomh); + goto repeat; + } + spin_unlock_zlock(&node->lock); + spin_unlock_atom(atomh); +} + +/* This is the interface to capture unformatted nodes via their struct page + reference. Currently it is only used in reiser4_invalidatepage */ +int try_capture_page_to_invalidate(struct page *pg) +{ + int ret; + jnode *node; + + assert("umka-292", pg != NULL); + assert("nikita-2597", PageLocked(pg)); + + if (IS_ERR(node = jnode_of_page(pg))) { + return PTR_ERR(node); + } + + spin_lock_jnode(node); + unlock_page(pg); + + ret = reiser4_try_capture(node, ZNODE_WRITE_LOCK, 0); + spin_unlock_jnode(node); + jput(node); + lock_page(pg); + return ret; +} + +/* This informs the transaction manager when a node is deleted. Add the block to the + atom's delete set and uncapture the block. + +VS-FIXME-HANS: this E_REPEAT paradigm clutters the code and creates a need for +explanations. find all the functions that use it, and unless there is some very +good reason to use it (I have not noticed one so far and I doubt it exists, but maybe somewhere somehow....), +move the loop to inside the function. + +VS-FIXME-HANS: can this code be at all streamlined? In particular, can you lock and unlock the jnode fewer times? + */ +void reiser4_uncapture_page(struct page *pg) +{ + jnode *node; + txn_atom *atom; + + assert("umka-199", pg != NULL); + assert("nikita-3155", PageLocked(pg)); + + clear_page_dirty_for_io(pg); + + reiser4_wait_page_writeback(pg); + + node = jprivate(pg); + BUG_ON(node == NULL); + + spin_lock_jnode(node); + + atom = jnode_get_atom(node); + if (atom == NULL) { + assert("jmacd-7111", !JF_ISSET(node, JNODE_DIRTY)); + spin_unlock_jnode(node); + return; + } + + /* We can remove jnode from transaction even if it is on flush queue + * prepped list, we only need to be sure that flush queue is not being + * written by reiser4_write_fq(). reiser4_write_fq() does not use atom + * spin lock for protection of the prepped nodes list, instead + * write_fq() increments atom's nr_running_queues counters for the time + * when prepped list is not protected by spin lock. Here we check this + * counter if we want to remove jnode from flush queue and, if the + * counter is not zero, wait all reiser4_write_fq() for this atom to + * complete. This is not significant overhead. */ + while (JF_ISSET(node, JNODE_FLUSH_QUEUED) && atom->nr_running_queues) { + spin_unlock_jnode(node); + /* + * at this moment we want to wait for "atom event", viz. wait + * until @node can be removed from flush queue. But + * reiser4_atom_wait_event() cannot be called with page locked, + * because it deadlocks with jnode_extent_write(). Unlock page, + * after making sure (through page_cache_get()) that it cannot + * be released from memory. + */ + page_cache_get(pg); + unlock_page(pg); + reiser4_atom_wait_event(atom); + lock_page(pg); + /* + * page may has been detached by ->writepage()->releasepage(). + */ + reiser4_wait_page_writeback(pg); + spin_lock_jnode(node); + page_cache_release(pg); + atom = jnode_get_atom(node); +/* VS-FIXME-HANS: improve the commenting in this function */ + if (atom == NULL) { + spin_unlock_jnode(node); + return; + } + } + reiser4_uncapture_block(node); + spin_unlock_atom(atom); + jput(node); +} + +/* this is used in extent's kill hook to uncapture and unhash jnodes attached to + * inode's tree of jnodes */ +void reiser4_uncapture_jnode(jnode * node) +{ + txn_atom *atom; + + assert_spin_locked(&(node->guard)); + assert("", node->pg == 0); + + atom = jnode_get_atom(node); + if (atom == NULL) { + assert("jmacd-7111", !JF_ISSET(node, JNODE_DIRTY)); + spin_unlock_jnode(node); + return; + } + + reiser4_uncapture_block(node); + spin_unlock_atom(atom); + jput(node); +} + +/* No-locking version of assign_txnh. Sets the transaction handle's atom pointer, + increases atom refcount and txnh_count, adds to txnh_list. */ +static void capture_assign_txnh_nolock(txn_atom *atom, txn_handle *txnh) +{ + assert("umka-200", atom != NULL); + assert("umka-201", txnh != NULL); + + assert_spin_locked(&(txnh->hlock)); + assert_spin_locked(&(atom->alock)); + assert("jmacd-824", txnh->atom == NULL); + assert("nikita-3540", atom_isopen(atom)); + BUG_ON(txnh->atom != NULL); + + atomic_inc(&atom->refcount); + txnh->atom = atom; + reiser4_ctx_gfp_mask_set(); + list_add_tail(&txnh->txnh_link, &atom->txnh_list); + atom->txnh_count += 1; +} + +/* No-locking version of assign_block. Sets the block's atom pointer, references the + block, adds it to the clean or dirty capture_jnode list, increments capture_count. */ +static void capture_assign_block_nolock(txn_atom *atom, jnode *node) +{ + assert("umka-202", atom != NULL); + assert("umka-203", node != NULL); + assert_spin_locked(&(node->guard)); + assert_spin_locked(&(atom->alock)); + assert("jmacd-323", node->atom == NULL); + BUG_ON(!list_empty_careful(&node->capture_link)); + assert("nikita-3470", !JF_ISSET(node, JNODE_DIRTY)); + + /* Pointer from jnode to atom is not counted in atom->refcount. */ + node->atom = atom; + + list_add_tail(&node->capture_link, ATOM_CLEAN_LIST(atom)); + atom->capture_count += 1; + /* reference to jnode is acquired by atom. */ + jref(node); + + ON_DEBUG(count_jnode(atom, node, NOT_CAPTURED, CLEAN_LIST, 1)); + + LOCK_CNT_INC(t_refs); +} + +/* common code for dirtying both unformatted jnodes and formatted znodes. */ +static void do_jnode_make_dirty(jnode * node, txn_atom * atom) +{ + assert_spin_locked(&(node->guard)); + assert_spin_locked(&(atom->alock)); + assert("jmacd-3981", !JF_ISSET(node, JNODE_DIRTY)); + + JF_SET(node, JNODE_DIRTY); + + get_current_context()->nr_marked_dirty++; + + /* We grab2flush_reserve one additional block only if node was + not CREATED and jnode_flush did not sort it into neither + relocate set nor overwrite one. If node is in overwrite or + relocate set we assume that atom's flush reserved counter was + already adjusted. */ + if (!JF_ISSET(node, JNODE_CREATED) && !JF_ISSET(node, JNODE_RELOC) + && !JF_ISSET(node, JNODE_OVRWR) && jnode_is_leaf(node) + && !jnode_is_cluster_page(node)) { + assert("vs-1093", !reiser4_blocknr_is_fake(&node->blocknr)); + assert("vs-1506", *jnode_get_block(node) != 0); + grabbed2flush_reserved_nolock(atom, (__u64) 1); + JF_SET(node, JNODE_FLUSH_RESERVED); + } + + if (!JF_ISSET(node, JNODE_FLUSH_QUEUED)) { + /* If the atom is not set yet, it will be added to the appropriate list in + capture_assign_block_nolock. */ + /* Sometimes a node is set dirty before being captured -- the case for new + jnodes. In that case the jnode will be added to the appropriate list + in capture_assign_block_nolock. Another reason not to re-link jnode is + that jnode is on a flush queue (see flush.c for details) */ + + int level = jnode_get_level(node); + + assert("nikita-3152", !JF_ISSET(node, JNODE_OVRWR)); + assert("zam-654", atom->stage < ASTAGE_PRE_COMMIT); + assert("nikita-2607", 0 <= level); + assert("nikita-2606", level <= REAL_MAX_ZTREE_HEIGHT); + + /* move node to atom's dirty list */ + list_move_tail(&node->capture_link, ATOM_DIRTY_LIST(atom, level)); + ON_DEBUG(count_jnode + (atom, node, NODE_LIST(node), DIRTY_LIST, 1)); + } +} + +/* Set the dirty status for this (spin locked) jnode. */ +void jnode_make_dirty_locked(jnode * node) +{ + assert("umka-204", node != NULL); + assert_spin_locked(&(node->guard)); + + if (REISER4_DEBUG && rofs_jnode(node)) { + warning("nikita-3365", "Dirtying jnode on rofs"); + dump_stack(); + } + + /* Fast check for already dirty node */ + if (!JF_ISSET(node, JNODE_DIRTY)) { + txn_atom *atom; + + atom = jnode_get_atom(node); + assert("vs-1094", atom); + /* Check jnode dirty status again because node spin lock might + * be released inside jnode_get_atom(). */ + if (likely(!JF_ISSET(node, JNODE_DIRTY))) + do_jnode_make_dirty(node, atom); + spin_unlock_atom(atom); + } +} + +/* Set the dirty status for this znode. */ +void znode_make_dirty(znode * z) +{ + jnode *node; + struct page *page; + + assert("umka-204", z != NULL); + assert("nikita-3290", znode_above_root(z) || znode_is_loaded(z)); + assert("nikita-3560", znode_is_write_locked(z)); + + node = ZJNODE(z); + /* znode is longterm locked, we can check dirty bit without spinlock */ + if (JF_ISSET(node, JNODE_DIRTY)) { + /* znode is dirty already. All we have to do is to change znode version */ + z->version = znode_build_version(jnode_get_tree(node)); + return; + } + + spin_lock_jnode(node); + jnode_make_dirty_locked(node); + page = jnode_page(node); + if (page != NULL) { + /* this is useful assertion (allows one to check that no + * modifications are lost due to update of in-flight page), + * but it requires locking on page to check PG_writeback + * bit. */ + /* assert("nikita-3292", + !PageWriteback(page) || ZF_ISSET(z, JNODE_WRITEBACK)); */ + page_cache_get(page); + + /* jnode lock is not needed for the rest of + * znode_set_dirty(). */ + spin_unlock_jnode(node); + /* reiser4 file write code calls set_page_dirty for + * unformatted nodes, for formatted nodes we do it here. */ + set_page_dirty_notag(page); + page_cache_release(page); + /* bump version counter in znode */ + z->version = znode_build_version(jnode_get_tree(node)); + } else { + assert("zam-596", znode_above_root(JZNODE(node))); + spin_unlock_jnode(node); + } + + assert("nikita-1900", znode_is_write_locked(z)); + assert("jmacd-9777", node->atom != NULL); +} + +int reiser4_sync_atom(txn_atom * atom) +{ + int result; + txn_handle *txnh; + + txnh = get_current_context()->trans; + + result = 0; + if (atom != NULL) { + if (atom->stage < ASTAGE_PRE_COMMIT) { + spin_lock_txnh(txnh); + capture_assign_txnh_nolock(atom, txnh); + result = force_commit_atom(txnh); + } else if (atom->stage < ASTAGE_POST_COMMIT) { + /* wait atom commit */ + reiser4_atom_wait_event(atom); + /* try once more */ + result = RETERR(-E_REPEAT); + } else + spin_unlock_atom(atom); + } + return result; +} + +#if REISER4_DEBUG + +/* move jnode form one list to another + call this after atom->capture_count is updated */ +void +count_jnode(txn_atom * atom, jnode * node, atom_list old_list, + atom_list new_list, int check_lists) +{ + struct list_head *pos; + + assert("zam-1018", atom_is_protected(atom)); + assert_spin_locked(&(node->guard)); + assert("", NODE_LIST(node) == old_list); + + switch (NODE_LIST(node)) { + case NOT_CAPTURED: + break; + case DIRTY_LIST: + assert("", atom->dirty > 0); + atom->dirty--; + break; + case CLEAN_LIST: + assert("", atom->clean > 0); + atom->clean--; + break; + case FQ_LIST: + assert("", atom->fq > 0); + atom->fq--; + break; + case WB_LIST: + assert("", atom->wb > 0); + atom->wb--; + break; + case OVRWR_LIST: + assert("", atom->ovrwr > 0); + atom->ovrwr--; + break; + default: + impossible("", ""); + } + + switch (new_list) { + case NOT_CAPTURED: + break; + case DIRTY_LIST: + atom->dirty++; + break; + case CLEAN_LIST: + atom->clean++; + break; + case FQ_LIST: + atom->fq++; + break; + case WB_LIST: + atom->wb++; + break; + case OVRWR_LIST: + atom->ovrwr++; + break; + default: + impossible("", ""); + } + ASSIGN_NODE_LIST(node, new_list); + if (0 && check_lists) { + int count; + tree_level level; + + count = 0; + + /* flush queue list */ + /* reiser4_check_fq(atom); */ + + /* dirty list */ + count = 0; + for (level = 0; level < REAL_MAX_ZTREE_HEIGHT + 1; level += 1) { + list_for_each(pos, ATOM_DIRTY_LIST(atom, level)) + count++; + } + if (count != atom->dirty) + warning("", "dirty counter %d, real %d\n", atom->dirty, + count); + + /* clean list */ + count = 0; + list_for_each(pos, ATOM_CLEAN_LIST(atom)) + count++; + if (count != atom->clean) + warning("", "clean counter %d, real %d\n", atom->clean, + count); + + /* wb list */ + count = 0; + list_for_each(pos, ATOM_WB_LIST(atom)) + count++; + if (count != atom->wb) + warning("", "wb counter %d, real %d\n", atom->wb, + count); + + /* overwrite list */ + count = 0; + list_for_each(pos, ATOM_OVRWR_LIST(atom)) + count++; + + if (count != atom->ovrwr) + warning("", "ovrwr counter %d, real %d\n", atom->ovrwr, + count); + } + assert("vs-1624", atom->num_queued == atom->fq); + if (atom->capture_count != + atom->dirty + atom->clean + atom->ovrwr + atom->wb + atom->fq) { + printk + ("count %d, dirty %d clean %d ovrwr %d wb %d fq %d\n", + atom->capture_count, atom->dirty, atom->clean, atom->ovrwr, + atom->wb, atom->fq); + assert("vs-1622", + atom->capture_count == + atom->dirty + atom->clean + atom->ovrwr + atom->wb + + atom->fq); + } +} + +#endif + +/* Make node OVRWR and put it on atom->overwrite_nodes list, atom lock and jnode + * lock should be taken before calling this function. */ +void jnode_make_wander_nolock(jnode * node) +{ + txn_atom *atom; + + assert("nikita-2431", node != NULL); + assert("nikita-2432", !JF_ISSET(node, JNODE_RELOC)); + assert("nikita-3153", JF_ISSET(node, JNODE_DIRTY)); + assert("zam-897", !JF_ISSET(node, JNODE_FLUSH_QUEUED)); + assert("nikita-3367", !reiser4_blocknr_is_fake(jnode_get_block(node))); + + atom = node->atom; + + assert("zam-895", atom != NULL); + assert("zam-894", atom_is_protected(atom)); + + JF_SET(node, JNODE_OVRWR); + /* move node to atom's overwrite list */ + list_move_tail(&node->capture_link, ATOM_OVRWR_LIST(atom)); + ON_DEBUG(count_jnode(atom, node, DIRTY_LIST, OVRWR_LIST, 1)); +} + +/* Same as jnode_make_wander_nolock, but all necessary locks are taken inside + * this function. */ +void jnode_make_wander(jnode * node) +{ + txn_atom *atom; + + spin_lock_jnode(node); + atom = jnode_get_atom(node); + assert("zam-913", atom != NULL); + assert("zam-914", !JF_ISSET(node, JNODE_RELOC)); + + jnode_make_wander_nolock(node); + spin_unlock_atom(atom); + spin_unlock_jnode(node); +} + +/* this just sets RELOC bit */ +static void jnode_make_reloc_nolock(flush_queue_t * fq, jnode * node) +{ + assert_spin_locked(&(node->guard)); + assert("zam-916", JF_ISSET(node, JNODE_DIRTY)); + assert("zam-917", !JF_ISSET(node, JNODE_RELOC)); + assert("zam-918", !JF_ISSET(node, JNODE_OVRWR)); + assert("zam-920", !JF_ISSET(node, JNODE_FLUSH_QUEUED)); + assert("nikita-3367", !reiser4_blocknr_is_fake(jnode_get_block(node))); + jnode_set_reloc(node); +} + +/* Make znode RELOC and put it on flush queue */ +void znode_make_reloc(znode * z, flush_queue_t * fq) +{ + jnode *node; + txn_atom *atom; + + node = ZJNODE(z); + spin_lock_jnode(node); + + atom = jnode_get_atom(node); + assert("zam-919", atom != NULL); + + jnode_make_reloc_nolock(fq, node); + queue_jnode(fq, node); + + spin_unlock_atom(atom); + spin_unlock_jnode(node); + +} + +/* Make unformatted node RELOC and put it on flush queue */ +void unformatted_make_reloc(jnode *node, flush_queue_t *fq) +{ + assert("vs-1479", jnode_is_unformatted(node)); + + jnode_make_reloc_nolock(fq, node); + queue_jnode(fq, node); +} + +int reiser4_capture_super_block(struct super_block *s) +{ + int result; + znode *uber; + lock_handle lh; + + init_lh(&lh); + result = get_uber_znode(reiser4_get_tree(s), + ZNODE_WRITE_LOCK, ZNODE_LOCK_LOPRI, &lh); + if (result) + return result; + + uber = lh.node; + /* Grabbing one block for superblock */ + result = reiser4_grab_space_force((__u64) 1, BA_RESERVED); + if (result != 0) + return result; + + znode_make_dirty(uber); + + done_lh(&lh); + return 0; +} + +/* Wakeup every handle on the atom's WAITFOR list */ +static void wakeup_atom_waitfor_list(txn_atom * atom) +{ + txn_wait_links *wlinks; + + assert("umka-210", atom != NULL); + + /* atom is locked */ + list_for_each_entry(wlinks, &atom->fwaitfor_list, _fwaitfor_link) { + if (wlinks->waitfor_cb == NULL || + wlinks->waitfor_cb(atom, wlinks)) + /* Wake up. */ + reiser4_wake_up(wlinks->_lock_stack); + } +} + +/* Wakeup every handle on the atom's WAITING list */ +static void wakeup_atom_waiting_list(txn_atom * atom) +{ + txn_wait_links *wlinks; + + assert("umka-211", atom != NULL); + + /* atom is locked */ + list_for_each_entry(wlinks, &atom->fwaiting_list, _fwaiting_link) { + if (wlinks->waiting_cb == NULL || + wlinks->waiting_cb(atom, wlinks)) + /* Wake up. */ + reiser4_wake_up(wlinks->_lock_stack); + } +} + +/* helper function used by capture_fuse_wait() to avoid "spurious wake-ups" */ +static int wait_for_fusion(txn_atom * atom, txn_wait_links * wlinks) +{ + assert("nikita-3330", atom != NULL); + assert_spin_locked(&(atom->alock)); + + /* atom->txnh_count == 1 is for waking waiters up if we are releasing + * last transaction handle. */ + return atom->stage != ASTAGE_CAPTURE_WAIT || atom->txnh_count == 1; +} + +/* The general purpose of this function is to wait on the first of two possible events. + The situation is that a handle (and its atom atomh) is blocked trying to capture a + block (i.e., node) but the node's atom (atomf) is in the CAPTURE_WAIT state. The + handle's atom (atomh) is not in the CAPTURE_WAIT state. However, atomh could fuse with + another atom or, due to age, enter the CAPTURE_WAIT state itself, at which point it + needs to unblock the handle to avoid deadlock. When the txnh is unblocked it will + proceed and fuse the two atoms in the CAPTURE_WAIT state. + + In other words, if either atomh or atomf change state, the handle will be awakened, + thus there are two lists per atom: WAITING and WAITFOR. + + This is also called by capture_assign_txnh with (atomh == NULL) to wait for atomf to + close but it is not assigned to an atom of its own. + + Lock ordering in this method: all four locks are held: JNODE_LOCK, TXNH_LOCK, + BOTH_ATOM_LOCKS. Result: all four locks are released. +*/ +static int capture_fuse_wait(txn_handle * txnh, txn_atom * atomf, + txn_atom * atomh, txn_capture mode) +{ + int ret; + txn_wait_links wlinks; + + assert("umka-213", txnh != NULL); + assert("umka-214", atomf != NULL); + + if ((mode & TXN_CAPTURE_NONBLOCKING) != 0) { + spin_unlock_txnh(txnh); + spin_unlock_atom(atomf); + + if (atomh) { + spin_unlock_atom(atomh); + } + + return RETERR(-E_BLOCK); + } + + /* Initialize the waiting list links. */ + init_wlinks(&wlinks); + + /* Add txnh to atomf's waitfor list, unlock atomf. */ + list_add_tail(&wlinks._fwaitfor_link, &atomf->fwaitfor_list); + wlinks.waitfor_cb = wait_for_fusion; + atomic_inc(&atomf->refcount); + spin_unlock_atom(atomf); + + if (atomh) { + /* Add txnh to atomh's waiting list, unlock atomh. */ + list_add_tail(&wlinks._fwaiting_link, &atomh->fwaiting_list); + atomic_inc(&atomh->refcount); + spin_unlock_atom(atomh); + } + + /* Go to sleep. */ + spin_unlock_txnh(txnh); + + ret = reiser4_prepare_to_sleep(wlinks._lock_stack); + if (ret == 0) { + reiser4_go_to_sleep(wlinks._lock_stack); + ret = RETERR(-E_REPEAT); + } + + /* Remove from the waitfor list. */ + spin_lock_atom(atomf); + + list_del(&wlinks._fwaitfor_link); + atom_dec_and_unlock(atomf); + + if (atomh) { + /* Remove from the waiting list. */ + spin_lock_atom(atomh); + list_del(&wlinks._fwaiting_link); + atom_dec_and_unlock(atomh); + } + return ret; +} + +static void lock_two_atoms(txn_atom * one, txn_atom * two) +{ + assert("zam-1067", one != two); + + /* lock the atom with lesser address first */ + if (one < two) { + spin_lock_atom(one); + spin_lock_atom_nested(two); + } else { + spin_lock_atom(two); + spin_lock_atom_nested(one); + } +} + +/* Perform the necessary work to prepare for fusing two atoms, which involves + * acquiring two atom locks in the proper order. If one of the node's atom is + * blocking fusion (i.e., it is in the CAPTURE_WAIT stage) and the handle's + * atom is not then the handle's request is put to sleep. If the node's atom + * is committing, then the node can be copy-on-captured. Otherwise, pick the + * atom with fewer pointers to be fused into the atom with more pointer and + * call capture_fuse_into. + */ +static int capture_init_fusion(jnode *node, txn_handle *txnh, txn_capture mode) +{ + txn_atom * txnh_atom = txnh->atom; + txn_atom * block_atom = node->atom; + + atomic_inc(&txnh_atom->refcount); + atomic_inc(&block_atom->refcount); + + spin_unlock_txnh(txnh); + spin_unlock_jnode(node); + + lock_two_atoms(txnh_atom, block_atom); + + if (txnh->atom != txnh_atom || node->atom != block_atom ) { + release_two_atoms(txnh_atom, block_atom); + return RETERR(-E_REPEAT); + } + + atomic_dec(&txnh_atom->refcount); + atomic_dec(&block_atom->refcount); + + assert ("zam-1066", atom_isopen(txnh_atom)); + + if (txnh_atom->stage >= block_atom->stage || + (block_atom->stage == ASTAGE_CAPTURE_WAIT && block_atom->txnh_count == 0)) { + capture_fuse_into(txnh_atom, block_atom); + return RETERR(-E_REPEAT); + } + spin_lock_txnh(txnh); + return capture_fuse_wait(txnh, block_atom, txnh_atom, mode); +} + +/* This function splices together two jnode lists (small and large) and sets all jnodes in + the small list to point to the large atom. Returns the length of the list. */ +static int +capture_fuse_jnode_lists(txn_atom *large, struct list_head *large_head, + struct list_head *small_head) +{ + int count = 0; + jnode *node; + + assert("umka-218", large != NULL); + assert("umka-219", large_head != NULL); + assert("umka-220", small_head != NULL); + /* small atom should be locked also. */ + assert_spin_locked(&(large->alock)); + + /* For every jnode on small's capture list... */ + list_for_each_entry(node, small_head, capture_link) { + count += 1; + + /* With the jnode lock held, update atom pointer. */ + spin_lock_jnode(node); + node->atom = large; + spin_unlock_jnode(node); + } + + /* Splice the lists. */ + list_splice_init(small_head, large_head->prev); + + return count; +} + +/* This function splices together two txnh lists (small and large) and sets all txn handles in + the small list to point to