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 "block-group.h"
21 #include "space-info.h"
24 /* magic values for the inode_only field in btrfs_log_inode:
26 * LOG_INODE_ALL means to log everything
27 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 * directory trouble cases
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
47 * rename foo/some_dir foo2/some_dir
49 * fsync foo/some_dir/some_file
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
81 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
91 LOG_WALK_REPLAY_INODES,
92 LOG_WALK_REPLAY_DIR_INDEX,
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 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);
108 static void wait_log_commit(struct btrfs_root *root, int transid);
111 * tree logging is a special write ahead log used to make sure that
112 * fsyncs and O_SYNCs can happen without doing full tree commits.
114 * Full tree commits are expensive because they require commonly
115 * modified blocks to be recowed, creating many dirty pages in the
116 * extent tree an 4x-6x higher write load than ext3.
118 * Instead of doing a tree commit on every fsync, we use the
119 * key ranges and transaction ids to find items for a given file or directory
120 * that have changed in this transaction. Those items are copied into
121 * a special tree (one per subvolume root), that tree is written to disk
122 * and then the fsync is considered complete.
124 * After a crash, items are copied out of the log-tree back into the
125 * subvolume tree. Any file data extents found are recorded in the extent
126 * allocation tree, and the log-tree freed.
128 * The log tree is read three times, once to pin down all the extents it is
129 * using in ram and once, once to create all the inodes logged in the tree
130 * and once to do all the other items.
134 * start a sub transaction and setup the log tree
135 * this increments the log tree writer count to make the people
136 * syncing the tree wait for us to finish
138 static int start_log_trans(struct btrfs_trans_handle *trans,
139 struct btrfs_root *root,
140 struct btrfs_log_ctx *ctx)
142 struct btrfs_fs_info *fs_info = root->fs_info;
143 struct btrfs_root *tree_root = fs_info->tree_root;
144 const bool zoned = btrfs_is_zoned(fs_info);
146 bool created = false;
149 * First check if the log root tree was already created. If not, create
150 * it before locking the root's log_mutex, just to keep lockdep happy.
152 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
153 mutex_lock(&tree_root->log_mutex);
154 if (!fs_info->log_root_tree) {
155 ret = btrfs_init_log_root_tree(trans, fs_info);
157 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
161 mutex_unlock(&tree_root->log_mutex);
166 mutex_lock(&root->log_mutex);
169 if (root->log_root) {
170 int index = (root->log_transid + 1) % 2;
172 if (btrfs_need_log_full_commit(trans)) {
177 if (zoned && atomic_read(&root->log_commit[index])) {
178 wait_log_commit(root, root->log_transid - 1);
182 if (!root->log_start_pid) {
183 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
184 root->log_start_pid = current->pid;
185 } else if (root->log_start_pid != current->pid) {
186 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
190 * This means fs_info->log_root_tree was already created
191 * for some other FS trees. Do the full commit not to mix
192 * nodes from multiple log transactions to do sequential
195 if (zoned && !created) {
200 ret = btrfs_add_log_tree(trans, root);
204 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
205 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
206 root->log_start_pid = current->pid;
209 atomic_inc(&root->log_writers);
210 if (ctx && !ctx->logging_new_name) {
211 int index = root->log_transid % 2;
212 list_add_tail(&ctx->list, &root->log_ctxs[index]);
213 ctx->log_transid = root->log_transid;
217 mutex_unlock(&root->log_mutex);
222 * returns 0 if there was a log transaction running and we were able
223 * to join, or returns -ENOENT if there were not transactions
226 static int join_running_log_trans(struct btrfs_root *root)
228 const bool zoned = btrfs_is_zoned(root->fs_info);
231 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
234 mutex_lock(&root->log_mutex);
236 if (root->log_root) {
237 int index = (root->log_transid + 1) % 2;
240 if (zoned && atomic_read(&root->log_commit[index])) {
241 wait_log_commit(root, root->log_transid - 1);
244 atomic_inc(&root->log_writers);
246 mutex_unlock(&root->log_mutex);
251 * This either makes the current running log transaction wait
252 * until you call btrfs_end_log_trans() or it makes any future
253 * log transactions wait until you call btrfs_end_log_trans()
255 void btrfs_pin_log_trans(struct btrfs_root *root)
257 atomic_inc(&root->log_writers);
261 * indicate we're done making changes to the log tree
262 * and wake up anyone waiting to do a sync
264 void btrfs_end_log_trans(struct btrfs_root *root)
266 if (atomic_dec_and_test(&root->log_writers)) {
267 /* atomic_dec_and_test implies a barrier */
268 cond_wake_up_nomb(&root->log_writer_wait);
272 static int btrfs_write_tree_block(struct extent_buffer *buf)
274 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
275 buf->start + buf->len - 1);
278 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
280 filemap_fdatawait_range(buf->pages[0]->mapping,
281 buf->start, buf->start + buf->len - 1);
285 * the walk control struct is used to pass state down the chain when
286 * processing the log tree. The stage field tells us which part
287 * of the log tree processing we are currently doing. The others
288 * are state fields used for that specific part
290 struct walk_control {
291 /* should we free the extent on disk when done? This is used
292 * at transaction commit time while freeing a log tree
296 /* should we write out the extent buffer? This is used
297 * while flushing the log tree to disk during a sync
301 /* should we wait for the extent buffer io to finish? Also used
302 * while flushing the log tree to disk for a sync
306 /* pin only walk, we record which extents on disk belong to the
311 /* what stage of the replay code we're currently in */
315 * Ignore any items from the inode currently being processed. Needs
316 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
317 * the LOG_WALK_REPLAY_INODES stage.
319 bool ignore_cur_inode;
321 /* the root we are currently replaying */
322 struct btrfs_root *replay_dest;
324 /* the trans handle for the current replay */
325 struct btrfs_trans_handle *trans;
327 /* the function that gets used to process blocks we find in the
328 * tree. Note the extent_buffer might not be up to date when it is
329 * passed in, and it must be checked or read if you need the data
332 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
333 struct walk_control *wc, u64 gen, int level);
337 * process_func used to pin down extents, write them or wait on them
339 static int process_one_buffer(struct btrfs_root *log,
340 struct extent_buffer *eb,
341 struct walk_control *wc, u64 gen, int level)
343 struct btrfs_fs_info *fs_info = log->fs_info;
347 * If this fs is mixed then we need to be able to process the leaves to
348 * pin down any logged extents, so we have to read the block.
350 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
351 ret = btrfs_read_buffer(eb, gen, level, NULL);
357 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
360 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
361 if (wc->pin && btrfs_header_level(eb) == 0)
362 ret = btrfs_exclude_logged_extents(eb);
364 btrfs_write_tree_block(eb);
366 btrfs_wait_tree_block_writeback(eb);
372 * Item overwrite used by replay and tree logging. eb, slot and key all refer
373 * to the src data we are copying out.
375 * root is the tree we are copying into, and path is a scratch
376 * path for use in this function (it should be released on entry and
377 * will be released on exit).
379 * If the key is already in the destination tree the existing item is
380 * overwritten. If the existing item isn't big enough, it is extended.
381 * If it is too large, it is truncated.
383 * If the key isn't in the destination yet, a new item is inserted.
385 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
386 struct btrfs_root *root,
387 struct btrfs_path *path,
388 struct extent_buffer *eb, int slot,
389 struct btrfs_key *key)
393 u64 saved_i_size = 0;
394 int save_old_i_size = 0;
395 unsigned long src_ptr;
396 unsigned long dst_ptr;
397 int overwrite_root = 0;
398 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
400 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
403 item_size = btrfs_item_size_nr(eb, slot);
404 src_ptr = btrfs_item_ptr_offset(eb, slot);
406 /* look for the key in the destination tree */
407 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
414 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
416 if (dst_size != item_size)
419 if (item_size == 0) {
420 btrfs_release_path(path);
423 dst_copy = kmalloc(item_size, GFP_NOFS);
424 src_copy = kmalloc(item_size, GFP_NOFS);
425 if (!dst_copy || !src_copy) {
426 btrfs_release_path(path);
432 read_extent_buffer(eb, src_copy, src_ptr, item_size);
434 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
435 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
437 ret = memcmp(dst_copy, src_copy, item_size);
442 * they have the same contents, just return, this saves
443 * us from cowing blocks in the destination tree and doing
444 * extra writes that may not have been done by a previous
448 btrfs_release_path(path);
453 * We need to load the old nbytes into the inode so when we
454 * replay the extents we've logged we get the right nbytes.
457 struct btrfs_inode_item *item;
461 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
462 struct btrfs_inode_item);
463 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
464 item = btrfs_item_ptr(eb, slot,
465 struct btrfs_inode_item);
466 btrfs_set_inode_nbytes(eb, item, nbytes);
469 * If this is a directory we need to reset the i_size to
470 * 0 so that we can set it up properly when replaying
471 * the rest of the items in this log.
473 mode = btrfs_inode_mode(eb, item);
475 btrfs_set_inode_size(eb, item, 0);
477 } else if (inode_item) {
478 struct btrfs_inode_item *item;
482 * New inode, set nbytes to 0 so that the nbytes comes out
483 * properly when we replay the extents.
485 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
486 btrfs_set_inode_nbytes(eb, item, 0);
489 * If this is a directory we need to reset the i_size to 0 so
490 * that we can set it up properly when replaying the rest of
491 * the items in this log.
493 mode = btrfs_inode_mode(eb, item);
495 btrfs_set_inode_size(eb, item, 0);
498 btrfs_release_path(path);
499 /* try to insert the key into the destination tree */
500 path->skip_release_on_error = 1;
501 ret = btrfs_insert_empty_item(trans, root, path,
503 path->skip_release_on_error = 0;
505 /* make sure any existing item is the correct size */
506 if (ret == -EEXIST || ret == -EOVERFLOW) {
508 found_size = btrfs_item_size_nr(path->nodes[0],
510 if (found_size > item_size)
511 btrfs_truncate_item(path, item_size, 1);
512 else if (found_size < item_size)
513 btrfs_extend_item(path, item_size - found_size);
517 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
520 /* don't overwrite an existing inode if the generation number
521 * was logged as zero. This is done when the tree logging code
522 * is just logging an inode to make sure it exists after recovery.
524 * Also, don't overwrite i_size on directories during replay.
525 * log replay inserts and removes directory items based on the
526 * state of the tree found in the subvolume, and i_size is modified
529 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
530 struct btrfs_inode_item *src_item;
531 struct btrfs_inode_item *dst_item;
533 src_item = (struct btrfs_inode_item *)src_ptr;
534 dst_item = (struct btrfs_inode_item *)dst_ptr;
536 if (btrfs_inode_generation(eb, src_item) == 0) {
537 struct extent_buffer *dst_eb = path->nodes[0];
538 const u64 ino_size = btrfs_inode_size(eb, src_item);
541 * For regular files an ino_size == 0 is used only when
542 * logging that an inode exists, as part of a directory
543 * fsync, and the inode wasn't fsynced before. In this
544 * case don't set the size of the inode in the fs/subvol
545 * tree, otherwise we would be throwing valid data away.
547 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
548 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
550 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
554 if (overwrite_root &&
555 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
556 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
558 saved_i_size = btrfs_inode_size(path->nodes[0],
563 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
566 if (save_old_i_size) {
567 struct btrfs_inode_item *dst_item;
568 dst_item = (struct btrfs_inode_item *)dst_ptr;
569 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
572 /* make sure the generation is filled in */
573 if (key->type == BTRFS_INODE_ITEM_KEY) {
574 struct btrfs_inode_item *dst_item;
575 dst_item = (struct btrfs_inode_item *)dst_ptr;
576 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
577 btrfs_set_inode_generation(path->nodes[0], dst_item,
582 btrfs_mark_buffer_dirty(path->nodes[0]);
583 btrfs_release_path(path);
588 * simple helper to read an inode off the disk from a given root
589 * This can only be called for subvolume roots and not for the log
591 static noinline struct inode *read_one_inode(struct btrfs_root *root,
596 inode = btrfs_iget(root->fs_info->sb, objectid, root);
602 /* replays a single extent in 'eb' at 'slot' with 'key' into the
603 * subvolume 'root'. path is released on entry and should be released
606 * extents in the log tree have not been allocated out of the extent
607 * tree yet. So, this completes the allocation, taking a reference
608 * as required if the extent already exists or creating a new extent
609 * if it isn't in the extent allocation tree yet.
611 * The extent is inserted into the file, dropping any existing extents
612 * from the file that overlap the new one.
614 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
615 struct btrfs_root *root,
616 struct btrfs_path *path,
617 struct extent_buffer *eb, int slot,
618 struct btrfs_key *key)
620 struct btrfs_drop_extents_args drop_args = { 0 };
621 struct btrfs_fs_info *fs_info = root->fs_info;
624 u64 start = key->offset;
626 struct btrfs_file_extent_item *item;
627 struct inode *inode = NULL;
631 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
632 found_type = btrfs_file_extent_type(eb, item);
634 if (found_type == BTRFS_FILE_EXTENT_REG ||
635 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
636 nbytes = btrfs_file_extent_num_bytes(eb, item);
637 extent_end = start + nbytes;
640 * We don't add to the inodes nbytes if we are prealloc or a
643 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
645 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
646 size = btrfs_file_extent_ram_bytes(eb, item);
647 nbytes = btrfs_file_extent_ram_bytes(eb, item);
648 extent_end = ALIGN(start + size,
649 fs_info->sectorsize);
655 inode = read_one_inode(root, key->objectid);
662 * first check to see if we already have this extent in the
663 * file. This must be done before the btrfs_drop_extents run
664 * so we don't try to drop this extent.
666 ret = btrfs_lookup_file_extent(trans, root, path,
667 btrfs_ino(BTRFS_I(inode)), start, 0);
670 (found_type == BTRFS_FILE_EXTENT_REG ||
671 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
672 struct btrfs_file_extent_item cmp1;
673 struct btrfs_file_extent_item cmp2;
674 struct btrfs_file_extent_item *existing;
675 struct extent_buffer *leaf;
677 leaf = path->nodes[0];
678 existing = btrfs_item_ptr(leaf, path->slots[0],
679 struct btrfs_file_extent_item);
681 read_extent_buffer(eb, &cmp1, (unsigned long)item,
683 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
687 * we already have a pointer to this exact extent,
688 * we don't have to do anything
690 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
691 btrfs_release_path(path);
695 btrfs_release_path(path);
697 /* drop any overlapping extents */
698 drop_args.start = start;
699 drop_args.end = extent_end;
700 drop_args.drop_cache = true;
701 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
705 if (found_type == BTRFS_FILE_EXTENT_REG ||
706 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
708 unsigned long dest_offset;
709 struct btrfs_key ins;
711 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
712 btrfs_fs_incompat(fs_info, NO_HOLES))
715 ret = btrfs_insert_empty_item(trans, root, path, key,
719 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
721 copy_extent_buffer(path->nodes[0], eb, dest_offset,
722 (unsigned long)item, sizeof(*item));
724 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
725 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
726 ins.type = BTRFS_EXTENT_ITEM_KEY;
727 offset = key->offset - btrfs_file_extent_offset(eb, item);
730 * Manually record dirty extent, as here we did a shallow
731 * file extent item copy and skip normal backref update,
732 * but modifying extent tree all by ourselves.
733 * So need to manually record dirty extent for qgroup,
734 * as the owner of the file extent changed from log tree
735 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
737 ret = btrfs_qgroup_trace_extent(trans,
738 btrfs_file_extent_disk_bytenr(eb, item),
739 btrfs_file_extent_disk_num_bytes(eb, item),
744 if (ins.objectid > 0) {
745 struct btrfs_ref ref = { 0 };
748 LIST_HEAD(ordered_sums);
751 * is this extent already allocated in the extent
752 * allocation tree? If so, just add a reference
754 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
757 btrfs_init_generic_ref(&ref,
758 BTRFS_ADD_DELAYED_REF,
759 ins.objectid, ins.offset, 0);
760 btrfs_init_data_ref(&ref,
761 root->root_key.objectid,
762 key->objectid, offset);
763 ret = btrfs_inc_extent_ref(trans, &ref);
768 * insert the extent pointer in the extent
771 ret = btrfs_alloc_logged_file_extent(trans,
772 root->root_key.objectid,
773 key->objectid, offset, &ins);
777 btrfs_release_path(path);
779 if (btrfs_file_extent_compression(eb, item)) {
780 csum_start = ins.objectid;
781 csum_end = csum_start + ins.offset;
783 csum_start = ins.objectid +
784 btrfs_file_extent_offset(eb, item);
785 csum_end = csum_start +
786 btrfs_file_extent_num_bytes(eb, item);
789 ret = btrfs_lookup_csums_range(root->log_root,
790 csum_start, csum_end - 1,
795 * Now delete all existing cums in the csum root that
796 * cover our range. We do this because we can have an
797 * extent that is completely referenced by one file
798 * extent item and partially referenced by another
799 * file extent item (like after using the clone or
800 * extent_same ioctls). In this case if we end up doing
801 * the replay of the one that partially references the
802 * extent first, and we do not do the csum deletion
803 * below, we can get 2 csum items in the csum tree that
804 * overlap each other. For example, imagine our log has
805 * the two following file extent items:
807 * key (257 EXTENT_DATA 409600)
808 * extent data disk byte 12845056 nr 102400
809 * extent data offset 20480 nr 20480 ram 102400
811 * key (257 EXTENT_DATA 819200)
812 * extent data disk byte 12845056 nr 102400
813 * extent data offset 0 nr 102400 ram 102400
815 * Where the second one fully references the 100K extent
816 * that starts at disk byte 12845056, and the log tree
817 * has a single csum item that covers the entire range
820 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
822 * After the first file extent item is replayed, the
823 * csum tree gets the following csum item:
825 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
827 * Which covers the 20K sub-range starting at offset 20K
828 * of our extent. Now when we replay the second file
829 * extent item, if we do not delete existing csum items
830 * that cover any of its blocks, we end up getting two
831 * csum items in our csum tree that overlap each other:
833 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
834 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
836 * Which is a problem, because after this anyone trying
837 * to lookup up for the checksum of any block of our
838 * extent starting at an offset of 40K or higher, will
839 * end up looking at the second csum item only, which
840 * does not contain the checksum for any block starting
841 * at offset 40K or higher of our extent.
