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"
23 /* magic values for the inode_only field in btrfs_log_inode:
25 * LOG_INODE_ALL means to log everything
26 * LOG_INODE_EXISTS means to log just enough to recreate the inode
37 * directory trouble cases
39 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
40 * log, we must force a full commit before doing an fsync of the directory
41 * where the unlink was done.
42 * ---> record transid of last unlink/rename per directory
46 * rename foo/some_dir foo2/some_dir
48 * fsync foo/some_dir/some_file
50 * The fsync above will unlink the original some_dir without recording
51 * it in its new location (foo2). After a crash, some_dir will be gone
52 * unless the fsync of some_file forces a full commit
54 * 2) we must log any new names for any file or dir that is in the fsync
55 * log. ---> check inode while renaming/linking.
57 * 2a) we must log any new names for any file or dir during rename
58 * when the directory they are being removed from was logged.
59 * ---> check inode and old parent dir during rename
61 * 2a is actually the more important variant. With the extra logging
62 * a crash might unlink the old name without recreating the new one
64 * 3) after a crash, we must go through any directories with a link count
65 * of zero and redo the rm -rf
72 * The directory f1 was fully removed from the FS, but fsync was never
73 * called on f1, only its parent dir. After a crash the rm -rf must
74 * be replayed. This must be able to recurse down the entire
75 * directory tree. The inode link count fixup code takes care of the
80 * stages for the tree walking. The first
81 * stage (0) is to only pin down the blocks we find
82 * the second stage (1) is to make sure that all the inodes
83 * we find in the log are created in the subvolume.
85 * The last stage is to deal with directories and links and extents
86 * and all the other fun semantics
90 LOG_WALK_REPLAY_INODES,
91 LOG_WALK_REPLAY_DIR_INDEX,
95 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root, struct btrfs_inode *inode,
98 struct btrfs_log_ctx *ctx);
99 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root,
101 struct btrfs_path *path, u64 objectid);
102 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_root *log,
105 struct btrfs_path *path,
106 u64 dirid, int del_all);
109 * tree logging is a special write ahead log used to make sure that
110 * fsyncs and O_SYNCs can happen without doing full tree commits.
112 * Full tree commits are expensive because they require commonly
113 * modified blocks to be recowed, creating many dirty pages in the
114 * extent tree an 4x-6x higher write load than ext3.
116 * Instead of doing a tree commit on every fsync, we use the
117 * key ranges and transaction ids to find items for a given file or directory
118 * that have changed in this transaction. Those items are copied into
119 * a special tree (one per subvolume root), that tree is written to disk
120 * and then the fsync is considered complete.
122 * After a crash, items are copied out of the log-tree back into the
123 * subvolume tree. Any file data extents found are recorded in the extent
124 * allocation tree, and the log-tree freed.
126 * The log tree is read three times, once to pin down all the extents it is
127 * using in ram and once, once to create all the inodes logged in the tree
128 * and once to do all the other items.
132 * start a sub transaction and setup the log tree
133 * this increments the log tree writer count to make the people
134 * syncing the tree wait for us to finish
136 static int start_log_trans(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_log_ctx *ctx)
140 struct btrfs_fs_info *fs_info = root->fs_info;
141 struct btrfs_root *tree_root = fs_info->tree_root;
145 * First check if the log root tree was already created. If not, create
146 * it before locking the root's log_mutex, just to keep lockdep happy.
148 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
149 mutex_lock(&tree_root->log_mutex);
150 if (!fs_info->log_root_tree) {
151 ret = btrfs_init_log_root_tree(trans, fs_info);
153 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
155 mutex_unlock(&tree_root->log_mutex);
160 mutex_lock(&root->log_mutex);
162 if (root->log_root) {
163 if (btrfs_need_log_full_commit(trans)) {
168 if (!root->log_start_pid) {
169 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
170 root->log_start_pid = current->pid;
171 } else if (root->log_start_pid != current->pid) {
172 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
175 ret = btrfs_add_log_tree(trans, root);
179 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
180 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
181 root->log_start_pid = current->pid;
184 atomic_inc(&root->log_writers);
185 if (ctx && !ctx->logging_new_name) {
186 int index = root->log_transid % 2;
187 list_add_tail(&ctx->list, &root->log_ctxs[index]);
188 ctx->log_transid = root->log_transid;
192 mutex_unlock(&root->log_mutex);
197 * returns 0 if there was a log transaction running and we were able
198 * to join, or returns -ENOENT if there were not transactions
201 static int join_running_log_trans(struct btrfs_root *root)
205 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
208 mutex_lock(&root->log_mutex);
209 if (root->log_root) {
211 atomic_inc(&root->log_writers);
213 mutex_unlock(&root->log_mutex);
218 * This either makes the current running log transaction wait
219 * until you call btrfs_end_log_trans() or it makes any future
220 * log transactions wait until you call btrfs_end_log_trans()
222 void btrfs_pin_log_trans(struct btrfs_root *root)
224 atomic_inc(&root->log_writers);
228 * indicate we're done making changes to the log tree
229 * and wake up anyone waiting to do a sync
231 void btrfs_end_log_trans(struct btrfs_root *root)
233 if (atomic_dec_and_test(&root->log_writers)) {
234 /* atomic_dec_and_test implies a barrier */
235 cond_wake_up_nomb(&root->log_writer_wait);
239 static int btrfs_write_tree_block(struct extent_buffer *buf)
241 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
242 buf->start + buf->len - 1);
245 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
247 filemap_fdatawait_range(buf->pages[0]->mapping,
248 buf->start, buf->start + buf->len - 1);
252 * the walk control struct is used to pass state down the chain when
253 * processing the log tree. The stage field tells us which part
254 * of the log tree processing we are currently doing. The others
255 * are state fields used for that specific part
257 struct walk_control {
258 /* should we free the extent on disk when done? This is used
259 * at transaction commit time while freeing a log tree
263 /* should we write out the extent buffer? This is used
264 * while flushing the log tree to disk during a sync
268 /* should we wait for the extent buffer io to finish? Also used
269 * while flushing the log tree to disk for a sync
273 /* pin only walk, we record which extents on disk belong to the
278 /* what stage of the replay code we're currently in */
282 * Ignore any items from the inode currently being processed. Needs
283 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
284 * the LOG_WALK_REPLAY_INODES stage.
286 bool ignore_cur_inode;
288 /* the root we are currently replaying */
289 struct btrfs_root *replay_dest;
291 /* the trans handle for the current replay */
292 struct btrfs_trans_handle *trans;
294 /* the function that gets used to process blocks we find in the
295 * tree. Note the extent_buffer might not be up to date when it is
296 * passed in, and it must be checked or read if you need the data
299 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
300 struct walk_control *wc, u64 gen, int level);
304 * process_func used to pin down extents, write them or wait on them
306 static int process_one_buffer(struct btrfs_root *log,
307 struct extent_buffer *eb,
308 struct walk_control *wc, u64 gen, int level)
310 struct btrfs_fs_info *fs_info = log->fs_info;
314 * If this fs is mixed then we need to be able to process the leaves to
315 * pin down any logged extents, so we have to read the block.
317 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
318 ret = btrfs_read_buffer(eb, gen, level, NULL);
324 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
327 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
328 if (wc->pin && btrfs_header_level(eb) == 0)
329 ret = btrfs_exclude_logged_extents(eb);
331 btrfs_write_tree_block(eb);
333 btrfs_wait_tree_block_writeback(eb);
339 * Item overwrite used by replay and tree logging. eb, slot and key all refer
340 * to the src data we are copying out.
342 * root is the tree we are copying into, and path is a scratch
343 * path for use in this function (it should be released on entry and
344 * will be released on exit).
346 * If the key is already in the destination tree the existing item is
347 * overwritten. If the existing item isn't big enough, it is extended.
348 * If it is too large, it is truncated.
350 * If the key isn't in the destination yet, a new item is inserted.
352 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
353 struct btrfs_root *root,
354 struct btrfs_path *path,
355 struct extent_buffer *eb, int slot,
356 struct btrfs_key *key)
360 u64 saved_i_size = 0;
361 int save_old_i_size = 0;
362 unsigned long src_ptr;
363 unsigned long dst_ptr;
364 int overwrite_root = 0;
365 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
367 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
370 item_size = btrfs_item_size_nr(eb, slot);
371 src_ptr = btrfs_item_ptr_offset(eb, slot);
373 /* look for the key in the destination tree */
374 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
381 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
383 if (dst_size != item_size)
386 if (item_size == 0) {
387 btrfs_release_path(path);
390 dst_copy = kmalloc(item_size, GFP_NOFS);
391 src_copy = kmalloc(item_size, GFP_NOFS);
392 if (!dst_copy || !src_copy) {
393 btrfs_release_path(path);
399 read_extent_buffer(eb, src_copy, src_ptr, item_size);
401 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
402 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
404 ret = memcmp(dst_copy, src_copy, item_size);
409 * they have the same contents, just return, this saves
410 * us from cowing blocks in the destination tree and doing
411 * extra writes that may not have been done by a previous
415 btrfs_release_path(path);
420 * We need to load the old nbytes into the inode so when we
421 * replay the extents we've logged we get the right nbytes.
424 struct btrfs_inode_item *item;
428 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
429 struct btrfs_inode_item);
430 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
431 item = btrfs_item_ptr(eb, slot,
432 struct btrfs_inode_item);
433 btrfs_set_inode_nbytes(eb, item, nbytes);
436 * If this is a directory we need to reset the i_size to
437 * 0 so that we can set it up properly when replaying
438 * the rest of the items in this log.
440 mode = btrfs_inode_mode(eb, item);
442 btrfs_set_inode_size(eb, item, 0);
444 } else if (inode_item) {
445 struct btrfs_inode_item *item;
449 * New inode, set nbytes to 0 so that the nbytes comes out
450 * properly when we replay the extents.
452 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
453 btrfs_set_inode_nbytes(eb, item, 0);
456 * If this is a directory we need to reset the i_size to 0 so
457 * that we can set it up properly when replaying the rest of
458 * the items in this log.
460 mode = btrfs_inode_mode(eb, item);
462 btrfs_set_inode_size(eb, item, 0);
465 btrfs_release_path(path);
466 /* try to insert the key into the destination tree */
467 path->skip_release_on_error = 1;
468 ret = btrfs_insert_empty_item(trans, root, path,
470 path->skip_release_on_error = 0;
472 /* make sure any existing item is the correct size */
473 if (ret == -EEXIST || ret == -EOVERFLOW) {
475 found_size = btrfs_item_size_nr(path->nodes[0],
477 if (found_size > item_size)
478 btrfs_truncate_item(path, item_size, 1);
479 else if (found_size < item_size)
480 btrfs_extend_item(path, item_size - found_size);
484 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
487 /* don't overwrite an existing inode if the generation number
488 * was logged as zero. This is done when the tree logging code
489 * is just logging an inode to make sure it exists after recovery.
491 * Also, don't overwrite i_size on directories during replay.
492 * log replay inserts and removes directory items based on the
493 * state of the tree found in the subvolume, and i_size is modified
496 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
497 struct btrfs_inode_item *src_item;
498 struct btrfs_inode_item *dst_item;
500 src_item = (struct btrfs_inode_item *)src_ptr;
501 dst_item = (struct btrfs_inode_item *)dst_ptr;
503 if (btrfs_inode_generation(eb, src_item) == 0) {
504 struct extent_buffer *dst_eb = path->nodes[0];
505 const u64 ino_size = btrfs_inode_size(eb, src_item);
508 * For regular files an ino_size == 0 is used only when
509 * logging that an inode exists, as part of a directory
510 * fsync, and the inode wasn't fsynced before. In this
511 * case don't set the size of the inode in the fs/subvol
512 * tree, otherwise we would be throwing valid data away.
514 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
515 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
517 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
521 if (overwrite_root &&
522 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
523 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
525 saved_i_size = btrfs_inode_size(path->nodes[0],
530 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
533 if (save_old_i_size) {
534 struct btrfs_inode_item *dst_item;
535 dst_item = (struct btrfs_inode_item *)dst_ptr;
536 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
539 /* make sure the generation is filled in */
540 if (key->type == BTRFS_INODE_ITEM_KEY) {
541 struct btrfs_inode_item *dst_item;
542 dst_item = (struct btrfs_inode_item *)dst_ptr;
543 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
544 btrfs_set_inode_generation(path->nodes[0], dst_item,
549 btrfs_mark_buffer_dirty(path->nodes[0]);
550 btrfs_release_path(path);
555 * simple helper to read an inode off the disk from a given root
556 * This can only be called for subvolume roots and not for the log
558 static noinline struct inode *read_one_inode(struct btrfs_root *root,
563 inode = btrfs_iget(root->fs_info->sb, objectid, root);
569 /* replays a single extent in 'eb' at 'slot' with 'key' into the
570 * subvolume 'root'. path is released on entry and should be released
573 * extents in the log tree have not been allocated out of the extent
574 * tree yet. So, this completes the allocation, taking a reference
575 * as required if the extent already exists or creating a new extent
576 * if it isn't in the extent allocation tree yet.
578 * The extent is inserted into the file, dropping any existing extents
579 * from the file that overlap the new one.
581 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
582 struct btrfs_root *root,
583 struct btrfs_path *path,
584 struct extent_buffer *eb, int slot,
585 struct btrfs_key *key)
587 struct btrfs_drop_extents_args drop_args = { 0 };
588 struct btrfs_fs_info *fs_info = root->fs_info;
591 u64 start = key->offset;
593 struct btrfs_file_extent_item *item;
594 struct inode *inode = NULL;
598 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
599 found_type = btrfs_file_extent_type(eb, item);
601 if (found_type == BTRFS_FILE_EXTENT_REG ||
602 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
603 nbytes = btrfs_file_extent_num_bytes(eb, item);
604 extent_end = start + nbytes;
607 * We don't add to the inodes nbytes if we are prealloc or a
610 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
612 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
613 size = btrfs_file_extent_ram_bytes(eb, item);
614 nbytes = btrfs_file_extent_ram_bytes(eb, item);
615 extent_end = ALIGN(start + size,
616 fs_info->sectorsize);
622 inode = read_one_inode(root, key->objectid);
629 * first check to see if we already have this extent in the
630 * file. This must be done before the btrfs_drop_extents run
631 * so we don't try to drop this extent.
633 ret = btrfs_lookup_file_extent(trans, root, path,
634 btrfs_ino(BTRFS_I(inode)), start, 0);
637 (found_type == BTRFS_FILE_EXTENT_REG ||
638 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
639 struct btrfs_file_extent_item cmp1;
640 struct btrfs_file_extent_item cmp2;
641 struct btrfs_file_extent_item *existing;
642 struct extent_buffer *leaf;
644 leaf = path->nodes[0];
645 existing = btrfs_item_ptr(leaf, path->slots[0],
646 struct btrfs_file_extent_item);
648 read_extent_buffer(eb, &cmp1, (unsigned long)item,
650 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
654 * we already have a pointer to this exact extent,
655 * we don't have to do anything
657 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
658 btrfs_release_path(path);
662 btrfs_release_path(path);
664 /* drop any overlapping extents */
665 drop_args.start = start;
666 drop_args.end = extent_end;
667 drop_args.drop_cache = true;
668 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
672 if (found_type == BTRFS_FILE_EXTENT_REG ||
673 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
675 unsigned long dest_offset;
676 struct btrfs_key ins;
678 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
679 btrfs_fs_incompat(fs_info, NO_HOLES))
682 ret = btrfs_insert_empty_item(trans, root, path, key,
686 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
688 copy_extent_buffer(path->nodes[0], eb, dest_offset,
689 (unsigned long)item, sizeof(*item));
691 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
692 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
693 ins.type = BTRFS_EXTENT_ITEM_KEY;
694 offset = key->offset - btrfs_file_extent_offset(eb, item);
697 * Manually record dirty extent, as here we did a shallow
698 * file extent item copy and skip normal backref update,
699 * but modifying extent tree all by ourselves.
700 * So need to manually record dirty extent for qgroup,
701 * as the owner of the file extent changed from log tree
702 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
704 ret = btrfs_qgroup_trace_extent(trans,
705 btrfs_file_extent_disk_bytenr(eb, item),
706 btrfs_file_extent_disk_num_bytes(eb, item),
711 if (ins.objectid > 0) {
712 struct btrfs_ref ref = { 0 };
715 LIST_HEAD(ordered_sums);
718 * is this extent already allocated in the extent
719 * allocation tree? If so, just add a reference
721 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
724 btrfs_init_generic_ref(&ref,
725 BTRFS_ADD_DELAYED_REF,
726 ins.objectid, ins.offset, 0);
727 btrfs_init_data_ref(&ref,
728 root->root_key.objectid,
729 key->objectid, offset);
730 ret = btrfs_inc_extent_ref(trans, &ref);
735 * insert the extent pointer in the extent
738 ret = btrfs_alloc_logged_file_extent(trans,
739 root->root_key.objectid,
740 key->objectid, offset, &ins);
744 btrfs_release_path(path);
746 if (btrfs_file_extent_compression(eb, item)) {
747 csum_start = ins.objectid;
748 csum_end = csum_start + ins.offset;
750 csum_start = ins.objectid +
751 btrfs_file_extent_offset(eb, item);
752 csum_end = csum_start +
753 btrfs_file_extent_num_bytes(eb, item);
756 ret = btrfs_lookup_csums_range(root->log_root,
757 csum_start, csum_end - 1,
762 * Now delete all existing cums in the csum root that
763 * cover our range. We do this because we can have an
764 * extent that is completely referenced by one file
765 * extent item and partially referenced by another
766 * file extent item (like after using the clone or
767 * extent_same ioctls). In this case if we end up doing
768 * the replay of the one that partially references the
769 * extent first, and we do not do the csum deletion
770 * below, we can get 2 csum items in the csum tree that
771 * overlap each other. For example, imagine our log has
772 * the two following file extent items:
774 * key (257 EXTENT_DATA 409600)
775 * extent data disk byte 12845056 nr 102400
776 * extent data offset 20480 nr 20480 ram 102400
778 * key (257 EXTENT_DATA 819200)
779 * extent data disk byte 12845056 nr 102400
780 * extent data offset 0 nr 102400 ram 102400
782 * Where the second one fully references the 100K extent
783 * that starts at disk byte 12845056, and the log tree
784 * has a single csum item that covers the entire range
787 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
789 * After the first file extent item is replayed, the
790 * csum tree gets the following csum item:
792 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
794 * Which covers the 20K sub-range starting at offset 20K
795 * of our extent. Now when we replay the second file
796 * extent item, if we do not delete existing csum items
797 * that cover any of its blocks, we end up getting two
798 * csum items in our csum tree that overlap each other:
800 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
801 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
803 * Which is a problem, because after this anyone trying
804 * to lookup up for the checksum of any block of our
805 * extent starting at an offset of 40K or higher, will
806 * end up looking at the second csum item only, which
807 * does not contain the checksum for any block starting
808 * at offset 40K or higher of our extent.