the large atom. Returns the length of the list. */ +static int +capture_fuse_txnh_lists(txn_atom *large, struct list_head *large_head, + struct list_head *small_head) +{ + int count = 0; + txn_handle *txnh; + + assert("umka-221", large != NULL); + assert("umka-222", large_head != NULL); + assert("umka-223", small_head != NULL); + + /* Adjust every txnh to the new atom. */ + list_for_each_entry(txnh, small_head, txnh_link) { + count += 1; + + /* With the txnh lock held, update atom pointer. */ + spin_lock_txnh(txnh); + txnh->atom = large; + spin_unlock_txnh(txnh); + } + + /* Splice the txn_handle list. */ + list_splice_init(small_head, large_head->prev); + + return count; +} + +/* This function fuses two atoms. The captured nodes and handles belonging to SMALL are + added to LARGE and their ->atom pointers are all updated. The associated counts are + updated as well, and any waiting handles belonging to either are awakened. Finally the + smaller atom's refcount is decremented. +*/ +static void capture_fuse_into(txn_atom * small, txn_atom * large) +{ + int level; + unsigned zcount = 0; + unsigned tcount = 0; + + assert("umka-224", small != NULL); + assert("umka-225", small != NULL); + + assert_spin_locked(&(large->alock)); + assert_spin_locked(&(small->alock)); + + assert("jmacd-201", atom_isopen(small)); + assert("jmacd-202", atom_isopen(large)); + + /* Splice and update the per-level dirty jnode lists */ + for (level = 0; level < REAL_MAX_ZTREE_HEIGHT + 1; level += 1) { + zcount += + capture_fuse_jnode_lists(large, + ATOM_DIRTY_LIST(large, level), + ATOM_DIRTY_LIST(small, level)); + } + + /* Splice and update the [clean,dirty] jnode and txnh lists */ + zcount += + capture_fuse_jnode_lists(large, ATOM_CLEAN_LIST(large), + ATOM_CLEAN_LIST(small)); + zcount += + capture_fuse_jnode_lists(large, ATOM_OVRWR_LIST(large), + ATOM_OVRWR_LIST(small)); + zcount += + capture_fuse_jnode_lists(large, ATOM_WB_LIST(large), + ATOM_WB_LIST(small)); + zcount += + capture_fuse_jnode_lists(large, &large->inodes, &small->inodes); + tcount += + capture_fuse_txnh_lists(large, &large->txnh_list, + &small->txnh_list); + + /* Check our accounting. */ + assert("jmacd-1063", + zcount + small->num_queued == small->capture_count); + assert("jmacd-1065", tcount == small->txnh_count); + + /* sum numbers of waiters threads */ + large->nr_waiters += small->nr_waiters; + small->nr_waiters = 0; + + /* splice flush queues */ + reiser4_fuse_fq(large, small); + + /* update counter of jnode on every atom' list */ + ON_DEBUG(large->dirty += small->dirty; + small->dirty = 0; + large->clean += small->clean; + small->clean = 0; + large->ovrwr += small->ovrwr; + small->ovrwr = 0; + large->wb += small->wb; + small->wb = 0; + large->fq += small->fq; + small->fq = 0;); + + /* count flushers in result atom */ + large->nr_flushers += small->nr_flushers; + small->nr_flushers = 0; + + /* update counts of flushed nodes */ + large->flushed += small->flushed; + small->flushed = 0; + + /* Transfer list counts to large. */ + large->txnh_count += small->txnh_count; + large->capture_count += small->capture_count; + + /* Add all txnh references to large. */ + atomic_add(small->txnh_count, &large->refcount); + atomic_sub(small->txnh_count, &small->refcount); + + /* Reset small counts */ + small->txnh_count = 0; + small->capture_count = 0; + + /* Assign the oldest start_time, merge flags. */ + large->start_time = min(large->start_time, small->start_time); + large->flags |= small->flags; + + /* Merge blocknr sets. */ + blocknr_set_merge(&small->delete_set, &large->delete_set); + blocknr_set_merge(&small->wandered_map, &large->wandered_map); + + /* Merge allocated/deleted file counts */ + large->nr_objects_deleted += small->nr_objects_deleted; + large->nr_objects_created += small->nr_objects_created; + + small->nr_objects_deleted = 0; + small->nr_objects_created = 0; + + /* Merge allocated blocks counts */ + large->nr_blocks_allocated += small->nr_blocks_allocated; + + large->nr_running_queues += small->nr_running_queues; + small->nr_running_queues = 0; + + /* Merge blocks reserved for overwrite set. */ + large->flush_reserved += small->flush_reserved; + small->flush_reserved = 0; + + if (large->stage < small->stage) { + /* Large only needs to notify if it has changed state. */ + reiser4_atom_set_stage(large, small->stage); + wakeup_atom_waiting_list(large); + } + + reiser4_atom_set_stage(small, ASTAGE_INVALID); + + /* Notify any waiters--small needs to unload its wait lists. Waiters + actually remove themselves from the list before returning from the + fuse_wait function. */ + wakeup_atom_waiting_list(small); + + /* Unlock atoms */ + spin_unlock_atom(large); + atom_dec_and_unlock(small); +} + +/* TXNMGR STUFF */ + +/* Release a block from the atom, reversing the effects of being captured, + do not release atom's reference to jnode due to holding spin-locks. + Currently this is only called when the atom commits. + + NOTE: this function does not release a (journal) reference to jnode + due to locking optimizations, you should call jput() somewhere after + calling reiser4_uncapture_block(). */ +void reiser4_uncapture_block(jnode * node) +{ + txn_atom *atom; + + assert("umka-226", node != NULL); + atom = node->atom; + assert("umka-228", atom != NULL); + + assert("jmacd-1021", node->atom == atom); + assert_spin_locked(&(node->guard)); + assert("jmacd-1023", atom_is_protected(atom)); + + JF_CLR(node, JNODE_DIRTY); + JF_CLR(node, JNODE_RELOC); + JF_CLR(node, JNODE_OVRWR); + JF_CLR(node, JNODE_CREATED); + JF_CLR(node, JNODE_WRITEBACK); + JF_CLR(node, JNODE_REPACK); + + list_del_init(&node->capture_link); + if (JF_ISSET(node, JNODE_FLUSH_QUEUED)) { + assert("zam-925", atom_isopen(atom)); + assert("vs-1623", NODE_LIST(node) == FQ_LIST); + ON_DEBUG(atom->num_queued--); + JF_CLR(node, JNODE_FLUSH_QUEUED); + } + atom->capture_count -= 1; + ON_DEBUG(count_jnode(atom, node, NODE_LIST(node), NOT_CAPTURED, 1)); + node->atom = NULL; + + spin_unlock_jnode(node); + LOCK_CNT_DEC(t_refs); +} + +/* Unconditional insert of jnode into atom's overwrite list. Currently used in + bitmap-based allocator code for adding modified bitmap blocks the + transaction. @atom and @node are spin locked */ +void insert_into_atom_ovrwr_list(txn_atom * atom, jnode * node) +{ + assert("zam-538", atom_is_protected(atom)); + assert_spin_locked(&(node->guard)); + assert("zam-899", JF_ISSET(node, JNODE_OVRWR)); + assert("zam-543", node->atom == NULL); + assert("vs-1433", !jnode_is_unformatted(node) && !jnode_is_znode(node)); + + list_add(&node->capture_link, ATOM_OVRWR_LIST(atom)); + jref(node); + node->atom = atom; + atom->capture_count++; + ON_DEBUG(count_jnode(atom, node, NODE_LIST(node), OVRWR_LIST, 1)); +} + +static int count_deleted_blocks_actor(txn_atom * atom, + const reiser4_block_nr * a, + const reiser4_block_nr * b, void *data) +{ + reiser4_block_nr *counter = data; + + assert("zam-995", data != NULL); + assert("zam-996", a != NULL); + if (b == NULL) + *counter += 1; + else + *counter += *b; + return 0; +} + +reiser4_block_nr txnmgr_count_deleted_blocks(void) +{ + reiser4_block_nr result; + txn_mgr *tmgr = &get_super_private(reiser4_get_current_sb())->tmgr; + txn_atom *atom; + + result = 0; + + spin_lock_txnmgr(tmgr); + list_for_each_entry(atom, &tmgr->atoms_list, atom_link) { + spin_lock_atom(atom); + if (atom_isopen(atom)) + blocknr_set_iterator( + atom, &atom->delete_set, + count_deleted_blocks_actor, &result, 0); + spin_unlock_atom(atom); + } + spin_unlock_txnmgr(tmgr); + + return result; +} + +/* + * Local variables: + * c-indentation-style: "K&R" + * mode-name: "LC" + * c-basic-offset: 8 + * tab-width: 8 + * fill-column: 79 + * End: + */ diff -urN linux-2.6.30.orig/fs/reiser4/txnmgr.h linux-2.6.30/fs/reiser4/txnmgr.h --- linux-2.6.30.orig/fs/reiser4/txnmgr.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/txnmgr.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,701 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* data-types and function declarations for transaction manager. See txnmgr.c + * for details. */ + +#ifndef __REISER4_TXNMGR_H__ +#define __REISER4_TXNMGR_H__ + +#include "forward.h" +#include "dformat.h" + +#include +#include +#include +#include +#include +#include + +/* TYPE DECLARATIONS */ + +/* This enumeration describes the possible types of a capture request (reiser4_try_capture). + A capture request dynamically assigns a block to the calling thread's transaction + handle. */ +typedef enum { + /* A READ_ATOMIC request indicates that a block will be read and that the caller's + atom should fuse in order to ensure that the block commits atomically with the + caller. */ + TXN_CAPTURE_READ_ATOMIC = (1 << 0), + + /* A READ_NONCOM request indicates that a block will be read and that the caller is + willing to read a non-committed block without causing atoms to fuse. */ + TXN_CAPTURE_READ_NONCOM = (1 << 1), + + /* A READ_MODIFY request indicates that a block will be read but that the caller + wishes for the block to be captured as it will be written. This capture request + mode is not currently used, but eventually it will be useful for preventing + deadlock in read-modify-write cycles. */ + TXN_CAPTURE_READ_MODIFY = (1 << 2), + + /* A WRITE capture request indicates that a block will be modified and that atoms + should fuse to make the commit atomic. */ + TXN_CAPTURE_WRITE = (1 << 3), + + /* CAPTURE_TYPES is a mask of the four above capture types, used to separate the + exclusive type designation from extra bits that may be supplied -- see + below. */ + TXN_CAPTURE_TYPES = (TXN_CAPTURE_READ_ATOMIC | + TXN_CAPTURE_READ_NONCOM | TXN_CAPTURE_READ_MODIFY | + TXN_CAPTURE_WRITE), + + /* A subset of CAPTURE_TYPES, CAPTURE_WTYPES is a mask of request types that + indicate modification will occur. */ + TXN_CAPTURE_WTYPES = (TXN_CAPTURE_READ_MODIFY | TXN_CAPTURE_WRITE), + + /* An option to reiser4_try_capture, NONBLOCKING indicates that the caller would + prefer not to sleep waiting for an aging atom to commit. */ + TXN_CAPTURE_NONBLOCKING = (1 << 4), + + /* An option to reiser4_try_capture to prevent atom fusion, just simple + capturing is allowed */ + TXN_CAPTURE_DONT_FUSE = (1 << 5) + + /* This macro selects only the exclusive capture request types, stripping out any + options that were supplied (i.e., NONBLOCKING). */ +#define CAPTURE_TYPE(x) ((x) & TXN_CAPTURE_TYPES) +} txn_capture; + +/* There are two kinds of transaction handle: WRITE_FUSING and READ_FUSING, the only + difference is in the handling of read requests. A WRITE_FUSING transaction handle + defaults read capture requests to TXN_CAPTURE_READ_NONCOM whereas a READ_FUSIONG + transaction handle defaults to TXN_CAPTURE_READ_ATOMIC. */ +typedef enum { + TXN_WRITE_FUSING = (1 << 0), + TXN_READ_FUSING = (1 << 1) | TXN_WRITE_FUSING, /* READ implies WRITE */ +} txn_mode; + +/* Every atom has a stage, which is one of these exclusive values: */ +typedef enum { + /* Initially an atom is free. */ + ASTAGE_FREE = 0, + + /* An atom begins by entering the CAPTURE_FUSE stage, where it proceeds to capture + blocks and fuse with other atoms. */ + ASTAGE_CAPTURE_FUSE = 1, + + /* We need to have a ASTAGE_CAPTURE_SLOW in which an atom fuses with one node for every X nodes it flushes to disk where X > 1. */ + + /* When an atom reaches a certain age it must do all it can to commit. An atom in + the CAPTURE_WAIT stage refuses new transaction handles and prevents fusion from + atoms in the CAPTURE_FUSE stage. */ + ASTAGE_CAPTURE_WAIT = 2, + + /* Waiting for I/O before commit. Copy-on-capture (see + http://namesys.com/v4/v4.html). */ + ASTAGE_PRE_COMMIT = 3, + + /* Post-commit overwrite I/O. Steal-on-capture. */ + ASTAGE_POST_COMMIT = 4, + + /* Atom which waits for the removal of the last reference to (it? ) to + * be deleted from memory */ + ASTAGE_DONE = 5, + + /* invalid atom. */ + ASTAGE_INVALID = 6, + +} txn_stage; + +/* Certain flags may be set in the txn_atom->flags field. */ +typedef enum { + /* Indicates that the atom should commit as soon as possible. */ + ATOM_FORCE_COMMIT = (1 << 0), + /* to avoid endless loop, mark the atom (which was considered as too + * small) after failed attempt to fuse it. */ + ATOM_CANCEL_FUSION = (1 << 1) +} txn_flags; + +/* Flags for controlling commit_txnh */ +typedef enum { + /* Wait commit atom completion in commit_txnh */ + TXNH_WAIT_COMMIT = 0x2, + /* Don't commit atom when this handle is closed */ + TXNH_DONT_COMMIT = 0x4 +} txn_handle_flags_t; + +/* TYPE DEFINITIONS */ + +/* A note on lock ordering: the handle & jnode spinlock protects reading of their ->atom + fields, so typically an operation on the atom through either of these objects must (1) + lock the object, (2) read the atom pointer, (3) lock the atom. + + During atom fusion, the process holds locks on both atoms at once. Then, it iterates + through the list of handles and pages held by the smaller of the two atoms. For each + handle and page referencing the smaller atom, the fusing process must: (1) lock the + object, and (2) update the atom pointer. + + You can see that there is a conflict of lock ordering here, so the more-complex + procedure should have priority, i.e., the fusing process has priority so that it is + guaranteed to make progress and to avoid restarts. + + This decision, however, means additional complexity for aquiring the atom lock in the + first place. + + The general original procedure followed in the code was: + + TXN_OBJECT *obj = ...; + TXN_ATOM *atom; + + spin_lock (& obj->_lock); + + atom = obj->_atom; + + if (! spin_trylock_atom (atom)) + { + spin_unlock (& obj->_lock); + RESTART OPERATION, THERE WAS A RACE; + } + + ELSE YOU HAVE BOTH ATOM AND OBJ LOCKED + + It has however been found that this wastes CPU a lot in a manner that is + hard to profile. So, proper refcounting was added to atoms, and new + standard locking sequence is like following: + + TXN_OBJECT *obj = ...; + TXN_ATOM *atom; + + spin_lock (& obj->_lock); + + atom = obj->_atom; + + if (! spin_trylock_atom (atom)) + { + atomic_inc (& atom->refcount); + spin_unlock (& obj->_lock); + spin_lock (&atom->_lock); + atomic_dec (& atom->refcount); + // HERE atom is locked + spin_unlock (&atom->_lock); + RESTART OPERATION, THERE WAS A RACE; + } + + ELSE YOU HAVE BOTH ATOM AND OBJ LOCKED + + (core of this is implemented in trylock_throttle() function) + + See the jnode_get_atom() function for a common case. + + As an additional (and important) optimization allowing to avoid restarts, + it is possible to re-check required pre-conditions at the HERE point in + code above and proceed without restarting if they are still satisfied. +*/ + +/* An atomic transaction: this is the underlying system representation + of a transaction, not the one seen by clients. + + Invariants involving this data-type: + + [sb-fake-allocated] +*/ +struct txn_atom { + /* The spinlock protecting the atom, held during fusion and various other state + changes. */ + spinlock_t alock; + + /* The atom's reference counter, increasing (in case of a duplication + of an existing reference or when we are sure that some other + reference exists) may be done without taking spinlock, decrementing + of the ref. counter requires a spinlock to be held. + + Each transaction handle counts in ->refcount. All jnodes count as + one reference acquired in atom_begin_andlock(), released in + commit_current_atom(). + */ + atomic_t refcount; + + /* The atom_id identifies the atom in persistent records such as the log. */ + __u32 atom_id; + + /* Flags holding any of the txn_flags enumerated values (e.g., + ATOM_FORCE_COMMIT). */ + __u32 flags; + + /* Number of open handles. */ + __u32 txnh_count; + + /* The number of znodes captured by this atom. Equal to the sum of lengths of the + dirty_nodes[level] and clean_nodes lists. */ + __u32 capture_count; + +#if REISER4_DEBUG + int clean; + int dirty; + int ovrwr; + int wb; + int fq; +#endif + + __u32 flushed; + + /* Current transaction stage. */ + txn_stage stage; + + /* Start time. */ + unsigned long start_time; + + /* The atom's delete set. It collects block numbers of the nodes + which were deleted during the transaction. */ + struct list_head delete_set; + + /* The atom's wandered_block mapping. */ + struct list_head wandered_map; + + /* The transaction's list of dirty captured nodes--per level. Index + by (level). dirty_nodes[0] is for znode-above-root */ + struct list_head dirty_nodes[REAL_MAX_ZTREE_HEIGHT + 1]; + + /* The transaction's list of clean captured nodes. */ + struct list_head clean_nodes; + + /* The atom's overwrite set */ + struct list_head ovrwr_nodes; + + /* nodes which are being written to disk */ + struct list_head writeback_nodes; + + /* list of inodes */ + struct list_head inodes; + + /* List of handles associated with this atom. */ + struct list_head txnh_list; + + /* Transaction list link: list of atoms in the transaction manager. */ + struct list_head atom_link; + + /* List of handles waiting FOR this atom: see 'capture_fuse_wait' comment. */ + struct list_head fwaitfor_list; + + /* List of this atom's handles that are waiting: see 'capture_fuse_wait' comment. */ + struct list_head fwaiting_list; + + /* Numbers of objects which were deleted/created in this transaction + thereby numbers of objects IDs which were released/deallocated. */ + int nr_objects_deleted; + int nr_objects_created; + /* number of blocks allocated during the transaction */ + __u64 nr_blocks_allocated; + /* All atom's flush queue objects are on this list */ + struct list_head flush_queues; +#if REISER4_DEBUG + /* number of flush queues for this atom. */ + int nr_flush_queues; + /* Number of jnodes which were removed from atom's lists and put + on flush_queue */ + int num_queued; +#endif + /* number of threads who wait for this atom to complete commit */ + int nr_waiters; + /* number of threads which do jnode_flush() over this atom */ + int nr_flushers; + /* number of flush queues which are IN_USE and jnodes from fq->prepped + are submitted to disk by the reiser4_write_fq() routine. */ + int nr_running_queues; + /* A counter of grabbed unformatted nodes, see a description of the + * reiser4 space reservation scheme at block_alloc.c */ + reiser4_block_nr flush_reserved; +#if REISER4_DEBUG + void *committer; +#endif + struct super_block *super; +}; + +#define ATOM_DIRTY_LIST(atom, level) (&(atom)->dirty_nodes[level]) +#define ATOM_CLEAN_LIST(atom) (&(atom)->clean_nodes) +#define ATOM_OVRWR_LIST(atom) (&(atom)->ovrwr_nodes) +#define ATOM_WB_LIST(atom) (&(atom)->writeback_nodes) +#define ATOM_FQ_LIST(fq) (&(fq)->prepped) + +#define NODE_LIST(node) (node)->list +#define ASSIGN_NODE_LIST(node, list) ON_DEBUG(NODE_LIST(node) = list) +ON_DEBUG(void + count_jnode(txn_atom *, jnode *, atom_list old_list, + atom_list new_list, int check_lists)); + +/* A transaction handle: the client obtains and commits this handle which is assigned by + the system to a txn_atom. */ +struct txn_handle { + /* Spinlock protecting ->atom pointer */ + spinlock_t hlock; + + /* Flags for controlling commit_txnh() behavior */ + /* from txn_handle_flags_t */ + txn_handle_flags_t flags; + + /* Whether it is READ_FUSING or WRITE_FUSING. */ + txn_mode mode; + + /* If assigned, the atom it is part of. */ + txn_atom *atom; + + /* Transaction list link. Head is in txn_atom. */ + struct list_head txnh_link; +}; + +/* The transaction manager: one is contained in the reiser4_super_info_data */ +struct txn_mgr { + /* A spinlock protecting the atom list, id_count, flush_control */ + spinlock_t tmgr_lock; + + /* List of atoms. */ + struct list_head atoms_list; + + /* Number of atoms. */ + int atom_count; + + /* A counter used to assign atom->atom_id values. */ + __u32 id_count; + + /* a mutex object for commit serialization */ + struct mutex commit_mutex; + + /* a list of all txnmrgs served by particular daemon. */ + struct list_head linkage; + + /* description of daemon for this txnmgr */ + ktxnmgrd_context *daemon; + + /* parameters. Adjustable through mount options. */ + unsigned int atom_max_size; + unsigned int atom_max_age; + unsigned int atom_min_size; + /* max number of concurrent flushers for one atom, 0 - unlimited. */ + unsigned int atom_max_flushers; + struct dentry *debugfs_atom_count; + struct dentry *debugfs_id_count; +}; + +/* FUNCTION DECLARATIONS */ + +/* These are the externally (within Reiser4) visible transaction functions, therefore they + are prefixed with "txn_". For comments, see txnmgr.c. */ + +extern int init_txnmgr_static(void); +extern void done_txnmgr_static(void); + +extern void reiser4_init_txnmgr(txn_mgr *); +extern void reiser4_done_txnmgr(txn_mgr *); + +extern int reiser4_txn_reserve(int reserved); + +extern void reiser4_txn_begin(reiser4_context * context); +extern int reiser4_txn_end(reiser4_context * context); + +extern void reiser4_txn_restart(reiser4_context * context); +extern void reiser4_txn_restart_current(void); + +extern int txnmgr_force_commit_all(struct super_block *, int); +extern int current_atom_should_commit(void); + +extern jnode *find_first_dirty_jnode(txn_atom *, int); + +extern int commit_some_atoms(txn_mgr *); +extern int force_commit_atom(txn_handle *); +extern int flush_current_atom(int, long, long *, txn_atom **, jnode *); + +extern int flush_some_atom(jnode *, long *, const struct writeback_control *, int); + +extern void reiser4_atom_set_stage(txn_atom * atom, txn_stage stage); + +extern int same_slum_check(jnode * base, jnode * check, int alloc_check, + int alloc_value); +extern void atom_dec_and_unlock(txn_atom * atom); + +extern int reiser4_try_capture(jnode * node, znode_lock_mode mode, txn_capture flags); +extern int try_capture_page_to_invalidate(struct page *pg); + +extern void reiser4_uncapture_page(struct page *pg); +extern void reiser4_uncapture_block(jnode *); +extern void reiser4_uncapture_jnode(jnode *); + +extern int reiser4_capture_inode(struct inode *); +extern int reiser4_uncapture_inode(struct inode *); + +extern txn_atom *get_current_atom_locked_nocheck(void); + +#if REISER4_DEBUG + +/** + * atom_is_protected - make sure that nobody but us can do anything with atom + * @atom: atom to be checked + * + * This is used to assert that atom either entered commit stages or is spin + * locked. + */ +static inline int atom_is_protected(txn_atom *atom) +{ + if (atom->stage >= ASTAGE_PRE_COMMIT) + return 1; + assert_spin_locked(&(atom->alock)); + return 1; +} + +#endif + +/* Get the current atom and spinlock it if current atom present. May not return NULL */ +static inline txn_atom *get_current_atom_locked(void) +{ + txn_atom *atom; + + atom = get_current_atom_locked_nocheck(); + assert("zam-761", atom != NULL); + + return atom; +} + +extern txn_atom *jnode_get_atom(jnode *); + +extern void reiser4_atom_wait_event(txn_atom *); +extern void reiser4_atom_send_event(txn_atom *); + +extern void insert_into_atom_ovrwr_list(txn_atom * atom, jnode * node); +extern int reiser4_capture_super_block(struct super_block *s); +int capture_bulk(jnode **, int count); + +/* See the comment on the function blocknrset.c:blocknr_set_add for the + calling convention of these three routines. */ +extern void blocknr_set_init(struct list_head * bset); +extern void blocknr_set_destroy(struct list_head * bset); +extern void blocknr_set_merge(struct list_head * from, struct list_head * into); +extern int blocknr_set_add_extent(txn_atom * atom, + struct list_head * bset, + blocknr_set_entry ** new_bsep, + const reiser4_block_nr * start, + const reiser4_block_nr * len); +extern int blocknr_set_add_pair(txn_atom * atom, struct list_head * bset, + blocknr_set_entry ** new_bsep, + const reiser4_block_nr * a, + const reiser4_block_nr * b); + +typedef int (*blocknr_set_actor_f) (txn_atom *, const reiser4_block_nr *, + const reiser4_block_nr *, void *); + +extern int blocknr_set_iterator(txn_atom * atom, struct list_head * bset, + blocknr_set_actor_f actor, void *data, + int delete); + +/* flush code takes care about how to fuse flush queues */ +extern void flush_init_atom(txn_atom * atom); +extern void flush_fuse_queues(txn_atom * large, txn_atom * small); + +static inline void spin_lock_atom(txn_atom *atom) +{ + /* check that spinlocks of lower priorities are not held */ + assert("", (LOCK_CNT_NIL(spin_locked_txnh) && + LOCK_CNT_NIL(spin_locked_atom) && + LOCK_CNT_NIL(spin_locked_jnode) && + LOCK_CNT_NIL(spin_locked_zlock) && + LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(rw_locked_tree))); + + spin_lock(&(atom->alock)); + + LOCK_CNT_INC(spin_locked_atom); + LOCK_CNT_INC(spin_locked); +} + +static inline void spin_lock_atom_nested(txn_atom *atom) +{ + assert("", (LOCK_CNT_NIL(spin_locked_txnh) && + LOCK_CNT_NIL(spin_locked_jnode) && + LOCK_CNT_NIL(spin_locked_zlock) && + LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(rw_locked_tree))); + + spin_lock_nested(&(atom->alock), SINGLE_DEPTH_NESTING); + + LOCK_CNT_INC(spin_locked_atom); + LOCK_CNT_INC(spin_locked); +} + +static inline int spin_trylock_atom(txn_atom *atom) +{ + if (spin_trylock(&(atom->alock))) { + LOCK_CNT_INC(spin_locked_atom); + LOCK_CNT_INC(spin_locked); + return 1; + } + return 0; +} + +static inline void spin_unlock_atom(txn_atom *atom) +{ + assert_spin_locked(&(atom->alock)); + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_atom)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(spin_locked_atom); + LOCK_CNT_DEC(spin_locked); + + spin_unlock(&(atom->alock)); +} + +static inline void spin_lock_txnh(txn_handle *txnh) +{ + /* check that spinlocks of lower priorities are not held */ + assert("", (LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(spin_locked_zlock) && + LOCK_CNT_NIL(rw_locked_tree))); + + spin_lock(&(txnh->hlock)); + + LOCK_CNT_INC(spin_locked_txnh); + LOCK_CNT_INC(spin_locked); +} + +static inline int spin_trylock_txnh(txn_handle *txnh) +{ + if (spin_trylock(&(txnh->hlock))) { + LOCK_CNT_INC(spin_locked_txnh); + LOCK_CNT_INC(spin_locked); + return 1; + } + return 0; +} + +static inline void spin_unlock_txnh(txn_handle *txnh) +{ + assert_spin_locked(&(txnh->hlock)); + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_txnh)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(spin_locked_txnh); + LOCK_CNT_DEC(spin_locked); + + spin_unlock(&(txnh->hlock)); +} + +#define spin_ordering_pred_txnmgr(tmgr) \ + ( LOCK_CNT_NIL(spin_locked_atom) && \ + LOCK_CNT_NIL(spin_locked_txnh) && \ + LOCK_CNT_NIL(spin_locked_jnode) && \ + LOCK_CNT_NIL(rw_locked_zlock) && \ + LOCK_CNT_NIL(rw_locked_dk) && \ + LOCK_CNT_NIL(rw_locked_tree) ) + +static inline void spin_lock_txnmgr(txn_mgr *mgr) +{ + /* check that spinlocks of lower priorities are not held */ + assert("", (LOCK_CNT_NIL(spin_locked_atom) && + LOCK_CNT_NIL(spin_locked_txnh) && + LOCK_CNT_NIL(spin_locked_jnode) && + LOCK_CNT_NIL(spin_locked_zlock) && + LOCK_CNT_NIL(rw_locked_dk) && + LOCK_CNT_NIL(rw_locked_tree))); + + spin_lock(&(mgr->tmgr_lock)); + + LOCK_CNT_INC(spin_locked_txnmgr); + LOCK_CNT_INC(spin_locked); +} + +static inline int spin_trylock_txnmgr(txn_mgr *mgr) +{ + if (spin_trylock(&(mgr->tmgr_lock))) { + LOCK_CNT_INC(spin_locked_txnmgr); + LOCK_CNT_INC(spin_locked); + return 1; + } + return 0; +} + +static inline void spin_unlock_txnmgr(txn_mgr *mgr) +{ + assert_spin_locked(&(mgr->tmgr_lock)); + assert("nikita-1375", LOCK_CNT_GTZ(spin_locked_txnmgr)); + assert("nikita-1376", LOCK_CNT_GTZ(spin_locked)); + + LOCK_CNT_DEC(spin_locked_txnmgr); + LOCK_CNT_DEC(spin_locked); + + spin_unlock(&(mgr->tmgr_lock)); +} + +typedef enum { + FQ_IN_USE = 0x1 +} flush_queue_state_t; + +typedef struct flush_queue flush_queue_t; + +/* This is an accumulator for jnodes prepared for writing to disk. A flush queue + is filled by the jnode_flush() routine, and written to disk under memory + pressure or at atom commit time. */ +/* LOCKING: fq state and fq->atom are protected by guard spinlock, fq->nr_queued + field and fq->prepped list can be modified if atom is spin-locked and fq + object is "in-use" state. For read-only traversal of the fq->prepped list + and reading of the fq->nr_queued field it is enough to keep fq "in-use" or + only have atom spin-locked. */ +struct flush_queue { + /* linkage element is the first in this structure to make debugging + easier. See field in atom struct for description of list. */ + struct list_head alink; + /* A spinlock to protect changes of fq state and fq->atom pointer */ + spinlock_t guard; + /* flush_queue state: [in_use | ready] */ + flush_queue_state_t state; + /* A list which contains queued nodes, queued nodes are removed from any + * atom's list and put on this ->prepped one. */ + struct list_head prepped; + /* number of submitted i/o requests */ + atomic_t nr_submitted; + /* number of i/o errors */ + atomic_t nr_errors; + /* An atom this flush queue is attached to */ + txn_atom *atom; + /* A wait queue head to wait on i/o completion */ + wait_queue_head_t wait; +#if REISER4_DEBUG + /* A thread which took this fq in exclusive use, NULL if fq is free, + * used for debugging. */ + struct task_struct *owner; +#endif +}; + +extern int reiser4_fq_by_atom(txn_atom *, flush_queue_t **); +extern void reiser4_fq_put_nolock(flush_queue_t *); +extern void reiser4_fq_put(flush_queue_t *); +extern void reiser4_fuse_fq(txn_atom * to, txn_atom * from); +extern void queue_jnode(flush_queue_t *, jnode *); + +extern int reiser4_write_fq(flush_queue_t *, long *, int); +extern int current_atom_finish_all_fq(void); +extern void init_atom_fq_parts(txn_atom *); + +extern reiser4_block_nr txnmgr_count_deleted_blocks(void); + +extern void znode_make_dirty(znode * node); +extern void jnode_make_dirty_locked(jnode * node); + +extern int reiser4_sync_atom(txn_atom * atom); + +#if REISER4_DEBUG +extern int atom_fq_parts_are_clean(txn_atom *); +#endif + +extern void add_fq_to_bio(flush_queue_t *, struct bio *); +extern flush_queue_t *get_fq_for_current_atom(void); + +void reiser4_invalidate_list(struct list_head * head); + +# endif /* __REISER4_TXNMGR_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/type_safe_hash.h linux-2.6.30/fs/reiser4/type_safe_hash.h --- linux-2.6.30.orig/fs/reiser4/type_safe_hash.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/type_safe_hash.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,320 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* A hash table class that uses hash chains (singly-linked) and is + parametrized to provide type safety. */ + +#ifndef __REISER4_TYPE_SAFE_HASH_H__ +#define __REISER4_TYPE_SAFE_HASH_H__ + +#include "debug.h" + +#include +/* Step 1: Use TYPE_SAFE_HASH_DECLARE() to define the TABLE and LINK objects + based on the object type. You need to declare the item type before + this definition, define it after this definition. */ +#define TYPE_SAFE_HASH_DECLARE(PREFIX,ITEM_TYPE) \ + \ +typedef struct PREFIX##_hash_table_ PREFIX##_hash_table; \ +typedef struct PREFIX##_hash_link_ PREFIX##_hash_link; \ + \ +struct PREFIX##_hash_table_ \ +{ \ + ITEM_TYPE **_table; \ + __u32 _buckets; \ +}; \ + \ +struct PREFIX##_hash_link_ \ +{ \ + ITEM_TYPE *_next; \ +} + +/* Step 2: Define the object type of the hash: give it field of type + PREFIX_hash_link. */ + +/* Step 3: Use TYPE_SAFE_HASH_DEFINE to define the hash table interface using + the type and field name used in step 3. The arguments are: + + ITEM_TYPE The item type being hashed + KEY_TYPE The type of key being hashed + KEY_NAME The name of the key field within the item + LINK_NAME The name of the link field within the item, which you must make type PREFIX_hash_link) + HASH_FUNC The name of the hash function (or macro, takes const pointer to key) + EQ_FUNC The name of the equality function (or macro, takes const pointer to two keys) + + It implements these functions: + + prefix_hash_init Initialize the table given its size. + prefix_hash_insert Insert an item + prefix_hash_insert_index Insert an item w/ precomputed hash_index + prefix_hash_find Find an item by key + prefix_hash_find_index Find an item w/ precomputed hash_index + prefix_hash_remove Remove an item, returns 1 if found, 0 if not found + prefix_hash_remove_index Remove an item w/ precomputed hash_index + + If you'd like something to be done differently, feel free to ask me + for modifications. Additional features that could be added but + have not been: + + prefix_hash_remove_key Find and remove an item by key + prefix_hash_remove_key_index Find and remove an item by key w/ precomputed hash_index + + The hash_function currently receives only the key as an argument, + meaning it must somehow know the number of buckets. If this is a + problem let me know. + + This hash table uses a single-linked hash chain. This means + insertion is fast but deletion requires searching the chain. + + There is also the doubly-linked hash chain approach, under which + deletion requires no search but the code is longer and it takes two + pointers per item. + + The circularly-linked approach has the shortest code but requires + two pointers per bucket, doubling the size of the bucket array (in + addition to two pointers per item). +*/ +#define TYPE_SAFE_HASH_DEFINE(PREFIX,ITEM_TYPE,KEY_TYPE,KEY_NAME,LINK_NAME,HASH_FUNC,EQ_FUNC) \ + \ +static __inline__ void \ +PREFIX##_check_hash (PREFIX##_hash_table *table UNUSED_ARG, \ + __u32 hash UNUSED_ARG) \ +{ \ + assert("nikita-2780", hash < table->_buckets); \ +} \ + \ +static __inline__ int \ +PREFIX##_hash_init (PREFIX##_hash_table *hash, \ + __u32 buckets) \ +{ \ + hash->_table = (ITEM_TYPE**) KMALLOC (sizeof (ITEM_TYPE*) * buckets); \ + hash->_buckets = buckets; \ + if (hash->_table == NULL) \ + { \ + return RETERR(-ENOMEM); \ + } \ + memset (hash->_table, 0, sizeof (ITEM_TYPE*) * buckets); \ + ON_DEBUG(printk(#PREFIX "_hash_table: %i buckets\n", buckets)); \ + return 0; \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_done (PREFIX##_hash_table *hash) \ +{ \ + if (REISER4_DEBUG && hash->_table != NULL) { \ + __u32 i; \ + for (i = 0 ; i < hash->_buckets ; ++ i) \ + assert("nikita-2905", hash->_table[i] == NULL); \ + } \ + if (hash->_table != NULL) \ + KFREE (hash->_table, sizeof (ITEM_TYPE*) * hash->_buckets); \ + hash->_table = NULL; \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_prefetch_next (ITEM_TYPE *item) \ +{ \ + prefetch(item->LINK_NAME._next); \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_prefetch_bucket (PREFIX##_hash_table *hash, \ + __u32 index) \ +{ \ + prefetch(hash->_table[index]); \ +} \ + \ +static __inline__ ITEM_TYPE* \ +PREFIX##_hash_find_index (PREFIX##_hash_table *hash, \ + __u32 hash_index, \ + KEY_TYPE const *find_key) \ +{ \ + ITEM_TYPE *item; \ + \ + PREFIX##_check_hash(hash, hash_index); \ + \ + for (item = hash->_table[hash_index]; \ + item != NULL; \ + item = item->LINK_NAME._next) \ + { \ + prefetch(item->LINK_NAME._next); \ + prefetch(item->LINK_NAME._next + offsetof(ITEM_TYPE, KEY_NAME)); \ + if (EQ_FUNC (& item->KEY_NAME, find_key)) \ + { \ + return item; \ + } \ + } \ + \ + return NULL; \ +} \ + \ +static __inline__ ITEM_TYPE* \ +PREFIX##_hash_find_index_lru (PREFIX##_hash_table *hash, \ + __u32 hash_index, \ + KEY_TYPE const *find_key) \ +{ \ + ITEM_TYPE ** item = &hash->_table[hash_index]; \ + \ + PREFIX##_check_hash(hash, hash_index); \ + \ + while (*item != NULL) { \ + prefetch(&(*item)->LINK_NAME._next); \ + if (EQ_FUNC (&(*item)->KEY_NAME, find_key)) { \ + ITEM_TYPE *found; \ + \ + found = *item; \ + *item = found->LINK_NAME._next; \ + found->LINK_NAME._next = hash->_table[hash_index]; \ + hash->_table[hash_index] = found; \ + return found; \ + } \ + item = &(*item)->LINK_NAME._next; \ + } \ + return NULL; \ +} \ + \ +static __inline__ int \ +PREFIX##_hash_remove_index (PREFIX##_hash_table *hash, \ + __u32 hash_index, \ + ITEM_TYPE *del_item) \ +{ \ + ITEM_TYPE ** hash_item_p = &hash->_table[hash_index]; \ + \ + PREFIX##_check_hash(hash, hash_index); \ + \ + while (*hash_item_p != NULL) { \ + prefetch(&(*hash_item_p)->LINK_NAME._next); \ + if (*hash_item_p == del_item) { \ + *hash_item_p = (*hash_item_p)->LINK_NAME._next; \ + return 1; \ + } \ + hash_item_p = &(*hash_item_p)->LINK_NAME._next; \ + } \ + return 0; \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_insert_index (PREFIX##_hash_table *hash, \ + __u32 hash_index, \ + ITEM_TYPE *ins_item) \ +{ \ + PREFIX##_check_hash(hash, hash_index); \ + \ + ins_item->LINK_NAME._next = hash->_table[hash_index]; \ + hash->_table[hash_index] = ins_item; \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_insert_index_rcu (PREFIX##_hash_table *hash, \ + __u32 hash_index, \ + ITEM_TYPE *ins_item) \ +{ \ + PREFIX##_check_hash(hash, hash_index); \ + \ + ins_item->LINK_NAME._next = hash->_table[hash_index]; \ + smp_wmb(); \ + hash->_table[hash_index] = ins_item; \ +} \ + \ +static __inline__ ITEM_TYPE* \ +PREFIX##_hash_find (PREFIX##_hash_table *hash, \ + KEY_TYPE const *find_key) \ +{ \ + return PREFIX##_hash_find_index (hash, HASH_FUNC(hash, find_key), find_key); \ +} \ + \ +static __inline__ ITEM_TYPE* \ +PREFIX##_hash_find_lru (PREFIX##_hash_table *hash, \ + KEY_TYPE const *find_key) \ +{ \ + return PREFIX##_hash_find_index_lru (hash, HASH_FUNC(hash, find_key), find_key); \ +} \ + \ +static __inline__ int \ +PREFIX##_hash_remove (PREFIX##_hash_table *hash, \ + ITEM_TYPE *del_item) \ +{ \ + return PREFIX##_hash_remove_index (hash, \ + HASH_FUNC(hash, &del_item->KEY_NAME), del_item); \ +} \ + \ +static __inline__ int \ +PREFIX##_hash_remove_rcu (PREFIX##_hash_table *hash, \ + ITEM_TYPE *del_item) \ +{ \ + return PREFIX##_hash_remove (hash, del_item); \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_insert (PREFIX##_hash_table *hash, \ + ITEM_TYPE *ins_item) \ +{ \ + return PREFIX##_hash_insert_index (hash, \ + HASH_FUNC(hash, &ins_item->KEY_NAME), ins_item); \ +} \ + \ +static __inline__ void \ +PREFIX##_hash_insert_rcu (PREFIX##_hash_table *hash, \ + ITEM_TYPE *ins_item) \ +{ \ + return PREFIX##_hash_insert_index_rcu (hash, HASH_FUNC(hash, &ins_item->KEY_NAME), \ + ins_item); \ +} \ + \ +static __inline__ ITEM_TYPE * \ +PREFIX##_hash_first (PREFIX##_hash_table *hash, __u32 ind) \ +{ \ + ITEM_TYPE *first; \ + \ + for (first = NULL; ind < hash->_buckets; ++ ind) { \ + first = hash->_table[ind]; \ + if (first != NULL) \ + break; \ + } \ + return first; \ +} \ + \ +static __inline__ ITEM_TYPE * \ +PREFIX##_hash_next (PREFIX##_hash_table *hash, \ + ITEM_TYPE *item) \ +{ \ + ITEM_TYPE *next; \ + \ + if (item == NULL) \ + return NULL; \ + next = item->LINK_NAME._next; \ + if (next == NULL) \ + next = PREFIX##_hash_first (hash, HASH_FUNC(hash, &item->KEY_NAME) + 1); \ + return next; \ +} \ + \ +typedef struct {} PREFIX##_hash_dummy + +#define for_all_ht_buckets(table, head) \ +for ((head) = &(table) -> _table[ 0 ] ; \ + (head) != &(table) -> _table[ (table) -> _buckets ] ; ++ (head)) + +#define for_all_in_bucket(bucket, item, next, field) \ +for ((item) = *(bucket), (next) = (item) ? (item) -> field._next : NULL ; \ + (item) != NULL ; \ + (item) = (next), (next) = (item) ? (item) -> field._next : NULL ) + +#define for_all_in_htable(table, prefix, item, next) \ +for ((item) = prefix ## _hash_first ((table), 0), \ + (next) = prefix ## _hash_next ((table), (item)) ; \ + (item) != NULL ; \ + (item) = (next), \ + (next) = prefix ## _hash_next ((table), (item))) + +/* __REISER4_TYPE_SAFE_HASH_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/vfs_ops.c linux-2.6.30/fs/reiser4/vfs_ops.c --- linux-2.6.30.orig/fs/reiser4/vfs_ops.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/vfs_ops.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,259 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Interface to VFS. Reiser4 {super|export|dentry}_operations are defined + here. */ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "coord.h" +#include "plugin/item/item.h" +#include "plugin/file/file.h" +#include "plugin/security/perm.h" +#include "plugin/disk_format/disk_format.h" +#include "plugin/plugin.h" +#include "plugin/plugin_set.h" +#include "plugin/object.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree.h" +#include "vfs_ops.h" +#include "inode.h" +#include "page_cache.h" +#include "ktxnmgrd.h" +#include "super.h" +#include "reiser4.h" +#include "entd.h" +#include "status_flags.h" +#include "flush.h" +#include "dscale.