2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
22 #include "transaction.h"
25 #include "print-tree.h"
29 /* magic values for the inode_only field in btrfs_log_inode:
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
35 #define LOG_INODE_ALL 0
36 #define LOG_INODE_EXISTS 1
39 * directory trouble cases
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
48 * rename foo/some_dir foo2/some_dir
50 * fsync foo/some_dir/some_file
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
82 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
90 #define LOG_WALK_PIN_ONLY 0
91 #define LOG_WALK_REPLAY_INODES 1
92 #define LOG_WALK_REPLAY_ALL 2
94 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root, struct inode *inode,
97 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root,
99 struct btrfs_path *path, u64 objectid);
100 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_root *log,
103 struct btrfs_path *path,
104 u64 dirid, int del_all);
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction. Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree. Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
134 static int start_log_trans(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root)
140 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
142 if (!root->log_start_pid) {
143 root->log_start_pid = current->pid;
144 root->log_multiple_pids = false;
145 } else if (root->log_start_pid != current->pid) {
146 root->log_multiple_pids = true;
150 atomic_inc(&root->log_writers);
151 mutex_unlock(&root->log_mutex);
154 root->log_multiple_pids = false;
155 root->log_start_pid = current->pid;
156 mutex_lock(&root->fs_info->tree_log_mutex);
157 if (!root->fs_info->log_root_tree) {
158 ret = btrfs_init_log_root_tree(trans, root->fs_info);
162 if (err == 0 && !root->log_root) {
163 ret = btrfs_add_log_tree(trans, root);
167 mutex_unlock(&root->fs_info->tree_log_mutex);
169 atomic_inc(&root->log_writers);
170 mutex_unlock(&root->log_mutex);
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
179 static int join_running_log_trans(struct btrfs_root *root)
187 mutex_lock(&root->log_mutex);
188 if (root->log_root) {
190 atomic_inc(&root->log_writers);
192 mutex_unlock(&root->log_mutex);
197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
201 int btrfs_pin_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
215 int btrfs_end_log_trans(struct btrfs_root *root)
217 if (atomic_dec_and_test(&root->log_writers)) {
219 if (waitqueue_active(&root->log_writer_wait))
220 wake_up(&root->log_writer_wait);
227 * the walk control struct is used to pass state down the chain when
228 * processing the log tree. The stage field tells us which part
229 * of the log tree processing we are currently doing. The others
230 * are state fields used for that specific part
232 struct walk_control {
233 /* should we free the extent on disk when done? This is used
234 * at transaction commit time while freeing a log tree
238 /* should we write out the extent buffer? This is used
239 * while flushing the log tree to disk during a sync
243 /* should we wait for the extent buffer io to finish? Also used
244 * while flushing the log tree to disk for a sync
248 /* pin only walk, we record which extents on disk belong to the
253 /* what stage of the replay code we're currently in */
256 /* the root we are currently replaying */
257 struct btrfs_root *replay_dest;
259 /* the trans handle for the current replay */
260 struct btrfs_trans_handle *trans;
262 /* the function that gets used to process blocks we find in the
263 * tree. Note the extent_buffer might not be up to date when it is
264 * passed in, and it must be checked or read if you need the data
267 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
268 struct walk_control *wc, u64 gen);
272 * process_func used to pin down extents, write them or wait on them
274 static int process_one_buffer(struct btrfs_root *log,
275 struct extent_buffer *eb,
276 struct walk_control *wc, u64 gen)
279 btrfs_pin_extent(log->fs_info->extent_root,
280 eb->start, eb->len, 0);
282 if (btrfs_buffer_uptodate(eb, gen)) {
284 btrfs_write_tree_block(eb);
286 btrfs_wait_tree_block_writeback(eb);
292 * Item overwrite used by replay and tree logging. eb, slot and key all refer
293 * to the src data we are copying out.
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
299 * If the key is already in the destination tree the existing item is
300 * overwritten. If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
303 * If the key isn't in the destination yet, a new item is inserted.
305 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
306 struct btrfs_root *root,
307 struct btrfs_path *path,
308 struct extent_buffer *eb, int slot,
309 struct btrfs_key *key)
313 u64 saved_i_size = 0;
314 int save_old_i_size = 0;
315 unsigned long src_ptr;
316 unsigned long dst_ptr;
317 int overwrite_root = 0;
319 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
322 item_size = btrfs_item_size_nr(eb, slot);
323 src_ptr = btrfs_item_ptr_offset(eb, slot);
325 /* look for the key in the destination tree */
326 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
330 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
332 if (dst_size != item_size)
335 if (item_size == 0) {
336 btrfs_release_path(root, path);
339 dst_copy = kmalloc(item_size, GFP_NOFS);
340 src_copy = kmalloc(item_size, GFP_NOFS);
341 if (!dst_copy || !src_copy) {
342 btrfs_release_path(root, path);
348 read_extent_buffer(eb, src_copy, src_ptr, item_size);
350 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
351 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
353 ret = memcmp(dst_copy, src_copy, item_size);
358 * they have the same contents, just return, this saves
359 * us from cowing blocks in the destination tree and doing
360 * extra writes that may not have been done by a previous
364 btrfs_release_path(root, path);
370 btrfs_release_path(root, path);
371 /* try to insert the key into the destination tree */
372 ret = btrfs_insert_empty_item(trans, root, path,
375 /* make sure any existing item is the correct size */
376 if (ret == -EEXIST) {
378 found_size = btrfs_item_size_nr(path->nodes[0],
380 if (found_size > item_size) {
381 btrfs_truncate_item(trans, root, path, item_size, 1);
382 } else if (found_size < item_size) {
383 ret = btrfs_extend_item(trans, root, path,
384 item_size - found_size);
390 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
393 /* don't overwrite an existing inode if the generation number
394 * was logged as zero. This is done when the tree logging code
395 * is just logging an inode to make sure it exists after recovery.
397 * Also, don't overwrite i_size on directories during replay.
398 * log replay inserts and removes directory items based on the
399 * state of the tree found in the subvolume, and i_size is modified
402 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
403 struct btrfs_inode_item *src_item;
404 struct btrfs_inode_item *dst_item;
406 src_item = (struct btrfs_inode_item *)src_ptr;
407 dst_item = (struct btrfs_inode_item *)dst_ptr;
409 if (btrfs_inode_generation(eb, src_item) == 0)
412 if (overwrite_root &&
413 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
414 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
416 saved_i_size = btrfs_inode_size(path->nodes[0],
421 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
424 if (save_old_i_size) {
425 struct btrfs_inode_item *dst_item;
426 dst_item = (struct btrfs_inode_item *)dst_ptr;
427 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
430 /* make sure the generation is filled in */
431 if (key->type == BTRFS_INODE_ITEM_KEY) {
432 struct btrfs_inode_item *dst_item;
433 dst_item = (struct btrfs_inode_item *)dst_ptr;
434 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
435 btrfs_set_inode_generation(path->nodes[0], dst_item,
440 btrfs_mark_buffer_dirty(path->nodes[0]);
441 btrfs_release_path(root, path);
446 * simple helper to read an inode off the disk from a given root
447 * This can only be called for subvolume roots and not for the log
449 static noinline struct inode *read_one_inode(struct btrfs_root *root,
452 struct btrfs_key key;
455 key.objectid = objectid;
456 key.type = BTRFS_INODE_ITEM_KEY;
458 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
461 } else if (is_bad_inode(inode)) {
468 /* replays a single extent in 'eb' at 'slot' with 'key' into the
469 * subvolume 'root'. path is released on entry and should be released
472 * extents in the log tree have not been allocated out of the extent
473 * tree yet. So, this completes the allocation, taking a reference
474 * as required if the extent already exists or creating a new extent
475 * if it isn't in the extent allocation tree yet.
477 * The extent is inserted into the file, dropping any existing extents
478 * from the file that overlap the new one.
480 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root,
482 struct btrfs_path *path,
483 struct extent_buffer *eb, int slot,
484 struct btrfs_key *key)
487 u64 mask = root->sectorsize - 1;
490 u64 start = key->offset;
492 struct btrfs_file_extent_item *item;
493 struct inode *inode = NULL;
497 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
498 found_type = btrfs_file_extent_type(eb, item);
500 if (found_type == BTRFS_FILE_EXTENT_REG ||
501 found_type == BTRFS_FILE_EXTENT_PREALLOC)
502 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
503 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
504 size = btrfs_file_extent_inline_len(eb, item);
505 extent_end = (start + size + mask) & ~mask;
511 inode = read_one_inode(root, key->objectid);
518 * first check to see if we already have this extent in the
519 * file. This must be done before the btrfs_drop_extents run
520 * so we don't try to drop this extent.
