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 * helper function to see if a given name and sequence number found
943 * in an inode back reference are already in a directory and correctly
944 * point to this inode
946 static noinline int inode_in_dir(struct btrfs_root *root,
947 struct btrfs_path *path,
948 u64 dirid, u64 objectid, u64 index,
949 const char *name, int name_len)
951 struct btrfs_dir_item *di;
952 struct btrfs_key location;
955 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
956 index, name, name_len, 0);
957 if (di && !IS_ERR(di)) {
958 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
959 if (location.objectid != objectid)
963 btrfs_release_path(path);
965 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
966 if (di && !IS_ERR(di)) {
967 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
968 if (location.objectid != objectid)
974 btrfs_release_path(path);
979 * helper function to check a log tree for a named back reference in
980 * an inode. This is used to decide if a back reference that is
981 * found in the subvolume conflicts with what we find in the log.
983 * inode backreferences may have multiple refs in a single item,
984 * during replay we process one reference at a time, and we don't
985 * want to delete valid links to a file from the subvolume if that
986 * link is also in the log.
988 static noinline int backref_in_log(struct btrfs_root *log,
989 struct btrfs_key *key,
991 const char *name, int namelen)
993 struct btrfs_path *path;
996 path = btrfs_alloc_path();
1000 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1003 } else if (ret == 1) {
1008 if (key->type == BTRFS_INODE_EXTREF_KEY)
1009 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1014 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1018 btrfs_free_path(path);
1022 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1023 struct btrfs_root *root,
1024 struct btrfs_path *path,
1025 struct btrfs_root *log_root,
1026 struct btrfs_inode *dir,
1027 struct btrfs_inode *inode,
1028 u64 inode_objectid, u64 parent_objectid,
1029 u64 ref_index, char *name, int namelen,
1034 int victim_name_len;
1035 struct extent_buffer *leaf;
1036 struct btrfs_dir_item *di;
1037 struct btrfs_key search_key;
1038 struct btrfs_inode_extref *extref;
1041 /* Search old style refs */
1042 search_key.objectid = inode_objectid;
1043 search_key.type = BTRFS_INODE_REF_KEY;
1044 search_key.offset = parent_objectid;
1045 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1047 struct btrfs_inode_ref *victim_ref;
1049 unsigned long ptr_end;
1051 leaf = path->nodes[0];
1053 /* are we trying to overwrite a back ref for the root directory
1054 * if so, just jump out, we're done
1056 if (search_key.objectid == search_key.offset)
1059 /* check all the names in this back reference to see
1060 * if they are in the log. if so, we allow them to stay
1061 * otherwise they must be unlinked as a conflict
1063 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1064 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1065 while (ptr < ptr_end) {
1066 victim_ref = (struct btrfs_inode_ref *)ptr;
1067 victim_name_len = btrfs_inode_ref_name_len(leaf,
1069 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1073 read_extent_buffer(leaf, victim_name,
1074 (unsigned long)(victim_ref + 1),
1077 ret = backref_in_log(log_root, &search_key,
1078 parent_objectid, victim_name,
1084 inc_nlink(&inode->vfs_inode);
1085 btrfs_release_path(path);
1087 ret = btrfs_unlink_inode(trans, root, dir, inode,
1088 victim_name, victim_name_len);
1092 ret = btrfs_run_delayed_items(trans);
1100 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1104 * NOTE: we have searched root tree and checked the
1105 * corresponding ref, it does not need to check again.
1109 btrfs_release_path(path);
1111 /* Same search but for extended refs */
1112 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1113 inode_objectid, parent_objectid, 0,
1115 if (!IS_ERR_OR_NULL(extref)) {
1119 struct inode *victim_parent;
1121 leaf = path->nodes[0];
1123 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1124 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1126 while (cur_offset < item_size) {
1127 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1129 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1131 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1134 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1137 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1140 search_key.objectid = inode_objectid;
1141 search_key.type = BTRFS_INODE_EXTREF_KEY;
1142 search_key.offset = btrfs_extref_hash(parent_objectid,
1145 ret = backref_in_log(log_root, &search_key,
1146 parent_objectid, victim_name,
1152 victim_parent = read_one_inode(root,
1154 if (victim_parent) {
1155 inc_nlink(&inode->vfs_inode);
1156 btrfs_release_path(path);
1158 ret = btrfs_unlink_inode(trans, root,
1159 BTRFS_I(victim_parent),
1164 ret = btrfs_run_delayed_items(
1167 iput(victim_parent);
1176 cur_offset += victim_name_len + sizeof(*extref);
1180 btrfs_release_path(path);
1182 /* look for a conflicting sequence number */
1183 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1184 ref_index, name, namelen, 0);
1185 if (di && !IS_ERR(di)) {
1186 ret = drop_one_dir_item(trans, root, path, dir, di);
1190 btrfs_release_path(path);
1192 /* look for a conflicting name */
1193 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1195 if (di && !IS_ERR(di)) {
1196 ret = drop_one_dir_item(trans, root, path, dir, di);
1200 btrfs_release_path(path);
1205 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1206 u32 *namelen, char **name, u64 *index,
1207 u64 *parent_objectid)
1209 struct btrfs_inode_extref *extref;
1211 extref = (struct btrfs_inode_extref *)ref_ptr;
1213 *namelen = btrfs_inode_extref_name_len(eb, extref);
1214 *name = kmalloc(*namelen, GFP_NOFS);
1218 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1222 *index = btrfs_inode_extref_index(eb, extref);
1223 if (parent_objectid)
1224 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1229 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1230 u32 *namelen, char **name, u64 *index)
1232 struct btrfs_inode_ref *ref;
1234 ref = (struct btrfs_inode_ref *)ref_ptr;
1236 *namelen = btrfs_inode_ref_name_len(eb, ref);
1237 *name = kmalloc(*namelen, GFP_NOFS);
1241 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1244 *index = btrfs_inode_ref_index(eb, ref);
1250 * Take an inode reference item from the log tree and iterate all names from the
1251 * inode reference item in the subvolume tree with the same key (if it exists).
1252 * For any name that is not in the inode reference item from the log tree, do a
1253 * proper unlink of that name (that is, remove its entry from the inode
1254 * reference item and both dir index keys).
1256 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1257 struct btrfs_root *root,
1258 struct btrfs_path *path,
1259 struct btrfs_inode *inode,
1260 struct extent_buffer *log_eb,
1262 struct btrfs_key *key)
1265 unsigned long ref_ptr;
1266 unsigned long ref_end;
1267 struct extent_buffer *eb;
1270 btrfs_release_path(path);
1271 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1279 eb = path->nodes[0];
1280 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1281 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1282 while (ref_ptr < ref_end) {
1287 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1288 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1291 parent_id = key->offset;
1292 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1298 if (key->type == BTRFS_INODE_EXTREF_KEY)
1299 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1303 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1309 btrfs_release_path(path);
1310 dir = read_one_inode(root, parent_id);
1316 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1317 inode, name, namelen);
1327 if (key->type == BTRFS_INODE_EXTREF_KEY)
1328 ref_ptr += sizeof(struct btrfs_inode_extref);
1330 ref_ptr += sizeof(struct btrfs_inode_ref);
1334 btrfs_release_path(path);
1338 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1339 const u8 ref_type, const char *name,
1342 struct btrfs_key key;
1343 struct btrfs_path *path;
1344 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1347 path = btrfs_alloc_path();
1351 key.objectid = btrfs_ino(BTRFS_I(inode));
1352 key.type = ref_type;
1353 if (key.type == BTRFS_INODE_REF_KEY)
1354 key.offset = parent_id;
1356 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1358 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1365 if (key.type == BTRFS_INODE_EXTREF_KEY)
1366 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1367 path->slots[0], parent_id, name, namelen);
1369 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1373 btrfs_free_path(path);
1377 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1378 struct inode *dir, struct inode *inode, const char *name,
1379 int namelen, u64 ref_index)
1381 struct btrfs_dir_item *dir_item;
1382 struct btrfs_key key;
1383 struct btrfs_path *path;
1384 struct inode *other_inode = NULL;
1387 path = btrfs_alloc_path();
1391 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1392 btrfs_ino(BTRFS_I(dir)),
1395 btrfs_release_path(path);
1397 } else if (IS_ERR(dir_item)) {
1398 ret = PTR_ERR(dir_item);
1403 * Our inode's dentry collides with the dentry of another inode which is
1404 * in the log but not yet processed since it has a higher inode number.
1405 * So delete that other dentry.
1407 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1408 btrfs_release_path(path);
1409 other_inode = read_one_inode(root, key.objectid);
1414 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1419 * If we dropped the link count to 0, bump it so that later the iput()
1420 * on the inode will not free it. We will fixup the link count later.
1422 if (other_inode->i_nlink == 0)
1423 inc_nlink(other_inode);
1425 ret = btrfs_run_delayed_items(trans);
1429 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1430 name, namelen, 0, ref_index);
1433 btrfs_free_path(path);
1439 * replay one inode back reference item found in the log tree.
1440 * eb, slot and key refer to the buffer and key found in the log tree.
1441 * root is the destination we are replaying into, and path is for temp
1442 * use by this function. (it should be released on return).
1444 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1445 struct btrfs_root *root,
1446 struct btrfs_root *log,
1447 struct btrfs_path *path,
1448 struct extent_buffer *eb, int slot,
1449 struct btrfs_key *key)
1451 struct inode *dir = NULL;
1452 struct inode *inode = NULL;
1453 unsigned long ref_ptr;
1454 unsigned long ref_end;
1458 int search_done = 0;
1459 int log_ref_ver = 0;
1460 u64 parent_objectid;
1463 int ref_struct_size;
1465 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1466 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1468 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1469 struct btrfs_inode_extref *r;
1471 ref_struct_size = sizeof(struct btrfs_inode_extref);
1473 r = (struct btrfs_inode_extref *)ref_ptr;
1474 parent_objectid = btrfs_inode_extref_parent(eb, r);
1476 ref_struct_size = sizeof(struct btrfs_inode_ref);
1477 parent_objectid = key->offset;
1479 inode_objectid = key->objectid;
1482 * it is possible that we didn't log all the parent directories
1483 * for a given inode. If we don't find the dir, just don't
1484 * copy the back ref in. The link count fixup code will take
1487 dir = read_one_inode(root, parent_objectid);
1493 inode = read_one_inode(root, inode_objectid);
1499 while (ref_ptr < ref_end) {
1501 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1502 &ref_index, &parent_objectid);
1504 * parent object can change from one array
1508 dir = read_one_inode(root, parent_objectid);
1514 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1520 /* if we already have a perfect match, we're done */
1521 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1522 btrfs_ino(BTRFS_I(inode)), ref_index,
1525 * look for a conflicting back reference in the
1526 * metadata. if we find one we have to unlink that name
1527 * of the file before we add our new link. Later on, we
1528 * overwrite any existing back reference, and we don't
1529 * want to create dangling pointers in the directory.
1533 ret = __add_inode_ref(trans, root, path, log,
1538 ref_index, name, namelen,
1548 * If a reference item already exists for this inode
1549 * with the same parent and name, but different index,
1550 * drop it and the corresponding directory index entries
1551 * from the parent before adding the new reference item
1552 * and dir index entries, otherwise we would fail with
1553 * -EEXIST returned from btrfs_add_link() below.
1555 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1558 ret = btrfs_unlink_inode(trans, root,
1563 * If we dropped the link count to 0, bump it so
1564 * that later the iput() on the inode will not
1565 * free it. We will fixup the link count later.
1567 if (!ret && inode->i_nlink == 0)
1573 /* insert our name */
1574 ret = add_link(trans, root, dir, inode, name, namelen,
1579 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1584 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1594 * Before we overwrite the inode reference item in the subvolume tree
1595 * with the item from the log tree, we must unlink all names from the
1596 * parent directory that are in the subvolume's tree inode reference
1597 * item, otherwise we end up with an inconsistent subvolume tree where
1598 * dir index entries exist for a name but there is no inode reference
1599 * item with the same name.
1601 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1606 /* finally write the back reference in the inode */
1607 ret = overwrite_item(trans, root, path, eb, slot, key);
1609 btrfs_release_path(path);
1616 static int count_inode_extrefs(struct btrfs_root *root,
1617 struct btrfs_inode *inode, struct btrfs_path *path)
1621 unsigned int nlink = 0;
1624 u64 inode_objectid = btrfs_ino(inode);
1627 struct btrfs_inode_extref *extref;
1628 struct extent_buffer *leaf;
1631 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1636 leaf = path->nodes[0];
1637 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1638 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1641 while (cur_offset < item_size) {
1642 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1643 name_len = btrfs_inode_extref_name_len(leaf, extref);
1647 cur_offset += name_len + sizeof(*extref);
1651 btrfs_release_path(path);
1653 btrfs_release_path(path);
1655 if (ret < 0 && ret != -ENOENT)
1660 static int count_inode_refs(struct btrfs_root *root,
1661 struct btrfs_inode *inode, struct btrfs_path *path)
1664 struct btrfs_key key;
1665 unsigned int nlink = 0;
1667 unsigned long ptr_end;
1669 u64 ino = btrfs_ino(inode);
1672 key.type = BTRFS_INODE_REF_KEY;
1673 key.offset = (u64)-1;
1676 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1680 if (path->slots[0] == 0)
1685 btrfs_item_key_to_cpu(path->nodes[0], &key,
1687 if (key.objectid != ino ||
1688 key.type != BTRFS_INODE_REF_KEY)
1690 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1691 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1693 while (ptr < ptr_end) {
1694 struct btrfs_inode_ref *ref;
1696 ref = (struct btrfs_inode_ref *)ptr;
1697 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1699 ptr = (unsigned long)(ref + 1) + name_len;
1703 if (key.offset == 0)
1705 if (path->slots[0] > 0) {
1710 btrfs_release_path(path);
1712 btrfs_release_path(path);
1718 * There are a few corners where the link count of the file can't
1719 * be properly maintained during replay. So, instead of adding
1720 * lots of complexity to the log code, we just scan the backrefs
1721 * for any file that has been through replay.
1723 * The scan will update the link count on the inode to reflect the
1724 * number of back refs found. If it goes down to zero, the iput
1725 * will free the inode.