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* update inode stat-data by calling plugin */ +int reiser4_update_sd(struct inode *object) +{ + file_plugin *fplug; + + assert("nikita-2338", object != NULL); + /* check for read-only file system. */ + if (IS_RDONLY(object)) + return 0; + + fplug = inode_file_plugin(object); + assert("nikita-2339", fplug != NULL); + return fplug->write_sd_by_inode(object); +} + +/* helper function: increase inode nlink count and call plugin method to save + updated stat-data. + + Used by link/create and during creation of dot and dotdot in mkdir +*/ +int reiser4_add_nlink(struct inode *object /* object to which link is added */ , + struct inode *parent /* parent where new entry will be */ + , + int write_sd_p /* true if stat-data has to be + * updated */ ) +{ + file_plugin *fplug; + int result; + + assert("nikita-1351", object != NULL); + + fplug = inode_file_plugin(object); + assert("nikita-1445", fplug != NULL); + + /* ask plugin whether it can add yet another link to this + object */ + if (!fplug->can_add_link(object)) + return RETERR(-EMLINK); + + assert("nikita-2211", fplug->add_link != NULL); + /* call plugin to do actual addition of link */ + result = fplug->add_link(object, parent); + + /* optionally update stat data */ + if (result == 0 && write_sd_p) + result = fplug->write_sd_by_inode(object); + return result; +} + +/* helper function: decrease inode nlink count and call plugin method to save + updated stat-data. + + Used by unlink/create +*/ +int reiser4_del_nlink(struct inode *object /* object from which link is + * removed */ , + struct inode *parent /* parent where entry was */ , + int write_sd_p /* true is stat-data has to be + * updated */ ) +{ + file_plugin *fplug; + int result; + + assert("nikita-1349", object != NULL); + + fplug = inode_file_plugin(object); + assert("nikita-1350", fplug != NULL); + assert("nikita-1446", object->i_nlink > 0); + assert("nikita-2210", fplug->rem_link != NULL); + + /* call plugin to do actual deletion of link */ + result = fplug->rem_link(object, parent); + + /* optionally update stat data */ + if (result == 0 && write_sd_p) + result = fplug->write_sd_by_inode(object); + return result; +} + +/* Release reiser4 dentry. This is d_op->d_release() method. */ +static void reiser4_d_release(struct dentry *dentry /* dentry released */ ) +{ + reiser4_free_dentry_fsdata(dentry); +} + +/* + * Called by reiser4_sync_inodes(), during speculative write-back (through + * pdflush, or balance_dirty_pages()). + */ +void reiser4_writeout(struct super_block *sb, struct writeback_control *wbc) +{ + long written = 0; + int repeats = 0; + int result; + struct address_space *mapping; + + /* + * Performs early flushing, trying to free some memory. If there is + * nothing to flush, commits some atoms. + */ + + /* Commit all atoms if reiser4_writepages() is called from sys_sync() or + sys_fsync(). */ + if (wbc->sync_mode != WB_SYNC_NONE) { + txnmgr_force_commit_all(sb, 0); + return; + } + + BUG_ON(reiser4_get_super_fake(sb) == NULL); + mapping = reiser4_get_super_fake(sb)->i_mapping; + do { + long nr_submitted = 0; + jnode *node = NULL; + + /* do not put more requests to overload write queue */ + if (wbc->nonblocking && + bdi_write_congested(mapping->backing_dev_info)) { + blk_run_address_space(mapping); + wbc->encountered_congestion = 1; + break; + } + repeats++; + BUG_ON(wbc->nr_to_write <= 0); + + if (get_current_context()->entd) { + entd_context *ent = get_entd_context(sb); + + if (ent->cur_request->node) + /* + * this is ent thread and it managed to capture + * requested page itself - start flush from + * that page + */ + node = jref(ent->cur_request->node); + } + + result = flush_some_atom(node, &nr_submitted, wbc, + JNODE_FLUSH_WRITE_BLOCKS); + if (result != 0) + warning("nikita-31001", "Flush failed: %i", result); + if (node) + jput(node); + if (!nr_submitted) + break; + + wbc->nr_to_write -= nr_submitted; + written += nr_submitted; + } while (wbc->nr_to_write > 0); +} + +void reiser4_throttle_write(struct inode *inode) +{ + reiser4_txn_restart_current(); + balance_dirty_pages_ratelimited(inode->i_mapping); +} + +const char *REISER4_SUPER_MAGIC_STRING = "ReIsEr4"; +const int REISER4_MAGIC_OFFSET = 16 * 4096; /* offset to magic string from the + * beginning of device */ + +/* + * Reiser4 initialization/shutdown. + * + * Code below performs global reiser4 initialization that is done either as + * part of kernel initialization (when reiser4 is statically built-in), or + * during reiser4 module load (when compiled as module). + */ + +void reiser4_handle_error(void) +{ + struct super_block *sb = reiser4_get_current_sb(); + + if (!sb) + return; + reiser4_status_write(REISER4_STATUS_DAMAGED, 0, + "Filesystem error occured"); + switch (get_super_private(sb)->onerror) { + case 0: + reiser4_panic("foobar-42", "Filesystem error occured\n"); + case 1: + default: + if (sb->s_flags & MS_RDONLY) + return; + sb->s_flags |= MS_RDONLY; + break; + } +} + +struct dentry_operations reiser4_dentry_operations = { + .d_revalidate = NULL, + .d_hash = NULL, + .d_compare = NULL, + .d_delete = NULL, + .d_release = reiser4_d_release, + .d_iput = NULL, +}; + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/vfs_ops.h linux-2.6.30/fs/reiser4/vfs_ops.h --- linux-2.6.30.orig/fs/reiser4/vfs_ops.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/vfs_ops.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,53 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* vfs_ops.c's exported symbols */ + +#if !defined( __FS_REISER4_VFS_OPS_H__ ) +#define __FS_REISER4_VFS_OPS_H__ + +#include "forward.h" +#include "coord.h" +#include "seal.h" +#include "plugin/file/file.h" +#include "super.h" +#include "readahead.h" + +#include /* for loff_t */ +#include /* for struct address_space */ +#include /* for struct dentry */ +#include +#include + +/* address space operations */ +int reiser4_writepage(struct page *, struct writeback_control *); +int reiser4_set_page_dirty(struct page *); +void reiser4_invalidatepage(struct page *, unsigned long offset); +int reiser4_releasepage(struct page *, gfp_t); + +extern int reiser4_update_sd(struct inode *); +extern int reiser4_add_nlink(struct inode *, struct inode *, int); +extern int reiser4_del_nlink(struct inode *, struct inode *, int); + +extern int reiser4_start_up_io(struct page *page); +extern void reiser4_throttle_write(struct inode *); +extern int jnode_is_releasable(jnode *); + +#define CAPTURE_APAGE_BURST (1024l) +void reiser4_writeout(struct super_block *, struct writeback_control *); + +extern void reiser4_handle_error(void); + +/* __FS_REISER4_VFS_OPS_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/wander.c linux-2.6.30/fs/reiser4/wander.c --- linux-2.6.30.orig/fs/reiser4/wander.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/wander.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1798 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Reiser4 Wandering Log */ + +/* You should read http://www.namesys.com/txn-doc.html + + That describes how filesystem operations are performed as atomic + transactions, and how we try to arrange it so that we can write most of the + data only once while performing the operation atomically. + + For the purposes of this code, it is enough for it to understand that it + has been told a given block should be written either once, or twice (if + twice then once to the wandered location and once to the real location). + + This code guarantees that those blocks that are defined to be part of an + atom either all take effect or none of them take effect. + + The "relocate set" of nodes are submitted to write by the jnode_flush() + routine, and the "overwrite set" is submitted by reiser4_write_log(). + This is because with the overwrite set we seek to optimize writes, and + with the relocate set we seek to cause disk order to correlate with the + "parent first order" (preorder). + + reiser4_write_log() allocates and writes wandered blocks and maintains + additional on-disk structures of the atom as wander records (each wander + record occupies one block) for storing of the "wandered map" (a table which + contains a relation between wandered and real block numbers) and other + information which might be needed at transaction recovery time. + + The wander records are unidirectionally linked into a circle: each wander + record contains a block number of the next wander record, the last wander + record points to the first one. + + One wander record (named "tx head" in this file) has a format which is + different from the other wander records. The "tx head" has a reference to the + "tx head" block of the previously committed atom. Also, "tx head" contains + fs information (the free blocks counter, and the oid allocator state) which + is logged in a special way . + + There are two journal control blocks, named journal header and journal + footer which have fixed on-disk locations. The journal header has a + reference to the "tx head" block of the last committed atom. The journal + footer points to the "tx head" of the last flushed atom. The atom is + "played" when all blocks from its overwrite set are written to disk the + second time (i.e. written to their real locations). + + NOTE: People who know reiserfs internals and its journal structure might be + confused with these terms journal footer and journal header. There is a table + with terms of similar semantics in reiserfs (reiser3) and reiser4: + + REISER3 TERM | REISER4 TERM | DESCRIPTION + --------------------+-----------------------+---------------------------- + commit record | journal header | atomic write of this record + | | ends transaction commit + --------------------+-----------------------+---------------------------- + journal header | journal footer | atomic write of this record + | | ends post-commit writes. + | | After successful + | | writing of this journal + | | blocks (in reiser3) or + | | wandered blocks/records are + | | free for re-use. + --------------------+-----------------------+---------------------------- + + The atom commit process is the following: + + 1. The overwrite set is taken from atom's clean list, and its size is + counted. + + 2. The number of necessary wander records (including tx head) is calculated, + and the wander record blocks are allocated. + + 3. Allocate wandered blocks and populate wander records by wandered map. + + 4. submit write requests for wander records and wandered blocks. + + 5. wait until submitted write requests complete. + + 6. update journal header: change the pointer to the block number of just + written tx head, submit an i/o for modified journal header block and wait + for i/o completion. + + NOTE: The special logging for bitmap blocks and some reiser4 super block + fields makes processes of atom commit, flush and recovering a bit more + complex (see comments in the source code for details). + + The atom playing process is the following: + + 1. Write atom's overwrite set in-place. + + 2. Wait on i/o. + + 3. Update journal footer: change the pointer to block number of tx head + block of the atom we currently flushing, submit an i/o, wait on i/o + completion. + + 4. Free disk space which was used for wandered blocks and wander records. + + After the freeing of wandered blocks and wander records we have that journal + footer points to the on-disk structure which might be overwritten soon. + Neither the log writer nor the journal recovery procedure use that pointer + for accessing the data. When the journal recovery procedure finds the oldest + transaction it compares the journal footer pointer value with the "prev_tx" + pointer value in tx head, if values are equal the oldest not flushed + transaction is found. + + NOTE on disk space leakage: the information about of what blocks and how many + blocks are allocated for wandered blocks, wandered records is not written to + the disk because of special logging for bitmaps and some super blocks + counters. After a system crash we the reiser4 does not remember those + objects allocation, thus we have no such a kind of disk space leakage. +*/ + +/* Special logging of reiser4 super block fields. */ + +/* There are some reiser4 super block fields (free block count and OID allocator + state (number of files and next free OID) which are logged separately from + super block to avoid unnecessary atom fusion. + + So, the reiser4 super block can be not captured by a transaction with + allocates/deallocates disk blocks or create/delete file objects. Moreover, + the reiser4 on-disk super block is not touched when such a transaction is + committed and flushed. Those "counters logged specially" are logged in "tx + head" blocks and in the journal footer block. + + A step-by-step description of special logging: + + 0. The per-atom information about deleted or created files and allocated or + freed blocks is collected during the transaction. The atom's + ->nr_objects_created and ->nr_objects_deleted are for object + deletion/creation tracking, the numbers of allocated and freed blocks are + calculated using atom's delete set and atom's capture list -- all new and + relocated nodes should be on atom's clean list and should have JNODE_RELOC + bit set. + + 1. The "logged specially" reiser4 super block fields have their "committed" + versions in the reiser4 in-memory super block. They get modified only at + atom commit time. The atom's commit thread has an exclusive access to those + "committed" fields because the log writer implementation supports only one + atom commit a time (there is a per-fs "commit" mutex). At + that time "committed" counters are modified using per-atom information + collected during the transaction. These counters are stored on disk as a + part of tx head block when atom is committed. + + 2. When the atom is flushed the value of the free block counter and the OID + allocator state get written to the journal footer block. A special journal + procedure (journal_recover_sb_data()) takes those values from the journal + footer and updates the reiser4 in-memory super block. + + NOTE: That means free block count and OID allocator state are logged + separately from the reiser4 super block regardless of the fact that the + reiser4 super block has fields to store both the free block counter and the + OID allocator. + + Writing the whole super block at commit time requires knowing true values of + all its fields without changes made by not yet committed transactions. It is + possible by having their "committed" version of the super block like the + reiser4 bitmap blocks have "committed" and "working" versions. However, + another scheme was implemented which stores special logged values in the + unused free space inside transaction head block. In my opinion it has an + advantage of not writing whole super block when only part of it was + modified. */ + +#include "debug.h" +#include "dformat.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "page_cache.h" +#include "wander.h" +#include "reiser4.h" +#include "super.h" +#include "vfs_ops.h" +#include "writeout.h" +#include "inode.h" +#include "entd.h" + +#include +#include /* for struct super_block */ +#include /* for struct page */ +#include +#include /* for struct bio */ +#include + +static int write_jnodes_to_disk_extent( + jnode *, int, const reiser4_block_nr *, flush_queue_t *, int); + +/* The commit_handle is a container for objects needed at atom commit time */ +struct commit_handle { + /* A pointer to atom's list of OVRWR nodes */ + struct list_head *overwrite_set; + /* atom's overwrite set size */ + int overwrite_set_size; + /* jnodes for wander record blocks */ + struct list_head tx_list; + /* number of wander records */ + __u32 tx_size; + /* 'committed' sb counters are saved here until atom is completely + flushed */ + __u64 free_blocks; + __u64 nr_files; + __u64 next_oid; + /* A pointer to the atom which is being committed */ + txn_atom *atom; + /* A pointer to current super block */ + struct super_block *super; + /* The counter of modified bitmaps */ + reiser4_block_nr nr_bitmap; +}; + +static void init_commit_handle(struct commit_handle *ch, txn_atom *atom) +{ + memset(ch, 0, sizeof(struct commit_handle)); + INIT_LIST_HEAD(&ch->tx_list); + + ch->atom = atom; + ch->super = reiser4_get_current_sb(); +} + +static void done_commit_handle(struct commit_handle *ch) +{ + assert("zam-690", list_empty(&ch->tx_list)); +} + +static inline int reiser4_use_write_barrier(struct super_block * s) +{ + return !reiser4_is_set(s, REISER4_NO_WRITE_BARRIER); +} + +static void disable_write_barrier(struct super_block * s) +{ + notice("zam-1055", "%s does not support write barriers," + " using synchronous write instead.", s->s_id); + set_bit((int)REISER4_NO_WRITE_BARRIER, &get_super_private(s)->fs_flags); +} + +/* fill journal header block data */ +static void format_journal_header(struct commit_handle *ch) +{ + struct reiser4_super_info_data *sbinfo; + struct journal_header *header; + jnode *txhead; + + sbinfo = get_super_private(ch->super); + assert("zam-479", sbinfo != NULL); + assert("zam-480", sbinfo->journal_header != NULL); + + txhead = list_entry(ch->tx_list.next, jnode, capture_link); + + jload(sbinfo->journal_header); + + header = (struct journal_header *)jdata(sbinfo->journal_header); + assert("zam-484", header != NULL); + + put_unaligned(cpu_to_le64(*jnode_get_block(txhead)), + &header->last_committed_tx); + + jrelse(sbinfo->journal_header); +} + +/* fill journal footer block data */ +static void format_journal_footer(struct commit_handle *ch) +{ + struct reiser4_super_info_data *sbinfo; + struct journal_footer *footer; + jnode *tx_head; + + sbinfo = get_super_private(ch->super); + + tx_head = list_entry(ch->tx_list.next, jnode, capture_link); + + assert("zam-493", sbinfo != NULL); + assert("zam-494", sbinfo->journal_header != NULL); + + check_me("zam-691", jload(sbinfo->journal_footer) == 0); + + footer = (struct journal_footer *)jdata(sbinfo->journal_footer); + assert("zam-495", footer != NULL); + + put_unaligned(cpu_to_le64(*jnode_get_block(tx_head)), + &footer->last_flushed_tx); + put_unaligned(cpu_to_le64(ch->free_blocks), &footer->free_blocks); + + put_unaligned(cpu_to_le64(ch->nr_files), &footer->nr_files); + put_unaligned(cpu_to_le64(ch->next_oid), &footer->next_oid); + + jrelse(sbinfo->journal_footer); +} + +/* wander record capacity depends on current block size */ +static int wander_record_capacity(const struct super_block *super) +{ + return (super->s_blocksize - + sizeof(struct wander_record_header)) / + sizeof(struct wander_entry); +} + +/* Fill first wander record (tx head) in accordance with supplied given data */ +static void format_tx_head(struct commit_handle *ch) +{ + jnode *tx_head; + jnode *next; + struct tx_header *header; + + tx_head = list_entry(ch->tx_list.next, jnode, capture_link); + assert("zam-692", &ch->tx_list != &tx_head->capture_link); + + next = list_entry(tx_head->capture_link.next, jnode, capture_link); + if (&ch->tx_list == &next->capture_link) + next = tx_head; + + header = (struct tx_header *)jdata(tx_head); + + assert("zam-460", header != NULL); + assert("zam-462", ch->super->s_blocksize >= sizeof(struct tx_header)); + + memset(jdata(tx_head), 0, (size_t) ch->super->s_blocksize); + memcpy(jdata(tx_head), TX_HEADER_MAGIC, TX_HEADER_MAGIC_SIZE); + + put_unaligned(cpu_to_le32(ch->tx_size), &header->total); + put_unaligned(cpu_to_le64(get_super_private(ch->super)->last_committed_tx), + &header->prev_tx); + put_unaligned(cpu_to_le64(*jnode_get_block(next)), &header->next_block); + put_unaligned(cpu_to_le64(ch->free_blocks), &header->free_blocks); + put_unaligned(cpu_to_le64(ch->nr_files), &header->nr_files); + put_unaligned(cpu_to_le64(ch->next_oid), &header->next_oid); +} + +/* prepare ordinary wander record block (fill all service fields) */ +static void +format_wander_record(struct commit_handle *ch, jnode *node, __u32 serial) +{ + struct wander_record_header *LRH; + jnode *next; + + assert("zam-464", node != NULL); + + LRH = (struct wander_record_header *)jdata(node); + next = list_entry(node->capture_link.next, jnode, capture_link); + + if (&ch->tx_list == &next->capture_link) + next = list_entry(ch->tx_list.next, jnode, capture_link); + + assert("zam-465", LRH != NULL); + assert("zam-463", + ch->super->s_blocksize > sizeof(struct wander_record_header)); + + memset(jdata(node), 0, (size_t) ch->super->s_blocksize); + memcpy(jdata(node), WANDER_RECORD_MAGIC, WANDER_RECORD_MAGIC_SIZE); + + put_unaligned(cpu_to_le32(ch->tx_size), &LRH->total); + put_unaligned(cpu_to_le32(serial), &LRH->serial); + put_unaligned(cpu_to_le64(*jnode_get_block(next)), &LRH->next_block); +} + +/* add one wandered map entry to formatted wander record */ +static void +store_entry(jnode * node, int index, const reiser4_block_nr * a, + const reiser4_block_nr * b) +{ + char *data; + struct wander_entry *pairs; + + data = jdata(node); + assert("zam-451", data != NULL); + + pairs = + (struct wander_entry *)(data + sizeof(struct wander_record_header)); + + put_unaligned(cpu_to_le64(*a), &pairs[index].original); + put_unaligned(cpu_to_le64(*b), &pairs[index].wandered); +} + +/* currently, wander records contains contain only wandered map, which depend on + overwrite set size */ +static void get_tx_size(struct commit_handle *ch) +{ + assert("zam-440", ch->overwrite_set_size != 0); + assert("zam-695", ch->tx_size == 0); + + /* count all ordinary wander records + ( - 1) / + 1 and add one + for tx head block */ + ch->tx_size = + (ch->overwrite_set_size - 1) / wander_record_capacity(ch->super) + + 2; +} + +/* A special structure for using in store_wmap_actor() for saving its state + between calls */ +struct store_wmap_params { + jnode *cur; /* jnode of current wander record to fill */ + int idx; /* free element index in wander record */ + int capacity; /* capacity */ + +#if REISER4_DEBUG + struct list_head *tx_list; +#endif +}; + +/* an actor for use in blocknr_set_iterator routine which populates the list + of pre-formatted wander records by wandered map info */ +static int +store_wmap_actor(txn_atom * atom UNUSED_ARG, const reiser4_block_nr * a, + const reiser4_block_nr * b, void *data) +{ + struct store_wmap_params *params = data; + + if (params->idx >= params->capacity) { + /* a new wander record should be taken from the tx_list */ + params->cur = list_entry(params->cur->capture_link.next, jnode, capture_link); + assert("zam-454", + params->tx_list != ¶ms->cur->capture_link); + + params->idx = 0; + } + + store_entry(params->cur, params->idx, a, b); + params->idx++; + + return 0; +} + +/* This function is called after Relocate set gets written to disk, Overwrite + set is written to wandered locations and all wander records are written + also. Updated journal header blocks contains a pointer (block number) to + first wander record of the just written transaction */ +static int update_journal_header(struct commit_handle *ch, int use_barrier) +{ + struct reiser4_super_info_data *sbinfo = get_super_private(ch->super); + jnode *jh = sbinfo->journal_header; + jnode *head = list_entry(ch->tx_list.next, jnode, capture_link); + int ret; + + format_journal_header(ch); + + ret = write_jnodes_to_disk_extent(jh, 1, jnode_get_block(jh), NULL, + use_barrier ? WRITEOUT_BARRIER : 0); + if (ret) + return ret; + + /* blk_run_address_space(sbinfo->fake->i_mapping); + * blk_run_queues(); */ + + ret = jwait_io(jh, WRITE); + + if (ret) + return ret; + + sbinfo->last_committed_tx = *jnode_get_block(head); + + return 0; +} + +/* This function is called after write-back is finished. We update journal + footer block and free blocks which were occupied by wandered blocks and + transaction wander records */ +static int update_journal_footer(struct commit_handle *ch, int use_barrier) +{ + reiser4_super_info_data *sbinfo = get_super_private(ch->super); + + jnode *jf = sbinfo->journal_footer; + + int ret; + + format_journal_footer(ch); + + ret = write_jnodes_to_disk_extent(jf, 1, jnode_get_block(jf), NULL, + use_barrier ? WRITEOUT_BARRIER : 0); + if (ret) + return