522 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
526 (found_type == BTRFS_FILE_EXTENT_REG ||
527 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
528 struct btrfs_file_extent_item cmp1;
529 struct btrfs_file_extent_item cmp2;
530 struct btrfs_file_extent_item *existing;
531 struct extent_buffer *leaf;
533 leaf = path->nodes[0];
534 existing = btrfs_item_ptr(leaf, path->slots[0],
535 struct btrfs_file_extent_item);
537 read_extent_buffer(eb, &cmp1, (unsigned long)item,
539 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
543 * we already have a pointer to this exact extent,
544 * we don't have to do anything
546 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
547 btrfs_release_path(root, path);
551 btrfs_release_path(root, path);
553 saved_nbytes = inode_get_bytes(inode);
554 /* drop any overlapping extents */
555 ret = btrfs_drop_extents(trans, inode, start, extent_end,
559 if (found_type == BTRFS_FILE_EXTENT_REG ||
560 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
562 unsigned long dest_offset;
563 struct btrfs_key ins;
565 ret = btrfs_insert_empty_item(trans, root, path, key,
568 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
570 copy_extent_buffer(path->nodes[0], eb, dest_offset,
571 (unsigned long)item, sizeof(*item));
573 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
574 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
575 ins.type = BTRFS_EXTENT_ITEM_KEY;
576 offset = key->offset - btrfs_file_extent_offset(eb, item);
578 if (ins.objectid > 0) {
581 LIST_HEAD(ordered_sums);
583 * is this extent already allocated in the extent
584 * allocation tree? If so, just add a reference
586 ret = btrfs_lookup_extent(root, ins.objectid,
589 ret = btrfs_inc_extent_ref(trans, root,
590 ins.objectid, ins.offset,
591 0, root->root_key.objectid,
592 key->objectid, offset);
595 * insert the extent pointer in the extent
598 ret = btrfs_alloc_logged_file_extent(trans,
599 root, root->root_key.objectid,
600 key->objectid, offset, &ins);
603 btrfs_release_path(root, path);
605 if (btrfs_file_extent_compression(eb, item)) {
606 csum_start = ins.objectid;
607 csum_end = csum_start + ins.offset;
609 csum_start = ins.objectid +
610 btrfs_file_extent_offset(eb, item);
611 csum_end = csum_start +
612 btrfs_file_extent_num_bytes(eb, item);
615 ret = btrfs_lookup_csums_range(root->log_root,
616 csum_start, csum_end - 1,
619 while (!list_empty(&ordered_sums)) {
620 struct btrfs_ordered_sum *sums;
621 sums = list_entry(ordered_sums.next,
622 struct btrfs_ordered_sum,
624 ret = btrfs_csum_file_blocks(trans,
625 root->fs_info->csum_root,
628 list_del(&sums->list);
632 btrfs_release_path(root, path);
634 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
635 /* inline extents are easy, we just overwrite them */
636 ret = overwrite_item(trans, root, path, eb, slot, key);
640 inode_set_bytes(inode, saved_nbytes);
641 btrfs_update_inode(trans, root, inode);
649 * when cleaning up conflicts between the directory names in the
650 * subvolume, directory names in the log and directory names in the
651 * inode back references, we may have to unlink inodes from directories.
653 * This is a helper function to do the unlink of a specific directory
656 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
657 struct btrfs_root *root,
658 struct btrfs_path *path,
660 struct btrfs_dir_item *di)
665 struct extent_buffer *leaf;
666 struct btrfs_key location;
669 leaf = path->nodes[0];
671 btrfs_dir_item_key_to_cpu(leaf, di, &location);
672 name_len = btrfs_dir_name_len(leaf, di);
673 name = kmalloc(name_len, GFP_NOFS);
677 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
678 btrfs_release_path(root, path);
680 inode = read_one_inode(root, location.objectid);
683 ret = link_to_fixup_dir(trans, root, path, location.objectid);
686 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
695 * helper function to see if a given name and sequence number found
696 * in an inode back reference are already in a directory and correctly
697 * point to this inode
699 static noinline int inode_in_dir(struct btrfs_root *root,
700 struct btrfs_path *path,
701 u64 dirid, u64 objectid, u64 index,
702 const char *name, int name_len)
704 struct btrfs_dir_item *di;
705 struct btrfs_key location;
708 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
709 index, name, name_len, 0);
710 if (di && !IS_ERR(di)) {
711 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
712 if (location.objectid != objectid)
716 btrfs_release_path(root, path);
718 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
719 if (di && !IS_ERR(di)) {
720 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
721 if (location.objectid != objectid)
727 btrfs_release_path(root, path);
732 * helper function to check a log tree for a named back reference in
733 * an inode. This is used to decide if a back reference that is
734 * found in the subvolume conflicts with what we find in the log.
736 * inode backreferences may have multiple refs in a single item,
737 * during replay we process one reference at a time, and we don't
738 * want to delete valid links to a file from the subvolume if that
739 * link is also in the log.
741 static noinline int backref_in_log(struct btrfs_root *log,
742 struct btrfs_key *key,
743 char *name, int namelen)
745 struct btrfs_path *path;
746 struct btrfs_inode_ref *ref;
748 unsigned long ptr_end;
749 unsigned long name_ptr;
755 path = btrfs_alloc_path();
759 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
763 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
764 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
765 ptr_end = ptr + item_size;
766 while (ptr < ptr_end) {
767 ref = (struct btrfs_inode_ref *)ptr;
768 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
769 if (found_name_len == namelen) {
770 name_ptr = (unsigned long)(ref + 1);
771 ret = memcmp_extent_buffer(path->nodes[0], name,
778 ptr = (unsigned long)(ref + 1) + found_name_len;
781 btrfs_free_path(path);
787 * replay one inode back reference item found in the log tree.
788 * eb, slot and key refer to the buffer and key found in the log tree.
789 * root is the destination we are replaying into, and path is for temp
790 * use by this function. (it should be released on return).
792 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
793 struct btrfs_root *root,
794 struct btrfs_root *log,
795 struct btrfs_path *path,
796 struct extent_buffer *eb, int slot,
797 struct btrfs_key *key)
801 struct btrfs_inode_ref *ref;
805 unsigned long ref_ptr;
806 unsigned long ref_end;
810 * it is possible that we didn't log all the parent directories
811 * for a given inode. If we don't find the dir, just don't
812 * copy the back ref in. The link count fixup code will take
815 dir = read_one_inode(root, key->offset);
819 inode = read_one_inode(root, key->objectid);
822 ref_ptr = btrfs_item_ptr_offset(eb, slot);
823 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
826 ref = (struct btrfs_inode_ref *)ref_ptr;
828 namelen = btrfs_inode_ref_name_len(eb, ref);
829 name = kmalloc(namelen, GFP_NOFS);
832 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
834 /* if we already have a perfect match, we're done */
835 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
836 btrfs_inode_ref_index(eb, ref),
842 * look for a conflicting back reference in the metadata.
843 * if we find one we have to unlink that name of the file
844 * before we add our new link. Later on, we overwrite any
845 * existing back reference, and we don't want to create
846 * dangling pointers in the directory.
852 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
856 struct btrfs_inode_ref *victim_ref;
858 unsigned long ptr_end;
859 struct extent_buffer *leaf = path->nodes[0];
861 /* are we trying to overwrite a back ref for the root directory
862 * if so, just jump out, we're done
864 if (key->objectid == key->offset)
867 /* check all the names in this back reference to see
868 * if they are in the log. if so, we allow them to stay
869 * otherwise they must be unlinked as a conflict
871 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
872 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
873 while (ptr < ptr_end) {
874 victim_ref = (struct btrfs_inode_ref *)ptr;
875 victim_name_len = btrfs_inode_ref_name_len(leaf,
877 victim_name = kmalloc(victim_name_len, GFP_NOFS);
878 BUG_ON(!victim_name);
880 read_extent_buffer(leaf, victim_name,
881 (unsigned long)(victim_ref + 1),
884 if (!backref_in_log(log, key, victim_name,
886 btrfs_inc_nlink(inode);
887 btrfs_release_path(root, path);
889 ret = btrfs_unlink_inode(trans, root, dir,
894 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
899 * NOTE: we have searched root tree and checked the
900 * coresponding ref, it does not need to check again.
