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 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1582 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1592 * Before we overwrite the inode reference item in the subvolume tree
1593 * with the item from the log tree, we must unlink all names from the
1594 * parent directory that are in the subvolume's tree inode reference
1595 * item, otherwise we end up with an inconsistent subvolume tree where
1596 * dir index entries exist for a name but there is no inode reference
1597 * item with the same name.
1599 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1604 /* finally write the back reference in the inode */
1605 ret = overwrite_item(trans, root, path, eb, slot, key);
1607 btrfs_release_path(path);
1614 static int count_inode_extrefs(struct btrfs_root *root,
1615 struct btrfs_inode *inode, struct btrfs_path *path)
1619 unsigned int nlink = 0;
1622 u64 inode_objectid = btrfs_ino(inode);
1625 struct btrfs_inode_extref *extref;
1626 struct extent_buffer *leaf;
1629 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1634 leaf = path->nodes[0];
1635 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1636 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1639 while (cur_offset < item_size) {
1640 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1641 name_len = btrfs_inode_extref_name_len(leaf, extref);
1645 cur_offset += name_len + sizeof(*extref);
1649 btrfs_release_path(path);
1651 btrfs_release_path(path);
1653 if (ret < 0 && ret != -ENOENT)
1658 static int count_inode_refs(struct btrfs_root *root,
1659 struct btrfs_inode *inode, struct btrfs_path *path)
1662 struct btrfs_key key;
1663 unsigned int nlink = 0;
1665 unsigned long ptr_end;
1667 u64 ino = btrfs_ino(inode);
1670 key.type = BTRFS_INODE_REF_KEY;
1671 key.offset = (u64)-1;
1674 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1678 if (path->slots[0] == 0)
1683 btrfs_item_key_to_cpu(path->nodes[0], &key,
1685 if (key.objectid != ino ||
1686 key.type != BTRFS_INODE_REF_KEY)
1688 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1689 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1691 while (ptr < ptr_end) {
1692 struct btrfs_inode_ref *ref;
1694 ref = (struct btrfs_inode_ref *)ptr;
1695 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1697 ptr = (unsigned long)(ref + 1) + name_len;
1701 if (key.offset == 0)
1703 if (path->slots[0] > 0) {
1708 btrfs_release_path(path);
1710 btrfs_release_path(path);
1716 * There are a few corners where the link count of the file can't
1717 * be properly maintained during replay. So, instead of adding
1718 * lots of complexity to the log code, we just scan the backrefs
1719 * for any file that has been through replay.
1721 * The scan will update the link count on the inode to reflect the
1722 * number of back refs found. If it goes down to zero, the iput
1723 * will free the inode.
1725 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1726 struct btrfs_root *root,
1727 struct inode *inode)
1729 struct btrfs_path *path;
1732 u64 ino = btrfs_ino(BTRFS_I(inode));
1734 path = btrfs_alloc_path();
1738 ret = count_inode_refs(root, BTRFS_I(inode), path);
1744 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1752 if (nlink != inode->i_nlink) {
1753 set_nlink(inode, nlink);
1754 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1758 BTRFS_I(inode)->index_cnt = (u64)-1;
1760 if (inode->i_nlink == 0) {
1761 if (S_ISDIR(inode->i_mode)) {
1762 ret = replay_dir_deletes(trans, root, NULL, path,
1767 ret = btrfs_insert_orphan_item(trans, root, ino);
1773 btrfs_free_path(path);
1777 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1778 struct btrfs_root *root,
1779 struct btrfs_path *path)
1782 struct btrfs_key key;
1783 struct inode *inode;
1785 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1786 key.type = BTRFS_ORPHAN_ITEM_KEY;
1787 key.offset = (u64)-1;
1789 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1795 if (path->slots[0] == 0)
1800 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1801 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1802 key.type != BTRFS_ORPHAN_ITEM_KEY)
1805 ret = btrfs_del_item(trans, root, path);
1809 btrfs_release_path(path);
1810 inode = read_one_inode(root, key.offset);
1816 ret = fixup_inode_link_count(trans, root, inode);
1822 * fixup on a directory may create new entries,
1823 * make sure we always look for the highset possible
1826 key.offset = (u64)-1;
1828 btrfs_release_path(path);
1834 * record a given inode in the fixup dir so we can check its link
1835 * count when replay is done. The link count is incremented here
1836 * so the inode won't go away until we check it
1838 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root,
1840 struct btrfs_path *path,
1843 struct btrfs_key key;
1845 struct inode *inode;
1847 inode = read_one_inode(root, objectid);
1851 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1852 key.type = BTRFS_ORPHAN_ITEM_KEY;
1853 key.offset = objectid;
1855 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1857 btrfs_release_path(path);
1859 if (!inode->i_nlink)
1860 set_nlink(inode, 1);
1863 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1864 } else if (ret == -EEXIST) {
1873 * when replaying the log for a directory, we only insert names
1874 * for inodes that actually exist. This means an fsync on a directory
1875 * does not implicitly fsync all the new files in it
1877 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1878 struct btrfs_root *root,
1879 u64 dirid, u64 index,
1880 char *name, int name_len,
1881 struct btrfs_key *location)
1883 struct inode *inode;
1887 inode = read_one_inode(root, location->objectid);
1891 dir = read_one_inode(root, dirid);
1897 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1898 name_len, 1, index);
1900 /* FIXME, put inode into FIXUP list */
1908 * take a single entry in a log directory item and replay it into
1911 * if a conflicting item exists in the subdirectory already,
1912 * the inode it points to is unlinked and put into the link count
1915 * If a name from the log points to a file or directory that does
1916 * not exist in the FS, it is skipped. fsyncs on directories
1917 * do not force down inodes inside that directory, just changes to the
1918 * names or unlinks in a directory.
1920 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1921 * non-existing inode) and 1 if the name was replayed.
1923 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1924 struct btrfs_root *root,
1925 struct btrfs_path *path,
1926 struct extent_buffer *eb,
1927 struct btrfs_dir_item *di,
1928 struct btrfs_key *key)
1932 struct btrfs_dir_item *dst_di;
1933 struct btrfs_key found_key;
1934 struct btrfs_key log_key;
1939 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1940 bool name_added = false;
1942 dir = read_one_inode(root, key->objectid);
1946 name_len = btrfs_dir_name_len(eb, di);
1947 name = kmalloc(name_len, GFP_NOFS);
1953 log_type = btrfs_dir_type(eb, di);
1954 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1957 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1958 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1963 btrfs_release_path(path);
1965 if (key->type == BTRFS_DIR_ITEM_KEY) {
1966 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1968 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1969 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1978 if (IS_ERR_OR_NULL(dst_di)) {
1979 /* we need a sequence number to insert, so we only
1980 * do inserts for the BTRFS_DIR_INDEX_KEY types
1982 if (key->type != BTRFS_DIR_INDEX_KEY)
1987 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1988 /* the existing item matches the logged item */
1989 if (found_key.objectid == log_key.objectid &&
1990 found_key.type == log_key.type &&
1991 found_key.offset == log_key.offset &&
1992 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1993 update_size = false;
1998 * don't drop the conflicting directory entry if the inode
1999 * for the new entry doesn't exist
2004 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2008 if (key->type == BTRFS_DIR_INDEX_KEY)
2011 btrfs_release_path(path);
2012 if (!ret && update_size) {
2013 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2014 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
2018 if (!ret && name_added)
2024 * Check if the inode reference exists in the log for the given name,
2025 * inode and parent inode
2027 found_key.objectid = log_key.objectid;
2028 found_key.type = BTRFS_INODE_REF_KEY;
2029 found_key.offset = key->objectid;
2030 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2034 /* The dentry will be added later. */
2036 update_size = false;
2040 found_key.objectid = log_key.objectid;
2041 found_key.type = BTRFS_INODE_EXTREF_KEY;
2042 found_key.offset = key->objectid;
2043 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2048 /* The dentry will be added later. */
2050 update_size = false;
2053 btrfs_release_path(path);
2054 ret = insert_one_name(trans, root, key->objectid, key->offset,
2055 name, name_len, &log_key);
2056 if (ret && ret != -ENOENT && ret != -EEXIST)
2060 update_size = false;
2066 * find all the names in a directory item and reconcile them into
2067 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2068 * one name in a directory item, but the same code gets used for
2069 * both directory index types
2071 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2072 struct btrfs_root *root,
2073 struct btrfs_path *path,
2074 struct extent_buffer *eb, int slot,
2075 struct btrfs_key *key)
2078 u32 item_size = btrfs_item_size_nr(eb, slot);
2079 struct btrfs_dir_item *di;
2082 unsigned long ptr_end;
2083 struct btrfs_path *fixup_path = NULL;
2085 ptr = btrfs_item_ptr_offset(eb, slot);
2086 ptr_end = ptr + item_size;
2087 while (ptr < ptr_end) {
2088 di = (struct btrfs_dir_item *)ptr;
2089 name_len = btrfs_dir_name_len(eb, di);
2090 ret = replay_one_name(trans, root, path, eb, di, key);
2093 ptr = (unsigned long)(di + 1);
2097 * If this entry refers to a non-directory (directories can not
2098 * have a link count > 1) and it was added in the transaction
2099 * that was not committed, make sure we fixup the link count of
2100 * the inode it the entry points to. Otherwise something like
2101 * the following would result in a directory pointing to an
2102 * inode with a wrong link that does not account for this dir
2110 * ln testdir/bar testdir/bar_link
2111 * ln testdir/foo testdir/foo_link
2112 * xfs_io -c "fsync" testdir/bar
2116 * mount fs, log replay happens
2118 * File foo would remain with a link count of 1 when it has two
2119 * entries pointing to it in the directory testdir. This would
2120 * make it impossible to ever delete the parent directory has
2121 * it would result in stale dentries that can never be deleted.
2123 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2124 struct btrfs_key di_key;
2127 fixup_path = btrfs_alloc_path();
2134 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2135 ret = link_to_fixup_dir(trans, root, fixup_path,
2142 btrfs_free_path(fixup_path);
2147 * directory replay has two parts. There are the standard directory
2148 * items in the log copied from the subvolume, and range items
2149 * created in the log while the subvolume was logged.
2151 * The range items tell us which parts of the key space the log
2152 * is authoritative for. During replay, if a key in the subvolume
2153 * directory is in a logged range item, but not actually in the log
2154 * that means it was deleted from the directory before the fsync
2155 * and should be removed.
2157 static noinline int find_dir_range(struct btrfs_root *root,
2158 struct btrfs_path *path,
2159 u64 dirid, int key_type,
2160 u64 *start_ret, u64 *end_ret)
2162 struct btrfs_key key;
2164 struct btrfs_dir_log_item *item;
2168 if (*start_ret == (u64)-1)
2171 key.objectid = dirid;
2172 key.type = key_type;
2173 key.offset = *start_ret;
2175 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2179 if (path->slots[0] == 0)
2184 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2186 if (key.type != key_type || key.objectid != dirid) {
2190 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2191 struct btrfs_dir_log_item);
2192 found_end = btrfs_dir_log_end(path->nodes[0], item);
2194 if (*start_ret >= key.offset && *start_ret <= found_end) {
2196 *start_ret = key.offset;
2197 *end_ret = found_end;
2202 /* check the next slot in the tree to see if it is a valid item */
2203 nritems = btrfs_header_nritems(path->nodes[0]);
2205 if (path->slots[0] >= nritems) {
2206 ret = btrfs_next_leaf(root, path);
2211 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2213 if (key.type != key_type || key.objectid != dirid) {
2217 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2218 struct btrfs_dir_log_item);
2219 found_end = btrfs_dir_log_end(path->nodes[0], item);
2220 *start_ret = key.offset;
2221 *end_ret = found_end;
2224 btrfs_release_path(path);
2229 * this looks for a given directory item in the log. If the directory
2230 * item is not in the log, the item is removed and the inode it points
2233 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2234 struct btrfs_root *root,
2235 struct btrfs_root *log,
2236 struct btrfs_path *path,
2237 struct btrfs_path *log_path,
2239 struct btrfs_key *dir_key)
2242 struct extent_buffer *eb;
2245 struct btrfs_dir_item *di;
2246 struct btrfs_dir_item *log_di;
2249 unsigned long ptr_end;
2251 struct inode *inode;
2252 struct btrfs_key location;
2255 eb = path->nodes[0];
2256 slot = path->slots[0];
2257 item_size = btrfs_item_size_nr(eb, slot);
2258 ptr = btrfs_item_ptr_offset(eb, slot);
2259 ptr_end = ptr + item_size;
2260 while (ptr < ptr_end) {
2261 di = (struct btrfs_dir_item *)ptr;
2262 name_len = btrfs_dir_name_len(eb, di);
2263 name = kmalloc(name_len, GFP_NOFS);
2268 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2271 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2272 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2275 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2276 log_di = btrfs_lookup_dir_index_item(trans, log,
2282 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2283 btrfs_dir_item_key_to_cpu(eb, di, &location);
2284 btrfs_release_path(path);
2285 btrfs_release_path(log_path);
2286 inode = read_one_inode(root, location.objectid);
2292 ret = link_to_fixup_dir(trans, root,
2293 path, location.objectid);
2301 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2302 BTRFS_I(inode), name, name_len);
2304 ret = btrfs_run_delayed_items(trans);
2310 /* there might still be more names under this key
2311 * check and repeat if required
2313 ret = btrfs_search_slot(NULL, root, dir_key, path,
2319 } else if (IS_ERR(log_di)) {
2321 return PTR_ERR(log_di);
2323 btrfs_release_path(log_path);
2326 ptr = (unsigned long)(di + 1);
2331 btrfs_release_path(path);
2332 btrfs_release_path(log_path);
2336 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2337 struct btrfs_root *root,
2338 struct btrfs_root *log,
2339 struct btrfs_path *path,
2342 struct btrfs_key search_key;
2343 struct btrfs_path *log_path;
2348 log_path = btrfs_alloc_path();
2352 search_key.objectid = ino;
2353 search_key.type = BTRFS_XATTR_ITEM_KEY;
2354 search_key.offset = 0;
2356 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2360 nritems = btrfs_header_nritems(path->nodes[0]);
2361 for (i = path->slots[0]; i < nritems; i++) {
2362 struct btrfs_key key;
2363 struct btrfs_dir_item *di;
2364 struct btrfs_dir_item *log_di;
2368 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2369 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2374 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2375 total_size = btrfs_item_size_nr(path->nodes[0], i);
2377 while (cur < total_size) {
2378 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2379 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2380 u32 this_len = sizeof(*di) + name_len + data_len;
2383 name = kmalloc(name_len, GFP_NOFS);
2388 read_extent_buffer(path->nodes[0], name,
2389 (unsigned long)(di + 1), name_len);
2391 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2393 btrfs_release_path(log_path);
2395 /* Doesn't exist in log tree, so delete it. */
2396 btrfs_release_path(path);
2397 di = btrfs_lookup_xattr(trans, root, path, ino,
2398 name, name_len, -1);
2405 ret = btrfs_delete_one_dir_name(trans, root,
2409 btrfs_release_path(path);
2414 if (IS_ERR(log_di)) {
2415 ret = PTR_ERR(log_di);
2419 di = (struct btrfs_dir_item *)((char *)di + this_len);
2422 ret = btrfs_next_leaf(root, path);
2428 btrfs_free_path(log_path);
2429 btrfs_release_path(path);
2435 * deletion replay happens before we copy any new directory items
2436 * out of the log or out of backreferences from inodes. It
2437 * scans the log to find ranges of keys that log is authoritative for,
2438 * and then scans the directory to find items in those ranges that are
2439 * not present in the log.
