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,
758 } else if (ret == 0) {
759 btrfs_init_generic_ref(&ref,
760 BTRFS_ADD_DELAYED_REF,
761 ins.objectid, ins.offset, 0);
762 btrfs_init_data_ref(&ref,
763 root->root_key.objectid,
764 key->objectid, offset);
765 ret = btrfs_inc_extent_ref(trans, &ref);
770 * insert the extent pointer in the extent
773 ret = btrfs_alloc_logged_file_extent(trans,
774 root->root_key.objectid,
775 key->objectid, offset, &ins);
779 btrfs_release_path(path);
781 if (btrfs_file_extent_compression(eb, item)) {
782 csum_start = ins.objectid;
783 csum_end = csum_start + ins.offset;
785 csum_start = ins.objectid +
786 btrfs_file_extent_offset(eb, item);
787 csum_end = csum_start +
788 btrfs_file_extent_num_bytes(eb, item);
791 ret = btrfs_lookup_csums_range(root->log_root,
792 csum_start, csum_end - 1,
797 * Now delete all existing cums in the csum root that
798 * cover our range. We do this because we can have an
799 * extent that is completely referenced by one file
800 * extent item and partially referenced by another
801 * file extent item (like after using the clone or
802 * extent_same ioctls). In this case if we end up doing
803 * the replay of the one that partially references the
804 * extent first, and we do not do the csum deletion
805 * below, we can get 2 csum items in the csum tree that
806 * overlap each other. For example, imagine our log has
807 * the two following file extent items:
809 * key (257 EXTENT_DATA 409600)
810 * extent data disk byte 12845056 nr 102400
811 * extent data offset 20480 nr 20480 ram 102400
813 * key (257 EXTENT_DATA 819200)
814 * extent data disk byte 12845056 nr 102400
815 * extent data offset 0 nr 102400 ram 102400
817 * Where the second one fully references the 100K extent
818 * that starts at disk byte 12845056, and the log tree
819 * has a single csum item that covers the entire range
822 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
824 * After the first file extent item is replayed, the
825 * csum tree gets the following csum item:
827 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
829 * Which covers the 20K sub-range starting at offset 20K
830 * of our extent. Now when we replay the second file
831 * extent item, if we do not delete existing csum items
832 * that cover any of its blocks, we end up getting two
833 * csum items in our csum tree that overlap each other:
835 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
836 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
838 * Which is a problem, because after this anyone trying
839 * to lookup up for the checksum of any block of our
840 * extent starting at an offset of 40K or higher, will
841 * end up looking at the second csum item only, which
842 * does not contain the checksum for any block starting
843 * at offset 40K or higher of our extent.
845 while (!list_empty(&ordered_sums)) {
846 struct btrfs_ordered_sum *sums;
847 sums = list_entry(ordered_sums.next,
848 struct btrfs_ordered_sum,
851 ret = btrfs_del_csums(trans,
856 ret = btrfs_csum_file_blocks(trans,
857 fs_info->csum_root, sums);
858 list_del(&sums->list);
864 btrfs_release_path(path);
866 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
867 /* inline extents are easy, we just overwrite them */
868 ret = overwrite_item(trans, root, path, eb, slot, key);
873 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
879 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
880 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
888 * when cleaning up conflicts between the directory names in the
889 * subvolume, directory names in the log and directory names in the
890 * inode back references, we may have to unlink inodes from directories.
892 * This is a helper function to do the unlink of a specific directory
895 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
896 struct btrfs_root *root,
897 struct btrfs_path *path,
898 struct btrfs_inode *dir,
899 struct btrfs_dir_item *di)
904 struct extent_buffer *leaf;
905 struct btrfs_key location;
908 leaf = path->nodes[0];
910 btrfs_dir_item_key_to_cpu(leaf, di, &location);
911 name_len = btrfs_dir_name_len(leaf, di);
912 name = kmalloc(name_len, GFP_NOFS);
916 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
917 btrfs_release_path(path);
919 inode = read_one_inode(root, location.objectid);
925 ret = link_to_fixup_dir(trans, root, path, location.objectid);
929 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
934 ret = btrfs_run_delayed_items(trans);
942 * See if a given name and sequence number found in an inode back reference are
943 * already in a directory and correctly point to this inode.
945 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
948 static noinline int inode_in_dir(struct btrfs_root *root,
949 struct btrfs_path *path,
950 u64 dirid, u64 objectid, u64 index,
951 const char *name, int name_len)
953 struct btrfs_dir_item *di;
954 struct btrfs_key location;
957 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
958 index, name, name_len, 0);
963 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
964 if (location.objectid != objectid)
970 btrfs_release_path(path);
971 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
976 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
977 if (location.objectid == objectid)
981 btrfs_release_path(path);
986 * helper function to check a log tree for a named back reference in
987 * an inode. This is used to decide if a back reference that is
988 * found in the subvolume conflicts with what we find in the log.
990 * inode backreferences may have multiple refs in a single item,
991 * during replay we process one reference at a time, and we don't
992 * want to delete valid links to a file from the subvolume if that
993 * link is also in the log.
995 static noinline int backref_in_log(struct btrfs_root *log,
996 struct btrfs_key *key,
998 const char *name, int namelen)
1000 struct btrfs_path *path;
1003 path = btrfs_alloc_path();
1007 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1010 } else if (ret == 1) {
1015 if (key->type == BTRFS_INODE_EXTREF_KEY)
1016 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1021 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1025 btrfs_free_path(path);
1029 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1030 struct btrfs_root *root,
1031 struct btrfs_path *path,
1032 struct btrfs_root *log_root,
1033 struct btrfs_inode *dir,
1034 struct btrfs_inode *inode,
1035 u64 inode_objectid, u64 parent_objectid,
1036 u64 ref_index, char *name, int namelen,
1041 int victim_name_len;
1042 struct extent_buffer *leaf;
1043 struct btrfs_dir_item *di;
1044 struct btrfs_key search_key;
1045 struct btrfs_inode_extref *extref;
1048 /* Search old style refs */
1049 search_key.objectid = inode_objectid;
1050 search_key.type = BTRFS_INODE_REF_KEY;
1051 search_key.offset = parent_objectid;
1052 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1054 struct btrfs_inode_ref *victim_ref;
1056 unsigned long ptr_end;
1058 leaf = path->nodes[0];
1060 /* are we trying to overwrite a back ref for the root directory
1061 * if so, just jump out, we're done
1063 if (search_key.objectid == search_key.offset)
1066 /* check all the names in this back reference to see
1067 * if they are in the log. if so, we allow them to stay
1068 * otherwise they must be unlinked as a conflict
1070 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1071 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1072 while (ptr < ptr_end) {
1073 victim_ref = (struct btrfs_inode_ref *)ptr;
1074 victim_name_len = btrfs_inode_ref_name_len(leaf,
1076 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1080 read_extent_buffer(leaf, victim_name,
1081 (unsigned long)(victim_ref + 1),
1084 ret = backref_in_log(log_root, &search_key,
1085 parent_objectid, victim_name,
1091 inc_nlink(&inode->vfs_inode);
1092 btrfs_release_path(path);
1094 ret = btrfs_unlink_inode(trans, root, dir, inode,
1095 victim_name, victim_name_len);
1099 ret = btrfs_run_delayed_items(trans);
1107 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1111 * NOTE: we have searched root tree and checked the
1112 * corresponding ref, it does not need to check again.
1116 btrfs_release_path(path);
1118 /* Same search but for extended refs */
1119 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1120 inode_objectid, parent_objectid, 0,
1122 if (!IS_ERR_OR_NULL(extref)) {
1126 struct inode *victim_parent;
1128 leaf = path->nodes[0];
1130 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1131 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1133 while (cur_offset < item_size) {
1134 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1136 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1138 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1141 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1144 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1147 search_key.objectid = inode_objectid;
1148 search_key.type = BTRFS_INODE_EXTREF_KEY;
1149 search_key.offset = btrfs_extref_hash(parent_objectid,
1152 ret = backref_in_log(log_root, &search_key,
1153 parent_objectid, victim_name,
1160 victim_parent = read_one_inode(root,
1162 if (victim_parent) {
1163 inc_nlink(&inode->vfs_inode);
1164 btrfs_release_path(path);
1166 ret = btrfs_unlink_inode(trans, root,
1167 BTRFS_I(victim_parent),
1172 ret = btrfs_run_delayed_items(
1175 iput(victim_parent);
1184 cur_offset += victim_name_len + sizeof(*extref);
1188 btrfs_release_path(path);
1190 /* look for a conflicting sequence number */
1191 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1192 ref_index, name, namelen, 0);
1196 ret = drop_one_dir_item(trans, root, path, dir, di);
1200 btrfs_release_path(path);
1202 /* look for a conflicting name */
1203 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1208 ret = drop_one_dir_item(trans, root, path, dir, di);
1212 btrfs_release_path(path);
1217 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1218 u32 *namelen, char **name, u64 *index,
1219 u64 *parent_objectid)
1221 struct btrfs_inode_extref *extref;
1223 extref = (struct btrfs_inode_extref *)ref_ptr;
1225 *namelen = btrfs_inode_extref_name_len(eb, extref);
1226 *name = kmalloc(*namelen, GFP_NOFS);
1230 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1234 *index = btrfs_inode_extref_index(eb, extref);
1235 if (parent_objectid)
1236 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1241 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1242 u32 *namelen, char **name, u64 *index)
1244 struct btrfs_inode_ref *ref;
1246 ref = (struct btrfs_inode_ref *)ref_ptr;
1248 *namelen = btrfs_inode_ref_name_len(eb, ref);
1249 *name = kmalloc(*namelen, GFP_NOFS);
1253 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1256 *index = btrfs_inode_ref_index(eb, ref);
1262 * Take an inode reference item from the log tree and iterate all names from the
1263 * inode reference item in the subvolume tree with the same key (if it exists).
1264 * For any name that is not in the inode reference item from the log tree, do a
1265 * proper unlink of that name (that is, remove its entry from the inode
1266 * reference item and both dir index keys).
1268 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1269 struct btrfs_root *root,
1270 struct btrfs_path *path,
1271 struct btrfs_inode *inode,
1272 struct extent_buffer *log_eb,
1274 struct btrfs_key *key)
1277 unsigned long ref_ptr;
1278 unsigned long ref_end;
1279 struct extent_buffer *eb;
1282 btrfs_release_path(path);
1283 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1291 eb = path->nodes[0];
1292 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1293 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1294 while (ref_ptr < ref_end) {
1299 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1300 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1303 parent_id = key->offset;
1304 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1310 if (key->type == BTRFS_INODE_EXTREF_KEY)
1311 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1315 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1321 btrfs_release_path(path);
1322 dir = read_one_inode(root, parent_id);
1328 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1329 inode, name, namelen);
1333 * Whenever we need to check if a name exists or not, we
1334 * check the subvolume tree. So after an unlink we must
1335 * run delayed items, so that future checks for a name
1336 * during log replay see that the name does not exists
1340 ret = btrfs_run_delayed_items(trans);
1348 if (key->type == BTRFS_INODE_EXTREF_KEY)
1349 ref_ptr += sizeof(struct btrfs_inode_extref);
1351 ref_ptr += sizeof(struct btrfs_inode_ref);
1355 btrfs_release_path(path);
1359 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1360 const u8 ref_type, const char *name,
1363 struct btrfs_key key;
1364 struct btrfs_path *path;
1365 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1368 path = btrfs_alloc_path();
1372 key.objectid = btrfs_ino(BTRFS_I(inode));
1373 key.type = ref_type;
1374 if (key.type == BTRFS_INODE_REF_KEY)
1375 key.offset = parent_id;
1377 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1379 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1386 if (key.type == BTRFS_INODE_EXTREF_KEY)
1387 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1388 path->slots[0], parent_id, name, namelen);
1390 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1394 btrfs_free_path(path);
1398 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1399 struct inode *dir, struct inode *inode, const char *name,
1400 int namelen, u64 ref_index)
1402 struct btrfs_dir_item *dir_item;
1403 struct btrfs_key key;
1404 struct btrfs_path *path;
1405 struct inode *other_inode = NULL;
1408 path = btrfs_alloc_path();
1412 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1413 btrfs_ino(BTRFS_I(dir)),
1416 btrfs_release_path(path);
1418 } else if (IS_ERR(dir_item)) {
1419 ret = PTR_ERR(dir_item);
1424 * Our inode's dentry collides with the dentry of another inode which is
1425 * in the log but not yet processed since it has a higher inode number.
1426 * So delete that other dentry.
1428 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1429 btrfs_release_path(path);
1430 other_inode = read_one_inode(root, key.objectid);
1435 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1440 * If we dropped the link count to 0, bump it so that later the iput()
1441 * on the inode will not free it. We will fixup the link count later.
1443 if (other_inode->i_nlink == 0)
1444 inc_nlink(other_inode);
1446 ret = btrfs_run_delayed_items(trans);
1450 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1451 name, namelen, 0, ref_index);
1454 btrfs_free_path(path);
1460 * replay one inode back reference item found in the log tree.
1461 * eb, slot and key refer to the buffer and key found in the log tree.
1462 * root is the destination we are replaying into, and path is for temp
1463 * use by this function. (it should be released on return).
1465 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1466 struct btrfs_root *root,
1467 struct btrfs_root *log,
1468 struct btrfs_path *path,
1469 struct extent_buffer *eb, int slot,
1470 struct btrfs_key *key)
1472 struct inode *dir = NULL;
1473 struct inode *inode = NULL;
1474 unsigned long ref_ptr;
1475 unsigned long ref_end;
1479 int search_done = 0;
1480 int log_ref_ver = 0;
1481 u64 parent_objectid;
1484 int ref_struct_size;
1486 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1487 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1489 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1490 struct btrfs_inode_extref *r;
1492 ref_struct_size = sizeof(struct btrfs_inode_extref);
1494 r = (struct btrfs_inode_extref *)ref_ptr;
1495 parent_objectid = btrfs_inode_extref_parent(eb, r);
1497 ref_struct_size = sizeof(struct btrfs_inode_ref);
1498 parent_objectid = key->offset;
1500 inode_objectid = key->objectid;
1503 * it is possible that we didn't log all the parent directories
1504 * for a given inode. If we don't find the dir, just don't
1505 * copy the back ref in. The link count fixup code will take
1508 dir = read_one_inode(root, parent_objectid);
1514 inode = read_one_inode(root, inode_objectid);
1520 while (ref_ptr < ref_end) {
1522 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1523 &ref_index, &parent_objectid);
1525 * parent object can change from one array
1529 dir = read_one_inode(root, parent_objectid);
1535 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1541 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1542 btrfs_ino(BTRFS_I(inode)), ref_index,
1546 } else if (ret == 0) {
1548 * look for a conflicting back reference in the
1549 * metadata. if we find one we have to unlink that name
1550 * of the file before we add our new link. Later on, we
1551 * overwrite any existing back reference, and we don't
1552 * want to create dangling pointers in the directory.
1556 ret = __add_inode_ref(trans, root, path, log,
1561 ref_index, name, namelen,
1571 * If a reference item already exists for this inode
1572 * with the same parent and name, but different index,
1573 * drop it and the corresponding directory index entries
1574 * from the parent before adding the new reference item
1575 * and dir index entries, otherwise we would fail with
1576 * -EEXIST returned from btrfs_add_link() below.
1578 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1581 ret = btrfs_unlink_inode(trans, root,
1586 * If we dropped the link count to 0, bump it so
1587 * that later the iput() on the inode will not
1588 * free it. We will fixup the link count later.
1590 if (!ret && inode->i_nlink == 0)
1593 * Whenever we need to check if a name exists or
1594 * not, we check the subvolume tree. So after an
1595 * unlink we must run delayed items, so that future
1596 * checks for a name during log replay see that the
1597 * name does not exists anymore.
1600 ret = btrfs_run_delayed_items(trans);
1605 /* insert our name */
1606 ret = add_link(trans, root, dir, inode, name, namelen,
1611 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1615 /* Else, ret == 1, we already have a perfect match, we're done. */
1617 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1627 * Before we overwrite the inode reference item in the subvolume tree
1628 * with the item from the log tree, we must unlink all names from the
1629 * parent directory that are in the subvolume's tree inode reference
1630 * item, otherwise we end up with an inconsistent subvolume tree where
1631 * dir index entries exist for a name but there is no inode reference
1632 * item with the same name.