904 btrfs_release_path(root, path);
907 /* insert our name */
908 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
909 btrfs_inode_ref_index(eb, ref));
912 btrfs_update_inode(trans, root, inode);
915 ref_ptr = (unsigned long)(ref + 1) + namelen;
917 if (ref_ptr < ref_end)
920 /* finally write the back reference in the inode */
921 ret = overwrite_item(trans, root, path, eb, slot, key);
925 btrfs_release_path(root, path);
931 static int insert_orphan_item(struct btrfs_trans_handle *trans,
932 struct btrfs_root *root, u64 offset)
935 ret = btrfs_find_orphan_item(root, offset);
937 ret = btrfs_insert_orphan_item(trans, root, offset);
943 * There are a few corners where the link count of the file can't
944 * be properly maintained during replay. So, instead of adding
945 * lots of complexity to the log code, we just scan the backrefs
946 * for any file that has been through replay.
948 * The scan will update the link count on the inode to reflect the
949 * number of back refs found. If it goes down to zero, the iput
950 * will free the inode.
952 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
953 struct btrfs_root *root,
956 struct btrfs_path *path;
958 struct btrfs_key key;
961 unsigned long ptr_end;
964 key.objectid = inode->i_ino;
965 key.type = BTRFS_INODE_REF_KEY;
966 key.offset = (u64)-1;
968 path = btrfs_alloc_path();
973 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
977 if (path->slots[0] == 0)
981 btrfs_item_key_to_cpu(path->nodes[0], &key,
983 if (key.objectid != inode->i_ino ||
984 key.type != BTRFS_INODE_REF_KEY)
986 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
987 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
989 while (ptr < ptr_end) {
990 struct btrfs_inode_ref *ref;
992 ref = (struct btrfs_inode_ref *)ptr;
993 name_len = btrfs_inode_ref_name_len(path->nodes[0],
995 ptr = (unsigned long)(ref + 1) + name_len;
1002 btrfs_release_path(root, path);
1004 btrfs_release_path(root, path);
1005 if (nlink != inode->i_nlink) {
1006 inode->i_nlink = nlink;
1007 btrfs_update_inode(trans, root, inode);
1009 BTRFS_I(inode)->index_cnt = (u64)-1;
1011 if (inode->i_nlink == 0) {
1012 if (S_ISDIR(inode->i_mode)) {
1013 ret = replay_dir_deletes(trans, root, NULL, path,
1017 ret = insert_orphan_item(trans, root, inode->i_ino);
1020 btrfs_free_path(path);
1025 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1026 struct btrfs_root *root,
1027 struct btrfs_path *path)
1030 struct btrfs_key key;
1031 struct inode *inode;
1033 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1034 key.type = BTRFS_ORPHAN_ITEM_KEY;
1035 key.offset = (u64)-1;
1037 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1042 if (path->slots[0] == 0)
1047 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1048 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1049 key.type != BTRFS_ORPHAN_ITEM_KEY)
1052 ret = btrfs_del_item(trans, root, path);
1055 btrfs_release_path(root, path);
1056 inode = read_one_inode(root, key.offset);
1059 ret = fixup_inode_link_count(trans, root, inode);
1065 * fixup on a directory may create new entries,
1066 * make sure we always look for the highset possible
1069 key.offset = (u64)-1;
1071 btrfs_release_path(root, path);
1077 * record a given inode in the fixup dir so we can check its link
1078 * count when replay is done. The link count is incremented here
1079 * so the inode won't go away until we check it
1081 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1082 struct btrfs_root *root,
1083 struct btrfs_path *path,
1086 struct btrfs_key key;
1088 struct inode *inode;
1090 inode = read_one_inode(root, objectid);
1093 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1094 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1095 key.offset = objectid;
1097 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1099 btrfs_release_path(root, path);
1101 btrfs_inc_nlink(inode);
1102 btrfs_update_inode(trans, root, inode);
1103 } else if (ret == -EEXIST) {
1114 * when replaying the log for a directory, we only insert names
1115 * for inodes that actually exist. This means an fsync on a directory
1116 * does not implicitly fsync all the new files in it
1118 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1119 struct btrfs_root *root,
1120 struct btrfs_path *path,
1121 u64 dirid, u64 index,
1122 char *name, int name_len, u8 type,
1123 struct btrfs_key *location)
1125 struct inode *inode;
1129 inode = read_one_inode(root, location->objectid);
1133 dir = read_one_inode(root, dirid);
1138 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1140 /* FIXME, put inode into FIXUP list */
1148 * take a single entry in a log directory item and replay it into
1151 * if a conflicting item exists in the subdirectory already,
1152 * the inode it points to is unlinked and put into the link count
1155 * If a name from the log points to a file or directory that does
1156 * not exist in the FS, it is skipped. fsyncs on directories
1157 * do not force down inodes inside that directory, just changes to the
1158 * names or unlinks in a directory.
1160 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 struct extent_buffer *eb,
1164 struct btrfs_dir_item *di,
1165 struct btrfs_key *key)
1169 struct btrfs_dir_item *dst_di;
1170 struct btrfs_key found_key;
1171 struct btrfs_key log_key;
1177 dir = read_one_inode(root, key->objectid);
1180 name_len = btrfs_dir_name_len(eb, di);
1181 name = kmalloc(name_len, GFP_NOFS);
1185 log_type = btrfs_dir_type(eb, di);
1186 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1189 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1190 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1195 btrfs_release_path(root, path);
1197 if (key->type == BTRFS_DIR_ITEM_KEY) {
1198 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1200 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1201 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1208 if (!dst_di || IS_ERR(dst_di)) {
1209 /* we need a sequence number to insert, so we only
1210 * do inserts for the BTRFS_DIR_INDEX_KEY types
1212 if (key->type != BTRFS_DIR_INDEX_KEY)
1217 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1218 /* the existing item matches the logged item */
1219 if (found_key.objectid == log_key.objectid &&
1220 found_key.type == log_key.type &&
1221 found_key.offset == log_key.offset &&
1222 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1227 * don't drop the conflicting directory entry if the inode
1228 * for the new entry doesn't exist
1233 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1236 if (key->type == BTRFS_DIR_INDEX_KEY)
1239 btrfs_release_path(root, path);
1245 btrfs_release_path(root, path);
1246 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1247 name, name_len, log_type, &log_key);
1249 BUG_ON(ret && ret != -ENOENT);
1254 * find all the names in a directory item and reconcile them into
1255 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1256 * one name in a directory item, but the same code gets used for
1257 * both directory index types
1259 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1260 struct btrfs_root *root,
1261 struct btrfs_path *path,
1262 struct extent_buffer *eb, int slot,
1263 struct btrfs_key *key)
1266 u32 item_size = btrfs_item_size_nr(eb, slot);
1267 struct btrfs_dir_item *di;
1270 unsigned long ptr_end;
1272 ptr = btrfs_item_ptr_offset(eb, slot);
1273 ptr_end = ptr + item_size;
1274 while (ptr < ptr_end) {
1275 di = (struct btrfs_dir_item *)ptr;
1276 if (verify_dir_item(root, eb, di))
1278 name_len = btrfs_dir_name_len(eb, di);
1279 ret = replay_one_name(trans, root, path, eb, di, key);
1281 ptr = (unsigned long)(di + 1);
1288 * directory replay has two parts. There are the standard directory
1289 * items in the log copied from the subvolume, and range items
1290 * created in the log while the subvolume was logged.
1292 * The range items tell us which parts of the key space the log
1293 * is authoritative for. During replay, if a key in the subvolume
1294 * directory is in a logged range item, but not actually in the log
1295 * that means it was deleted from the directory before the fsync
1296 * and should be removed.