2441 * Anything we don't find in the log is unlinked and removed from the
2444 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2445 struct btrfs_root *root,
2446 struct btrfs_root *log,
2447 struct btrfs_path *path,
2448 u64 dirid, int del_all)
2452 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2454 struct btrfs_key dir_key;
2455 struct btrfs_key found_key;
2456 struct btrfs_path *log_path;
2459 dir_key.objectid = dirid;
2460 dir_key.type = BTRFS_DIR_ITEM_KEY;
2461 log_path = btrfs_alloc_path();
2465 dir = read_one_inode(root, dirid);
2466 /* it isn't an error if the inode isn't there, that can happen
2467 * because we replay the deletes before we copy in the inode item
2471 btrfs_free_path(log_path);
2479 range_end = (u64)-1;
2481 ret = find_dir_range(log, path, dirid, key_type,
2482 &range_start, &range_end);
2487 dir_key.offset = range_start;
2490 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2495 nritems = btrfs_header_nritems(path->nodes[0]);
2496 if (path->slots[0] >= nritems) {
2497 ret = btrfs_next_leaf(root, path);
2503 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2505 if (found_key.objectid != dirid ||
2506 found_key.type != dir_key.type)
2509 if (found_key.offset > range_end)
2512 ret = check_item_in_log(trans, root, log, path,
2517 if (found_key.offset == (u64)-1)
2519 dir_key.offset = found_key.offset + 1;
2521 btrfs_release_path(path);
2522 if (range_end == (u64)-1)
2524 range_start = range_end + 1;
2529 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2530 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2531 dir_key.type = BTRFS_DIR_INDEX_KEY;
2532 btrfs_release_path(path);
2536 btrfs_release_path(path);
2537 btrfs_free_path(log_path);
2543 * the process_func used to replay items from the log tree. This
2544 * gets called in two different stages. The first stage just looks
2545 * for inodes and makes sure they are all copied into the subvolume.
2547 * The second stage copies all the other item types from the log into
2548 * the subvolume. The two stage approach is slower, but gets rid of
2549 * lots of complexity around inodes referencing other inodes that exist
2550 * only in the log (references come from either directory items or inode
2553 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2554 struct walk_control *wc, u64 gen, int level)
2557 struct btrfs_path *path;
2558 struct btrfs_root *root = wc->replay_dest;
2559 struct btrfs_key key;
2563 ret = btrfs_read_buffer(eb, gen, level, NULL);
2567 level = btrfs_header_level(eb);
2572 path = btrfs_alloc_path();
2576 nritems = btrfs_header_nritems(eb);
2577 for (i = 0; i < nritems; i++) {
2578 btrfs_item_key_to_cpu(eb, &key, i);
2580 /* inode keys are done during the first stage */
2581 if (key.type == BTRFS_INODE_ITEM_KEY &&
2582 wc->stage == LOG_WALK_REPLAY_INODES) {
2583 struct btrfs_inode_item *inode_item;
2586 inode_item = btrfs_item_ptr(eb, i,
2587 struct btrfs_inode_item);
2589 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2590 * and never got linked before the fsync, skip it, as
2591 * replaying it is pointless since it would be deleted
2592 * later. We skip logging tmpfiles, but it's always
2593 * possible we are replaying a log created with a kernel
2594 * that used to log tmpfiles.
2596 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2597 wc->ignore_cur_inode = true;
2600 wc->ignore_cur_inode = false;
2602 ret = replay_xattr_deletes(wc->trans, root, log,
2603 path, key.objectid);
2606 mode = btrfs_inode_mode(eb, inode_item);
2607 if (S_ISDIR(mode)) {
2608 ret = replay_dir_deletes(wc->trans,
2609 root, log, path, key.objectid, 0);
2613 ret = overwrite_item(wc->trans, root, path,
2619 * Before replaying extents, truncate the inode to its
2620 * size. We need to do it now and not after log replay
2621 * because before an fsync we can have prealloc extents
2622 * added beyond the inode's i_size. If we did it after,
2623 * through orphan cleanup for example, we would drop
2624 * those prealloc extents just after replaying them.
2626 if (S_ISREG(mode)) {
2627 struct btrfs_drop_extents_args drop_args = { 0 };
2628 struct inode *inode;
2631 inode = read_one_inode(root, key.objectid);
2636 from = ALIGN(i_size_read(inode),
2637 root->fs_info->sectorsize);
2638 drop_args.start = from;
2639 drop_args.end = (u64)-1;
2640 drop_args.drop_cache = true;
2641 ret = btrfs_drop_extents(wc->trans, root,
2645 inode_sub_bytes(inode,
2646 drop_args.bytes_found);
2647 /* Update the inode's nbytes. */
2648 ret = btrfs_update_inode(wc->trans,
2649 root, BTRFS_I(inode));
2656 ret = link_to_fixup_dir(wc->trans, root,
2657 path, key.objectid);
2662 if (wc->ignore_cur_inode)
2665 if (key.type == BTRFS_DIR_INDEX_KEY &&
2666 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2667 ret = replay_one_dir_item(wc->trans, root, path,
2673 if (wc->stage < LOG_WALK_REPLAY_ALL)
2676 /* these keys are simply copied */
2677 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2678 ret = overwrite_item(wc->trans, root, path,
2682 } else if (key.type == BTRFS_INODE_REF_KEY ||
2683 key.type == BTRFS_INODE_EXTREF_KEY) {
2684 ret = add_inode_ref(wc->trans, root, log, path,
2686 if (ret && ret != -ENOENT)
2689 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2690 ret = replay_one_extent(wc->trans, root, path,
2694 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2695 ret = replay_one_dir_item(wc->trans, root, path,
2701 btrfs_free_path(path);
2706 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2708 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2710 struct btrfs_block_group *cache;
2712 cache = btrfs_lookup_block_group(fs_info, start);
2714 btrfs_err(fs_info, "unable to find block group for %llu", start);
2718 spin_lock(&cache->space_info->lock);
2719 spin_lock(&cache->lock);
2720 cache->reserved -= fs_info->nodesize;
2721 cache->space_info->bytes_reserved -= fs_info->nodesize;
2722 spin_unlock(&cache->lock);
2723 spin_unlock(&cache->space_info->lock);
2725 btrfs_put_block_group(cache);
2728 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2729 struct btrfs_root *root,
2730 struct btrfs_path *path, int *level,
2731 struct walk_control *wc)
2733 struct btrfs_fs_info *fs_info = root->fs_info;
2736 struct extent_buffer *next;
2737 struct extent_buffer *cur;
2741 while (*level > 0) {
2742 struct btrfs_key first_key;
2744 cur = path->nodes[*level];
2746 WARN_ON(btrfs_header_level(cur) != *level);
2748 if (path->slots[*level] >=
2749 btrfs_header_nritems(cur))
2752 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2753 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2754 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2755 blocksize = fs_info->nodesize;
2757 next = btrfs_find_create_tree_block(fs_info, bytenr,
2758 btrfs_header_owner(cur),
2761 return PTR_ERR(next);
2764 ret = wc->process_func(root, next, wc, ptr_gen,
2767 free_extent_buffer(next);
2771 path->slots[*level]++;
2773 ret = btrfs_read_buffer(next, ptr_gen,
2774 *level - 1, &first_key);
2776 free_extent_buffer(next);
2781 btrfs_tree_lock(next);
2782 btrfs_clean_tree_block(next);
2783 btrfs_wait_tree_block_writeback(next);
2784 btrfs_tree_unlock(next);
2785 ret = btrfs_pin_reserved_extent(trans,
2788 free_extent_buffer(next);
2791 btrfs_redirty_list_add(
2792 trans->transaction, next);
2794 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2795 clear_extent_buffer_dirty(next);
2796 unaccount_log_buffer(fs_info, bytenr);
2799 free_extent_buffer(next);
2802 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2804 free_extent_buffer(next);
2808 if (path->nodes[*level-1])
2809 free_extent_buffer(path->nodes[*level-1]);
2810 path->nodes[*level-1] = next;
2811 *level = btrfs_header_level(next);
2812 path->slots[*level] = 0;
2815 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2821 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2822 struct btrfs_root *root,
2823 struct btrfs_path *path, int *level,
2824 struct walk_control *wc)
2826 struct btrfs_fs_info *fs_info = root->fs_info;
2831 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2832 slot = path->slots[i];
2833 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2836 WARN_ON(*level == 0);
2839 ret = wc->process_func(root, path->nodes[*level], wc,
2840 btrfs_header_generation(path->nodes[*level]),
2846 struct extent_buffer *next;
2848 next = path->nodes[*level];
2851 btrfs_tree_lock(next);
2852 btrfs_clean_tree_block(next);
2853 btrfs_wait_tree_block_writeback(next);
2854 btrfs_tree_unlock(next);
2855 ret = btrfs_pin_reserved_extent(trans,
2856 path->nodes[*level]->start,
2857 path->nodes[*level]->len);
2861 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2862 clear_extent_buffer_dirty(next);
2864 unaccount_log_buffer(fs_info,
2865 path->nodes[*level]->start);
2868 free_extent_buffer(path->nodes[*level]);
2869 path->nodes[*level] = NULL;
2877 * drop the reference count on the tree rooted at 'snap'. This traverses
2878 * the tree freeing any blocks that have a ref count of zero after being
2881 static int walk_log_tree(struct btrfs_trans_handle *trans,
2882 struct btrfs_root *log, struct walk_control *wc)
2884 struct btrfs_fs_info *fs_info = log->fs_info;
2888 struct btrfs_path *path;
2891 path = btrfs_alloc_path();
2895 level = btrfs_header_level(log->node);
2897 path->nodes[level] = log->node;
2898 atomic_inc(&log->node->refs);
2899 path->slots[level] = 0;
2902 wret = walk_down_log_tree(trans, log, path, &level, wc);
2910 wret = walk_up_log_tree(trans, log, path, &level, wc);
2919 /* was the root node processed? if not, catch it here */
2920 if (path->nodes[orig_level]) {
2921 ret = wc->process_func(log, path->nodes[orig_level], wc,
2922 btrfs_header_generation(path->nodes[orig_level]),
2927 struct extent_buffer *next;
2929 next = path->nodes[orig_level];
2932 btrfs_tree_lock(next);
2933 btrfs_clean_tree_block(next);
2934 btrfs_wait_tree_block_writeback(next);
2935 btrfs_tree_unlock(next);
2936 ret = btrfs_pin_reserved_extent(trans,
2937 next->start, next->len);
2941 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2942 clear_extent_buffer_dirty(next);
2943 unaccount_log_buffer(fs_info, next->start);
2949 btrfs_free_path(path);
2954 * helper function to update the item for a given subvolumes log root
2955 * in the tree of log roots
2957 static int update_log_root(struct btrfs_trans_handle *trans,
2958 struct btrfs_root *log,
2959 struct btrfs_root_item *root_item)
2961 struct btrfs_fs_info *fs_info = log->fs_info;
2964 if (log->log_transid == 1) {
2965 /* insert root item on the first sync */
2966 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2967 &log->root_key, root_item);
2969 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2970 &log->root_key, root_item);
2975 static void wait_log_commit(struct btrfs_root *root, int transid)
2978 int index = transid % 2;
2981 * we only allow two pending log transactions at a time,
2982 * so we know that if ours is more than 2 older than the
2983 * current transaction, we're done
2986 prepare_to_wait(&root->log_commit_wait[index],
2987 &wait, TASK_UNINTERRUPTIBLE);
2989 if (!(root->log_transid_committed < transid &&
2990 atomic_read(&root->log_commit[index])))
2993 mutex_unlock(&root->log_mutex);
2995 mutex_lock(&root->log_mutex);
2997 finish_wait(&root->log_commit_wait[index], &wait);
3000 static void wait_for_writer(struct btrfs_root *root)
3005 prepare_to_wait(&root->log_writer_wait, &wait,
3006 TASK_UNINTERRUPTIBLE);
3007 if (!atomic_read(&root->log_writers))
3010 mutex_unlock(&root->log_mutex);
3012 mutex_lock(&root->log_mutex);
3014 finish_wait(&root->log_writer_wait, &wait);
3017 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3018 struct btrfs_log_ctx *ctx)
3023 mutex_lock(&root->log_mutex);
3024 list_del_init(&ctx->list);
3025 mutex_unlock(&root->log_mutex);
3029 * Invoked in log mutex context, or be sure there is no other task which
3030 * can access the list.
3032 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3033 int index, int error)
3035 struct btrfs_log_ctx *ctx;
3036 struct btrfs_log_ctx *safe;
3038 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3039 list_del_init(&ctx->list);
3040 ctx->log_ret = error;
3045 * btrfs_sync_log does sends a given tree log down to the disk and
3046 * updates the super blocks to record it. When this call is done,
3047 * you know that any inodes previously logged are safely on disk only
3050 * Any other return value means you need to call btrfs_commit_transaction.