810 while (!list_empty(&ordered_sums)) {
811 struct btrfs_ordered_sum *sums;
812 sums = list_entry(ordered_sums.next,
813 struct btrfs_ordered_sum,
816 ret = btrfs_del_csums(trans,
821 ret = btrfs_csum_file_blocks(trans,
822 fs_info->csum_root, sums);
823 list_del(&sums->list);
829 btrfs_release_path(path);
831 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
832 /* inline extents are easy, we just overwrite them */
833 ret = overwrite_item(trans, root, path, eb, slot, key);
838 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
844 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
845 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
853 * when cleaning up conflicts between the directory names in the
854 * subvolume, directory names in the log and directory names in the
855 * inode back references, we may have to unlink inodes from directories.
857 * This is a helper function to do the unlink of a specific directory
860 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
861 struct btrfs_root *root,
862 struct btrfs_path *path,
863 struct btrfs_inode *dir,
864 struct btrfs_dir_item *di)
869 struct extent_buffer *leaf;
870 struct btrfs_key location;
873 leaf = path->nodes[0];
875 btrfs_dir_item_key_to_cpu(leaf, di, &location);
876 name_len = btrfs_dir_name_len(leaf, di);
877 name = kmalloc(name_len, GFP_NOFS);
881 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
882 btrfs_release_path(path);
884 inode = read_one_inode(root, location.objectid);
890 ret = link_to_fixup_dir(trans, root, path, location.objectid);
894 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
899 ret = btrfs_run_delayed_items(trans);
907 * helper function to see if a given name and sequence number found
908 * in an inode back reference are already in a directory and correctly
909 * point to this inode
911 static noinline int inode_in_dir(struct btrfs_root *root,
912 struct btrfs_path *path,
913 u64 dirid, u64 objectid, u64 index,
914 const char *name, int name_len)
916 struct btrfs_dir_item *di;
917 struct btrfs_key location;
920 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
921 index, name, name_len, 0);
922 if (di && !IS_ERR(di)) {
923 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
924 if (location.objectid != objectid)
928 btrfs_release_path(path);
930 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
931 if (di && !IS_ERR(di)) {
932 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
933 if (location.objectid != objectid)
939 btrfs_release_path(path);
944 * helper function to check a log tree for a named back reference in
945 * an inode. This is used to decide if a back reference that is
946 * found in the subvolume conflicts with what we find in the log.
948 * inode backreferences may have multiple refs in a single item,
949 * during replay we process one reference at a time, and we don't
950 * want to delete valid links to a file from the subvolume if that
951 * link is also in the log.
953 static noinline int backref_in_log(struct btrfs_root *log,
954 struct btrfs_key *key,
956 const char *name, int namelen)
958 struct btrfs_path *path;
961 path = btrfs_alloc_path();
965 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
968 } else if (ret == 1) {
973 if (key->type == BTRFS_INODE_EXTREF_KEY)
974 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
979 ret = !!btrfs_find_name_in_backref(path->nodes[0],
983 btrfs_free_path(path);
987 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
989 struct btrfs_path *path,
990 struct btrfs_root *log_root,
991 struct btrfs_inode *dir,
992 struct btrfs_inode *inode,
993 u64 inode_objectid, u64 parent_objectid,
994 u64 ref_index, char *name, int namelen,
1000 struct extent_buffer *leaf;
1001 struct btrfs_dir_item *di;
1002 struct btrfs_key search_key;
1003 struct btrfs_inode_extref *extref;
1006 /* Search old style refs */
1007 search_key.objectid = inode_objectid;
1008 search_key.type = BTRFS_INODE_REF_KEY;
1009 search_key.offset = parent_objectid;
1010 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1012 struct btrfs_inode_ref *victim_ref;
1014 unsigned long ptr_end;
1016 leaf = path->nodes[0];
1018 /* are we trying to overwrite a back ref for the root directory
1019 * if so, just jump out, we're done
1021 if (search_key.objectid == search_key.offset)
1024 /* check all the names in this back reference to see
1025 * if they are in the log. if so, we allow them to stay
1026 * otherwise they must be unlinked as a conflict
1028 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1029 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1030 while (ptr < ptr_end) {
1031 victim_ref = (struct btrfs_inode_ref *)ptr;
1032 victim_name_len = btrfs_inode_ref_name_len(leaf,
1034 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1038 read_extent_buffer(leaf, victim_name,
1039 (unsigned long)(victim_ref + 1),
1042 ret = backref_in_log(log_root, &search_key,
1043 parent_objectid, victim_name,
1049 inc_nlink(&inode->vfs_inode);
1050 btrfs_release_path(path);
1052 ret = btrfs_unlink_inode(trans, root, dir, inode,
1053 victim_name, victim_name_len);
1057 ret = btrfs_run_delayed_items(trans);
1065 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1069 * NOTE: we have searched root tree and checked the
1070 * corresponding ref, it does not need to check again.
1074 btrfs_release_path(path);
1076 /* Same search but for extended refs */
1077 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1078 inode_objectid, parent_objectid, 0,
1080 if (!IS_ERR_OR_NULL(extref)) {
1084 struct inode *victim_parent;
1086 leaf = path->nodes[0];
1088 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1089 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1091 while (cur_offset < item_size) {
1092 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1094 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1096 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1099 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1102 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1105 search_key.objectid = inode_objectid;
1106 search_key.type = BTRFS_INODE_EXTREF_KEY;
1107 search_key.offset = btrfs_extref_hash(parent_objectid,
1110 ret = backref_in_log(log_root, &search_key,
1111 parent_objectid, victim_name,
1117 victim_parent = read_one_inode(root,
1119 if (victim_parent) {
1120 inc_nlink(&inode->vfs_inode);
1121 btrfs_release_path(path);
1123 ret = btrfs_unlink_inode(trans, root,
1124 BTRFS_I(victim_parent),
1129 ret = btrfs_run_delayed_items(
1132 iput(victim_parent);
1141 cur_offset += victim_name_len + sizeof(*extref);
1145 btrfs_release_path(path);
1147 /* look for a conflicting sequence number */
1148 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1149 ref_index, name, namelen, 0);
1150 if (di && !IS_ERR(di)) {
1151 ret = drop_one_dir_item(trans, root, path, dir, di);
1155 btrfs_release_path(path);
1157 /* look for a conflicting name */
1158 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1160 if (di && !IS_ERR(di)) {
1161 ret = drop_one_dir_item(trans, root, path, dir, di);
1165 btrfs_release_path(path);
1170 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1171 u32 *namelen, char **name, u64 *index,
1172 u64 *parent_objectid)
1174 struct btrfs_inode_extref *extref;
1176 extref = (struct btrfs_inode_extref *)ref_ptr;
1178 *namelen = btrfs_inode_extref_name_len(eb, extref);
1179 *name = kmalloc(*namelen, GFP_NOFS);
1183 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1187 *index = btrfs_inode_extref_index(eb, extref);
1188 if (parent_objectid)
1189 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1194 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1195 u32 *namelen, char **name, u64 *index)
1197 struct btrfs_inode_ref *ref;
1199 ref = (struct btrfs_inode_ref *)ref_ptr;
1201 *namelen = btrfs_inode_ref_name_len(eb, ref);
1202 *name = kmalloc(*namelen, GFP_NOFS);
1206 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1209 *index = btrfs_inode_ref_index(eb, ref);
1215 * Take an inode reference item from the log tree and iterate all names from the
1216 * inode reference item in the subvolume tree with the same key (if it exists).
1217 * For any name that is not in the inode reference item from the log tree, do a
1218 * proper unlink of that name (that is, remove its entry from the inode
1219 * reference item and both dir index keys).
1221 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1222 struct btrfs_root *root,
1223 struct btrfs_path *path,
1224 struct btrfs_inode *inode,
1225 struct extent_buffer *log_eb,
1227 struct btrfs_key *key)
1230 unsigned long ref_ptr;
1231 unsigned long ref_end;
1232 struct extent_buffer *eb;
1235 btrfs_release_path(path);
1236 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1244 eb = path->nodes[0];
1245 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1246 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1247 while (ref_ptr < ref_end) {
1252 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1253 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1256 parent_id = key->offset;
1257 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1263 if (key->type == BTRFS_INODE_EXTREF_KEY)
1264 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1268 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1274 btrfs_release_path(path);
1275 dir = read_one_inode(root, parent_id);
1281 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1282 inode, name, namelen);
1292 if (key->type == BTRFS_INODE_EXTREF_KEY)
1293 ref_ptr += sizeof(struct btrfs_inode_extref);
1295 ref_ptr += sizeof(struct btrfs_inode_ref);
1299 btrfs_release_path(path);
1303 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1304 const u8 ref_type, const char *name,
1307 struct btrfs_key key;
1308 struct btrfs_path *path;
1309 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1312 path = btrfs_alloc_path();
1316 key.objectid = btrfs_ino(BTRFS_I(inode));
1317 key.type = ref_type;
1318 if (key.type == BTRFS_INODE_REF_KEY)
1319 key.offset = parent_id;
1321 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1323 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1330 if (key.type == BTRFS_INODE_EXTREF_KEY)
1331 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1332 path->slots[0], parent_id, name, namelen);
1334 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1338 btrfs_free_path(path);
1342 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1343 struct inode *dir, struct inode *inode, const char *name,
1344 int namelen, u64 ref_index)
1346 struct btrfs_dir_item *dir_item;
1347 struct btrfs_key key;
1348 struct btrfs_path *path;
1349 struct inode *other_inode = NULL;
1352 path = btrfs_alloc_path();
1356 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1357 btrfs_ino(BTRFS_I(dir)),
1360 btrfs_release_path(path);
1362 } else if (IS_ERR(dir_item)) {
1363 ret = PTR_ERR(dir_item);
1368 * Our inode's dentry collides with the dentry of another inode which is
1369 * in the log but not yet processed since it has a higher inode number.
1370 * So delete that other dentry.
1372 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1373 btrfs_release_path(path);
1374 other_inode = read_one_inode(root, key.objectid);
1379 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1384 * If we dropped the link count to 0, bump it so that later the iput()
1385 * on the inode will not free it. We will fixup the link count later.
1387 if (other_inode->i_nlink == 0)
1388 inc_nlink(other_inode);
1390 ret = btrfs_run_delayed_items(trans);
1394 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1395 name, namelen, 0, ref_index);
1398 btrfs_free_path(path);
1404 * replay one inode back reference item found in the log tree.
1405 * eb, slot and key refer to the buffer and key found in the log tree.
1406 * root is the destination we are replaying into, and path is for temp
1407 * use by this function. (it should be released on return).
1409 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1410 struct btrfs_root *root,
1411 struct btrfs_root *log,
1412 struct btrfs_path *path,
1413 struct extent_buffer *eb, int slot,
1414 struct btrfs_key *key)
1416 struct inode *dir = NULL;
1417 struct inode *inode = NULL;
1418 unsigned long ref_ptr;
1419 unsigned long ref_end;
1423 int search_done = 0;
1424 int log_ref_ver = 0;
1425 u64 parent_objectid;
1428 int ref_struct_size;
1430 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1431 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1433 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1434 struct btrfs_inode_extref *r;
1436 ref_struct_size = sizeof(struct btrfs_inode_extref);
1438 r = (struct btrfs_inode_extref *)ref_ptr;
1439 parent_objectid = btrfs_inode_extref_parent(eb, r);
1441 ref_struct_size = sizeof(struct btrfs_inode_ref);
1442 parent_objectid = key->offset;
1444 inode_objectid = key->objectid;
1447 * it is possible that we didn't log all the parent directories
1448 * for a given inode. If we don't find the dir, just don't
1449 * copy the back ref in. The link count fixup code will take
1452 dir = read_one_inode(root, parent_objectid);
1458 inode = read_one_inode(root, inode_objectid);
1464 while (ref_ptr < ref_end) {
1466 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1467 &ref_index, &parent_objectid);
1469 * parent object can change from one array
1473 dir = read_one_inode(root, parent_objectid);
1479 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1485 /* if we already have a perfect match, we're done */
1486 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1487 btrfs_ino(BTRFS_I(inode)), ref_index,
1490 * look for a conflicting back reference in the
1491 * metadata. if we find one we have to unlink that name
1492 * of the file before we add our new link. Later on, we
1493 * overwrite any existing back reference, and we don't
1494 * want to create dangling pointers in the directory.
1498 ret = __add_inode_ref(trans, root, path, log,
1503 ref_index, name, namelen,
1513 * If a reference item already exists for this inode
1514 * with the same parent and name, but different index,
1515 * drop it and the corresponding directory index entries
1516 * from the parent before adding the new reference item
1517 * and dir index entries, otherwise we would fail with
1518 * -EEXIST returned from btrfs_add_link() below.
1520 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1523 ret = btrfs_unlink_inode(trans, root,
1528 * If we dropped the link count to 0, bump it so
1529 * that later the iput() on the inode will not
1530 * free it. We will fixup the link count later.
1532 if (!ret && inode->i_nlink == 0)
1538 /* insert our name */
1539 ret = add_link(trans, root, dir, inode, name, namelen,
1544 btrfs_update_inode(trans, root, BTRFS_I(inode));
1547 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1557 * Before we overwrite the inode reference item in the subvolume tree
1558 * with the item from the log tree, we must unlink all names from the
1559 * parent directory that are in the subvolume's tree inode reference
1560 * item, otherwise we end up with an inconsistent subvolume tree where
1561 * dir index entries exist for a name but there is no inode reference
1562 * item with the same name.
1564 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1569 /* finally write the back reference in the inode */
1570 ret = overwrite_item(trans, root, path, eb, slot, key);
1572 btrfs_release_path(path);
1579 static int count_inode_extrefs(struct btrfs_root *root,
1580 struct btrfs_inode *inode, struct btrfs_path *path)
1584 unsigned int nlink = 0;
1587 u64 inode_objectid = btrfs_ino(inode);
1590 struct btrfs_inode_extref *extref;
1591 struct extent_buffer *leaf;
1594 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1599 leaf = path->nodes[0];
1600 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1601 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1604 while (cur_offset < item_size) {
1605 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1606 name_len = btrfs_inode_extref_name_len(leaf, extref);
1610 cur_offset += name_len + sizeof(*extref);
1614 btrfs_release_path(path);
1616 btrfs_release_path(path);
1618 if (ret < 0 && ret != -ENOENT)
1623 static int count_inode_refs(struct btrfs_root *root,
1624 struct btrfs_inode *inode, struct btrfs_path *path)
1627 struct btrfs_key key;
1628 unsigned int nlink = 0;
1630 unsigned long ptr_end;
1632 u64 ino = btrfs_ino(inode);
1635 key.type = BTRFS_INODE_REF_KEY;
1636 key.offset = (u64)-1;
1639 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1643 if (path->slots[0] == 0)
1648 btrfs_item_key_to_cpu(path->nodes[0], &key,
1650 if (key.objectid != ino ||
1651 key.type != BTRFS_INODE_REF_KEY)
1653 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1654 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1656 while (ptr < ptr_end) {
1657 struct btrfs_inode_ref *ref;
1659 ref = (struct btrfs_inode_ref *)ptr;
1660 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1662 ptr = (unsigned long)(ref + 1) + name_len;
1666 if (key.offset == 0)
1668 if (path->slots[0] > 0) {
1673 btrfs_release_path(path);
1675 btrfs_release_path(path);
1681 * There are a few corners where the link count of the file can't
1682 * be properly maintained during replay. So, instead of adding
1683 * lots of complexity to the log code, we just scan the backrefs
1684 * for any file that has been through replay.
1686 * The scan will update the link count on the inode to reflect the
1687 * number of back refs found. If it goes down to zero, the iput
1688 * will free the inode.