1634 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1639 /* finally write the back reference in the inode */
1640 ret = overwrite_item(trans, root, path, eb, slot, key);
1642 btrfs_release_path(path);
1649 static int count_inode_extrefs(struct btrfs_root *root,
1650 struct btrfs_inode *inode, struct btrfs_path *path)
1654 unsigned int nlink = 0;
1657 u64 inode_objectid = btrfs_ino(inode);
1660 struct btrfs_inode_extref *extref;
1661 struct extent_buffer *leaf;
1664 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1669 leaf = path->nodes[0];
1670 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1671 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1674 while (cur_offset < item_size) {
1675 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1676 name_len = btrfs_inode_extref_name_len(leaf, extref);
1680 cur_offset += name_len + sizeof(*extref);
1684 btrfs_release_path(path);
1686 btrfs_release_path(path);
1688 if (ret < 0 && ret != -ENOENT)
1693 static int count_inode_refs(struct btrfs_root *root,
1694 struct btrfs_inode *inode, struct btrfs_path *path)
1697 struct btrfs_key key;
1698 unsigned int nlink = 0;
1700 unsigned long ptr_end;
1702 u64 ino = btrfs_ino(inode);
1705 key.type = BTRFS_INODE_REF_KEY;
1706 key.offset = (u64)-1;
1709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1713 if (path->slots[0] == 0)
1718 btrfs_item_key_to_cpu(path->nodes[0], &key,
1720 if (key.objectid != ino ||
1721 key.type != BTRFS_INODE_REF_KEY)
1723 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1724 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1726 while (ptr < ptr_end) {
1727 struct btrfs_inode_ref *ref;
1729 ref = (struct btrfs_inode_ref *)ptr;
1730 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1732 ptr = (unsigned long)(ref + 1) + name_len;
1736 if (key.offset == 0)
1738 if (path->slots[0] > 0) {
1743 btrfs_release_path(path);
1745 btrfs_release_path(path);
1751 * There are a few corners where the link count of the file can't
1752 * be properly maintained during replay. So, instead of adding
1753 * lots of complexity to the log code, we just scan the backrefs
1754 * for any file that has been through replay.
1756 * The scan will update the link count on the inode to reflect the
1757 * number of back refs found. If it goes down to zero, the iput
1758 * will free the inode.
1760 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1761 struct btrfs_root *root,
1762 struct inode *inode)
1764 struct btrfs_path *path;
1767 u64 ino = btrfs_ino(BTRFS_I(inode));
1769 path = btrfs_alloc_path();
1773 ret = count_inode_refs(root, BTRFS_I(inode), path);
1779 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1787 if (nlink != inode->i_nlink) {
1788 set_nlink(inode, nlink);
1789 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1793 BTRFS_I(inode)->index_cnt = (u64)-1;
1795 if (inode->i_nlink == 0) {
1796 if (S_ISDIR(inode->i_mode)) {
1797 ret = replay_dir_deletes(trans, root, NULL, path,
1802 ret = btrfs_insert_orphan_item(trans, root, ino);
1808 btrfs_free_path(path);
1812 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1813 struct btrfs_root *root,
1814 struct btrfs_path *path)
1817 struct btrfs_key key;
1818 struct inode *inode;
1820 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1821 key.type = BTRFS_ORPHAN_ITEM_KEY;
1822 key.offset = (u64)-1;
1824 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1830 if (path->slots[0] == 0)
1835 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1836 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1837 key.type != BTRFS_ORPHAN_ITEM_KEY)
1840 ret = btrfs_del_item(trans, root, path);
1844 btrfs_release_path(path);
1845 inode = read_one_inode(root, key.offset);
1851 ret = fixup_inode_link_count(trans, root, inode);
1857 * fixup on a directory may create new entries,
1858 * make sure we always look for the highset possible
1861 key.offset = (u64)-1;
1863 btrfs_release_path(path);
1869 * record a given inode in the fixup dir so we can check its link
1870 * count when replay is done. The link count is incremented here
1871 * so the inode won't go away until we check it
1873 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1874 struct btrfs_root *root,
1875 struct btrfs_path *path,
1878 struct btrfs_key key;
1880 struct inode *inode;
1882 inode = read_one_inode(root, objectid);
1886 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1887 key.type = BTRFS_ORPHAN_ITEM_KEY;
1888 key.offset = objectid;
1890 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1892 btrfs_release_path(path);
1894 if (!inode->i_nlink)
1895 set_nlink(inode, 1);
1898 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1899 } else if (ret == -EEXIST) {
1908 * when replaying the log for a directory, we only insert names
1909 * for inodes that actually exist. This means an fsync on a directory
1910 * does not implicitly fsync all the new files in it
1912 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1913 struct btrfs_root *root,
1914 u64 dirid, u64 index,
1915 char *name, int name_len,
1916 struct btrfs_key *location)
1918 struct inode *inode;
1922 inode = read_one_inode(root, location->objectid);
1926 dir = read_one_inode(root, dirid);
1932 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1933 name_len, 1, index);
1935 /* FIXME, put inode into FIXUP list */
1943 * take a single entry in a log directory item and replay it into
1946 * if a conflicting item exists in the subdirectory already,
1947 * the inode it points to is unlinked and put into the link count
1950 * If a name from the log points to a file or directory that does
1951 * not exist in the FS, it is skipped. fsyncs on directories
1952 * do not force down inodes inside that directory, just changes to the
1953 * names or unlinks in a directory.
1955 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1956 * non-existing inode) and 1 if the name was replayed.
1958 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1959 struct btrfs_root *root,
1960 struct btrfs_path *path,
1961 struct extent_buffer *eb,
1962 struct btrfs_dir_item *di,
1963 struct btrfs_key *key)
1967 struct btrfs_dir_item *dst_di;
1968 struct btrfs_key found_key;
1969 struct btrfs_key log_key;
1974 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1975 bool name_added = false;
1977 dir = read_one_inode(root, key->objectid);
1981 name_len = btrfs_dir_name_len(eb, di);
1982 name = kmalloc(name_len, GFP_NOFS);
1988 log_type = btrfs_dir_type(eb, di);
1989 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1992 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1993 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1994 btrfs_release_path(path);
1997 exists = (ret == 0);
2000 if (key->type == BTRFS_DIR_ITEM_KEY) {
2001 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
2003 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
2004 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
2014 if (IS_ERR(dst_di)) {
2015 ret = PTR_ERR(dst_di);
2017 } else if (!dst_di) {
2018 /* we need a sequence number to insert, so we only
2019 * do inserts for the BTRFS_DIR_INDEX_KEY types
2021 if (key->type != BTRFS_DIR_INDEX_KEY)
2026 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
2027 /* the existing item matches the logged item */
2028 if (found_key.objectid == log_key.objectid &&
2029 found_key.type == log_key.type &&
2030 found_key.offset == log_key.offset &&
2031 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
2032 update_size = false;
2037 * don't drop the conflicting directory entry if the inode
2038 * for the new entry doesn't exist
2043 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2047 if (key->type == BTRFS_DIR_INDEX_KEY)
2050 btrfs_release_path(path);
2051 if (!ret && update_size) {
2052 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2053 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
2057 if (!ret && name_added)
2063 * Check if the inode reference exists in the log for the given name,
2064 * inode and parent inode
2066 found_key.objectid = log_key.objectid;
2067 found_key.type = BTRFS_INODE_REF_KEY;
2068 found_key.offset = key->objectid;
2069 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2073 /* The dentry will be added later. */
2075 update_size = false;
2079 found_key.objectid = log_key.objectid;
2080 found_key.type = BTRFS_INODE_EXTREF_KEY;
2081 found_key.offset = key->objectid;
2082 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2087 /* The dentry will be added later. */
2089 update_size = false;
2092 btrfs_release_path(path);
2093 ret = insert_one_name(trans, root, key->objectid, key->offset,
2094 name, name_len, &log_key);
2095 if (ret && ret != -ENOENT && ret != -EEXIST)
2099 update_size = false;
2105 * find all the names in a directory item and reconcile them into
2106 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2107 * one name in a directory item, but the same code gets used for
2108 * both directory index types
2110 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2111 struct btrfs_root *root,
2112 struct btrfs_path *path,
2113 struct extent_buffer *eb, int slot,
2114 struct btrfs_key *key)
2117 u32 item_size = btrfs_item_size_nr(eb, slot);
2118 struct btrfs_dir_item *di;
2121 unsigned long ptr_end;
2122 struct btrfs_path *fixup_path = NULL;
2124 ptr = btrfs_item_ptr_offset(eb, slot);
2125 ptr_end = ptr + item_size;
2126 while (ptr < ptr_end) {
2127 di = (struct btrfs_dir_item *)ptr;
2128 name_len = btrfs_dir_name_len(eb, di);
2129 ret = replay_one_name(trans, root, path, eb, di, key);
2132 ptr = (unsigned long)(di + 1);
2136 * If this entry refers to a non-directory (directories can not
2137 * have a link count > 1) and it was added in the transaction
2138 * that was not committed, make sure we fixup the link count of
2139 * the inode it the entry points to. Otherwise something like
2140 * the following would result in a directory pointing to an
2141 * inode with a wrong link that does not account for this dir
2149 * ln testdir/bar testdir/bar_link
2150 * ln testdir/foo testdir/foo_link
2151 * xfs_io -c "fsync" testdir/bar
2155 * mount fs, log replay happens
2157 * File foo would remain with a link count of 1 when it has two
2158 * entries pointing to it in the directory testdir. This would
2159 * make it impossible to ever delete the parent directory has
2160 * it would result in stale dentries that can never be deleted.
2162 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2163 struct btrfs_key di_key;
2166 fixup_path = btrfs_alloc_path();
2173 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2174 ret = link_to_fixup_dir(trans, root, fixup_path,
2181 btrfs_free_path(fixup_path);
2186 * directory replay has two parts. There are the standard directory
2187 * items in the log copied from the subvolume, and range items
2188 * created in the log while the subvolume was logged.
2190 * The range items tell us which parts of the key space the log
2191 * is authoritative for. During replay, if a key in the subvolume
2192 * directory is in a logged range item, but not actually in the log
2193 * that means it was deleted from the directory before the fsync
2194 * and should be removed.
2196 static noinline int find_dir_range(struct btrfs_root *root,
2197 struct btrfs_path *path,
2198 u64 dirid, int key_type,
2199 u64 *start_ret, u64 *end_ret)
2201 struct btrfs_key key;
2203 struct btrfs_dir_log_item *item;
2207 if (*start_ret == (u64)-1)
2210 key.objectid = dirid;
2211 key.type = key_type;
2212 key.offset = *start_ret;
2214 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2218 if (path->slots[0] == 0)
2223 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2225 if (key.type != key_type || key.objectid != dirid) {
2229 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2230 struct btrfs_dir_log_item);
2231 found_end = btrfs_dir_log_end(path->nodes[0], item);
2233 if (*start_ret >= key.offset && *start_ret <= found_end) {
2235 *start_ret = key.offset;
2236 *end_ret = found_end;
2241 /* check the next slot in the tree to see if it is a valid item */
2242 nritems = btrfs_header_nritems(path->nodes[0]);
2244 if (path->slots[0] >= nritems) {
2245 ret = btrfs_next_leaf(root, path);
2250 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2252 if (key.type != key_type || key.objectid != dirid) {
2256 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2257 struct btrfs_dir_log_item);
2258 found_end = btrfs_dir_log_end(path->nodes[0], item);
2259 *start_ret = key.offset;
2260 *end_ret = found_end;
2263 btrfs_release_path(path);
2268 * this looks for a given directory item in the log. If the directory
2269 * item is not in the log, the item is removed and the inode it points
2272 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *root,
2274 struct btrfs_root *log,
2275 struct btrfs_path *path,
2276 struct btrfs_path *log_path,
2278 struct btrfs_key *dir_key)
2281 struct extent_buffer *eb;
2284 struct btrfs_dir_item *di;
2285 struct btrfs_dir_item *log_di;
2288 unsigned long ptr_end;
2290 struct inode *inode;
2291 struct btrfs_key location;
2294 eb = path->nodes[0];
2295 slot = path->slots[0];
2296 item_size = btrfs_item_size_nr(eb, slot);
2297 ptr = btrfs_item_ptr_offset(eb, slot);
2298 ptr_end = ptr + item_size;
2299 while (ptr < ptr_end) {
2300 di = (struct btrfs_dir_item *)ptr;
2301 name_len = btrfs_dir_name_len(eb, di);
2302 name = kmalloc(name_len, GFP_NOFS);
2307 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2310 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2311 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2314 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2315 log_di = btrfs_lookup_dir_index_item(trans, log,
2322 btrfs_dir_item_key_to_cpu(eb, di, &location);
2323 btrfs_release_path(path);
2324 btrfs_release_path(log_path);
2325 inode = read_one_inode(root, location.objectid);
2331 ret = link_to_fixup_dir(trans, root,
2332 path, location.objectid);
2340 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2341 BTRFS_I(inode), name, name_len);
2343 ret = btrfs_run_delayed_items(trans);
2349 /* there might still be more names under this key
2350 * check and repeat if required
2352 ret = btrfs_search_slot(NULL, root, dir_key, path,
2358 } else if (IS_ERR(log_di)) {
2360 return PTR_ERR(log_di);
2362 btrfs_release_path(log_path);
2365 ptr = (unsigned long)(di + 1);
2370 btrfs_release_path(path);
2371 btrfs_release_path(log_path);
2375 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *root,
2377 struct btrfs_root *log,
2378 struct btrfs_path *path,
2381 struct btrfs_key search_key;
2382 struct btrfs_path *log_path;
2387 log_path = btrfs_alloc_path();
2391 search_key.objectid = ino;
2392 search_key.type = BTRFS_XATTR_ITEM_KEY;
2393 search_key.offset = 0;
2395 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2399 nritems = btrfs_header_nritems(path->nodes[0]);
2400 for (i = path->slots[0]; i < nritems; i++) {
2401 struct btrfs_key key;
2402 struct btrfs_dir_item *di;
2403 struct btrfs_dir_item *log_di;
2407 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2408 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2413 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2414 total_size = btrfs_item_size_nr(path->nodes[0], i);
2416 while (cur < total_size) {
2417 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2418 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2419 u32 this_len = sizeof(*di) + name_len + data_len;
2422 name = kmalloc(name_len, GFP_NOFS);
2427 read_extent_buffer(path->nodes[0], name,
2428 (unsigned long)(di + 1), name_len);
2430 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2432 btrfs_release_path(log_path);
2434 /* Doesn't exist in log tree, so delete it. */
2435 btrfs_release_path(path);
2436 di = btrfs_lookup_xattr(trans, root, path, ino,
2437 name, name_len, -1);
2444 ret = btrfs_delete_one_dir_name(trans, root,
2448 btrfs_release_path(path);
2453 if (IS_ERR(log_di)) {
2454 ret = PTR_ERR(log_di);
2458 di = (struct btrfs_dir_item *)((char *)di + this_len);
2461 ret = btrfs_next_leaf(root, path);
2467 btrfs_free_path(log_path);
2468 btrfs_release_path(path);
2474 * deletion replay happens before we copy any new directory items
2475 * out of the log or out of backreferences from inodes. It
2476 * scans the log to find ranges of keys that log is authoritative for,
2477 * and then scans the directory to find items in those ranges that are
2478 * not present in the log.
