1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
16 #include "print-tree.h"
18 #include "compression.h"
20 #include "block-group.h"
21 #include "space-info.h"
23 #include "inode-item.h"
25 #include "accessors.h"
26 #include "extent-tree.h"
27 #include "root-tree.h"
29 #include "file-item.h"
32 #include "tree-checker.h"
34 #define MAX_CONFLICT_INODES 10
36 /* magic values for the inode_only field in btrfs_log_inode:
38 * LOG_INODE_ALL means to log everything
39 * LOG_INODE_EXISTS means to log just enough to recreate the inode
48 * directory trouble cases
50 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
51 * log, we must force a full commit before doing an fsync of the directory
52 * where the unlink was done.
53 * ---> record transid of last unlink/rename per directory
57 * rename foo/some_dir foo2/some_dir
59 * fsync foo/some_dir/some_file
61 * The fsync above will unlink the original some_dir without recording
62 * it in its new location (foo2). After a crash, some_dir will be gone
63 * unless the fsync of some_file forces a full commit
65 * 2) we must log any new names for any file or dir that is in the fsync
66 * log. ---> check inode while renaming/linking.
68 * 2a) we must log any new names for any file or dir during rename
69 * when the directory they are being removed from was logged.
70 * ---> check inode and old parent dir during rename
72 * 2a is actually the more important variant. With the extra logging
73 * a crash might unlink the old name without recreating the new one
75 * 3) after a crash, we must go through any directories with a link count
76 * of zero and redo the rm -rf
83 * The directory f1 was fully removed from the FS, but fsync was never
84 * called on f1, only its parent dir. After a crash the rm -rf must
85 * be replayed. This must be able to recurse down the entire
86 * directory tree. The inode link count fixup code takes care of the
91 * stages for the tree walking. The first
92 * stage (0) is to only pin down the blocks we find
93 * the second stage (1) is to make sure that all the inodes
94 * we find in the log are created in the subvolume.
96 * The last stage is to deal with directories and links and extents
97 * and all the other fun semantics
101 LOG_WALK_REPLAY_INODES,
102 LOG_WALK_REPLAY_DIR_INDEX,
106 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
107 struct btrfs_inode *inode,
109 struct btrfs_log_ctx *ctx);
110 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
111 struct btrfs_root *root,
112 struct btrfs_path *path, u64 objectid);
113 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
114 struct btrfs_root *root,
115 struct btrfs_root *log,
116 struct btrfs_path *path,
117 u64 dirid, int del_all);
118 static void wait_log_commit(struct btrfs_root *root, int transid);
121 * tree logging is a special write ahead log used to make sure that
122 * fsyncs and O_SYNCs can happen without doing full tree commits.
124 * Full tree commits are expensive because they require commonly
125 * modified blocks to be recowed, creating many dirty pages in the
126 * extent tree an 4x-6x higher write load than ext3.
128 * Instead of doing a tree commit on every fsync, we use the
129 * key ranges and transaction ids to find items for a given file or directory
130 * that have changed in this transaction. Those items are copied into
131 * a special tree (one per subvolume root), that tree is written to disk
132 * and then the fsync is considered complete.
134 * After a crash, items are copied out of the log-tree back into the
135 * subvolume tree. Any file data extents found are recorded in the extent
136 * allocation tree, and the log-tree freed.
138 * The log tree is read three times, once to pin down all the extents it is
139 * using in ram and once, once to create all the inodes logged in the tree
140 * and once to do all the other items.
144 * start a sub transaction and setup the log tree
145 * this increments the log tree writer count to make the people
146 * syncing the tree wait for us to finish
148 static int start_log_trans(struct btrfs_trans_handle *trans,
149 struct btrfs_root *root,
150 struct btrfs_log_ctx *ctx)
152 struct btrfs_fs_info *fs_info = root->fs_info;
153 struct btrfs_root *tree_root = fs_info->tree_root;
154 const bool zoned = btrfs_is_zoned(fs_info);
156 bool created = false;
159 * First check if the log root tree was already created. If not, create
160 * it before locking the root's log_mutex, just to keep lockdep happy.
162 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
163 mutex_lock(&tree_root->log_mutex);
164 if (!fs_info->log_root_tree) {
165 ret = btrfs_init_log_root_tree(trans, fs_info);
167 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
171 mutex_unlock(&tree_root->log_mutex);
176 mutex_lock(&root->log_mutex);
179 if (root->log_root) {
180 int index = (root->log_transid + 1) % 2;
182 if (btrfs_need_log_full_commit(trans)) {
183 ret = BTRFS_LOG_FORCE_COMMIT;
187 if (zoned && atomic_read(&root->log_commit[index])) {
188 wait_log_commit(root, root->log_transid - 1);
192 if (!root->log_start_pid) {
193 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
194 root->log_start_pid = current->pid;
195 } else if (root->log_start_pid != current->pid) {
196 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
200 * This means fs_info->log_root_tree was already created
201 * for some other FS trees. Do the full commit not to mix
202 * nodes from multiple log transactions to do sequential
205 if (zoned && !created) {
206 ret = BTRFS_LOG_FORCE_COMMIT;
210 ret = btrfs_add_log_tree(trans, root);
214 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
215 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
216 root->log_start_pid = current->pid;
219 atomic_inc(&root->log_writers);
220 if (!ctx->logging_new_name) {
221 int index = root->log_transid % 2;
222 list_add_tail(&ctx->list, &root->log_ctxs[index]);
223 ctx->log_transid = root->log_transid;
227 mutex_unlock(&root->log_mutex);
232 * returns 0 if there was a log transaction running and we were able
233 * to join, or returns -ENOENT if there were not transactions
236 static int join_running_log_trans(struct btrfs_root *root)
238 const bool zoned = btrfs_is_zoned(root->fs_info);
241 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
244 mutex_lock(&root->log_mutex);
246 if (root->log_root) {
247 int index = (root->log_transid + 1) % 2;
250 if (zoned && atomic_read(&root->log_commit[index])) {
251 wait_log_commit(root, root->log_transid - 1);
254 atomic_inc(&root->log_writers);
256 mutex_unlock(&root->log_mutex);
261 * This either makes the current running log transaction wait
262 * until you call btrfs_end_log_trans() or it makes any future
263 * log transactions wait until you call btrfs_end_log_trans()
265 void btrfs_pin_log_trans(struct btrfs_root *root)
267 atomic_inc(&root->log_writers);
271 * indicate we're done making changes to the log tree
272 * and wake up anyone waiting to do a sync
274 void btrfs_end_log_trans(struct btrfs_root *root)
276 if (atomic_dec_and_test(&root->log_writers)) {
277 /* atomic_dec_and_test implies a barrier */
278 cond_wake_up_nomb(&root->log_writer_wait);
283 * the walk control struct is used to pass state down the chain when
284 * processing the log tree. The stage field tells us which part
285 * of the log tree processing we are currently doing. The others
286 * are state fields used for that specific part
288 struct walk_control {
289 /* should we free the extent on disk when done? This is used
290 * at transaction commit time while freeing a log tree
294 /* pin only walk, we record which extents on disk belong to the
299 /* what stage of the replay code we're currently in */
303 * Ignore any items from the inode currently being processed. Needs
304 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
305 * the LOG_WALK_REPLAY_INODES stage.
307 bool ignore_cur_inode;
309 /* the root we are currently replaying */
310 struct btrfs_root *replay_dest;
312 /* the trans handle for the current replay */
313 struct btrfs_trans_handle *trans;
315 /* the function that gets used to process blocks we find in the
316 * tree. Note the extent_buffer might not be up to date when it is
317 * passed in, and it must be checked or read if you need the data
320 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
321 struct walk_control *wc, u64 gen, int level);
325 * process_func used to pin down extents, write them or wait on them
327 static int process_one_buffer(struct btrfs_root *log,
328 struct extent_buffer *eb,
329 struct walk_control *wc, u64 gen, int level)
331 struct btrfs_fs_info *fs_info = log->fs_info;
335 * If this fs is mixed then we need to be able to process the leaves to
336 * pin down any logged extents, so we have to read the block.
338 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
339 struct btrfs_tree_parent_check check = {
344 ret = btrfs_read_extent_buffer(eb, &check);
350 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb);
354 if (btrfs_buffer_uptodate(eb, gen, 0) &&
355 btrfs_header_level(eb) == 0)
356 ret = btrfs_exclude_logged_extents(eb);
362 * Item overwrite used by replay and tree logging. eb, slot and key all refer
363 * to the src data we are copying out.
365 * root is the tree we are copying into, and path is a scratch
366 * path for use in this function (it should be released on entry and
367 * will be released on exit).
369 * If the key is already in the destination tree the existing item is
370 * overwritten. If the existing item isn't big enough, it is extended.
371 * If it is too large, it is truncated.
373 * If the key isn't in the destination yet, a new item is inserted.
375 static int overwrite_item(struct btrfs_trans_handle *trans,
376 struct btrfs_root *root,
377 struct btrfs_path *path,
378 struct extent_buffer *eb, int slot,
379 struct btrfs_key *key)
383 u64 saved_i_size = 0;
384 int save_old_i_size = 0;
385 unsigned long src_ptr;
386 unsigned long dst_ptr;
387 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
390 * This is only used during log replay, so the root is always from a
391 * fs/subvolume tree. In case we ever need to support a log root, then
392 * we'll have to clone the leaf in the path, release the path and use
393 * the leaf before writing into the log tree. See the comments at
394 * copy_items() for more details.
396 ASSERT(root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
398 item_size = btrfs_item_size(eb, slot);
399 src_ptr = btrfs_item_ptr_offset(eb, slot);
401 /* Look for the key in the destination tree. */
402 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
409 u32 dst_size = btrfs_item_size(path->nodes[0],
411 if (dst_size != item_size)
414 if (item_size == 0) {
415 btrfs_release_path(path);
418 dst_copy = kmalloc(item_size, GFP_NOFS);
419 src_copy = kmalloc(item_size, GFP_NOFS);
420 if (!dst_copy || !src_copy) {
421 btrfs_release_path(path);
427 read_extent_buffer(eb, src_copy, src_ptr, item_size);
429 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
430 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
432 ret = memcmp(dst_copy, src_copy, item_size);
437 * they have the same contents, just return, this saves
438 * us from cowing blocks in the destination tree and doing
439 * extra writes that may not have been done by a previous
443 btrfs_release_path(path);
448 * We need to load the old nbytes into the inode so when we
449 * replay the extents we've logged we get the right nbytes.
