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 "inode-map.h"
21 #include "block-group.h"
22 #include "space-info.h"
24 /* magic values for the inode_only field in btrfs_log_inode:
26 * LOG_INODE_ALL means to log everything
27 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 * directory trouble cases
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
47 * rename foo/some_dir foo2/some_dir
49 * fsync foo/some_dir/some_file
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
81 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
91 LOG_WALK_REPLAY_INODES,
92 LOG_WALK_REPLAY_DIR_INDEX,
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct btrfs_inode *inode,
101 struct btrfs_log_ctx *ctx);
102 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_path *path, u64 objectid);
105 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_root *log,
108 struct btrfs_path *path,
109 u64 dirid, int del_all);
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
139 static int start_log_trans(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root,
141 struct btrfs_log_ctx *ctx)
143 struct btrfs_fs_info *fs_info = root->fs_info;
146 mutex_lock(&root->log_mutex);
148 if (root->log_root) {
149 if (btrfs_need_log_full_commit(trans)) {
154 if (!root->log_start_pid) {
155 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
156 root->log_start_pid = current->pid;
157 } else if (root->log_start_pid != current->pid) {
158 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
161 mutex_lock(&fs_info->tree_log_mutex);
162 if (!fs_info->log_root_tree)
163 ret = btrfs_init_log_root_tree(trans, fs_info);
164 mutex_unlock(&fs_info->tree_log_mutex);
168 ret = btrfs_add_log_tree(trans, root);
172 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
173 root->log_start_pid = current->pid;
176 atomic_inc(&root->log_batch);
177 atomic_inc(&root->log_writers);
179 int index = root->log_transid % 2;
180 list_add_tail(&ctx->list, &root->log_ctxs[index]);
181 ctx->log_transid = root->log_transid;
185 mutex_unlock(&root->log_mutex);
190 * returns 0 if there was a log transaction running and we were able
191 * to join, or returns -ENOENT if there were not transactions
194 static int join_running_log_trans(struct btrfs_root *root)
198 mutex_lock(&root->log_mutex);
199 if (root->log_root) {
201 atomic_inc(&root->log_writers);
203 mutex_unlock(&root->log_mutex);
208 * This either makes the current running log transaction wait
209 * until you call btrfs_end_log_trans() or it makes any future
210 * log transactions wait until you call btrfs_end_log_trans()
212 void btrfs_pin_log_trans(struct btrfs_root *root)
214 mutex_lock(&root->log_mutex);
215 atomic_inc(&root->log_writers);
216 mutex_unlock(&root->log_mutex);
220 * indicate we're done making changes to the log tree
221 * and wake up anyone waiting to do a sync
223 void btrfs_end_log_trans(struct btrfs_root *root)
225 if (atomic_dec_and_test(&root->log_writers)) {
226 /* atomic_dec_and_test implies a barrier */
227 cond_wake_up_nomb(&root->log_writer_wait);
231 static int btrfs_write_tree_block(struct extent_buffer *buf)
233 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
234 buf->start + buf->len - 1);
237 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
239 filemap_fdatawait_range(buf->pages[0]->mapping,
240 buf->start, buf->start + buf->len - 1);
244 * the walk control struct is used to pass state down the chain when
245 * processing the log tree. The stage field tells us which part
246 * of the log tree processing we are currently doing. The others
247 * are state fields used for that specific part
249 struct walk_control {
250 /* should we free the extent on disk when done? This is used
251 * at transaction commit time while freeing a log tree
255 /* should we write out the extent buffer? This is used
256 * while flushing the log tree to disk during a sync
260 /* should we wait for the extent buffer io to finish? Also used
261 * while flushing the log tree to disk for a sync
265 /* pin only walk, we record which extents on disk belong to the
270 /* what stage of the replay code we're currently in */
274 * Ignore any items from the inode currently being processed. Needs
275 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
276 * the LOG_WALK_REPLAY_INODES stage.
278 bool ignore_cur_inode;
280 /* the root we are currently replaying */
281 struct btrfs_root *replay_dest;
283 /* the trans handle for the current replay */
284 struct btrfs_trans_handle *trans;
286 /* the function that gets used to process blocks we find in the
287 * tree. Note the extent_buffer might not be up to date when it is
288 * passed in, and it must be checked or read if you need the data
291 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
292 struct walk_control *wc, u64 gen, int level);
296 * process_func used to pin down extents, write them or wait on them
298 static int process_one_buffer(struct btrfs_root *log,
299 struct extent_buffer *eb,
300 struct walk_control *wc, u64 gen, int level)
302 struct btrfs_fs_info *fs_info = log->fs_info;
306 * If this fs is mixed then we need to be able to process the leaves to
307 * pin down any logged extents, so we have to read the block.
309 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
310 ret = btrfs_read_buffer(eb, gen, level, NULL);
316 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
319 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
320 if (wc->pin && btrfs_header_level(eb) == 0)
321 ret = btrfs_exclude_logged_extents(eb);
323 btrfs_write_tree_block(eb);
325 btrfs_wait_tree_block_writeback(eb);
331 * Item overwrite used by replay and tree logging. eb, slot and key all refer
332 * to the src data we are copying out.
334 * root is the tree we are copying into, and path is a scratch
335 * path for use in this function (it should be released on entry and
336 * will be released on exit).
338 * If the key is already in the destination tree the existing item is
339 * overwritten. If the existing item isn't big enough, it is extended.
340 * If it is too large, it is truncated.
342 * If the key isn't in the destination yet, a new item is inserted.
344 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
345 struct btrfs_root *root,
346 struct btrfs_path *path,
347 struct extent_buffer *eb, int slot,
348 struct btrfs_key *key)
352 u64 saved_i_size = 0;
353 int save_old_i_size = 0;
354 unsigned long src_ptr;
355 unsigned long dst_ptr;
356 int overwrite_root = 0;
357 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
359 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
362 item_size = btrfs_item_size_nr(eb, slot);
363 src_ptr = btrfs_item_ptr_offset(eb, slot);
365 /* look for the key in the destination tree */
366 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
373 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
375 if (dst_size != item_size)
378 if (item_size == 0) {
379 btrfs_release_path(path);
382 dst_copy = kmalloc(item_size, GFP_NOFS);
383 src_copy = kmalloc(item_size, GFP_NOFS);
384 if (!dst_copy || !src_copy) {
385 btrfs_release_path(path);
391 read_extent_buffer(eb, src_copy, src_ptr, item_size);
393 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
394 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
396 ret = memcmp(dst_copy, src_copy, item_size);
401 * they have the same contents, just return, this saves
402 * us from cowing blocks in the destination tree and doing
403 * extra writes that may not have been done by a previous
407 btrfs_release_path(path);
412 * We need to load the old nbytes into the inode so when we
413 * replay the extents we've logged we get the right nbytes.
416 struct btrfs_inode_item *item;
420 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
421 struct btrfs_inode_item);
422 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
423 item = btrfs_item_ptr(eb, slot,
424 struct btrfs_inode_item);
425 btrfs_set_inode_nbytes(eb, item, nbytes);
428 * If this is a directory we need to reset the i_size to
429 * 0 so that we can set it up properly when replaying
430 * the rest of the items in this log.
432 mode = btrfs_inode_mode(eb, item);
434 btrfs_set_inode_size(eb, item, 0);
436 } else if (inode_item) {
437 struct btrfs_inode_item *item;
441 * New inode, set nbytes to 0 so that the nbytes comes out
442 * properly when we replay the extents.
444 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
445 btrfs_set_inode_nbytes(eb, item, 0);
448 * If this is a directory we need to reset the i_size to 0 so
449 * that we can set it up properly when replaying the rest of
450 * the items in this log.
452 mode = btrfs_inode_mode(eb, item);
454 btrfs_set_inode_size(eb, item, 0);
457 btrfs_release_path(path);
458 /* try to insert the key into the destination tree */
459 path->skip_release_on_error = 1;
460 ret = btrfs_insert_empty_item(trans, root, path,
462 path->skip_release_on_error = 0;
464 /* make sure any existing item is the correct size */
465 if (ret == -EEXIST || ret == -EOVERFLOW) {
467 found_size = btrfs_item_size_nr(path->nodes[0],
469 if (found_size > item_size)
470 btrfs_truncate_item(path, item_size, 1);
471 else if (found_size < item_size)
472 btrfs_extend_item(path, item_size - found_size);
476 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
479 /* don't overwrite an existing inode if the generation number
480 * was logged as zero. This is done when the tree logging code
481 * is just logging an inode to make sure it exists after recovery.
483 * Also, don't overwrite i_size on directories during replay.
484 * log replay inserts and removes directory items based on the
485 * state of the tree found in the subvolume, and i_size is modified
488 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
489 struct btrfs_inode_item *src_item;
490 struct btrfs_inode_item *dst_item;
492 src_item = (struct btrfs_inode_item *)src_ptr;
493 dst_item = (struct btrfs_inode_item *)dst_ptr;
495 if (btrfs_inode_generation(eb, src_item) == 0) {
496 struct extent_buffer *dst_eb = path->nodes[0];
497 const u64 ino_size = btrfs_inode_size(eb, src_item);
500 * For regular files an ino_size == 0 is used only when
501 * logging that an inode exists, as part of a directory
502 * fsync, and the inode wasn't fsynced before. In this
503 * case don't set the size of the inode in the fs/subvol
504 * tree, otherwise we would be throwing valid data away.
506 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
507 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
509 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
513 if (overwrite_root &&
514 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
515 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
517 saved_i_size = btrfs_inode_size(path->nodes[0],
522 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
525 if (save_old_i_size) {
526 struct btrfs_inode_item *dst_item;
527 dst_item = (struct btrfs_inode_item *)dst_ptr;
528 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
531 /* make sure the generation is filled in */
532 if (key->type == BTRFS_INODE_ITEM_KEY) {
533 struct btrfs_inode_item *dst_item;
534 dst_item = (struct btrfs_inode_item *)dst_ptr;
535 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
536 btrfs_set_inode_generation(path->nodes[0], dst_item,
541 btrfs_mark_buffer_dirty(path->nodes[0]);
542 btrfs_release_path(path);
547 * simple helper to read an inode off the disk from a given root
548 * This can only be called for subvolume roots and not for the log
550 static noinline struct inode *read_one_inode(struct btrfs_root *root,
555 inode = btrfs_iget(root->fs_info->sb, objectid, root);
561 /* replays a single extent in 'eb' at 'slot' with 'key' into the
562 * subvolume 'root'. path is released on entry and should be released
565 * extents in the log tree have not been allocated out of the extent
566 * tree yet. So, this completes the allocation, taking a reference
567 * as required if the extent already exists or creating a new extent
568 * if it isn't in the extent allocation tree yet.
570 * The extent is inserted into the file, dropping any existing extents
571 * from the file that overlap the new one.
573 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
574 struct btrfs_root *root,
575 struct btrfs_path *path,
576 struct extent_buffer *eb, int slot,
577 struct btrfs_key *key)
579 struct btrfs_fs_info *fs_info = root->fs_info;
582 u64 start = key->offset;
584 struct btrfs_file_extent_item *item;
585 struct inode *inode = NULL;
589 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
590 found_type = btrfs_file_extent_type(eb, item);
592 if (found_type == BTRFS_FILE_EXTENT_REG ||
593 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
594 nbytes = btrfs_file_extent_num_bytes(eb, item);
595 extent_end = start + nbytes;
598 * We don't add to the inodes nbytes if we are prealloc or a
601 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
603 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
604 size = btrfs_file_extent_ram_bytes(eb, item);
605 nbytes = btrfs_file_extent_ram_bytes(eb, item);
606 extent_end = ALIGN(start + size,
607 fs_info->sectorsize);
613 inode = read_one_inode(root, key->objectid);
620 * first check to see if we already have this extent in the
621 * file. This must be done before the btrfs_drop_extents run
622 * so we don't try to drop this extent.
624 ret = btrfs_lookup_file_extent(trans, root, path,
625 btrfs_ino(BTRFS_I(inode)), start, 0);
628 (found_type == BTRFS_FILE_EXTENT_REG ||
629 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
630 struct btrfs_file_extent_item cmp1;
631 struct btrfs_file_extent_item cmp2;
632 struct btrfs_file_extent_item *existing;
633 struct extent_buffer *leaf;
635 leaf = path->nodes[0];
636 existing = btrfs_item_ptr(leaf, path->slots[0],
637 struct btrfs_file_extent_item);
639 read_extent_buffer(eb, &cmp1, (unsigned long)item,
641 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
645 * we already have a pointer to this exact extent,
646 * we don't have to do anything
648 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
649 btrfs_release_path(path);
653 btrfs_release_path(path);
655 /* drop any overlapping extents */
656 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
660 if (found_type == BTRFS_FILE_EXTENT_REG ||
661 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
663 unsigned long dest_offset;
664 struct btrfs_key ins;
666 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
667 btrfs_fs_incompat(fs_info, NO_HOLES))
670 ret = btrfs_insert_empty_item(trans, root, path, key,
674 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
676 copy_extent_buffer(path->nodes[0], eb, dest_offset,
677 (unsigned long)item, sizeof(*item));
679 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
680 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
681 ins.type = BTRFS_EXTENT_ITEM_KEY;
682 offset = key->offset - btrfs_file_extent_offset(eb, item);
685 * Manually record dirty extent, as here we did a shallow
686 * file extent item copy and skip normal backref update,
687 * but modifying extent tree all by ourselves.
688 * So need to manually record dirty extent for qgroup,
689 * as the owner of the file extent changed from log tree
690 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
692 ret = btrfs_qgroup_trace_extent(trans,
693 btrfs_file_extent_disk_bytenr(eb, item),
694 btrfs_file_extent_disk_num_bytes(eb, item),
699 if (ins.objectid > 0) {
700 struct btrfs_ref ref = { 0 };
703 LIST_HEAD(ordered_sums);
706 * is this extent already allocated in the extent
707 * allocation tree? If so, just add a reference
709 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
712 btrfs_init_generic_ref(&ref,
713 BTRFS_ADD_DELAYED_REF,
714 ins.objectid, ins.offset, 0);
715 btrfs_init_data_ref(&ref,
716 root->root_key.objectid,
717 key->objectid, offset);
718 ret = btrfs_inc_extent_ref(trans, &ref);
723 * insert the extent pointer in the extent
726 ret = btrfs_alloc_logged_file_extent(trans,
727 root->root_key.objectid,
728 key->objectid, offset, &ins);
732 btrfs_release_path(path);
734 if (btrfs_file_extent_compression(eb, item)) {
735 csum_start = ins.objectid;
736 csum_end = csum_start + ins.offset;
738 csum_start = ins.objectid +
739 btrfs_file_extent_offset(eb, item);
740 csum_end = csum_start +
741 btrfs_file_extent_num_bytes(eb, item);
744 ret = btrfs_lookup_csums_range(root->log_root,
745 csum_start, csum_end - 1,
750 * Now delete all existing cums in the csum root that
751 * cover our range. We do this because we can have an
752 * extent that is completely referenced by one file
753 * extent item and partially referenced by another
754 * file extent item (like after using the clone or
755 * extent_same ioctls). In this case if we end up doing
756 * the replay of the one that partially references the
757 * extent first, and we do not do the csum deletion
758 * below, we can get 2 csum items in the csum tree that
759 * overlap each other. For example, imagine our log has
760 * the two following file extent items:
762 * key (257 EXTENT_DATA 409600)
763 * extent data disk byte 12845056 nr 102400
764 * extent data offset 20480 nr 20480 ram 102400
766 * key (257 EXTENT_DATA 819200)
767 * extent data disk byte 12845056 nr 102400
768 * extent data offset 0 nr 102400 ram 102400
770 * Where the second one fully references the 100K extent
771 * that starts at disk byte 12845056, and the log tree
772 * has a single csum item that covers the entire range
775 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
777 * After the first file extent item is replayed, the
778 * csum tree gets the following csum item:
780 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
782 * Which covers the 20K sub-range starting at offset 20K
783 * of our extent. Now when we replay the second file
784 * extent item, if we do not delete existing csum items
785 * that cover any of its blocks, we end up getting two
786 * csum items in our csum tree that overlap each other:
788 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
789 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
791 * Which is a problem, because after this anyone trying
792 * to lookup up for the checksum of any block of our
793 * extent starting at an offset of 40K or higher, will
794 * end up looking at the second csum item only, which
795 * does not contain the checksum for any block starting
796 * at offset 40K or higher of our extent.
798 while (!list_empty(&ordered_sums)) {
799 struct btrfs_ordered_sum *sums;
800 sums = list_entry(ordered_sums.next,
801 struct btrfs_ordered_sum,
804 ret = btrfs_del_csums(trans,
809 ret = btrfs_csum_file_blocks(trans,
810 fs_info->csum_root, sums);
811 list_del(&sums->list);
817 btrfs_release_path(path);
819 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
820 /* inline extents are easy, we just overwrite them */
821 ret = overwrite_item(trans, root, path, eb, slot, key);
826 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
831 inode_add_bytes(inode, nbytes);
833 ret = btrfs_update_inode(trans, root, inode);
841 * when cleaning up conflicts between the directory names in the
842 * subvolume, directory names in the log and directory names in the
843 * inode back references, we may have to unlink inodes from directories.