ret; + + /* blk_run_address_space(sbinfo->fake->i_mapping); + * blk_run_queue(); */ + + ret = jwait_io(jf, WRITE); + if (ret) + return ret; + + return 0; +} + +/* free block numbers of wander records of already written in place transaction */ +static void dealloc_tx_list(struct commit_handle *ch) +{ + while (!list_empty(&ch->tx_list)) { + jnode *cur = list_entry(ch->tx_list.next, jnode, capture_link); + list_del(&cur->capture_link); + ON_DEBUG(INIT_LIST_HEAD(&cur->capture_link)); + reiser4_dealloc_block(jnode_get_block(cur), BLOCK_NOT_COUNTED, + BA_FORMATTED); + + unpin_jnode_data(cur); + reiser4_drop_io_head(cur); + } +} + +/* An actor for use in block_nr_iterator() routine which frees wandered blocks + from atom's overwrite set. */ +static int +dealloc_wmap_actor(txn_atom * atom UNUSED_ARG, + const reiser4_block_nr * a UNUSED_ARG, + const reiser4_block_nr * b, void *data UNUSED_ARG) +{ + + assert("zam-499", b != NULL); + assert("zam-500", *b != 0); + assert("zam-501", !reiser4_blocknr_is_fake(b)); + + reiser4_dealloc_block(b, BLOCK_NOT_COUNTED, BA_FORMATTED); + return 0; +} + +/* free wandered block locations of already written in place transaction */ +static void dealloc_wmap(struct commit_handle *ch) +{ + assert("zam-696", ch->atom != NULL); + + blocknr_set_iterator(ch->atom, &ch->atom->wandered_map, + dealloc_wmap_actor, NULL, 1); +} + +/* helper function for alloc wandered blocks, which refill set of block + numbers needed for wandered blocks */ +static int +get_more_wandered_blocks(int count, reiser4_block_nr * start, int *len) +{ + reiser4_blocknr_hint hint; + int ret; + + reiser4_block_nr wide_len = count; + + /* FIXME-ZAM: A special policy needed for allocation of wandered blocks + ZAM-FIXME-HANS: yes, what happened to our discussion of using a fixed + reserved allocation area so as to get the best qualities of fixed + journals? */ + reiser4_blocknr_hint_init(&hint); + hint.block_stage = BLOCK_GRABBED; + + ret = reiser4_alloc_blocks(&hint, start, &wide_len, + BA_FORMATTED | BA_USE_DEFAULT_SEARCH_START); + *len = (int)wide_len; + + return ret; +} + +/* + * roll back changes made before issuing BIO in the case of IO error. + */ +static void undo_bio(struct bio *bio) +{ + int i; + + for (i = 0; i < bio->bi_vcnt; ++i) { + struct page *pg; + jnode *node; + + pg = bio->bi_io_vec[i].bv_page; + end_page_writeback(pg); + node = jprivate(pg); + spin_lock_jnode(node); + JF_CLR(node, JNODE_WRITEBACK); + JF_SET(node, JNODE_DIRTY); + spin_unlock_jnode(node); + } + bio_put(bio); +} + +/* put overwrite set back to atom's clean list */ +static void put_overwrite_set(struct commit_handle *ch) +{ + jnode *cur; + + list_for_each_entry(cur, ch->overwrite_set, capture_link) + jrelse_tail(cur); +} + +/* Count overwrite set size, grab disk space for wandered blocks allocation. + Since we have a separate list for atom's overwrite set we just scan the list, + count bitmap and other not leaf nodes which wandered blocks allocation we + have to grab space for. */ +static int get_overwrite_set(struct commit_handle *ch) +{ + int ret; + jnode *cur; + __u64 nr_not_leaves = 0; +#if REISER4_DEBUG + __u64 nr_formatted_leaves = 0; + __u64 nr_unformatted_leaves = 0; +#endif + + assert("zam-697", ch->overwrite_set_size == 0); + + ch->overwrite_set = ATOM_OVRWR_LIST(ch->atom); + cur = list_entry(ch->overwrite_set->next, jnode, capture_link); + + while (ch->overwrite_set != &cur->capture_link) { + jnode *next = list_entry(cur->capture_link.next, jnode, capture_link); + + /* Count bitmap locks for getting correct statistics what number + * of blocks were cleared by the transaction commit. */ + if (jnode_get_type(cur) == JNODE_BITMAP) + ch->nr_bitmap++; + + assert("zam-939", JF_ISSET(cur, JNODE_OVRWR) + || jnode_get_type(cur) == JNODE_BITMAP); + + if (jnode_is_znode(cur) && znode_above_root(JZNODE(cur))) { + /* we replace fake znode by another (real) + znode which is suggested by disk_layout + plugin */ + + /* FIXME: it looks like fake znode should be + replaced by jnode supplied by + disk_layout. */ + + struct super_block *s = reiser4_get_current_sb(); + reiser4_super_info_data *sbinfo = + get_current_super_private(); + + if (sbinfo->df_plug->log_super) { + jnode *sj = sbinfo->df_plug->log_super(s); + + assert("zam-593", sj != NULL); + + if (IS_ERR(sj)) + return PTR_ERR(sj); + + spin_lock_jnode(sj); + JF_SET(sj, JNODE_OVRWR); + insert_into_atom_ovrwr_list(ch->atom, sj); + spin_unlock_jnode(sj); + + /* jload it as the rest of overwrite set */ + jload_gfp(sj, reiser4_ctx_gfp_mask_get(), 0); + + ch->overwrite_set_size++; + } + spin_lock_jnode(cur); + reiser4_uncapture_block(cur); + jput(cur); + + } else { + int ret; + ch->overwrite_set_size++; + ret = jload_gfp(cur, reiser4_ctx_gfp_mask_get(), 0); + if (ret) + reiser4_panic("zam-783", + "cannot load e-flushed jnode back (ret = %d)\n", + ret); + } + + /* Count not leaves here because we have to grab disk space + * for wandered blocks. They were not counted as "flush + * reserved". Counting should be done _after_ nodes are pinned + * into memory by jload(). */ + if (!jnode_is_leaf(cur)) + nr_not_leaves++; + else { +#if REISER4_DEBUG + /* at this point @cur either has JNODE_FLUSH_RESERVED + * or is eflushed. Locking is not strong enough to + * write an assertion checking for this. */ + if (jnode_is_znode(cur)) + nr_formatted_leaves++; + else + nr_unformatted_leaves++; +#endif + JF_CLR(cur, JNODE_FLUSH_RESERVED); + } + + cur = next; + } + + /* Grab space for writing (wandered blocks) of not leaves found in + * overwrite set. */ + ret = reiser4_grab_space_force(nr_not_leaves, BA_RESERVED); + if (ret) + return ret; + + /* Disk space for allocation of wandered blocks of leaf nodes already + * reserved as "flush reserved", move it to grabbed space counter. */ + spin_lock_atom(ch->atom); + assert("zam-940", + nr_formatted_leaves + nr_unformatted_leaves <= + ch->atom->flush_reserved); + flush_reserved2grabbed(ch->atom, ch->atom->flush_reserved); + spin_unlock_atom(ch->atom); + + return ch->overwrite_set_size; +} + +/** + * write_jnodes_to_disk_extent - submit write request + * @head: + * @first: first jnode of the list + * @nr: number of jnodes on the list + * @block_p: + * @fq: + * @flags: used to decide whether page is to get PG_reclaim flag + * + * Submits a write request for @nr jnodes beginning from the @first, other + * jnodes are after the @first on the double-linked "capture" list. All jnodes + * will be written to the disk region of @nr blocks starting with @block_p block + * number. If @fq is not NULL it means that waiting for i/o completion will be + * done more efficiently by using flush_queue_t objects. + * This function is the one which writes list of jnodes in batch mode. It does + * all low-level things as bio construction and page states manipulation. + * + * ZAM-FIXME-HANS: brief me on why this function exists, and why bios are + * aggregated in this function instead of being left to the layers below + * + * FIXME: ZAM->HANS: What layer are you talking about? Can you point me to that? + * Why that layer needed? Why BIOs cannot be constructed here? + */ +static int write_jnodes_to_disk_extent( + jnode *first, int nr, const reiser4_block_nr *block_p, + flush_queue_t *fq, int flags) +{ + struct super_block *super = reiser4_get_current_sb(); + int write_op = ( flags & WRITEOUT_BARRIER ) ? WRITE_BARRIER : WRITE; + int max_blocks; + jnode *cur = first; + reiser4_block_nr block; + + assert("zam-571", first != NULL); + assert("zam-572", block_p != NULL); + assert("zam-570", nr > 0); + + block = *block_p; + max_blocks = min(bio_get_nr_vecs(super->s_bdev), BIO_MAX_PAGES); + + while (nr > 0) { + struct bio *bio; + int nr_blocks = min(nr, max_blocks); + int i; + int nr_used; + + bio = bio_alloc(GFP_NOIO, nr_blocks); + if (!bio) + return RETERR(-ENOMEM); + + bio->bi_bdev = super->s_bdev; + bio->bi_sector = block * (super->s_blocksize >> 9); + for (nr_used = 0, i = 0; i < nr_blocks; i++) { + struct page *pg; + + pg = jnode_page(cur); + assert("zam-573", pg != NULL); + + page_cache_get(pg); + + lock_and_wait_page_writeback(pg); + + if (!bio_add_page(bio, pg, super->s_blocksize, 0)) { + /* + * underlying device is satiated. Stop adding + * pages to the bio. + */ + unlock_page(pg); + page_cache_release(pg); + break; + } + + spin_lock_jnode(cur); + assert("nikita-3166", + pg->mapping == jnode_get_mapping(cur)); + assert("zam-912", !JF_ISSET(cur, JNODE_WRITEBACK)); +#if REISER4_DEBUG + spin_lock(&cur->load); + assert("nikita-3165", !jnode_is_releasable(cur)); + spin_unlock(&cur->load); +#endif + JF_SET(cur, JNODE_WRITEBACK); + JF_CLR(cur, JNODE_DIRTY); + ON_DEBUG(cur->written++); + spin_unlock_jnode(cur); + + ClearPageError(pg); + set_page_writeback(pg); + + if (get_current_context()->entd) { + /* this is ent thread */ + entd_context *ent = get_entd_context(super); + struct wbq *rq, *next; + + spin_lock(&ent->guard); + + if (pg == ent->cur_request->page) { + /* + * entd is called for this page. This + * request is not in th etodo list + */ + ent->cur_request->written = 1; + } else { + /* + * if we have written a page for which writepage + * is called for - move request to another list. + */ + list_for_each_entry_safe(rq, next, &ent->todo_list, link) { + assert("", rq->magic == WBQ_MAGIC); + if (pg == rq->page) { + /* + * remove request from + * entd's queue, but do + * not wake up a thread + * which put this + * request + */ + list_del_init(&rq->link); + ent->nr_todo_reqs --; + list_add_tail(&rq->link, &ent->done_list); + ent->nr_done_reqs ++; + rq->written = 1; + break; + } + } + } + spin_unlock(&ent->guard); + } + + clear_page_dirty_for_io(pg); + + unlock_page(pg); + + cur = list_entry(cur->capture_link.next, jnode, capture_link); + nr_used++; + } + if (nr_used > 0) { + assert("nikita-3453", + bio->bi_size == super->s_blocksize * nr_used); + assert("nikita-3454", bio->bi_vcnt == nr_used); + + /* Check if we are allowed to write at all */ + if (super->s_flags & MS_RDONLY) + undo_bio(bio); + else { + int not_supported; + + add_fq_to_bio(fq, bio); + bio_get(bio); + reiser4_submit_bio(write_op, bio); + not_supported = bio_flagged(bio, BIO_EOPNOTSUPP); + bio_put(bio); + if (not_supported) + return -EOPNOTSUPP; + } + + block += nr_used - 1; + update_blocknr_hint_default(super, &block); + block += 1; + } else { + bio_put(bio); + } + nr -= nr_used; + } + + return 0; +} + +/* This is a procedure which recovers a contiguous sequences of disk block + numbers in the given list of j-nodes and submits write requests on this + per-sequence basis */ +int +write_jnode_list(struct list_head *head, flush_queue_t *fq, + long *nr_submitted, int flags) +{ + int ret; + jnode *beg = list_entry(head->next, jnode, capture_link); + + while (head != &beg->capture_link) { + int nr = 1; + jnode *cur = list_entry(beg->capture_link.next, jnode, capture_link); + + while (head != &cur->capture_link) { + if (*jnode_get_block(cur) != *jnode_get_block(beg) + nr) + break; + ++nr; + cur = list_entry(cur->capture_link.next, jnode, capture_link); + } + + ret = write_jnodes_to_disk_extent( + beg, nr, jnode_get_block(beg), fq, flags); + if (ret) + return ret; + + if (nr_submitted) + *nr_submitted += nr; + + beg = cur; + } + + return 0; +} + +/* add given wandered mapping to atom's wandered map */ +static int +add_region_to_wmap(jnode * cur, int len, const reiser4_block_nr * block_p) +{ + int ret; + blocknr_set_entry *new_bsep = NULL; + reiser4_block_nr block; + + txn_atom *atom; + + assert("zam-568", block_p != NULL); + block = *block_p; + assert("zam-569", len > 0); + + while ((len--) > 0) { + do { + atom = get_current_atom_locked(); + assert("zam-536", + !reiser4_blocknr_is_fake(jnode_get_block(cur))); + ret = + blocknr_set_add_pair(atom, &atom->wandered_map, + &new_bsep, + jnode_get_block(cur), &block); + } while (ret == -E_REPEAT); + + if (ret) { + /* deallocate blocks which were not added to wandered + map */ + reiser4_block_nr wide_len = len; + + reiser4_dealloc_blocks(&block, &wide_len, + BLOCK_NOT_COUNTED, + BA_FORMATTED + /* formatted, without defer */ ); + + return ret; + } + + spin_unlock_atom(atom); + + cur = list_entry(cur->capture_link.next, jnode, capture_link); + ++block; + } + + return 0; +} + +/* Allocate wandered blocks for current atom's OVERWRITE SET and immediately + submit IO for allocated blocks. We assume that current atom is in a stage + when any atom fusion is impossible and atom is unlocked and it is safe. */ +static int alloc_wandered_blocks(struct commit_handle *ch, flush_queue_t *fq) +{ + reiser4_block_nr block; + + int rest; + int len; + int ret; + + jnode *cur; + + assert("zam-534", ch->overwrite_set_size > 0); + + rest = ch->overwrite_set_size; + + cur = list_entry(ch->overwrite_set->next, jnode, capture_link); + while (ch->overwrite_set != &cur->capture_link) { + assert("zam-567", JF_ISSET(cur, JNODE_OVRWR)); + + ret = get_more_wandered_blocks(rest, &block, &len); + if (ret) + return ret; + + rest -= len; + + ret = add_region_to_wmap(cur, len, &block); + if (ret) + return ret; + + ret = write_jnodes_to_disk_extent(cur, len, &block, fq, 0); + if (ret) + return ret; + + while ((len--) > 0) { + assert("zam-604", + ch->overwrite_set != &cur->capture_link); + cur = list_entry(cur->capture_link.next, jnode, capture_link); + } + } + + return 0; +} + +/* allocate given number of nodes over the journal area and link them into a + list, return pointer to the first jnode in the list */ +static int alloc_tx(struct commit_handle *ch, flush_queue_t * fq) +{ + reiser4_blocknr_hint hint; + reiser4_block_nr allocated = 0; + reiser4_block_nr first, len; + jnode *cur; + jnode *txhead; + int ret; + reiser4_context *ctx; + reiser4_super_info_data *sbinfo; + + assert("zam-698", ch->tx_size > 0); + assert("zam-699", list_empty_careful(&ch->tx_list)); + + ctx = get_current_context(); + sbinfo = get_super_private(ctx->super); + + while (allocated < (unsigned)ch->tx_size) { + len = (ch->tx_size - allocated); + + reiser4_blocknr_hint_init(&hint); + + hint.block_stage = BLOCK_GRABBED; + + /* FIXME: there should be some block allocation policy for + nodes which contain wander records */ + + /* We assume that disk space for wandered record blocks can be + * taken from reserved area. */ + ret = reiser4_alloc_blocks(&hint, &first, &len, + BA_FORMATTED | BA_RESERVED | + BA_USE_DEFAULT_SEARCH_START); + reiser4_blocknr_hint_done(&hint); + + if (ret) + return ret; + + allocated += len; + + /* create jnodes for all wander records */ + while (len--) { + cur = reiser4_alloc_io_head(&first); + + if (cur == NULL) { + ret = RETERR(-ENOMEM); + goto free_not_assigned; + } + + ret = jinit_new(cur, reiser4_ctx_gfp_mask_get()); + + if (ret != 0) { + jfree(cur); + goto free_not_assigned; + } + + pin_jnode_data(cur); + + list_add_tail(&cur->capture_link, &ch->tx_list); + + first++; + } + } + + { /* format a on-disk linked list of wander records */ + int serial = 1; + + txhead = list_entry(ch->tx_list.next, jnode, capture_link); + format_tx_head(ch); + + cur = list_entry(txhead->capture_link.next, jnode, capture_link); + while (&ch->tx_list != &cur->capture_link) { + format_wander_record(ch, cur, serial++); + cur = list_entry(cur->capture_link.next, jnode, capture_link); + } + } + + { /* Fill wander records with Wandered Set */ + struct store_wmap_params params; + txn_atom *atom; + + params.cur = list_entry(txhead->capture_link.next, jnode, capture_link); + + params.idx = 0; + params.capacity = + wander_record_capacity(reiser4_get_current_sb()); + + atom = get_current_atom_locked(); + blocknr_set_iterator(atom, &atom->wandered_map, + &store_wmap_actor, ¶ms, 0); + spin_unlock_atom(atom); + } + + { /* relse all jnodes from tx_list */ + cur = list_entry(ch->tx_list.next, jnode, capture_link); + while (&ch->tx_list != &cur->capture_link) { + jrelse(cur); + cur = list_entry(cur->capture_link.next, jnode, capture_link); + } + } + + ret = write_jnode_list(&ch->tx_list, fq, NULL, 0); + + return ret; + + free_not_assigned: + /* We deallocate blocks not yet assigned to jnodes on tx_list. The + caller takes care about invalidating of tx list */ + reiser4_dealloc_blocks(&first, &len, BLOCK_NOT_COUNTED, BA_FORMATTED); + + return ret; +} + +static int commit_tx(struct commit_handle *ch) +{ + flush_queue_t *fq; + int barrier; + int ret; + + /* Grab more space for wandered records. */ + ret = reiser4_grab_space_force((__u64) (ch->tx_size), BA_RESERVED); + if (ret) + return ret; + + fq = get_fq_for_current_atom(); + if (IS_ERR(fq)) + return PTR_ERR(fq); + + spin_unlock_atom(fq->atom); + do { + ret = alloc_wandered_blocks(ch, fq); + if (ret) + break; + ret = alloc_tx(ch, fq); + if (ret) + break; + } while (0); + + reiser4_fq_put(fq); + if (ret) + return ret; + repeat_wo_barrier: + barrier = reiser4_use_write_barrier(ch->super); + if (!barrier) { + ret = current_atom_finish_all_fq(); + if (ret) + return ret; + } + ret = update_journal_header(ch, barrier); + if (barrier) { + if (ret) { + if (ret == -EOPNOTSUPP) { + disable_write_barrier(ch->super); + goto repeat_wo_barrier; + } + return ret; + } + ret = current_atom_finish_all_fq(); + } + return ret; +} + +static int write_tx_back(struct commit_handle * ch) +{ + flush_queue_t *fq; + int ret; + int barrier; + + reiser4_post_commit_hook(); + fq = get_fq_for_current_atom(); + if (IS_ERR(fq)) + return PTR_ERR(fq); + spin_unlock_atom(fq->atom); + ret = write_jnode_list( + ch->overwrite_set, fq, NULL, WRITEOUT_FOR_PAGE_RECLAIM); + reiser4_fq_put(fq); + if (ret) + return ret; + repeat_wo_barrier: + barrier = reiser4_use_write_barrier(ch->super); + if (!barrier) { + ret = current_atom_finish_all_fq(); + if (ret) + return ret; + } + ret = update_journal_footer(ch, barrier); + if (barrier) { + if (ret) { + if (ret == -EOPNOTSUPP) { + disable_write_barrier(ch->super); + goto repeat_wo_barrier; + } + return ret; + } + ret = current_atom_finish_all_fq(); + } + if (ret) + return ret; + reiser4_post_write_back_hook(); + return 0; +} + +/* We assume that at this moment all captured blocks are marked as RELOC or + WANDER (belong to Relocate o Overwrite set), all nodes from Relocate set + are submitted to write. +*/ + +int reiser4_write_logs(long *nr_submitted) +{ + txn_atom *atom; + struct super_block *super = reiser4_get_current_sb(); + reiser4_super_info_data *sbinfo = get_super_private(super); + struct commit_handle ch; + int ret; + + writeout_mode_enable(); + + /* block allocator may add j-nodes to the clean_list */ + ret = reiser4_pre_commit_hook(); + if (ret) + return ret; + + /* No locks are required if we take atom which stage >= + * ASTAGE_PRE_COMMIT */ + atom = get_current_context()->trans->atom; + assert("zam-965", atom != NULL); + + /* relocate set is on the atom->clean_nodes list after + * current_atom_complete_writes() finishes. It can be safely + * uncaptured after commit_mutex is locked, because any atom that + * captures these nodes is guaranteed to commit after current one. + * + * This can only be done after reiser4_pre_commit_hook(), because it is where + * early flushed jnodes with CREATED bit are transferred to the + * overwrite list. */ + reiser4_invalidate_list(ATOM_CLEAN_LIST(atom)); + spin_lock_atom(atom); + /* There might be waiters for the relocate nodes which we have + * released, wake them up. */ + reiser4_atom_send_event(atom); + spin_unlock_atom(atom); + + if (REISER4_DEBUG) { + int level; + + for (level = 0; level < REAL_MAX_ZTREE_HEIGHT + 1; ++level) + assert("nikita-3352", + list_empty_careful(ATOM_DIRTY_LIST(atom, level))); + } + + sbinfo->nr_files_committed += (unsigned)atom->nr_objects_created; + sbinfo->nr_files_committed -= (unsigned)atom->nr_objects_deleted; + + init_commit_handle(&ch, atom); + + ch.free_blocks = sbinfo->blocks_free_committed; + ch.nr_files = sbinfo->nr_files_committed; + /* ZAM-FIXME-HANS: email me what the contention level is for the super + * lock. */ + ch.next_oid = oid_next(super); + + /* count overwrite set and place it in a separate list */ + ret = get_overwrite_set(&ch); + + if (ret <= 0) { + /* It is possible that overwrite set is empty here, it means + all captured nodes are clean */ + goto up_and_ret; + } + + /* Inform the caller about what number of dirty pages will be + * submitted to disk. */ + *nr_submitted += ch.overwrite_set_size - ch.nr_bitmap; + + /* count all records needed for storing of the wandered set */ + get_tx_size(&ch); + + ret = commit_tx(&ch); + if (ret) + goto up_and_ret; + + spin_lock_atom(atom); + reiser4_atom_set_stage(atom, ASTAGE_POST_COMMIT); + spin_unlock_atom(atom); + + ret = write_tx_back(&ch); + reiser4_post_write_back_hook(); + + up_and_ret: + if (ret) { + /* there could be fq attached to current atom; the only way to + remove them is: */ + current_atom_finish_all_fq(); + } + + /* free blocks of flushed transaction */ + dealloc_tx_list(&ch); + dealloc_wmap(&ch); + + put_overwrite_set(&ch); + + done_commit_handle(&ch); + + writeout_mode_disable(); + + return ret; +} + +/* consistency checks for journal data/control blocks: header, footer, log + records, transactions head blocks. All functions return zero on success. */ + +static int check_journal_header(const jnode * node UNUSED_ARG) +{ + /* FIXME: journal header has no magic field yet. */ + return 0; +} + +/* wait for write completion for all jnodes from given list */ +static int wait_on_jnode_list(struct list_head *head) +{ + jnode *scan; + int ret = 0; + + list_for_each_entry(scan, head, capture_link) { + struct page *pg = jnode_page(scan); + + if (pg) { + if (PageWriteback(pg)) + wait_on_page_writeback(pg); + + if (PageError(pg)) + ret++; + } + } + + return ret; +} + +static int check_journal_footer(const jnode * node UNUSED_ARG) +{ + /* FIXME: journal footer has no magic field yet. */ + return 0; +} + +static int check_tx_head(const jnode * node) +{ + struct tx_header *header = (struct tx_header *)jdata(node); + + if (memcmp(&header->magic, TX_HEADER_MAGIC, TX_HEADER_MAGIC_SIZE) != 0) { + warning("zam-627", "tx head at block %s corrupted\n", + sprint_address(jnode_get_block(node))); + return RETERR(-EIO); + } + + return 0; +} + +static int check_wander_record(const jnode * node) +{ + struct wander_record_header *RH = + (struct wander_record_header *)jdata(node); + + if (memcmp(&RH->magic, WANDER_RECORD_MAGIC, WANDER_RECORD_MAGIC_SIZE) != + 0) { + warning("zam-628", "wander record at block %s corrupted\n", + sprint_address(jnode_get_block(node))); + return RETERR(-EIO); + } + + return 0; +} + +/* fill commit_handler structure by everything what is needed for update_journal_footer */ +static int restore_commit_handle(struct commit_handle *ch, jnode *tx_head) +{ + struct tx_header *TXH; + int ret; + + ret = jload(tx_head); + if (ret) + return ret; + + TXH = (struct tx_header *)jdata(tx_head); + + ch->free_blocks = le64_to_cpu(get_unaligned(&TXH->free_blocks)); + ch->nr_files = le64_to_cpu(get_unaligned(&TXH->nr_files)); + ch->next_oid = le64_to_cpu(get_unaligned(&TXH->next_oid)); + + jrelse(tx_head); + + list_add(&tx_head->capture_link, &ch->tx_list); + + return 0; +} + +/* replay one transaction: restore and write overwrite set in place */ +static int replay_transaction(const struct super_block *s, + jnode * tx_head, + const reiser4_block_nr * log_rec_block_p, + const reiser4_block_nr * end_block, + unsigned int nr_wander_records) +{ + reiser4_block_nr log_rec_block = *log_rec_block_p; + struct commit_handle ch; + LIST_HEAD(overwrite_set); + jnode *log; + int ret; + + init_commit_handle(&ch, NULL); + ch.overwrite_set = &overwrite_set; + + restore_commit_handle(&ch, tx_head); + + while (log_rec_block != *end_block) { + struct wander_record_header *header; + struct wander_entry *entry; + + int i; + + if (nr_wander_records == 0) { + warning("zam-631", + "number of wander records in the linked list" + " greater than number stored in tx head.