1690 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1691 struct btrfs_root *root,
1692 struct inode *inode)
1694 struct btrfs_path *path;
1697 u64 ino = btrfs_ino(BTRFS_I(inode));
1699 path = btrfs_alloc_path();
1703 ret = count_inode_refs(root, BTRFS_I(inode), path);
1709 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1717 if (nlink != inode->i_nlink) {
1718 set_nlink(inode, nlink);
1719 btrfs_update_inode(trans, root, BTRFS_I(inode));
1721 BTRFS_I(inode)->index_cnt = (u64)-1;
1723 if (inode->i_nlink == 0) {
1724 if (S_ISDIR(inode->i_mode)) {
1725 ret = replay_dir_deletes(trans, root, NULL, path,
1730 ret = btrfs_insert_orphan_item(trans, root, ino);
1736 btrfs_free_path(path);
1740 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path)
1745 struct btrfs_key key;
1746 struct inode *inode;
1748 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1749 key.type = BTRFS_ORPHAN_ITEM_KEY;
1750 key.offset = (u64)-1;
1752 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1757 if (path->slots[0] == 0)
1762 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1763 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1764 key.type != BTRFS_ORPHAN_ITEM_KEY)
1767 ret = btrfs_del_item(trans, root, path);
1771 btrfs_release_path(path);
1772 inode = read_one_inode(root, key.offset);
1776 ret = fixup_inode_link_count(trans, root, inode);
1782 * fixup on a directory may create new entries,
1783 * make sure we always look for the highset possible
1786 key.offset = (u64)-1;
1790 btrfs_release_path(path);
1796 * record a given inode in the fixup dir so we can check its link
1797 * count when replay is done. The link count is incremented here
1798 * so the inode won't go away until we check it
1800 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1805 struct btrfs_key key;
1807 struct inode *inode;
1809 inode = read_one_inode(root, objectid);
1813 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1814 key.type = BTRFS_ORPHAN_ITEM_KEY;
1815 key.offset = objectid;
1817 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1819 btrfs_release_path(path);
1821 if (!inode->i_nlink)
1822 set_nlink(inode, 1);
1825 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1826 } else if (ret == -EEXIST) {
1829 BUG(); /* Logic Error */
1837 * when replaying the log for a directory, we only insert names
1838 * for inodes that actually exist. This means an fsync on a directory
1839 * does not implicitly fsync all the new files in it
1841 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1842 struct btrfs_root *root,
1843 u64 dirid, u64 index,
1844 char *name, int name_len,
1845 struct btrfs_key *location)
1847 struct inode *inode;
1851 inode = read_one_inode(root, location->objectid);
1855 dir = read_one_inode(root, dirid);
1861 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1862 name_len, 1, index);
1864 /* FIXME, put inode into FIXUP list */
1872 * take a single entry in a log directory item and replay it into
1875 * if a conflicting item exists in the subdirectory already,
1876 * the inode it points to is unlinked and put into the link count
1879 * If a name from the log points to a file or directory that does
1880 * not exist in the FS, it is skipped. fsyncs on directories
1881 * do not force down inodes inside that directory, just changes to the
1882 * names or unlinks in a directory.
1884 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1885 * non-existing inode) and 1 if the name was replayed.
1887 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1888 struct btrfs_root *root,
1889 struct btrfs_path *path,
1890 struct extent_buffer *eb,
1891 struct btrfs_dir_item *di,
1892 struct btrfs_key *key)
1896 struct btrfs_dir_item *dst_di;
1897 struct btrfs_key found_key;
1898 struct btrfs_key log_key;
1903 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1904 bool name_added = false;
1906 dir = read_one_inode(root, key->objectid);
1910 name_len = btrfs_dir_name_len(eb, di);
1911 name = kmalloc(name_len, GFP_NOFS);
1917 log_type = btrfs_dir_type(eb, di);
1918 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1921 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1922 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1927 btrfs_release_path(path);
1929 if (key->type == BTRFS_DIR_ITEM_KEY) {
1930 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1932 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1933 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1942 if (IS_ERR_OR_NULL(dst_di)) {
1943 /* we need a sequence number to insert, so we only
1944 * do inserts for the BTRFS_DIR_INDEX_KEY types
1946 if (key->type != BTRFS_DIR_INDEX_KEY)
1951 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1952 /* the existing item matches the logged item */
1953 if (found_key.objectid == log_key.objectid &&
1954 found_key.type == log_key.type &&
1955 found_key.offset == log_key.offset &&
1956 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1957 update_size = false;
1962 * don't drop the conflicting directory entry if the inode
1963 * for the new entry doesn't exist
1968 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1972 if (key->type == BTRFS_DIR_INDEX_KEY)
1975 btrfs_release_path(path);
1976 if (!ret && update_size) {
1977 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1978 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
1982 if (!ret && name_added)
1988 * Check if the inode reference exists in the log for the given name,
1989 * inode and parent inode
1991 found_key.objectid = log_key.objectid;
1992 found_key.type = BTRFS_INODE_REF_KEY;
1993 found_key.offset = key->objectid;
1994 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
1998 /* The dentry will be added later. */
2000 update_size = false;
2004 found_key.objectid = log_key.objectid;
2005 found_key.type = BTRFS_INODE_EXTREF_KEY;
2006 found_key.offset = key->objectid;
2007 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2012 /* The dentry will be added later. */
2014 update_size = false;
2017 btrfs_release_path(path);
2018 ret = insert_one_name(trans, root, key->objectid, key->offset,
2019 name, name_len, &log_key);
2020 if (ret && ret != -ENOENT && ret != -EEXIST)
2024 update_size = false;
2030 * find all the names in a directory item and reconcile them into
2031 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2032 * one name in a directory item, but the same code gets used for
2033 * both directory index types
2035 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2036 struct btrfs_root *root,
2037 struct btrfs_path *path,
2038 struct extent_buffer *eb, int slot,
2039 struct btrfs_key *key)
2042 u32 item_size = btrfs_item_size_nr(eb, slot);
2043 struct btrfs_dir_item *di;
2046 unsigned long ptr_end;
2047 struct btrfs_path *fixup_path = NULL;
2049 ptr = btrfs_item_ptr_offset(eb, slot);
2050 ptr_end = ptr + item_size;
2051 while (ptr < ptr_end) {
2052 di = (struct btrfs_dir_item *)ptr;
2053 name_len = btrfs_dir_name_len(eb, di);
2054 ret = replay_one_name(trans, root, path, eb, di, key);
2057 ptr = (unsigned long)(di + 1);
2061 * If this entry refers to a non-directory (directories can not
2062 * have a link count > 1) and it was added in the transaction
2063 * that was not committed, make sure we fixup the link count of
2064 * the inode it the entry points to. Otherwise something like
2065 * the following would result in a directory pointing to an
2066 * inode with a wrong link that does not account for this dir
2074 * ln testdir/bar testdir/bar_link
2075 * ln testdir/foo testdir/foo_link
2076 * xfs_io -c "fsync" testdir/bar
2080 * mount fs, log replay happens
2082 * File foo would remain with a link count of 1 when it has two
2083 * entries pointing to it in the directory testdir. This would
2084 * make it impossible to ever delete the parent directory has
2085 * it would result in stale dentries that can never be deleted.
2087 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2088 struct btrfs_key di_key;
2091 fixup_path = btrfs_alloc_path();
2098 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2099 ret = link_to_fixup_dir(trans, root, fixup_path,
2106 btrfs_free_path(fixup_path);
2111 * directory replay has two parts. There are the standard directory
2112 * items in the log copied from the subvolume, and range items
2113 * created in the log while the subvolume was logged.
2115 * The range items tell us which parts of the key space the log
2116 * is authoritative for. During replay, if a key in the subvolume
2117 * directory is in a logged range item, but not actually in the log
2118 * that means it was deleted from the directory before the fsync
2119 * and should be removed.
2121 static noinline int find_dir_range(struct btrfs_root *root,
2122 struct btrfs_path *path,
2123 u64 dirid, int key_type,
2124 u64 *start_ret, u64 *end_ret)
2126 struct btrfs_key key;
2128 struct btrfs_dir_log_item *item;
2132 if (*start_ret == (u64)-1)
2135 key.objectid = dirid;
2136 key.type = key_type;
2137 key.offset = *start_ret;
2139 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2143 if (path->slots[0] == 0)
2148 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2150 if (key.type != key_type || key.objectid != dirid) {
2154 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2155 struct btrfs_dir_log_item);
2156 found_end = btrfs_dir_log_end(path->nodes[0], item);
2158 if (*start_ret >= key.offset && *start_ret <= found_end) {
2160 *start_ret = key.offset;
2161 *end_ret = found_end;
2166 /* check the next slot in the tree to see if it is a valid item */
2167 nritems = btrfs_header_nritems(path->nodes[0]);
2169 if (path->slots[0] >= nritems) {
2170 ret = btrfs_next_leaf(root, path);
2175 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2177 if (key.type != key_type || key.objectid != dirid) {
2181 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2182 struct btrfs_dir_log_item);
2183 found_end = btrfs_dir_log_end(path->nodes[0], item);
2184 *start_ret = key.offset;
2185 *end_ret = found_end;
2188 btrfs_release_path(path);
2193 * this looks for a given directory item in the log. If the directory
2194 * item is not in the log, the item is removed and the inode it points
2197 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2198 struct btrfs_root *root,
2199 struct btrfs_root *log,
2200 struct btrfs_path *path,
2201 struct btrfs_path *log_path,
2203 struct btrfs_key *dir_key)
2206 struct extent_buffer *eb;
2209 struct btrfs_dir_item *di;
2210 struct btrfs_dir_item *log_di;
2213 unsigned long ptr_end;
2215 struct inode *inode;
2216 struct btrfs_key location;
2219 eb = path->nodes[0];
2220 slot = path->slots[0];
2221 item_size = btrfs_item_size_nr(eb, slot);
2222 ptr = btrfs_item_ptr_offset(eb, slot);
2223 ptr_end = ptr + item_size;
2224 while (ptr < ptr_end) {
2225 di = (struct btrfs_dir_item *)ptr;
2226 name_len = btrfs_dir_name_len(eb, di);
2227 name = kmalloc(name_len, GFP_NOFS);
2232 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2235 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2236 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2239 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2240 log_di = btrfs_lookup_dir_index_item(trans, log,
2246 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2247 btrfs_dir_item_key_to_cpu(eb, di, &location);
2248 btrfs_release_path(path);
2249 btrfs_release_path(log_path);
2250 inode = read_one_inode(root, location.objectid);
2256 ret = link_to_fixup_dir(trans, root,
2257 path, location.objectid);
2265 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2266 BTRFS_I(inode), name, name_len);
2268 ret = btrfs_run_delayed_items(trans);
2274 /* there might still be more names under this key
2275 * check and repeat if required
2277 ret = btrfs_search_slot(NULL, root, dir_key, path,
2283 } else if (IS_ERR(log_di)) {
2285 return PTR_ERR(log_di);
2287 btrfs_release_path(log_path);
2290 ptr = (unsigned long)(di + 1);
2295 btrfs_release_path(path);
2296 btrfs_release_path(log_path);
2300 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2301 struct btrfs_root *root,
2302 struct btrfs_root *log,
2303 struct btrfs_path *path,
2306 struct btrfs_key search_key;
2307 struct btrfs_path *log_path;
2312 log_path = btrfs_alloc_path();
2316 search_key.objectid = ino;
2317 search_key.type = BTRFS_XATTR_ITEM_KEY;
2318 search_key.offset = 0;
2320 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2324 nritems = btrfs_header_nritems(path->nodes[0]);
2325 for (i = path->slots[0]; i < nritems; i++) {
2326 struct btrfs_key key;
2327 struct btrfs_dir_item *di;
2328 struct btrfs_dir_item *log_di;
2332 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2333 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2338 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2339 total_size = btrfs_item_size_nr(path->nodes[0], i);
2341 while (cur < total_size) {
2342 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2343 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2344 u32 this_len = sizeof(*di) + name_len + data_len;
2347 name = kmalloc(name_len, GFP_NOFS);
2352 read_extent_buffer(path->nodes[0], name,
2353 (unsigned long)(di + 1), name_len);
2355 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2357 btrfs_release_path(log_path);
2359 /* Doesn't exist in log tree, so delete it. */
2360 btrfs_release_path(path);
2361 di = btrfs_lookup_xattr(trans, root, path, ino,
2362 name, name_len, -1);
2369 ret = btrfs_delete_one_dir_name(trans, root,
2373 btrfs_release_path(path);
2378 if (IS_ERR(log_di)) {
2379 ret = PTR_ERR(log_di);
2383 di = (struct btrfs_dir_item *)((char *)di + this_len);
2386 ret = btrfs_next_leaf(root, path);
2392 btrfs_free_path(log_path);
2393 btrfs_release_path(path);
2399 * deletion replay happens before we copy any new directory items
2400 * out of the log or out of backreferences from inodes. It
2401 * scans the log to find ranges of keys that log is authoritative for,
2402 * and then scans the directory to find items in those ranges that are
2403 * not present in the log.
2405 * Anything we don't find in the log is unlinked and removed from the
2408 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2409 struct btrfs_root *root,
2410 struct btrfs_root *log,
2411 struct btrfs_path *path,
2412 u64 dirid, int del_all)
2416 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2418 struct btrfs_key dir_key;
2419 struct btrfs_key found_key;
2420 struct btrfs_path *log_path;
2423 dir_key.objectid = dirid;
2424 dir_key.type = BTRFS_DIR_ITEM_KEY;
2425 log_path = btrfs_alloc_path();
2429 dir = read_one_inode(root, dirid);
2430 /* it isn't an error if the inode isn't there, that can happen
2431 * because we replay the deletes before we copy in the inode item
2435 btrfs_free_path(log_path);
2443 range_end = (u64)-1;
2445 ret = find_dir_range(log, path, dirid, key_type,
2446 &range_start, &range_end);
2451 dir_key.offset = range_start;
2454 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2459 nritems = btrfs_header_nritems(path->nodes[0]);
2460 if (path->slots[0] >= nritems) {
2461 ret = btrfs_next_leaf(root, path);
2467 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2469 if (found_key.objectid != dirid ||
2470 found_key.type != dir_key.type)
2473 if (found_key.offset > range_end)
2476 ret = check_item_in_log(trans, root, log, path,
2481 if (found_key.offset == (u64)-1)
2483 dir_key.offset = found_key.offset + 1;
2485 btrfs_release_path(path);
2486 if (range_end == (u64)-1)
2488 range_start = range_end + 1;
2493 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2494 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2495 dir_key.type = BTRFS_DIR_INDEX_KEY;
2496 btrfs_release_path(path);
2500 btrfs_release_path(path);
2501 btrfs_free_path(log_path);
2507 * the process_func used to replay items from the log tree. This
2508 * gets called in two different stages. The first stage just looks
2509 * for inodes and makes sure they are all copied into the subvolume.
2511 * The second stage copies all the other item types from the log into
2512 * the subvolume. The two stage approach is slower, but gets rid of
2513 * lots of complexity around inodes referencing other inodes that exist
2514 * only in the log (references come from either directory items or inode
2517 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2518 struct walk_control *wc, u64 gen, int level)
2521 struct btrfs_path *path;
2522 struct btrfs_root *root = wc->replay_dest;
2523 struct btrfs_key key;
2527 ret = btrfs_read_buffer(eb, gen, level, NULL);
2531 level = btrfs_header_level(eb);
2536 path = btrfs_alloc_path();
2540 nritems = btrfs_header_nritems(eb);
2541 for (i = 0; i < nritems; i++) {
2542 btrfs_item_key_to_cpu(eb, &key, i);
2544 /* inode keys are done during the first stage */
2545 if (key.type == BTRFS_INODE_ITEM_KEY &&
2546 wc->stage == LOG_WALK_REPLAY_INODES) {
2547 struct btrfs_inode_item *inode_item;
2550 inode_item = btrfs_item_ptr(eb, i,
2551 struct btrfs_inode_item);
2553 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2554 * and never got linked before the fsync, skip it, as
2555 * replaying it is pointless since it would be deleted
2556 * later. We skip logging tmpfiles, but it's always
2557 * possible we are replaying a log created with a kernel
2558 * that used to log tmpfiles.
2560 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2561 wc->ignore_cur_inode = true;
2564 wc->ignore_cur_inode = false;
2566 ret = replay_xattr_deletes(wc->trans, root, log,
2567 path, key.objectid);
2570 mode = btrfs_inode_mode(eb, inode_item);
2571 if (S_ISDIR(mode)) {
2572 ret = replay_dir_deletes(wc->trans,
2573 root, log, path, key.objectid, 0);
2577 ret = overwrite_item(wc->trans, root, path,
2583 * Before replaying extents, truncate the inode to its
2584 * size. We need to do it now and not after log replay
2585 * because before an fsync we can have prealloc extents
2586 * added beyond the inode's i_size. If we did it after,
2587 * through orphan cleanup for example, we would drop
2588 * those prealloc extents just after replaying them.
