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, 0, false);
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));
887 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
888 struct btrfs_inode *dir,
889 struct btrfs_inode *inode,
895 ret = btrfs_unlink_inode(trans, dir, inode, name, name_len);
899 * Whenever we need to check if a name exists or not, we check the
900 * fs/subvolume tree. So after an unlink we must run delayed items, so
901 * that future checks for a name during log replay see that the name
902 * does not exists anymore.
904 return btrfs_run_delayed_items(trans);
908 * when cleaning up conflicts between the directory names in the
909 * subvolume, directory names in the log and directory names in the
910 * inode back references, we may have to unlink inodes from directories.
912 * This is a helper function to do the unlink of a specific directory
915 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
916 struct btrfs_root *root,
917 struct btrfs_path *path,
918 struct btrfs_inode *dir,
919 struct btrfs_dir_item *di)
924 struct extent_buffer *leaf;
925 struct btrfs_key location;
928 leaf = path->nodes[0];
930 btrfs_dir_item_key_to_cpu(leaf, di, &location);
931 name_len = btrfs_dir_name_len(leaf, di);
932 name = kmalloc(name_len, GFP_NOFS);
936 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
937 btrfs_release_path(path);
939 inode = read_one_inode(root, location.objectid);
945 ret = link_to_fixup_dir(trans, root, path, location.objectid);
949 ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), name,
958 * See if a given name and sequence number found in an inode back reference are
959 * already in a directory and correctly point to this inode.
961 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
964 static noinline int inode_in_dir(struct btrfs_root *root,
965 struct btrfs_path *path,
966 u64 dirid, u64 objectid, u64 index,
967 const char *name, int name_len)
969 struct btrfs_dir_item *di;
970 struct btrfs_key location;
973 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
974 index, name, name_len, 0);
979 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
980 if (location.objectid != objectid)
986 btrfs_release_path(path);
987 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
992 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
993 if (location.objectid == objectid)
997 btrfs_release_path(path);
1002 * helper function to check a log tree for a named back reference in
1003 * an inode. This is used to decide if a back reference that is
1004 * found in the subvolume conflicts with what we find in the log.
1006 * inode backreferences may have multiple refs in a single item,
1007 * during replay we process one reference at a time, and we don't
1008 * want to delete valid links to a file from the subvolume if that
1009 * link is also in the log.
1011 static noinline int backref_in_log(struct btrfs_root *log,
1012 struct btrfs_key *key,
1014 const char *name, int namelen)
1016 struct btrfs_path *path;
1019 path = btrfs_alloc_path();
1023 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1026 } else if (ret == 1) {
1031 if (key->type == BTRFS_INODE_EXTREF_KEY)
1032 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1037 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1041 btrfs_free_path(path);
1045 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1046 struct btrfs_root *root,
1047 struct btrfs_path *path,
1048 struct btrfs_root *log_root,
1049 struct btrfs_inode *dir,
1050 struct btrfs_inode *inode,
1051 u64 inode_objectid, u64 parent_objectid,
1052 u64 ref_index, char *name, int namelen,
1057 int victim_name_len;
1058 struct extent_buffer *leaf;
1059 struct btrfs_dir_item *di;
1060 struct btrfs_key search_key;
1061 struct btrfs_inode_extref *extref;
1064 /* Search old style refs */
1065 search_key.objectid = inode_objectid;
1066 search_key.type = BTRFS_INODE_REF_KEY;
1067 search_key.offset = parent_objectid;
1068 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1070 struct btrfs_inode_ref *victim_ref;
1072 unsigned long ptr_end;
1074 leaf = path->nodes[0];
1076 /* are we trying to overwrite a back ref for the root directory
1077 * if so, just jump out, we're done
1079 if (search_key.objectid == search_key.offset)
1082 /* check all the names in this back reference to see
1083 * if they are in the log. if so, we allow them to stay
1084 * otherwise they must be unlinked as a conflict
1086 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1087 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1088 while (ptr < ptr_end) {
1089 victim_ref = (struct btrfs_inode_ref *)ptr;
1090 victim_name_len = btrfs_inode_ref_name_len(leaf,
1092 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1096 read_extent_buffer(leaf, victim_name,
1097 (unsigned long)(victim_ref + 1),
1100 ret = backref_in_log(log_root, &search_key,
1101 parent_objectid, victim_name,
1107 inc_nlink(&inode->vfs_inode);
1108 btrfs_release_path(path);
1110 ret = unlink_inode_for_log_replay(trans, dir, inode,
1111 victim_name, victim_name_len);
1120 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1124 * NOTE: we have searched root tree and checked the
1125 * corresponding ref, it does not need to check again.
1129 btrfs_release_path(path);
1131 /* Same search but for extended refs */
1132 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1133 inode_objectid, parent_objectid, 0,
1135 if (IS_ERR(extref)) {
1136 return PTR_ERR(extref);
1137 } else if (extref) {
1141 struct inode *victim_parent;
1143 leaf = path->nodes[0];
1145 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1146 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1148 while (cur_offset < item_size) {
1149 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1151 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1153 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1156 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1159 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1162 search_key.objectid = inode_objectid;
1163 search_key.type = BTRFS_INODE_EXTREF_KEY;
1164 search_key.offset = btrfs_extref_hash(parent_objectid,
1167 ret = backref_in_log(log_root, &search_key,
1168 parent_objectid, victim_name,
1175 victim_parent = read_one_inode(root,
1177 if (victim_parent) {
1178 inc_nlink(&inode->vfs_inode);
1179 btrfs_release_path(path);
1181 ret = unlink_inode_for_log_replay(trans,
1182 BTRFS_I(victim_parent),
1187 iput(victim_parent);
1196 cur_offset += victim_name_len + sizeof(*extref);
1200 btrfs_release_path(path);
1202 /* look for a conflicting sequence number */
1203 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1204 ref_index, name, namelen, 0);
1208 ret = drop_one_dir_item(trans, root, path, dir, di);
1212 btrfs_release_path(path);
1214 /* look for a conflicting name */
1215 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1220 ret = drop_one_dir_item(trans, root, path, dir, di);
1224 btrfs_release_path(path);
1229 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1230 u32 *namelen, char **name, u64 *index,
1231 u64 *parent_objectid)
1233 struct btrfs_inode_extref *extref;
1235 extref = (struct btrfs_inode_extref *)ref_ptr;
1237 *namelen = btrfs_inode_extref_name_len(eb, extref);
1238 *name = kmalloc(*namelen, GFP_NOFS);
1242 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1246 *index = btrfs_inode_extref_index(eb, extref);
1247 if (parent_objectid)
1248 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1253 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1254 u32 *namelen, char **name, u64 *index)
1256 struct btrfs_inode_ref *ref;
1258 ref = (struct btrfs_inode_ref *)ref_ptr;
1260 *namelen = btrfs_inode_ref_name_len(eb, ref);
1261 *name = kmalloc(*namelen, GFP_NOFS);
1265 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1268 *index = btrfs_inode_ref_index(eb, ref);
1274 * Take an inode reference item from the log tree and iterate all names from the
1275 * inode reference item in the subvolume tree with the same key (if it exists).
1276 * For any name that is not in the inode reference item from the log tree, do a
1277 * proper unlink of that name (that is, remove its entry from the inode
1278 * reference item and both dir index keys).
1280 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1281 struct btrfs_root *root,
1282 struct btrfs_path *path,
1283 struct btrfs_inode *inode,
1284 struct extent_buffer *log_eb,
1286 struct btrfs_key *key)
1289 unsigned long ref_ptr;
1290 unsigned long ref_end;
1291 struct extent_buffer *eb;
1294 btrfs_release_path(path);
1295 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1303 eb = path->nodes[0];
1304 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1305 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1306 while (ref_ptr < ref_end) {
1311 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1312 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1315 parent_id = key->offset;
1316 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1322 if (key->type == BTRFS_INODE_EXTREF_KEY)
1323 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1327 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1333 btrfs_release_path(path);
1334 dir = read_one_inode(root, parent_id);
1340 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1341 inode, name, namelen);
1351 if (key->type == BTRFS_INODE_EXTREF_KEY)
1352 ref_ptr += sizeof(struct btrfs_inode_extref);
1354 ref_ptr += sizeof(struct btrfs_inode_ref);
1358 btrfs_release_path(path);
1362 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1363 const u8 ref_type, const char *name,
1366 struct btrfs_key key;
1367 struct btrfs_path *path;
1368 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1371 path = btrfs_alloc_path();
1375 key.objectid = btrfs_ino(BTRFS_I(inode));
1376 key.type = ref_type;
1377 if (key.type == BTRFS_INODE_REF_KEY)
1378 key.offset = parent_id;
1380 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1382 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1389 if (key.type == BTRFS_INODE_EXTREF_KEY)
1390 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1391 path->slots[0], parent_id, name, namelen);
1393 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1397 btrfs_free_path(path);
1401 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1402 struct inode *dir, struct inode *inode, const char *name,
1403 int namelen, u64 ref_index)
1405 struct btrfs_dir_item *dir_item;
1406 struct btrfs_key key;
1407 struct btrfs_path *path;
1408 struct inode *other_inode = NULL;
1411 path = btrfs_alloc_path();
1415 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1416 btrfs_ino(BTRFS_I(dir)),
1419 btrfs_release_path(path);
1421 } else if (IS_ERR(dir_item)) {
1422 ret = PTR_ERR(dir_item);
1427 * Our inode's dentry collides with the dentry of another inode which is
1428 * in the log but not yet processed since it has a higher inode number.
1429 * So delete that other dentry.
1431 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1432 btrfs_release_path(path);
1433 other_inode = read_one_inode(root, key.objectid);
1438 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(other_inode),
1443 * If we dropped the link count to 0, bump it so that later the iput()
1444 * on the inode will not free it. We will fixup the link count later.
1446 if (other_inode->i_nlink == 0)
1447 set_nlink(other_inode, 1);
1449 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1450 name, namelen, 0, ref_index);
1453 btrfs_free_path(path);
1459 * replay one inode back reference item found in the log tree.
1460 * eb, slot and key refer to the buffer and key found in the log tree.
1461 * root is the destination we are replaying into, and path is for temp
1462 * use by this function. (it should be released on return).
1464 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1465 struct btrfs_root *root,
1466 struct btrfs_root *log,
1467 struct btrfs_path *path,
1468 struct extent_buffer *eb, int slot,
1469 struct btrfs_key *key)
1471 struct inode *dir = NULL;
1472 struct inode *inode = NULL;
1473 unsigned long ref_ptr;
1474 unsigned long ref_end;
1478 int search_done = 0;
1479 int log_ref_ver = 0;
1480 u64 parent_objectid;
1483 int ref_struct_size;
1485 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1486 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1488 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1489 struct btrfs_inode_extref *r;
1491 ref_struct_size = sizeof(struct btrfs_inode_extref);
1493 r = (struct btrfs_inode_extref *)ref_ptr;
1494 parent_objectid = btrfs_inode_extref_parent(eb, r);
1496 ref_struct_size = sizeof(struct btrfs_inode_ref);
1497 parent_objectid = key->offset;
1499 inode_objectid = key->objectid;
1502 * it is possible that we didn't log all the parent directories
1503 * for a given inode. If we don't find the dir, just don't
1504 * copy the back ref in. The link count fixup code will take
1507 dir = read_one_inode(root, parent_objectid);
1513 inode = read_one_inode(root, inode_objectid);
1519 while (ref_ptr < ref_end) {
1521 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1522 &ref_index, &parent_objectid);
1524 * parent object can change from one array
1528 dir = read_one_inode(root, parent_objectid);
1534 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1540 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1541 btrfs_ino(BTRFS_I(inode)), ref_index,
1545 } else if (ret == 0) {
1547 * look for a conflicting back reference in the
1548 * metadata. if we find one we have to unlink that name
1549 * of the file before we add our new link. Later on, we
1550 * overwrite any existing back reference, and we don't
1551 * want to create dangling pointers in the directory.
1555 ret = __add_inode_ref(trans, root, path, log,
1560 ref_index, name, namelen,
1570 * If a reference item already exists for this inode
1571 * with the same parent and name, but different index,
1572 * drop it and the corresponding directory index entries
1573 * from the parent before adding the new reference item
1574 * and dir index entries, otherwise we would fail with
1575 * -EEXIST returned from btrfs_add_link() below.
1577 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1580 ret = unlink_inode_for_log_replay(trans,
1585 * If we dropped the link count to 0, bump it so
1586 * that later the iput() on the inode will not
1587 * free it. We will fixup the link count later.
1589 if (!ret && inode->i_nlink == 0)
1590 set_nlink(inode, 1);
1595 /* insert our name */
1596 ret = add_link(trans, root, dir, inode, name, namelen,
1601 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1605 /* Else, ret == 1, we already have a perfect match, we're done. */
1607 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1617 * Before we overwrite the inode reference item in the subvolume tree
1618 * with the item from the log tree, we must unlink all names from the
1619 * parent directory that are in the subvolume's tree inode reference
1620 * item, otherwise we end up with an inconsistent subvolume tree where
1621 * dir index entries exist for a name but there is no inode reference
1622 * item with the same name.
1624 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1629 /* finally write the back reference in the inode */
1630 ret = overwrite_item(trans, root, path, eb, slot, key);
1632 btrfs_release_path(path);
1639 static int count_inode_extrefs(struct btrfs_root *root,
1640 struct btrfs_inode *inode, struct btrfs_path *path)
1644 unsigned int nlink = 0;
1647 u64 inode_objectid = btrfs_ino(inode);
1650 struct btrfs_inode_extref *extref;
1651 struct extent_buffer *leaf;
1654 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1659 leaf = path->nodes[0];
1660 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1661 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1664 while (cur_offset < item_size) {
1665 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1666 name_len = btrfs_inode_extref_name_len(leaf, extref);
1670 cur_offset += name_len + sizeof(*extref);
1674 btrfs_release_path(path);
1676 btrfs_release_path(path);
1678 if (ret < 0 && ret != -ENOENT)
1683 static int count_inode_refs(struct btrfs_root *root,
1684 struct btrfs_inode *inode, struct btrfs_path *path)
1687 struct btrfs_key key;
1688 unsigned int nlink = 0;
1690 unsigned long ptr_end;
1692 u64 ino = btrfs_ino(inode);
1695 key.type = BTRFS_INODE_REF_KEY;
1696 key.offset = (u64)-1;
1699 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1703 if (path->slots[0] == 0)
1708 btrfs_item_key_to_cpu(path->nodes[0], &key,
1710 if (key.objectid != ino ||
1711 key.type != BTRFS_INODE_REF_KEY)
1713 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1714 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1716 while (ptr < ptr_end) {
1717 struct btrfs_inode_ref *ref;
1719 ref = (struct btrfs_inode_ref *)ptr;
1720 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1722 ptr = (unsigned long)(ref + 1) + name_len;
1726 if (key.offset == 0)
1728 if (path->slots[0] > 0) {
1733 btrfs_release_path(path);
1735 btrfs_release_path(path);
1741 * There are a few corners where the link count of the file can't
1742 * be properly maintained during replay. So, instead of adding
1743 * lots of complexity to the log code, we just scan the backrefs
1744 * for any file that has been through replay.
1746 * The scan will update the link count on the inode to reflect the
1747 * number of back refs found. If it goes down to zero, the iput
1748 * will free the inode.
