1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
16 #include "print-tree.h"
18 #include "compression.h"
20 #include "inode-map.h"
22 /* magic values for the inode_only field in btrfs_log_inode:
24 * LOG_INODE_ALL means to log everything
25 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 * directory trouble cases
38 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
39 * log, we must force a full commit before doing an fsync of the directory
40 * where the unlink was done.
41 * ---> record transid of last unlink/rename per directory
45 * rename foo/some_dir foo2/some_dir
47 * fsync foo/some_dir/some_file
49 * The fsync above will unlink the original some_dir without recording
50 * it in its new location (foo2). After a crash, some_dir will be gone
51 * unless the fsync of some_file forces a full commit
53 * 2) we must log any new names for any file or dir that is in the fsync
54 * log. ---> check inode while renaming/linking.
56 * 2a) we must log any new names for any file or dir during rename
57 * when the directory they are being removed from was logged.
58 * ---> check inode and old parent dir during rename
60 * 2a is actually the more important variant. With the extra logging
61 * a crash might unlink the old name without recreating the new one
63 * 3) after a crash, we must go through any directories with a link count
64 * of zero and redo the rm -rf
71 * The directory f1 was fully removed from the FS, but fsync was never
72 * called on f1, only its parent dir. After a crash the rm -rf must
73 * be replayed. This must be able to recurse down the entire
74 * directory tree. The inode link count fixup code takes care of the
79 * stages for the tree walking. The first
80 * stage (0) is to only pin down the blocks we find
81 * the second stage (1) is to make sure that all the inodes
82 * we find in the log are created in the subvolume.
84 * The last stage is to deal with directories and links and extents
85 * and all the other fun semantics
89 LOG_WALK_REPLAY_INODES,
90 LOG_WALK_REPLAY_DIR_INDEX,
94 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root, struct btrfs_inode *inode,
99 struct btrfs_log_ctx *ctx);
100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_path *path, u64 objectid);
103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_root *log,
106 struct btrfs_path *path,
107 u64 dirid, int del_all);
110 * tree logging is a special write ahead log used to make sure that
111 * fsyncs and O_SYNCs can happen without doing full tree commits.
113 * Full tree commits are expensive because they require commonly
114 * modified blocks to be recowed, creating many dirty pages in the
115 * extent tree an 4x-6x higher write load than ext3.
117 * Instead of doing a tree commit on every fsync, we use the
118 * key ranges and transaction ids to find items for a given file or directory
119 * that have changed in this transaction. Those items are copied into
120 * a special tree (one per subvolume root), that tree is written to disk
121 * and then the fsync is considered complete.
123 * After a crash, items are copied out of the log-tree back into the
124 * subvolume tree. Any file data extents found are recorded in the extent
125 * allocation tree, and the log-tree freed.
127 * The log tree is read three times, once to pin down all the extents it is
128 * using in ram and once, once to create all the inodes logged in the tree
129 * and once to do all the other items.
133 * start a sub transaction and setup the log tree
134 * this increments the log tree writer count to make the people
135 * syncing the tree wait for us to finish
137 static int start_log_trans(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root,
139 struct btrfs_log_ctx *ctx)
141 struct btrfs_fs_info *fs_info = root->fs_info;
144 mutex_lock(&root->log_mutex);
146 if (root->log_root) {
147 if (btrfs_need_log_full_commit(trans)) {
152 if (!root->log_start_pid) {
153 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
154 root->log_start_pid = current->pid;
155 } else if (root->log_start_pid != current->pid) {
156 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 mutex_lock(&fs_info->tree_log_mutex);
160 if (!fs_info->log_root_tree)
161 ret = btrfs_init_log_root_tree(trans, fs_info);
162 mutex_unlock(&fs_info->tree_log_mutex);
166 ret = btrfs_add_log_tree(trans, root);
170 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
171 root->log_start_pid = current->pid;
174 atomic_inc(&root->log_batch);
175 atomic_inc(&root->log_writers);
177 int index = root->log_transid % 2;
178 list_add_tail(&ctx->list, &root->log_ctxs[index]);
179 ctx->log_transid = root->log_transid;
183 mutex_unlock(&root->log_mutex);
188 * returns 0 if there was a log transaction running and we were able
189 * to join, or returns -ENOENT if there were not transactions
192 static int join_running_log_trans(struct btrfs_root *root)
196 mutex_lock(&root->log_mutex);
197 if (root->log_root) {
199 atomic_inc(&root->log_writers);
201 mutex_unlock(&root->log_mutex);
206 * This either makes the current running log transaction wait
207 * until you call btrfs_end_log_trans() or it makes any future
208 * log transactions wait until you call btrfs_end_log_trans()
210 void btrfs_pin_log_trans(struct btrfs_root *root)
212 mutex_lock(&root->log_mutex);
213 atomic_inc(&root->log_writers);
214 mutex_unlock(&root->log_mutex);
218 * indicate we're done making changes to the log tree
219 * and wake up anyone waiting to do a sync
221 void btrfs_end_log_trans(struct btrfs_root *root)
223 if (atomic_dec_and_test(&root->log_writers)) {
224 /* atomic_dec_and_test implies a barrier */
225 cond_wake_up_nomb(&root->log_writer_wait);
229 static int btrfs_write_tree_block(struct extent_buffer *buf)
231 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
232 buf->start + buf->len - 1);
235 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
237 filemap_fdatawait_range(buf->pages[0]->mapping,
238 buf->start, buf->start + buf->len - 1);
242 * the walk control struct is used to pass state down the chain when
243 * processing the log tree. The stage field tells us which part
244 * of the log tree processing we are currently doing. The others
245 * are state fields used for that specific part
247 struct walk_control {
248 /* should we free the extent on disk when done? This is used
249 * at transaction commit time while freeing a log tree
253 /* should we write out the extent buffer? This is used
254 * while flushing the log tree to disk during a sync
258 /* should we wait for the extent buffer io to finish? Also used
259 * while flushing the log tree to disk for a sync
263 /* pin only walk, we record which extents on disk belong to the
268 /* what stage of the replay code we're currently in */
272 * Ignore any items from the inode currently being processed. Needs
273 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
274 * the LOG_WALK_REPLAY_INODES stage.
276 bool ignore_cur_inode;
278 /* the root we are currently replaying */
279 struct btrfs_root *replay_dest;
281 /* the trans handle for the current replay */
282 struct btrfs_trans_handle *trans;
284 /* the function that gets used to process blocks we find in the
285 * tree. Note the extent_buffer might not be up to date when it is
286 * passed in, and it must be checked or read if you need the data
289 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
290 struct walk_control *wc, u64 gen, int level);
294 * process_func used to pin down extents, write them or wait on them
296 static int process_one_buffer(struct btrfs_root *log,
297 struct extent_buffer *eb,
298 struct walk_control *wc, u64 gen, int level)
300 struct btrfs_fs_info *fs_info = log->fs_info;
304 * If this fs is mixed then we need to be able to process the leaves to
305 * pin down any logged extents, so we have to read the block.
307 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
308 ret = btrfs_read_buffer(eb, gen, level, NULL);
314 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
317 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
318 if (wc->pin && btrfs_header_level(eb) == 0)
319 ret = btrfs_exclude_logged_extents(eb);
321 btrfs_write_tree_block(eb);
323 btrfs_wait_tree_block_writeback(eb);
329 * Item overwrite used by replay and tree logging. eb, slot and key all refer
330 * to the src data we are copying out.
332 * root is the tree we are copying into, and path is a scratch
333 * path for use in this function (it should be released on entry and
334 * will be released on exit).
336 * If the key is already in the destination tree the existing item is
337 * overwritten. If the existing item isn't big enough, it is extended.
338 * If it is too large, it is truncated.
340 * If the key isn't in the destination yet, a new item is inserted.
342 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
343 struct btrfs_root *root,
344 struct btrfs_path *path,
345 struct extent_buffer *eb, int slot,
346 struct btrfs_key *key)
350 u64 saved_i_size = 0;
351 int save_old_i_size = 0;
352 unsigned long src_ptr;
353 unsigned long dst_ptr;
354 int overwrite_root = 0;
355 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
357 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
360 item_size = btrfs_item_size_nr(eb, slot);
361 src_ptr = btrfs_item_ptr_offset(eb, slot);
363 /* look for the key in the destination tree */
364 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
371 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
373 if (dst_size != item_size)
376 if (item_size == 0) {
377 btrfs_release_path(path);
380 dst_copy = kmalloc(item_size, GFP_NOFS);
381 src_copy = kmalloc(item_size, GFP_NOFS);
382 if (!dst_copy || !src_copy) {
383 btrfs_release_path(path);
389 read_extent_buffer(eb, src_copy, src_ptr, item_size);
391 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
392 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
394 ret = memcmp(dst_copy, src_copy, item_size);
399 * they have the same contents, just return, this saves
400 * us from cowing blocks in the destination tree and doing
401 * extra writes that may not have been done by a previous
405 btrfs_release_path(path);
410 * We need to load the old nbytes into the inode so when we
411 * replay the extents we've logged we get the right nbytes.
414 struct btrfs_inode_item *item;
418 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
419 struct btrfs_inode_item);
420 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
421 item = btrfs_item_ptr(eb, slot,
422 struct btrfs_inode_item);
423 btrfs_set_inode_nbytes(eb, item, nbytes);
426 * If this is a directory we need to reset the i_size to
427 * 0 so that we can set it up properly when replaying
428 * the rest of the items in this log.
430 mode = btrfs_inode_mode(eb, item);
432 btrfs_set_inode_size(eb, item, 0);
434 } else if (inode_item) {
435 struct btrfs_inode_item *item;
439 * New inode, set nbytes to 0 so that the nbytes comes out
440 * properly when we replay the extents.
442 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
443 btrfs_set_inode_nbytes(eb, item, 0);
446 * If this is a directory we need to reset the i_size to 0 so
447 * that we can set it up properly when replaying the rest of
448 * the items in this log.
450 mode = btrfs_inode_mode(eb, item);
452 btrfs_set_inode_size(eb, item, 0);
455 btrfs_release_path(path);
456 /* try to insert the key into the destination tree */
457 path->skip_release_on_error = 1;
458 ret = btrfs_insert_empty_item(trans, root, path,
460 path->skip_release_on_error = 0;
462 /* make sure any existing item is the correct size */
463 if (ret == -EEXIST || ret == -EOVERFLOW) {
465 found_size = btrfs_item_size_nr(path->nodes[0],
467 if (found_size > item_size)
468 btrfs_truncate_item(path, item_size, 1);
469 else if (found_size < item_size)
470 btrfs_extend_item(path, item_size - found_size);
474 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
477 /* don't overwrite an existing inode if the generation number
478 * was logged as zero. This is done when the tree logging code
479 * is just logging an inode to make sure it exists after recovery.
481 * Also, don't overwrite i_size on directories during replay.
482 * log replay inserts and removes directory items based on the
483 * state of the tree found in the subvolume, and i_size is modified
486 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
487 struct btrfs_inode_item *src_item;
488 struct btrfs_inode_item *dst_item;
490 src_item = (struct btrfs_inode_item *)src_ptr;
491 dst_item = (struct btrfs_inode_item *)dst_ptr;
493 if (btrfs_inode_generation(eb, src_item) == 0) {
494 struct extent_buffer *dst_eb = path->nodes[0];
495 const u64 ino_size = btrfs_inode_size(eb, src_item);
498 * For regular files an ino_size == 0 is used only when
499 * logging that an inode exists, as part of a directory
500 * fsync, and the inode wasn't fsynced before. In this
501 * case don't set the size of the inode in the fs/subvol
502 * tree, otherwise we would be throwing valid data away.
504 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
505 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
507 struct btrfs_map_token token;
509 btrfs_init_map_token(&token, dst_eb);
510 btrfs_set_token_inode_size(dst_eb, dst_item,
516 if (overwrite_root &&
517 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
518 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
520 saved_i_size = btrfs_inode_size(path->nodes[0],
525 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
528 if (save_old_i_size) {
529 struct btrfs_inode_item *dst_item;
530 dst_item = (struct btrfs_inode_item *)dst_ptr;
531 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
534 /* make sure the generation is filled in */
535 if (key->type == BTRFS_INODE_ITEM_KEY) {
536 struct btrfs_inode_item *dst_item;
537 dst_item = (struct btrfs_inode_item *)dst_ptr;
538 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
539 btrfs_set_inode_generation(path->nodes[0], dst_item,
544 btrfs_mark_buffer_dirty(path->nodes[0]);
545 btrfs_release_path(path);
550 * simple helper to read an inode off the disk from a given root
551 * This can only be called for subvolume roots and not for the log
553 static noinline struct inode *read_one_inode(struct btrfs_root *root,
556 struct btrfs_key key;
559 key.objectid = objectid;
560 key.type = BTRFS_INODE_ITEM_KEY;
562 inode = btrfs_iget(root->fs_info->sb, &key, root);
568 /* replays a single extent in 'eb' at 'slot' with 'key' into the
569 * subvolume 'root'. path is released on entry and should be released
572 * extents in the log tree have not been allocated out of the extent
573 * tree yet. So, this completes the allocation, taking a reference
574 * as required if the extent already exists or creating a new extent
575 * if it isn't in the extent allocation tree yet.
577 * The extent is inserted into the file, dropping any existing extents
578 * from the file that overlap the new one.
580 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
581 struct btrfs_root *root,
582 struct btrfs_path *path,
583 struct extent_buffer *eb, int slot,
584 struct btrfs_key *key)
586 struct btrfs_fs_info *fs_info = root->fs_info;
589 u64 start = key->offset;
591 struct btrfs_file_extent_item *item;
592 struct inode *inode = NULL;
596 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
597 found_type = btrfs_file_extent_type(eb, item);
599 if (found_type == BTRFS_FILE_EXTENT_REG ||
600 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
601 nbytes = btrfs_file_extent_num_bytes(eb, item);
602 extent_end = start + nbytes;
605 * We don't add to the inodes nbytes if we are prealloc or a
608 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
610 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
611 size = btrfs_file_extent_ram_bytes(eb, item);
612 nbytes = btrfs_file_extent_ram_bytes(eb, item);
613 extent_end = ALIGN(start + size,
614 fs_info->sectorsize);
620 inode = read_one_inode(root, key->objectid);
627 * first check to see if we already have this extent in the
628 * file. This must be done before the btrfs_drop_extents run
629 * so we don't try to drop this extent.
631 ret = btrfs_lookup_file_extent(trans, root, path,
632 btrfs_ino(BTRFS_I(inode)), start, 0);
635 (found_type == BTRFS_FILE_EXTENT_REG ||
636 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
637 struct btrfs_file_extent_item cmp1;
638 struct btrfs_file_extent_item cmp2;
639 struct btrfs_file_extent_item *existing;
640 struct extent_buffer *leaf;
642 leaf = path->nodes[0];
643 existing = btrfs_item_ptr(leaf, path->slots[0],
644 struct btrfs_file_extent_item);
646 read_extent_buffer(eb, &cmp1, (unsigned long)item,
648 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
652 * we already have a pointer to this exact extent,
653 * we don't have to do anything
655 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
656 btrfs_release_path(path);
660 btrfs_release_path(path);
662 /* drop any overlapping extents */
663 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
667 if (found_type == BTRFS_FILE_EXTENT_REG ||
668 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
670 unsigned long dest_offset;
671 struct btrfs_key ins;
673 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
674 btrfs_fs_incompat(fs_info, NO_HOLES))
677 ret = btrfs_insert_empty_item(trans, root, path, key,
681 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
683 copy_extent_buffer(path->nodes[0], eb, dest_offset,
684 (unsigned long)item, sizeof(*item));
686 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
687 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
688 ins.type = BTRFS_EXTENT_ITEM_KEY;
689 offset = key->offset - btrfs_file_extent_offset(eb, item);
692 * Manually record dirty extent, as here we did a shallow
693 * file extent item copy and skip normal backref update,
694 * but modifying extent tree all by ourselves.