452 struct btrfs_inode_item *item;
456 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
457 struct btrfs_inode_item);
458 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
459 item = btrfs_item_ptr(eb, slot,
460 struct btrfs_inode_item);
461 btrfs_set_inode_nbytes(eb, item, nbytes);
464 * If this is a directory we need to reset the i_size to
465 * 0 so that we can set it up properly when replaying
466 * the rest of the items in this log.
468 mode = btrfs_inode_mode(eb, item);
470 btrfs_set_inode_size(eb, item, 0);
472 } else if (inode_item) {
473 struct btrfs_inode_item *item;
477 * New inode, set nbytes to 0 so that the nbytes comes out
478 * properly when we replay the extents.
480 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
481 btrfs_set_inode_nbytes(eb, item, 0);
484 * If this is a directory we need to reset the i_size to 0 so
485 * that we can set it up properly when replaying the rest of
486 * the items in this log.
488 mode = btrfs_inode_mode(eb, item);
490 btrfs_set_inode_size(eb, item, 0);
493 btrfs_release_path(path);
494 /* try to insert the key into the destination tree */
495 path->skip_release_on_error = 1;
496 ret = btrfs_insert_empty_item(trans, root, path,
498 path->skip_release_on_error = 0;
500 /* make sure any existing item is the correct size */
501 if (ret == -EEXIST || ret == -EOVERFLOW) {
503 found_size = btrfs_item_size(path->nodes[0],
505 if (found_size > item_size)
506 btrfs_truncate_item(trans, path, item_size, 1);
507 else if (found_size < item_size)
508 btrfs_extend_item(trans, path, item_size - found_size);
512 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
515 /* don't overwrite an existing inode if the generation number
516 * was logged as zero. This is done when the tree logging code
517 * is just logging an inode to make sure it exists after recovery.
519 * Also, don't overwrite i_size on directories during replay.
520 * log replay inserts and removes directory items based on the
521 * state of the tree found in the subvolume, and i_size is modified
524 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
525 struct btrfs_inode_item *src_item;
526 struct btrfs_inode_item *dst_item;
528 src_item = (struct btrfs_inode_item *)src_ptr;
529 dst_item = (struct btrfs_inode_item *)dst_ptr;
531 if (btrfs_inode_generation(eb, src_item) == 0) {
532 struct extent_buffer *dst_eb = path->nodes[0];
533 const u64 ino_size = btrfs_inode_size(eb, src_item);
536 * For regular files an ino_size == 0 is used only when
537 * logging that an inode exists, as part of a directory
538 * fsync, and the inode wasn't fsynced before. In this
539 * case don't set the size of the inode in the fs/subvol
540 * tree, otherwise we would be throwing valid data away.
542 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
543 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
545 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
549 if (S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
550 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
552 saved_i_size = btrfs_inode_size(path->nodes[0],
557 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
560 if (save_old_i_size) {
561 struct btrfs_inode_item *dst_item;
562 dst_item = (struct btrfs_inode_item *)dst_ptr;
563 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
566 /* make sure the generation is filled in */
567 if (key->type == BTRFS_INODE_ITEM_KEY) {
568 struct btrfs_inode_item *dst_item;
569 dst_item = (struct btrfs_inode_item *)dst_ptr;
570 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
571 btrfs_set_inode_generation(path->nodes[0], dst_item,
576 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
577 btrfs_release_path(path);
581 static int read_alloc_one_name(struct extent_buffer *eb, void *start, int len,
582 struct fscrypt_str *name)
586 buf = kmalloc(len, GFP_NOFS);
590 read_extent_buffer(eb, buf, (unsigned long)start, len);
597 * simple helper to read an inode off the disk from a given root
598 * This can only be called for subvolume roots and not for the log
600 static noinline struct inode *read_one_inode(struct btrfs_root *root,
605 inode = btrfs_iget(root->fs_info->sb, objectid, root);
611 /* replays a single extent in 'eb' at 'slot' with 'key' into the
612 * subvolume 'root'. path is released on entry and should be released
615 * extents in the log tree have not been allocated out of the extent
616 * tree yet. So, this completes the allocation, taking a reference
617 * as required if the extent already exists or creating a new extent
618 * if it isn't in the extent allocation tree yet.
620 * The extent is inserted into the file, dropping any existing extents
621 * from the file that overlap the new one.
623 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
624 struct btrfs_root *root,
625 struct btrfs_path *path,
626 struct extent_buffer *eb, int slot,
627 struct btrfs_key *key)
629 struct btrfs_drop_extents_args drop_args = { 0 };
630 struct btrfs_fs_info *fs_info = root->fs_info;
633 u64 start = key->offset;
635 struct btrfs_file_extent_item *item;
636 struct inode *inode = NULL;
640 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
641 found_type = btrfs_file_extent_type(eb, item);
643 if (found_type == BTRFS_FILE_EXTENT_REG ||
644 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
645 nbytes = btrfs_file_extent_num_bytes(eb, item);
646 extent_end = start + nbytes;
649 * We don't add to the inodes nbytes if we are prealloc or a
652 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
654 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
655 size = btrfs_file_extent_ram_bytes(eb, item);
656 nbytes = btrfs_file_extent_ram_bytes(eb, item);
657 extent_end = ALIGN(start + size,
658 fs_info->sectorsize);
664 inode = read_one_inode(root, key->objectid);
671 * first check to see if we already have this extent in the
672 * file. This must be done before the btrfs_drop_extents run
673 * so we don't try to drop this extent.
675 ret = btrfs_lookup_file_extent(trans, root, path,
676 btrfs_ino(BTRFS_I(inode)), start, 0);
679 (found_type == BTRFS_FILE_EXTENT_REG ||
680 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
681 struct btrfs_file_extent_item cmp1;
682 struct btrfs_file_extent_item cmp2;
683 struct btrfs_file_extent_item *existing;
684 struct extent_buffer *leaf;
686 leaf = path->nodes[0];
687 existing = btrfs_item_ptr(leaf, path->slots[0],
688 struct btrfs_file_extent_item);
690 read_extent_buffer(eb, &cmp1, (unsigned long)item,
692 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
696 * we already have a pointer to this exact extent,
697 * we don't have to do anything
699 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
700 btrfs_release_path(path);
704 btrfs_release_path(path);
706 /* drop any overlapping extents */
707 drop_args.start = start;
708 drop_args.end = extent_end;
709 drop_args.drop_cache = true;
710 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
714 if (found_type == BTRFS_FILE_EXTENT_REG ||
715 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
717 unsigned long dest_offset;
718 struct btrfs_key ins;
720 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
721 btrfs_fs_incompat(fs_info, NO_HOLES))
724 ret = btrfs_insert_empty_item(trans, root, path, key,
728 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
730 copy_extent_buffer(path->nodes[0], eb, dest_offset,
731 (unsigned long)item, sizeof(*item));
733 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
734 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
735 ins.type = BTRFS_EXTENT_ITEM_KEY;
736 offset = key->offset - btrfs_file_extent_offset(eb, item);
739 * Manually record dirty extent, as here we did a shallow
740 * file extent item copy and skip normal backref update,
741 * but modifying extent tree all by ourselves.
742 * So need to manually record dirty extent for qgroup,
743 * as the owner of the file extent changed from log tree
744 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
746 ret = btrfs_qgroup_trace_extent(trans,
747 btrfs_file_extent_disk_bytenr(eb, item),
748 btrfs_file_extent_disk_num_bytes(eb, item));
752 if (ins.objectid > 0) {
753 struct btrfs_ref ref = { 0 };
756 LIST_HEAD(ordered_sums);
759 * is this extent already allocated in the extent
760 * allocation tree? If so, just add a reference
762 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
766 } else if (ret == 0) {
767 btrfs_init_generic_ref(&ref,
768 BTRFS_ADD_DELAYED_REF,
769 ins.objectid, ins.offset, 0,
770 root->root_key.objectid);
771 btrfs_init_data_ref(&ref,
772 root->root_key.objectid,
773 key->objectid, offset, 0, false);
774 ret = btrfs_inc_extent_ref(trans, &ref);
779 * insert the extent pointer in the extent
782 ret = btrfs_alloc_logged_file_extent(trans,
783 root->root_key.objectid,
784 key->objectid, offset, &ins);
788 btrfs_release_path(path);
790 if (btrfs_file_extent_compression(eb, item)) {
791 csum_start = ins.objectid;
792 csum_end = csum_start + ins.offset;
794 csum_start = ins.objectid +
795 btrfs_file_extent_offset(eb, item);
796 csum_end = csum_start +
797 btrfs_file_extent_num_bytes(eb, item);
800 ret = btrfs_lookup_csums_list(root->log_root,
801 csum_start, csum_end - 1,
802 &ordered_sums, 0, false);
806 * Now delete all existing cums in the csum root that
807 * cover our range. We do this because we can have an
808 * extent that is completely referenced by one file
809 * extent item and partially referenced by another
810 * file extent item (like after using the clone or
811 * extent_same ioctls). In this case if we end up doing
812 * the replay of the one that partially references the
813 * extent first, and we do not do the csum deletion
814 * below, we can get 2 csum items in the csum tree that
815 * overlap each other. For example, imagine our log has
816 * the two following file extent items:
818 * key (257 EXTENT_DATA 409600)
819 * extent data disk byte 12845056 nr 102400
820 * extent data offset 20480 nr 20480 ram 102400
822 * key (257 EXTENT_DATA 819200)
823 * extent data disk byte 12845056 nr 102400
824 * extent data offset 0 nr 102400 ram 102400
826 * Where the second one fully references the 100K extent
827 * that starts at disk byte 12845056, and the log tree
828 * has a single csum item that covers the entire range
831 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
833 * After the first file extent item is replayed, the
834 * csum tree gets the following csum item:
836 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
838 * Which covers the 20K sub-range starting at offset 20K
839 * of our extent. Now when we replay the second file
840 * extent item, if we do not delete existing csum items
841 * that cover any of its blocks, we end up getting two
842 * csum items in our csum tree that overlap each other:
844 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
845 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
847 * Which is a problem, because after this anyone trying
848 * to lookup up for the checksum of any block of our
849 * extent starting at an offset of 40K or higher, will
850 * end up looking at the second csum item only, which
851 * does not contain the checksum for any block starting
852 * at offset 40K or higher of our extent.