843 while (!list_empty(&ordered_sums)) {
844 struct btrfs_ordered_sum *sums;
845 sums = list_entry(ordered_sums.next,
846 struct btrfs_ordered_sum,
849 ret = btrfs_del_csums(trans,
854 ret = btrfs_csum_file_blocks(trans,
855 fs_info->csum_root, sums);
856 list_del(&sums->list);
862 btrfs_release_path(path);
864 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
865 /* inline extents are easy, we just overwrite them */
866 ret = overwrite_item(trans, root, path, eb, slot, key);
871 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
877 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
878 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
886 * when cleaning up conflicts between the directory names in the
887 * subvolume, directory names in the log and directory names in the
888 * inode back references, we may have to unlink inodes from directories.
890 * This is a helper function to do the unlink of a specific directory
893 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
894 struct btrfs_root *root,
895 struct btrfs_path *path,
896 struct btrfs_inode *dir,
897 struct btrfs_dir_item *di)
902 struct extent_buffer *leaf;
903 struct btrfs_key location;
906 leaf = path->nodes[0];
908 btrfs_dir_item_key_to_cpu(leaf, di, &location);
909 name_len = btrfs_dir_name_len(leaf, di);
910 name = kmalloc(name_len, GFP_NOFS);
914 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
915 btrfs_release_path(path);
917 inode = read_one_inode(root, location.objectid);
923 ret = link_to_fixup_dir(trans, root, path, location.objectid);
927 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
932 ret = btrfs_run_delayed_items(trans);
940 * helper function to see if a given name and sequence number found
941 * in an inode back reference are already in a directory and correctly
942 * point to this inode
944 static noinline int inode_in_dir(struct btrfs_root *root,
945 struct btrfs_path *path,
946 u64 dirid, u64 objectid, u64 index,
947 const char *name, int name_len)
949 struct btrfs_dir_item *di;
950 struct btrfs_key location;
953 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
954 index, name, name_len, 0);
955 if (di && !IS_ERR(di)) {
956 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
957 if (location.objectid != objectid)
961 btrfs_release_path(path);
963 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
964 if (di && !IS_ERR(di)) {
965 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
966 if (location.objectid != objectid)
972 btrfs_release_path(path);
977 * helper function to check a log tree for a named back reference in
978 * an inode. This is used to decide if a back reference that is
979 * found in the subvolume conflicts with what we find in the log.
981 * inode backreferences may have multiple refs in a single item,
982 * during replay we process one reference at a time, and we don't
983 * want to delete valid links to a file from the subvolume if that
984 * link is also in the log.
986 static noinline int backref_in_log(struct btrfs_root *log,
987 struct btrfs_key *key,
989 const char *name, int namelen)
991 struct btrfs_path *path;
994 path = btrfs_alloc_path();
998 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1001 } else if (ret == 1) {
1006 if (key->type == BTRFS_INODE_EXTREF_KEY)
1007 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1012 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1016 btrfs_free_path(path);
1020 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1021 struct btrfs_root *root,
1022 struct btrfs_path *path,
1023 struct btrfs_root *log_root,
1024 struct btrfs_inode *dir,
1025 struct btrfs_inode *inode,
1026 u64 inode_objectid, u64 parent_objectid,
1027 u64 ref_index, char *name, int namelen,
1032 int victim_name_len;
1033 struct extent_buffer *leaf;
1034 struct btrfs_dir_item *di;
1035 struct btrfs_key search_key;
1036 struct btrfs_inode_extref *extref;
1039 /* Search old style refs */
1040 search_key.objectid = inode_objectid;
1041 search_key.type = BTRFS_INODE_REF_KEY;
1042 search_key.offset = parent_objectid;
1043 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1045 struct btrfs_inode_ref *victim_ref;
1047 unsigned long ptr_end;
1049 leaf = path->nodes[0];
1051 /* are we trying to overwrite a back ref for the root directory
1052 * if so, just jump out, we're done
1054 if (search_key.objectid == search_key.offset)
1057 /* check all the names in this back reference to see
1058 * if they are in the log. if so, we allow them to stay
1059 * otherwise they must be unlinked as a conflict
1061 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1062 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1063 while (ptr < ptr_end) {
1064 victim_ref = (struct btrfs_inode_ref *)ptr;
1065 victim_name_len = btrfs_inode_ref_name_len(leaf,
1067 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1071 read_extent_buffer(leaf, victim_name,
1072 (unsigned long)(victim_ref + 1),
1075 ret = backref_in_log(log_root, &search_key,
1076 parent_objectid, victim_name,
1082 inc_nlink(&inode->vfs_inode);
1083 btrfs_release_path(path);
1085 ret = btrfs_unlink_inode(trans, root, dir, inode,
1086 victim_name, victim_name_len);
1090 ret = btrfs_run_delayed_items(trans);
1098 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1102 * NOTE: we have searched root tree and checked the
1103 * corresponding ref, it does not need to check again.
1107 btrfs_release_path(path);
1109 /* Same search but for extended refs */
1110 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1111 inode_objectid, parent_objectid, 0,
1113 if (!IS_ERR_OR_NULL(extref)) {
1117 struct inode *victim_parent;
1119 leaf = path->nodes[0];
1121 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1122 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1124 while (cur_offset < item_size) {
1125 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1127 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1129 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1132 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1135 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1138 search_key.objectid = inode_objectid;
1139 search_key.type = BTRFS_INODE_EXTREF_KEY;
1140 search_key.offset = btrfs_extref_hash(parent_objectid,
1143 ret = backref_in_log(log_root, &search_key,
1144 parent_objectid, victim_name,
1150 victim_parent = read_one_inode(root,
1152 if (victim_parent) {
1153 inc_nlink(&inode->vfs_inode);
1154 btrfs_release_path(path);
1156 ret = btrfs_unlink_inode(trans, root,
1157 BTRFS_I(victim_parent),
1162 ret = btrfs_run_delayed_items(
1165 iput(victim_parent);
1174 cur_offset += victim_name_len + sizeof(*extref);
1178 btrfs_release_path(path);
1180 /* look for a conflicting sequence number */
1181 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1182 ref_index, name, namelen, 0);
1183 if (di && !IS_ERR(di)) {
1184 ret = drop_one_dir_item(trans, root, path, dir, di);
1188 btrfs_release_path(path);
1190 /* look for a conflicting name */
1191 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1193 if (di && !IS_ERR(di)) {
1194 ret = drop_one_dir_item(trans, root, path, dir, di);
1198 btrfs_release_path(path);
1203 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1204 u32 *namelen, char **name, u64 *index,
1205 u64 *parent_objectid)
1207 struct btrfs_inode_extref *extref;
1209 extref = (struct btrfs_inode_extref *)ref_ptr;
1211 *namelen = btrfs_inode_extref_name_len(eb, extref);
1212 *name = kmalloc(*namelen, GFP_NOFS);
1216 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1220 *index = btrfs_inode_extref_index(eb, extref);
1221 if (parent_objectid)
1222 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1227 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1228 u32 *namelen, char **name, u64 *index)
1230 struct btrfs_inode_ref *ref;
1232 ref = (struct btrfs_inode_ref *)ref_ptr;
1234 *namelen = btrfs_inode_ref_name_len(eb, ref);
1235 *name = kmalloc(*namelen, GFP_NOFS);
1239 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1242 *index = btrfs_inode_ref_index(eb, ref);
1248 * Take an inode reference item from the log tree and iterate all names from the
1249 * inode reference item in the subvolume tree with the same key (if it exists).
1250 * For any name that is not in the inode reference item from the log tree, do a
1251 * proper unlink of that name (that is, remove its entry from the inode
1252 * reference item and both dir index keys).
1254 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1255 struct btrfs_root *root,
1256 struct btrfs_path *path,
1257 struct btrfs_inode *inode,
1258 struct extent_buffer *log_eb,
1260 struct btrfs_key *key)
1263 unsigned long ref_ptr;
1264 unsigned long ref_end;
1265 struct extent_buffer *eb;
1268 btrfs_release_path(path);
1269 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1277 eb = path->nodes[0];
1278 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1279 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1280 while (ref_ptr < ref_end) {
1285 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1286 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1289 parent_id = key->offset;
1290 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1296 if (key->type == BTRFS_INODE_EXTREF_KEY)
1297 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1301 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1307 btrfs_release_path(path);
1308 dir = read_one_inode(root, parent_id);
1314 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1315 inode, name, namelen);
1325 if (key->type == BTRFS_INODE_EXTREF_KEY)
1326 ref_ptr += sizeof(struct btrfs_inode_extref);
1328 ref_ptr += sizeof(struct btrfs_inode_ref);
1332 btrfs_release_path(path);
1336 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1337 const u8 ref_type, const char *name,
1340 struct btrfs_key key;
1341 struct btrfs_path *path;
1342 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1345 path = btrfs_alloc_path();
1349 key.objectid = btrfs_ino(BTRFS_I(inode));
1350 key.type = ref_type;
1351 if (key.type == BTRFS_INODE_REF_KEY)
1352 key.offset = parent_id;
1354 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1356 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1363 if (key.type == BTRFS_INODE_EXTREF_KEY)
1364 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1365 path->slots[0], parent_id, name, namelen);
1367 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1371 btrfs_free_path(path);
1375 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1376 struct inode *dir, struct inode *inode, const char *name,
1377 int namelen, u64 ref_index)
1379 struct btrfs_dir_item *dir_item;
1380 struct btrfs_key key;
1381 struct btrfs_path *path;
1382 struct inode *other_inode = NULL;
1385 path = btrfs_alloc_path();
1389 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1390 btrfs_ino(BTRFS_I(dir)),
1393 btrfs_release_path(path);
1395 } else if (IS_ERR(dir_item)) {
1396 ret = PTR_ERR(dir_item);
1401 * Our inode's dentry collides with the dentry of another inode which is
1402 * in the log but not yet processed since it has a higher inode number.
1403 * So delete that other dentry.
1405 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1406 btrfs_release_path(path);
1407 other_inode = read_one_inode(root, key.objectid);
1412 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1417 * If we dropped the link count to 0, bump it so that later the iput()
1418 * on the inode will not free it. We will fixup the link count later.
1420 if (other_inode->i_nlink == 0)
1421 inc_nlink(other_inode);
1423 ret = btrfs_run_delayed_items(trans);
1427 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1428 name, namelen, 0, ref_index);
1431 btrfs_free_path(path);
1437 * replay one inode back reference item found in the log tree.
1438 * eb, slot and key refer to the buffer and key found in the log tree.
1439 * root is the destination we are replaying into, and path is for temp
1440 * use by this function. (it should be released on return).
1442 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1443 struct btrfs_root *root,
1444 struct btrfs_root *log,
1445 struct btrfs_path *path,
1446 struct extent_buffer *eb, int slot,
1447 struct btrfs_key *key)
1449 struct inode *dir = NULL;
1450 struct inode *inode = NULL;
1451 unsigned long ref_ptr;
1452 unsigned long ref_end;
1456 int search_done = 0;
1457 int log_ref_ver = 0;
1458 u64 parent_objectid;
1461 int ref_struct_size;
1463 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1464 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1466 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1467 struct btrfs_inode_extref *r;
1469 ref_struct_size = sizeof(struct btrfs_inode_extref);
1471 r = (struct btrfs_inode_extref *)ref_ptr;
1472 parent_objectid = btrfs_inode_extref_parent(eb, r);
1474 ref_struct_size = sizeof(struct btrfs_inode_ref);
1475 parent_objectid = key->offset;
1477 inode_objectid = key->objectid;
1480 * it is possible that we didn't log all the parent directories
1481 * for a given inode. If we don't find the dir, just don't
1482 * copy the back ref in. The link count fixup code will take
1485 dir = read_one_inode(root, parent_objectid);
1491 inode = read_one_inode(root, inode_objectid);
1497 while (ref_ptr < ref_end) {
1499 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1500 &ref_index, &parent_objectid);
1502 * parent object can change from one array
1506 dir = read_one_inode(root, parent_objectid);
1512 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1518 /* if we already have a perfect match, we're done */
1519 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1520 btrfs_ino(BTRFS_I(inode)), ref_index,
1523 * look for a conflicting back reference in the
1524 * metadata. if we find one we have to unlink that name
1525 * of the file before we add our new link. Later on, we
1526 * overwrite any existing back reference, and we don't
1527 * want to create dangling pointers in the directory.
1531 ret = __add_inode_ref(trans, root, path, log,
1536 ref_index, name, namelen,
1546 * If a reference item already exists for this inode
1547 * with the same parent and name, but different index,
1548 * drop it and the corresponding directory index entries
1549 * from the parent before adding the new reference item
1550 * and dir index entries, otherwise we would fail with
1551 * -EEXIST returned from btrfs_add_link() below.
1553 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1556 ret = btrfs_unlink_inode(trans, root,
1561 * If we dropped the link count to 0, bump it so
1562 * that later the iput() on the inode will not
1563 * free it. We will fixup the link count later.
1565 if (!ret && inode->i_nlink == 0)
1571 /* insert our name */
1572 ret = add_link(trans, root, dir, inode, name, namelen,
1577 btrfs_update_inode(trans, root, BTRFS_I(inode));
1580 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1590 * Before we overwrite the inode reference item in the subvolume tree
1591 * with the item from the log tree, we must unlink all names from the
1592 * parent directory that are in the subvolume's tree inode reference
1593 * item, otherwise we end up with an inconsistent subvolume tree where
1594 * dir index entries exist for a name but there is no inode reference
1595 * item with the same name.
1597 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1602 /* finally write the back reference in the inode */
1603 ret = overwrite_item(trans, root, path, eb, slot, key);
1605 btrfs_release_path(path);
1612 static int count_inode_extrefs(struct btrfs_root *root,
1613 struct btrfs_inode *inode, struct btrfs_path *path)
1617 unsigned int nlink = 0;
1620 u64 inode_objectid = btrfs_ino(inode);
1623 struct btrfs_inode_extref *extref;
1624 struct extent_buffer *leaf;
1627 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1632 leaf = path->nodes[0];
1633 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1634 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1637 while (cur_offset < item_size) {
1638 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1639 name_len = btrfs_inode_extref_name_len(leaf, extref);
1643 cur_offset += name_len + sizeof(*extref);
1647 btrfs_release_path(path);
1649 btrfs_release_path(path);
1651 if (ret < 0 && ret != -ENOENT)
1656 static int count_inode_refs(struct btrfs_root *root,
1657 struct btrfs_inode *inode, struct btrfs_path *path)
1660 struct btrfs_key key;
1661 unsigned int nlink = 0;
1663 unsigned long ptr_end;
1665 u64 ino = btrfs_ino(inode);
1668 key.type = BTRFS_INODE_REF_KEY;
1669 key.offset = (u64)-1;
1672 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1676 if (path->slots[0] == 0)
1681 btrfs_item_key_to_cpu(path->nodes[0], &key,
1683 if (key.objectid != ino ||
1684 key.type != BTRFS_INODE_REF_KEY)
1686 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1687 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1689 while (ptr < ptr_end) {
1690 struct btrfs_inode_ref *ref;
1692 ref = (struct btrfs_inode_ref *)ptr;
1693 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1695 ptr = (unsigned long)(ref + 1) + name_len;
1699 if (key.offset == 0)
1701 if (path->slots[0] > 0) {
1706 btrfs_release_path(path);
1708 btrfs_release_path(path);
1714 * There are a few corners where the link count of the file can't
1715 * be properly maintained during replay. So, instead of adding
1716 * lots of complexity to the log code, we just scan the backrefs
1717 * for any file that has been through replay.
1719 * The scan will update the link count on the inode to reflect the
1720 * number of back refs found. If it goes down to zero, the iput
1721 * will free the inode.