695 * So need to manually record dirty extent for qgroup,
696 * as the owner of the file extent changed from log tree
697 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
699 ret = btrfs_qgroup_trace_extent(trans,
700 btrfs_file_extent_disk_bytenr(eb, item),
701 btrfs_file_extent_disk_num_bytes(eb, item),
706 if (ins.objectid > 0) {
707 struct btrfs_ref ref = { 0 };
710 LIST_HEAD(ordered_sums);
713 * is this extent already allocated in the extent
714 * allocation tree? If so, just add a reference
716 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
719 btrfs_init_generic_ref(&ref,
720 BTRFS_ADD_DELAYED_REF,
721 ins.objectid, ins.offset, 0);
722 btrfs_init_data_ref(&ref,
723 root->root_key.objectid,
724 key->objectid, offset);
725 ret = btrfs_inc_extent_ref(trans, &ref);
730 * insert the extent pointer in the extent
733 ret = btrfs_alloc_logged_file_extent(trans,
734 root->root_key.objectid,
735 key->objectid, offset, &ins);
739 btrfs_release_path(path);
741 if (btrfs_file_extent_compression(eb, item)) {
742 csum_start = ins.objectid;
743 csum_end = csum_start + ins.offset;
745 csum_start = ins.objectid +
746 btrfs_file_extent_offset(eb, item);
747 csum_end = csum_start +
748 btrfs_file_extent_num_bytes(eb, item);
751 ret = btrfs_lookup_csums_range(root->log_root,
752 csum_start, csum_end - 1,
757 * Now delete all existing cums in the csum root that
758 * cover our range. We do this because we can have an
759 * extent that is completely referenced by one file
760 * extent item and partially referenced by another
761 * file extent item (like after using the clone or
762 * extent_same ioctls). In this case if we end up doing
763 * the replay of the one that partially references the
764 * extent first, and we do not do the csum deletion
765 * below, we can get 2 csum items in the csum tree that
766 * overlap each other. For example, imagine our log has
767 * the two following file extent items:
769 * key (257 EXTENT_DATA 409600)
770 * extent data disk byte 12845056 nr 102400
771 * extent data offset 20480 nr 20480 ram 102400
773 * key (257 EXTENT_DATA 819200)
774 * extent data disk byte 12845056 nr 102400
775 * extent data offset 0 nr 102400 ram 102400
777 * Where the second one fully references the 100K extent
778 * that starts at disk byte 12845056, and the log tree
779 * has a single csum item that covers the entire range
782 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
784 * After the first file extent item is replayed, the
785 * csum tree gets the following csum item:
787 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
789 * Which covers the 20K sub-range starting at offset 20K
790 * of our extent. Now when we replay the second file
791 * extent item, if we do not delete existing csum items
792 * that cover any of its blocks, we end up getting two
793 * csum items in our csum tree that overlap each other:
795 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
796 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
798 * Which is a problem, because after this anyone trying
799 * to lookup up for the checksum of any block of our
800 * extent starting at an offset of 40K or higher, will
801 * end up looking at the second csum item only, which
802 * does not contain the checksum for any block starting
803 * at offset 40K or higher of our extent.
805 while (!list_empty(&ordered_sums)) {
806 struct btrfs_ordered_sum *sums;
807 sums = list_entry(ordered_sums.next,
808 struct btrfs_ordered_sum,
811 ret = btrfs_del_csums(trans,
816 ret = btrfs_csum_file_blocks(trans,
817 fs_info->csum_root, sums);
818 list_del(&sums->list);
824 btrfs_release_path(path);
826 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
827 /* inline extents are easy, we just overwrite them */
828 ret = overwrite_item(trans, root, path, eb, slot, key);
833 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
838 inode_add_bytes(inode, nbytes);
840 ret = btrfs_update_inode(trans, root, inode);
848 * when cleaning up conflicts between the directory names in the
849 * subvolume, directory names in the log and directory names in the
850 * inode back references, we may have to unlink inodes from directories.
852 * This is a helper function to do the unlink of a specific directory
855 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
856 struct btrfs_root *root,
857 struct btrfs_path *path,
858 struct btrfs_inode *dir,
859 struct btrfs_dir_item *di)
864 struct extent_buffer *leaf;
865 struct btrfs_key location;
868 leaf = path->nodes[0];
870 btrfs_dir_item_key_to_cpu(leaf, di, &location);
871 name_len = btrfs_dir_name_len(leaf, di);
872 name = kmalloc(name_len, GFP_NOFS);
876 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
877 btrfs_release_path(path);
879 inode = read_one_inode(root, location.objectid);
885 ret = link_to_fixup_dir(trans, root, path, location.objectid);
889 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
894 ret = btrfs_run_delayed_items(trans);
902 * helper function to see if a given name and sequence number found
903 * in an inode back reference are already in a directory and correctly
904 * point to this inode
906 static noinline int inode_in_dir(struct btrfs_root *root,
907 struct btrfs_path *path,
908 u64 dirid, u64 objectid, u64 index,
909 const char *name, int name_len)
911 struct btrfs_dir_item *di;
912 struct btrfs_key location;
915 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
916 index, name, name_len, 0);
917 if (di && !IS_ERR(di)) {
918 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
919 if (location.objectid != objectid)
923 btrfs_release_path(path);
925 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
926 if (di && !IS_ERR(di)) {
927 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
928 if (location.objectid != objectid)
934 btrfs_release_path(path);
939 * helper function to check a log tree for a named back reference in
940 * an inode. This is used to decide if a back reference that is
941 * found in the subvolume conflicts with what we find in the log.
943 * inode backreferences may have multiple refs in a single item,
944 * during replay we process one reference at a time, and we don't
945 * want to delete valid links to a file from the subvolume if that
946 * link is also in the log.
948 static noinline int backref_in_log(struct btrfs_root *log,
949 struct btrfs_key *key,
951 const char *name, int namelen)
953 struct btrfs_path *path;
956 path = btrfs_alloc_path();
960 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
963 } else if (ret == 1) {
968 if (key->type == BTRFS_INODE_EXTREF_KEY)
969 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
974 ret = !!btrfs_find_name_in_backref(path->nodes[0],
978 btrfs_free_path(path);
982 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
983 struct btrfs_root *root,
984 struct btrfs_path *path,
985 struct btrfs_root *log_root,
986 struct btrfs_inode *dir,
987 struct btrfs_inode *inode,
988 u64 inode_objectid, u64 parent_objectid,
989 u64 ref_index, char *name, int namelen,
995 struct extent_buffer *leaf;
996 struct btrfs_dir_item *di;
997 struct btrfs_key search_key;
998 struct btrfs_inode_extref *extref;
1001 /* Search old style refs */
1002 search_key.objectid = inode_objectid;
1003 search_key.type = BTRFS_INODE_REF_KEY;
1004 search_key.offset = parent_objectid;
1005 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1007 struct btrfs_inode_ref *victim_ref;
1009 unsigned long ptr_end;
1011 leaf = path->nodes[0];
1013 /* are we trying to overwrite a back ref for the root directory
1014 * if so, just jump out, we're done
1016 if (search_key.objectid == search_key.offset)
1019 /* check all the names in this back reference to see
1020 * if they are in the log. if so, we allow them to stay
1021 * otherwise they must be unlinked as a conflict
1023 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1024 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1025 while (ptr < ptr_end) {
1026 victim_ref = (struct btrfs_inode_ref *)ptr;
1027 victim_name_len = btrfs_inode_ref_name_len(leaf,
1029 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1033 read_extent_buffer(leaf, victim_name,
1034 (unsigned long)(victim_ref + 1),
1037 ret = backref_in_log(log_root, &search_key,
1038 parent_objectid, victim_name,
1044 inc_nlink(&inode->vfs_inode);
1045 btrfs_release_path(path);
1047 ret = btrfs_unlink_inode(trans, root, dir, inode,
1048 victim_name, victim_name_len);
1052 ret = btrfs_run_delayed_items(trans);
1060 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1064 * NOTE: we have searched root tree and checked the
1065 * corresponding ref, it does not need to check again.
1069 btrfs_release_path(path);
1071 /* Same search but for extended refs */
1072 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1073 inode_objectid, parent_objectid, 0,
1075 if (!IS_ERR_OR_NULL(extref)) {
1079 struct inode *victim_parent;
1081 leaf = path->nodes[0];
1083 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1084 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1086 while (cur_offset < item_size) {
1087 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1089 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1091 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1094 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1097 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1100 search_key.objectid = inode_objectid;
1101 search_key.type = BTRFS_INODE_EXTREF_KEY;
1102 search_key.offset = btrfs_extref_hash(parent_objectid,
1105 ret = backref_in_log(log_root, &search_key,
1106 parent_objectid, victim_name,
1112 victim_parent = read_one_inode(root,
1114 if (victim_parent) {
1115 inc_nlink(&inode->vfs_inode);
1116 btrfs_release_path(path);
1118 ret = btrfs_unlink_inode(trans, root,
1119 BTRFS_I(victim_parent),
1124 ret = btrfs_run_delayed_items(
1127 iput(victim_parent);
1136 cur_offset += victim_name_len + sizeof(*extref);
1140 btrfs_release_path(path);
1142 /* look for a conflicting sequence number */
1143 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1144 ref_index, name, namelen, 0);
1145 if (di && !IS_ERR(di)) {
1146 ret = drop_one_dir_item(trans, root, path, dir, di);
1150 btrfs_release_path(path);
1152 /* look for a conflicting name */
1153 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1155 if (di && !IS_ERR(di)) {
1156 ret = drop_one_dir_item(trans, root, path, dir, di);
1160 btrfs_release_path(path);
1165 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1166 u32 *namelen, char **name, u64 *index,
1167 u64 *parent_objectid)
1169 struct btrfs_inode_extref *extref;
1171 extref = (struct btrfs_inode_extref *)ref_ptr;
1173 *namelen = btrfs_inode_extref_name_len(eb, extref);
1174 *name = kmalloc(*namelen, GFP_NOFS);
1178 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1182 *index = btrfs_inode_extref_index(eb, extref);
1183 if (parent_objectid)
1184 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1189 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1190 u32 *namelen, char **name, u64 *index)
1192 struct btrfs_inode_ref *ref;
1194 ref = (struct btrfs_inode_ref *)ref_ptr;
1196 *namelen = btrfs_inode_ref_name_len(eb, ref);
1197 *name = kmalloc(*namelen, GFP_NOFS);
1201 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1204 *index = btrfs_inode_ref_index(eb, ref);
1210 * Take an inode reference item from the log tree and iterate all names from the
1211 * inode reference item in the subvolume tree with the same key (if it exists).
1212 * For any name that is not in the inode reference item from the log tree, do a
1213 * proper unlink of that name (that is, remove its entry from the inode
1214 * reference item and both dir index keys).
1216 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1217 struct btrfs_root *root,
1218 struct btrfs_path *path,
1219 struct btrfs_inode *inode,
1220 struct extent_buffer *log_eb,
1222 struct btrfs_key *key)
1225 unsigned long ref_ptr;
1226 unsigned long ref_end;
1227 struct extent_buffer *eb;
1230 btrfs_release_path(path);
1231 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1239 eb = path->nodes[0];
1240 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1241 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1242 while (ref_ptr < ref_end) {
1247 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1248 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1251 parent_id = key->offset;
1252 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1258 if (key->type == BTRFS_INODE_EXTREF_KEY)
1259 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1263 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1269 btrfs_release_path(path);
1270 dir = read_one_inode(root, parent_id);
1276 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1277 inode, name, namelen);
1287 if (key->type == BTRFS_INODE_EXTREF_KEY)
1288 ref_ptr += sizeof(struct btrfs_inode_extref);
1290 ref_ptr += sizeof(struct btrfs_inode_ref);
1294 btrfs_release_path(path);
1298 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1299 const u8 ref_type, const char *name,
1302 struct btrfs_key key;
1303 struct btrfs_path *path;
1304 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1307 path = btrfs_alloc_path();
1311 key.objectid = btrfs_ino(BTRFS_I(inode));
1312 key.type = ref_type;
1313 if (key.type == BTRFS_INODE_REF_KEY)
1314 key.offset = parent_id;
1316 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1318 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1325 if (key.type == BTRFS_INODE_EXTREF_KEY)
1326 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1327 path->slots[0], parent_id, name, namelen);
1329 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1333 btrfs_free_path(path);
1337 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1338 struct inode *dir, struct inode *inode, const char *name,
1339 int namelen, u64 ref_index)
1341 struct btrfs_dir_item *dir_item;
1342 struct btrfs_key key;
1343 struct btrfs_path *path;
1344 struct inode *other_inode = NULL;
1347 path = btrfs_alloc_path();
1351 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1352 btrfs_ino(BTRFS_I(dir)),
1355 btrfs_release_path(path);
1357 } else if (IS_ERR(dir_item)) {
1358 ret = PTR_ERR(dir_item);
1363 * Our inode's dentry collides with the dentry of another inode which is
1364 * in the log but not yet processed since it has a higher inode number.
1365 * So delete that other dentry.
1367 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1368 btrfs_release_path(path);
1369 other_inode = read_one_inode(root, key.objectid);
1374 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1379 * If we dropped the link count to 0, bump it so that later the iput()
1380 * on the inode will not free it. We will fixup the link count later.
1382 if (other_inode->i_nlink == 0)
1383 inc_nlink(other_inode);
1385 ret = btrfs_run_delayed_items(trans);
1389 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1390 name, namelen, 0, ref_index);
1393 btrfs_free_path(path);
1399 * replay one inode back reference item found in the log tree.
1400 * eb, slot and key refer to the buffer and key found in the log tree.
1401 * root is the destination we are replaying into, and path is for temp
1402 * use by this function. (it should be released on return).
1404 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1405 struct btrfs_root *root,
1406 struct btrfs_root *log,
1407 struct btrfs_path *path,
1408 struct extent_buffer *eb, int slot,
1409 struct btrfs_key *key)
1411 struct inode *dir = NULL;
1412 struct inode *inode = NULL;
1413 unsigned long ref_ptr;
1414 unsigned long ref_end;
1418 int search_done = 0;
1419 int log_ref_ver = 0;
1420 u64 parent_objectid;
1423 int ref_struct_size;
1425 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1426 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1428 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1429 struct btrfs_inode_extref *r;
1431 ref_struct_size = sizeof(struct btrfs_inode_extref);
1433 r = (struct btrfs_inode_extref *)ref_ptr;
1434 parent_objectid = btrfs_inode_extref_parent(eb, r);
1436 ref_struct_size = sizeof(struct btrfs_inode_ref);
1437 parent_objectid = key->offset;
1439 inode_objectid = key->objectid;
1442 * it is possible that we didn't log all the parent directories
1443 * for a given inode. If we don't find the dir, just don't
1444 * copy the back ref in. The link count fixup code will take
1447 dir = read_one_inode(root, parent_objectid);
1453 inode = read_one_inode(root, inode_objectid);
1459 while (ref_ptr < ref_end) {
1461 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1462 &ref_index, &parent_objectid);
1464 * parent object can change from one array
1468 dir = read_one_inode(root, parent_objectid);
1474 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1480 /* if we already have a perfect match, we're done */
1481 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1482 btrfs_ino(BTRFS_I(inode)), ref_index,
1485 * look for a conflicting back reference in the
1486 * metadata. if we find one we have to unlink that name
1487 * of the file before we add our new link. Later on, we
1488 * overwrite any existing back reference, and we don't
1489 * want to create dangling pointers in the directory.
1493 ret = __add_inode_ref(trans, root, path, log,
1498 ref_index, name, namelen,
1508 * If a reference item already exists for this inode
1509 * with the same parent and name, but different index,
1510 * drop it and the corresponding directory index entries
1511 * from the parent before adding the new reference item
1512 * and dir index entries, otherwise we would fail with
1513 * -EEXIST returned from btrfs_add_link() below.
1515 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1518 ret = btrfs_unlink_inode(trans, root,
1523 * If we dropped the link count to 0, bump it so
1524 * that later the iput() on the inode will not
1525 * free it. We will fixup the link count later.
1527 if (!ret && inode->i_nlink == 0)
1533 /* insert our name */
1534 ret = add_link(trans, root, dir, inode, name, namelen,
1539 btrfs_update_inode(trans, root, inode);
1542 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1552 * Before we overwrite the inode reference item in the subvolume tree
1553 * with the item from the log tree, we must unlink all names from the
1554 * parent directory that are in the subvolume's tree inode reference
1555 * item, otherwise we end up with an inconsistent subvolume tree where
1556 * dir index entries exist for a name but there is no inode reference
1557 * item with the same name.
1559 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1564 /* finally write the back reference in the inode */
1565 ret = overwrite_item(trans, root, path, eb, slot, key);
1567 btrfs_release_path(path);
1574 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1575 struct btrfs_root *root, u64 ino)
1579 ret = btrfs_insert_orphan_item(trans, root, ino);
1586 static int count_inode_extrefs(struct btrfs_root *root,
1587 struct btrfs_inode *inode, struct btrfs_path *path)
1591 unsigned int nlink = 0;
1594 u64 inode_objectid = btrfs_ino(inode);
1597 struct btrfs_inode_extref *extref;
1598 struct extent_buffer *leaf;
1601 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1606 leaf = path->nodes[0];
1607 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1608 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1611 while (cur_offset < item_size) {
1612 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1613 name_len = btrfs_inode_extref_name_len(leaf, extref);
1617 cur_offset += name_len + sizeof(*extref);
1621 btrfs_release_path(path);
1623 btrfs_release_path(path);
1625 if (ret < 0 && ret != -ENOENT)
1630 static int count_inode_refs(struct btrfs_root *root,
1631 struct btrfs_inode *inode, struct btrfs_path *path)
1634 struct btrfs_key key;
1635 unsigned int nlink = 0;
1637 unsigned long ptr_end;
1639 u64 ino = btrfs_ino(inode);
1642 key.type = BTRFS_INODE_REF_KEY;
1643 key.offset = (u64)-1;
1646 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1650 if (path->slots[0] == 0)
1655 btrfs_item_key_to_cpu(path->nodes[0], &key,
1657 if (key.objectid != ino ||
1658 key.type != BTRFS_INODE_REF_KEY)
1660 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1661 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1663 while (ptr < ptr_end) {
1664 struct btrfs_inode_ref *ref;
1666 ref = (struct btrfs_inode_ref *)ptr;
1667 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1669 ptr = (unsigned long)(ref + 1) + name_len;
1673 if (key.offset == 0)
1675 if (path->slots[0] > 0) {
1680 btrfs_release_path(path);
1682 btrfs_release_path(path);
1688 * There are a few corners where the link count of the file can't
1689 * be properly maintained during replay. So, instead of adding
1690 * lots of complexity to the log code, we just scan the backrefs
1691 * for any file that has been through replay.
1693 * The scan will update the link count on the inode to reflect the
1694 * number of back refs found. If it goes down to zero, the iput
1695 * will free the inode.