854 while (!list_empty(&ordered_sums)) {
855 struct btrfs_ordered_sum *sums;
856 struct btrfs_root *csum_root;
858 sums = list_entry(ordered_sums.next,
859 struct btrfs_ordered_sum,
861 csum_root = btrfs_csum_root(fs_info,
864 ret = btrfs_del_csums(trans, csum_root,
868 ret = btrfs_csum_file_blocks(trans,
871 list_del(&sums->list);
877 btrfs_release_path(path);
879 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
880 /* inline extents are easy, we just overwrite them */
881 ret = overwrite_item(trans, root, path, eb, slot, key);
886 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
892 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
893 ret = btrfs_update_inode(trans, BTRFS_I(inode));
899 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
900 struct btrfs_inode *dir,
901 struct btrfs_inode *inode,
902 const struct fscrypt_str *name)
906 ret = btrfs_unlink_inode(trans, dir, inode, name);
910 * Whenever we need to check if a name exists or not, we check the
911 * fs/subvolume tree. So after an unlink we must run delayed items, so
912 * that future checks for a name during log replay see that the name
913 * does not exists anymore.
915 return btrfs_run_delayed_items(trans);
919 * when cleaning up conflicts between the directory names in the
920 * subvolume, directory names in the log and directory names in the
921 * inode back references, we may have to unlink inodes from directories.
923 * This is a helper function to do the unlink of a specific directory
926 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
927 struct btrfs_path *path,
928 struct btrfs_inode *dir,
929 struct btrfs_dir_item *di)
931 struct btrfs_root *root = dir->root;
933 struct fscrypt_str name;
934 struct extent_buffer *leaf;
935 struct btrfs_key location;
938 leaf = path->nodes[0];
940 btrfs_dir_item_key_to_cpu(leaf, di, &location);
941 ret = read_alloc_one_name(leaf, di + 1, btrfs_dir_name_len(leaf, di), &name);
945 btrfs_release_path(path);
947 inode = read_one_inode(root, location.objectid);
953 ret = link_to_fixup_dir(trans, root, path, location.objectid);
957 ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), &name);
965 * See if a given name and sequence number found in an inode back reference are
966 * already in a directory and correctly point to this inode.
968 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
971 static noinline int inode_in_dir(struct btrfs_root *root,
972 struct btrfs_path *path,
973 u64 dirid, u64 objectid, u64 index,
974 struct fscrypt_str *name)
976 struct btrfs_dir_item *di;
977 struct btrfs_key location;
980 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
986 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
987 if (location.objectid != objectid)
993 btrfs_release_path(path);
994 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, 0);
999 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1000 if (location.objectid == objectid)
1004 btrfs_release_path(path);
1009 * helper function to check a log tree for a named back reference in
1010 * an inode. This is used to decide if a back reference that is
1011 * found in the subvolume conflicts with what we find in the log.
1013 * inode backreferences may have multiple refs in a single item,
1014 * during replay we process one reference at a time, and we don't
1015 * want to delete valid links to a file from the subvolume if that
1016 * link is also in the log.
1018 static noinline int backref_in_log(struct btrfs_root *log,
1019 struct btrfs_key *key,
1021 const struct fscrypt_str *name)
1023 struct btrfs_path *path;
1026 path = btrfs_alloc_path();
1030 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1033 } else if (ret == 1) {
1038 if (key->type == BTRFS_INODE_EXTREF_KEY)
1039 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1041 ref_objectid, name);
1043 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1044 path->slots[0], name);
1046 btrfs_free_path(path);
1050 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1051 struct btrfs_root *root,
1052 struct btrfs_path *path,
1053 struct btrfs_root *log_root,
1054 struct btrfs_inode *dir,
1055 struct btrfs_inode *inode,
1056 u64 inode_objectid, u64 parent_objectid,
1057 u64 ref_index, struct fscrypt_str *name)
1060 struct extent_buffer *leaf;
1061 struct btrfs_dir_item *di;
1062 struct btrfs_key search_key;
1063 struct btrfs_inode_extref *extref;
1066 /* Search old style refs */
1067 search_key.objectid = inode_objectid;
1068 search_key.type = BTRFS_INODE_REF_KEY;
1069 search_key.offset = parent_objectid;
1070 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1072 struct btrfs_inode_ref *victim_ref;
1074 unsigned long ptr_end;
1076 leaf = path->nodes[0];
1078 /* are we trying to overwrite a back ref for the root directory
1079 * if so, just jump out, we're done
1081 if (search_key.objectid == search_key.offset)
1084 /* check all the names in this back reference to see
1085 * if they are in the log. if so, we allow them to stay
1086 * otherwise they must be unlinked as a conflict
1088 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1089 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1090 while (ptr < ptr_end) {
1091 struct fscrypt_str victim_name;
1093 victim_ref = (struct btrfs_inode_ref *)ptr;
1094 ret = read_alloc_one_name(leaf, (victim_ref + 1),
1095 btrfs_inode_ref_name_len(leaf, victim_ref),
1100 ret = backref_in_log(log_root, &search_key,
1101 parent_objectid, &victim_name);
1103 kfree(victim_name.name);
1106 inc_nlink(&inode->vfs_inode);
1107 btrfs_release_path(path);
1109 ret = unlink_inode_for_log_replay(trans, dir, inode,
1111 kfree(victim_name.name);
1116 kfree(victim_name.name);
1118 ptr = (unsigned long)(victim_ref + 1) + victim_name.len;
1121 btrfs_release_path(path);
1123 /* Same search but for extended refs */
1124 extref = btrfs_lookup_inode_extref(NULL, root, path, name,
1125 inode_objectid, parent_objectid, 0,
1127 if (IS_ERR(extref)) {
1128 return PTR_ERR(extref);
1129 } else if (extref) {
1133 struct inode *victim_parent;
1135 leaf = path->nodes[0];
1137 item_size = btrfs_item_size(leaf, path->slots[0]);
1138 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1140 while (cur_offset < item_size) {
1141 struct fscrypt_str victim_name;
1143 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1145 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1148 ret = read_alloc_one_name(leaf, &extref->name,
1149 btrfs_inode_extref_name_len(leaf, extref),
1154 search_key.objectid = inode_objectid;
1155 search_key.type = BTRFS_INODE_EXTREF_KEY;
1156 search_key.offset = btrfs_extref_hash(parent_objectid,
1159 ret = backref_in_log(log_root, &search_key,
1160 parent_objectid, &victim_name);
1162 kfree(victim_name.name);
1166 victim_parent = read_one_inode(root,
1168 if (victim_parent) {
1169 inc_nlink(&inode->vfs_inode);
1170 btrfs_release_path(path);
1172 ret = unlink_inode_for_log_replay(trans,
1173 BTRFS_I(victim_parent),
1174 inode, &victim_name);
1176 iput(victim_parent);
1177 kfree(victim_name.name);
1182 kfree(victim_name.name);
1184 cur_offset += victim_name.len + sizeof(*extref);
1187 btrfs_release_path(path);
1189 /* look for a conflicting sequence number */
1190 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1191 ref_index, name, 0);
1195 ret = drop_one_dir_item(trans, path, dir, di);
1199 btrfs_release_path(path);
1201 /* look for a conflicting name */
1202 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), name, 0);
1206 ret = drop_one_dir_item(trans, path, dir, di);
1210 btrfs_release_path(path);
1215 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1216 struct fscrypt_str *name, u64 *index,
1217 u64 *parent_objectid)
1219 struct btrfs_inode_extref *extref;
1222 extref = (struct btrfs_inode_extref *)ref_ptr;
1224 ret = read_alloc_one_name(eb, &extref->name,
1225 btrfs_inode_extref_name_len(eb, extref), name);
1230 *index = btrfs_inode_extref_index(eb, extref);
1231 if (parent_objectid)
1232 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1237 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1238 struct fscrypt_str *name, u64 *index)
1240 struct btrfs_inode_ref *ref;
1243 ref = (struct btrfs_inode_ref *)ref_ptr;
1245 ret = read_alloc_one_name(eb, ref + 1, btrfs_inode_ref_name_len(eb, ref),
1251 *index = btrfs_inode_ref_index(eb, ref);
1257 * Take an inode reference item from the log tree and iterate all names from the
1258 * inode reference item in the subvolume tree with the same key (if it exists).
1259 * For any name that is not in the inode reference item from the log tree, do a
1260 * proper unlink of that name (that is, remove its entry from the inode
1261 * reference item and both dir index keys).
1263 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1264 struct btrfs_root *root,
1265 struct btrfs_path *path,
1266 struct btrfs_inode *inode,
1267 struct extent_buffer *log_eb,
1269 struct btrfs_key *key)
1272 unsigned long ref_ptr;
1273 unsigned long ref_end;
1274 struct extent_buffer *eb;
1277 btrfs_release_path(path);
1278 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1286 eb = path->nodes[0];
1287 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1288 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1289 while (ref_ptr < ref_end) {
1290 struct fscrypt_str name;
1293 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1294 ret = extref_get_fields(eb, ref_ptr, &name,
1297 parent_id = key->offset;
1298 ret = ref_get_fields(eb, ref_ptr, &name, NULL);
1303 if (key->type == BTRFS_INODE_EXTREF_KEY)
1304 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1307 ret = !!btrfs_find_name_in_backref(log_eb, log_slot, &name);
1312 btrfs_release_path(path);
1313 dir = read_one_inode(root, parent_id);
1319 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1329 ref_ptr += name.len;
1330 if (key->type == BTRFS_INODE_EXTREF_KEY)
1331 ref_ptr += sizeof(struct btrfs_inode_extref);
1333 ref_ptr += sizeof(struct btrfs_inode_ref);
1337 btrfs_release_path(path);
1342 * replay one inode back reference item found in the log tree.