1723 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct inode *inode)
1727 struct btrfs_path *path;
1730 u64 ino = btrfs_ino(BTRFS_I(inode));
1732 path = btrfs_alloc_path();
1736 ret = count_inode_refs(root, BTRFS_I(inode), path);
1742 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1750 if (nlink != inode->i_nlink) {
1751 set_nlink(inode, nlink);
1752 btrfs_update_inode(trans, root, BTRFS_I(inode));
1754 BTRFS_I(inode)->index_cnt = (u64)-1;
1756 if (inode->i_nlink == 0) {
1757 if (S_ISDIR(inode->i_mode)) {
1758 ret = replay_dir_deletes(trans, root, NULL, path,
1763 ret = btrfs_insert_orphan_item(trans, root, ino);
1769 btrfs_free_path(path);
1773 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1774 struct btrfs_root *root,
1775 struct btrfs_path *path)
1778 struct btrfs_key key;
1779 struct inode *inode;
1781 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1782 key.type = BTRFS_ORPHAN_ITEM_KEY;
1783 key.offset = (u64)-1;
1785 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1790 if (path->slots[0] == 0)
1795 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1796 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1797 key.type != BTRFS_ORPHAN_ITEM_KEY)
1800 ret = btrfs_del_item(trans, root, path);
1804 btrfs_release_path(path);
1805 inode = read_one_inode(root, key.offset);
1809 ret = fixup_inode_link_count(trans, root, inode);
1815 * fixup on a directory may create new entries,
1816 * make sure we always look for the highset possible
1819 key.offset = (u64)-1;
1823 btrfs_release_path(path);
1829 * record a given inode in the fixup dir so we can check its link
1830 * count when replay is done. The link count is incremented here
1831 * so the inode won't go away until we check it
1833 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1834 struct btrfs_root *root,
1835 struct btrfs_path *path,
1838 struct btrfs_key key;
1840 struct inode *inode;
1842 inode = read_one_inode(root, objectid);
1846 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1847 key.type = BTRFS_ORPHAN_ITEM_KEY;
1848 key.offset = objectid;
1850 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1852 btrfs_release_path(path);
1854 if (!inode->i_nlink)
1855 set_nlink(inode, 1);
1858 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1859 } else if (ret == -EEXIST) {
1862 BUG(); /* Logic Error */
1870 * when replaying the log for a directory, we only insert names
1871 * for inodes that actually exist. This means an fsync on a directory
1872 * does not implicitly fsync all the new files in it
1874 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1875 struct btrfs_root *root,
1876 u64 dirid, u64 index,
1877 char *name, int name_len,
1878 struct btrfs_key *location)
1880 struct inode *inode;
1884 inode = read_one_inode(root, location->objectid);
1888 dir = read_one_inode(root, dirid);
1894 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1895 name_len, 1, index);
1897 /* FIXME, put inode into FIXUP list */
1905 * take a single entry in a log directory item and replay it into
1908 * if a conflicting item exists in the subdirectory already,
1909 * the inode it points to is unlinked and put into the link count
1912 * If a name from the log points to a file or directory that does
1913 * not exist in the FS, it is skipped. fsyncs on directories
1914 * do not force down inodes inside that directory, just changes to the
1915 * names or unlinks in a directory.
1917 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1918 * non-existing inode) and 1 if the name was replayed.
1920 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1921 struct btrfs_root *root,
1922 struct btrfs_path *path,
1923 struct extent_buffer *eb,
1924 struct btrfs_dir_item *di,
1925 struct btrfs_key *key)
1929 struct btrfs_dir_item *dst_di;
1930 struct btrfs_key found_key;
1931 struct btrfs_key log_key;
1936 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1937 bool name_added = false;
1939 dir = read_one_inode(root, key->objectid);
1943 name_len = btrfs_dir_name_len(eb, di);
1944 name = kmalloc(name_len, GFP_NOFS);
1950 log_type = btrfs_dir_type(eb, di);
1951 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1954 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1955 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1960 btrfs_release_path(path);
1962 if (key->type == BTRFS_DIR_ITEM_KEY) {
1963 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1965 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1966 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1975 if (IS_ERR_OR_NULL(dst_di)) {
1976 /* we need a sequence number to insert, so we only
1977 * do inserts for the BTRFS_DIR_INDEX_KEY types
1979 if (key->type != BTRFS_DIR_INDEX_KEY)
1984 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1985 /* the existing item matches the logged item */
1986 if (found_key.objectid == log_key.objectid &&
1987 found_key.type == log_key.type &&
1988 found_key.offset == log_key.offset &&
1989 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1990 update_size = false;
1995 * don't drop the conflicting directory entry if the inode
1996 * for the new entry doesn't exist
2001 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2005 if (key->type == BTRFS_DIR_INDEX_KEY)
2008 btrfs_release_path(path);
2009 if (!ret && update_size) {
2010 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2011 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
2015 if (!ret && name_added)
2021 * Check if the inode reference exists in the log for the given name,
2022 * inode and parent inode
2024 found_key.objectid = log_key.objectid;
2025 found_key.type = BTRFS_INODE_REF_KEY;
2026 found_key.offset = key->objectid;
2027 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2031 /* The dentry will be added later. */
2033 update_size = false;
2037 found_key.objectid = log_key.objectid;
2038 found_key.type = BTRFS_INODE_EXTREF_KEY;
2039 found_key.offset = key->objectid;
2040 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2045 /* The dentry will be added later. */
2047 update_size = false;
2050 btrfs_release_path(path);
2051 ret = insert_one_name(trans, root, key->objectid, key->offset,
2052 name, name_len, &log_key);
2053 if (ret && ret != -ENOENT && ret != -EEXIST)
2057 update_size = false;
2063 * find all the names in a directory item and reconcile them into
2064 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2065 * one name in a directory item, but the same code gets used for
2066 * both directory index types
2068 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2069 struct btrfs_root *root,
2070 struct btrfs_path *path,
2071 struct extent_buffer *eb, int slot,
2072 struct btrfs_key *key)
2075 u32 item_size = btrfs_item_size_nr(eb, slot);
2076 struct btrfs_dir_item *di;
2079 unsigned long ptr_end;
2080 struct btrfs_path *fixup_path = NULL;
2082 ptr = btrfs_item_ptr_offset(eb, slot);
2083 ptr_end = ptr + item_size;
2084 while (ptr < ptr_end) {
2085 di = (struct btrfs_dir_item *)ptr;
2086 name_len = btrfs_dir_name_len(eb, di);
2087 ret = replay_one_name(trans, root, path, eb, di, key);
2090 ptr = (unsigned long)(di + 1);
2094 * If this entry refers to a non-directory (directories can not
2095 * have a link count > 1) and it was added in the transaction
2096 * that was not committed, make sure we fixup the link count of
2097 * the inode it the entry points to. Otherwise something like
2098 * the following would result in a directory pointing to an
2099 * inode with a wrong link that does not account for this dir
2107 * ln testdir/bar testdir/bar_link
2108 * ln testdir/foo testdir/foo_link
2109 * xfs_io -c "fsync" testdir/bar
2113 * mount fs, log replay happens
2115 * File foo would remain with a link count of 1 when it has two
2116 * entries pointing to it in the directory testdir. This would
2117 * make it impossible to ever delete the parent directory has
2118 * it would result in stale dentries that can never be deleted.
2120 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2121 struct btrfs_key di_key;
2124 fixup_path = btrfs_alloc_path();
2131 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2132 ret = link_to_fixup_dir(trans, root, fixup_path,
2139 btrfs_free_path(fixup_path);
2144 * directory replay has two parts. There are the standard directory
2145 * items in the log copied from the subvolume, and range items
2146 * created in the log while the subvolume was logged.
2148 * The range items tell us which parts of the key space the log
2149 * is authoritative for. During replay, if a key in the subvolume
2150 * directory is in a logged range item, but not actually in the log
2151 * that means it was deleted from the directory before the fsync
2152 * and should be removed.
2154 static noinline int find_dir_range(struct btrfs_root *root,
2155 struct btrfs_path *path,
2156 u64 dirid, int key_type,
2157 u64 *start_ret, u64 *end_ret)
2159 struct btrfs_key key;
2161 struct btrfs_dir_log_item *item;
2165 if (*start_ret == (u64)-1)
2168 key.objectid = dirid;
2169 key.type = key_type;
2170 key.offset = *start_ret;
2172 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2176 if (path->slots[0] == 0)
2181 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2183 if (key.type != key_type || key.objectid != dirid) {
2187 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2188 struct btrfs_dir_log_item);
2189 found_end = btrfs_dir_log_end(path->nodes[0], item);
2191 if (*start_ret >= key.offset && *start_ret <= found_end) {
2193 *start_ret = key.offset;
2194 *end_ret = found_end;
2199 /* check the next slot in the tree to see if it is a valid item */
2200 nritems = btrfs_header_nritems(path->nodes[0]);
2202 if (path->slots[0] >= nritems) {
2203 ret = btrfs_next_leaf(root, path);
2208 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2210 if (key.type != key_type || key.objectid != dirid) {
2214 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2215 struct btrfs_dir_log_item);
2216 found_end = btrfs_dir_log_end(path->nodes[0], item);
2217 *start_ret = key.offset;
2218 *end_ret = found_end;
2221 btrfs_release_path(path);
2226 * this looks for a given directory item in the log. If the directory
2227 * item is not in the log, the item is removed and the inode it points
2230 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2231 struct btrfs_root *root,
2232 struct btrfs_root *log,
2233 struct btrfs_path *path,
2234 struct btrfs_path *log_path,
2236 struct btrfs_key *dir_key)
2239 struct extent_buffer *eb;
2242 struct btrfs_dir_item *di;
2243 struct btrfs_dir_item *log_di;
2246 unsigned long ptr_end;
2248 struct inode *inode;
2249 struct btrfs_key location;
2252 eb = path->nodes[0];
2253 slot = path->slots[0];
2254 item_size = btrfs_item_size_nr(eb, slot);
2255 ptr = btrfs_item_ptr_offset(eb, slot);
2256 ptr_end = ptr + item_size;
2257 while (ptr < ptr_end) {
2258 di = (struct btrfs_dir_item *)ptr;
2259 name_len = btrfs_dir_name_len(eb, di);
2260 name = kmalloc(name_len, GFP_NOFS);
2265 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2268 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2269 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2272 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2273 log_di = btrfs_lookup_dir_index_item(trans, log,
2279 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2280 btrfs_dir_item_key_to_cpu(eb, di, &location);
2281 btrfs_release_path(path);
2282 btrfs_release_path(log_path);
2283 inode = read_one_inode(root, location.objectid);
2289 ret = link_to_fixup_dir(trans, root,
2290 path, location.objectid);
2298 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2299 BTRFS_I(inode), name, name_len);
2301 ret = btrfs_run_delayed_items(trans);
2307 /* there might still be more names under this key
2308 * check and repeat if required
2310 ret = btrfs_search_slot(NULL, root, dir_key, path,
2316 } else if (IS_ERR(log_di)) {
2318 return PTR_ERR(log_di);
2320 btrfs_release_path(log_path);
2323 ptr = (unsigned long)(di + 1);
2328 btrfs_release_path(path);
2329 btrfs_release_path(log_path);
2333 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *root,
2335 struct btrfs_root *log,
2336 struct btrfs_path *path,
2339 struct btrfs_key search_key;
2340 struct btrfs_path *log_path;
2345 log_path = btrfs_alloc_path();
2349 search_key.objectid = ino;
2350 search_key.type = BTRFS_XATTR_ITEM_KEY;
2351 search_key.offset = 0;
2353 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2357 nritems = btrfs_header_nritems(path->nodes[0]);
2358 for (i = path->slots[0]; i < nritems; i++) {
2359 struct btrfs_key key;
2360 struct btrfs_dir_item *di;
2361 struct btrfs_dir_item *log_di;
2365 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2366 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2371 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2372 total_size = btrfs_item_size_nr(path->nodes[0], i);
2374 while (cur < total_size) {
2375 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2376 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2377 u32 this_len = sizeof(*di) + name_len + data_len;
2380 name = kmalloc(name_len, GFP_NOFS);
2385 read_extent_buffer(path->nodes[0], name,
2386 (unsigned long)(di + 1), name_len);
2388 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2390 btrfs_release_path(log_path);
2392 /* Doesn't exist in log tree, so delete it. */
2393 btrfs_release_path(path);
2394 di = btrfs_lookup_xattr(trans, root, path, ino,
2395 name, name_len, -1);
2402 ret = btrfs_delete_one_dir_name(trans, root,
2406 btrfs_release_path(path);
2411 if (IS_ERR(log_di)) {
2412 ret = PTR_ERR(log_di);
2416 di = (struct btrfs_dir_item *)((char *)di + this_len);
2419 ret = btrfs_next_leaf(root, path);
2425 btrfs_free_path(log_path);
2426 btrfs_release_path(path);
2432 * deletion replay happens before we copy any new directory items
2433 * out of the log or out of backreferences from inodes. It
2434 * scans the log to find ranges of keys that log is authoritative for,
2435 * and then scans the directory to find items in those ranges that are
2436 * not present in the log.
2438 * Anything we don't find in the log is unlinked and removed from the
2441 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2442 struct btrfs_root *root,
2443 struct btrfs_root *log,
2444 struct btrfs_path *path,
2445 u64 dirid, int del_all)
2449 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2451 struct btrfs_key dir_key;
2452 struct btrfs_key found_key;
2453 struct btrfs_path *log_path;
2456 dir_key.objectid = dirid;
2457 dir_key.type = BTRFS_DIR_ITEM_KEY;
2458 log_path = btrfs_alloc_path();
2462 dir = read_one_inode(root, dirid);
2463 /* it isn't an error if the inode isn't there, that can happen
2464 * because we replay the deletes before we copy in the inode item
2468 btrfs_free_path(log_path);
2476 range_end = (u64)-1;
2478 ret = find_dir_range(log, path, dirid, key_type,
2479 &range_start, &range_end);
2484 dir_key.offset = range_start;
2487 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2492 nritems = btrfs_header_nritems(path->nodes[0]);
2493 if (path->slots[0] >= nritems) {
2494 ret = btrfs_next_leaf(root, path);
2500 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2502 if (found_key.objectid != dirid ||
2503 found_key.type != dir_key.type)
2506 if (found_key.offset > range_end)
2509 ret = check_item_in_log(trans, root, log, path,
2514 if (found_key.offset == (u64)-1)
2516 dir_key.offset = found_key.offset + 1;
2518 btrfs_release_path(path);
2519 if (range_end == (u64)-1)
2521 range_start = range_end + 1;
2526 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2527 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2528 dir_key.type = BTRFS_DIR_INDEX_KEY;
2529 btrfs_release_path(path);
2533 btrfs_release_path(path);
2534 btrfs_free_path(log_path);
2540 * the process_func used to replay items from the log tree. This
2541 * gets called in two different stages. The first stage just looks
2542 * for inodes and makes sure they are all copied into the subvolume.
2544 * The second stage copies all the other item types from the log into
2545 * the subvolume. The two stage approach is slower, but gets rid of
2546 * lots of complexity around inodes referencing other inodes that exist
2547 * only in the log (references come from either directory items or inode
2550 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2551 struct walk_control *wc, u64 gen, int level)
2554 struct btrfs_path *path;
2555 struct btrfs_root *root = wc->replay_dest;
2556 struct btrfs_key key;
2560 ret = btrfs_read_buffer(eb, gen, level, NULL);
2564 level = btrfs_header_level(eb);
2569 path = btrfs_alloc_path();
2573 nritems = btrfs_header_nritems(eb);
2574 for (i = 0; i < nritems; i++) {
2575 btrfs_item_key_to_cpu(eb, &key, i);
2577 /* inode keys are done during the first stage */
2578 if (key.type == BTRFS_INODE_ITEM_KEY &&
2579 wc->stage == LOG_WALK_REPLAY_INODES) {
2580 struct btrfs_inode_item *inode_item;
2583 inode_item = btrfs_item_ptr(eb, i,
2584 struct btrfs_inode_item);
2586 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2587 * and never got linked before the fsync, skip it, as
2588 * replaying it is pointless since it would be deleted
2589 * later. We skip logging tmpfiles, but it's always
2590 * possible we are replaying a log created with a kernel
2591 * that used to log tmpfiles.
2593 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2594 wc->ignore_cur_inode = true;
2597 wc->ignore_cur_inode = false;
2599 ret = replay_xattr_deletes(wc->trans, root, log,
2600 path, key.objectid);
2603 mode = btrfs_inode_mode(eb, inode_item);
2604 if (S_ISDIR(mode)) {
2605 ret = replay_dir_deletes(wc->trans,
2606 root, log, path, key.objectid, 0);
2610 ret = overwrite_item(wc->trans, root, path,
2616 * Before replaying extents, truncate the inode to its
2617 * size. We need to do it now and not after log replay
2618 * because before an fsync we can have prealloc extents
2619 * added beyond the inode's i_size. If we did it after,
2620 * through orphan cleanup for example, we would drop
2621 * those prealloc extents just after replaying them.