1750 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1751 struct btrfs_root *root,
1752 struct inode *inode)
1754 struct btrfs_path *path;
1757 u64 ino = btrfs_ino(BTRFS_I(inode));
1759 path = btrfs_alloc_path();
1763 ret = count_inode_refs(root, BTRFS_I(inode), path);
1769 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1777 if (nlink != inode->i_nlink) {
1778 set_nlink(inode, nlink);
1779 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1783 BTRFS_I(inode)->index_cnt = (u64)-1;
1785 if (inode->i_nlink == 0) {
1786 if (S_ISDIR(inode->i_mode)) {
1787 ret = replay_dir_deletes(trans, root, NULL, path,
1792 ret = btrfs_insert_orphan_item(trans, root, ino);
1798 btrfs_free_path(path);
1802 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path)
1807 struct btrfs_key key;
1808 struct inode *inode;
1810 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1811 key.type = BTRFS_ORPHAN_ITEM_KEY;
1812 key.offset = (u64)-1;
1814 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1820 if (path->slots[0] == 0)
1825 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1826 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1827 key.type != BTRFS_ORPHAN_ITEM_KEY)
1830 ret = btrfs_del_item(trans, root, path);
1834 btrfs_release_path(path);
1835 inode = read_one_inode(root, key.offset);
1841 ret = fixup_inode_link_count(trans, root, inode);
1847 * fixup on a directory may create new entries,
1848 * make sure we always look for the highset possible
1851 key.offset = (u64)-1;
1853 btrfs_release_path(path);
1859 * record a given inode in the fixup dir so we can check its link
1860 * count when replay is done. The link count is incremented here
1861 * so the inode won't go away until we check it
1863 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1864 struct btrfs_root *root,
1865 struct btrfs_path *path,
1868 struct btrfs_key key;
1870 struct inode *inode;
1872 inode = read_one_inode(root, objectid);
1876 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1877 key.type = BTRFS_ORPHAN_ITEM_KEY;
1878 key.offset = objectid;
1880 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1882 btrfs_release_path(path);
1884 if (!inode->i_nlink)
1885 set_nlink(inode, 1);
1888 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1889 } else if (ret == -EEXIST) {
1898 * when replaying the log for a directory, we only insert names
1899 * for inodes that actually exist. This means an fsync on a directory
1900 * does not implicitly fsync all the new files in it
1902 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1903 struct btrfs_root *root,
1904 u64 dirid, u64 index,
1905 char *name, int name_len,
1906 struct btrfs_key *location)
1908 struct inode *inode;
1912 inode = read_one_inode(root, location->objectid);
1916 dir = read_one_inode(root, dirid);
1922 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1923 name_len, 1, index);
1925 /* FIXME, put inode into FIXUP list */
1933 * take a single entry in a log directory item and replay it into
1936 * if a conflicting item exists in the subdirectory already,
1937 * the inode it points to is unlinked and put into the link count
1940 * If a name from the log points to a file or directory that does
1941 * not exist in the FS, it is skipped. fsyncs on directories
1942 * do not force down inodes inside that directory, just changes to the
1943 * names or unlinks in a directory.
1945 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1946 * non-existing inode) and 1 if the name was replayed.
1948 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct btrfs_path *path,
1951 struct extent_buffer *eb,
1952 struct btrfs_dir_item *di,
1953 struct btrfs_key *key)
1957 struct btrfs_dir_item *dst_di;
1958 struct btrfs_key found_key;
1959 struct btrfs_key log_key;
1964 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1965 bool name_added = false;
1967 dir = read_one_inode(root, key->objectid);
1971 name_len = btrfs_dir_name_len(eb, di);
1972 name = kmalloc(name_len, GFP_NOFS);
1978 log_type = btrfs_dir_type(eb, di);
1979 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1982 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1983 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1984 btrfs_release_path(path);
1987 exists = (ret == 0);
1990 if (key->type == BTRFS_DIR_ITEM_KEY) {
1991 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1993 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1994 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
2004 if (IS_ERR(dst_di)) {
2005 ret = PTR_ERR(dst_di);
2007 } else if (!dst_di) {
2008 /* we need a sequence number to insert, so we only
2009 * do inserts for the BTRFS_DIR_INDEX_KEY types
2011 if (key->type != BTRFS_DIR_INDEX_KEY)
2016 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
2017 /* the existing item matches the logged item */
2018 if (found_key.objectid == log_key.objectid &&
2019 found_key.type == log_key.type &&
2020 found_key.offset == log_key.offset &&
2021 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
2022 update_size = false;
2027 * don't drop the conflicting directory entry if the inode
2028 * for the new entry doesn't exist
2033 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2037 if (key->type == BTRFS_DIR_INDEX_KEY)
2040 btrfs_release_path(path);
2041 if (!ret && update_size) {
2042 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2043 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
2047 if (!ret && name_added)
2053 * Check if the inode reference exists in the log for the given name,
2054 * inode and parent inode
2056 found_key.objectid = log_key.objectid;
2057 found_key.type = BTRFS_INODE_REF_KEY;
2058 found_key.offset = key->objectid;
2059 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2063 /* The dentry will be added later. */
2065 update_size = false;
2069 found_key.objectid = log_key.objectid;
2070 found_key.type = BTRFS_INODE_EXTREF_KEY;
2071 found_key.offset = key->objectid;
2072 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2077 /* The dentry will be added later. */
2079 update_size = false;
2082 btrfs_release_path(path);
2083 ret = insert_one_name(trans, root, key->objectid, key->offset,
2084 name, name_len, &log_key);
2085 if (ret && ret != -ENOENT && ret != -EEXIST)
2089 update_size = false;
2095 * find all the names in a directory item and reconcile them into
2096 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2097 * one name in a directory item, but the same code gets used for
2098 * both directory index types
2100 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2101 struct btrfs_root *root,
2102 struct btrfs_path *path,
2103 struct extent_buffer *eb, int slot,
2104 struct btrfs_key *key)
2107 u32 item_size = btrfs_item_size_nr(eb, slot);
2108 struct btrfs_dir_item *di;
2111 unsigned long ptr_end;
2112 struct btrfs_path *fixup_path = NULL;
2114 ptr = btrfs_item_ptr_offset(eb, slot);
2115 ptr_end = ptr + item_size;
2116 while (ptr < ptr_end) {
2117 di = (struct btrfs_dir_item *)ptr;
2118 name_len = btrfs_dir_name_len(eb, di);
2119 ret = replay_one_name(trans, root, path, eb, di, key);
2122 ptr = (unsigned long)(di + 1);
2126 * If this entry refers to a non-directory (directories can not
2127 * have a link count > 1) and it was added in the transaction
2128 * that was not committed, make sure we fixup the link count of
2129 * the inode it the entry points to. Otherwise something like
2130 * the following would result in a directory pointing to an
2131 * inode with a wrong link that does not account for this dir
2139 * ln testdir/bar testdir/bar_link
2140 * ln testdir/foo testdir/foo_link
2141 * xfs_io -c "fsync" testdir/bar
2145 * mount fs, log replay happens
2147 * File foo would remain with a link count of 1 when it has two
2148 * entries pointing to it in the directory testdir. This would
2149 * make it impossible to ever delete the parent directory has
2150 * it would result in stale dentries that can never be deleted.
2152 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2153 struct btrfs_key di_key;
2156 fixup_path = btrfs_alloc_path();
2163 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2164 ret = link_to_fixup_dir(trans, root, fixup_path,
2171 btrfs_free_path(fixup_path);
2176 * directory replay has two parts. There are the standard directory
2177 * items in the log copied from the subvolume, and range items
2178 * created in the log while the subvolume was logged.
2180 * The range items tell us which parts of the key space the log
2181 * is authoritative for. During replay, if a key in the subvolume
2182 * directory is in a logged range item, but not actually in the log
2183 * that means it was deleted from the directory before the fsync
2184 * and should be removed.
2186 static noinline int find_dir_range(struct btrfs_root *root,
2187 struct btrfs_path *path,
2189 u64 *start_ret, u64 *end_ret)
2191 struct btrfs_key key;
2193 struct btrfs_dir_log_item *item;
2197 if (*start_ret == (u64)-1)
2200 key.objectid = dirid;
2201 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2202 key.offset = *start_ret;
2204 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2208 if (path->slots[0] == 0)
2213 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2215 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || 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);
2223 if (*start_ret >= key.offset && *start_ret <= found_end) {
2225 *start_ret = key.offset;
2226 *end_ret = found_end;
2231 /* check the next slot in the tree to see if it is a valid item */
2232 nritems = btrfs_header_nritems(path->nodes[0]);
2234 if (path->slots[0] >= nritems) {
2235 ret = btrfs_next_leaf(root, path);
2240 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2242 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2246 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2247 struct btrfs_dir_log_item);
2248 found_end = btrfs_dir_log_end(path->nodes[0], item);
2249 *start_ret = key.offset;
2250 *end_ret = found_end;
2253 btrfs_release_path(path);
2258 * this looks for a given directory item in the log. If the directory
2259 * item is not in the log, the item is removed and the inode it points
2262 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2263 struct btrfs_root *root,
2264 struct btrfs_root *log,
2265 struct btrfs_path *path,
2266 struct btrfs_path *log_path,
2268 struct btrfs_key *dir_key)
2271 struct extent_buffer *eb;
2273 struct btrfs_dir_item *di;
2276 struct inode *inode = NULL;
2277 struct btrfs_key location;
2280 * Currenly we only log dir index keys. Even if we replay a log created
2281 * by an older kernel that logged both dir index and dir item keys, all
2282 * we need to do is process the dir index keys, we (and our caller) can
2283 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2285 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2287 eb = path->nodes[0];
2288 slot = path->slots[0];
2289 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2290 name_len = btrfs_dir_name_len(eb, di);
2291 name = kmalloc(name_len, GFP_NOFS);
2297 read_extent_buffer(eb, name, (unsigned long)(di + 1), name_len);
2300 struct btrfs_dir_item *log_di;
2302 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2306 if (IS_ERR(log_di)) {
2307 ret = PTR_ERR(log_di);
2309 } else if (log_di) {
2310 /* The dentry exists in the log, we have nothing to do. */
2316 btrfs_dir_item_key_to_cpu(eb, di, &location);
2317 btrfs_release_path(path);
2318 btrfs_release_path(log_path);
2319 inode = read_one_inode(root, location.objectid);
2325 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2330 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2333 * Unlike dir item keys, dir index keys can only have one name (entry) in
2334 * them, as there are no key collisions since each key has a unique offset
2335 * (an index number), so we're done.
2338 btrfs_release_path(path);
2339 btrfs_release_path(log_path);
2345 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2346 struct btrfs_root *root,
2347 struct btrfs_root *log,
2348 struct btrfs_path *path,
2351 struct btrfs_key search_key;
2352 struct btrfs_path *log_path;
2357 log_path = btrfs_alloc_path();
2361 search_key.objectid = ino;
2362 search_key.type = BTRFS_XATTR_ITEM_KEY;
2363 search_key.offset = 0;
2365 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2369 nritems = btrfs_header_nritems(path->nodes[0]);
2370 for (i = path->slots[0]; i < nritems; i++) {
2371 struct btrfs_key key;
2372 struct btrfs_dir_item *di;
2373 struct btrfs_dir_item *log_di;
2377 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2378 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2383 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2384 total_size = btrfs_item_size_nr(path->nodes[0], i);
2386 while (cur < total_size) {
2387 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2388 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2389 u32 this_len = sizeof(*di) + name_len + data_len;
2392 name = kmalloc(name_len, GFP_NOFS);
2397 read_extent_buffer(path->nodes[0], name,
2398 (unsigned long)(di + 1), name_len);
2400 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2402 btrfs_release_path(log_path);
2404 /* Doesn't exist in log tree, so delete it. */
2405 btrfs_release_path(path);
2406 di = btrfs_lookup_xattr(trans, root, path, ino,
2407 name, name_len, -1);
2414 ret = btrfs_delete_one_dir_name(trans, root,
2418 btrfs_release_path(path);
2423 if (IS_ERR(log_di)) {
2424 ret = PTR_ERR(log_di);
2428 di = (struct btrfs_dir_item *)((char *)di + this_len);
2431 ret = btrfs_next_leaf(root, path);
2437 btrfs_free_path(log_path);
2438 btrfs_release_path(path);
2444 * deletion replay happens before we copy any new directory items
2445 * out of the log or out of backreferences from inodes. It
2446 * scans the log to find ranges of keys that log is authoritative for,
2447 * and then scans the directory to find items in those ranges that are
2448 * not present in the log.
2450 * Anything we don't find in the log is unlinked and removed from the
2453 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2454 struct btrfs_root *root,
2455 struct btrfs_root *log,
2456 struct btrfs_path *path,
2457 u64 dirid, int del_all)
2462 struct btrfs_key dir_key;
2463 struct btrfs_key found_key;
2464 struct btrfs_path *log_path;
2467 dir_key.objectid = dirid;
2468 dir_key.type = BTRFS_DIR_INDEX_KEY;
2469 log_path = btrfs_alloc_path();
2473 dir = read_one_inode(root, dirid);
2474 /* it isn't an error if the inode isn't there, that can happen
2475 * because we replay the deletes before we copy in the inode item
2479 btrfs_free_path(log_path);
2487 range_end = (u64)-1;
2489 ret = find_dir_range(log, path, dirid,
2490 &range_start, &range_end);
2497 dir_key.offset = range_start;
2500 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2505 nritems = btrfs_header_nritems(path->nodes[0]);
2506 if (path->slots[0] >= nritems) {
2507 ret = btrfs_next_leaf(root, path);
2513 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2515 if (found_key.objectid != dirid ||
2516 found_key.type != dir_key.type) {
2521 if (found_key.offset > range_end)
2524 ret = check_item_in_log(trans, root, log, path,
2529 if (found_key.offset == (u64)-1)
2531 dir_key.offset = found_key.offset + 1;
2533 btrfs_release_path(path);
2534 if (range_end == (u64)-1)
2536 range_start = range_end + 1;
2540 btrfs_release_path(path);
2541 btrfs_free_path(log_path);
2547 * the process_func used to replay items from the log tree. This
2548 * gets called in two different stages. The first stage just looks
2549 * for inodes and makes sure they are all copied into the subvolume.
2551 * The second stage copies all the other item types from the log into
2552 * the subvolume. The two stage approach is slower, but gets rid of
2553 * lots of complexity around inodes referencing other inodes that exist
2554 * only in the log (references come from either directory items or inode
2557 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2558 struct walk_control *wc, u64 gen, int level)
2561 struct btrfs_path *path;
2562 struct btrfs_root *root = wc->replay_dest;
2563 struct btrfs_key key;
2567 ret = btrfs_read_buffer(eb, gen, level, NULL);
2571 level = btrfs_header_level(eb);
2576 path = btrfs_alloc_path();
2580 nritems = btrfs_header_nritems(eb);
2581 for (i = 0; i < nritems; i++) {
2582 btrfs_item_key_to_cpu(eb, &key, i);
2584 /* inode keys are done during the first stage */
2585 if (key.type == BTRFS_INODE_ITEM_KEY &&
2586 wc->stage == LOG_WALK_REPLAY_INODES) {
2587 struct btrfs_inode_item *inode_item;
2590 inode_item = btrfs_item_ptr(eb, i,
2591 struct btrfs_inode_item);
2593 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2594 * and never got linked before the fsync, skip it, as
2595 * replaying it is pointless since it would be deleted
2596 * later. We skip logging tmpfiles, but it's always
2597 * possible we are replaying a log created with a kernel
2598 * that used to log tmpfiles.
2600 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2601 wc->ignore_cur_inode = true;
2604 wc->ignore_cur_inode = false;
2606 ret = replay_xattr_deletes(wc->trans, root, log,
2607 path, key.objectid);
2610 mode = btrfs_inode_mode(eb, inode_item);
2611 if (S_ISDIR(mode)) {
2612 ret = replay_dir_deletes(wc->trans,
2613 root, log, path, key.objectid, 0);
2617 ret = overwrite_item(wc->trans, root, path,
2623 * Before replaying extents, truncate the inode to its
2624 * size. We need to do it now and not after log replay
2625 * because before an fsync we can have prealloc extents
2626 * added beyond the inode's i_size. If we did it after,
2627 * through orphan cleanup for example, we would drop
2628 * those prealloc extents just after replaying them.