2480 * Anything we don't find in the log is unlinked and removed from the
2483 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2484 struct btrfs_root *root,
2485 struct btrfs_root *log,
2486 struct btrfs_path *path,
2487 u64 dirid, int del_all)
2491 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2493 struct btrfs_key dir_key;
2494 struct btrfs_key found_key;
2495 struct btrfs_path *log_path;
2498 dir_key.objectid = dirid;
2499 dir_key.type = BTRFS_DIR_ITEM_KEY;
2500 log_path = btrfs_alloc_path();
2504 dir = read_one_inode(root, dirid);
2505 /* it isn't an error if the inode isn't there, that can happen
2506 * because we replay the deletes before we copy in the inode item
2510 btrfs_free_path(log_path);
2518 range_end = (u64)-1;
2520 ret = find_dir_range(log, path, dirid, key_type,
2521 &range_start, &range_end);
2528 dir_key.offset = range_start;
2531 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2536 nritems = btrfs_header_nritems(path->nodes[0]);
2537 if (path->slots[0] >= nritems) {
2538 ret = btrfs_next_leaf(root, path);
2544 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2546 if (found_key.objectid != dirid ||
2547 found_key.type != dir_key.type)
2550 if (found_key.offset > range_end)
2553 ret = check_item_in_log(trans, root, log, path,
2558 if (found_key.offset == (u64)-1)
2560 dir_key.offset = found_key.offset + 1;
2562 btrfs_release_path(path);
2563 if (range_end == (u64)-1)
2565 range_start = range_end + 1;
2570 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2571 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2572 dir_key.type = BTRFS_DIR_INDEX_KEY;
2573 btrfs_release_path(path);
2577 btrfs_release_path(path);
2578 btrfs_free_path(log_path);
2584 * the process_func used to replay items from the log tree. This
2585 * gets called in two different stages. The first stage just looks
2586 * for inodes and makes sure they are all copied into the subvolume.
2588 * The second stage copies all the other item types from the log into
2589 * the subvolume. The two stage approach is slower, but gets rid of
2590 * lots of complexity around inodes referencing other inodes that exist
2591 * only in the log (references come from either directory items or inode
2594 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2595 struct walk_control *wc, u64 gen, int level)
2598 struct btrfs_path *path;
2599 struct btrfs_root *root = wc->replay_dest;
2600 struct btrfs_key key;
2604 ret = btrfs_read_buffer(eb, gen, level, NULL);
2608 level = btrfs_header_level(eb);
2613 path = btrfs_alloc_path();
2617 nritems = btrfs_header_nritems(eb);
2618 for (i = 0; i < nritems; i++) {
2619 btrfs_item_key_to_cpu(eb, &key, i);
2621 /* inode keys are done during the first stage */
2622 if (key.type == BTRFS_INODE_ITEM_KEY &&
2623 wc->stage == LOG_WALK_REPLAY_INODES) {
2624 struct btrfs_inode_item *inode_item;
2627 inode_item = btrfs_item_ptr(eb, i,
2628 struct btrfs_inode_item);
2630 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2631 * and never got linked before the fsync, skip it, as
2632 * replaying it is pointless since it would be deleted
2633 * later. We skip logging tmpfiles, but it's always
2634 * possible we are replaying a log created with a kernel
2635 * that used to log tmpfiles.
2637 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2638 wc->ignore_cur_inode = true;
2641 wc->ignore_cur_inode = false;
2643 ret = replay_xattr_deletes(wc->trans, root, log,
2644 path, key.objectid);
2647 mode = btrfs_inode_mode(eb, inode_item);
2648 if (S_ISDIR(mode)) {
2649 ret = replay_dir_deletes(wc->trans,
2650 root, log, path, key.objectid, 0);
2654 ret = overwrite_item(wc->trans, root, path,
2660 * Before replaying extents, truncate the inode to its
2661 * size. We need to do it now and not after log replay
2662 * because before an fsync we can have prealloc extents
2663 * added beyond the inode's i_size. If we did it after,
2664 * through orphan cleanup for example, we would drop
2665 * those prealloc extents just after replaying them.
2667 if (S_ISREG(mode)) {
2668 struct btrfs_drop_extents_args drop_args = { 0 };
2669 struct inode *inode;
2672 inode = read_one_inode(root, key.objectid);
2677 from = ALIGN(i_size_read(inode),
2678 root->fs_info->sectorsize);
2679 drop_args.start = from;
2680 drop_args.end = (u64)-1;
2681 drop_args.drop_cache = true;
2682 ret = btrfs_drop_extents(wc->trans, root,
2686 inode_sub_bytes(inode,
2687 drop_args.bytes_found);
2688 /* Update the inode's nbytes. */
2689 ret = btrfs_update_inode(wc->trans,
2690 root, BTRFS_I(inode));
2697 ret = link_to_fixup_dir(wc->trans, root,
2698 path, key.objectid);
2703 if (wc->ignore_cur_inode)
2706 if (key.type == BTRFS_DIR_INDEX_KEY &&
2707 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2708 ret = replay_one_dir_item(wc->trans, root, path,
2714 if (wc->stage < LOG_WALK_REPLAY_ALL)
2717 /* these keys are simply copied */
2718 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2719 ret = overwrite_item(wc->trans, root, path,
2723 } else if (key.type == BTRFS_INODE_REF_KEY ||
2724 key.type == BTRFS_INODE_EXTREF_KEY) {
2725 ret = add_inode_ref(wc->trans, root, log, path,
2727 if (ret && ret != -ENOENT)
2730 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2731 ret = replay_one_extent(wc->trans, root, path,
2735 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2736 ret = replay_one_dir_item(wc->trans, root, path,
2742 btrfs_free_path(path);
2747 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2749 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2751 struct btrfs_block_group *cache;
2753 cache = btrfs_lookup_block_group(fs_info, start);
2755 btrfs_err(fs_info, "unable to find block group for %llu", start);
2759 spin_lock(&cache->space_info->lock);
2760 spin_lock(&cache->lock);
2761 cache->reserved -= fs_info->nodesize;
2762 cache->space_info->bytes_reserved -= fs_info->nodesize;
2763 spin_unlock(&cache->lock);
2764 spin_unlock(&cache->space_info->lock);
2766 btrfs_put_block_group(cache);
2769 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2770 struct btrfs_root *root,
2771 struct btrfs_path *path, int *level,
2772 struct walk_control *wc)
2774 struct btrfs_fs_info *fs_info = root->fs_info;
2777 struct extent_buffer *next;
2778 struct extent_buffer *cur;
2782 while (*level > 0) {
2783 struct btrfs_key first_key;
2785 cur = path->nodes[*level];
2787 WARN_ON(btrfs_header_level(cur) != *level);
2789 if (path->slots[*level] >=
2790 btrfs_header_nritems(cur))
2793 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2794 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2795 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2796 blocksize = fs_info->nodesize;
2798 next = btrfs_find_create_tree_block(fs_info, bytenr,
2799 btrfs_header_owner(cur),
2802 return PTR_ERR(next);
2805 ret = wc->process_func(root, next, wc, ptr_gen,
2808 free_extent_buffer(next);
2812 path->slots[*level]++;
2814 ret = btrfs_read_buffer(next, ptr_gen,
2815 *level - 1, &first_key);
2817 free_extent_buffer(next);
2822 btrfs_tree_lock(next);
2823 btrfs_clean_tree_block(next);
2824 btrfs_wait_tree_block_writeback(next);
2825 btrfs_tree_unlock(next);
2826 ret = btrfs_pin_reserved_extent(trans,
2829 free_extent_buffer(next);
2832 btrfs_redirty_list_add(
2833 trans->transaction, next);
2835 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2836 clear_extent_buffer_dirty(next);
2837 unaccount_log_buffer(fs_info, bytenr);
2840 free_extent_buffer(next);
2843 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2845 free_extent_buffer(next);
2849 if (path->nodes[*level-1])
2850 free_extent_buffer(path->nodes[*level-1]);
2851 path->nodes[*level-1] = next;
2852 *level = btrfs_header_level(next);
2853 path->slots[*level] = 0;
2856 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2862 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2863 struct btrfs_root *root,
2864 struct btrfs_path *path, int *level,
2865 struct walk_control *wc)
2867 struct btrfs_fs_info *fs_info = root->fs_info;
2872 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2873 slot = path->slots[i];
2874 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2877 WARN_ON(*level == 0);
2880 ret = wc->process_func(root, path->nodes[*level], wc,
2881 btrfs_header_generation(path->nodes[*level]),
2887 struct extent_buffer *next;
2889 next = path->nodes[*level];
2892 btrfs_tree_lock(next);
2893 btrfs_clean_tree_block(next);
2894 btrfs_wait_tree_block_writeback(next);
2895 btrfs_tree_unlock(next);
2896 ret = btrfs_pin_reserved_extent(trans,
2897 path->nodes[*level]->start,
2898 path->nodes[*level]->len);
2901 btrfs_redirty_list_add(trans->transaction,
2904 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2905 clear_extent_buffer_dirty(next);
2907 unaccount_log_buffer(fs_info,
2908 path->nodes[*level]->start);
2911 free_extent_buffer(path->nodes[*level]);
2912 path->nodes[*level] = NULL;
2920 * drop the reference count on the tree rooted at 'snap'. This traverses
2921 * the tree freeing any blocks that have a ref count of zero after being
2924 static int walk_log_tree(struct btrfs_trans_handle *trans,
2925 struct btrfs_root *log, struct walk_control *wc)
2927 struct btrfs_fs_info *fs_info = log->fs_info;
2931 struct btrfs_path *path;
2934 path = btrfs_alloc_path();
2938 level = btrfs_header_level(log->node);
2940 path->nodes[level] = log->node;
2941 atomic_inc(&log->node->refs);
2942 path->slots[level] = 0;
2945 wret = walk_down_log_tree(trans, log, path, &level, wc);
2953 wret = walk_up_log_tree(trans, log, path, &level, wc);
2962 /* was the root node processed? if not, catch it here */
2963 if (path->nodes[orig_level]) {
2964 ret = wc->process_func(log, path->nodes[orig_level], wc,
2965 btrfs_header_generation(path->nodes[orig_level]),
2970 struct extent_buffer *next;
2972 next = path->nodes[orig_level];
2975 btrfs_tree_lock(next);
2976 btrfs_clean_tree_block(next);
2977 btrfs_wait_tree_block_writeback(next);
2978 btrfs_tree_unlock(next);
2979 ret = btrfs_pin_reserved_extent(trans,
2980 next->start, next->len);
2983 btrfs_redirty_list_add(trans->transaction, next);
2985 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2986 clear_extent_buffer_dirty(next);
2987 unaccount_log_buffer(fs_info, next->start);
2993 btrfs_free_path(path);
2998 * helper function to update the item for a given subvolumes log root
2999 * in the tree of log roots
3001 static int update_log_root(struct btrfs_trans_handle *trans,
3002 struct btrfs_root *log,
3003 struct btrfs_root_item *root_item)
3005 struct btrfs_fs_info *fs_info = log->fs_info;
3008 if (log->log_transid == 1) {
3009 /* insert root item on the first sync */
3010 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
3011 &log->root_key, root_item);
3013 ret = btrfs_update_root(trans, fs_info->log_root_tree,
3014 &log->root_key, root_item);
3019 static void wait_log_commit(struct btrfs_root *root, int transid)
3022 int index = transid % 2;
3025 * we only allow two pending log transactions at a time,
3026 * so we know that if ours is more than 2 older than the
3027 * current transaction, we're done
3030 prepare_to_wait(&root->log_commit_wait[index],
3031 &wait, TASK_UNINTERRUPTIBLE);
3033 if (!(root->log_transid_committed < transid &&
3034 atomic_read(&root->log_commit[index])))
3037 mutex_unlock(&root->log_mutex);
3039 mutex_lock(&root->log_mutex);
3041 finish_wait(&root->log_commit_wait[index], &wait);
3044 static void wait_for_writer(struct btrfs_root *root)
3049 prepare_to_wait(&root->log_writer_wait, &wait,
3050 TASK_UNINTERRUPTIBLE);
3051 if (!atomic_read(&root->log_writers))
3054 mutex_unlock(&root->log_mutex);
3056 mutex_lock(&root->log_mutex);
3058 finish_wait(&root->log_writer_wait, &wait);
3061 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3062 struct btrfs_log_ctx *ctx)
3067 mutex_lock(&root->log_mutex);
3068 list_del_init(&ctx->list);
3069 mutex_unlock(&root->log_mutex);
3073 * Invoked in log mutex context, or be sure there is no other task which
3074 * can access the list.
3076 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3077 int index, int error)
3079 struct btrfs_log_ctx *ctx;
3080 struct btrfs_log_ctx *safe;
3082 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3083 list_del_init(&ctx->list);
3084 ctx->log_ret = error;
3089 * btrfs_sync_log does sends a given tree log down to the disk and
3090 * updates the super blocks to record it. When this call is done,
3091 * you know that any inodes previously logged are safely on disk only
3094 * Any other return value means you need to call btrfs_commit_transaction.
3095 * Some of the edge cases for fsyncing directories that have had unlinks
3096 * or renames done in the past mean that sometimes the only safe
3097 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3098 * that has happened.
3100 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3101 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3107 struct btrfs_fs_info *fs_info = root->fs_info;
3108 struct btrfs_root *log = root->log_root;
3109 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3110 struct btrfs_root_item new_root_item;
3111 int log_transid = 0;
3112 struct btrfs_log_ctx root_log_ctx;
3113 struct blk_plug plug;
3117 mutex_lock(&root->log_mutex);
3118 log_transid = ctx->log_transid;
3119 if (root->log_transid_committed >= log_transid) {
3120 mutex_unlock(&root->log_mutex);
3121 return ctx->log_ret;
3124 index1 = log_transid % 2;
3125 if (atomic_read(&root->log_commit[index1])) {
3126 wait_log_commit(root, log_transid);
3127 mutex_unlock(&root->log_mutex);
3128 return ctx->log_ret;
3130 ASSERT(log_transid == root->log_transid);
3131 atomic_set(&root->log_commit[index1], 1);
3133 /* wait for previous tree log sync to complete */
3134 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3135 wait_log_commit(root, log_transid - 1);
3138 int batch = atomic_read(&root->log_batch);
3139 /* when we're on an ssd, just kick the log commit out */
3140 if (!btrfs_test_opt(fs_info, SSD) &&
3141 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3142 mutex_unlock(&root->log_mutex);
3143 schedule_timeout_uninterruptible(1);
3144 mutex_lock(&root->log_mutex);
3146 wait_for_writer(root);
3147 if (batch == atomic_read(&root->log_batch))
3151 /* bail out if we need to do a full commit */
3152 if (btrfs_need_log_full_commit(trans)) {
3154 mutex_unlock(&root->log_mutex);
3158 if (log_transid % 2 == 0)
3159 mark = EXTENT_DIRTY;
3163 /* we start IO on all the marked extents here, but we don't actually
3164 * wait for them until later.
3166 blk_start_plug(&plug);
3167 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3169 * -EAGAIN happens when someone, e.g., a concurrent transaction
3170 * commit, writes a dirty extent in this tree-log commit. This
3171 * concurrent write will create a hole writing out the extents,
3172 * and we cannot proceed on a zoned filesystem, requiring
3173 * sequential writing. While we can bail out to a full commit
3174 * here, but we can continue hoping the concurrent writing fills
3177 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3180 blk_finish_plug(&plug);
3181 btrfs_abort_transaction(trans, ret);
3182 btrfs_set_log_full_commit(trans);
3183 mutex_unlock(&root->log_mutex);
3188 * We _must_ update under the root->log_mutex in order to make sure we
3189 * have a consistent view of the log root we are trying to commit at
3192 * We _must_ copy this into a local copy, because we are not holding the
3193 * log_root_tree->log_mutex yet. This is important because when we
3194 * commit the log_root_tree we must have a consistent view of the
3195 * log_root_tree when we update the super block to point at the
3196 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3197 * with the commit and possibly point at the new block which we may not
3200 btrfs_set_root_node(&log->root_item, log->node);
3201 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3203 root->log_transid++;
3204 log->log_transid = root->log_transid;
3205 root->log_start_pid = 0;
3207 * IO has been started, blocks of the log tree have WRITTEN flag set
3208 * in their headers. new modifications of the log will be written to
3209 * new positions. so it's safe to allow log writers to go in.
3211 mutex_unlock(&root->log_mutex);
3213 if (btrfs_is_zoned(fs_info)) {
3214 mutex_lock(&fs_info->tree_root->log_mutex);
3215 if (!log_root_tree->node) {
3216 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3218 mutex_unlock(&fs_info->tree_root->log_mutex);
3222 mutex_unlock(&fs_info->tree_root->log_mutex);
3225 btrfs_init_log_ctx(&root_log_ctx, NULL);
3227 mutex_lock(&log_root_tree->log_mutex);
3229 index2 = log_root_tree->log_transid % 2;
3230 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3231 root_log_ctx.log_transid = log_root_tree->log_transid;
3234 * Now we are safe to update the log_root_tree because we're under the
3235 * log_mutex, and we're a current writer so we're holding the commit
3236 * open until we drop the log_mutex.