1343 * eb, slot and key refer to the buffer and key found in the log tree.
1344 * root is the destination we are replaying into, and path is for temp
1345 * use by this function. (it should be released on return).
1347 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1348 struct btrfs_root *root,
1349 struct btrfs_root *log,
1350 struct btrfs_path *path,
1351 struct extent_buffer *eb, int slot,
1352 struct btrfs_key *key)
1354 struct inode *dir = NULL;
1355 struct inode *inode = NULL;
1356 unsigned long ref_ptr;
1357 unsigned long ref_end;
1358 struct fscrypt_str name;
1360 int log_ref_ver = 0;
1361 u64 parent_objectid;
1364 int ref_struct_size;
1366 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1367 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1369 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1370 struct btrfs_inode_extref *r;
1372 ref_struct_size = sizeof(struct btrfs_inode_extref);
1374 r = (struct btrfs_inode_extref *)ref_ptr;
1375 parent_objectid = btrfs_inode_extref_parent(eb, r);
1377 ref_struct_size = sizeof(struct btrfs_inode_ref);
1378 parent_objectid = key->offset;
1380 inode_objectid = key->objectid;
1383 * it is possible that we didn't log all the parent directories
1384 * for a given inode. If we don't find the dir, just don't
1385 * copy the back ref in. The link count fixup code will take
1388 dir = read_one_inode(root, parent_objectid);
1394 inode = read_one_inode(root, inode_objectid);
1400 while (ref_ptr < ref_end) {
1402 ret = extref_get_fields(eb, ref_ptr, &name,
1403 &ref_index, &parent_objectid);
1405 * parent object can change from one array
1409 dir = read_one_inode(root, parent_objectid);
1415 ret = ref_get_fields(eb, ref_ptr, &name, &ref_index);
1420 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1421 btrfs_ino(BTRFS_I(inode)), ref_index, &name);
1424 } else if (ret == 0) {
1426 * look for a conflicting back reference in the
1427 * metadata. if we find one we have to unlink that name
1428 * of the file before we add our new link. Later on, we
1429 * overwrite any existing back reference, and we don't
1430 * want to create dangling pointers in the directory.
1432 ret = __add_inode_ref(trans, root, path, log,
1433 BTRFS_I(dir), BTRFS_I(inode),
1434 inode_objectid, parent_objectid,
1442 /* insert our name */
1443 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1444 &name, 0, ref_index);
1448 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1452 /* Else, ret == 1, we already have a perfect match, we're done. */
1454 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + name.len;
1464 * Before we overwrite the inode reference item in the subvolume tree
1465 * with the item from the log tree, we must unlink all names from the
1466 * parent directory that are in the subvolume's tree inode reference
1467 * item, otherwise we end up with an inconsistent subvolume tree where
1468 * dir index entries exist for a name but there is no inode reference
1469 * item with the same name.
1471 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1476 /* finally write the back reference in the inode */
1477 ret = overwrite_item(trans, root, path, eb, slot, key);
1479 btrfs_release_path(path);
1486 static int count_inode_extrefs(struct btrfs_inode *inode, struct btrfs_path *path)
1490 unsigned int nlink = 0;
1493 u64 inode_objectid = btrfs_ino(inode);
1496 struct btrfs_inode_extref *extref;
1497 struct extent_buffer *leaf;
1500 ret = btrfs_find_one_extref(inode->root, inode_objectid, offset,
1501 path, &extref, &offset);
1505 leaf = path->nodes[0];
1506 item_size = btrfs_item_size(leaf, path->slots[0]);
1507 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1510 while (cur_offset < item_size) {
1511 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1512 name_len = btrfs_inode_extref_name_len(leaf, extref);
1516 cur_offset += name_len + sizeof(*extref);
1520 btrfs_release_path(path);
1522 btrfs_release_path(path);
1524 if (ret < 0 && ret != -ENOENT)
1529 static int count_inode_refs(struct btrfs_inode *inode, struct btrfs_path *path)
1532 struct btrfs_key key;
1533 unsigned int nlink = 0;
1535 unsigned long ptr_end;
1537 u64 ino = btrfs_ino(inode);
1540 key.type = BTRFS_INODE_REF_KEY;
1541 key.offset = (u64)-1;
1544 ret = btrfs_search_slot(NULL, inode->root, &key, path, 0, 0);
1548 if (path->slots[0] == 0)
1553 btrfs_item_key_to_cpu(path->nodes[0], &key,
1555 if (key.objectid != ino ||
1556 key.type != BTRFS_INODE_REF_KEY)
1558 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1559 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1561 while (ptr < ptr_end) {
1562 struct btrfs_inode_ref *ref;
1564 ref = (struct btrfs_inode_ref *)ptr;
1565 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1567 ptr = (unsigned long)(ref + 1) + name_len;
1571 if (key.offset == 0)
1573 if (path->slots[0] > 0) {
1578 btrfs_release_path(path);
1580 btrfs_release_path(path);
1586 * There are a few corners where the link count of the file can't
1587 * be properly maintained during replay. So, instead of adding
1588 * lots of complexity to the log code, we just scan the backrefs
1589 * for any file that has been through replay.
1591 * The scan will update the link count on the inode to reflect the
1592 * number of back refs found. If it goes down to zero, the iput
1593 * will free the inode.
1595 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1596 struct inode *inode)
1598 struct btrfs_root *root = BTRFS_I(inode)->root;
1599 struct btrfs_path *path;
1602 u64 ino = btrfs_ino(BTRFS_I(inode));
1604 path = btrfs_alloc_path();
1608 ret = count_inode_refs(BTRFS_I(inode), path);
1614 ret = count_inode_extrefs(BTRFS_I(inode), path);
1622 if (nlink != inode->i_nlink) {
1623 set_nlink(inode, nlink);
1624 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1628 BTRFS_I(inode)->index_cnt = (u64)-1;
1630 if (inode->i_nlink == 0) {
1631 if (S_ISDIR(inode->i_mode)) {
1632 ret = replay_dir_deletes(trans, root, NULL, path,
1637 ret = btrfs_insert_orphan_item(trans, root, ino);
1643 btrfs_free_path(path);
1647 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1648 struct btrfs_root *root,
1649 struct btrfs_path *path)
1652 struct btrfs_key key;
1653 struct inode *inode;
1655 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1656 key.type = BTRFS_ORPHAN_ITEM_KEY;
1657 key.offset = (u64)-1;
1659 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1665 if (path->slots[0] == 0)
1670 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1671 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1672 key.type != BTRFS_ORPHAN_ITEM_KEY)
1675 ret = btrfs_del_item(trans, root, path);
1679 btrfs_release_path(path);
1680 inode = read_one_inode(root, key.offset);
1686 ret = fixup_inode_link_count(trans, inode);
1692 * fixup on a directory may create new entries,
1693 * make sure we always look for the highset possible
1696 key.offset = (u64)-1;
1698 btrfs_release_path(path);
1704 * record a given inode in the fixup dir so we can check its link
1705 * count when replay is done. The link count is incremented here
1706 * so the inode won't go away until we check it
1708 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root,
1710 struct btrfs_path *path,
1713 struct btrfs_key key;
1715 struct inode *inode;
1717 inode = read_one_inode(root, objectid);
1721 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1722 key.type = BTRFS_ORPHAN_ITEM_KEY;
1723 key.offset = objectid;
1725 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1727 btrfs_release_path(path);
1729 if (!inode->i_nlink)
1730 set_nlink(inode, 1);
1733 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1734 } else if (ret == -EEXIST) {
1743 * when replaying the log for a directory, we only insert names
1744 * for inodes that actually exist. This means an fsync on a directory
1745 * does not implicitly fsync all the new files in it
1747 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1748 struct btrfs_root *root,
1749 u64 dirid, u64 index,
1750 const struct fscrypt_str *name,
1751 struct btrfs_key *location)
1753 struct inode *inode;
1757 inode = read_one_inode(root, location->objectid);
1761 dir = read_one_inode(root, dirid);
1767 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1770 /* FIXME, put inode into FIXUP list */
1777 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1778 struct btrfs_inode *dir,
1779 struct btrfs_path *path,
1780 struct btrfs_dir_item *dst_di,
1781 const struct btrfs_key *log_key,
1785 struct btrfs_key found_key;
1787 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1788 /* The existing dentry points to the same inode, don't delete it. */
1789 if (found_key.objectid == log_key->objectid &&
1790 found_key.type == log_key->type &&
1791 found_key.offset == log_key->offset &&
1792 btrfs_dir_flags(path->nodes[0], dst_di) == log_flags)
1796 * Don't drop the conflicting directory entry if the inode for the new
1797 * entry doesn't exist.
1802 return drop_one_dir_item(trans, path, dir, dst_di);
1806 * take a single entry in a log directory item and replay it into
1809 * if a conflicting item exists in the subdirectory already,
1810 * the inode it points to is unlinked and put into the link count
1813 * If a name from the log points to a file or directory that does
1814 * not exist in the FS, it is skipped. fsyncs on directories
1815 * do not force down inodes inside that directory, just changes to the
1816 * names or unlinks in a directory.
1818 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1819 * non-existing inode) and 1 if the name was replayed.