3051 * Some of the edge cases for fsyncing directories that have had unlinks
3052 * or renames done in the past mean that sometimes the only safe
3053 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3054 * that has happened.
3056 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3057 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3063 struct btrfs_fs_info *fs_info = root->fs_info;
3064 struct btrfs_root *log = root->log_root;
3065 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3066 struct btrfs_root_item new_root_item;
3067 int log_transid = 0;
3068 struct btrfs_log_ctx root_log_ctx;
3069 struct blk_plug plug;
3073 mutex_lock(&root->log_mutex);
3074 log_transid = ctx->log_transid;
3075 if (root->log_transid_committed >= log_transid) {
3076 mutex_unlock(&root->log_mutex);
3077 return ctx->log_ret;
3080 index1 = log_transid % 2;
3081 if (atomic_read(&root->log_commit[index1])) {
3082 wait_log_commit(root, log_transid);
3083 mutex_unlock(&root->log_mutex);
3084 return ctx->log_ret;
3086 ASSERT(log_transid == root->log_transid);
3087 atomic_set(&root->log_commit[index1], 1);
3089 /* wait for previous tree log sync to complete */
3090 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3091 wait_log_commit(root, log_transid - 1);
3094 int batch = atomic_read(&root->log_batch);
3095 /* when we're on an ssd, just kick the log commit out */
3096 if (!btrfs_test_opt(fs_info, SSD) &&
3097 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3098 mutex_unlock(&root->log_mutex);
3099 schedule_timeout_uninterruptible(1);
3100 mutex_lock(&root->log_mutex);
3102 wait_for_writer(root);
3103 if (batch == atomic_read(&root->log_batch))
3107 /* bail out if we need to do a full commit */
3108 if (btrfs_need_log_full_commit(trans)) {
3110 mutex_unlock(&root->log_mutex);
3114 if (log_transid % 2 == 0)
3115 mark = EXTENT_DIRTY;
3119 /* we start IO on all the marked extents here, but we don't actually
3120 * wait for them until later.
3122 blk_start_plug(&plug);
3123 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3125 * -EAGAIN happens when someone, e.g., a concurrent transaction
3126 * commit, writes a dirty extent in this tree-log commit. This
3127 * concurrent write will create a hole writing out the extents,
3128 * and we cannot proceed on a zoned filesystem, requiring
3129 * sequential writing. While we can bail out to a full commit
3130 * here, but we can continue hoping the concurrent writing fills
3133 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3136 blk_finish_plug(&plug);
3137 btrfs_abort_transaction(trans, ret);
3138 btrfs_set_log_full_commit(trans);
3139 mutex_unlock(&root->log_mutex);
3144 * We _must_ update under the root->log_mutex in order to make sure we
3145 * have a consistent view of the log root we are trying to commit at
3148 * We _must_ copy this into a local copy, because we are not holding the
3149 * log_root_tree->log_mutex yet. This is important because when we
3150 * commit the log_root_tree we must have a consistent view of the
3151 * log_root_tree when we update the super block to point at the
3152 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3153 * with the commit and possibly point at the new block which we may not
3156 btrfs_set_root_node(&log->root_item, log->node);
3157 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3159 root->log_transid++;
3160 log->log_transid = root->log_transid;
3161 root->log_start_pid = 0;
3163 * IO has been started, blocks of the log tree have WRITTEN flag set
3164 * in their headers. new modifications of the log will be written to
3165 * new positions. so it's safe to allow log writers to go in.
3167 mutex_unlock(&root->log_mutex);
3169 if (btrfs_is_zoned(fs_info)) {
3170 mutex_lock(&fs_info->tree_root->log_mutex);
3171 if (!log_root_tree->node) {
3172 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3174 mutex_unlock(&fs_info->tree_root->log_mutex);
3178 mutex_unlock(&fs_info->tree_root->log_mutex);
3181 btrfs_init_log_ctx(&root_log_ctx, NULL);
3183 mutex_lock(&log_root_tree->log_mutex);
3185 index2 = log_root_tree->log_transid % 2;
3186 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3187 root_log_ctx.log_transid = log_root_tree->log_transid;
3190 * Now we are safe to update the log_root_tree because we're under the
3191 * log_mutex, and we're a current writer so we're holding the commit
3192 * open until we drop the log_mutex.
3194 ret = update_log_root(trans, log, &new_root_item);
3196 if (!list_empty(&root_log_ctx.list))
3197 list_del_init(&root_log_ctx.list);
3199 blk_finish_plug(&plug);
3200 btrfs_set_log_full_commit(trans);
3202 if (ret != -ENOSPC) {
3203 btrfs_abort_transaction(trans, ret);
3204 mutex_unlock(&log_root_tree->log_mutex);
3207 btrfs_wait_tree_log_extents(log, mark);
3208 mutex_unlock(&log_root_tree->log_mutex);
3213 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3214 blk_finish_plug(&plug);
3215 list_del_init(&root_log_ctx.list);
3216 mutex_unlock(&log_root_tree->log_mutex);
3217 ret = root_log_ctx.log_ret;
3221 index2 = root_log_ctx.log_transid % 2;
3222 if (atomic_read(&log_root_tree->log_commit[index2])) {
3223 blk_finish_plug(&plug);
3224 ret = btrfs_wait_tree_log_extents(log, mark);
3225 wait_log_commit(log_root_tree,
3226 root_log_ctx.log_transid);
3227 mutex_unlock(&log_root_tree->log_mutex);
3229 ret = root_log_ctx.log_ret;
3232 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3233 atomic_set(&log_root_tree->log_commit[index2], 1);
3235 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3236 wait_log_commit(log_root_tree,
3237 root_log_ctx.log_transid - 1);
3241 * now that we've moved on to the tree of log tree roots,
3242 * check the full commit flag again
3244 if (btrfs_need_log_full_commit(trans)) {
3245 blk_finish_plug(&plug);
3246 btrfs_wait_tree_log_extents(log, mark);
3247 mutex_unlock(&log_root_tree->log_mutex);
3249 goto out_wake_log_root;
3252 ret = btrfs_write_marked_extents(fs_info,
3253 &log_root_tree->dirty_log_pages,
3254 EXTENT_DIRTY | EXTENT_NEW);
3255 blk_finish_plug(&plug);
3257 * As described above, -EAGAIN indicates a hole in the extents. We
3258 * cannot wait for these write outs since the waiting cause a
3259 * deadlock. Bail out to the full commit instead.
3261 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3262 btrfs_set_log_full_commit(trans);
3263 btrfs_wait_tree_log_extents(log, mark);
3264 mutex_unlock(&log_root_tree->log_mutex);
3265 goto out_wake_log_root;
3267 btrfs_set_log_full_commit(trans);
3268 btrfs_abort_transaction(trans, ret);
3269 mutex_unlock(&log_root_tree->log_mutex);
3270 goto out_wake_log_root;
3272 ret = btrfs_wait_tree_log_extents(log, mark);
3274 ret = btrfs_wait_tree_log_extents(log_root_tree,
3275 EXTENT_NEW | EXTENT_DIRTY);
3277 btrfs_set_log_full_commit(trans);
3278 mutex_unlock(&log_root_tree->log_mutex);
3279 goto out_wake_log_root;
3282 log_root_start = log_root_tree->node->start;
3283 log_root_level = btrfs_header_level(log_root_tree->node);
3284 log_root_tree->log_transid++;
3285 mutex_unlock(&log_root_tree->log_mutex);
3288 * Here we are guaranteed that nobody is going to write the superblock
3289 * for the current transaction before us and that neither we do write
3290 * our superblock before the previous transaction finishes its commit
3291 * and writes its superblock, because:
3293 * 1) We are holding a handle on the current transaction, so no body
3294 * can commit it until we release the handle;
3296 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3297 * if the previous transaction is still committing, and hasn't yet
3298 * written its superblock, we wait for it to do it, because a
3299 * transaction commit acquires the tree_log_mutex when the commit
3300 * begins and releases it only after writing its superblock.
3302 mutex_lock(&fs_info->tree_log_mutex);
3305 * The previous transaction writeout phase could have failed, and thus
3306 * marked the fs in an error state. We must not commit here, as we
3307 * could have updated our generation in the super_for_commit and
3308 * writing the super here would result in transid mismatches. If there
3309 * is an error here just bail.
3311 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3313 btrfs_set_log_full_commit(trans);
3314 btrfs_abort_transaction(trans, ret);
3315 mutex_unlock(&fs_info->tree_log_mutex);
3316 goto out_wake_log_root;
3319 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3320 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3321 ret = write_all_supers(fs_info, 1);
3322 mutex_unlock(&fs_info->tree_log_mutex);
3324 btrfs_set_log_full_commit(trans);
3325 btrfs_abort_transaction(trans, ret);
3326 goto out_wake_log_root;
3330 * We know there can only be one task here, since we have not yet set
3331 * root->log_commit[index1] to 0 and any task attempting to sync the
3332 * log must wait for the previous log transaction to commit if it's
3333 * still in progress or wait for the current log transaction commit if
3334 * someone else already started it. We use <= and not < because the
3335 * first log transaction has an ID of 0.
3337 ASSERT(root->last_log_commit <= log_transid);
3338 root->last_log_commit = log_transid;
3341 mutex_lock(&log_root_tree->log_mutex);
3342 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3344 log_root_tree->log_transid_committed++;
3345 atomic_set(&log_root_tree->log_commit[index2], 0);
3346 mutex_unlock(&log_root_tree->log_mutex);
3349 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3350 * all the updates above are seen by the woken threads. It might not be
3351 * necessary, but proving that seems to be hard.
3353 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3355 mutex_lock(&root->log_mutex);
3356 btrfs_remove_all_log_ctxs(root, index1, ret);
3357 root->log_transid_committed++;
3358 atomic_set(&root->log_commit[index1], 0);
3359 mutex_unlock(&root->log_mutex);
3362 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3363 * all the updates above are seen by the woken threads. It might not be
3364 * necessary, but proving that seems to be hard.
3366 cond_wake_up(&root->log_commit_wait[index1]);
3370 static void free_log_tree(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *log)
3374 struct walk_control wc = {
3376 .process_func = process_one_buffer
3380 ret = walk_log_tree(trans, log, &wc);
3383 btrfs_abort_transaction(trans, ret);
3385 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3389 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3390 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3391 extent_io_tree_release(&log->log_csum_range);
3393 if (trans && log->node)
3394 btrfs_redirty_list_add(trans->transaction, log->node);
3395 btrfs_put_root(log);
3399 * free all the extents used by the tree log. This should be called
3400 * at commit time of the full transaction
3402 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3404 if (root->log_root) {
3405 free_log_tree(trans, root->log_root);
3406 root->log_root = NULL;
3407 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3412 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3413 struct btrfs_fs_info *fs_info)
3415 if (fs_info->log_root_tree) {
3416 free_log_tree(trans, fs_info->log_root_tree);
3417 fs_info->log_root_tree = NULL;
3418 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3424 * Check if an inode was logged in the current transaction. This may often
3425 * return some false positives, because logged_trans is an in memory only field,
3426 * not persisted anywhere. This is meant to be used in contexts where a false
3427 * positive has no functional consequences.
3429 static bool inode_logged(struct btrfs_trans_handle *trans,
3430 struct btrfs_inode *inode)
3432 if (inode->logged_trans == trans->transid)
3436 * The inode's logged_trans is always 0 when we load it (because it is
3437 * not persisted in the inode item or elsewhere). So if it is 0, the
3438 * inode was last modified in the current transaction then the inode may
3439 * have been logged before in the current transaction, then evicted and
3440 * loaded again in the current transaction - or may have never been logged
3441 * in the current transaction, but since we can not be sure, we have to
3442 * assume it was, otherwise our callers can leave an inconsistent log.
3444 if (inode->logged_trans == 0 &&
3445 inode->last_trans == trans->transid &&
3446 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3453 * If both a file and directory are logged, and unlinks or renames are
3454 * mixed in, we have a few interesting corners:
3456 * create file X in dir Y
3457 * link file X to X.link in dir Y
3459 * unlink file X but leave X.link
3462 * After a crash we would expect only X.link to exist. But file X
3463 * didn't get fsync'd again so the log has back refs for X and X.link.
3465 * We solve this by removing directory entries and inode backrefs from the
3466 * log when a file that was logged in the current transaction is
3467 * unlinked. Any later fsync will include the updated log entries, and
3468 * we'll be able to reconstruct the proper directory items from backrefs.
3470 * This optimizations allows us to avoid relogging the entire inode
3471 * or the entire directory.
3473 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *root,
3475 const char *name, int name_len,
3476 struct btrfs_inode *dir, u64 index)
3478 struct btrfs_root *log;
3479 struct btrfs_dir_item *di;
3480 struct btrfs_path *path;
3483 u64 dir_ino = btrfs_ino(dir);
3485 if (!inode_logged(trans, dir))
3488 ret = join_running_log_trans(root);
3492 mutex_lock(&dir->log_mutex);
3494 log = root->log_root;
3495 path = btrfs_alloc_path();
3501 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3502 name, name_len, -1);
3508 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3514 btrfs_release_path(path);
3515 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3516 index, name, name_len, -1);
3522 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3530 * We do not need to update the size field of the directory's inode item
3531 * because on log replay we update the field to reflect all existing
3532 * entries in the directory (see overwrite_item()).
3535 btrfs_free_path(path);
3537 mutex_unlock(&dir->log_mutex);
3538 if (err == -ENOSPC) {
3539 btrfs_set_log_full_commit(trans);
3541 } else if (err < 0 && err != -ENOENT) {
3542 /* ENOENT can be returned if the entry hasn't been fsynced yet */
3543 btrfs_abort_transaction(trans, err);
3546 btrfs_end_log_trans(root);
3551 /* see comments for btrfs_del_dir_entries_in_log */
3552 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3553 struct btrfs_root *root,
3554 const char *name, int name_len,
3555 struct btrfs_inode *inode, u64 dirid)
3557 struct btrfs_root *log;
3561 if (!inode_logged(trans, inode))
3564 ret = join_running_log_trans(root);
3567 log = root->log_root;
3568 mutex_lock(&inode->log_mutex);
3570 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3572 mutex_unlock(&inode->log_mutex);
3573 if (ret == -ENOSPC) {
3574 btrfs_set_log_full_commit(trans);
3576 } else if (ret < 0 && ret != -ENOENT)
3577 btrfs_abort_transaction(trans, ret);
3578 btrfs_end_log_trans(root);
3584 * creates a range item in the log for 'dirid'. first_offset and
3585 * last_offset tell us which parts of the key space the log should
3586 * be considered authoritative for.