1298 static noinline int find_dir_range(struct btrfs_root *root,
1299 struct btrfs_path *path,
1300 u64 dirid, int key_type,
1301 u64 *start_ret, u64 *end_ret)
1303 struct btrfs_key key;
1305 struct btrfs_dir_log_item *item;
1309 if (*start_ret == (u64)-1)
1312 key.objectid = dirid;
1313 key.type = key_type;
1314 key.offset = *start_ret;
1316 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1320 if (path->slots[0] == 0)
1325 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1327 if (key.type != key_type || key.objectid != dirid) {
1331 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1332 struct btrfs_dir_log_item);
1333 found_end = btrfs_dir_log_end(path->nodes[0], item);
1335 if (*start_ret >= key.offset && *start_ret <= found_end) {
1337 *start_ret = key.offset;
1338 *end_ret = found_end;
1343 /* check the next slot in the tree to see if it is a valid item */
1344 nritems = btrfs_header_nritems(path->nodes[0]);
1345 if (path->slots[0] >= nritems) {
1346 ret = btrfs_next_leaf(root, path);
1353 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1355 if (key.type != key_type || key.objectid != dirid) {
1359 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1360 struct btrfs_dir_log_item);
1361 found_end = btrfs_dir_log_end(path->nodes[0], item);
1362 *start_ret = key.offset;
1363 *end_ret = found_end;
1366 btrfs_release_path(root, path);
1371 * this looks for a given directory item in the log. If the directory
1372 * item is not in the log, the item is removed and the inode it points
1375 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1376 struct btrfs_root *root,
1377 struct btrfs_root *log,
1378 struct btrfs_path *path,
1379 struct btrfs_path *log_path,
1381 struct btrfs_key *dir_key)
1384 struct extent_buffer *eb;
1387 struct btrfs_dir_item *di;
1388 struct btrfs_dir_item *log_di;
1391 unsigned long ptr_end;
1393 struct inode *inode;
1394 struct btrfs_key location;
1397 eb = path->nodes[0];
1398 slot = path->slots[0];
1399 item_size = btrfs_item_size_nr(eb, slot);
1400 ptr = btrfs_item_ptr_offset(eb, slot);
1401 ptr_end = ptr + item_size;
1402 while (ptr < ptr_end) {
1403 di = (struct btrfs_dir_item *)ptr;
1404 if (verify_dir_item(root, eb, di)) {
1409 name_len = btrfs_dir_name_len(eb, di);
1410 name = kmalloc(name_len, GFP_NOFS);
1415 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1418 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1419 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1422 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1423 log_di = btrfs_lookup_dir_index_item(trans, log,
1429 if (!log_di || IS_ERR(log_di)) {
1430 btrfs_dir_item_key_to_cpu(eb, di, &location);
1431 btrfs_release_path(root, path);
1432 btrfs_release_path(log, log_path);
1433 inode = read_one_inode(root, location.objectid);
1436 ret = link_to_fixup_dir(trans, root,
1437 path, location.objectid);
1439 btrfs_inc_nlink(inode);
1440 ret = btrfs_unlink_inode(trans, root, dir, inode,
1446 /* there might still be more names under this key
1447 * check and repeat if required
1449 ret = btrfs_search_slot(NULL, root, dir_key, path,
1456 btrfs_release_path(log, log_path);
1459 ptr = (unsigned long)(di + 1);
1464 btrfs_release_path(root, path);
1465 btrfs_release_path(log, log_path);
1470 * deletion replay happens before we copy any new directory items
1471 * out of the log or out of backreferences from inodes. It
1472 * scans the log to find ranges of keys that log is authoritative for,
1473 * and then scans the directory to find items in those ranges that are
1474 * not present in the log.
1476 * Anything we don't find in the log is unlinked and removed from the
1479 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1480 struct btrfs_root *root,
1481 struct btrfs_root *log,
1482 struct btrfs_path *path,
1483 u64 dirid, int del_all)
1487 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1489 struct btrfs_key dir_key;
1490 struct btrfs_key found_key;
1491 struct btrfs_path *log_path;
1494 dir_key.objectid = dirid;
1495 dir_key.type = BTRFS_DIR_ITEM_KEY;
1496 log_path = btrfs_alloc_path();
1500 dir = read_one_inode(root, dirid);
1501 /* it isn't an error if the inode isn't there, that can happen
1502 * because we replay the deletes before we copy in the inode item
1506 btrfs_free_path(log_path);
1514 range_end = (u64)-1;
1516 ret = find_dir_range(log, path, dirid, key_type,
1517 &range_start, &range_end);
1522 dir_key.offset = range_start;
1525 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1530 nritems = btrfs_header_nritems(path->nodes[0]);
1531 if (path->slots[0] >= nritems) {
1532 ret = btrfs_next_leaf(root, path);
1536 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1538 if (found_key.objectid != dirid ||
1539 found_key.type != dir_key.type)
1542 if (found_key.offset > range_end)
1545 ret = check_item_in_log(trans, root, log, path,
1549 if (found_key.offset == (u64)-1)
1551 dir_key.offset = found_key.offset + 1;
1553 btrfs_release_path(root, path);
1554 if (range_end == (u64)-1)
1556 range_start = range_end + 1;
1561 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1562 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1563 dir_key.type = BTRFS_DIR_INDEX_KEY;
1564 btrfs_release_path(root, path);
1568 btrfs_release_path(root, path);
1569 btrfs_free_path(log_path);
1575 * the process_func used to replay items from the log tree. This
1576 * gets called in two different stages. The first stage just looks
1577 * for inodes and makes sure they are all copied into the subvolume.
1579 * The second stage copies all the other item types from the log into
1580 * the subvolume. The two stage approach is slower, but gets rid of
1581 * lots of complexity around inodes referencing other inodes that exist
1582 * only in the log (references come from either directory items or inode
1585 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1586 struct walk_control *wc, u64 gen)
1589 struct btrfs_path *path;
1590 struct btrfs_root *root = wc->replay_dest;
1591 struct btrfs_key key;
1596 btrfs_read_buffer(eb, gen);
1598 level = btrfs_header_level(eb);
1603 path = btrfs_alloc_path();
1606 nritems = btrfs_header_nritems(eb);
1607 for (i = 0; i < nritems; i++) {
1608 btrfs_item_key_to_cpu(eb, &key, i);
1610 /* inode keys are done during the first stage */
1611 if (key.type == BTRFS_INODE_ITEM_KEY &&
1612 wc->stage == LOG_WALK_REPLAY_INODES) {
1613 struct btrfs_inode_item *inode_item;
1616 inode_item = btrfs_item_ptr(eb, i,
1617 struct btrfs_inode_item);
1618 mode = btrfs_inode_mode(eb, inode_item);
1619 if (S_ISDIR(mode)) {
1620 ret = replay_dir_deletes(wc->trans,
1621 root, log, path, key.objectid, 0);
1624 ret = overwrite_item(wc->trans, root, path,
1628 /* for regular files, make sure corresponding
1629 * orhpan item exist. extents past the new EOF
1630 * will be truncated later by orphan cleanup.