2623 if (S_ISREG(mode)) {
2624 struct btrfs_drop_extents_args drop_args = { 0 };
2625 struct inode *inode;
2628 inode = read_one_inode(root, key.objectid);
2633 from = ALIGN(i_size_read(inode),
2634 root->fs_info->sectorsize);
2635 drop_args.start = from;
2636 drop_args.end = (u64)-1;
2637 drop_args.drop_cache = true;
2638 ret = btrfs_drop_extents(wc->trans, root,
2642 inode_sub_bytes(inode,
2643 drop_args.bytes_found);
2644 /* Update the inode's nbytes. */
2645 ret = btrfs_update_inode(wc->trans,
2646 root, BTRFS_I(inode));
2653 ret = link_to_fixup_dir(wc->trans, root,
2654 path, key.objectid);
2659 if (wc->ignore_cur_inode)
2662 if (key.type == BTRFS_DIR_INDEX_KEY &&
2663 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2664 ret = replay_one_dir_item(wc->trans, root, path,
2670 if (wc->stage < LOG_WALK_REPLAY_ALL)
2673 /* these keys are simply copied */
2674 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2675 ret = overwrite_item(wc->trans, root, path,
2679 } else if (key.type == BTRFS_INODE_REF_KEY ||
2680 key.type == BTRFS_INODE_EXTREF_KEY) {
2681 ret = add_inode_ref(wc->trans, root, log, path,
2683 if (ret && ret != -ENOENT)
2686 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2687 ret = replay_one_extent(wc->trans, root, path,
2691 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2692 ret = replay_one_dir_item(wc->trans, root, path,
2698 btrfs_free_path(path);
2703 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2705 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2707 struct btrfs_block_group *cache;
2709 cache = btrfs_lookup_block_group(fs_info, start);
2711 btrfs_err(fs_info, "unable to find block group for %llu", start);
2715 spin_lock(&cache->space_info->lock);
2716 spin_lock(&cache->lock);
2717 cache->reserved -= fs_info->nodesize;
2718 cache->space_info->bytes_reserved -= fs_info->nodesize;
2719 spin_unlock(&cache->lock);
2720 spin_unlock(&cache->space_info->lock);
2722 btrfs_put_block_group(cache);
2725 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2726 struct btrfs_root *root,
2727 struct btrfs_path *path, int *level,
2728 struct walk_control *wc)
2730 struct btrfs_fs_info *fs_info = root->fs_info;
2733 struct extent_buffer *next;
2734 struct extent_buffer *cur;
2738 while (*level > 0) {
2739 struct btrfs_key first_key;
2741 cur = path->nodes[*level];
2743 WARN_ON(btrfs_header_level(cur) != *level);
2745 if (path->slots[*level] >=
2746 btrfs_header_nritems(cur))
2749 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2750 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2751 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2752 blocksize = fs_info->nodesize;
2754 next = btrfs_find_create_tree_block(fs_info, bytenr,
2755 btrfs_header_owner(cur),
2758 return PTR_ERR(next);
2761 ret = wc->process_func(root, next, wc, ptr_gen,
2764 free_extent_buffer(next);
2768 path->slots[*level]++;
2770 ret = btrfs_read_buffer(next, ptr_gen,
2771 *level - 1, &first_key);
2773 free_extent_buffer(next);
2778 btrfs_tree_lock(next);
2779 btrfs_clean_tree_block(next);
2780 btrfs_wait_tree_block_writeback(next);
2781 btrfs_tree_unlock(next);
2782 ret = btrfs_pin_reserved_extent(trans,
2785 free_extent_buffer(next);
2788 btrfs_redirty_list_add(
2789 trans->transaction, next);
2791 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2792 clear_extent_buffer_dirty(next);
2793 unaccount_log_buffer(fs_info, bytenr);
2796 free_extent_buffer(next);
2799 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2801 free_extent_buffer(next);
2805 if (path->nodes[*level-1])
2806 free_extent_buffer(path->nodes[*level-1]);
2807 path->nodes[*level-1] = next;
2808 *level = btrfs_header_level(next);
2809 path->slots[*level] = 0;
2812 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2818 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2819 struct btrfs_root *root,
2820 struct btrfs_path *path, int *level,
2821 struct walk_control *wc)
2823 struct btrfs_fs_info *fs_info = root->fs_info;
2828 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2829 slot = path->slots[i];
2830 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2833 WARN_ON(*level == 0);
2836 ret = wc->process_func(root, path->nodes[*level], wc,
2837 btrfs_header_generation(path->nodes[*level]),
2843 struct extent_buffer *next;
2845 next = path->nodes[*level];
2848 btrfs_tree_lock(next);
2849 btrfs_clean_tree_block(next);
2850 btrfs_wait_tree_block_writeback(next);
2851 btrfs_tree_unlock(next);
2852 ret = btrfs_pin_reserved_extent(trans,
2853 path->nodes[*level]->start,
2854 path->nodes[*level]->len);
2858 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2859 clear_extent_buffer_dirty(next);
2861 unaccount_log_buffer(fs_info,
2862 path->nodes[*level]->start);
2865 free_extent_buffer(path->nodes[*level]);
2866 path->nodes[*level] = NULL;
2874 * drop the reference count on the tree rooted at 'snap'. This traverses
2875 * the tree freeing any blocks that have a ref count of zero after being
2878 static int walk_log_tree(struct btrfs_trans_handle *trans,
2879 struct btrfs_root *log, struct walk_control *wc)
2881 struct btrfs_fs_info *fs_info = log->fs_info;
2885 struct btrfs_path *path;
2888 path = btrfs_alloc_path();
2892 level = btrfs_header_level(log->node);
2894 path->nodes[level] = log->node;
2895 atomic_inc(&log->node->refs);
2896 path->slots[level] = 0;
2899 wret = walk_down_log_tree(trans, log, path, &level, wc);
2907 wret = walk_up_log_tree(trans, log, path, &level, wc);
2916 /* was the root node processed? if not, catch it here */
2917 if (path->nodes[orig_level]) {
2918 ret = wc->process_func(log, path->nodes[orig_level], wc,
2919 btrfs_header_generation(path->nodes[orig_level]),
2924 struct extent_buffer *next;
2926 next = path->nodes[orig_level];
2929 btrfs_tree_lock(next);
2930 btrfs_clean_tree_block(next);
2931 btrfs_wait_tree_block_writeback(next);
2932 btrfs_tree_unlock(next);
2933 ret = btrfs_pin_reserved_extent(trans,
2934 next->start, next->len);
2938 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2939 clear_extent_buffer_dirty(next);
2940 unaccount_log_buffer(fs_info, next->start);
2946 btrfs_free_path(path);
2951 * helper function to update the item for a given subvolumes log root
2952 * in the tree of log roots
2954 static int update_log_root(struct btrfs_trans_handle *trans,
2955 struct btrfs_root *log,
2956 struct btrfs_root_item *root_item)
2958 struct btrfs_fs_info *fs_info = log->fs_info;
2961 if (log->log_transid == 1) {
2962 /* insert root item on the first sync */
2963 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2964 &log->root_key, root_item);
2966 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2967 &log->root_key, root_item);
2972 static void wait_log_commit(struct btrfs_root *root, int transid)
2975 int index = transid % 2;
2978 * we only allow two pending log transactions at a time,
2979 * so we know that if ours is more than 2 older than the
2980 * current transaction, we're done
2983 prepare_to_wait(&root->log_commit_wait[index],
2984 &wait, TASK_UNINTERRUPTIBLE);
2986 if (!(root->log_transid_committed < transid &&
2987 atomic_read(&root->log_commit[index])))
2990 mutex_unlock(&root->log_mutex);
2992 mutex_lock(&root->log_mutex);
2994 finish_wait(&root->log_commit_wait[index], &wait);
2997 static void wait_for_writer(struct btrfs_root *root)
3002 prepare_to_wait(&root->log_writer_wait, &wait,
3003 TASK_UNINTERRUPTIBLE);
3004 if (!atomic_read(&root->log_writers))
3007 mutex_unlock(&root->log_mutex);
3009 mutex_lock(&root->log_mutex);
3011 finish_wait(&root->log_writer_wait, &wait);
3014 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3015 struct btrfs_log_ctx *ctx)
3020 mutex_lock(&root->log_mutex);
3021 list_del_init(&ctx->list);
3022 mutex_unlock(&root->log_mutex);
3026 * Invoked in log mutex context, or be sure there is no other task which
3027 * can access the list.
3029 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3030 int index, int error)
3032 struct btrfs_log_ctx *ctx;
3033 struct btrfs_log_ctx *safe;
3035 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3036 list_del_init(&ctx->list);
3037 ctx->log_ret = error;
3040 INIT_LIST_HEAD(&root->log_ctxs[index]);
3044 * btrfs_sync_log does sends a given tree log down to the disk and
3045 * updates the super blocks to record it. When this call is done,
3046 * you know that any inodes previously logged are safely on disk only
3049 * Any other return value means you need to call btrfs_commit_transaction.
3050 * Some of the edge cases for fsyncing directories that have had unlinks
3051 * or renames done in the past mean that sometimes the only safe
3052 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3053 * that has happened.
3055 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3056 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3062 struct btrfs_fs_info *fs_info = root->fs_info;
3063 struct btrfs_root *log = root->log_root;
3064 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3065 struct btrfs_root_item new_root_item;
3066 int log_transid = 0;
3067 struct btrfs_log_ctx root_log_ctx;
3068 struct blk_plug plug;
3072 mutex_lock(&root->log_mutex);
3073 log_transid = ctx->log_transid;
3074 if (root->log_transid_committed >= log_transid) {
3075 mutex_unlock(&root->log_mutex);
3076 return ctx->log_ret;
3079 index1 = log_transid % 2;
3080 if (atomic_read(&root->log_commit[index1])) {
3081 wait_log_commit(root, log_transid);
3082 mutex_unlock(&root->log_mutex);
3083 return ctx->log_ret;
3085 ASSERT(log_transid == root->log_transid);
3086 atomic_set(&root->log_commit[index1], 1);
3088 /* wait for previous tree log sync to complete */
3089 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3090 wait_log_commit(root, log_transid - 1);
3093 int batch = atomic_read(&root->log_batch);
3094 /* when we're on an ssd, just kick the log commit out */
3095 if (!btrfs_test_opt(fs_info, SSD) &&
3096 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3097 mutex_unlock(&root->log_mutex);
3098 schedule_timeout_uninterruptible(1);
3099 mutex_lock(&root->log_mutex);
3101 wait_for_writer(root);
3102 if (batch == atomic_read(&root->log_batch))
3106 /* bail out if we need to do a full commit */
3107 if (btrfs_need_log_full_commit(trans)) {
3109 mutex_unlock(&root->log_mutex);
3113 if (log_transid % 2 == 0)
3114 mark = EXTENT_DIRTY;
3118 /* we start IO on all the marked extents here, but we don't actually
3119 * wait for them until later.
3121 blk_start_plug(&plug);
3122 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3124 * -EAGAIN happens when someone, e.g., a concurrent transaction
3125 * commit, writes a dirty extent in this tree-log commit. This
3126 * concurrent write will create a hole writing out the extents,
3127 * and we cannot proceed on a zoned filesystem, requiring
3128 * sequential writing. While we can bail out to a full commit
3129 * here, but we can continue hoping the concurrent writing fills
3132 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3135 blk_finish_plug(&plug);
3136 btrfs_abort_transaction(trans, ret);
3137 btrfs_set_log_full_commit(trans);
3138 mutex_unlock(&root->log_mutex);
3143 * We _must_ update under the root->log_mutex in order to make sure we
3144 * have a consistent view of the log root we are trying to commit at
3147 * We _must_ copy this into a local copy, because we are not holding the
3148 * log_root_tree->log_mutex yet. This is important because when we
3149 * commit the log_root_tree we must have a consistent view of the
3150 * log_root_tree when we update the super block to point at the
3151 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3152 * with the commit and possibly point at the new block which we may not
3155 btrfs_set_root_node(&log->root_item, log->node);
3156 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3158 root->log_transid++;
3159 log->log_transid = root->log_transid;
3160 root->log_start_pid = 0;
3162 * IO has been started, blocks of the log tree have WRITTEN flag set
3163 * in their headers. new modifications of the log will be written to
3164 * new positions. so it's safe to allow log writers to go in.
3166 mutex_unlock(&root->log_mutex);
3168 btrfs_init_log_ctx(&root_log_ctx, NULL);
3170 mutex_lock(&log_root_tree->log_mutex);
3172 if (btrfs_is_zoned(fs_info)) {
3173 if (!log_root_tree->node) {
3174 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3176 mutex_unlock(&log_root_tree->log_mutex);
3182 index2 = log_root_tree->log_transid % 2;
3183 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3184 root_log_ctx.log_transid = log_root_tree->log_transid;
3187 * Now we are safe to update the log_root_tree because we're under the
3188 * log_mutex, and we're a current writer so we're holding the commit
3189 * open until we drop the log_mutex.
3191 ret = update_log_root(trans, log, &new_root_item);
3193 if (!list_empty(&root_log_ctx.list))
3194 list_del_init(&root_log_ctx.list);
3196 blk_finish_plug(&plug);
3197 btrfs_set_log_full_commit(trans);
3199 if (ret != -ENOSPC) {
3200 btrfs_abort_transaction(trans, ret);
3201 mutex_unlock(&log_root_tree->log_mutex);
3204 btrfs_wait_tree_log_extents(log, mark);
3205 mutex_unlock(&log_root_tree->log_mutex);
3210 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3211 blk_finish_plug(&plug);
3212 list_del_init(&root_log_ctx.list);
3213 mutex_unlock(&log_root_tree->log_mutex);
3214 ret = root_log_ctx.log_ret;
3218 index2 = root_log_ctx.log_transid % 2;
3219 if (atomic_read(&log_root_tree->log_commit[index2])) {
3220 blk_finish_plug(&plug);
3221 ret = btrfs_wait_tree_log_extents(log, mark);
3222 wait_log_commit(log_root_tree,
3223 root_log_ctx.log_transid);
3224 mutex_unlock(&log_root_tree->log_mutex);
3226 ret = root_log_ctx.log_ret;
3229 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3230 atomic_set(&log_root_tree->log_commit[index2], 1);
3232 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3233 wait_log_commit(log_root_tree,
3234 root_log_ctx.log_transid - 1);
3238 * now that we've moved on to the tree of log tree roots,
3239 * check the full commit flag again
3241 if (btrfs_need_log_full_commit(trans)) {
3242 blk_finish_plug(&plug);
3243 btrfs_wait_tree_log_extents(log, mark);
3244 mutex_unlock(&log_root_tree->log_mutex);
3246 goto out_wake_log_root;
3249 ret = btrfs_write_marked_extents(fs_info,
3250 &log_root_tree->dirty_log_pages,
3251 EXTENT_DIRTY | EXTENT_NEW);
3252 blk_finish_plug(&plug);
3254 * As described above, -EAGAIN indicates a hole in the extents. We
3255 * cannot wait for these write outs since the waiting cause a
3256 * deadlock. Bail out to the full commit instead.
3258 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3259 btrfs_set_log_full_commit(trans);
3260 btrfs_wait_tree_log_extents(log, mark);
3261 mutex_unlock(&log_root_tree->log_mutex);
3262 goto out_wake_log_root;
3264 btrfs_set_log_full_commit(trans);
3265 btrfs_abort_transaction(trans, ret);
3266 mutex_unlock(&log_root_tree->log_mutex);
3267 goto out_wake_log_root;
3269 ret = btrfs_wait_tree_log_extents(log, mark);
3271 ret = btrfs_wait_tree_log_extents(log_root_tree,
3272 EXTENT_NEW | EXTENT_DIRTY);
3274 btrfs_set_log_full_commit(trans);
3275 mutex_unlock(&log_root_tree->log_mutex);
3276 goto out_wake_log_root;
3279 log_root_start = log_root_tree->node->start;
3280 log_root_level = btrfs_header_level(log_root_tree->node);
3281 log_root_tree->log_transid++;
3282 mutex_unlock(&log_root_tree->log_mutex);
3285 * Here we are guaranteed that nobody is going to write the superblock
3286 * for the current transaction before us and that neither we do write
3287 * our superblock before the previous transaction finishes its commit
3288 * and writes its superblock, because:
3290 * 1) We are holding a handle on the current transaction, so no body
3291 * can commit it until we release the handle;
3293 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3294 * if the previous transaction is still committing, and hasn't yet
3295 * written its superblock, we wait for it to do it, because a
3296 * transaction commit acquires the tree_log_mutex when the commit
3297 * begins and releases it only after writing its superblock.
3299 mutex_lock(&fs_info->tree_log_mutex);
3300 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3301 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3302 ret = write_all_supers(fs_info, 1);
3303 mutex_unlock(&fs_info->tree_log_mutex);
3305 btrfs_set_log_full_commit(trans);
3306 btrfs_abort_transaction(trans, ret);
3307 goto out_wake_log_root;
3310 mutex_lock(&root->log_mutex);
3311 if (root->last_log_commit < log_transid)
3312 root->last_log_commit = log_transid;
3313 mutex_unlock(&root->log_mutex);
3316 mutex_lock(&log_root_tree->log_mutex);
3317 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3319 log_root_tree->log_transid_committed++;
3320 atomic_set(&log_root_tree->log_commit[index2], 0);
3321 mutex_unlock(&log_root_tree->log_mutex);
3324 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3325 * all the updates above are seen by the woken threads. It might not be
3326 * necessary, but proving that seems to be hard.
3328 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3330 mutex_lock(&root->log_mutex);
3331 btrfs_remove_all_log_ctxs(root, index1, ret);
3332 root->log_transid_committed++;
3333 atomic_set(&root->log_commit[index1], 0);
3334 mutex_unlock(&root->log_mutex);
3337 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3338 * all the updates above are seen by the woken threads. It might not be
3339 * necessary, but proving that seems to be hard.
3341 cond_wake_up(&root->log_commit_wait[index1]);
3345 static void free_log_tree(struct btrfs_trans_handle *trans,
3346 struct btrfs_root *log)
3349 struct walk_control wc = {
3351 .process_func = process_one_buffer
3355 ret = walk_log_tree(trans, log, &wc);
3358 btrfs_abort_transaction(trans, ret);
3360 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3364 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3365 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3366 extent_io_tree_release(&log->log_csum_range);
3368 if (trans && log->node)
3369 btrfs_redirty_list_add(trans->transaction, log->node);
3370 btrfs_put_root(log);
3374 * free all the extents used by the tree log. This should be called
3375 * at commit time of the full transaction
3377 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3379 if (root->log_root) {
3380 free_log_tree(trans, root->log_root);
3381 root->log_root = NULL;
3382 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3387 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3388 struct btrfs_fs_info *fs_info)
3390 if (fs_info->log_root_tree) {
3391 free_log_tree(trans, fs_info->log_root_tree);
3392 fs_info->log_root_tree = NULL;
3393 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3399 * Check if an inode was logged in the current transaction. We can't always rely
3400 * on an inode's logged_trans value, because it's an in-memory only field and
3401 * therefore not persisted. This means that its value is lost if the inode gets
3402 * evicted and loaded again from disk (in which case it has a value of 0, and
3403 * certainly it is smaller then any possible transaction ID), when that happens
3404 * the full_sync flag is set in the inode's runtime flags, so on that case we
3405 * assume eviction happened and ignore the logged_trans value, assuming the
3406 * worst case, that the inode was logged before in the current transaction.