3238 ret = update_log_root(trans, log, &new_root_item);
3240 if (!list_empty(&root_log_ctx.list))
3241 list_del_init(&root_log_ctx.list);
3243 blk_finish_plug(&plug);
3244 btrfs_set_log_full_commit(trans);
3246 if (ret != -ENOSPC) {
3247 btrfs_abort_transaction(trans, ret);
3248 mutex_unlock(&log_root_tree->log_mutex);
3251 btrfs_wait_tree_log_extents(log, mark);
3252 mutex_unlock(&log_root_tree->log_mutex);
3257 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3258 blk_finish_plug(&plug);
3259 list_del_init(&root_log_ctx.list);
3260 mutex_unlock(&log_root_tree->log_mutex);
3261 ret = root_log_ctx.log_ret;
3265 index2 = root_log_ctx.log_transid % 2;
3266 if (atomic_read(&log_root_tree->log_commit[index2])) {
3267 blk_finish_plug(&plug);
3268 ret = btrfs_wait_tree_log_extents(log, mark);
3269 wait_log_commit(log_root_tree,
3270 root_log_ctx.log_transid);
3271 mutex_unlock(&log_root_tree->log_mutex);
3273 ret = root_log_ctx.log_ret;
3276 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3277 atomic_set(&log_root_tree->log_commit[index2], 1);
3279 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3280 wait_log_commit(log_root_tree,
3281 root_log_ctx.log_transid - 1);
3285 * now that we've moved on to the tree of log tree roots,
3286 * check the full commit flag again
3288 if (btrfs_need_log_full_commit(trans)) {
3289 blk_finish_plug(&plug);
3290 btrfs_wait_tree_log_extents(log, mark);
3291 mutex_unlock(&log_root_tree->log_mutex);
3293 goto out_wake_log_root;
3296 ret = btrfs_write_marked_extents(fs_info,
3297 &log_root_tree->dirty_log_pages,
3298 EXTENT_DIRTY | EXTENT_NEW);
3299 blk_finish_plug(&plug);
3301 * As described above, -EAGAIN indicates a hole in the extents. We
3302 * cannot wait for these write outs since the waiting cause a
3303 * deadlock. Bail out to the full commit instead.
3305 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3306 btrfs_set_log_full_commit(trans);
3307 btrfs_wait_tree_log_extents(log, mark);
3308 mutex_unlock(&log_root_tree->log_mutex);
3309 goto out_wake_log_root;
3311 btrfs_set_log_full_commit(trans);
3312 btrfs_abort_transaction(trans, ret);
3313 mutex_unlock(&log_root_tree->log_mutex);
3314 goto out_wake_log_root;
3316 ret = btrfs_wait_tree_log_extents(log, mark);
3318 ret = btrfs_wait_tree_log_extents(log_root_tree,
3319 EXTENT_NEW | EXTENT_DIRTY);
3321 btrfs_set_log_full_commit(trans);
3322 mutex_unlock(&log_root_tree->log_mutex);
3323 goto out_wake_log_root;
3326 log_root_start = log_root_tree->node->start;
3327 log_root_level = btrfs_header_level(log_root_tree->node);
3328 log_root_tree->log_transid++;
3329 mutex_unlock(&log_root_tree->log_mutex);
3332 * Here we are guaranteed that nobody is going to write the superblock
3333 * for the current transaction before us and that neither we do write
3334 * our superblock before the previous transaction finishes its commit
3335 * and writes its superblock, because:
3337 * 1) We are holding a handle on the current transaction, so no body
3338 * can commit it until we release the handle;
3340 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3341 * if the previous transaction is still committing, and hasn't yet
3342 * written its superblock, we wait for it to do it, because a
3343 * transaction commit acquires the tree_log_mutex when the commit
3344 * begins and releases it only after writing its superblock.
3346 mutex_lock(&fs_info->tree_log_mutex);
3349 * The previous transaction writeout phase could have failed, and thus
3350 * marked the fs in an error state. We must not commit here, as we
3351 * could have updated our generation in the super_for_commit and
3352 * writing the super here would result in transid mismatches. If there
3353 * is an error here just bail.
3355 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3357 btrfs_set_log_full_commit(trans);
3358 btrfs_abort_transaction(trans, ret);
3359 mutex_unlock(&fs_info->tree_log_mutex);
3360 goto out_wake_log_root;
3363 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3364 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3365 ret = write_all_supers(fs_info, 1);
3366 mutex_unlock(&fs_info->tree_log_mutex);
3368 btrfs_set_log_full_commit(trans);
3369 btrfs_abort_transaction(trans, ret);
3370 goto out_wake_log_root;
3374 * We know there can only be one task here, since we have not yet set
3375 * root->log_commit[index1] to 0 and any task attempting to sync the
3376 * log must wait for the previous log transaction to commit if it's
3377 * still in progress or wait for the current log transaction commit if
3378 * someone else already started it. We use <= and not < because the
3379 * first log transaction has an ID of 0.
3381 ASSERT(root->last_log_commit <= log_transid);
3382 root->last_log_commit = log_transid;
3385 mutex_lock(&log_root_tree->log_mutex);
3386 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3388 log_root_tree->log_transid_committed++;
3389 atomic_set(&log_root_tree->log_commit[index2], 0);
3390 mutex_unlock(&log_root_tree->log_mutex);
3393 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3394 * all the updates above are seen by the woken threads. It might not be
3395 * necessary, but proving that seems to be hard.
3397 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3399 mutex_lock(&root->log_mutex);
3400 btrfs_remove_all_log_ctxs(root, index1, ret);
3401 root->log_transid_committed++;
3402 atomic_set(&root->log_commit[index1], 0);
3403 mutex_unlock(&root->log_mutex);
3406 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3407 * all the updates above are seen by the woken threads. It might not be
3408 * necessary, but proving that seems to be hard.
3410 cond_wake_up(&root->log_commit_wait[index1]);
3414 static void free_log_tree(struct btrfs_trans_handle *trans,
3415 struct btrfs_root *log)
3418 struct walk_control wc = {
3420 .process_func = process_one_buffer
3424 ret = walk_log_tree(trans, log, &wc);
3427 * We weren't able to traverse the entire log tree, the
3428 * typical scenario is getting an -EIO when reading an
3429 * extent buffer of the tree, due to a previous writeback
3432 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3433 &log->fs_info->fs_state);
3436 * Some extent buffers of the log tree may still be dirty
3437 * and not yet written back to storage, because we may
3438 * have updates to a log tree without syncing a log tree,
3439 * such as during rename and link operations. So flush
3440 * them out and wait for their writeback to complete, so
3441 * that we properly cleanup their state and pages.
3443 btrfs_write_marked_extents(log->fs_info,
3444 &log->dirty_log_pages,
3445 EXTENT_DIRTY | EXTENT_NEW);
3446 btrfs_wait_tree_log_extents(log,
3447 EXTENT_DIRTY | EXTENT_NEW);
3450 btrfs_abort_transaction(trans, ret);
3452 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3456 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3457 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3458 extent_io_tree_release(&log->log_csum_range);
3460 btrfs_put_root(log);
3464 * free all the extents used by the tree log. This should be called
3465 * at commit time of the full transaction
3467 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3469 if (root->log_root) {
3470 free_log_tree(trans, root->log_root);
3471 root->log_root = NULL;
3472 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3477 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3478 struct btrfs_fs_info *fs_info)
3480 if (fs_info->log_root_tree) {
3481 free_log_tree(trans, fs_info->log_root_tree);
3482 fs_info->log_root_tree = NULL;
3483 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3489 * Check if an inode was logged in the current transaction. This may often
3490 * return some false positives, because logged_trans is an in memory only field,
3491 * not persisted anywhere. This is meant to be used in contexts where a false
3492 * positive has no functional consequences.
3494 static bool inode_logged(struct btrfs_trans_handle *trans,
3495 struct btrfs_inode *inode)
3497 if (inode->logged_trans == trans->transid)
3501 * The inode's logged_trans is always 0 when we load it (because it is
3502 * not persisted in the inode item or elsewhere). So if it is 0, the
3503 * inode was last modified in the current transaction then the inode may
3504 * have been logged before in the current transaction, then evicted and
3505 * loaded again in the current transaction - or may have never been logged
3506 * in the current transaction, but since we can not be sure, we have to
3507 * assume it was, otherwise our callers can leave an inconsistent log.
3509 if (inode->logged_trans == 0 &&
3510 inode->last_trans == trans->transid &&
3511 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3518 * If both a file and directory are logged, and unlinks or renames are
3519 * mixed in, we have a few interesting corners:
3521 * create file X in dir Y
3522 * link file X to X.link in dir Y
3524 * unlink file X but leave X.link
3527 * After a crash we would expect only X.link to exist. But file X
3528 * didn't get fsync'd again so the log has back refs for X and X.link.
3530 * We solve this by removing directory entries and inode backrefs from the
3531 * log when a file that was logged in the current transaction is
3532 * unlinked. Any later fsync will include the updated log entries, and
3533 * we'll be able to reconstruct the proper directory items from backrefs.
3535 * This optimizations allows us to avoid relogging the entire inode
3536 * or the entire directory.
3538 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3539 struct btrfs_root *root,
3540 const char *name, int name_len,
3541 struct btrfs_inode *dir, u64 index)
3543 struct btrfs_root *log;
3544 struct btrfs_dir_item *di;
3545 struct btrfs_path *path;
3548 u64 dir_ino = btrfs_ino(dir);
3550 if (!inode_logged(trans, dir))
3553 ret = join_running_log_trans(root);
3557 mutex_lock(&dir->log_mutex);
3559 log = root->log_root;
3560 path = btrfs_alloc_path();
3566 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3567 name, name_len, -1);
3573 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3579 btrfs_release_path(path);
3580 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3581 index, name, name_len, -1);
3587 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3595 * We do not need to update the size field of the directory's inode item
3596 * because on log replay we update the field to reflect all existing
3597 * entries in the directory (see overwrite_item()).
3600 btrfs_free_path(path);
3602 mutex_unlock(&dir->log_mutex);
3603 if (err == -ENOSPC) {
3604 btrfs_set_log_full_commit(trans);
3606 } else if (err < 0) {
3607 btrfs_abort_transaction(trans, err);
3610 btrfs_end_log_trans(root);
3615 /* see comments for btrfs_del_dir_entries_in_log */
3616 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *root,
3618 const char *name, int name_len,
3619 struct btrfs_inode *inode, u64 dirid)
3621 struct btrfs_root *log;
3625 if (!inode_logged(trans, inode))
3628 ret = join_running_log_trans(root);
3631 log = root->log_root;
3632 mutex_lock(&inode->log_mutex);
3634 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3636 mutex_unlock(&inode->log_mutex);
3637 if (ret == -ENOSPC) {
3638 btrfs_set_log_full_commit(trans);
3640 } else if (ret < 0 && ret != -ENOENT)
3641 btrfs_abort_transaction(trans, ret);
3642 btrfs_end_log_trans(root);
3648 * creates a range item in the log for 'dirid'. first_offset and
3649 * last_offset tell us which parts of the key space the log should
3650 * be considered authoritative for.
3652 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3653 struct btrfs_root *log,
3654 struct btrfs_path *path,
3655 int key_type, u64 dirid,
3656 u64 first_offset, u64 last_offset)
3659 struct btrfs_key key;
3660 struct btrfs_dir_log_item *item;
3662 key.objectid = dirid;
3663 key.offset = first_offset;
3664 if (key_type == BTRFS_DIR_ITEM_KEY)
3665 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3667 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3668 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3672 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3673 struct btrfs_dir_log_item);
3674 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3675 btrfs_mark_buffer_dirty(path->nodes[0]);
3676 btrfs_release_path(path);
3681 * log all the items included in the current transaction for a given
3682 * directory. This also creates the range items in the log tree required
3683 * to replay anything deleted before the fsync
3685 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3686 struct btrfs_root *root, struct btrfs_inode *inode,
3687 struct btrfs_path *path,
3688 struct btrfs_path *dst_path, int key_type,
3689 struct btrfs_log_ctx *ctx,
3690 u64 min_offset, u64 *last_offset_ret)
3692 struct btrfs_key min_key;
3693 struct btrfs_root *log = root->log_root;
3694 struct extent_buffer *src;
3699 u64 first_offset = min_offset;
3700 u64 last_offset = (u64)-1;
3701 u64 ino = btrfs_ino(inode);
3703 log = root->log_root;
3705 min_key.objectid = ino;
3706 min_key.type = key_type;
3707 min_key.offset = min_offset;
3709 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3712 * we didn't find anything from this transaction, see if there
3713 * is anything at all
3715 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3716 min_key.objectid = ino;
3717 min_key.type = key_type;
3718 min_key.offset = (u64)-1;
3719 btrfs_release_path(path);
3720 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3722 btrfs_release_path(path);
3725 ret = btrfs_previous_item(root, path, ino, key_type);
3727 /* if ret == 0 there are items for this type,
3728 * create a range to tell us the last key of this type.
3729 * otherwise, there are no items in this directory after
3730 * *min_offset, and we create a range to indicate that.
3733 struct btrfs_key tmp;
3734 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3736 if (key_type == tmp.type)
3737 first_offset = max(min_offset, tmp.offset) + 1;
3742 /* go backward to find any previous key */
3743 ret = btrfs_previous_item(root, path, ino, key_type);
3745 struct btrfs_key tmp;
3746 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3747 if (key_type == tmp.type) {
3748 first_offset = tmp.offset;
3749 ret = overwrite_item(trans, log, dst_path,
3750 path->nodes[0], path->slots[0],
3758 btrfs_release_path(path);
3761 * Find the first key from this transaction again. See the note for
3762 * log_new_dir_dentries, if we're logging a directory recursively we
3763 * won't be holding its i_mutex, which means we can modify the directory
3764 * while we're logging it. If we remove an entry between our first
3765 * search and this search we'll not find the key again and can just
3769 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3774 * we have a block from this transaction, log every item in it
3775 * from our directory
3778 struct btrfs_key tmp;
3779 src = path->nodes[0];
3780 nritems = btrfs_header_nritems(src);
3781 for (i = path->slots[0]; i < nritems; i++) {
3782 struct btrfs_dir_item *di;
3784 btrfs_item_key_to_cpu(src, &min_key, i);
3786 if (min_key.objectid != ino || min_key.type != key_type)
3789 if (need_resched()) {
3790 btrfs_release_path(path);
3795 ret = overwrite_item(trans, log, dst_path, src, i,
3803 * We must make sure that when we log a directory entry,
3804 * the corresponding inode, after log replay, has a
3805 * matching link count. For example:
3811 * xfs_io -c "fsync" mydir
3813 * <mount fs and log replay>
3815 * Would result in a fsync log that when replayed, our
3816 * file inode would have a link count of 1, but we get
3817 * two directory entries pointing to the same inode.
3818 * After removing one of the names, it would not be
3819 * possible to remove the other name, which resulted
3820 * always in stale file handle errors, and would not
3821 * be possible to rmdir the parent directory, since
3822 * its i_size could never decrement to the value
3823 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3825 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3826 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3828 (btrfs_dir_transid(src, di) == trans->transid ||
3829 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3830 tmp.type != BTRFS_ROOT_ITEM_KEY)
3831 ctx->log_new_dentries = true;
3833 path->slots[0] = nritems;
3836 * look ahead to the next item and see if it is also
3837 * from this directory and from this transaction
3839 ret = btrfs_next_leaf(root, path);
3842 last_offset = (u64)-1;
3847 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3848 if (tmp.objectid != ino || tmp.type != key_type) {
3849 last_offset = (u64)-1;
3852 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3853 ret = overwrite_item(trans, log, dst_path,
3854 path->nodes[0], path->slots[0],
3859 last_offset = tmp.offset;
3864 btrfs_release_path(path);
3865 btrfs_release_path(dst_path);
3868 *last_offset_ret = last_offset;
3870 * insert the log range keys to indicate where the log
3873 ret = insert_dir_log_key(trans, log, path, key_type,
3874 ino, first_offset, last_offset);
3882 * logging directories is very similar to logging inodes, We find all the items
3883 * from the current transaction and write them to the log.
3885 * The recovery code scans the directory in the subvolume, and if it finds a
3886 * key in the range logged that is not present in the log tree, then it means
3887 * that dir entry was unlinked during the transaction.