3588 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3589 struct btrfs_root *log,
3590 struct btrfs_path *path,
3591 int key_type, u64 dirid,
3592 u64 first_offset, u64 last_offset)
3595 struct btrfs_key key;
3596 struct btrfs_dir_log_item *item;
3598 key.objectid = dirid;
3599 key.offset = first_offset;
3600 if (key_type == BTRFS_DIR_ITEM_KEY)
3601 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3603 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3604 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3608 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3609 struct btrfs_dir_log_item);
3610 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3611 btrfs_mark_buffer_dirty(path->nodes[0]);
3612 btrfs_release_path(path);
3617 * log all the items included in the current transaction for a given
3618 * directory. This also creates the range items in the log tree required
3619 * to replay anything deleted before the fsync
3621 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3622 struct btrfs_root *root, struct btrfs_inode *inode,
3623 struct btrfs_path *path,
3624 struct btrfs_path *dst_path, int key_type,
3625 struct btrfs_log_ctx *ctx,
3626 u64 min_offset, u64 *last_offset_ret)
3628 struct btrfs_key min_key;
3629 struct btrfs_root *log = root->log_root;
3630 struct extent_buffer *src;
3635 u64 first_offset = min_offset;
3636 u64 last_offset = (u64)-1;
3637 u64 ino = btrfs_ino(inode);
3639 log = root->log_root;
3641 min_key.objectid = ino;
3642 min_key.type = key_type;
3643 min_key.offset = min_offset;
3645 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3648 * we didn't find anything from this transaction, see if there
3649 * is anything at all
3651 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3652 min_key.objectid = ino;
3653 min_key.type = key_type;
3654 min_key.offset = (u64)-1;
3655 btrfs_release_path(path);
3656 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3658 btrfs_release_path(path);
3661 ret = btrfs_previous_item(root, path, ino, key_type);
3663 /* if ret == 0 there are items for this type,
3664 * create a range to tell us the last key of this type.
3665 * otherwise, there are no items in this directory after
3666 * *min_offset, and we create a range to indicate that.
3669 struct btrfs_key tmp;
3670 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3672 if (key_type == tmp.type)
3673 first_offset = max(min_offset, tmp.offset) + 1;
3678 /* go backward to find any previous key */
3679 ret = btrfs_previous_item(root, path, ino, key_type);
3681 struct btrfs_key tmp;
3682 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3683 if (key_type == tmp.type) {
3684 first_offset = tmp.offset;
3685 ret = overwrite_item(trans, log, dst_path,
3686 path->nodes[0], path->slots[0],
3694 btrfs_release_path(path);
3697 * Find the first key from this transaction again. See the note for
3698 * log_new_dir_dentries, if we're logging a directory recursively we
3699 * won't be holding its i_mutex, which means we can modify the directory
3700 * while we're logging it. If we remove an entry between our first
3701 * search and this search we'll not find the key again and can just
3705 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3710 * we have a block from this transaction, log every item in it
3711 * from our directory
3714 struct btrfs_key tmp;
3715 src = path->nodes[0];
3716 nritems = btrfs_header_nritems(src);
3717 for (i = path->slots[0]; i < nritems; i++) {
3718 struct btrfs_dir_item *di;
3720 btrfs_item_key_to_cpu(src, &min_key, i);
3722 if (min_key.objectid != ino || min_key.type != key_type)
3725 if (need_resched()) {
3726 btrfs_release_path(path);
3731 ret = overwrite_item(trans, log, dst_path, src, i,
3739 * We must make sure that when we log a directory entry,
3740 * the corresponding inode, after log replay, has a
3741 * matching link count. For example:
3747 * xfs_io -c "fsync" mydir
3749 * <mount fs and log replay>
3751 * Would result in a fsync log that when replayed, our
3752 * file inode would have a link count of 1, but we get
3753 * two directory entries pointing to the same inode.
3754 * After removing one of the names, it would not be
3755 * possible to remove the other name, which resulted
3756 * always in stale file handle errors, and would not
3757 * be possible to rmdir the parent directory, since
3758 * its i_size could never decrement to the value
3759 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3761 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3762 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3764 (btrfs_dir_transid(src, di) == trans->transid ||
3765 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3766 tmp.type != BTRFS_ROOT_ITEM_KEY)
3767 ctx->log_new_dentries = true;
3769 path->slots[0] = nritems;
3772 * look ahead to the next item and see if it is also
3773 * from this directory and from this transaction
3775 ret = btrfs_next_leaf(root, path);
3778 last_offset = (u64)-1;
3783 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3784 if (tmp.objectid != ino || tmp.type != key_type) {
3785 last_offset = (u64)-1;
3788 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3789 ret = overwrite_item(trans, log, dst_path,
3790 path->nodes[0], path->slots[0],
3795 last_offset = tmp.offset;
3800 btrfs_release_path(path);
3801 btrfs_release_path(dst_path);
3804 *last_offset_ret = last_offset;
3806 * insert the log range keys to indicate where the log
3809 ret = insert_dir_log_key(trans, log, path, key_type,
3810 ino, first_offset, last_offset);
3818 * logging directories is very similar to logging inodes, We find all the items
3819 * from the current transaction and write them to the log.
3821 * The recovery code scans the directory in the subvolume, and if it finds a
3822 * key in the range logged that is not present in the log tree, then it means
3823 * that dir entry was unlinked during the transaction.
3825 * In order for that scan to work, we must include one key smaller than
3826 * the smallest logged by this transaction and one key larger than the largest
3827 * key logged by this transaction.
3829 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3830 struct btrfs_root *root, struct btrfs_inode *inode,
3831 struct btrfs_path *path,
3832 struct btrfs_path *dst_path,
3833 struct btrfs_log_ctx *ctx)
3838 int key_type = BTRFS_DIR_ITEM_KEY;
3844 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3845 ctx, min_key, &max_key);
3848 if (max_key == (u64)-1)
3850 min_key = max_key + 1;
3853 if (key_type == BTRFS_DIR_ITEM_KEY) {
3854 key_type = BTRFS_DIR_INDEX_KEY;
3861 * a helper function to drop items from the log before we relog an
3862 * inode. max_key_type indicates the highest item type to remove.
3863 * This cannot be run for file data extents because it does not
3864 * free the extents they point to.
3866 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3867 struct btrfs_root *log,
3868 struct btrfs_path *path,
3869 u64 objectid, int max_key_type)
3872 struct btrfs_key key;
3873 struct btrfs_key found_key;
3876 key.objectid = objectid;
3877 key.type = max_key_type;
3878 key.offset = (u64)-1;
3881 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3882 BUG_ON(ret == 0); /* Logic error */
3886 if (path->slots[0] == 0)
3890 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3893 if (found_key.objectid != objectid)
3896 found_key.offset = 0;
3898 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3902 ret = btrfs_del_items(trans, log, path, start_slot,
3903 path->slots[0] - start_slot + 1);
3905 * If start slot isn't 0 then we don't need to re-search, we've
3906 * found the last guy with the objectid in this tree.
3908 if (ret || start_slot != 0)
3910 btrfs_release_path(path);
3912 btrfs_release_path(path);
3918 static void fill_inode_item(struct btrfs_trans_handle *trans,
3919 struct extent_buffer *leaf,
3920 struct btrfs_inode_item *item,
3921 struct inode *inode, int log_inode_only,
3924 struct btrfs_map_token token;
3927 btrfs_init_map_token(&token, leaf);
3929 if (log_inode_only) {
3930 /* set the generation to zero so the recover code
3931 * can tell the difference between an logging
3932 * just to say 'this inode exists' and a logging
3933 * to say 'update this inode with these values'
3935 btrfs_set_token_inode_generation(&token, item, 0);
3936 btrfs_set_token_inode_size(&token, item, logged_isize);
3938 btrfs_set_token_inode_generation(&token, item,
3939 BTRFS_I(inode)->generation);
3940 btrfs_set_token_inode_size(&token, item, inode->i_size);
3943 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3944 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3945 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3946 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3948 btrfs_set_token_timespec_sec(&token, &item->atime,
3949 inode->i_atime.tv_sec);
3950 btrfs_set_token_timespec_nsec(&token, &item->atime,
3951 inode->i_atime.tv_nsec);
3953 btrfs_set_token_timespec_sec(&token, &item->mtime,
3954 inode->i_mtime.tv_sec);
3955 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3956 inode->i_mtime.tv_nsec);
3958 btrfs_set_token_timespec_sec(&token, &item->ctime,
3959 inode->i_ctime.tv_sec);
3960 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3961 inode->i_ctime.tv_nsec);
3964 * We do not need to set the nbytes field, in fact during a fast fsync
3965 * its value may not even be correct, since a fast fsync does not wait
3966 * for ordered extent completion, which is where we update nbytes, it
3967 * only waits for writeback to complete. During log replay as we find
3968 * file extent items and replay them, we adjust the nbytes field of the
3969 * inode item in subvolume tree as needed (see overwrite_item()).
3972 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3973 btrfs_set_token_inode_transid(&token, item, trans->transid);
3974 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3975 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3976 BTRFS_I(inode)->ro_flags);
3977 btrfs_set_token_inode_flags(&token, item, flags);
3978 btrfs_set_token_inode_block_group(&token, item, 0);
3981 static int log_inode_item(struct btrfs_trans_handle *trans,
3982 struct btrfs_root *log, struct btrfs_path *path,
3983 struct btrfs_inode *inode, bool inode_item_dropped)
3985 struct btrfs_inode_item *inode_item;
3989 * If we are doing a fast fsync and the inode was logged before in the
3990 * current transaction, then we know the inode was previously logged and
3991 * it exists in the log tree. For performance reasons, in this case use
3992 * btrfs_search_slot() directly with ins_len set to 0 so that we never
3993 * attempt a write lock on the leaf's parent, which adds unnecessary lock
3994 * contention in case there are concurrent fsyncs for other inodes of the
3995 * same subvolume. Using btrfs_insert_empty_item() when the inode item
3996 * already exists can also result in unnecessarily splitting a leaf.
3998 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
3999 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4005 * This means it is the first fsync in the current transaction,
4006 * so the inode item is not in the log and we need to insert it.
4007 * We can never get -EEXIST because we are only called for a fast
4008 * fsync and in case an inode eviction happens after the inode was
4009 * logged before in the current transaction, when we load again
4010 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4011 * flags and set ->logged_trans to 0.
4013 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4014 sizeof(*inode_item));
4015 ASSERT(ret != -EEXIST);
4019 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4020 struct btrfs_inode_item);
4021 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4023 btrfs_release_path(path);
4027 static int log_csums(struct btrfs_trans_handle *trans,
4028 struct btrfs_inode *inode,
4029 struct btrfs_root *log_root,
4030 struct btrfs_ordered_sum *sums)
4032 const u64 lock_end = sums->bytenr + sums->len - 1;
4033 struct extent_state *cached_state = NULL;
4037 * If this inode was not used for reflink operations in the current
4038 * transaction with new extents, then do the fast path, no need to
4039 * worry about logging checksum items with overlapping ranges.
4041 if (inode->last_reflink_trans < trans->transid)
4042 return btrfs_csum_file_blocks(trans, log_root, sums);
4045 * Serialize logging for checksums. This is to avoid racing with the
4046 * same checksum being logged by another task that is logging another
4047 * file which happens to refer to the same extent as well. Such races
4048 * can leave checksum items in the log with overlapping ranges.
4050 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
4051 lock_end, &cached_state);
4055 * Due to extent cloning, we might have logged a csum item that covers a
4056 * subrange of a cloned extent, and later we can end up logging a csum
4057 * item for a larger subrange of the same extent or the entire range.
4058 * This would leave csum items in the log tree that cover the same range
4059 * and break the searches for checksums in the log tree, resulting in
4060 * some checksums missing in the fs/subvolume tree. So just delete (or
4061 * trim and adjust) any existing csum items in the log for this range.
4063 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4065 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4067 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
4073 static noinline int copy_items(struct btrfs_trans_handle *trans,
4074 struct btrfs_inode *inode,
4075 struct btrfs_path *dst_path,
4076 struct btrfs_path *src_path,
4077 int start_slot, int nr, int inode_only,
4080 struct btrfs_fs_info *fs_info = trans->fs_info;
4081 unsigned long src_offset;
4082 unsigned long dst_offset;
4083 struct btrfs_root *log = inode->root->log_root;
4084 struct btrfs_file_extent_item *extent;
4085 struct btrfs_inode_item *inode_item;
4086 struct extent_buffer *src = src_path->nodes[0];
4088 struct btrfs_key *ins_keys;
4092 struct list_head ordered_sums;
4093 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4095 INIT_LIST_HEAD(&ordered_sums);
4097 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4098 nr * sizeof(u32), GFP_NOFS);
4102 ins_sizes = (u32 *)ins_data;
4103 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4105 for (i = 0; i < nr; i++) {
4106 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4107 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4109 ret = btrfs_insert_empty_items(trans, log, dst_path,
4110 ins_keys, ins_sizes, nr);
4116 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4117 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4118 dst_path->slots[0]);
4120 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4122 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4123 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4125 struct btrfs_inode_item);
4126 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4128 inode_only == LOG_INODE_EXISTS,
4131 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4132 src_offset, ins_sizes[i]);
4135 /* take a reference on file data extents so that truncates
4136 * or deletes of this inode don't have to relog the inode
4139 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4142 extent = btrfs_item_ptr(src, start_slot + i,
4143 struct btrfs_file_extent_item);
4145 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4148 found_type = btrfs_file_extent_type(src, extent);
4149 if (found_type == BTRFS_FILE_EXTENT_REG) {
4151 ds = btrfs_file_extent_disk_bytenr(src,
4153 /* ds == 0 is a hole */
4157 dl = btrfs_file_extent_disk_num_bytes(src,
4159 cs = btrfs_file_extent_offset(src, extent);
4160 cl = btrfs_file_extent_num_bytes(src,
4162 if (btrfs_file_extent_compression(src,
4168 ret = btrfs_lookup_csums_range(
4170 ds + cs, ds + cs + cl - 1,
4178 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4179 btrfs_release_path(dst_path);
4183 * we have to do this after the loop above to avoid changing the
4184 * log tree while trying to change the log tree.