1821 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1822 struct btrfs_root *root,
1823 struct btrfs_path *path,
1824 struct extent_buffer *eb,
1825 struct btrfs_dir_item *di,
1826 struct btrfs_key *key)
1828 struct fscrypt_str name;
1829 struct btrfs_dir_item *dir_dst_di;
1830 struct btrfs_dir_item *index_dst_di;
1831 bool dir_dst_matches = false;
1832 bool index_dst_matches = false;
1833 struct btrfs_key log_key;
1834 struct btrfs_key search_key;
1839 bool update_size = true;
1840 bool name_added = false;
1842 dir = read_one_inode(root, key->objectid);
1846 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
1850 log_flags = btrfs_dir_flags(eb, di);
1851 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1852 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1853 btrfs_release_path(path);
1856 exists = (ret == 0);
1859 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1861 if (IS_ERR(dir_dst_di)) {
1862 ret = PTR_ERR(dir_dst_di);
1864 } else if (dir_dst_di) {
1865 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1866 dir_dst_di, &log_key,
1870 dir_dst_matches = (ret == 1);
1873 btrfs_release_path(path);
1875 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1876 key->objectid, key->offset,
1878 if (IS_ERR(index_dst_di)) {
1879 ret = PTR_ERR(index_dst_di);
1881 } else if (index_dst_di) {
1882 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1883 index_dst_di, &log_key,
1887 index_dst_matches = (ret == 1);
1890 btrfs_release_path(path);
1892 if (dir_dst_matches && index_dst_matches) {
1894 update_size = false;
1899 * Check if the inode reference exists in the log for the given name,
1900 * inode and parent inode
1902 search_key.objectid = log_key.objectid;
1903 search_key.type = BTRFS_INODE_REF_KEY;
1904 search_key.offset = key->objectid;
1905 ret = backref_in_log(root->log_root, &search_key, 0, &name);
1909 /* The dentry will be added later. */
1911 update_size = false;
1915 search_key.objectid = log_key.objectid;
1916 search_key.type = BTRFS_INODE_EXTREF_KEY;
1917 search_key.offset = key->objectid;
1918 ret = backref_in_log(root->log_root, &search_key, key->objectid, &name);
1922 /* The dentry will be added later. */
1924 update_size = false;
1927 btrfs_release_path(path);
1928 ret = insert_one_name(trans, root, key->objectid, key->offset,
1930 if (ret && ret != -ENOENT && ret != -EEXIST)
1934 update_size = false;
1938 if (!ret && update_size) {
1939 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2);
1940 ret = btrfs_update_inode(trans, BTRFS_I(dir));
1944 if (!ret && name_added)
1949 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1950 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1951 struct btrfs_root *root,
1952 struct btrfs_path *path,
1953 struct extent_buffer *eb, int slot,
1954 struct btrfs_key *key)
1957 struct btrfs_dir_item *di;
1959 /* We only log dir index keys, which only contain a single dir item. */
1960 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
1962 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1963 ret = replay_one_name(trans, root, path, eb, di, key);
1968 * If this entry refers to a non-directory (directories can not have a
1969 * link count > 1) and it was added in the transaction that was not
1970 * committed, make sure we fixup the link count of the inode the entry
1971 * points to. Otherwise something like the following would result in a
1972 * directory pointing to an inode with a wrong link that does not account
1973 * for this dir entry:
1980 * ln testdir/bar testdir/bar_link
1981 * ln testdir/foo testdir/foo_link
1982 * xfs_io -c "fsync" testdir/bar
1986 * mount fs, log replay happens
1988 * File foo would remain with a link count of 1 when it has two entries
1989 * pointing to it in the directory testdir. This would make it impossible
1990 * to ever delete the parent directory has it would result in stale
1991 * dentries that can never be deleted.
1993 if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) {
1994 struct btrfs_path *fixup_path;
1995 struct btrfs_key di_key;
1997 fixup_path = btrfs_alloc_path();
2001 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2002 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2003 btrfs_free_path(fixup_path);
2010 * directory replay has two parts. There are the standard directory
2011 * items in the log copied from the subvolume, and range items
2012 * created in the log while the subvolume was logged.
2014 * The range items tell us which parts of the key space the log
2015 * is authoritative for. During replay, if a key in the subvolume
2016 * directory is in a logged range item, but not actually in the log
2017 * that means it was deleted from the directory before the fsync
2018 * and should be removed.
2020 static noinline int find_dir_range(struct btrfs_root *root,
2021 struct btrfs_path *path,
2023 u64 *start_ret, u64 *end_ret)
2025 struct btrfs_key key;
2027 struct btrfs_dir_log_item *item;
2031 if (*start_ret == (u64)-1)
2034 key.objectid = dirid;
2035 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2036 key.offset = *start_ret;
2038 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2042 if (path->slots[0] == 0)
2047 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2049 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2053 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2054 struct btrfs_dir_log_item);
2055 found_end = btrfs_dir_log_end(path->nodes[0], item);
2057 if (*start_ret >= key.offset && *start_ret <= found_end) {
2059 *start_ret = key.offset;
2060 *end_ret = found_end;
2065 /* check the next slot in the tree to see if it is a valid item */
2066 nritems = btrfs_header_nritems(path->nodes[0]);
2068 if (path->slots[0] >= nritems) {
2069 ret = btrfs_next_leaf(root, path);
2074 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2076 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2080 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2081 struct btrfs_dir_log_item);
2082 found_end = btrfs_dir_log_end(path->nodes[0], item);
2083 *start_ret = key.offset;
2084 *end_ret = found_end;
2087 btrfs_release_path(path);
2092 * this looks for a given directory item in the log. If the directory
2093 * item is not in the log, the item is removed and the inode it points
2096 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *log,
2098 struct btrfs_path *path,
2099 struct btrfs_path *log_path,
2101 struct btrfs_key *dir_key)
2103 struct btrfs_root *root = BTRFS_I(dir)->root;
2105 struct extent_buffer *eb;
2107 struct btrfs_dir_item *di;
2108 struct fscrypt_str name;
2109 struct inode *inode = NULL;
2110 struct btrfs_key location;
2113 * Currently we only log dir index keys. Even if we replay a log created
2114 * by an older kernel that logged both dir index and dir item keys, all
2115 * we need to do is process the dir index keys, we (and our caller) can
2116 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2118 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2120 eb = path->nodes[0];
2121 slot = path->slots[0];
2122 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2123 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
2128 struct btrfs_dir_item *log_di;
2130 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2132 dir_key->offset, &name, 0);
2133 if (IS_ERR(log_di)) {
2134 ret = PTR_ERR(log_di);
2136 } else if (log_di) {
2137 /* The dentry exists in the log, we have nothing to do. */
2143 btrfs_dir_item_key_to_cpu(eb, di, &location);
2144 btrfs_release_path(path);
2145 btrfs_release_path(log_path);
2146 inode = read_one_inode(root, location.objectid);
2152 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2157 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2160 * Unlike dir item keys, dir index keys can only have one name (entry) in
2161 * them, as there are no key collisions since each key has a unique offset
2162 * (an index number), so we're done.
2165 btrfs_release_path(path);
2166 btrfs_release_path(log_path);
2172 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2173 struct btrfs_root *root,
2174 struct btrfs_root *log,
2175 struct btrfs_path *path,
2178 struct btrfs_key search_key;
2179 struct btrfs_path *log_path;
2184 log_path = btrfs_alloc_path();
2188 search_key.objectid = ino;
2189 search_key.type = BTRFS_XATTR_ITEM_KEY;
2190 search_key.offset = 0;
2192 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2196 nritems = btrfs_header_nritems(path->nodes[0]);
2197 for (i = path->slots[0]; i < nritems; i++) {
2198 struct btrfs_key key;
2199 struct btrfs_dir_item *di;
2200 struct btrfs_dir_item *log_di;
2204 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2205 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2210 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2211 total_size = btrfs_item_size(path->nodes[0], i);
2213 while (cur < total_size) {
2214 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2215 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2216 u32 this_len = sizeof(*di) + name_len + data_len;
2219 name = kmalloc(name_len, GFP_NOFS);
2224 read_extent_buffer(path->nodes[0], name,
2225 (unsigned long)(di + 1), name_len);
2227 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2229 btrfs_release_path(log_path);
2231 /* Doesn't exist in log tree, so delete it. */
2232 btrfs_release_path(path);
2233 di = btrfs_lookup_xattr(trans, root, path, ino,
2234 name, name_len, -1);
2241 ret = btrfs_delete_one_dir_name(trans, root,
2245 btrfs_release_path(path);
2250 if (IS_ERR(log_di)) {
2251 ret = PTR_ERR(log_di);
2255 di = (struct btrfs_dir_item *)((char *)di + this_len);
2258 ret = btrfs_next_leaf(root, path);
2264 btrfs_free_path(log_path);
2265 btrfs_release_path(path);
2271 * deletion replay happens before we copy any new directory items
2272 * out of the log or out of backreferences from inodes. It
2273 * scans the log to find ranges of keys that log is authoritative for,
2274 * and then scans the directory to find items in those ranges that are
2275 * not present in the log.
2277 * Anything we don't find in the log is unlinked and removed from the
2280 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2281 struct btrfs_root *root,
2282 struct btrfs_root *log,
2283 struct btrfs_path *path,
2284 u64 dirid, int del_all)
2289 struct btrfs_key dir_key;
2290 struct btrfs_key found_key;
2291 struct btrfs_path *log_path;
2294 dir_key.objectid = dirid;
2295 dir_key.type = BTRFS_DIR_INDEX_KEY;
2296 log_path = btrfs_alloc_path();
2300 dir = read_one_inode(root, dirid);
2301 /* it isn't an error if the inode isn't there, that can happen
2302 * because we replay the deletes before we copy in the inode item
2306 btrfs_free_path(log_path);
2314 range_end = (u64)-1;
2316 ret = find_dir_range(log, path, dirid,
2317 &range_start, &range_end);
2324 dir_key.offset = range_start;
2327 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2332 nritems = btrfs_header_nritems(path->nodes[0]);
2333 if (path->slots[0] >= nritems) {
2334 ret = btrfs_next_leaf(root, path);
2340 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2342 if (found_key.objectid != dirid ||
2343 found_key.type != dir_key.type) {
2348 if (found_key.offset > range_end)
2351 ret = check_item_in_log(trans, log, path,
2356 if (found_key.offset == (u64)-1)
2358 dir_key.offset = found_key.offset + 1;
2360 btrfs_release_path(path);
2361 if (range_end == (u64)-1)
2363 range_start = range_end + 1;
2367 btrfs_release_path(path);
2368 btrfs_free_path(log_path);
2374 * the process_func used to replay items from the log tree. This
2375 * gets called in two different stages. The first stage just looks
2376 * for inodes and makes sure they are all copied into the subvolume.