1632 if (S_ISREG(mode)) {
1633 ret = insert_orphan_item(wc->trans, root,
1638 ret = link_to_fixup_dir(wc->trans, root,
1639 path, key.objectid);
1642 if (wc->stage < LOG_WALK_REPLAY_ALL)
1645 /* these keys are simply copied */
1646 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1647 ret = overwrite_item(wc->trans, root, path,
1650 } else if (key.type == BTRFS_INODE_REF_KEY) {
1651 ret = add_inode_ref(wc->trans, root, log, path,
1653 BUG_ON(ret && ret != -ENOENT);
1654 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1655 ret = replay_one_extent(wc->trans, root, path,
1658 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1659 key.type == BTRFS_DIR_INDEX_KEY) {
1660 ret = replay_one_dir_item(wc->trans, root, path,
1665 btrfs_free_path(path);
1669 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1670 struct btrfs_root *root,
1671 struct btrfs_path *path, int *level,
1672 struct walk_control *wc)
1677 struct extent_buffer *next;
1678 struct extent_buffer *cur;
1679 struct extent_buffer *parent;
1683 WARN_ON(*level < 0);
1684 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1686 while (*level > 0) {
1687 WARN_ON(*level < 0);
1688 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1689 cur = path->nodes[*level];
1691 if (btrfs_header_level(cur) != *level)
1694 if (path->slots[*level] >=
1695 btrfs_header_nritems(cur))
1698 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1699 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1700 blocksize = btrfs_level_size(root, *level - 1);
1702 parent = path->nodes[*level];
1703 root_owner = btrfs_header_owner(parent);
1705 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1710 wc->process_func(root, next, wc, ptr_gen);
1712 path->slots[*level]++;
1714 btrfs_read_buffer(next, ptr_gen);
1716 btrfs_tree_lock(next);
1717 clean_tree_block(trans, root, next);
1718 btrfs_set_lock_blocking(next);
1719 btrfs_wait_tree_block_writeback(next);
1720 btrfs_tree_unlock(next);
1722 WARN_ON(root_owner !=
1723 BTRFS_TREE_LOG_OBJECTID);
1724 ret = btrfs_free_reserved_extent(root,
1728 free_extent_buffer(next);
1731 btrfs_read_buffer(next, ptr_gen);
1733 WARN_ON(*level <= 0);
1734 if (path->nodes[*level-1])
1735 free_extent_buffer(path->nodes[*level-1]);
1736 path->nodes[*level-1] = next;
1737 *level = btrfs_header_level(next);
1738 path->slots[*level] = 0;
1741 WARN_ON(*level < 0);
1742 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1744 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1750 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1751 struct btrfs_root *root,
1752 struct btrfs_path *path, int *level,
1753 struct walk_control *wc)
1760 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1761 slot = path->slots[i];
1762 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
1765 WARN_ON(*level == 0);
1768 struct extent_buffer *parent;
1769 if (path->nodes[*level] == root->node)
1770 parent = path->nodes[*level];
1772 parent = path->nodes[*level + 1];
1774 root_owner = btrfs_header_owner(parent);
1775 wc->process_func(root, path->nodes[*level], wc,
1776 btrfs_header_generation(path->nodes[*level]));
1778 struct extent_buffer *next;
1780 next = path->nodes[*level];
1782 btrfs_tree_lock(next);
1783 clean_tree_block(trans, root, next);
1784 btrfs_set_lock_blocking(next);
1785 btrfs_wait_tree_block_writeback(next);
1786 btrfs_tree_unlock(next);
1788 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1789 ret = btrfs_free_reserved_extent(root,
1790 path->nodes[*level]->start,
1791 path->nodes[*level]->len);
1794 free_extent_buffer(path->nodes[*level]);
1795 path->nodes[*level] = NULL;
1803 * drop the reference count on the tree rooted at 'snap'. This traverses
1804 * the tree freeing any blocks that have a ref count of zero after being
1807 static int walk_log_tree(struct btrfs_trans_handle *trans,
1808 struct btrfs_root *log, struct walk_control *wc)
1813 struct btrfs_path *path;
1817 path = btrfs_alloc_path();
1821 level = btrfs_header_level(log->node);
1823 path->nodes[level] = log->node;
1824 extent_buffer_get(log->node);
1825 path->slots[level] = 0;
1828 wret = walk_down_log_tree(trans, log, path, &level, wc);
1834 wret = walk_up_log_tree(trans, log, path, &level, wc);
1841 /* was the root node processed? if not, catch it here */
1842 if (path->nodes[orig_level]) {
1843 wc->process_func(log, path->nodes[orig_level], wc,
1844 btrfs_header_generation(path->nodes[orig_level]));
1846 struct extent_buffer *next;
1848 next = path->nodes[orig_level];
1850 btrfs_tree_lock(next);
1851 clean_tree_block(trans, log, next);
1852 btrfs_set_lock_blocking(next);
1853 btrfs_wait_tree_block_writeback(next);
1854 btrfs_tree_unlock(next);
1856 WARN_ON(log->root_key.objectid !=
1857 BTRFS_TREE_LOG_OBJECTID);
1858 ret = btrfs_free_reserved_extent(log, next->start,
1864 for (i = 0; i <= orig_level; i++) {
1865 if (path->nodes[i]) {
1866 free_extent_buffer(path->nodes[i]);
1867 path->nodes[i] = NULL;
1870 btrfs_free_path(path);
1875 * helper function to update the item for a given subvolumes log root
1876 * in the tree of log roots
1878 static int update_log_root(struct btrfs_trans_handle *trans,
1879 struct btrfs_root *log)
1883 if (log->log_transid == 1) {
1884 /* insert root item on the first sync */
1885 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1886 &log->root_key, &log->root_item);
1888 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1889 &log->root_key, &log->root_item);
1894 static int wait_log_commit(struct btrfs_trans_handle *trans,
1895 struct btrfs_root *root, unsigned long transid)
1898 int index = transid % 2;
1901 * we only allow two pending log transactions at a time,
1902 * so we know that if ours is more than 2 older than the
1903 * current transaction, we're done
1906 prepare_to_wait(&root->log_commit_wait[index],
1907 &wait, TASK_UNINTERRUPTIBLE);
1908 mutex_unlock(&root->log_mutex);
1910 if (root->fs_info->last_trans_log_full_commit !=
1911 trans->transid && root->log_transid < transid + 2 &&
1912 atomic_read(&root->log_commit[index]))
1915 finish_wait(&root->log_commit_wait[index], &wait);
1916 mutex_lock(&root->log_mutex);
1917 } while (root->log_transid < transid + 2 &&
1918 atomic_read(&root->log_commit[index]));
1922 static int wait_for_writer(struct btrfs_trans_handle *trans,
1923 struct btrfs_root *root)
1926 while (atomic_read(&root->log_writers)) {
1927 prepare_to_wait(&root->log_writer_wait,
1928 &wait, TASK_UNINTERRUPTIBLE);
1929 mutex_unlock(&root->log_mutex);
1930 if (root->fs_info->last_trans_log_full_commit !=
1931 trans->transid && atomic_read(&root->log_writers))
1933 mutex_lock(&root->log_mutex);
1934 finish_wait(&root->log_writer_wait, &wait);
1940 * btrfs_sync_log does sends a given tree log down to the disk and
1941 * updates the super blocks to record it. When this call is done,
1942 * you know that any inodes previously logged are safely on disk only
1945 * Any other return value means you need to call btrfs_commit_transaction.
1946 * Some of the edge cases for fsyncing directories that have had unlinks
1947 * or renames done in the past mean that sometimes the only safe
1948 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1949 * that has happened.
1951 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1952 struct btrfs_root *root)
1958 struct btrfs_root *log = root->log_root;
1959 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
1960 unsigned long log_transid = 0;
1962 mutex_lock(&root->log_mutex);
1963 index1 = root->log_transid % 2;
1964 if (atomic_read(&root->log_commit[index1])) {
1965 wait_log_commit(trans, root, root->log_transid);
1966 mutex_unlock(&root->log_mutex);
1969 atomic_set(&root->log_commit[index1], 1);
1971 /* wait for previous tree log sync to complete */
1972 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
1973 wait_log_commit(trans, root, root->log_transid - 1);
1976 unsigned long batch = root->log_batch;
1977 if (root->log_multiple_pids) {
1978 mutex_unlock(&root->log_mutex);
1979 schedule_timeout_uninterruptible(1);
1980 mutex_lock(&root->log_mutex);
1982 wait_for_writer(trans, root);
1983 if (batch == root->log_batch)
1987 /* bail out if we need to do a full commit */
1988 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
1990 mutex_unlock(&root->log_mutex);
1994 log_transid = root->log_transid;
1995 if (log_transid % 2 == 0)
1996 mark = EXTENT_DIRTY;
2000 /* we start IO on all the marked extents here, but we don't actually
2001 * wait for them until later.
2003 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2006 btrfs_set_root_node(&log->root_item, log->node);
2008 root->log_batch = 0;
2009 root->log_transid++;
2010 log->log_transid = root->log_transid;
2011 root->log_start_pid = 0;
2014 * IO has been started, blocks of the log tree have WRITTEN flag set
2015 * in their headers. new modifications of the log will be written to
2016 * new positions. so it's safe to allow log writers to go in.
2018 mutex_unlock(&root->log_mutex);
2020 mutex_lock(&log_root_tree->log_mutex);
2021 log_root_tree->log_batch++;
2022 atomic_inc(&log_root_tree->log_writers);
2023 mutex_unlock(&log_root_tree->log_mutex);
2025 ret = update_log_root(trans, log);
2027 mutex_lock(&log_root_tree->log_mutex);
2028 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2030 if (waitqueue_active(&log_root_tree->log_writer_wait))
2031 wake_up(&log_root_tree->log_writer_wait);
2035 BUG_ON(ret != -ENOSPC);
2036 root->fs_info->last_trans_log_full_commit = trans->transid;
2037 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2038 mutex_unlock(&log_root_tree->log_mutex);
2043 index2 = log_root_tree->log_transid % 2;
2044 if (atomic_read(&log_root_tree->log_commit[index2])) {
2045 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2046 wait_log_commit(trans, log_root_tree,
2047 log_root_tree->log_transid);
2048 mutex_unlock(&log_root_tree->log_mutex);
2052 atomic_set(&log_root_tree->log_commit[index2], 1);
2054 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2055 wait_log_commit(trans, log_root_tree,
2056 log_root_tree->log_transid - 1);
2059 wait_for_writer(trans, log_root_tree);
2062 * now that we've moved on to the tree of log tree roots,
2063 * check the full commit flag again
2065 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2066 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2067 mutex_unlock(&log_root_tree->log_mutex);
2069 goto out_wake_log_root;
2072 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2073 &log_root_tree->dirty_log_pages,
2074 EXTENT_DIRTY | EXTENT_NEW);
2076 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2078 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2079 log_root_tree->node->start);
2080 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2081 btrfs_header_level(log_root_tree->node));
2083 log_root_tree->log_batch = 0;
2084 log_root_tree->log_transid++;
2087 mutex_unlock(&log_root_tree->log_mutex);
2090 * nobody else is going to jump in and write the the ctree
2091 * super here because the log_commit atomic below is protecting
2092 * us. We must be called with a transaction handle pinning
2093 * the running transaction open, so a full commit can't hop
2094 * in and cause problems either.