3408 static bool inode_logged(struct btrfs_trans_handle *trans,
3409 struct btrfs_inode *inode)
3411 if (inode->logged_trans == trans->transid)
3414 if (inode->last_trans == trans->transid &&
3415 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3416 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3423 * If both a file and directory are logged, and unlinks or renames are
3424 * mixed in, we have a few interesting corners:
3426 * create file X in dir Y
3427 * link file X to X.link in dir Y
3429 * unlink file X but leave X.link
3432 * After a crash we would expect only X.link to exist. But file X
3433 * didn't get fsync'd again so the log has back refs for X and X.link.
3435 * We solve this by removing directory entries and inode backrefs from the
3436 * log when a file that was logged in the current transaction is
3437 * unlinked. Any later fsync will include the updated log entries, and
3438 * we'll be able to reconstruct the proper directory items from backrefs.
3440 * This optimizations allows us to avoid relogging the entire inode
3441 * or the entire directory.
3443 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3444 struct btrfs_root *root,
3445 const char *name, int name_len,
3446 struct btrfs_inode *dir, u64 index)
3448 struct btrfs_root *log;
3449 struct btrfs_dir_item *di;
3450 struct btrfs_path *path;
3453 u64 dir_ino = btrfs_ino(dir);
3455 if (!inode_logged(trans, dir))
3458 ret = join_running_log_trans(root);
3462 mutex_lock(&dir->log_mutex);
3464 log = root->log_root;
3465 path = btrfs_alloc_path();
3471 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3472 name, name_len, -1);
3478 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3484 btrfs_release_path(path);
3485 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3486 index, name, name_len, -1);
3492 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3500 * We do not need to update the size field of the directory's inode item
3501 * because on log replay we update the field to reflect all existing
3502 * entries in the directory (see overwrite_item()).
3505 btrfs_free_path(path);
3507 mutex_unlock(&dir->log_mutex);
3508 if (err == -ENOSPC) {
3509 btrfs_set_log_full_commit(trans);
3511 } else if (err < 0 && err != -ENOENT) {
3512 /* ENOENT can be returned if the entry hasn't been fsynced yet */
3513 btrfs_abort_transaction(trans, err);
3516 btrfs_end_log_trans(root);
3521 /* see comments for btrfs_del_dir_entries_in_log */
3522 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3523 struct btrfs_root *root,
3524 const char *name, int name_len,
3525 struct btrfs_inode *inode, u64 dirid)
3527 struct btrfs_root *log;
3531 if (!inode_logged(trans, inode))
3534 ret = join_running_log_trans(root);
3537 log = root->log_root;
3538 mutex_lock(&inode->log_mutex);
3540 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3542 mutex_unlock(&inode->log_mutex);
3543 if (ret == -ENOSPC) {
3544 btrfs_set_log_full_commit(trans);
3546 } else if (ret < 0 && ret != -ENOENT)
3547 btrfs_abort_transaction(trans, ret);
3548 btrfs_end_log_trans(root);
3554 * creates a range item in the log for 'dirid'. first_offset and
3555 * last_offset tell us which parts of the key space the log should
3556 * be considered authoritative for.
3558 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3559 struct btrfs_root *log,
3560 struct btrfs_path *path,
3561 int key_type, u64 dirid,
3562 u64 first_offset, u64 last_offset)
3565 struct btrfs_key key;
3566 struct btrfs_dir_log_item *item;
3568 key.objectid = dirid;
3569 key.offset = first_offset;
3570 if (key_type == BTRFS_DIR_ITEM_KEY)
3571 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3573 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3574 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3578 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3579 struct btrfs_dir_log_item);
3580 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3581 btrfs_mark_buffer_dirty(path->nodes[0]);
3582 btrfs_release_path(path);
3587 * log all the items included in the current transaction for a given
3588 * directory. This also creates the range items in the log tree required
3589 * to replay anything deleted before the fsync
3591 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3592 struct btrfs_root *root, struct btrfs_inode *inode,
3593 struct btrfs_path *path,
3594 struct btrfs_path *dst_path, int key_type,
3595 struct btrfs_log_ctx *ctx,
3596 u64 min_offset, u64 *last_offset_ret)
3598 struct btrfs_key min_key;
3599 struct btrfs_root *log = root->log_root;
3600 struct extent_buffer *src;
3605 u64 first_offset = min_offset;
3606 u64 last_offset = (u64)-1;
3607 u64 ino = btrfs_ino(inode);
3609 log = root->log_root;
3611 min_key.objectid = ino;
3612 min_key.type = key_type;
3613 min_key.offset = min_offset;
3615 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3618 * we didn't find anything from this transaction, see if there
3619 * is anything at all
3621 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3622 min_key.objectid = ino;
3623 min_key.type = key_type;
3624 min_key.offset = (u64)-1;
3625 btrfs_release_path(path);
3626 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3628 btrfs_release_path(path);
3631 ret = btrfs_previous_item(root, path, ino, key_type);
3633 /* if ret == 0 there are items for this type,
3634 * create a range to tell us the last key of this type.
3635 * otherwise, there are no items in this directory after
3636 * *min_offset, and we create a range to indicate that.
3639 struct btrfs_key tmp;
3640 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3642 if (key_type == tmp.type)
3643 first_offset = max(min_offset, tmp.offset) + 1;
3648 /* go backward to find any previous key */
3649 ret = btrfs_previous_item(root, path, ino, key_type);
3651 struct btrfs_key tmp;
3652 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3653 if (key_type == tmp.type) {
3654 first_offset = tmp.offset;
3655 ret = overwrite_item(trans, log, dst_path,
3656 path->nodes[0], path->slots[0],
3664 btrfs_release_path(path);
3667 * Find the first key from this transaction again. See the note for
3668 * log_new_dir_dentries, if we're logging a directory recursively we
3669 * won't be holding its i_mutex, which means we can modify the directory
3670 * while we're logging it. If we remove an entry between our first
3671 * search and this search we'll not find the key again and can just
3675 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3680 * we have a block from this transaction, log every item in it
3681 * from our directory
3684 struct btrfs_key tmp;
3685 src = path->nodes[0];
3686 nritems = btrfs_header_nritems(src);
3687 for (i = path->slots[0]; i < nritems; i++) {
3688 struct btrfs_dir_item *di;
3690 btrfs_item_key_to_cpu(src, &min_key, i);
3692 if (min_key.objectid != ino || min_key.type != key_type)
3695 if (need_resched()) {
3696 btrfs_release_path(path);
3701 ret = overwrite_item(trans, log, dst_path, src, i,
3709 * We must make sure that when we log a directory entry,
3710 * the corresponding inode, after log replay, has a
3711 * matching link count. For example:
3717 * xfs_io -c "fsync" mydir
3719 * <mount fs and log replay>
3721 * Would result in a fsync log that when replayed, our
3722 * file inode would have a link count of 1, but we get
3723 * two directory entries pointing to the same inode.
3724 * After removing one of the names, it would not be
3725 * possible to remove the other name, which resulted
3726 * always in stale file handle errors, and would not
3727 * be possible to rmdir the parent directory, since
3728 * its i_size could never decrement to the value
3729 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3731 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3732 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3734 (btrfs_dir_transid(src, di) == trans->transid ||
3735 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3736 tmp.type != BTRFS_ROOT_ITEM_KEY)
3737 ctx->log_new_dentries = true;
3739 path->slots[0] = nritems;
3742 * look ahead to the next item and see if it is also
3743 * from this directory and from this transaction
3745 ret = btrfs_next_leaf(root, path);
3748 last_offset = (u64)-1;
3753 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3754 if (tmp.objectid != ino || tmp.type != key_type) {
3755 last_offset = (u64)-1;
3758 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3759 ret = overwrite_item(trans, log, dst_path,
3760 path->nodes[0], path->slots[0],
3765 last_offset = tmp.offset;
3770 btrfs_release_path(path);
3771 btrfs_release_path(dst_path);
3774 *last_offset_ret = last_offset;
3776 * insert the log range keys to indicate where the log
3779 ret = insert_dir_log_key(trans, log, path, key_type,
3780 ino, first_offset, last_offset);
3788 * logging directories is very similar to logging inodes, We find all the items
3789 * from the current transaction and write them to the log.
3791 * The recovery code scans the directory in the subvolume, and if it finds a
3792 * key in the range logged that is not present in the log tree, then it means
3793 * that dir entry was unlinked during the transaction.
3795 * In order for that scan to work, we must include one key smaller than
3796 * the smallest logged by this transaction and one key larger than the largest
3797 * key logged by this transaction.
3799 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3800 struct btrfs_root *root, struct btrfs_inode *inode,
3801 struct btrfs_path *path,
3802 struct btrfs_path *dst_path,
3803 struct btrfs_log_ctx *ctx)
3808 int key_type = BTRFS_DIR_ITEM_KEY;
3814 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3815 ctx, min_key, &max_key);
3818 if (max_key == (u64)-1)
3820 min_key = max_key + 1;
3823 if (key_type == BTRFS_DIR_ITEM_KEY) {
3824 key_type = BTRFS_DIR_INDEX_KEY;
3831 * a helper function to drop items from the log before we relog an
3832 * inode. max_key_type indicates the highest item type to remove.
3833 * This cannot be run for file data extents because it does not
3834 * free the extents they point to.
3836 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3837 struct btrfs_root *log,
3838 struct btrfs_path *path,
3839 u64 objectid, int max_key_type)
3842 struct btrfs_key key;
3843 struct btrfs_key found_key;
3846 key.objectid = objectid;
3847 key.type = max_key_type;
3848 key.offset = (u64)-1;
3851 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3852 BUG_ON(ret == 0); /* Logic error */
3856 if (path->slots[0] == 0)
3860 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3863 if (found_key.objectid != objectid)
3866 found_key.offset = 0;
3868 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3872 ret = btrfs_del_items(trans, log, path, start_slot,
3873 path->slots[0] - start_slot + 1);
3875 * If start slot isn't 0 then we don't need to re-search, we've
3876 * found the last guy with the objectid in this tree.
3878 if (ret || start_slot != 0)
3880 btrfs_release_path(path);
3882 btrfs_release_path(path);
3888 static void fill_inode_item(struct btrfs_trans_handle *trans,
3889 struct extent_buffer *leaf,
3890 struct btrfs_inode_item *item,
3891 struct inode *inode, int log_inode_only,
3894 struct btrfs_map_token token;
3896 btrfs_init_map_token(&token, leaf);
3898 if (log_inode_only) {
3899 /* set the generation to zero so the recover code
3900 * can tell the difference between an logging
3901 * just to say 'this inode exists' and a logging
3902 * to say 'update this inode with these values'
3904 btrfs_set_token_inode_generation(&token, item, 0);
3905 btrfs_set_token_inode_size(&token, item, logged_isize);
3907 btrfs_set_token_inode_generation(&token, item,
3908 BTRFS_I(inode)->generation);
3909 btrfs_set_token_inode_size(&token, item, inode->i_size);
3912 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3913 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3914 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3915 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3917 btrfs_set_token_timespec_sec(&token, &item->atime,
3918 inode->i_atime.tv_sec);
3919 btrfs_set_token_timespec_nsec(&token, &item->atime,
3920 inode->i_atime.tv_nsec);
3922 btrfs_set_token_timespec_sec(&token, &item->mtime,
3923 inode->i_mtime.tv_sec);
3924 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3925 inode->i_mtime.tv_nsec);
3927 btrfs_set_token_timespec_sec(&token, &item->ctime,
3928 inode->i_ctime.tv_sec);
3929 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3930 inode->i_ctime.tv_nsec);
3933 * We do not need to set the nbytes field, in fact during a fast fsync
3934 * its value may not even be correct, since a fast fsync does not wait
3935 * for ordered extent completion, which is where we update nbytes, it
3936 * only waits for writeback to complete. During log replay as we find
3937 * file extent items and replay them, we adjust the nbytes field of the
3938 * inode item in subvolume tree as needed (see overwrite_item()).
3941 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3942 btrfs_set_token_inode_transid(&token, item, trans->transid);
3943 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3944 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3945 btrfs_set_token_inode_block_group(&token, item, 0);
3948 static int log_inode_item(struct btrfs_trans_handle *trans,
3949 struct btrfs_root *log, struct btrfs_path *path,
3950 struct btrfs_inode *inode)
3952 struct btrfs_inode_item *inode_item;
3955 ret = btrfs_insert_empty_item(trans, log, path,
3956 &inode->location, sizeof(*inode_item));
3957 if (ret && ret != -EEXIST)
3959 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3960 struct btrfs_inode_item);
3961 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3963 btrfs_release_path(path);
3967 static int log_csums(struct btrfs_trans_handle *trans,
3968 struct btrfs_inode *inode,
3969 struct btrfs_root *log_root,
3970 struct btrfs_ordered_sum *sums)
3972 const u64 lock_end = sums->bytenr + sums->len - 1;
3973 struct extent_state *cached_state = NULL;
3977 * If this inode was not used for reflink operations in the current
3978 * transaction with new extents, then do the fast path, no need to
3979 * worry about logging checksum items with overlapping ranges.
3981 if (inode->last_reflink_trans < trans->transid)
3982 return btrfs_csum_file_blocks(trans, log_root, sums);
3985 * Serialize logging for checksums. This is to avoid racing with the
3986 * same checksum being logged by another task that is logging another
3987 * file which happens to refer to the same extent as well. Such races
3988 * can leave checksum items in the log with overlapping ranges.
3990 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
3991 lock_end, &cached_state);
3995 * Due to extent cloning, we might have logged a csum item that covers a
3996 * subrange of a cloned extent, and later we can end up logging a csum
3997 * item for a larger subrange of the same extent or the entire range.
3998 * This would leave csum items in the log tree that cover the same range
3999 * and break the searches for checksums in the log tree, resulting in
4000 * some checksums missing in the fs/subvolume tree. So just delete (or
4001 * trim and adjust) any existing csum items in the log for this range.
4003 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4005 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4007 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
4013 static noinline int copy_items(struct btrfs_trans_handle *trans,
4014 struct btrfs_inode *inode,
4015 struct btrfs_path *dst_path,
4016 struct btrfs_path *src_path,
4017 int start_slot, int nr, int inode_only,
4020 struct btrfs_fs_info *fs_info = trans->fs_info;
4021 unsigned long src_offset;
4022 unsigned long dst_offset;
4023 struct btrfs_root *log = inode->root->log_root;
4024 struct btrfs_file_extent_item *extent;
4025 struct btrfs_inode_item *inode_item;
4026 struct extent_buffer *src = src_path->nodes[0];
4028 struct btrfs_key *ins_keys;
4032 struct list_head ordered_sums;
4033 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4035 INIT_LIST_HEAD(&ordered_sums);
4037 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4038 nr * sizeof(u32), GFP_NOFS);
4042 ins_sizes = (u32 *)ins_data;
4043 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4045 for (i = 0; i < nr; i++) {
4046 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4047 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4049 ret = btrfs_insert_empty_items(trans, log, dst_path,
4050 ins_keys, ins_sizes, nr);
4056 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4057 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4058 dst_path->slots[0]);
4060 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4062 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4063 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4065 struct btrfs_inode_item);
4066 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4068 inode_only == LOG_INODE_EXISTS,
4071 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4072 src_offset, ins_sizes[i]);
4075 /* take a reference on file data extents so that truncates
4076 * or deletes of this inode don't have to relog the inode
4079 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4082 extent = btrfs_item_ptr(src, start_slot + i,
4083 struct btrfs_file_extent_item);
4085 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4088 found_type = btrfs_file_extent_type(src, extent);
4089 if (found_type == BTRFS_FILE_EXTENT_REG) {
4091 ds = btrfs_file_extent_disk_bytenr(src,
4093 /* ds == 0 is a hole */
4097 dl = btrfs_file_extent_disk_num_bytes(src,
4099 cs = btrfs_file_extent_offset(src, extent);
4100 cl = btrfs_file_extent_num_bytes(src,
4102 if (btrfs_file_extent_compression(src,
4108 ret = btrfs_lookup_csums_range(
4110 ds + cs, ds + cs + cl - 1,
4118 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4119 btrfs_release_path(dst_path);
4123 * we have to do this after the loop above to avoid changing the
4124 * log tree while trying to change the log tree.
4126 while (!list_empty(&ordered_sums)) {
4127 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4128 struct btrfs_ordered_sum,
4131 ret = log_csums(trans, inode, log, sums);
4132 list_del(&sums->list);
4139 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4141 struct extent_map *em1, *em2;
4143 em1 = list_entry(a, struct extent_map, list);
4144 em2 = list_entry(b, struct extent_map, list);
4146 if (em1->start < em2->start)
4148 else if (em1->start > em2->start)
4153 static int log_extent_csums(struct btrfs_trans_handle *trans,
4154 struct btrfs_inode *inode,
4155 struct btrfs_root *log_root,
4156 const struct extent_map *em,
4157 struct btrfs_log_ctx *ctx)
4159 struct btrfs_ordered_extent *ordered;
4162 u64 mod_start = em->mod_start;
4163 u64 mod_len = em->mod_len;
4164 LIST_HEAD(ordered_sums);
4167 if (inode->flags & BTRFS_INODE_NODATASUM ||
4168 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4169 em->block_start == EXTENT_MAP_HOLE)
4172 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4173 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4174 const u64 mod_end = mod_start + mod_len;
4175 struct btrfs_ordered_sum *sums;
4180 if (ordered_end <= mod_start)
4182 if (mod_end <= ordered->file_offset)
4186 * We are going to copy all the csums on this ordered extent, so
4187 * go ahead and adjust mod_start and mod_len in case this ordered
4188 * extent has already been logged.