2590 if (S_ISREG(mode)) {
2591 struct btrfs_drop_extents_args drop_args = { 0 };
2592 struct inode *inode;
2595 inode = read_one_inode(root, key.objectid);
2600 from = ALIGN(i_size_read(inode),
2601 root->fs_info->sectorsize);
2602 drop_args.start = from;
2603 drop_args.end = (u64)-1;
2604 drop_args.drop_cache = true;
2605 ret = btrfs_drop_extents(wc->trans, root,
2609 inode_sub_bytes(inode,
2610 drop_args.bytes_found);
2611 /* Update the inode's nbytes. */
2612 ret = btrfs_update_inode(wc->trans,
2613 root, BTRFS_I(inode));
2620 ret = link_to_fixup_dir(wc->trans, root,
2621 path, key.objectid);
2626 if (wc->ignore_cur_inode)
2629 if (key.type == BTRFS_DIR_INDEX_KEY &&
2630 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2631 ret = replay_one_dir_item(wc->trans, root, path,
2637 if (wc->stage < LOG_WALK_REPLAY_ALL)
2640 /* these keys are simply copied */
2641 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2642 ret = overwrite_item(wc->trans, root, path,
2646 } else if (key.type == BTRFS_INODE_REF_KEY ||
2647 key.type == BTRFS_INODE_EXTREF_KEY) {
2648 ret = add_inode_ref(wc->trans, root, log, path,
2650 if (ret && ret != -ENOENT)
2653 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2654 ret = replay_one_extent(wc->trans, root, path,
2658 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2659 ret = replay_one_dir_item(wc->trans, root, path,
2665 btrfs_free_path(path);
2670 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2672 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2674 struct btrfs_block_group *cache;
2676 cache = btrfs_lookup_block_group(fs_info, start);
2678 btrfs_err(fs_info, "unable to find block group for %llu", start);
2682 spin_lock(&cache->space_info->lock);
2683 spin_lock(&cache->lock);
2684 cache->reserved -= fs_info->nodesize;
2685 cache->space_info->bytes_reserved -= fs_info->nodesize;
2686 spin_unlock(&cache->lock);
2687 spin_unlock(&cache->space_info->lock);
2689 btrfs_put_block_group(cache);
2692 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2693 struct btrfs_root *root,
2694 struct btrfs_path *path, int *level,
2695 struct walk_control *wc)
2697 struct btrfs_fs_info *fs_info = root->fs_info;
2700 struct extent_buffer *next;
2701 struct extent_buffer *cur;
2705 while (*level > 0) {
2706 struct btrfs_key first_key;
2708 cur = path->nodes[*level];
2710 WARN_ON(btrfs_header_level(cur) != *level);
2712 if (path->slots[*level] >=
2713 btrfs_header_nritems(cur))
2716 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2717 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2718 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2719 blocksize = fs_info->nodesize;
2721 next = btrfs_find_create_tree_block(fs_info, bytenr,
2722 btrfs_header_owner(cur),
2725 return PTR_ERR(next);
2728 ret = wc->process_func(root, next, wc, ptr_gen,
2731 free_extent_buffer(next);
2735 path->slots[*level]++;
2737 ret = btrfs_read_buffer(next, ptr_gen,
2738 *level - 1, &first_key);
2740 free_extent_buffer(next);
2745 btrfs_tree_lock(next);
2746 btrfs_clean_tree_block(next);
2747 btrfs_wait_tree_block_writeback(next);
2748 btrfs_tree_unlock(next);
2749 ret = btrfs_pin_reserved_extent(trans,
2752 free_extent_buffer(next);
2756 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2757 clear_extent_buffer_dirty(next);
2758 unaccount_log_buffer(fs_info, bytenr);
2761 free_extent_buffer(next);
2764 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2766 free_extent_buffer(next);
2770 if (path->nodes[*level-1])
2771 free_extent_buffer(path->nodes[*level-1]);
2772 path->nodes[*level-1] = next;
2773 *level = btrfs_header_level(next);
2774 path->slots[*level] = 0;
2777 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2783 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root,
2785 struct btrfs_path *path, int *level,
2786 struct walk_control *wc)
2788 struct btrfs_fs_info *fs_info = root->fs_info;
2793 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2794 slot = path->slots[i];
2795 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2798 WARN_ON(*level == 0);
2801 ret = wc->process_func(root, path->nodes[*level], wc,
2802 btrfs_header_generation(path->nodes[*level]),
2808 struct extent_buffer *next;
2810 next = path->nodes[*level];
2813 btrfs_tree_lock(next);
2814 btrfs_clean_tree_block(next);
2815 btrfs_wait_tree_block_writeback(next);
2816 btrfs_tree_unlock(next);
2817 ret = btrfs_pin_reserved_extent(trans,
2818 path->nodes[*level]->start,
2819 path->nodes[*level]->len);
2823 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2824 clear_extent_buffer_dirty(next);
2826 unaccount_log_buffer(fs_info,
2827 path->nodes[*level]->start);
2830 free_extent_buffer(path->nodes[*level]);
2831 path->nodes[*level] = NULL;
2839 * drop the reference count on the tree rooted at 'snap'. This traverses
2840 * the tree freeing any blocks that have a ref count of zero after being
2843 static int walk_log_tree(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *log, struct walk_control *wc)
2846 struct btrfs_fs_info *fs_info = log->fs_info;
2850 struct btrfs_path *path;
2853 path = btrfs_alloc_path();
2857 level = btrfs_header_level(log->node);
2859 path->nodes[level] = log->node;
2860 atomic_inc(&log->node->refs);
2861 path->slots[level] = 0;
2864 wret = walk_down_log_tree(trans, log, path, &level, wc);
2872 wret = walk_up_log_tree(trans, log, path, &level, wc);
2881 /* was the root node processed? if not, catch it here */
2882 if (path->nodes[orig_level]) {
2883 ret = wc->process_func(log, path->nodes[orig_level], wc,
2884 btrfs_header_generation(path->nodes[orig_level]),
2889 struct extent_buffer *next;
2891 next = path->nodes[orig_level];
2894 btrfs_tree_lock(next);
2895 btrfs_clean_tree_block(next);
2896 btrfs_wait_tree_block_writeback(next);
2897 btrfs_tree_unlock(next);
2898 ret = btrfs_pin_reserved_extent(trans,
2899 next->start, next->len);
2903 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2904 clear_extent_buffer_dirty(next);
2905 unaccount_log_buffer(fs_info, next->start);
2911 btrfs_free_path(path);
2916 * helper function to update the item for a given subvolumes log root
2917 * in the tree of log roots
2919 static int update_log_root(struct btrfs_trans_handle *trans,
2920 struct btrfs_root *log,
2921 struct btrfs_root_item *root_item)
2923 struct btrfs_fs_info *fs_info = log->fs_info;
2926 if (log->log_transid == 1) {
2927 /* insert root item on the first sync */
2928 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2929 &log->root_key, root_item);
2931 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2932 &log->root_key, root_item);
2937 static void wait_log_commit(struct btrfs_root *root, int transid)
2940 int index = transid % 2;
2943 * we only allow two pending log transactions at a time,
2944 * so we know that if ours is more than 2 older than the
2945 * current transaction, we're done
2948 prepare_to_wait(&root->log_commit_wait[index],
2949 &wait, TASK_UNINTERRUPTIBLE);
2951 if (!(root->log_transid_committed < transid &&
2952 atomic_read(&root->log_commit[index])))
2955 mutex_unlock(&root->log_mutex);
2957 mutex_lock(&root->log_mutex);
2959 finish_wait(&root->log_commit_wait[index], &wait);
2962 static void wait_for_writer(struct btrfs_root *root)
2967 prepare_to_wait(&root->log_writer_wait, &wait,
2968 TASK_UNINTERRUPTIBLE);
2969 if (!atomic_read(&root->log_writers))
2972 mutex_unlock(&root->log_mutex);
2974 mutex_lock(&root->log_mutex);
2976 finish_wait(&root->log_writer_wait, &wait);
2979 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2980 struct btrfs_log_ctx *ctx)
2985 mutex_lock(&root->log_mutex);
2986 list_del_init(&ctx->list);
2987 mutex_unlock(&root->log_mutex);
2991 * Invoked in log mutex context, or be sure there is no other task which
2992 * can access the list.
2994 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2995 int index, int error)
2997 struct btrfs_log_ctx *ctx;
2998 struct btrfs_log_ctx *safe;
3000 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3001 list_del_init(&ctx->list);
3002 ctx->log_ret = error;
3005 INIT_LIST_HEAD(&root->log_ctxs[index]);
3009 * btrfs_sync_log does sends a given tree log down to the disk and
3010 * updates the super blocks to record it. When this call is done,
3011 * you know that any inodes previously logged are safely on disk only
3014 * Any other return value means you need to call btrfs_commit_transaction.
3015 * Some of the edge cases for fsyncing directories that have had unlinks
3016 * or renames done in the past mean that sometimes the only safe
3017 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3018 * that has happened.
3020 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3021 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3027 struct btrfs_fs_info *fs_info = root->fs_info;
3028 struct btrfs_root *log = root->log_root;
3029 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3030 struct btrfs_root_item new_root_item;
3031 int log_transid = 0;
3032 struct btrfs_log_ctx root_log_ctx;
3033 struct blk_plug plug;
3037 mutex_lock(&root->log_mutex);
3038 log_transid = ctx->log_transid;
3039 if (root->log_transid_committed >= log_transid) {
3040 mutex_unlock(&root->log_mutex);
3041 return ctx->log_ret;
3044 index1 = log_transid % 2;
3045 if (atomic_read(&root->log_commit[index1])) {
3046 wait_log_commit(root, log_transid);
3047 mutex_unlock(&root->log_mutex);
3048 return ctx->log_ret;
3050 ASSERT(log_transid == root->log_transid);
3051 atomic_set(&root->log_commit[index1], 1);
3053 /* wait for previous tree log sync to complete */
3054 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3055 wait_log_commit(root, log_transid - 1);
3058 int batch = atomic_read(&root->log_batch);
3059 /* when we're on an ssd, just kick the log commit out */
3060 if (!btrfs_test_opt(fs_info, SSD) &&
3061 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3062 mutex_unlock(&root->log_mutex);
3063 schedule_timeout_uninterruptible(1);
3064 mutex_lock(&root->log_mutex);
3066 wait_for_writer(root);
3067 if (batch == atomic_read(&root->log_batch))
3071 /* bail out if we need to do a full commit */
3072 if (btrfs_need_log_full_commit(trans)) {
3074 mutex_unlock(&root->log_mutex);
3078 if (log_transid % 2 == 0)
3079 mark = EXTENT_DIRTY;
3083 /* we start IO on all the marked extents here, but we don't actually
3084 * wait for them until later.
3086 blk_start_plug(&plug);
3087 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3089 blk_finish_plug(&plug);
3090 btrfs_abort_transaction(trans, ret);
3091 btrfs_set_log_full_commit(trans);
3092 mutex_unlock(&root->log_mutex);
3097 * We _must_ update under the root->log_mutex in order to make sure we
3098 * have a consistent view of the log root we are trying to commit at
3101 * We _must_ copy this into a local copy, because we are not holding the
3102 * log_root_tree->log_mutex yet. This is important because when we
3103 * commit the log_root_tree we must have a consistent view of the
3104 * log_root_tree when we update the super block to point at the
3105 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3106 * with the commit and possibly point at the new block which we may not
3109 btrfs_set_root_node(&log->root_item, log->node);
3110 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3112 root->log_transid++;
3113 log->log_transid = root->log_transid;
3114 root->log_start_pid = 0;
3116 * IO has been started, blocks of the log tree have WRITTEN flag set
3117 * in their headers. new modifications of the log will be written to
3118 * new positions. so it's safe to allow log writers to go in.
3120 mutex_unlock(&root->log_mutex);
3122 btrfs_init_log_ctx(&root_log_ctx, NULL);
3124 mutex_lock(&log_root_tree->log_mutex);
3126 index2 = log_root_tree->log_transid % 2;
3127 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3128 root_log_ctx.log_transid = log_root_tree->log_transid;
3131 * Now we are safe to update the log_root_tree because we're under the
3132 * log_mutex, and we're a current writer so we're holding the commit
3133 * open until we drop the log_mutex.
3135 ret = update_log_root(trans, log, &new_root_item);
3137 if (!list_empty(&root_log_ctx.list))
3138 list_del_init(&root_log_ctx.list);
3140 blk_finish_plug(&plug);
3141 btrfs_set_log_full_commit(trans);
3143 if (ret != -ENOSPC) {
3144 btrfs_abort_transaction(trans, ret);
3145 mutex_unlock(&log_root_tree->log_mutex);
3148 btrfs_wait_tree_log_extents(log, mark);
3149 mutex_unlock(&log_root_tree->log_mutex);
3154 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3155 blk_finish_plug(&plug);
3156 list_del_init(&root_log_ctx.list);
3157 mutex_unlock(&log_root_tree->log_mutex);
3158 ret = root_log_ctx.log_ret;
3162 index2 = root_log_ctx.log_transid % 2;
3163 if (atomic_read(&log_root_tree->log_commit[index2])) {
3164 blk_finish_plug(&plug);
3165 ret = btrfs_wait_tree_log_extents(log, mark);
3166 wait_log_commit(log_root_tree,
3167 root_log_ctx.log_transid);
3168 mutex_unlock(&log_root_tree->log_mutex);
3170 ret = root_log_ctx.log_ret;
3173 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3174 atomic_set(&log_root_tree->log_commit[index2], 1);
3176 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3177 wait_log_commit(log_root_tree,
3178 root_log_ctx.log_transid - 1);
3182 * now that we've moved on to the tree of log tree roots,
3183 * check the full commit flag again
3185 if (btrfs_need_log_full_commit(trans)) {
3186 blk_finish_plug(&plug);
3187 btrfs_wait_tree_log_extents(log, mark);
3188 mutex_unlock(&log_root_tree->log_mutex);
3190 goto out_wake_log_root;
3193 ret = btrfs_write_marked_extents(fs_info,
3194 &log_root_tree->dirty_log_pages,
3195 EXTENT_DIRTY | EXTENT_NEW);
3196 blk_finish_plug(&plug);
3198 btrfs_set_log_full_commit(trans);
3199 btrfs_abort_transaction(trans, ret);
3200 mutex_unlock(&log_root_tree->log_mutex);
3201 goto out_wake_log_root;
3203 ret = btrfs_wait_tree_log_extents(log, mark);
3205 ret = btrfs_wait_tree_log_extents(log_root_tree,
3206 EXTENT_NEW | EXTENT_DIRTY);
3208 btrfs_set_log_full_commit(trans);
3209 mutex_unlock(&log_root_tree->log_mutex);
3210 goto out_wake_log_root;
3213 log_root_start = log_root_tree->node->start;
3214 log_root_level = btrfs_header_level(log_root_tree->node);
3215 log_root_tree->log_transid++;
3216 mutex_unlock(&log_root_tree->log_mutex);
3219 * Here we are guaranteed that nobody is going to write the superblock
3220 * for the current transaction before us and that neither we do write
3221 * our superblock before the previous transaction finishes its commit
3222 * and writes its superblock, because:
3224 * 1) We are holding a handle on the current transaction, so no body
3225 * can commit it until we release the handle;
3227 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3228 * if the previous transaction is still committing, and hasn't yet
3229 * written its superblock, we wait for it to do it, because a
3230 * transaction commit acquires the tree_log_mutex when the commit
3231 * begins and releases it only after writing its superblock.
3233 mutex_lock(&fs_info->tree_log_mutex);
3234 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3235 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3236 ret = write_all_supers(fs_info, 1);
3237 mutex_unlock(&fs_info->tree_log_mutex);
3239 btrfs_set_log_full_commit(trans);
3240 btrfs_abort_transaction(trans, ret);
3241 goto out_wake_log_root;
3244 mutex_lock(&root->log_mutex);
3245 if (root->last_log_commit < log_transid)
3246 root->last_log_commit = log_transid;
3247 mutex_unlock(&root->log_mutex);
3250 mutex_lock(&log_root_tree->log_mutex);
3251 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3253 log_root_tree->log_transid_committed++;
3254 atomic_set(&log_root_tree->log_commit[index2], 0);
3255 mutex_unlock(&log_root_tree->log_mutex);
3258 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3259 * all the updates above are seen by the woken threads. It might not be
3260 * necessary, but proving that seems to be hard.
3262 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3264 mutex_lock(&root->log_mutex);
3265 btrfs_remove_all_log_ctxs(root, index1, ret);
3266 root->log_transid_committed++;
3267 atomic_set(&root->log_commit[index1], 0);
3268 mutex_unlock(&root->log_mutex);
3271 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3272 * all the updates above are seen by the woken threads. It might not be
3273 * necessary, but proving that seems to be hard.
3275 cond_wake_up(&root->log_commit_wait[index1]);
3279 static void free_log_tree(struct btrfs_trans_handle *trans,
3280 struct btrfs_root *log)
3283 struct walk_control wc = {
3285 .process_func = process_one_buffer
3288 ret = walk_log_tree(trans, log, &wc);
3291 btrfs_abort_transaction(trans, ret);
3293 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3296 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3297 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3298 extent_io_tree_release(&log->log_csum_range);
3299 btrfs_put_root(log);
3303 * free all the extents used by the tree log. This should be called
3304 * at commit time of the full transaction
3306 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3308 if (root->log_root) {
3309 free_log_tree(trans, root->log_root);
3310 root->log_root = NULL;
3311 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3316 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3317 struct btrfs_fs_info *fs_info)
3319 if (fs_info->log_root_tree) {
3320 free_log_tree(trans, fs_info->log_root_tree);
3321 fs_info->log_root_tree = NULL;
3322 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3328 * Check if an inode was logged in the current transaction. We can't always rely
3329 * on an inode's logged_trans value, because it's an in-memory only field and
3330 * therefore not persisted. This means that its value is lost if the inode gets
3331 * evicted and loaded again from disk (in which case it has a value of 0, and
3332 * certainly it is smaller then any possible transaction ID), when that happens
3333 * the full_sync flag is set in the inode's runtime flags, so on that case we
3334 * assume eviction happened and ignore the logged_trans value, assuming the
3335 * worst case, that the inode was logged before in the current transaction.
3337 static bool inode_logged(struct btrfs_trans_handle *trans,
3338 struct btrfs_inode *inode)
3340 if (inode->logged_trans == trans->transid)
3343 if (inode->last_trans == trans->transid &&
3344 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3345 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3352 * If both a file and directory are logged, and unlinks or renames are
3353 * mixed in, we have a few interesting corners:
3355 * create file X in dir Y
3356 * link file X to X.link in dir Y
3358 * unlink file X but leave X.link
3361 * After a crash we would expect only X.link to exist. But file X
3362 * didn't get fsync'd again so the log has back refs for X and X.link.
3364 * We solve this by removing directory entries and inode backrefs from the
3365 * log when a file that was logged in the current transaction is
3366 * unlinked. Any later fsync will include the updated log entries, and
3367 * we'll be able to reconstruct the proper directory items from backrefs.
3369 * This optimizations allows us to avoid relogging the entire inode
3370 * or the entire directory.