2630 if (S_ISREG(mode)) {
2631 struct btrfs_drop_extents_args drop_args = { 0 };
2632 struct inode *inode;
2635 inode = read_one_inode(root, key.objectid);
2640 from = ALIGN(i_size_read(inode),
2641 root->fs_info->sectorsize);
2642 drop_args.start = from;
2643 drop_args.end = (u64)-1;
2644 drop_args.drop_cache = true;
2645 ret = btrfs_drop_extents(wc->trans, root,
2649 inode_sub_bytes(inode,
2650 drop_args.bytes_found);
2651 /* Update the inode's nbytes. */
2652 ret = btrfs_update_inode(wc->trans,
2653 root, BTRFS_I(inode));
2660 ret = link_to_fixup_dir(wc->trans, root,
2661 path, key.objectid);
2666 if (wc->ignore_cur_inode)
2669 if (key.type == BTRFS_DIR_INDEX_KEY &&
2670 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2671 ret = replay_one_dir_item(wc->trans, root, path,
2677 if (wc->stage < LOG_WALK_REPLAY_ALL)
2680 /* these keys are simply copied */
2681 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2682 ret = overwrite_item(wc->trans, root, path,
2686 } else if (key.type == BTRFS_INODE_REF_KEY ||
2687 key.type == BTRFS_INODE_EXTREF_KEY) {
2688 ret = add_inode_ref(wc->trans, root, log, path,
2690 if (ret && ret != -ENOENT)
2693 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2694 ret = replay_one_extent(wc->trans, root, path,
2698 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2699 ret = replay_one_dir_item(wc->trans, root, path,
2705 btrfs_free_path(path);
2710 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2712 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2714 struct btrfs_block_group *cache;
2716 cache = btrfs_lookup_block_group(fs_info, start);
2718 btrfs_err(fs_info, "unable to find block group for %llu", start);
2722 spin_lock(&cache->space_info->lock);
2723 spin_lock(&cache->lock);
2724 cache->reserved -= fs_info->nodesize;
2725 cache->space_info->bytes_reserved -= fs_info->nodesize;
2726 spin_unlock(&cache->lock);
2727 spin_unlock(&cache->space_info->lock);
2729 btrfs_put_block_group(cache);
2732 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2733 struct btrfs_root *root,
2734 struct btrfs_path *path, int *level,
2735 struct walk_control *wc)
2737 struct btrfs_fs_info *fs_info = root->fs_info;
2740 struct extent_buffer *next;
2741 struct extent_buffer *cur;
2745 while (*level > 0) {
2746 struct btrfs_key first_key;
2748 cur = path->nodes[*level];
2750 WARN_ON(btrfs_header_level(cur) != *level);
2752 if (path->slots[*level] >=
2753 btrfs_header_nritems(cur))
2756 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2757 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2758 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2759 blocksize = fs_info->nodesize;
2761 next = btrfs_find_create_tree_block(fs_info, bytenr,
2762 btrfs_header_owner(cur),
2765 return PTR_ERR(next);
2768 ret = wc->process_func(root, next, wc, ptr_gen,
2771 free_extent_buffer(next);
2775 path->slots[*level]++;
2777 ret = btrfs_read_buffer(next, ptr_gen,
2778 *level - 1, &first_key);
2780 free_extent_buffer(next);
2785 btrfs_tree_lock(next);
2786 btrfs_clean_tree_block(next);
2787 btrfs_wait_tree_block_writeback(next);
2788 btrfs_tree_unlock(next);
2789 ret = btrfs_pin_reserved_extent(trans,
2792 free_extent_buffer(next);
2795 btrfs_redirty_list_add(
2796 trans->transaction, next);
2798 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2799 clear_extent_buffer_dirty(next);
2800 unaccount_log_buffer(fs_info, bytenr);
2803 free_extent_buffer(next);
2806 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2808 free_extent_buffer(next);
2812 if (path->nodes[*level-1])
2813 free_extent_buffer(path->nodes[*level-1]);
2814 path->nodes[*level-1] = next;
2815 *level = btrfs_header_level(next);
2816 path->slots[*level] = 0;
2819 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2825 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2826 struct btrfs_root *root,
2827 struct btrfs_path *path, int *level,
2828 struct walk_control *wc)
2830 struct btrfs_fs_info *fs_info = root->fs_info;
2835 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2836 slot = path->slots[i];
2837 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2840 WARN_ON(*level == 0);
2843 ret = wc->process_func(root, path->nodes[*level], wc,
2844 btrfs_header_generation(path->nodes[*level]),
2850 struct extent_buffer *next;
2852 next = path->nodes[*level];
2855 btrfs_tree_lock(next);
2856 btrfs_clean_tree_block(next);
2857 btrfs_wait_tree_block_writeback(next);
2858 btrfs_tree_unlock(next);
2859 ret = btrfs_pin_reserved_extent(trans,
2860 path->nodes[*level]->start,
2861 path->nodes[*level]->len);
2864 btrfs_redirty_list_add(trans->transaction,
2867 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2868 clear_extent_buffer_dirty(next);
2870 unaccount_log_buffer(fs_info,
2871 path->nodes[*level]->start);
2874 free_extent_buffer(path->nodes[*level]);
2875 path->nodes[*level] = NULL;
2883 * drop the reference count on the tree rooted at 'snap'. This traverses
2884 * the tree freeing any blocks that have a ref count of zero after being
2887 static int walk_log_tree(struct btrfs_trans_handle *trans,
2888 struct btrfs_root *log, struct walk_control *wc)
2890 struct btrfs_fs_info *fs_info = log->fs_info;
2894 struct btrfs_path *path;
2897 path = btrfs_alloc_path();
2901 level = btrfs_header_level(log->node);
2903 path->nodes[level] = log->node;
2904 atomic_inc(&log->node->refs);
2905 path->slots[level] = 0;
2908 wret = walk_down_log_tree(trans, log, path, &level, wc);
2916 wret = walk_up_log_tree(trans, log, path, &level, wc);
2925 /* was the root node processed? if not, catch it here */
2926 if (path->nodes[orig_level]) {
2927 ret = wc->process_func(log, path->nodes[orig_level], wc,
2928 btrfs_header_generation(path->nodes[orig_level]),
2933 struct extent_buffer *next;
2935 next = path->nodes[orig_level];
2938 btrfs_tree_lock(next);
2939 btrfs_clean_tree_block(next);
2940 btrfs_wait_tree_block_writeback(next);
2941 btrfs_tree_unlock(next);
2942 ret = btrfs_pin_reserved_extent(trans,
2943 next->start, next->len);
2946 btrfs_redirty_list_add(trans->transaction, next);
2948 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2949 clear_extent_buffer_dirty(next);
2950 unaccount_log_buffer(fs_info, next->start);
2956 btrfs_free_path(path);
2961 * helper function to update the item for a given subvolumes log root
2962 * in the tree of log roots
2964 static int update_log_root(struct btrfs_trans_handle *trans,
2965 struct btrfs_root *log,
2966 struct btrfs_root_item *root_item)
2968 struct btrfs_fs_info *fs_info = log->fs_info;
2971 if (log->log_transid == 1) {
2972 /* insert root item on the first sync */
2973 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2974 &log->root_key, root_item);
2976 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2977 &log->root_key, root_item);
2982 static void wait_log_commit(struct btrfs_root *root, int transid)
2985 int index = transid % 2;
2988 * we only allow two pending log transactions at a time,
2989 * so we know that if ours is more than 2 older than the
2990 * current transaction, we're done
2993 prepare_to_wait(&root->log_commit_wait[index],
2994 &wait, TASK_UNINTERRUPTIBLE);
2996 if (!(root->log_transid_committed < transid &&
2997 atomic_read(&root->log_commit[index])))
3000 mutex_unlock(&root->log_mutex);
3002 mutex_lock(&root->log_mutex);
3004 finish_wait(&root->log_commit_wait[index], &wait);
3007 static void wait_for_writer(struct btrfs_root *root)
3012 prepare_to_wait(&root->log_writer_wait, &wait,
3013 TASK_UNINTERRUPTIBLE);
3014 if (!atomic_read(&root->log_writers))
3017 mutex_unlock(&root->log_mutex);
3019 mutex_lock(&root->log_mutex);
3021 finish_wait(&root->log_writer_wait, &wait);
3024 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3025 struct btrfs_log_ctx *ctx)
3030 mutex_lock(&root->log_mutex);
3031 list_del_init(&ctx->list);
3032 mutex_unlock(&root->log_mutex);
3036 * Invoked in log mutex context, or be sure there is no other task which
3037 * can access the list.
3039 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3040 int index, int error)
3042 struct btrfs_log_ctx *ctx;
3043 struct btrfs_log_ctx *safe;
3045 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3046 list_del_init(&ctx->list);
3047 ctx->log_ret = error;
3052 * btrfs_sync_log does sends a given tree log down to the disk and
3053 * updates the super blocks to record it. When this call is done,
3054 * you know that any inodes previously logged are safely on disk only
3057 * Any other return value means you need to call btrfs_commit_transaction.
3058 * Some of the edge cases for fsyncing directories that have had unlinks
3059 * or renames done in the past mean that sometimes the only safe
3060 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3061 * that has happened.
3063 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3064 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3070 struct btrfs_fs_info *fs_info = root->fs_info;
3071 struct btrfs_root *log = root->log_root;
3072 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3073 struct btrfs_root_item new_root_item;
3074 int log_transid = 0;
3075 struct btrfs_log_ctx root_log_ctx;
3076 struct blk_plug plug;
3080 mutex_lock(&root->log_mutex);
3081 log_transid = ctx->log_transid;
3082 if (root->log_transid_committed >= log_transid) {
3083 mutex_unlock(&root->log_mutex);
3084 return ctx->log_ret;
3087 index1 = log_transid % 2;
3088 if (atomic_read(&root->log_commit[index1])) {
3089 wait_log_commit(root, log_transid);
3090 mutex_unlock(&root->log_mutex);
3091 return ctx->log_ret;
3093 ASSERT(log_transid == root->log_transid);
3094 atomic_set(&root->log_commit[index1], 1);
3096 /* wait for previous tree log sync to complete */
3097 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3098 wait_log_commit(root, log_transid - 1);
3101 int batch = atomic_read(&root->log_batch);
3102 /* when we're on an ssd, just kick the log commit out */
3103 if (!btrfs_test_opt(fs_info, SSD) &&
3104 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3105 mutex_unlock(&root->log_mutex);
3106 schedule_timeout_uninterruptible(1);
3107 mutex_lock(&root->log_mutex);
3109 wait_for_writer(root);
3110 if (batch == atomic_read(&root->log_batch))
3114 /* bail out if we need to do a full commit */
3115 if (btrfs_need_log_full_commit(trans)) {
3117 mutex_unlock(&root->log_mutex);
3121 if (log_transid % 2 == 0)
3122 mark = EXTENT_DIRTY;
3126 /* we start IO on all the marked extents here, but we don't actually
3127 * wait for them until later.
3129 blk_start_plug(&plug);
3130 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3132 * -EAGAIN happens when someone, e.g., a concurrent transaction
3133 * commit, writes a dirty extent in this tree-log commit. This
3134 * concurrent write will create a hole writing out the extents,
3135 * and we cannot proceed on a zoned filesystem, requiring
3136 * sequential writing. While we can bail out to a full commit
3137 * here, but we can continue hoping the concurrent writing fills
3140 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3143 blk_finish_plug(&plug);
3144 btrfs_abort_transaction(trans, ret);
3145 btrfs_set_log_full_commit(trans);
3146 mutex_unlock(&root->log_mutex);
3151 * We _must_ update under the root->log_mutex in order to make sure we
3152 * have a consistent view of the log root we are trying to commit at
3155 * We _must_ copy this into a local copy, because we are not holding the
3156 * log_root_tree->log_mutex yet. This is important because when we
3157 * commit the log_root_tree we must have a consistent view of the
3158 * log_root_tree when we update the super block to point at the
3159 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3160 * with the commit and possibly point at the new block which we may not
3163 btrfs_set_root_node(&log->root_item, log->node);
3164 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3166 root->log_transid++;
3167 log->log_transid = root->log_transid;
3168 root->log_start_pid = 0;
3170 * IO has been started, blocks of the log tree have WRITTEN flag set
3171 * in their headers. new modifications of the log will be written to
3172 * new positions. so it's safe to allow log writers to go in.
3174 mutex_unlock(&root->log_mutex);
3176 if (btrfs_is_zoned(fs_info)) {
3177 mutex_lock(&fs_info->tree_root->log_mutex);
3178 if (!log_root_tree->node) {
3179 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3181 mutex_unlock(&fs_info->tree_root->log_mutex);
3182 blk_finish_plug(&plug);
3186 mutex_unlock(&fs_info->tree_root->log_mutex);
3189 btrfs_init_log_ctx(&root_log_ctx, NULL);
3191 mutex_lock(&log_root_tree->log_mutex);
3193 index2 = log_root_tree->log_transid % 2;
3194 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3195 root_log_ctx.log_transid = log_root_tree->log_transid;
3198 * Now we are safe to update the log_root_tree because we're under the
3199 * log_mutex, and we're a current writer so we're holding the commit
3200 * open until we drop the log_mutex.
3202 ret = update_log_root(trans, log, &new_root_item);
3204 if (!list_empty(&root_log_ctx.list))
3205 list_del_init(&root_log_ctx.list);
3207 blk_finish_plug(&plug);
3208 btrfs_set_log_full_commit(trans);
3210 if (ret != -ENOSPC) {
3211 btrfs_abort_transaction(trans, ret);
3212 mutex_unlock(&log_root_tree->log_mutex);
3215 btrfs_wait_tree_log_extents(log, mark);
3216 mutex_unlock(&log_root_tree->log_mutex);
3221 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3222 blk_finish_plug(&plug);
3223 list_del_init(&root_log_ctx.list);
3224 mutex_unlock(&log_root_tree->log_mutex);
3225 ret = root_log_ctx.log_ret;
3229 index2 = root_log_ctx.log_transid % 2;
3230 if (atomic_read(&log_root_tree->log_commit[index2])) {
3231 blk_finish_plug(&plug);
3232 ret = btrfs_wait_tree_log_extents(log, mark);
3233 wait_log_commit(log_root_tree,
3234 root_log_ctx.log_transid);
3235 mutex_unlock(&log_root_tree->log_mutex);
3237 ret = root_log_ctx.log_ret;
3240 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3241 atomic_set(&log_root_tree->log_commit[index2], 1);
3243 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3244 wait_log_commit(log_root_tree,
3245 root_log_ctx.log_transid - 1);
3249 * now that we've moved on to the tree of log tree roots,
3250 * check the full commit flag again
3252 if (btrfs_need_log_full_commit(trans)) {
3253 blk_finish_plug(&plug);
3254 btrfs_wait_tree_log_extents(log, mark);
3255 mutex_unlock(&log_root_tree->log_mutex);
3257 goto out_wake_log_root;
3260 ret = btrfs_write_marked_extents(fs_info,
3261 &log_root_tree->dirty_log_pages,
3262 EXTENT_DIRTY | EXTENT_NEW);
3263 blk_finish_plug(&plug);
3265 * As described above, -EAGAIN indicates a hole in the extents. We
3266 * cannot wait for these write outs since the waiting cause a
3267 * deadlock. Bail out to the full commit instead.