1727 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1728 struct btrfs_root *root,
1729 struct inode *inode)
1731 struct btrfs_path *path;
1734 u64 ino = btrfs_ino(BTRFS_I(inode));
1736 path = btrfs_alloc_path();
1740 ret = count_inode_refs(root, BTRFS_I(inode), path);
1746 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1754 if (nlink != inode->i_nlink) {
1755 set_nlink(inode, nlink);
1756 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1760 BTRFS_I(inode)->index_cnt = (u64)-1;
1762 if (inode->i_nlink == 0) {
1763 if (S_ISDIR(inode->i_mode)) {
1764 ret = replay_dir_deletes(trans, root, NULL, path,
1769 ret = btrfs_insert_orphan_item(trans, root, ino);
1775 btrfs_free_path(path);
1779 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1780 struct btrfs_root *root,
1781 struct btrfs_path *path)
1784 struct btrfs_key key;
1785 struct inode *inode;
1787 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1788 key.type = BTRFS_ORPHAN_ITEM_KEY;
1789 key.offset = (u64)-1;
1791 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1797 if (path->slots[0] == 0)
1802 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1803 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1804 key.type != BTRFS_ORPHAN_ITEM_KEY)
1807 ret = btrfs_del_item(trans, root, path);
1811 btrfs_release_path(path);
1812 inode = read_one_inode(root, key.offset);
1818 ret = fixup_inode_link_count(trans, root, inode);
1824 * fixup on a directory may create new entries,
1825 * make sure we always look for the highset possible
1828 key.offset = (u64)-1;
1830 btrfs_release_path(path);
1836 * record a given inode in the fixup dir so we can check its link
1837 * count when replay is done. The link count is incremented here
1838 * so the inode won't go away until we check it
1840 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1841 struct btrfs_root *root,
1842 struct btrfs_path *path,
1845 struct btrfs_key key;
1847 struct inode *inode;
1849 inode = read_one_inode(root, objectid);
1853 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1854 key.type = BTRFS_ORPHAN_ITEM_KEY;
1855 key.offset = objectid;
1857 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1859 btrfs_release_path(path);
1861 if (!inode->i_nlink)
1862 set_nlink(inode, 1);
1865 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1866 } else if (ret == -EEXIST) {
1875 * when replaying the log for a directory, we only insert names
1876 * for inodes that actually exist. This means an fsync on a directory
1877 * does not implicitly fsync all the new files in it
1879 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1880 struct btrfs_root *root,
1881 u64 dirid, u64 index,
1882 char *name, int name_len,
1883 struct btrfs_key *location)
1885 struct inode *inode;
1889 inode = read_one_inode(root, location->objectid);
1893 dir = read_one_inode(root, dirid);
1899 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1900 name_len, 1, index);
1902 /* FIXME, put inode into FIXUP list */
1910 * take a single entry in a log directory item and replay it into
1913 * if a conflicting item exists in the subdirectory already,
1914 * the inode it points to is unlinked and put into the link count
1917 * If a name from the log points to a file or directory that does
1918 * not exist in the FS, it is skipped. fsyncs on directories
1919 * do not force down inodes inside that directory, just changes to the
1920 * names or unlinks in a directory.
1922 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1923 * non-existing inode) and 1 if the name was replayed.
1925 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1926 struct btrfs_root *root,
1927 struct btrfs_path *path,
1928 struct extent_buffer *eb,
1929 struct btrfs_dir_item *di,
1930 struct btrfs_key *key)
1934 struct btrfs_dir_item *dst_di;
1935 struct btrfs_key found_key;
1936 struct btrfs_key log_key;
1941 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1942 bool name_added = false;
1944 dir = read_one_inode(root, key->objectid);
1948 name_len = btrfs_dir_name_len(eb, di);
1949 name = kmalloc(name_len, GFP_NOFS);
1955 log_type = btrfs_dir_type(eb, di);
1956 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1959 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1960 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1965 btrfs_release_path(path);
1967 if (key->type == BTRFS_DIR_ITEM_KEY) {
1968 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1970 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1971 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1980 if (IS_ERR_OR_NULL(dst_di)) {
1981 /* we need a sequence number to insert, so we only
1982 * do inserts for the BTRFS_DIR_INDEX_KEY types
1984 if (key->type != BTRFS_DIR_INDEX_KEY)
1989 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1990 /* the existing item matches the logged item */
1991 if (found_key.objectid == log_key.objectid &&
1992 found_key.type == log_key.type &&
1993 found_key.offset == log_key.offset &&
1994 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1995 update_size = false;
2000 * don't drop the conflicting directory entry if the inode
2001 * for the new entry doesn't exist
2006 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2010 if (key->type == BTRFS_DIR_INDEX_KEY)
2013 btrfs_release_path(path);
2014 if (!ret && update_size) {
2015 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2016 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
2020 if (!ret && name_added)
2026 * Check if the inode reference exists in the log for the given name,
2027 * inode and parent inode
2029 found_key.objectid = log_key.objectid;
2030 found_key.type = BTRFS_INODE_REF_KEY;
2031 found_key.offset = key->objectid;
2032 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2036 /* The dentry will be added later. */
2038 update_size = false;
2042 found_key.objectid = log_key.objectid;
2043 found_key.type = BTRFS_INODE_EXTREF_KEY;
2044 found_key.offset = key->objectid;
2045 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2050 /* The dentry will be added later. */
2052 update_size = false;
2055 btrfs_release_path(path);
2056 ret = insert_one_name(trans, root, key->objectid, key->offset,
2057 name, name_len, &log_key);
2058 if (ret && ret != -ENOENT && ret != -EEXIST)
2062 update_size = false;
2068 * find all the names in a directory item and reconcile them into
2069 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2070 * one name in a directory item, but the same code gets used for
2071 * both directory index types
2073 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2074 struct btrfs_root *root,
2075 struct btrfs_path *path,
2076 struct extent_buffer *eb, int slot,
2077 struct btrfs_key *key)
2080 u32 item_size = btrfs_item_size_nr(eb, slot);
2081 struct btrfs_dir_item *di;
2084 unsigned long ptr_end;
2085 struct btrfs_path *fixup_path = NULL;
2087 ptr = btrfs_item_ptr_offset(eb, slot);
2088 ptr_end = ptr + item_size;
2089 while (ptr < ptr_end) {
2090 di = (struct btrfs_dir_item *)ptr;
2091 name_len = btrfs_dir_name_len(eb, di);
2092 ret = replay_one_name(trans, root, path, eb, di, key);
2095 ptr = (unsigned long)(di + 1);
2099 * If this entry refers to a non-directory (directories can not
2100 * have a link count > 1) and it was added in the transaction
2101 * that was not committed, make sure we fixup the link count of
2102 * the inode it the entry points to. Otherwise something like
2103 * the following would result in a directory pointing to an
2104 * inode with a wrong link that does not account for this dir
2112 * ln testdir/bar testdir/bar_link
2113 * ln testdir/foo testdir/foo_link
2114 * xfs_io -c "fsync" testdir/bar
2118 * mount fs, log replay happens
2120 * File foo would remain with a link count of 1 when it has two
2121 * entries pointing to it in the directory testdir. This would
2122 * make it impossible to ever delete the parent directory has
2123 * it would result in stale dentries that can never be deleted.
2125 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2126 struct btrfs_key di_key;
2129 fixup_path = btrfs_alloc_path();
2136 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2137 ret = link_to_fixup_dir(trans, root, fixup_path,
2144 btrfs_free_path(fixup_path);
2149 * directory replay has two parts. There are the standard directory
2150 * items in the log copied from the subvolume, and range items
2151 * created in the log while the subvolume was logged.
2153 * The range items tell us which parts of the key space the log
2154 * is authoritative for. During replay, if a key in the subvolume
2155 * directory is in a logged range item, but not actually in the log
2156 * that means it was deleted from the directory before the fsync
2157 * and should be removed.
2159 static noinline int find_dir_range(struct btrfs_root *root,
2160 struct btrfs_path *path,
2161 u64 dirid, int key_type,
2162 u64 *start_ret, u64 *end_ret)
2164 struct btrfs_key key;
2166 struct btrfs_dir_log_item *item;
2170 if (*start_ret == (u64)-1)
2173 key.objectid = dirid;
2174 key.type = key_type;
2175 key.offset = *start_ret;
2177 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2181 if (path->slots[0] == 0)
2186 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2188 if (key.type != key_type || key.objectid != dirid) {
2192 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2193 struct btrfs_dir_log_item);
2194 found_end = btrfs_dir_log_end(path->nodes[0], item);
2196 if (*start_ret >= key.offset && *start_ret <= found_end) {
2198 *start_ret = key.offset;
2199 *end_ret = found_end;
2204 /* check the next slot in the tree to see if it is a valid item */
2205 nritems = btrfs_header_nritems(path->nodes[0]);
2207 if (path->slots[0] >= nritems) {
2208 ret = btrfs_next_leaf(root, path);
2213 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2215 if (key.type != key_type || key.objectid != dirid) {
2219 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2220 struct btrfs_dir_log_item);
2221 found_end = btrfs_dir_log_end(path->nodes[0], item);
2222 *start_ret = key.offset;
2223 *end_ret = found_end;
2226 btrfs_release_path(path);
2231 * this looks for a given directory item in the log. If the directory
2232 * item is not in the log, the item is removed and the inode it points
2235 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2236 struct btrfs_root *root,
2237 struct btrfs_root *log,
2238 struct btrfs_path *path,
2239 struct btrfs_path *log_path,
2241 struct btrfs_key *dir_key)
2244 struct extent_buffer *eb;
2247 struct btrfs_dir_item *di;
2248 struct btrfs_dir_item *log_di;
2251 unsigned long ptr_end;
2253 struct inode *inode;
2254 struct btrfs_key location;
2257 eb = path->nodes[0];
2258 slot = path->slots[0];
2259 item_size = btrfs_item_size_nr(eb, slot);
2260 ptr = btrfs_item_ptr_offset(eb, slot);
2261 ptr_end = ptr + item_size;
2262 while (ptr < ptr_end) {
2263 di = (struct btrfs_dir_item *)ptr;
2264 name_len = btrfs_dir_name_len(eb, di);
2265 name = kmalloc(name_len, GFP_NOFS);
2270 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2273 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2274 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2277 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2278 log_di = btrfs_lookup_dir_index_item(trans, log,
2284 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2285 btrfs_dir_item_key_to_cpu(eb, di, &location);
2286 btrfs_release_path(path);
2287 btrfs_release_path(log_path);
2288 inode = read_one_inode(root, location.objectid);
2294 ret = link_to_fixup_dir(trans, root,
2295 path, location.objectid);
2303 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2304 BTRFS_I(inode), name, name_len);
2306 ret = btrfs_run_delayed_items(trans);
2312 /* there might still be more names under this key
2313 * check and repeat if required
2315 ret = btrfs_search_slot(NULL, root, dir_key, path,
2321 } else if (IS_ERR(log_di)) {
2323 return PTR_ERR(log_di);
2325 btrfs_release_path(log_path);
2328 ptr = (unsigned long)(di + 1);
2333 btrfs_release_path(path);
2334 btrfs_release_path(log_path);
2338 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2339 struct btrfs_root *root,
2340 struct btrfs_root *log,
2341 struct btrfs_path *path,
2344 struct btrfs_key search_key;
2345 struct btrfs_path *log_path;
2350 log_path = btrfs_alloc_path();
2354 search_key.objectid = ino;
2355 search_key.type = BTRFS_XATTR_ITEM_KEY;
2356 search_key.offset = 0;
2358 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2362 nritems = btrfs_header_nritems(path->nodes[0]);
2363 for (i = path->slots[0]; i < nritems; i++) {
2364 struct btrfs_key key;
2365 struct btrfs_dir_item *di;
2366 struct btrfs_dir_item *log_di;
2370 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2371 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2376 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2377 total_size = btrfs_item_size_nr(path->nodes[0], i);
2379 while (cur < total_size) {
2380 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2381 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2382 u32 this_len = sizeof(*di) + name_len + data_len;
2385 name = kmalloc(name_len, GFP_NOFS);
2390 read_extent_buffer(path->nodes[0], name,
2391 (unsigned long)(di + 1), name_len);
2393 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2395 btrfs_release_path(log_path);
2397 /* Doesn't exist in log tree, so delete it. */
2398 btrfs_release_path(path);
2399 di = btrfs_lookup_xattr(trans, root, path, ino,
2400 name, name_len, -1);
2407 ret = btrfs_delete_one_dir_name(trans, root,
2411 btrfs_release_path(path);
2416 if (IS_ERR(log_di)) {
2417 ret = PTR_ERR(log_di);
2421 di = (struct btrfs_dir_item *)((char *)di + this_len);
2424 ret = btrfs_next_leaf(root, path);
2430 btrfs_free_path(log_path);
2431 btrfs_release_path(path);
2437 * deletion replay happens before we copy any new directory items
2438 * out of the log or out of backreferences from inodes. It
2439 * scans the log to find ranges of keys that log is authoritative for,
2440 * and then scans the directory to find items in those ranges that are
2441 * not present in the log.
2443 * Anything we don't find in the log is unlinked and removed from the
2446 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2447 struct btrfs_root *root,
2448 struct btrfs_root *log,
2449 struct btrfs_path *path,
2450 u64 dirid, int del_all)
2454 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2456 struct btrfs_key dir_key;
2457 struct btrfs_key found_key;
2458 struct btrfs_path *log_path;
2461 dir_key.objectid = dirid;
2462 dir_key.type = BTRFS_DIR_ITEM_KEY;
2463 log_path = btrfs_alloc_path();
2467 dir = read_one_inode(root, dirid);
2468 /* it isn't an error if the inode isn't there, that can happen
2469 * because we replay the deletes before we copy in the inode item
2473 btrfs_free_path(log_path);
2481 range_end = (u64)-1;
2483 ret = find_dir_range(log, path, dirid, key_type,
2484 &range_start, &range_end);
2489 dir_key.offset = range_start;
2492 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2497 nritems = btrfs_header_nritems(path->nodes[0]);
2498 if (path->slots[0] >= nritems) {
2499 ret = btrfs_next_leaf(root, path);
2505 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2507 if (found_key.objectid != dirid ||
2508 found_key.type != dir_key.type)
2511 if (found_key.offset > range_end)
2514 ret = check_item_in_log(trans, root, log, path,
2519 if (found_key.offset == (u64)-1)
2521 dir_key.offset = found_key.offset + 1;
2523 btrfs_release_path(path);
2524 if (range_end == (u64)-1)
2526 range_start = range_end + 1;
2531 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2532 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2533 dir_key.type = BTRFS_DIR_INDEX_KEY;
2534 btrfs_release_path(path);
2538 btrfs_release_path(path);
2539 btrfs_free_path(log_path);
2545 * the process_func used to replay items from the log tree. This
2546 * gets called in two different stages. The first stage just looks
2547 * for inodes and makes sure they are all copied into the subvolume.
2549 * The second stage copies all the other item types from the log into
2550 * the subvolume. The two stage approach is slower, but gets rid of
2551 * lots of complexity around inodes referencing other inodes that exist
2552 * only in the log (references come from either directory items or inode
2555 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2556 struct walk_control *wc, u64 gen, int level)
2559 struct btrfs_path *path;
2560 struct btrfs_root *root = wc->replay_dest;
2561 struct btrfs_key key;
2565 ret = btrfs_read_buffer(eb, gen, level, NULL);
2569 level = btrfs_header_level(eb);
2574 path = btrfs_alloc_path();
2578 nritems = btrfs_header_nritems(eb);
2579 for (i = 0; i < nritems; i++) {
2580 btrfs_item_key_to_cpu(eb, &key, i);
2582 /* inode keys are done during the first stage */
2583 if (key.type == BTRFS_INODE_ITEM_KEY &&
2584 wc->stage == LOG_WALK_REPLAY_INODES) {
2585 struct btrfs_inode_item *inode_item;
2588 inode_item = btrfs_item_ptr(eb, i,
2589 struct btrfs_inode_item);
2591 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2592 * and never got linked before the fsync, skip it, as
2593 * replaying it is pointless since it would be deleted
2594 * later. We skip logging tmpfiles, but it's always
2595 * possible we are replaying a log created with a kernel
2596 * that used to log tmpfiles.
2598 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2599 wc->ignore_cur_inode = true;
2602 wc->ignore_cur_inode = false;
2604 ret = replay_xattr_deletes(wc->trans, root, log,
2605 path, key.objectid);
2608 mode = btrfs_inode_mode(eb, inode_item);
2609 if (S_ISDIR(mode)) {
2610 ret = replay_dir_deletes(wc->trans,
2611 root, log, path, key.objectid, 0);
2615 ret = overwrite_item(wc->trans, root, path,
2621 * Before replaying extents, truncate the inode to its
2622 * size. We need to do it now and not after log replay
2623 * because before an fsync we can have prealloc extents
2624 * added beyond the inode's i_size. If we did it after,
2625 * through orphan cleanup for example, we would drop
2626 * those prealloc extents just after replaying them.