\n"); + ret = RETERR(-EIO); + goto free_ow_set; + } + + log = reiser4_alloc_io_head(&log_rec_block); + if (log == NULL) + return RETERR(-ENOMEM); + + ret = jload(log); + if (ret < 0) { + reiser4_drop_io_head(log); + return ret; + } + + ret = check_wander_record(log); + if (ret) { + jrelse(log); + reiser4_drop_io_head(log); + return ret; + } + + header = (struct wander_record_header *)jdata(log); + log_rec_block = le64_to_cpu(get_unaligned(&header->next_block)); + + entry = (struct wander_entry *)(header + 1); + + /* restore overwrite set from wander record content */ + for (i = 0; i < wander_record_capacity(s); i++) { + reiser4_block_nr block; + jnode *node; + + block = le64_to_cpu(get_unaligned(&entry->wandered)); + if (block == 0) + break; + + node = reiser4_alloc_io_head(&block); + if (node == NULL) { + ret = RETERR(-ENOMEM); + /* + * FIXME-VS:??? + */ + jrelse(log); + reiser4_drop_io_head(log); + goto free_ow_set; + } + + ret = jload(node); + + if (ret < 0) { + reiser4_drop_io_head(node); + /* + * FIXME-VS:??? + */ + jrelse(log); + reiser4_drop_io_head(log); + goto free_ow_set; + } + + block = le64_to_cpu(get_unaligned(&entry->original)); + + assert("zam-603", block != 0); + + jnode_set_block(node, &block); + + list_add_tail(&node->capture_link, ch.overwrite_set); + + ++entry; + } + + jrelse(log); + reiser4_drop_io_head(log); + + --nr_wander_records; + } + + if (nr_wander_records != 0) { + warning("zam-632", "number of wander records in the linked list" + " less than number stored in tx head.\n"); + ret = RETERR(-EIO); + goto free_ow_set; + } + + { /* write wandered set in place */ + write_jnode_list(ch.overwrite_set, NULL, NULL, 0); + ret = wait_on_jnode_list(ch.overwrite_set); + + if (ret) { + ret = RETERR(-EIO); + goto free_ow_set; + } + } + + ret = update_journal_footer(&ch, 0); + + free_ow_set: + + while (!list_empty(ch.overwrite_set)) { + jnode *cur = list_entry(ch.overwrite_set->next, jnode, capture_link); + list_del_init(&cur->capture_link); + jrelse(cur); + reiser4_drop_io_head(cur); + } + + list_del_init(&tx_head->capture_link); + + done_commit_handle(&ch); + + return ret; +} + +/* find oldest committed and not played transaction and play it. The transaction + * was committed and journal header block was updated but the blocks from the + * process of writing the atom's overwrite set in-place and updating of journal + * footer block were not completed. This function completes the process by + * recovering the atom's overwrite set from their wandered locations and writes + * them in-place and updating the journal footer. */ +static int replay_oldest_transaction(struct super_block *s) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + jnode *jf = sbinfo->journal_footer; + unsigned int total; + struct journal_footer *F; + struct tx_header *T; + + reiser4_block_nr prev_tx; + reiser4_block_nr last_flushed_tx; + reiser4_block_nr log_rec_block = 0; + + jnode *tx_head; + + int ret; + + if ((ret = jload(jf)) < 0) + return ret; + + F = (struct journal_footer *)jdata(jf); + + last_flushed_tx = le64_to_cpu(get_unaligned(&F->last_flushed_tx)); + + jrelse(jf); + + if (sbinfo->last_committed_tx == last_flushed_tx) { + /* all transactions are replayed */ + return 0; + } + + prev_tx = sbinfo->last_committed_tx; + + /* searching for oldest not flushed transaction */ + while (1) { + tx_head = reiser4_alloc_io_head(&prev_tx); + if (!tx_head) + return RETERR(-ENOMEM); + + ret = jload(tx_head); + if (ret < 0) { + reiser4_drop_io_head(tx_head); + return ret; + } + + ret = check_tx_head(tx_head); + if (ret) { + jrelse(tx_head); + reiser4_drop_io_head(tx_head); + return ret; + } + + T = (struct tx_header *)jdata(tx_head); + + prev_tx = le64_to_cpu(get_unaligned(&T->prev_tx)); + + if (prev_tx == last_flushed_tx) + break; + + jrelse(tx_head); + reiser4_drop_io_head(tx_head); + } + + total = le32_to_cpu(get_unaligned(&T->total)); + log_rec_block = le64_to_cpu(get_unaligned(&T->next_block)); + + pin_jnode_data(tx_head); + jrelse(tx_head); + + ret = + replay_transaction(s, tx_head, &log_rec_block, + jnode_get_block(tx_head), total - 1); + + unpin_jnode_data(tx_head); + reiser4_drop_io_head(tx_head); + + if (ret) + return ret; + return -E_REPEAT; +} + +/* The reiser4 journal current implementation was optimized to not to capture + super block if certain super blocks fields are modified. Currently, the set + is (, ). These fields are logged by + special way which includes storing them in each transaction head block at + atom commit time and writing that information to journal footer block at + atom flush time. For getting info from journal footer block to the + in-memory super block there is a special function + reiser4_journal_recover_sb_data() which should be called after disk format + plugin re-reads super block after journal replaying. +*/ + +/* get the information from journal footer in-memory super block */ +int reiser4_journal_recover_sb_data(struct super_block *s) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + struct journal_footer *jf; + int ret; + + assert("zam-673", sbinfo->journal_footer != NULL); + + ret = jload(sbinfo->journal_footer); + if (ret != 0) + return ret; + + ret = check_journal_footer(sbinfo->journal_footer); + if (ret != 0) + goto out; + + jf = (struct journal_footer *)jdata(sbinfo->journal_footer); + + /* was there at least one flushed transaction? */ + if (jf->last_flushed_tx) { + + /* restore free block counter logged in this transaction */ + reiser4_set_free_blocks(s, le64_to_cpu(get_unaligned(&jf->free_blocks))); + + /* restore oid allocator state */ + oid_init_allocator(s, + le64_to_cpu(get_unaligned(&jf->nr_files)), + le64_to_cpu(get_unaligned(&jf->next_oid))); + } + out: + jrelse(sbinfo->journal_footer); + return ret; +} + +/* reiser4 replay journal procedure */ +int reiser4_journal_replay(struct super_block *s) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + jnode *jh, *jf; + struct journal_header *header; + int nr_tx_replayed = 0; + int ret; + + assert("zam-582", sbinfo != NULL); + + jh = sbinfo->journal_header; + jf = sbinfo->journal_footer; + + if (!jh || !jf) { + /* it is possible that disk layout does not support journal + structures, we just warn about this */ + warning("zam-583", + "journal control blocks were not loaded by disk layout plugin. " + "journal replaying is not possible.\n"); + return 0; + } + + /* Take free block count from journal footer block. The free block + counter value corresponds the last flushed transaction state */ + ret = jload(jf); + if (ret < 0) + return ret; + + ret = check_journal_footer(jf); + if (ret) { + jrelse(jf); + return ret; + } + + jrelse(jf); + + /* store last committed transaction info in reiser4 in-memory super + block */ + ret = jload(jh); + if (ret < 0) + return ret; + + ret = check_journal_header(jh); + if (ret) { + jrelse(jh); + return ret; + } + + header = (struct journal_header *)jdata(jh); + sbinfo->last_committed_tx = le64_to_cpu(get_unaligned(&header->last_committed_tx)); + + jrelse(jh); + + /* replay committed transactions */ + while ((ret = replay_oldest_transaction(s)) == -E_REPEAT) + nr_tx_replayed++; + + return ret; +} + +/* load journal control block (either journal header or journal footer block) */ +static int +load_journal_control_block(jnode ** node, const reiser4_block_nr * block) +{ + int ret; + + *node = reiser4_alloc_io_head(block); + if (!(*node)) + return RETERR(-ENOMEM); + + ret = jload(*node); + + if (ret) { + reiser4_drop_io_head(*node); + *node = NULL; + return ret; + } + + pin_jnode_data(*node); + jrelse(*node); + + return 0; +} + +/* unload journal header or footer and free jnode */ +static void unload_journal_control_block(jnode ** node) +{ + if (*node) { + unpin_jnode_data(*node); + reiser4_drop_io_head(*node); + *node = NULL; + } +} + +/* release journal control blocks */ +void reiser4_done_journal_info(struct super_block *s) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + + assert("zam-476", sbinfo != NULL); + + unload_journal_control_block(&sbinfo->journal_header); + unload_journal_control_block(&sbinfo->journal_footer); + rcu_barrier(); +} + +/* load journal control blocks */ +int reiser4_init_journal_info(struct super_block *s) +{ + reiser4_super_info_data *sbinfo = get_super_private(s); + journal_location *loc; + int ret; + + loc = &sbinfo->jloc; + + assert("zam-651", loc != NULL); + assert("zam-652", loc->header != 0); + assert("zam-653", loc->footer != 0); + + ret = load_journal_control_block(&sbinfo->journal_header, &loc->header); + + if (ret) + return ret; + + ret = load_journal_control_block(&sbinfo->journal_footer, &loc->footer); + + if (ret) { + unload_journal_control_block(&sbinfo->journal_header); + } + + return ret; +} + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/wander.h linux-2.6.30/fs/reiser4/wander.h --- linux-2.6.30.orig/fs/reiser4/wander.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/wander.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,135 @@ +/* Copyright 2002, 2003 by Hans Reiser, licensing governed by reiser4/README */ + +#if !defined (__FS_REISER4_WANDER_H__) +#define __FS_REISER4_WANDER_H__ + +#include "dformat.h" + +#include /* for struct super_block */ + +/* REISER4 JOURNAL ON-DISK DATA STRUCTURES */ + +#define TX_HEADER_MAGIC "TxMagic4" +#define WANDER_RECORD_MAGIC "LogMagc4" + +#define TX_HEADER_MAGIC_SIZE (8) +#define WANDER_RECORD_MAGIC_SIZE (8) + +/* journal header block format */ +struct journal_header { + /* last written transaction head location */ + d64 last_committed_tx; +}; + +typedef struct journal_location { + reiser4_block_nr footer; + reiser4_block_nr header; +} journal_location; + +/* The wander.c head comment describes usage and semantic of all these structures */ +/* journal footer block format */ +struct journal_footer { + /* last flushed transaction location. */ + /* This block number is no more valid after the transaction it points + to gets flushed, this number is used only at journal replaying time + for detection of the end of on-disk list of committed transactions + which were not flushed completely */ + d64 last_flushed_tx; + + /* free block counter is written in journal footer at transaction + flushing , not in super block because free blocks counter is logged + by another way than super block fields (root pointer, for + example). */ + d64 free_blocks; + + /* number of used OIDs and maximal used OID are logged separately from + super block */ + d64 nr_files; + d64 next_oid; +}; + +/* Each wander record (except the first one) has unified format with wander + record header followed by an array of log entries */ +struct wander_record_header { + /* when there is no predefined location for wander records, this magic + string should help reiser4fsck. */ + char magic[WANDER_RECORD_MAGIC_SIZE]; + + /* transaction id */ + d64 id; + + /* total number of wander records in current transaction */ + d32 total; + + /* this block number in transaction */ + d32 serial; + + /* number of previous block in commit */ + d64 next_block; +}; + +/* The first wander record (transaction head) of written transaction has the + special format */ +struct tx_header { + /* magic string makes first block in transaction different from other + logged blocks, it should help fsck. */ + char magic[TX_HEADER_MAGIC_SIZE]; + + /* transaction id */ + d64 id; + + /* total number of records (including this first tx head) in the + transaction */ + d32 total; + + /* align next field to 8-byte boundary; this field always is zero */ + d32 padding; + + /* block number of previous transaction head */ + d64 prev_tx; + + /* next wander record location */ + d64 next_block; + + /* committed versions of free blocks counter */ + d64 free_blocks; + + /* number of used OIDs (nr_files) and maximal used OID are logged + separately from super block */ + d64 nr_files; + d64 next_oid; +}; + +/* A transaction gets written to disk as a set of wander records (each wander + record size is fs block) */ + +/* As it was told above a wander The rest of wander record is filled by these log entries, unused space filled + by zeroes */ +struct wander_entry { + d64 original; /* block original location */ + d64 wandered; /* block wandered location */ +}; + +/* REISER4 JOURNAL WRITER FUNCTIONS */ + +extern int reiser4_write_logs(long *); +extern int reiser4_journal_replay(struct super_block *); +extern int reiser4_journal_recover_sb_data(struct super_block *); + +extern int reiser4_init_journal_info(struct super_block *); +extern void reiser4_done_journal_info(struct super_block *); + +extern int write_jnode_list(struct list_head *, flush_queue_t *, long *, int); + +#endif /* __FS_REISER4_WANDER_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + scroll-step: 1 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/writeout.h linux-2.6.30/fs/reiser4/writeout.h --- linux-2.6.30.orig/fs/reiser4/writeout.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/writeout.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,21 @@ +/* Copyright 2002, 2003, 2004 by Hans Reiser, licensing governed by reiser4/README */ + +#if !defined (__FS_REISER4_WRITEOUT_H__) + +#define WRITEOUT_SINGLE_STREAM (0x1) +#define WRITEOUT_FOR_PAGE_RECLAIM (0x2) +#define WRITEOUT_BARRIER (0x4) + +extern int reiser4_get_writeout_flags(void); + +#endif /* __FS_REISER4_WRITEOUT_H__ */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 80 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/znode.c linux-2.6.30/fs/reiser4/znode.c --- linux-2.6.30.orig/fs/reiser4/znode.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/znode.c 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,1029 @@ +/* Copyright 2001, 2002, 2003 by Hans Reiser, licensing governed by + * reiser4/README */ +/* Znode manipulation functions. */ +/* Znode is the in-memory header for a tree node. It is stored + separately from the node itself so that it does not get written to + disk. In this respect znode is like buffer head or page head. We + also use znodes for additional reiser4 specific purposes: + + . they are organized into tree structure which is a part of whole + reiser4 tree. + . they are used to implement node grained locking + . they are used to keep additional state associated with a + node + . they contain links to lists used by the transaction manager + + Znode is attached to some variable "block number" which is instance of + fs/reiser4/tree.h:reiser4_block_nr type. Znode can exist without + appropriate node being actually loaded in memory. Existence of znode itself + is regulated by reference count (->x_count) in it. Each time thread + acquires reference to znode through call to zget(), ->x_count is + incremented and decremented on call to zput(). Data (content of node) are + brought in memory through call to zload(), which also increments ->d_count + reference counter. zload can block waiting on IO. Call to zrelse() + decreases this counter. Also, ->c_count keeps track of number of child + znodes and prevents parent znode from being recycled until all of its + children are. ->c_count is decremented whenever child goes out of existence + (being actually recycled in zdestroy()) which can be some time after last + reference to this child dies if we support some form of LRU cache for + znodes. + +*/ +/* EVERY ZNODE'S STORY + + 1. His infancy. + + Once upon a time, the znode was born deep inside of zget() by call to + zalloc(). At the return from zget() znode had: + + . reference counter (x_count) of 1 + . assigned block number, marked as used in bitmap + . pointer to parent znode. Root znode parent pointer points + to its father: "fake" znode. This, in turn, has NULL parent pointer. + . hash table linkage + . no data loaded from disk + . no node plugin + . no sibling linkage + + 2. His childhood + + Each node is either brought into memory as a result of tree traversal, or + created afresh, creation of the root being a special case of the latter. In + either case it's inserted into sibling list. This will typically require + some ancillary tree traversing, but ultimately both sibling pointers will + exist and JNODE_LEFT_CONNECTED and JNODE_RIGHT_CONNECTED will be true in + zjnode.state. + + 3. His youth. + + If znode is bound to already existing node in a tree, its content is read + from the disk by call to zload(). At that moment, JNODE_LOADED bit is set + in zjnode.state and zdata() function starts to return non null for this + znode. zload() further calls zparse() that determines which node layout + this node is rendered in, and sets ->nplug on success. + + If znode is for new node just created, memory for it is allocated and + zinit_new() function is called to initialise data, according to selected + node layout. + + 4. His maturity. + + After this point, znode lingers in memory for some time. Threads can + acquire references to znode either by blocknr through call to zget(), or by + following a pointer to unallocated znode from internal item. Each time + reference to znode is obtained, x_count is increased. Thread can read/write + lock znode. Znode data can be loaded through calls to zload(), d_count will + be increased appropriately. If all references to znode are released + (x_count drops to 0), znode is not recycled immediately. Rather, it is + still cached in the hash table in the hope that it will be accessed + shortly. + + There are two ways in which znode existence can be terminated: + + . sudden death: node bound to this znode is removed from the tree + . overpopulation: znode is purged out of memory due to memory pressure + + 5. His death. + + Death is complex process. + + When we irrevocably commit ourselves to decision to remove node from the + tree, JNODE_HEARD_BANSHEE bit is set in zjnode.state of corresponding + znode. This is done either in ->kill_hook() of internal item or in + reiser4_kill_root() function when tree root is removed. + + At this moment znode still has: + + . locks held on it, necessary write ones + . references to it + . disk block assigned to it + . data loaded from the disk + . pending requests for lock + + But once JNODE_HEARD_BANSHEE bit set, last call to unlock_znode() does node + deletion. Node deletion includes two phases. First all ways to get + references to that znode (sibling and parent links and hash lookup using + block number stored in parent node) should be deleted -- it is done through + sibling_list_remove(), also we assume that nobody uses down link from + parent node due to its nonexistence or proper parent node locking and + nobody uses parent pointers from children due to absence of them. Second we + invalidate all pending lock requests which still are on znode's lock + request queue, this is done by reiser4_invalidate_lock(). Another + JNODE_IS_DYING znode status bit is used to invalidate pending lock requests. + Once it set all requesters are forced to return -EINVAL from + longterm_lock_znode(). Future locking attempts are not possible because all + ways to get references to that znode are removed already. Last, node is + uncaptured from transaction. + + When last reference to the dying znode is just about to be released, + block number for this