845 * This is a helper function to do the unlink of a specific directory
848 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
849 struct btrfs_root *root,
850 struct btrfs_path *path,
851 struct btrfs_inode *dir,
852 struct btrfs_dir_item *di)
857 struct extent_buffer *leaf;
858 struct btrfs_key location;
861 leaf = path->nodes[0];
863 btrfs_dir_item_key_to_cpu(leaf, di, &location);
864 name_len = btrfs_dir_name_len(leaf, di);
865 name = kmalloc(name_len, GFP_NOFS);
869 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
870 btrfs_release_path(path);
872 inode = read_one_inode(root, location.objectid);
878 ret = link_to_fixup_dir(trans, root, path, location.objectid);
882 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
887 ret = btrfs_run_delayed_items(trans);
895 * helper function to see if a given name and sequence number found
896 * in an inode back reference are already in a directory and correctly
897 * point to this inode
899 static noinline int inode_in_dir(struct btrfs_root *root,
900 struct btrfs_path *path,
901 u64 dirid, u64 objectid, u64 index,
902 const char *name, int name_len)
904 struct btrfs_dir_item *di;
905 struct btrfs_key location;
908 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
909 index, name, name_len, 0);
910 if (di && !IS_ERR(di)) {
911 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
912 if (location.objectid != objectid)
916 btrfs_release_path(path);
918 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
919 if (di && !IS_ERR(di)) {
920 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
921 if (location.objectid != objectid)
927 btrfs_release_path(path);
932 * helper function to check a log tree for a named back reference in
933 * an inode. This is used to decide if a back reference that is
934 * found in the subvolume conflicts with what we find in the log.
936 * inode backreferences may have multiple refs in a single item,
937 * during replay we process one reference at a time, and we don't
938 * want to delete valid links to a file from the subvolume if that
939 * link is also in the log.
941 static noinline int backref_in_log(struct btrfs_root *log,
942 struct btrfs_key *key,
944 const char *name, int namelen)
946 struct btrfs_path *path;
949 path = btrfs_alloc_path();
953 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
956 } else if (ret == 1) {
961 if (key->type == BTRFS_INODE_EXTREF_KEY)
962 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
967 ret = !!btrfs_find_name_in_backref(path->nodes[0],
971 btrfs_free_path(path);
975 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
976 struct btrfs_root *root,
977 struct btrfs_path *path,
978 struct btrfs_root *log_root,
979 struct btrfs_inode *dir,
980 struct btrfs_inode *inode,
981 u64 inode_objectid, u64 parent_objectid,
982 u64 ref_index, char *name, int namelen,
988 struct extent_buffer *leaf;
989 struct btrfs_dir_item *di;
990 struct btrfs_key search_key;
991 struct btrfs_inode_extref *extref;
994 /* Search old style refs */
995 search_key.objectid = inode_objectid;
996 search_key.type = BTRFS_INODE_REF_KEY;
997 search_key.offset = parent_objectid;
998 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1000 struct btrfs_inode_ref *victim_ref;
1002 unsigned long ptr_end;
1004 leaf = path->nodes[0];
1006 /* are we trying to overwrite a back ref for the root directory
1007 * if so, just jump out, we're done
1009 if (search_key.objectid == search_key.offset)
1012 /* check all the names in this back reference to see
1013 * if they are in the log. if so, we allow them to stay
1014 * otherwise they must be unlinked as a conflict
1016 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1017 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1018 while (ptr < ptr_end) {
1019 victim_ref = (struct btrfs_inode_ref *)ptr;
1020 victim_name_len = btrfs_inode_ref_name_len(leaf,
1022 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1026 read_extent_buffer(leaf, victim_name,
1027 (unsigned long)(victim_ref + 1),
1030 ret = backref_in_log(log_root, &search_key,
1031 parent_objectid, victim_name,
1037 inc_nlink(&inode->vfs_inode);
1038 btrfs_release_path(path);
1040 ret = btrfs_unlink_inode(trans, root, dir, inode,
1041 victim_name, victim_name_len);
1045 ret = btrfs_run_delayed_items(trans);
1053 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1057 * NOTE: we have searched root tree and checked the
1058 * corresponding ref, it does not need to check again.
1062 btrfs_release_path(path);
1064 /* Same search but for extended refs */
1065 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1066 inode_objectid, parent_objectid, 0,
1068 if (!IS_ERR_OR_NULL(extref)) {
1072 struct inode *victim_parent;
1074 leaf = path->nodes[0];
1076 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1077 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1079 while (cur_offset < item_size) {
1080 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1082 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1084 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1087 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1090 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1093 search_key.objectid = inode_objectid;
1094 search_key.type = BTRFS_INODE_EXTREF_KEY;
1095 search_key.offset = btrfs_extref_hash(parent_objectid,
1098 ret = backref_in_log(log_root, &search_key,
1099 parent_objectid, victim_name,
1105 victim_parent = read_one_inode(root,
1107 if (victim_parent) {
1108 inc_nlink(&inode->vfs_inode);
1109 btrfs_release_path(path);
1111 ret = btrfs_unlink_inode(trans, root,
1112 BTRFS_I(victim_parent),
1117 ret = btrfs_run_delayed_items(
1120 iput(victim_parent);
1129 cur_offset += victim_name_len + sizeof(*extref);
1133 btrfs_release_path(path);
1135 /* look for a conflicting sequence number */
1136 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1137 ref_index, name, namelen, 0);
1138 if (di && !IS_ERR(di)) {
1139 ret = drop_one_dir_item(trans, root, path, dir, di);
1143 btrfs_release_path(path);
1145 /* look for a conflicting name */
1146 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1148 if (di && !IS_ERR(di)) {
1149 ret = drop_one_dir_item(trans, root, path, dir, di);
1153 btrfs_release_path(path);
1158 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1159 u32 *namelen, char **name, u64 *index,
1160 u64 *parent_objectid)
1162 struct btrfs_inode_extref *extref;
1164 extref = (struct btrfs_inode_extref *)ref_ptr;
1166 *namelen = btrfs_inode_extref_name_len(eb, extref);
1167 *name = kmalloc(*namelen, GFP_NOFS);
1171 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1175 *index = btrfs_inode_extref_index(eb, extref);
1176 if (parent_objectid)
1177 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1182 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1183 u32 *namelen, char **name, u64 *index)
1185 struct btrfs_inode_ref *ref;
1187 ref = (struct btrfs_inode_ref *)ref_ptr;
1189 *namelen = btrfs_inode_ref_name_len(eb, ref);
1190 *name = kmalloc(*namelen, GFP_NOFS);
1194 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1197 *index = btrfs_inode_ref_index(eb, ref);
1203 * Take an inode reference item from the log tree and iterate all names from the
1204 * inode reference item in the subvolume tree with the same key (if it exists).
1205 * For any name that is not in the inode reference item from the log tree, do a
1206 * proper unlink of that name (that is, remove its entry from the inode
1207 * reference item and both dir index keys).
1209 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1210 struct btrfs_root *root,
1211 struct btrfs_path *path,
1212 struct btrfs_inode *inode,
1213 struct extent_buffer *log_eb,
1215 struct btrfs_key *key)
1218 unsigned long ref_ptr;
1219 unsigned long ref_end;
1220 struct extent_buffer *eb;
1223 btrfs_release_path(path);
1224 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1232 eb = path->nodes[0];
1233 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1234 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1235 while (ref_ptr < ref_end) {
1240 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1241 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1244 parent_id = key->offset;
1245 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1251 if (key->type == BTRFS_INODE_EXTREF_KEY)
1252 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1256 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1262 btrfs_release_path(path);
1263 dir = read_one_inode(root, parent_id);
1269 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1270 inode, name, namelen);
1280 if (key->type == BTRFS_INODE_EXTREF_KEY)
1281 ref_ptr += sizeof(struct btrfs_inode_extref);
1283 ref_ptr += sizeof(struct btrfs_inode_ref);
1287 btrfs_release_path(path);
1291 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1292 const u8 ref_type, const char *name,
1295 struct btrfs_key key;
1296 struct btrfs_path *path;
1297 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1300 path = btrfs_alloc_path();
1304 key.objectid = btrfs_ino(BTRFS_I(inode));
1305 key.type = ref_type;
1306 if (key.type == BTRFS_INODE_REF_KEY)
1307 key.offset = parent_id;
1309 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1311 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1318 if (key.type == BTRFS_INODE_EXTREF_KEY)
1319 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1320 path->slots[0], parent_id, name, namelen);
1322 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1326 btrfs_free_path(path);
1330 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1331 struct inode *dir, struct inode *inode, const char *name,
1332 int namelen, u64 ref_index)
1334 struct btrfs_dir_item *dir_item;
1335 struct btrfs_key key;
1336 struct btrfs_path *path;
1337 struct inode *other_inode = NULL;
1340 path = btrfs_alloc_path();
1344 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1345 btrfs_ino(BTRFS_I(dir)),
1348 btrfs_release_path(path);
1350 } else if (IS_ERR(dir_item)) {
1351 ret = PTR_ERR(dir_item);
1356 * Our inode's dentry collides with the dentry of another inode which is
1357 * in the log but not yet processed since it has a higher inode number.
1358 * So delete that other dentry.
1360 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1361 btrfs_release_path(path);
1362 other_inode = read_one_inode(root, key.objectid);
1367 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1372 * If we dropped the link count to 0, bump it so that later the iput()
1373 * on the inode will not free it. We will fixup the link count later.
1375 if (other_inode->i_nlink == 0)
1376 inc_nlink(other_inode);
1378 ret = btrfs_run_delayed_items(trans);
1382 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1383 name, namelen, 0, ref_index);
1386 btrfs_free_path(path);
1392 * replay one inode back reference item found in the log tree.
1393 * eb, slot and key refer to the buffer and key found in the log tree.
1394 * root is the destination we are replaying into, and path is for temp
1395 * use by this function. (it should be released on return).
1397 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1398 struct btrfs_root *root,
1399 struct btrfs_root *log,
1400 struct btrfs_path *path,
1401 struct extent_buffer *eb, int slot,
1402 struct btrfs_key *key)
1404 struct inode *dir = NULL;
1405 struct inode *inode = NULL;
1406 unsigned long ref_ptr;
1407 unsigned long ref_end;
1411 int search_done = 0;
1412 int log_ref_ver = 0;
1413 u64 parent_objectid;
1416 int ref_struct_size;
1418 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1419 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1421 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1422 struct btrfs_inode_extref *r;
1424 ref_struct_size = sizeof(struct btrfs_inode_extref);
1426 r = (struct btrfs_inode_extref *)ref_ptr;
1427 parent_objectid = btrfs_inode_extref_parent(eb, r);
1429 ref_struct_size = sizeof(struct btrfs_inode_ref);
1430 parent_objectid = key->offset;
1432 inode_objectid = key->objectid;
1435 * it is possible that we didn't log all the parent directories
1436 * for a given inode. If we don't find the dir, just don't
1437 * copy the back ref in. The link count fixup code will take
1440 dir = read_one_inode(root, parent_objectid);
1446 inode = read_one_inode(root, inode_objectid);
1452 while (ref_ptr < ref_end) {
1454 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1455 &ref_index, &parent_objectid);
1457 * parent object can change from one array
1461 dir = read_one_inode(root, parent_objectid);
1467 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1473 /* if we already have a perfect match, we're done */
1474 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1475 btrfs_ino(BTRFS_I(inode)), ref_index,
1478 * look for a conflicting back reference in the
1479 * metadata. if we find one we have to unlink that name
1480 * of the file before we add our new link. Later on, we
1481 * overwrite any existing back reference, and we don't
1482 * want to create dangling pointers in the directory.
1486 ret = __add_inode_ref(trans, root, path, log,
1491 ref_index, name, namelen,
1501 * If a reference item already exists for this inode
1502 * with the same parent and name, but different index,
1503 * drop it and the corresponding directory index entries
1504 * from the parent before adding the new reference item
1505 * and dir index entries, otherwise we would fail with
1506 * -EEXIST returned from btrfs_add_link() below.
1508 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1511 ret = btrfs_unlink_inode(trans, root,
1516 * If we dropped the link count to 0, bump it so
1517 * that later the iput() on the inode will not
1518 * free it. We will fixup the link count later.
1520 if (!ret && inode->i_nlink == 0)
1526 /* insert our name */
1527 ret = add_link(trans, root, dir, inode, name, namelen,
1532 btrfs_update_inode(trans, root, inode);
1535 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1545 * Before we overwrite the inode reference item in the subvolume tree
1546 * with the item from the log tree, we must unlink all names from the
1547 * parent directory that are in the subvolume's tree inode reference
1548 * item, otherwise we end up with an inconsistent subvolume tree where
1549 * dir index entries exist for a name but there is no inode reference
1550 * item with the same name.
1552 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1557 /* finally write the back reference in the inode */
1558 ret = overwrite_item(trans, root, path, eb, slot, key);
1560 btrfs_release_path(path);
1567 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1568 struct btrfs_root *root, u64 ino)
1572 ret = btrfs_insert_orphan_item(trans, root, ino);
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, 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 = insert_orphan_item(trans, root, ino);
1734 btrfs_free_path(path);
1738 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1739 struct btrfs_root *root,
1740 struct btrfs_path *path)
1743 struct btrfs_key key;
1744 struct inode *inode;
1746 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1747 key.type = BTRFS_ORPHAN_ITEM_KEY;
1748 key.offset = (u64)-1;
1750 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1755 if (path->slots[0] == 0)
1760 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1761 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1762 key.type != BTRFS_ORPHAN_ITEM_KEY)
1765 ret = btrfs_del_item(trans, root, path);
1769 btrfs_release_path(path);
1770 inode = read_one_inode(root, key.offset);
1774 ret = fixup_inode_link_count(trans, root, inode);
1780 * fixup on a directory may create new entries,
1781 * make sure we always look for the highset possible
1784 key.offset = (u64)-1;
1788 btrfs_release_path(path);
1794 * record a given inode in the fixup dir so we can check its link
1795 * count when replay is done. The link count is incremented here
1796 * so the inode won't go away until we check it
1798 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1799 struct btrfs_root *root,
1800 struct btrfs_path *path,
1803 struct btrfs_key key;
1805 struct inode *inode;
1807 inode = read_one_inode(root, objectid);
1811 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1812 key.type = BTRFS_ORPHAN_ITEM_KEY;
1813 key.offset = objectid;
1815 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1817 btrfs_release_path(path);
1819 if (!inode->i_nlink)
1820 set_nlink(inode, 1);
1823 ret = btrfs_update_inode(trans, root, inode);
1824 } else if (ret == -EEXIST) {
1827 BUG(); /* Logic Error */
1835 * when replaying the log for a directory, we only insert names
1836 * for inodes that actually exist. This means an fsync on a directory
1837 * does not implicitly fsync all the new files in it
1839 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1840 struct btrfs_root *root,
1841 u64 dirid, u64 index,
1842 char *name, int name_len,
1843 struct btrfs_key *location)
1845 struct inode *inode;
1849 inode = read_one_inode(root, location->objectid);
1853 dir = read_one_inode(root, dirid);
1859 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1860 name_len, 1, index);
1862 /* FIXME, put inode into FIXUP list */
1870 * take a single entry in a log directory item and replay it into
1873 * if a conflicting item exists in the subdirectory already,
1874 * the inode it points to is unlinked and put into the link count
1877 * If a name from the log points to a file or directory that does
1878 * not exist in the FS, it is skipped. fsyncs on directories
1879 * do not force down inodes inside that directory, just changes to the
1880 * names or unlinks in a directory.
1882 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1883 * non-existing inode) and 1 if the name was replayed.