4186 while (!list_empty(&ordered_sums)) {
4187 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4188 struct btrfs_ordered_sum,
4191 ret = log_csums(trans, inode, log, sums);
4192 list_del(&sums->list);
4199 static int extent_cmp(void *priv, const struct list_head *a,
4200 const struct list_head *b)
4202 const struct extent_map *em1, *em2;
4204 em1 = list_entry(a, struct extent_map, list);
4205 em2 = list_entry(b, struct extent_map, list);
4207 if (em1->start < em2->start)
4209 else if (em1->start > em2->start)
4214 static int log_extent_csums(struct btrfs_trans_handle *trans,
4215 struct btrfs_inode *inode,
4216 struct btrfs_root *log_root,
4217 const struct extent_map *em,
4218 struct btrfs_log_ctx *ctx)
4220 struct btrfs_ordered_extent *ordered;
4223 u64 mod_start = em->mod_start;
4224 u64 mod_len = em->mod_len;
4225 LIST_HEAD(ordered_sums);
4228 if (inode->flags & BTRFS_INODE_NODATASUM ||
4229 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4230 em->block_start == EXTENT_MAP_HOLE)
4233 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4234 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4235 const u64 mod_end = mod_start + mod_len;
4236 struct btrfs_ordered_sum *sums;
4241 if (ordered_end <= mod_start)
4243 if (mod_end <= ordered->file_offset)
4247 * We are going to copy all the csums on this ordered extent, so
4248 * go ahead and adjust mod_start and mod_len in case this ordered
4249 * extent has already been logged.
4251 if (ordered->file_offset > mod_start) {
4252 if (ordered_end >= mod_end)
4253 mod_len = ordered->file_offset - mod_start;
4255 * If we have this case
4257 * |--------- logged extent ---------|
4258 * |----- ordered extent ----|
4260 * Just don't mess with mod_start and mod_len, we'll
4261 * just end up logging more csums than we need and it
4265 if (ordered_end < mod_end) {
4266 mod_len = mod_end - ordered_end;
4267 mod_start = ordered_end;
4274 * To keep us from looping for the above case of an ordered
4275 * extent that falls inside of the logged extent.
4277 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4280 list_for_each_entry(sums, &ordered->list, list) {
4281 ret = log_csums(trans, inode, log_root, sums);
4287 /* We're done, found all csums in the ordered extents. */
4291 /* If we're compressed we have to save the entire range of csums. */
4292 if (em->compress_type) {
4294 csum_len = max(em->block_len, em->orig_block_len);
4296 csum_offset = mod_start - em->start;
4300 /* block start is already adjusted for the file extent offset. */
4301 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4302 em->block_start + csum_offset,
4303 em->block_start + csum_offset +
4304 csum_len - 1, &ordered_sums, 0);
4308 while (!list_empty(&ordered_sums)) {
4309 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4310 struct btrfs_ordered_sum,
4313 ret = log_csums(trans, inode, log_root, sums);
4314 list_del(&sums->list);
4321 static int log_one_extent(struct btrfs_trans_handle *trans,
4322 struct btrfs_inode *inode, struct btrfs_root *root,
4323 const struct extent_map *em,
4324 struct btrfs_path *path,
4325 struct btrfs_log_ctx *ctx)
4327 struct btrfs_drop_extents_args drop_args = { 0 };
4328 struct btrfs_root *log = root->log_root;
4329 struct btrfs_file_extent_item *fi;
4330 struct extent_buffer *leaf;
4331 struct btrfs_map_token token;
4332 struct btrfs_key key;
4333 u64 extent_offset = em->start - em->orig_start;
4337 ret = log_extent_csums(trans, inode, log, em, ctx);
4341 drop_args.path = path;
4342 drop_args.start = em->start;
4343 drop_args.end = em->start + em->len;
4344 drop_args.replace_extent = true;
4345 drop_args.extent_item_size = sizeof(*fi);
4346 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4350 if (!drop_args.extent_inserted) {
4351 key.objectid = btrfs_ino(inode);
4352 key.type = BTRFS_EXTENT_DATA_KEY;
4353 key.offset = em->start;
4355 ret = btrfs_insert_empty_item(trans, log, path, &key,
4360 leaf = path->nodes[0];
4361 btrfs_init_map_token(&token, leaf);
4362 fi = btrfs_item_ptr(leaf, path->slots[0],
4363 struct btrfs_file_extent_item);
4365 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4366 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4367 btrfs_set_token_file_extent_type(&token, fi,
4368 BTRFS_FILE_EXTENT_PREALLOC);
4370 btrfs_set_token_file_extent_type(&token, fi,
4371 BTRFS_FILE_EXTENT_REG);
4373 block_len = max(em->block_len, em->orig_block_len);
4374 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4375 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4377 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4378 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4379 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4382 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4384 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4385 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4388 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4389 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4390 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4391 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4392 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4393 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4394 btrfs_mark_buffer_dirty(leaf);
4396 btrfs_release_path(path);
4402 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4403 * lose them after doing a fast fsync and replaying the log. We scan the
4404 * subvolume's root instead of iterating the inode's extent map tree because
4405 * otherwise we can log incorrect extent items based on extent map conversion.
4406 * That can happen due to the fact that extent maps are merged when they
4407 * are not in the extent map tree's list of modified extents.
4409 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4410 struct btrfs_inode *inode,
4411 struct btrfs_path *path)
4413 struct btrfs_root *root = inode->root;
4414 struct btrfs_key key;
4415 const u64 i_size = i_size_read(&inode->vfs_inode);
4416 const u64 ino = btrfs_ino(inode);
4417 struct btrfs_path *dst_path = NULL;
4418 bool dropped_extents = false;
4419 u64 truncate_offset = i_size;
4420 struct extent_buffer *leaf;
4426 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4430 key.type = BTRFS_EXTENT_DATA_KEY;
4431 key.offset = i_size;
4432 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4437 * We must check if there is a prealloc extent that starts before the
4438 * i_size and crosses the i_size boundary. This is to ensure later we
4439 * truncate down to the end of that extent and not to the i_size, as
4440 * otherwise we end up losing part of the prealloc extent after a log
4441 * replay and with an implicit hole if there is another prealloc extent
4442 * that starts at an offset beyond i_size.
4444 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4449 struct btrfs_file_extent_item *ei;
4451 leaf = path->nodes[0];
4452 slot = path->slots[0];
4453 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4455 if (btrfs_file_extent_type(leaf, ei) ==
4456 BTRFS_FILE_EXTENT_PREALLOC) {
4459 btrfs_item_key_to_cpu(leaf, &key, slot);
4460 extent_end = key.offset +
4461 btrfs_file_extent_num_bytes(leaf, ei);
4463 if (extent_end > i_size)
4464 truncate_offset = extent_end;
4471 leaf = path->nodes[0];
4472 slot = path->slots[0];
4474 if (slot >= btrfs_header_nritems(leaf)) {
4476 ret = copy_items(trans, inode, dst_path, path,
4477 start_slot, ins_nr, 1, 0);
4482 ret = btrfs_next_leaf(root, path);
4492 btrfs_item_key_to_cpu(leaf, &key, slot);
4493 if (key.objectid > ino)
4495 if (WARN_ON_ONCE(key.objectid < ino) ||
4496 key.type < BTRFS_EXTENT_DATA_KEY ||
4497 key.offset < i_size) {
4501 if (!dropped_extents) {
4503 * Avoid logging extent items logged in past fsync calls
4504 * and leading to duplicate keys in the log tree.
4507 ret = btrfs_truncate_inode_items(trans,
4509 inode, truncate_offset,
4510 BTRFS_EXTENT_DATA_KEY,
4512 } while (ret == -EAGAIN);
4515 dropped_extents = true;
4522 dst_path = btrfs_alloc_path();
4530 ret = copy_items(trans, inode, dst_path, path,
4531 start_slot, ins_nr, 1, 0);
4533 btrfs_release_path(path);
4534 btrfs_free_path(dst_path);
4538 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4539 struct btrfs_root *root,
4540 struct btrfs_inode *inode,
4541 struct btrfs_path *path,
4542 struct btrfs_log_ctx *ctx)
4544 struct btrfs_ordered_extent *ordered;
4545 struct btrfs_ordered_extent *tmp;
4546 struct extent_map *em, *n;
4547 struct list_head extents;
4548 struct extent_map_tree *tree = &inode->extent_tree;
4552 INIT_LIST_HEAD(&extents);
4554 write_lock(&tree->lock);
4556 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4557 list_del_init(&em->list);
4559 * Just an arbitrary number, this can be really CPU intensive
4560 * once we start getting a lot of extents, and really once we
4561 * have a bunch of extents we just want to commit since it will
4564 if (++num > 32768) {
4565 list_del_init(&tree->modified_extents);
4570 if (em->generation < trans->transid)
4573 /* We log prealloc extents beyond eof later. */
4574 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4575 em->start >= i_size_read(&inode->vfs_inode))
4578 /* Need a ref to keep it from getting evicted from cache */
4579 refcount_inc(&em->refs);
4580 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4581 list_add_tail(&em->list, &extents);
4585 list_sort(NULL, &extents, extent_cmp);
4587 while (!list_empty(&extents)) {
4588 em = list_entry(extents.next, struct extent_map, list);
4590 list_del_init(&em->list);
4593 * If we had an error we just need to delete everybody from our
4597 clear_em_logging(tree, em);
4598 free_extent_map(em);
4602 write_unlock(&tree->lock);
4604 ret = log_one_extent(trans, inode, root, em, path, ctx);
4605 write_lock(&tree->lock);
4606 clear_em_logging(tree, em);
4607 free_extent_map(em);
4609 WARN_ON(!list_empty(&extents));
4610 write_unlock(&tree->lock);
4612 btrfs_release_path(path);
4614 ret = btrfs_log_prealloc_extents(trans, inode, path);
4619 * We have logged all extents successfully, now make sure the commit of
4620 * the current transaction waits for the ordered extents to complete
4621 * before it commits and wipes out the log trees, otherwise we would
4622 * lose data if an ordered extents completes after the transaction
4623 * commits and a power failure happens after the transaction commit.
4625 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4626 list_del_init(&ordered->log_list);
4627 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4629 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4630 spin_lock_irq(&inode->ordered_tree.lock);
4631 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4632 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4633 atomic_inc(&trans->transaction->pending_ordered);
4635 spin_unlock_irq(&inode->ordered_tree.lock);
4637 btrfs_put_ordered_extent(ordered);
4643 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4644 struct btrfs_path *path, u64 *size_ret)
4646 struct btrfs_key key;
4649 key.objectid = btrfs_ino(inode);
4650 key.type = BTRFS_INODE_ITEM_KEY;
4653 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4656 } else if (ret > 0) {
4659 struct btrfs_inode_item *item;
4661 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4662 struct btrfs_inode_item);
4663 *size_ret = btrfs_inode_size(path->nodes[0], item);
4665 * If the in-memory inode's i_size is smaller then the inode
4666 * size stored in the btree, return the inode's i_size, so
4667 * that we get a correct inode size after replaying the log
4668 * when before a power failure we had a shrinking truncate
4669 * followed by addition of a new name (rename / new hard link).
4670 * Otherwise return the inode size from the btree, to avoid
4671 * data loss when replaying a log due to previously doing a
4672 * write that expands the inode's size and logging a new name
4673 * immediately after.
4675 if (*size_ret > inode->vfs_inode.i_size)
4676 *size_ret = inode->vfs_inode.i_size;
4679 btrfs_release_path(path);
4684 * At the moment we always log all xattrs. This is to figure out at log replay
4685 * time which xattrs must have their deletion replayed. If a xattr is missing
4686 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4687 * because if a xattr is deleted, the inode is fsynced and a power failure
4688 * happens, causing the log to be replayed the next time the fs is mounted,
4689 * we want the xattr to not exist anymore (same behaviour as other filesystems
4690 * with a journal, ext3/4, xfs, f2fs, etc).
4692 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4693 struct btrfs_root *root,
4694 struct btrfs_inode *inode,
4695 struct btrfs_path *path,
4696 struct btrfs_path *dst_path)
4699 struct btrfs_key key;
4700 const u64 ino = btrfs_ino(inode);
4703 bool found_xattrs = false;
4705 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4709 key.type = BTRFS_XATTR_ITEM_KEY;
4712 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4717 int slot = path->slots[0];
4718 struct extent_buffer *leaf = path->nodes[0];
4719 int nritems = btrfs_header_nritems(leaf);
4721 if (slot >= nritems) {
4723 ret = copy_items(trans, inode, dst_path, path,
4724 start_slot, ins_nr, 1, 0);
4729 ret = btrfs_next_leaf(root, path);
4737 btrfs_item_key_to_cpu(leaf, &key, slot);
4738 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4745 found_xattrs = true;
4749 ret = copy_items(trans, inode, dst_path, path,
4750 start_slot, ins_nr, 1, 0);
4756 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4762 * When using the NO_HOLES feature if we punched a hole that causes the
4763 * deletion of entire leafs or all the extent items of the first leaf (the one
4764 * that contains the inode item and references) we may end up not processing
4765 * any extents, because there are no leafs with a generation matching the
4766 * current transaction that have extent items for our inode. So we need to find
4767 * if any holes exist and then log them. We also need to log holes after any
4768 * truncate operation that changes the inode's size.