2628 if (S_ISREG(mode)) {
2629 struct btrfs_drop_extents_args drop_args = { 0 };
2630 struct inode *inode;
2633 inode = read_one_inode(root, key.objectid);
2638 from = ALIGN(i_size_read(inode),
2639 root->fs_info->sectorsize);
2640 drop_args.start = from;
2641 drop_args.end = (u64)-1;
2642 drop_args.drop_cache = true;
2643 ret = btrfs_drop_extents(wc->trans, root,
2647 inode_sub_bytes(inode,
2648 drop_args.bytes_found);
2649 /* Update the inode's nbytes. */
2650 ret = btrfs_update_inode(wc->trans,
2651 root, BTRFS_I(inode));
2658 ret = link_to_fixup_dir(wc->trans, root,
2659 path, key.objectid);
2664 if (wc->ignore_cur_inode)
2667 if (key.type == BTRFS_DIR_INDEX_KEY &&
2668 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2669 ret = replay_one_dir_item(wc->trans, root, path,
2675 if (wc->stage < LOG_WALK_REPLAY_ALL)
2678 /* these keys are simply copied */
2679 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2680 ret = overwrite_item(wc->trans, root, path,
2684 } else if (key.type == BTRFS_INODE_REF_KEY ||
2685 key.type == BTRFS_INODE_EXTREF_KEY) {
2686 ret = add_inode_ref(wc->trans, root, log, path,
2688 if (ret && ret != -ENOENT)
2691 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2692 ret = replay_one_extent(wc->trans, root, path,
2696 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2697 ret = replay_one_dir_item(wc->trans, root, path,
2703 btrfs_free_path(path);
2708 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2710 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2712 struct btrfs_block_group *cache;
2714 cache = btrfs_lookup_block_group(fs_info, start);
2716 btrfs_err(fs_info, "unable to find block group for %llu", start);
2720 spin_lock(&cache->space_info->lock);
2721 spin_lock(&cache->lock);
2722 cache->reserved -= fs_info->nodesize;
2723 cache->space_info->bytes_reserved -= fs_info->nodesize;
2724 spin_unlock(&cache->lock);
2725 spin_unlock(&cache->space_info->lock);
2727 btrfs_put_block_group(cache);
2730 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2731 struct btrfs_root *root,
2732 struct btrfs_path *path, int *level,
2733 struct walk_control *wc)
2735 struct btrfs_fs_info *fs_info = root->fs_info;
2738 struct extent_buffer *next;
2739 struct extent_buffer *cur;
2743 while (*level > 0) {
2744 struct btrfs_key first_key;
2746 cur = path->nodes[*level];
2748 WARN_ON(btrfs_header_level(cur) != *level);
2750 if (path->slots[*level] >=
2751 btrfs_header_nritems(cur))
2754 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2755 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2756 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2757 blocksize = fs_info->nodesize;
2759 next = btrfs_find_create_tree_block(fs_info, bytenr,
2760 btrfs_header_owner(cur),
2763 return PTR_ERR(next);
2766 ret = wc->process_func(root, next, wc, ptr_gen,
2769 free_extent_buffer(next);
2773 path->slots[*level]++;
2775 ret = btrfs_read_buffer(next, ptr_gen,
2776 *level - 1, &first_key);
2778 free_extent_buffer(next);
2783 btrfs_tree_lock(next);
2784 btrfs_clean_tree_block(next);
2785 btrfs_wait_tree_block_writeback(next);
2786 btrfs_tree_unlock(next);
2787 ret = btrfs_pin_reserved_extent(trans,
2790 free_extent_buffer(next);
2793 btrfs_redirty_list_add(
2794 trans->transaction, next);
2796 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2797 clear_extent_buffer_dirty(next);
2798 unaccount_log_buffer(fs_info, bytenr);
2801 free_extent_buffer(next);
2804 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2806 free_extent_buffer(next);
2810 if (path->nodes[*level-1])
2811 free_extent_buffer(path->nodes[*level-1]);
2812 path->nodes[*level-1] = next;
2813 *level = btrfs_header_level(next);
2814 path->slots[*level] = 0;
2817 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2823 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2824 struct btrfs_root *root,
2825 struct btrfs_path *path, int *level,
2826 struct walk_control *wc)
2828 struct btrfs_fs_info *fs_info = root->fs_info;
2833 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2834 slot = path->slots[i];
2835 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2838 WARN_ON(*level == 0);
2841 ret = wc->process_func(root, path->nodes[*level], wc,
2842 btrfs_header_generation(path->nodes[*level]),
2848 struct extent_buffer *next;
2850 next = path->nodes[*level];
2853 btrfs_tree_lock(next);
2854 btrfs_clean_tree_block(next);
2855 btrfs_wait_tree_block_writeback(next);
2856 btrfs_tree_unlock(next);
2857 ret = btrfs_pin_reserved_extent(trans,
2858 path->nodes[*level]->start,
2859 path->nodes[*level]->len);
2863 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2864 clear_extent_buffer_dirty(next);
2866 unaccount_log_buffer(fs_info,
2867 path->nodes[*level]->start);
2870 free_extent_buffer(path->nodes[*level]);
2871 path->nodes[*level] = NULL;
2879 * drop the reference count on the tree rooted at 'snap'. This traverses
2880 * the tree freeing any blocks that have a ref count of zero after being
2883 static int walk_log_tree(struct btrfs_trans_handle *trans,
2884 struct btrfs_root *log, struct walk_control *wc)
2886 struct btrfs_fs_info *fs_info = log->fs_info;
2890 struct btrfs_path *path;
2893 path = btrfs_alloc_path();
2897 level = btrfs_header_level(log->node);
2899 path->nodes[level] = log->node;
2900 atomic_inc(&log->node->refs);
2901 path->slots[level] = 0;
2904 wret = walk_down_log_tree(trans, log, path, &level, wc);
2912 wret = walk_up_log_tree(trans, log, path, &level, wc);
2921 /* was the root node processed? if not, catch it here */
2922 if (path->nodes[orig_level]) {
2923 ret = wc->process_func(log, path->nodes[orig_level], wc,
2924 btrfs_header_generation(path->nodes[orig_level]),
2929 struct extent_buffer *next;
2931 next = path->nodes[orig_level];
2934 btrfs_tree_lock(next);
2935 btrfs_clean_tree_block(next);
2936 btrfs_wait_tree_block_writeback(next);
2937 btrfs_tree_unlock(next);
2938 ret = btrfs_pin_reserved_extent(trans,
2939 next->start, next->len);
2943 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2944 clear_extent_buffer_dirty(next);
2945 unaccount_log_buffer(fs_info, next->start);
2951 btrfs_free_path(path);
2956 * helper function to update the item for a given subvolumes log root
2957 * in the tree of log roots
2959 static int update_log_root(struct btrfs_trans_handle *trans,
2960 struct btrfs_root *log,
2961 struct btrfs_root_item *root_item)
2963 struct btrfs_fs_info *fs_info = log->fs_info;
2966 if (log->log_transid == 1) {
2967 /* insert root item on the first sync */
2968 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2969 &log->root_key, root_item);
2971 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2972 &log->root_key, root_item);
2977 static void wait_log_commit(struct btrfs_root *root, int transid)
2980 int index = transid % 2;
2983 * we only allow two pending log transactions at a time,
2984 * so we know that if ours is more than 2 older than the
2985 * current transaction, we're done
2988 prepare_to_wait(&root->log_commit_wait[index],
2989 &wait, TASK_UNINTERRUPTIBLE);
2991 if (!(root->log_transid_committed < transid &&
2992 atomic_read(&root->log_commit[index])))
2995 mutex_unlock(&root->log_mutex);
2997 mutex_lock(&root->log_mutex);
2999 finish_wait(&root->log_commit_wait[index], &wait);
3002 static void wait_for_writer(struct btrfs_root *root)
3007 prepare_to_wait(&root->log_writer_wait, &wait,
3008 TASK_UNINTERRUPTIBLE);
3009 if (!atomic_read(&root->log_writers))
3012 mutex_unlock(&root->log_mutex);
3014 mutex_lock(&root->log_mutex);
3016 finish_wait(&root->log_writer_wait, &wait);
3019 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3020 struct btrfs_log_ctx *ctx)
3025 mutex_lock(&root->log_mutex);
3026 list_del_init(&ctx->list);
3027 mutex_unlock(&root->log_mutex);
3031 * Invoked in log mutex context, or be sure there is no other task which
3032 * can access the list.
3034 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3035 int index, int error)
3037 struct btrfs_log_ctx *ctx;
3038 struct btrfs_log_ctx *safe;
3040 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3041 list_del_init(&ctx->list);
3042 ctx->log_ret = error;
3047 * btrfs_sync_log does sends a given tree log down to the disk and
3048 * updates the super blocks to record it. When this call is done,
3049 * you know that any inodes previously logged are safely on disk only
3052 * Any other return value means you need to call btrfs_commit_transaction.
3053 * Some of the edge cases for fsyncing directories that have had unlinks
3054 * or renames done in the past mean that sometimes the only safe
3055 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3056 * that has happened.
3058 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3059 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3065 struct btrfs_fs_info *fs_info = root->fs_info;
3066 struct btrfs_root *log = root->log_root;
3067 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3068 struct btrfs_root_item new_root_item;
3069 int log_transid = 0;
3070 struct btrfs_log_ctx root_log_ctx;
3071 struct blk_plug plug;
3075 mutex_lock(&root->log_mutex);
3076 log_transid = ctx->log_transid;
3077 if (root->log_transid_committed >= log_transid) {
3078 mutex_unlock(&root->log_mutex);
3079 return ctx->log_ret;
3082 index1 = log_transid % 2;
3083 if (atomic_read(&root->log_commit[index1])) {
3084 wait_log_commit(root, log_transid);
3085 mutex_unlock(&root->log_mutex);
3086 return ctx->log_ret;
3088 ASSERT(log_transid == root->log_transid);
3089 atomic_set(&root->log_commit[index1], 1);
3091 /* wait for previous tree log sync to complete */
3092 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3093 wait_log_commit(root, log_transid - 1);
3096 int batch = atomic_read(&root->log_batch);
3097 /* when we're on an ssd, just kick the log commit out */
3098 if (!btrfs_test_opt(fs_info, SSD) &&
3099 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3100 mutex_unlock(&root->log_mutex);
3101 schedule_timeout_uninterruptible(1);
3102 mutex_lock(&root->log_mutex);
3104 wait_for_writer(root);
3105 if (batch == atomic_read(&root->log_batch))
3109 /* bail out if we need to do a full commit */
3110 if (btrfs_need_log_full_commit(trans)) {
3112 mutex_unlock(&root->log_mutex);
3116 if (log_transid % 2 == 0)
3117 mark = EXTENT_DIRTY;
3121 /* we start IO on all the marked extents here, but we don't actually
3122 * wait for them until later.
3124 blk_start_plug(&plug);
3125 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3127 * -EAGAIN happens when someone, e.g., a concurrent transaction
3128 * commit, writes a dirty extent in this tree-log commit. This
3129 * concurrent write will create a hole writing out the extents,
3130 * and we cannot proceed on a zoned filesystem, requiring
3131 * sequential writing. While we can bail out to a full commit
3132 * here, but we can continue hoping the concurrent writing fills
3135 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3138 blk_finish_plug(&plug);
3139 btrfs_abort_transaction(trans, ret);
3140 btrfs_set_log_full_commit(trans);
3141 mutex_unlock(&root->log_mutex);
3146 * We _must_ update under the root->log_mutex in order to make sure we
3147 * have a consistent view of the log root we are trying to commit at
3150 * We _must_ copy this into a local copy, because we are not holding the
3151 * log_root_tree->log_mutex yet. This is important because when we
3152 * commit the log_root_tree we must have a consistent view of the
3153 * log_root_tree when we update the super block to point at the
3154 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3155 * with the commit and possibly point at the new block which we may not
3158 btrfs_set_root_node(&log->root_item, log->node);
3159 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3161 root->log_transid++;
3162 log->log_transid = root->log_transid;
3163 root->log_start_pid = 0;
3165 * IO has been started, blocks of the log tree have WRITTEN flag set
3166 * in their headers. new modifications of the log will be written to
3167 * new positions. so it's safe to allow log writers to go in.
3169 mutex_unlock(&root->log_mutex);
3171 if (btrfs_is_zoned(fs_info)) {
3172 mutex_lock(&fs_info->tree_root->log_mutex);
3173 if (!log_root_tree->node) {
3174 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3176 mutex_unlock(&fs_info->tree_root->log_mutex);
3180 mutex_unlock(&fs_info->tree_root->log_mutex);
3183 btrfs_init_log_ctx(&root_log_ctx, NULL);
3185 mutex_lock(&log_root_tree->log_mutex);
3187 index2 = log_root_tree->log_transid % 2;
3188 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3189 root_log_ctx.log_transid = log_root_tree->log_transid;
3192 * Now we are safe to update the log_root_tree because we're under the
3193 * log_mutex, and we're a current writer so we're holding the commit
3194 * open until we drop the log_mutex.
3196 ret = update_log_root(trans, log, &new_root_item);
3198 if (!list_empty(&root_log_ctx.list))
3199 list_del_init(&root_log_ctx.list);
3201 blk_finish_plug(&plug);
3202 btrfs_set_log_full_commit(trans);
3204 if (ret != -ENOSPC) {
3205 btrfs_abort_transaction(trans, ret);
3206 mutex_unlock(&log_root_tree->log_mutex);
3209 btrfs_wait_tree_log_extents(log, mark);
3210 mutex_unlock(&log_root_tree->log_mutex);
3215 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3216 blk_finish_plug(&plug);
3217 list_del_init(&root_log_ctx.list);
3218 mutex_unlock(&log_root_tree->log_mutex);
3219 ret = root_log_ctx.log_ret;
3223 index2 = root_log_ctx.log_transid % 2;
3224 if (atomic_read(&log_root_tree->log_commit[index2])) {
3225 blk_finish_plug(&plug);
3226 ret = btrfs_wait_tree_log_extents(log, mark);
3227 wait_log_commit(log_root_tree,
3228 root_log_ctx.log_transid);
3229 mutex_unlock(&log_root_tree->log_mutex);
3231 ret = root_log_ctx.log_ret;
3234 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3235 atomic_set(&log_root_tree->log_commit[index2], 1);
3237 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3238 wait_log_commit(log_root_tree,
3239 root_log_ctx.log_transid - 1);
3243 * now that we've moved on to the tree of log tree roots,
3244 * check the full commit flag again
3246 if (btrfs_need_log_full_commit(trans)) {
3247 blk_finish_plug(&plug);
3248 btrfs_wait_tree_log_extents(log, mark);
3249 mutex_unlock(&log_root_tree->log_mutex);
3251 goto out_wake_log_root;
3254 ret = btrfs_write_marked_extents(fs_info,
3255 &log_root_tree->dirty_log_pages,
3256 EXTENT_DIRTY | EXTENT_NEW);
3257 blk_finish_plug(&plug);
3259 * As described above, -EAGAIN indicates a hole in the extents. We
3260 * cannot wait for these write outs since the waiting cause a
3261 * deadlock. Bail out to the full commit instead.