1697 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1698 struct btrfs_root *root,
1699 struct inode *inode)
1701 struct btrfs_path *path;
1704 u64 ino = btrfs_ino(BTRFS_I(inode));
1706 path = btrfs_alloc_path();
1710 ret = count_inode_refs(root, BTRFS_I(inode), path);
1716 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1724 if (nlink != inode->i_nlink) {
1725 set_nlink(inode, nlink);
1726 btrfs_update_inode(trans, root, inode);
1728 BTRFS_I(inode)->index_cnt = (u64)-1;
1730 if (inode->i_nlink == 0) {
1731 if (S_ISDIR(inode->i_mode)) {
1732 ret = replay_dir_deletes(trans, root, NULL, path,
1737 ret = insert_orphan_item(trans, root, ino);
1741 btrfs_free_path(path);
1745 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1746 struct btrfs_root *root,
1747 struct btrfs_path *path)
1750 struct btrfs_key key;
1751 struct inode *inode;
1753 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1754 key.type = BTRFS_ORPHAN_ITEM_KEY;
1755 key.offset = (u64)-1;
1757 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762 if (path->slots[0] == 0)
1767 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1768 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1769 key.type != BTRFS_ORPHAN_ITEM_KEY)
1772 ret = btrfs_del_item(trans, root, path);
1776 btrfs_release_path(path);
1777 inode = read_one_inode(root, key.offset);
1781 ret = fixup_inode_link_count(trans, root, inode);
1787 * fixup on a directory may create new entries,
1788 * make sure we always look for the highset possible
1791 key.offset = (u64)-1;
1795 btrfs_release_path(path);
1801 * record a given inode in the fixup dir so we can check its link
1802 * count when replay is done. The link count is incremented here
1803 * so the inode won't go away until we check it
1805 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1806 struct btrfs_root *root,
1807 struct btrfs_path *path,
1810 struct btrfs_key key;
1812 struct inode *inode;
1814 inode = read_one_inode(root, objectid);
1818 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1819 key.type = BTRFS_ORPHAN_ITEM_KEY;
1820 key.offset = objectid;
1822 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1824 btrfs_release_path(path);
1826 if (!inode->i_nlink)
1827 set_nlink(inode, 1);
1830 ret = btrfs_update_inode(trans, root, inode);
1831 } else if (ret == -EEXIST) {
1834 BUG(); /* Logic Error */
1842 * when replaying the log for a directory, we only insert names
1843 * for inodes that actually exist. This means an fsync on a directory
1844 * does not implicitly fsync all the new files in it
1846 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1847 struct btrfs_root *root,
1848 u64 dirid, u64 index,
1849 char *name, int name_len,
1850 struct btrfs_key *location)
1852 struct inode *inode;
1856 inode = read_one_inode(root, location->objectid);
1860 dir = read_one_inode(root, dirid);
1866 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1867 name_len, 1, index);
1869 /* FIXME, put inode into FIXUP list */
1877 * take a single entry in a log directory item and replay it into
1880 * if a conflicting item exists in the subdirectory already,
1881 * the inode it points to is unlinked and put into the link count
1884 * If a name from the log points to a file or directory that does
1885 * not exist in the FS, it is skipped. fsyncs on directories
1886 * do not force down inodes inside that directory, just changes to the
1887 * names or unlinks in a directory.
1889 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1890 * non-existing inode) and 1 if the name was replayed.
1892 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1893 struct btrfs_root *root,
1894 struct btrfs_path *path,
1895 struct extent_buffer *eb,
1896 struct btrfs_dir_item *di,
1897 struct btrfs_key *key)
1901 struct btrfs_dir_item *dst_di;
1902 struct btrfs_key found_key;
1903 struct btrfs_key log_key;
1908 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1909 bool name_added = false;
1911 dir = read_one_inode(root, key->objectid);
1915 name_len = btrfs_dir_name_len(eb, di);
1916 name = kmalloc(name_len, GFP_NOFS);
1922 log_type = btrfs_dir_type(eb, di);
1923 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1926 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1927 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1932 btrfs_release_path(path);
1934 if (key->type == BTRFS_DIR_ITEM_KEY) {
1935 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1937 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1938 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1947 if (IS_ERR_OR_NULL(dst_di)) {
1948 /* we need a sequence number to insert, so we only
1949 * do inserts for the BTRFS_DIR_INDEX_KEY types
1951 if (key->type != BTRFS_DIR_INDEX_KEY)
1956 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1957 /* the existing item matches the logged item */
1958 if (found_key.objectid == log_key.objectid &&
1959 found_key.type == log_key.type &&
1960 found_key.offset == log_key.offset &&
1961 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1962 update_size = false;
1967 * don't drop the conflicting directory entry if the inode
1968 * for the new entry doesn't exist
1973 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1977 if (key->type == BTRFS_DIR_INDEX_KEY)
1980 btrfs_release_path(path);
1981 if (!ret && update_size) {
1982 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1983 ret = btrfs_update_inode(trans, root, dir);
1987 if (!ret && name_added)
1993 * Check if the inode reference exists in the log for the given name,
1994 * inode and parent inode
1996 found_key.objectid = log_key.objectid;
1997 found_key.type = BTRFS_INODE_REF_KEY;
1998 found_key.offset = key->objectid;
1999 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2003 /* The dentry will be added later. */
2005 update_size = false;
2009 found_key.objectid = log_key.objectid;
2010 found_key.type = BTRFS_INODE_EXTREF_KEY;
2011 found_key.offset = key->objectid;
2012 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2017 /* The dentry will be added later. */
2019 update_size = false;
2022 btrfs_release_path(path);
2023 ret = insert_one_name(trans, root, key->objectid, key->offset,
2024 name, name_len, &log_key);
2025 if (ret && ret != -ENOENT && ret != -EEXIST)
2029 update_size = false;
2035 * find all the names in a directory item and reconcile them into
2036 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2037 * one name in a directory item, but the same code gets used for
2038 * both directory index types
2040 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2041 struct btrfs_root *root,
2042 struct btrfs_path *path,
2043 struct extent_buffer *eb, int slot,
2044 struct btrfs_key *key)
2047 u32 item_size = btrfs_item_size_nr(eb, slot);
2048 struct btrfs_dir_item *di;
2051 unsigned long ptr_end;
2052 struct btrfs_path *fixup_path = NULL;
2054 ptr = btrfs_item_ptr_offset(eb, slot);
2055 ptr_end = ptr + item_size;
2056 while (ptr < ptr_end) {
2057 di = (struct btrfs_dir_item *)ptr;
2058 name_len = btrfs_dir_name_len(eb, di);
2059 ret = replay_one_name(trans, root, path, eb, di, key);
2062 ptr = (unsigned long)(di + 1);
2066 * If this entry refers to a non-directory (directories can not
2067 * have a link count > 1) and it was added in the transaction
2068 * that was not committed, make sure we fixup the link count of
2069 * the inode it the entry points to. Otherwise something like
2070 * the following would result in a directory pointing to an
2071 * inode with a wrong link that does not account for this dir
2079 * ln testdir/bar testdir/bar_link
2080 * ln testdir/foo testdir/foo_link
2081 * xfs_io -c "fsync" testdir/bar
2085 * mount fs, log replay happens
2087 * File foo would remain with a link count of 1 when it has two
2088 * entries pointing to it in the directory testdir. This would
2089 * make it impossible to ever delete the parent directory has
2090 * it would result in stale dentries that can never be deleted.
2092 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2093 struct btrfs_key di_key;
2096 fixup_path = btrfs_alloc_path();
2103 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2104 ret = link_to_fixup_dir(trans, root, fixup_path,
2111 btrfs_free_path(fixup_path);
2116 * directory replay has two parts. There are the standard directory
2117 * items in the log copied from the subvolume, and range items
2118 * created in the log while the subvolume was logged.
2120 * The range items tell us which parts of the key space the log
2121 * is authoritative for. During replay, if a key in the subvolume
2122 * directory is in a logged range item, but not actually in the log
2123 * that means it was deleted from the directory before the fsync
2124 * and should be removed.
2126 static noinline int find_dir_range(struct btrfs_root *root,
2127 struct btrfs_path *path,
2128 u64 dirid, int key_type,
2129 u64 *start_ret, u64 *end_ret)
2131 struct btrfs_key key;
2133 struct btrfs_dir_log_item *item;
2137 if (*start_ret == (u64)-1)
2140 key.objectid = dirid;
2141 key.type = key_type;
2142 key.offset = *start_ret;
2144 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2148 if (path->slots[0] == 0)
2153 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2155 if (key.type != key_type || key.objectid != dirid) {
2159 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2160 struct btrfs_dir_log_item);
2161 found_end = btrfs_dir_log_end(path->nodes[0], item);
2163 if (*start_ret >= key.offset && *start_ret <= found_end) {
2165 *start_ret = key.offset;
2166 *end_ret = found_end;
2171 /* check the next slot in the tree to see if it is a valid item */
2172 nritems = btrfs_header_nritems(path->nodes[0]);
2174 if (path->slots[0] >= nritems) {
2175 ret = btrfs_next_leaf(root, path);
2180 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2182 if (key.type != key_type || key.objectid != dirid) {
2186 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2187 struct btrfs_dir_log_item);
2188 found_end = btrfs_dir_log_end(path->nodes[0], item);
2189 *start_ret = key.offset;
2190 *end_ret = found_end;
2193 btrfs_release_path(path);
2198 * this looks for a given directory item in the log. If the directory
2199 * item is not in the log, the item is removed and the inode it points
2202 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2203 struct btrfs_root *root,
2204 struct btrfs_root *log,
2205 struct btrfs_path *path,
2206 struct btrfs_path *log_path,
2208 struct btrfs_key *dir_key)
2211 struct extent_buffer *eb;
2214 struct btrfs_dir_item *di;
2215 struct btrfs_dir_item *log_di;
2218 unsigned long ptr_end;
2220 struct inode *inode;
2221 struct btrfs_key location;
2224 eb = path->nodes[0];
2225 slot = path->slots[0];
2226 item_size = btrfs_item_size_nr(eb, slot);
2227 ptr = btrfs_item_ptr_offset(eb, slot);
2228 ptr_end = ptr + item_size;
2229 while (ptr < ptr_end) {
2230 di = (struct btrfs_dir_item *)ptr;
2231 name_len = btrfs_dir_name_len(eb, di);
2232 name = kmalloc(name_len, GFP_NOFS);
2237 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2240 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2241 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2244 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2245 log_di = btrfs_lookup_dir_index_item(trans, log,
2251 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2252 btrfs_dir_item_key_to_cpu(eb, di, &location);
2253 btrfs_release_path(path);
2254 btrfs_release_path(log_path);
2255 inode = read_one_inode(root, location.objectid);
2261 ret = link_to_fixup_dir(trans, root,
2262 path, location.objectid);
2270 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2271 BTRFS_I(inode), name, name_len);
2273 ret = btrfs_run_delayed_items(trans);
2279 /* there might still be more names under this key
2280 * check and repeat if required
2282 ret = btrfs_search_slot(NULL, root, dir_key, path,
2288 } else if (IS_ERR(log_di)) {
2290 return PTR_ERR(log_di);
2292 btrfs_release_path(log_path);
2295 ptr = (unsigned long)(di + 1);
2300 btrfs_release_path(path);
2301 btrfs_release_path(log_path);
2305 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2306 struct btrfs_root *root,
2307 struct btrfs_root *log,
2308 struct btrfs_path *path,
2311 struct btrfs_key search_key;
2312 struct btrfs_path *log_path;
2317 log_path = btrfs_alloc_path();
2321 search_key.objectid = ino;
2322 search_key.type = BTRFS_XATTR_ITEM_KEY;
2323 search_key.offset = 0;
2325 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2329 nritems = btrfs_header_nritems(path->nodes[0]);
2330 for (i = path->slots[0]; i < nritems; i++) {
2331 struct btrfs_key key;
2332 struct btrfs_dir_item *di;
2333 struct btrfs_dir_item *log_di;
2337 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2338 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2343 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2344 total_size = btrfs_item_size_nr(path->nodes[0], i);
2346 while (cur < total_size) {
2347 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2348 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2349 u32 this_len = sizeof(*di) + name_len + data_len;
2352 name = kmalloc(name_len, GFP_NOFS);
2357 read_extent_buffer(path->nodes[0], name,
2358 (unsigned long)(di + 1), name_len);
2360 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2362 btrfs_release_path(log_path);
2364 /* Doesn't exist in log tree, so delete it. */
2365 btrfs_release_path(path);
2366 di = btrfs_lookup_xattr(trans, root, path, ino,
2367 name, name_len, -1);
2374 ret = btrfs_delete_one_dir_name(trans, root,
2378 btrfs_release_path(path);
2383 if (IS_ERR(log_di)) {
2384 ret = PTR_ERR(log_di);
2388 di = (struct btrfs_dir_item *)((char *)di + this_len);
2391 ret = btrfs_next_leaf(root, path);
2397 btrfs_free_path(log_path);
2398 btrfs_release_path(path);
2404 * deletion replay happens before we copy any new directory items
2405 * out of the log or out of backreferences from inodes. It
2406 * scans the log to find ranges of keys that log is authoritative for,
2407 * and then scans the directory to find items in those ranges that are
2408 * not present in the log.
2410 * Anything we don't find in the log is unlinked and removed from the
2413 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2414 struct btrfs_root *root,
2415 struct btrfs_root *log,
2416 struct btrfs_path *path,
2417 u64 dirid, int del_all)
2421 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2423 struct btrfs_key dir_key;
2424 struct btrfs_key found_key;
2425 struct btrfs_path *log_path;
2428 dir_key.objectid = dirid;
2429 dir_key.type = BTRFS_DIR_ITEM_KEY;
2430 log_path = btrfs_alloc_path();
2434 dir = read_one_inode(root, dirid);
2435 /* it isn't an error if the inode isn't there, that can happen
2436 * because we replay the deletes before we copy in the inode item
2440 btrfs_free_path(log_path);
2448 range_end = (u64)-1;
2450 ret = find_dir_range(log, path, dirid, key_type,
2451 &range_start, &range_end);
2456 dir_key.offset = range_start;
2459 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2464 nritems = btrfs_header_nritems(path->nodes[0]);
2465 if (path->slots[0] >= nritems) {
2466 ret = btrfs_next_leaf(root, path);
2472 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2474 if (found_key.objectid != dirid ||
2475 found_key.type != dir_key.type)
2478 if (found_key.offset > range_end)
2481 ret = check_item_in_log(trans, root, log, path,
2486 if (found_key.offset == (u64)-1)
2488 dir_key.offset = found_key.offset + 1;
2490 btrfs_release_path(path);
2491 if (range_end == (u64)-1)
2493 range_start = range_end + 1;
2498 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2499 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2500 dir_key.type = BTRFS_DIR_INDEX_KEY;
2501 btrfs_release_path(path);
2505 btrfs_release_path(path);
2506 btrfs_free_path(log_path);
2512 * the process_func used to replay items from the log tree. This
2513 * gets called in two different stages. The first stage just looks
2514 * for inodes and makes sure they are all copied into the subvolume.
2516 * The second stage copies all the other item types from the log into
2517 * the subvolume. The two stage approach is slower, but gets rid of
2518 * lots of complexity around inodes referencing other inodes that exist
2519 * only in the log (references come from either directory items or inode
2522 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2523 struct walk_control *wc, u64 gen, int level)
2526 struct btrfs_path *path;
2527 struct btrfs_root *root = wc->replay_dest;
2528 struct btrfs_key key;
2532 ret = btrfs_read_buffer(eb, gen, level, NULL);
2536 level = btrfs_header_level(eb);
2541 path = btrfs_alloc_path();
2545 nritems = btrfs_header_nritems(eb);
2546 for (i = 0; i < nritems; i++) {
2547 btrfs_item_key_to_cpu(eb, &key, i);
2549 /* inode keys are done during the first stage */
2550 if (key.type == BTRFS_INODE_ITEM_KEY &&
2551 wc->stage == LOG_WALK_REPLAY_INODES) {
2552 struct btrfs_inode_item *inode_item;
2555 inode_item = btrfs_item_ptr(eb, i,
2556 struct btrfs_inode_item);
2558 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2559 * and never got linked before the fsync, skip it, as
2560 * replaying it is pointless since it would be deleted
2561 * later. We skip logging tmpfiles, but it's always
2562 * possible we are replaying a log created with a kernel
2563 * that used to log tmpfiles.
2565 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2566 wc->ignore_cur_inode = true;
2569 wc->ignore_cur_inode = false;
2571 ret = replay_xattr_deletes(wc->trans, root, log,
2572 path, key.objectid);
2575 mode = btrfs_inode_mode(eb, inode_item);
2576 if (S_ISDIR(mode)) {
2577 ret = replay_dir_deletes(wc->trans,
2578 root, log, path, key.objectid, 0);
2582 ret = overwrite_item(wc->trans, root, path,
2588 * Before replaying extents, truncate the inode to its
2589 * size. We need to do it now and not after log replay
2590 * because before an fsync we can have prealloc extents
2591 * added beyond the inode's i_size. If we did it after,
2592 * through orphan cleanup for example, we would drop
2593 * those prealloc extents just after replaying them.