3889 * In order for that scan to work, we must include one key smaller than
3890 * the smallest logged by this transaction and one key larger than the largest
3891 * key logged by this transaction.
3893 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3894 struct btrfs_root *root, struct btrfs_inode *inode,
3895 struct btrfs_path *path,
3896 struct btrfs_path *dst_path,
3897 struct btrfs_log_ctx *ctx)
3902 int key_type = BTRFS_DIR_ITEM_KEY;
3908 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3909 ctx, min_key, &max_key);
3912 if (max_key == (u64)-1)
3914 min_key = max_key + 1;
3917 if (key_type == BTRFS_DIR_ITEM_KEY) {
3918 key_type = BTRFS_DIR_INDEX_KEY;
3925 * a helper function to drop items from the log before we relog an
3926 * inode. max_key_type indicates the highest item type to remove.
3927 * This cannot be run for file data extents because it does not
3928 * free the extents they point to.
3930 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3931 struct btrfs_root *log,
3932 struct btrfs_path *path,
3933 u64 objectid, int max_key_type)
3936 struct btrfs_key key;
3937 struct btrfs_key found_key;
3940 key.objectid = objectid;
3941 key.type = max_key_type;
3942 key.offset = (u64)-1;
3945 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3946 BUG_ON(ret == 0); /* Logic error */
3950 if (path->slots[0] == 0)
3954 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3957 if (found_key.objectid != objectid)
3960 found_key.offset = 0;
3962 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3966 ret = btrfs_del_items(trans, log, path, start_slot,
3967 path->slots[0] - start_slot + 1);
3969 * If start slot isn't 0 then we don't need to re-search, we've
3970 * found the last guy with the objectid in this tree.
3972 if (ret || start_slot != 0)
3974 btrfs_release_path(path);
3976 btrfs_release_path(path);
3982 static void fill_inode_item(struct btrfs_trans_handle *trans,
3983 struct extent_buffer *leaf,
3984 struct btrfs_inode_item *item,
3985 struct inode *inode, int log_inode_only,
3988 struct btrfs_map_token token;
3991 btrfs_init_map_token(&token, leaf);
3993 if (log_inode_only) {
3994 /* set the generation to zero so the recover code
3995 * can tell the difference between an logging
3996 * just to say 'this inode exists' and a logging
3997 * to say 'update this inode with these values'
3999 btrfs_set_token_inode_generation(&token, item, 0);
4000 btrfs_set_token_inode_size(&token, item, logged_isize);
4002 btrfs_set_token_inode_generation(&token, item,
4003 BTRFS_I(inode)->generation);
4004 btrfs_set_token_inode_size(&token, item, inode->i_size);
4007 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4008 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4009 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4010 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4012 btrfs_set_token_timespec_sec(&token, &item->atime,
4013 inode->i_atime.tv_sec);
4014 btrfs_set_token_timespec_nsec(&token, &item->atime,
4015 inode->i_atime.tv_nsec);
4017 btrfs_set_token_timespec_sec(&token, &item->mtime,
4018 inode->i_mtime.tv_sec);
4019 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4020 inode->i_mtime.tv_nsec);
4022 btrfs_set_token_timespec_sec(&token, &item->ctime,
4023 inode->i_ctime.tv_sec);
4024 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4025 inode->i_ctime.tv_nsec);
4028 * We do not need to set the nbytes field, in fact during a fast fsync
4029 * its value may not even be correct, since a fast fsync does not wait
4030 * for ordered extent completion, which is where we update nbytes, it
4031 * only waits for writeback to complete. During log replay as we find
4032 * file extent items and replay them, we adjust the nbytes field of the
4033 * inode item in subvolume tree as needed (see overwrite_item()).
4036 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4037 btrfs_set_token_inode_transid(&token, item, trans->transid);
4038 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4039 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4040 BTRFS_I(inode)->ro_flags);
4041 btrfs_set_token_inode_flags(&token, item, flags);
4042 btrfs_set_token_inode_block_group(&token, item, 0);
4045 static int log_inode_item(struct btrfs_trans_handle *trans,
4046 struct btrfs_root *log, struct btrfs_path *path,
4047 struct btrfs_inode *inode, bool inode_item_dropped)
4049 struct btrfs_inode_item *inode_item;
4053 * If we are doing a fast fsync and the inode was logged before in the
4054 * current transaction, then we know the inode was previously logged and
4055 * it exists in the log tree. For performance reasons, in this case use
4056 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4057 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4058 * contention in case there are concurrent fsyncs for other inodes of the
4059 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4060 * already exists can also result in unnecessarily splitting a leaf.
4062 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4063 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4069 * This means it is the first fsync in the current transaction,
4070 * so the inode item is not in the log and we need to insert it.
4071 * We can never get -EEXIST because we are only called for a fast
4072 * fsync and in case an inode eviction happens after the inode was
4073 * logged before in the current transaction, when we load again
4074 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4075 * flags and set ->logged_trans to 0.
4077 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4078 sizeof(*inode_item));
4079 ASSERT(ret != -EEXIST);
4083 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4084 struct btrfs_inode_item);
4085 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4087 btrfs_release_path(path);
4091 static int log_csums(struct btrfs_trans_handle *trans,
4092 struct btrfs_inode *inode,
4093 struct btrfs_root *log_root,
4094 struct btrfs_ordered_sum *sums)
4096 const u64 lock_end = sums->bytenr + sums->len - 1;
4097 struct extent_state *cached_state = NULL;
4101 * If this inode was not used for reflink operations in the current
4102 * transaction with new extents, then do the fast path, no need to
4103 * worry about logging checksum items with overlapping ranges.
4105 if (inode->last_reflink_trans < trans->transid)
4106 return btrfs_csum_file_blocks(trans, log_root, sums);
4109 * Serialize logging for checksums. This is to avoid racing with the
4110 * same checksum being logged by another task that is logging another
4111 * file which happens to refer to the same extent as well. Such races
4112 * can leave checksum items in the log with overlapping ranges.
4114 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
4115 lock_end, &cached_state);
4119 * Due to extent cloning, we might have logged a csum item that covers a
4120 * subrange of a cloned extent, and later we can end up logging a csum
4121 * item for a larger subrange of the same extent or the entire range.
4122 * This would leave csum items in the log tree that cover the same range
4123 * and break the searches for checksums in the log tree, resulting in
4124 * some checksums missing in the fs/subvolume tree. So just delete (or
4125 * trim and adjust) any existing csum items in the log for this range.
4127 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4129 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4131 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
4137 static noinline int copy_items(struct btrfs_trans_handle *trans,
4138 struct btrfs_inode *inode,
4139 struct btrfs_path *dst_path,
4140 struct btrfs_path *src_path,
4141 int start_slot, int nr, int inode_only,
4144 struct btrfs_fs_info *fs_info = trans->fs_info;
4145 unsigned long src_offset;
4146 unsigned long dst_offset;
4147 struct btrfs_root *log = inode->root->log_root;
4148 struct btrfs_file_extent_item *extent;
4149 struct btrfs_inode_item *inode_item;
4150 struct extent_buffer *src = src_path->nodes[0];
4152 struct btrfs_key *ins_keys;
4156 struct list_head ordered_sums;
4157 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4159 INIT_LIST_HEAD(&ordered_sums);
4161 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4162 nr * sizeof(u32), GFP_NOFS);
4166 ins_sizes = (u32 *)ins_data;
4167 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4169 for (i = 0; i < nr; i++) {
4170 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4171 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4173 ret = btrfs_insert_empty_items(trans, log, dst_path,
4174 ins_keys, ins_sizes, nr);
4180 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4181 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4182 dst_path->slots[0]);
4184 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4186 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4187 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4189 struct btrfs_inode_item);
4190 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4192 inode_only == LOG_INODE_EXISTS,
4195 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4196 src_offset, ins_sizes[i]);
4199 /* take a reference on file data extents so that truncates
4200 * or deletes of this inode don't have to relog the inode
4203 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4206 extent = btrfs_item_ptr(src, start_slot + i,
4207 struct btrfs_file_extent_item);
4209 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4212 found_type = btrfs_file_extent_type(src, extent);
4213 if (found_type == BTRFS_FILE_EXTENT_REG) {
4215 ds = btrfs_file_extent_disk_bytenr(src,
4217 /* ds == 0 is a hole */
4221 dl = btrfs_file_extent_disk_num_bytes(src,
4223 cs = btrfs_file_extent_offset(src, extent);
4224 cl = btrfs_file_extent_num_bytes(src,
4226 if (btrfs_file_extent_compression(src,
4232 ret = btrfs_lookup_csums_range(
4234 ds + cs, ds + cs + cl - 1,
4242 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4243 btrfs_release_path(dst_path);
4247 * we have to do this after the loop above to avoid changing the
4248 * log tree while trying to change the log tree.
4250 while (!list_empty(&ordered_sums)) {
4251 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4252 struct btrfs_ordered_sum,
4255 ret = log_csums(trans, inode, log, sums);
4256 list_del(&sums->list);
4263 static int extent_cmp(void *priv, const struct list_head *a,
4264 const struct list_head *b)
4266 const struct extent_map *em1, *em2;
4268 em1 = list_entry(a, struct extent_map, list);
4269 em2 = list_entry(b, struct extent_map, list);
4271 if (em1->start < em2->start)
4273 else if (em1->start > em2->start)
4278 static int log_extent_csums(struct btrfs_trans_handle *trans,
4279 struct btrfs_inode *inode,
4280 struct btrfs_root *log_root,
4281 const struct extent_map *em,
4282 struct btrfs_log_ctx *ctx)
4284 struct btrfs_ordered_extent *ordered;
4287 u64 mod_start = em->mod_start;
4288 u64 mod_len = em->mod_len;
4289 LIST_HEAD(ordered_sums);
4292 if (inode->flags & BTRFS_INODE_NODATASUM ||
4293 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4294 em->block_start == EXTENT_MAP_HOLE)
4297 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4298 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4299 const u64 mod_end = mod_start + mod_len;
4300 struct btrfs_ordered_sum *sums;
4305 if (ordered_end <= mod_start)
4307 if (mod_end <= ordered->file_offset)
4311 * We are going to copy all the csums on this ordered extent, so
4312 * go ahead and adjust mod_start and mod_len in case this ordered
4313 * extent has already been logged.
4315 if (ordered->file_offset > mod_start) {
4316 if (ordered_end >= mod_end)
4317 mod_len = ordered->file_offset - mod_start;
4319 * If we have this case
4321 * |--------- logged extent ---------|
4322 * |----- ordered extent ----|
4324 * Just don't mess with mod_start and mod_len, we'll
4325 * just end up logging more csums than we need and it
4329 if (ordered_end < mod_end) {
4330 mod_len = mod_end - ordered_end;
4331 mod_start = ordered_end;
4338 * To keep us from looping for the above case of an ordered
4339 * extent that falls inside of the logged extent.
4341 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4344 list_for_each_entry(sums, &ordered->list, list) {
4345 ret = log_csums(trans, inode, log_root, sums);
4351 /* We're done, found all csums in the ordered extents. */
4355 /* If we're compressed we have to save the entire range of csums. */
4356 if (em->compress_type) {
4358 csum_len = max(em->block_len, em->orig_block_len);
4360 csum_offset = mod_start - em->start;
4364 /* block start is already adjusted for the file extent offset. */
4365 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4366 em->block_start + csum_offset,
4367 em->block_start + csum_offset +
4368 csum_len - 1, &ordered_sums, 0);
4372 while (!list_empty(&ordered_sums)) {
4373 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4374 struct btrfs_ordered_sum,
4377 ret = log_csums(trans, inode, log_root, sums);
4378 list_del(&sums->list);
4385 static int log_one_extent(struct btrfs_trans_handle *trans,
4386 struct btrfs_inode *inode, struct btrfs_root *root,
4387 const struct extent_map *em,
4388 struct btrfs_path *path,
4389 struct btrfs_log_ctx *ctx)
4391 struct btrfs_drop_extents_args drop_args = { 0 };
4392 struct btrfs_root *log = root->log_root;
4393 struct btrfs_file_extent_item *fi;
4394 struct extent_buffer *leaf;
4395 struct btrfs_map_token token;
4396 struct btrfs_key key;
4397 u64 extent_offset = em->start - em->orig_start;
4401 ret = log_extent_csums(trans, inode, log, em, ctx);
4405 drop_args.path = path;
4406 drop_args.start = em->start;
4407 drop_args.end = em->start + em->len;
4408 drop_args.replace_extent = true;
4409 drop_args.extent_item_size = sizeof(*fi);
4410 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4414 if (!drop_args.extent_inserted) {
4415 key.objectid = btrfs_ino(inode);
4416 key.type = BTRFS_EXTENT_DATA_KEY;
4417 key.offset = em->start;
4419 ret = btrfs_insert_empty_item(trans, log, path, &key,
4424 leaf = path->nodes[0];
4425 btrfs_init_map_token(&token, leaf);
4426 fi = btrfs_item_ptr(leaf, path->slots[0],
4427 struct btrfs_file_extent_item);
4429 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4430 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4431 btrfs_set_token_file_extent_type(&token, fi,
4432 BTRFS_FILE_EXTENT_PREALLOC);
4434 btrfs_set_token_file_extent_type(&token, fi,
4435 BTRFS_FILE_EXTENT_REG);
4437 block_len = max(em->block_len, em->orig_block_len);
4438 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4439 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4441 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4442 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4443 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4446 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4448 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4449 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4452 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4453 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4454 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4455 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4456 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4457 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4458 btrfs_mark_buffer_dirty(leaf);
4460 btrfs_release_path(path);
4466 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4467 * lose them after doing a full/fast fsync and replaying the log. We scan the
4468 * subvolume's root instead of iterating the inode's extent map tree because
4469 * otherwise we can log incorrect extent items based on extent map conversion.
4470 * That can happen due to the fact that extent maps are merged when they
4471 * are not in the extent map tree's list of modified extents.
4473 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4474 struct btrfs_inode *inode,
4475 struct btrfs_path *path)
4477 struct btrfs_root *root = inode->root;
4478 struct btrfs_key key;
4479 const u64 i_size = i_size_read(&inode->vfs_inode);
4480 const u64 ino = btrfs_ino(inode);
4481 struct btrfs_path *dst_path = NULL;
4482 bool dropped_extents = false;
4483 u64 truncate_offset = i_size;
4484 struct extent_buffer *leaf;
4490 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4494 key.type = BTRFS_EXTENT_DATA_KEY;
4495 key.offset = i_size;
4496 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4501 * We must check if there is a prealloc extent that starts before the
4502 * i_size and crosses the i_size boundary. This is to ensure later we
4503 * truncate down to the end of that extent and not to the i_size, as
4504 * otherwise we end up losing part of the prealloc extent after a log
4505 * replay and with an implicit hole if there is another prealloc extent
4506 * that starts at an offset beyond i_size.
4508 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4513 struct btrfs_file_extent_item *ei;
4515 leaf = path->nodes[0];
4516 slot = path->slots[0];
4517 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4519 if (btrfs_file_extent_type(leaf, ei) ==
4520 BTRFS_FILE_EXTENT_PREALLOC) {
4523 btrfs_item_key_to_cpu(leaf, &key, slot);
4524 extent_end = key.offset +
4525 btrfs_file_extent_num_bytes(leaf, ei);
4527 if (extent_end > i_size)
4528 truncate_offset = extent_end;
4535 leaf = path->nodes[0];
4536 slot = path->slots[0];
4538 if (slot >= btrfs_header_nritems(leaf)) {
4540 ret = copy_items(trans, inode, dst_path, path,
4541 start_slot, ins_nr, 1, 0);
4546 ret = btrfs_next_leaf(root, path);
4556 btrfs_item_key_to_cpu(leaf, &key, slot);
4557 if (key.objectid > ino)
4559 if (WARN_ON_ONCE(key.objectid < ino) ||
4560 key.type < BTRFS_EXTENT_DATA_KEY ||
4561 key.offset < i_size) {
4565 if (!dropped_extents) {
4567 * Avoid logging extent items logged in past fsync calls
4568 * and leading to duplicate keys in the log tree.