3263 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3264 btrfs_set_log_full_commit(trans);
3265 btrfs_wait_tree_log_extents(log, mark);
3266 mutex_unlock(&log_root_tree->log_mutex);
3267 goto out_wake_log_root;
3269 btrfs_set_log_full_commit(trans);
3270 btrfs_abort_transaction(trans, ret);
3271 mutex_unlock(&log_root_tree->log_mutex);
3272 goto out_wake_log_root;
3274 ret = btrfs_wait_tree_log_extents(log, mark);
3276 ret = btrfs_wait_tree_log_extents(log_root_tree,
3277 EXTENT_NEW | EXTENT_DIRTY);
3279 btrfs_set_log_full_commit(trans);
3280 mutex_unlock(&log_root_tree->log_mutex);
3281 goto out_wake_log_root;
3284 log_root_start = log_root_tree->node->start;
3285 log_root_level = btrfs_header_level(log_root_tree->node);
3286 log_root_tree->log_transid++;
3287 mutex_unlock(&log_root_tree->log_mutex);
3290 * Here we are guaranteed that nobody is going to write the superblock
3291 * for the current transaction before us and that neither we do write
3292 * our superblock before the previous transaction finishes its commit
3293 * and writes its superblock, because:
3295 * 1) We are holding a handle on the current transaction, so no body
3296 * can commit it until we release the handle;
3298 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3299 * if the previous transaction is still committing, and hasn't yet
3300 * written its superblock, we wait for it to do it, because a
3301 * transaction commit acquires the tree_log_mutex when the commit
3302 * begins and releases it only after writing its superblock.
3304 mutex_lock(&fs_info->tree_log_mutex);
3307 * The previous transaction writeout phase could have failed, and thus
3308 * marked the fs in an error state. We must not commit here, as we
3309 * could have updated our generation in the super_for_commit and
3310 * writing the super here would result in transid mismatches. If there
3311 * is an error here just bail.
3313 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3315 btrfs_set_log_full_commit(trans);
3316 btrfs_abort_transaction(trans, ret);
3317 mutex_unlock(&fs_info->tree_log_mutex);
3318 goto out_wake_log_root;
3321 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3322 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3323 ret = write_all_supers(fs_info, 1);
3324 mutex_unlock(&fs_info->tree_log_mutex);
3326 btrfs_set_log_full_commit(trans);
3327 btrfs_abort_transaction(trans, ret);
3328 goto out_wake_log_root;
3332 * We know there can only be one task here, since we have not yet set
3333 * root->log_commit[index1] to 0 and any task attempting to sync the
3334 * log must wait for the previous log transaction to commit if it's
3335 * still in progress or wait for the current log transaction commit if
3336 * someone else already started it. We use <= and not < because the
3337 * first log transaction has an ID of 0.
3339 ASSERT(root->last_log_commit <= log_transid);
3340 root->last_log_commit = log_transid;
3343 mutex_lock(&log_root_tree->log_mutex);
3344 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3346 log_root_tree->log_transid_committed++;
3347 atomic_set(&log_root_tree->log_commit[index2], 0);
3348 mutex_unlock(&log_root_tree->log_mutex);
3351 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3352 * all the updates above are seen by the woken threads. It might not be
3353 * necessary, but proving that seems to be hard.
3355 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3357 mutex_lock(&root->log_mutex);
3358 btrfs_remove_all_log_ctxs(root, index1, ret);
3359 root->log_transid_committed++;
3360 atomic_set(&root->log_commit[index1], 0);
3361 mutex_unlock(&root->log_mutex);
3364 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3365 * all the updates above are seen by the woken threads. It might not be
3366 * necessary, but proving that seems to be hard.
3368 cond_wake_up(&root->log_commit_wait[index1]);
3372 static void free_log_tree(struct btrfs_trans_handle *trans,
3373 struct btrfs_root *log)
3376 struct walk_control wc = {
3378 .process_func = process_one_buffer
3382 ret = walk_log_tree(trans, log, &wc);
3385 btrfs_abort_transaction(trans, ret);
3387 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3391 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3392 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3393 extent_io_tree_release(&log->log_csum_range);
3395 if (trans && log->node)
3396 btrfs_redirty_list_add(trans->transaction, log->node);
3397 btrfs_put_root(log);
3401 * free all the extents used by the tree log. This should be called
3402 * at commit time of the full transaction
3404 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3406 if (root->log_root) {
3407 free_log_tree(trans, root->log_root);
3408 root->log_root = NULL;
3409 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3414 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3415 struct btrfs_fs_info *fs_info)
3417 if (fs_info->log_root_tree) {
3418 free_log_tree(trans, fs_info->log_root_tree);
3419 fs_info->log_root_tree = NULL;
3420 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3426 * Check if an inode was logged in the current transaction. This may often
3427 * return some false positives, because logged_trans is an in memory only field,
3428 * not persisted anywhere. This is meant to be used in contexts where a false
3429 * positive has no functional consequences.
3431 static bool inode_logged(struct btrfs_trans_handle *trans,
3432 struct btrfs_inode *inode)
3434 if (inode->logged_trans == trans->transid)
3438 * The inode's logged_trans is always 0 when we load it (because it is
3439 * not persisted in the inode item or elsewhere). So if it is 0, the
3440 * inode was last modified in the current transaction then the inode may
3441 * have been logged before in the current transaction, then evicted and
3442 * loaded again in the current transaction - or may have never been logged
3443 * in the current transaction, but since we can not be sure, we have to
3444 * assume it was, otherwise our callers can leave an inconsistent log.
3446 if (inode->logged_trans == 0 &&
3447 inode->last_trans == trans->transid &&
3448 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3455 * If both a file and directory are logged, and unlinks or renames are
3456 * mixed in, we have a few interesting corners:
3458 * create file X in dir Y
3459 * link file X to X.link in dir Y
3461 * unlink file X but leave X.link
3464 * After a crash we would expect only X.link to exist. But file X
3465 * didn't get fsync'd again so the log has back refs for X and X.link.
3467 * We solve this by removing directory entries and inode backrefs from the
3468 * log when a file that was logged in the current transaction is
3469 * unlinked. Any later fsync will include the updated log entries, and
3470 * we'll be able to reconstruct the proper directory items from backrefs.
3472 * This optimizations allows us to avoid relogging the entire inode
3473 * or the entire directory.
3475 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3476 struct btrfs_root *root,
3477 const char *name, int name_len,
3478 struct btrfs_inode *dir, u64 index)
3480 struct btrfs_root *log;
3481 struct btrfs_dir_item *di;
3482 struct btrfs_path *path;
3485 u64 dir_ino = btrfs_ino(dir);
3487 if (!inode_logged(trans, dir))
3490 ret = join_running_log_trans(root);
3494 mutex_lock(&dir->log_mutex);
3496 log = root->log_root;
3497 path = btrfs_alloc_path();
3503 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3504 name, name_len, -1);
3510 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3516 btrfs_release_path(path);
3517 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3518 index, name, name_len, -1);
3524 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3532 * We do not need to update the size field of the directory's inode item
3533 * because on log replay we update the field to reflect all existing
3534 * entries in the directory (see overwrite_item()).
3537 btrfs_free_path(path);
3539 mutex_unlock(&dir->log_mutex);
3540 if (err == -ENOSPC) {
3541 btrfs_set_log_full_commit(trans);
3543 } else if (err < 0 && err != -ENOENT) {
3544 /* ENOENT can be returned if the entry hasn't been fsynced yet */
3545 btrfs_abort_transaction(trans, err);
3548 btrfs_end_log_trans(root);
3553 /* see comments for btrfs_del_dir_entries_in_log */
3554 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3555 struct btrfs_root *root,
3556 const char *name, int name_len,
3557 struct btrfs_inode *inode, u64 dirid)
3559 struct btrfs_root *log;
3563 if (!inode_logged(trans, inode))
3566 ret = join_running_log_trans(root);
3569 log = root->log_root;
3570 mutex_lock(&inode->log_mutex);
3572 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3574 mutex_unlock(&inode->log_mutex);
3575 if (ret == -ENOSPC) {
3576 btrfs_set_log_full_commit(trans);
3578 } else if (ret < 0 && ret != -ENOENT)
3579 btrfs_abort_transaction(trans, ret);
3580 btrfs_end_log_trans(root);
3586 * creates a range item in the log for 'dirid'. first_offset and
3587 * last_offset tell us which parts of the key space the log should
3588 * be considered authoritative for.
3590 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3591 struct btrfs_root *log,
3592 struct btrfs_path *path,
3593 int key_type, u64 dirid,
3594 u64 first_offset, u64 last_offset)
3597 struct btrfs_key key;
3598 struct btrfs_dir_log_item *item;
3600 key.objectid = dirid;
3601 key.offset = first_offset;
3602 if (key_type == BTRFS_DIR_ITEM_KEY)
3603 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3605 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3606 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3610 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3611 struct btrfs_dir_log_item);
3612 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3613 btrfs_mark_buffer_dirty(path->nodes[0]);
3614 btrfs_release_path(path);
3619 * log all the items included in the current transaction for a given
3620 * directory. This also creates the range items in the log tree required
3621 * to replay anything deleted before the fsync
3623 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3624 struct btrfs_root *root, struct btrfs_inode *inode,
3625 struct btrfs_path *path,
3626 struct btrfs_path *dst_path, int key_type,
3627 struct btrfs_log_ctx *ctx,
3628 u64 min_offset, u64 *last_offset_ret)
3630 struct btrfs_key min_key;
3631 struct btrfs_root *log = root->log_root;
3632 struct extent_buffer *src;
3637 u64 first_offset = min_offset;
3638 u64 last_offset = (u64)-1;
3639 u64 ino = btrfs_ino(inode);
3641 log = root->log_root;
3643 min_key.objectid = ino;
3644 min_key.type = key_type;
3645 min_key.offset = min_offset;
3647 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3650 * we didn't find anything from this transaction, see if there
3651 * is anything at all
3653 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3654 min_key.objectid = ino;
3655 min_key.type = key_type;
3656 min_key.offset = (u64)-1;
3657 btrfs_release_path(path);
3658 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3660 btrfs_release_path(path);
3663 ret = btrfs_previous_item(root, path, ino, key_type);
3665 /* if ret == 0 there are items for this type,
3666 * create a range to tell us the last key of this type.
3667 * otherwise, there are no items in this directory after
3668 * *min_offset, and we create a range to indicate that.
3671 struct btrfs_key tmp;
3672 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3674 if (key_type == tmp.type)
3675 first_offset = max(min_offset, tmp.offset) + 1;
3680 /* go backward to find any previous key */
3681 ret = btrfs_previous_item(root, path, ino, key_type);
3683 struct btrfs_key tmp;
3684 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3685 if (key_type == tmp.type) {
3686 first_offset = tmp.offset;
3687 ret = overwrite_item(trans, log, dst_path,
3688 path->nodes[0], path->slots[0],
3696 btrfs_release_path(path);
3699 * Find the first key from this transaction again. See the note for
3700 * log_new_dir_dentries, if we're logging a directory recursively we
3701 * won't be holding its i_mutex, which means we can modify the directory
3702 * while we're logging it. If we remove an entry between our first
3703 * search and this search we'll not find the key again and can just
3707 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3712 * we have a block from this transaction, log every item in it
3713 * from our directory
3716 struct btrfs_key tmp;
3717 src = path->nodes[0];
3718 nritems = btrfs_header_nritems(src);
3719 for (i = path->slots[0]; i < nritems; i++) {
3720 struct btrfs_dir_item *di;
3722 btrfs_item_key_to_cpu(src, &min_key, i);
3724 if (min_key.objectid != ino || min_key.type != key_type)
3727 if (need_resched()) {
3728 btrfs_release_path(path);
3733 ret = overwrite_item(trans, log, dst_path, src, i,
3741 * We must make sure that when we log a directory entry,
3742 * the corresponding inode, after log replay, has a
3743 * matching link count. For example:
3749 * xfs_io -c "fsync" mydir
3751 * <mount fs and log replay>
3753 * Would result in a fsync log that when replayed, our
3754 * file inode would have a link count of 1, but we get
3755 * two directory entries pointing to the same inode.
3756 * After removing one of the names, it would not be
3757 * possible to remove the other name, which resulted
3758 * always in stale file handle errors, and would not
3759 * be possible to rmdir the parent directory, since
3760 * its i_size could never decrement to the value
3761 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3763 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3764 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3766 (btrfs_dir_transid(src, di) == trans->transid ||
3767 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3768 tmp.type != BTRFS_ROOT_ITEM_KEY)
3769 ctx->log_new_dentries = true;
3771 path->slots[0] = nritems;
3774 * look ahead to the next item and see if it is also
3775 * from this directory and from this transaction
3777 ret = btrfs_next_leaf(root, path);
3780 last_offset = (u64)-1;
3785 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3786 if (tmp.objectid != ino || tmp.type != key_type) {
3787 last_offset = (u64)-1;
3790 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3791 ret = overwrite_item(trans, log, dst_path,
3792 path->nodes[0], path->slots[0],
3797 last_offset = tmp.offset;
3802 btrfs_release_path(path);
3803 btrfs_release_path(dst_path);
3806 *last_offset_ret = last_offset;
3808 * insert the log range keys to indicate where the log
3811 ret = insert_dir_log_key(trans, log, path, key_type,
3812 ino, first_offset, last_offset);
3820 * logging directories is very similar to logging inodes, We find all the items
3821 * from the current transaction and write them to the log.
3823 * The recovery code scans the directory in the subvolume, and if it finds a
3824 * key in the range logged that is not present in the log tree, then it means
3825 * that dir entry was unlinked during the transaction.
3827 * In order for that scan to work, we must include one key smaller than
3828 * the smallest logged by this transaction and one key larger than the largest
3829 * key logged by this transaction.
3831 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3832 struct btrfs_root *root, struct btrfs_inode *inode,
3833 struct btrfs_path *path,
3834 struct btrfs_path *dst_path,
3835 struct btrfs_log_ctx *ctx)
3840 int key_type = BTRFS_DIR_ITEM_KEY;
3846 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3847 ctx, min_key, &max_key);
3850 if (max_key == (u64)-1)
3852 min_key = max_key + 1;
3855 if (key_type == BTRFS_DIR_ITEM_KEY) {
3856 key_type = BTRFS_DIR_INDEX_KEY;
3863 * a helper function to drop items from the log before we relog an
3864 * inode. max_key_type indicates the highest item type to remove.
3865 * This cannot be run for file data extents because it does not
3866 * free the extents they point to.
3868 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3869 struct btrfs_root *log,
3870 struct btrfs_path *path,
3871 u64 objectid, int max_key_type)
3874 struct btrfs_key key;
3875 struct btrfs_key found_key;
3878 key.objectid = objectid;
3879 key.type = max_key_type;
3880 key.offset = (u64)-1;
3883 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3884 BUG_ON(ret == 0); /* Logic error */
3888 if (path->slots[0] == 0)
3892 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3895 if (found_key.objectid != objectid)
3898 found_key.offset = 0;
3900 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3904 ret = btrfs_del_items(trans, log, path, start_slot,
3905 path->slots[0] - start_slot + 1);
3907 * If start slot isn't 0 then we don't need to re-search, we've
3908 * found the last guy with the objectid in this tree.