2378 * The second stage copies all the other item types from the log into
2379 * the subvolume. The two stage approach is slower, but gets rid of
2380 * lots of complexity around inodes referencing other inodes that exist
2381 * only in the log (references come from either directory items or inode
2384 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2385 struct walk_control *wc, u64 gen, int level)
2388 struct btrfs_tree_parent_check check = {
2392 struct btrfs_path *path;
2393 struct btrfs_root *root = wc->replay_dest;
2394 struct btrfs_key key;
2398 ret = btrfs_read_extent_buffer(eb, &check);
2402 level = btrfs_header_level(eb);
2407 path = btrfs_alloc_path();
2411 nritems = btrfs_header_nritems(eb);
2412 for (i = 0; i < nritems; i++) {
2413 btrfs_item_key_to_cpu(eb, &key, i);
2415 /* inode keys are done during the first stage */
2416 if (key.type == BTRFS_INODE_ITEM_KEY &&
2417 wc->stage == LOG_WALK_REPLAY_INODES) {
2418 struct btrfs_inode_item *inode_item;
2421 inode_item = btrfs_item_ptr(eb, i,
2422 struct btrfs_inode_item);
2424 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2425 * and never got linked before the fsync, skip it, as
2426 * replaying it is pointless since it would be deleted
2427 * later. We skip logging tmpfiles, but it's always
2428 * possible we are replaying a log created with a kernel
2429 * that used to log tmpfiles.
2431 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2432 wc->ignore_cur_inode = true;
2435 wc->ignore_cur_inode = false;
2437 ret = replay_xattr_deletes(wc->trans, root, log,
2438 path, key.objectid);
2441 mode = btrfs_inode_mode(eb, inode_item);
2442 if (S_ISDIR(mode)) {
2443 ret = replay_dir_deletes(wc->trans,
2444 root, log, path, key.objectid, 0);
2448 ret = overwrite_item(wc->trans, root, path,
2454 * Before replaying extents, truncate the inode to its
2455 * size. We need to do it now and not after log replay
2456 * because before an fsync we can have prealloc extents
2457 * added beyond the inode's i_size. If we did it after,
2458 * through orphan cleanup for example, we would drop
2459 * those prealloc extents just after replaying them.
2461 if (S_ISREG(mode)) {
2462 struct btrfs_drop_extents_args drop_args = { 0 };
2463 struct inode *inode;
2466 inode = read_one_inode(root, key.objectid);
2471 from = ALIGN(i_size_read(inode),
2472 root->fs_info->sectorsize);
2473 drop_args.start = from;
2474 drop_args.end = (u64)-1;
2475 drop_args.drop_cache = true;
2476 ret = btrfs_drop_extents(wc->trans, root,
2480 inode_sub_bytes(inode,
2481 drop_args.bytes_found);
2482 /* Update the inode's nbytes. */
2483 ret = btrfs_update_inode(wc->trans,
2491 ret = link_to_fixup_dir(wc->trans, root,
2492 path, key.objectid);
2497 if (wc->ignore_cur_inode)
2500 if (key.type == BTRFS_DIR_INDEX_KEY &&
2501 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2502 ret = replay_one_dir_item(wc->trans, root, path,
2508 if (wc->stage < LOG_WALK_REPLAY_ALL)
2511 /* these keys are simply copied */
2512 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2513 ret = overwrite_item(wc->trans, root, path,
2517 } else if (key.type == BTRFS_INODE_REF_KEY ||
2518 key.type == BTRFS_INODE_EXTREF_KEY) {
2519 ret = add_inode_ref(wc->trans, root, log, path,
2521 if (ret && ret != -ENOENT)
2524 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2525 ret = replay_one_extent(wc->trans, root, path,
2531 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2532 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2533 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2534 * older kernel with such keys, ignore them.
2537 btrfs_free_path(path);
2542 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2544 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2546 struct btrfs_block_group *cache;
2548 cache = btrfs_lookup_block_group(fs_info, start);
2550 btrfs_err(fs_info, "unable to find block group for %llu", start);
2554 spin_lock(&cache->space_info->lock);
2555 spin_lock(&cache->lock);
2556 cache->reserved -= fs_info->nodesize;
2557 cache->space_info->bytes_reserved -= fs_info->nodesize;
2558 spin_unlock(&cache->lock);
2559 spin_unlock(&cache->space_info->lock);
2561 btrfs_put_block_group(cache);
2564 static int clean_log_buffer(struct btrfs_trans_handle *trans,
2565 struct extent_buffer *eb)
2569 btrfs_tree_lock(eb);
2570 btrfs_clear_buffer_dirty(trans, eb);
2571 wait_on_extent_buffer_writeback(eb);
2572 btrfs_tree_unlock(eb);
2575 ret = btrfs_pin_reserved_extent(trans, eb);
2578 btrfs_redirty_list_add(trans->transaction, eb);
2580 unaccount_log_buffer(eb->fs_info, eb->start);
2586 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2587 struct btrfs_root *root,
2588 struct btrfs_path *path, int *level,
2589 struct walk_control *wc)
2591 struct btrfs_fs_info *fs_info = root->fs_info;
2594 struct extent_buffer *next;
2595 struct extent_buffer *cur;
2598 while (*level > 0) {
2599 struct btrfs_tree_parent_check check = { 0 };
2601 cur = path->nodes[*level];
2603 WARN_ON(btrfs_header_level(cur) != *level);
2605 if (path->slots[*level] >=
2606 btrfs_header_nritems(cur))
2609 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2610 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2611 check.transid = ptr_gen;
2612 check.level = *level - 1;
2613 check.has_first_key = true;
2614 btrfs_node_key_to_cpu(cur, &check.first_key, path->slots[*level]);
2616 next = btrfs_find_create_tree_block(fs_info, bytenr,
2617 btrfs_header_owner(cur),
2620 return PTR_ERR(next);
2623 ret = wc->process_func(root, next, wc, ptr_gen,
2626 free_extent_buffer(next);
2630 path->slots[*level]++;
2632 ret = btrfs_read_extent_buffer(next, &check);
2634 free_extent_buffer(next);
2638 ret = clean_log_buffer(trans, next);
2640 free_extent_buffer(next);
2644 free_extent_buffer(next);
2647 ret = btrfs_read_extent_buffer(next, &check);
2649 free_extent_buffer(next);
2653 if (path->nodes[*level-1])
2654 free_extent_buffer(path->nodes[*level-1]);
2655 path->nodes[*level-1] = next;
2656 *level = btrfs_header_level(next);
2657 path->slots[*level] = 0;
2660 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2666 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2667 struct btrfs_root *root,
2668 struct btrfs_path *path, int *level,
2669 struct walk_control *wc)
2675 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2676 slot = path->slots[i];
2677 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2680 WARN_ON(*level == 0);
2683 ret = wc->process_func(root, path->nodes[*level], wc,
2684 btrfs_header_generation(path->nodes[*level]),
2690 ret = clean_log_buffer(trans, path->nodes[*level]);
2694 free_extent_buffer(path->nodes[*level]);
2695 path->nodes[*level] = NULL;
2703 * drop the reference count on the tree rooted at 'snap'. This traverses
2704 * the tree freeing any blocks that have a ref count of zero after being
2707 static int walk_log_tree(struct btrfs_trans_handle *trans,
2708 struct btrfs_root *log, struct walk_control *wc)
2713 struct btrfs_path *path;
2716 path = btrfs_alloc_path();
2720 level = btrfs_header_level(log->node);
2722 path->nodes[level] = log->node;
2723 atomic_inc(&log->node->refs);
2724 path->slots[level] = 0;
2727 wret = walk_down_log_tree(trans, log, path, &level, wc);
2735 wret = walk_up_log_tree(trans, log, path, &level, wc);
2744 /* was the root node processed? if not, catch it here */
2745 if (path->nodes[orig_level]) {
2746 ret = wc->process_func(log, path->nodes[orig_level], wc,
2747 btrfs_header_generation(path->nodes[orig_level]),
2752 ret = clean_log_buffer(trans, path->nodes[orig_level]);
2756 btrfs_free_path(path);
2761 * helper function to update the item for a given subvolumes log root
2762 * in the tree of log roots
2764 static int update_log_root(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *log,
2766 struct btrfs_root_item *root_item)
2768 struct btrfs_fs_info *fs_info = log->fs_info;
2771 if (log->log_transid == 1) {
2772 /* insert root item on the first sync */
2773 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2774 &log->root_key, root_item);
2776 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2777 &log->root_key, root_item);
2782 static void wait_log_commit(struct btrfs_root *root, int transid)
2785 int index = transid % 2;
2788 * we only allow two pending log transactions at a time,
2789 * so we know that if ours is more than 2 older than the
2790 * current transaction, we're done
2793 prepare_to_wait(&root->log_commit_wait[index],
2794 &wait, TASK_UNINTERRUPTIBLE);
2796 if (!(root->log_transid_committed < transid &&
2797 atomic_read(&root->log_commit[index])))
2800 mutex_unlock(&root->log_mutex);
2802 mutex_lock(&root->log_mutex);
2804 finish_wait(&root->log_commit_wait[index], &wait);
2807 static void wait_for_writer(struct btrfs_root *root)
2812 prepare_to_wait(&root->log_writer_wait, &wait,
2813 TASK_UNINTERRUPTIBLE);
2814 if (!atomic_read(&root->log_writers))
2817 mutex_unlock(&root->log_mutex);
2819 mutex_lock(&root->log_mutex);
2821 finish_wait(&root->log_writer_wait, &wait);
2824 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2825 struct btrfs_log_ctx *ctx)
2827 mutex_lock(&root->log_mutex);
2828 list_del_init(&ctx->list);
2829 mutex_unlock(&root->log_mutex);
2833 * Invoked in log mutex context, or be sure there is no other task which
2834 * can access the list.
2836 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2837 int index, int error)
2839 struct btrfs_log_ctx *ctx;
2840 struct btrfs_log_ctx *safe;
2842 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2843 list_del_init(&ctx->list);
2844 ctx->log_ret = error;
2849 * Sends a given tree log down to the disk and updates the super blocks to
2850 * record it. When this call is done, you know that any inodes previously
2851 * logged are safely on disk only if it returns 0.
2853 * Any other return value means you need to call btrfs_commit_transaction.
2854 * Some of the edge cases for fsyncing directories that have had unlinks
2855 * or renames done in the past mean that sometimes the only safe
2856 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2857 * that has happened.