4770 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4771 struct btrfs_root *root,
4772 struct btrfs_inode *inode,
4773 struct btrfs_path *path)
4775 struct btrfs_fs_info *fs_info = root->fs_info;
4776 struct btrfs_key key;
4777 const u64 ino = btrfs_ino(inode);
4778 const u64 i_size = i_size_read(&inode->vfs_inode);
4779 u64 prev_extent_end = 0;
4782 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4786 key.type = BTRFS_EXTENT_DATA_KEY;
4789 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4794 struct extent_buffer *leaf = path->nodes[0];
4796 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4797 ret = btrfs_next_leaf(root, path);
4804 leaf = path->nodes[0];
4807 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4808 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4811 /* We have a hole, log it. */
4812 if (prev_extent_end < key.offset) {
4813 const u64 hole_len = key.offset - prev_extent_end;
4816 * Release the path to avoid deadlocks with other code
4817 * paths that search the root while holding locks on
4818 * leafs from the log root.
4820 btrfs_release_path(path);
4821 ret = btrfs_insert_file_extent(trans, root->log_root,
4822 ino, prev_extent_end, 0,
4823 0, hole_len, 0, hole_len,
4829 * Search for the same key again in the root. Since it's
4830 * an extent item and we are holding the inode lock, the
4831 * key must still exist. If it doesn't just emit warning
4832 * and return an error to fall back to a transaction
4835 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4838 if (WARN_ON(ret > 0))
4840 leaf = path->nodes[0];
4843 prev_extent_end = btrfs_file_extent_end(path);
4848 if (prev_extent_end < i_size) {
4851 btrfs_release_path(path);
4852 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4853 ret = btrfs_insert_file_extent(trans, root->log_root,
4854 ino, prev_extent_end, 0, 0,
4855 hole_len, 0, hole_len,
4865 * When we are logging a new inode X, check if it doesn't have a reference that
4866 * matches the reference from some other inode Y created in a past transaction
4867 * and that was renamed in the current transaction. If we don't do this, then at
4868 * log replay time we can lose inode Y (and all its files if it's a directory):
4871 * echo "hello world" > /mnt/x/foobar
4874 * mkdir /mnt/x # or touch /mnt/x
4875 * xfs_io -c fsync /mnt/x
4877 * mount fs, trigger log replay
4879 * After the log replay procedure, we would lose the first directory and all its
4880 * files (file foobar).
4881 * For the case where inode Y is not a directory we simply end up losing it:
4883 * echo "123" > /mnt/foo
4885 * mv /mnt/foo /mnt/bar
4886 * echo "abc" > /mnt/foo
4887 * xfs_io -c fsync /mnt/foo
4890 * We also need this for cases where a snapshot entry is replaced by some other
4891 * entry (file or directory) otherwise we end up with an unreplayable log due to
4892 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4893 * if it were a regular entry:
4896 * btrfs subvolume snapshot /mnt /mnt/x/snap
4897 * btrfs subvolume delete /mnt/x/snap
4900 * fsync /mnt/x or fsync some new file inside it
4903 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4904 * the same transaction.
4906 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4908 const struct btrfs_key *key,
4909 struct btrfs_inode *inode,
4910 u64 *other_ino, u64 *other_parent)
4913 struct btrfs_path *search_path;
4916 u32 item_size = btrfs_item_size_nr(eb, slot);
4918 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4920 search_path = btrfs_alloc_path();
4923 search_path->search_commit_root = 1;
4924 search_path->skip_locking = 1;
4926 while (cur_offset < item_size) {
4930 unsigned long name_ptr;
4931 struct btrfs_dir_item *di;
4933 if (key->type == BTRFS_INODE_REF_KEY) {
4934 struct btrfs_inode_ref *iref;
4936 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4937 parent = key->offset;
4938 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4939 name_ptr = (unsigned long)(iref + 1);
4940 this_len = sizeof(*iref) + this_name_len;
4942 struct btrfs_inode_extref *extref;
4944 extref = (struct btrfs_inode_extref *)(ptr +
4946 parent = btrfs_inode_extref_parent(eb, extref);
4947 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4948 name_ptr = (unsigned long)&extref->name;
4949 this_len = sizeof(*extref) + this_name_len;
4952 if (this_name_len > name_len) {
4955 new_name = krealloc(name, this_name_len, GFP_NOFS);
4960 name_len = this_name_len;
4964 read_extent_buffer(eb, name, name_ptr, this_name_len);
4965 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4966 parent, name, this_name_len, 0);
4967 if (di && !IS_ERR(di)) {
4968 struct btrfs_key di_key;
4970 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4972 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4973 if (di_key.objectid != key->objectid) {
4975 *other_ino = di_key.objectid;
4976 *other_parent = parent;
4984 } else if (IS_ERR(di)) {
4988 btrfs_release_path(search_path);
4990 cur_offset += this_len;
4994 btrfs_free_path(search_path);
4999 struct btrfs_ino_list {
5002 struct list_head list;
5005 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5006 struct btrfs_root *root,
5007 struct btrfs_path *path,
5008 struct btrfs_log_ctx *ctx,
5009 u64 ino, u64 parent)
5011 struct btrfs_ino_list *ino_elem;
5012 LIST_HEAD(inode_list);
5015 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5018 ino_elem->ino = ino;
5019 ino_elem->parent = parent;
5020 list_add_tail(&ino_elem->list, &inode_list);
5022 while (!list_empty(&inode_list)) {
5023 struct btrfs_fs_info *fs_info = root->fs_info;
5024 struct btrfs_key key;
5025 struct inode *inode;
5027 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
5029 ino = ino_elem->ino;
5030 parent = ino_elem->parent;
5031 list_del(&ino_elem->list);
5036 btrfs_release_path(path);
5038 inode = btrfs_iget(fs_info->sb, ino, root);
5040 * If the other inode that had a conflicting dir entry was
5041 * deleted in the current transaction, we need to log its parent
5044 if (IS_ERR(inode)) {
5045 ret = PTR_ERR(inode);
5046 if (ret == -ENOENT) {
5047 inode = btrfs_iget(fs_info->sb, parent, root);
5048 if (IS_ERR(inode)) {
5049 ret = PTR_ERR(inode);
5051 ret = btrfs_log_inode(trans, root,
5053 LOG_OTHER_INODE_ALL,
5055 btrfs_add_delayed_iput(inode);
5061 * If the inode was already logged skip it - otherwise we can
5062 * hit an infinite loop. Example:
5064 * From the commit root (previous transaction) we have the
5067 * inode 257 a directory
5068 * inode 258 with references "zz" and "zz_link" on inode 257
5069 * inode 259 with reference "a" on inode 257
5071 * And in the current (uncommitted) transaction we have:
5073 * inode 257 a directory, unchanged
5074 * inode 258 with references "a" and "a2" on inode 257
5075 * inode 259 with reference "zz_link" on inode 257
5076 * inode 261 with reference "zz" on inode 257
5078 * When logging inode 261 the following infinite loop could
5079 * happen if we don't skip already logged inodes:
5081 * - we detect inode 258 as a conflicting inode, with inode 261
5082 * on reference "zz", and log it;
5084 * - we detect inode 259 as a conflicting inode, with inode 258
5085 * on reference "a", and log it;
5087 * - we detect inode 258 as a conflicting inode, with inode 259
5088 * on reference "zz_link", and log it - again! After this we
5089 * repeat the above steps forever.
5091 spin_lock(&BTRFS_I(inode)->lock);
5093 * Check the inode's logged_trans only instead of
5094 * btrfs_inode_in_log(). This is because the last_log_commit of
5095 * the inode is not updated when we only log that it exists (see
5096 * btrfs_log_inode()).
5098 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5099 spin_unlock(&BTRFS_I(inode)->lock);
5100 btrfs_add_delayed_iput(inode);
5103 spin_unlock(&BTRFS_I(inode)->lock);
5105 * We are safe logging the other inode without acquiring its
5106 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5107 * are safe against concurrent renames of the other inode as
5108 * well because during a rename we pin the log and update the
5109 * log with the new name before we unpin it.
5111 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5112 LOG_OTHER_INODE, ctx);
5114 btrfs_add_delayed_iput(inode);
5119 key.type = BTRFS_INODE_REF_KEY;
5121 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5123 btrfs_add_delayed_iput(inode);
5128 struct extent_buffer *leaf = path->nodes[0];
5129 int slot = path->slots[0];
5131 u64 other_parent = 0;
5133 if (slot >= btrfs_header_nritems(leaf)) {
5134 ret = btrfs_next_leaf(root, path);
5137 } else if (ret > 0) {
5144 btrfs_item_key_to_cpu(leaf, &key, slot);
5145 if (key.objectid != ino ||
5146 (key.type != BTRFS_INODE_REF_KEY &&
5147 key.type != BTRFS_INODE_EXTREF_KEY)) {
5152 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5153 BTRFS_I(inode), &other_ino,
5158 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5163 ino_elem->ino = other_ino;
5164 ino_elem->parent = other_parent;
5165 list_add_tail(&ino_elem->list, &inode_list);
5170 btrfs_add_delayed_iput(inode);
5176 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5177 struct btrfs_inode *inode,
5178 struct btrfs_key *min_key,
5179 const struct btrfs_key *max_key,
5180 struct btrfs_path *path,
5181 struct btrfs_path *dst_path,
5182 const u64 logged_isize,
5183 const bool recursive_logging,
5184 const int inode_only,
5185 struct btrfs_log_ctx *ctx,
5186 bool *need_log_inode_item)
5188 struct btrfs_root *root = inode->root;
5189 int ins_start_slot = 0;
5194 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5202 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5203 if (min_key->objectid != max_key->objectid)
5205 if (min_key->type > max_key->type)
5208 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5209 *need_log_inode_item = false;
5211 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5212 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5213 inode->generation == trans->transid &&
5214 !recursive_logging) {
5216 u64 other_parent = 0;
5218 ret = btrfs_check_ref_name_override(path->nodes[0],
5219 path->slots[0], min_key, inode,
5220 &other_ino, &other_parent);
5223 } else if (ret > 0 && ctx &&
5224 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5229 ins_start_slot = path->slots[0];
5231 ret = copy_items(trans, inode, dst_path, path,
5232 ins_start_slot, ins_nr,
5233 inode_only, logged_isize);
5238 ret = log_conflicting_inodes(trans, root, path,
5239 ctx, other_ino, other_parent);
5242 btrfs_release_path(path);
5247 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5248 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5251 ret = copy_items(trans, inode, dst_path, path,
5253 ins_nr, inode_only, logged_isize);
5260 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5263 } else if (!ins_nr) {
5264 ins_start_slot = path->slots[0];
5269 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5270 ins_nr, inode_only, logged_isize);
5274 ins_start_slot = path->slots[0];
5277 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5278 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5283 ret = copy_items(trans, inode, dst_path, path,
5284 ins_start_slot, ins_nr, inode_only,
5290 btrfs_release_path(path);
5292 if (min_key->offset < (u64)-1) {
5294 } else if (min_key->type < max_key->type) {
5296 min_key->offset = 0;
5302 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5303 ins_nr, inode_only, logged_isize);
5308 /* log a single inode in the tree log.
5309 * At least one parent directory for this inode must exist in the tree
5310 * or be logged already.
5312 * Any items from this inode changed by the current transaction are copied
5313 * to the log tree. An extra reference is taken on any extents in this
5314 * file, allowing us to avoid a whole pile of corner cases around logging
5315 * blocks that have been removed from the tree.
5317 * See LOG_INODE_ALL and related defines for a description of what inode_only
5320 * This handles both files and directories.
5322 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5323 struct btrfs_root *root, struct btrfs_inode *inode,
5325 struct btrfs_log_ctx *ctx)
5327 struct btrfs_path *path;
5328 struct btrfs_path *dst_path;
5329 struct btrfs_key min_key;
5330 struct btrfs_key max_key;
5331 struct btrfs_root *log = root->log_root;
5334 bool fast_search = false;
5335 u64 ino = btrfs_ino(inode);
5336 struct extent_map_tree *em_tree = &inode->extent_tree;
5337 u64 logged_isize = 0;
5338 bool need_log_inode_item = true;
5339 bool xattrs_logged = false;
5340 bool recursive_logging = false;
5341 bool inode_item_dropped = true;
5343 path = btrfs_alloc_path();
5346 dst_path = btrfs_alloc_path();
5348 btrfs_free_path(path);
5352 min_key.objectid = ino;
5353 min_key.type = BTRFS_INODE_ITEM_KEY;
5356 max_key.objectid = ino;
5359 /* today the code can only do partial logging of directories */
5360 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5361 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5362 &inode->runtime_flags) &&
5363 inode_only >= LOG_INODE_EXISTS))
5364 max_key.type = BTRFS_XATTR_ITEM_KEY;
5366 max_key.type = (u8)-1;
5367 max_key.offset = (u64)-1;
5370 * Only run delayed items if we are a directory. We want to make sure
5371 * all directory indexes hit the fs/subvolume tree so we can find them
5372 * and figure out which index ranges have to be logged.
5374 * Otherwise commit the delayed inode only if the full sync flag is set,
5375 * as we want to make sure an up to date version is in the subvolume
5376 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5377 * it to the log tree. For a non full sync, we always log the inode item
5378 * based on the in-memory struct btrfs_inode which is always up to date.
5380 if (S_ISDIR(inode->vfs_inode.i_mode))
5381 ret = btrfs_commit_inode_delayed_items(trans, inode);
5382 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5383 ret = btrfs_commit_inode_delayed_inode(inode);
5386 btrfs_free_path(path);
5387 btrfs_free_path(dst_path);
5391 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5392 recursive_logging = true;
5393 if (inode_only == LOG_OTHER_INODE)
5394 inode_only = LOG_INODE_EXISTS;
5396 inode_only = LOG_INODE_ALL;
5397 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5399 mutex_lock(&inode->log_mutex);
5403 * This is for cases where logging a directory could result in losing a
5404 * a file after replaying the log. For example, if we move a file from a
5405 * directory A to a directory B, then fsync directory A, we have no way
5406 * to known the file was moved from A to B, so logging just A would
5407 * result in losing the file after a log replay.
5409 if (S_ISDIR(inode->vfs_inode.i_mode) &&
5410 inode_only == LOG_INODE_ALL &&
5411 inode->last_unlink_trans >= trans->transid) {
5412 btrfs_set_log_full_commit(trans);
5418 * a brute force approach to making sure we get the most uptodate
5419 * copies of everything.