3910 if (ret || start_slot != 0)
3912 btrfs_release_path(path);
3914 btrfs_release_path(path);
3920 static void fill_inode_item(struct btrfs_trans_handle *trans,
3921 struct extent_buffer *leaf,
3922 struct btrfs_inode_item *item,
3923 struct inode *inode, int log_inode_only,
3926 struct btrfs_map_token token;
3929 btrfs_init_map_token(&token, leaf);
3931 if (log_inode_only) {
3932 /* set the generation to zero so the recover code
3933 * can tell the difference between an logging
3934 * just to say 'this inode exists' and a logging
3935 * to say 'update this inode with these values'
3937 btrfs_set_token_inode_generation(&token, item, 0);
3938 btrfs_set_token_inode_size(&token, item, logged_isize);
3940 btrfs_set_token_inode_generation(&token, item,
3941 BTRFS_I(inode)->generation);
3942 btrfs_set_token_inode_size(&token, item, inode->i_size);
3945 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3946 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3947 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3948 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3950 btrfs_set_token_timespec_sec(&token, &item->atime,
3951 inode->i_atime.tv_sec);
3952 btrfs_set_token_timespec_nsec(&token, &item->atime,
3953 inode->i_atime.tv_nsec);
3955 btrfs_set_token_timespec_sec(&token, &item->mtime,
3956 inode->i_mtime.tv_sec);
3957 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3958 inode->i_mtime.tv_nsec);
3960 btrfs_set_token_timespec_sec(&token, &item->ctime,
3961 inode->i_ctime.tv_sec);
3962 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3963 inode->i_ctime.tv_nsec);
3966 * We do not need to set the nbytes field, in fact during a fast fsync
3967 * its value may not even be correct, since a fast fsync does not wait
3968 * for ordered extent completion, which is where we update nbytes, it
3969 * only waits for writeback to complete. During log replay as we find
3970 * file extent items and replay them, we adjust the nbytes field of the
3971 * inode item in subvolume tree as needed (see overwrite_item()).
3974 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3975 btrfs_set_token_inode_transid(&token, item, trans->transid);
3976 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3977 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3978 BTRFS_I(inode)->ro_flags);
3979 btrfs_set_token_inode_flags(&token, item, flags);
3980 btrfs_set_token_inode_block_group(&token, item, 0);
3983 static int log_inode_item(struct btrfs_trans_handle *trans,
3984 struct btrfs_root *log, struct btrfs_path *path,
3985 struct btrfs_inode *inode, bool inode_item_dropped)
3987 struct btrfs_inode_item *inode_item;
3991 * If we are doing a fast fsync and the inode was logged before in the
3992 * current transaction, then we know the inode was previously logged and
3993 * it exists in the log tree. For performance reasons, in this case use
3994 * btrfs_search_slot() directly with ins_len set to 0 so that we never
3995 * attempt a write lock on the leaf's parent, which adds unnecessary lock
3996 * contention in case there are concurrent fsyncs for other inodes of the
3997 * same subvolume. Using btrfs_insert_empty_item() when the inode item
3998 * already exists can also result in unnecessarily splitting a leaf.
4000 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4001 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4007 * This means it is the first fsync in the current transaction,
4008 * so the inode item is not in the log and we need to insert it.
4009 * We can never get -EEXIST because we are only called for a fast
4010 * fsync and in case an inode eviction happens after the inode was
4011 * logged before in the current transaction, when we load again
4012 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4013 * flags and set ->logged_trans to 0.
4015 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4016 sizeof(*inode_item));
4017 ASSERT(ret != -EEXIST);
4021 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4022 struct btrfs_inode_item);
4023 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4025 btrfs_release_path(path);
4029 static int log_csums(struct btrfs_trans_handle *trans,
4030 struct btrfs_inode *inode,
4031 struct btrfs_root *log_root,
4032 struct btrfs_ordered_sum *sums)
4034 const u64 lock_end = sums->bytenr + sums->len - 1;
4035 struct extent_state *cached_state = NULL;
4039 * If this inode was not used for reflink operations in the current
4040 * transaction with new extents, then do the fast path, no need to
4041 * worry about logging checksum items with overlapping ranges.
4043 if (inode->last_reflink_trans < trans->transid)
4044 return btrfs_csum_file_blocks(trans, log_root, sums);
4047 * Serialize logging for checksums. This is to avoid racing with the
4048 * same checksum being logged by another task that is logging another
4049 * file which happens to refer to the same extent as well. Such races
4050 * can leave checksum items in the log with overlapping ranges.
4052 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
4053 lock_end, &cached_state);
4057 * Due to extent cloning, we might have logged a csum item that covers a
4058 * subrange of a cloned extent, and later we can end up logging a csum
4059 * item for a larger subrange of the same extent or the entire range.
4060 * This would leave csum items in the log tree that cover the same range
4061 * and break the searches for checksums in the log tree, resulting in
4062 * some checksums missing in the fs/subvolume tree. So just delete (or
4063 * trim and adjust) any existing csum items in the log for this range.
4065 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4067 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4069 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
4075 static noinline int copy_items(struct btrfs_trans_handle *trans,
4076 struct btrfs_inode *inode,
4077 struct btrfs_path *dst_path,
4078 struct btrfs_path *src_path,
4079 int start_slot, int nr, int inode_only,
4082 struct btrfs_fs_info *fs_info = trans->fs_info;
4083 unsigned long src_offset;
4084 unsigned long dst_offset;
4085 struct btrfs_root *log = inode->root->log_root;
4086 struct btrfs_file_extent_item *extent;
4087 struct btrfs_inode_item *inode_item;
4088 struct extent_buffer *src = src_path->nodes[0];
4090 struct btrfs_key *ins_keys;
4094 struct list_head ordered_sums;
4095 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4097 INIT_LIST_HEAD(&ordered_sums);
4099 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4100 nr * sizeof(u32), GFP_NOFS);
4104 ins_sizes = (u32 *)ins_data;
4105 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4107 for (i = 0; i < nr; i++) {
4108 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4109 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4111 ret = btrfs_insert_empty_items(trans, log, dst_path,
4112 ins_keys, ins_sizes, nr);
4118 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4119 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4120 dst_path->slots[0]);
4122 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4124 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4125 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4127 struct btrfs_inode_item);
4128 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4130 inode_only == LOG_INODE_EXISTS,
4133 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4134 src_offset, ins_sizes[i]);
4137 /* take a reference on file data extents so that truncates
4138 * or deletes of this inode don't have to relog the inode
4141 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4144 extent = btrfs_item_ptr(src, start_slot + i,
4145 struct btrfs_file_extent_item);
4147 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4150 found_type = btrfs_file_extent_type(src, extent);
4151 if (found_type == BTRFS_FILE_EXTENT_REG) {
4153 ds = btrfs_file_extent_disk_bytenr(src,
4155 /* ds == 0 is a hole */
4159 dl = btrfs_file_extent_disk_num_bytes(src,
4161 cs = btrfs_file_extent_offset(src, extent);
4162 cl = btrfs_file_extent_num_bytes(src,
4164 if (btrfs_file_extent_compression(src,
4170 ret = btrfs_lookup_csums_range(
4172 ds + cs, ds + cs + cl - 1,
4180 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4181 btrfs_release_path(dst_path);
4185 * we have to do this after the loop above to avoid changing the
4186 * log tree while trying to change the log tree.
4188 while (!list_empty(&ordered_sums)) {
4189 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4190 struct btrfs_ordered_sum,
4193 ret = log_csums(trans, inode, log, sums);
4194 list_del(&sums->list);
4201 static int extent_cmp(void *priv, const struct list_head *a,
4202 const struct list_head *b)
4204 const struct extent_map *em1, *em2;
4206 em1 = list_entry(a, struct extent_map, list);
4207 em2 = list_entry(b, struct extent_map, list);
4209 if (em1->start < em2->start)
4211 else if (em1->start > em2->start)
4216 static int log_extent_csums(struct btrfs_trans_handle *trans,
4217 struct btrfs_inode *inode,
4218 struct btrfs_root *log_root,
4219 const struct extent_map *em,
4220 struct btrfs_log_ctx *ctx)
4222 struct btrfs_ordered_extent *ordered;
4225 u64 mod_start = em->mod_start;
4226 u64 mod_len = em->mod_len;
4227 LIST_HEAD(ordered_sums);
4230 if (inode->flags & BTRFS_INODE_NODATASUM ||
4231 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4232 em->block_start == EXTENT_MAP_HOLE)
4235 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4236 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4237 const u64 mod_end = mod_start + mod_len;
4238 struct btrfs_ordered_sum *sums;
4243 if (ordered_end <= mod_start)
4245 if (mod_end <= ordered->file_offset)
4249 * We are going to copy all the csums on this ordered extent, so
4250 * go ahead and adjust mod_start and mod_len in case this ordered
4251 * extent has already been logged.
4253 if (ordered->file_offset > mod_start) {
4254 if (ordered_end >= mod_end)
4255 mod_len = ordered->file_offset - mod_start;
4257 * If we have this case
4259 * |--------- logged extent ---------|
4260 * |----- ordered extent ----|
4262 * Just don't mess with mod_start and mod_len, we'll
4263 * just end up logging more csums than we need and it
4267 if (ordered_end < mod_end) {
4268 mod_len = mod_end - ordered_end;
4269 mod_start = ordered_end;
4276 * To keep us from looping for the above case of an ordered
4277 * extent that falls inside of the logged extent.
4279 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4282 list_for_each_entry(sums, &ordered->list, list) {
4283 ret = log_csums(trans, inode, log_root, sums);
4289 /* We're done, found all csums in the ordered extents. */
4293 /* If we're compressed we have to save the entire range of csums. */
4294 if (em->compress_type) {
4296 csum_len = max(em->block_len, em->orig_block_len);
4298 csum_offset = mod_start - em->start;
4302 /* block start is already adjusted for the file extent offset. */
4303 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4304 em->block_start + csum_offset,
4305 em->block_start + csum_offset +
4306 csum_len - 1, &ordered_sums, 0);
4310 while (!list_empty(&ordered_sums)) {
4311 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4312 struct btrfs_ordered_sum,
4315 ret = log_csums(trans, inode, log_root, sums);
4316 list_del(&sums->list);
4323 static int log_one_extent(struct btrfs_trans_handle *trans,
4324 struct btrfs_inode *inode, struct btrfs_root *root,
4325 const struct extent_map *em,
4326 struct btrfs_path *path,
4327 struct btrfs_log_ctx *ctx)
4329 struct btrfs_drop_extents_args drop_args = { 0 };
4330 struct btrfs_root *log = root->log_root;
4331 struct btrfs_file_extent_item *fi;
4332 struct extent_buffer *leaf;
4333 struct btrfs_map_token token;
4334 struct btrfs_key key;
4335 u64 extent_offset = em->start - em->orig_start;
4339 ret = log_extent_csums(trans, inode, log, em, ctx);
4343 drop_args.path = path;
4344 drop_args.start = em->start;
4345 drop_args.end = em->start + em->len;
4346 drop_args.replace_extent = true;
4347 drop_args.extent_item_size = sizeof(*fi);
4348 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4352 if (!drop_args.extent_inserted) {
4353 key.objectid = btrfs_ino(inode);
4354 key.type = BTRFS_EXTENT_DATA_KEY;
4355 key.offset = em->start;
4357 ret = btrfs_insert_empty_item(trans, log, path, &key,
4362 leaf = path->nodes[0];
4363 btrfs_init_map_token(&token, leaf);
4364 fi = btrfs_item_ptr(leaf, path->slots[0],
4365 struct btrfs_file_extent_item);
4367 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4368 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4369 btrfs_set_token_file_extent_type(&token, fi,
4370 BTRFS_FILE_EXTENT_PREALLOC);
4372 btrfs_set_token_file_extent_type(&token, fi,
4373 BTRFS_FILE_EXTENT_REG);
4375 block_len = max(em->block_len, em->orig_block_len);
4376 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4377 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4379 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4380 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4381 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4384 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4386 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4387 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4390 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4391 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4392 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4393 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4394 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4395 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4396 btrfs_mark_buffer_dirty(leaf);
4398 btrfs_release_path(path);
4404 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4405 * lose them after doing a fast fsync and replaying the log. We scan the
4406 * subvolume's root instead of iterating the inode's extent map tree because
4407 * otherwise we can log incorrect extent items based on extent map conversion.
4408 * That can happen due to the fact that extent maps are merged when they
4409 * are not in the extent map tree's list of modified extents.
4411 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4412 struct btrfs_inode *inode,
4413 struct btrfs_path *path)
4415 struct btrfs_root *root = inode->root;
4416 struct btrfs_key key;
4417 const u64 i_size = i_size_read(&inode->vfs_inode);
4418 const u64 ino = btrfs_ino(inode);
4419 struct btrfs_path *dst_path = NULL;
4420 bool dropped_extents = false;
4421 u64 truncate_offset = i_size;
4422 struct extent_buffer *leaf;
4428 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4432 key.type = BTRFS_EXTENT_DATA_KEY;
4433 key.offset = i_size;
4434 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4439 * We must check if there is a prealloc extent that starts before the
4440 * i_size and crosses the i_size boundary. This is to ensure later we
4441 * truncate down to the end of that extent and not to the i_size, as
4442 * otherwise we end up losing part of the prealloc extent after a log
4443 * replay and with an implicit hole if there is another prealloc extent
4444 * that starts at an offset beyond i_size.
4446 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4451 struct btrfs_file_extent_item *ei;
4453 leaf = path->nodes[0];
4454 slot = path->slots[0];
4455 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4457 if (btrfs_file_extent_type(leaf, ei) ==
4458 BTRFS_FILE_EXTENT_PREALLOC) {
4461 btrfs_item_key_to_cpu(leaf, &key, slot);
4462 extent_end = key.offset +
4463 btrfs_file_extent_num_bytes(leaf, ei);
4465 if (extent_end > i_size)
4466 truncate_offset = extent_end;
4473 leaf = path->nodes[0];
4474 slot = path->slots[0];
4476 if (slot >= btrfs_header_nritems(leaf)) {
4478 ret = copy_items(trans, inode, dst_path, path,
4479 start_slot, ins_nr, 1, 0);
4484 ret = btrfs_next_leaf(root, path);
4494 btrfs_item_key_to_cpu(leaf, &key, slot);
4495 if (key.objectid > ino)
4497 if (WARN_ON_ONCE(key.objectid < ino) ||
4498 key.type < BTRFS_EXTENT_DATA_KEY ||
4499 key.offset < i_size) {
4503 if (!dropped_extents) {
4505 * Avoid logging extent items logged in past fsync calls
4506 * and leading to duplicate keys in the log tree.
4509 ret = btrfs_truncate_inode_items(trans,
4511 inode, truncate_offset,
4512 BTRFS_EXTENT_DATA_KEY,
4514 } while (ret == -EAGAIN);
4517 dropped_extents = true;
4524 dst_path = btrfs_alloc_path();
4532 ret = copy_items(trans, inode, dst_path, path,
4533 start_slot, ins_nr, 1, 0);
4535 btrfs_release_path(path);
4536 btrfs_free_path(dst_path);
4540 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4541 struct btrfs_root *root,
4542 struct btrfs_inode *inode,
4543 struct btrfs_path *path,
4544 struct btrfs_log_ctx *ctx)
4546 struct btrfs_ordered_extent *ordered;
4547 struct btrfs_ordered_extent *tmp;
4548 struct extent_map *em, *n;
4549 struct list_head extents;
4550 struct extent_map_tree *tree = &inode->extent_tree;
4554 INIT_LIST_HEAD(&extents);
4556 write_lock(&tree->lock);
4558 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4559 list_del_init(&em->list);
4561 * Just an arbitrary number, this can be really CPU intensive
4562 * once we start getting a lot of extents, and really once we
4563 * have a bunch of extents we just want to commit since it will
4566 if (++num > 32768) {
4567 list_del_init(&tree->modified_extents);
4572 if (em->generation < trans->transid)
4575 /* We log prealloc extents beyond eof later. */
4576 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4577 em->start >= i_size_read(&inode->vfs_inode))
4580 /* Need a ref to keep it from getting evicted from cache */
4581 refcount_inc(&em->refs);
4582 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4583 list_add_tail(&em->list, &extents);
4587 list_sort(NULL, &extents, extent_cmp);
4589 while (!list_empty(&extents)) {
4590 em = list_entry(extents.next, struct extent_map, list);
4592 list_del_init(&em->list);
4595 * If we had an error we just need to delete everybody from our
4599 clear_em_logging(tree, em);
4600 free_extent_map(em);
4604 write_unlock(&tree->lock);
4606 ret = log_one_extent(trans, inode, root, em, path, ctx);
4607 write_lock(&tree->lock);
4608 clear_em_logging(tree, em);
4609 free_extent_map(em);
4611 WARN_ON(!list_empty(&extents));
4612 write_unlock(&tree->lock);
4614 btrfs_release_path(path);
4616 ret = btrfs_log_prealloc_extents(trans, inode, path);
4621 * We have logged all extents successfully, now make sure the commit of
4622 * the current transaction waits for the ordered extents to complete
4623 * before it commits and wipes out the log trees, otherwise we would
4624 * lose data if an ordered extents completes after the transaction
4625 * commits and a power failure happens after the transaction commit.