4190 if (ordered->file_offset > mod_start) {
4191 if (ordered_end >= mod_end)
4192 mod_len = ordered->file_offset - mod_start;
4194 * If we have this case
4196 * |--------- logged extent ---------|
4197 * |----- ordered extent ----|
4199 * Just don't mess with mod_start and mod_len, we'll
4200 * just end up logging more csums than we need and it
4204 if (ordered_end < mod_end) {
4205 mod_len = mod_end - ordered_end;
4206 mod_start = ordered_end;
4213 * To keep us from looping for the above case of an ordered
4214 * extent that falls inside of the logged extent.
4216 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4219 list_for_each_entry(sums, &ordered->list, list) {
4220 ret = log_csums(trans, inode, log_root, sums);
4226 /* We're done, found all csums in the ordered extents. */
4230 /* If we're compressed we have to save the entire range of csums. */
4231 if (em->compress_type) {
4233 csum_len = max(em->block_len, em->orig_block_len);
4235 csum_offset = mod_start - em->start;
4239 /* block start is already adjusted for the file extent offset. */
4240 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4241 em->block_start + csum_offset,
4242 em->block_start + csum_offset +
4243 csum_len - 1, &ordered_sums, 0);
4247 while (!list_empty(&ordered_sums)) {
4248 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4249 struct btrfs_ordered_sum,
4252 ret = log_csums(trans, inode, log_root, sums);
4253 list_del(&sums->list);
4260 static int log_one_extent(struct btrfs_trans_handle *trans,
4261 struct btrfs_inode *inode, struct btrfs_root *root,
4262 const struct extent_map *em,
4263 struct btrfs_path *path,
4264 struct btrfs_log_ctx *ctx)
4266 struct btrfs_drop_extents_args drop_args = { 0 };
4267 struct btrfs_root *log = root->log_root;
4268 struct btrfs_file_extent_item *fi;
4269 struct extent_buffer *leaf;
4270 struct btrfs_map_token token;
4271 struct btrfs_key key;
4272 u64 extent_offset = em->start - em->orig_start;
4276 ret = log_extent_csums(trans, inode, log, em, ctx);
4280 drop_args.path = path;
4281 drop_args.start = em->start;
4282 drop_args.end = em->start + em->len;
4283 drop_args.replace_extent = true;
4284 drop_args.extent_item_size = sizeof(*fi);
4285 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4289 if (!drop_args.extent_inserted) {
4290 key.objectid = btrfs_ino(inode);
4291 key.type = BTRFS_EXTENT_DATA_KEY;
4292 key.offset = em->start;
4294 ret = btrfs_insert_empty_item(trans, log, path, &key,
4299 leaf = path->nodes[0];
4300 btrfs_init_map_token(&token, leaf);
4301 fi = btrfs_item_ptr(leaf, path->slots[0],
4302 struct btrfs_file_extent_item);
4304 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4305 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4306 btrfs_set_token_file_extent_type(&token, fi,
4307 BTRFS_FILE_EXTENT_PREALLOC);
4309 btrfs_set_token_file_extent_type(&token, fi,
4310 BTRFS_FILE_EXTENT_REG);
4312 block_len = max(em->block_len, em->orig_block_len);
4313 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4314 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4316 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4317 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4318 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4321 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4323 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4324 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4327 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4328 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4329 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4330 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4331 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4332 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4333 btrfs_mark_buffer_dirty(leaf);
4335 btrfs_release_path(path);
4341 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4342 * lose them after doing a fast fsync and replaying the log. We scan the
4343 * subvolume's root instead of iterating the inode's extent map tree because
4344 * otherwise we can log incorrect extent items based on extent map conversion.
4345 * That can happen due to the fact that extent maps are merged when they
4346 * are not in the extent map tree's list of modified extents.
4348 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4349 struct btrfs_inode *inode,
4350 struct btrfs_path *path)
4352 struct btrfs_root *root = inode->root;
4353 struct btrfs_key key;
4354 const u64 i_size = i_size_read(&inode->vfs_inode);
4355 const u64 ino = btrfs_ino(inode);
4356 struct btrfs_path *dst_path = NULL;
4357 bool dropped_extents = false;
4358 u64 truncate_offset = i_size;
4359 struct extent_buffer *leaf;
4365 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4369 key.type = BTRFS_EXTENT_DATA_KEY;
4370 key.offset = i_size;
4371 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4376 * We must check if there is a prealloc extent that starts before the
4377 * i_size and crosses the i_size boundary. This is to ensure later we
4378 * truncate down to the end of that extent and not to the i_size, as
4379 * otherwise we end up losing part of the prealloc extent after a log
4380 * replay and with an implicit hole if there is another prealloc extent
4381 * that starts at an offset beyond i_size.
4383 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4388 struct btrfs_file_extent_item *ei;
4390 leaf = path->nodes[0];
4391 slot = path->slots[0];
4392 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4394 if (btrfs_file_extent_type(leaf, ei) ==
4395 BTRFS_FILE_EXTENT_PREALLOC) {
4398 btrfs_item_key_to_cpu(leaf, &key, slot);
4399 extent_end = key.offset +
4400 btrfs_file_extent_num_bytes(leaf, ei);
4402 if (extent_end > i_size)
4403 truncate_offset = extent_end;
4410 leaf = path->nodes[0];
4411 slot = path->slots[0];
4413 if (slot >= btrfs_header_nritems(leaf)) {
4415 ret = copy_items(trans, inode, dst_path, path,
4416 start_slot, ins_nr, 1, 0);
4421 ret = btrfs_next_leaf(root, path);
4431 btrfs_item_key_to_cpu(leaf, &key, slot);
4432 if (key.objectid > ino)
4434 if (WARN_ON_ONCE(key.objectid < ino) ||
4435 key.type < BTRFS_EXTENT_DATA_KEY ||
4436 key.offset < i_size) {
4440 if (!dropped_extents) {
4442 * Avoid logging extent items logged in past fsync calls
4443 * and leading to duplicate keys in the log tree.
4446 ret = btrfs_truncate_inode_items(trans,
4448 inode, truncate_offset,
4449 BTRFS_EXTENT_DATA_KEY);
4450 } while (ret == -EAGAIN);
4453 dropped_extents = true;
4460 dst_path = btrfs_alloc_path();
4468 ret = copy_items(trans, inode, dst_path, path,
4469 start_slot, ins_nr, 1, 0);
4471 btrfs_release_path(path);
4472 btrfs_free_path(dst_path);
4476 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4477 struct btrfs_root *root,
4478 struct btrfs_inode *inode,
4479 struct btrfs_path *path,
4480 struct btrfs_log_ctx *ctx)
4482 struct btrfs_ordered_extent *ordered;
4483 struct btrfs_ordered_extent *tmp;
4484 struct extent_map *em, *n;
4485 struct list_head extents;
4486 struct extent_map_tree *tree = &inode->extent_tree;
4490 INIT_LIST_HEAD(&extents);
4492 write_lock(&tree->lock);
4494 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4495 list_del_init(&em->list);
4497 * Just an arbitrary number, this can be really CPU intensive
4498 * once we start getting a lot of extents, and really once we
4499 * have a bunch of extents we just want to commit since it will
4502 if (++num > 32768) {
4503 list_del_init(&tree->modified_extents);
4508 if (em->generation < trans->transid)
4511 /* We log prealloc extents beyond eof later. */
4512 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4513 em->start >= i_size_read(&inode->vfs_inode))
4516 /* Need a ref to keep it from getting evicted from cache */
4517 refcount_inc(&em->refs);
4518 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4519 list_add_tail(&em->list, &extents);
4523 list_sort(NULL, &extents, extent_cmp);
4525 while (!list_empty(&extents)) {
4526 em = list_entry(extents.next, struct extent_map, list);
4528 list_del_init(&em->list);
4531 * If we had an error we just need to delete everybody from our
4535 clear_em_logging(tree, em);
4536 free_extent_map(em);
4540 write_unlock(&tree->lock);
4542 ret = log_one_extent(trans, inode, root, em, path, ctx);
4543 write_lock(&tree->lock);
4544 clear_em_logging(tree, em);
4545 free_extent_map(em);
4547 WARN_ON(!list_empty(&extents));
4548 write_unlock(&tree->lock);
4550 btrfs_release_path(path);
4552 ret = btrfs_log_prealloc_extents(trans, inode, path);
4557 * We have logged all extents successfully, now make sure the commit of
4558 * the current transaction waits for the ordered extents to complete
4559 * before it commits and wipes out the log trees, otherwise we would
4560 * lose data if an ordered extents completes after the transaction
4561 * commits and a power failure happens after the transaction commit.
4563 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4564 list_del_init(&ordered->log_list);
4565 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4567 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4568 spin_lock_irq(&inode->ordered_tree.lock);
4569 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4570 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4571 atomic_inc(&trans->transaction->pending_ordered);
4573 spin_unlock_irq(&inode->ordered_tree.lock);
4575 btrfs_put_ordered_extent(ordered);
4581 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4582 struct btrfs_path *path, u64 *size_ret)
4584 struct btrfs_key key;
4587 key.objectid = btrfs_ino(inode);
4588 key.type = BTRFS_INODE_ITEM_KEY;
4591 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4594 } else if (ret > 0) {
4597 struct btrfs_inode_item *item;
4599 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4600 struct btrfs_inode_item);
4601 *size_ret = btrfs_inode_size(path->nodes[0], item);
4603 * If the in-memory inode's i_size is smaller then the inode
4604 * size stored in the btree, return the inode's i_size, so
4605 * that we get a correct inode size after replaying the log
4606 * when before a power failure we had a shrinking truncate
4607 * followed by addition of a new name (rename / new hard link).
4608 * Otherwise return the inode size from the btree, to avoid
4609 * data loss when replaying a log due to previously doing a
4610 * write that expands the inode's size and logging a new name
4611 * immediately after.
4613 if (*size_ret > inode->vfs_inode.i_size)
4614 *size_ret = inode->vfs_inode.i_size;
4617 btrfs_release_path(path);
4622 * At the moment we always log all xattrs. This is to figure out at log replay
4623 * time which xattrs must have their deletion replayed. If a xattr is missing
4624 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4625 * because if a xattr is deleted, the inode is fsynced and a power failure
4626 * happens, causing the log to be replayed the next time the fs is mounted,
4627 * we want the xattr to not exist anymore (same behaviour as other filesystems
4628 * with a journal, ext3/4, xfs, f2fs, etc).
4630 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4631 struct btrfs_root *root,
4632 struct btrfs_inode *inode,
4633 struct btrfs_path *path,
4634 struct btrfs_path *dst_path)
4637 struct btrfs_key key;
4638 const u64 ino = btrfs_ino(inode);
4641 bool found_xattrs = false;
4643 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4647 key.type = BTRFS_XATTR_ITEM_KEY;
4650 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4655 int slot = path->slots[0];
4656 struct extent_buffer *leaf = path->nodes[0];
4657 int nritems = btrfs_header_nritems(leaf);
4659 if (slot >= nritems) {
4661 ret = copy_items(trans, inode, dst_path, path,
4662 start_slot, ins_nr, 1, 0);
4667 ret = btrfs_next_leaf(root, path);
4675 btrfs_item_key_to_cpu(leaf, &key, slot);
4676 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4683 found_xattrs = true;
4687 ret = copy_items(trans, inode, dst_path, path,
4688 start_slot, ins_nr, 1, 0);
4694 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4700 * When using the NO_HOLES feature if we punched a hole that causes the
4701 * deletion of entire leafs or all the extent items of the first leaf (the one
4702 * that contains the inode item and references) we may end up not processing
4703 * any extents, because there are no leafs with a generation matching the
4704 * current transaction that have extent items for our inode. So we need to find
4705 * if any holes exist and then log them. We also need to log holes after any
4706 * truncate operation that changes the inode's size.
4708 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4709 struct btrfs_root *root,
4710 struct btrfs_inode *inode,
4711 struct btrfs_path *path)
4713 struct btrfs_fs_info *fs_info = root->fs_info;
4714 struct btrfs_key key;
4715 const u64 ino = btrfs_ino(inode);
4716 const u64 i_size = i_size_read(&inode->vfs_inode);
4717 u64 prev_extent_end = 0;
4720 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4724 key.type = BTRFS_EXTENT_DATA_KEY;
4727 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4732 struct extent_buffer *leaf = path->nodes[0];
4734 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4735 ret = btrfs_next_leaf(root, path);
4742 leaf = path->nodes[0];
4745 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4746 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4749 /* We have a hole, log it. */
4750 if (prev_extent_end < key.offset) {
4751 const u64 hole_len = key.offset - prev_extent_end;
4754 * Release the path to avoid deadlocks with other code
4755 * paths that search the root while holding locks on
4756 * leafs from the log root.
4758 btrfs_release_path(path);
4759 ret = btrfs_insert_file_extent(trans, root->log_root,
4760 ino, prev_extent_end, 0,
4761 0, hole_len, 0, hole_len,
4767 * Search for the same key again in the root. Since it's
4768 * an extent item and we are holding the inode lock, the
4769 * key must still exist. If it doesn't just emit warning
4770 * and return an error to fall back to a transaction
4773 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4776 if (WARN_ON(ret > 0))
4778 leaf = path->nodes[0];
4781 prev_extent_end = btrfs_file_extent_end(path);
4786 if (prev_extent_end < i_size) {
4789 btrfs_release_path(path);
4790 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4791 ret = btrfs_insert_file_extent(trans, root->log_root,
4792 ino, prev_extent_end, 0, 0,
4793 hole_len, 0, hole_len,
4803 * When we are logging a new inode X, check if it doesn't have a reference that
4804 * matches the reference from some other inode Y created in a past transaction
4805 * and that was renamed in the current transaction. If we don't do this, then at
4806 * log replay time we can lose inode Y (and all its files if it's a directory):
4809 * echo "hello world" > /mnt/x/foobar
4812 * mkdir /mnt/x # or touch /mnt/x
4813 * xfs_io -c fsync /mnt/x
4815 * mount fs, trigger log replay
4817 * After the log replay procedure, we would lose the first directory and all its
4818 * files (file foobar).
4819 * For the case where inode Y is not a directory we simply end up losing it:
4821 * echo "123" > /mnt/foo
4823 * mv /mnt/foo /mnt/bar
4824 * echo "abc" > /mnt/foo
4825 * xfs_io -c fsync /mnt/foo
4828 * We also need this for cases where a snapshot entry is replaced by some other
4829 * entry (file or directory) otherwise we end up with an unreplayable log due to
4830 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4831 * if it were a regular entry:
4834 * btrfs subvolume snapshot /mnt /mnt/x/snap
4835 * btrfs subvolume delete /mnt/x/snap
4838 * fsync /mnt/x or fsync some new file inside it
4841 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4842 * the same transaction.
4844 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4846 const struct btrfs_key *key,
4847 struct btrfs_inode *inode,
4848 u64 *other_ino, u64 *other_parent)
4851 struct btrfs_path *search_path;
4854 u32 item_size = btrfs_item_size_nr(eb, slot);
4856 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4858 search_path = btrfs_alloc_path();
4861 search_path->search_commit_root = 1;
4862 search_path->skip_locking = 1;
4864 while (cur_offset < item_size) {
4868 unsigned long name_ptr;
4869 struct btrfs_dir_item *di;
4871 if (key->type == BTRFS_INODE_REF_KEY) {
4872 struct btrfs_inode_ref *iref;
4874 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4875 parent = key->offset;
4876 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4877 name_ptr = (unsigned long)(iref + 1);
4878 this_len = sizeof(*iref) + this_name_len;
4880 struct btrfs_inode_extref *extref;
4882 extref = (struct btrfs_inode_extref *)(ptr +
4884 parent = btrfs_inode_extref_parent(eb, extref);
4885 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4886 name_ptr = (unsigned long)&extref->name;
4887 this_len = sizeof(*extref) + this_name_len;
4890 if (this_name_len > name_len) {
4893 new_name = krealloc(name, this_name_len, GFP_NOFS);
4898 name_len = this_name_len;
4902 read_extent_buffer(eb, name, name_ptr, this_name_len);
4903 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4904 parent, name, this_name_len, 0);
4905 if (di && !IS_ERR(di)) {
4906 struct btrfs_key di_key;
4908 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4910 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4911 if (di_key.objectid != key->objectid) {
4913 *other_ino = di_key.objectid;
4914 *other_parent = parent;
4922 } else if (IS_ERR(di)) {
4926 btrfs_release_path(search_path);
4928 cur_offset += this_len;
4932 btrfs_free_path(search_path);
4937 struct btrfs_ino_list {
4940 struct list_head list;
4943 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4944 struct btrfs_root *root,
4945 struct btrfs_path *path,
4946 struct btrfs_log_ctx *ctx,
4947 u64 ino, u64 parent)
4949 struct btrfs_ino_list *ino_elem;
4950 LIST_HEAD(inode_list);
4953 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4956 ino_elem->ino = ino;
4957 ino_elem->parent = parent;
4958 list_add_tail(&ino_elem->list, &inode_list);
4960 while (!list_empty(&inode_list)) {
4961 struct btrfs_fs_info *fs_info = root->fs_info;
4962 struct btrfs_key key;
4963 struct inode *inode;
4965 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4967 ino = ino_elem->ino;
4968 parent = ino_elem->parent;
4969 list_del(&ino_elem->list);
4974 btrfs_release_path(path);
4976 inode = btrfs_iget(fs_info->sb, ino, root);
4978 * If the other inode that had a conflicting dir entry was
4979 * deleted in the current transaction, we need to log its parent
4982 if (IS_ERR(inode)) {
4983 ret = PTR_ERR(inode);
4984 if (ret == -ENOENT) {
4985 inode = btrfs_iget(fs_info->sb, parent, root);
4986 if (IS_ERR(inode)) {
4987 ret = PTR_ERR(inode);
4989 ret = btrfs_log_inode(trans, root,
4991 LOG_OTHER_INODE_ALL,
4993 btrfs_add_delayed_iput(inode);
4999 * If the inode was already logged skip it - otherwise we can
5000 * hit an infinite loop. Example:
5002 * From the commit root (previous transaction) we have the
5005 * inode 257 a directory
5006 * inode 258 with references "zz" and "zz_link" on inode 257
5007 * inode 259 with reference "a" on inode 257
5009 * And in the current (uncommitted) transaction we have:
5011 * inode 257 a directory, unchanged
5012 * inode 258 with references "a" and "a2" on inode 257
5013 * inode 259 with reference "zz_link" on inode 257
5014 * inode 261 with reference "zz" on inode 257
5016 * When logging inode 261 the following infinite loop could
5017 * happen if we don't skip already logged inodes:
5019 * - we detect inode 258 as a conflicting inode, with inode 261
5020 * on reference "zz", and log it;
5022 * - we detect inode 259 as a conflicting inode, with inode 258
5023 * on reference "a", and log it;
5025 * - we detect inode 258 as a conflicting inode, with inode 259
5026 * on reference "zz_link", and log it - again! After this we
5027 * repeat the above steps forever.