2595 if (S_ISREG(mode)) {
2596 struct inode *inode;
2599 inode = read_one_inode(root, key.objectid);
2604 from = ALIGN(i_size_read(inode),
2605 root->fs_info->sectorsize);
2606 ret = btrfs_drop_extents(wc->trans, root, inode,
2609 /* Update the inode's nbytes. */
2610 ret = btrfs_update_inode(wc->trans,
2618 ret = link_to_fixup_dir(wc->trans, root,
2619 path, key.objectid);
2624 if (wc->ignore_cur_inode)
2627 if (key.type == BTRFS_DIR_INDEX_KEY &&
2628 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2629 ret = replay_one_dir_item(wc->trans, root, path,
2635 if (wc->stage < LOG_WALK_REPLAY_ALL)
2638 /* these keys are simply copied */
2639 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2640 ret = overwrite_item(wc->trans, root, path,
2644 } else if (key.type == BTRFS_INODE_REF_KEY ||
2645 key.type == BTRFS_INODE_EXTREF_KEY) {
2646 ret = add_inode_ref(wc->trans, root, log, path,
2648 if (ret && ret != -ENOENT)
2651 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2652 ret = replay_one_extent(wc->trans, root, path,
2656 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2657 ret = replay_one_dir_item(wc->trans, root, path,
2663 btrfs_free_path(path);
2667 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2668 struct btrfs_root *root,
2669 struct btrfs_path *path, int *level,
2670 struct walk_control *wc)
2672 struct btrfs_fs_info *fs_info = root->fs_info;
2676 struct extent_buffer *next;
2677 struct extent_buffer *cur;
2678 struct extent_buffer *parent;
2682 while (*level > 0) {
2683 struct btrfs_key first_key;
2685 cur = path->nodes[*level];
2687 WARN_ON(btrfs_header_level(cur) != *level);
2689 if (path->slots[*level] >=
2690 btrfs_header_nritems(cur))
2693 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2694 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2695 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2696 blocksize = fs_info->nodesize;
2698 parent = path->nodes[*level];
2699 root_owner = btrfs_header_owner(parent);
2701 next = btrfs_find_create_tree_block(fs_info, bytenr);
2703 return PTR_ERR(next);
2706 ret = wc->process_func(root, next, wc, ptr_gen,
2709 free_extent_buffer(next);
2713 path->slots[*level]++;
2715 ret = btrfs_read_buffer(next, ptr_gen,
2716 *level - 1, &first_key);
2718 free_extent_buffer(next);
2723 btrfs_tree_lock(next);
2724 btrfs_set_lock_blocking_write(next);
2725 btrfs_clean_tree_block(next);
2726 btrfs_wait_tree_block_writeback(next);
2727 btrfs_tree_unlock(next);
2729 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2730 clear_extent_buffer_dirty(next);
2733 WARN_ON(root_owner !=
2734 BTRFS_TREE_LOG_OBJECTID);
2735 ret = btrfs_pin_reserved_extent(fs_info,
2738 free_extent_buffer(next);
2742 free_extent_buffer(next);
2745 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2747 free_extent_buffer(next);
2751 if (path->nodes[*level-1])
2752 free_extent_buffer(path->nodes[*level-1]);
2753 path->nodes[*level-1] = next;
2754 *level = btrfs_header_level(next);
2755 path->slots[*level] = 0;
2758 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2764 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 struct btrfs_path *path, int *level,
2767 struct walk_control *wc)
2769 struct btrfs_fs_info *fs_info = root->fs_info;
2775 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2776 slot = path->slots[i];
2777 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2780 WARN_ON(*level == 0);
2783 struct extent_buffer *parent;
2784 if (path->nodes[*level] == root->node)
2785 parent = path->nodes[*level];
2787 parent = path->nodes[*level + 1];
2789 root_owner = btrfs_header_owner(parent);
2790 ret = wc->process_func(root, path->nodes[*level], wc,
2791 btrfs_header_generation(path->nodes[*level]),
2797 struct extent_buffer *next;
2799 next = path->nodes[*level];
2802 btrfs_tree_lock(next);
2803 btrfs_set_lock_blocking_write(next);
2804 btrfs_clean_tree_block(next);
2805 btrfs_wait_tree_block_writeback(next);
2806 btrfs_tree_unlock(next);
2808 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2809 clear_extent_buffer_dirty(next);
2812 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2813 ret = btrfs_pin_reserved_extent(fs_info,
2814 path->nodes[*level]->start,
2815 path->nodes[*level]->len);
2819 free_extent_buffer(path->nodes[*level]);
2820 path->nodes[*level] = NULL;
2828 * drop the reference count on the tree rooted at 'snap'. This traverses
2829 * the tree freeing any blocks that have a ref count of zero after being
2832 static int walk_log_tree(struct btrfs_trans_handle *trans,
2833 struct btrfs_root *log, struct walk_control *wc)
2835 struct btrfs_fs_info *fs_info = log->fs_info;
2839 struct btrfs_path *path;
2842 path = btrfs_alloc_path();
2846 level = btrfs_header_level(log->node);
2848 path->nodes[level] = log->node;
2849 atomic_inc(&log->node->refs);
2850 path->slots[level] = 0;
2853 wret = walk_down_log_tree(trans, log, path, &level, wc);
2861 wret = walk_up_log_tree(trans, log, path, &level, wc);
2870 /* was the root node processed? if not, catch it here */
2871 if (path->nodes[orig_level]) {
2872 ret = wc->process_func(log, path->nodes[orig_level], wc,
2873 btrfs_header_generation(path->nodes[orig_level]),
2878 struct extent_buffer *next;
2880 next = path->nodes[orig_level];
2883 btrfs_tree_lock(next);
2884 btrfs_set_lock_blocking_write(next);
2885 btrfs_clean_tree_block(next);
2886 btrfs_wait_tree_block_writeback(next);
2887 btrfs_tree_unlock(next);
2889 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2890 clear_extent_buffer_dirty(next);
2893 ret = btrfs_pin_reserved_extent(fs_info, next->start,
2901 btrfs_free_path(path);
2906 * helper function to update the item for a given subvolumes log root
2907 * in the tree of log roots
2909 static int update_log_root(struct btrfs_trans_handle *trans,
2910 struct btrfs_root *log,
2911 struct btrfs_root_item *root_item)
2913 struct btrfs_fs_info *fs_info = log->fs_info;
2916 if (log->log_transid == 1) {
2917 /* insert root item on the first sync */
2918 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2919 &log->root_key, root_item);
2921 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2922 &log->root_key, root_item);
2927 static void wait_log_commit(struct btrfs_root *root, int transid)
2930 int index = transid % 2;
2933 * we only allow two pending log transactions at a time,
2934 * so we know that if ours is more than 2 older than the
2935 * current transaction, we're done
2938 prepare_to_wait(&root->log_commit_wait[index],
2939 &wait, TASK_UNINTERRUPTIBLE);
2941 if (!(root->log_transid_committed < transid &&
2942 atomic_read(&root->log_commit[index])))
2945 mutex_unlock(&root->log_mutex);
2947 mutex_lock(&root->log_mutex);
2949 finish_wait(&root->log_commit_wait[index], &wait);
2952 static void wait_for_writer(struct btrfs_root *root)
2957 prepare_to_wait(&root->log_writer_wait, &wait,
2958 TASK_UNINTERRUPTIBLE);
2959 if (!atomic_read(&root->log_writers))
2962 mutex_unlock(&root->log_mutex);
2964 mutex_lock(&root->log_mutex);
2966 finish_wait(&root->log_writer_wait, &wait);
2969 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2970 struct btrfs_log_ctx *ctx)
2975 mutex_lock(&root->log_mutex);
2976 list_del_init(&ctx->list);
2977 mutex_unlock(&root->log_mutex);
2981 * Invoked in log mutex context, or be sure there is no other task which
2982 * can access the list.
2984 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2985 int index, int error)
2987 struct btrfs_log_ctx *ctx;
2988 struct btrfs_log_ctx *safe;
2990 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2991 list_del_init(&ctx->list);
2992 ctx->log_ret = error;
2995 INIT_LIST_HEAD(&root->log_ctxs[index]);
2999 * btrfs_sync_log does sends a given tree log down to the disk and
3000 * updates the super blocks to record it. When this call is done,
3001 * you know that any inodes previously logged are safely on disk only
3004 * Any other return value means you need to call btrfs_commit_transaction.
3005 * Some of the edge cases for fsyncing directories that have had unlinks
3006 * or renames done in the past mean that sometimes the only safe
3007 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3008 * that has happened.
3010 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3011 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3017 struct btrfs_fs_info *fs_info = root->fs_info;
3018 struct btrfs_root *log = root->log_root;
3019 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3020 struct btrfs_root_item new_root_item;
3021 int log_transid = 0;
3022 struct btrfs_log_ctx root_log_ctx;
3023 struct blk_plug plug;
3025 mutex_lock(&root->log_mutex);
3026 log_transid = ctx->log_transid;
3027 if (root->log_transid_committed >= log_transid) {
3028 mutex_unlock(&root->log_mutex);
3029 return ctx->log_ret;
3032 index1 = log_transid % 2;
3033 if (atomic_read(&root->log_commit[index1])) {
3034 wait_log_commit(root, log_transid);
3035 mutex_unlock(&root->log_mutex);
3036 return ctx->log_ret;
3038 ASSERT(log_transid == root->log_transid);
3039 atomic_set(&root->log_commit[index1], 1);
3041 /* wait for previous tree log sync to complete */
3042 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3043 wait_log_commit(root, log_transid - 1);
3046 int batch = atomic_read(&root->log_batch);
3047 /* when we're on an ssd, just kick the log commit out */
3048 if (!btrfs_test_opt(fs_info, SSD) &&
3049 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3050 mutex_unlock(&root->log_mutex);
3051 schedule_timeout_uninterruptible(1);
3052 mutex_lock(&root->log_mutex);
3054 wait_for_writer(root);
3055 if (batch == atomic_read(&root->log_batch))
3059 /* bail out if we need to do a full commit */
3060 if (btrfs_need_log_full_commit(trans)) {
3062 mutex_unlock(&root->log_mutex);
3066 if (log_transid % 2 == 0)
3067 mark = EXTENT_DIRTY;
3071 /* we start IO on all the marked extents here, but we don't actually
3072 * wait for them until later.
3074 blk_start_plug(&plug);
3075 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3077 blk_finish_plug(&plug);
3078 btrfs_abort_transaction(trans, ret);
3079 btrfs_set_log_full_commit(trans);
3080 mutex_unlock(&root->log_mutex);
3085 * We _must_ update under the root->log_mutex in order to make sure we
3086 * have a consistent view of the log root we are trying to commit at
3089 * We _must_ copy this into a local copy, because we are not holding the
3090 * log_root_tree->log_mutex yet. This is important because when we
3091 * commit the log_root_tree we must have a consistent view of the
3092 * log_root_tree when we update the super block to point at the
3093 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3094 * with the commit and possibly point at the new block which we may not
3097 btrfs_set_root_node(&log->root_item, log->node);
3098 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3100 root->log_transid++;
3101 log->log_transid = root->log_transid;
3102 root->log_start_pid = 0;
3104 * IO has been started, blocks of the log tree have WRITTEN flag set
3105 * in their headers. new modifications of the log will be written to
3106 * new positions. so it's safe to allow log writers to go in.
3108 mutex_unlock(&root->log_mutex);
3110 btrfs_init_log_ctx(&root_log_ctx, NULL);
3112 mutex_lock(&log_root_tree->log_mutex);
3113 atomic_inc(&log_root_tree->log_batch);
3114 atomic_inc(&log_root_tree->log_writers);
3116 index2 = log_root_tree->log_transid % 2;
3117 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3118 root_log_ctx.log_transid = log_root_tree->log_transid;
3120 mutex_unlock(&log_root_tree->log_mutex);
3122 mutex_lock(&log_root_tree->log_mutex);
3125 * Now we are safe to update the log_root_tree because we're under the
3126 * log_mutex, and we're a current writer so we're holding the commit
3127 * open until we drop the log_mutex.
3129 ret = update_log_root(trans, log, &new_root_item);
3131 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3132 /* atomic_dec_and_test implies a barrier */
3133 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3137 if (!list_empty(&root_log_ctx.list))
3138 list_del_init(&root_log_ctx.list);
3140 blk_finish_plug(&plug);
3141 btrfs_set_log_full_commit(trans);
3143 if (ret != -ENOSPC) {
3144 btrfs_abort_transaction(trans, ret);
3145 mutex_unlock(&log_root_tree->log_mutex);
3148 btrfs_wait_tree_log_extents(log, mark);
3149 mutex_unlock(&log_root_tree->log_mutex);
3154 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3155 blk_finish_plug(&plug);
3156 list_del_init(&root_log_ctx.list);
3157 mutex_unlock(&log_root_tree->log_mutex);
3158 ret = root_log_ctx.log_ret;
3162 index2 = root_log_ctx.log_transid % 2;
3163 if (atomic_read(&log_root_tree->log_commit[index2])) {
3164 blk_finish_plug(&plug);
3165 ret = btrfs_wait_tree_log_extents(log, mark);
3166 wait_log_commit(log_root_tree,
3167 root_log_ctx.log_transid);
3168 mutex_unlock(&log_root_tree->log_mutex);
3170 ret = root_log_ctx.log_ret;
3173 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3174 atomic_set(&log_root_tree->log_commit[index2], 1);
3176 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3177 wait_log_commit(log_root_tree,
3178 root_log_ctx.log_transid - 1);
3181 wait_for_writer(log_root_tree);
3184 * now that we've moved on to the tree of log tree roots,
3185 * check the full commit flag again
3187 if (btrfs_need_log_full_commit(trans)) {
3188 blk_finish_plug(&plug);
3189 btrfs_wait_tree_log_extents(log, mark);
3190 mutex_unlock(&log_root_tree->log_mutex);
3192 goto out_wake_log_root;
3195 ret = btrfs_write_marked_extents(fs_info,
3196 &log_root_tree->dirty_log_pages,
3197 EXTENT_DIRTY | EXTENT_NEW);
3198 blk_finish_plug(&plug);
3200 btrfs_set_log_full_commit(trans);
3201 btrfs_abort_transaction(trans, ret);
3202 mutex_unlock(&log_root_tree->log_mutex);
3203 goto out_wake_log_root;
3205 ret = btrfs_wait_tree_log_extents(log, mark);
3207 ret = btrfs_wait_tree_log_extents(log_root_tree,
3208 EXTENT_NEW | EXTENT_DIRTY);
3210 btrfs_set_log_full_commit(trans);
3211 mutex_unlock(&log_root_tree->log_mutex);
3212 goto out_wake_log_root;
3215 btrfs_set_super_log_root(fs_info->super_for_commit,
3216 log_root_tree->node->start);
3217 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3218 btrfs_header_level(log_root_tree->node));
3220 log_root_tree->log_transid++;
3221 mutex_unlock(&log_root_tree->log_mutex);
3224 * Nobody else is going to jump in and write the ctree
3225 * super here because the log_commit atomic below is protecting
3226 * us. We must be called with a transaction handle pinning
3227 * the running transaction open, so a full commit can't hop
3228 * in and cause problems either.
3230 ret = write_all_supers(fs_info, 1);
3232 btrfs_set_log_full_commit(trans);
3233 btrfs_abort_transaction(trans, ret);
3234 goto out_wake_log_root;
3237 mutex_lock(&root->log_mutex);
3238 if (root->last_log_commit < log_transid)
3239 root->last_log_commit = log_transid;
3240 mutex_unlock(&root->log_mutex);
3243 mutex_lock(&log_root_tree->log_mutex);
3244 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3246 log_root_tree->log_transid_committed++;
3247 atomic_set(&log_root_tree->log_commit[index2], 0);
3248 mutex_unlock(&log_root_tree->log_mutex);
3251 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3252 * all the updates above are seen by the woken threads. It might not be
3253 * necessary, but proving that seems to be hard.
3255 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3257 mutex_lock(&root->log_mutex);
3258 btrfs_remove_all_log_ctxs(root, index1, ret);
3259 root->log_transid_committed++;
3260 atomic_set(&root->log_commit[index1], 0);
3261 mutex_unlock(&root->log_mutex);
3264 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3265 * all the updates above are seen by the woken threads. It might not be
3266 * necessary, but proving that seems to be hard.
3268 cond_wake_up(&root->log_commit_wait[index1]);
3272 static void free_log_tree(struct btrfs_trans_handle *trans,
3273 struct btrfs_root *log)
3276 struct walk_control wc = {
3278 .process_func = process_one_buffer
3281 ret = walk_log_tree(trans, log, &wc);
3284 btrfs_abort_transaction(trans, ret);
3286 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3289 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3290 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3291 free_extent_buffer(log->node);
3292 btrfs_put_fs_root(log);
3296 * free all the extents used by the tree log. This should be called
3297 * at commit time of the full transaction
3299 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3301 if (root->log_root) {
3302 free_log_tree(trans, root->log_root);
3303 root->log_root = NULL;
3308 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3309 struct btrfs_fs_info *fs_info)
3311 if (fs_info->log_root_tree) {
3312 free_log_tree(trans, fs_info->log_root_tree);
3313 fs_info->log_root_tree = NULL;
3319 * Check if an inode was logged in the current transaction. We can't always rely
3320 * on an inode's logged_trans value, because it's an in-memory only field and
3321 * therefore not persisted. This means that its value is lost if the inode gets
3322 * evicted and loaded again from disk (in which case it has a value of 0, and
3323 * certainly it is smaller then any possible transaction ID), when that happens
3324 * the full_sync flag is set in the inode's runtime flags, so on that case we
3325 * assume eviction happened and ignore the logged_trans value, assuming the
3326 * worst case, that the inode was logged before in the current transaction.