4627 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4628 list_del_init(&ordered->log_list);
4629 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4631 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4632 spin_lock_irq(&inode->ordered_tree.lock);
4633 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4634 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4635 atomic_inc(&trans->transaction->pending_ordered);
4637 spin_unlock_irq(&inode->ordered_tree.lock);
4639 btrfs_put_ordered_extent(ordered);
4645 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4646 struct btrfs_path *path, u64 *size_ret)
4648 struct btrfs_key key;
4651 key.objectid = btrfs_ino(inode);
4652 key.type = BTRFS_INODE_ITEM_KEY;
4655 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4658 } else if (ret > 0) {
4661 struct btrfs_inode_item *item;
4663 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4664 struct btrfs_inode_item);
4665 *size_ret = btrfs_inode_size(path->nodes[0], item);
4667 * If the in-memory inode's i_size is smaller then the inode
4668 * size stored in the btree, return the inode's i_size, so
4669 * that we get a correct inode size after replaying the log
4670 * when before a power failure we had a shrinking truncate
4671 * followed by addition of a new name (rename / new hard link).
4672 * Otherwise return the inode size from the btree, to avoid
4673 * data loss when replaying a log due to previously doing a
4674 * write that expands the inode's size and logging a new name
4675 * immediately after.
4677 if (*size_ret > inode->vfs_inode.i_size)
4678 *size_ret = inode->vfs_inode.i_size;
4681 btrfs_release_path(path);
4686 * At the moment we always log all xattrs. This is to figure out at log replay
4687 * time which xattrs must have their deletion replayed. If a xattr is missing
4688 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4689 * because if a xattr is deleted, the inode is fsynced and a power failure
4690 * happens, causing the log to be replayed the next time the fs is mounted,
4691 * we want the xattr to not exist anymore (same behaviour as other filesystems
4692 * with a journal, ext3/4, xfs, f2fs, etc).
4694 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4695 struct btrfs_root *root,
4696 struct btrfs_inode *inode,
4697 struct btrfs_path *path,
4698 struct btrfs_path *dst_path)
4701 struct btrfs_key key;
4702 const u64 ino = btrfs_ino(inode);
4705 bool found_xattrs = false;
4707 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4711 key.type = BTRFS_XATTR_ITEM_KEY;
4714 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4719 int slot = path->slots[0];
4720 struct extent_buffer *leaf = path->nodes[0];
4721 int nritems = btrfs_header_nritems(leaf);
4723 if (slot >= nritems) {
4725 ret = copy_items(trans, inode, dst_path, path,
4726 start_slot, ins_nr, 1, 0);
4731 ret = btrfs_next_leaf(root, path);
4739 btrfs_item_key_to_cpu(leaf, &key, slot);
4740 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4747 found_xattrs = true;
4751 ret = copy_items(trans, inode, dst_path, path,
4752 start_slot, ins_nr, 1, 0);
4758 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4764 * When using the NO_HOLES feature if we punched a hole that causes the
4765 * deletion of entire leafs or all the extent items of the first leaf (the one
4766 * that contains the inode item and references) we may end up not processing
4767 * any extents, because there are no leafs with a generation matching the
4768 * current transaction that have extent items for our inode. So we need to find
4769 * if any holes exist and then log them. We also need to log holes after any
4770 * truncate operation that changes the inode's size.
4772 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4773 struct btrfs_root *root,
4774 struct btrfs_inode *inode,
4775 struct btrfs_path *path)
4777 struct btrfs_fs_info *fs_info = root->fs_info;
4778 struct btrfs_key key;
4779 const u64 ino = btrfs_ino(inode);
4780 const u64 i_size = i_size_read(&inode->vfs_inode);
4781 u64 prev_extent_end = 0;
4784 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4788 key.type = BTRFS_EXTENT_DATA_KEY;
4791 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4796 struct extent_buffer *leaf = path->nodes[0];
4798 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4799 ret = btrfs_next_leaf(root, path);
4806 leaf = path->nodes[0];
4809 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4810 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4813 /* We have a hole, log it. */
4814 if (prev_extent_end < key.offset) {
4815 const u64 hole_len = key.offset - prev_extent_end;
4818 * Release the path to avoid deadlocks with other code
4819 * paths that search the root while holding locks on
4820 * leafs from the log root.
4822 btrfs_release_path(path);
4823 ret = btrfs_insert_file_extent(trans, root->log_root,
4824 ino, prev_extent_end, 0,
4825 0, hole_len, 0, hole_len,
4831 * Search for the same key again in the root. Since it's
4832 * an extent item and we are holding the inode lock, the
4833 * key must still exist. If it doesn't just emit warning
4834 * and return an error to fall back to a transaction
4837 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4840 if (WARN_ON(ret > 0))
4842 leaf = path->nodes[0];
4845 prev_extent_end = btrfs_file_extent_end(path);
4850 if (prev_extent_end < i_size) {
4853 btrfs_release_path(path);
4854 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4855 ret = btrfs_insert_file_extent(trans, root->log_root,
4856 ino, prev_extent_end, 0, 0,
4857 hole_len, 0, hole_len,
4867 * When we are logging a new inode X, check if it doesn't have a reference that
4868 * matches the reference from some other inode Y created in a past transaction
4869 * and that was renamed in the current transaction. If we don't do this, then at
4870 * log replay time we can lose inode Y (and all its files if it's a directory):
4873 * echo "hello world" > /mnt/x/foobar
4876 * mkdir /mnt/x # or touch /mnt/x
4877 * xfs_io -c fsync /mnt/x
4879 * mount fs, trigger log replay
4881 * After the log replay procedure, we would lose the first directory and all its
4882 * files (file foobar).
4883 * For the case where inode Y is not a directory we simply end up losing it:
4885 * echo "123" > /mnt/foo
4887 * mv /mnt/foo /mnt/bar
4888 * echo "abc" > /mnt/foo
4889 * xfs_io -c fsync /mnt/foo
4892 * We also need this for cases where a snapshot entry is replaced by some other
4893 * entry (file or directory) otherwise we end up with an unreplayable log due to
4894 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4895 * if it were a regular entry:
4898 * btrfs subvolume snapshot /mnt /mnt/x/snap
4899 * btrfs subvolume delete /mnt/x/snap
4902 * fsync /mnt/x or fsync some new file inside it
4905 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4906 * the same transaction.
4908 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4910 const struct btrfs_key *key,
4911 struct btrfs_inode *inode,
4912 u64 *other_ino, u64 *other_parent)
4915 struct btrfs_path *search_path;
4918 u32 item_size = btrfs_item_size_nr(eb, slot);
4920 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4922 search_path = btrfs_alloc_path();
4925 search_path->search_commit_root = 1;
4926 search_path->skip_locking = 1;
4928 while (cur_offset < item_size) {
4932 unsigned long name_ptr;
4933 struct btrfs_dir_item *di;
4935 if (key->type == BTRFS_INODE_REF_KEY) {
4936 struct btrfs_inode_ref *iref;
4938 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4939 parent = key->offset;
4940 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4941 name_ptr = (unsigned long)(iref + 1);
4942 this_len = sizeof(*iref) + this_name_len;
4944 struct btrfs_inode_extref *extref;
4946 extref = (struct btrfs_inode_extref *)(ptr +
4948 parent = btrfs_inode_extref_parent(eb, extref);
4949 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4950 name_ptr = (unsigned long)&extref->name;
4951 this_len = sizeof(*extref) + this_name_len;
4954 if (this_name_len > name_len) {
4957 new_name = krealloc(name, this_name_len, GFP_NOFS);
4962 name_len = this_name_len;
4966 read_extent_buffer(eb, name, name_ptr, this_name_len);
4967 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4968 parent, name, this_name_len, 0);
4969 if (di && !IS_ERR(di)) {
4970 struct btrfs_key di_key;
4972 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4974 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4975 if (di_key.objectid != key->objectid) {
4977 *other_ino = di_key.objectid;
4978 *other_parent = parent;
4986 } else if (IS_ERR(di)) {
4990 btrfs_release_path(search_path);
4992 cur_offset += this_len;
4996 btrfs_free_path(search_path);
5001 struct btrfs_ino_list {
5004 struct list_head list;
5007 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5008 struct btrfs_root *root,
5009 struct btrfs_path *path,
5010 struct btrfs_log_ctx *ctx,
5011 u64 ino, u64 parent)
5013 struct btrfs_ino_list *ino_elem;
5014 LIST_HEAD(inode_list);
5017 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5020 ino_elem->ino = ino;
5021 ino_elem->parent = parent;
5022 list_add_tail(&ino_elem->list, &inode_list);
5024 while (!list_empty(&inode_list)) {
5025 struct btrfs_fs_info *fs_info = root->fs_info;
5026 struct btrfs_key key;
5027 struct inode *inode;
5029 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
5031 ino = ino_elem->ino;
5032 parent = ino_elem->parent;
5033 list_del(&ino_elem->list);
5038 btrfs_release_path(path);
5040 inode = btrfs_iget(fs_info->sb, ino, root);
5042 * If the other inode that had a conflicting dir entry was
5043 * deleted in the current transaction, we need to log its parent
5046 if (IS_ERR(inode)) {
5047 ret = PTR_ERR(inode);
5048 if (ret == -ENOENT) {
5049 inode = btrfs_iget(fs_info->sb, parent, root);
5050 if (IS_ERR(inode)) {
5051 ret = PTR_ERR(inode);
5053 ret = btrfs_log_inode(trans, root,
5055 LOG_OTHER_INODE_ALL,
5057 btrfs_add_delayed_iput(inode);
5063 * If the inode was already logged skip it - otherwise we can
5064 * hit an infinite loop. Example:
5066 * From the commit root (previous transaction) we have the
5069 * inode 257 a directory
5070 * inode 258 with references "zz" and "zz_link" on inode 257
5071 * inode 259 with reference "a" on inode 257
5073 * And in the current (uncommitted) transaction we have:
5075 * inode 257 a directory, unchanged
5076 * inode 258 with references "a" and "a2" on inode 257
5077 * inode 259 with reference "zz_link" on inode 257
5078 * inode 261 with reference "zz" on inode 257
5080 * When logging inode 261 the following infinite loop could
5081 * happen if we don't skip already logged inodes:
5083 * - we detect inode 258 as a conflicting inode, with inode 261
5084 * on reference "zz", and log it;
5086 * - we detect inode 259 as a conflicting inode, with inode 258
5087 * on reference "a", and log it;
5089 * - we detect inode 258 as a conflicting inode, with inode 259
5090 * on reference "zz_link", and log it - again! After this we
5091 * repeat the above steps forever.
5093 spin_lock(&BTRFS_I(inode)->lock);
5095 * Check the inode's logged_trans only instead of
5096 * btrfs_inode_in_log(). This is because the last_log_commit of
5097 * the inode is not updated when we only log that it exists (see
5098 * btrfs_log_inode()).
5100 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5101 spin_unlock(&BTRFS_I(inode)->lock);
5102 btrfs_add_delayed_iput(inode);
5105 spin_unlock(&BTRFS_I(inode)->lock);
5107 * We are safe logging the other inode without acquiring its
5108 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5109 * are safe against concurrent renames of the other inode as
5110 * well because during a rename we pin the log and update the
5111 * log with the new name before we unpin it.
5113 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5114 LOG_OTHER_INODE, ctx);
5116 btrfs_add_delayed_iput(inode);
5121 key.type = BTRFS_INODE_REF_KEY;
5123 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5125 btrfs_add_delayed_iput(inode);
5130 struct extent_buffer *leaf = path->nodes[0];
5131 int slot = path->slots[0];
5133 u64 other_parent = 0;
5135 if (slot >= btrfs_header_nritems(leaf)) {
5136 ret = btrfs_next_leaf(root, path);
5139 } else if (ret > 0) {
5146 btrfs_item_key_to_cpu(leaf, &key, slot);
5147 if (key.objectid != ino ||
5148 (key.type != BTRFS_INODE_REF_KEY &&
5149 key.type != BTRFS_INODE_EXTREF_KEY)) {
5154 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5155 BTRFS_I(inode), &other_ino,
5160 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5165 ino_elem->ino = other_ino;
5166 ino_elem->parent = other_parent;
5167 list_add_tail(&ino_elem->list, &inode_list);
5172 btrfs_add_delayed_iput(inode);
5178 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5179 struct btrfs_inode *inode,
5180 struct btrfs_key *min_key,
5181 const struct btrfs_key *max_key,
5182 struct btrfs_path *path,
5183 struct btrfs_path *dst_path,
5184 const u64 logged_isize,
5185 const bool recursive_logging,
5186 const int inode_only,
5187 struct btrfs_log_ctx *ctx,
5188 bool *need_log_inode_item)
5190 struct btrfs_root *root = inode->root;
5191 int ins_start_slot = 0;
5196 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5204 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5205 if (min_key->objectid != max_key->objectid)
5207 if (min_key->type > max_key->type)
5210 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5211 *need_log_inode_item = false;
5213 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5214 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5215 inode->generation == trans->transid &&
5216 !recursive_logging) {
5218 u64 other_parent = 0;
5220 ret = btrfs_check_ref_name_override(path->nodes[0],
5221 path->slots[0], min_key, inode,
5222 &other_ino, &other_parent);
5225 } else if (ret > 0 && ctx &&
5226 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5231 ins_start_slot = path->slots[0];
5233 ret = copy_items(trans, inode, dst_path, path,
5234 ins_start_slot, ins_nr,
5235 inode_only, logged_isize);
5240 ret = log_conflicting_inodes(trans, root, path,
5241 ctx, other_ino, other_parent);
5244 btrfs_release_path(path);
5249 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5250 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5253 ret = copy_items(trans, inode, dst_path, path,
5255 ins_nr, inode_only, logged_isize);
5262 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5265 } else if (!ins_nr) {
5266 ins_start_slot = path->slots[0];
5271 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5272 ins_nr, inode_only, logged_isize);
5276 ins_start_slot = path->slots[0];
5279 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5280 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5285 ret = copy_items(trans, inode, dst_path, path,
5286 ins_start_slot, ins_nr, inode_only,
5292 btrfs_release_path(path);
5294 if (min_key->offset < (u64)-1) {
5296 } else if (min_key->type < max_key->type) {
5298 min_key->offset = 0;
5304 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5305 ins_nr, inode_only, logged_isize);
5310 /* log a single inode in the tree log.
5311 * At least one parent directory for this inode must exist in the tree
5312 * or be logged already.
5314 * Any items from this inode changed by the current transaction are copied
5315 * to the log tree. An extra reference is taken on any extents in this
5316 * file, allowing us to avoid a whole pile of corner cases around logging
5317 * blocks that have been removed from the tree.
5319 * See LOG_INODE_ALL and related defines for a description of what inode_only
5322 * This handles both files and directories.