1885 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1886 struct btrfs_root *root,
1887 struct btrfs_path *path,
1888 struct extent_buffer *eb,
1889 struct btrfs_dir_item *di,
1890 struct btrfs_key *key)
1894 struct btrfs_dir_item *dst_di;
1895 struct btrfs_key found_key;
1896 struct btrfs_key log_key;
1901 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1902 bool name_added = false;
1904 dir = read_one_inode(root, key->objectid);
1908 name_len = btrfs_dir_name_len(eb, di);
1909 name = kmalloc(name_len, GFP_NOFS);
1915 log_type = btrfs_dir_type(eb, di);
1916 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1919 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1920 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1925 btrfs_release_path(path);
1927 if (key->type == BTRFS_DIR_ITEM_KEY) {
1928 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1930 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1931 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1940 if (IS_ERR_OR_NULL(dst_di)) {
1941 /* we need a sequence number to insert, so we only
1942 * do inserts for the BTRFS_DIR_INDEX_KEY types
1944 if (key->type != BTRFS_DIR_INDEX_KEY)
1949 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1950 /* the existing item matches the logged item */
1951 if (found_key.objectid == log_key.objectid &&
1952 found_key.type == log_key.type &&
1953 found_key.offset == log_key.offset &&
1954 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1955 update_size = false;
1960 * don't drop the conflicting directory entry if the inode
1961 * for the new entry doesn't exist
1966 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1970 if (key->type == BTRFS_DIR_INDEX_KEY)
1973 btrfs_release_path(path);
1974 if (!ret && update_size) {
1975 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1976 ret = btrfs_update_inode(trans, root, dir);
1980 if (!ret && name_added)
1986 * Check if the inode reference exists in the log for the given name,
1987 * inode and parent inode
1989 found_key.objectid = log_key.objectid;
1990 found_key.type = BTRFS_INODE_REF_KEY;
1991 found_key.offset = key->objectid;
1992 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
1996 /* The dentry will be added later. */
1998 update_size = false;
2002 found_key.objectid = log_key.objectid;
2003 found_key.type = BTRFS_INODE_EXTREF_KEY;
2004 found_key.offset = key->objectid;
2005 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2010 /* The dentry will be added later. */
2012 update_size = false;
2015 btrfs_release_path(path);
2016 ret = insert_one_name(trans, root, key->objectid, key->offset,
2017 name, name_len, &log_key);
2018 if (ret && ret != -ENOENT && ret != -EEXIST)
2022 update_size = false;
2028 * find all the names in a directory item and reconcile them into
2029 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2030 * one name in a directory item, but the same code gets used for
2031 * both directory index types
2033 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2034 struct btrfs_root *root,
2035 struct btrfs_path *path,
2036 struct extent_buffer *eb, int slot,
2037 struct btrfs_key *key)
2040 u32 item_size = btrfs_item_size_nr(eb, slot);
2041 struct btrfs_dir_item *di;
2044 unsigned long ptr_end;
2045 struct btrfs_path *fixup_path = NULL;
2047 ptr = btrfs_item_ptr_offset(eb, slot);
2048 ptr_end = ptr + item_size;
2049 while (ptr < ptr_end) {
2050 di = (struct btrfs_dir_item *)ptr;
2051 name_len = btrfs_dir_name_len(eb, di);
2052 ret = replay_one_name(trans, root, path, eb, di, key);
2055 ptr = (unsigned long)(di + 1);
2059 * If this entry refers to a non-directory (directories can not
2060 * have a link count > 1) and it was added in the transaction
2061 * that was not committed, make sure we fixup the link count of
2062 * the inode it the entry points to. Otherwise something like
2063 * the following would result in a directory pointing to an
2064 * inode with a wrong link that does not account for this dir
2072 * ln testdir/bar testdir/bar_link
2073 * ln testdir/foo testdir/foo_link
2074 * xfs_io -c "fsync" testdir/bar
2078 * mount fs, log replay happens
2080 * File foo would remain with a link count of 1 when it has two
2081 * entries pointing to it in the directory testdir. This would
2082 * make it impossible to ever delete the parent directory has
2083 * it would result in stale dentries that can never be deleted.
2085 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2086 struct btrfs_key di_key;
2089 fixup_path = btrfs_alloc_path();
2096 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2097 ret = link_to_fixup_dir(trans, root, fixup_path,
2104 btrfs_free_path(fixup_path);
2109 * directory replay has two parts. There are the standard directory
2110 * items in the log copied from the subvolume, and range items
2111 * created in the log while the subvolume was logged.
2113 * The range items tell us which parts of the key space the log
2114 * is authoritative for. During replay, if a key in the subvolume
2115 * directory is in a logged range item, but not actually in the log
2116 * that means it was deleted from the directory before the fsync
2117 * and should be removed.
2119 static noinline int find_dir_range(struct btrfs_root *root,
2120 struct btrfs_path *path,
2121 u64 dirid, int key_type,
2122 u64 *start_ret, u64 *end_ret)
2124 struct btrfs_key key;
2126 struct btrfs_dir_log_item *item;
2130 if (*start_ret == (u64)-1)
2133 key.objectid = dirid;
2134 key.type = key_type;
2135 key.offset = *start_ret;
2137 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2141 if (path->slots[0] == 0)
2146 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2148 if (key.type != key_type || key.objectid != dirid) {
2152 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2153 struct btrfs_dir_log_item);
2154 found_end = btrfs_dir_log_end(path->nodes[0], item);
2156 if (*start_ret >= key.offset && *start_ret <= found_end) {
2158 *start_ret = key.offset;
2159 *end_ret = found_end;
2164 /* check the next slot in the tree to see if it is a valid item */
2165 nritems = btrfs_header_nritems(path->nodes[0]);
2167 if (path->slots[0] >= nritems) {
2168 ret = btrfs_next_leaf(root, path);
2173 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2175 if (key.type != key_type || key.objectid != dirid) {
2179 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2180 struct btrfs_dir_log_item);
2181 found_end = btrfs_dir_log_end(path->nodes[0], item);
2182 *start_ret = key.offset;
2183 *end_ret = found_end;
2186 btrfs_release_path(path);
2191 * this looks for a given directory item in the log. If the directory
2192 * item is not in the log, the item is removed and the inode it points
2195 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2196 struct btrfs_root *root,
2197 struct btrfs_root *log,
2198 struct btrfs_path *path,
2199 struct btrfs_path *log_path,
2201 struct btrfs_key *dir_key)
2204 struct extent_buffer *eb;
2207 struct btrfs_dir_item *di;
2208 struct btrfs_dir_item *log_di;
2211 unsigned long ptr_end;
2213 struct inode *inode;
2214 struct btrfs_key location;
2217 eb = path->nodes[0];
2218 slot = path->slots[0];
2219 item_size = btrfs_item_size_nr(eb, slot);
2220 ptr = btrfs_item_ptr_offset(eb, slot);
2221 ptr_end = ptr + item_size;
2222 while (ptr < ptr_end) {
2223 di = (struct btrfs_dir_item *)ptr;
2224 name_len = btrfs_dir_name_len(eb, di);
2225 name = kmalloc(name_len, GFP_NOFS);
2230 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2233 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2234 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2237 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2238 log_di = btrfs_lookup_dir_index_item(trans, log,
2244 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2245 btrfs_dir_item_key_to_cpu(eb, di, &location);
2246 btrfs_release_path(path);
2247 btrfs_release_path(log_path);
2248 inode = read_one_inode(root, location.objectid);
2254 ret = link_to_fixup_dir(trans, root,
2255 path, location.objectid);
2263 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2264 BTRFS_I(inode), name, name_len);
2266 ret = btrfs_run_delayed_items(trans);
2272 /* there might still be more names under this key
2273 * check and repeat if required
2275 ret = btrfs_search_slot(NULL, root, dir_key, path,
2281 } else if (IS_ERR(log_di)) {
2283 return PTR_ERR(log_di);
2285 btrfs_release_path(log_path);
2288 ptr = (unsigned long)(di + 1);
2293 btrfs_release_path(path);
2294 btrfs_release_path(log_path);
2298 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2299 struct btrfs_root *root,
2300 struct btrfs_root *log,
2301 struct btrfs_path *path,
2304 struct btrfs_key search_key;
2305 struct btrfs_path *log_path;
2310 log_path = btrfs_alloc_path();
2314 search_key.objectid = ino;
2315 search_key.type = BTRFS_XATTR_ITEM_KEY;
2316 search_key.offset = 0;
2318 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2322 nritems = btrfs_header_nritems(path->nodes[0]);
2323 for (i = path->slots[0]; i < nritems; i++) {
2324 struct btrfs_key key;
2325 struct btrfs_dir_item *di;
2326 struct btrfs_dir_item *log_di;
2330 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2331 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2336 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2337 total_size = btrfs_item_size_nr(path->nodes[0], i);
2339 while (cur < total_size) {
2340 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2341 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2342 u32 this_len = sizeof(*di) + name_len + data_len;
2345 name = kmalloc(name_len, GFP_NOFS);
2350 read_extent_buffer(path->nodes[0], name,
2351 (unsigned long)(di + 1), name_len);
2353 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2355 btrfs_release_path(log_path);
2357 /* Doesn't exist in log tree, so delete it. */
2358 btrfs_release_path(path);
2359 di = btrfs_lookup_xattr(trans, root, path, ino,
2360 name, name_len, -1);
2367 ret = btrfs_delete_one_dir_name(trans, root,
2371 btrfs_release_path(path);
2376 if (IS_ERR(log_di)) {
2377 ret = PTR_ERR(log_di);
2381 di = (struct btrfs_dir_item *)((char *)di + this_len);
2384 ret = btrfs_next_leaf(root, path);
2390 btrfs_free_path(log_path);
2391 btrfs_release_path(path);
2397 * deletion replay happens before we copy any new directory items
2398 * out of the log or out of backreferences from inodes. It
2399 * scans the log to find ranges of keys that log is authoritative for,
2400 * and then scans the directory to find items in those ranges that are
2401 * not present in the log.
2403 * Anything we don't find in the log is unlinked and removed from the
2406 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2407 struct btrfs_root *root,
2408 struct btrfs_root *log,
2409 struct btrfs_path *path,
2410 u64 dirid, int del_all)
2414 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2416 struct btrfs_key dir_key;
2417 struct btrfs_key found_key;
2418 struct btrfs_path *log_path;
2421 dir_key.objectid = dirid;
2422 dir_key.type = BTRFS_DIR_ITEM_KEY;
2423 log_path = btrfs_alloc_path();
2427 dir = read_one_inode(root, dirid);
2428 /* it isn't an error if the inode isn't there, that can happen
2429 * because we replay the deletes before we copy in the inode item
2433 btrfs_free_path(log_path);
2441 range_end = (u64)-1;
2443 ret = find_dir_range(log, path, dirid, key_type,
2444 &range_start, &range_end);
2449 dir_key.offset = range_start;
2452 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2457 nritems = btrfs_header_nritems(path->nodes[0]);
2458 if (path->slots[0] >= nritems) {
2459 ret = btrfs_next_leaf(root, path);
2465 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2467 if (found_key.objectid != dirid ||
2468 found_key.type != dir_key.type)
2471 if (found_key.offset > range_end)
2474 ret = check_item_in_log(trans, root, log, path,
2479 if (found_key.offset == (u64)-1)
2481 dir_key.offset = found_key.offset + 1;
2483 btrfs_release_path(path);
2484 if (range_end == (u64)-1)
2486 range_start = range_end + 1;
2491 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2492 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2493 dir_key.type = BTRFS_DIR_INDEX_KEY;
2494 btrfs_release_path(path);
2498 btrfs_release_path(path);
2499 btrfs_free_path(log_path);
2505 * the process_func used to replay items from the log tree. This
2506 * gets called in two different stages. The first stage just looks
2507 * for inodes and makes sure they are all copied into the subvolume.
2509 * The second stage copies all the other item types from the log into
2510 * the subvolume. The two stage approach is slower, but gets rid of
2511 * lots of complexity around inodes referencing other inodes that exist
2512 * only in the log (references come from either directory items or inode
2515 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2516 struct walk_control *wc, u64 gen, int level)
2519 struct btrfs_path *path;
2520 struct btrfs_root *root = wc->replay_dest;
2521 struct btrfs_key key;
2525 ret = btrfs_read_buffer(eb, gen, level, NULL);
2529 level = btrfs_header_level(eb);
2534 path = btrfs_alloc_path();
2538 nritems = btrfs_header_nritems(eb);
2539 for (i = 0; i < nritems; i++) {
2540 btrfs_item_key_to_cpu(eb, &key, i);
2542 /* inode keys are done during the first stage */
2543 if (key.type == BTRFS_INODE_ITEM_KEY &&
2544 wc->stage == LOG_WALK_REPLAY_INODES) {
2545 struct btrfs_inode_item *inode_item;
2548 inode_item = btrfs_item_ptr(eb, i,
2549 struct btrfs_inode_item);
2551 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2552 * and never got linked before the fsync, skip it, as
2553 * replaying it is pointless since it would be deleted
2554 * later. We skip logging tmpfiles, but it's always
2555 * possible we are replaying a log created with a kernel
2556 * that used to log tmpfiles.
2558 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2559 wc->ignore_cur_inode = true;
2562 wc->ignore_cur_inode = false;
2564 ret = replay_xattr_deletes(wc->trans, root, log,
2565 path, key.objectid);
2568 mode = btrfs_inode_mode(eb, inode_item);
2569 if (S_ISDIR(mode)) {
2570 ret = replay_dir_deletes(wc->trans,
2571 root, log, path, key.objectid, 0);
2575 ret = overwrite_item(wc->trans, root, path,
2581 * Before replaying extents, truncate the inode to its
2582 * size. We need to do it now and not after log replay
2583 * because before an fsync we can have prealloc extents
2584 * added beyond the inode's i_size. If we did it after,
2585 * through orphan cleanup for example, we would drop
2586 * those prealloc extents just after replaying them.