5029 spin_lock(&BTRFS_I(inode)->lock);
5031 * Check the inode's logged_trans only instead of
5032 * btrfs_inode_in_log(). This is because the last_log_commit of
5033 * the inode is not updated when we only log that it exists and
5034 * it has the full sync bit set (see btrfs_log_inode()).
5036 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5037 spin_unlock(&BTRFS_I(inode)->lock);
5038 btrfs_add_delayed_iput(inode);
5041 spin_unlock(&BTRFS_I(inode)->lock);
5043 * We are safe logging the other inode without acquiring its
5044 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5045 * are safe against concurrent renames of the other inode as
5046 * well because during a rename we pin the log and update the
5047 * log with the new name before we unpin it.
5049 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5050 LOG_OTHER_INODE, ctx);
5052 btrfs_add_delayed_iput(inode);
5057 key.type = BTRFS_INODE_REF_KEY;
5059 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5061 btrfs_add_delayed_iput(inode);
5066 struct extent_buffer *leaf = path->nodes[0];
5067 int slot = path->slots[0];
5069 u64 other_parent = 0;
5071 if (slot >= btrfs_header_nritems(leaf)) {
5072 ret = btrfs_next_leaf(root, path);
5075 } else if (ret > 0) {
5082 btrfs_item_key_to_cpu(leaf, &key, slot);
5083 if (key.objectid != ino ||
5084 (key.type != BTRFS_INODE_REF_KEY &&
5085 key.type != BTRFS_INODE_EXTREF_KEY)) {
5090 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5091 BTRFS_I(inode), &other_ino,
5096 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5101 ino_elem->ino = other_ino;
5102 ino_elem->parent = other_parent;
5103 list_add_tail(&ino_elem->list, &inode_list);
5108 btrfs_add_delayed_iput(inode);
5114 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5115 struct btrfs_inode *inode,
5116 struct btrfs_key *min_key,
5117 const struct btrfs_key *max_key,
5118 struct btrfs_path *path,
5119 struct btrfs_path *dst_path,
5120 const u64 logged_isize,
5121 const bool recursive_logging,
5122 const int inode_only,
5123 struct btrfs_log_ctx *ctx,
5124 bool *need_log_inode_item)
5126 struct btrfs_root *root = inode->root;
5127 int ins_start_slot = 0;
5132 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5140 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5141 if (min_key->objectid != max_key->objectid)
5143 if (min_key->type > max_key->type)
5146 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5147 *need_log_inode_item = false;
5149 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5150 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5151 inode->generation == trans->transid &&
5152 !recursive_logging) {
5154 u64 other_parent = 0;
5156 ret = btrfs_check_ref_name_override(path->nodes[0],
5157 path->slots[0], min_key, inode,
5158 &other_ino, &other_parent);
5161 } else if (ret > 0 && ctx &&
5162 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5167 ins_start_slot = path->slots[0];
5169 ret = copy_items(trans, inode, dst_path, path,
5170 ins_start_slot, ins_nr,
5171 inode_only, logged_isize);
5176 ret = log_conflicting_inodes(trans, root, path,
5177 ctx, other_ino, other_parent);
5180 btrfs_release_path(path);
5185 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5186 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5189 ret = copy_items(trans, inode, dst_path, path,
5191 ins_nr, inode_only, logged_isize);
5198 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5201 } else if (!ins_nr) {
5202 ins_start_slot = path->slots[0];
5207 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5208 ins_nr, inode_only, logged_isize);
5212 ins_start_slot = path->slots[0];
5215 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5216 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5221 ret = copy_items(trans, inode, dst_path, path,
5222 ins_start_slot, ins_nr, inode_only,
5228 btrfs_release_path(path);
5230 if (min_key->offset < (u64)-1) {
5232 } else if (min_key->type < max_key->type) {
5234 min_key->offset = 0;
5240 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5241 ins_nr, inode_only, logged_isize);
5246 /* log a single inode in the tree log.
5247 * At least one parent directory for this inode must exist in the tree
5248 * or be logged already.
5250 * Any items from this inode changed by the current transaction are copied
5251 * to the log tree. An extra reference is taken on any extents in this
5252 * file, allowing us to avoid a whole pile of corner cases around logging
5253 * blocks that have been removed from the tree.
5255 * See LOG_INODE_ALL and related defines for a description of what inode_only
5258 * This handles both files and directories.
5260 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5261 struct btrfs_root *root, struct btrfs_inode *inode,
5263 struct btrfs_log_ctx *ctx)
5265 struct btrfs_path *path;
5266 struct btrfs_path *dst_path;
5267 struct btrfs_key min_key;
5268 struct btrfs_key max_key;
5269 struct btrfs_root *log = root->log_root;
5272 bool fast_search = false;
5273 u64 ino = btrfs_ino(inode);
5274 struct extent_map_tree *em_tree = &inode->extent_tree;
5275 u64 logged_isize = 0;
5276 bool need_log_inode_item = true;
5277 bool xattrs_logged = false;
5278 bool recursive_logging = false;
5280 path = btrfs_alloc_path();
5283 dst_path = btrfs_alloc_path();
5285 btrfs_free_path(path);
5289 min_key.objectid = ino;
5290 min_key.type = BTRFS_INODE_ITEM_KEY;
5293 max_key.objectid = ino;
5296 /* today the code can only do partial logging of directories */
5297 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5298 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5299 &inode->runtime_flags) &&
5300 inode_only >= LOG_INODE_EXISTS))
5301 max_key.type = BTRFS_XATTR_ITEM_KEY;
5303 max_key.type = (u8)-1;
5304 max_key.offset = (u64)-1;
5307 * Only run delayed items if we are a directory. We want to make sure
5308 * all directory indexes hit the fs/subvolume tree so we can find them
5309 * and figure out which index ranges have to be logged.
5311 * Otherwise commit the delayed inode only if the full sync flag is set,
5312 * as we want to make sure an up to date version is in the subvolume
5313 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5314 * it to the log tree. For a non full sync, we always log the inode item
5315 * based on the in-memory struct btrfs_inode which is always up to date.
5317 if (S_ISDIR(inode->vfs_inode.i_mode))
5318 ret = btrfs_commit_inode_delayed_items(trans, inode);
5319 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5320 ret = btrfs_commit_inode_delayed_inode(inode);
5323 btrfs_free_path(path);
5324 btrfs_free_path(dst_path);
5328 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5329 recursive_logging = true;
5330 if (inode_only == LOG_OTHER_INODE)
5331 inode_only = LOG_INODE_EXISTS;
5333 inode_only = LOG_INODE_ALL;
5334 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5336 mutex_lock(&inode->log_mutex);
5340 * This is for cases where logging a directory could result in losing a
5341 * a file after replaying the log. For example, if we move a file from a
5342 * directory A to a directory B, then fsync directory A, we have no way
5343 * to known the file was moved from A to B, so logging just A would
5344 * result in losing the file after a log replay.
5346 if (S_ISDIR(inode->vfs_inode.i_mode) &&
5347 inode_only == LOG_INODE_ALL &&
5348 inode->last_unlink_trans >= trans->transid) {
5349 btrfs_set_log_full_commit(trans);
5355 * a brute force approach to making sure we get the most uptodate
5356 * copies of everything.
5358 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5359 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5361 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5362 if (inode_only == LOG_INODE_EXISTS)
5363 max_key_type = BTRFS_XATTR_ITEM_KEY;
5364 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5366 if (inode_only == LOG_INODE_EXISTS) {
5368 * Make sure the new inode item we write to the log has
5369 * the same isize as the current one (if it exists).
5370 * This is necessary to prevent data loss after log
5371 * replay, and also to prevent doing a wrong expanding
5372 * truncate - for e.g. create file, write 4K into offset
5373 * 0, fsync, write 4K into offset 4096, add hard link,
5374 * fsync some other file (to sync log), power fail - if
5375 * we use the inode's current i_size, after log replay
5376 * we get a 8Kb file, with the last 4Kb extent as a hole
5377 * (zeroes), as if an expanding truncate happened,
5378 * instead of getting a file of 4Kb only.
5380 err = logged_inode_size(log, inode, path, &logged_isize);
5384 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5385 &inode->runtime_flags)) {
5386 if (inode_only == LOG_INODE_EXISTS) {
5387 max_key.type = BTRFS_XATTR_ITEM_KEY;
5388 ret = drop_objectid_items(trans, log, path, ino,
5391 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5392 &inode->runtime_flags);
5393 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5394 &inode->runtime_flags);
5396 ret = btrfs_truncate_inode_items(trans,
5402 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5403 &inode->runtime_flags) ||
5404 inode_only == LOG_INODE_EXISTS) {
5405 if (inode_only == LOG_INODE_ALL)
5407 max_key.type = BTRFS_XATTR_ITEM_KEY;
5408 ret = drop_objectid_items(trans, log, path, ino,
5411 if (inode_only == LOG_INODE_ALL)
5422 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5423 path, dst_path, logged_isize,
5424 recursive_logging, inode_only, ctx,
5425 &need_log_inode_item);
5429 btrfs_release_path(path);
5430 btrfs_release_path(dst_path);
5431 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5434 xattrs_logged = true;
5435 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5436 btrfs_release_path(path);
5437 btrfs_release_path(dst_path);
5438 err = btrfs_log_holes(trans, root, inode, path);
5443 btrfs_release_path(path);
5444 btrfs_release_path(dst_path);
5445 if (need_log_inode_item) {
5446 err = log_inode_item(trans, log, dst_path, inode);
5447 if (!err && !xattrs_logged) {
5448 err = btrfs_log_all_xattrs(trans, root, inode, path,
5450 btrfs_release_path(path);
5456 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5462 } else if (inode_only == LOG_INODE_ALL) {
5463 struct extent_map *em, *n;
5465 write_lock(&em_tree->lock);
5466 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5467 list_del_init(&em->list);
5468 write_unlock(&em_tree->lock);
5471 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5472 ret = log_directory_changes(trans, root, inode, path, dst_path,
5481 * If we are logging that an ancestor inode exists as part of logging a
5482 * new name from a link or rename operation, don't mark the inode as
5483 * logged - otherwise if an explicit fsync is made against an ancestor,
5484 * the fsync considers the inode in the log and doesn't sync the log,
5485 * resulting in the ancestor missing after a power failure unless the
5486 * log was synced as part of an fsync against any other unrelated inode.
5487 * So keep it simple for this case and just don't flag the ancestors as
5491 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5492 &inode->vfs_inode != ctx->inode)) {
5493 spin_lock(&inode->lock);
5494 inode->logged_trans = trans->transid;
5496 * Don't update last_log_commit if we logged that an inode exists
5497 * after it was loaded to memory (full_sync bit set).
5498 * This is to prevent data loss when we do a write to the inode,
5499 * then the inode gets evicted after all delalloc was flushed,
5500 * then we log it exists (due to a rename for example) and then
5501 * fsync it. This last fsync would do nothing (not logging the
5502 * extents previously written).
5504 if (inode_only != LOG_INODE_EXISTS ||
5505 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5506 inode->last_log_commit = inode->last_sub_trans;
5507 spin_unlock(&inode->lock);
5510 mutex_unlock(&inode->log_mutex);
5512 btrfs_free_path(path);
5513 btrfs_free_path(dst_path);
5518 * Check if we need to log an inode. This is used in contexts where while
5519 * logging an inode we need to log another inode (either that it exists or in
5520 * full mode). This is used instead of btrfs_inode_in_log() because the later
5521 * requires the inode to be in the log and have the log transaction committed,
5522 * while here we do not care if the log transaction was already committed - our
5523 * caller will commit the log later - and we want to avoid logging an inode
5524 * multiple times when multiple tasks have joined the same log transaction.
5526 static bool need_log_inode(struct btrfs_trans_handle *trans,
5527 struct btrfs_inode *inode)
5530 * If this inode does not have new/updated/deleted xattrs since the last
5531 * time it was logged and is flagged as logged in the current transaction,
5532 * we can skip logging it. As for new/deleted names, those are updated in
5533 * the log by link/unlink/rename operations.
5534 * In case the inode was logged and then evicted and reloaded, its
5535 * logged_trans will be 0, in which case we have to fully log it since
5536 * logged_trans is a transient field, not persisted.
5538 if (inode->logged_trans == trans->transid &&
5539 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5545 struct btrfs_dir_list {
5547 struct list_head list;
5551 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5552 * details about the why it is needed.
5553 * This is a recursive operation - if an existing dentry corresponds to a
5554 * directory, that directory's new entries are logged too (same behaviour as
5555 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5556 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5557 * complains about the following circular lock dependency / possible deadlock:
5561 * lock(&type->i_mutex_dir_key#3/2);
5562 * lock(sb_internal#2);
5563 * lock(&type->i_mutex_dir_key#3/2);
5564 * lock(&sb->s_type->i_mutex_key#14);
5566 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5567 * sb_start_intwrite() in btrfs_start_transaction().
5568 * Not locking i_mutex of the inodes is still safe because:
5570 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5571 * that while logging the inode new references (names) are added or removed
5572 * from the inode, leaving the logged inode item with a link count that does
5573 * not match the number of logged inode reference items. This is fine because
5574 * at log replay time we compute the real number of links and correct the
5575 * link count in the inode item (see replay_one_buffer() and
5576 * link_to_fixup_dir());
5578 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5579 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5580 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5581 * has a size that doesn't match the sum of the lengths of all the logged
5582 * names. This does not result in a problem because if a dir_item key is
5583 * logged but its matching dir_index key is not logged, at log replay time we
5584 * don't use it to replay the respective name (see replay_one_name()). On the
5585 * other hand if only the dir_index key ends up being logged, the respective
5586 * name is added to the fs/subvol tree with both the dir_item and dir_index
5587 * keys created (see replay_one_name()).
5588 * The directory's inode item with a wrong i_size is not a problem as well,
5589 * since we don't use it at log replay time to set the i_size in the inode
5590 * item of the fs/subvol tree (see overwrite_item()).