4571 ret = btrfs_truncate_inode_items(trans,
4573 inode, truncate_offset,
4574 BTRFS_EXTENT_DATA_KEY,
4576 } while (ret == -EAGAIN);
4579 dropped_extents = true;
4586 dst_path = btrfs_alloc_path();
4594 ret = copy_items(trans, inode, dst_path, path,
4595 start_slot, ins_nr, 1, 0);
4597 btrfs_release_path(path);
4598 btrfs_free_path(dst_path);
4602 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4603 struct btrfs_root *root,
4604 struct btrfs_inode *inode,
4605 struct btrfs_path *path,
4606 struct btrfs_log_ctx *ctx)
4608 struct btrfs_ordered_extent *ordered;
4609 struct btrfs_ordered_extent *tmp;
4610 struct extent_map *em, *n;
4611 struct list_head extents;
4612 struct extent_map_tree *tree = &inode->extent_tree;
4616 INIT_LIST_HEAD(&extents);
4618 write_lock(&tree->lock);
4620 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4621 list_del_init(&em->list);
4623 * Just an arbitrary number, this can be really CPU intensive
4624 * once we start getting a lot of extents, and really once we
4625 * have a bunch of extents we just want to commit since it will
4628 if (++num > 32768) {
4629 list_del_init(&tree->modified_extents);
4634 if (em->generation < trans->transid)
4637 /* We log prealloc extents beyond eof later. */
4638 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4639 em->start >= i_size_read(&inode->vfs_inode))
4642 /* Need a ref to keep it from getting evicted from cache */
4643 refcount_inc(&em->refs);
4644 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4645 list_add_tail(&em->list, &extents);
4649 list_sort(NULL, &extents, extent_cmp);
4651 while (!list_empty(&extents)) {
4652 em = list_entry(extents.next, struct extent_map, list);
4654 list_del_init(&em->list);
4657 * If we had an error we just need to delete everybody from our
4661 clear_em_logging(tree, em);
4662 free_extent_map(em);
4666 write_unlock(&tree->lock);
4668 ret = log_one_extent(trans, inode, root, em, path, ctx);
4669 write_lock(&tree->lock);
4670 clear_em_logging(tree, em);
4671 free_extent_map(em);
4673 WARN_ON(!list_empty(&extents));
4674 write_unlock(&tree->lock);
4676 btrfs_release_path(path);
4678 ret = btrfs_log_prealloc_extents(trans, inode, path);
4683 * We have logged all extents successfully, now make sure the commit of
4684 * the current transaction waits for the ordered extents to complete
4685 * before it commits and wipes out the log trees, otherwise we would
4686 * lose data if an ordered extents completes after the transaction
4687 * commits and a power failure happens after the transaction commit.
4689 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4690 list_del_init(&ordered->log_list);
4691 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4693 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4694 spin_lock_irq(&inode->ordered_tree.lock);
4695 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4696 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4697 atomic_inc(&trans->transaction->pending_ordered);
4699 spin_unlock_irq(&inode->ordered_tree.lock);
4701 btrfs_put_ordered_extent(ordered);
4707 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4708 struct btrfs_path *path, u64 *size_ret)
4710 struct btrfs_key key;
4713 key.objectid = btrfs_ino(inode);
4714 key.type = BTRFS_INODE_ITEM_KEY;
4717 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4720 } else if (ret > 0) {
4723 struct btrfs_inode_item *item;
4725 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4726 struct btrfs_inode_item);
4727 *size_ret = btrfs_inode_size(path->nodes[0], item);
4729 * If the in-memory inode's i_size is smaller then the inode
4730 * size stored in the btree, return the inode's i_size, so
4731 * that we get a correct inode size after replaying the log
4732 * when before a power failure we had a shrinking truncate
4733 * followed by addition of a new name (rename / new hard link).
4734 * Otherwise return the inode size from the btree, to avoid
4735 * data loss when replaying a log due to previously doing a
4736 * write that expands the inode's size and logging a new name
4737 * immediately after.
4739 if (*size_ret > inode->vfs_inode.i_size)
4740 *size_ret = inode->vfs_inode.i_size;
4743 btrfs_release_path(path);
4748 * At the moment we always log all xattrs. This is to figure out at log replay
4749 * time which xattrs must have their deletion replayed. If a xattr is missing
4750 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4751 * because if a xattr is deleted, the inode is fsynced and a power failure
4752 * happens, causing the log to be replayed the next time the fs is mounted,
4753 * we want the xattr to not exist anymore (same behaviour as other filesystems
4754 * with a journal, ext3/4, xfs, f2fs, etc).
4756 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4757 struct btrfs_root *root,
4758 struct btrfs_inode *inode,
4759 struct btrfs_path *path,
4760 struct btrfs_path *dst_path)
4763 struct btrfs_key key;
4764 const u64 ino = btrfs_ino(inode);
4767 bool found_xattrs = false;
4769 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4773 key.type = BTRFS_XATTR_ITEM_KEY;
4776 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4781 int slot = path->slots[0];
4782 struct extent_buffer *leaf = path->nodes[0];
4783 int nritems = btrfs_header_nritems(leaf);
4785 if (slot >= nritems) {
4787 ret = copy_items(trans, inode, dst_path, path,
4788 start_slot, ins_nr, 1, 0);
4793 ret = btrfs_next_leaf(root, path);
4801 btrfs_item_key_to_cpu(leaf, &key, slot);
4802 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4809 found_xattrs = true;
4813 ret = copy_items(trans, inode, dst_path, path,
4814 start_slot, ins_nr, 1, 0);
4820 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4826 * When using the NO_HOLES feature if we punched a hole that causes the
4827 * deletion of entire leafs or all the extent items of the first leaf (the one
4828 * that contains the inode item and references) we may end up not processing
4829 * any extents, because there are no leafs with a generation matching the
4830 * current transaction that have extent items for our inode. So we need to find
4831 * if any holes exist and then log them. We also need to log holes after any
4832 * truncate operation that changes the inode's size.
4834 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4835 struct btrfs_root *root,
4836 struct btrfs_inode *inode,
4837 struct btrfs_path *path)
4839 struct btrfs_fs_info *fs_info = root->fs_info;
4840 struct btrfs_key key;
4841 const u64 ino = btrfs_ino(inode);
4842 const u64 i_size = i_size_read(&inode->vfs_inode);
4843 u64 prev_extent_end = 0;
4846 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4850 key.type = BTRFS_EXTENT_DATA_KEY;
4853 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4858 struct extent_buffer *leaf = path->nodes[0];
4860 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4861 ret = btrfs_next_leaf(root, path);
4868 leaf = path->nodes[0];
4871 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4872 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4875 /* We have a hole, log it. */
4876 if (prev_extent_end < key.offset) {
4877 const u64 hole_len = key.offset - prev_extent_end;
4880 * Release the path to avoid deadlocks with other code
4881 * paths that search the root while holding locks on
4882 * leafs from the log root.
4884 btrfs_release_path(path);
4885 ret = btrfs_insert_file_extent(trans, root->log_root,
4886 ino, prev_extent_end, 0,
4887 0, hole_len, 0, hole_len,
4893 * Search for the same key again in the root. Since it's
4894 * an extent item and we are holding the inode lock, the
4895 * key must still exist. If it doesn't just emit warning
4896 * and return an error to fall back to a transaction
4899 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4902 if (WARN_ON(ret > 0))
4904 leaf = path->nodes[0];
4907 prev_extent_end = btrfs_file_extent_end(path);
4912 if (prev_extent_end < i_size) {
4915 btrfs_release_path(path);
4916 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4917 ret = btrfs_insert_file_extent(trans, root->log_root,
4918 ino, prev_extent_end, 0, 0,
4919 hole_len, 0, hole_len,
4929 * When we are logging a new inode X, check if it doesn't have a reference that
4930 * matches the reference from some other inode Y created in a past transaction
4931 * and that was renamed in the current transaction. If we don't do this, then at
4932 * log replay time we can lose inode Y (and all its files if it's a directory):
4935 * echo "hello world" > /mnt/x/foobar
4938 * mkdir /mnt/x # or touch /mnt/x
4939 * xfs_io -c fsync /mnt/x
4941 * mount fs, trigger log replay
4943 * After the log replay procedure, we would lose the first directory and all its
4944 * files (file foobar).
4945 * For the case where inode Y is not a directory we simply end up losing it:
4947 * echo "123" > /mnt/foo
4949 * mv /mnt/foo /mnt/bar
4950 * echo "abc" > /mnt/foo
4951 * xfs_io -c fsync /mnt/foo
4954 * We also need this for cases where a snapshot entry is replaced by some other
4955 * entry (file or directory) otherwise we end up with an unreplayable log due to
4956 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4957 * if it were a regular entry:
4960 * btrfs subvolume snapshot /mnt /mnt/x/snap
4961 * btrfs subvolume delete /mnt/x/snap
4964 * fsync /mnt/x or fsync some new file inside it
4967 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4968 * the same transaction.
4970 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4972 const struct btrfs_key *key,
4973 struct btrfs_inode *inode,
4974 u64 *other_ino, u64 *other_parent)
4977 struct btrfs_path *search_path;
4980 u32 item_size = btrfs_item_size_nr(eb, slot);
4982 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4984 search_path = btrfs_alloc_path();
4987 search_path->search_commit_root = 1;
4988 search_path->skip_locking = 1;
4990 while (cur_offset < item_size) {
4994 unsigned long name_ptr;
4995 struct btrfs_dir_item *di;
4997 if (key->type == BTRFS_INODE_REF_KEY) {
4998 struct btrfs_inode_ref *iref;
5000 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5001 parent = key->offset;
5002 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5003 name_ptr = (unsigned long)(iref + 1);
5004 this_len = sizeof(*iref) + this_name_len;
5006 struct btrfs_inode_extref *extref;
5008 extref = (struct btrfs_inode_extref *)(ptr +
5010 parent = btrfs_inode_extref_parent(eb, extref);
5011 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5012 name_ptr = (unsigned long)&extref->name;
5013 this_len = sizeof(*extref) + this_name_len;
5016 if (this_name_len > name_len) {
5019 new_name = krealloc(name, this_name_len, GFP_NOFS);
5024 name_len = this_name_len;
5028 read_extent_buffer(eb, name, name_ptr, this_name_len);
5029 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5030 parent, name, this_name_len, 0);
5031 if (di && !IS_ERR(di)) {
5032 struct btrfs_key di_key;
5034 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5036 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5037 if (di_key.objectid != key->objectid) {
5039 *other_ino = di_key.objectid;
5040 *other_parent = parent;
5048 } else if (IS_ERR(di)) {
5052 btrfs_release_path(search_path);
5054 cur_offset += this_len;
5058 btrfs_free_path(search_path);
5063 struct btrfs_ino_list {
5066 struct list_head list;
5069 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5070 struct btrfs_root *root,
5071 struct btrfs_path *path,
5072 struct btrfs_log_ctx *ctx,
5073 u64 ino, u64 parent)
5075 struct btrfs_ino_list *ino_elem;
5076 LIST_HEAD(inode_list);
5079 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5082 ino_elem->ino = ino;
5083 ino_elem->parent = parent;
5084 list_add_tail(&ino_elem->list, &inode_list);
5086 while (!list_empty(&inode_list)) {
5087 struct btrfs_fs_info *fs_info = root->fs_info;
5088 struct btrfs_key key;
5089 struct inode *inode;
5091 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
5093 ino = ino_elem->ino;
5094 parent = ino_elem->parent;
5095 list_del(&ino_elem->list);
5100 btrfs_release_path(path);
5102 inode = btrfs_iget(fs_info->sb, ino, root);
5104 * If the other inode that had a conflicting dir entry was
5105 * deleted in the current transaction, we need to log its parent
5108 if (IS_ERR(inode)) {
5109 ret = PTR_ERR(inode);
5110 if (ret == -ENOENT) {
5111 inode = btrfs_iget(fs_info->sb, parent, root);
5112 if (IS_ERR(inode)) {
5113 ret = PTR_ERR(inode);
5115 ret = btrfs_log_inode(trans, root,
5117 LOG_OTHER_INODE_ALL,
5119 btrfs_add_delayed_iput(inode);
5125 * If the inode was already logged skip it - otherwise we can
5126 * hit an infinite loop. Example:
5128 * From the commit root (previous transaction) we have the
5131 * inode 257 a directory
5132 * inode 258 with references "zz" and "zz_link" on inode 257
5133 * inode 259 with reference "a" on inode 257
5135 * And in the current (uncommitted) transaction we have:
5137 * inode 257 a directory, unchanged
5138 * inode 258 with references "a" and "a2" on inode 257
5139 * inode 259 with reference "zz_link" on inode 257
5140 * inode 261 with reference "zz" on inode 257
5142 * When logging inode 261 the following infinite loop could
5143 * happen if we don't skip already logged inodes:
5145 * - we detect inode 258 as a conflicting inode, with inode 261
5146 * on reference "zz", and log it;
5148 * - we detect inode 259 as a conflicting inode, with inode 258
5149 * on reference "a", and log it;
5151 * - we detect inode 258 as a conflicting inode, with inode 259
5152 * on reference "zz_link", and log it - again! After this we
5153 * repeat the above steps forever.
5155 spin_lock(&BTRFS_I(inode)->lock);
5157 * Check the inode's logged_trans only instead of
5158 * btrfs_inode_in_log(). This is because the last_log_commit of
5159 * the inode is not updated when we only log that it exists (see
5160 * btrfs_log_inode()).
5162 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5163 spin_unlock(&BTRFS_I(inode)->lock);
5164 btrfs_add_delayed_iput(inode);
5167 spin_unlock(&BTRFS_I(inode)->lock);
5169 * We are safe logging the other inode without acquiring its
5170 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5171 * are safe against concurrent renames of the other inode as
5172 * well because during a rename we pin the log and update the
5173 * log with the new name before we unpin it.
5175 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5176 LOG_OTHER_INODE, ctx);
5178 btrfs_add_delayed_iput(inode);
5183 key.type = BTRFS_INODE_REF_KEY;
5185 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5187 btrfs_add_delayed_iput(inode);
5192 struct extent_buffer *leaf = path->nodes[0];
5193 int slot = path->slots[0];
5195 u64 other_parent = 0;
5197 if (slot >= btrfs_header_nritems(leaf)) {
5198 ret = btrfs_next_leaf(root, path);
5201 } else if (ret > 0) {
5208 btrfs_item_key_to_cpu(leaf, &key, slot);
5209 if (key.objectid != ino ||
5210 (key.type != BTRFS_INODE_REF_KEY &&
5211 key.type != BTRFS_INODE_EXTREF_KEY)) {
5216 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5217 BTRFS_I(inode), &other_ino,
5222 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5227 ino_elem->ino = other_ino;
5228 ino_elem->parent = other_parent;
5229 list_add_tail(&ino_elem->list, &inode_list);
5234 btrfs_add_delayed_iput(inode);
5240 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5241 struct btrfs_inode *inode,
5242 struct btrfs_key *min_key,
5243 const struct btrfs_key *max_key,
5244 struct btrfs_path *path,
5245 struct btrfs_path *dst_path,
5246 const u64 logged_isize,
5247 const bool recursive_logging,
5248 const int inode_only,
5249 struct btrfs_log_ctx *ctx,
5250 bool *need_log_inode_item)
5252 const u64 i_size = i_size_read(&inode->vfs_inode);
5253 struct btrfs_root *root = inode->root;
5254 int ins_start_slot = 0;
5259 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5267 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5268 if (min_key->objectid != max_key->objectid)
5270 if (min_key->type > max_key->type)
5273 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5274 *need_log_inode_item = false;
5275 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5276 min_key->offset >= i_size) {
5278 * Extents at and beyond eof are logged with
5279 * btrfs_log_prealloc_extents().
5280 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5281 * and no keys greater than that, so bail out.