5421 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5422 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5424 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5425 if (inode_only == LOG_INODE_EXISTS)
5426 max_key_type = BTRFS_XATTR_ITEM_KEY;
5427 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5429 if (inode_only == LOG_INODE_EXISTS) {
5431 * Make sure the new inode item we write to the log has
5432 * the same isize as the current one (if it exists).
5433 * This is necessary to prevent data loss after log
5434 * replay, and also to prevent doing a wrong expanding
5435 * truncate - for e.g. create file, write 4K into offset
5436 * 0, fsync, write 4K into offset 4096, add hard link,
5437 * fsync some other file (to sync log), power fail - if
5438 * we use the inode's current i_size, after log replay
5439 * we get a 8Kb file, with the last 4Kb extent as a hole
5440 * (zeroes), as if an expanding truncate happened,
5441 * instead of getting a file of 4Kb only.
5443 err = logged_inode_size(log, inode, path, &logged_isize);
5447 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5448 &inode->runtime_flags)) {
5449 if (inode_only == LOG_INODE_EXISTS) {
5450 max_key.type = BTRFS_XATTR_ITEM_KEY;
5451 ret = drop_objectid_items(trans, log, path, ino,
5454 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5455 &inode->runtime_flags);
5456 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5457 &inode->runtime_flags);
5459 ret = btrfs_truncate_inode_items(trans,
5460 log, inode, 0, 0, NULL);
5465 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5466 &inode->runtime_flags) ||
5467 inode_only == LOG_INODE_EXISTS) {
5468 if (inode_only == LOG_INODE_ALL)
5470 max_key.type = BTRFS_XATTR_ITEM_KEY;
5471 ret = drop_objectid_items(trans, log, path, ino,
5474 if (inode_only == LOG_INODE_ALL)
5476 inode_item_dropped = false;
5486 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5487 path, dst_path, logged_isize,
5488 recursive_logging, inode_only, ctx,
5489 &need_log_inode_item);
5493 btrfs_release_path(path);
5494 btrfs_release_path(dst_path);
5495 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5498 xattrs_logged = true;
5499 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5500 btrfs_release_path(path);
5501 btrfs_release_path(dst_path);
5502 err = btrfs_log_holes(trans, root, inode, path);
5507 btrfs_release_path(path);
5508 btrfs_release_path(dst_path);
5509 if (need_log_inode_item) {
5510 err = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
5514 * If we are doing a fast fsync and the inode was logged before
5515 * in this transaction, we don't need to log the xattrs because
5516 * they were logged before. If xattrs were added, changed or
5517 * deleted since the last time we logged the inode, then we have
5518 * already logged them because the inode had the runtime flag
5519 * BTRFS_INODE_COPY_EVERYTHING set.
5521 if (!xattrs_logged && inode->logged_trans < trans->transid) {
5522 err = btrfs_log_all_xattrs(trans, root, inode, path,
5526 btrfs_release_path(path);
5530 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5536 } else if (inode_only == LOG_INODE_ALL) {
5537 struct extent_map *em, *n;
5539 write_lock(&em_tree->lock);
5540 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5541 list_del_init(&em->list);
5542 write_unlock(&em_tree->lock);
5545 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5546 ret = log_directory_changes(trans, root, inode, path, dst_path,
5555 * If we are logging that an ancestor inode exists as part of logging a
5556 * new name from a link or rename operation, don't mark the inode as
5557 * logged - otherwise if an explicit fsync is made against an ancestor,
5558 * the fsync considers the inode in the log and doesn't sync the log,
5559 * resulting in the ancestor missing after a power failure unless the
5560 * log was synced as part of an fsync against any other unrelated inode.
5561 * So keep it simple for this case and just don't flag the ancestors as
5565 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5566 &inode->vfs_inode != ctx->inode)) {
5567 spin_lock(&inode->lock);
5568 inode->logged_trans = trans->transid;
5570 * Don't update last_log_commit if we logged that an inode exists.
5571 * We do this for two reasons:
5573 * 1) We might have had buffered writes to this inode that were
5574 * flushed and had their ordered extents completed in this
5575 * transaction, but we did not previously log the inode with
5576 * LOG_INODE_ALL. Later the inode was evicted and after that
5577 * it was loaded again and this LOG_INODE_EXISTS log operation
5578 * happened. We must make sure that if an explicit fsync against
5579 * the inode is performed later, it logs the new extents, an
5580 * updated inode item, etc, and syncs the log. The same logic
5581 * applies to direct IO writes instead of buffered writes.
5583 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
5584 * is logged with an i_size of 0 or whatever value was logged
5585 * before. If later the i_size of the inode is increased by a
5586 * truncate operation, the log is synced through an fsync of
5587 * some other inode and then finally an explicit fsync against
5588 * this inode is made, we must make sure this fsync logs the
5589 * inode with the new i_size, the hole between old i_size and
5590 * the new i_size, and syncs the log.
5592 if (inode_only != LOG_INODE_EXISTS)
5593 inode->last_log_commit = inode->last_sub_trans;
5594 spin_unlock(&inode->lock);
5597 mutex_unlock(&inode->log_mutex);
5599 btrfs_free_path(path);
5600 btrfs_free_path(dst_path);
5605 * Check if we need to log an inode. This is used in contexts where while
5606 * logging an inode we need to log another inode (either that it exists or in
5607 * full mode). This is used instead of btrfs_inode_in_log() because the later
5608 * requires the inode to be in the log and have the log transaction committed,
5609 * while here we do not care if the log transaction was already committed - our
5610 * caller will commit the log later - and we want to avoid logging an inode
5611 * multiple times when multiple tasks have joined the same log transaction.
5613 static bool need_log_inode(struct btrfs_trans_handle *trans,
5614 struct btrfs_inode *inode)
5617 * If this inode does not have new/updated/deleted xattrs since the last
5618 * time it was logged and is flagged as logged in the current transaction,
5619 * we can skip logging it. As for new/deleted names, those are updated in
5620 * the log by link/unlink/rename operations.
5621 * In case the inode was logged and then evicted and reloaded, its
5622 * logged_trans will be 0, in which case we have to fully log it since
5623 * logged_trans is a transient field, not persisted.
5625 if (inode->logged_trans == trans->transid &&
5626 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5632 struct btrfs_dir_list {
5634 struct list_head list;
5638 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5639 * details about the why it is needed.
5640 * This is a recursive operation - if an existing dentry corresponds to a
5641 * directory, that directory's new entries are logged too (same behaviour as
5642 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5643 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5644 * complains about the following circular lock dependency / possible deadlock:
5648 * lock(&type->i_mutex_dir_key#3/2);
5649 * lock(sb_internal#2);
5650 * lock(&type->i_mutex_dir_key#3/2);
5651 * lock(&sb->s_type->i_mutex_key#14);
5653 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5654 * sb_start_intwrite() in btrfs_start_transaction().
5655 * Not locking i_mutex of the inodes is still safe because:
5657 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5658 * that while logging the inode new references (names) are added or removed
5659 * from the inode, leaving the logged inode item with a link count that does
5660 * not match the number of logged inode reference items. This is fine because
5661 * at log replay time we compute the real number of links and correct the
5662 * link count in the inode item (see replay_one_buffer() and
5663 * link_to_fixup_dir());
5665 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5666 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5667 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5668 * has a size that doesn't match the sum of the lengths of all the logged
5669 * names. This does not result in a problem because if a dir_item key is
5670 * logged but its matching dir_index key is not logged, at log replay time we
5671 * don't use it to replay the respective name (see replay_one_name()). On the
5672 * other hand if only the dir_index key ends up being logged, the respective
5673 * name is added to the fs/subvol tree with both the dir_item and dir_index
5674 * keys created (see replay_one_name()).
5675 * The directory's inode item with a wrong i_size is not a problem as well,
5676 * since we don't use it at log replay time to set the i_size in the inode
5677 * item of the fs/subvol tree (see overwrite_item()).
5679 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5680 struct btrfs_root *root,
5681 struct btrfs_inode *start_inode,
5682 struct btrfs_log_ctx *ctx)
5684 struct btrfs_fs_info *fs_info = root->fs_info;
5685 struct btrfs_root *log = root->log_root;
5686 struct btrfs_path *path;
5687 LIST_HEAD(dir_list);
5688 struct btrfs_dir_list *dir_elem;
5691 path = btrfs_alloc_path();
5695 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5697 btrfs_free_path(path);
5700 dir_elem->ino = btrfs_ino(start_inode);
5701 list_add_tail(&dir_elem->list, &dir_list);
5703 while (!list_empty(&dir_list)) {
5704 struct extent_buffer *leaf;
5705 struct btrfs_key min_key;
5709 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5712 goto next_dir_inode;
5714 min_key.objectid = dir_elem->ino;
5715 min_key.type = BTRFS_DIR_ITEM_KEY;
5718 btrfs_release_path(path);
5719 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5721 goto next_dir_inode;
5722 } else if (ret > 0) {
5724 goto next_dir_inode;
5728 leaf = path->nodes[0];
5729 nritems = btrfs_header_nritems(leaf);
5730 for (i = path->slots[0]; i < nritems; i++) {
5731 struct btrfs_dir_item *di;
5732 struct btrfs_key di_key;
5733 struct inode *di_inode;
5734 struct btrfs_dir_list *new_dir_elem;
5735 int log_mode = LOG_INODE_EXISTS;
5738 btrfs_item_key_to_cpu(leaf, &min_key, i);
5739 if (min_key.objectid != dir_elem->ino ||
5740 min_key.type != BTRFS_DIR_ITEM_KEY)
5741 goto next_dir_inode;
5743 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5744 type = btrfs_dir_type(leaf, di);
5745 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5746 type != BTRFS_FT_DIR)
5748 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5749 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5752 btrfs_release_path(path);
5753 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5754 if (IS_ERR(di_inode)) {
5755 ret = PTR_ERR(di_inode);
5756 goto next_dir_inode;
5759 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5760 btrfs_add_delayed_iput(di_inode);
5764 ctx->log_new_dentries = false;
5765 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5766 log_mode = LOG_INODE_ALL;
5767 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5769 btrfs_add_delayed_iput(di_inode);
5771 goto next_dir_inode;
5772 if (ctx->log_new_dentries) {
5773 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5775 if (!new_dir_elem) {
5777 goto next_dir_inode;
5779 new_dir_elem->ino = di_key.objectid;
5780 list_add_tail(&new_dir_elem->list, &dir_list);
5785 ret = btrfs_next_leaf(log, path);
5787 goto next_dir_inode;
5788 } else if (ret > 0) {
5790 goto next_dir_inode;
5794 if (min_key.offset < (u64)-1) {
5799 list_del(&dir_elem->list);
5803 btrfs_free_path(path);
5807 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5808 struct btrfs_inode *inode,
5809 struct btrfs_log_ctx *ctx)
5811 struct btrfs_fs_info *fs_info = trans->fs_info;
5813 struct btrfs_path *path;
5814 struct btrfs_key key;
5815 struct btrfs_root *root = inode->root;
5816 const u64 ino = btrfs_ino(inode);
5818 path = btrfs_alloc_path();
5821 path->skip_locking = 1;
5822 path->search_commit_root = 1;
5825 key.type = BTRFS_INODE_REF_KEY;
5827 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5832 struct extent_buffer *leaf = path->nodes[0];
5833 int slot = path->slots[0];
5838 if (slot >= btrfs_header_nritems(leaf)) {
5839 ret = btrfs_next_leaf(root, path);
5847 btrfs_item_key_to_cpu(leaf, &key, slot);
5848 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5849 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5852 item_size = btrfs_item_size_nr(leaf, slot);
5853 ptr = btrfs_item_ptr_offset(leaf, slot);
5854 while (cur_offset < item_size) {
5855 struct btrfs_key inode_key;
5856 struct inode *dir_inode;
5858 inode_key.type = BTRFS_INODE_ITEM_KEY;
5859 inode_key.offset = 0;
5861 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5862 struct btrfs_inode_extref *extref;
5864 extref = (struct btrfs_inode_extref *)
5866 inode_key.objectid = btrfs_inode_extref_parent(
5868 cur_offset += sizeof(*extref);
5869 cur_offset += btrfs_inode_extref_name_len(leaf,
5872 inode_key.objectid = key.offset;
5873 cur_offset = item_size;
5876 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5879 * If the parent inode was deleted, return an error to
5880 * fallback to a transaction commit. This is to prevent
5881 * getting an inode that was moved from one parent A to
5882 * a parent B, got its former parent A deleted and then
5883 * it got fsync'ed, from existing at both parents after
5884 * a log replay (and the old parent still existing).