2096 write_ctree_super(trans, root->fs_info->tree_root, 1);
2099 mutex_lock(&root->log_mutex);
2100 if (root->last_log_commit < log_transid)
2101 root->last_log_commit = log_transid;
2102 mutex_unlock(&root->log_mutex);
2105 atomic_set(&log_root_tree->log_commit[index2], 0);
2107 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2108 wake_up(&log_root_tree->log_commit_wait[index2]);
2110 atomic_set(&root->log_commit[index1], 0);
2112 if (waitqueue_active(&root->log_commit_wait[index1]))
2113 wake_up(&root->log_commit_wait[index1]);
2117 static void free_log_tree(struct btrfs_trans_handle *trans,
2118 struct btrfs_root *log)
2123 struct walk_control wc = {
2125 .process_func = process_one_buffer
2128 ret = walk_log_tree(trans, log, &wc);
2132 ret = find_first_extent_bit(&log->dirty_log_pages,
2133 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2137 clear_extent_bits(&log->dirty_log_pages, start, end,
2138 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2141 free_extent_buffer(log->node);
2146 * free all the extents used by the tree log. This should be called
2147 * at commit time of the full transaction
2149 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2151 if (root->log_root) {
2152 free_log_tree(trans, root->log_root);
2153 root->log_root = NULL;
2158 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2159 struct btrfs_fs_info *fs_info)
2161 if (fs_info->log_root_tree) {
2162 free_log_tree(trans, fs_info->log_root_tree);
2163 fs_info->log_root_tree = NULL;
2169 * If both a file and directory are logged, and unlinks or renames are
2170 * mixed in, we have a few interesting corners:
2172 * create file X in dir Y
2173 * link file X to X.link in dir Y
2175 * unlink file X but leave X.link
2178 * After a crash we would expect only X.link to exist. But file X
2179 * didn't get fsync'd again so the log has back refs for X and X.link.
2181 * We solve this by removing directory entries and inode backrefs from the
2182 * log when a file that was logged in the current transaction is
2183 * unlinked. Any later fsync will include the updated log entries, and
2184 * we'll be able to reconstruct the proper directory items from backrefs.
2186 * This optimizations allows us to avoid relogging the entire inode
2187 * or the entire directory.
2189 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 const char *name, int name_len,
2192 struct inode *dir, u64 index)
2194 struct btrfs_root *log;
2195 struct btrfs_dir_item *di;
2196 struct btrfs_path *path;
2201 if (BTRFS_I(dir)->logged_trans < trans->transid)
2204 ret = join_running_log_trans(root);
2208 mutex_lock(&BTRFS_I(dir)->log_mutex);
2210 log = root->log_root;
2211 path = btrfs_alloc_path();
2215 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2216 name, name_len, -1);
2222 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2223 bytes_del += name_len;
2226 btrfs_release_path(log, path);
2227 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2228 index, name, name_len, -1);
2234 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2235 bytes_del += name_len;
2239 /* update the directory size in the log to reflect the names
2243 struct btrfs_key key;
2245 key.objectid = dir->i_ino;
2247 key.type = BTRFS_INODE_ITEM_KEY;
2248 btrfs_release_path(log, path);
2250 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2256 struct btrfs_inode_item *item;
2259 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2260 struct btrfs_inode_item);
2261 i_size = btrfs_inode_size(path->nodes[0], item);
2262 if (i_size > bytes_del)
2263 i_size -= bytes_del;
2266 btrfs_set_inode_size(path->nodes[0], item, i_size);
2267 btrfs_mark_buffer_dirty(path->nodes[0]);
2270 btrfs_release_path(log, path);
2273 btrfs_free_path(path);
2274 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2275 if (ret == -ENOSPC) {
2276 root->fs_info->last_trans_log_full_commit = trans->transid;
2279 btrfs_end_log_trans(root);
2284 /* see comments for btrfs_del_dir_entries_in_log */
2285 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2286 struct btrfs_root *root,
2287 const char *name, int name_len,
2288 struct inode *inode, u64 dirid)
2290 struct btrfs_root *log;
2294 if (BTRFS_I(inode)->logged_trans < trans->transid)
2297 ret = join_running_log_trans(root);
2300 log = root->log_root;
2301 mutex_lock(&BTRFS_I(inode)->log_mutex);
2303 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2305 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2306 if (ret == -ENOSPC) {
2307 root->fs_info->last_trans_log_full_commit = trans->transid;
2310 btrfs_end_log_trans(root);
2316 * creates a range item in the log for 'dirid'. first_offset and
2317 * last_offset tell us which parts of the key space the log should
2318 * be considered authoritative for.
2320 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2321 struct btrfs_root *log,
2322 struct btrfs_path *path,
2323 int key_type, u64 dirid,
2324 u64 first_offset, u64 last_offset)
2327 struct btrfs_key key;
2328 struct btrfs_dir_log_item *item;
2330 key.objectid = dirid;
2331 key.offset = first_offset;
2332 if (key_type == BTRFS_DIR_ITEM_KEY)
2333 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2335 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2336 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2340 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2341 struct btrfs_dir_log_item);
2342 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2343 btrfs_mark_buffer_dirty(path->nodes[0]);
2344 btrfs_release_path(log, path);
2349 * log all the items included in the current transaction for a given
2350 * directory. This also creates the range items in the log tree required
2351 * to replay anything deleted before the fsync
2353 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2354 struct btrfs_root *root, struct inode *inode,
2355 struct btrfs_path *path,
2356 struct btrfs_path *dst_path, int key_type,
2357 u64 min_offset, u64 *last_offset_ret)
2359 struct btrfs_key min_key;
2360 struct btrfs_key max_key;
2361 struct btrfs_root *log = root->log_root;
2362 struct extent_buffer *src;
2367 u64 first_offset = min_offset;
2368 u64 last_offset = (u64)-1;
2370 log = root->log_root;
2371 max_key.objectid = inode->i_ino;
2372 max_key.offset = (u64)-1;
2373 max_key.type = key_type;
2375 min_key.objectid = inode->i_ino;
2376 min_key.type = key_type;
2377 min_key.offset = min_offset;
2379 path->keep_locks = 1;
2381 ret = btrfs_search_forward(root, &min_key, &max_key,
2382 path, 0, trans->transid);
2385 * we didn't find anything from this transaction, see if there
2386 * is anything at all
2388 if (ret != 0 || min_key.objectid != inode->i_ino ||
2389 min_key.type != key_type) {
2390 min_key.objectid = inode->i_ino;
2391 min_key.type = key_type;
2392 min_key.offset = (u64)-1;
2393 btrfs_release_path(root, path);
2394 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2396 btrfs_release_path(root, path);
2399 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2401 /* if ret == 0 there are items for this type,
2402 * create a range to tell us the last key of this type.
2403 * otherwise, there are no items in this directory after
2404 * *min_offset, and we create a range to indicate that.