3328 static bool inode_logged(struct btrfs_trans_handle *trans,
3329 struct btrfs_inode *inode)
3331 if (inode->logged_trans == trans->transid)
3334 if (inode->last_trans == trans->transid &&
3335 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3336 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3343 * If both a file and directory are logged, and unlinks or renames are
3344 * mixed in, we have a few interesting corners:
3346 * create file X in dir Y
3347 * link file X to X.link in dir Y
3349 * unlink file X but leave X.link
3352 * After a crash we would expect only X.link to exist. But file X
3353 * didn't get fsync'd again so the log has back refs for X and X.link.
3355 * We solve this by removing directory entries and inode backrefs from the
3356 * log when a file that was logged in the current transaction is
3357 * unlinked. Any later fsync will include the updated log entries, and
3358 * we'll be able to reconstruct the proper directory items from backrefs.
3360 * This optimizations allows us to avoid relogging the entire inode
3361 * or the entire directory.
3363 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3364 struct btrfs_root *root,
3365 const char *name, int name_len,
3366 struct btrfs_inode *dir, u64 index)
3368 struct btrfs_root *log;
3369 struct btrfs_dir_item *di;
3370 struct btrfs_path *path;
3374 u64 dir_ino = btrfs_ino(dir);
3376 if (!inode_logged(trans, dir))
3379 ret = join_running_log_trans(root);
3383 mutex_lock(&dir->log_mutex);
3385 log = root->log_root;
3386 path = btrfs_alloc_path();
3392 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3393 name, name_len, -1);
3399 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3400 bytes_del += name_len;
3406 btrfs_release_path(path);
3407 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3408 index, name, name_len, -1);
3414 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3415 bytes_del += name_len;
3422 /* update the directory size in the log to reflect the names
3426 struct btrfs_key key;
3428 key.objectid = dir_ino;
3430 key.type = BTRFS_INODE_ITEM_KEY;
3431 btrfs_release_path(path);
3433 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3439 struct btrfs_inode_item *item;
3442 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3443 struct btrfs_inode_item);
3444 i_size = btrfs_inode_size(path->nodes[0], item);
3445 if (i_size > bytes_del)
3446 i_size -= bytes_del;
3449 btrfs_set_inode_size(path->nodes[0], item, i_size);
3450 btrfs_mark_buffer_dirty(path->nodes[0]);
3453 btrfs_release_path(path);
3456 btrfs_free_path(path);
3458 mutex_unlock(&dir->log_mutex);
3459 if (ret == -ENOSPC) {
3460 btrfs_set_log_full_commit(trans);
3463 btrfs_abort_transaction(trans, ret);
3465 btrfs_end_log_trans(root);
3470 /* see comments for btrfs_del_dir_entries_in_log */
3471 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3472 struct btrfs_root *root,
3473 const char *name, int name_len,
3474 struct btrfs_inode *inode, u64 dirid)
3476 struct btrfs_root *log;
3480 if (!inode_logged(trans, inode))
3483 ret = join_running_log_trans(root);
3486 log = root->log_root;
3487 mutex_lock(&inode->log_mutex);
3489 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3491 mutex_unlock(&inode->log_mutex);
3492 if (ret == -ENOSPC) {
3493 btrfs_set_log_full_commit(trans);
3495 } else if (ret < 0 && ret != -ENOENT)
3496 btrfs_abort_transaction(trans, ret);
3497 btrfs_end_log_trans(root);
3503 * creates a range item in the log for 'dirid'. first_offset and
3504 * last_offset tell us which parts of the key space the log should
3505 * be considered authoritative for.
3507 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3508 struct btrfs_root *log,
3509 struct btrfs_path *path,
3510 int key_type, u64 dirid,
3511 u64 first_offset, u64 last_offset)
3514 struct btrfs_key key;
3515 struct btrfs_dir_log_item *item;
3517 key.objectid = dirid;
3518 key.offset = first_offset;
3519 if (key_type == BTRFS_DIR_ITEM_KEY)
3520 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3522 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3523 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3527 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3528 struct btrfs_dir_log_item);
3529 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3530 btrfs_mark_buffer_dirty(path->nodes[0]);
3531 btrfs_release_path(path);
3536 * log all the items included in the current transaction for a given
3537 * directory. This also creates the range items in the log tree required
3538 * to replay anything deleted before the fsync
3540 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root, struct btrfs_inode *inode,
3542 struct btrfs_path *path,
3543 struct btrfs_path *dst_path, int key_type,
3544 struct btrfs_log_ctx *ctx,
3545 u64 min_offset, u64 *last_offset_ret)
3547 struct btrfs_key min_key;
3548 struct btrfs_root *log = root->log_root;
3549 struct extent_buffer *src;
3554 u64 first_offset = min_offset;
3555 u64 last_offset = (u64)-1;
3556 u64 ino = btrfs_ino(inode);
3558 log = root->log_root;
3560 min_key.objectid = ino;
3561 min_key.type = key_type;
3562 min_key.offset = min_offset;
3564 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3567 * we didn't find anything from this transaction, see if there
3568 * is anything at all
3570 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3571 min_key.objectid = ino;
3572 min_key.type = key_type;
3573 min_key.offset = (u64)-1;
3574 btrfs_release_path(path);
3575 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3577 btrfs_release_path(path);
3580 ret = btrfs_previous_item(root, path, ino, key_type);
3582 /* if ret == 0 there are items for this type,
3583 * create a range to tell us the last key of this type.
3584 * otherwise, there are no items in this directory after
3585 * *min_offset, and we create a range to indicate that.
3588 struct btrfs_key tmp;
3589 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3591 if (key_type == tmp.type)
3592 first_offset = max(min_offset, tmp.offset) + 1;
3597 /* go backward to find any previous key */
3598 ret = btrfs_previous_item(root, path, ino, key_type);
3600 struct btrfs_key tmp;
3601 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3602 if (key_type == tmp.type) {
3603 first_offset = tmp.offset;
3604 ret = overwrite_item(trans, log, dst_path,
3605 path->nodes[0], path->slots[0],
3613 btrfs_release_path(path);
3616 * Find the first key from this transaction again. See the note for
3617 * log_new_dir_dentries, if we're logging a directory recursively we
3618 * won't be holding its i_mutex, which means we can modify the directory
3619 * while we're logging it. If we remove an entry between our first
3620 * search and this search we'll not find the key again and can just
3623 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3628 * we have a block from this transaction, log every item in it
3629 * from our directory
3632 struct btrfs_key tmp;
3633 src = path->nodes[0];
3634 nritems = btrfs_header_nritems(src);
3635 for (i = path->slots[0]; i < nritems; i++) {
3636 struct btrfs_dir_item *di;
3638 btrfs_item_key_to_cpu(src, &min_key, i);
3640 if (min_key.objectid != ino || min_key.type != key_type)
3642 ret = overwrite_item(trans, log, dst_path, src, i,
3650 * We must make sure that when we log a directory entry,
3651 * the corresponding inode, after log replay, has a
3652 * matching link count. For example:
3658 * xfs_io -c "fsync" mydir
3660 * <mount fs and log replay>
3662 * Would result in a fsync log that when replayed, our
3663 * file inode would have a link count of 1, but we get
3664 * two directory entries pointing to the same inode.
3665 * After removing one of the names, it would not be
3666 * possible to remove the other name, which resulted
3667 * always in stale file handle errors, and would not
3668 * be possible to rmdir the parent directory, since
3669 * its i_size could never decrement to the value
3670 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3672 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3673 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3675 (btrfs_dir_transid(src, di) == trans->transid ||
3676 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3677 tmp.type != BTRFS_ROOT_ITEM_KEY)
3678 ctx->log_new_dentries = true;
3680 path->slots[0] = nritems;
3683 * look ahead to the next item and see if it is also
3684 * from this directory and from this transaction
3686 ret = btrfs_next_leaf(root, path);
3689 last_offset = (u64)-1;
3694 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3695 if (tmp.objectid != ino || tmp.type != key_type) {
3696 last_offset = (u64)-1;
3699 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3700 ret = overwrite_item(trans, log, dst_path,
3701 path->nodes[0], path->slots[0],
3706 last_offset = tmp.offset;
3711 btrfs_release_path(path);
3712 btrfs_release_path(dst_path);
3715 *last_offset_ret = last_offset;
3717 * insert the log range keys to indicate where the log
3720 ret = insert_dir_log_key(trans, log, path, key_type,
3721 ino, first_offset, last_offset);
3729 * logging directories is very similar to logging inodes, We find all the items
3730 * from the current transaction and write them to the log.
3732 * The recovery code scans the directory in the subvolume, and if it finds a
3733 * key in the range logged that is not present in the log tree, then it means
3734 * that dir entry was unlinked during the transaction.
3736 * In order for that scan to work, we must include one key smaller than
3737 * the smallest logged by this transaction and one key larger than the largest
3738 * key logged by this transaction.
3740 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3741 struct btrfs_root *root, struct btrfs_inode *inode,
3742 struct btrfs_path *path,
3743 struct btrfs_path *dst_path,
3744 struct btrfs_log_ctx *ctx)
3749 int key_type = BTRFS_DIR_ITEM_KEY;
3755 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3756 ctx, min_key, &max_key);
3759 if (max_key == (u64)-1)
3761 min_key = max_key + 1;
3764 if (key_type == BTRFS_DIR_ITEM_KEY) {
3765 key_type = BTRFS_DIR_INDEX_KEY;
3772 * a helper function to drop items from the log before we relog an
3773 * inode. max_key_type indicates the highest item type to remove.
3774 * This cannot be run for file data extents because it does not
3775 * free the extents they point to.
3777 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3778 struct btrfs_root *log,
3779 struct btrfs_path *path,
3780 u64 objectid, int max_key_type)
3783 struct btrfs_key key;
3784 struct btrfs_key found_key;
3787 key.objectid = objectid;
3788 key.type = max_key_type;
3789 key.offset = (u64)-1;
3792 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3793 BUG_ON(ret == 0); /* Logic error */
3797 if (path->slots[0] == 0)
3801 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3804 if (found_key.objectid != objectid)
3807 found_key.offset = 0;
3809 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3814 ret = btrfs_del_items(trans, log, path, start_slot,
3815 path->slots[0] - start_slot + 1);
3817 * If start slot isn't 0 then we don't need to re-search, we've
3818 * found the last guy with the objectid in this tree.
3820 if (ret || start_slot != 0)
3822 btrfs_release_path(path);
3824 btrfs_release_path(path);
3830 static void fill_inode_item(struct btrfs_trans_handle *trans,
3831 struct extent_buffer *leaf,
3832 struct btrfs_inode_item *item,
3833 struct inode *inode, int log_inode_only,
3836 struct btrfs_map_token token;
3838 btrfs_init_map_token(&token, leaf);
3840 if (log_inode_only) {
3841 /* set the generation to zero so the recover code
3842 * can tell the difference between an logging
3843 * just to say 'this inode exists' and a logging
3844 * to say 'update this inode with these values'
3846 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3847 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3849 btrfs_set_token_inode_generation(leaf, item,
3850 BTRFS_I(inode)->generation,
3852 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3855 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3856 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3857 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3858 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3860 btrfs_set_token_timespec_sec(leaf, &item->atime,
3861 inode->i_atime.tv_sec, &token);
3862 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3863 inode->i_atime.tv_nsec, &token);
3865 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3866 inode->i_mtime.tv_sec, &token);
3867 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3868 inode->i_mtime.tv_nsec, &token);
3870 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3871 inode->i_ctime.tv_sec, &token);
3872 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3873 inode->i_ctime.tv_nsec, &token);
3875 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3878 btrfs_set_token_inode_sequence(leaf, item,
3879 inode_peek_iversion(inode), &token);
3880 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3881 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3882 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3883 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3886 static int log_inode_item(struct btrfs_trans_handle *trans,
3887 struct btrfs_root *log, struct btrfs_path *path,
3888 struct btrfs_inode *inode)
3890 struct btrfs_inode_item *inode_item;
3893 ret = btrfs_insert_empty_item(trans, log, path,
3894 &inode->location, sizeof(*inode_item));
3895 if (ret && ret != -EEXIST)
3897 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3898 struct btrfs_inode_item);
3899 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3901 btrfs_release_path(path);
3905 static int log_csums(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *log_root,
3907 struct btrfs_ordered_sum *sums)
3912 * Due to extent cloning, we might have logged a csum item that covers a
3913 * subrange of a cloned extent, and later we can end up logging a csum
3914 * item for a larger subrange of the same extent or the entire range.
3915 * This would leave csum items in the log tree that cover the same range
3916 * and break the searches for checksums in the log tree, resulting in
3917 * some checksums missing in the fs/subvolume tree. So just delete (or
3918 * trim and adjust) any existing csum items in the log for this range.
3920 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3924 return btrfs_csum_file_blocks(trans, log_root, sums);
3927 static noinline int copy_items(struct btrfs_trans_handle *trans,
3928 struct btrfs_inode *inode,
3929 struct btrfs_path *dst_path,
3930 struct btrfs_path *src_path,
3931 int start_slot, int nr, int inode_only,
3934 struct btrfs_fs_info *fs_info = trans->fs_info;
3935 unsigned long src_offset;
3936 unsigned long dst_offset;
3937 struct btrfs_root *log = inode->root->log_root;
3938 struct btrfs_file_extent_item *extent;
3939 struct btrfs_inode_item *inode_item;
3940 struct extent_buffer *src = src_path->nodes[0];
3942 struct btrfs_key *ins_keys;
3946 struct list_head ordered_sums;
3947 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3949 INIT_LIST_HEAD(&ordered_sums);
3951 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3952 nr * sizeof(u32), GFP_NOFS);
3956 ins_sizes = (u32 *)ins_data;
3957 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3959 for (i = 0; i < nr; i++) {
3960 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3961 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3963 ret = btrfs_insert_empty_items(trans, log, dst_path,
3964 ins_keys, ins_sizes, nr);
3970 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3971 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3972 dst_path->slots[0]);
3974 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3976 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3977 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3979 struct btrfs_inode_item);
3980 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3982 inode_only == LOG_INODE_EXISTS,
3985 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3986 src_offset, ins_sizes[i]);
3989 /* take a reference on file data extents so that truncates
3990 * or deletes of this inode don't have to relog the inode
3993 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3996 extent = btrfs_item_ptr(src, start_slot + i,
3997 struct btrfs_file_extent_item);
3999 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4002 found_type = btrfs_file_extent_type(src, extent);
4003 if (found_type == BTRFS_FILE_EXTENT_REG) {
4005 ds = btrfs_file_extent_disk_bytenr(src,
4007 /* ds == 0 is a hole */
4011 dl = btrfs_file_extent_disk_num_bytes(src,
4013 cs = btrfs_file_extent_offset(src, extent);
4014 cl = btrfs_file_extent_num_bytes(src,
4016 if (btrfs_file_extent_compression(src,
4022 ret = btrfs_lookup_csums_range(
4024 ds + cs, ds + cs + cl - 1,
4027 btrfs_release_path(dst_path);
4035 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4036 btrfs_release_path(dst_path);
4040 * we have to do this after the loop above to avoid changing the
4041 * log tree while trying to change the log tree.
4044 while (!list_empty(&ordered_sums)) {
4045 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4046 struct btrfs_ordered_sum,
4049 ret = log_csums(trans, log, sums);
4050 list_del(&sums->list);
4057 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4059 struct extent_map *em1, *em2;
4061 em1 = list_entry(a, struct extent_map, list);
4062 em2 = list_entry(b, struct extent_map, list);
4064 if (em1->start < em2->start)
4066 else if (em1->start > em2->start)
4071 static int log_extent_csums(struct btrfs_trans_handle *trans,
4072 struct btrfs_inode *inode,
4073 struct btrfs_root *log_root,
4074 const struct extent_map *em)
4078 LIST_HEAD(ordered_sums);
4081 if (inode->flags & BTRFS_INODE_NODATASUM ||
4082 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4083 em->block_start == EXTENT_MAP_HOLE)
4086 /* If we're compressed we have to save the entire range of csums. */
4087 if (em->compress_type) {
4089 csum_len = max(em->block_len, em->orig_block_len);
4091 csum_offset = em->mod_start - em->start;
4092 csum_len = em->mod_len;
4095 /* block start is already adjusted for the file extent offset. */
4096 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4097 em->block_start + csum_offset,
4098 em->block_start + csum_offset +
4099 csum_len - 1, &ordered_sums, 0);
4103 while (!list_empty(&ordered_sums)) {
4104 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4105 struct btrfs_ordered_sum,
4108 ret = log_csums(trans, log_root, sums);
4109 list_del(&sums->list);
4116 static int log_one_extent(struct btrfs_trans_handle *trans,
4117 struct btrfs_inode *inode, struct btrfs_root *root,
4118 const struct extent_map *em,
4119 struct btrfs_path *path,
4120 struct btrfs_log_ctx *ctx)
4122 struct btrfs_root *log = root->log_root;
4123 struct btrfs_file_extent_item *fi;
4124 struct extent_buffer *leaf;
4125 struct btrfs_map_token token;
4126 struct btrfs_key key;
4127 u64 extent_offset = em->start - em->orig_start;
4130 int extent_inserted = 0;
4132 ret = log_extent_csums(trans, inode, log, em);
4136 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4137 em->start + em->len, NULL, 0, 1,
4138 sizeof(*fi), &extent_inserted);
4142 if (!extent_inserted) {
4143 key.objectid = btrfs_ino(inode);
4144 key.type = BTRFS_EXTENT_DATA_KEY;
4145 key.offset = em->start;
4147 ret = btrfs_insert_empty_item(trans, log, path, &key,
4152 leaf = path->nodes[0];
4153 btrfs_init_map_token(&token, leaf);
4154 fi = btrfs_item_ptr(leaf, path->slots[0],
4155 struct btrfs_file_extent_item);
4157 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4159 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4160 btrfs_set_token_file_extent_type(leaf, fi,
4161 BTRFS_FILE_EXTENT_PREALLOC,
4164 btrfs_set_token_file_extent_type(leaf, fi,
4165 BTRFS_FILE_EXTENT_REG,
4168 block_len = max(em->block_len, em->orig_block_len);
4169 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4170 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4173 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4175 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4176 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4178 extent_offset, &token);
4179 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4182 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4183 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4187 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4188 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4189 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4190 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4192 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4193 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4194 btrfs_mark_buffer_dirty(leaf);
4196 btrfs_release_path(path);
4202 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4203 * lose them after doing a fast fsync and replaying the log. We scan the
4204 * subvolume's root instead of iterating the inode's extent map tree because
4205 * otherwise we can log incorrect extent items based on extent map conversion.