3269 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3270 btrfs_set_log_full_commit(trans);
3271 btrfs_wait_tree_log_extents(log, mark);
3272 mutex_unlock(&log_root_tree->log_mutex);
3273 goto out_wake_log_root;
3275 btrfs_set_log_full_commit(trans);
3276 btrfs_abort_transaction(trans, ret);
3277 mutex_unlock(&log_root_tree->log_mutex);
3278 goto out_wake_log_root;
3280 ret = btrfs_wait_tree_log_extents(log, mark);
3282 ret = btrfs_wait_tree_log_extents(log_root_tree,
3283 EXTENT_NEW | EXTENT_DIRTY);
3285 btrfs_set_log_full_commit(trans);
3286 mutex_unlock(&log_root_tree->log_mutex);
3287 goto out_wake_log_root;
3290 log_root_start = log_root_tree->node->start;
3291 log_root_level = btrfs_header_level(log_root_tree->node);
3292 log_root_tree->log_transid++;
3293 mutex_unlock(&log_root_tree->log_mutex);
3296 * Here we are guaranteed that nobody is going to write the superblock
3297 * for the current transaction before us and that neither we do write
3298 * our superblock before the previous transaction finishes its commit
3299 * and writes its superblock, because:
3301 * 1) We are holding a handle on the current transaction, so no body
3302 * can commit it until we release the handle;
3304 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3305 * if the previous transaction is still committing, and hasn't yet
3306 * written its superblock, we wait for it to do it, because a
3307 * transaction commit acquires the tree_log_mutex when the commit
3308 * begins and releases it only after writing its superblock.
3310 mutex_lock(&fs_info->tree_log_mutex);
3313 * The previous transaction writeout phase could have failed, and thus
3314 * marked the fs in an error state. We must not commit here, as we
3315 * could have updated our generation in the super_for_commit and
3316 * writing the super here would result in transid mismatches. If there
3317 * is an error here just bail.
3319 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321 btrfs_set_log_full_commit(trans);
3322 btrfs_abort_transaction(trans, ret);
3323 mutex_unlock(&fs_info->tree_log_mutex);
3324 goto out_wake_log_root;
3327 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3328 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3329 ret = write_all_supers(fs_info, 1);
3330 mutex_unlock(&fs_info->tree_log_mutex);
3332 btrfs_set_log_full_commit(trans);
3333 btrfs_abort_transaction(trans, ret);
3334 goto out_wake_log_root;
3338 * We know there can only be one task here, since we have not yet set
3339 * root->log_commit[index1] to 0 and any task attempting to sync the
3340 * log must wait for the previous log transaction to commit if it's
3341 * still in progress or wait for the current log transaction commit if
3342 * someone else already started it. We use <= and not < because the
3343 * first log transaction has an ID of 0.
3345 ASSERT(root->last_log_commit <= log_transid);
3346 root->last_log_commit = log_transid;
3349 mutex_lock(&log_root_tree->log_mutex);
3350 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3352 log_root_tree->log_transid_committed++;
3353 atomic_set(&log_root_tree->log_commit[index2], 0);
3354 mutex_unlock(&log_root_tree->log_mutex);
3357 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3358 * all the updates above are seen by the woken threads. It might not be
3359 * necessary, but proving that seems to be hard.
3361 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3363 mutex_lock(&root->log_mutex);
3364 btrfs_remove_all_log_ctxs(root, index1, ret);
3365 root->log_transid_committed++;
3366 atomic_set(&root->log_commit[index1], 0);
3367 mutex_unlock(&root->log_mutex);
3370 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3371 * all the updates above are seen by the woken threads. It might not be
3372 * necessary, but proving that seems to be hard.
3374 cond_wake_up(&root->log_commit_wait[index1]);
3378 static void free_log_tree(struct btrfs_trans_handle *trans,
3379 struct btrfs_root *log)
3382 struct walk_control wc = {
3384 .process_func = process_one_buffer
3388 ret = walk_log_tree(trans, log, &wc);
3391 * We weren't able to traverse the entire log tree, the
3392 * typical scenario is getting an -EIO when reading an
3393 * extent buffer of the tree, due to a previous writeback
3396 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3397 &log->fs_info->fs_state);
3400 * Some extent buffers of the log tree may still be dirty
3401 * and not yet written back to storage, because we may
3402 * have updates to a log tree without syncing a log tree,
3403 * such as during rename and link operations. So flush
3404 * them out and wait for their writeback to complete, so
3405 * that we properly cleanup their state and pages.
3407 btrfs_write_marked_extents(log->fs_info,
3408 &log->dirty_log_pages,
3409 EXTENT_DIRTY | EXTENT_NEW);
3410 btrfs_wait_tree_log_extents(log,
3411 EXTENT_DIRTY | EXTENT_NEW);
3414 btrfs_abort_transaction(trans, ret);
3416 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3420 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3421 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3422 extent_io_tree_release(&log->log_csum_range);
3424 btrfs_put_root(log);
3428 * free all the extents used by the tree log. This should be called
3429 * at commit time of the full transaction
3431 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3433 if (root->log_root) {
3434 free_log_tree(trans, root->log_root);
3435 root->log_root = NULL;
3436 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3441 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3442 struct btrfs_fs_info *fs_info)
3444 if (fs_info->log_root_tree) {
3445 free_log_tree(trans, fs_info->log_root_tree);
3446 fs_info->log_root_tree = NULL;
3447 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3453 * Check if an inode was logged in the current transaction. This may often
3454 * return some false positives, because logged_trans is an in memory only field,
3455 * not persisted anywhere. This is meant to be used in contexts where a false
3456 * positive has no functional consequences.
3458 static bool inode_logged(struct btrfs_trans_handle *trans,
3459 struct btrfs_inode *inode)
3461 if (inode->logged_trans == trans->transid)
3465 * The inode's logged_trans is always 0 when we load it (because it is
3466 * not persisted in the inode item or elsewhere). So if it is 0, the
3467 * inode was last modified in the current transaction then the inode may
3468 * have been logged before in the current transaction, then evicted and
3469 * loaded again in the current transaction - or may have never been logged
3470 * in the current transaction, but since we can not be sure, we have to
3471 * assume it was, otherwise our callers can leave an inconsistent log.
3473 if (inode->logged_trans == 0 &&
3474 inode->last_trans == trans->transid &&
3475 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3482 * If both a file and directory are logged, and unlinks or renames are
3483 * mixed in, we have a few interesting corners:
3485 * create file X in dir Y
3486 * link file X to X.link in dir Y
3488 * unlink file X but leave X.link
3491 * After a crash we would expect only X.link to exist. But file X
3492 * didn't get fsync'd again so the log has back refs for X and X.link.
3494 * We solve this by removing directory entries and inode backrefs from the
3495 * log when a file that was logged in the current transaction is
3496 * unlinked. Any later fsync will include the updated log entries, and
3497 * we'll be able to reconstruct the proper directory items from backrefs.
3499 * This optimizations allows us to avoid relogging the entire inode
3500 * or the entire directory.
3502 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3503 struct btrfs_root *root,
3504 const char *name, int name_len,
3505 struct btrfs_inode *dir, u64 index)
3507 struct btrfs_root *log;
3508 struct btrfs_dir_item *di;
3509 struct btrfs_path *path;
3512 u64 dir_ino = btrfs_ino(dir);
3514 if (!inode_logged(trans, dir))
3517 ret = join_running_log_trans(root);
3521 mutex_lock(&dir->log_mutex);
3523 log = root->log_root;
3524 path = btrfs_alloc_path();
3530 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3531 name, name_len, -1);
3537 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3543 btrfs_release_path(path);
3544 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3545 index, name, name_len, -1);
3551 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3559 * We do not need to update the size field of the directory's inode item
3560 * because on log replay we update the field to reflect all existing
3561 * entries in the directory (see overwrite_item()).
3564 btrfs_free_path(path);
3566 mutex_unlock(&dir->log_mutex);
3567 if (err == -ENOSPC) {
3568 btrfs_set_log_full_commit(trans);
3570 } else if (err < 0) {
3571 btrfs_abort_transaction(trans, err);
3574 btrfs_end_log_trans(root);
3579 /* see comments for btrfs_del_dir_entries_in_log */
3580 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3581 struct btrfs_root *root,
3582 const char *name, int name_len,
3583 struct btrfs_inode *inode, u64 dirid)
3585 struct btrfs_root *log;
3589 if (!inode_logged(trans, inode))
3592 ret = join_running_log_trans(root);
3595 log = root->log_root;
3596 mutex_lock(&inode->log_mutex);
3598 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3600 mutex_unlock(&inode->log_mutex);
3601 if (ret == -ENOSPC) {
3602 btrfs_set_log_full_commit(trans);
3604 } else if (ret < 0 && ret != -ENOENT)
3605 btrfs_abort_transaction(trans, ret);
3606 btrfs_end_log_trans(root);
3612 * creates a range item in the log for 'dirid'. first_offset and
3613 * last_offset tell us which parts of the key space the log should
3614 * be considered authoritative for.
3616 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *log,
3618 struct btrfs_path *path,
3619 int key_type, u64 dirid,
3620 u64 first_offset, u64 last_offset)
3623 struct btrfs_key key;
3624 struct btrfs_dir_log_item *item;
3626 key.objectid = dirid;
3627 key.offset = first_offset;
3628 if (key_type == BTRFS_DIR_ITEM_KEY)
3629 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3631 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3632 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3636 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3637 struct btrfs_dir_log_item);
3638 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3639 btrfs_mark_buffer_dirty(path->nodes[0]);
3640 btrfs_release_path(path);
3645 * log all the items included in the current transaction for a given
3646 * directory. This also creates the range items in the log tree required
3647 * to replay anything deleted before the fsync
3649 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3650 struct btrfs_root *root, struct btrfs_inode *inode,
3651 struct btrfs_path *path,
3652 struct btrfs_path *dst_path, int key_type,
3653 struct btrfs_log_ctx *ctx,
3654 u64 min_offset, u64 *last_offset_ret)
3656 struct btrfs_key min_key;
3657 struct btrfs_root *log = root->log_root;
3658 struct extent_buffer *src;
3663 u64 first_offset = min_offset;
3664 u64 last_offset = (u64)-1;
3665 u64 ino = btrfs_ino(inode);
3667 log = root->log_root;
3669 min_key.objectid = ino;
3670 min_key.type = key_type;
3671 min_key.offset = min_offset;
3673 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3676 * we didn't find anything from this transaction, see if there
3677 * is anything at all
3679 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3680 min_key.objectid = ino;
3681 min_key.type = key_type;
3682 min_key.offset = (u64)-1;
3683 btrfs_release_path(path);
3684 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3686 btrfs_release_path(path);
3689 ret = btrfs_previous_item(root, path, ino, key_type);
3691 /* if ret == 0 there are items for this type,
3692 * create a range to tell us the last key of this type.
3693 * otherwise, there are no items in this directory after
3694 * *min_offset, and we create a range to indicate that.
3697 struct btrfs_key tmp;
3698 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3700 if (key_type == tmp.type)
3701 first_offset = max(min_offset, tmp.offset) + 1;
3706 /* go backward to find any previous key */
3707 ret = btrfs_previous_item(root, path, ino, key_type);
3709 struct btrfs_key tmp;
3710 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3711 if (key_type == tmp.type) {
3712 first_offset = tmp.offset;
3713 ret = overwrite_item(trans, log, dst_path,
3714 path->nodes[0], path->slots[0],
3722 btrfs_release_path(path);
3725 * Find the first key from this transaction again. See the note for
3726 * log_new_dir_dentries, if we're logging a directory recursively we
3727 * won't be holding its i_mutex, which means we can modify the directory
3728 * while we're logging it. If we remove an entry between our first
3729 * search and this search we'll not find the key again and can just
3733 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3738 * we have a block from this transaction, log every item in it
3739 * from our directory
3742 struct btrfs_key tmp;
3743 src = path->nodes[0];
3744 nritems = btrfs_header_nritems(src);
3745 for (i = path->slots[0]; i < nritems; i++) {
3746 struct btrfs_dir_item *di;
3748 btrfs_item_key_to_cpu(src, &min_key, i);
3750 if (min_key.objectid != ino || min_key.type != key_type)
3753 if (need_resched()) {
3754 btrfs_release_path(path);
3759 ret = overwrite_item(trans, log, dst_path, src, i,
3767 * We must make sure that when we log a directory entry,
3768 * the corresponding inode, after log replay, has a
3769 * matching link count. For example:
3775 * xfs_io -c "fsync" mydir
3777 * <mount fs and log replay>
3779 * Would result in a fsync log that when replayed, our
3780 * file inode would have a link count of 1, but we get
3781 * two directory entries pointing to the same inode.
3782 * After removing one of the names, it would not be
3783 * possible to remove the other name, which resulted
3784 * always in stale file handle errors, and would not
3785 * be possible to rmdir the parent directory, since
3786 * its i_size could never decrement to the value
3787 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3789 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3790 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3792 (btrfs_dir_transid(src, di) == trans->transid ||
3793 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3794 tmp.type != BTRFS_ROOT_ITEM_KEY)
3795 ctx->log_new_dentries = true;
3797 path->slots[0] = nritems;
3800 * look ahead to the next item and see if it is also
3801 * from this directory and from this transaction
3803 ret = btrfs_next_leaf(root, path);
3806 last_offset = (u64)-1;
3811 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3812 if (tmp.objectid != ino || tmp.type != key_type) {
3813 last_offset = (u64)-1;
3816 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3817 ret = overwrite_item(trans, log, dst_path,
3818 path->nodes[0], path->slots[0],
3823 last_offset = tmp.offset;
3828 btrfs_release_path(path);
3829 btrfs_release_path(dst_path);
3832 *last_offset_ret = last_offset;
3834 * insert the log range keys to indicate where the log
3837 ret = insert_dir_log_key(trans, log, path, key_type,
3838 ino, first_offset, last_offset);
3846 * logging directories is very similar to logging inodes, We find all the items
3847 * from the current transaction and write them to the log.
3849 * The recovery code scans the directory in the subvolume, and if it finds a
3850 * key in the range logged that is not present in the log tree, then it means
3851 * that dir entry was unlinked during the transaction.
3853 * In order for that scan to work, we must include one key smaller than
3854 * the smallest logged by this transaction and one key larger than the largest
3855 * key logged by this transaction.
3857 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3858 struct btrfs_root *root, struct btrfs_inode *inode,
3859 struct btrfs_path *path,
3860 struct btrfs_path *dst_path,
3861 struct btrfs_log_ctx *ctx)
3866 int key_type = BTRFS_DIR_ITEM_KEY;
3872 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3873 ctx, min_key, &max_key);
3876 if (max_key == (u64)-1)
3878 min_key = max_key + 1;
3881 if (key_type == BTRFS_DIR_ITEM_KEY) {
3882 key_type = BTRFS_DIR_INDEX_KEY;
3889 * a helper function to drop items from the log before we relog an
3890 * inode. max_key_type indicates the highest item type to remove.
3891 * This cannot be run for file data extents because it does not
3892 * free the extents they point to.
3894 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3895 struct btrfs_root *log,
3896 struct btrfs_path *path,
3897 u64 objectid, int max_key_type)
3900 struct btrfs_key key;
3901 struct btrfs_key found_key;
3904 key.objectid = objectid;
3905 key.type = max_key_type;
3906 key.offset = (u64)-1;
3909 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3910 BUG_ON(ret == 0); /* Logic error */
3914 if (path->slots[0] == 0)
3918 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3921 if (found_key.objectid != objectid)
3924 found_key.offset = 0;
3926 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3930 ret = btrfs_del_items(trans, log, path, start_slot,
3931 path->slots[0] - start_slot + 1);
3933 * If start slot isn't 0 then we don't need to re-search, we've
3934 * found the last guy with the objectid in this tree.