2859 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2860 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2866 struct btrfs_fs_info *fs_info = root->fs_info;
2867 struct btrfs_root *log = root->log_root;
2868 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2869 struct btrfs_root_item new_root_item;
2870 int log_transid = 0;
2871 struct btrfs_log_ctx root_log_ctx;
2872 struct blk_plug plug;
2876 mutex_lock(&root->log_mutex);
2877 log_transid = ctx->log_transid;
2878 if (root->log_transid_committed >= log_transid) {
2879 mutex_unlock(&root->log_mutex);
2880 return ctx->log_ret;
2883 index1 = log_transid % 2;
2884 if (atomic_read(&root->log_commit[index1])) {
2885 wait_log_commit(root, log_transid);
2886 mutex_unlock(&root->log_mutex);
2887 return ctx->log_ret;
2889 ASSERT(log_transid == root->log_transid);
2890 atomic_set(&root->log_commit[index1], 1);
2892 /* wait for previous tree log sync to complete */
2893 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2894 wait_log_commit(root, log_transid - 1);
2897 int batch = atomic_read(&root->log_batch);
2898 /* when we're on an ssd, just kick the log commit out */
2899 if (!btrfs_test_opt(fs_info, SSD) &&
2900 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2901 mutex_unlock(&root->log_mutex);
2902 schedule_timeout_uninterruptible(1);
2903 mutex_lock(&root->log_mutex);
2905 wait_for_writer(root);
2906 if (batch == atomic_read(&root->log_batch))
2910 /* bail out if we need to do a full commit */
2911 if (btrfs_need_log_full_commit(trans)) {
2912 ret = BTRFS_LOG_FORCE_COMMIT;
2913 mutex_unlock(&root->log_mutex);
2917 if (log_transid % 2 == 0)
2918 mark = EXTENT_DIRTY;
2922 /* we start IO on all the marked extents here, but we don't actually
2923 * wait for them until later.
2925 blk_start_plug(&plug);
2926 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2928 * -EAGAIN happens when someone, e.g., a concurrent transaction
2929 * commit, writes a dirty extent in this tree-log commit. This
2930 * concurrent write will create a hole writing out the extents,
2931 * and we cannot proceed on a zoned filesystem, requiring
2932 * sequential writing. While we can bail out to a full commit
2933 * here, but we can continue hoping the concurrent writing fills
2936 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
2939 blk_finish_plug(&plug);
2940 btrfs_set_log_full_commit(trans);
2941 mutex_unlock(&root->log_mutex);
2946 * We _must_ update under the root->log_mutex in order to make sure we
2947 * have a consistent view of the log root we are trying to commit at
2950 * We _must_ copy this into a local copy, because we are not holding the
2951 * log_root_tree->log_mutex yet. This is important because when we
2952 * commit the log_root_tree we must have a consistent view of the
2953 * log_root_tree when we update the super block to point at the
2954 * log_root_tree bytenr. If we update the log_root_tree here we'll race
2955 * with the commit and possibly point at the new block which we may not
2958 btrfs_set_root_node(&log->root_item, log->node);
2959 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
2961 btrfs_set_root_log_transid(root, root->log_transid + 1);
2962 log->log_transid = root->log_transid;
2963 root->log_start_pid = 0;
2965 * IO has been started, blocks of the log tree have WRITTEN flag set
2966 * in their headers. new modifications of the log will be written to
2967 * new positions. so it's safe to allow log writers to go in.
2969 mutex_unlock(&root->log_mutex);
2971 if (btrfs_is_zoned(fs_info)) {
2972 mutex_lock(&fs_info->tree_root->log_mutex);
2973 if (!log_root_tree->node) {
2974 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
2976 mutex_unlock(&fs_info->tree_root->log_mutex);
2977 blk_finish_plug(&plug);
2981 mutex_unlock(&fs_info->tree_root->log_mutex);
2984 btrfs_init_log_ctx(&root_log_ctx, NULL);
2986 mutex_lock(&log_root_tree->log_mutex);
2988 index2 = log_root_tree->log_transid % 2;
2989 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2990 root_log_ctx.log_transid = log_root_tree->log_transid;
2993 * Now we are safe to update the log_root_tree because we're under the
2994 * log_mutex, and we're a current writer so we're holding the commit
2995 * open until we drop the log_mutex.
2997 ret = update_log_root(trans, log, &new_root_item);
2999 list_del_init(&root_log_ctx.list);
3000 blk_finish_plug(&plug);
3001 btrfs_set_log_full_commit(trans);
3004 "failed to update log for root %llu ret %d",
3005 root->root_key.objectid, ret);
3006 btrfs_wait_tree_log_extents(log, mark);
3007 mutex_unlock(&log_root_tree->log_mutex);
3011 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3012 blk_finish_plug(&plug);
3013 list_del_init(&root_log_ctx.list);
3014 mutex_unlock(&log_root_tree->log_mutex);
3015 ret = root_log_ctx.log_ret;
3019 if (atomic_read(&log_root_tree->log_commit[index2])) {
3020 blk_finish_plug(&plug);
3021 ret = btrfs_wait_tree_log_extents(log, mark);
3022 wait_log_commit(log_root_tree,
3023 root_log_ctx.log_transid);
3024 mutex_unlock(&log_root_tree->log_mutex);
3026 ret = root_log_ctx.log_ret;
3029 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3030 atomic_set(&log_root_tree->log_commit[index2], 1);
3032 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3033 wait_log_commit(log_root_tree,
3034 root_log_ctx.log_transid - 1);
3038 * now that we've moved on to the tree of log tree roots,
3039 * check the full commit flag again
3041 if (btrfs_need_log_full_commit(trans)) {
3042 blk_finish_plug(&plug);
3043 btrfs_wait_tree_log_extents(log, mark);
3044 mutex_unlock(&log_root_tree->log_mutex);
3045 ret = BTRFS_LOG_FORCE_COMMIT;
3046 goto out_wake_log_root;
3049 ret = btrfs_write_marked_extents(fs_info,
3050 &log_root_tree->dirty_log_pages,
3051 EXTENT_DIRTY | EXTENT_NEW);
3052 blk_finish_plug(&plug);
3054 * As described above, -EAGAIN indicates a hole in the extents. We
3055 * cannot wait for these write outs since the waiting cause a
3056 * deadlock. Bail out to the full commit instead.
3058 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3059 btrfs_set_log_full_commit(trans);
3060 btrfs_wait_tree_log_extents(log, mark);
3061 mutex_unlock(&log_root_tree->log_mutex);
3062 goto out_wake_log_root;
3064 btrfs_set_log_full_commit(trans);
3065 mutex_unlock(&log_root_tree->log_mutex);
3066 goto out_wake_log_root;
3068 ret = btrfs_wait_tree_log_extents(log, mark);
3070 ret = btrfs_wait_tree_log_extents(log_root_tree,
3071 EXTENT_NEW | EXTENT_DIRTY);
3073 btrfs_set_log_full_commit(trans);
3074 mutex_unlock(&log_root_tree->log_mutex);
3075 goto out_wake_log_root;
3078 log_root_start = log_root_tree->node->start;
3079 log_root_level = btrfs_header_level(log_root_tree->node);
3080 log_root_tree->log_transid++;
3081 mutex_unlock(&log_root_tree->log_mutex);
3084 * Here we are guaranteed that nobody is going to write the superblock
3085 * for the current transaction before us and that neither we do write
3086 * our superblock before the previous transaction finishes its commit
3087 * and writes its superblock, because:
3089 * 1) We are holding a handle on the current transaction, so no body
3090 * can commit it until we release the handle;
3092 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3093 * if the previous transaction is still committing, and hasn't yet
3094 * written its superblock, we wait for it to do it, because a
3095 * transaction commit acquires the tree_log_mutex when the commit
3096 * begins and releases it only after writing its superblock.
3098 mutex_lock(&fs_info->tree_log_mutex);
3101 * The previous transaction writeout phase could have failed, and thus
3102 * marked the fs in an error state. We must not commit here, as we
3103 * could have updated our generation in the super_for_commit and
3104 * writing the super here would result in transid mismatches. If there
3105 * is an error here just bail.
3107 if (BTRFS_FS_ERROR(fs_info)) {
3109 btrfs_set_log_full_commit(trans);
3110 btrfs_abort_transaction(trans, ret);
3111 mutex_unlock(&fs_info->tree_log_mutex);
3112 goto out_wake_log_root;
3115 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3116 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3117 ret = write_all_supers(fs_info, 1);
3118 mutex_unlock(&fs_info->tree_log_mutex);
3120 btrfs_set_log_full_commit(trans);
3121 btrfs_abort_transaction(trans, ret);
3122 goto out_wake_log_root;
3126 * We know there can only be one task here, since we have not yet set
3127 * root->log_commit[index1] to 0 and any task attempting to sync the
3128 * log must wait for the previous log transaction to commit if it's
3129 * still in progress or wait for the current log transaction commit if
3130 * someone else already started it. We use <= and not < because the
3131 * first log transaction has an ID of 0.
3133 ASSERT(btrfs_get_root_last_log_commit(root) <= log_transid);
3134 btrfs_set_root_last_log_commit(root, log_transid);
3137 mutex_lock(&log_root_tree->log_mutex);
3138 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3140 log_root_tree->log_transid_committed++;
3141 atomic_set(&log_root_tree->log_commit[index2], 0);
3142 mutex_unlock(&log_root_tree->log_mutex);
3145 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3146 * all the updates above are seen by the woken threads. It might not be
3147 * necessary, but proving that seems to be hard.
3149 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3151 mutex_lock(&root->log_mutex);
3152 btrfs_remove_all_log_ctxs(root, index1, ret);
3153 root->log_transid_committed++;
3154 atomic_set(&root->log_commit[index1], 0);
3155 mutex_unlock(&root->log_mutex);
3158 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3159 * all the updates above are seen by the woken threads. It might not be
3160 * necessary, but proving that seems to be hard.