2588 if (S_ISREG(mode)) {
2589 struct inode *inode;
2592 inode = read_one_inode(root, key.objectid);
2597 from = ALIGN(i_size_read(inode),
2598 root->fs_info->sectorsize);
2599 ret = btrfs_drop_extents(wc->trans, root, inode,
2602 /* Update the inode's nbytes. */
2603 ret = btrfs_update_inode(wc->trans,
2611 ret = link_to_fixup_dir(wc->trans, root,
2612 path, key.objectid);
2617 if (wc->ignore_cur_inode)
2620 if (key.type == BTRFS_DIR_INDEX_KEY &&
2621 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2622 ret = replay_one_dir_item(wc->trans, root, path,
2628 if (wc->stage < LOG_WALK_REPLAY_ALL)
2631 /* these keys are simply copied */
2632 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2633 ret = overwrite_item(wc->trans, root, path,
2637 } else if (key.type == BTRFS_INODE_REF_KEY ||
2638 key.type == BTRFS_INODE_EXTREF_KEY) {
2639 ret = add_inode_ref(wc->trans, root, log, path,
2641 if (ret && ret != -ENOENT)
2644 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2645 ret = replay_one_extent(wc->trans, root, path,
2649 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2650 ret = replay_one_dir_item(wc->trans, root, path,
2656 btrfs_free_path(path);
2661 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2663 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2665 struct btrfs_block_group *cache;
2667 cache = btrfs_lookup_block_group(fs_info, start);
2669 btrfs_err(fs_info, "unable to find block group for %llu", start);
2673 spin_lock(&cache->space_info->lock);
2674 spin_lock(&cache->lock);
2675 cache->reserved -= fs_info->nodesize;
2676 cache->space_info->bytes_reserved -= fs_info->nodesize;
2677 spin_unlock(&cache->lock);
2678 spin_unlock(&cache->space_info->lock);
2680 btrfs_put_block_group(cache);
2683 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root,
2685 struct btrfs_path *path, int *level,
2686 struct walk_control *wc)
2688 struct btrfs_fs_info *fs_info = root->fs_info;
2691 struct extent_buffer *next;
2692 struct extent_buffer *cur;
2696 while (*level > 0) {
2697 struct btrfs_key first_key;
2699 cur = path->nodes[*level];
2701 WARN_ON(btrfs_header_level(cur) != *level);
2703 if (path->slots[*level] >=
2704 btrfs_header_nritems(cur))
2707 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2708 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2709 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2710 blocksize = fs_info->nodesize;
2712 next = btrfs_find_create_tree_block(fs_info, bytenr);
2714 return PTR_ERR(next);
2717 ret = wc->process_func(root, next, wc, ptr_gen,
2720 free_extent_buffer(next);
2724 path->slots[*level]++;
2726 ret = btrfs_read_buffer(next, ptr_gen,
2727 *level - 1, &first_key);
2729 free_extent_buffer(next);
2734 btrfs_tree_lock(next);
2735 btrfs_set_lock_blocking_write(next);
2736 btrfs_clean_tree_block(next);
2737 btrfs_wait_tree_block_writeback(next);
2738 btrfs_tree_unlock(next);
2739 ret = btrfs_pin_reserved_extent(trans,
2742 free_extent_buffer(next);
2746 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2747 clear_extent_buffer_dirty(next);
2748 unaccount_log_buffer(fs_info, bytenr);
2751 free_extent_buffer(next);
2754 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2756 free_extent_buffer(next);
2760 if (path->nodes[*level-1])
2761 free_extent_buffer(path->nodes[*level-1]);
2762 path->nodes[*level-1] = next;
2763 *level = btrfs_header_level(next);
2764 path->slots[*level] = 0;
2767 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2773 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2774 struct btrfs_root *root,
2775 struct btrfs_path *path, int *level,
2776 struct walk_control *wc)
2778 struct btrfs_fs_info *fs_info = root->fs_info;
2783 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2784 slot = path->slots[i];
2785 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2788 WARN_ON(*level == 0);
2791 ret = wc->process_func(root, path->nodes[*level], wc,
2792 btrfs_header_generation(path->nodes[*level]),
2798 struct extent_buffer *next;
2800 next = path->nodes[*level];
2803 btrfs_tree_lock(next);
2804 btrfs_set_lock_blocking_write(next);
2805 btrfs_clean_tree_block(next);
2806 btrfs_wait_tree_block_writeback(next);
2807 btrfs_tree_unlock(next);
2808 ret = btrfs_pin_reserved_extent(trans,
2809 path->nodes[*level]->start,
2810 path->nodes[*level]->len);
2814 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2815 clear_extent_buffer_dirty(next);
2817 unaccount_log_buffer(fs_info,
2818 path->nodes[*level]->start);
2821 free_extent_buffer(path->nodes[*level]);
2822 path->nodes[*level] = NULL;
2830 * drop the reference count on the tree rooted at 'snap'. This traverses
2831 * the tree freeing any blocks that have a ref count of zero after being
2834 static int walk_log_tree(struct btrfs_trans_handle *trans,
2835 struct btrfs_root *log, struct walk_control *wc)
2837 struct btrfs_fs_info *fs_info = log->fs_info;
2841 struct btrfs_path *path;
2844 path = btrfs_alloc_path();
2848 level = btrfs_header_level(log->node);
2850 path->nodes[level] = log->node;
2851 atomic_inc(&log->node->refs);
2852 path->slots[level] = 0;
2855 wret = walk_down_log_tree(trans, log, path, &level, wc);
2863 wret = walk_up_log_tree(trans, log, path, &level, wc);
2872 /* was the root node processed? if not, catch it here */
2873 if (path->nodes[orig_level]) {
2874 ret = wc->process_func(log, path->nodes[orig_level], wc,
2875 btrfs_header_generation(path->nodes[orig_level]),
2880 struct extent_buffer *next;
2882 next = path->nodes[orig_level];
2885 btrfs_tree_lock(next);
2886 btrfs_set_lock_blocking_write(next);
2887 btrfs_clean_tree_block(next);
2888 btrfs_wait_tree_block_writeback(next);
2889 btrfs_tree_unlock(next);
2890 ret = btrfs_pin_reserved_extent(trans,
2891 next->start, next->len);
2895 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2896 clear_extent_buffer_dirty(next);
2897 unaccount_log_buffer(fs_info, next->start);
2903 btrfs_free_path(path);
2908 * helper function to update the item for a given subvolumes log root
2909 * in the tree of log roots
2911 static int update_log_root(struct btrfs_trans_handle *trans,
2912 struct btrfs_root *log,
2913 struct btrfs_root_item *root_item)
2915 struct btrfs_fs_info *fs_info = log->fs_info;
2918 if (log->log_transid == 1) {
2919 /* insert root item on the first sync */
2920 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2921 &log->root_key, root_item);
2923 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2924 &log->root_key, root_item);
2929 static void wait_log_commit(struct btrfs_root *root, int transid)
2932 int index = transid % 2;
2935 * we only allow two pending log transactions at a time,
2936 * so we know that if ours is more than 2 older than the
2937 * current transaction, we're done
2940 prepare_to_wait(&root->log_commit_wait[index],
2941 &wait, TASK_UNINTERRUPTIBLE);
2943 if (!(root->log_transid_committed < transid &&
2944 atomic_read(&root->log_commit[index])))
2947 mutex_unlock(&root->log_mutex);
2949 mutex_lock(&root->log_mutex);
2951 finish_wait(&root->log_commit_wait[index], &wait);
2954 static void wait_for_writer(struct btrfs_root *root)
2959 prepare_to_wait(&root->log_writer_wait, &wait,
2960 TASK_UNINTERRUPTIBLE);
2961 if (!atomic_read(&root->log_writers))
2964 mutex_unlock(&root->log_mutex);
2966 mutex_lock(&root->log_mutex);
2968 finish_wait(&root->log_writer_wait, &wait);
2971 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2972 struct btrfs_log_ctx *ctx)
2977 mutex_lock(&root->log_mutex);
2978 list_del_init(&ctx->list);
2979 mutex_unlock(&root->log_mutex);
2983 * Invoked in log mutex context, or be sure there is no other task which
2984 * can access the list.
2986 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2987 int index, int error)
2989 struct btrfs_log_ctx *ctx;
2990 struct btrfs_log_ctx *safe;
2992 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2993 list_del_init(&ctx->list);
2994 ctx->log_ret = error;
2997 INIT_LIST_HEAD(&root->log_ctxs[index]);
3001 * btrfs_sync_log does sends a given tree log down to the disk and
3002 * updates the super blocks to record it. When this call is done,
3003 * you know that any inodes previously logged are safely on disk only
3006 * Any other return value means you need to call btrfs_commit_transaction.
3007 * Some of the edge cases for fsyncing directories that have had unlinks
3008 * or renames done in the past mean that sometimes the only safe
3009 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3010 * that has happened.
3012 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3013 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3019 struct btrfs_fs_info *fs_info = root->fs_info;
3020 struct btrfs_root *log = root->log_root;
3021 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3022 struct btrfs_root_item new_root_item;
3023 int log_transid = 0;
3024 struct btrfs_log_ctx root_log_ctx;
3025 struct blk_plug plug;
3027 mutex_lock(&root->log_mutex);
3028 log_transid = ctx->log_transid;
3029 if (root->log_transid_committed >= log_transid) {
3030 mutex_unlock(&root->log_mutex);
3031 return ctx->log_ret;
3034 index1 = log_transid % 2;
3035 if (atomic_read(&root->log_commit[index1])) {
3036 wait_log_commit(root, log_transid);
3037 mutex_unlock(&root->log_mutex);
3038 return ctx->log_ret;
3040 ASSERT(log_transid == root->log_transid);
3041 atomic_set(&root->log_commit[index1], 1);
3043 /* wait for previous tree log sync to complete */
3044 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3045 wait_log_commit(root, log_transid - 1);
3048 int batch = atomic_read(&root->log_batch);
3049 /* when we're on an ssd, just kick the log commit out */
3050 if (!btrfs_test_opt(fs_info, SSD) &&
3051 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3052 mutex_unlock(&root->log_mutex);
3053 schedule_timeout_uninterruptible(1);
3054 mutex_lock(&root->log_mutex);
3056 wait_for_writer(root);
3057 if (batch == atomic_read(&root->log_batch))
3061 /* bail out if we need to do a full commit */
3062 if (btrfs_need_log_full_commit(trans)) {
3064 mutex_unlock(&root->log_mutex);
3068 if (log_transid % 2 == 0)
3069 mark = EXTENT_DIRTY;
3073 /* we start IO on all the marked extents here, but we don't actually
3074 * wait for them until later.
3076 blk_start_plug(&plug);
3077 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3079 blk_finish_plug(&plug);
3080 btrfs_abort_transaction(trans, ret);
3081 btrfs_set_log_full_commit(trans);
3082 mutex_unlock(&root->log_mutex);
3087 * We _must_ update under the root->log_mutex in order to make sure we
3088 * have a consistent view of the log root we are trying to commit at
3091 * We _must_ copy this into a local copy, because we are not holding the
3092 * log_root_tree->log_mutex yet. This is important because when we
3093 * commit the log_root_tree we must have a consistent view of the
3094 * log_root_tree when we update the super block to point at the
3095 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3096 * with the commit and possibly point at the new block which we may not
3099 btrfs_set_root_node(&log->root_item, log->node);
3100 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3102 root->log_transid++;
3103 log->log_transid = root->log_transid;
3104 root->log_start_pid = 0;
3106 * IO has been started, blocks of the log tree have WRITTEN flag set
3107 * in their headers. new modifications of the log will be written to
3108 * new positions. so it's safe to allow log writers to go in.
3110 mutex_unlock(&root->log_mutex);
3112 btrfs_init_log_ctx(&root_log_ctx, NULL);
3114 mutex_lock(&log_root_tree->log_mutex);
3115 atomic_inc(&log_root_tree->log_batch);
3116 atomic_inc(&log_root_tree->log_writers);
3118 index2 = log_root_tree->log_transid % 2;
3119 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3120 root_log_ctx.log_transid = log_root_tree->log_transid;
3122 mutex_unlock(&log_root_tree->log_mutex);
3124 mutex_lock(&log_root_tree->log_mutex);
3127 * Now we are safe to update the log_root_tree because we're under the
3128 * log_mutex, and we're a current writer so we're holding the commit
3129 * open until we drop the log_mutex.
3131 ret = update_log_root(trans, log, &new_root_item);
3133 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3134 /* atomic_dec_and_test implies a barrier */
3135 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3139 if (!list_empty(&root_log_ctx.list))
3140 list_del_init(&root_log_ctx.list);
3142 blk_finish_plug(&plug);
3143 btrfs_set_log_full_commit(trans);
3145 if (ret != -ENOSPC) {
3146 btrfs_abort_transaction(trans, ret);
3147 mutex_unlock(&log_root_tree->log_mutex);
3150 btrfs_wait_tree_log_extents(log, mark);
3151 mutex_unlock(&log_root_tree->log_mutex);
3156 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3157 blk_finish_plug(&plug);
3158 list_del_init(&root_log_ctx.list);
3159 mutex_unlock(&log_root_tree->log_mutex);
3160 ret = root_log_ctx.log_ret;
3164 index2 = root_log_ctx.log_transid % 2;
3165 if (atomic_read(&log_root_tree->log_commit[index2])) {
3166 blk_finish_plug(&plug);
3167 ret = btrfs_wait_tree_log_extents(log, mark);
3168 wait_log_commit(log_root_tree,
3169 root_log_ctx.log_transid);
3170 mutex_unlock(&log_root_tree->log_mutex);
3172 ret = root_log_ctx.log_ret;
3175 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3176 atomic_set(&log_root_tree->log_commit[index2], 1);
3178 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3179 wait_log_commit(log_root_tree,
3180 root_log_ctx.log_transid - 1);
3183 wait_for_writer(log_root_tree);
3186 * now that we've moved on to the tree of log tree roots,
3187 * check the full commit flag again
3189 if (btrfs_need_log_full_commit(trans)) {
3190 blk_finish_plug(&plug);
3191 btrfs_wait_tree_log_extents(log, mark);
3192 mutex_unlock(&log_root_tree->log_mutex);
3194 goto out_wake_log_root;
3197 ret = btrfs_write_marked_extents(fs_info,
3198 &log_root_tree->dirty_log_pages,
3199 EXTENT_DIRTY | EXTENT_NEW);
3200 blk_finish_plug(&plug);
3202 btrfs_set_log_full_commit(trans);
3203 btrfs_abort_transaction(trans, ret);
3204 mutex_unlock(&log_root_tree->log_mutex);
3205 goto out_wake_log_root;
3207 ret = btrfs_wait_tree_log_extents(log, mark);
3209 ret = btrfs_wait_tree_log_extents(log_root_tree,
3210 EXTENT_NEW | EXTENT_DIRTY);
3212 btrfs_set_log_full_commit(trans);
3213 mutex_unlock(&log_root_tree->log_mutex);
3214 goto out_wake_log_root;
3217 btrfs_set_super_log_root(fs_info->super_for_commit,
3218 log_root_tree->node->start);
3219 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3220 btrfs_header_level(log_root_tree->node));
3222 log_root_tree->log_transid++;
3223 mutex_unlock(&log_root_tree->log_mutex);
3226 * Nobody else is going to jump in and write the ctree
3227 * super here because the log_commit atomic below is protecting
3228 * us. We must be called with a transaction handle pinning
3229 * the running transaction open, so a full commit can't hop
3230 * in and cause problems either.
3232 ret = write_all_supers(fs_info, 1);
3234 btrfs_set_log_full_commit(trans);
3235 btrfs_abort_transaction(trans, ret);
3236 goto out_wake_log_root;
3239 mutex_lock(&root->log_mutex);
3240 if (root->last_log_commit < log_transid)
3241 root->last_log_commit = log_transid;
3242 mutex_unlock(&root->log_mutex);
3245 mutex_lock(&log_root_tree->log_mutex);
3246 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3248 log_root_tree->log_transid_committed++;
3249 atomic_set(&log_root_tree->log_commit[index2], 0);
3250 mutex_unlock(&log_root_tree->log_mutex);
3253 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3254 * all the updates above are seen by the woken threads. It might not be
3255 * necessary, but proving that seems to be hard.
3257 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3259 mutex_lock(&root->log_mutex);
3260 btrfs_remove_all_log_ctxs(root, index1, ret);
3261 root->log_transid_committed++;
3262 atomic_set(&root->log_commit[index1], 0);
3263 mutex_unlock(&root->log_mutex);
3266 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3267 * all the updates above are seen by the woken threads. It might not be
3268 * necessary, but proving that seems to be hard.
3270 cond_wake_up(&root->log_commit_wait[index1]);
3274 static void free_log_tree(struct btrfs_trans_handle *trans,
3275 struct btrfs_root *log)
3278 struct walk_control wc = {
3280 .process_func = process_one_buffer
3283 ret = walk_log_tree(trans, log, &wc);
3286 btrfs_abort_transaction(trans, ret);
3288 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3291 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3292 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3293 extent_io_tree_release(&log->log_csum_range);
3294 btrfs_put_root(log);
3298 * free all the extents used by the tree log. This should be called
3299 * at commit time of the full transaction
3301 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3303 if (root->log_root) {
3304 free_log_tree(trans, root->log_root);
3305 root->log_root = NULL;
3310 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3311 struct btrfs_fs_info *fs_info)
3313 if (fs_info->log_root_tree) {
3314 free_log_tree(trans, fs_info->log_root_tree);
3315 fs_info->log_root_tree = NULL;
3321 * Check if an inode was logged in the current transaction. We can't always rely
3322 * on an inode's logged_trans value, because it's an in-memory only field and
3323 * therefore not persisted. This means that its value is lost if the inode gets
3324 * evicted and loaded again from disk (in which case it has a value of 0, and
3325 * certainly it is smaller then any possible transaction ID), when that happens
3326 * the full_sync flag is set in the inode's runtime flags, so on that case we
3327 * assume eviction happened and ignore the logged_trans value, assuming the
3328 * worst case, that the inode was logged before in the current transaction.
3330 static bool inode_logged(struct btrfs_trans_handle *trans,
3331 struct btrfs_inode *inode)
3333 if (inode->logged_trans == trans->transid)
3336 if (inode->last_trans == trans->transid &&
3337 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3338 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3345 * If both a file and directory are logged, and unlinks or renames are
3346 * mixed in, we have a few interesting corners:
3348 * create file X in dir Y
3349 * link file X to X.link in dir Y
3351 * unlink file X but leave X.link
3354 * After a crash we would expect only X.link to exist. But file X
3355 * didn't get fsync'd again so the log has back refs for X and X.link.
3357 * We solve this by removing directory entries and inode backrefs from the
3358 * log when a file that was logged in the current transaction is
3359 * unlinked. Any later fsync will include the updated log entries, and
3360 * we'll be able to reconstruct the proper directory items from backrefs.
3362 * This optimizations allows us to avoid relogging the entire inode
3363 * or the entire directory.
3365 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3366 struct btrfs_root *root,
3367 const char *name, int name_len,
3368 struct btrfs_inode *dir, u64 index)
3370 struct btrfs_root *log;
3371 struct btrfs_dir_item *di;
3372 struct btrfs_path *path;
3376 u64 dir_ino = btrfs_ino(dir);
3378 if (!inode_logged(trans, dir))
3381 ret = join_running_log_trans(root);
3385 mutex_lock(&dir->log_mutex);
3387 log = root->log_root;
3388 path = btrfs_alloc_path();
3394 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3395 name, name_len, -1);
3401 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3402 bytes_del += name_len;
3408 btrfs_release_path(path);
3409 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3410 index, name, name_len, -1);
3416 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3417 bytes_del += name_len;
3424 /* update the directory size in the log to reflect the names
3428 struct btrfs_key key;
3430 key.objectid = dir_ino;
3432 key.type = BTRFS_INODE_ITEM_KEY;
3433 btrfs_release_path(path);
3435 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3441 struct btrfs_inode_item *item;
3444 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3445 struct btrfs_inode_item);
3446 i_size = btrfs_inode_size(path->nodes[0], item);
3447 if (i_size > bytes_del)
3448 i_size -= bytes_del;
3451 btrfs_set_inode_size(path->nodes[0], item, i_size);
3452 btrfs_mark_buffer_dirty(path->nodes[0]);
3455 btrfs_release_path(path);
3458 btrfs_free_path(path);
3460 mutex_unlock(&dir->log_mutex);
3461 if (ret == -ENOSPC) {
3462 btrfs_set_log_full_commit(trans);
3465 btrfs_abort_transaction(trans, ret);
3467 btrfs_end_log_trans(root);
3472 /* see comments for btrfs_del_dir_entries_in_log */
3473 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *root,
3475 const char *name, int name_len,
3476 struct btrfs_inode *inode, u64 dirid)
3478 struct btrfs_root *log;
3482 if (!inode_logged(trans, inode))
3485 ret = join_running_log_trans(root);
3488 log = root->log_root;
3489 mutex_lock(&inode->log_mutex);
3491 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3493 mutex_unlock(&inode->log_mutex);
3494 if (ret == -ENOSPC) {
3495 btrfs_set_log_full_commit(trans);
3497 } else if (ret < 0 && ret != -ENOENT)
3498 btrfs_abort_transaction(trans, ret);
3499 btrfs_end_log_trans(root);
3505 * creates a range item in the log for 'dirid'. first_offset and
3506 * last_offset tell us which parts of the key space the log should
3507 * be considered authoritative for.