3936 if (ret || start_slot != 0)
3938 btrfs_release_path(path);
3940 btrfs_release_path(path);
3946 static void fill_inode_item(struct btrfs_trans_handle *trans,
3947 struct extent_buffer *leaf,
3948 struct btrfs_inode_item *item,
3949 struct inode *inode, int log_inode_only,
3952 struct btrfs_map_token token;
3955 btrfs_init_map_token(&token, leaf);
3957 if (log_inode_only) {
3958 /* set the generation to zero so the recover code
3959 * can tell the difference between an logging
3960 * just to say 'this inode exists' and a logging
3961 * to say 'update this inode with these values'
3963 btrfs_set_token_inode_generation(&token, item, 0);
3964 btrfs_set_token_inode_size(&token, item, logged_isize);
3966 btrfs_set_token_inode_generation(&token, item,
3967 BTRFS_I(inode)->generation);
3968 btrfs_set_token_inode_size(&token, item, inode->i_size);
3971 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3972 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3973 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3974 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3976 btrfs_set_token_timespec_sec(&token, &item->atime,
3977 inode->i_atime.tv_sec);
3978 btrfs_set_token_timespec_nsec(&token, &item->atime,
3979 inode->i_atime.tv_nsec);
3981 btrfs_set_token_timespec_sec(&token, &item->mtime,
3982 inode->i_mtime.tv_sec);
3983 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3984 inode->i_mtime.tv_nsec);
3986 btrfs_set_token_timespec_sec(&token, &item->ctime,
3987 inode->i_ctime.tv_sec);
3988 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3989 inode->i_ctime.tv_nsec);
3992 * We do not need to set the nbytes field, in fact during a fast fsync
3993 * its value may not even be correct, since a fast fsync does not wait
3994 * for ordered extent completion, which is where we update nbytes, it
3995 * only waits for writeback to complete. During log replay as we find
3996 * file extent items and replay them, we adjust the nbytes field of the
3997 * inode item in subvolume tree as needed (see overwrite_item()).
4000 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4001 btrfs_set_token_inode_transid(&token, item, trans->transid);
4002 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4003 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4004 BTRFS_I(inode)->ro_flags);
4005 btrfs_set_token_inode_flags(&token, item, flags);
4006 btrfs_set_token_inode_block_group(&token, item, 0);
4009 static int log_inode_item(struct btrfs_trans_handle *trans,
4010 struct btrfs_root *log, struct btrfs_path *path,
4011 struct btrfs_inode *inode, bool inode_item_dropped)
4013 struct btrfs_inode_item *inode_item;
4017 * If we are doing a fast fsync and the inode was logged before in the
4018 * current transaction, then we know the inode was previously logged and
4019 * it exists in the log tree. For performance reasons, in this case use
4020 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4021 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4022 * contention in case there are concurrent fsyncs for other inodes of the
4023 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4024 * already exists can also result in unnecessarily splitting a leaf.
4026 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4027 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4033 * This means it is the first fsync in the current transaction,
4034 * so the inode item is not in the log and we need to insert it.
4035 * We can never get -EEXIST because we are only called for a fast
4036 * fsync and in case an inode eviction happens after the inode was
4037 * logged before in the current transaction, when we load again
4038 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4039 * flags and set ->logged_trans to 0.
4041 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4042 sizeof(*inode_item));
4043 ASSERT(ret != -EEXIST);
4047 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4048 struct btrfs_inode_item);
4049 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4051 btrfs_release_path(path);
4055 static int log_csums(struct btrfs_trans_handle *trans,
4056 struct btrfs_inode *inode,
4057 struct btrfs_root *log_root,
4058 struct btrfs_ordered_sum *sums)
4060 const u64 lock_end = sums->bytenr + sums->len - 1;
4061 struct extent_state *cached_state = NULL;
4065 * If this inode was not used for reflink operations in the current
4066 * transaction with new extents, then do the fast path, no need to
4067 * worry about logging checksum items with overlapping ranges.
4069 if (inode->last_reflink_trans < trans->transid)
4070 return btrfs_csum_file_blocks(trans, log_root, sums);
4073 * Serialize logging for checksums. This is to avoid racing with the
4074 * same checksum being logged by another task that is logging another
4075 * file which happens to refer to the same extent as well. Such races
4076 * can leave checksum items in the log with overlapping ranges.
4078 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
4079 lock_end, &cached_state);
4083 * Due to extent cloning, we might have logged a csum item that covers a
4084 * subrange of a cloned extent, and later we can end up logging a csum
4085 * item for a larger subrange of the same extent or the entire range.
4086 * This would leave csum items in the log tree that cover the same range
4087 * and break the searches for checksums in the log tree, resulting in
4088 * some checksums missing in the fs/subvolume tree. So just delete (or
4089 * trim and adjust) any existing csum items in the log for this range.
4091 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4093 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4095 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
4101 static noinline int copy_items(struct btrfs_trans_handle *trans,
4102 struct btrfs_inode *inode,
4103 struct btrfs_path *dst_path,
4104 struct btrfs_path *src_path,
4105 int start_slot, int nr, int inode_only,
4108 struct btrfs_fs_info *fs_info = trans->fs_info;
4109 unsigned long src_offset;
4110 unsigned long dst_offset;
4111 struct btrfs_root *log = inode->root->log_root;
4112 struct btrfs_file_extent_item *extent;
4113 struct btrfs_inode_item *inode_item;
4114 struct extent_buffer *src = src_path->nodes[0];
4116 struct btrfs_key *ins_keys;
4120 struct list_head ordered_sums;
4121 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4123 INIT_LIST_HEAD(&ordered_sums);
4125 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4126 nr * sizeof(u32), GFP_NOFS);
4130 ins_sizes = (u32 *)ins_data;
4131 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4133 for (i = 0; i < nr; i++) {
4134 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4135 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4137 ret = btrfs_insert_empty_items(trans, log, dst_path,
4138 ins_keys, ins_sizes, nr);
4144 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4145 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4146 dst_path->slots[0]);
4148 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4150 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4151 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4153 struct btrfs_inode_item);
4154 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4156 inode_only == LOG_INODE_EXISTS,
4159 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4160 src_offset, ins_sizes[i]);
4163 /* take a reference on file data extents so that truncates
4164 * or deletes of this inode don't have to relog the inode
4167 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4170 extent = btrfs_item_ptr(src, start_slot + i,
4171 struct btrfs_file_extent_item);
4173 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4176 found_type = btrfs_file_extent_type(src, extent);
4177 if (found_type == BTRFS_FILE_EXTENT_REG) {
4179 ds = btrfs_file_extent_disk_bytenr(src,
4181 /* ds == 0 is a hole */
4185 dl = btrfs_file_extent_disk_num_bytes(src,
4187 cs = btrfs_file_extent_offset(src, extent);
4188 cl = btrfs_file_extent_num_bytes(src,
4190 if (btrfs_file_extent_compression(src,
4196 ret = btrfs_lookup_csums_range(
4198 ds + cs, ds + cs + cl - 1,
4206 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4207 btrfs_release_path(dst_path);
4211 * we have to do this after the loop above to avoid changing the
4212 * log tree while trying to change the log tree.
4214 while (!list_empty(&ordered_sums)) {
4215 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4216 struct btrfs_ordered_sum,
4219 ret = log_csums(trans, inode, log, sums);
4220 list_del(&sums->list);
4227 static int extent_cmp(void *priv, const struct list_head *a,
4228 const struct list_head *b)
4230 const struct extent_map *em1, *em2;
4232 em1 = list_entry(a, struct extent_map, list);
4233 em2 = list_entry(b, struct extent_map, list);
4235 if (em1->start < em2->start)
4237 else if (em1->start > em2->start)
4242 static int log_extent_csums(struct btrfs_trans_handle *trans,
4243 struct btrfs_inode *inode,
4244 struct btrfs_root *log_root,
4245 const struct extent_map *em,
4246 struct btrfs_log_ctx *ctx)
4248 struct btrfs_ordered_extent *ordered;
4251 u64 mod_start = em->mod_start;
4252 u64 mod_len = em->mod_len;
4253 LIST_HEAD(ordered_sums);
4256 if (inode->flags & BTRFS_INODE_NODATASUM ||
4257 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4258 em->block_start == EXTENT_MAP_HOLE)
4261 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4262 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4263 const u64 mod_end = mod_start + mod_len;
4264 struct btrfs_ordered_sum *sums;
4269 if (ordered_end <= mod_start)
4271 if (mod_end <= ordered->file_offset)
4275 * We are going to copy all the csums on this ordered extent, so
4276 * go ahead and adjust mod_start and mod_len in case this ordered
4277 * extent has already been logged.
4279 if (ordered->file_offset > mod_start) {
4280 if (ordered_end >= mod_end)
4281 mod_len = ordered->file_offset - mod_start;
4283 * If we have this case
4285 * |--------- logged extent ---------|
4286 * |----- ordered extent ----|
4288 * Just don't mess with mod_start and mod_len, we'll
4289 * just end up logging more csums than we need and it
4293 if (ordered_end < mod_end) {
4294 mod_len = mod_end - ordered_end;
4295 mod_start = ordered_end;
4302 * To keep us from looping for the above case of an ordered
4303 * extent that falls inside of the logged extent.
4305 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4308 list_for_each_entry(sums, &ordered->list, list) {
4309 ret = log_csums(trans, inode, log_root, sums);
4315 /* We're done, found all csums in the ordered extents. */
4319 /* If we're compressed we have to save the entire range of csums. */
4320 if (em->compress_type) {
4322 csum_len = max(em->block_len, em->orig_block_len);
4324 csum_offset = mod_start - em->start;
4328 /* block start is already adjusted for the file extent offset. */
4329 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4330 em->block_start + csum_offset,
4331 em->block_start + csum_offset +
4332 csum_len - 1, &ordered_sums, 0);
4336 while (!list_empty(&ordered_sums)) {
4337 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4338 struct btrfs_ordered_sum,
4341 ret = log_csums(trans, inode, log_root, sums);
4342 list_del(&sums->list);
4349 static int log_one_extent(struct btrfs_trans_handle *trans,
4350 struct btrfs_inode *inode, struct btrfs_root *root,
4351 const struct extent_map *em,
4352 struct btrfs_path *path,
4353 struct btrfs_log_ctx *ctx)
4355 struct btrfs_drop_extents_args drop_args = { 0 };
4356 struct btrfs_root *log = root->log_root;
4357 struct btrfs_file_extent_item *fi;
4358 struct extent_buffer *leaf;
4359 struct btrfs_map_token token;
4360 struct btrfs_key key;
4361 u64 extent_offset = em->start - em->orig_start;
4365 ret = log_extent_csums(trans, inode, log, em, ctx);
4369 drop_args.path = path;
4370 drop_args.start = em->start;
4371 drop_args.end = em->start + em->len;
4372 drop_args.replace_extent = true;
4373 drop_args.extent_item_size = sizeof(*fi);
4374 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4378 if (!drop_args.extent_inserted) {
4379 key.objectid = btrfs_ino(inode);
4380 key.type = BTRFS_EXTENT_DATA_KEY;
4381 key.offset = em->start;
4383 ret = btrfs_insert_empty_item(trans, log, path, &key,
4388 leaf = path->nodes[0];
4389 btrfs_init_map_token(&token, leaf);
4390 fi = btrfs_item_ptr(leaf, path->slots[0],
4391 struct btrfs_file_extent_item);
4393 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4394 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4395 btrfs_set_token_file_extent_type(&token, fi,
4396 BTRFS_FILE_EXTENT_PREALLOC);
4398 btrfs_set_token_file_extent_type(&token, fi,
4399 BTRFS_FILE_EXTENT_REG);
4401 block_len = max(em->block_len, em->orig_block_len);
4402 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4403 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4405 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4406 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4407 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4410 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4412 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4413 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4416 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4417 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4418 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4419 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4420 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4421 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4422 btrfs_mark_buffer_dirty(leaf);
4424 btrfs_release_path(path);
4430 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4431 * lose them after doing a full/fast fsync and replaying the log. We scan the
4432 * subvolume's root instead of iterating the inode's extent map tree because
4433 * otherwise we can log incorrect extent items based on extent map conversion.
4434 * That can happen due to the fact that extent maps are merged when they
4435 * are not in the extent map tree's list of modified extents.
4437 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4438 struct btrfs_inode *inode,
4439 struct btrfs_path *path)
4441 struct btrfs_root *root = inode->root;
4442 struct btrfs_key key;
4443 const u64 i_size = i_size_read(&inode->vfs_inode);
4444 const u64 ino = btrfs_ino(inode);
4445 struct btrfs_path *dst_path = NULL;
4446 bool dropped_extents = false;
4447 u64 truncate_offset = i_size;
4448 struct extent_buffer *leaf;
4454 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4458 key.type = BTRFS_EXTENT_DATA_KEY;
4459 key.offset = i_size;
4460 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4465 * We must check if there is a prealloc extent that starts before the
4466 * i_size and crosses the i_size boundary. This is to ensure later we
4467 * truncate down to the end of that extent and not to the i_size, as
4468 * otherwise we end up losing part of the prealloc extent after a log
4469 * replay and with an implicit hole if there is another prealloc extent
4470 * that starts at an offset beyond i_size.
4472 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4477 struct btrfs_file_extent_item *ei;
4479 leaf = path->nodes[0];
4480 slot = path->slots[0];
4481 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4483 if (btrfs_file_extent_type(leaf, ei) ==
4484 BTRFS_FILE_EXTENT_PREALLOC) {
4487 btrfs_item_key_to_cpu(leaf, &key, slot);
4488 extent_end = key.offset +
4489 btrfs_file_extent_num_bytes(leaf, ei);
4491 if (extent_end > i_size)
4492 truncate_offset = extent_end;
4499 leaf = path->nodes[0];
4500 slot = path->slots[0];
4502 if (slot >= btrfs_header_nritems(leaf)) {
4504 ret = copy_items(trans, inode, dst_path, path,
4505 start_slot, ins_nr, 1, 0);
4510 ret = btrfs_next_leaf(root, path);
4520 btrfs_item_key_to_cpu(leaf, &key, slot);
4521 if (key.objectid > ino)
4523 if (WARN_ON_ONCE(key.objectid < ino) ||
4524 key.type < BTRFS_EXTENT_DATA_KEY ||
4525 key.offset < i_size) {
4529 if (!dropped_extents) {
4531 * Avoid logging extent items logged in past fsync calls
4532 * and leading to duplicate keys in the log tree.
4535 ret = btrfs_truncate_inode_items(trans,
4537 inode, truncate_offset,
4538 BTRFS_EXTENT_DATA_KEY,
4540 } while (ret == -EAGAIN);
4543 dropped_extents = true;
4550 dst_path = btrfs_alloc_path();
4558 ret = copy_items(trans, inode, dst_path, path,
4559 start_slot, ins_nr, 1, 0);
4561 btrfs_release_path(path);
4562 btrfs_free_path(dst_path);
4566 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4567 struct btrfs_root *root,
4568 struct btrfs_inode *inode,
4569 struct btrfs_path *path,
4570 struct btrfs_log_ctx *ctx)
4572 struct btrfs_ordered_extent *ordered;
4573 struct btrfs_ordered_extent *tmp;
4574 struct extent_map *em, *n;
4575 struct list_head extents;
4576 struct extent_map_tree *tree = &inode->extent_tree;
4580 INIT_LIST_HEAD(&extents);
4582 write_lock(&tree->lock);
4584 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4585 list_del_init(&em->list);
4587 * Just an arbitrary number, this can be really CPU intensive
4588 * once we start getting a lot of extents, and really once we
4589 * have a bunch of extents we just want to commit since it will
4592 if (++num > 32768) {
4593 list_del_init(&tree->modified_extents);
4598 if (em->generation < trans->transid)
4601 /* We log prealloc extents beyond eof later. */
4602 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4603 em->start >= i_size_read(&inode->vfs_inode))
4606 /* Need a ref to keep it from getting evicted from cache */
4607 refcount_inc(&em->refs);
4608 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4609 list_add_tail(&em->list, &extents);
4613 list_sort(NULL, &extents, extent_cmp);
4615 while (!list_empty(&extents)) {
4616 em = list_entry(extents.next, struct extent_map, list);
4618 list_del_init(&em->list);
4621 * If we had an error we just need to delete everybody from our
4625 clear_em_logging(tree, em);
4626 free_extent_map(em);
4630 write_unlock(&tree->lock);
4632 ret = log_one_extent(trans, inode, root, em, path, ctx);
4633 write_lock(&tree->lock);
4634 clear_em_logging(tree, em);
4635 free_extent_map(em);
4637 WARN_ON(!list_empty(&extents));
4638 write_unlock(&tree->lock);
4640 btrfs_release_path(path);
4642 ret = btrfs_log_prealloc_extents(trans, inode, path);
4647 * We have logged all extents successfully, now make sure the commit of
4648 * the current transaction waits for the ordered extents to complete
4649 * before it commits and wipes out the log trees, otherwise we would
4650 * lose data if an ordered extents completes after the transaction
4651 * commits and a power failure happens after the transaction commit.