3372 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3373 struct btrfs_root *root,
3374 const char *name, int name_len,
3375 struct btrfs_inode *dir, u64 index)
3377 struct btrfs_root *log;
3378 struct btrfs_dir_item *di;
3379 struct btrfs_path *path;
3382 u64 dir_ino = btrfs_ino(dir);
3384 if (!inode_logged(trans, dir))
3387 ret = join_running_log_trans(root);
3391 mutex_lock(&dir->log_mutex);
3393 log = root->log_root;
3394 path = btrfs_alloc_path();
3400 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3401 name, name_len, -1);
3407 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3413 btrfs_release_path(path);
3414 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3415 index, name, name_len, -1);
3421 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3429 * We do not need to update the size field of the directory's inode item
3430 * because on log replay we update the field to reflect all existing
3431 * entries in the directory (see overwrite_item()).
3434 btrfs_free_path(path);
3436 mutex_unlock(&dir->log_mutex);
3437 if (err == -ENOSPC) {
3438 btrfs_set_log_full_commit(trans);
3440 } else if (err < 0 && err != -ENOENT) {
3441 /* ENOENT can be returned if the entry hasn't been fsynced yet */
3442 btrfs_abort_transaction(trans, err);
3445 btrfs_end_log_trans(root);
3450 /* see comments for btrfs_del_dir_entries_in_log */
3451 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3452 struct btrfs_root *root,
3453 const char *name, int name_len,
3454 struct btrfs_inode *inode, u64 dirid)
3456 struct btrfs_root *log;
3460 if (!inode_logged(trans, inode))
3463 ret = join_running_log_trans(root);
3466 log = root->log_root;
3467 mutex_lock(&inode->log_mutex);
3469 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3471 mutex_unlock(&inode->log_mutex);
3472 if (ret == -ENOSPC) {
3473 btrfs_set_log_full_commit(trans);
3475 } else if (ret < 0 && ret != -ENOENT)
3476 btrfs_abort_transaction(trans, ret);
3477 btrfs_end_log_trans(root);
3483 * creates a range item in the log for 'dirid'. first_offset and
3484 * last_offset tell us which parts of the key space the log should
3485 * be considered authoritative for.
3487 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3488 struct btrfs_root *log,
3489 struct btrfs_path *path,
3490 int key_type, u64 dirid,
3491 u64 first_offset, u64 last_offset)
3494 struct btrfs_key key;
3495 struct btrfs_dir_log_item *item;
3497 key.objectid = dirid;
3498 key.offset = first_offset;
3499 if (key_type == BTRFS_DIR_ITEM_KEY)
3500 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3502 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3503 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3507 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3508 struct btrfs_dir_log_item);
3509 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3510 btrfs_mark_buffer_dirty(path->nodes[0]);
3511 btrfs_release_path(path);
3516 * log all the items included in the current transaction for a given
3517 * directory. This also creates the range items in the log tree required
3518 * to replay anything deleted before the fsync
3520 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3521 struct btrfs_root *root, struct btrfs_inode *inode,
3522 struct btrfs_path *path,
3523 struct btrfs_path *dst_path, int key_type,
3524 struct btrfs_log_ctx *ctx,
3525 u64 min_offset, u64 *last_offset_ret)
3527 struct btrfs_key min_key;
3528 struct btrfs_root *log = root->log_root;
3529 struct extent_buffer *src;
3534 u64 first_offset = min_offset;
3535 u64 last_offset = (u64)-1;
3536 u64 ino = btrfs_ino(inode);
3538 log = root->log_root;
3540 min_key.objectid = ino;
3541 min_key.type = key_type;
3542 min_key.offset = min_offset;
3544 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3547 * we didn't find anything from this transaction, see if there
3548 * is anything at all
3550 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3551 min_key.objectid = ino;
3552 min_key.type = key_type;
3553 min_key.offset = (u64)-1;
3554 btrfs_release_path(path);
3555 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3557 btrfs_release_path(path);
3560 ret = btrfs_previous_item(root, path, ino, key_type);
3562 /* if ret == 0 there are items for this type,
3563 * create a range to tell us the last key of this type.
3564 * otherwise, there are no items in this directory after
3565 * *min_offset, and we create a range to indicate that.
3568 struct btrfs_key tmp;
3569 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3571 if (key_type == tmp.type)
3572 first_offset = max(min_offset, tmp.offset) + 1;
3577 /* go backward to find any previous key */
3578 ret = btrfs_previous_item(root, path, ino, key_type);
3580 struct btrfs_key tmp;
3581 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3582 if (key_type == tmp.type) {
3583 first_offset = tmp.offset;
3584 ret = overwrite_item(trans, log, dst_path,
3585 path->nodes[0], path->slots[0],
3593 btrfs_release_path(path);
3596 * Find the first key from this transaction again. See the note for
3597 * log_new_dir_dentries, if we're logging a directory recursively we
3598 * won't be holding its i_mutex, which means we can modify the directory
3599 * while we're logging it. If we remove an entry between our first
3600 * search and this search we'll not find the key again and can just
3604 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3609 * we have a block from this transaction, log every item in it
3610 * from our directory
3613 struct btrfs_key tmp;
3614 src = path->nodes[0];
3615 nritems = btrfs_header_nritems(src);
3616 for (i = path->slots[0]; i < nritems; i++) {
3617 struct btrfs_dir_item *di;
3619 btrfs_item_key_to_cpu(src, &min_key, i);
3621 if (min_key.objectid != ino || min_key.type != key_type)
3624 if (need_resched()) {
3625 btrfs_release_path(path);
3630 ret = overwrite_item(trans, log, dst_path, src, i,
3638 * We must make sure that when we log a directory entry,
3639 * the corresponding inode, after log replay, has a
3640 * matching link count. For example:
3646 * xfs_io -c "fsync" mydir
3648 * <mount fs and log replay>
3650 * Would result in a fsync log that when replayed, our
3651 * file inode would have a link count of 1, but we get
3652 * two directory entries pointing to the same inode.
3653 * After removing one of the names, it would not be
3654 * possible to remove the other name, which resulted
3655 * always in stale file handle errors, and would not
3656 * be possible to rmdir the parent directory, since
3657 * its i_size could never decrement to the value
3658 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3660 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3661 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3663 (btrfs_dir_transid(src, di) == trans->transid ||
3664 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3665 tmp.type != BTRFS_ROOT_ITEM_KEY)
3666 ctx->log_new_dentries = true;
3668 path->slots[0] = nritems;
3671 * look ahead to the next item and see if it is also
3672 * from this directory and from this transaction
3674 ret = btrfs_next_leaf(root, path);
3677 last_offset = (u64)-1;
3682 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3683 if (tmp.objectid != ino || tmp.type != key_type) {
3684 last_offset = (u64)-1;
3687 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3688 ret = overwrite_item(trans, log, dst_path,
3689 path->nodes[0], path->slots[0],
3694 last_offset = tmp.offset;
3699 btrfs_release_path(path);
3700 btrfs_release_path(dst_path);
3703 *last_offset_ret = last_offset;
3705 * insert the log range keys to indicate where the log
3708 ret = insert_dir_log_key(trans, log, path, key_type,
3709 ino, first_offset, last_offset);
3717 * logging directories is very similar to logging inodes, We find all the items
3718 * from the current transaction and write them to the log.
3720 * The recovery code scans the directory in the subvolume, and if it finds a
3721 * key in the range logged that is not present in the log tree, then it means
3722 * that dir entry was unlinked during the transaction.
3724 * In order for that scan to work, we must include one key smaller than
3725 * the smallest logged by this transaction and one key larger than the largest
3726 * key logged by this transaction.
3728 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3729 struct btrfs_root *root, struct btrfs_inode *inode,
3730 struct btrfs_path *path,
3731 struct btrfs_path *dst_path,
3732 struct btrfs_log_ctx *ctx)
3737 int key_type = BTRFS_DIR_ITEM_KEY;
3743 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3744 ctx, min_key, &max_key);
3747 if (max_key == (u64)-1)
3749 min_key = max_key + 1;
3752 if (key_type == BTRFS_DIR_ITEM_KEY) {
3753 key_type = BTRFS_DIR_INDEX_KEY;
3760 * a helper function to drop items from the log before we relog an
3761 * inode. max_key_type indicates the highest item type to remove.
3762 * This cannot be run for file data extents because it does not
3763 * free the extents they point to.
3765 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3766 struct btrfs_root *log,
3767 struct btrfs_path *path,
3768 u64 objectid, int max_key_type)
3771 struct btrfs_key key;
3772 struct btrfs_key found_key;
3775 key.objectid = objectid;
3776 key.type = max_key_type;
3777 key.offset = (u64)-1;
3780 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3781 BUG_ON(ret == 0); /* Logic error */
3785 if (path->slots[0] == 0)
3789 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3792 if (found_key.objectid != objectid)
3795 found_key.offset = 0;
3797 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3801 ret = btrfs_del_items(trans, log, path, start_slot,
3802 path->slots[0] - start_slot + 1);
3804 * If start slot isn't 0 then we don't need to re-search, we've
3805 * found the last guy with the objectid in this tree.
3807 if (ret || start_slot != 0)
3809 btrfs_release_path(path);
3811 btrfs_release_path(path);
3817 static void fill_inode_item(struct btrfs_trans_handle *trans,
3818 struct extent_buffer *leaf,
3819 struct btrfs_inode_item *item,
3820 struct inode *inode, int log_inode_only,
3823 struct btrfs_map_token token;
3825 btrfs_init_map_token(&token, leaf);
3827 if (log_inode_only) {
3828 /* set the generation to zero so the recover code
3829 * can tell the difference between an logging
3830 * just to say 'this inode exists' and a logging
3831 * to say 'update this inode with these values'
3833 btrfs_set_token_inode_generation(&token, item, 0);
3834 btrfs_set_token_inode_size(&token, item, logged_isize);
3836 btrfs_set_token_inode_generation(&token, item,
3837 BTRFS_I(inode)->generation);
3838 btrfs_set_token_inode_size(&token, item, inode->i_size);
3841 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3842 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3843 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3844 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3846 btrfs_set_token_timespec_sec(&token, &item->atime,
3847 inode->i_atime.tv_sec);
3848 btrfs_set_token_timespec_nsec(&token, &item->atime,
3849 inode->i_atime.tv_nsec);
3851 btrfs_set_token_timespec_sec(&token, &item->mtime,
3852 inode->i_mtime.tv_sec);
3853 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3854 inode->i_mtime.tv_nsec);
3856 btrfs_set_token_timespec_sec(&token, &item->ctime,
3857 inode->i_ctime.tv_sec);
3858 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3859 inode->i_ctime.tv_nsec);
3862 * We do not need to set the nbytes field, in fact during a fast fsync
3863 * its value may not even be correct, since a fast fsync does not wait
3864 * for ordered extent completion, which is where we update nbytes, it
3865 * only waits for writeback to complete. During log replay as we find
3866 * file extent items and replay them, we adjust the nbytes field of the
3867 * inode item in subvolume tree as needed (see overwrite_item()).
3870 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3871 btrfs_set_token_inode_transid(&token, item, trans->transid);
3872 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3873 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3874 btrfs_set_token_inode_block_group(&token, item, 0);
3877 static int log_inode_item(struct btrfs_trans_handle *trans,
3878 struct btrfs_root *log, struct btrfs_path *path,
3879 struct btrfs_inode *inode)
3881 struct btrfs_inode_item *inode_item;
3884 ret = btrfs_insert_empty_item(trans, log, path,
3885 &inode->location, sizeof(*inode_item));
3886 if (ret && ret != -EEXIST)
3888 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3889 struct btrfs_inode_item);
3890 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3892 btrfs_release_path(path);
3896 static int log_csums(struct btrfs_trans_handle *trans,
3897 struct btrfs_inode *inode,
3898 struct btrfs_root *log_root,
3899 struct btrfs_ordered_sum *sums)
3901 const u64 lock_end = sums->bytenr + sums->len - 1;
3902 struct extent_state *cached_state = NULL;
3906 * If this inode was not used for reflink operations in the current
3907 * transaction with new extents, then do the fast path, no need to
3908 * worry about logging checksum items with overlapping ranges.
3910 if (inode->last_reflink_trans < trans->transid)
3911 return btrfs_csum_file_blocks(trans, log_root, sums);
3914 * Serialize logging for checksums. This is to avoid racing with the
3915 * same checksum being logged by another task that is logging another
3916 * file which happens to refer to the same extent as well. Such races
3917 * can leave checksum items in the log with overlapping ranges.
3919 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
3920 lock_end, &cached_state);
3924 * Due to extent cloning, we might have logged a csum item that covers a
3925 * subrange of a cloned extent, and later we can end up logging a csum
3926 * item for a larger subrange of the same extent or the entire range.
3927 * This would leave csum items in the log tree that cover the same range
3928 * and break the searches for checksums in the log tree, resulting in
3929 * some checksums missing in the fs/subvolume tree. So just delete (or
3930 * trim and adjust) any existing csum items in the log for this range.
3932 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3934 ret = btrfs_csum_file_blocks(trans, log_root, sums);
3936 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
3942 static noinline int copy_items(struct btrfs_trans_handle *trans,
3943 struct btrfs_inode *inode,
3944 struct btrfs_path *dst_path,
3945 struct btrfs_path *src_path,
3946 int start_slot, int nr, int inode_only,
3949 struct btrfs_fs_info *fs_info = trans->fs_info;
3950 unsigned long src_offset;
3951 unsigned long dst_offset;
3952 struct btrfs_root *log = inode->root->log_root;
3953 struct btrfs_file_extent_item *extent;
3954 struct btrfs_inode_item *inode_item;
3955 struct extent_buffer *src = src_path->nodes[0];
3957 struct btrfs_key *ins_keys;
3961 struct list_head ordered_sums;
3962 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3964 INIT_LIST_HEAD(&ordered_sums);
3966 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3967 nr * sizeof(u32), GFP_NOFS);
3971 ins_sizes = (u32 *)ins_data;
3972 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3974 for (i = 0; i < nr; i++) {
3975 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3976 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3978 ret = btrfs_insert_empty_items(trans, log, dst_path,
3979 ins_keys, ins_sizes, nr);
3985 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3986 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3987 dst_path->slots[0]);
3989 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3991 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3992 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3994 struct btrfs_inode_item);
3995 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3997 inode_only == LOG_INODE_EXISTS,
4000 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4001 src_offset, ins_sizes[i]);
4004 /* take a reference on file data extents so that truncates
4005 * or deletes of this inode don't have to relog the inode
4008 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4011 extent = btrfs_item_ptr(src, start_slot + i,
4012 struct btrfs_file_extent_item);
4014 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4017 found_type = btrfs_file_extent_type(src, extent);
4018 if (found_type == BTRFS_FILE_EXTENT_REG) {
4020 ds = btrfs_file_extent_disk_bytenr(src,
4022 /* ds == 0 is a hole */
4026 dl = btrfs_file_extent_disk_num_bytes(src,
4028 cs = btrfs_file_extent_offset(src, extent);
4029 cl = btrfs_file_extent_num_bytes(src,
4031 if (btrfs_file_extent_compression(src,
4037 ret = btrfs_lookup_csums_range(
4039 ds + cs, ds + cs + cl - 1,
4047 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4048 btrfs_release_path(dst_path);
4052 * we have to do this after the loop above to avoid changing the
4053 * log tree while trying to change the log tree.
4055 while (!list_empty(&ordered_sums)) {
4056 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4057 struct btrfs_ordered_sum,
4060 ret = log_csums(trans, inode, log, sums);
4061 list_del(&sums->list);
4068 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4070 struct extent_map *em1, *em2;
4072 em1 = list_entry(a, struct extent_map, list);
4073 em2 = list_entry(b, struct extent_map, list);
4075 if (em1->start < em2->start)
4077 else if (em1->start > em2->start)
4082 static int log_extent_csums(struct btrfs_trans_handle *trans,
4083 struct btrfs_inode *inode,
4084 struct btrfs_root *log_root,
4085 const struct extent_map *em,
4086 struct btrfs_log_ctx *ctx)
4088 struct btrfs_ordered_extent *ordered;
4091 u64 mod_start = em->mod_start;
4092 u64 mod_len = em->mod_len;
4093 LIST_HEAD(ordered_sums);
4096 if (inode->flags & BTRFS_INODE_NODATASUM ||
4097 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4098 em->block_start == EXTENT_MAP_HOLE)
4101 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4102 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4103 const u64 mod_end = mod_start + mod_len;
4104 struct btrfs_ordered_sum *sums;
4109 if (ordered_end <= mod_start)
4111 if (mod_end <= ordered->file_offset)
4115 * We are going to copy all the csums on this ordered extent, so
4116 * go ahead and adjust mod_start and mod_len in case this ordered
4117 * extent has already been logged.
4119 if (ordered->file_offset > mod_start) {
4120 if (ordered_end >= mod_end)
4121 mod_len = ordered->file_offset - mod_start;
4123 * If we have this case
4125 * |--------- logged extent ---------|
4126 * |----- ordered extent ----|
4128 * Just don't mess with mod_start and mod_len, we'll
4129 * just end up logging more csums than we need and it
4133 if (ordered_end < mod_end) {
4134 mod_len = mod_end - ordered_end;
4135 mod_start = ordered_end;
4142 * To keep us from looping for the above case of an ordered
4143 * extent that falls inside of the logged extent.