5324 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5325 struct btrfs_root *root, struct btrfs_inode *inode,
5327 struct btrfs_log_ctx *ctx)
5329 struct btrfs_path *path;
5330 struct btrfs_path *dst_path;
5331 struct btrfs_key min_key;
5332 struct btrfs_key max_key;
5333 struct btrfs_root *log = root->log_root;
5336 bool fast_search = false;
5337 u64 ino = btrfs_ino(inode);
5338 struct extent_map_tree *em_tree = &inode->extent_tree;
5339 u64 logged_isize = 0;
5340 bool need_log_inode_item = true;
5341 bool xattrs_logged = false;
5342 bool recursive_logging = false;
5343 bool inode_item_dropped = true;
5345 path = btrfs_alloc_path();
5348 dst_path = btrfs_alloc_path();
5350 btrfs_free_path(path);
5354 min_key.objectid = ino;
5355 min_key.type = BTRFS_INODE_ITEM_KEY;
5358 max_key.objectid = ino;
5361 /* today the code can only do partial logging of directories */
5362 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5363 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5364 &inode->runtime_flags) &&
5365 inode_only >= LOG_INODE_EXISTS))
5366 max_key.type = BTRFS_XATTR_ITEM_KEY;
5368 max_key.type = (u8)-1;
5369 max_key.offset = (u64)-1;
5372 * Only run delayed items if we are a directory. We want to make sure
5373 * all directory indexes hit the fs/subvolume tree so we can find them
5374 * and figure out which index ranges have to be logged.
5376 * Otherwise commit the delayed inode only if the full sync flag is set,
5377 * as we want to make sure an up to date version is in the subvolume
5378 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5379 * it to the log tree. For a non full sync, we always log the inode item
5380 * based on the in-memory struct btrfs_inode which is always up to date.
5382 if (S_ISDIR(inode->vfs_inode.i_mode))
5383 ret = btrfs_commit_inode_delayed_items(trans, inode);
5384 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5385 ret = btrfs_commit_inode_delayed_inode(inode);
5388 btrfs_free_path(path);
5389 btrfs_free_path(dst_path);
5393 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5394 recursive_logging = true;
5395 if (inode_only == LOG_OTHER_INODE)
5396 inode_only = LOG_INODE_EXISTS;
5398 inode_only = LOG_INODE_ALL;
5399 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5401 mutex_lock(&inode->log_mutex);
5405 * This is for cases where logging a directory could result in losing a
5406 * a file after replaying the log. For example, if we move a file from a
5407 * directory A to a directory B, then fsync directory A, we have no way
5408 * to known the file was moved from A to B, so logging just A would
5409 * result in losing the file after a log replay.
5411 if (S_ISDIR(inode->vfs_inode.i_mode) &&
5412 inode_only == LOG_INODE_ALL &&
5413 inode->last_unlink_trans >= trans->transid) {
5414 btrfs_set_log_full_commit(trans);
5420 * a brute force approach to making sure we get the most uptodate
5421 * copies of everything.
5423 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5424 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5426 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5427 if (inode_only == LOG_INODE_EXISTS)
5428 max_key_type = BTRFS_XATTR_ITEM_KEY;
5429 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5431 if (inode_only == LOG_INODE_EXISTS) {
5433 * Make sure the new inode item we write to the log has
5434 * the same isize as the current one (if it exists).
5435 * This is necessary to prevent data loss after log
5436 * replay, and also to prevent doing a wrong expanding
5437 * truncate - for e.g. create file, write 4K into offset
5438 * 0, fsync, write 4K into offset 4096, add hard link,
5439 * fsync some other file (to sync log), power fail - if
5440 * we use the inode's current i_size, after log replay
5441 * we get a 8Kb file, with the last 4Kb extent as a hole
5442 * (zeroes), as if an expanding truncate happened,
5443 * instead of getting a file of 4Kb only.
5445 err = logged_inode_size(log, inode, path, &logged_isize);
5449 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5450 &inode->runtime_flags)) {
5451 if (inode_only == LOG_INODE_EXISTS) {
5452 max_key.type = BTRFS_XATTR_ITEM_KEY;
5453 ret = drop_objectid_items(trans, log, path, ino,
5456 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5457 &inode->runtime_flags);
5458 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5459 &inode->runtime_flags);
5461 ret = btrfs_truncate_inode_items(trans,
5462 log, inode, 0, 0, NULL);
5467 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5468 &inode->runtime_flags) ||
5469 inode_only == LOG_INODE_EXISTS) {
5470 if (inode_only == LOG_INODE_ALL)
5472 max_key.type = BTRFS_XATTR_ITEM_KEY;
5473 ret = drop_objectid_items(trans, log, path, ino,
5476 if (inode_only == LOG_INODE_ALL)
5478 inode_item_dropped = false;
5488 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5489 path, dst_path, logged_isize,
5490 recursive_logging, inode_only, ctx,
5491 &need_log_inode_item);
5495 btrfs_release_path(path);
5496 btrfs_release_path(dst_path);
5497 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5500 xattrs_logged = true;
5501 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5502 btrfs_release_path(path);
5503 btrfs_release_path(dst_path);
5504 err = btrfs_log_holes(trans, root, inode, path);
5509 btrfs_release_path(path);
5510 btrfs_release_path(dst_path);
5511 if (need_log_inode_item) {
5512 err = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
5516 * If we are doing a fast fsync and the inode was logged before
5517 * in this transaction, we don't need to log the xattrs because
5518 * they were logged before. If xattrs were added, changed or
5519 * deleted since the last time we logged the inode, then we have
5520 * already logged them because the inode had the runtime flag
5521 * BTRFS_INODE_COPY_EVERYTHING set.
5523 if (!xattrs_logged && inode->logged_trans < trans->transid) {
5524 err = btrfs_log_all_xattrs(trans, root, inode, path,
5528 btrfs_release_path(path);
5532 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5538 } else if (inode_only == LOG_INODE_ALL) {
5539 struct extent_map *em, *n;
5541 write_lock(&em_tree->lock);
5542 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5543 list_del_init(&em->list);
5544 write_unlock(&em_tree->lock);
5547 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5548 ret = log_directory_changes(trans, root, inode, path, dst_path,
5557 * If we are logging that an ancestor inode exists as part of logging a
5558 * new name from a link or rename operation, don't mark the inode as
5559 * logged - otherwise if an explicit fsync is made against an ancestor,
5560 * the fsync considers the inode in the log and doesn't sync the log,
5561 * resulting in the ancestor missing after a power failure unless the
5562 * log was synced as part of an fsync against any other unrelated inode.
5563 * So keep it simple for this case and just don't flag the ancestors as
5567 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5568 &inode->vfs_inode != ctx->inode)) {
5569 spin_lock(&inode->lock);
5570 inode->logged_trans = trans->transid;
5572 * Don't update last_log_commit if we logged that an inode exists.
5573 * We do this for two reasons:
5575 * 1) We might have had buffered writes to this inode that were
5576 * flushed and had their ordered extents completed in this
5577 * transaction, but we did not previously log the inode with
5578 * LOG_INODE_ALL. Later the inode was evicted and after that
5579 * it was loaded again and this LOG_INODE_EXISTS log operation
5580 * happened. We must make sure that if an explicit fsync against
5581 * the inode is performed later, it logs the new extents, an
5582 * updated inode item, etc, and syncs the log. The same logic
5583 * applies to direct IO writes instead of buffered writes.
5585 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
5586 * is logged with an i_size of 0 or whatever value was logged
5587 * before. If later the i_size of the inode is increased by a
5588 * truncate operation, the log is synced through an fsync of
5589 * some other inode and then finally an explicit fsync against
5590 * this inode is made, we must make sure this fsync logs the
5591 * inode with the new i_size, the hole between old i_size and
5592 * the new i_size, and syncs the log.
5594 if (inode_only != LOG_INODE_EXISTS)
5595 inode->last_log_commit = inode->last_sub_trans;
5596 spin_unlock(&inode->lock);
5599 mutex_unlock(&inode->log_mutex);
5601 btrfs_free_path(path);
5602 btrfs_free_path(dst_path);
5607 * Check if we need to log an inode. This is used in contexts where while
5608 * logging an inode we need to log another inode (either that it exists or in
5609 * full mode). This is used instead of btrfs_inode_in_log() because the later
5610 * requires the inode to be in the log and have the log transaction committed,
5611 * while here we do not care if the log transaction was already committed - our
5612 * caller will commit the log later - and we want to avoid logging an inode
5613 * multiple times when multiple tasks have joined the same log transaction.
5615 static bool need_log_inode(struct btrfs_trans_handle *trans,
5616 struct btrfs_inode *inode)
5619 * If a directory was not modified, no dentries added or removed, we can
5620 * and should avoid logging it.
5622 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5626 * If this inode does not have new/updated/deleted xattrs since the last
5627 * time it was logged and is flagged as logged in the current transaction,
5628 * we can skip logging it. As for new/deleted names, those are updated in
5629 * the log by link/unlink/rename operations.
5630 * In case the inode was logged and then evicted and reloaded, its
5631 * logged_trans will be 0, in which case we have to fully log it since
5632 * logged_trans is a transient field, not persisted.
5634 if (inode->logged_trans == trans->transid &&
5635 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5641 struct btrfs_dir_list {
5643 struct list_head list;
5647 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5648 * details about the why it is needed.
5649 * This is a recursive operation - if an existing dentry corresponds to a
5650 * directory, that directory's new entries are logged too (same behaviour as
5651 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5652 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5653 * complains about the following circular lock dependency / possible deadlock:
5657 * lock(&type->i_mutex_dir_key#3/2);
5658 * lock(sb_internal#2);
5659 * lock(&type->i_mutex_dir_key#3/2);
5660 * lock(&sb->s_type->i_mutex_key#14);
5662 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5663 * sb_start_intwrite() in btrfs_start_transaction().
5664 * Not locking i_mutex of the inodes is still safe because:
5666 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5667 * that while logging the inode new references (names) are added or removed
5668 * from the inode, leaving the logged inode item with a link count that does
5669 * not match the number of logged inode reference items. This is fine because
5670 * at log replay time we compute the real number of links and correct the
5671 * link count in the inode item (see replay_one_buffer() and
5672 * link_to_fixup_dir());
5674 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5675 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5676 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5677 * has a size that doesn't match the sum of the lengths of all the logged
5678 * names. This does not result in a problem because if a dir_item key is
5679 * logged but its matching dir_index key is not logged, at log replay time we
5680 * don't use it to replay the respective name (see replay_one_name()). On the
5681 * other hand if only the dir_index key ends up being logged, the respective
5682 * name is added to the fs/subvol tree with both the dir_item and dir_index
5683 * keys created (see replay_one_name()).
5684 * The directory's inode item with a wrong i_size is not a problem as well,
5685 * since we don't use it at log replay time to set the i_size in the inode
5686 * item of the fs/subvol tree (see overwrite_item()).
5688 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5689 struct btrfs_root *root,
5690 struct btrfs_inode *start_inode,
5691 struct btrfs_log_ctx *ctx)
5693 struct btrfs_fs_info *fs_info = root->fs_info;
5694 struct btrfs_root *log = root->log_root;
5695 struct btrfs_path *path;
5696 LIST_HEAD(dir_list);
5697 struct btrfs_dir_list *dir_elem;
5700 path = btrfs_alloc_path();
5704 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5706 btrfs_free_path(path);
5709 dir_elem->ino = btrfs_ino(start_inode);
5710 list_add_tail(&dir_elem->list, &dir_list);
5712 while (!list_empty(&dir_list)) {
5713 struct extent_buffer *leaf;
5714 struct btrfs_key min_key;
5718 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5721 goto next_dir_inode;
5723 min_key.objectid = dir_elem->ino;
5724 min_key.type = BTRFS_DIR_ITEM_KEY;
5727 btrfs_release_path(path);
5728 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5730 goto next_dir_inode;
5731 } else if (ret > 0) {
5733 goto next_dir_inode;
5737 leaf = path->nodes[0];
5738 nritems = btrfs_header_nritems(leaf);
5739 for (i = path->slots[0]; i < nritems; i++) {
5740 struct btrfs_dir_item *di;
5741 struct btrfs_key di_key;
5742 struct inode *di_inode;
5743 struct btrfs_dir_list *new_dir_elem;
5744 int log_mode = LOG_INODE_EXISTS;
5747 btrfs_item_key_to_cpu(leaf, &min_key, i);
5748 if (min_key.objectid != dir_elem->ino ||
5749 min_key.type != BTRFS_DIR_ITEM_KEY)
5750 goto next_dir_inode;
5752 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5753 type = btrfs_dir_type(leaf, di);
5754 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5755 type != BTRFS_FT_DIR)
5757 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5758 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5761 btrfs_release_path(path);
5762 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5763 if (IS_ERR(di_inode)) {
5764 ret = PTR_ERR(di_inode);
5765 goto next_dir_inode;
5768 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5769 btrfs_add_delayed_iput(di_inode);
5773 ctx->log_new_dentries = false;
5774 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5775 log_mode = LOG_INODE_ALL;
5776 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5778 btrfs_add_delayed_iput(di_inode);
5780 goto next_dir_inode;
5781 if (ctx->log_new_dentries) {
5782 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5784 if (!new_dir_elem) {
5786 goto next_dir_inode;
5788 new_dir_elem->ino = di_key.objectid;
5789 list_add_tail(&new_dir_elem->list, &dir_list);
5794 ret = btrfs_next_leaf(log, path);
5796 goto next_dir_inode;
5797 } else if (ret > 0) {
5799 goto next_dir_inode;
5803 if (min_key.offset < (u64)-1) {
5808 list_del(&dir_elem->list);
5812 btrfs_free_path(path);
5816 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5817 struct btrfs_inode *inode,
5818 struct btrfs_log_ctx *ctx)
5820 struct btrfs_fs_info *fs_info = trans->fs_info;
5822 struct btrfs_path *path;
5823 struct btrfs_key key;
5824 struct btrfs_root *root = inode->root;
5825 const u64 ino = btrfs_ino(inode);
5827 path = btrfs_alloc_path();
5830 path->skip_locking = 1;
5831 path->search_commit_root = 1;
5834 key.type = BTRFS_INODE_REF_KEY;
5836 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5841 struct extent_buffer *leaf = path->nodes[0];
5842 int slot = path->slots[0];
5847 if (slot >= btrfs_header_nritems(leaf)) {
5848 ret = btrfs_next_leaf(root, path);
5856 btrfs_item_key_to_cpu(leaf, &key, slot);
5857 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5858 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5861 item_size = btrfs_item_size_nr(leaf, slot);
5862 ptr = btrfs_item_ptr_offset(leaf, slot);
5863 while (cur_offset < item_size) {
5864 struct btrfs_key inode_key;
5865 struct inode *dir_inode;
5867 inode_key.type = BTRFS_INODE_ITEM_KEY;
5868 inode_key.offset = 0;
5870 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5871 struct btrfs_inode_extref *extref;
5873 extref = (struct btrfs_inode_extref *)
5875 inode_key.objectid = btrfs_inode_extref_parent(
5877 cur_offset += sizeof(*extref);
5878 cur_offset += btrfs_inode_extref_name_len(leaf,
5881 inode_key.objectid = key.offset;
5882 cur_offset = item_size;
5885 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5888 * If the parent inode was deleted, return an error to
5889 * fallback to a transaction commit. This is to prevent
5890 * getting an inode that was moved from one parent A to
5891 * a parent B, got its former parent A deleted and then
5892 * it got fsync'ed, from existing at both parents after
5893 * a log replay (and the old parent still existing).