2407 struct btrfs_key tmp;
2408 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2410 if (key_type == tmp.type)
2411 first_offset = max(min_offset, tmp.offset) + 1;
2416 /* go backward to find any previous key */
2417 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2419 struct btrfs_key tmp;
2420 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2421 if (key_type == tmp.type) {
2422 first_offset = tmp.offset;
2423 ret = overwrite_item(trans, log, dst_path,
2424 path->nodes[0], path->slots[0],
2432 btrfs_release_path(root, path);
2434 /* find the first key from this transaction again */
2435 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2442 * we have a block from this transaction, log every item in it
2443 * from our directory
2446 struct btrfs_key tmp;
2447 src = path->nodes[0];
2448 nritems = btrfs_header_nritems(src);
2449 for (i = path->slots[0]; i < nritems; i++) {
2450 btrfs_item_key_to_cpu(src, &min_key, i);
2452 if (min_key.objectid != inode->i_ino ||
2453 min_key.type != key_type)
2455 ret = overwrite_item(trans, log, dst_path, src, i,
2462 path->slots[0] = nritems;
2465 * look ahead to the next item and see if it is also
2466 * from this directory and from this transaction
2468 ret = btrfs_next_leaf(root, path);
2470 last_offset = (u64)-1;
2473 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2474 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2475 last_offset = (u64)-1;
2478 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2479 ret = overwrite_item(trans, log, dst_path,
2480 path->nodes[0], path->slots[0],
2485 last_offset = tmp.offset;
2490 btrfs_release_path(root, path);
2491 btrfs_release_path(log, dst_path);
2494 *last_offset_ret = last_offset;
2496 * insert the log range keys to indicate where the log
2499 ret = insert_dir_log_key(trans, log, path, key_type,
2500 inode->i_ino, first_offset,
2509 * logging directories is very similar to logging inodes, We find all the items
2510 * from the current transaction and write them to the log.
2512 * The recovery code scans the directory in the subvolume, and if it finds a
2513 * key in the range logged that is not present in the log tree, then it means
2514 * that dir entry was unlinked during the transaction.
2516 * In order for that scan to work, we must include one key smaller than
2517 * the smallest logged by this transaction and one key larger than the largest
2518 * key logged by this transaction.
2520 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2521 struct btrfs_root *root, struct inode *inode,
2522 struct btrfs_path *path,
2523 struct btrfs_path *dst_path)
2528 int key_type = BTRFS_DIR_ITEM_KEY;
2534 ret = log_dir_items(trans, root, inode, path,
2535 dst_path, key_type, min_key,
2539 if (max_key == (u64)-1)
2541 min_key = max_key + 1;
2544 if (key_type == BTRFS_DIR_ITEM_KEY) {
2545 key_type = BTRFS_DIR_INDEX_KEY;
2552 * a helper function to drop items from the log before we relog an
2553 * inode. max_key_type indicates the highest item type to remove.
2554 * This cannot be run for file data extents because it does not
2555 * free the extents they point to.
2557 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2558 struct btrfs_root *log,
2559 struct btrfs_path *path,
2560 u64 objectid, int max_key_type)
2563 struct btrfs_key key;
2564 struct btrfs_key found_key;
2566 key.objectid = objectid;
2567 key.type = max_key_type;
2568 key.offset = (u64)-1;
2571 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2576 if (path->slots[0] == 0)
2580 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2583 if (found_key.objectid != objectid)
2586 ret = btrfs_del_item(trans, log, path);
2588 btrfs_release_path(log, path);
2590 btrfs_release_path(log, path);
2594 static noinline int copy_items(struct btrfs_trans_handle *trans,
2595 struct btrfs_root *log,
2596 struct btrfs_path *dst_path,
2597 struct extent_buffer *src,
2598 int start_slot, int nr, int inode_only)
2600 unsigned long src_offset;
2601 unsigned long dst_offset;
2602 struct btrfs_file_extent_item *extent;
2603 struct btrfs_inode_item *inode_item;
2605 struct btrfs_key *ins_keys;
2609 struct list_head ordered_sums;
2611 INIT_LIST_HEAD(&ordered_sums);
2613 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2614 nr * sizeof(u32), GFP_NOFS);
2618 ins_sizes = (u32 *)ins_data;
2619 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2621 for (i = 0; i < nr; i++) {
2622 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2623 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2625 ret = btrfs_insert_empty_items(trans, log, dst_path,
2626 ins_keys, ins_sizes, nr);
2632 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2633 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2634 dst_path->slots[0]);
2636 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2638 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2639 src_offset, ins_sizes[i]);
2641 if (inode_only == LOG_INODE_EXISTS &&
2642 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2643 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2645 struct btrfs_inode_item);
2646 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2648 /* set the generation to zero so the recover code
2649 * can tell the difference between an logging
2650 * just to say 'this inode exists' and a logging
2651 * to say 'update this inode with these values'
2653 btrfs_set_inode_generation(dst_path->nodes[0],
2656 /* take a reference on file data extents so that truncates
2657 * or deletes of this inode don't have to relog the inode
2660 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2662 extent = btrfs_item_ptr(src, start_slot + i,
2663 struct btrfs_file_extent_item);
2665 found_type = btrfs_file_extent_type(src, extent);
2666 if (found_type == BTRFS_FILE_EXTENT_REG ||
2667 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2669 ds = btrfs_file_extent_disk_bytenr(src,
2671 /* ds == 0 is a hole */
2675 dl = btrfs_file_extent_disk_num_bytes(src,
2677 cs = btrfs_file_extent_offset(src, extent);
2678 cl = btrfs_file_extent_num_bytes(src,
2680 if (btrfs_file_extent_compression(src,
2686 ret = btrfs_lookup_csums_range(
2687 log->fs_info->csum_root,
2688 ds + cs, ds + cs + cl - 1,
2695 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2696 btrfs_release_path(log, dst_path);
2700 * we have to do this after the loop above to avoid changing the
2701 * log tree while trying to change the log tree.
2704 while (!list_empty(&ordered_sums)) {
2705 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2706 struct btrfs_ordered_sum,
2709 ret = btrfs_csum_file_blocks(trans, log, sums);
2710 list_del(&sums->list);
2716 /* log a single inode in the tree log.
2717 * At least one parent directory for this inode must exist in the tree
2718 * or be logged already.
2720 * Any items from this inode changed by the current transaction are copied
2721 * to the log tree. An extra reference is taken on any extents in this
2722 * file, allowing us to avoid a whole pile of corner cases around logging
2723 * blocks that have been removed from the tree.
2725 * See LOG_INODE_ALL and related defines for a description of what inode_only
2728 * This handles both files and directories.
2730 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2731 struct btrfs_root *root, struct inode *inode,
2734 struct btrfs_path *path;
2735 struct btrfs_path *dst_path;
2736 struct btrfs_key min_key;
2737 struct btrfs_key max_key;
2738 struct btrfs_root *log = root->log_root;
2739 struct extent_buffer *src = NULL;
2743 int ins_start_slot = 0;
2746 log = root->log_root;
2748 path = btrfs_alloc_path();
2751 dst_path = btrfs_alloc_path();
2753 btrfs_free_path(path);
2757 min_key.objectid = inode->i_ino;
2758 min_key.type = BTRFS_INODE_ITEM_KEY;
2761 max_key.objectid = inode->i_ino;
2763 /* today the code can only do partial logging of directories */
2764 if (!S_ISDIR(inode->i_mode))
2765 inode_only = LOG_INODE_ALL;
2767 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2768 max_key.type = BTRFS_XATTR_ITEM_KEY;
2770 max_key.type = (u8)-1;
2771 max_key.offset = (u64)-1;
2773 mutex_lock(&BTRFS_I(inode)->log_mutex);
2776 * a brute force approach to making sure we get the most uptodate
2777 * copies of everything.
2779 if (S_ISDIR(inode->i_mode)) {
2780 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2782 if (inode_only == LOG_INODE_EXISTS)
2783 max_key_type = BTRFS_XATTR_ITEM_KEY;
2784 ret = drop_objectid_items(trans, log, path,
2785 inode->i_ino, max_key_type);
2787 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2793 path->keep_locks = 1;
2797 ret = btrfs_search_forward(root, &min_key, &max_key,
2798 path, 0, trans->transid);
2802 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2803 if (min_key.objectid != inode->i_ino)
2805 if (min_key.type > max_key.type)
2808 src = path->nodes[0];
2809 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2812 } else if (!ins_nr) {
2813 ins_start_slot = path->slots[0];
2818 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2819 ins_nr, inode_only);
2825 ins_start_slot = path->slots[0];
2828 nritems = btrfs_header_nritems(path->nodes[0]);
2830 if (path->slots[0] < nritems) {
2831 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2836 ret = copy_items(trans, log, dst_path, src,
2838 ins_nr, inode_only);
2845 btrfs_release_path(root, path);
2847 if (min_key.offset < (u64)-1)
2849 else if (min_key.type < (u8)-1)
2851 else if (min_key.objectid < (u64)-1)
2857 ret = copy_items(trans, log, dst_path, src,
2859 ins_nr, inode_only);
2867 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2868 btrfs_release_path(root, path);
2869 btrfs_release_path(log, dst_path);
2870 ret = log_directory_changes(trans, root, inode, path, dst_path);
2876 BTRFS_I(inode)->logged_trans = trans->transid;
2878 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2880 btrfs_free_path(path);
2881 btrfs_free_path(dst_path);
2886 * follow the dentry parent pointers up the chain and see if any
2887 * of the directories in it require a full commit before they can
2888 * be logged. Returns zero if nothing special needs to be done or 1 if
2889 * a full commit is required.