5284 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5285 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5286 inode->generation == trans->transid &&
5287 !recursive_logging) {
5289 u64 other_parent = 0;
5291 ret = btrfs_check_ref_name_override(path->nodes[0],
5292 path->slots[0], min_key, inode,
5293 &other_ino, &other_parent);
5296 } else if (ret > 0 && ctx &&
5297 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5302 ins_start_slot = path->slots[0];
5304 ret = copy_items(trans, inode, dst_path, path,
5305 ins_start_slot, ins_nr,
5306 inode_only, logged_isize);
5311 ret = log_conflicting_inodes(trans, root, path,
5312 ctx, other_ino, other_parent);
5315 btrfs_release_path(path);
5318 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5319 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5322 ret = copy_items(trans, inode, dst_path, path,
5324 ins_nr, inode_only, logged_isize);
5331 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5334 } else if (!ins_nr) {
5335 ins_start_slot = path->slots[0];
5340 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5341 ins_nr, inode_only, logged_isize);
5345 ins_start_slot = path->slots[0];
5348 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5349 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5354 ret = copy_items(trans, inode, dst_path, path,
5355 ins_start_slot, ins_nr, inode_only,
5361 btrfs_release_path(path);
5363 if (min_key->offset < (u64)-1) {
5365 } else if (min_key->type < max_key->type) {
5367 min_key->offset = 0;
5373 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5374 ins_nr, inode_only, logged_isize);
5379 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5381 * Release the path because otherwise we might attempt to double
5382 * lock the same leaf with btrfs_log_prealloc_extents() below.
5384 btrfs_release_path(path);
5385 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5391 /* log a single inode in the tree log.
5392 * At least one parent directory for this inode must exist in the tree
5393 * or be logged already.
5395 * Any items from this inode changed by the current transaction are copied
5396 * to the log tree. An extra reference is taken on any extents in this
5397 * file, allowing us to avoid a whole pile of corner cases around logging
5398 * blocks that have been removed from the tree.
5400 * See LOG_INODE_ALL and related defines for a description of what inode_only
5403 * This handles both files and directories.
5405 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5406 struct btrfs_root *root, struct btrfs_inode *inode,
5408 struct btrfs_log_ctx *ctx)
5410 struct btrfs_path *path;
5411 struct btrfs_path *dst_path;
5412 struct btrfs_key min_key;
5413 struct btrfs_key max_key;
5414 struct btrfs_root *log = root->log_root;
5417 bool fast_search = false;
5418 u64 ino = btrfs_ino(inode);
5419 struct extent_map_tree *em_tree = &inode->extent_tree;
5420 u64 logged_isize = 0;
5421 bool need_log_inode_item = true;
5422 bool xattrs_logged = false;
5423 bool recursive_logging = false;
5424 bool inode_item_dropped = true;
5426 path = btrfs_alloc_path();
5429 dst_path = btrfs_alloc_path();
5431 btrfs_free_path(path);
5435 min_key.objectid = ino;
5436 min_key.type = BTRFS_INODE_ITEM_KEY;
5439 max_key.objectid = ino;
5442 /* today the code can only do partial logging of directories */
5443 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5444 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5445 &inode->runtime_flags) &&
5446 inode_only >= LOG_INODE_EXISTS))
5447 max_key.type = BTRFS_XATTR_ITEM_KEY;
5449 max_key.type = (u8)-1;
5450 max_key.offset = (u64)-1;
5453 * Only run delayed items if we are a directory. We want to make sure
5454 * all directory indexes hit the fs/subvolume tree so we can find them
5455 * and figure out which index ranges have to be logged.
5457 * Otherwise commit the delayed inode only if the full sync flag is set,
5458 * as we want to make sure an up to date version is in the subvolume
5459 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5460 * it to the log tree. For a non full sync, we always log the inode item
5461 * based on the in-memory struct btrfs_inode which is always up to date.
5463 if (S_ISDIR(inode->vfs_inode.i_mode))
5464 ret = btrfs_commit_inode_delayed_items(trans, inode);
5465 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5466 ret = btrfs_commit_inode_delayed_inode(inode);
5469 btrfs_free_path(path);
5470 btrfs_free_path(dst_path);
5474 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5475 recursive_logging = true;
5476 if (inode_only == LOG_OTHER_INODE)
5477 inode_only = LOG_INODE_EXISTS;
5479 inode_only = LOG_INODE_ALL;
5480 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5482 mutex_lock(&inode->log_mutex);
5486 * This is for cases where logging a directory could result in losing a
5487 * a file after replaying the log. For example, if we move a file from a
5488 * directory A to a directory B, then fsync directory A, we have no way
5489 * to known the file was moved from A to B, so logging just A would
5490 * result in losing the file after a log replay.
5492 if (S_ISDIR(inode->vfs_inode.i_mode) &&
5493 inode_only == LOG_INODE_ALL &&
5494 inode->last_unlink_trans >= trans->transid) {
5495 btrfs_set_log_full_commit(trans);
5501 * a brute force approach to making sure we get the most uptodate
5502 * copies of everything.
5504 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5505 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5507 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5508 if (inode_only == LOG_INODE_EXISTS)
5509 max_key_type = BTRFS_XATTR_ITEM_KEY;
5510 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5512 if (inode_only == LOG_INODE_EXISTS) {
5514 * Make sure the new inode item we write to the log has
5515 * the same isize as the current one (if it exists).
5516 * This is necessary to prevent data loss after log
5517 * replay, and also to prevent doing a wrong expanding
5518 * truncate - for e.g. create file, write 4K into offset
5519 * 0, fsync, write 4K into offset 4096, add hard link,
5520 * fsync some other file (to sync log), power fail - if
5521 * we use the inode's current i_size, after log replay
5522 * we get a 8Kb file, with the last 4Kb extent as a hole
5523 * (zeroes), as if an expanding truncate happened,
5524 * instead of getting a file of 4Kb only.
5526 err = logged_inode_size(log, inode, path, &logged_isize);
5530 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5531 &inode->runtime_flags)) {
5532 if (inode_only == LOG_INODE_EXISTS) {
5533 max_key.type = BTRFS_XATTR_ITEM_KEY;
5534 ret = drop_objectid_items(trans, log, path, ino,
5537 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5538 &inode->runtime_flags);
5539 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5540 &inode->runtime_flags);
5542 ret = btrfs_truncate_inode_items(trans,
5543 log, inode, 0, 0, NULL);
5548 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5549 &inode->runtime_flags) ||
5550 inode_only == LOG_INODE_EXISTS) {
5551 if (inode_only == LOG_INODE_ALL)
5553 max_key.type = BTRFS_XATTR_ITEM_KEY;
5554 ret = drop_objectid_items(trans, log, path, ino,
5557 if (inode_only == LOG_INODE_ALL)
5559 inode_item_dropped = false;
5569 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5570 path, dst_path, logged_isize,
5571 recursive_logging, inode_only, ctx,
5572 &need_log_inode_item);
5576 btrfs_release_path(path);
5577 btrfs_release_path(dst_path);
5578 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5581 xattrs_logged = true;
5582 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5583 btrfs_release_path(path);
5584 btrfs_release_path(dst_path);
5585 err = btrfs_log_holes(trans, root, inode, path);
5590 btrfs_release_path(path);
5591 btrfs_release_path(dst_path);
5592 if (need_log_inode_item) {
5593 err = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
5597 * If we are doing a fast fsync and the inode was logged before
5598 * in this transaction, we don't need to log the xattrs because
5599 * they were logged before. If xattrs were added, changed or
5600 * deleted since the last time we logged the inode, then we have
5601 * already logged them because the inode had the runtime flag
5602 * BTRFS_INODE_COPY_EVERYTHING set.
5604 if (!xattrs_logged && inode->logged_trans < trans->transid) {
5605 err = btrfs_log_all_xattrs(trans, root, inode, path,
5609 btrfs_release_path(path);
5613 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5619 } else if (inode_only == LOG_INODE_ALL) {
5620 struct extent_map *em, *n;
5622 write_lock(&em_tree->lock);
5623 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5624 list_del_init(&em->list);
5625 write_unlock(&em_tree->lock);
5628 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5629 ret = log_directory_changes(trans, root, inode, path, dst_path,
5638 * If we are logging that an ancestor inode exists as part of logging a
5639 * new name from a link or rename operation, don't mark the inode as
5640 * logged - otherwise if an explicit fsync is made against an ancestor,
5641 * the fsync considers the inode in the log and doesn't sync the log,
5642 * resulting in the ancestor missing after a power failure unless the
5643 * log was synced as part of an fsync against any other unrelated inode.
5644 * So keep it simple for this case and just don't flag the ancestors as
5648 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5649 &inode->vfs_inode != ctx->inode)) {
5650 spin_lock(&inode->lock);
5651 inode->logged_trans = trans->transid;
5653 * Don't update last_log_commit if we logged that an inode exists.
5654 * We do this for two reasons:
5656 * 1) We might have had buffered writes to this inode that were
5657 * flushed and had their ordered extents completed in this
5658 * transaction, but we did not previously log the inode with
5659 * LOG_INODE_ALL. Later the inode was evicted and after that
5660 * it was loaded again and this LOG_INODE_EXISTS log operation
5661 * happened. We must make sure that if an explicit fsync against
5662 * the inode is performed later, it logs the new extents, an
5663 * updated inode item, etc, and syncs the log. The same logic
5664 * applies to direct IO writes instead of buffered writes.
5666 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
5667 * is logged with an i_size of 0 or whatever value was logged
5668 * before. If later the i_size of the inode is increased by a
5669 * truncate operation, the log is synced through an fsync of
5670 * some other inode and then finally an explicit fsync against
5671 * this inode is made, we must make sure this fsync logs the
5672 * inode with the new i_size, the hole between old i_size and
5673 * the new i_size, and syncs the log.
5675 if (inode_only != LOG_INODE_EXISTS)
5676 inode->last_log_commit = inode->last_sub_trans;
5677 spin_unlock(&inode->lock);
5680 mutex_unlock(&inode->log_mutex);
5682 btrfs_free_path(path);
5683 btrfs_free_path(dst_path);
5688 * Check if we need to log an inode. This is used in contexts where while
5689 * logging an inode we need to log another inode (either that it exists or in
5690 * full mode). This is used instead of btrfs_inode_in_log() because the later
5691 * requires the inode to be in the log and have the log transaction committed,
5692 * while here we do not care if the log transaction was already committed - our
5693 * caller will commit the log later - and we want to avoid logging an inode
5694 * multiple times when multiple tasks have joined the same log transaction.
5696 static bool need_log_inode(struct btrfs_trans_handle *trans,
5697 struct btrfs_inode *inode)
5700 * If a directory was not modified, no dentries added or removed, we can
5701 * and should avoid logging it.
5703 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5707 * If this inode does not have new/updated/deleted xattrs since the last
5708 * time it was logged and is flagged as logged in the current transaction,
5709 * we can skip logging it. As for new/deleted names, those are updated in
5710 * the log by link/unlink/rename operations.
5711 * In case the inode was logged and then evicted and reloaded, its
5712 * logged_trans will be 0, in which case we have to fully log it since
5713 * logged_trans is a transient field, not persisted.
5715 if (inode->logged_trans == trans->transid &&
5716 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5722 struct btrfs_dir_list {
5724 struct list_head list;
5728 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5729 * details about the why it is needed.
5730 * This is a recursive operation - if an existing dentry corresponds to a
5731 * directory, that directory's new entries are logged too (same behaviour as
5732 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5733 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5734 * complains about the following circular lock dependency / possible deadlock:
5738 * lock(&type->i_mutex_dir_key#3/2);
5739 * lock(sb_internal#2);
5740 * lock(&type->i_mutex_dir_key#3/2);
5741 * lock(&sb->s_type->i_mutex_key#14);
5743 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5744 * sb_start_intwrite() in btrfs_start_transaction().
5745 * Not locking i_mutex of the inodes is still safe because:
5747 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5748 * that while logging the inode new references (names) are added or removed
5749 * from the inode, leaving the logged inode item with a link count that does
5750 * not match the number of logged inode reference items. This is fine because
5751 * at log replay time we compute the real number of links and correct the
5752 * link count in the inode item (see replay_one_buffer() and
5753 * link_to_fixup_dir());
5755 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5756 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5757 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5758 * has a size that doesn't match the sum of the lengths of all the logged
5759 * names. This does not result in a problem because if a dir_item key is
5760 * logged but its matching dir_index key is not logged, at log replay time we
5761 * don't use it to replay the respective name (see replay_one_name()). On the
5762 * other hand if only the dir_index key ends up being logged, the respective
5763 * name is added to the fs/subvol tree with both the dir_item and dir_index
5764 * keys created (see replay_one_name()).
5765 * The directory's inode item with a wrong i_size is not a problem as well,
5766 * since we don't use it at log replay time to set the i_size in the inode
5767 * item of the fs/subvol tree (see overwrite_item()).
5769 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5770 struct btrfs_root *root,
5771 struct btrfs_inode *start_inode,
5772 struct btrfs_log_ctx *ctx)
5774 struct btrfs_fs_info *fs_info = root->fs_info;
5775 struct btrfs_root *log = root->log_root;
5776 struct btrfs_path *path;
5777 LIST_HEAD(dir_list);
5778 struct btrfs_dir_list *dir_elem;
5781 path = btrfs_alloc_path();
5785 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5787 btrfs_free_path(path);
5790 dir_elem->ino = btrfs_ino(start_inode);
5791 list_add_tail(&dir_elem->list, &dir_list);
5793 while (!list_empty(&dir_list)) {
5794 struct extent_buffer *leaf;
5795 struct btrfs_key min_key;
5799 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5802 goto next_dir_inode;
5804 min_key.objectid = dir_elem->ino;
5805 min_key.type = BTRFS_DIR_ITEM_KEY;
5808 btrfs_release_path(path);
5809 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5811 goto next_dir_inode;
5812 } else if (ret > 0) {
5814 goto next_dir_inode;
5818 leaf = path->nodes[0];
5819 nritems = btrfs_header_nritems(leaf);
5820 for (i = path->slots[0]; i < nritems; i++) {
5821 struct btrfs_dir_item *di;
5822 struct btrfs_key di_key;
5823 struct inode *di_inode;
5824 struct btrfs_dir_list *new_dir_elem;
5825 int log_mode = LOG_INODE_EXISTS;
5828 btrfs_item_key_to_cpu(leaf, &min_key, i);
5829 if (min_key.objectid != dir_elem->ino ||
5830 min_key.type != BTRFS_DIR_ITEM_KEY)
5831 goto next_dir_inode;
5833 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5834 type = btrfs_dir_type(leaf, di);
5835 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5836 type != BTRFS_FT_DIR)
5838 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5839 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5842 btrfs_release_path(path);
5843 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5844 if (IS_ERR(di_inode)) {
5845 ret = PTR_ERR(di_inode);
5846 goto next_dir_inode;
5849 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5850 btrfs_add_delayed_iput(di_inode);
5854 ctx->log_new_dentries = false;
5855 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5856 log_mode = LOG_INODE_ALL;
5857 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5859 btrfs_add_delayed_iput(di_inode);
5861 goto next_dir_inode;
5862 if (ctx->log_new_dentries) {
5863 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5865 if (!new_dir_elem) {
5867 goto next_dir_inode;
5869 new_dir_elem->ino = di_key.objectid;
5870 list_add_tail(&new_dir_elem->list, &dir_list);
5875 ret = btrfs_next_leaf(log, path);
5877 goto next_dir_inode;
5878 } else if (ret > 0) {
5880 goto next_dir_inode;
5884 if (min_key.offset < (u64)-1) {
5889 list_del(&dir_elem->list);
5893 btrfs_free_path(path);
5897 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5898 struct btrfs_inode *inode,
5899 struct btrfs_log_ctx *ctx)
5901 struct btrfs_fs_info *fs_info = trans->fs_info;
5903 struct btrfs_path *path;
5904 struct btrfs_key key;
5905 struct btrfs_root *root = inode->root;
5906 const u64 ino = btrfs_ino(inode);
5908 path = btrfs_alloc_path();
5911 path->skip_locking = 1;
5912 path->search_commit_root = 1;
5915 key.type = BTRFS_INODE_REF_KEY;
5917 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5922 struct extent_buffer *leaf = path->nodes[0];
5923 int slot = path->slots[0];
5928 if (slot >= btrfs_header_nritems(leaf)) {
5929 ret = btrfs_next_leaf(root, path);
5937 btrfs_item_key_to_cpu(leaf, &key, slot);
5938 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5939 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5942 item_size = btrfs_item_size_nr(leaf, slot);
5943 ptr = btrfs_item_ptr_offset(leaf, slot);
5944 while (cur_offset < item_size) {
5945 struct btrfs_key inode_key;
5946 struct inode *dir_inode;
5948 inode_key.type = BTRFS_INODE_ITEM_KEY;
5949 inode_key.offset = 0;
5951 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5952 struct btrfs_inode_extref *extref;
5954 extref = (struct btrfs_inode_extref *)
5956 inode_key.objectid = btrfs_inode_extref_parent(
5958 cur_offset += sizeof(*extref);
5959 cur_offset += btrfs_inode_extref_name_len(leaf,
5962 inode_key.objectid = key.offset;
5963 cur_offset = item_size;
5966 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5969 * If the parent inode was deleted, return an error to
5970 * fallback to a transaction commit. This is to prevent
5971 * getting an inode that was moved from one parent A to
5972 * a parent B, got its former parent A deleted and then
5973 * it got fsync'ed, from existing at both parents after
5974 * a log replay (and the old parent still existing).