4653 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4654 list_del_init(&ordered->log_list);
4655 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4657 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4658 spin_lock_irq(&inode->ordered_tree.lock);
4659 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4660 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4661 atomic_inc(&trans->transaction->pending_ordered);
4663 spin_unlock_irq(&inode->ordered_tree.lock);
4665 btrfs_put_ordered_extent(ordered);
4671 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4672 struct btrfs_path *path, u64 *size_ret)
4674 struct btrfs_key key;
4677 key.objectid = btrfs_ino(inode);
4678 key.type = BTRFS_INODE_ITEM_KEY;
4681 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4684 } else if (ret > 0) {
4687 struct btrfs_inode_item *item;
4689 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4690 struct btrfs_inode_item);
4691 *size_ret = btrfs_inode_size(path->nodes[0], item);
4693 * If the in-memory inode's i_size is smaller then the inode
4694 * size stored in the btree, return the inode's i_size, so
4695 * that we get a correct inode size after replaying the log
4696 * when before a power failure we had a shrinking truncate
4697 * followed by addition of a new name (rename / new hard link).
4698 * Otherwise return the inode size from the btree, to avoid
4699 * data loss when replaying a log due to previously doing a
4700 * write that expands the inode's size and logging a new name
4701 * immediately after.
4703 if (*size_ret > inode->vfs_inode.i_size)
4704 *size_ret = inode->vfs_inode.i_size;
4707 btrfs_release_path(path);
4712 * At the moment we always log all xattrs. This is to figure out at log replay
4713 * time which xattrs must have their deletion replayed. If a xattr is missing
4714 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4715 * because if a xattr is deleted, the inode is fsynced and a power failure
4716 * happens, causing the log to be replayed the next time the fs is mounted,
4717 * we want the xattr to not exist anymore (same behaviour as other filesystems
4718 * with a journal, ext3/4, xfs, f2fs, etc).
4720 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4721 struct btrfs_root *root,
4722 struct btrfs_inode *inode,
4723 struct btrfs_path *path,
4724 struct btrfs_path *dst_path)
4727 struct btrfs_key key;
4728 const u64 ino = btrfs_ino(inode);
4731 bool found_xattrs = false;
4733 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4737 key.type = BTRFS_XATTR_ITEM_KEY;
4740 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4745 int slot = path->slots[0];
4746 struct extent_buffer *leaf = path->nodes[0];
4747 int nritems = btrfs_header_nritems(leaf);
4749 if (slot >= nritems) {
4751 ret = copy_items(trans, inode, dst_path, path,
4752 start_slot, ins_nr, 1, 0);
4757 ret = btrfs_next_leaf(root, path);
4765 btrfs_item_key_to_cpu(leaf, &key, slot);
4766 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4773 found_xattrs = true;
4777 ret = copy_items(trans, inode, dst_path, path,
4778 start_slot, ins_nr, 1, 0);
4784 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4790 * When using the NO_HOLES feature if we punched a hole that causes the
4791 * deletion of entire leafs or all the extent items of the first leaf (the one
4792 * that contains the inode item and references) we may end up not processing
4793 * any extents, because there are no leafs with a generation matching the
4794 * current transaction that have extent items for our inode. So we need to find
4795 * if any holes exist and then log them. We also need to log holes after any
4796 * truncate operation that changes the inode's size.
4798 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4799 struct btrfs_root *root,
4800 struct btrfs_inode *inode,
4801 struct btrfs_path *path)
4803 struct btrfs_fs_info *fs_info = root->fs_info;
4804 struct btrfs_key key;
4805 const u64 ino = btrfs_ino(inode);
4806 const u64 i_size = i_size_read(&inode->vfs_inode);
4807 u64 prev_extent_end = 0;
4810 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4814 key.type = BTRFS_EXTENT_DATA_KEY;
4817 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4822 struct extent_buffer *leaf = path->nodes[0];
4824 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4825 ret = btrfs_next_leaf(root, path);
4832 leaf = path->nodes[0];
4835 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4836 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4839 /* We have a hole, log it. */
4840 if (prev_extent_end < key.offset) {
4841 const u64 hole_len = key.offset - prev_extent_end;
4844 * Release the path to avoid deadlocks with other code
4845 * paths that search the root while holding locks on
4846 * leafs from the log root.
4848 btrfs_release_path(path);
4849 ret = btrfs_insert_file_extent(trans, root->log_root,
4850 ino, prev_extent_end, 0,
4851 0, hole_len, 0, hole_len,
4857 * Search for the same key again in the root. Since it's
4858 * an extent item and we are holding the inode lock, the
4859 * key must still exist. If it doesn't just emit warning
4860 * and return an error to fall back to a transaction
4863 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4866 if (WARN_ON(ret > 0))
4868 leaf = path->nodes[0];
4871 prev_extent_end = btrfs_file_extent_end(path);
4876 if (prev_extent_end < i_size) {
4879 btrfs_release_path(path);
4880 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4881 ret = btrfs_insert_file_extent(trans, root->log_root,
4882 ino, prev_extent_end, 0, 0,
4883 hole_len, 0, hole_len,
4893 * When we are logging a new inode X, check if it doesn't have a reference that
4894 * matches the reference from some other inode Y created in a past transaction
4895 * and that was renamed in the current transaction. If we don't do this, then at
4896 * log replay time we can lose inode Y (and all its files if it's a directory):
4899 * echo "hello world" > /mnt/x/foobar
4902 * mkdir /mnt/x # or touch /mnt/x
4903 * xfs_io -c fsync /mnt/x
4905 * mount fs, trigger log replay
4907 * After the log replay procedure, we would lose the first directory and all its
4908 * files (file foobar).
4909 * For the case where inode Y is not a directory we simply end up losing it:
4911 * echo "123" > /mnt/foo
4913 * mv /mnt/foo /mnt/bar
4914 * echo "abc" > /mnt/foo
4915 * xfs_io -c fsync /mnt/foo
4918 * We also need this for cases where a snapshot entry is replaced by some other
4919 * entry (file or directory) otherwise we end up with an unreplayable log due to
4920 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4921 * if it were a regular entry:
4924 * btrfs subvolume snapshot /mnt /mnt/x/snap
4925 * btrfs subvolume delete /mnt/x/snap
4928 * fsync /mnt/x or fsync some new file inside it
4931 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4932 * the same transaction.
4934 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4936 const struct btrfs_key *key,
4937 struct btrfs_inode *inode,
4938 u64 *other_ino, u64 *other_parent)
4941 struct btrfs_path *search_path;
4944 u32 item_size = btrfs_item_size_nr(eb, slot);
4946 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4948 search_path = btrfs_alloc_path();
4951 search_path->search_commit_root = 1;
4952 search_path->skip_locking = 1;
4954 while (cur_offset < item_size) {
4958 unsigned long name_ptr;
4959 struct btrfs_dir_item *di;
4961 if (key->type == BTRFS_INODE_REF_KEY) {
4962 struct btrfs_inode_ref *iref;
4964 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4965 parent = key->offset;
4966 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4967 name_ptr = (unsigned long)(iref + 1);
4968 this_len = sizeof(*iref) + this_name_len;
4970 struct btrfs_inode_extref *extref;
4972 extref = (struct btrfs_inode_extref *)(ptr +
4974 parent = btrfs_inode_extref_parent(eb, extref);
4975 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4976 name_ptr = (unsigned long)&extref->name;
4977 this_len = sizeof(*extref) + this_name_len;
4980 if (this_name_len > name_len) {
4983 new_name = krealloc(name, this_name_len, GFP_NOFS);
4988 name_len = this_name_len;
4992 read_extent_buffer(eb, name, name_ptr, this_name_len);
4993 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4994 parent, name, this_name_len, 0);
4995 if (di && !IS_ERR(di)) {
4996 struct btrfs_key di_key;
4998 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5000 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5001 if (di_key.objectid != key->objectid) {
5003 *other_ino = di_key.objectid;
5004 *other_parent = parent;
5012 } else if (IS_ERR(di)) {
5016 btrfs_release_path(search_path);
5018 cur_offset += this_len;
5022 btrfs_free_path(search_path);
5027 struct btrfs_ino_list {
5030 struct list_head list;
5033 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5034 struct btrfs_root *root,
5035 struct btrfs_path *path,
5036 struct btrfs_log_ctx *ctx,
5037 u64 ino, u64 parent)
5039 struct btrfs_ino_list *ino_elem;
5040 LIST_HEAD(inode_list);
5043 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5046 ino_elem->ino = ino;
5047 ino_elem->parent = parent;
5048 list_add_tail(&ino_elem->list, &inode_list);
5050 while (!list_empty(&inode_list)) {
5051 struct btrfs_fs_info *fs_info = root->fs_info;
5052 struct btrfs_key key;
5053 struct inode *inode;
5055 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
5057 ino = ino_elem->ino;
5058 parent = ino_elem->parent;
5059 list_del(&ino_elem->list);
5064 btrfs_release_path(path);
5066 inode = btrfs_iget(fs_info->sb, ino, root);
5068 * If the other inode that had a conflicting dir entry was
5069 * deleted in the current transaction, we need to log its parent
5072 if (IS_ERR(inode)) {
5073 ret = PTR_ERR(inode);
5074 if (ret == -ENOENT) {
5075 inode = btrfs_iget(fs_info->sb, parent, root);
5076 if (IS_ERR(inode)) {
5077 ret = PTR_ERR(inode);
5079 ret = btrfs_log_inode(trans, root,
5081 LOG_OTHER_INODE_ALL,
5083 btrfs_add_delayed_iput(inode);
5089 * If the inode was already logged skip it - otherwise we can
5090 * hit an infinite loop. Example:
5092 * From the commit root (previous transaction) we have the
5095 * inode 257 a directory
5096 * inode 258 with references "zz" and "zz_link" on inode 257
5097 * inode 259 with reference "a" on inode 257
5099 * And in the current (uncommitted) transaction we have:
5101 * inode 257 a directory, unchanged
5102 * inode 258 with references "a" and "a2" on inode 257
5103 * inode 259 with reference "zz_link" on inode 257
5104 * inode 261 with reference "zz" on inode 257
5106 * When logging inode 261 the following infinite loop could
5107 * happen if we don't skip already logged inodes:
5109 * - we detect inode 258 as a conflicting inode, with inode 261
5110 * on reference "zz", and log it;
5112 * - we detect inode 259 as a conflicting inode, with inode 258
5113 * on reference "a", and log it;
5115 * - we detect inode 258 as a conflicting inode, with inode 259
5116 * on reference "zz_link", and log it - again! After this we
5117 * repeat the above steps forever.
5119 spin_lock(&BTRFS_I(inode)->lock);
5121 * Check the inode's logged_trans only instead of
5122 * btrfs_inode_in_log(). This is because the last_log_commit of
5123 * the inode is not updated when we only log that it exists (see
5124 * btrfs_log_inode()).
5126 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5127 spin_unlock(&BTRFS_I(inode)->lock);
5128 btrfs_add_delayed_iput(inode);
5131 spin_unlock(&BTRFS_I(inode)->lock);
5133 * We are safe logging the other inode without acquiring its
5134 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5135 * are safe against concurrent renames of the other inode as
5136 * well because during a rename we pin the log and update the
5137 * log with the new name before we unpin it.
5139 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5140 LOG_OTHER_INODE, ctx);
5142 btrfs_add_delayed_iput(inode);
5147 key.type = BTRFS_INODE_REF_KEY;
5149 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5151 btrfs_add_delayed_iput(inode);
5156 struct extent_buffer *leaf = path->nodes[0];
5157 int slot = path->slots[0];
5159 u64 other_parent = 0;
5161 if (slot >= btrfs_header_nritems(leaf)) {
5162 ret = btrfs_next_leaf(root, path);
5165 } else if (ret > 0) {
5172 btrfs_item_key_to_cpu(leaf, &key, slot);
5173 if (key.objectid != ino ||
5174 (key.type != BTRFS_INODE_REF_KEY &&
5175 key.type != BTRFS_INODE_EXTREF_KEY)) {
5180 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5181 BTRFS_I(inode), &other_ino,
5186 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5191 ino_elem->ino = other_ino;
5192 ino_elem->parent = other_parent;
5193 list_add_tail(&ino_elem->list, &inode_list);
5198 btrfs_add_delayed_iput(inode);
5204 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5205 struct btrfs_inode *inode,
5206 struct btrfs_key *min_key,
5207 const struct btrfs_key *max_key,
5208 struct btrfs_path *path,
5209 struct btrfs_path *dst_path,
5210 const u64 logged_isize,
5211 const bool recursive_logging,
5212 const int inode_only,
5213 struct btrfs_log_ctx *ctx,
5214 bool *need_log_inode_item)
5216 const u64 i_size = i_size_read(&inode->vfs_inode);
5217 struct btrfs_root *root = inode->root;
5218 int ins_start_slot = 0;
5223 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5231 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5232 if (min_key->objectid != max_key->objectid)
5234 if (min_key->type > max_key->type)
5237 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5238 *need_log_inode_item = false;
5239 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5240 min_key->offset >= i_size) {
5242 * Extents at and beyond eof are logged with
5243 * btrfs_log_prealloc_extents().
5244 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5245 * and no keys greater than that, so bail out.
5248 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5249 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5250 inode->generation == trans->transid &&
5251 !recursive_logging) {
5253 u64 other_parent = 0;
5255 ret = btrfs_check_ref_name_override(path->nodes[0],
5256 path->slots[0], min_key, inode,
5257 &other_ino, &other_parent);
5260 } else if (ret > 0 && ctx &&
5261 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5266 ins_start_slot = path->slots[0];
5268 ret = copy_items(trans, inode, dst_path, path,
5269 ins_start_slot, ins_nr,
5270 inode_only, logged_isize);
5275 ret = log_conflicting_inodes(trans, root, path,
5276 ctx, other_ino, other_parent);
5279 btrfs_release_path(path);
5282 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5283 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5286 ret = copy_items(trans, inode, dst_path, path,
5288 ins_nr, inode_only, logged_isize);
5295 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5298 } else if (!ins_nr) {
5299 ins_start_slot = path->slots[0];
5304 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5305 ins_nr, inode_only, logged_isize);
5309 ins_start_slot = path->slots[0];
5312 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5313 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5318 ret = copy_items(trans, inode, dst_path, path,
5319 ins_start_slot, ins_nr, inode_only,
5325 btrfs_release_path(path);
5327 if (min_key->offset < (u64)-1) {
5329 } else if (min_key->type < max_key->type) {
5331 min_key->offset = 0;
5337 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5338 ins_nr, inode_only, logged_isize);
5343 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5345 * Release the path because otherwise we might attempt to double
5346 * lock the same leaf with btrfs_log_prealloc_extents() below.
5348 btrfs_release_path(path);
5349 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5355 /* log a single inode in the tree log.
5356 * At least one parent directory for this inode must exist in the tree
5357 * or be logged already.
5359 * Any items from this inode changed by the current transaction are copied
5360 * to the log tree. An extra reference is taken on any extents in this
5361 * file, allowing us to avoid a whole pile of corner cases around logging
5362 * blocks that have been removed from the tree.
5364 * See LOG_INODE_ALL and related defines for a description of what inode_only
5367 * This handles both files and directories.