5891 * mv /mnt/B/bar /mnt/A/bar
5892 * mv -T /mnt/A /mnt/B
5896 * If we ignore the old parent B which got deleted,
5897 * after a log replay we would have file bar linked
5898 * at both parents and the old parent B would still
5901 if (IS_ERR(dir_inode)) {
5902 ret = PTR_ERR(dir_inode);
5906 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
5907 btrfs_add_delayed_iput(dir_inode);
5912 ctx->log_new_dentries = false;
5913 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5914 LOG_INODE_ALL, ctx);
5915 if (!ret && ctx && ctx->log_new_dentries)
5916 ret = log_new_dir_dentries(trans, root,
5917 BTRFS_I(dir_inode), ctx);
5918 btrfs_add_delayed_iput(dir_inode);
5926 btrfs_free_path(path);
5930 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5931 struct btrfs_root *root,
5932 struct btrfs_path *path,
5933 struct btrfs_log_ctx *ctx)
5935 struct btrfs_key found_key;
5937 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5940 struct btrfs_fs_info *fs_info = root->fs_info;
5941 struct extent_buffer *leaf = path->nodes[0];
5942 int slot = path->slots[0];
5943 struct btrfs_key search_key;
5944 struct inode *inode;
5948 btrfs_release_path(path);
5950 ino = found_key.offset;
5952 search_key.objectid = found_key.offset;
5953 search_key.type = BTRFS_INODE_ITEM_KEY;
5954 search_key.offset = 0;
5955 inode = btrfs_iget(fs_info->sb, ino, root);
5957 return PTR_ERR(inode);
5959 if (BTRFS_I(inode)->generation >= trans->transid &&
5960 need_log_inode(trans, BTRFS_I(inode)))
5961 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5962 LOG_INODE_EXISTS, ctx);
5963 btrfs_add_delayed_iput(inode);
5967 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5970 search_key.type = BTRFS_INODE_REF_KEY;
5971 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5975 leaf = path->nodes[0];
5976 slot = path->slots[0];
5977 if (slot >= btrfs_header_nritems(leaf)) {
5978 ret = btrfs_next_leaf(root, path);
5983 leaf = path->nodes[0];
5984 slot = path->slots[0];
5987 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5988 if (found_key.objectid != search_key.objectid ||
5989 found_key.type != BTRFS_INODE_REF_KEY)
5995 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5996 struct btrfs_inode *inode,
5997 struct dentry *parent,
5998 struct btrfs_log_ctx *ctx)
6000 struct btrfs_root *root = inode->root;
6001 struct dentry *old_parent = NULL;
6002 struct super_block *sb = inode->vfs_inode.i_sb;
6006 if (!parent || d_really_is_negative(parent) ||
6010 inode = BTRFS_I(d_inode(parent));
6011 if (root != inode->root)
6014 if (inode->generation >= trans->transid &&
6015 need_log_inode(trans, inode)) {
6016 ret = btrfs_log_inode(trans, root, inode,
6017 LOG_INODE_EXISTS, ctx);
6021 if (IS_ROOT(parent))
6024 parent = dget_parent(parent);
6026 old_parent = parent;
6033 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6034 struct btrfs_inode *inode,
6035 struct dentry *parent,
6036 struct btrfs_log_ctx *ctx)
6038 struct btrfs_root *root = inode->root;
6039 const u64 ino = btrfs_ino(inode);
6040 struct btrfs_path *path;
6041 struct btrfs_key search_key;
6045 * For a single hard link case, go through a fast path that does not
6046 * need to iterate the fs/subvolume tree.
6048 if (inode->vfs_inode.i_nlink < 2)
6049 return log_new_ancestors_fast(trans, inode, parent, ctx);
6051 path = btrfs_alloc_path();
6055 search_key.objectid = ino;
6056 search_key.type = BTRFS_INODE_REF_KEY;
6057 search_key.offset = 0;
6059 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6066 struct extent_buffer *leaf = path->nodes[0];
6067 int slot = path->slots[0];
6068 struct btrfs_key found_key;
6070 if (slot >= btrfs_header_nritems(leaf)) {
6071 ret = btrfs_next_leaf(root, path);
6079 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6080 if (found_key.objectid != ino ||
6081 found_key.type > BTRFS_INODE_EXTREF_KEY)
6085 * Don't deal with extended references because they are rare
6086 * cases and too complex to deal with (we would need to keep
6087 * track of which subitem we are processing for each item in
6088 * this loop, etc). So just return some error to fallback to
6089 * a transaction commit.
6091 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6097 * Logging ancestors needs to do more searches on the fs/subvol
6098 * tree, so it releases the path as needed to avoid deadlocks.
6099 * Keep track of the last inode ref key and resume from that key
6100 * after logging all new ancestors for the current hard link.
6102 memcpy(&search_key, &found_key, sizeof(search_key));
6104 ret = log_new_ancestors(trans, root, path, ctx);
6107 btrfs_release_path(path);
6112 btrfs_free_path(path);
6117 * helper function around btrfs_log_inode to make sure newly created
6118 * parent directories also end up in the log. A minimal inode and backref
6119 * only logging is done of any parent directories that are older than
6120 * the last committed transaction
6122 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6123 struct btrfs_inode *inode,
6124 struct dentry *parent,
6126 struct btrfs_log_ctx *ctx)
6128 struct btrfs_root *root = inode->root;
6129 struct btrfs_fs_info *fs_info = root->fs_info;
6131 bool log_dentries = false;
6133 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6138 if (btrfs_root_refs(&root->root_item) == 0) {
6144 * Skip already logged inodes or inodes corresponding to tmpfiles
6145 * (since logging them is pointless, a link count of 0 means they
6146 * will never be accessible).
6148 if ((btrfs_inode_in_log(inode, trans->transid) &&
6149 list_empty(&ctx->ordered_extents)) ||
6150 inode->vfs_inode.i_nlink == 0) {
6151 ret = BTRFS_NO_LOG_SYNC;
6155 ret = start_log_trans(trans, root, ctx);
6159 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6164 * for regular files, if its inode is already on disk, we don't
6165 * have to worry about the parents at all. This is because
6166 * we can use the last_unlink_trans field to record renames
6167 * and other fun in this file.
6169 if (S_ISREG(inode->vfs_inode.i_mode) &&
6170 inode->generation < trans->transid &&
6171 inode->last_unlink_trans < trans->transid) {
6176 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6177 log_dentries = true;
6180 * On unlink we must make sure all our current and old parent directory
6181 * inodes are fully logged. This is to prevent leaving dangling
6182 * directory index entries in directories that were our parents but are
6183 * not anymore. Not doing this results in old parent directory being
6184 * impossible to delete after log replay (rmdir will always fail with
6185 * error -ENOTEMPTY).
6191 * ln testdir/foo testdir/bar
6193 * unlink testdir/bar
6194 * xfs_io -c fsync testdir/foo
6196 * mount fs, triggers log replay
6198 * If we don't log the parent directory (testdir), after log replay the
6199 * directory still has an entry pointing to the file inode using the bar
6200 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6201 * the file inode has a link count of 1.
6207 * ln foo testdir/foo2
6208 * ln foo testdir/foo3
6210 * unlink testdir/foo3
6211 * xfs_io -c fsync foo
6213 * mount fs, triggers log replay
6215 * Similar as the first example, after log replay the parent directory
6216 * testdir still has an entry pointing to the inode file with name foo3
6217 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6218 * and has a link count of 2.
6220 if (inode->last_unlink_trans >= trans->transid) {
6221 ret = btrfs_log_all_parents(trans, inode, ctx);
6226 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6231 ret = log_new_dir_dentries(trans, root, inode, ctx);
6236 btrfs_set_log_full_commit(trans);
6241 btrfs_remove_log_ctx(root, ctx);
6242 btrfs_end_log_trans(root);
6248 * it is not safe to log dentry if the chunk root has added new
6249 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6250 * If this returns 1, you must commit the transaction to safely get your
6253 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6254 struct dentry *dentry,
6255 struct btrfs_log_ctx *ctx)
6257 struct dentry *parent = dget_parent(dentry);
6260 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6261 LOG_INODE_ALL, ctx);
6268 * should be called during mount to recover any replay any log trees
6271 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6274 struct btrfs_path *path;
6275 struct btrfs_trans_handle *trans;
6276 struct btrfs_key key;
6277 struct btrfs_key found_key;
6278 struct btrfs_root *log;
6279 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6280 struct walk_control wc = {
6281 .process_func = process_one_buffer,
6282 .stage = LOG_WALK_PIN_ONLY,
6285 path = btrfs_alloc_path();
6289 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6291 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6292 if (IS_ERR(trans)) {
6293 ret = PTR_ERR(trans);
6300 ret = walk_log_tree(trans, log_root_tree, &wc);
6302 btrfs_handle_fs_error(fs_info, ret,
6303 "Failed to pin buffers while recovering log root tree.");
6308 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6309 key.offset = (u64)-1;
6310 key.type = BTRFS_ROOT_ITEM_KEY;
6313 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6316 btrfs_handle_fs_error(fs_info, ret,
6317 "Couldn't find tree log root.");
6321 if (path->slots[0] == 0)
6325 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6327 btrfs_release_path(path);
6328 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6331 log = btrfs_read_tree_root(log_root_tree, &found_key);
6334 btrfs_handle_fs_error(fs_info, ret,
6335 "Couldn't read tree log root.");
6339 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6341 if (IS_ERR(wc.replay_dest)) {
6342 ret = PTR_ERR(wc.replay_dest);
6345 * We didn't find the subvol, likely because it was
6346 * deleted. This is ok, simply skip this log and go to
6349 * We need to exclude the root because we can't have
6350 * other log replays overwriting this log as we'll read
6351 * it back in a few more times. This will keep our
6352 * block from being modified, and we'll just bail for
6353 * each subsequent pass.
6356 ret = btrfs_pin_extent_for_log_replay(trans,
6359 btrfs_put_root(log);
6363 btrfs_handle_fs_error(fs_info, ret,
6364 "Couldn't read target root for tree log recovery.");
6368 wc.replay_dest->log_root = log;
6369 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
6371 /* The loop needs to continue due to the root refs */
6372 btrfs_handle_fs_error(fs_info, ret,
6373 "failed to record the log root in transaction");
6375 ret = walk_log_tree(trans, log, &wc);
6377 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6378 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6382 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6383 struct btrfs_root *root = wc.replay_dest;
6385 btrfs_release_path(path);
6388 * We have just replayed everything, and the highest
6389 * objectid of fs roots probably has changed in case
6390 * some inode_item's got replayed.
6392 * root->objectid_mutex is not acquired as log replay
6393 * could only happen during mount.
6395 ret = btrfs_init_root_free_objectid(root);
6398 wc.replay_dest->log_root = NULL;
6399 btrfs_put_root(wc.replay_dest);
6400 btrfs_put_root(log);
6405 if (found_key.offset == 0)
6407 key.offset = found_key.offset - 1;
6409 btrfs_release_path(path);
6411 /* step one is to pin it all, step two is to replay just inodes */
6414 wc.process_func = replay_one_buffer;
6415 wc.stage = LOG_WALK_REPLAY_INODES;
6418 /* step three is to replay everything */
6419 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6424 btrfs_free_path(path);
6426 /* step 4: commit the transaction, which also unpins the blocks */
6427 ret = btrfs_commit_transaction(trans);
6431 log_root_tree->log_root = NULL;
6432 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6433 btrfs_put_root(log_root_tree);
6438 btrfs_end_transaction(wc.trans);
6439 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6440 btrfs_free_path(path);
6445 * there are some corner cases where we want to force a full
6446 * commit instead of allowing a directory to be logged.
6448 * They revolve around files there were unlinked from the directory, and
6449 * this function updates the parent directory so that a full commit is
6450 * properly done if it is fsync'd later after the unlinks are done.
6452 * Must be called before the unlink operations (updates to the subvolume tree,
6453 * inodes, etc) are done.
6455 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6456 struct btrfs_inode *dir, struct btrfs_inode *inode,
6460 * when we're logging a file, if it hasn't been renamed
6461 * or unlinked, and its inode is fully committed on disk,
6462 * we don't have to worry about walking up the directory chain
6463 * to log its parents.
6465 * So, we use the last_unlink_trans field to put this transid
6466 * into the file. When the file is logged we check it and
6467 * don't log the parents if the file is fully on disk.
6469 mutex_lock(&inode->log_mutex);
6470 inode->last_unlink_trans = trans->transid;
6471 mutex_unlock(&inode->log_mutex);
6474 * if this directory was already logged any new
6475 * names for this file/dir will get recorded
6477 if (dir->logged_trans == trans->transid)
6481 * if the inode we're about to unlink was logged,
6482 * the log will be properly updated for any new names
6484 if (inode->logged_trans == trans->transid)
6488 * when renaming files across directories, if the directory
6489 * there we're unlinking from gets fsync'd later on, there's
6490 * no way to find the destination directory later and fsync it
6491 * properly. So, we have to be conservative and force commits
6492 * so the new name gets discovered.
6497 /* we can safely do the unlink without any special recording */
6501 mutex_lock(&dir->log_mutex);
6502 dir->last_unlink_trans = trans->transid;
6503 mutex_unlock(&dir->log_mutex);
6507 * Make sure that if someone attempts to fsync the parent directory of a deleted
6508 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6509 * that after replaying the log tree of the parent directory's root we will not
6510 * see the snapshot anymore and at log replay time we will not see any log tree
6511 * corresponding to the deleted snapshot's root, which could lead to replaying
6512 * it after replaying the log tree of the parent directory (which would replay
6513 * the snapshot delete operation).
6515 * Must be called before the actual snapshot destroy operation (updates to the
6516 * parent root and tree of tree roots trees, etc) are done.
6518 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6519 struct btrfs_inode *dir)
6521 mutex_lock(&dir->log_mutex);
6522 dir->last_unlink_trans = trans->transid;
6523 mutex_unlock(&dir->log_mutex);
6527 * Call this after adding a new name for a file and it will properly
6528 * update the log to reflect the new name.
6530 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6531 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6532 struct dentry *parent)
6534 struct btrfs_log_ctx ctx;
6537 * this will force the logging code to walk the dentry chain
6540 if (!S_ISDIR(inode->vfs_inode.i_mode))
6541 inode->last_unlink_trans = trans->transid;
6544 * if this inode hasn't been logged and directory we're renaming it
6545 * from hasn't been logged, we don't need to log it
6547 if (!inode_logged(trans, inode) &&
6548 (!old_dir || !inode_logged(trans, old_dir)))
6552 * If we are doing a rename (old_dir is not NULL) from a directory that
6553 * was previously logged, make sure the next log attempt on the directory
6554 * is not skipped and logs the inode again. This is because the log may
6555 * not currently be authoritative for a range including the old
6556 * BTRFS_DIR_ITEM_KEY and BTRFS_DIR_INDEX_KEY keys, so we want to make
6557 * sure after a log replay we do not end up with both the new and old
6558 * dentries around (in case the inode is a directory we would have a
6559 * directory with two hard links and 2 inode references for different
6560 * parents). The next log attempt of old_dir will happen at
6561 * btrfs_log_all_parents(), called through btrfs_log_inode_parent()
6562 * below, because we have previously set inode->last_unlink_trans to the
6563 * current transaction ID, either here or at btrfs_record_unlink_dir() in
6564 * case inode is a directory.
6567 old_dir->logged_trans = 0;
6569 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6570 ctx.logging_new_name = true;
6572 * We don't care about the return value. If we fail to log the new name
6573 * then we know the next attempt to sync the log will fallback to a full
6574 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6575 * we don't need to worry about getting a log committed that has an
6576 * inconsistent state after a rename operation.
6578 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);