5981 * mv /mnt/B/bar /mnt/A/bar
5982 * mv -T /mnt/A /mnt/B
5986 * If we ignore the old parent B which got deleted,
5987 * after a log replay we would have file bar linked
5988 * at both parents and the old parent B would still
5991 if (IS_ERR(dir_inode)) {
5992 ret = PTR_ERR(dir_inode);
5996 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
5997 btrfs_add_delayed_iput(dir_inode);
6002 ctx->log_new_dentries = false;
6003 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
6004 LOG_INODE_ALL, ctx);
6005 if (!ret && ctx && ctx->log_new_dentries)
6006 ret = log_new_dir_dentries(trans, root,
6007 BTRFS_I(dir_inode), ctx);
6008 btrfs_add_delayed_iput(dir_inode);
6016 btrfs_free_path(path);
6020 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6021 struct btrfs_root *root,
6022 struct btrfs_path *path,
6023 struct btrfs_log_ctx *ctx)
6025 struct btrfs_key found_key;
6027 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6030 struct btrfs_fs_info *fs_info = root->fs_info;
6031 struct extent_buffer *leaf = path->nodes[0];
6032 int slot = path->slots[0];
6033 struct btrfs_key search_key;
6034 struct inode *inode;
6038 btrfs_release_path(path);
6040 ino = found_key.offset;
6042 search_key.objectid = found_key.offset;
6043 search_key.type = BTRFS_INODE_ITEM_KEY;
6044 search_key.offset = 0;
6045 inode = btrfs_iget(fs_info->sb, ino, root);
6047 return PTR_ERR(inode);
6049 if (BTRFS_I(inode)->generation >= trans->transid &&
6050 need_log_inode(trans, BTRFS_I(inode)))
6051 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
6052 LOG_INODE_EXISTS, ctx);
6053 btrfs_add_delayed_iput(inode);
6057 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6060 search_key.type = BTRFS_INODE_REF_KEY;
6061 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6065 leaf = path->nodes[0];
6066 slot = path->slots[0];
6067 if (slot >= btrfs_header_nritems(leaf)) {
6068 ret = btrfs_next_leaf(root, path);
6073 leaf = path->nodes[0];
6074 slot = path->slots[0];
6077 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6078 if (found_key.objectid != search_key.objectid ||
6079 found_key.type != BTRFS_INODE_REF_KEY)
6085 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6086 struct btrfs_inode *inode,
6087 struct dentry *parent,
6088 struct btrfs_log_ctx *ctx)
6090 struct btrfs_root *root = inode->root;
6091 struct dentry *old_parent = NULL;
6092 struct super_block *sb = inode->vfs_inode.i_sb;
6096 if (!parent || d_really_is_negative(parent) ||
6100 inode = BTRFS_I(d_inode(parent));
6101 if (root != inode->root)
6104 if (inode->generation >= trans->transid &&
6105 need_log_inode(trans, inode)) {
6106 ret = btrfs_log_inode(trans, root, inode,
6107 LOG_INODE_EXISTS, ctx);
6111 if (IS_ROOT(parent))
6114 parent = dget_parent(parent);
6116 old_parent = parent;
6123 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6124 struct btrfs_inode *inode,
6125 struct dentry *parent,
6126 struct btrfs_log_ctx *ctx)
6128 struct btrfs_root *root = inode->root;
6129 const u64 ino = btrfs_ino(inode);
6130 struct btrfs_path *path;
6131 struct btrfs_key search_key;
6135 * For a single hard link case, go through a fast path that does not
6136 * need to iterate the fs/subvolume tree.
6138 if (inode->vfs_inode.i_nlink < 2)
6139 return log_new_ancestors_fast(trans, inode, parent, ctx);
6141 path = btrfs_alloc_path();
6145 search_key.objectid = ino;
6146 search_key.type = BTRFS_INODE_REF_KEY;
6147 search_key.offset = 0;
6149 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6156 struct extent_buffer *leaf = path->nodes[0];
6157 int slot = path->slots[0];
6158 struct btrfs_key found_key;
6160 if (slot >= btrfs_header_nritems(leaf)) {
6161 ret = btrfs_next_leaf(root, path);
6169 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6170 if (found_key.objectid != ino ||
6171 found_key.type > BTRFS_INODE_EXTREF_KEY)
6175 * Don't deal with extended references because they are rare
6176 * cases and too complex to deal with (we would need to keep
6177 * track of which subitem we are processing for each item in
6178 * this loop, etc). So just return some error to fallback to
6179 * a transaction commit.
6181 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6187 * Logging ancestors needs to do more searches on the fs/subvol
6188 * tree, so it releases the path as needed to avoid deadlocks.
6189 * Keep track of the last inode ref key and resume from that key
6190 * after logging all new ancestors for the current hard link.
6192 memcpy(&search_key, &found_key, sizeof(search_key));
6194 ret = log_new_ancestors(trans, root, path, ctx);
6197 btrfs_release_path(path);
6202 btrfs_free_path(path);
6207 * helper function around btrfs_log_inode to make sure newly created
6208 * parent directories also end up in the log. A minimal inode and backref
6209 * only logging is done of any parent directories that are older than
6210 * the last committed transaction
6212 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6213 struct btrfs_inode *inode,
6214 struct dentry *parent,
6216 struct btrfs_log_ctx *ctx)
6218 struct btrfs_root *root = inode->root;
6219 struct btrfs_fs_info *fs_info = root->fs_info;
6221 bool log_dentries = false;
6223 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6228 if (btrfs_root_refs(&root->root_item) == 0) {
6234 * Skip already logged inodes or inodes corresponding to tmpfiles
6235 * (since logging them is pointless, a link count of 0 means they
6236 * will never be accessible).
6238 if ((btrfs_inode_in_log(inode, trans->transid) &&
6239 list_empty(&ctx->ordered_extents)) ||
6240 inode->vfs_inode.i_nlink == 0) {
6241 ret = BTRFS_NO_LOG_SYNC;
6245 ret = start_log_trans(trans, root, ctx);
6249 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6254 * for regular files, if its inode is already on disk, we don't
6255 * have to worry about the parents at all. This is because
6256 * we can use the last_unlink_trans field to record renames
6257 * and other fun in this file.
6259 if (S_ISREG(inode->vfs_inode.i_mode) &&
6260 inode->generation < trans->transid &&
6261 inode->last_unlink_trans < trans->transid) {
6266 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6267 log_dentries = true;
6270 * On unlink we must make sure all our current and old parent directory
6271 * inodes are fully logged. This is to prevent leaving dangling
6272 * directory index entries in directories that were our parents but are
6273 * not anymore. Not doing this results in old parent directory being
6274 * impossible to delete after log replay (rmdir will always fail with
6275 * error -ENOTEMPTY).
6281 * ln testdir/foo testdir/bar
6283 * unlink testdir/bar
6284 * xfs_io -c fsync testdir/foo
6286 * mount fs, triggers log replay
6288 * If we don't log the parent directory (testdir), after log replay the
6289 * directory still has an entry pointing to the file inode using the bar
6290 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6291 * the file inode has a link count of 1.
6297 * ln foo testdir/foo2
6298 * ln foo testdir/foo3
6300 * unlink testdir/foo3
6301 * xfs_io -c fsync foo
6303 * mount fs, triggers log replay
6305 * Similar as the first example, after log replay the parent directory
6306 * testdir still has an entry pointing to the inode file with name foo3
6307 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6308 * and has a link count of 2.
6310 if (inode->last_unlink_trans >= trans->transid) {
6311 ret = btrfs_log_all_parents(trans, inode, ctx);
6316 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6321 ret = log_new_dir_dentries(trans, root, inode, ctx);
6326 btrfs_set_log_full_commit(trans);
6331 btrfs_remove_log_ctx(root, ctx);
6332 btrfs_end_log_trans(root);
6338 * it is not safe to log dentry if the chunk root has added new
6339 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6340 * If this returns 1, you must commit the transaction to safely get your
6343 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6344 struct dentry *dentry,
6345 struct btrfs_log_ctx *ctx)
6347 struct dentry *parent = dget_parent(dentry);
6350 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6351 LOG_INODE_ALL, ctx);
6358 * should be called during mount to recover any replay any log trees
6361 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6364 struct btrfs_path *path;
6365 struct btrfs_trans_handle *trans;
6366 struct btrfs_key key;
6367 struct btrfs_key found_key;
6368 struct btrfs_root *log;
6369 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6370 struct walk_control wc = {
6371 .process_func = process_one_buffer,
6372 .stage = LOG_WALK_PIN_ONLY,
6375 path = btrfs_alloc_path();
6379 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6381 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6382 if (IS_ERR(trans)) {
6383 ret = PTR_ERR(trans);
6390 ret = walk_log_tree(trans, log_root_tree, &wc);
6392 btrfs_handle_fs_error(fs_info, ret,
6393 "Failed to pin buffers while recovering log root tree.");
6398 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6399 key.offset = (u64)-1;
6400 key.type = BTRFS_ROOT_ITEM_KEY;
6403 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6406 btrfs_handle_fs_error(fs_info, ret,
6407 "Couldn't find tree log root.");
6411 if (path->slots[0] == 0)
6415 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6417 btrfs_release_path(path);
6418 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6421 log = btrfs_read_tree_root(log_root_tree, &found_key);
6424 btrfs_handle_fs_error(fs_info, ret,
6425 "Couldn't read tree log root.");
6429 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6431 if (IS_ERR(wc.replay_dest)) {
6432 ret = PTR_ERR(wc.replay_dest);
6435 * We didn't find the subvol, likely because it was
6436 * deleted. This is ok, simply skip this log and go to
6439 * We need to exclude the root because we can't have
6440 * other log replays overwriting this log as we'll read
6441 * it back in a few more times. This will keep our
6442 * block from being modified, and we'll just bail for
6443 * each subsequent pass.
6446 ret = btrfs_pin_extent_for_log_replay(trans,
6449 btrfs_put_root(log);
6453 btrfs_handle_fs_error(fs_info, ret,
6454 "Couldn't read target root for tree log recovery.");
6458 wc.replay_dest->log_root = log;
6459 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
6461 /* The loop needs to continue due to the root refs */
6462 btrfs_handle_fs_error(fs_info, ret,
6463 "failed to record the log root in transaction");
6465 ret = walk_log_tree(trans, log, &wc);
6467 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6468 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6472 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6473 struct btrfs_root *root = wc.replay_dest;
6475 btrfs_release_path(path);
6478 * We have just replayed everything, and the highest
6479 * objectid of fs roots probably has changed in case
6480 * some inode_item's got replayed.
6482 * root->objectid_mutex is not acquired as log replay
6483 * could only happen during mount.
6485 ret = btrfs_init_root_free_objectid(root);
6488 wc.replay_dest->log_root = NULL;
6489 btrfs_put_root(wc.replay_dest);
6490 btrfs_put_root(log);
6495 if (found_key.offset == 0)
6497 key.offset = found_key.offset - 1;
6499 btrfs_release_path(path);
6501 /* step one is to pin it all, step two is to replay just inodes */
6504 wc.process_func = replay_one_buffer;
6505 wc.stage = LOG_WALK_REPLAY_INODES;
6508 /* step three is to replay everything */
6509 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6514 btrfs_free_path(path);
6516 /* step 4: commit the transaction, which also unpins the blocks */
6517 ret = btrfs_commit_transaction(trans);
6521 log_root_tree->log_root = NULL;
6522 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6523 btrfs_put_root(log_root_tree);
6528 btrfs_end_transaction(wc.trans);
6529 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6530 btrfs_free_path(path);
6535 * there are some corner cases where we want to force a full
6536 * commit instead of allowing a directory to be logged.
6538 * They revolve around files there were unlinked from the directory, and
6539 * this function updates the parent directory so that a full commit is
6540 * properly done if it is fsync'd later after the unlinks are done.
6542 * Must be called before the unlink operations (updates to the subvolume tree,
6543 * inodes, etc) are done.
6545 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6546 struct btrfs_inode *dir, struct btrfs_inode *inode,
6550 * when we're logging a file, if it hasn't been renamed
6551 * or unlinked, and its inode is fully committed on disk,
6552 * we don't have to worry about walking up the directory chain
6553 * to log its parents.
6555 * So, we use the last_unlink_trans field to put this transid
6556 * into the file. When the file is logged we check it and
6557 * don't log the parents if the file is fully on disk.
6559 mutex_lock(&inode->log_mutex);
6560 inode->last_unlink_trans = trans->transid;
6561 mutex_unlock(&inode->log_mutex);
6564 * if this directory was already logged any new
6565 * names for this file/dir will get recorded
6567 if (dir->logged_trans == trans->transid)
6571 * if the inode we're about to unlink was logged,
6572 * the log will be properly updated for any new names
6574 if (inode->logged_trans == trans->transid)
6578 * when renaming files across directories, if the directory
6579 * there we're unlinking from gets fsync'd later on, there's
6580 * no way to find the destination directory later and fsync it
6581 * properly. So, we have to be conservative and force commits
6582 * so the new name gets discovered.
6587 /* we can safely do the unlink without any special recording */
6591 mutex_lock(&dir->log_mutex);
6592 dir->last_unlink_trans = trans->transid;
6593 mutex_unlock(&dir->log_mutex);
6597 * Make sure that if someone attempts to fsync the parent directory of a deleted
6598 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6599 * that after replaying the log tree of the parent directory's root we will not
6600 * see the snapshot anymore and at log replay time we will not see any log tree
6601 * corresponding to the deleted snapshot's root, which could lead to replaying
6602 * it after replaying the log tree of the parent directory (which would replay
6603 * the snapshot delete operation).
6605 * Must be called before the actual snapshot destroy operation (updates to the
6606 * parent root and tree of tree roots trees, etc) are done.
6608 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6609 struct btrfs_inode *dir)
6611 mutex_lock(&dir->log_mutex);
6612 dir->last_unlink_trans = trans->transid;
6613 mutex_unlock(&dir->log_mutex);
6617 * Call this after adding a new name for a file and it will properly
6618 * update the log to reflect the new name.
6620 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6621 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6622 struct dentry *parent)
6624 struct btrfs_log_ctx ctx;
6627 * this will force the logging code to walk the dentry chain
6630 if (!S_ISDIR(inode->vfs_inode.i_mode))
6631 inode->last_unlink_trans = trans->transid;
6634 * if this inode hasn't been logged and directory we're renaming it
6635 * from hasn't been logged, we don't need to log it
6637 if (!inode_logged(trans, inode) &&
6638 (!old_dir || !inode_logged(trans, old_dir)))
6642 * If we are doing a rename (old_dir is not NULL) from a directory that
6643 * was previously logged, make sure the next log attempt on the directory
6644 * is not skipped and logs the inode again. This is because the log may
6645 * not currently be authoritative for a range including the old
6646 * BTRFS_DIR_ITEM_KEY and BTRFS_DIR_INDEX_KEY keys, so we want to make
6647 * sure after a log replay we do not end up with both the new and old
6648 * dentries around (in case the inode is a directory we would have a
6649 * directory with two hard links and 2 inode references for different
6650 * parents). The next log attempt of old_dir will happen at
6651 * btrfs_log_all_parents(), called through btrfs_log_inode_parent()
6652 * below, because we have previously set inode->last_unlink_trans to the
6653 * current transaction ID, either here or at btrfs_record_unlink_dir() in
6654 * case inode is a directory.
6657 old_dir->logged_trans = 0;
6659 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6660 ctx.logging_new_name = true;
6662 * We don't care about the return value. If we fail to log the new name
6663 * then we know the next attempt to sync the log will fallback to a full
6664 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6665 * we don't need to worry about getting a log committed that has an
6666 * inconsistent state after a rename operation.
6668 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);