5369 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5370 struct btrfs_root *root, struct btrfs_inode *inode,
5372 struct btrfs_log_ctx *ctx)
5374 struct btrfs_path *path;
5375 struct btrfs_path *dst_path;
5376 struct btrfs_key min_key;
5377 struct btrfs_key max_key;
5378 struct btrfs_root *log = root->log_root;
5381 bool fast_search = false;
5382 u64 ino = btrfs_ino(inode);
5383 struct extent_map_tree *em_tree = &inode->extent_tree;
5384 u64 logged_isize = 0;
5385 bool need_log_inode_item = true;
5386 bool xattrs_logged = false;
5387 bool recursive_logging = false;
5388 bool inode_item_dropped = true;
5390 path = btrfs_alloc_path();
5393 dst_path = btrfs_alloc_path();
5395 btrfs_free_path(path);
5399 min_key.objectid = ino;
5400 min_key.type = BTRFS_INODE_ITEM_KEY;
5403 max_key.objectid = ino;
5406 /* today the code can only do partial logging of directories */
5407 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5408 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5409 &inode->runtime_flags) &&
5410 inode_only >= LOG_INODE_EXISTS))
5411 max_key.type = BTRFS_XATTR_ITEM_KEY;
5413 max_key.type = (u8)-1;
5414 max_key.offset = (u64)-1;
5417 * Only run delayed items if we are a directory. We want to make sure
5418 * all directory indexes hit the fs/subvolume tree so we can find them
5419 * and figure out which index ranges have to be logged.
5421 * Otherwise commit the delayed inode only if the full sync flag is set,
5422 * as we want to make sure an up to date version is in the subvolume
5423 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5424 * it to the log tree. For a non full sync, we always log the inode item
5425 * based on the in-memory struct btrfs_inode which is always up to date.
5427 if (S_ISDIR(inode->vfs_inode.i_mode))
5428 ret = btrfs_commit_inode_delayed_items(trans, inode);
5429 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5430 ret = btrfs_commit_inode_delayed_inode(inode);
5433 btrfs_free_path(path);
5434 btrfs_free_path(dst_path);
5438 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5439 recursive_logging = true;
5440 if (inode_only == LOG_OTHER_INODE)
5441 inode_only = LOG_INODE_EXISTS;
5443 inode_only = LOG_INODE_ALL;
5444 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5446 mutex_lock(&inode->log_mutex);
5450 * For symlinks, we must always log their content, which is stored in an
5451 * inline extent, otherwise we could end up with an empty symlink after
5452 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
5453 * one attempts to create an empty symlink).
5454 * We don't need to worry about flushing delalloc, because when we create
5455 * the inline extent when the symlink is created (we never have delalloc
5458 if (S_ISLNK(inode->vfs_inode.i_mode))
5459 inode_only = LOG_INODE_ALL;
5462 * This is for cases where logging a directory could result in losing a
5463 * a file after replaying the log. For example, if we move a file from a
5464 * directory A to a directory B, then fsync directory A, we have no way
5465 * to known the file was moved from A to B, so logging just A would
5466 * result in losing the file after a log replay.
5468 if (S_ISDIR(inode->vfs_inode.i_mode) &&
5469 inode_only == LOG_INODE_ALL &&
5470 inode->last_unlink_trans >= trans->transid) {
5471 btrfs_set_log_full_commit(trans);
5477 * a brute force approach to making sure we get the most uptodate
5478 * copies of everything.
5480 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5481 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5483 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5484 if (inode_only == LOG_INODE_EXISTS)
5485 max_key_type = BTRFS_XATTR_ITEM_KEY;
5486 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5488 if (inode_only == LOG_INODE_EXISTS) {
5490 * Make sure the new inode item we write to the log has
5491 * the same isize as the current one (if it exists).
5492 * This is necessary to prevent data loss after log
5493 * replay, and also to prevent doing a wrong expanding
5494 * truncate - for e.g. create file, write 4K into offset
5495 * 0, fsync, write 4K into offset 4096, add hard link,
5496 * fsync some other file (to sync log), power fail - if
5497 * we use the inode's current i_size, after log replay
5498 * we get a 8Kb file, with the last 4Kb extent as a hole
5499 * (zeroes), as if an expanding truncate happened,
5500 * instead of getting a file of 4Kb only.
5502 err = logged_inode_size(log, inode, path, &logged_isize);
5506 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5507 &inode->runtime_flags)) {
5508 if (inode_only == LOG_INODE_EXISTS) {
5509 max_key.type = BTRFS_XATTR_ITEM_KEY;
5510 ret = drop_objectid_items(trans, log, path, ino,
5513 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5514 &inode->runtime_flags);
5515 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5516 &inode->runtime_flags);
5518 ret = btrfs_truncate_inode_items(trans,
5519 log, inode, 0, 0, NULL);
5524 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5525 &inode->runtime_flags) ||
5526 inode_only == LOG_INODE_EXISTS) {
5527 if (inode_only == LOG_INODE_ALL)
5529 max_key.type = BTRFS_XATTR_ITEM_KEY;
5530 ret = drop_objectid_items(trans, log, path, ino,
5533 if (inode_only == LOG_INODE_ALL)
5535 inode_item_dropped = false;
5545 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5546 path, dst_path, logged_isize,
5547 recursive_logging, inode_only, ctx,
5548 &need_log_inode_item);
5552 btrfs_release_path(path);
5553 btrfs_release_path(dst_path);
5554 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5557 xattrs_logged = true;
5558 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5559 btrfs_release_path(path);
5560 btrfs_release_path(dst_path);
5561 err = btrfs_log_holes(trans, root, inode, path);
5566 btrfs_release_path(path);
5567 btrfs_release_path(dst_path);
5568 if (need_log_inode_item) {
5569 err = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
5573 * If we are doing a fast fsync and the inode was logged before
5574 * in this transaction, we don't need to log the xattrs because
5575 * they were logged before. If xattrs were added, changed or
5576 * deleted since the last time we logged the inode, then we have
5577 * already logged them because the inode had the runtime flag
5578 * BTRFS_INODE_COPY_EVERYTHING set.
5580 if (!xattrs_logged && inode->logged_trans < trans->transid) {
5581 err = btrfs_log_all_xattrs(trans, root, inode, path,
5585 btrfs_release_path(path);
5589 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5595 } else if (inode_only == LOG_INODE_ALL) {
5596 struct extent_map *em, *n;
5598 write_lock(&em_tree->lock);
5599 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5600 list_del_init(&em->list);
5601 write_unlock(&em_tree->lock);
5604 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5605 ret = log_directory_changes(trans, root, inode, path, dst_path,
5614 * If we are logging that an ancestor inode exists as part of logging a
5615 * new name from a link or rename operation, don't mark the inode as
5616 * logged - otherwise if an explicit fsync is made against an ancestor,
5617 * the fsync considers the inode in the log and doesn't sync the log,
5618 * resulting in the ancestor missing after a power failure unless the
5619 * log was synced as part of an fsync against any other unrelated inode.
5620 * So keep it simple for this case and just don't flag the ancestors as
5624 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5625 &inode->vfs_inode != ctx->inode)) {
5626 spin_lock(&inode->lock);
5627 inode->logged_trans = trans->transid;
5629 * Don't update last_log_commit if we logged that an inode exists.
5630 * We do this for two reasons:
5632 * 1) We might have had buffered writes to this inode that were
5633 * flushed and had their ordered extents completed in this
5634 * transaction, but we did not previously log the inode with
5635 * LOG_INODE_ALL. Later the inode was evicted and after that
5636 * it was loaded again and this LOG_INODE_EXISTS log operation
5637 * happened. We must make sure that if an explicit fsync against
5638 * the inode is performed later, it logs the new extents, an
5639 * updated inode item, etc, and syncs the log. The same logic
5640 * applies to direct IO writes instead of buffered writes.
5642 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
5643 * is logged with an i_size of 0 or whatever value was logged
5644 * before. If later the i_size of the inode is increased by a
5645 * truncate operation, the log is synced through an fsync of
5646 * some other inode and then finally an explicit fsync against
5647 * this inode is made, we must make sure this fsync logs the
5648 * inode with the new i_size, the hole between old i_size and
5649 * the new i_size, and syncs the log.
5651 if (inode_only != LOG_INODE_EXISTS)
5652 inode->last_log_commit = inode->last_sub_trans;
5653 spin_unlock(&inode->lock);
5656 mutex_unlock(&inode->log_mutex);
5658 btrfs_free_path(path);
5659 btrfs_free_path(dst_path);
5664 * Check if we need to log an inode. This is used in contexts where while
5665 * logging an inode we need to log another inode (either that it exists or in
5666 * full mode). This is used instead of btrfs_inode_in_log() because the later
5667 * requires the inode to be in the log and have the log transaction committed,
5668 * while here we do not care if the log transaction was already committed - our
5669 * caller will commit the log later - and we want to avoid logging an inode
5670 * multiple times when multiple tasks have joined the same log transaction.
5672 static bool need_log_inode(struct btrfs_trans_handle *trans,
5673 struct btrfs_inode *inode)
5676 * If a directory was not modified, no dentries added or removed, we can
5677 * and should avoid logging it.
5679 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5683 * If this inode does not have new/updated/deleted xattrs since the last
5684 * time it was logged and is flagged as logged in the current transaction,
5685 * we can skip logging it. As for new/deleted names, those are updated in
5686 * the log by link/unlink/rename operations.
5687 * In case the inode was logged and then evicted and reloaded, its
5688 * logged_trans will be 0, in which case we have to fully log it since
5689 * logged_trans is a transient field, not persisted.
5691 if (inode->logged_trans == trans->transid &&
5692 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5698 struct btrfs_dir_list {
5700 struct list_head list;
5704 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5705 * details about the why it is needed.
5706 * This is a recursive operation - if an existing dentry corresponds to a
5707 * directory, that directory's new entries are logged too (same behaviour as
5708 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5709 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5710 * complains about the following circular lock dependency / possible deadlock:
5714 * lock(&type->i_mutex_dir_key#3/2);
5715 * lock(sb_internal#2);
5716 * lock(&type->i_mutex_dir_key#3/2);
5717 * lock(&sb->s_type->i_mutex_key#14);
5719 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5720 * sb_start_intwrite() in btrfs_start_transaction().
5721 * Not locking i_mutex of the inodes is still safe because:
5723 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5724 * that while logging the inode new references (names) are added or removed
5725 * from the inode, leaving the logged inode item with a link count that does
5726 * not match the number of logged inode reference items. This is fine because
5727 * at log replay time we compute the real number of links and correct the
5728 * link count in the inode item (see replay_one_buffer() and
5729 * link_to_fixup_dir());
5731 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5732 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5733 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5734 * has a size that doesn't match the sum of the lengths of all the logged
5735 * names. This does not result in a problem because if a dir_item key is
5736 * logged but its matching dir_index key is not logged, at log replay time we
5737 * don't use it to replay the respective name (see replay_one_name()). On the
5738 * other hand if only the dir_index key ends up being logged, the respective
5739 * name is added to the fs/subvol tree with both the dir_item and dir_index
5740 * keys created (see replay_one_name()).
5741 * The directory's inode item with a wrong i_size is not a problem as well,
5742 * since we don't use it at log replay time to set the i_size in the inode
5743 * item of the fs/subvol tree (see overwrite_item()).
5745 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5746 struct btrfs_root *root,
5747 struct btrfs_inode *start_inode,
5748 struct btrfs_log_ctx *ctx)
5750 struct btrfs_fs_info *fs_info = root->fs_info;
5751 struct btrfs_root *log = root->log_root;
5752 struct btrfs_path *path;
5753 LIST_HEAD(dir_list);
5754 struct btrfs_dir_list *dir_elem;
5757 path = btrfs_alloc_path();
5761 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5763 btrfs_free_path(path);
5766 dir_elem->ino = btrfs_ino(start_inode);
5767 list_add_tail(&dir_elem->list, &dir_list);
5769 while (!list_empty(&dir_list)) {
5770 struct extent_buffer *leaf;
5771 struct btrfs_key min_key;
5775 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5778 goto next_dir_inode;
5780 min_key.objectid = dir_elem->ino;
5781 min_key.type = BTRFS_DIR_ITEM_KEY;
5784 btrfs_release_path(path);
5785 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5787 goto next_dir_inode;
5788 } else if (ret > 0) {
5790 goto next_dir_inode;
5794 leaf = path->nodes[0];
5795 nritems = btrfs_header_nritems(leaf);
5796 for (i = path->slots[0]; i < nritems; i++) {
5797 struct btrfs_dir_item *di;
5798 struct btrfs_key di_key;
5799 struct inode *di_inode;
5800 struct btrfs_dir_list *new_dir_elem;
5801 int log_mode = LOG_INODE_EXISTS;
5804 btrfs_item_key_to_cpu(leaf, &min_key, i);
5805 if (min_key.objectid != dir_elem->ino ||
5806 min_key.type != BTRFS_DIR_ITEM_KEY)
5807 goto next_dir_inode;
5809 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5810 type = btrfs_dir_type(leaf, di);
5811 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5812 type != BTRFS_FT_DIR)
5814 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5815 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5818 btrfs_release_path(path);
5819 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5820 if (IS_ERR(di_inode)) {
5821 ret = PTR_ERR(di_inode);
5822 goto next_dir_inode;
5825 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5826 btrfs_add_delayed_iput(di_inode);
5830 ctx->log_new_dentries = false;
5831 if (type == BTRFS_FT_DIR)
5832 log_mode = LOG_INODE_ALL;
5833 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5835 btrfs_add_delayed_iput(di_inode);
5837 goto next_dir_inode;
5838 if (ctx->log_new_dentries) {
5839 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5841 if (!new_dir_elem) {
5843 goto next_dir_inode;
5845 new_dir_elem->ino = di_key.objectid;
5846 list_add_tail(&new_dir_elem->list, &dir_list);
5851 ret = btrfs_next_leaf(log, path);
5853 goto next_dir_inode;
5854 } else if (ret > 0) {
5856 goto next_dir_inode;
5860 if (min_key.offset < (u64)-1) {
5865 list_del(&dir_elem->list);
5869 btrfs_free_path(path);
5873 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5874 struct btrfs_inode *inode,
5875 struct btrfs_log_ctx *ctx)
5877 struct btrfs_fs_info *fs_info = trans->fs_info;
5879 struct btrfs_path *path;
5880 struct btrfs_key key;
5881 struct btrfs_root *root = inode->root;
5882 const u64 ino = btrfs_ino(inode);
5884 path = btrfs_alloc_path();
5887 path->skip_locking = 1;
5888 path->search_commit_root = 1;
5891 key.type = BTRFS_INODE_REF_KEY;
5893 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5898 struct extent_buffer *leaf = path->nodes[0];
5899 int slot = path->slots[0];
5904 if (slot >= btrfs_header_nritems(leaf)) {
5905 ret = btrfs_next_leaf(root, path);
5913 btrfs_item_key_to_cpu(leaf, &key, slot);
5914 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5915 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5918 item_size = btrfs_item_size_nr(leaf, slot);
5919 ptr = btrfs_item_ptr_offset(leaf, slot);
5920 while (cur_offset < item_size) {
5921 struct btrfs_key inode_key;
5922 struct inode *dir_inode;
5924 inode_key.type = BTRFS_INODE_ITEM_KEY;
5925 inode_key.offset = 0;
5927 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5928 struct btrfs_inode_extref *extref;
5930 extref = (struct btrfs_inode_extref *)
5932 inode_key.objectid = btrfs_inode_extref_parent(
5934 cur_offset += sizeof(*extref);
5935 cur_offset += btrfs_inode_extref_name_len(leaf,
5938 inode_key.objectid = key.offset;
5939 cur_offset = item_size;
5942 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5945 * If the parent inode was deleted, return an error to
5946 * fallback to a transaction commit. This is to prevent
5947 * getting an inode that was moved from one parent A to
5948 * a parent B, got its former parent A deleted and then
5949 * it got fsync'ed, from existing at both parents after
5950 * a log replay (and the old parent still existing).