4206 * That can happen due to the fact that extent maps are merged when they
4207 * are not in the extent map tree's list of modified extents.
4209 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4210 struct btrfs_inode *inode,
4211 struct btrfs_path *path)
4213 struct btrfs_root *root = inode->root;
4214 struct btrfs_key key;
4215 const u64 i_size = i_size_read(&inode->vfs_inode);
4216 const u64 ino = btrfs_ino(inode);
4217 struct btrfs_path *dst_path = NULL;
4218 bool dropped_extents = false;
4223 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4227 key.type = BTRFS_EXTENT_DATA_KEY;
4228 key.offset = i_size;
4229 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4234 struct extent_buffer *leaf = path->nodes[0];
4235 int slot = path->slots[0];
4237 if (slot >= btrfs_header_nritems(leaf)) {
4239 ret = copy_items(trans, inode, dst_path, path,
4240 start_slot, ins_nr, 1, 0);
4245 ret = btrfs_next_leaf(root, path);
4255 btrfs_item_key_to_cpu(leaf, &key, slot);
4256 if (key.objectid > ino)
4258 if (WARN_ON_ONCE(key.objectid < ino) ||
4259 key.type < BTRFS_EXTENT_DATA_KEY ||
4260 key.offset < i_size) {
4264 if (!dropped_extents) {
4266 * Avoid logging extent items logged in past fsync calls
4267 * and leading to duplicate keys in the log tree.
4270 ret = btrfs_truncate_inode_items(trans,
4274 BTRFS_EXTENT_DATA_KEY);
4275 } while (ret == -EAGAIN);
4278 dropped_extents = true;
4285 dst_path = btrfs_alloc_path();
4293 ret = copy_items(trans, inode, dst_path, path,
4294 start_slot, ins_nr, 1, 0);
4299 btrfs_release_path(path);
4300 btrfs_free_path(dst_path);
4304 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4305 struct btrfs_root *root,
4306 struct btrfs_inode *inode,
4307 struct btrfs_path *path,
4308 struct btrfs_log_ctx *ctx,
4312 struct extent_map *em, *n;
4313 struct list_head extents;
4314 struct extent_map_tree *tree = &inode->extent_tree;
4319 INIT_LIST_HEAD(&extents);
4321 write_lock(&tree->lock);
4322 test_gen = root->fs_info->last_trans_committed;
4324 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4326 * Skip extents outside our logging range. It's important to do
4327 * it for correctness because if we don't ignore them, we may
4328 * log them before their ordered extent completes, and therefore
4329 * we could log them without logging their respective checksums
4330 * (the checksum items are added to the csum tree at the very
4331 * end of btrfs_finish_ordered_io()). Also leave such extents
4332 * outside of our range in the list, since we may have another
4333 * ranged fsync in the near future that needs them. If an extent
4334 * outside our range corresponds to a hole, log it to avoid
4335 * leaving gaps between extents (fsck will complain when we are
4336 * not using the NO_HOLES feature).
4338 if ((em->start > end || em->start + em->len <= start) &&
4339 em->block_start != EXTENT_MAP_HOLE)
4342 list_del_init(&em->list);
4344 * Just an arbitrary number, this can be really CPU intensive
4345 * once we start getting a lot of extents, and really once we
4346 * have a bunch of extents we just want to commit since it will
4349 if (++num > 32768) {
4350 list_del_init(&tree->modified_extents);
4355 if (em->generation <= test_gen)
4358 /* We log prealloc extents beyond eof later. */
4359 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4360 em->start >= i_size_read(&inode->vfs_inode))
4363 /* Need a ref to keep it from getting evicted from cache */
4364 refcount_inc(&em->refs);
4365 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4366 list_add_tail(&em->list, &extents);
4370 list_sort(NULL, &extents, extent_cmp);
4372 while (!list_empty(&extents)) {
4373 em = list_entry(extents.next, struct extent_map, list);
4375 list_del_init(&em->list);
4378 * If we had an error we just need to delete everybody from our
4382 clear_em_logging(tree, em);
4383 free_extent_map(em);
4387 write_unlock(&tree->lock);
4389 ret = log_one_extent(trans, inode, root, em, path, ctx);
4390 write_lock(&tree->lock);
4391 clear_em_logging(tree, em);
4392 free_extent_map(em);
4394 WARN_ON(!list_empty(&extents));
4395 write_unlock(&tree->lock);
4397 btrfs_release_path(path);
4399 ret = btrfs_log_prealloc_extents(trans, inode, path);
4404 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4405 struct btrfs_path *path, u64 *size_ret)
4407 struct btrfs_key key;
4410 key.objectid = btrfs_ino(inode);
4411 key.type = BTRFS_INODE_ITEM_KEY;
4414 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4417 } else if (ret > 0) {
4420 struct btrfs_inode_item *item;
4422 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4423 struct btrfs_inode_item);
4424 *size_ret = btrfs_inode_size(path->nodes[0], item);
4426 * If the in-memory inode's i_size is smaller then the inode
4427 * size stored in the btree, return the inode's i_size, so
4428 * that we get a correct inode size after replaying the log
4429 * when before a power failure we had a shrinking truncate
4430 * followed by addition of a new name (rename / new hard link).
4431 * Otherwise return the inode size from the btree, to avoid
4432 * data loss when replaying a log due to previously doing a
4433 * write that expands the inode's size and logging a new name
4434 * immediately after.
4436 if (*size_ret > inode->vfs_inode.i_size)
4437 *size_ret = inode->vfs_inode.i_size;
4440 btrfs_release_path(path);
4445 * At the moment we always log all xattrs. This is to figure out at log replay
4446 * time which xattrs must have their deletion replayed. If a xattr is missing
4447 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4448 * because if a xattr is deleted, the inode is fsynced and a power failure
4449 * happens, causing the log to be replayed the next time the fs is mounted,
4450 * we want the xattr to not exist anymore (same behaviour as other filesystems
4451 * with a journal, ext3/4, xfs, f2fs, etc).
4453 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4454 struct btrfs_root *root,
4455 struct btrfs_inode *inode,
4456 struct btrfs_path *path,
4457 struct btrfs_path *dst_path)
4460 struct btrfs_key key;
4461 const u64 ino = btrfs_ino(inode);
4466 key.type = BTRFS_XATTR_ITEM_KEY;
4469 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4474 int slot = path->slots[0];
4475 struct extent_buffer *leaf = path->nodes[0];
4476 int nritems = btrfs_header_nritems(leaf);
4478 if (slot >= nritems) {
4480 ret = copy_items(trans, inode, dst_path, path,
4481 start_slot, ins_nr, 1, 0);
4486 ret = btrfs_next_leaf(root, path);
4494 btrfs_item_key_to_cpu(leaf, &key, slot);
4495 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4505 ret = copy_items(trans, inode, dst_path, path,
4506 start_slot, ins_nr, 1, 0);
4515 * When using the NO_HOLES feature if we punched a hole that causes the
4516 * deletion of entire leafs or all the extent items of the first leaf (the one
4517 * that contains the inode item and references) we may end up not processing
4518 * any extents, because there are no leafs with a generation matching the
4519 * current transaction that have extent items for our inode. So we need to find
4520 * if any holes exist and then log them. We also need to log holes after any
4521 * truncate operation that changes the inode's size.
4523 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4524 struct btrfs_root *root,
4525 struct btrfs_inode *inode,
4526 struct btrfs_path *path)
4528 struct btrfs_fs_info *fs_info = root->fs_info;
4529 struct btrfs_key key;
4530 const u64 ino = btrfs_ino(inode);
4531 const u64 i_size = i_size_read(&inode->vfs_inode);
4532 u64 prev_extent_end = 0;
4535 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4539 key.type = BTRFS_EXTENT_DATA_KEY;
4542 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4547 struct btrfs_file_extent_item *extent;
4548 struct extent_buffer *leaf = path->nodes[0];
4551 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4552 ret = btrfs_next_leaf(root, path);
4559 leaf = path->nodes[0];
4562 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4563 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4566 /* We have a hole, log it. */
4567 if (prev_extent_end < key.offset) {
4568 const u64 hole_len = key.offset - prev_extent_end;
4571 * Release the path to avoid deadlocks with other code
4572 * paths that search the root while holding locks on
4573 * leafs from the log root.
4575 btrfs_release_path(path);
4576 ret = btrfs_insert_file_extent(trans, root->log_root,
4577 ino, prev_extent_end, 0,
4578 0, hole_len, 0, hole_len,
4584 * Search for the same key again in the root. Since it's
4585 * an extent item and we are holding the inode lock, the
4586 * key must still exist. If it doesn't just emit warning
4587 * and return an error to fall back to a transaction
4590 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4593 if (WARN_ON(ret > 0))
4595 leaf = path->nodes[0];
4598 extent = btrfs_item_ptr(leaf, path->slots[0],
4599 struct btrfs_file_extent_item);
4600 if (btrfs_file_extent_type(leaf, extent) ==
4601 BTRFS_FILE_EXTENT_INLINE) {
4602 len = btrfs_file_extent_ram_bytes(leaf, extent);
4603 prev_extent_end = ALIGN(key.offset + len,
4604 fs_info->sectorsize);
4606 len = btrfs_file_extent_num_bytes(leaf, extent);
4607 prev_extent_end = key.offset + len;
4614 if (prev_extent_end < i_size) {
4617 btrfs_release_path(path);
4618 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4619 ret = btrfs_insert_file_extent(trans, root->log_root,
4620 ino, prev_extent_end, 0, 0,
4621 hole_len, 0, hole_len,
4631 * When we are logging a new inode X, check if it doesn't have a reference that
4632 * matches the reference from some other inode Y created in a past transaction
4633 * and that was renamed in the current transaction. If we don't do this, then at
4634 * log replay time we can lose inode Y (and all its files if it's a directory):
4637 * echo "hello world" > /mnt/x/foobar
4640 * mkdir /mnt/x # or touch /mnt/x
4641 * xfs_io -c fsync /mnt/x
4643 * mount fs, trigger log replay
4645 * After the log replay procedure, we would lose the first directory and all its
4646 * files (file foobar).
4647 * For the case where inode Y is not a directory we simply end up losing it:
4649 * echo "123" > /mnt/foo
4651 * mv /mnt/foo /mnt/bar
4652 * echo "abc" > /mnt/foo
4653 * xfs_io -c fsync /mnt/foo
4656 * We also need this for cases where a snapshot entry is replaced by some other
4657 * entry (file or directory) otherwise we end up with an unreplayable log due to
4658 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4659 * if it were a regular entry:
4662 * btrfs subvolume snapshot /mnt /mnt/x/snap
4663 * btrfs subvolume delete /mnt/x/snap
4666 * fsync /mnt/x or fsync some new file inside it
4669 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4670 * the same transaction.
4672 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4674 const struct btrfs_key *key,
4675 struct btrfs_inode *inode,
4676 u64 *other_ino, u64 *other_parent)
4679 struct btrfs_path *search_path;
4682 u32 item_size = btrfs_item_size_nr(eb, slot);
4684 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4686 search_path = btrfs_alloc_path();
4689 search_path->search_commit_root = 1;
4690 search_path->skip_locking = 1;
4692 while (cur_offset < item_size) {
4696 unsigned long name_ptr;
4697 struct btrfs_dir_item *di;
4699 if (key->type == BTRFS_INODE_REF_KEY) {
4700 struct btrfs_inode_ref *iref;
4702 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4703 parent = key->offset;
4704 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4705 name_ptr = (unsigned long)(iref + 1);
4706 this_len = sizeof(*iref) + this_name_len;
4708 struct btrfs_inode_extref *extref;
4710 extref = (struct btrfs_inode_extref *)(ptr +
4712 parent = btrfs_inode_extref_parent(eb, extref);
4713 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4714 name_ptr = (unsigned long)&extref->name;
4715 this_len = sizeof(*extref) + this_name_len;
4718 if (this_name_len > name_len) {
4721 new_name = krealloc(name, this_name_len, GFP_NOFS);
4726 name_len = this_name_len;
4730 read_extent_buffer(eb, name, name_ptr, this_name_len);
4731 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4732 parent, name, this_name_len, 0);
4733 if (di && !IS_ERR(di)) {
4734 struct btrfs_key di_key;
4736 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4738 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4739 if (di_key.objectid != key->objectid) {
4741 *other_ino = di_key.objectid;
4742 *other_parent = parent;
4750 } else if (IS_ERR(di)) {
4754 btrfs_release_path(search_path);
4756 cur_offset += this_len;
4760 btrfs_free_path(search_path);
4765 struct btrfs_ino_list {
4768 struct list_head list;
4771 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4772 struct btrfs_root *root,
4773 struct btrfs_path *path,
4774 struct btrfs_log_ctx *ctx,
4775 u64 ino, u64 parent)
4777 struct btrfs_ino_list *ino_elem;
4778 LIST_HEAD(inode_list);
4781 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4784 ino_elem->ino = ino;
4785 ino_elem->parent = parent;
4786 list_add_tail(&ino_elem->list, &inode_list);
4788 while (!list_empty(&inode_list)) {
4789 struct btrfs_fs_info *fs_info = root->fs_info;
4790 struct btrfs_key key;
4791 struct inode *inode;
4793 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4795 ino = ino_elem->ino;
4796 parent = ino_elem->parent;
4797 list_del(&ino_elem->list);
4802 btrfs_release_path(path);
4805 key.type = BTRFS_INODE_ITEM_KEY;
4807 inode = btrfs_iget(fs_info->sb, &key, root);
4809 * If the other inode that had a conflicting dir entry was
4810 * deleted in the current transaction, we need to log its parent
4813 if (IS_ERR(inode)) {
4814 ret = PTR_ERR(inode);
4815 if (ret == -ENOENT) {
4816 key.objectid = parent;
4817 inode = btrfs_iget(fs_info->sb, &key, root);
4818 if (IS_ERR(inode)) {
4819 ret = PTR_ERR(inode);
4821 ret = btrfs_log_inode(trans, root,
4823 LOG_OTHER_INODE_ALL,
4825 btrfs_add_delayed_iput(inode);
4831 * If the inode was already logged skip it - otherwise we can
4832 * hit an infinite loop. Example:
4834 * From the commit root (previous transaction) we have the
4837 * inode 257 a directory
4838 * inode 258 with references "zz" and "zz_link" on inode 257
4839 * inode 259 with reference "a" on inode 257
4841 * And in the current (uncommitted) transaction we have:
4843 * inode 257 a directory, unchanged
4844 * inode 258 with references "a" and "a2" on inode 257
4845 * inode 259 with reference "zz_link" on inode 257
4846 * inode 261 with reference "zz" on inode 257
4848 * When logging inode 261 the following infinite loop could
4849 * happen if we don't skip already logged inodes:
4851 * - we detect inode 258 as a conflicting inode, with inode 261
4852 * on reference "zz", and log it;
4854 * - we detect inode 259 as a conflicting inode, with inode 258
4855 * on reference "a", and log it;
4857 * - we detect inode 258 as a conflicting inode, with inode 259
4858 * on reference "zz_link", and log it - again! After this we
4859 * repeat the above steps forever.
4861 spin_lock(&BTRFS_I(inode)->lock);
4863 * Check the inode's logged_trans only instead of
4864 * btrfs_inode_in_log(). This is because the last_log_commit of
4865 * the inode is not updated when we only log that it exists and
4866 * and it has the full sync bit set (see btrfs_log_inode()).
4868 if (BTRFS_I(inode)->logged_trans == trans->transid) {
4869 spin_unlock(&BTRFS_I(inode)->lock);
4870 btrfs_add_delayed_iput(inode);
4873 spin_unlock(&BTRFS_I(inode)->lock);
4875 * We are safe logging the other inode without acquiring its
4876 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4877 * are safe against concurrent renames of the other inode as
4878 * well because during a rename we pin the log and update the
4879 * log with the new name before we unpin it.