3509 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3510 struct btrfs_root *log,
3511 struct btrfs_path *path,
3512 int key_type, u64 dirid,
3513 u64 first_offset, u64 last_offset)
3516 struct btrfs_key key;
3517 struct btrfs_dir_log_item *item;
3519 key.objectid = dirid;
3520 key.offset = first_offset;
3521 if (key_type == BTRFS_DIR_ITEM_KEY)
3522 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3524 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3525 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3529 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3530 struct btrfs_dir_log_item);
3531 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3532 btrfs_mark_buffer_dirty(path->nodes[0]);
3533 btrfs_release_path(path);
3538 * log all the items included in the current transaction for a given
3539 * directory. This also creates the range items in the log tree required
3540 * to replay anything deleted before the fsync
3542 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3543 struct btrfs_root *root, struct btrfs_inode *inode,
3544 struct btrfs_path *path,
3545 struct btrfs_path *dst_path, int key_type,
3546 struct btrfs_log_ctx *ctx,
3547 u64 min_offset, u64 *last_offset_ret)
3549 struct btrfs_key min_key;
3550 struct btrfs_root *log = root->log_root;
3551 struct extent_buffer *src;
3556 u64 first_offset = min_offset;
3557 u64 last_offset = (u64)-1;
3558 u64 ino = btrfs_ino(inode);
3560 log = root->log_root;
3562 min_key.objectid = ino;
3563 min_key.type = key_type;
3564 min_key.offset = min_offset;
3566 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3569 * we didn't find anything from this transaction, see if there
3570 * is anything at all
3572 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3573 min_key.objectid = ino;
3574 min_key.type = key_type;
3575 min_key.offset = (u64)-1;
3576 btrfs_release_path(path);
3577 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3579 btrfs_release_path(path);
3582 ret = btrfs_previous_item(root, path, ino, key_type);
3584 /* if ret == 0 there are items for this type,
3585 * create a range to tell us the last key of this type.
3586 * otherwise, there are no items in this directory after
3587 * *min_offset, and we create a range to indicate that.
3590 struct btrfs_key tmp;
3591 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3593 if (key_type == tmp.type)
3594 first_offset = max(min_offset, tmp.offset) + 1;
3599 /* go backward to find any previous key */
3600 ret = btrfs_previous_item(root, path, ino, key_type);
3602 struct btrfs_key tmp;
3603 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3604 if (key_type == tmp.type) {
3605 first_offset = tmp.offset;
3606 ret = overwrite_item(trans, log, dst_path,
3607 path->nodes[0], path->slots[0],
3615 btrfs_release_path(path);
3618 * Find the first key from this transaction again. See the note for
3619 * log_new_dir_dentries, if we're logging a directory recursively we
3620 * won't be holding its i_mutex, which means we can modify the directory
3621 * while we're logging it. If we remove an entry between our first
3622 * search and this search we'll not find the key again and can just
3625 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3630 * we have a block from this transaction, log every item in it
3631 * from our directory
3634 struct btrfs_key tmp;
3635 src = path->nodes[0];
3636 nritems = btrfs_header_nritems(src);
3637 for (i = path->slots[0]; i < nritems; i++) {
3638 struct btrfs_dir_item *di;
3640 btrfs_item_key_to_cpu(src, &min_key, i);
3642 if (min_key.objectid != ino || min_key.type != key_type)
3644 ret = overwrite_item(trans, log, dst_path, src, i,
3652 * We must make sure that when we log a directory entry,
3653 * the corresponding inode, after log replay, has a
3654 * matching link count. For example:
3660 * xfs_io -c "fsync" mydir
3662 * <mount fs and log replay>
3664 * Would result in a fsync log that when replayed, our
3665 * file inode would have a link count of 1, but we get
3666 * two directory entries pointing to the same inode.
3667 * After removing one of the names, it would not be
3668 * possible to remove the other name, which resulted
3669 * always in stale file handle errors, and would not
3670 * be possible to rmdir the parent directory, since
3671 * its i_size could never decrement to the value
3672 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3674 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3675 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3677 (btrfs_dir_transid(src, di) == trans->transid ||
3678 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3679 tmp.type != BTRFS_ROOT_ITEM_KEY)
3680 ctx->log_new_dentries = true;
3682 path->slots[0] = nritems;
3685 * look ahead to the next item and see if it is also
3686 * from this directory and from this transaction
3688 ret = btrfs_next_leaf(root, path);
3691 last_offset = (u64)-1;
3696 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3697 if (tmp.objectid != ino || tmp.type != key_type) {
3698 last_offset = (u64)-1;
3701 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3702 ret = overwrite_item(trans, log, dst_path,
3703 path->nodes[0], path->slots[0],
3708 last_offset = tmp.offset;
3713 btrfs_release_path(path);
3714 btrfs_release_path(dst_path);
3717 *last_offset_ret = last_offset;
3719 * insert the log range keys to indicate where the log
3722 ret = insert_dir_log_key(trans, log, path, key_type,
3723 ino, first_offset, last_offset);
3731 * logging directories is very similar to logging inodes, We find all the items
3732 * from the current transaction and write them to the log.
3734 * The recovery code scans the directory in the subvolume, and if it finds a
3735 * key in the range logged that is not present in the log tree, then it means
3736 * that dir entry was unlinked during the transaction.
3738 * In order for that scan to work, we must include one key smaller than
3739 * the smallest logged by this transaction and one key larger than the largest
3740 * key logged by this transaction.
3742 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3743 struct btrfs_root *root, struct btrfs_inode *inode,
3744 struct btrfs_path *path,
3745 struct btrfs_path *dst_path,
3746 struct btrfs_log_ctx *ctx)
3751 int key_type = BTRFS_DIR_ITEM_KEY;
3757 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3758 ctx, min_key, &max_key);
3761 if (max_key == (u64)-1)
3763 min_key = max_key + 1;
3766 if (key_type == BTRFS_DIR_ITEM_KEY) {
3767 key_type = BTRFS_DIR_INDEX_KEY;
3774 * a helper function to drop items from the log before we relog an
3775 * inode. max_key_type indicates the highest item type to remove.
3776 * This cannot be run for file data extents because it does not
3777 * free the extents they point to.
3779 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3780 struct btrfs_root *log,
3781 struct btrfs_path *path,
3782 u64 objectid, int max_key_type)
3785 struct btrfs_key key;
3786 struct btrfs_key found_key;
3789 key.objectid = objectid;
3790 key.type = max_key_type;
3791 key.offset = (u64)-1;
3794 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3795 BUG_ON(ret == 0); /* Logic error */
3799 if (path->slots[0] == 0)
3803 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3806 if (found_key.objectid != objectid)
3809 found_key.offset = 0;
3811 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3815 ret = btrfs_del_items(trans, log, path, start_slot,
3816 path->slots[0] - start_slot + 1);
3818 * If start slot isn't 0 then we don't need to re-search, we've
3819 * found the last guy with the objectid in this tree.
3821 if (ret || start_slot != 0)
3823 btrfs_release_path(path);
3825 btrfs_release_path(path);
3831 static void fill_inode_item(struct btrfs_trans_handle *trans,
3832 struct extent_buffer *leaf,
3833 struct btrfs_inode_item *item,
3834 struct inode *inode, int log_inode_only,
3837 struct btrfs_map_token token;
3839 btrfs_init_map_token(&token, leaf);
3841 if (log_inode_only) {
3842 /* set the generation to zero so the recover code
3843 * can tell the difference between an logging
3844 * just to say 'this inode exists' and a logging
3845 * to say 'update this inode with these values'
3847 btrfs_set_token_inode_generation(&token, item, 0);
3848 btrfs_set_token_inode_size(&token, item, logged_isize);
3850 btrfs_set_token_inode_generation(&token, item,
3851 BTRFS_I(inode)->generation);
3852 btrfs_set_token_inode_size(&token, item, inode->i_size);
3855 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3856 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3857 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3858 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3860 btrfs_set_token_timespec_sec(&token, &item->atime,
3861 inode->i_atime.tv_sec);
3862 btrfs_set_token_timespec_nsec(&token, &item->atime,
3863 inode->i_atime.tv_nsec);
3865 btrfs_set_token_timespec_sec(&token, &item->mtime,
3866 inode->i_mtime.tv_sec);
3867 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3868 inode->i_mtime.tv_nsec);
3870 btrfs_set_token_timespec_sec(&token, &item->ctime,
3871 inode->i_ctime.tv_sec);
3872 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3873 inode->i_ctime.tv_nsec);
3875 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3877 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3878 btrfs_set_token_inode_transid(&token, item, trans->transid);
3879 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3880 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3881 btrfs_set_token_inode_block_group(&token, item, 0);
3884 static int log_inode_item(struct btrfs_trans_handle *trans,
3885 struct btrfs_root *log, struct btrfs_path *path,
3886 struct btrfs_inode *inode)
3888 struct btrfs_inode_item *inode_item;
3891 ret = btrfs_insert_empty_item(trans, log, path,
3892 &inode->location, sizeof(*inode_item));
3893 if (ret && ret != -EEXIST)
3895 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3896 struct btrfs_inode_item);
3897 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3899 btrfs_release_path(path);
3903 static int log_csums(struct btrfs_trans_handle *trans,
3904 struct btrfs_root *log_root,
3905 struct btrfs_ordered_sum *sums)
3907 const u64 lock_end = sums->bytenr + sums->len - 1;
3908 struct extent_state *cached_state = NULL;
3912 * Serialize logging for checksums. This is to avoid racing with the
3913 * same checksum being logged by another task that is logging another
3914 * file which happens to refer to the same extent as well. Such races
3915 * can leave checksum items in the log with overlapping ranges.
3917 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
3918 lock_end, &cached_state);
3922 * Due to extent cloning, we might have logged a csum item that covers a
3923 * subrange of a cloned extent, and later we can end up logging a csum
3924 * item for a larger subrange of the same extent or the entire range.
3925 * This would leave csum items in the log tree that cover the same range
3926 * and break the searches for checksums in the log tree, resulting in
3927 * some checksums missing in the fs/subvolume tree. So just delete (or
3928 * trim and adjust) any existing csum items in the log for this range.
3930 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3932 ret = btrfs_csum_file_blocks(trans, log_root, sums);
3934 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
3940 static noinline int copy_items(struct btrfs_trans_handle *trans,
3941 struct btrfs_inode *inode,
3942 struct btrfs_path *dst_path,
3943 struct btrfs_path *src_path,
3944 int start_slot, int nr, int inode_only,
3947 struct btrfs_fs_info *fs_info = trans->fs_info;
3948 unsigned long src_offset;
3949 unsigned long dst_offset;
3950 struct btrfs_root *log = inode->root->log_root;
3951 struct btrfs_file_extent_item *extent;
3952 struct btrfs_inode_item *inode_item;
3953 struct extent_buffer *src = src_path->nodes[0];
3955 struct btrfs_key *ins_keys;
3959 struct list_head ordered_sums;
3960 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3962 INIT_LIST_HEAD(&ordered_sums);
3964 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3965 nr * sizeof(u32), GFP_NOFS);
3969 ins_sizes = (u32 *)ins_data;
3970 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3972 for (i = 0; i < nr; i++) {
3973 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3974 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3976 ret = btrfs_insert_empty_items(trans, log, dst_path,
3977 ins_keys, ins_sizes, nr);
3983 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3984 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3985 dst_path->slots[0]);
3987 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3989 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3990 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3992 struct btrfs_inode_item);
3993 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3995 inode_only == LOG_INODE_EXISTS,
3998 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3999 src_offset, ins_sizes[i]);
4002 /* take a reference on file data extents so that truncates
4003 * or deletes of this inode don't have to relog the inode
4006 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4009 extent = btrfs_item_ptr(src, start_slot + i,
4010 struct btrfs_file_extent_item);
4012 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4015 found_type = btrfs_file_extent_type(src, extent);
4016 if (found_type == BTRFS_FILE_EXTENT_REG) {
4018 ds = btrfs_file_extent_disk_bytenr(src,
4020 /* ds == 0 is a hole */
4024 dl = btrfs_file_extent_disk_num_bytes(src,
4026 cs = btrfs_file_extent_offset(src, extent);
4027 cl = btrfs_file_extent_num_bytes(src,
4029 if (btrfs_file_extent_compression(src,
4035 ret = btrfs_lookup_csums_range(
4037 ds + cs, ds + cs + cl - 1,
4040 btrfs_release_path(dst_path);
4048 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4049 btrfs_release_path(dst_path);
4053 * we have to do this after the loop above to avoid changing the
4054 * log tree while trying to change the log tree.
4057 while (!list_empty(&ordered_sums)) {
4058 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4059 struct btrfs_ordered_sum,
4062 ret = log_csums(trans, log, sums);
4063 list_del(&sums->list);
4070 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4072 struct extent_map *em1, *em2;
4074 em1 = list_entry(a, struct extent_map, list);
4075 em2 = list_entry(b, struct extent_map, list);
4077 if (em1->start < em2->start)
4079 else if (em1->start > em2->start)
4084 static int log_extent_csums(struct btrfs_trans_handle *trans,
4085 struct btrfs_inode *inode,
4086 struct btrfs_root *log_root,
4087 const struct extent_map *em)
4091 LIST_HEAD(ordered_sums);
4094 if (inode->flags & BTRFS_INODE_NODATASUM ||
4095 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4096 em->block_start == EXTENT_MAP_HOLE)
4099 /* If we're compressed we have to save the entire range of csums. */
4100 if (em->compress_type) {
4102 csum_len = max(em->block_len, em->orig_block_len);
4104 csum_offset = em->mod_start - em->start;
4105 csum_len = em->mod_len;
4108 /* block start is already adjusted for the file extent offset. */
4109 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4110 em->block_start + csum_offset,
4111 em->block_start + csum_offset +
4112 csum_len - 1, &ordered_sums, 0);
4116 while (!list_empty(&ordered_sums)) {
4117 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4118 struct btrfs_ordered_sum,
4121 ret = log_csums(trans, log_root, sums);
4122 list_del(&sums->list);
4129 static int log_one_extent(struct btrfs_trans_handle *trans,
4130 struct btrfs_inode *inode, struct btrfs_root *root,
4131 const struct extent_map *em,
4132 struct btrfs_path *path,
4133 struct btrfs_log_ctx *ctx)
4135 struct btrfs_root *log = root->log_root;
4136 struct btrfs_file_extent_item *fi;
4137 struct extent_buffer *leaf;
4138 struct btrfs_map_token token;
4139 struct btrfs_key key;
4140 u64 extent_offset = em->start - em->orig_start;
4143 int extent_inserted = 0;
4145 ret = log_extent_csums(trans, inode, log, em);
4149 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4150 em->start + em->len, NULL, 0, 1,
4151 sizeof(*fi), &extent_inserted);
4155 if (!extent_inserted) {
4156 key.objectid = btrfs_ino(inode);
4157 key.type = BTRFS_EXTENT_DATA_KEY;
4158 key.offset = em->start;
4160 ret = btrfs_insert_empty_item(trans, log, path, &key,
4165 leaf = path->nodes[0];
4166 btrfs_init_map_token(&token, leaf);
4167 fi = btrfs_item_ptr(leaf, path->slots[0],
4168 struct btrfs_file_extent_item);
4170 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4171 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4172 btrfs_set_token_file_extent_type(&token, fi,
4173 BTRFS_FILE_EXTENT_PREALLOC);
4175 btrfs_set_token_file_extent_type(&token, fi,
4176 BTRFS_FILE_EXTENT_REG);
4178 block_len = max(em->block_len, em->orig_block_len);
4179 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4180 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4182 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4183 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4184 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4187 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4189 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4190 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4193 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4194 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4195 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4196 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4197 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4198 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4199 btrfs_mark_buffer_dirty(leaf);
4201 btrfs_release_path(path);
4207 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4208 * lose them after doing a fast fsync and replaying the log. We scan the
4209 * subvolume's root instead of iterating the inode's extent map tree because
4210 * otherwise we can log incorrect extent items based on extent map conversion.
4211 * That can happen due to the fact that extent maps are merged when they
4212 * are not in the extent map tree's list of modified extents.