5900 * mv /mnt/B/bar /mnt/A/bar
5901 * mv -T /mnt/A /mnt/B
5905 * If we ignore the old parent B which got deleted,
5906 * after a log replay we would have file bar linked
5907 * at both parents and the old parent B would still
5910 if (IS_ERR(dir_inode)) {
5911 ret = PTR_ERR(dir_inode);
5915 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
5916 btrfs_add_delayed_iput(dir_inode);
5921 ctx->log_new_dentries = false;
5922 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5923 LOG_INODE_ALL, ctx);
5924 if (!ret && ctx && ctx->log_new_dentries)
5925 ret = log_new_dir_dentries(trans, root,
5926 BTRFS_I(dir_inode), ctx);
5927 btrfs_add_delayed_iput(dir_inode);
5935 btrfs_free_path(path);
5939 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5940 struct btrfs_root *root,
5941 struct btrfs_path *path,
5942 struct btrfs_log_ctx *ctx)
5944 struct btrfs_key found_key;
5946 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5949 struct btrfs_fs_info *fs_info = root->fs_info;
5950 struct extent_buffer *leaf = path->nodes[0];
5951 int slot = path->slots[0];
5952 struct btrfs_key search_key;
5953 struct inode *inode;
5957 btrfs_release_path(path);
5959 ino = found_key.offset;
5961 search_key.objectid = found_key.offset;
5962 search_key.type = BTRFS_INODE_ITEM_KEY;
5963 search_key.offset = 0;
5964 inode = btrfs_iget(fs_info->sb, ino, root);
5966 return PTR_ERR(inode);
5968 if (BTRFS_I(inode)->generation >= trans->transid &&
5969 need_log_inode(trans, BTRFS_I(inode)))
5970 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5971 LOG_INODE_EXISTS, ctx);
5972 btrfs_add_delayed_iput(inode);
5976 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5979 search_key.type = BTRFS_INODE_REF_KEY;
5980 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5984 leaf = path->nodes[0];
5985 slot = path->slots[0];
5986 if (slot >= btrfs_header_nritems(leaf)) {
5987 ret = btrfs_next_leaf(root, path);
5992 leaf = path->nodes[0];
5993 slot = path->slots[0];
5996 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5997 if (found_key.objectid != search_key.objectid ||
5998 found_key.type != BTRFS_INODE_REF_KEY)
6004 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6005 struct btrfs_inode *inode,
6006 struct dentry *parent,
6007 struct btrfs_log_ctx *ctx)
6009 struct btrfs_root *root = inode->root;
6010 struct dentry *old_parent = NULL;
6011 struct super_block *sb = inode->vfs_inode.i_sb;
6015 if (!parent || d_really_is_negative(parent) ||
6019 inode = BTRFS_I(d_inode(parent));
6020 if (root != inode->root)
6023 if (inode->generation >= trans->transid &&
6024 need_log_inode(trans, inode)) {
6025 ret = btrfs_log_inode(trans, root, inode,
6026 LOG_INODE_EXISTS, ctx);
6030 if (IS_ROOT(parent))
6033 parent = dget_parent(parent);
6035 old_parent = parent;
6042 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6043 struct btrfs_inode *inode,
6044 struct dentry *parent,
6045 struct btrfs_log_ctx *ctx)
6047 struct btrfs_root *root = inode->root;
6048 const u64 ino = btrfs_ino(inode);
6049 struct btrfs_path *path;
6050 struct btrfs_key search_key;
6054 * For a single hard link case, go through a fast path that does not
6055 * need to iterate the fs/subvolume tree.
6057 if (inode->vfs_inode.i_nlink < 2)
6058 return log_new_ancestors_fast(trans, inode, parent, ctx);
6060 path = btrfs_alloc_path();
6064 search_key.objectid = ino;
6065 search_key.type = BTRFS_INODE_REF_KEY;
6066 search_key.offset = 0;
6068 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6075 struct extent_buffer *leaf = path->nodes[0];
6076 int slot = path->slots[0];
6077 struct btrfs_key found_key;
6079 if (slot >= btrfs_header_nritems(leaf)) {
6080 ret = btrfs_next_leaf(root, path);
6088 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6089 if (found_key.objectid != ino ||
6090 found_key.type > BTRFS_INODE_EXTREF_KEY)
6094 * Don't deal with extended references because they are rare
6095 * cases and too complex to deal with (we would need to keep
6096 * track of which subitem we are processing for each item in
6097 * this loop, etc). So just return some error to fallback to
6098 * a transaction commit.
6100 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6106 * Logging ancestors needs to do more searches on the fs/subvol
6107 * tree, so it releases the path as needed to avoid deadlocks.
6108 * Keep track of the last inode ref key and resume from that key
6109 * after logging all new ancestors for the current hard link.
6111 memcpy(&search_key, &found_key, sizeof(search_key));
6113 ret = log_new_ancestors(trans, root, path, ctx);
6116 btrfs_release_path(path);
6121 btrfs_free_path(path);
6126 * helper function around btrfs_log_inode to make sure newly created
6127 * parent directories also end up in the log. A minimal inode and backref
6128 * only logging is done of any parent directories that are older than
6129 * the last committed transaction
6131 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6132 struct btrfs_inode *inode,
6133 struct dentry *parent,
6135 struct btrfs_log_ctx *ctx)
6137 struct btrfs_root *root = inode->root;
6138 struct btrfs_fs_info *fs_info = root->fs_info;
6140 bool log_dentries = false;
6142 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6147 if (btrfs_root_refs(&root->root_item) == 0) {
6153 * Skip already logged inodes or inodes corresponding to tmpfiles
6154 * (since logging them is pointless, a link count of 0 means they
6155 * will never be accessible).
6157 if ((btrfs_inode_in_log(inode, trans->transid) &&
6158 list_empty(&ctx->ordered_extents)) ||
6159 inode->vfs_inode.i_nlink == 0) {
6160 ret = BTRFS_NO_LOG_SYNC;
6164 ret = start_log_trans(trans, root, ctx);
6168 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6173 * for regular files, if its inode is already on disk, we don't
6174 * have to worry about the parents at all. This is because
6175 * we can use the last_unlink_trans field to record renames
6176 * and other fun in this file.
6178 if (S_ISREG(inode->vfs_inode.i_mode) &&
6179 inode->generation < trans->transid &&
6180 inode->last_unlink_trans < trans->transid) {
6185 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6186 log_dentries = true;
6189 * On unlink we must make sure all our current and old parent directory
6190 * inodes are fully logged. This is to prevent leaving dangling
6191 * directory index entries in directories that were our parents but are
6192 * not anymore. Not doing this results in old parent directory being
6193 * impossible to delete after log replay (rmdir will always fail with
6194 * error -ENOTEMPTY).
6200 * ln testdir/foo testdir/bar
6202 * unlink testdir/bar
6203 * xfs_io -c fsync testdir/foo
6205 * mount fs, triggers log replay
6207 * If we don't log the parent directory (testdir), after log replay the
6208 * directory still has an entry pointing to the file inode using the bar
6209 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6210 * the file inode has a link count of 1.
6216 * ln foo testdir/foo2
6217 * ln foo testdir/foo3
6219 * unlink testdir/foo3
6220 * xfs_io -c fsync foo
6222 * mount fs, triggers log replay
6224 * Similar as the first example, after log replay the parent directory
6225 * testdir still has an entry pointing to the inode file with name foo3
6226 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6227 * and has a link count of 2.
6229 if (inode->last_unlink_trans >= trans->transid) {
6230 ret = btrfs_log_all_parents(trans, inode, ctx);
6235 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6240 ret = log_new_dir_dentries(trans, root, inode, ctx);
6245 btrfs_set_log_full_commit(trans);
6250 btrfs_remove_log_ctx(root, ctx);
6251 btrfs_end_log_trans(root);
6257 * it is not safe to log dentry if the chunk root has added new
6258 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6259 * If this returns 1, you must commit the transaction to safely get your
6262 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6263 struct dentry *dentry,
6264 struct btrfs_log_ctx *ctx)
6266 struct dentry *parent = dget_parent(dentry);
6269 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6270 LOG_INODE_ALL, ctx);
6277 * should be called during mount to recover any replay any log trees
6280 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6283 struct btrfs_path *path;
6284 struct btrfs_trans_handle *trans;
6285 struct btrfs_key key;
6286 struct btrfs_key found_key;
6287 struct btrfs_root *log;
6288 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6289 struct walk_control wc = {
6290 .process_func = process_one_buffer,
6291 .stage = LOG_WALK_PIN_ONLY,
6294 path = btrfs_alloc_path();
6298 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6300 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6301 if (IS_ERR(trans)) {
6302 ret = PTR_ERR(trans);
6309 ret = walk_log_tree(trans, log_root_tree, &wc);
6311 btrfs_handle_fs_error(fs_info, ret,
6312 "Failed to pin buffers while recovering log root tree.");
6317 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6318 key.offset = (u64)-1;
6319 key.type = BTRFS_ROOT_ITEM_KEY;
6322 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6325 btrfs_handle_fs_error(fs_info, ret,
6326 "Couldn't find tree log root.");
6330 if (path->slots[0] == 0)
6334 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6336 btrfs_release_path(path);
6337 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6340 log = btrfs_read_tree_root(log_root_tree, &found_key);
6343 btrfs_handle_fs_error(fs_info, ret,
6344 "Couldn't read tree log root.");
6348 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6350 if (IS_ERR(wc.replay_dest)) {
6351 ret = PTR_ERR(wc.replay_dest);
6354 * We didn't find the subvol, likely because it was
6355 * deleted. This is ok, simply skip this log and go to
6358 * We need to exclude the root because we can't have
6359 * other log replays overwriting this log as we'll read
6360 * it back in a few more times. This will keep our
6361 * block from being modified, and we'll just bail for
6362 * each subsequent pass.
6365 ret = btrfs_pin_extent_for_log_replay(trans,
6368 btrfs_put_root(log);
6372 btrfs_handle_fs_error(fs_info, ret,
6373 "Couldn't read target root for tree log recovery.");
6377 wc.replay_dest->log_root = log;
6378 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
6380 /* The loop needs to continue due to the root refs */
6381 btrfs_handle_fs_error(fs_info, ret,
6382 "failed to record the log root in transaction");
6384 ret = walk_log_tree(trans, log, &wc);
6386 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6387 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6391 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6392 struct btrfs_root *root = wc.replay_dest;
6394 btrfs_release_path(path);
6397 * We have just replayed everything, and the highest
6398 * objectid of fs roots probably has changed in case
6399 * some inode_item's got replayed.
6401 * root->objectid_mutex is not acquired as log replay
6402 * could only happen during mount.
6404 ret = btrfs_init_root_free_objectid(root);
6407 wc.replay_dest->log_root = NULL;
6408 btrfs_put_root(wc.replay_dest);
6409 btrfs_put_root(log);
6414 if (found_key.offset == 0)
6416 key.offset = found_key.offset - 1;
6418 btrfs_release_path(path);
6420 /* step one is to pin it all, step two is to replay just inodes */
6423 wc.process_func = replay_one_buffer;
6424 wc.stage = LOG_WALK_REPLAY_INODES;
6427 /* step three is to replay everything */
6428 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6433 btrfs_free_path(path);
6435 /* step 4: commit the transaction, which also unpins the blocks */
6436 ret = btrfs_commit_transaction(trans);
6440 log_root_tree->log_root = NULL;
6441 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6442 btrfs_put_root(log_root_tree);
6447 btrfs_end_transaction(wc.trans);
6448 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6449 btrfs_free_path(path);
6454 * there are some corner cases where we want to force a full
6455 * commit instead of allowing a directory to be logged.
6457 * They revolve around files there were unlinked from the directory, and
6458 * this function updates the parent directory so that a full commit is
6459 * properly done if it is fsync'd later after the unlinks are done.
6461 * Must be called before the unlink operations (updates to the subvolume tree,
6462 * inodes, etc) are done.
6464 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6465 struct btrfs_inode *dir, struct btrfs_inode *inode,
6469 * when we're logging a file, if it hasn't been renamed
6470 * or unlinked, and its inode is fully committed on disk,
6471 * we don't have to worry about walking up the directory chain
6472 * to log its parents.
6474 * So, we use the last_unlink_trans field to put this transid
6475 * into the file. When the file is logged we check it and
6476 * don't log the parents if the file is fully on disk.
6478 mutex_lock(&inode->log_mutex);
6479 inode->last_unlink_trans = trans->transid;
6480 mutex_unlock(&inode->log_mutex);
6483 * if this directory was already logged any new
6484 * names for this file/dir will get recorded
6486 if (dir->logged_trans == trans->transid)
6490 * if the inode we're about to unlink was logged,
6491 * the log will be properly updated for any new names
6493 if (inode->logged_trans == trans->transid)
6497 * when renaming files across directories, if the directory
6498 * there we're unlinking from gets fsync'd later on, there's
6499 * no way to find the destination directory later and fsync it
6500 * properly. So, we have to be conservative and force commits
6501 * so the new name gets discovered.
6506 /* we can safely do the unlink without any special recording */
6510 mutex_lock(&dir->log_mutex);
6511 dir->last_unlink_trans = trans->transid;
6512 mutex_unlock(&dir->log_mutex);
6516 * Make sure that if someone attempts to fsync the parent directory of a deleted
6517 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6518 * that after replaying the log tree of the parent directory's root we will not
6519 * see the snapshot anymore and at log replay time we will not see any log tree
6520 * corresponding to the deleted snapshot's root, which could lead to replaying
6521 * it after replaying the log tree of the parent directory (which would replay
6522 * the snapshot delete operation).
6524 * Must be called before the actual snapshot destroy operation (updates to the
6525 * parent root and tree of tree roots trees, etc) are done.
6527 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6528 struct btrfs_inode *dir)
6530 mutex_lock(&dir->log_mutex);
6531 dir->last_unlink_trans = trans->transid;
6532 mutex_unlock(&dir->log_mutex);
6536 * Call this after adding a new name for a file and it will properly
6537 * update the log to reflect the new name.
6539 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6540 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6541 struct dentry *parent)
6543 struct btrfs_log_ctx ctx;
6546 * this will force the logging code to walk the dentry chain
6549 if (!S_ISDIR(inode->vfs_inode.i_mode))
6550 inode->last_unlink_trans = trans->transid;
6553 * if this inode hasn't been logged and directory we're renaming it
6554 * from hasn't been logged, we don't need to log it
6556 if (!inode_logged(trans, inode) &&
6557 (!old_dir || !inode_logged(trans, old_dir)))
6561 * If we are doing a rename (old_dir is not NULL) from a directory that
6562 * was previously logged, make sure the next log attempt on the directory
6563 * is not skipped and logs the inode again. This is because the log may
6564 * not currently be authoritative for a range including the old
6565 * BTRFS_DIR_ITEM_KEY and BTRFS_DIR_INDEX_KEY keys, so we want to make
6566 * sure after a log replay we do not end up with both the new and old
6567 * dentries around (in case the inode is a directory we would have a
6568 * directory with two hard links and 2 inode references for different
6569 * parents). The next log attempt of old_dir will happen at
6570 * btrfs_log_all_parents(), called through btrfs_log_inode_parent()
6571 * below, because we have previously set inode->last_unlink_trans to the
6572 * current transaction ID, either here or at btrfs_record_unlink_dir() in
6573 * case inode is a directory.
6576 old_dir->logged_trans = 0;
6578 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6579 ctx.logging_new_name = true;
6581 * We don't care about the return value. If we fail to log the new name
6582 * then we know the next attempt to sync the log will fallback to a full
6583 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6584 * we don't need to worry about getting a log committed that has an
6585 * inconsistent state after a rename operation.
6587 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);