4145 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4148 list_for_each_entry(sums, &ordered->list, list) {
4149 ret = log_csums(trans, inode, log_root, sums);
4155 /* We're done, found all csums in the ordered extents. */
4159 /* If we're compressed we have to save the entire range of csums. */
4160 if (em->compress_type) {
4162 csum_len = max(em->block_len, em->orig_block_len);
4164 csum_offset = mod_start - em->start;
4168 /* block start is already adjusted for the file extent offset. */
4169 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4170 em->block_start + csum_offset,
4171 em->block_start + csum_offset +
4172 csum_len - 1, &ordered_sums, 0);
4176 while (!list_empty(&ordered_sums)) {
4177 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4178 struct btrfs_ordered_sum,
4181 ret = log_csums(trans, inode, log_root, sums);
4182 list_del(&sums->list);
4189 static int log_one_extent(struct btrfs_trans_handle *trans,
4190 struct btrfs_inode *inode, struct btrfs_root *root,
4191 const struct extent_map *em,
4192 struct btrfs_path *path,
4193 struct btrfs_log_ctx *ctx)
4195 struct btrfs_drop_extents_args drop_args = { 0 };
4196 struct btrfs_root *log = root->log_root;
4197 struct btrfs_file_extent_item *fi;
4198 struct extent_buffer *leaf;
4199 struct btrfs_map_token token;
4200 struct btrfs_key key;
4201 u64 extent_offset = em->start - em->orig_start;
4205 ret = log_extent_csums(trans, inode, log, em, ctx);
4209 drop_args.path = path;
4210 drop_args.start = em->start;
4211 drop_args.end = em->start + em->len;
4212 drop_args.replace_extent = true;
4213 drop_args.extent_item_size = sizeof(*fi);
4214 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4218 if (!drop_args.extent_inserted) {
4219 key.objectid = btrfs_ino(inode);
4220 key.type = BTRFS_EXTENT_DATA_KEY;
4221 key.offset = em->start;
4223 ret = btrfs_insert_empty_item(trans, log, path, &key,
4228 leaf = path->nodes[0];
4229 btrfs_init_map_token(&token, leaf);
4230 fi = btrfs_item_ptr(leaf, path->slots[0],
4231 struct btrfs_file_extent_item);
4233 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4234 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4235 btrfs_set_token_file_extent_type(&token, fi,
4236 BTRFS_FILE_EXTENT_PREALLOC);
4238 btrfs_set_token_file_extent_type(&token, fi,
4239 BTRFS_FILE_EXTENT_REG);
4241 block_len = max(em->block_len, em->orig_block_len);
4242 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4243 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4245 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4246 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4247 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4250 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4252 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4253 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4256 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4257 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4258 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4259 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4260 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4261 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4262 btrfs_mark_buffer_dirty(leaf);
4264 btrfs_release_path(path);
4270 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4271 * lose them after doing a fast fsync and replaying the log. We scan the
4272 * subvolume's root instead of iterating the inode's extent map tree because
4273 * otherwise we can log incorrect extent items based on extent map conversion.
4274 * That can happen due to the fact that extent maps are merged when they
4275 * are not in the extent map tree's list of modified extents.
4277 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4278 struct btrfs_inode *inode,
4279 struct btrfs_path *path)
4281 struct btrfs_root *root = inode->root;
4282 struct btrfs_key key;
4283 const u64 i_size = i_size_read(&inode->vfs_inode);
4284 const u64 ino = btrfs_ino(inode);
4285 struct btrfs_path *dst_path = NULL;
4286 bool dropped_extents = false;
4287 u64 truncate_offset = i_size;
4288 struct extent_buffer *leaf;
4294 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4298 key.type = BTRFS_EXTENT_DATA_KEY;
4299 key.offset = i_size;
4300 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4305 * We must check if there is a prealloc extent that starts before the
4306 * i_size and crosses the i_size boundary. This is to ensure later we
4307 * truncate down to the end of that extent and not to the i_size, as
4308 * otherwise we end up losing part of the prealloc extent after a log
4309 * replay and with an implicit hole if there is another prealloc extent
4310 * that starts at an offset beyond i_size.
4312 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4317 struct btrfs_file_extent_item *ei;
4319 leaf = path->nodes[0];
4320 slot = path->slots[0];
4321 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4323 if (btrfs_file_extent_type(leaf, ei) ==
4324 BTRFS_FILE_EXTENT_PREALLOC) {
4327 btrfs_item_key_to_cpu(leaf, &key, slot);
4328 extent_end = key.offset +
4329 btrfs_file_extent_num_bytes(leaf, ei);
4331 if (extent_end > i_size)
4332 truncate_offset = extent_end;
4339 leaf = path->nodes[0];
4340 slot = path->slots[0];
4342 if (slot >= btrfs_header_nritems(leaf)) {
4344 ret = copy_items(trans, inode, dst_path, path,
4345 start_slot, ins_nr, 1, 0);
4350 ret = btrfs_next_leaf(root, path);
4360 btrfs_item_key_to_cpu(leaf, &key, slot);
4361 if (key.objectid > ino)
4363 if (WARN_ON_ONCE(key.objectid < ino) ||
4364 key.type < BTRFS_EXTENT_DATA_KEY ||
4365 key.offset < i_size) {
4369 if (!dropped_extents) {
4371 * Avoid logging extent items logged in past fsync calls
4372 * and leading to duplicate keys in the log tree.
4375 ret = btrfs_truncate_inode_items(trans,
4377 inode, truncate_offset,
4378 BTRFS_EXTENT_DATA_KEY);
4379 } while (ret == -EAGAIN);
4382 dropped_extents = true;
4389 dst_path = btrfs_alloc_path();
4397 ret = copy_items(trans, inode, dst_path, path,
4398 start_slot, ins_nr, 1, 0);
4400 btrfs_release_path(path);
4401 btrfs_free_path(dst_path);
4405 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4406 struct btrfs_root *root,
4407 struct btrfs_inode *inode,
4408 struct btrfs_path *path,
4409 struct btrfs_log_ctx *ctx)
4411 struct btrfs_ordered_extent *ordered;
4412 struct btrfs_ordered_extent *tmp;
4413 struct extent_map *em, *n;
4414 struct list_head extents;
4415 struct extent_map_tree *tree = &inode->extent_tree;
4419 INIT_LIST_HEAD(&extents);
4421 write_lock(&tree->lock);
4423 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4424 list_del_init(&em->list);
4426 * Just an arbitrary number, this can be really CPU intensive
4427 * once we start getting a lot of extents, and really once we
4428 * have a bunch of extents we just want to commit since it will
4431 if (++num > 32768) {
4432 list_del_init(&tree->modified_extents);
4437 if (em->generation < trans->transid)
4440 /* We log prealloc extents beyond eof later. */
4441 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4442 em->start >= i_size_read(&inode->vfs_inode))
4445 /* Need a ref to keep it from getting evicted from cache */
4446 refcount_inc(&em->refs);
4447 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4448 list_add_tail(&em->list, &extents);
4452 list_sort(NULL, &extents, extent_cmp);
4454 while (!list_empty(&extents)) {
4455 em = list_entry(extents.next, struct extent_map, list);
4457 list_del_init(&em->list);
4460 * If we had an error we just need to delete everybody from our
4464 clear_em_logging(tree, em);
4465 free_extent_map(em);
4469 write_unlock(&tree->lock);
4471 ret = log_one_extent(trans, inode, root, em, path, ctx);
4472 write_lock(&tree->lock);
4473 clear_em_logging(tree, em);
4474 free_extent_map(em);
4476 WARN_ON(!list_empty(&extents));
4477 write_unlock(&tree->lock);
4479 btrfs_release_path(path);
4481 ret = btrfs_log_prealloc_extents(trans, inode, path);
4486 * We have logged all extents successfully, now make sure the commit of
4487 * the current transaction waits for the ordered extents to complete
4488 * before it commits and wipes out the log trees, otherwise we would
4489 * lose data if an ordered extents completes after the transaction
4490 * commits and a power failure happens after the transaction commit.
4492 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4493 list_del_init(&ordered->log_list);
4494 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4496 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4497 spin_lock_irq(&inode->ordered_tree.lock);
4498 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4499 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4500 atomic_inc(&trans->transaction->pending_ordered);
4502 spin_unlock_irq(&inode->ordered_tree.lock);
4504 btrfs_put_ordered_extent(ordered);
4510 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4511 struct btrfs_path *path, u64 *size_ret)
4513 struct btrfs_key key;
4516 key.objectid = btrfs_ino(inode);
4517 key.type = BTRFS_INODE_ITEM_KEY;
4520 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4523 } else if (ret > 0) {
4526 struct btrfs_inode_item *item;
4528 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4529 struct btrfs_inode_item);
4530 *size_ret = btrfs_inode_size(path->nodes[0], item);
4532 * If the in-memory inode's i_size is smaller then the inode
4533 * size stored in the btree, return the inode's i_size, so
4534 * that we get a correct inode size after replaying the log
4535 * when before a power failure we had a shrinking truncate
4536 * followed by addition of a new name (rename / new hard link).
4537 * Otherwise return the inode size from the btree, to avoid
4538 * data loss when replaying a log due to previously doing a
4539 * write that expands the inode's size and logging a new name
4540 * immediately after.
4542 if (*size_ret > inode->vfs_inode.i_size)
4543 *size_ret = inode->vfs_inode.i_size;
4546 btrfs_release_path(path);
4551 * At the moment we always log all xattrs. This is to figure out at log replay
4552 * time which xattrs must have their deletion replayed. If a xattr is missing
4553 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4554 * because if a xattr is deleted, the inode is fsynced and a power failure
4555 * happens, causing the log to be replayed the next time the fs is mounted,
4556 * we want the xattr to not exist anymore (same behaviour as other filesystems
4557 * with a journal, ext3/4, xfs, f2fs, etc).
4559 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4560 struct btrfs_root *root,
4561 struct btrfs_inode *inode,
4562 struct btrfs_path *path,
4563 struct btrfs_path *dst_path)
4566 struct btrfs_key key;
4567 const u64 ino = btrfs_ino(inode);
4570 bool found_xattrs = false;
4572 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4576 key.type = BTRFS_XATTR_ITEM_KEY;
4579 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4584 int slot = path->slots[0];
4585 struct extent_buffer *leaf = path->nodes[0];
4586 int nritems = btrfs_header_nritems(leaf);
4588 if (slot >= nritems) {
4590 ret = copy_items(trans, inode, dst_path, path,
4591 start_slot, ins_nr, 1, 0);
4596 ret = btrfs_next_leaf(root, path);
4604 btrfs_item_key_to_cpu(leaf, &key, slot);
4605 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4612 found_xattrs = true;
4616 ret = copy_items(trans, inode, dst_path, path,
4617 start_slot, ins_nr, 1, 0);
4623 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4629 * When using the NO_HOLES feature if we punched a hole that causes the
4630 * deletion of entire leafs or all the extent items of the first leaf (the one
4631 * that contains the inode item and references) we may end up not processing
4632 * any extents, because there are no leafs with a generation matching the
4633 * current transaction that have extent items for our inode. So we need to find
4634 * if any holes exist and then log them. We also need to log holes after any
4635 * truncate operation that changes the inode's size.
4637 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4638 struct btrfs_root *root,
4639 struct btrfs_inode *inode,
4640 struct btrfs_path *path)
4642 struct btrfs_fs_info *fs_info = root->fs_info;
4643 struct btrfs_key key;
4644 const u64 ino = btrfs_ino(inode);
4645 const u64 i_size = i_size_read(&inode->vfs_inode);
4646 u64 prev_extent_end = 0;
4649 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4653 key.type = BTRFS_EXTENT_DATA_KEY;
4656 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4661 struct extent_buffer *leaf = path->nodes[0];
4663 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4664 ret = btrfs_next_leaf(root, path);
4671 leaf = path->nodes[0];
4674 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4675 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4678 /* We have a hole, log it. */
4679 if (prev_extent_end < key.offset) {
4680 const u64 hole_len = key.offset - prev_extent_end;
4683 * Release the path to avoid deadlocks with other code
4684 * paths that search the root while holding locks on
4685 * leafs from the log root.
4687 btrfs_release_path(path);
4688 ret = btrfs_insert_file_extent(trans, root->log_root,
4689 ino, prev_extent_end, 0,
4690 0, hole_len, 0, hole_len,
4696 * Search for the same key again in the root. Since it's
4697 * an extent item and we are holding the inode lock, the
4698 * key must still exist. If it doesn't just emit warning
4699 * and return an error to fall back to a transaction
4702 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4705 if (WARN_ON(ret > 0))
4707 leaf = path->nodes[0];
4710 prev_extent_end = btrfs_file_extent_end(path);
4715 if (prev_extent_end < i_size) {
4718 btrfs_release_path(path);
4719 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4720 ret = btrfs_insert_file_extent(trans, root->log_root,
4721 ino, prev_extent_end, 0, 0,
4722 hole_len, 0, hole_len,
4732 * When we are logging a new inode X, check if it doesn't have a reference that
4733 * matches the reference from some other inode Y created in a past transaction
4734 * and that was renamed in the current transaction. If we don't do this, then at
4735 * log replay time we can lose inode Y (and all its files if it's a directory):
4738 * echo "hello world" > /mnt/x/foobar
4741 * mkdir /mnt/x # or touch /mnt/x
4742 * xfs_io -c fsync /mnt/x
4744 * mount fs, trigger log replay
4746 * After the log replay procedure, we would lose the first directory and all its
4747 * files (file foobar).
4748 * For the case where inode Y is not a directory we simply end up losing it:
4750 * echo "123" > /mnt/foo
4752 * mv /mnt/foo /mnt/bar
4753 * echo "abc" > /mnt/foo
4754 * xfs_io -c fsync /mnt/foo
4757 * We also need this for cases where a snapshot entry is replaced by some other
4758 * entry (file or directory) otherwise we end up with an unreplayable log due to
4759 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4760 * if it were a regular entry:
4763 * btrfs subvolume snapshot /mnt /mnt/x/snap
4764 * btrfs subvolume delete /mnt/x/snap
4767 * fsync /mnt/x or fsync some new file inside it
4770 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4771 * the same transaction.
4773 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4775 const struct btrfs_key *key,
4776 struct btrfs_inode *inode,
4777 u64 *other_ino, u64 *other_parent)
4780 struct btrfs_path *search_path;
4783 u32 item_size = btrfs_item_size_nr(eb, slot);
4785 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4787 search_path = btrfs_alloc_path();
4790 search_path->search_commit_root = 1;
4791 search_path->skip_locking = 1;
4793 while (cur_offset < item_size) {
4797 unsigned long name_ptr;
4798 struct btrfs_dir_item *di;
4800 if (key->type == BTRFS_INODE_REF_KEY) {
4801 struct btrfs_inode_ref *iref;
4803 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4804 parent = key->offset;
4805 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4806 name_ptr = (unsigned long)(iref + 1);
4807 this_len = sizeof(*iref) + this_name_len;
4809 struct btrfs_inode_extref *extref;
4811 extref = (struct btrfs_inode_extref *)(ptr +
4813 parent = btrfs_inode_extref_parent(eb, extref);
4814 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4815 name_ptr = (unsigned long)&extref->name;
4816 this_len = sizeof(*extref) + this_name_len;
4819 if (this_name_len > name_len) {
4822 new_name = krealloc(name, this_name_len, GFP_NOFS);
4827 name_len = this_name_len;
4831 read_extent_buffer(eb, name, name_ptr, this_name_len);
4832 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4833 parent, name, this_name_len, 0);
4834 if (di && !IS_ERR(di)) {
4835 struct btrfs_key di_key;
4837 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4839 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4840 if (di_key.objectid != key->objectid) {
4842 *other_ino = di_key.objectid;
4843 *other_parent = parent;
4851 } else if (IS_ERR(di)) {
4855 btrfs_release_path(search_path);
4857 cur_offset += this_len;
4861 btrfs_free_path(search_path);
4866 struct btrfs_ino_list {
4869 struct list_head list;
4872 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4873 struct btrfs_root *root,
4874 struct btrfs_path *path,
4875 struct btrfs_log_ctx *ctx,
4876 u64 ino, u64 parent)
4878 struct btrfs_ino_list *ino_elem;
4879 LIST_HEAD(inode_list);
4882 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4885 ino_elem->ino = ino;
4886 ino_elem->parent = parent;
4887 list_add_tail(&ino_elem->list, &inode_list);
4889 while (!list_empty(&inode_list)) {
4890 struct btrfs_fs_info *fs_info = root->fs_info;
4891 struct btrfs_key key;
4892 struct inode *inode;
4894 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4896 ino = ino_elem->ino;
4897 parent = ino_elem->parent;
4898 list_del(&ino_elem->list);
4903 btrfs_release_path(path);
4905 inode = btrfs_iget(fs_info->sb, ino, root);
4907 * If the other inode that had a conflicting dir entry was
4908 * deleted in the current transaction, we need to log its parent
4911 if (IS_ERR(inode)) {
4912 ret = PTR_ERR(inode);
4913 if (ret == -ENOENT) {
4914 inode = btrfs_iget(fs_info->sb, parent, root);
4915 if (IS_ERR(inode)) {
4916 ret = PTR_ERR(inode);
4918 ret = btrfs_log_inode(trans, root,
4920 LOG_OTHER_INODE_ALL,
4922 btrfs_add_delayed_iput(inode);
4928 * If the inode was already logged skip it - otherwise we can
4929 * hit an infinite loop. Example:
4931 * From the commit root (previous transaction) we have the
4934 * inode 257 a directory
4935 * inode 258 with references "zz" and "zz_link" on inode 257
4936 * inode 259 with reference "a" on inode 257
4938 * And in the current (uncommitted) transaction we have:
4940 * inode 257 a directory, unchanged
4941 * inode 258 with references "a" and "a2" on inode 257
4942 * inode 259 with reference "zz_link" on inode 257
4943 * inode 261 with reference "zz" on inode 257
4945 * When logging inode 261 the following infinite loop could
4946 * happen if we don't skip already logged inodes:
4948 * - we detect inode 258 as a conflicting inode, with inode 261
4949 * on reference "zz", and log it;
4951 * - we detect inode 259 as a conflicting inode, with inode 258
4952 * on reference "a", and log it;
4954 * - we detect inode 258 as a conflicting inode, with inode 259
4955 * on reference "zz_link", and log it - again! After this we
4956 * repeat the above steps forever.