4881 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4882 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4884 btrfs_add_delayed_iput(inode);
4889 key.type = BTRFS_INODE_REF_KEY;
4891 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4893 btrfs_add_delayed_iput(inode);
4898 struct extent_buffer *leaf = path->nodes[0];
4899 int slot = path->slots[0];
4901 u64 other_parent = 0;
4903 if (slot >= btrfs_header_nritems(leaf)) {
4904 ret = btrfs_next_leaf(root, path);
4907 } else if (ret > 0) {
4914 btrfs_item_key_to_cpu(leaf, &key, slot);
4915 if (key.objectid != ino ||
4916 (key.type != BTRFS_INODE_REF_KEY &&
4917 key.type != BTRFS_INODE_EXTREF_KEY)) {
4922 ret = btrfs_check_ref_name_override(leaf, slot, &key,
4923 BTRFS_I(inode), &other_ino,
4928 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4933 ino_elem->ino = other_ino;
4934 ino_elem->parent = other_parent;
4935 list_add_tail(&ino_elem->list, &inode_list);
4940 btrfs_add_delayed_iput(inode);
4946 /* log a single inode in the tree log.
4947 * At least one parent directory for this inode must exist in the tree
4948 * or be logged already.
4950 * Any items from this inode changed by the current transaction are copied
4951 * to the log tree. An extra reference is taken on any extents in this
4952 * file, allowing us to avoid a whole pile of corner cases around logging
4953 * blocks that have been removed from the tree.
4955 * See LOG_INODE_ALL and related defines for a description of what inode_only
4958 * This handles both files and directories.
4960 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4961 struct btrfs_root *root, struct btrfs_inode *inode,
4965 struct btrfs_log_ctx *ctx)
4967 struct btrfs_fs_info *fs_info = root->fs_info;
4968 struct btrfs_path *path;
4969 struct btrfs_path *dst_path;
4970 struct btrfs_key min_key;
4971 struct btrfs_key max_key;
4972 struct btrfs_root *log = root->log_root;
4976 int ins_start_slot = 0;
4978 bool fast_search = false;
4979 u64 ino = btrfs_ino(inode);
4980 struct extent_map_tree *em_tree = &inode->extent_tree;
4981 u64 logged_isize = 0;
4982 bool need_log_inode_item = true;
4983 bool xattrs_logged = false;
4984 bool recursive_logging = false;
4986 path = btrfs_alloc_path();
4989 dst_path = btrfs_alloc_path();
4991 btrfs_free_path(path);
4995 min_key.objectid = ino;
4996 min_key.type = BTRFS_INODE_ITEM_KEY;
4999 max_key.objectid = ino;
5002 /* today the code can only do partial logging of directories */
5003 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5004 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5005 &inode->runtime_flags) &&
5006 inode_only >= LOG_INODE_EXISTS))
5007 max_key.type = BTRFS_XATTR_ITEM_KEY;
5009 max_key.type = (u8)-1;
5010 max_key.offset = (u64)-1;
5013 * Only run delayed items if we are a dir or a new file.
5014 * Otherwise commit the delayed inode only, which is needed in
5015 * order for the log replay code to mark inodes for link count
5016 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5018 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5019 inode->generation > fs_info->last_trans_committed)
5020 ret = btrfs_commit_inode_delayed_items(trans, inode);
5022 ret = btrfs_commit_inode_delayed_inode(inode);
5025 btrfs_free_path(path);
5026 btrfs_free_path(dst_path);
5030 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5031 recursive_logging = true;
5032 if (inode_only == LOG_OTHER_INODE)
5033 inode_only = LOG_INODE_EXISTS;
5035 inode_only = LOG_INODE_ALL;
5036 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5038 mutex_lock(&inode->log_mutex);
5042 * a brute force approach to making sure we get the most uptodate
5043 * copies of everything.
5045 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5046 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5048 if (inode_only == LOG_INODE_EXISTS)
5049 max_key_type = BTRFS_XATTR_ITEM_KEY;
5050 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5052 if (inode_only == LOG_INODE_EXISTS) {
5054 * Make sure the new inode item we write to the log has
5055 * the same isize as the current one (if it exists).
5056 * This is necessary to prevent data loss after log
5057 * replay, and also to prevent doing a wrong expanding
5058 * truncate - for e.g. create file, write 4K into offset
5059 * 0, fsync, write 4K into offset 4096, add hard link,
5060 * fsync some other file (to sync log), power fail - if
5061 * we use the inode's current i_size, after log replay
5062 * we get a 8Kb file, with the last 4Kb extent as a hole
5063 * (zeroes), as if an expanding truncate happened,
5064 * instead of getting a file of 4Kb only.
5066 err = logged_inode_size(log, inode, path, &logged_isize);
5070 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5071 &inode->runtime_flags)) {
5072 if (inode_only == LOG_INODE_EXISTS) {
5073 max_key.type = BTRFS_XATTR_ITEM_KEY;
5074 ret = drop_objectid_items(trans, log, path, ino,
5077 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5078 &inode->runtime_flags);
5079 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5080 &inode->runtime_flags);
5082 ret = btrfs_truncate_inode_items(trans,
5083 log, &inode->vfs_inode, 0, 0);
5088 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5089 &inode->runtime_flags) ||
5090 inode_only == LOG_INODE_EXISTS) {
5091 if (inode_only == LOG_INODE_ALL)
5093 max_key.type = BTRFS_XATTR_ITEM_KEY;
5094 ret = drop_objectid_items(trans, log, path, ino,
5097 if (inode_only == LOG_INODE_ALL)
5110 ret = btrfs_search_forward(root, &min_key,
5111 path, trans->transid);
5119 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5120 if (min_key.objectid != ino)
5122 if (min_key.type > max_key.type)
5125 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5126 need_log_inode_item = false;
5128 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5129 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5130 inode->generation == trans->transid &&
5131 !recursive_logging) {
5133 u64 other_parent = 0;
5135 ret = btrfs_check_ref_name_override(path->nodes[0],
5136 path->slots[0], &min_key, inode,
5137 &other_ino, &other_parent);
5141 } else if (ret > 0 && ctx &&
5142 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5147 ins_start_slot = path->slots[0];
5149 ret = copy_items(trans, inode, dst_path, path,
5159 err = log_conflicting_inodes(trans, root, path,
5160 ctx, other_ino, other_parent);
5163 btrfs_release_path(path);
5168 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5169 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5172 ret = copy_items(trans, inode, dst_path, path,
5174 ins_nr, inode_only, logged_isize);
5183 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5186 } else if (!ins_nr) {
5187 ins_start_slot = path->slots[0];
5192 ret = copy_items(trans, inode, dst_path, path,
5193 ins_start_slot, ins_nr, inode_only,
5200 ins_start_slot = path->slots[0];
5203 nritems = btrfs_header_nritems(path->nodes[0]);
5205 if (path->slots[0] < nritems) {
5206 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5211 ret = copy_items(trans, inode, dst_path, path,
5213 ins_nr, inode_only, logged_isize);
5220 btrfs_release_path(path);
5222 if (min_key.offset < (u64)-1) {
5224 } else if (min_key.type < max_key.type) {
5232 ret = copy_items(trans, inode, dst_path, path,
5233 ins_start_slot, ins_nr, inode_only,
5242 btrfs_release_path(path);
5243 btrfs_release_path(dst_path);
5244 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5247 xattrs_logged = true;
5248 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5249 btrfs_release_path(path);
5250 btrfs_release_path(dst_path);
5251 err = btrfs_log_holes(trans, root, inode, path);
5256 btrfs_release_path(path);
5257 btrfs_release_path(dst_path);
5258 if (need_log_inode_item) {
5259 err = log_inode_item(trans, log, dst_path, inode);
5260 if (!err && !xattrs_logged) {
5261 err = btrfs_log_all_xattrs(trans, root, inode, path,
5263 btrfs_release_path(path);
5269 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5275 } else if (inode_only == LOG_INODE_ALL) {
5276 struct extent_map *em, *n;
5278 write_lock(&em_tree->lock);
5280 * We can't just remove every em if we're called for a ranged
5281 * fsync - that is, one that doesn't cover the whole possible
5282 * file range (0 to LLONG_MAX). This is because we can have
5283 * em's that fall outside the range we're logging and therefore
5284 * their ordered operations haven't completed yet
5285 * (btrfs_finish_ordered_io() not invoked yet). This means we
5286 * didn't get their respective file extent item in the fs/subvol
5287 * tree yet, and need to let the next fast fsync (one which
5288 * consults the list of modified extent maps) find the em so
5289 * that it logs a matching file extent item and waits for the
5290 * respective ordered operation to complete (if it's still
5293 * Removing every em outside the range we're logging would make
5294 * the next fast fsync not log their matching file extent items,
5295 * therefore making us lose data after a log replay.
5297 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5299 const u64 mod_end = em->mod_start + em->mod_len - 1;
5301 if (em->mod_start >= start && mod_end <= end)
5302 list_del_init(&em->list);
5304 write_unlock(&em_tree->lock);
5307 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5308 ret = log_directory_changes(trans, root, inode, path, dst_path,
5317 * Don't update last_log_commit if we logged that an inode exists after
5318 * it was loaded to memory (full_sync bit set).
5319 * This is to prevent data loss when we do a write to the inode, then
5320 * the inode gets evicted after all delalloc was flushed, then we log
5321 * it exists (due to a rename for example) and then fsync it. This last
5322 * fsync would do nothing (not logging the extents previously written).
5324 spin_lock(&inode->lock);
5325 inode->logged_trans = trans->transid;
5326 if (inode_only != LOG_INODE_EXISTS ||
5327 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5328 inode->last_log_commit = inode->last_sub_trans;
5329 spin_unlock(&inode->lock);
5331 mutex_unlock(&inode->log_mutex);
5333 btrfs_free_path(path);
5334 btrfs_free_path(dst_path);
5339 * Check if we must fallback to a transaction commit when logging an inode.
5340 * This must be called after logging the inode and is used only in the context
5341 * when fsyncing an inode requires the need to log some other inode - in which
5342 * case we can't lock the i_mutex of each other inode we need to log as that
5343 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5344 * log inodes up or down in the hierarchy) or rename operations for example. So
5345 * we take the log_mutex of the inode after we have logged it and then check for
5346 * its last_unlink_trans value - this is safe because any task setting
5347 * last_unlink_trans must take the log_mutex and it must do this before it does
5348 * the actual unlink operation, so if we do this check before a concurrent task
5349 * sets last_unlink_trans it means we've logged a consistent version/state of
5350 * all the inode items, otherwise we are not sure and must do a transaction
5351 * commit (the concurrent task might have only updated last_unlink_trans before
5352 * we logged the inode or it might have also done the unlink).
5354 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5355 struct btrfs_inode *inode)
5357 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5360 mutex_lock(&inode->log_mutex);
5361 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5363 * Make sure any commits to the log are forced to be full
5366 btrfs_set_log_full_commit(trans);
5369 mutex_unlock(&inode->log_mutex);
5375 * follow the dentry parent pointers up the chain and see if any
5376 * of the directories in it require a full commit before they can
5377 * be logged. Returns zero if nothing special needs to be done or 1 if
5378 * a full commit is required.
5380 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5381 struct btrfs_inode *inode,
5382 struct dentry *parent,
5383 struct super_block *sb,
5387 struct dentry *old_parent = NULL;
5390 * for regular files, if its inode is already on disk, we don't
5391 * have to worry about the parents at all. This is because
5392 * we can use the last_unlink_trans field to record renames
5393 * and other fun in this file.
5395 if (S_ISREG(inode->vfs_inode.i_mode) &&
5396 inode->generation <= last_committed &&
5397 inode->last_unlink_trans <= last_committed)
5400 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5401 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5403 inode = BTRFS_I(d_inode(parent));
5407 if (btrfs_must_commit_transaction(trans, inode)) {
5412 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5415 if (IS_ROOT(parent)) {
5416 inode = BTRFS_I(d_inode(parent));
5417 if (btrfs_must_commit_transaction(trans, inode))
5422 parent = dget_parent(parent);
5424 old_parent = parent;
5425 inode = BTRFS_I(d_inode(parent));
5433 struct btrfs_dir_list {
5435 struct list_head list;
5439 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5440 * details about the why it is needed.
5441 * This is a recursive operation - if an existing dentry corresponds to a
5442 * directory, that directory's new entries are logged too (same behaviour as
5443 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5444 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5445 * complains about the following circular lock dependency / possible deadlock:
5449 * lock(&type->i_mutex_dir_key#3/2);
5450 * lock(sb_internal#2);
5451 * lock(&type->i_mutex_dir_key#3/2);
5452 * lock(&sb->s_type->i_mutex_key#14);
5454 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5455 * sb_start_intwrite() in btrfs_start_transaction().
5456 * Not locking i_mutex of the inodes is still safe because:
5458 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5459 * that while logging the inode new references (names) are added or removed
5460 * from the inode, leaving the logged inode item with a link count that does
5461 * not match the number of logged inode reference items. This is fine because
5462 * at log replay time we compute the real number of links and correct the
5463 * link count in the inode item (see replay_one_buffer() and
5464 * link_to_fixup_dir());
5466 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5467 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5468 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5469 * has a size that doesn't match the sum of the lengths of all the logged
5470 * names. This does not result in a problem because if a dir_item key is
5471 * logged but its matching dir_index key is not logged, at log replay time we
5472 * don't use it to replay the respective name (see replay_one_name()). On the
5473 * other hand if only the dir_index key ends up being logged, the respective
5474 * name is added to the fs/subvol tree with both the dir_item and dir_index
5475 * keys created (see replay_one_name()).
5476 * The directory's inode item with a wrong i_size is not a problem as well,
5477 * since we don't use it at log replay time to set the i_size in the inode
5478 * item of the fs/subvol tree (see overwrite_item()).
5480 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5481 struct btrfs_root *root,
5482 struct btrfs_inode *start_inode,
5483 struct btrfs_log_ctx *ctx)
5485 struct btrfs_fs_info *fs_info = root->fs_info;
5486 struct btrfs_root *log = root->log_root;
5487 struct btrfs_path *path;
5488 LIST_HEAD(dir_list);
5489 struct btrfs_dir_list *dir_elem;
5492 path = btrfs_alloc_path();
5496 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5498 btrfs_free_path(path);
5501 dir_elem->ino = btrfs_ino(start_inode);
5502 list_add_tail(&dir_elem->list, &dir_list);
5504 while (!list_empty(&dir_list)) {
5505 struct extent_buffer *leaf;
5506 struct btrfs_key min_key;
5510 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5513 goto next_dir_inode;
5515 min_key.objectid = dir_elem->ino;
5516 min_key.type = BTRFS_DIR_ITEM_KEY;
5519 btrfs_release_path(path);
5520 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5522 goto next_dir_inode;
5523 } else if (ret > 0) {
5525 goto next_dir_inode;
5529 leaf = path->nodes[0];
5530 nritems = btrfs_header_nritems(leaf);
5531 for (i = path->slots[0]; i < nritems; i++) {
5532 struct btrfs_dir_item *di;
5533 struct btrfs_key di_key;
5534 struct inode *di_inode;
5535 struct btrfs_dir_list *new_dir_elem;
5536 int log_mode = LOG_INODE_EXISTS;
5539 btrfs_item_key_to_cpu(leaf, &min_key, i);
5540 if (min_key.objectid != dir_elem->ino ||
5541 min_key.type != BTRFS_DIR_ITEM_KEY)
5542 goto next_dir_inode;
5544 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5545 type = btrfs_dir_type(leaf, di);
5546 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5547 type != BTRFS_FT_DIR)
5549 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5550 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5553 btrfs_release_path(path);
5554 di_inode = btrfs_iget(fs_info->sb, &di_key, root);
5555 if (IS_ERR(di_inode)) {
5556 ret = PTR_ERR(di_inode);
5557 goto next_dir_inode;
5560 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5561 btrfs_add_delayed_iput(di_inode);
5565 ctx->log_new_dentries = false;
5566 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5567 log_mode = LOG_INODE_ALL;
5568 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5569 log_mode, 0, LLONG_MAX, ctx);
5571 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5573 btrfs_add_delayed_iput(di_inode);
5575 goto next_dir_inode;
5576 if (ctx->log_new_dentries) {
5577 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5579 if (!new_dir_elem) {
5581 goto next_dir_inode;
5583 new_dir_elem->ino = di_key.objectid;
5584 list_add_tail(&new_dir_elem->list, &dir_list);
5589 ret = btrfs_next_leaf(log, path);
5591 goto next_dir_inode;
5592 } else if (ret > 0) {
5594 goto next_dir_inode;
5598 if (min_key.offset < (u64)-1) {
5603 list_del(&dir_elem->list);
5607 btrfs_free_path(path);
5611 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5612 struct btrfs_inode *inode,
5613 struct btrfs_log_ctx *ctx)
5615 struct btrfs_fs_info *fs_info = trans->fs_info;
5617 struct btrfs_path *path;
5618 struct btrfs_key key;
5619 struct btrfs_root *root = inode->root;
5620 const u64 ino = btrfs_ino(inode);
5622 path = btrfs_alloc_path();
5625 path->skip_locking = 1;
5626 path->search_commit_root = 1;
5629 key.type = BTRFS_INODE_REF_KEY;
5631 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5636 struct extent_buffer *leaf = path->nodes[0];
5637 int slot = path->slots[0];
5642 if (slot >= btrfs_header_nritems(leaf)) {
5643 ret = btrfs_next_leaf(root, path);
5651 btrfs_item_key_to_cpu(leaf, &key, slot);
5652 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5653 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5656 item_size = btrfs_item_size_nr(leaf, slot);
5657 ptr = btrfs_item_ptr_offset(leaf, slot);
5658 while (cur_offset < item_size) {
5659 struct btrfs_key inode_key;
5660 struct inode *dir_inode;
5662 inode_key.type = BTRFS_INODE_ITEM_KEY;
5663 inode_key.offset = 0;
5665 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5666 struct btrfs_inode_extref *extref;
5668 extref = (struct btrfs_inode_extref *)
5670 inode_key.objectid = btrfs_inode_extref_parent(
5672 cur_offset += sizeof(*extref);
5673 cur_offset += btrfs_inode_extref_name_len(leaf,
5676 inode_key.objectid = key.offset;
5677 cur_offset = item_size;
5680 dir_inode = btrfs_iget(fs_info->sb, &inode_key, root);
5682 * If the parent inode was deleted, return an error to
5683 * fallback to a transaction commit. This is to prevent
5684 * getting an inode that was moved from one parent A to
5685 * a parent B, got its former parent A deleted and then
5686 * it got fsync'ed, from existing at both parents after
5687 * a log replay (and the old parent still existing).