4214 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4215 struct btrfs_inode *inode,
4216 struct btrfs_path *path)
4218 struct btrfs_root *root = inode->root;
4219 struct btrfs_key key;
4220 const u64 i_size = i_size_read(&inode->vfs_inode);
4221 const u64 ino = btrfs_ino(inode);
4222 struct btrfs_path *dst_path = NULL;
4223 bool dropped_extents = false;
4224 u64 truncate_offset = i_size;
4225 struct extent_buffer *leaf;
4231 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4235 key.type = BTRFS_EXTENT_DATA_KEY;
4236 key.offset = i_size;
4237 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4242 * We must check if there is a prealloc extent that starts before the
4243 * i_size and crosses the i_size boundary. This is to ensure later we
4244 * truncate down to the end of that extent and not to the i_size, as
4245 * otherwise we end up losing part of the prealloc extent after a log
4246 * replay and with an implicit hole if there is another prealloc extent
4247 * that starts at an offset beyond i_size.
4249 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4254 struct btrfs_file_extent_item *ei;
4256 leaf = path->nodes[0];
4257 slot = path->slots[0];
4258 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4260 if (btrfs_file_extent_type(leaf, ei) ==
4261 BTRFS_FILE_EXTENT_PREALLOC) {
4264 btrfs_item_key_to_cpu(leaf, &key, slot);
4265 extent_end = key.offset +
4266 btrfs_file_extent_num_bytes(leaf, ei);
4268 if (extent_end > i_size)
4269 truncate_offset = extent_end;
4276 leaf = path->nodes[0];
4277 slot = path->slots[0];
4279 if (slot >= btrfs_header_nritems(leaf)) {
4281 ret = copy_items(trans, inode, dst_path, path,
4282 start_slot, ins_nr, 1, 0);
4287 ret = btrfs_next_leaf(root, path);
4297 btrfs_item_key_to_cpu(leaf, &key, slot);
4298 if (key.objectid > ino)
4300 if (WARN_ON_ONCE(key.objectid < ino) ||
4301 key.type < BTRFS_EXTENT_DATA_KEY ||
4302 key.offset < i_size) {
4306 if (!dropped_extents) {
4308 * Avoid logging extent items logged in past fsync calls
4309 * and leading to duplicate keys in the log tree.
4312 ret = btrfs_truncate_inode_items(trans,
4316 BTRFS_EXTENT_DATA_KEY);
4317 } while (ret == -EAGAIN);
4320 dropped_extents = true;
4327 dst_path = btrfs_alloc_path();
4335 ret = copy_items(trans, inode, dst_path, path,
4336 start_slot, ins_nr, 1, 0);
4338 btrfs_release_path(path);
4339 btrfs_free_path(dst_path);
4343 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4344 struct btrfs_root *root,
4345 struct btrfs_inode *inode,
4346 struct btrfs_path *path,
4347 struct btrfs_log_ctx *ctx,
4351 struct extent_map *em, *n;
4352 struct list_head extents;
4353 struct extent_map_tree *tree = &inode->extent_tree;
4358 INIT_LIST_HEAD(&extents);
4360 write_lock(&tree->lock);
4361 test_gen = root->fs_info->last_trans_committed;
4363 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4365 * Skip extents outside our logging range. It's important to do
4366 * it for correctness because if we don't ignore them, we may
4367 * log them before their ordered extent completes, and therefore
4368 * we could log them without logging their respective checksums
4369 * (the checksum items are added to the csum tree at the very
4370 * end of btrfs_finish_ordered_io()). Also leave such extents
4371 * outside of our range in the list, since we may have another
4372 * ranged fsync in the near future that needs them. If an extent
4373 * outside our range corresponds to a hole, log it to avoid
4374 * leaving gaps between extents (fsck will complain when we are
4375 * not using the NO_HOLES feature).
4377 if ((em->start > end || em->start + em->len <= start) &&
4378 em->block_start != EXTENT_MAP_HOLE)
4381 list_del_init(&em->list);
4383 * Just an arbitrary number, this can be really CPU intensive
4384 * once we start getting a lot of extents, and really once we
4385 * have a bunch of extents we just want to commit since it will
4388 if (++num > 32768) {
4389 list_del_init(&tree->modified_extents);
4394 if (em->generation <= test_gen)
4397 /* We log prealloc extents beyond eof later. */
4398 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4399 em->start >= i_size_read(&inode->vfs_inode))
4402 /* Need a ref to keep it from getting evicted from cache */
4403 refcount_inc(&em->refs);
4404 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4405 list_add_tail(&em->list, &extents);
4409 list_sort(NULL, &extents, extent_cmp);
4411 while (!list_empty(&extents)) {
4412 em = list_entry(extents.next, struct extent_map, list);
4414 list_del_init(&em->list);
4417 * If we had an error we just need to delete everybody from our
4421 clear_em_logging(tree, em);
4422 free_extent_map(em);
4426 write_unlock(&tree->lock);
4428 ret = log_one_extent(trans, inode, root, em, path, ctx);
4429 write_lock(&tree->lock);
4430 clear_em_logging(tree, em);
4431 free_extent_map(em);
4433 WARN_ON(!list_empty(&extents));
4434 write_unlock(&tree->lock);
4436 btrfs_release_path(path);
4438 ret = btrfs_log_prealloc_extents(trans, inode, path);
4443 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4444 struct btrfs_path *path, u64 *size_ret)
4446 struct btrfs_key key;
4449 key.objectid = btrfs_ino(inode);
4450 key.type = BTRFS_INODE_ITEM_KEY;
4453 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4456 } else if (ret > 0) {
4459 struct btrfs_inode_item *item;
4461 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4462 struct btrfs_inode_item);
4463 *size_ret = btrfs_inode_size(path->nodes[0], item);
4465 * If the in-memory inode's i_size is smaller then the inode
4466 * size stored in the btree, return the inode's i_size, so
4467 * that we get a correct inode size after replaying the log
4468 * when before a power failure we had a shrinking truncate
4469 * followed by addition of a new name (rename / new hard link).
4470 * Otherwise return the inode size from the btree, to avoid
4471 * data loss when replaying a log due to previously doing a
4472 * write that expands the inode's size and logging a new name
4473 * immediately after.
4475 if (*size_ret > inode->vfs_inode.i_size)
4476 *size_ret = inode->vfs_inode.i_size;
4479 btrfs_release_path(path);
4484 * At the moment we always log all xattrs. This is to figure out at log replay
4485 * time which xattrs must have their deletion replayed. If a xattr is missing
4486 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4487 * because if a xattr is deleted, the inode is fsynced and a power failure
4488 * happens, causing the log to be replayed the next time the fs is mounted,
4489 * we want the xattr to not exist anymore (same behaviour as other filesystems
4490 * with a journal, ext3/4, xfs, f2fs, etc).
4492 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4493 struct btrfs_root *root,
4494 struct btrfs_inode *inode,
4495 struct btrfs_path *path,
4496 struct btrfs_path *dst_path)
4499 struct btrfs_key key;
4500 const u64 ino = btrfs_ino(inode);
4505 key.type = BTRFS_XATTR_ITEM_KEY;
4508 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4513 int slot = path->slots[0];
4514 struct extent_buffer *leaf = path->nodes[0];
4515 int nritems = btrfs_header_nritems(leaf);
4517 if (slot >= nritems) {
4519 ret = copy_items(trans, inode, dst_path, path,
4520 start_slot, ins_nr, 1, 0);
4525 ret = btrfs_next_leaf(root, path);
4533 btrfs_item_key_to_cpu(leaf, &key, slot);
4534 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4544 ret = copy_items(trans, inode, dst_path, path,
4545 start_slot, ins_nr, 1, 0);
4554 * When using the NO_HOLES feature if we punched a hole that causes the
4555 * deletion of entire leafs or all the extent items of the first leaf (the one
4556 * that contains the inode item and references) we may end up not processing
4557 * any extents, because there are no leafs with a generation matching the
4558 * current transaction that have extent items for our inode. So we need to find
4559 * if any holes exist and then log them. We also need to log holes after any
4560 * truncate operation that changes the inode's size.
4562 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4563 struct btrfs_root *root,
4564 struct btrfs_inode *inode,
4565 struct btrfs_path *path)
4567 struct btrfs_fs_info *fs_info = root->fs_info;
4568 struct btrfs_key key;
4569 const u64 ino = btrfs_ino(inode);
4570 const u64 i_size = i_size_read(&inode->vfs_inode);
4571 u64 prev_extent_end = 0;
4574 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4578 key.type = BTRFS_EXTENT_DATA_KEY;
4581 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4586 struct extent_buffer *leaf = path->nodes[0];
4588 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4589 ret = btrfs_next_leaf(root, path);
4596 leaf = path->nodes[0];
4599 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4600 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4603 /* We have a hole, log it. */
4604 if (prev_extent_end < key.offset) {
4605 const u64 hole_len = key.offset - prev_extent_end;
4608 * Release the path to avoid deadlocks with other code
4609 * paths that search the root while holding locks on
4610 * leafs from the log root.
4612 btrfs_release_path(path);
4613 ret = btrfs_insert_file_extent(trans, root->log_root,
4614 ino, prev_extent_end, 0,
4615 0, hole_len, 0, hole_len,
4621 * Search for the same key again in the root. Since it's
4622 * an extent item and we are holding the inode lock, the
4623 * key must still exist. If it doesn't just emit warning
4624 * and return an error to fall back to a transaction
4627 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4630 if (WARN_ON(ret > 0))
4632 leaf = path->nodes[0];
4635 prev_extent_end = btrfs_file_extent_end(path);
4640 if (prev_extent_end < i_size) {
4643 btrfs_release_path(path);
4644 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4645 ret = btrfs_insert_file_extent(trans, root->log_root,
4646 ino, prev_extent_end, 0, 0,
4647 hole_len, 0, hole_len,
4657 * When we are logging a new inode X, check if it doesn't have a reference that
4658 * matches the reference from some other inode Y created in a past transaction
4659 * and that was renamed in the current transaction. If we don't do this, then at
4660 * log replay time we can lose inode Y (and all its files if it's a directory):
4663 * echo "hello world" > /mnt/x/foobar
4666 * mkdir /mnt/x # or touch /mnt/x
4667 * xfs_io -c fsync /mnt/x
4669 * mount fs, trigger log replay
4671 * After the log replay procedure, we would lose the first directory and all its
4672 * files (file foobar).
4673 * For the case where inode Y is not a directory we simply end up losing it:
4675 * echo "123" > /mnt/foo
4677 * mv /mnt/foo /mnt/bar
4678 * echo "abc" > /mnt/foo
4679 * xfs_io -c fsync /mnt/foo
4682 * We also need this for cases where a snapshot entry is replaced by some other
4683 * entry (file or directory) otherwise we end up with an unreplayable log due to
4684 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4685 * if it were a regular entry:
4688 * btrfs subvolume snapshot /mnt /mnt/x/snap
4689 * btrfs subvolume delete /mnt/x/snap
4692 * fsync /mnt/x or fsync some new file inside it
4695 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4696 * the same transaction.
4698 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4700 const struct btrfs_key *key,
4701 struct btrfs_inode *inode,
4702 u64 *other_ino, u64 *other_parent)
4705 struct btrfs_path *search_path;
4708 u32 item_size = btrfs_item_size_nr(eb, slot);
4710 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4712 search_path = btrfs_alloc_path();
4715 search_path->search_commit_root = 1;
4716 search_path->skip_locking = 1;
4718 while (cur_offset < item_size) {
4722 unsigned long name_ptr;
4723 struct btrfs_dir_item *di;
4725 if (key->type == BTRFS_INODE_REF_KEY) {
4726 struct btrfs_inode_ref *iref;
4728 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4729 parent = key->offset;
4730 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4731 name_ptr = (unsigned long)(iref + 1);
4732 this_len = sizeof(*iref) + this_name_len;
4734 struct btrfs_inode_extref *extref;
4736 extref = (struct btrfs_inode_extref *)(ptr +
4738 parent = btrfs_inode_extref_parent(eb, extref);
4739 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4740 name_ptr = (unsigned long)&extref->name;
4741 this_len = sizeof(*extref) + this_name_len;
4744 if (this_name_len > name_len) {
4747 new_name = krealloc(name, this_name_len, GFP_NOFS);
4752 name_len = this_name_len;
4756 read_extent_buffer(eb, name, name_ptr, this_name_len);
4757 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4758 parent, name, this_name_len, 0);
4759 if (di && !IS_ERR(di)) {
4760 struct btrfs_key di_key;
4762 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4764 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4765 if (di_key.objectid != key->objectid) {
4767 *other_ino = di_key.objectid;
4768 *other_parent = parent;
4776 } else if (IS_ERR(di)) {
4780 btrfs_release_path(search_path);
4782 cur_offset += this_len;
4786 btrfs_free_path(search_path);
4791 struct btrfs_ino_list {
4794 struct list_head list;
4797 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4798 struct btrfs_root *root,
4799 struct btrfs_path *path,
4800 struct btrfs_log_ctx *ctx,
4801 u64 ino, u64 parent)
4803 struct btrfs_ino_list *ino_elem;
4804 LIST_HEAD(inode_list);
4807 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4810 ino_elem->ino = ino;
4811 ino_elem->parent = parent;
4812 list_add_tail(&ino_elem->list, &inode_list);
4814 while (!list_empty(&inode_list)) {
4815 struct btrfs_fs_info *fs_info = root->fs_info;
4816 struct btrfs_key key;
4817 struct inode *inode;
4819 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4821 ino = ino_elem->ino;
4822 parent = ino_elem->parent;
4823 list_del(&ino_elem->list);
4828 btrfs_release_path(path);
4830 inode = btrfs_iget(fs_info->sb, ino, root);
4832 * If the other inode that had a conflicting dir entry was
4833 * deleted in the current transaction, we need to log its parent
4836 if (IS_ERR(inode)) {
4837 ret = PTR_ERR(inode);
4838 if (ret == -ENOENT) {
4839 inode = btrfs_iget(fs_info->sb, parent, root);
4840 if (IS_ERR(inode)) {
4841 ret = PTR_ERR(inode);
4843 ret = btrfs_log_inode(trans, root,
4845 LOG_OTHER_INODE_ALL,
4847 btrfs_add_delayed_iput(inode);
4853 * If the inode was already logged skip it - otherwise we can
4854 * hit an infinite loop. Example:
4856 * From the commit root (previous transaction) we have the
4859 * inode 257 a directory
4860 * inode 258 with references "zz" and "zz_link" on inode 257
4861 * inode 259 with reference "a" on inode 257
4863 * And in the current (uncommitted) transaction we have:
4865 * inode 257 a directory, unchanged
4866 * inode 258 with references "a" and "a2" on inode 257
4867 * inode 259 with reference "zz_link" on inode 257
4868 * inode 261 with reference "zz" on inode 257
4870 * When logging inode 261 the following infinite loop could
4871 * happen if we don't skip already logged inodes:
4873 * - we detect inode 258 as a conflicting inode, with inode 261
4874 * on reference "zz", and log it;
4876 * - we detect inode 259 as a conflicting inode, with inode 258
4877 * on reference "a", and log it;
4879 * - we detect inode 258 as a conflicting inode, with inode 259
4880 * on reference "zz_link", and log it - again! After this we
4881 * repeat the above steps forever.
4883 spin_lock(&BTRFS_I(inode)->lock);
4885 * Check the inode's logged_trans only instead of
4886 * btrfs_inode_in_log(). This is because the last_log_commit of
4887 * the inode is not updated when we only log that it exists and
4888 * and it has the full sync bit set (see btrfs_log_inode()).
4890 if (BTRFS_I(inode)->logged_trans == trans->transid) {
4891 spin_unlock(&BTRFS_I(inode)->lock);
4892 btrfs_add_delayed_iput(inode);
4895 spin_unlock(&BTRFS_I(inode)->lock);
4897 * We are safe logging the other inode without acquiring its
4898 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4899 * are safe against concurrent renames of the other inode as
4900 * well because during a rename we pin the log and update the
4901 * log with the new name before we unpin it.