5592 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5593 struct btrfs_root *root,
5594 struct btrfs_inode *start_inode,
5595 struct btrfs_log_ctx *ctx)
5597 struct btrfs_fs_info *fs_info = root->fs_info;
5598 struct btrfs_root *log = root->log_root;
5599 struct btrfs_path *path;
5600 LIST_HEAD(dir_list);
5601 struct btrfs_dir_list *dir_elem;
5604 path = btrfs_alloc_path();
5608 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5610 btrfs_free_path(path);
5613 dir_elem->ino = btrfs_ino(start_inode);
5614 list_add_tail(&dir_elem->list, &dir_list);
5616 while (!list_empty(&dir_list)) {
5617 struct extent_buffer *leaf;
5618 struct btrfs_key min_key;
5622 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5625 goto next_dir_inode;
5627 min_key.objectid = dir_elem->ino;
5628 min_key.type = BTRFS_DIR_ITEM_KEY;
5631 btrfs_release_path(path);
5632 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5634 goto next_dir_inode;
5635 } else if (ret > 0) {
5637 goto next_dir_inode;
5641 leaf = path->nodes[0];
5642 nritems = btrfs_header_nritems(leaf);
5643 for (i = path->slots[0]; i < nritems; i++) {
5644 struct btrfs_dir_item *di;
5645 struct btrfs_key di_key;
5646 struct inode *di_inode;
5647 struct btrfs_dir_list *new_dir_elem;
5648 int log_mode = LOG_INODE_EXISTS;
5651 btrfs_item_key_to_cpu(leaf, &min_key, i);
5652 if (min_key.objectid != dir_elem->ino ||
5653 min_key.type != BTRFS_DIR_ITEM_KEY)
5654 goto next_dir_inode;
5656 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5657 type = btrfs_dir_type(leaf, di);
5658 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5659 type != BTRFS_FT_DIR)
5661 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5662 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5665 btrfs_release_path(path);
5666 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5667 if (IS_ERR(di_inode)) {
5668 ret = PTR_ERR(di_inode);
5669 goto next_dir_inode;
5672 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5673 btrfs_add_delayed_iput(di_inode);
5677 ctx->log_new_dentries = false;
5678 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5679 log_mode = LOG_INODE_ALL;
5680 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5682 btrfs_add_delayed_iput(di_inode);
5684 goto next_dir_inode;
5685 if (ctx->log_new_dentries) {
5686 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5688 if (!new_dir_elem) {
5690 goto next_dir_inode;
5692 new_dir_elem->ino = di_key.objectid;
5693 list_add_tail(&new_dir_elem->list, &dir_list);
5698 ret = btrfs_next_leaf(log, path);
5700 goto next_dir_inode;
5701 } else if (ret > 0) {
5703 goto next_dir_inode;
5707 if (min_key.offset < (u64)-1) {
5712 list_del(&dir_elem->list);
5716 btrfs_free_path(path);
5720 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5721 struct btrfs_inode *inode,
5722 struct btrfs_log_ctx *ctx)
5724 struct btrfs_fs_info *fs_info = trans->fs_info;
5726 struct btrfs_path *path;
5727 struct btrfs_key key;
5728 struct btrfs_root *root = inode->root;
5729 const u64 ino = btrfs_ino(inode);
5731 path = btrfs_alloc_path();
5734 path->skip_locking = 1;
5735 path->search_commit_root = 1;
5738 key.type = BTRFS_INODE_REF_KEY;
5740 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5745 struct extent_buffer *leaf = path->nodes[0];
5746 int slot = path->slots[0];
5751 if (slot >= btrfs_header_nritems(leaf)) {
5752 ret = btrfs_next_leaf(root, path);
5760 btrfs_item_key_to_cpu(leaf, &key, slot);
5761 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5762 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5765 item_size = btrfs_item_size_nr(leaf, slot);
5766 ptr = btrfs_item_ptr_offset(leaf, slot);
5767 while (cur_offset < item_size) {
5768 struct btrfs_key inode_key;
5769 struct inode *dir_inode;
5771 inode_key.type = BTRFS_INODE_ITEM_KEY;
5772 inode_key.offset = 0;
5774 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5775 struct btrfs_inode_extref *extref;
5777 extref = (struct btrfs_inode_extref *)
5779 inode_key.objectid = btrfs_inode_extref_parent(
5781 cur_offset += sizeof(*extref);
5782 cur_offset += btrfs_inode_extref_name_len(leaf,
5785 inode_key.objectid = key.offset;
5786 cur_offset = item_size;
5789 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5792 * If the parent inode was deleted, return an error to
5793 * fallback to a transaction commit. This is to prevent
5794 * getting an inode that was moved from one parent A to
5795 * a parent B, got its former parent A deleted and then
5796 * it got fsync'ed, from existing at both parents after
5797 * a log replay (and the old parent still existing).
5804 * mv /mnt/B/bar /mnt/A/bar
5805 * mv -T /mnt/A /mnt/B
5809 * If we ignore the old parent B which got deleted,
5810 * after a log replay we would have file bar linked
5811 * at both parents and the old parent B would still
5814 if (IS_ERR(dir_inode)) {
5815 ret = PTR_ERR(dir_inode);
5819 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
5820 btrfs_add_delayed_iput(dir_inode);
5825 ctx->log_new_dentries = false;
5826 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5827 LOG_INODE_ALL, ctx);
5828 if (!ret && ctx && ctx->log_new_dentries)
5829 ret = log_new_dir_dentries(trans, root,
5830 BTRFS_I(dir_inode), ctx);
5831 btrfs_add_delayed_iput(dir_inode);
5839 btrfs_free_path(path);
5843 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5844 struct btrfs_root *root,
5845 struct btrfs_path *path,
5846 struct btrfs_log_ctx *ctx)
5848 struct btrfs_key found_key;
5850 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5853 struct btrfs_fs_info *fs_info = root->fs_info;
5854 struct extent_buffer *leaf = path->nodes[0];
5855 int slot = path->slots[0];
5856 struct btrfs_key search_key;
5857 struct inode *inode;
5861 btrfs_release_path(path);
5863 ino = found_key.offset;
5865 search_key.objectid = found_key.offset;
5866 search_key.type = BTRFS_INODE_ITEM_KEY;
5867 search_key.offset = 0;
5868 inode = btrfs_iget(fs_info->sb, ino, root);
5870 return PTR_ERR(inode);
5872 if (BTRFS_I(inode)->generation >= trans->transid &&
5873 need_log_inode(trans, BTRFS_I(inode)))
5874 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5875 LOG_INODE_EXISTS, ctx);
5876 btrfs_add_delayed_iput(inode);
5880 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5883 search_key.type = BTRFS_INODE_REF_KEY;
5884 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5888 leaf = path->nodes[0];
5889 slot = path->slots[0];
5890 if (slot >= btrfs_header_nritems(leaf)) {
5891 ret = btrfs_next_leaf(root, path);
5896 leaf = path->nodes[0];
5897 slot = path->slots[0];
5900 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5901 if (found_key.objectid != search_key.objectid ||
5902 found_key.type != BTRFS_INODE_REF_KEY)
5908 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5909 struct btrfs_inode *inode,
5910 struct dentry *parent,
5911 struct btrfs_log_ctx *ctx)
5913 struct btrfs_root *root = inode->root;
5914 struct dentry *old_parent = NULL;
5915 struct super_block *sb = inode->vfs_inode.i_sb;
5919 if (!parent || d_really_is_negative(parent) ||
5923 inode = BTRFS_I(d_inode(parent));
5924 if (root != inode->root)
5927 if (inode->generation >= trans->transid &&
5928 need_log_inode(trans, inode)) {
5929 ret = btrfs_log_inode(trans, root, inode,
5930 LOG_INODE_EXISTS, ctx);
5934 if (IS_ROOT(parent))
5937 parent = dget_parent(parent);
5939 old_parent = parent;
5946 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5947 struct btrfs_inode *inode,
5948 struct dentry *parent,
5949 struct btrfs_log_ctx *ctx)
5951 struct btrfs_root *root = inode->root;
5952 const u64 ino = btrfs_ino(inode);
5953 struct btrfs_path *path;
5954 struct btrfs_key search_key;
5958 * For a single hard link case, go through a fast path that does not
5959 * need to iterate the fs/subvolume tree.
5961 if (inode->vfs_inode.i_nlink < 2)
5962 return log_new_ancestors_fast(trans, inode, parent, ctx);
5964 path = btrfs_alloc_path();
5968 search_key.objectid = ino;
5969 search_key.type = BTRFS_INODE_REF_KEY;
5970 search_key.offset = 0;
5972 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5979 struct extent_buffer *leaf = path->nodes[0];
5980 int slot = path->slots[0];
5981 struct btrfs_key found_key;
5983 if (slot >= btrfs_header_nritems(leaf)) {
5984 ret = btrfs_next_leaf(root, path);
5992 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5993 if (found_key.objectid != ino ||
5994 found_key.type > BTRFS_INODE_EXTREF_KEY)
5998 * Don't deal with extended references because they are rare
5999 * cases and too complex to deal with (we would need to keep
6000 * track of which subitem we are processing for each item in
6001 * this loop, etc). So just return some error to fallback to
6002 * a transaction commit.
6004 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6010 * Logging ancestors needs to do more searches on the fs/subvol
6011 * tree, so it releases the path as needed to avoid deadlocks.
6012 * Keep track of the last inode ref key and resume from that key
6013 * after logging all new ancestors for the current hard link.
6015 memcpy(&search_key, &found_key, sizeof(search_key));
6017 ret = log_new_ancestors(trans, root, path, ctx);
6020 btrfs_release_path(path);
6025 btrfs_free_path(path);
6030 * helper function around btrfs_log_inode to make sure newly created
6031 * parent directories also end up in the log. A minimal inode and backref
6032 * only logging is done of any parent directories that are older than
6033 * the last committed transaction
6035 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6036 struct btrfs_inode *inode,
6037 struct dentry *parent,
6039 struct btrfs_log_ctx *ctx)
6041 struct btrfs_root *root = inode->root;
6042 struct btrfs_fs_info *fs_info = root->fs_info;
6044 bool log_dentries = false;
6046 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6051 if (btrfs_root_refs(&root->root_item) == 0) {
6057 * Skip already logged inodes or inodes corresponding to tmpfiles
6058 * (since logging them is pointless, a link count of 0 means they
6059 * will never be accessible).
6061 if (btrfs_inode_in_log(inode, trans->transid) ||
6062 inode->vfs_inode.i_nlink == 0) {
6063 ret = BTRFS_NO_LOG_SYNC;
6067 ret = start_log_trans(trans, root, ctx);
6071 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6076 * for regular files, if its inode is already on disk, we don't
6077 * have to worry about the parents at all. This is because
6078 * we can use the last_unlink_trans field to record renames
6079 * and other fun in this file.
6081 if (S_ISREG(inode->vfs_inode.i_mode) &&
6082 inode->generation < trans->transid &&
6083 inode->last_unlink_trans < trans->transid) {
6088 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6089 log_dentries = true;
6092 * On unlink we must make sure all our current and old parent directory
6093 * inodes are fully logged. This is to prevent leaving dangling
6094 * directory index entries in directories that were our parents but are
6095 * not anymore. Not doing this results in old parent directory being
6096 * impossible to delete after log replay (rmdir will always fail with
6097 * error -ENOTEMPTY).
6103 * ln testdir/foo testdir/bar
6105 * unlink testdir/bar
6106 * xfs_io -c fsync testdir/foo
6108 * mount fs, triggers log replay
6110 * If we don't log the parent directory (testdir), after log replay the
6111 * directory still has an entry pointing to the file inode using the bar
6112 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6113 * the file inode has a link count of 1.
6119 * ln foo testdir/foo2
6120 * ln foo testdir/foo3
6122 * unlink testdir/foo3
6123 * xfs_io -c fsync foo
6125 * mount fs, triggers log replay
6127 * Similar as the first example, after log replay the parent directory
6128 * testdir still has an entry pointing to the inode file with name foo3
6129 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6130 * and has a link count of 2.
6132 if (inode->last_unlink_trans >= trans->transid) {
6133 ret = btrfs_log_all_parents(trans, inode, ctx);
6138 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6143 ret = log_new_dir_dentries(trans, root, inode, ctx);
6148 btrfs_set_log_full_commit(trans);
6153 btrfs_remove_log_ctx(root, ctx);
6154 btrfs_end_log_trans(root);
6160 * it is not safe to log dentry if the chunk root has added new
6161 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6162 * If this returns 1, you must commit the transaction to safely get your
6165 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6166 struct dentry *dentry,
6167 struct btrfs_log_ctx *ctx)
6169 struct dentry *parent = dget_parent(dentry);
6172 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6173 LOG_INODE_ALL, ctx);
6180 * should be called during mount to recover any replay any log trees
6183 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6186 struct btrfs_path *path;
6187 struct btrfs_trans_handle *trans;
6188 struct btrfs_key key;
6189 struct btrfs_key found_key;
6190 struct btrfs_root *log;
6191 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6192 struct walk_control wc = {
6193 .process_func = process_one_buffer,
6194 .stage = LOG_WALK_PIN_ONLY,
6197 path = btrfs_alloc_path();
6201 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6203 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6204 if (IS_ERR(trans)) {
6205 ret = PTR_ERR(trans);
6212 ret = walk_log_tree(trans, log_root_tree, &wc);
6214 btrfs_handle_fs_error(fs_info, ret,
6215 "Failed to pin buffers while recovering log root tree.");
6220 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6221 key.offset = (u64)-1;
6222 key.type = BTRFS_ROOT_ITEM_KEY;
6225 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6228 btrfs_handle_fs_error(fs_info, ret,
6229 "Couldn't find tree log root.");
6233 if (path->slots[0] == 0)
6237 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6239 btrfs_release_path(path);
6240 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6243 log = btrfs_read_tree_root(log_root_tree, &found_key);
6246 btrfs_handle_fs_error(fs_info, ret,
6247 "Couldn't read tree log root.");
6251 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6253 if (IS_ERR(wc.replay_dest)) {
6254 ret = PTR_ERR(wc.replay_dest);
6257 * We didn't find the subvol, likely because it was
6258 * deleted. This is ok, simply skip this log and go to
6261 * We need to exclude the root because we can't have
6262 * other log replays overwriting this log as we'll read
6263 * it back in a few more times. This will keep our
6264 * block from being modified, and we'll just bail for
6265 * each subsequent pass.
6268 ret = btrfs_pin_extent_for_log_replay(trans,
6271 btrfs_put_root(log);
6275 btrfs_handle_fs_error(fs_info, ret,
6276 "Couldn't read target root for tree log recovery.");
6280 wc.replay_dest->log_root = log;
6281 btrfs_record_root_in_trans(trans, wc.replay_dest);
6282 ret = walk_log_tree(trans, log, &wc);
6284 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6285 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6289 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6290 struct btrfs_root *root = wc.replay_dest;
6292 btrfs_release_path(path);
6295 * We have just replayed everything, and the highest
6296 * objectid of fs roots probably has changed in case
6297 * some inode_item's got replayed.
6299 * root->objectid_mutex is not acquired as log replay
6300 * could only happen during mount.
6302 ret = btrfs_init_root_free_objectid(root);
6305 wc.replay_dest->log_root = NULL;
6306 btrfs_put_root(wc.replay_dest);
6307 btrfs_put_root(log);
6312 if (found_key.offset == 0)
6314 key.offset = found_key.offset - 1;
6316 btrfs_release_path(path);
6318 /* step one is to pin it all, step two is to replay just inodes */
6321 wc.process_func = replay_one_buffer;
6322 wc.stage = LOG_WALK_REPLAY_INODES;
6325 /* step three is to replay everything */
6326 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6331 btrfs_free_path(path);
6333 /* step 4: commit the transaction, which also unpins the blocks */
6334 ret = btrfs_commit_transaction(trans);
6338 log_root_tree->log_root = NULL;
6339 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6340 btrfs_put_root(log_root_tree);
6345 btrfs_end_transaction(wc.trans);
6346 btrfs_free_path(path);
6351 * there are some corner cases where we want to force a full
6352 * commit instead of allowing a directory to be logged.
6354 * They revolve around files there were unlinked from the directory, and
6355 * this function updates the parent directory so that a full commit is
6356 * properly done if it is fsync'd later after the unlinks are done.
6358 * Must be called before the unlink operations (updates to the subvolume tree,
6359 * inodes, etc) are done.
6361 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6362 struct btrfs_inode *dir, struct btrfs_inode *inode,
6366 * when we're logging a file, if it hasn't been renamed
6367 * or unlinked, and its inode is fully committed on disk,
6368 * we don't have to worry about walking up the directory chain
6369 * to log its parents.
6371 * So, we use the last_unlink_trans field to put this transid
6372 * into the file. When the file is logged we check it and
6373 * don't log the parents if the file is fully on disk.
6375 mutex_lock(&inode->log_mutex);
6376 inode->last_unlink_trans = trans->transid;
6377 mutex_unlock(&inode->log_mutex);
6380 * if this directory was already logged any new
6381 * names for this file/dir will get recorded
6383 if (dir->logged_trans == trans->transid)
6387 * if the inode we're about to unlink was logged,
6388 * the log will be properly updated for any new names
6390 if (inode->logged_trans == trans->transid)
6394 * when renaming files across directories, if the directory
6395 * there we're unlinking from gets fsync'd later on, there's
6396 * no way to find the destination directory later and fsync it
6397 * properly. So, we have to be conservative and force commits
6398 * so the new name gets discovered.
6403 /* we can safely do the unlink without any special recording */
6407 mutex_lock(&dir->log_mutex);
6408 dir->last_unlink_trans = trans->transid;
6409 mutex_unlock(&dir->log_mutex);
6413 * Make sure that if someone attempts to fsync the parent directory of a deleted
6414 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6415 * that after replaying the log tree of the parent directory's root we will not
6416 * see the snapshot anymore and at log replay time we will not see any log tree
6417 * corresponding to the deleted snapshot's root, which could lead to replaying
6418 * it after replaying the log tree of the parent directory (which would replay
6419 * the snapshot delete operation).
6421 * Must be called before the actual snapshot destroy operation (updates to the
6422 * parent root and tree of tree roots trees, etc) are done.
6424 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6425 struct btrfs_inode *dir)
6427 mutex_lock(&dir->log_mutex);
6428 dir->last_unlink_trans = trans->transid;
6429 mutex_unlock(&dir->log_mutex);
6433 * Call this after adding a new name for a file and it will properly
6434 * update the log to reflect the new name.
6436 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6437 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6438 struct dentry *parent)
6440 struct btrfs_log_ctx ctx;
6443 * this will force the logging code to walk the dentry chain
6446 if (!S_ISDIR(inode->vfs_inode.i_mode))
6447 inode->last_unlink_trans = trans->transid;
6450 * if this inode hasn't been logged and directory we're renaming it
6451 * from hasn't been logged, we don't need to log it
6453 if (inode->logged_trans < trans->transid &&
6454 (!old_dir || old_dir->logged_trans < trans->transid))
6457 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6458 ctx.logging_new_name = true;
6460 * We don't care about the return value. If we fail to log the new name
6461 * then we know the next attempt to sync the log will fallback to a full
6462 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6463 * we don't need to worry about getting a log committed that has an
6464 * inconsistent state after a rename operation.
6466 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);