2891 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2892 struct inode *inode,
2893 struct dentry *parent,
2894 struct super_block *sb,
2898 struct btrfs_root *root;
2899 struct dentry *old_parent = NULL;
2902 * for regular files, if its inode is already on disk, we don't
2903 * have to worry about the parents at all. This is because
2904 * we can use the last_unlink_trans field to record renames
2905 * and other fun in this file.
2907 if (S_ISREG(inode->i_mode) &&
2908 BTRFS_I(inode)->generation <= last_committed &&
2909 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2912 if (!S_ISDIR(inode->i_mode)) {
2913 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2915 inode = parent->d_inode;
2919 BTRFS_I(inode)->logged_trans = trans->transid;
2922 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2923 root = BTRFS_I(inode)->root;
2926 * make sure any commits to the log are forced
2927 * to be full commits
2929 root->fs_info->last_trans_log_full_commit =
2935 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2938 if (IS_ROOT(parent))
2941 parent = dget_parent(parent);
2943 old_parent = parent;
2944 inode = parent->d_inode;
2952 static int inode_in_log(struct btrfs_trans_handle *trans,
2953 struct inode *inode)
2955 struct btrfs_root *root = BTRFS_I(inode)->root;
2958 mutex_lock(&root->log_mutex);
2959 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2960 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2962 mutex_unlock(&root->log_mutex);
2968 * helper function around btrfs_log_inode to make sure newly created
2969 * parent directories also end up in the log. A minimal inode and backref
2970 * only logging is done of any parent directories that are older than
2971 * the last committed transaction
2973 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2974 struct btrfs_root *root, struct inode *inode,
2975 struct dentry *parent, int exists_only)
2977 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
2978 struct super_block *sb;
2979 struct dentry *old_parent = NULL;
2981 u64 last_committed = root->fs_info->last_trans_committed;
2985 if (btrfs_test_opt(root, NOTREELOG)) {
2990 if (root->fs_info->last_trans_log_full_commit >
2991 root->fs_info->last_trans_committed) {
2996 if (root != BTRFS_I(inode)->root ||
2997 btrfs_root_refs(&root->root_item) == 0) {
3002 ret = check_parent_dirs_for_sync(trans, inode, parent,
3003 sb, last_committed);
3007 if (inode_in_log(trans, inode)) {
3008 ret = BTRFS_NO_LOG_SYNC;
3012 ret = start_log_trans(trans, root);
3016 ret = btrfs_log_inode(trans, root, inode, inode_only);
3021 * for regular files, if its inode is already on disk, we don't
3022 * have to worry about the parents at all. This is because
3023 * we can use the last_unlink_trans field to record renames
3024 * and other fun in this file.
3026 if (S_ISREG(inode->i_mode) &&
3027 BTRFS_I(inode)->generation <= last_committed &&
3028 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3033 inode_only = LOG_INODE_EXISTS;
3035 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3038 inode = parent->d_inode;
3039 if (root != BTRFS_I(inode)->root)
3042 if (BTRFS_I(inode)->generation >
3043 root->fs_info->last_trans_committed) {
3044 ret = btrfs_log_inode(trans, root, inode, inode_only);
3048 if (IS_ROOT(parent))
3051 parent = dget_parent(parent);
3053 old_parent = parent;
3059 BUG_ON(ret != -ENOSPC);
3060 root->fs_info->last_trans_log_full_commit = trans->transid;
3063 btrfs_end_log_trans(root);
3069 * it is not safe to log dentry if the chunk root has added new
3070 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3071 * If this returns 1, you must commit the transaction to safely get your
3074 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3075 struct btrfs_root *root, struct dentry *dentry)
3077 struct dentry *parent = dget_parent(dentry);
3080 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3087 * should be called during mount to recover any replay any log trees
3090 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3093 struct btrfs_path *path;
3094 struct btrfs_trans_handle *trans;
3095 struct btrfs_key key;
3096 struct btrfs_key found_key;
3097 struct btrfs_key tmp_key;
3098 struct btrfs_root *log;
3099 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3100 struct walk_control wc = {
3101 .process_func = process_one_buffer,
3105 path = btrfs_alloc_path();
3109 fs_info->log_root_recovering = 1;
3111 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3112 BUG_ON(IS_ERR(trans));
3117 ret = walk_log_tree(trans, log_root_tree, &wc);
3121 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3122 key.offset = (u64)-1;
3123 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3126 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3130 if (path->slots[0] == 0)
3134 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3136 btrfs_release_path(log_root_tree, path);
3137 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3140 log = btrfs_read_fs_root_no_radix(log_root_tree,
3142 BUG_ON(IS_ERR(log));
3144 tmp_key.objectid = found_key.offset;
3145 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3146 tmp_key.offset = (u64)-1;
3148 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3149 BUG_ON(!wc.replay_dest);
3151 wc.replay_dest->log_root = log;
3152 btrfs_record_root_in_trans(trans, wc.replay_dest);
3153 ret = walk_log_tree(trans, log, &wc);
3156 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3157 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3162 key.offset = found_key.offset - 1;
3163 wc.replay_dest->log_root = NULL;
3164 free_extent_buffer(log->node);
3165 free_extent_buffer(log->commit_root);
3168 if (found_key.offset == 0)
3171 btrfs_release_path(log_root_tree, path);
3173 /* step one is to pin it all, step two is to replay just inodes */
3176 wc.process_func = replay_one_buffer;
3177 wc.stage = LOG_WALK_REPLAY_INODES;
3180 /* step three is to replay everything */
3181 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3186 btrfs_free_path(path);
3188 free_extent_buffer(log_root_tree->node);
3189 log_root_tree->log_root = NULL;
3190 fs_info->log_root_recovering = 0;
3192 /* step 4: commit the transaction, which also unpins the blocks */
3193 btrfs_commit_transaction(trans, fs_info->tree_root);
3195 kfree(log_root_tree);
3200 * there are some corner cases where we want to force a full
3201 * commit instead of allowing a directory to be logged.
3203 * They revolve around files there were unlinked from the directory, and
3204 * this function updates the parent directory so that a full commit is
3205 * properly done if it is fsync'd later after the unlinks are done.
3207 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3208 struct inode *dir, struct inode *inode,
3212 * when we're logging a file, if it hasn't been renamed
3213 * or unlinked, and its inode is fully committed on disk,
3214 * we don't have to worry about walking up the directory chain
3215 * to log its parents.
3217 * So, we use the last_unlink_trans field to put this transid
3218 * into the file. When the file is logged we check it and
3219 * don't log the parents if the file is fully on disk.
3221 if (S_ISREG(inode->i_mode))
3222 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3225 * if this directory was already logged any new
3226 * names for this file/dir will get recorded
3229 if (BTRFS_I(dir)->logged_trans == trans->transid)
3233 * if the inode we're about to unlink was logged,
3234 * the log will be properly updated for any new names
3236 if (BTRFS_I(inode)->logged_trans == trans->transid)
3240 * when renaming files across directories, if the directory
3241 * there we're unlinking from gets fsync'd later on, there's
3242 * no way to find the destination directory later and fsync it
3243 * properly. So, we have to be conservative and force commits
3244 * so the new name gets discovered.
3249 /* we can safely do the unlink without any special recording */
3253 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3257 * Call this after adding a new name for a file and it will properly
3258 * update the log to reflect the new name.
3260 * It will return zero if all goes well, and it will return 1 if a
3261 * full transaction commit is required.
3263 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3264 struct inode *inode, struct inode *old_dir,
3265 struct dentry *parent)
3267 struct btrfs_root * root = BTRFS_I(inode)->root;
3270 * this will force the logging code to walk the dentry chain
3273 if (S_ISREG(inode->i_mode))
3274 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3277 * if this inode hasn't been logged and directory we're renaming it
3278 * from hasn't been logged, we don't need to log it
3280 if (BTRFS_I(inode)->logged_trans <=
3281 root->fs_info->last_trans_committed &&
3282 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3283 root->fs_info->last_trans_committed))
3286 return btrfs_log_inode_parent(trans, root, inode, parent, 1);