4903 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4904 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4906 btrfs_add_delayed_iput(inode);
4911 key.type = BTRFS_INODE_REF_KEY;
4913 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4915 btrfs_add_delayed_iput(inode);
4920 struct extent_buffer *leaf = path->nodes[0];
4921 int slot = path->slots[0];
4923 u64 other_parent = 0;
4925 if (slot >= btrfs_header_nritems(leaf)) {
4926 ret = btrfs_next_leaf(root, path);
4929 } else if (ret > 0) {
4936 btrfs_item_key_to_cpu(leaf, &key, slot);
4937 if (key.objectid != ino ||
4938 (key.type != BTRFS_INODE_REF_KEY &&
4939 key.type != BTRFS_INODE_EXTREF_KEY)) {
4944 ret = btrfs_check_ref_name_override(leaf, slot, &key,
4945 BTRFS_I(inode), &other_ino,
4950 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4955 ino_elem->ino = other_ino;
4956 ino_elem->parent = other_parent;
4957 list_add_tail(&ino_elem->list, &inode_list);
4962 btrfs_add_delayed_iput(inode);
4968 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
4969 struct btrfs_inode *inode,
4970 struct btrfs_key *min_key,
4971 const struct btrfs_key *max_key,
4972 struct btrfs_path *path,
4973 struct btrfs_path *dst_path,
4974 const u64 logged_isize,
4975 const bool recursive_logging,
4976 const int inode_only,
4977 struct btrfs_log_ctx *ctx,
4978 bool *need_log_inode_item)
4980 struct btrfs_root *root = inode->root;
4981 int ins_start_slot = 0;
4986 ret = btrfs_search_forward(root, min_key, path, trans->transid);
4994 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
4995 if (min_key->objectid != max_key->objectid)
4997 if (min_key->type > max_key->type)
5000 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5001 *need_log_inode_item = false;
5003 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5004 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5005 inode->generation == trans->transid &&
5006 !recursive_logging) {
5008 u64 other_parent = 0;
5010 ret = btrfs_check_ref_name_override(path->nodes[0],
5011 path->slots[0], min_key, inode,
5012 &other_ino, &other_parent);
5015 } else if (ret > 0 && ctx &&
5016 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5021 ins_start_slot = path->slots[0];
5023 ret = copy_items(trans, inode, dst_path, path,
5024 ins_start_slot, ins_nr,
5025 inode_only, logged_isize);
5030 ret = log_conflicting_inodes(trans, root, path,
5031 ctx, other_ino, other_parent);
5034 btrfs_release_path(path);
5039 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5040 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5043 ret = copy_items(trans, inode, dst_path, path,
5045 ins_nr, inode_only, logged_isize);
5052 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5055 } else if (!ins_nr) {
5056 ins_start_slot = path->slots[0];
5061 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5062 ins_nr, inode_only, logged_isize);
5066 ins_start_slot = path->slots[0];
5069 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5070 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5075 ret = copy_items(trans, inode, dst_path, path,
5076 ins_start_slot, ins_nr, inode_only,
5082 btrfs_release_path(path);
5084 if (min_key->offset < (u64)-1) {
5086 } else if (min_key->type < max_key->type) {
5088 min_key->offset = 0;
5094 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5095 ins_nr, inode_only, logged_isize);
5100 /* log a single inode in the tree log.
5101 * At least one parent directory for this inode must exist in the tree
5102 * or be logged already.
5104 * Any items from this inode changed by the current transaction are copied
5105 * to the log tree. An extra reference is taken on any extents in this
5106 * file, allowing us to avoid a whole pile of corner cases around logging
5107 * blocks that have been removed from the tree.
5109 * See LOG_INODE_ALL and related defines for a description of what inode_only
5112 * This handles both files and directories.
5114 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5115 struct btrfs_root *root, struct btrfs_inode *inode,
5119 struct btrfs_log_ctx *ctx)
5121 struct btrfs_fs_info *fs_info = root->fs_info;
5122 struct btrfs_path *path;
5123 struct btrfs_path *dst_path;
5124 struct btrfs_key min_key;
5125 struct btrfs_key max_key;
5126 struct btrfs_root *log = root->log_root;
5129 bool fast_search = false;
5130 u64 ino = btrfs_ino(inode);
5131 struct extent_map_tree *em_tree = &inode->extent_tree;
5132 u64 logged_isize = 0;
5133 bool need_log_inode_item = true;
5134 bool xattrs_logged = false;
5135 bool recursive_logging = false;
5137 path = btrfs_alloc_path();
5140 dst_path = btrfs_alloc_path();
5142 btrfs_free_path(path);
5146 min_key.objectid = ino;
5147 min_key.type = BTRFS_INODE_ITEM_KEY;
5150 max_key.objectid = ino;
5153 /* today the code can only do partial logging of directories */
5154 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5155 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5156 &inode->runtime_flags) &&
5157 inode_only >= LOG_INODE_EXISTS))
5158 max_key.type = BTRFS_XATTR_ITEM_KEY;
5160 max_key.type = (u8)-1;
5161 max_key.offset = (u64)-1;
5164 * Only run delayed items if we are a dir or a new file.
5165 * Otherwise commit the delayed inode only, which is needed in
5166 * order for the log replay code to mark inodes for link count
5167 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5169 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5170 inode->generation > fs_info->last_trans_committed)
5171 ret = btrfs_commit_inode_delayed_items(trans, inode);
5173 ret = btrfs_commit_inode_delayed_inode(inode);
5176 btrfs_free_path(path);
5177 btrfs_free_path(dst_path);
5181 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5182 recursive_logging = true;
5183 if (inode_only == LOG_OTHER_INODE)
5184 inode_only = LOG_INODE_EXISTS;
5186 inode_only = LOG_INODE_ALL;
5187 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5189 mutex_lock(&inode->log_mutex);
5193 * a brute force approach to making sure we get the most uptodate
5194 * copies of everything.
5196 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5197 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5199 if (inode_only == LOG_INODE_EXISTS)
5200 max_key_type = BTRFS_XATTR_ITEM_KEY;
5201 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5203 if (inode_only == LOG_INODE_EXISTS) {
5205 * Make sure the new inode item we write to the log has
5206 * the same isize as the current one (if it exists).
5207 * This is necessary to prevent data loss after log
5208 * replay, and also to prevent doing a wrong expanding
5209 * truncate - for e.g. create file, write 4K into offset
5210 * 0, fsync, write 4K into offset 4096, add hard link,
5211 * fsync some other file (to sync log), power fail - if
5212 * we use the inode's current i_size, after log replay
5213 * we get a 8Kb file, with the last 4Kb extent as a hole
5214 * (zeroes), as if an expanding truncate happened,
5215 * instead of getting a file of 4Kb only.
5217 err = logged_inode_size(log, inode, path, &logged_isize);
5221 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5222 &inode->runtime_flags)) {
5223 if (inode_only == LOG_INODE_EXISTS) {
5224 max_key.type = BTRFS_XATTR_ITEM_KEY;
5225 ret = drop_objectid_items(trans, log, path, ino,
5228 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5229 &inode->runtime_flags);
5230 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5231 &inode->runtime_flags);
5233 ret = btrfs_truncate_inode_items(trans,
5234 log, &inode->vfs_inode, 0, 0);
5239 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5240 &inode->runtime_flags) ||
5241 inode_only == LOG_INODE_EXISTS) {
5242 if (inode_only == LOG_INODE_ALL)
5244 max_key.type = BTRFS_XATTR_ITEM_KEY;
5245 ret = drop_objectid_items(trans, log, path, ino,
5248 if (inode_only == LOG_INODE_ALL)
5259 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5260 path, dst_path, logged_isize,
5261 recursive_logging, inode_only, ctx,
5262 &need_log_inode_item);
5266 btrfs_release_path(path);
5267 btrfs_release_path(dst_path);
5268 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5271 xattrs_logged = true;
5272 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5273 btrfs_release_path(path);
5274 btrfs_release_path(dst_path);
5275 err = btrfs_log_holes(trans, root, inode, path);
5280 btrfs_release_path(path);
5281 btrfs_release_path(dst_path);
5282 if (need_log_inode_item) {
5283 err = log_inode_item(trans, log, dst_path, inode);
5284 if (!err && !xattrs_logged) {
5285 err = btrfs_log_all_xattrs(trans, root, inode, path,
5287 btrfs_release_path(path);
5293 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5299 } else if (inode_only == LOG_INODE_ALL) {
5300 struct extent_map *em, *n;
5302 write_lock(&em_tree->lock);
5304 * We can't just remove every em if we're called for a ranged
5305 * fsync - that is, one that doesn't cover the whole possible
5306 * file range (0 to LLONG_MAX). This is because we can have
5307 * em's that fall outside the range we're logging and therefore
5308 * their ordered operations haven't completed yet
5309 * (btrfs_finish_ordered_io() not invoked yet). This means we
5310 * didn't get their respective file extent item in the fs/subvol
5311 * tree yet, and need to let the next fast fsync (one which
5312 * consults the list of modified extent maps) find the em so
5313 * that it logs a matching file extent item and waits for the
5314 * respective ordered operation to complete (if it's still
5317 * Removing every em outside the range we're logging would make
5318 * the next fast fsync not log their matching file extent items,
5319 * therefore making us lose data after a log replay.
5321 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5323 const u64 mod_end = em->mod_start + em->mod_len - 1;
5325 if (em->mod_start >= start && mod_end <= end)
5326 list_del_init(&em->list);
5328 write_unlock(&em_tree->lock);
5331 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5332 ret = log_directory_changes(trans, root, inode, path, dst_path,
5341 * Don't update last_log_commit if we logged that an inode exists after
5342 * it was loaded to memory (full_sync bit set).
5343 * This is to prevent data loss when we do a write to the inode, then
5344 * the inode gets evicted after all delalloc was flushed, then we log
5345 * it exists (due to a rename for example) and then fsync it. This last
5346 * fsync would do nothing (not logging the extents previously written).
5348 spin_lock(&inode->lock);
5349 inode->logged_trans = trans->transid;
5350 if (inode_only != LOG_INODE_EXISTS ||
5351 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5352 inode->last_log_commit = inode->last_sub_trans;
5353 spin_unlock(&inode->lock);
5355 mutex_unlock(&inode->log_mutex);
5357 btrfs_free_path(path);
5358 btrfs_free_path(dst_path);
5363 * Check if we must fallback to a transaction commit when logging an inode.
5364 * This must be called after logging the inode and is used only in the context
5365 * when fsyncing an inode requires the need to log some other inode - in which
5366 * case we can't lock the i_mutex of each other inode we need to log as that
5367 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5368 * log inodes up or down in the hierarchy) or rename operations for example. So
5369 * we take the log_mutex of the inode after we have logged it and then check for
5370 * its last_unlink_trans value - this is safe because any task setting
5371 * last_unlink_trans must take the log_mutex and it must do this before it does
5372 * the actual unlink operation, so if we do this check before a concurrent task
5373 * sets last_unlink_trans it means we've logged a consistent version/state of
5374 * all the inode items, otherwise we are not sure and must do a transaction
5375 * commit (the concurrent task might have only updated last_unlink_trans before
5376 * we logged the inode or it might have also done the unlink).
5378 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5379 struct btrfs_inode *inode)
5381 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5384 mutex_lock(&inode->log_mutex);
5385 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5387 * Make sure any commits to the log are forced to be full
5390 btrfs_set_log_full_commit(trans);
5393 mutex_unlock(&inode->log_mutex);
5399 * follow the dentry parent pointers up the chain and see if any
5400 * of the directories in it require a full commit before they can
5401 * be logged. Returns zero if nothing special needs to be done or 1 if
5402 * a full commit is required.
5404 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5405 struct btrfs_inode *inode,
5406 struct dentry *parent,
5407 struct super_block *sb,
5411 struct dentry *old_parent = NULL;
5414 * for regular files, if its inode is already on disk, we don't
5415 * have to worry about the parents at all. This is because
5416 * we can use the last_unlink_trans field to record renames
5417 * and other fun in this file.
5419 if (S_ISREG(inode->vfs_inode.i_mode) &&
5420 inode->generation <= last_committed &&
5421 inode->last_unlink_trans <= last_committed)
5424 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5425 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5427 inode = BTRFS_I(d_inode(parent));
5431 if (btrfs_must_commit_transaction(trans, inode)) {
5436 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5439 if (IS_ROOT(parent)) {
5440 inode = BTRFS_I(d_inode(parent));
5441 if (btrfs_must_commit_transaction(trans, inode))
5446 parent = dget_parent(parent);
5448 old_parent = parent;
5449 inode = BTRFS_I(d_inode(parent));
5457 struct btrfs_dir_list {
5459 struct list_head list;
5463 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5464 * details about the why it is needed.
5465 * This is a recursive operation - if an existing dentry corresponds to a
5466 * directory, that directory's new entries are logged too (same behaviour as
5467 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5468 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5469 * complains about the following circular lock dependency / possible deadlock:
5473 * lock(&type->i_mutex_dir_key#3/2);
5474 * lock(sb_internal#2);
5475 * lock(&type->i_mutex_dir_key#3/2);
5476 * lock(&sb->s_type->i_mutex_key#14);
5478 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5479 * sb_start_intwrite() in btrfs_start_transaction().
5480 * Not locking i_mutex of the inodes is still safe because:
5482 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5483 * that while logging the inode new references (names) are added or removed
5484 * from the inode, leaving the logged inode item with a link count that does
5485 * not match the number of logged inode reference items. This is fine because
5486 * at log replay time we compute the real number of links and correct the
5487 * link count in the inode item (see replay_one_buffer() and
5488 * link_to_fixup_dir());
5490 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5491 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5492 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5493 * has a size that doesn't match the sum of the lengths of all the logged
5494 * names. This does not result in a problem because if a dir_item key is
5495 * logged but its matching dir_index key is not logged, at log replay time we
5496 * don't use it to replay the respective name (see replay_one_name()). On the
5497 * other hand if only the dir_index key ends up being logged, the respective
5498 * name is added to the fs/subvol tree with both the dir_item and dir_index
5499 * keys created (see replay_one_name()).
5500 * The directory's inode item with a wrong i_size is not a problem as well,
5501 * since we don't use it at log replay time to set the i_size in the inode
5502 * item of the fs/subvol tree (see overwrite_item()).
5504 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5505 struct btrfs_root *root,
5506 struct btrfs_inode *start_inode,
5507 struct btrfs_log_ctx *ctx)
5509 struct btrfs_fs_info *fs_info = root->fs_info;
5510 struct btrfs_root *log = root->log_root;
5511 struct btrfs_path *path;
5512 LIST_HEAD(dir_list);
5513 struct btrfs_dir_list *dir_elem;
5516 path = btrfs_alloc_path();
5520 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5522 btrfs_free_path(path);
5525 dir_elem->ino = btrfs_ino(start_inode);
5526 list_add_tail(&dir_elem->list, &dir_list);
5528 while (!list_empty(&dir_list)) {
5529 struct extent_buffer *leaf;
5530 struct btrfs_key min_key;
5534 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5537 goto next_dir_inode;
5539 min_key.objectid = dir_elem->ino;
5540 min_key.type = BTRFS_DIR_ITEM_KEY;
5543 btrfs_release_path(path);
5544 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5546 goto next_dir_inode;
5547 } else if (ret > 0) {
5549 goto next_dir_inode;
5553 leaf = path->nodes[0];
5554 nritems = btrfs_header_nritems(leaf);
5555 for (i = path->slots[0]; i < nritems; i++) {
5556 struct btrfs_dir_item *di;
5557 struct btrfs_key di_key;
5558 struct inode *di_inode;
5559 struct btrfs_dir_list *new_dir_elem;
5560 int log_mode = LOG_INODE_EXISTS;
5563 btrfs_item_key_to_cpu(leaf, &min_key, i);
5564 if (min_key.objectid != dir_elem->ino ||
5565 min_key.type != BTRFS_DIR_ITEM_KEY)
5566 goto next_dir_inode;
5568 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5569 type = btrfs_dir_type(leaf, di);
5570 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5571 type != BTRFS_FT_DIR)
5573 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5574 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5577 btrfs_release_path(path);
5578 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5579 if (IS_ERR(di_inode)) {
5580 ret = PTR_ERR(di_inode);
5581 goto next_dir_inode;
5584 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5585 btrfs_add_delayed_iput(di_inode);
5589 ctx->log_new_dentries = false;
5590 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5591 log_mode = LOG_INODE_ALL;
5592 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5593 log_mode, 0, LLONG_MAX, ctx);
5595 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5597 btrfs_add_delayed_iput(di_inode);
5599 goto next_dir_inode;
5600 if (ctx->log_new_dentries) {
5601 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5603 if (!new_dir_elem) {
5605 goto next_dir_inode;
5607 new_dir_elem->ino = di_key.objectid;
5608 list_add_tail(&new_dir_elem->list, &dir_list);
5613 ret = btrfs_next_leaf(log, path);
5615 goto next_dir_inode;
5616 } else if (ret > 0) {
5618 goto next_dir_inode;
5622 if (min_key.offset < (u64)-1) {
5627 list_del(&dir_elem->list);
5631 btrfs_free_path(path);
5635 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5636 struct btrfs_inode *inode,
5637 struct btrfs_log_ctx *ctx)
5639 struct btrfs_fs_info *fs_info = trans->fs_info;
5641 struct btrfs_path *path;
5642 struct btrfs_key key;
5643 struct btrfs_root *root = inode->root;
5644 const u64 ino = btrfs_ino(inode);
5646 path = btrfs_alloc_path();
5649 path->skip_locking = 1;
5650 path->search_commit_root = 1;
5653 key.type = BTRFS_INODE_REF_KEY;
5655 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5660 struct extent_buffer *leaf = path->nodes[0];
5661 int slot = path->slots[0];
5666 if (slot >= btrfs_header_nritems(leaf)) {
5667 ret = btrfs_next_leaf(root, path);
5675 btrfs_item_key_to_cpu(leaf, &key, slot);
5676 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5677 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5680 item_size = btrfs_item_size_nr(leaf, slot);
5681 ptr = btrfs_item_ptr_offset(leaf, slot);
5682 while (cur_offset < item_size) {
5683 struct btrfs_key inode_key;
5684 struct inode *dir_inode;
5686 inode_key.type = BTRFS_INODE_ITEM_KEY;
5687 inode_key.offset = 0;
5689 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5690 struct btrfs_inode_extref *extref;
5692 extref = (struct btrfs_inode_extref *)
5694 inode_key.objectid = btrfs_inode_extref_parent(
5696 cur_offset += sizeof(*extref);
5697 cur_offset += btrfs_inode_extref_name_len(leaf,
5700 inode_key.objectid = key.offset;
5701 cur_offset = item_size;
5704 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5707 * If the parent inode was deleted, return an error to
5708 * fallback to a transaction commit. This is to prevent
5709 * getting an inode that was moved from one parent A to
5710 * a parent B, got its former parent A deleted and then
5711 * it got fsync'ed, from existing at both parents after
5712 * a log replay (and the old parent still existing).