lock is released and znode is removed from the + hash table. + + Now znode can be recycled. + + [it's possible to free bitmap block and remove znode from the hash + table when last lock is released. This will result in having + referenced but completely orphaned znode] + + 6. Limbo + + As have been mentioned above znodes with reference counter 0 are + still cached in a hash table. Once memory pressure increases they are + purged out of there [this requires something like LRU list for + efficient implementation. LRU list would also greatly simplify + implementation of coord cache that would in this case morph to just + scanning some initial segment of LRU list]. Data loaded into + unreferenced znode are flushed back to the durable storage if + necessary and memory is freed. Znodes themselves can be recycled at + this point too. + +*/ + +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "plugin/plugin_header.h" +#include "plugin/node/node.h" +#include "plugin/plugin.h" +#include "txnmgr.h" +#include "jnode.h" +#include "znode.h" +#include "block_alloc.h" +#include "tree.h" +#include "tree_walk.h" +#include "super.h" +#include "reiser4.h" + +#include +#include +#include +#include + +static z_hash_table *get_htable(reiser4_tree *, + const reiser4_block_nr * const blocknr); +static z_hash_table *znode_get_htable(const znode *); +static void zdrop(znode *); + +/* hash table support */ + +/* compare two block numbers for equality. Used by hash-table macros */ +static inline int +blknreq(const reiser4_block_nr * b1, const reiser4_block_nr * b2) +{ + assert("nikita-534", b1 != NULL); + assert("nikita-535", b2 != NULL); + + return *b1 == *b2; +} + +/* Hash znode by block number. Used by hash-table macros */ +/* Audited by: umka (2002.06.11) */ +static inline __u32 +blknrhashfn(z_hash_table * table, const reiser4_block_nr * b) +{ + assert("nikita-536", b != NULL); + + return *b & (REISER4_ZNODE_HASH_TABLE_SIZE - 1); +} + +/* The hash table definition */ +#define KMALLOC(size) kmalloc((size), reiser4_ctx_gfp_mask_get()) +#define KFREE(ptr, size) kfree(ptr) +TYPE_SAFE_HASH_DEFINE(z, znode, reiser4_block_nr, zjnode.key.z, zjnode.link.z, + blknrhashfn, blknreq); +#undef KFREE +#undef KMALLOC + +/* slab for znodes */ +static struct kmem_cache *znode_cache; + +int znode_shift_order; + +/** + * init_znodes - create znode cache + * + * Initializes slab cache of znodes. It is part of reiser4 module initialization. + */ +int init_znodes(void) +{ + znode_cache = kmem_cache_create("znode", sizeof(znode), 0, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT, NULL); + if (znode_cache == NULL) + return RETERR(-ENOMEM); + + for (znode_shift_order = 0; (1 << znode_shift_order) < sizeof(znode); + ++znode_shift_order); + --znode_shift_order; + return 0; +} + +/** + * done_znodes - delete znode cache + * + * This is called on reiser4 module unloading or system shutdown. + */ +void done_znodes(void) +{ + destroy_reiser4_cache(&znode_cache); +} + +/* call this to initialise tree of znodes */ +int znodes_tree_init(reiser4_tree * tree /* tree to initialise znodes for */ ) +{ + int result; + assert("umka-050", tree != NULL); + + rwlock_init(&tree->dk_lock); + + result = z_hash_init(&tree->zhash_table, REISER4_ZNODE_HASH_TABLE_SIZE); + if (result != 0) + return result; + result = z_hash_init(&tree->zfake_table, REISER4_ZNODE_HASH_TABLE_SIZE); + return result; +} + +/* free this znode */ +void zfree(znode * node /* znode to free */ ) +{ + assert("nikita-465", node != NULL); + assert("nikita-2120", znode_page(node) == NULL); + assert("nikita-2301", list_empty_careful(&node->lock.owners)); + assert("nikita-2302", list_empty_careful(&node->lock.requestors)); + assert("nikita-2663", (list_empty_careful(&ZJNODE(node)->capture_link) && + NODE_LIST(ZJNODE(node)) == NOT_CAPTURED)); + assert("nikita-3220", list_empty(&ZJNODE(node)->jnodes)); + assert("nikita-3293", !znode_is_right_connected(node)); + assert("nikita-3294", !znode_is_left_connected(node)); + assert("nikita-3295", node->left == NULL); + assert("nikita-3296", node->right == NULL); + + /* not yet phash_jnode_destroy(ZJNODE(node)); */ + + kmem_cache_free(znode_cache, node); +} + +/* call this to free tree of znodes */ +void znodes_tree_done(reiser4_tree * tree /* tree to finish with znodes of */ ) +{ + znode *node; + znode *next; + z_hash_table *ztable; + + /* scan znode hash-tables and kill all znodes, then free hash tables + * themselves. */ + + assert("nikita-795", tree != NULL); + + ztable = &tree->zhash_table; + + if (ztable->_table != NULL) { + for_all_in_htable(ztable, z, node, next) { + node->c_count = 0; + node->in_parent.node = NULL; + assert("nikita-2179", atomic_read(&ZJNODE(node)->x_count) == 0); + zdrop(node); + } + + z_hash_done(&tree->zhash_table); + } + + ztable = &tree->zfake_table; + + if (ztable->_table != NULL) { + for_all_in_htable(ztable, z, node, next) { + node->c_count = 0; + node->in_parent.node = NULL; + assert("nikita-2179", atomic_read(&ZJNODE(node)->x_count) == 0); + zdrop(node); + } + + z_hash_done(&tree->zfake_table); + } +} + +/* ZNODE STRUCTURES */ + +/* allocate fresh znode */ +znode *zalloc(gfp_t gfp_flag /* allocation flag */ ) +{ + znode *node; + + node = kmem_cache_alloc(znode_cache, gfp_flag); + return node; +} + +/* Initialize fields of znode + @node: znode to initialize; + @parent: parent znode; + @tree: tree we are in. */ +void zinit(znode * node, const znode * parent, reiser4_tree * tree) +{ + assert("nikita-466", node != NULL); + assert("umka-268", current_tree != NULL); + + memset(node, 0, sizeof *node); + + assert("umka-051", tree != NULL); + + jnode_init(&node->zjnode, tree, JNODE_FORMATTED_BLOCK); + reiser4_init_lock(&node->lock); + init_parent_coord(&node->in_parent, parent); +} + +/* + * remove znode from indices. This is called jput() when last reference on + * znode is released. + */ +void znode_remove(znode * node /* znode to remove */ , reiser4_tree * tree) +{ + assert("nikita-2108", node != NULL); + assert("nikita-470", node->c_count == 0); + assert_rw_write_locked(&(tree->tree_lock)); + + /* remove reference to this znode from cbk cache */ + cbk_cache_invalidate(node, tree); + + /* update c_count of parent */ + if (znode_parent(node) != NULL) { + assert("nikita-472", znode_parent(node)->c_count > 0); + /* father, onto your hands I forward my spirit... */ + znode_parent(node)->c_count--; + node->in_parent.node = NULL; + } else { + /* orphaned znode?! Root? */ + } + + /* remove znode from hash-table */ + z_hash_remove_rcu(znode_get_htable(node), node); +} + +/* zdrop() -- Remove znode from the tree. + + This is called when znode is removed from the memory. */ +static void zdrop(znode * node /* znode to finish with */ ) +{ + jdrop(ZJNODE(node)); +} + +/* + * put znode into right place in the hash table. This is called by relocate + * code. + */ +int znode_rehash(znode * node /* node to rehash */ , + const reiser4_block_nr * new_block_nr /* new block number */ ) +{ + z_hash_table *oldtable; + z_hash_table *newtable; + reiser4_tree *tree; + + assert("nikita-2018", node != NULL); + + tree = znode_get_tree(node); + oldtable = znode_get_htable(node); + newtable = get_htable(tree, new_block_nr); + + write_lock_tree(tree); + /* remove znode from hash-table */ + z_hash_remove_rcu(oldtable, node); + + /* assertion no longer valid due to RCU */ + /* assert("nikita-2019", z_hash_find(newtable, new_block_nr) == NULL); */ + + /* update blocknr */ + znode_set_block(node, new_block_nr); + node->zjnode.key.z = *new_block_nr; + + /* insert it into hash */ + z_hash_insert_rcu(newtable, node); + write_unlock_tree(tree); + return 0; +} + +/* ZNODE LOOKUP, GET, PUT */ + +/* zlook() - get znode with given block_nr in a hash table or return NULL + + If result is non-NULL then the znode's x_count is incremented. Internal version + accepts pre-computed hash index. The hash table is accessed under caller's + tree->hash_lock. +*/ +znode *zlook(reiser4_tree * tree, const reiser4_block_nr * const blocknr) +{ + znode *result; + __u32 hash; + z_hash_table *htable; + + assert("jmacd-506", tree != NULL); + assert("jmacd-507", blocknr != NULL); + + htable = get_htable(tree, blocknr); + hash = blknrhashfn(htable, blocknr); + + rcu_read_lock(); + result = z_hash_find_index(htable, hash, blocknr); + + if (result != NULL) { + add_x_ref(ZJNODE(result)); + result = znode_rip_check(tree, result); + } + rcu_read_unlock(); + + return result; +} + +/* return hash table where znode with block @blocknr is (or should be) + * stored */ +static z_hash_table *get_htable(reiser4_tree * tree, + const reiser4_block_nr * const blocknr) +{ + z_hash_table *table; + if (is_disk_addr_unallocated(blocknr)) + table = &tree->zfake_table; + else + table = &tree->zhash_table; + return table; +} + +/* return hash table where znode @node is (or should be) stored */ +static z_hash_table *znode_get_htable(const znode * node) +{ + return get_htable(znode_get_tree(node), znode_get_block(node)); +} + +/* zget() - get znode from hash table, allocating it if necessary. + + First a call to zlook, locating a x-referenced znode if one + exists. If znode is not found, allocate new one and return. Result + is returned with x_count reference increased. + + LOCKS TAKEN: TREE_LOCK, ZNODE_LOCK + LOCK ORDERING: NONE +*/ +znode *zget(reiser4_tree * tree, + const reiser4_block_nr * const blocknr, + znode * parent, tree_level level, gfp_t gfp_flag) +{ + znode *result; + __u32 hashi; + + z_hash_table *zth; + + assert("jmacd-512", tree != NULL); + assert("jmacd-513", blocknr != NULL); + assert("jmacd-514", level < REISER4_MAX_ZTREE_HEIGHT); + + zth = get_htable(tree, blocknr); + hashi = blknrhashfn(zth, blocknr); + + /* NOTE-NIKITA address-as-unallocated-blocknr still is not + implemented. */ + + z_hash_prefetch_bucket(zth, hashi); + + rcu_read_lock(); + /* Find a matching BLOCKNR in the hash table. If the znode is found, + we obtain an reference (x_count) but the znode remains unlocked. + Have to worry about race conditions later. */ + result = z_hash_find_index(zth, hashi, blocknr); + /* According to the current design, the hash table lock protects new + znode references. */ + if (result != NULL) { + add_x_ref(ZJNODE(result)); + /* NOTE-NIKITA it should be so, but special case during + creation of new root makes such assertion highly + complicated. */ + assert("nikita-2131", 1 || znode_parent(result) == parent || + (ZF_ISSET(result, JNODE_ORPHAN) + && (znode_parent(result) == NULL))); + result = znode_rip_check(tree, result); + } + + rcu_read_unlock(); + + if (!result) { + znode *shadow; + + result = zalloc(gfp_flag); + if (!result) { + return ERR_PTR(RETERR(-ENOMEM)); + } + + zinit(result, parent, tree); + ZJNODE(result)->blocknr = *blocknr; + ZJNODE(result)->key.z = *blocknr; + result->level = level; + + write_lock_tree(tree); + + shadow = z_hash_find_index(zth, hashi, blocknr); + if (unlikely(shadow != NULL && !ZF_ISSET(shadow, JNODE_RIP))) { + jnode_list_remove(ZJNODE(result)); + zfree(result); + result = shadow; + } else { + result->version = znode_build_version(tree); + z_hash_insert_index_rcu(zth, hashi, result); + + if (parent != NULL) + ++parent->c_count; + } + + add_x_ref(ZJNODE(result)); + + write_unlock_tree(tree); + } +#if REISER4_DEBUG + if (!reiser4_blocknr_is_fake(blocknr) && *blocknr != 0) + reiser4_check_block(blocknr, 1); +#endif + /* Check for invalid tree level, return -EIO */ + if (unlikely(znode_get_level(result) != level)) { + warning("jmacd-504", + "Wrong level for cached block %llu: %i expecting %i", + (unsigned long long)(*blocknr), znode_get_level(result), + level); + zput(result); + return ERR_PTR(RETERR(-EIO)); + } + + assert("nikita-1227", znode_invariant(result)); + + return result; +} + +/* ZNODE PLUGINS/DATA */ + +/* "guess" plugin for node loaded from the disk. Plugin id of node plugin is + stored at the fixed offset from the beginning of the node. */ +static node_plugin *znode_guess_plugin(const znode * node /* znode to guess + * plugin of */ ) +{ + reiser4_tree *tree; + + assert("nikita-1053", node != NULL); + assert("nikita-1055", zdata(node) != NULL); + + tree = znode_get_tree(node); + assert("umka-053", tree != NULL); + + if (reiser4_is_set(tree->super, REISER4_ONE_NODE_PLUGIN)) { + return tree->nplug; + } else { + return node_plugin_by_disk_id + (tree, &((common_node_header *) zdata(node))->plugin_id); +#ifdef GUESS_EXISTS + reiser4_plugin *plugin; + + /* NOTE-NIKITA add locking here when dynamic plugins will be + * implemented */ + for_all_plugins(REISER4_NODE_PLUGIN_TYPE, plugin) { + if ((plugin->u.node.guess != NULL) + && plugin->u.node.guess(node)) + return plugin; + } + warning("nikita-1057", "Cannot guess node plugin"); + print_znode("node", node); + return NULL; +#endif + } +} + +/* parse node header and install ->node_plugin */ +int zparse(znode * node /* znode to parse */ ) +{ + int result; + + assert("nikita-1233", node != NULL); + assert("nikita-2370", zdata(node) != NULL); + + if (node->nplug == NULL) { + node_plugin *nplug; + + nplug = znode_guess_plugin(node); + if (likely(nplug != NULL)) { + result = nplug->parse(node); + if (likely(result == 0)) + node->nplug = nplug; + } else { + result = RETERR(-EIO); + } + } else + result = 0; + return result; +} + +/* zload with readahead */ +int zload_ra(znode * node /* znode to load */ , ra_info_t * info) +{ + int result; + + assert("nikita-484", node != NULL); + assert("nikita-1377", znode_invariant(node)); + assert("jmacd-7771", !znode_above_root(node)); + assert("nikita-2125", atomic_read(&ZJNODE(node)->x_count) > 0); + assert("nikita-3016", reiser4_schedulable()); + + if (info) + formatted_readahead(node, info); + + result = jload(ZJNODE(node)); + assert("nikita-1378", znode_invariant(node)); + return result; +} + +/* load content of node into memory */ +int zload(znode * node) +{ + return zload_ra(node, NULL); +} + +/* call node plugin to initialise newly allocated node. */ +int zinit_new(znode * node /* znode to initialise */ , gfp_t gfp_flags) +{ + return jinit_new(ZJNODE(node), gfp_flags); +} + +/* drop reference to node data. When last reference is dropped, data are + unloaded. */ +void zrelse(znode * node /* znode to release references to */ ) +{ + assert("nikita-1381", znode_invariant(node)); + + jrelse(ZJNODE(node)); +} + +/* returns free space in node */ +unsigned znode_free_space(znode * node /* znode to query */ ) +{ + assert("nikita-852", node != NULL); + return node_plugin_by_node(node)->free_space(node); +} + +/* left delimiting key of znode */ +reiser4_key *znode_get_rd_key(znode * node /* znode to query */ ) +{ + assert("nikita-958", node != NULL); + assert_rw_locked(&(znode_get_tree(node)->dk_lock)); + assert("nikita-3067", LOCK_CNT_GTZ(rw_locked_dk)); + assert("nikita-30671", node->rd_key_version != 0); + return &node->rd_key; +} + +/* right delimiting key of znode */ +reiser4_key *znode_get_ld_key(znode * node /* znode to query */ ) +{ + assert("nikita-974", node != NULL); + assert_rw_locked(&(znode_get_tree(node)->dk_lock)); + assert("nikita-3068", LOCK_CNT_GTZ(rw_locked_dk)); + assert("nikita-30681", node->ld_key_version != 0); + return &node->ld_key; +} + +ON_DEBUG(atomic_t delim_key_version = ATOMIC_INIT(0); + ) + +/* update right-delimiting key of @node */ +reiser4_key *znode_set_rd_key(znode * node, const reiser4_key * key) +{ + assert("nikita-2937", node != NULL); + assert("nikita-2939", key != NULL); + assert_rw_write_locked(&(znode_get_tree(node)->dk_lock)); + assert("nikita-3069", LOCK_CNT_GTZ(write_locked_dk)); + assert("nikita-2944", + znode_is_any_locked(node) || + znode_get_level(node) != LEAF_LEVEL || + keyge(key, &node->rd_key) || + keyeq(&node->rd_key, reiser4_min_key()) || + ZF_ISSET(node, JNODE_HEARD_BANSHEE)); + + node->rd_key = *key; + ON_DEBUG(node->rd_key_version = atomic_inc_return(&delim_key_version)); + return &node->rd_key; +} + +/* update left-delimiting key of @node */ +reiser4_key *znode_set_ld_key(znode * node, const reiser4_key * key) +{ + assert("nikita-2940", node != NULL); + assert("nikita-2941", key != NULL); + assert_rw_write_locked(&(znode_get_tree(node)->dk_lock)); + assert("nikita-3070", LOCK_CNT_GTZ(write_locked_dk)); + assert("nikita-2943", + znode_is_any_locked(node) || keyeq(&node->ld_key, + reiser4_min_key())); + + node->ld_key = *key; + ON_DEBUG(node->ld_key_version = atomic_inc_return(&delim_key_version)); + return &node->ld_key; +} + +/* true if @key is inside key range for @node */ +int znode_contains_key(znode * node /* znode to look in */ , + const reiser4_key * key /* key to look for */ ) +{ + assert("nikita-1237", node != NULL); + assert("nikita-1238", key != NULL); + + /* left_delimiting_key <= key <= right_delimiting_key */ + return keyle(znode_get_ld_key(node), key) + && keyle(key, znode_get_rd_key(node)); +} + +/* same as znode_contains_key(), but lock dk lock */ +int znode_contains_key_lock(znode * node /* znode to look in */ , + const reiser4_key * key /* key to look for */ ) +{ + int result; + + assert("umka-056", node != NULL); + assert("umka-057", key != NULL); + + read_lock_dk(znode_get_tree(node)); + result = znode_contains_key(node, key); + read_unlock_dk(znode_get_tree(node)); + return result; +} + +/* get parent pointer, assuming tree is not locked */ +znode *znode_parent_nolock(const znode * node /* child znode */ ) +{ + assert("nikita-1444", node != NULL); + return node->in_parent.node; +} + +/* get parent pointer of znode */ +znode *znode_parent(const znode * node /* child znode */ ) +{ + assert("nikita-1226", node != NULL); + assert("nikita-1406", LOCK_CNT_GTZ(rw_locked_tree)); + return znode_parent_nolock(node); +} + +/* detect uber znode used to protect in-superblock tree root pointer */ +int znode_above_root(const znode * node /* znode to query */ ) +{ + assert("umka-059", node != NULL); + + return disk_addr_eq(&ZJNODE(node)->blocknr, &UBER_TREE_ADDR); +} + +/* check that @node is root---that its block number is recorder in the tree as + that of root node */ +#if REISER4_DEBUG +static int znode_is_true_root(const znode * node /* znode to query */ ) +{ + assert("umka-060", node != NULL); + assert("umka-061", current_tree != NULL); + + return disk_addr_eq(znode_get_block(node), + &znode_get_tree(node)->root_block); +} +#endif + +/* check that @node is root */ +int znode_is_root(const znode * node /* znode to query */ ) +{ + assert("nikita-1206", node != NULL); + + return znode_get_level(node) == znode_get_tree(node)->height; +} + +/* Returns true is @node was just created by zget() and wasn't ever loaded + into memory. */ +/* NIKITA-HANS: yes */ +int znode_just_created(const znode * node) +{ + assert("nikita-2188", node != NULL); + return (znode_page(node) == NULL); +} + +/* obtain updated ->znode_epoch. See seal.c for description. */ +__u64 znode_build_version(reiser4_tree * tree) +{ + __u64 result; + + spin_lock(&tree->epoch_lock); + result = ++tree->znode_epoch; + spin_unlock(&tree->epoch_lock); + return result; +} + +void init_load_count(load_count * dh) +{ + assert("nikita-2105", dh != NULL); + memset(dh, 0, sizeof *dh); +} + +void done_load_count(load_count * dh) +{ + assert("nikita-2106", dh != NULL); + if (dh->node != NULL) { + for (; dh->d_ref > 0; --dh->d_ref) + zrelse(dh->node); + dh->node = NULL; + } +} + +static int incr_load_count(load_count * dh) +{ + int result; + + assert("nikita-2110", dh != NULL); + assert("nikita-2111", dh->node != NULL); + + result = zload(dh->node); + if (result == 0) + ++dh->d_ref; + return result; +} + +int incr_load_count_znode(load_count * dh, znode * node) +{ + assert("nikita-2107", dh != NULL); + assert("nikita-2158", node != NULL); + assert("nikita-2109", + ergo(dh->node != NULL, (dh->node == node) || (dh->d_ref == 0))); + + dh->node = node; + return incr_load_count(dh); +} + +int incr_load_count_jnode(load_count * dh, jnode * node) +{ + if (jnode_is_znode(node)) { + return incr_load_count_znode(dh, JZNODE(node)); + } + return 0; +} + +void copy_load_count(load_count * new, load_count * old) +{ + int ret = 0; + done_load_count(new); + new->node = old->node; + new->d_ref = 0; + + while ((new->d_ref < old->d_ref) && (ret = incr_load_count(new)) == 0) { + } + + assert("jmacd-87589", ret == 0); +} + +void move_load_count(load_count * new, load_count * old) +{ + done_load_count(new); + new->node = old->node; + new->d_ref = old->d_ref; + old->node = NULL; + old->d_ref = 0; +} + +/* convert parent pointer into coord */ +void parent_coord_to_coord(const parent_coord_t * pcoord, coord_t * coord) +{ + assert("nikita-3204", pcoord != NULL); + assert("nikita-3205", coord != NULL); + + coord_init_first_unit_nocheck(coord, pcoord->node); + coord_set_item_pos(coord, pcoord->item_pos); + coord->between = AT_UNIT; +} + +/* pack coord into parent_coord_t */ +void coord_to_parent_coord(const coord_t * coord, parent_coord_t * pcoord) +{ + assert("nikita-3206", pcoord != NULL); + assert("nikita-3207", coord != NULL); + + pcoord->node = coord->node; + pcoord->item_pos = coord->item_pos; +} + +/* Initialize a parent hint pointer. (parent hint pointer is a field in znode, + look for comments there) */ +void init_parent_coord(parent_coord_t * pcoord, const znode * node) +{ + pcoord->node = (znode *) node; + pcoord->item_pos = (unsigned short)~0; +} + +#if REISER4_DEBUG + +/* debugging aid: znode invariant */ +static int znode_invariant_f(const znode * node /* znode to check */ , + char const **msg /* where to store error + * message, if any */ ) +{ +#define _ergo(ant, con) \ + ((*msg) = "{" #ant "} ergo {" #con "}", ergo((ant), (con))) + +#define _equi(e1, e2) \ + ((*msg) = "{" #e1 "} <=> {" #e2 "}", equi((e1), (e2))) + +#define _check(exp) ((*msg) = #exp, (exp)) + + return jnode_invariant_f(ZJNODE(node), msg) && + /* [znode-fake] invariant */ + /* fake znode doesn't have a parent, and */ + _ergo(znode_get_level(node) == 0, znode_parent(node) == NULL) && + /* there is another way to express this very check, and */ + _ergo(znode_above_root(node), znode_parent(node) == NULL) && + /* it has special block number, and */ + _ergo(znode_get_level(node) == 0, + disk_addr_eq(znode_get_block(node), &UBER_TREE_ADDR)) && + /* it is the only znode with such block number, and */ + _ergo(!znode_above_root(node) && znode_is_loaded(node), + !disk_addr_eq(znode_get_block(node), &UBER_TREE_ADDR)) && + /* it is parent of the tree root node */ + _ergo(znode_is_true_root(node), + znode_above_root(znode_parent(node))) && + /* [znode-level] invariant */ + /* level of parent znode is one larger than that of child, + except for the fake znode, and */ + _ergo(znode_parent(node) && !znode_above_root(znode_parent(node)), + znode_get_level(znode_parent(node)) == + znode_get_level(node) + 1) && + /* left neighbor is at the same level, and */ + _ergo(znode_is_left_connected(node) && node->left != NULL, + znode_get_level(node) == znode_get_level(node->left)) && + /* right neighbor is at the same level */ + _ergo(znode_is_right_connected(node) && node->right != NULL, + znode_get_level(node) == znode_get_level(node->right)) && + /* [znode-connected] invariant */ + _ergo(node->left != NULL, znode_is_left_connected(node)) && + _ergo(node->right != NULL, znode_is_right_connected(node)) && + _ergo(!znode_is_root(node) && node->left != NULL, + znode_is_right_connected(node->left) && + node->left->right == node) && + _ergo(!znode_is_root(node) && node->right != NULL, + znode_is_left_connected(node->right) && + node->right->left == node) && + /* [znode-c_count] invariant */ + /* for any znode, c_count of its parent is greater than 0 */ + _ergo(znode_parent(node) != NULL && + !znode_above_root(znode_parent(node)), + znode_parent(node)->c_count > 0) && + /* leaves don't have children */ + _ergo(znode_get_level(node) == LEAF_LEVEL, + node->c_count == 0) && + _check(node->zjnode.jnodes.prev != NULL) && + _check(node->zjnode.jnodes.next != NULL) && + /* orphan doesn't have a parent */ + _ergo(ZF_ISSET(node, JNODE_ORPHAN), znode_parent(node) == 0) && + /* [znode-modify] invariant */ + /* if znode is not write-locked, its checksum remains + * invariant */ + /* unfortunately, zlock is unordered w.r.t. jnode_lock, so we + * cannot check this. */ + /* [znode-refs] invariant */ + /* only referenced znode can be long-term locked */ + _ergo(znode_is_locked(node), + atomic_read(&ZJNODE(node)->x_count) != 0); +} + +/* debugging aid: check znode invariant and panic if it doesn't hold */ +int znode_invariant(znode * node /* znode to check */ ) +{ + char const *failed_msg; + int result; + + assert("umka-063", node != NULL); + assert("umka-064", current_tree != NULL); + + spin_lock_znode(node); + read_lock_tree(znode_get_tree(node)); + result = znode_invariant_f(node, &failed_msg); + if (!result) { + /* print_znode("corrupted node", node); */ + warning("jmacd-555", "Condition %s failed", failed_msg); + } + read_unlock_tree(znode_get_tree(node)); + spin_unlock_znode(node); + return result; +} + +/* return non-0 iff data are loaded into znode */ +int znode_is_loaded(const znode * node /* znode to query */ ) +{ + assert("nikita-497", node != NULL); + return jnode_is_loaded(ZJNODE(node)); +} + +unsigned long znode_times_locked(const znode * z) +{ + return z->times_locked; +} + +#endif /* REISER4_DEBUG */ + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/fs/reiser4/znode.h linux-2.6.30/fs/reiser4/znode.h --- linux-2.6.30.orig/fs/reiser4/znode.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.30/fs/reiser4/znode.h 2009-06-22 16:08:13.000000000 +0200 @@ -0,0 +1,433 @@ +/* Copyright 2001, 2002, 2003, 2004 by Hans Reiser, licensing governed by + * reiser4/README */ + +/* Declaration of znode (Zam's node). See znode.c for more details. */ + +#ifndef __ZNODE_H__ +#define __ZNODE_H__ + +#include "forward.h" +#include "debug.h" +#include "dformat.h" +#include "key.h" +#include "coord.h" +#include "plugin/node/node.h" +#include "jnode.h" +#include "lock.h" +#include "readahead.h" + +#include +#include +#include /* for PAGE_CACHE_SIZE */ +#include + +/* znode tracks its position within parent (internal item in a parent node, + * that contains znode's block number). */ +typedef struct parent_coord { + znode *node; + pos_in_node_t item_pos; +} parent_coord_t; + +/* &znode - node in a reiser4 tree. + + NOTE-NIKITA fields in this struct have to be rearranged (later) to reduce + cacheline pressure. + + Locking: + + Long term: data in a disk node attached to this znode are protected + by long term, deadlock aware lock ->lock; + + Spin lock: the following fields are protected by the spin lock: + + ->lock + + Following fields are protected by the global tree lock: + + ->left + ->right + ->in_parent + ->c_count + + Following fields are protected by the global delimiting key lock (dk_lock): + + ->ld_key (to update ->ld_key long-term lock on the node is also required) + ->rd_key + + Following fields are protected by the long term lock: + + ->nr_items + + ->node_plugin is never changed once set. This means that after code made + itself sure that field is valid it can be accessed without any additional + locking. + + ->level is immutable. + + Invariants involving this data-type: + + [znode-fake] + [znode-level] + [znode-connected] + [znode-c_count] + [znode-refs] + [jnode-refs] + [jnode-queued] + [znode-modify] + + For this to be made into a clustering or NUMA filesystem, we would want to eliminate all of the global locks. + Suggestions for how to do that are desired.*/ +struct znode { + /* Embedded jnode. */ + jnode zjnode; + + /* contains three subfields, node, pos_in_node, and pos_in_unit. + + pos_in_node and pos_in_unit are only hints that are cached to + speed up lookups during balancing. They are not required to be up to + date. Synched in find_child_ptr(). + + This value allows us to avoid expensive binary searches. + + in_parent->node points to the parent of this node, and is NOT a + hint. + */ + parent_coord_t in_parent; + + /* + * sibling list pointers + */ + + /* left-neighbor */ + znode *left; + /* right-neighbor */ + znode *right; + + /* long term lock on node content. This lock supports deadlock + detection. See lock.c + */ + zlock lock; + + /* You cannot remove from memory a node that has children in + memory. This is because we rely on the fact that parent of given + node can always be reached without blocking for io. When reading a + node into memory you must increase the c_count of its parent, when + removing it from memory you must decrease the c_count. This makes + the code simpler, and the cases where it is suboptimal are truly + obscure. + */ + int c_count; + + /* plugin of node attached to this znode. NULL if znode is not + loaded. */ + node_plugin *nplug; + + /* version of znode data. This is increased on each modification. This + * is necessary to implement seals (see seal.[ch]) efficiently. */ + __u64 version; + + /* left delimiting key. Necessary to efficiently perform + balancing with node-level locking. Kept in memory only. */ + reiser4_key ld_key; + /* right delimiting key. */ + reiser4_key rd_key; + + /* znode's tree level */ + __u16 level; + /* number of items in this node. This field is modified by node + * plugin. */ + __u16 nr_items; + +#if REISER4_DEBUG + void *creator; + reiser4_key first_key; + unsigned long times_locked; + int left_version; /* when node->left was updated */ + int right_version; /* when node->right was updated */ + int ld_key_version; /* when node->ld_key was updated */ + int rd_key_version; /* when node->rd_key was updated */ +#endif + +} __attribute__ ((aligned(16))); + +ON_DEBUG(extern atomic_t delim_key_version; + ) + +/* In general I think these macros should not be exposed. */ +#define znode_is_locked(node) (lock_is_locked(&node->lock)) +#define znode_is_rlocked(node) (lock_is_rlocked(&node->lock)) +#define znode_is_wlocked(node) (lock_is_wlocked(&node->lock)) +#define znode_is_wlocked_once(node) (lock_is_wlocked_once(&node->lock)) +#define znode_can_be_rlocked(node) (lock_can_be_rlocked(&node->lock)) +#define is_lock_compatible(node, mode) (lock_mode_compatible(&node->lock, mode)) +/* Macros for accessing the znode state. */ +#define ZF_CLR(p,f) JF_CLR (ZJNODE(p), (f)) +#define ZF_ISSET(p,f) JF_ISSET(ZJNODE(p), (f)) +#define ZF_SET(p,f) JF_SET (ZJNODE(p), (f)) +extern znode *zget(reiser4_tree * tree, const reiser4_block_nr * const block, + znode * parent, tree_level level, gfp_t gfp_flag); +extern znode *zlook(reiser4_tree * tree, const reiser4_block_nr * const block); +extern int zload(znode * node); +extern int zload_ra(znode * node, ra_info_t * info); +extern int zinit_new(znode * node, gfp_t gfp_flags); +extern void zrelse(znode * node); +extern void znode_change_parent(znode * new_parent, reiser4_block_nr * block); + +/* size of data in znode */ +static inline unsigned +znode_size(const znode * node UNUSED_ARG /* znode to query */ ) +{ + assert("nikita-1416", node != NULL); + return PAGE_CACHE_SIZE; +} + +extern void parent_coord_to_coord(const parent_coord_t * pcoord, + coord_t * coord); +extern void coord_to_parent_coord(const coord_t * coord, + parent_coord_t * pcoord); +extern void init_parent_coord(parent_coord_t * pcoord, const znode * node); + +extern unsigned znode_free_space(znode * node); + +extern reiser4_key *znode_get_rd_key(znode * node); +extern reiser4_key *znode_get_ld_key(znode * node); + +extern reiser4_key *znode_set_rd_key(znode * node, const reiser4_key * key); +extern reiser4_key *znode_set_ld_key(znode * node, const reiser4_key * key); + +/* `connected' state checks */ +static inline int znode_is_right_connected(const znode * node) +{ + return ZF_ISSET(node, JNODE_RIGHT_CONNECTED); +} + +static inline int znode_is_left_connected(const znode * node) +{ + return ZF_ISSET(node, JNODE_LEFT_CONNECTED); +} + +static inline int znode_is_connected(const znode * node) +{ + return znode_is_right_connected(node) && znode_is_left_connected(node); +} + +extern int znode_shift_order; +extern int znode_rehash(znode * node, const reiser4_block_nr * new_block_nr); +extern void znode_remove(znode *, reiser4_tree *); +extern znode *znode_parent(const znode * node); +extern znode *znode_parent_nolock(const znode * node); +extern int znode_above_root(const znode * node); +extern int init_znodes(void); +extern void done_znodes(void); +extern int znodes_tree_init(reiser4_tree * ztree); +extern void znodes_tree_done(reiser4_tree * ztree); +extern int znode_contains_key(znode * node, const reiser4_key * key); +extern int znode_contains_key_lock(znode * node, const reiser4_key * key); +extern unsigned znode_save_free_space(znode * node); +extern unsigned znode_recover_free_space(znode * node); +extern znode *zalloc(gfp_t gfp_flag); +extern void zinit(znode *, const znode * parent, reiser4_tree *); +extern int zparse(znode * node); + +extern int znode_just_created(const znode * node); + +extern void zfree(znode * node); + +#if REISER4_DEBUG +extern void print_znode(const char *prefix, const znode * node); +#else +#define print_znode( p, n ) noop +#endif + +/* Make it look like various znode functions exist instead of treating znodes as + jnodes in znode-specific code. */ +#define znode_page(x) jnode_page ( ZJNODE(x) ) +#define zdata(x) jdata ( ZJNODE(x) ) +#define znode_get_block(x) jnode_get_block ( ZJNODE(x) ) +#define znode_created(x) jnode_created ( ZJNODE(x) ) +#define znode_set_created(x) jnode_set_created ( ZJNODE(x) ) +#define znode_convertible(x) jnode_convertible (ZJNODE(x)) +#define znode_set_convertible(x) jnode_set_convertible (ZJNODE(x)) + +#define znode_is_dirty(x) jnode_is_dirty ( ZJNODE(x) ) +#define znode_check_dirty(x) jnode_check_dirty ( ZJNODE(x) ) +#define znode_make_clean(x) jnode_make_clean ( ZJNODE(x) ) +#define znode_set_block(x, b) jnode_set_block ( ZJNODE(x), (b) ) + +#define spin_lock_znode(x) spin_lock_jnode ( ZJNODE(x) ) +#define spin_unlock_znode(x) spin_unlock_jnode ( ZJNODE(x) ) +#define spin_trylock_znode(x) spin_trylock_jnode ( ZJNODE(x) ) +#define spin_znode_is_locked(x) spin_jnode_is_locked ( ZJNODE(x) ) +#define spin_znode_is_not_locked(x) spin_jnode_is_not_locked ( ZJNODE(x) ) + +#if REISER4_DEBUG +extern int znode_x_count_is_protected(const znode * node); +extern int znode_invariant(znode * node); +#endif + +/* acquire reference to @node */ +static inline znode *zref(znode * node) +{ + /* change of x_count from 0 to 1 is protected by tree spin-lock */ + return JZNODE(jref(ZJNODE(node))); +} + +/* release reference to @node */ +static inline void zput(znode * node) +{ + assert("nikita-3564", znode_invariant(node)); + jput(ZJNODE(node)); +} + +/* get the level field for a znode */ +static inline tree_level znode_get_level(const znode * node) +{ + return node->level; +} + +/* get the level field for a jnode */ +static inline tree_level jnode_get_level(const jnode * node) +{ + if (jnode_is_znode(node)) + return znode_get_level(JZNODE(node)); + else + /* unformatted nodes are all at the LEAF_LEVEL and for + "semi-formatted" nodes like bitmaps, level doesn't matter. */ + return LEAF_LEVEL; +} + +/* true if jnode is on leaf level */ +static inline int jnode_is_leaf(const jnode * node) +{ + if (jnode_is_znode(node)) + return (znode_get_level(JZNODE(node)) == LEAF_LEVEL); + if (jnode_get_type(node) == JNODE_UNFORMATTED_BLOCK) + return 1; + return 0; +} + +/* return znode's tree */ +static inline reiser4_tree *znode_get_tree(const znode * node) +{ + assert("nikita-2692", node != NULL); + return jnode_get_tree(ZJNODE(node)); +} + +/* resolve race with zput */ +static inline znode *znode_rip_check(reiser4_tree * tree, znode * node) +{ + jnode *j; + + j = jnode_rip_sync(tree, ZJNODE(node)); + if (likely(j != NULL)) + node = JZNODE(j); + else + node = NULL; + return node; +} + +#if defined(REISER4_DEBUG) +int znode_is_loaded(const znode * node /* znode to query */ ); +#endif + +extern __u64 znode_build_version(reiser4_tree * tree); + +/* Data-handles. A data handle object manages pairing calls to zload() and zrelse(). We + must load the data for a node in many places. We could do this by simply calling + zload() everywhere, the difficulty arises when we must release the loaded data by + calling zrelse. In a function with many possible error/return paths, it requires extra + work to figure out which exit paths must call zrelse and those which do not. The data + handle automatically calls zrelse for every zload that it is responsible for. In that + sense, it acts much like a lock_handle. +*/ +typedef struct load_count { + znode *node; + int d_ref; +} load_count; + +extern void init_load_count(load_count * lc); /* Initialize a load_count set the current node to NULL. */ +extern void done_load_count(load_count * dh); /* Finalize a load_count: call zrelse() if necessary */ +extern int incr_load_count_znode(load_count * dh, znode * node); /* Set the argument znode to the current node, call zload(). */ +extern int incr_load_count_jnode(load_count * dh, jnode * node); /* If the argument jnode is formatted, do the same as + * incr_load_count_znode, otherwise do nothing (unformatted nodes + * don't require zload/zrelse treatment). */ +extern void move_load_count(load_count * new, load_count * old); /* Move the contents of a load_count. Old handle is released. */ +extern void copy_load_count(load_count * new, load_count * old); /* Copy the contents of a load_count. Old handle remains held. */ + +/* Variable initializers for load_count. */ +#define INIT_LOAD_COUNT ( load_count * ){ .node = NULL, .d_ref = 0 } +#define INIT_LOAD_COUNT_NODE( n ) ( load_count ){ .node = ( n ), .d_ref = 0 } +/* A convenience macro for use in assertions or debug-only code, where loaded + data is only required to perform the debugging check. This macro + encapsulates an expression inside a pair of calls to zload()/zrelse(). */ +#define WITH_DATA( node, exp ) \ +({ \ + long __with_dh_result; \ + znode *__with_dh_node; \ + \ + __with_dh_node = ( node ); \ + __with_dh_result = zload( __with_dh_node ); \ + if( __with_dh_result == 0 ) { \ + __with_dh_result = ( long )( exp ); \ + zrelse( __with_dh_node ); \ + } \ + __with_dh_result; \ +}) + +/* Same as above, but accepts a return value in case zload fails. */ +#define WITH_DATA_RET( node, ret, exp ) \ +({ \ + int __with_dh_result; \ + znode *__with_dh_node; \ + \ + __with_dh_node = ( node ); \ + __with_dh_result = zload( __with_dh_node ); \ + if( __with_dh_result == 0 ) { \ + __with_dh_result = ( int )( exp ); \ + zrelse( __with_dh_node ); \ + } else \ + __with_dh_result = ( ret ); \ + __with_dh_result; \ +}) + +#define WITH_COORD(coord, exp) \ +({ \ + coord_t *__coord; \ + \ + __coord = (coord); \ + coord_clear_iplug(__coord); \ + WITH_DATA(__coord->node, exp); \ +}) + +#if REISER4_DEBUG +#define STORE_COUNTERS \ + reiser4_lock_cnt_info __entry_counters = \ + *reiser4_lock_counters() +#define CHECK_COUNTERS \ +ON_DEBUG_CONTEXT( \ +({ \ + __entry_counters.x_refs = reiser4_lock_counters() -> x_refs; \ + __entry_counters.t_refs = reiser4_lock_counters() -> t_refs; \ + __entry_counters.d_refs = reiser4_lock_counters() -> d_refs; \ + assert("nikita-2159", \ + !memcmp(&__entry_counters, reiser4_lock_counters(), \ + sizeof __entry_counters)); \ +}) ) + +#else +#define STORE_COUNTERS +#define CHECK_COUNTERS noop +#endif + +/* __ZNODE_H__ */ +#endif + +/* Make Linus happy. + Local variables: + c-indentation-style: "K&R" + mode-name: "LC" + c-basic-offset: 8 + tab-width: 8 + fill-column: 120 + End: +*/ diff -urN linux-2.6.30.orig/include/linux/fs.h linux-2.6.30/include/linux/fs.h --- linux-2.6.30.orig/include/linux/fs.h 2009-06-23 00:20:41.000000000 +0200 +++ linux-2.6.30/include/linux/fs.h 2009-06-22 16:08:13.000000000 +0200 @@ -1571,6 +1571,8 @@ void (*clear_inode) (struct inode *); void (*umount_begin) (struct super_block *); + void (*sync_inodes) (struct super_block *sb, + struct writeback_control *wbc); int (*show_options)(struct seq_file *, struct vfsmount *); int (*show_stats)(struct seq_file *, struct vfsmount *); #ifdef CONFIG_QUOTA @@ -2067,6 +2069,7 @@ extern void generic_sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc); extern int write_inode_now(struct inode *, int); +extern void generic_sync_sb_inodes(struct super_block *, struct writeback_control *); extern int filemap_fdatawrite(struct address_space *); extern int filemap_flush(struct address_space *); extern int filemap_fdatawait(struct address_space *); diff -urN linux-2.6.30.orig/include/linux/mm.h linux-2.6.30/include/linux/mm.h --- linux-2.6.30.orig/include/linux/mm.h 2009-06-23 00:20:41.000000000 +0200 +++ linux-2.6.30/include/linux/mm.h 2009-06-22 16:17:44.000000000 +0200 @@ -838,6 +838,7 @@ void account_page_dirtied(struct page *page, struct address_space *mapping); int set_page_dirty(struct page *page); int set_page_dirty_lock(struct page *page); +int set_page_dirty_notag(struct page *page); int clear_page_dirty_for_io(struct page *page); extern unsigned long move_page_tables(struct vm_area_struct *vma, diff -urN linux-2.6.30.orig/mm/filemap.c linux-2.6.30/mm/filemap.c --- linux-2.6.30.orig/mm/filemap.c 2009-06-23 00:20:41.000000000 +0200 +++ linux-2.6.30/mm/filemap.c 2009-06-22 16:51:45.000000000 +0200 @@ -134,6 +134,7 @@ dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); } } +EXPORT_SYMBOL(__remove_from_page_cache); void remove_from_page_cache(struct page *page) { @@ -146,6 +147,7 @@ spin_unlock_irq(&mapping->tree_lock); mem_cgroup_uncharge_cache_page(page); } +EXPORT_SYMBOL(remove_from_page_cache); static int sync_page(void *word) { @@ -1009,6 +1011,7 @@ ra->ra_pages /= 4; } +EXPORT_SYMBOL(find_get_pages); /** * do_generic_file_read - generic file read routine diff -urN linux-2.6.30.orig/mm/page-writeback.c linux-2.6.30/mm/page-writeback.c --- linux-2.6.30.orig/mm/page-writeback.c 2009-06-23 00:20:41.000000000 +0200 +++ linux-2.6.30/mm/page-writeback.c 2009-06-22 16:53:49.000000000 +0200 @@ -1258,6 +1258,32 @@ EXPORT_SYMBOL(__set_page_dirty_nobuffers); /* + * set_page_dirty_notag() -- similar to __set_page_dirty_nobuffers() + * except it doesn't tag the page dirty in the page-cache radix tree. + * This means that the address space using this cannot use the regular + * filemap ->writepages() helpers and must provide its own means of + * tracking and finding non-tagged dirty pages. + * + * NOTE: furthermore, this version also doesn't handle truncate races. + */ +int set_page_dirty_notag(struct page *page) +{ + struct address_space *mapping = page->mapping; + + if (!TestSetPageDirty(page)) { + unsigned long flags; + WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); + local_irq_save(flags); + account_page_dirtied(page, mapping); + local_irq_restore(flags); + __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); + return 1; + } + return 0; +} +EXPORT_SYMBOL(set_page_dirty_notag); + +/* * When a writepage implementation decides that it doesn't want to write this * page for some reason, it should redirty the locked page via * redirty_page_for_writepage() and it should then unlock the page and return 0