4958 spin_lock(&BTRFS_I(inode)->lock);
4960 * Check the inode's logged_trans only instead of
4961 * btrfs_inode_in_log(). This is because the last_log_commit of
4962 * the inode is not updated when we only log that it exists and
4963 * it has the full sync bit set (see btrfs_log_inode()).
4965 if (BTRFS_I(inode)->logged_trans == trans->transid) {
4966 spin_unlock(&BTRFS_I(inode)->lock);
4967 btrfs_add_delayed_iput(inode);
4970 spin_unlock(&BTRFS_I(inode)->lock);
4972 * We are safe logging the other inode without acquiring its
4973 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4974 * are safe against concurrent renames of the other inode as
4975 * well because during a rename we pin the log and update the
4976 * log with the new name before we unpin it.
4978 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4979 LOG_OTHER_INODE, ctx);
4981 btrfs_add_delayed_iput(inode);
4986 key.type = BTRFS_INODE_REF_KEY;
4988 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4990 btrfs_add_delayed_iput(inode);
4995 struct extent_buffer *leaf = path->nodes[0];
4996 int slot = path->slots[0];
4998 u64 other_parent = 0;
5000 if (slot >= btrfs_header_nritems(leaf)) {
5001 ret = btrfs_next_leaf(root, path);
5004 } else if (ret > 0) {
5011 btrfs_item_key_to_cpu(leaf, &key, slot);
5012 if (key.objectid != ino ||
5013 (key.type != BTRFS_INODE_REF_KEY &&
5014 key.type != BTRFS_INODE_EXTREF_KEY)) {
5019 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5020 BTRFS_I(inode), &other_ino,
5025 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5030 ino_elem->ino = other_ino;
5031 ino_elem->parent = other_parent;
5032 list_add_tail(&ino_elem->list, &inode_list);
5037 btrfs_add_delayed_iput(inode);
5043 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5044 struct btrfs_inode *inode,
5045 struct btrfs_key *min_key,
5046 const struct btrfs_key *max_key,
5047 struct btrfs_path *path,
5048 struct btrfs_path *dst_path,
5049 const u64 logged_isize,
5050 const bool recursive_logging,
5051 const int inode_only,
5052 struct btrfs_log_ctx *ctx,
5053 bool *need_log_inode_item)
5055 struct btrfs_root *root = inode->root;
5056 int ins_start_slot = 0;
5061 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5069 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5070 if (min_key->objectid != max_key->objectid)
5072 if (min_key->type > max_key->type)
5075 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5076 *need_log_inode_item = false;
5078 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5079 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5080 inode->generation == trans->transid &&
5081 !recursive_logging) {
5083 u64 other_parent = 0;
5085 ret = btrfs_check_ref_name_override(path->nodes[0],
5086 path->slots[0], min_key, inode,
5087 &other_ino, &other_parent);
5090 } else if (ret > 0 && ctx &&
5091 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5096 ins_start_slot = path->slots[0];
5098 ret = copy_items(trans, inode, dst_path, path,
5099 ins_start_slot, ins_nr,
5100 inode_only, logged_isize);
5105 ret = log_conflicting_inodes(trans, root, path,
5106 ctx, other_ino, other_parent);
5109 btrfs_release_path(path);
5114 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5115 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5118 ret = copy_items(trans, inode, dst_path, path,
5120 ins_nr, inode_only, logged_isize);
5127 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5130 } else if (!ins_nr) {
5131 ins_start_slot = path->slots[0];
5136 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5137 ins_nr, inode_only, logged_isize);
5141 ins_start_slot = path->slots[0];
5144 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5145 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5150 ret = copy_items(trans, inode, dst_path, path,
5151 ins_start_slot, ins_nr, inode_only,
5157 btrfs_release_path(path);
5159 if (min_key->offset < (u64)-1) {
5161 } else if (min_key->type < max_key->type) {
5163 min_key->offset = 0;
5169 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5170 ins_nr, inode_only, logged_isize);
5175 /* log a single inode in the tree log.
5176 * At least one parent directory for this inode must exist in the tree
5177 * or be logged already.
5179 * Any items from this inode changed by the current transaction are copied
5180 * to the log tree. An extra reference is taken on any extents in this
5181 * file, allowing us to avoid a whole pile of corner cases around logging
5182 * blocks that have been removed from the tree.
5184 * See LOG_INODE_ALL and related defines for a description of what inode_only
5187 * This handles both files and directories.
5189 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5190 struct btrfs_root *root, struct btrfs_inode *inode,
5192 struct btrfs_log_ctx *ctx)
5194 struct btrfs_path *path;
5195 struct btrfs_path *dst_path;
5196 struct btrfs_key min_key;
5197 struct btrfs_key max_key;
5198 struct btrfs_root *log = root->log_root;
5201 bool fast_search = false;
5202 u64 ino = btrfs_ino(inode);
5203 struct extent_map_tree *em_tree = &inode->extent_tree;
5204 u64 logged_isize = 0;
5205 bool need_log_inode_item = true;
5206 bool xattrs_logged = false;
5207 bool recursive_logging = false;
5209 path = btrfs_alloc_path();
5212 dst_path = btrfs_alloc_path();
5214 btrfs_free_path(path);
5218 min_key.objectid = ino;
5219 min_key.type = BTRFS_INODE_ITEM_KEY;
5222 max_key.objectid = ino;
5225 /* today the code can only do partial logging of directories */
5226 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5227 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5228 &inode->runtime_flags) &&
5229 inode_only >= LOG_INODE_EXISTS))
5230 max_key.type = BTRFS_XATTR_ITEM_KEY;
5232 max_key.type = (u8)-1;
5233 max_key.offset = (u64)-1;
5236 * Only run delayed items if we are a directory. We want to make sure
5237 * all directory indexes hit the fs/subvolume tree so we can find them
5238 * and figure out which index ranges have to be logged.
5240 * Otherwise commit the delayed inode only if the full sync flag is set,
5241 * as we want to make sure an up to date version is in the subvolume
5242 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5243 * it to the log tree. For a non full sync, we always log the inode item
5244 * based on the in-memory struct btrfs_inode which is always up to date.
5246 if (S_ISDIR(inode->vfs_inode.i_mode))
5247 ret = btrfs_commit_inode_delayed_items(trans, inode);
5248 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5249 ret = btrfs_commit_inode_delayed_inode(inode);
5252 btrfs_free_path(path);
5253 btrfs_free_path(dst_path);
5257 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5258 recursive_logging = true;
5259 if (inode_only == LOG_OTHER_INODE)
5260 inode_only = LOG_INODE_EXISTS;
5262 inode_only = LOG_INODE_ALL;
5263 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5265 mutex_lock(&inode->log_mutex);
5269 * a brute force approach to making sure we get the most uptodate
5270 * copies of everything.
5272 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5273 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5275 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5276 if (inode_only == LOG_INODE_EXISTS)
5277 max_key_type = BTRFS_XATTR_ITEM_KEY;
5278 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5280 if (inode_only == LOG_INODE_EXISTS) {
5282 * Make sure the new inode item we write to the log has
5283 * the same isize as the current one (if it exists).
5284 * This is necessary to prevent data loss after log
5285 * replay, and also to prevent doing a wrong expanding
5286 * truncate - for e.g. create file, write 4K into offset
5287 * 0, fsync, write 4K into offset 4096, add hard link,
5288 * fsync some other file (to sync log), power fail - if
5289 * we use the inode's current i_size, after log replay
5290 * we get a 8Kb file, with the last 4Kb extent as a hole
5291 * (zeroes), as if an expanding truncate happened,
5292 * instead of getting a file of 4Kb only.
5294 err = logged_inode_size(log, inode, path, &logged_isize);
5298 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5299 &inode->runtime_flags)) {
5300 if (inode_only == LOG_INODE_EXISTS) {
5301 max_key.type = BTRFS_XATTR_ITEM_KEY;
5302 ret = drop_objectid_items(trans, log, path, ino,
5305 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5306 &inode->runtime_flags);
5307 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5308 &inode->runtime_flags);
5310 ret = btrfs_truncate_inode_items(trans,
5316 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5317 &inode->runtime_flags) ||
5318 inode_only == LOG_INODE_EXISTS) {
5319 if (inode_only == LOG_INODE_ALL)
5321 max_key.type = BTRFS_XATTR_ITEM_KEY;
5322 ret = drop_objectid_items(trans, log, path, ino,
5325 if (inode_only == LOG_INODE_ALL)
5336 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5337 path, dst_path, logged_isize,
5338 recursive_logging, inode_only, ctx,
5339 &need_log_inode_item);
5343 btrfs_release_path(path);
5344 btrfs_release_path(dst_path);
5345 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5348 xattrs_logged = true;
5349 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5350 btrfs_release_path(path);
5351 btrfs_release_path(dst_path);
5352 err = btrfs_log_holes(trans, root, inode, path);
5357 btrfs_release_path(path);
5358 btrfs_release_path(dst_path);
5359 if (need_log_inode_item) {
5360 err = log_inode_item(trans, log, dst_path, inode);
5361 if (!err && !xattrs_logged) {
5362 err = btrfs_log_all_xattrs(trans, root, inode, path,
5364 btrfs_release_path(path);
5370 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5376 } else if (inode_only == LOG_INODE_ALL) {
5377 struct extent_map *em, *n;
5379 write_lock(&em_tree->lock);
5380 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5381 list_del_init(&em->list);
5382 write_unlock(&em_tree->lock);
5385 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5386 ret = log_directory_changes(trans, root, inode, path, dst_path,
5395 * If we are logging that an ancestor inode exists as part of logging a
5396 * new name from a link or rename operation, don't mark the inode as
5397 * logged - otherwise if an explicit fsync is made against an ancestor,
5398 * the fsync considers the inode in the log and doesn't sync the log,
5399 * resulting in the ancestor missing after a power failure unless the
5400 * log was synced as part of an fsync against any other unrelated inode.
5401 * So keep it simple for this case and just don't flag the ancestors as
5405 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5406 &inode->vfs_inode != ctx->inode)) {
5407 spin_lock(&inode->lock);
5408 inode->logged_trans = trans->transid;
5410 * Don't update last_log_commit if we logged that an inode exists
5411 * after it was loaded to memory (full_sync bit set).
5412 * This is to prevent data loss when we do a write to the inode,
5413 * then the inode gets evicted after all delalloc was flushed,
5414 * then we log it exists (due to a rename for example) and then
5415 * fsync it. This last fsync would do nothing (not logging the
5416 * extents previously written).
5418 if (inode_only != LOG_INODE_EXISTS ||
5419 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5420 inode->last_log_commit = inode->last_sub_trans;
5421 spin_unlock(&inode->lock);
5424 mutex_unlock(&inode->log_mutex);
5426 btrfs_free_path(path);
5427 btrfs_free_path(dst_path);
5432 * Check if we must fallback to a transaction commit when logging an inode.
5433 * This must be called after logging the inode and is used only in the context
5434 * when fsyncing an inode requires the need to log some other inode - in which
5435 * case we can't lock the i_mutex of each other inode we need to log as that
5436 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5437 * log inodes up or down in the hierarchy) or rename operations for example. So
5438 * we take the log_mutex of the inode after we have logged it and then check for
5439 * its last_unlink_trans value - this is safe because any task setting
5440 * last_unlink_trans must take the log_mutex and it must do this before it does
5441 * the actual unlink operation, so if we do this check before a concurrent task
5442 * sets last_unlink_trans it means we've logged a consistent version/state of
5443 * all the inode items, otherwise we are not sure and must do a transaction
5444 * commit (the concurrent task might have only updated last_unlink_trans before
5445 * we logged the inode or it might have also done the unlink).
5447 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5448 struct btrfs_inode *inode)
5452 mutex_lock(&inode->log_mutex);
5453 if (inode->last_unlink_trans >= trans->transid) {
5455 * Make sure any commits to the log are forced to be full
5458 btrfs_set_log_full_commit(trans);
5461 mutex_unlock(&inode->log_mutex);
5467 * follow the dentry parent pointers up the chain and see if any
5468 * of the directories in it require a full commit before they can
5469 * be logged. Returns zero if nothing special needs to be done or 1 if
5470 * a full commit is required.
5472 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5473 struct btrfs_inode *inode,
5474 struct dentry *parent,
5475 struct super_block *sb)
5478 struct dentry *old_parent = NULL;
5481 * for regular files, if its inode is already on disk, we don't
5482 * have to worry about the parents at all. This is because
5483 * we can use the last_unlink_trans field to record renames
5484 * and other fun in this file.
5486 if (S_ISREG(inode->vfs_inode.i_mode) &&
5487 inode->generation < trans->transid &&
5488 inode->last_unlink_trans < trans->transid)
5491 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5492 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5494 inode = BTRFS_I(d_inode(parent));
5498 if (btrfs_must_commit_transaction(trans, inode)) {
5503 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5506 if (IS_ROOT(parent)) {
5507 inode = BTRFS_I(d_inode(parent));
5508 if (btrfs_must_commit_transaction(trans, inode))
5513 parent = dget_parent(parent);
5515 old_parent = parent;
5516 inode = BTRFS_I(d_inode(parent));
5525 * Check if we need to log an inode. This is used in contexts where while
5526 * logging an inode we need to log another inode (either that it exists or in
5527 * full mode). This is used instead of btrfs_inode_in_log() because the later
5528 * requires the inode to be in the log and have the log transaction committed,
5529 * while here we do not care if the log transaction was already committed - our
5530 * caller will commit the log later - and we want to avoid logging an inode
5531 * multiple times when multiple tasks have joined the same log transaction.
5533 static bool need_log_inode(struct btrfs_trans_handle *trans,
5534 struct btrfs_inode *inode)
5537 * If this inode does not have new/updated/deleted xattrs since the last
5538 * time it was logged and is flagged as logged in the current transaction,
5539 * we can skip logging it. As for new/deleted names, those are updated in
5540 * the log by link/unlink/rename operations.
5541 * In case the inode was logged and then evicted and reloaded, its
5542 * logged_trans will be 0, in which case we have to fully log it since
5543 * logged_trans is a transient field, not persisted.
5545 if (inode->logged_trans == trans->transid &&
5546 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5552 struct btrfs_dir_list {
5554 struct list_head list;
5558 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5559 * details about the why it is needed.
5560 * This is a recursive operation - if an existing dentry corresponds to a
5561 * directory, that directory's new entries are logged too (same behaviour as
5562 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5563 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5564 * complains about the following circular lock dependency / possible deadlock:
5568 * lock(&type->i_mutex_dir_key#3/2);
5569 * lock(sb_internal#2);
5570 * lock(&type->i_mutex_dir_key#3/2);
5571 * lock(&sb->s_type->i_mutex_key#14);
5573 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5574 * sb_start_intwrite() in btrfs_start_transaction().
5575 * Not locking i_mutex of the inodes is still safe because:
5577 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5578 * that while logging the inode new references (names) are added or removed
5579 * from the inode, leaving the logged inode item with a link count that does
5580 * not match the number of logged inode reference items. This is fine because
5581 * at log replay time we compute the real number of links and correct the
5582 * link count in the inode item (see replay_one_buffer() and
5583 * link_to_fixup_dir());
5585 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5586 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5587 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5588 * has a size that doesn't match the sum of the lengths of all the logged
5589 * names. This does not result in a problem because if a dir_item key is
5590 * logged but its matching dir_index key is not logged, at log replay time we
5591 * don't use it to replay the respective name (see replay_one_name()). On the
5592 * other hand if only the dir_index key ends up being logged, the respective
5593 * name is added to the fs/subvol tree with both the dir_item and dir_index
5594 * keys created (see replay_one_name()).
5595 * The directory's inode item with a wrong i_size is not a problem as well,
5596 * since we don't use it at log replay time to set the i_size in the inode
5597 * item of the fs/subvol tree (see overwrite_item()).