5719 * mv /mnt/B/bar /mnt/A/bar
5720 * mv -T /mnt/A /mnt/B
5724 * If we ignore the old parent B which got deleted,
5725 * after a log replay we would have file bar linked
5726 * at both parents and the old parent B would still
5729 if (IS_ERR(dir_inode)) {
5730 ret = PTR_ERR(dir_inode);
5735 ctx->log_new_dentries = false;
5736 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5737 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5739 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5741 if (!ret && ctx && ctx->log_new_dentries)
5742 ret = log_new_dir_dentries(trans, root,
5743 BTRFS_I(dir_inode), ctx);
5744 btrfs_add_delayed_iput(dir_inode);
5752 btrfs_free_path(path);
5756 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5757 struct btrfs_root *root,
5758 struct btrfs_path *path,
5759 struct btrfs_log_ctx *ctx)
5761 struct btrfs_key found_key;
5763 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5766 struct btrfs_fs_info *fs_info = root->fs_info;
5767 const u64 last_committed = fs_info->last_trans_committed;
5768 struct extent_buffer *leaf = path->nodes[0];
5769 int slot = path->slots[0];
5770 struct btrfs_key search_key;
5771 struct inode *inode;
5775 btrfs_release_path(path);
5777 ino = found_key.offset;
5779 search_key.objectid = found_key.offset;
5780 search_key.type = BTRFS_INODE_ITEM_KEY;
5781 search_key.offset = 0;
5782 inode = btrfs_iget(fs_info->sb, ino, root);
5784 return PTR_ERR(inode);
5786 if (BTRFS_I(inode)->generation > last_committed)
5787 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5790 btrfs_add_delayed_iput(inode);
5794 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5797 search_key.type = BTRFS_INODE_REF_KEY;
5798 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5802 leaf = path->nodes[0];
5803 slot = path->slots[0];
5804 if (slot >= btrfs_header_nritems(leaf)) {
5805 ret = btrfs_next_leaf(root, path);
5810 leaf = path->nodes[0];
5811 slot = path->slots[0];
5814 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5815 if (found_key.objectid != search_key.objectid ||
5816 found_key.type != BTRFS_INODE_REF_KEY)
5822 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5823 struct btrfs_inode *inode,
5824 struct dentry *parent,
5825 struct btrfs_log_ctx *ctx)
5827 struct btrfs_root *root = inode->root;
5828 struct btrfs_fs_info *fs_info = root->fs_info;
5829 struct dentry *old_parent = NULL;
5830 struct super_block *sb = inode->vfs_inode.i_sb;
5834 if (!parent || d_really_is_negative(parent) ||
5838 inode = BTRFS_I(d_inode(parent));
5839 if (root != inode->root)
5842 if (inode->generation > fs_info->last_trans_committed) {
5843 ret = btrfs_log_inode(trans, root, inode,
5844 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5848 if (IS_ROOT(parent))
5851 parent = dget_parent(parent);
5853 old_parent = parent;
5860 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5861 struct btrfs_inode *inode,
5862 struct dentry *parent,
5863 struct btrfs_log_ctx *ctx)
5865 struct btrfs_root *root = inode->root;
5866 const u64 ino = btrfs_ino(inode);
5867 struct btrfs_path *path;
5868 struct btrfs_key search_key;
5872 * For a single hard link case, go through a fast path that does not
5873 * need to iterate the fs/subvolume tree.
5875 if (inode->vfs_inode.i_nlink < 2)
5876 return log_new_ancestors_fast(trans, inode, parent, ctx);
5878 path = btrfs_alloc_path();
5882 search_key.objectid = ino;
5883 search_key.type = BTRFS_INODE_REF_KEY;
5884 search_key.offset = 0;
5886 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5893 struct extent_buffer *leaf = path->nodes[0];
5894 int slot = path->slots[0];
5895 struct btrfs_key found_key;
5897 if (slot >= btrfs_header_nritems(leaf)) {
5898 ret = btrfs_next_leaf(root, path);
5906 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5907 if (found_key.objectid != ino ||
5908 found_key.type > BTRFS_INODE_EXTREF_KEY)
5912 * Don't deal with extended references because they are rare
5913 * cases and too complex to deal with (we would need to keep
5914 * track of which subitem we are processing for each item in
5915 * this loop, etc). So just return some error to fallback to
5916 * a transaction commit.
5918 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
5924 * Logging ancestors needs to do more searches on the fs/subvol
5925 * tree, so it releases the path as needed to avoid deadlocks.
5926 * Keep track of the last inode ref key and resume from that key
5927 * after logging all new ancestors for the current hard link.
5929 memcpy(&search_key, &found_key, sizeof(search_key));
5931 ret = log_new_ancestors(trans, root, path, ctx);
5934 btrfs_release_path(path);
5939 btrfs_free_path(path);
5944 * helper function around btrfs_log_inode to make sure newly created
5945 * parent directories also end up in the log. A minimal inode and backref
5946 * only logging is done of any parent directories that are older than
5947 * the last committed transaction
5949 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5950 struct btrfs_inode *inode,
5951 struct dentry *parent,
5955 struct btrfs_log_ctx *ctx)
5957 struct btrfs_root *root = inode->root;
5958 struct btrfs_fs_info *fs_info = root->fs_info;
5959 struct super_block *sb;
5961 u64 last_committed = fs_info->last_trans_committed;
5962 bool log_dentries = false;
5964 sb = inode->vfs_inode.i_sb;
5966 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5972 * The prev transaction commit doesn't complete, we need do
5973 * full commit by ourselves.
5975 if (fs_info->last_trans_log_full_commit >
5976 fs_info->last_trans_committed) {
5981 if (btrfs_root_refs(&root->root_item) == 0) {
5986 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5992 * Skip already logged inodes or inodes corresponding to tmpfiles
5993 * (since logging them is pointless, a link count of 0 means they
5994 * will never be accessible).
5996 if (btrfs_inode_in_log(inode, trans->transid) ||
5997 inode->vfs_inode.i_nlink == 0) {
5998 ret = BTRFS_NO_LOG_SYNC;
6002 ret = start_log_trans(trans, root, ctx);
6006 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
6011 * for regular files, if its inode is already on disk, we don't
6012 * have to worry about the parents at all. This is because
6013 * we can use the last_unlink_trans field to record renames
6014 * and other fun in this file.
6016 if (S_ISREG(inode->vfs_inode.i_mode) &&
6017 inode->generation <= last_committed &&
6018 inode->last_unlink_trans <= last_committed) {
6023 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6024 log_dentries = true;
6027 * On unlink we must make sure all our current and old parent directory
6028 * inodes are fully logged. This is to prevent leaving dangling
6029 * directory index entries in directories that were our parents but are
6030 * not anymore. Not doing this results in old parent directory being
6031 * impossible to delete after log replay (rmdir will always fail with
6032 * error -ENOTEMPTY).
6038 * ln testdir/foo testdir/bar
6040 * unlink testdir/bar
6041 * xfs_io -c fsync testdir/foo
6043 * mount fs, triggers log replay
6045 * If we don't log the parent directory (testdir), after log replay the
6046 * directory still has an entry pointing to the file inode using the bar
6047 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6048 * the file inode has a link count of 1.
6054 * ln foo testdir/foo2
6055 * ln foo testdir/foo3
6057 * unlink testdir/foo3
6058 * xfs_io -c fsync foo
6060 * mount fs, triggers log replay
6062 * Similar as the first example, after log replay the parent directory
6063 * testdir still has an entry pointing to the inode file with name foo3
6064 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6065 * and has a link count of 2.
6067 if (inode->last_unlink_trans > last_committed) {
6068 ret = btrfs_log_all_parents(trans, inode, ctx);
6073 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6078 ret = log_new_dir_dentries(trans, root, inode, ctx);
6083 btrfs_set_log_full_commit(trans);
6088 btrfs_remove_log_ctx(root, ctx);
6089 btrfs_end_log_trans(root);
6095 * it is not safe to log dentry if the chunk root has added new
6096 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6097 * If this returns 1, you must commit the transaction to safely get your
6100 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6101 struct dentry *dentry,
6104 struct btrfs_log_ctx *ctx)
6106 struct dentry *parent = dget_parent(dentry);
6109 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6110 start, end, LOG_INODE_ALL, ctx);
6117 * should be called during mount to recover any replay any log trees
6120 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6123 struct btrfs_path *path;
6124 struct btrfs_trans_handle *trans;
6125 struct btrfs_key key;
6126 struct btrfs_key found_key;
6127 struct btrfs_root *log;
6128 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6129 struct walk_control wc = {
6130 .process_func = process_one_buffer,
6131 .stage = LOG_WALK_PIN_ONLY,
6134 path = btrfs_alloc_path();
6138 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6140 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6141 if (IS_ERR(trans)) {
6142 ret = PTR_ERR(trans);
6149 ret = walk_log_tree(trans, log_root_tree, &wc);
6151 btrfs_handle_fs_error(fs_info, ret,
6152 "Failed to pin buffers while recovering log root tree.");
6157 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6158 key.offset = (u64)-1;
6159 key.type = BTRFS_ROOT_ITEM_KEY;
6162 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6165 btrfs_handle_fs_error(fs_info, ret,
6166 "Couldn't find tree log root.");
6170 if (path->slots[0] == 0)
6174 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6176 btrfs_release_path(path);
6177 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6180 log = btrfs_read_tree_root(log_root_tree, &found_key);
6183 btrfs_handle_fs_error(fs_info, ret,
6184 "Couldn't read tree log root.");
6188 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6190 if (IS_ERR(wc.replay_dest)) {
6191 ret = PTR_ERR(wc.replay_dest);
6194 * We didn't find the subvol, likely because it was
6195 * deleted. This is ok, simply skip this log and go to
6198 * We need to exclude the root because we can't have
6199 * other log replays overwriting this log as we'll read
6200 * it back in a few more times. This will keep our
6201 * block from being modified, and we'll just bail for
6202 * each subsequent pass.
6205 ret = btrfs_pin_extent_for_log_replay(trans,
6208 btrfs_put_root(log);
6212 btrfs_handle_fs_error(fs_info, ret,
6213 "Couldn't read target root for tree log recovery.");
6217 wc.replay_dest->log_root = log;
6218 btrfs_record_root_in_trans(trans, wc.replay_dest);
6219 ret = walk_log_tree(trans, log, &wc);
6221 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6222 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6226 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6227 struct btrfs_root *root = wc.replay_dest;
6229 btrfs_release_path(path);
6232 * We have just replayed everything, and the highest
6233 * objectid of fs roots probably has changed in case
6234 * some inode_item's got replayed.
6236 * root->objectid_mutex is not acquired as log replay
6237 * could only happen during mount.
6239 ret = btrfs_find_highest_objectid(root,
6240 &root->highest_objectid);
6243 wc.replay_dest->log_root = NULL;
6244 btrfs_put_root(wc.replay_dest);
6245 btrfs_put_root(log);
6250 if (found_key.offset == 0)
6252 key.offset = found_key.offset - 1;
6254 btrfs_release_path(path);
6256 /* step one is to pin it all, step two is to replay just inodes */
6259 wc.process_func = replay_one_buffer;
6260 wc.stage = LOG_WALK_REPLAY_INODES;
6263 /* step three is to replay everything */
6264 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6269 btrfs_free_path(path);
6271 /* step 4: commit the transaction, which also unpins the blocks */
6272 ret = btrfs_commit_transaction(trans);
6276 log_root_tree->log_root = NULL;
6277 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6278 btrfs_put_root(log_root_tree);
6283 btrfs_end_transaction(wc.trans);
6284 btrfs_free_path(path);
6289 * there are some corner cases where we want to force a full
6290 * commit instead of allowing a directory to be logged.
6292 * They revolve around files there were unlinked from the directory, and
6293 * this function updates the parent directory so that a full commit is
6294 * properly done if it is fsync'd later after the unlinks are done.
6296 * Must be called before the unlink operations (updates to the subvolume tree,
6297 * inodes, etc) are done.
6299 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6300 struct btrfs_inode *dir, struct btrfs_inode *inode,
6304 * when we're logging a file, if it hasn't been renamed
6305 * or unlinked, and its inode is fully committed on disk,
6306 * we don't have to worry about walking up the directory chain
6307 * to log its parents.
6309 * So, we use the last_unlink_trans field to put this transid
6310 * into the file. When the file is logged we check it and
6311 * don't log the parents if the file is fully on disk.
6313 mutex_lock(&inode->log_mutex);
6314 inode->last_unlink_trans = trans->transid;
6315 mutex_unlock(&inode->log_mutex);
6318 * if this directory was already logged any new
6319 * names for this file/dir will get recorded
6321 if (dir->logged_trans == trans->transid)
6325 * if the inode we're about to unlink was logged,
6326 * the log will be properly updated for any new names
6328 if (inode->logged_trans == trans->transid)
6332 * when renaming files across directories, if the directory
6333 * there we're unlinking from gets fsync'd later on, there's
6334 * no way to find the destination directory later and fsync it
6335 * properly. So, we have to be conservative and force commits
6336 * so the new name gets discovered.
6341 /* we can safely do the unlink without any special recording */
6345 mutex_lock(&dir->log_mutex);
6346 dir->last_unlink_trans = trans->transid;
6347 mutex_unlock(&dir->log_mutex);
6351 * Make sure that if someone attempts to fsync the parent directory of a deleted
6352 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6353 * that after replaying the log tree of the parent directory's root we will not
6354 * see the snapshot anymore and at log replay time we will not see any log tree
6355 * corresponding to the deleted snapshot's root, which could lead to replaying
6356 * it after replaying the log tree of the parent directory (which would replay
6357 * the snapshot delete operation).
6359 * Must be called before the actual snapshot destroy operation (updates to the
6360 * parent root and tree of tree roots trees, etc) are done.
6362 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6363 struct btrfs_inode *dir)
6365 mutex_lock(&dir->log_mutex);
6366 dir->last_unlink_trans = trans->transid;
6367 mutex_unlock(&dir->log_mutex);
6371 * Call this after adding a new name for a file and it will properly
6372 * update the log to reflect the new name.
6374 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6375 * true (because it's not used).
6377 * Return value depends on whether @sync_log is true or false.
6378 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6379 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6381 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6382 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6383 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6384 * committed (without attempting to sync the log).
6386 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6387 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6388 struct dentry *parent,
6389 bool sync_log, struct btrfs_log_ctx *ctx)
6391 struct btrfs_fs_info *fs_info = trans->fs_info;
6395 * this will force the logging code to walk the dentry chain
6398 if (!S_ISDIR(inode->vfs_inode.i_mode))
6399 inode->last_unlink_trans = trans->transid;
6402 * if this inode hasn't been logged and directory we're renaming it
6403 * from hasn't been logged, we don't need to log it
6405 if (inode->logged_trans <= fs_info->last_trans_committed &&
6406 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6407 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6408 BTRFS_DONT_NEED_LOG_SYNC;
6411 struct btrfs_log_ctx ctx2;
6413 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6414 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6415 LOG_INODE_EXISTS, &ctx2);
6416 if (ret == BTRFS_NO_LOG_SYNC)
6417 return BTRFS_DONT_NEED_TRANS_COMMIT;
6419 return BTRFS_NEED_TRANS_COMMIT;
6421 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6423 return BTRFS_NEED_TRANS_COMMIT;
6424 return BTRFS_DONT_NEED_TRANS_COMMIT;
6428 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6429 LOG_INODE_EXISTS, ctx);
6430 if (ret == BTRFS_NO_LOG_SYNC)
6431 return BTRFS_DONT_NEED_LOG_SYNC;
6433 return BTRFS_NEED_TRANS_COMMIT;
6435 return BTRFS_NEED_LOG_SYNC;