5957 * mv /mnt/B/bar /mnt/A/bar
5958 * mv -T /mnt/A /mnt/B
5962 * If we ignore the old parent B which got deleted,
5963 * after a log replay we would have file bar linked
5964 * at both parents and the old parent B would still
5967 if (IS_ERR(dir_inode)) {
5968 ret = PTR_ERR(dir_inode);
5972 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
5973 btrfs_add_delayed_iput(dir_inode);
5978 ctx->log_new_dentries = false;
5979 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5980 LOG_INODE_ALL, ctx);
5981 if (!ret && ctx && ctx->log_new_dentries)
5982 ret = log_new_dir_dentries(trans, root,
5983 BTRFS_I(dir_inode), ctx);
5984 btrfs_add_delayed_iput(dir_inode);
5992 btrfs_free_path(path);
5996 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5997 struct btrfs_root *root,
5998 struct btrfs_path *path,
5999 struct btrfs_log_ctx *ctx)
6001 struct btrfs_key found_key;
6003 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6006 struct btrfs_fs_info *fs_info = root->fs_info;
6007 struct extent_buffer *leaf = path->nodes[0];
6008 int slot = path->slots[0];
6009 struct btrfs_key search_key;
6010 struct inode *inode;
6014 btrfs_release_path(path);
6016 ino = found_key.offset;
6018 search_key.objectid = found_key.offset;
6019 search_key.type = BTRFS_INODE_ITEM_KEY;
6020 search_key.offset = 0;
6021 inode = btrfs_iget(fs_info->sb, ino, root);
6023 return PTR_ERR(inode);
6025 if (BTRFS_I(inode)->generation >= trans->transid &&
6026 need_log_inode(trans, BTRFS_I(inode)))
6027 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
6028 LOG_INODE_EXISTS, ctx);
6029 btrfs_add_delayed_iput(inode);
6033 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6036 search_key.type = BTRFS_INODE_REF_KEY;
6037 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6041 leaf = path->nodes[0];
6042 slot = path->slots[0];
6043 if (slot >= btrfs_header_nritems(leaf)) {
6044 ret = btrfs_next_leaf(root, path);
6049 leaf = path->nodes[0];
6050 slot = path->slots[0];
6053 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6054 if (found_key.objectid != search_key.objectid ||
6055 found_key.type != BTRFS_INODE_REF_KEY)
6061 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6062 struct btrfs_inode *inode,
6063 struct dentry *parent,
6064 struct btrfs_log_ctx *ctx)
6066 struct btrfs_root *root = inode->root;
6067 struct dentry *old_parent = NULL;
6068 struct super_block *sb = inode->vfs_inode.i_sb;
6072 if (!parent || d_really_is_negative(parent) ||
6076 inode = BTRFS_I(d_inode(parent));
6077 if (root != inode->root)
6080 if (inode->generation >= trans->transid &&
6081 need_log_inode(trans, inode)) {
6082 ret = btrfs_log_inode(trans, root, inode,
6083 LOG_INODE_EXISTS, ctx);
6087 if (IS_ROOT(parent))
6090 parent = dget_parent(parent);
6092 old_parent = parent;
6099 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6100 struct btrfs_inode *inode,
6101 struct dentry *parent,
6102 struct btrfs_log_ctx *ctx)
6104 struct btrfs_root *root = inode->root;
6105 const u64 ino = btrfs_ino(inode);
6106 struct btrfs_path *path;
6107 struct btrfs_key search_key;
6111 * For a single hard link case, go through a fast path that does not
6112 * need to iterate the fs/subvolume tree.
6114 if (inode->vfs_inode.i_nlink < 2)
6115 return log_new_ancestors_fast(trans, inode, parent, ctx);
6117 path = btrfs_alloc_path();
6121 search_key.objectid = ino;
6122 search_key.type = BTRFS_INODE_REF_KEY;
6123 search_key.offset = 0;
6125 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6132 struct extent_buffer *leaf = path->nodes[0];
6133 int slot = path->slots[0];
6134 struct btrfs_key found_key;
6136 if (slot >= btrfs_header_nritems(leaf)) {
6137 ret = btrfs_next_leaf(root, path);
6145 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6146 if (found_key.objectid != ino ||
6147 found_key.type > BTRFS_INODE_EXTREF_KEY)
6151 * Don't deal with extended references because they are rare
6152 * cases and too complex to deal with (we would need to keep
6153 * track of which subitem we are processing for each item in
6154 * this loop, etc). So just return some error to fallback to
6155 * a transaction commit.
6157 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6163 * Logging ancestors needs to do more searches on the fs/subvol
6164 * tree, so it releases the path as needed to avoid deadlocks.
6165 * Keep track of the last inode ref key and resume from that key
6166 * after logging all new ancestors for the current hard link.
6168 memcpy(&search_key, &found_key, sizeof(search_key));
6170 ret = log_new_ancestors(trans, root, path, ctx);
6173 btrfs_release_path(path);
6178 btrfs_free_path(path);
6183 * helper function around btrfs_log_inode to make sure newly created
6184 * parent directories also end up in the log. A minimal inode and backref
6185 * only logging is done of any parent directories that are older than
6186 * the last committed transaction
6188 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6189 struct btrfs_inode *inode,
6190 struct dentry *parent,
6192 struct btrfs_log_ctx *ctx)
6194 struct btrfs_root *root = inode->root;
6195 struct btrfs_fs_info *fs_info = root->fs_info;
6197 bool log_dentries = false;
6199 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6204 if (btrfs_root_refs(&root->root_item) == 0) {
6210 * Skip already logged inodes or inodes corresponding to tmpfiles
6211 * (since logging them is pointless, a link count of 0 means they
6212 * will never be accessible).
6214 if ((btrfs_inode_in_log(inode, trans->transid) &&
6215 list_empty(&ctx->ordered_extents)) ||
6216 inode->vfs_inode.i_nlink == 0) {
6217 ret = BTRFS_NO_LOG_SYNC;
6221 ret = start_log_trans(trans, root, ctx);
6225 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6230 * for regular files, if its inode is already on disk, we don't
6231 * have to worry about the parents at all. This is because
6232 * we can use the last_unlink_trans field to record renames
6233 * and other fun in this file.
6235 if (S_ISREG(inode->vfs_inode.i_mode) &&
6236 inode->generation < trans->transid &&
6237 inode->last_unlink_trans < trans->transid) {
6242 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6243 log_dentries = true;
6246 * On unlink we must make sure all our current and old parent directory
6247 * inodes are fully logged. This is to prevent leaving dangling
6248 * directory index entries in directories that were our parents but are
6249 * not anymore. Not doing this results in old parent directory being
6250 * impossible to delete after log replay (rmdir will always fail with
6251 * error -ENOTEMPTY).
6257 * ln testdir/foo testdir/bar
6259 * unlink testdir/bar
6260 * xfs_io -c fsync testdir/foo
6262 * mount fs, triggers log replay
6264 * If we don't log the parent directory (testdir), after log replay the
6265 * directory still has an entry pointing to the file inode using the bar
6266 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6267 * the file inode has a link count of 1.
6273 * ln foo testdir/foo2
6274 * ln foo testdir/foo3
6276 * unlink testdir/foo3
6277 * xfs_io -c fsync foo
6279 * mount fs, triggers log replay
6281 * Similar as the first example, after log replay the parent directory
6282 * testdir still has an entry pointing to the inode file with name foo3
6283 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6284 * and has a link count of 2.
6286 if (inode->last_unlink_trans >= trans->transid) {
6287 ret = btrfs_log_all_parents(trans, inode, ctx);
6292 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6297 ret = log_new_dir_dentries(trans, root, inode, ctx);
6302 btrfs_set_log_full_commit(trans);
6307 btrfs_remove_log_ctx(root, ctx);
6308 btrfs_end_log_trans(root);
6314 * it is not safe to log dentry if the chunk root has added new
6315 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6316 * If this returns 1, you must commit the transaction to safely get your
6319 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6320 struct dentry *dentry,
6321 struct btrfs_log_ctx *ctx)
6323 struct dentry *parent = dget_parent(dentry);
6326 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6327 LOG_INODE_ALL, ctx);
6334 * should be called during mount to recover any replay any log trees
6337 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6340 struct btrfs_path *path;
6341 struct btrfs_trans_handle *trans;
6342 struct btrfs_key key;
6343 struct btrfs_key found_key;
6344 struct btrfs_root *log;
6345 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6346 struct walk_control wc = {
6347 .process_func = process_one_buffer,
6348 .stage = LOG_WALK_PIN_ONLY,
6351 path = btrfs_alloc_path();
6355 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6357 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6358 if (IS_ERR(trans)) {
6359 ret = PTR_ERR(trans);
6366 ret = walk_log_tree(trans, log_root_tree, &wc);
6368 btrfs_handle_fs_error(fs_info, ret,
6369 "Failed to pin buffers while recovering log root tree.");
6374 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6375 key.offset = (u64)-1;
6376 key.type = BTRFS_ROOT_ITEM_KEY;
6379 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6382 btrfs_handle_fs_error(fs_info, ret,
6383 "Couldn't find tree log root.");
6387 if (path->slots[0] == 0)
6391 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6393 btrfs_release_path(path);
6394 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6397 log = btrfs_read_tree_root(log_root_tree, &found_key);
6400 btrfs_handle_fs_error(fs_info, ret,
6401 "Couldn't read tree log root.");
6405 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6407 if (IS_ERR(wc.replay_dest)) {
6408 ret = PTR_ERR(wc.replay_dest);
6411 * We didn't find the subvol, likely because it was
6412 * deleted. This is ok, simply skip this log and go to
6415 * We need to exclude the root because we can't have
6416 * other log replays overwriting this log as we'll read
6417 * it back in a few more times. This will keep our
6418 * block from being modified, and we'll just bail for
6419 * each subsequent pass.
6422 ret = btrfs_pin_extent_for_log_replay(trans,
6425 btrfs_put_root(log);
6429 btrfs_handle_fs_error(fs_info, ret,
6430 "Couldn't read target root for tree log recovery.");
6434 wc.replay_dest->log_root = log;
6435 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
6437 /* The loop needs to continue due to the root refs */
6438 btrfs_handle_fs_error(fs_info, ret,
6439 "failed to record the log root in transaction");
6441 ret = walk_log_tree(trans, log, &wc);
6443 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6444 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6448 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6449 struct btrfs_root *root = wc.replay_dest;
6451 btrfs_release_path(path);
6454 * We have just replayed everything, and the highest
6455 * objectid of fs roots probably has changed in case
6456 * some inode_item's got replayed.
6458 * root->objectid_mutex is not acquired as log replay
6459 * could only happen during mount.
6461 ret = btrfs_init_root_free_objectid(root);
6464 wc.replay_dest->log_root = NULL;
6465 btrfs_put_root(wc.replay_dest);
6466 btrfs_put_root(log);
6471 if (found_key.offset == 0)
6473 key.offset = found_key.offset - 1;
6475 btrfs_release_path(path);
6477 /* step one is to pin it all, step two is to replay just inodes */
6480 wc.process_func = replay_one_buffer;
6481 wc.stage = LOG_WALK_REPLAY_INODES;
6484 /* step three is to replay everything */
6485 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6490 btrfs_free_path(path);
6492 /* step 4: commit the transaction, which also unpins the blocks */
6493 ret = btrfs_commit_transaction(trans);
6497 log_root_tree->log_root = NULL;
6498 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6499 btrfs_put_root(log_root_tree);
6504 btrfs_end_transaction(wc.trans);
6505 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6506 btrfs_free_path(path);
6511 * there are some corner cases where we want to force a full
6512 * commit instead of allowing a directory to be logged.
6514 * They revolve around files there were unlinked from the directory, and
6515 * this function updates the parent directory so that a full commit is
6516 * properly done if it is fsync'd later after the unlinks are done.
6518 * Must be called before the unlink operations (updates to the subvolume tree,
6519 * inodes, etc) are done.
6521 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6522 struct btrfs_inode *dir, struct btrfs_inode *inode,
6526 * when we're logging a file, if it hasn't been renamed
6527 * or unlinked, and its inode is fully committed on disk,
6528 * we don't have to worry about walking up the directory chain
6529 * to log its parents.
6531 * So, we use the last_unlink_trans field to put this transid
6532 * into the file. When the file is logged we check it and
6533 * don't log the parents if the file is fully on disk.
6535 mutex_lock(&inode->log_mutex);
6536 inode->last_unlink_trans = trans->transid;
6537 mutex_unlock(&inode->log_mutex);
6540 * if this directory was already logged any new
6541 * names for this file/dir will get recorded
6543 if (dir->logged_trans == trans->transid)
6547 * if the inode we're about to unlink was logged,
6548 * the log will be properly updated for any new names
6550 if (inode->logged_trans == trans->transid)
6554 * when renaming files across directories, if the directory
6555 * there we're unlinking from gets fsync'd later on, there's
6556 * no way to find the destination directory later and fsync it
6557 * properly. So, we have to be conservative and force commits
6558 * so the new name gets discovered.
6563 /* we can safely do the unlink without any special recording */
6567 mutex_lock(&dir->log_mutex);
6568 dir->last_unlink_trans = trans->transid;
6569 mutex_unlock(&dir->log_mutex);
6573 * Make sure that if someone attempts to fsync the parent directory of a deleted
6574 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6575 * that after replaying the log tree of the parent directory's root we will not
6576 * see the snapshot anymore and at log replay time we will not see any log tree
6577 * corresponding to the deleted snapshot's root, which could lead to replaying
6578 * it after replaying the log tree of the parent directory (which would replay
6579 * the snapshot delete operation).
6581 * Must be called before the actual snapshot destroy operation (updates to the
6582 * parent root and tree of tree roots trees, etc) are done.
6584 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6585 struct btrfs_inode *dir)
6587 mutex_lock(&dir->log_mutex);
6588 dir->last_unlink_trans = trans->transid;
6589 mutex_unlock(&dir->log_mutex);
6593 * Update the log after adding a new name for an inode.
6595 * @trans: Transaction handle.
6596 * @old_dentry: The dentry associated with the old name and the old
6598 * @old_dir: The inode of the previous parent directory for the case
6599 * of a rename. For a link operation, it must be NULL.
6600 * @parent: The dentry associated with the directory under which the
6601 * new name is located.
6603 * Call this after adding a new name for an inode, as a result of a link or
6604 * rename operation, and it will properly update the log to reflect the new name.
6606 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6607 struct dentry *old_dentry, struct btrfs_inode *old_dir,
6608 struct dentry *parent)
6610 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
6611 struct btrfs_log_ctx ctx;
6614 * this will force the logging code to walk the dentry chain
6617 if (!S_ISDIR(inode->vfs_inode.i_mode))
6618 inode->last_unlink_trans = trans->transid;
6621 * if this inode hasn't been logged and directory we're renaming it
6622 * from hasn't been logged, we don't need to log it
6624 if (!inode_logged(trans, inode) &&
6625 (!old_dir || !inode_logged(trans, old_dir)))
6629 * If we are doing a rename (old_dir is not NULL) from a directory that
6630 * was previously logged, make sure the next log attempt on the directory
6631 * is not skipped and logs the inode again. This is because the log may
6632 * not currently be authoritative for a range including the old
6633 * BTRFS_DIR_ITEM_KEY and BTRFS_DIR_INDEX_KEY keys, so we want to make
6634 * sure after a log replay we do not end up with both the new and old
6635 * dentries around (in case the inode is a directory we would have a
6636 * directory with two hard links and 2 inode references for different
6637 * parents). The next log attempt of old_dir will happen at
6638 * btrfs_log_all_parents(), called through btrfs_log_inode_parent()
6639 * below, because we have previously set inode->last_unlink_trans to the
6640 * current transaction ID, either here or at btrfs_record_unlink_dir() in
6641 * case inode is a directory.
6644 old_dir->logged_trans = 0;
6646 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6647 ctx.logging_new_name = true;
6649 * We don't care about the return value. If we fail to log the new name
6650 * then we know the next attempt to sync the log will fallback to a full
6651 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6652 * we don't need to worry about getting a log committed that has an
6653 * inconsistent state after a rename operation.
6655 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);