5694 * mv /mnt/B/bar /mnt/A/bar
5695 * mv -T /mnt/A /mnt/B
5699 * If we ignore the old parent B which got deleted,
5700 * after a log replay we would have file bar linked
5701 * at both parents and the old parent B would still
5704 if (IS_ERR(dir_inode)) {
5705 ret = PTR_ERR(dir_inode);
5710 ctx->log_new_dentries = false;
5711 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5712 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5714 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5716 if (!ret && ctx && ctx->log_new_dentries)
5717 ret = log_new_dir_dentries(trans, root,
5718 BTRFS_I(dir_inode), ctx);
5719 btrfs_add_delayed_iput(dir_inode);
5727 btrfs_free_path(path);
5731 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5732 struct btrfs_root *root,
5733 struct btrfs_path *path,
5734 struct btrfs_log_ctx *ctx)
5736 struct btrfs_key found_key;
5738 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5741 struct btrfs_fs_info *fs_info = root->fs_info;
5742 const u64 last_committed = fs_info->last_trans_committed;
5743 struct extent_buffer *leaf = path->nodes[0];
5744 int slot = path->slots[0];
5745 struct btrfs_key search_key;
5746 struct inode *inode;
5749 btrfs_release_path(path);
5751 search_key.objectid = found_key.offset;
5752 search_key.type = BTRFS_INODE_ITEM_KEY;
5753 search_key.offset = 0;
5754 inode = btrfs_iget(fs_info->sb, &search_key, root);
5756 return PTR_ERR(inode);
5758 if (BTRFS_I(inode)->generation > last_committed)
5759 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5762 btrfs_add_delayed_iput(inode);
5766 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5769 search_key.type = BTRFS_INODE_REF_KEY;
5770 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5774 leaf = path->nodes[0];
5775 slot = path->slots[0];
5776 if (slot >= btrfs_header_nritems(leaf)) {
5777 ret = btrfs_next_leaf(root, path);
5782 leaf = path->nodes[0];
5783 slot = path->slots[0];
5786 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5787 if (found_key.objectid != search_key.objectid ||
5788 found_key.type != BTRFS_INODE_REF_KEY)
5794 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5795 struct btrfs_inode *inode,
5796 struct dentry *parent,
5797 struct btrfs_log_ctx *ctx)
5799 struct btrfs_root *root = inode->root;
5800 struct btrfs_fs_info *fs_info = root->fs_info;
5801 struct dentry *old_parent = NULL;
5802 struct super_block *sb = inode->vfs_inode.i_sb;
5806 if (!parent || d_really_is_negative(parent) ||
5810 inode = BTRFS_I(d_inode(parent));
5811 if (root != inode->root)
5814 if (inode->generation > fs_info->last_trans_committed) {
5815 ret = btrfs_log_inode(trans, root, inode,
5816 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5820 if (IS_ROOT(parent))
5823 parent = dget_parent(parent);
5825 old_parent = parent;
5832 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5833 struct btrfs_inode *inode,
5834 struct dentry *parent,
5835 struct btrfs_log_ctx *ctx)
5837 struct btrfs_root *root = inode->root;
5838 const u64 ino = btrfs_ino(inode);
5839 struct btrfs_path *path;
5840 struct btrfs_key search_key;
5844 * For a single hard link case, go through a fast path that does not
5845 * need to iterate the fs/subvolume tree.
5847 if (inode->vfs_inode.i_nlink < 2)
5848 return log_new_ancestors_fast(trans, inode, parent, ctx);
5850 path = btrfs_alloc_path();
5854 search_key.objectid = ino;
5855 search_key.type = BTRFS_INODE_REF_KEY;
5856 search_key.offset = 0;
5858 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5865 struct extent_buffer *leaf = path->nodes[0];
5866 int slot = path->slots[0];
5867 struct btrfs_key found_key;
5869 if (slot >= btrfs_header_nritems(leaf)) {
5870 ret = btrfs_next_leaf(root, path);
5878 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5879 if (found_key.objectid != ino ||
5880 found_key.type > BTRFS_INODE_EXTREF_KEY)
5884 * Don't deal with extended references because they are rare
5885 * cases and too complex to deal with (we would need to keep
5886 * track of which subitem we are processing for each item in
5887 * this loop, etc). So just return some error to fallback to
5888 * a transaction commit.
5890 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
5896 * Logging ancestors needs to do more searches on the fs/subvol
5897 * tree, so it releases the path as needed to avoid deadlocks.
5898 * Keep track of the last inode ref key and resume from that key
5899 * after logging all new ancestors for the current hard link.
5901 memcpy(&search_key, &found_key, sizeof(search_key));
5903 ret = log_new_ancestors(trans, root, path, ctx);
5906 btrfs_release_path(path);
5911 btrfs_free_path(path);
5916 * helper function around btrfs_log_inode to make sure newly created
5917 * parent directories also end up in the log. A minimal inode and backref
5918 * only logging is done of any parent directories that are older than
5919 * the last committed transaction
5921 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5922 struct btrfs_inode *inode,
5923 struct dentry *parent,
5927 struct btrfs_log_ctx *ctx)
5929 struct btrfs_root *root = inode->root;
5930 struct btrfs_fs_info *fs_info = root->fs_info;
5931 struct super_block *sb;
5933 u64 last_committed = fs_info->last_trans_committed;
5934 bool log_dentries = false;
5936 sb = inode->vfs_inode.i_sb;
5938 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5944 * The prev transaction commit doesn't complete, we need do
5945 * full commit by ourselves.
5947 if (fs_info->last_trans_log_full_commit >
5948 fs_info->last_trans_committed) {
5953 if (btrfs_root_refs(&root->root_item) == 0) {
5958 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5964 * Skip already logged inodes or inodes corresponding to tmpfiles
5965 * (since logging them is pointless, a link count of 0 means they
5966 * will never be accessible).
5968 if (btrfs_inode_in_log(inode, trans->transid) ||
5969 inode->vfs_inode.i_nlink == 0) {
5970 ret = BTRFS_NO_LOG_SYNC;
5974 ret = start_log_trans(trans, root, ctx);
5978 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5983 * for regular files, if its inode is already on disk, we don't
5984 * have to worry about the parents at all. This is because
5985 * we can use the last_unlink_trans field to record renames
5986 * and other fun in this file.
5988 if (S_ISREG(inode->vfs_inode.i_mode) &&
5989 inode->generation <= last_committed &&
5990 inode->last_unlink_trans <= last_committed) {
5995 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5996 log_dentries = true;
5999 * On unlink we must make sure all our current and old parent directory
6000 * inodes are fully logged. This is to prevent leaving dangling
6001 * directory index entries in directories that were our parents but are
6002 * not anymore. Not doing this results in old parent directory being
6003 * impossible to delete after log replay (rmdir will always fail with
6004 * error -ENOTEMPTY).
6010 * ln testdir/foo testdir/bar
6012 * unlink testdir/bar
6013 * xfs_io -c fsync testdir/foo
6015 * mount fs, triggers log replay
6017 * If we don't log the parent directory (testdir), after log replay the
6018 * directory still has an entry pointing to the file inode using the bar
6019 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6020 * the file inode has a link count of 1.
6026 * ln foo testdir/foo2
6027 * ln foo testdir/foo3
6029 * unlink testdir/foo3
6030 * xfs_io -c fsync foo
6032 * mount fs, triggers log replay
6034 * Similar as the first example, after log replay the parent directory
6035 * testdir still has an entry pointing to the inode file with name foo3
6036 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6037 * and has a link count of 2.
6039 if (inode->last_unlink_trans > last_committed) {
6040 ret = btrfs_log_all_parents(trans, inode, ctx);
6045 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6050 ret = log_new_dir_dentries(trans, root, inode, ctx);
6055 btrfs_set_log_full_commit(trans);
6060 btrfs_remove_log_ctx(root, ctx);
6061 btrfs_end_log_trans(root);
6067 * it is not safe to log dentry if the chunk root has added new
6068 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6069 * If this returns 1, you must commit the transaction to safely get your
6072 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6073 struct dentry *dentry,
6076 struct btrfs_log_ctx *ctx)
6078 struct dentry *parent = dget_parent(dentry);
6081 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6082 start, end, LOG_INODE_ALL, ctx);
6089 * should be called during mount to recover any replay any log trees
6092 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6095 struct btrfs_path *path;
6096 struct btrfs_trans_handle *trans;
6097 struct btrfs_key key;
6098 struct btrfs_key found_key;
6099 struct btrfs_key tmp_key;
6100 struct btrfs_root *log;
6101 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6102 struct walk_control wc = {
6103 .process_func = process_one_buffer,
6104 .stage = LOG_WALK_PIN_ONLY,
6107 path = btrfs_alloc_path();
6111 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6113 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6114 if (IS_ERR(trans)) {
6115 ret = PTR_ERR(trans);
6122 ret = walk_log_tree(trans, log_root_tree, &wc);
6124 btrfs_handle_fs_error(fs_info, ret,
6125 "Failed to pin buffers while recovering log root tree.");
6130 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6131 key.offset = (u64)-1;
6132 key.type = BTRFS_ROOT_ITEM_KEY;
6135 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6138 btrfs_handle_fs_error(fs_info, ret,
6139 "Couldn't find tree log root.");
6143 if (path->slots[0] == 0)
6147 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6149 btrfs_release_path(path);
6150 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6153 log = btrfs_read_tree_root(log_root_tree, &found_key);
6156 btrfs_handle_fs_error(fs_info, ret,
6157 "Couldn't read tree log root.");
6161 tmp_key.objectid = found_key.offset;
6162 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6163 tmp_key.offset = (u64)-1;
6165 wc.replay_dest = btrfs_get_fs_root(fs_info, &tmp_key, true);
6166 if (!IS_ERR(wc.replay_dest)) {
6167 if (!btrfs_grab_fs_root(wc.replay_dest))
6168 wc.replay_dest = ERR_PTR(-ENOENT);
6170 if (IS_ERR(wc.replay_dest)) {
6171 ret = PTR_ERR(wc.replay_dest);
6174 * We didn't find the subvol, likely because it was
6175 * deleted. This is ok, simply skip this log and go to
6178 * We need to exclude the root because we can't have
6179 * other log replays overwriting this log as we'll read
6180 * it back in a few more times. This will keep our
6181 * block from being modified, and we'll just bail for
6182 * each subsequent pass.
6185 ret = btrfs_pin_extent_for_log_replay(fs_info,
6188 free_extent_buffer(log->node);
6189 free_extent_buffer(log->commit_root);
6190 btrfs_put_fs_root(log);
6194 btrfs_handle_fs_error(fs_info, ret,
6195 "Couldn't read target root for tree log recovery.");
6199 wc.replay_dest->log_root = log;
6200 btrfs_record_root_in_trans(trans, wc.replay_dest);
6201 ret = walk_log_tree(trans, log, &wc);
6203 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6204 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6208 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6209 struct btrfs_root *root = wc.replay_dest;
6211 btrfs_release_path(path);
6214 * We have just replayed everything, and the highest
6215 * objectid of fs roots probably has changed in case
6216 * some inode_item's got replayed.
6218 * root->objectid_mutex is not acquired as log replay
6219 * could only happen during mount.
6221 ret = btrfs_find_highest_objectid(root,
6222 &root->highest_objectid);
6225 wc.replay_dest->log_root = NULL;
6226 btrfs_put_fs_root(wc.replay_dest);
6227 free_extent_buffer(log->node);
6228 free_extent_buffer(log->commit_root);
6229 btrfs_put_fs_root(log);
6234 if (found_key.offset == 0)
6236 key.offset = found_key.offset - 1;
6238 btrfs_release_path(path);
6240 /* step one is to pin it all, step two is to replay just inodes */
6243 wc.process_func = replay_one_buffer;
6244 wc.stage = LOG_WALK_REPLAY_INODES;
6247 /* step three is to replay everything */
6248 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6253 btrfs_free_path(path);
6255 /* step 4: commit the transaction, which also unpins the blocks */
6256 ret = btrfs_commit_transaction(trans);
6260 free_extent_buffer(log_root_tree->node);
6261 log_root_tree->log_root = NULL;
6262 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6263 btrfs_put_fs_root(log_root_tree);
6268 btrfs_end_transaction(wc.trans);
6269 btrfs_free_path(path);
6274 * there are some corner cases where we want to force a full
6275 * commit instead of allowing a directory to be logged.
6277 * They revolve around files there were unlinked from the directory, and
6278 * this function updates the parent directory so that a full commit is
6279 * properly done if it is fsync'd later after the unlinks are done.
6281 * Must be called before the unlink operations (updates to the subvolume tree,
6282 * inodes, etc) are done.
6284 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6285 struct btrfs_inode *dir, struct btrfs_inode *inode,
6289 * when we're logging a file, if it hasn't been renamed
6290 * or unlinked, and its inode is fully committed on disk,
6291 * we don't have to worry about walking up the directory chain
6292 * to log its parents.
6294 * So, we use the last_unlink_trans field to put this transid
6295 * into the file. When the file is logged we check it and
6296 * don't log the parents if the file is fully on disk.
6298 mutex_lock(&inode->log_mutex);
6299 inode->last_unlink_trans = trans->transid;
6300 mutex_unlock(&inode->log_mutex);
6303 * if this directory was already logged any new
6304 * names for this file/dir will get recorded
6306 if (dir->logged_trans == trans->transid)
6310 * if the inode we're about to unlink was logged,
6311 * the log will be properly updated for any new names
6313 if (inode->logged_trans == trans->transid)
6317 * when renaming files across directories, if the directory
6318 * there we're unlinking from gets fsync'd later on, there's
6319 * no way to find the destination directory later and fsync it
6320 * properly. So, we have to be conservative and force commits
6321 * so the new name gets discovered.
6326 /* we can safely do the unlink without any special recording */
6330 mutex_lock(&dir->log_mutex);
6331 dir->last_unlink_trans = trans->transid;
6332 mutex_unlock(&dir->log_mutex);
6336 * Make sure that if someone attempts to fsync the parent directory of a deleted
6337 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6338 * that after replaying the log tree of the parent directory's root we will not
6339 * see the snapshot anymore and at log replay time we will not see any log tree
6340 * corresponding to the deleted snapshot's root, which could lead to replaying
6341 * it after replaying the log tree of the parent directory (which would replay
6342 * the snapshot delete operation).
6344 * Must be called before the actual snapshot destroy operation (updates to the
6345 * parent root and tree of tree roots trees, etc) are done.
6347 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6348 struct btrfs_inode *dir)
6350 mutex_lock(&dir->log_mutex);
6351 dir->last_unlink_trans = trans->transid;
6352 mutex_unlock(&dir->log_mutex);
6356 * Call this after adding a new name for a file and it will properly
6357 * update the log to reflect the new name.
6359 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6360 * true (because it's not used).
6362 * Return value depends on whether @sync_log is true or false.
6363 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6364 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6366 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6367 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6368 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6369 * committed (without attempting to sync the log).
6371 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6372 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6373 struct dentry *parent,
6374 bool sync_log, struct btrfs_log_ctx *ctx)
6376 struct btrfs_fs_info *fs_info = trans->fs_info;
6380 * this will force the logging code to walk the dentry chain
6383 if (!S_ISDIR(inode->vfs_inode.i_mode))
6384 inode->last_unlink_trans = trans->transid;
6387 * if this inode hasn't been logged and directory we're renaming it
6388 * from hasn't been logged, we don't need to log it
6390 if (inode->logged_trans <= fs_info->last_trans_committed &&
6391 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6392 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6393 BTRFS_DONT_NEED_LOG_SYNC;
6396 struct btrfs_log_ctx ctx2;
6398 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6399 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6400 LOG_INODE_EXISTS, &ctx2);
6401 if (ret == BTRFS_NO_LOG_SYNC)
6402 return BTRFS_DONT_NEED_TRANS_COMMIT;
6404 return BTRFS_NEED_TRANS_COMMIT;
6406 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6408 return BTRFS_NEED_TRANS_COMMIT;
6409 return BTRFS_DONT_NEED_TRANS_COMMIT;
6413 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6414 LOG_INODE_EXISTS, ctx);
6415 if (ret == BTRFS_NO_LOG_SYNC)
6416 return BTRFS_DONT_NEED_LOG_SYNC;
6418 return BTRFS_NEED_TRANS_COMMIT;
6420 return BTRFS_NEED_LOG_SYNC;