5599 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5600 struct btrfs_root *root,
5601 struct btrfs_inode *start_inode,
5602 struct btrfs_log_ctx *ctx)
5604 struct btrfs_fs_info *fs_info = root->fs_info;
5605 struct btrfs_root *log = root->log_root;
5606 struct btrfs_path *path;
5607 LIST_HEAD(dir_list);
5608 struct btrfs_dir_list *dir_elem;
5611 path = btrfs_alloc_path();
5615 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5617 btrfs_free_path(path);
5620 dir_elem->ino = btrfs_ino(start_inode);
5621 list_add_tail(&dir_elem->list, &dir_list);
5623 while (!list_empty(&dir_list)) {
5624 struct extent_buffer *leaf;
5625 struct btrfs_key min_key;
5629 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5632 goto next_dir_inode;
5634 min_key.objectid = dir_elem->ino;
5635 min_key.type = BTRFS_DIR_ITEM_KEY;
5638 btrfs_release_path(path);
5639 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5641 goto next_dir_inode;
5642 } else if (ret > 0) {
5644 goto next_dir_inode;
5648 leaf = path->nodes[0];
5649 nritems = btrfs_header_nritems(leaf);
5650 for (i = path->slots[0]; i < nritems; i++) {
5651 struct btrfs_dir_item *di;
5652 struct btrfs_key di_key;
5653 struct inode *di_inode;
5654 struct btrfs_dir_list *new_dir_elem;
5655 int log_mode = LOG_INODE_EXISTS;
5658 btrfs_item_key_to_cpu(leaf, &min_key, i);
5659 if (min_key.objectid != dir_elem->ino ||
5660 min_key.type != BTRFS_DIR_ITEM_KEY)
5661 goto next_dir_inode;
5663 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5664 type = btrfs_dir_type(leaf, di);
5665 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5666 type != BTRFS_FT_DIR)
5668 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5669 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5672 btrfs_release_path(path);
5673 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5674 if (IS_ERR(di_inode)) {
5675 ret = PTR_ERR(di_inode);
5676 goto next_dir_inode;
5679 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5680 btrfs_add_delayed_iput(di_inode);
5684 ctx->log_new_dentries = false;
5685 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5686 log_mode = LOG_INODE_ALL;
5687 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5690 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5692 btrfs_add_delayed_iput(di_inode);
5694 goto next_dir_inode;
5695 if (ctx->log_new_dentries) {
5696 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5698 if (!new_dir_elem) {
5700 goto next_dir_inode;
5702 new_dir_elem->ino = di_key.objectid;
5703 list_add_tail(&new_dir_elem->list, &dir_list);
5708 ret = btrfs_next_leaf(log, path);
5710 goto next_dir_inode;
5711 } else if (ret > 0) {
5713 goto next_dir_inode;
5717 if (min_key.offset < (u64)-1) {
5722 list_del(&dir_elem->list);
5726 btrfs_free_path(path);
5730 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5731 struct btrfs_inode *inode,
5732 struct btrfs_log_ctx *ctx)
5734 struct btrfs_fs_info *fs_info = trans->fs_info;
5736 struct btrfs_path *path;
5737 struct btrfs_key key;
5738 struct btrfs_root *root = inode->root;
5739 const u64 ino = btrfs_ino(inode);
5741 path = btrfs_alloc_path();
5744 path->skip_locking = 1;
5745 path->search_commit_root = 1;
5748 key.type = BTRFS_INODE_REF_KEY;
5750 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5755 struct extent_buffer *leaf = path->nodes[0];
5756 int slot = path->slots[0];
5761 if (slot >= btrfs_header_nritems(leaf)) {
5762 ret = btrfs_next_leaf(root, path);
5770 btrfs_item_key_to_cpu(leaf, &key, slot);
5771 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5772 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5775 item_size = btrfs_item_size_nr(leaf, slot);
5776 ptr = btrfs_item_ptr_offset(leaf, slot);
5777 while (cur_offset < item_size) {
5778 struct btrfs_key inode_key;
5779 struct inode *dir_inode;
5781 inode_key.type = BTRFS_INODE_ITEM_KEY;
5782 inode_key.offset = 0;
5784 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5785 struct btrfs_inode_extref *extref;
5787 extref = (struct btrfs_inode_extref *)
5789 inode_key.objectid = btrfs_inode_extref_parent(
5791 cur_offset += sizeof(*extref);
5792 cur_offset += btrfs_inode_extref_name_len(leaf,
5795 inode_key.objectid = key.offset;
5796 cur_offset = item_size;
5799 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5802 * If the parent inode was deleted, return an error to
5803 * fallback to a transaction commit. This is to prevent
5804 * getting an inode that was moved from one parent A to
5805 * a parent B, got its former parent A deleted and then
5806 * it got fsync'ed, from existing at both parents after
5807 * a log replay (and the old parent still existing).
5814 * mv /mnt/B/bar /mnt/A/bar
5815 * mv -T /mnt/A /mnt/B
5819 * If we ignore the old parent B which got deleted,
5820 * after a log replay we would have file bar linked
5821 * at both parents and the old parent B would still
5824 if (IS_ERR(dir_inode)) {
5825 ret = PTR_ERR(dir_inode);
5830 ctx->log_new_dentries = false;
5831 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5832 LOG_INODE_ALL, ctx);
5834 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5836 if (!ret && ctx && ctx->log_new_dentries)
5837 ret = log_new_dir_dentries(trans, root,
5838 BTRFS_I(dir_inode), ctx);
5839 btrfs_add_delayed_iput(dir_inode);
5847 btrfs_free_path(path);
5851 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5852 struct btrfs_root *root,
5853 struct btrfs_path *path,
5854 struct btrfs_log_ctx *ctx)
5856 struct btrfs_key found_key;
5858 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5861 struct btrfs_fs_info *fs_info = root->fs_info;
5862 struct extent_buffer *leaf = path->nodes[0];
5863 int slot = path->slots[0];
5864 struct btrfs_key search_key;
5865 struct inode *inode;
5869 btrfs_release_path(path);
5871 ino = found_key.offset;
5873 search_key.objectid = found_key.offset;
5874 search_key.type = BTRFS_INODE_ITEM_KEY;
5875 search_key.offset = 0;
5876 inode = btrfs_iget(fs_info->sb, ino, root);
5878 return PTR_ERR(inode);
5880 if (BTRFS_I(inode)->generation >= trans->transid &&
5881 need_log_inode(trans, BTRFS_I(inode)))
5882 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5883 LOG_INODE_EXISTS, ctx);
5884 btrfs_add_delayed_iput(inode);
5888 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5891 search_key.type = BTRFS_INODE_REF_KEY;
5892 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5896 leaf = path->nodes[0];
5897 slot = path->slots[0];
5898 if (slot >= btrfs_header_nritems(leaf)) {
5899 ret = btrfs_next_leaf(root, path);
5904 leaf = path->nodes[0];
5905 slot = path->slots[0];
5908 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5909 if (found_key.objectid != search_key.objectid ||
5910 found_key.type != BTRFS_INODE_REF_KEY)
5916 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5917 struct btrfs_inode *inode,
5918 struct dentry *parent,
5919 struct btrfs_log_ctx *ctx)
5921 struct btrfs_root *root = inode->root;
5922 struct dentry *old_parent = NULL;
5923 struct super_block *sb = inode->vfs_inode.i_sb;
5927 if (!parent || d_really_is_negative(parent) ||
5931 inode = BTRFS_I(d_inode(parent));
5932 if (root != inode->root)
5935 if (inode->generation >= trans->transid &&
5936 need_log_inode(trans, inode)) {
5937 ret = btrfs_log_inode(trans, root, inode,
5938 LOG_INODE_EXISTS, ctx);
5942 if (IS_ROOT(parent))
5945 parent = dget_parent(parent);
5947 old_parent = parent;
5954 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5955 struct btrfs_inode *inode,
5956 struct dentry *parent,
5957 struct btrfs_log_ctx *ctx)
5959 struct btrfs_root *root = inode->root;
5960 const u64 ino = btrfs_ino(inode);
5961 struct btrfs_path *path;
5962 struct btrfs_key search_key;
5966 * For a single hard link case, go through a fast path that does not
5967 * need to iterate the fs/subvolume tree.
5969 if (inode->vfs_inode.i_nlink < 2)
5970 return log_new_ancestors_fast(trans, inode, parent, ctx);
5972 path = btrfs_alloc_path();
5976 search_key.objectid = ino;
5977 search_key.type = BTRFS_INODE_REF_KEY;
5978 search_key.offset = 0;
5980 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5987 struct extent_buffer *leaf = path->nodes[0];
5988 int slot = path->slots[0];
5989 struct btrfs_key found_key;
5991 if (slot >= btrfs_header_nritems(leaf)) {
5992 ret = btrfs_next_leaf(root, path);
6000 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6001 if (found_key.objectid != ino ||
6002 found_key.type > BTRFS_INODE_EXTREF_KEY)
6006 * Don't deal with extended references because they are rare
6007 * cases and too complex to deal with (we would need to keep
6008 * track of which subitem we are processing for each item in
6009 * this loop, etc). So just return some error to fallback to
6010 * a transaction commit.
6012 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6018 * Logging ancestors needs to do more searches on the fs/subvol
6019 * tree, so it releases the path as needed to avoid deadlocks.
6020 * Keep track of the last inode ref key and resume from that key
6021 * after logging all new ancestors for the current hard link.
6023 memcpy(&search_key, &found_key, sizeof(search_key));
6025 ret = log_new_ancestors(trans, root, path, ctx);
6028 btrfs_release_path(path);
6033 btrfs_free_path(path);
6038 * helper function around btrfs_log_inode to make sure newly created
6039 * parent directories also end up in the log. A minimal inode and backref
6040 * only logging is done of any parent directories that are older than
6041 * the last committed transaction
6043 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6044 struct btrfs_inode *inode,
6045 struct dentry *parent,
6047 struct btrfs_log_ctx *ctx)
6049 struct btrfs_root *root = inode->root;
6050 struct btrfs_fs_info *fs_info = root->fs_info;
6051 struct super_block *sb;
6053 bool log_dentries = false;
6055 sb = inode->vfs_inode.i_sb;
6057 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6062 if (btrfs_root_refs(&root->root_item) == 0) {
6067 ret = check_parent_dirs_for_sync(trans, inode, parent, sb);
6072 * Skip already logged inodes or inodes corresponding to tmpfiles
6073 * (since logging them is pointless, a link count of 0 means they
6074 * will never be accessible).
6076 if (btrfs_inode_in_log(inode, trans->transid) ||
6077 inode->vfs_inode.i_nlink == 0) {
6078 ret = BTRFS_NO_LOG_SYNC;
6082 ret = start_log_trans(trans, root, ctx);
6086 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6091 * for regular files, if its inode is already on disk, we don't
6092 * have to worry about the parents at all. This is because
6093 * we can use the last_unlink_trans field to record renames
6094 * and other fun in this file.
6096 if (S_ISREG(inode->vfs_inode.i_mode) &&
6097 inode->generation < trans->transid &&
6098 inode->last_unlink_trans < trans->transid) {
6103 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6104 log_dentries = true;
6107 * On unlink we must make sure all our current and old parent directory
6108 * inodes are fully logged. This is to prevent leaving dangling
6109 * directory index entries in directories that were our parents but are
6110 * not anymore. Not doing this results in old parent directory being
6111 * impossible to delete after log replay (rmdir will always fail with
6112 * error -ENOTEMPTY).
6118 * ln testdir/foo testdir/bar
6120 * unlink testdir/bar
6121 * xfs_io -c fsync testdir/foo
6123 * mount fs, triggers log replay
6125 * If we don't log the parent directory (testdir), after log replay the
6126 * directory still has an entry pointing to the file inode using the bar
6127 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6128 * the file inode has a link count of 1.
6134 * ln foo testdir/foo2
6135 * ln foo testdir/foo3
6137 * unlink testdir/foo3
6138 * xfs_io -c fsync foo
6140 * mount fs, triggers log replay
6142 * Similar as the first example, after log replay the parent directory
6143 * testdir still has an entry pointing to the inode file with name foo3
6144 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6145 * and has a link count of 2.
6147 if (inode->last_unlink_trans >= trans->transid) {
6148 ret = btrfs_log_all_parents(trans, inode, ctx);
6153 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6158 ret = log_new_dir_dentries(trans, root, inode, ctx);
6163 btrfs_set_log_full_commit(trans);
6168 btrfs_remove_log_ctx(root, ctx);
6169 btrfs_end_log_trans(root);
6175 * it is not safe to log dentry if the chunk root has added new
6176 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6177 * If this returns 1, you must commit the transaction to safely get your
6180 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6181 struct dentry *dentry,
6182 struct btrfs_log_ctx *ctx)
6184 struct dentry *parent = dget_parent(dentry);
6187 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6188 LOG_INODE_ALL, ctx);
6195 * should be called during mount to recover any replay any log trees
6198 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6201 struct btrfs_path *path;
6202 struct btrfs_trans_handle *trans;
6203 struct btrfs_key key;
6204 struct btrfs_key found_key;
6205 struct btrfs_root *log;
6206 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6207 struct walk_control wc = {
6208 .process_func = process_one_buffer,
6209 .stage = LOG_WALK_PIN_ONLY,
6212 path = btrfs_alloc_path();
6216 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6218 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6219 if (IS_ERR(trans)) {
6220 ret = PTR_ERR(trans);
6227 ret = walk_log_tree(trans, log_root_tree, &wc);
6229 btrfs_handle_fs_error(fs_info, ret,
6230 "Failed to pin buffers while recovering log root tree.");
6235 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6236 key.offset = (u64)-1;
6237 key.type = BTRFS_ROOT_ITEM_KEY;
6240 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6243 btrfs_handle_fs_error(fs_info, ret,
6244 "Couldn't find tree log root.");
6248 if (path->slots[0] == 0)
6252 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6254 btrfs_release_path(path);
6255 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6258 log = btrfs_read_tree_root(log_root_tree, &found_key);
6261 btrfs_handle_fs_error(fs_info, ret,
6262 "Couldn't read tree log root.");
6266 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6268 if (IS_ERR(wc.replay_dest)) {
6269 ret = PTR_ERR(wc.replay_dest);
6272 * We didn't find the subvol, likely because it was
6273 * deleted. This is ok, simply skip this log and go to
6276 * We need to exclude the root because we can't have
6277 * other log replays overwriting this log as we'll read
6278 * it back in a few more times. This will keep our
6279 * block from being modified, and we'll just bail for
6280 * each subsequent pass.
6283 ret = btrfs_pin_extent_for_log_replay(trans,
6286 btrfs_put_root(log);
6290 btrfs_handle_fs_error(fs_info, ret,
6291 "Couldn't read target root for tree log recovery.");
6295 wc.replay_dest->log_root = log;
6296 btrfs_record_root_in_trans(trans, wc.replay_dest);
6297 ret = walk_log_tree(trans, log, &wc);
6299 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6300 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6304 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6305 struct btrfs_root *root = wc.replay_dest;
6307 btrfs_release_path(path);
6310 * We have just replayed everything, and the highest
6311 * objectid of fs roots probably has changed in case
6312 * some inode_item's got replayed.
6314 * root->objectid_mutex is not acquired as log replay
6315 * could only happen during mount.
6317 ret = btrfs_init_root_free_objectid(root);
6320 wc.replay_dest->log_root = NULL;
6321 btrfs_put_root(wc.replay_dest);
6322 btrfs_put_root(log);
6327 if (found_key.offset == 0)
6329 key.offset = found_key.offset - 1;
6331 btrfs_release_path(path);
6333 /* step one is to pin it all, step two is to replay just inodes */
6336 wc.process_func = replay_one_buffer;
6337 wc.stage = LOG_WALK_REPLAY_INODES;
6340 /* step three is to replay everything */
6341 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6346 btrfs_free_path(path);
6348 /* step 4: commit the transaction, which also unpins the blocks */
6349 ret = btrfs_commit_transaction(trans);
6353 log_root_tree->log_root = NULL;
6354 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6355 btrfs_put_root(log_root_tree);
6360 btrfs_end_transaction(wc.trans);
6361 btrfs_free_path(path);
6366 * there are some corner cases where we want to force a full
6367 * commit instead of allowing a directory to be logged.
6369 * They revolve around files there were unlinked from the directory, and
6370 * this function updates the parent directory so that a full commit is
6371 * properly done if it is fsync'd later after the unlinks are done.
6373 * Must be called before the unlink operations (updates to the subvolume tree,
6374 * inodes, etc) are done.
6376 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6377 struct btrfs_inode *dir, struct btrfs_inode *inode,
6381 * when we're logging a file, if it hasn't been renamed
6382 * or unlinked, and its inode is fully committed on disk,
6383 * we don't have to worry about walking up the directory chain
6384 * to log its parents.
6386 * So, we use the last_unlink_trans field to put this transid
6387 * into the file. When the file is logged we check it and
6388 * don't log the parents if the file is fully on disk.
6390 mutex_lock(&inode->log_mutex);
6391 inode->last_unlink_trans = trans->transid;
6392 mutex_unlock(&inode->log_mutex);
6395 * if this directory was already logged any new
6396 * names for this file/dir will get recorded
6398 if (dir->logged_trans == trans->transid)
6402 * if the inode we're about to unlink was logged,
6403 * the log will be properly updated for any new names
6405 if (inode->logged_trans == trans->transid)
6409 * when renaming files across directories, if the directory
6410 * there we're unlinking from gets fsync'd later on, there's
6411 * no way to find the destination directory later and fsync it
6412 * properly. So, we have to be conservative and force commits
6413 * so the new name gets discovered.
6418 /* we can safely do the unlink without any special recording */
6422 mutex_lock(&dir->log_mutex);
6423 dir->last_unlink_trans = trans->transid;
6424 mutex_unlock(&dir->log_mutex);
6428 * Make sure that if someone attempts to fsync the parent directory of a deleted
6429 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6430 * that after replaying the log tree of the parent directory's root we will not
6431 * see the snapshot anymore and at log replay time we will not see any log tree
6432 * corresponding to the deleted snapshot's root, which could lead to replaying
6433 * it after replaying the log tree of the parent directory (which would replay
6434 * the snapshot delete operation).
6436 * Must be called before the actual snapshot destroy operation (updates to the
6437 * parent root and tree of tree roots trees, etc) are done.
6439 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6440 struct btrfs_inode *dir)
6442 mutex_lock(&dir->log_mutex);
6443 dir->last_unlink_trans = trans->transid;
6444 mutex_unlock(&dir->log_mutex);
6448 * Call this after adding a new name for a file and it will properly
6449 * update the log to reflect the new name.
6451 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6452 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6453 struct dentry *parent)
6455 struct btrfs_log_ctx ctx;
6458 * this will force the logging code to walk the dentry chain
6461 if (!S_ISDIR(inode->vfs_inode.i_mode))
6462 inode->last_unlink_trans = trans->transid;
6465 * if this inode hasn't been logged and directory we're renaming it
6466 * from hasn't been logged, we don't need to log it
6468 if (inode->logged_trans < trans->transid &&
6469 (!old_dir || old_dir->logged_trans < trans->transid))
6472 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6473 ctx.logging_new_name = true;
6475 * We don't care about the return value. If we fail to log the new name
6476 * then we know the next attempt to sync the log will fallback to a full
6477 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6478 * we don't need to worry about getting a log committed that has an
6479 * inconsistent state after a rename operation.
6481 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);