3162 cond_wake_up(&root->log_commit_wait[index1]);
3166 static void free_log_tree(struct btrfs_trans_handle *trans,
3167 struct btrfs_root *log)
3170 struct walk_control wc = {
3172 .process_func = process_one_buffer
3176 ret = walk_log_tree(trans, log, &wc);
3179 * We weren't able to traverse the entire log tree, the
3180 * typical scenario is getting an -EIO when reading an
3181 * extent buffer of the tree, due to a previous writeback
3184 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3185 &log->fs_info->fs_state);
3188 * Some extent buffers of the log tree may still be dirty
3189 * and not yet written back to storage, because we may
3190 * have updates to a log tree without syncing a log tree,
3191 * such as during rename and link operations. So flush
3192 * them out and wait for their writeback to complete, so
3193 * that we properly cleanup their state and pages.
3195 btrfs_write_marked_extents(log->fs_info,
3196 &log->dirty_log_pages,
3197 EXTENT_DIRTY | EXTENT_NEW);
3198 btrfs_wait_tree_log_extents(log,
3199 EXTENT_DIRTY | EXTENT_NEW);
3202 btrfs_abort_transaction(trans, ret);
3204 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3208 extent_io_tree_release(&log->dirty_log_pages);
3209 extent_io_tree_release(&log->log_csum_range);
3211 btrfs_put_root(log);
3215 * free all the extents used by the tree log. This should be called
3216 * at commit time of the full transaction
3218 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3220 if (root->log_root) {
3221 free_log_tree(trans, root->log_root);
3222 root->log_root = NULL;
3223 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3228 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3229 struct btrfs_fs_info *fs_info)
3231 if (fs_info->log_root_tree) {
3232 free_log_tree(trans, fs_info->log_root_tree);
3233 fs_info->log_root_tree = NULL;
3234 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3240 * Check if an inode was logged in the current transaction. This correctly deals
3241 * with the case where the inode was logged but has a logged_trans of 0, which
3242 * happens if the inode is evicted and loaded again, as logged_trans is an in
3243 * memory only field (not persisted).
3245 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3248 static int inode_logged(const struct btrfs_trans_handle *trans,
3249 struct btrfs_inode *inode,
3250 struct btrfs_path *path_in)
3252 struct btrfs_path *path = path_in;
3253 struct btrfs_key key;
3256 if (inode->logged_trans == trans->transid)
3260 * If logged_trans is not 0, then we know the inode logged was not logged
3261 * in this transaction, so we can return false right away.
3263 if (inode->logged_trans > 0)
3267 * If no log tree was created for this root in this transaction, then
3268 * the inode can not have been logged in this transaction. In that case
3269 * set logged_trans to anything greater than 0 and less than the current
3270 * transaction's ID, to avoid the search below in a future call in case
3271 * a log tree gets created after this.
3273 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3274 inode->logged_trans = trans->transid - 1;
3279 * We have a log tree and the inode's logged_trans is 0. We can't tell
3280 * for sure if the inode was logged before in this transaction by looking
3281 * only at logged_trans. We could be pessimistic and assume it was, but
3282 * that can lead to unnecessarily logging an inode during rename and link
3283 * operations, and then further updating the log in followup rename and
3284 * link operations, specially if it's a directory, which adds latency
3285 * visible to applications doing a series of rename or link operations.
3287 * A logged_trans of 0 here can mean several things:
3289 * 1) The inode was never logged since the filesystem was mounted, and may
3290 * or may have not been evicted and loaded again;
3292 * 2) The inode was logged in a previous transaction, then evicted and
3293 * then loaded again;
3295 * 3) The inode was logged in the current transaction, then evicted and
3296 * then loaded again.
3298 * For cases 1) and 2) we don't want to return true, but we need to detect
3299 * case 3) and return true. So we do a search in the log root for the inode
3302 key.objectid = btrfs_ino(inode);
3303 key.type = BTRFS_INODE_ITEM_KEY;
3307 path = btrfs_alloc_path();
3312 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3315 btrfs_release_path(path);
3317 btrfs_free_path(path);
3320 * Logging an inode always results in logging its inode item. So if we
3321 * did not find the item we know the inode was not logged for sure.
3325 } else if (ret > 0) {
3327 * Set logged_trans to a value greater than 0 and less then the
3328 * current transaction to avoid doing the search in future calls.
3330 inode->logged_trans = trans->transid - 1;
3335 * The inode was previously logged and then evicted, set logged_trans to
3336 * the current transacion's ID, to avoid future tree searches as long as
3337 * the inode is not evicted again.
3339 inode->logged_trans = trans->transid;
3342 * If it's a directory, then we must set last_dir_index_offset to the
3343 * maximum possible value, so that the next attempt to log the inode does
3344 * not skip checking if dir index keys found in modified subvolume tree
3345 * leaves have been logged before, otherwise it would result in attempts
3346 * to insert duplicate dir index keys in the log tree. This must be done
3347 * because last_dir_index_offset is an in-memory only field, not persisted
3348 * in the inode item or any other on-disk structure, so its value is lost
3349 * once the inode is evicted.
3351 if (S_ISDIR(inode->vfs_inode.i_mode))
3352 inode->last_dir_index_offset = (u64)-1;
3358 * Delete a directory entry from the log if it exists.
3360 * Returns < 0 on error
3361 * 1 if the entry does not exists
3362 * 0 if the entry existed and was successfully deleted
3364 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3365 struct btrfs_root *log,
3366 struct btrfs_path *path,
3368 const struct fscrypt_str *name,
3371 struct btrfs_dir_item *di;
3374 * We only log dir index items of a directory, so we don't need to look
3375 * for dir item keys.
3377 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3385 * We do not need to update the size field of the directory's
3386 * inode item because on log replay we update the field to reflect
3387 * all existing entries in the directory (see overwrite_item()).
3389 return btrfs_delete_one_dir_name(trans, log, path, di);
3393 * If both a file and directory are logged, and unlinks or renames are
3394 * mixed in, we have a few interesting corners:
3396 * create file X in dir Y
3397 * link file X to X.link in dir Y
3399 * unlink file X but leave X.link
3402 * After a crash we would expect only X.link to exist. But file X
3403 * didn't get fsync'd again so the log has back refs for X and X.link.
3405 * We solve this by removing directory entries and inode backrefs from the
3406 * log when a file that was logged in the current transaction is
3407 * unlinked. Any later fsync will include the updated log entries, and
3408 * we'll be able to reconstruct the proper directory items from backrefs.
3410 * This optimizations allows us to avoid relogging the entire inode
3411 * or the entire directory.
3413 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3414 struct btrfs_root *root,
3415 const struct fscrypt_str *name,
3416 struct btrfs_inode *dir, u64 index)
3418 struct btrfs_path *path;
3421 ret = inode_logged(trans, dir, NULL);
3425 btrfs_set_log_full_commit(trans);
3429 ret = join_running_log_trans(root);
3433 mutex_lock(&dir->log_mutex);
3435 path = btrfs_alloc_path();
3441 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3443 btrfs_free_path(path);
3445 mutex_unlock(&dir->log_mutex);
3447 btrfs_set_log_full_commit(trans);
3448 btrfs_end_log_trans(root);
3451 /* see comments for btrfs_del_dir_entries_in_log */
3452 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3453 struct btrfs_root *root,
3454 const struct fscrypt_str *name,
3455 struct btrfs_inode *inode, u64 dirid)
3457 struct btrfs_root *log;
3461 ret = inode_logged(trans, inode, NULL);
3465 btrfs_set_log_full_commit(trans);
3469 ret = join_running_log_trans(root);
3472 log = root->log_root;
3473 mutex_lock(&inode->log_mutex);
3475 ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode),
3477 mutex_unlock(&inode->log_mutex);
3478 if (ret < 0 && ret != -ENOENT)
3479 btrfs_set_log_full_commit(trans);
3480 btrfs_end_log_trans(root);
3484 * creates a range item in the log for 'dirid'. first_offset and
3485 * last_offset tell us which parts of the key space the log should
3486 * be considered authoritative for.
3488 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *log,
3490 struct btrfs_path *path,
3492 u64 first_offset, u64 last_offset)
3495 struct btrfs_key key;
3496 struct btrfs_dir_log_item *item;
3498 key.objectid = dirid;
3499 key.offset = first_offset;
3500 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3501 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3503 * -EEXIST is fine and can happen sporadically when we are logging a
3504 * directory and have concurrent insertions in the subvolume's tree for
3505 * items from other inodes and that result in pushing off some dir items
3506 * from one leaf to another in order to accommodate for the new items.
3507 * This results in logging the same dir index range key.
3509 if (ret && ret != -EEXIST)
3512 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3513 struct btrfs_dir_log_item);
3514 if (ret == -EEXIST) {
3515 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3518 * btrfs_del_dir_entries_in_log() might have been called during
3519 * an unlink between the initial insertion of this key and the
3520 * current update, or we might be logging a single entry deletion
3521 * during a rename, so set the new last_offset to the max value.
3523 last_offset = max(last_offset, curr_end);
3525 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3526 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
3527 btrfs_release_path(path);
3531 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3532 struct btrfs_inode *inode,
3533 struct extent_buffer *src,
3534 struct btrfs_path *dst_path,
3538 struct btrfs_root *log = inode->root->log_root;
3539 char *ins_data = NULL;
3540 struct btrfs_item_batch batch;
3541 struct extent_buffer *dst;
3542 unsigned long src_offset;
3543 unsigned long dst_offset;
3545 struct btrfs_key key;
3554 btrfs_item_key_to_cpu(src, &key, start_slot);
3555 item_size = btrfs_item_size(src, start_slot);
3557 batch.data_sizes = &item_size;
3558 batch.total_data_size = item_size;
3560 struct btrfs_key *ins_keys;
3563 ins_data = kmalloc(count * sizeof(u32) +
3564 count * sizeof(struct btrfs_key), GFP_NOFS);
3568 ins_sizes = (u32 *)ins_data;
3569 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3570 batch.keys = ins_keys;
3571 batch.data_sizes = ins_sizes;
3572 batch.total_data_size = 0;
3574 for (i = 0; i < count; i++) {
3575 const int slot = start_slot + i;
3577 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3578 ins_sizes[i] = btrfs_item_size(src, slot);
3579 batch.total_data_size += ins_sizes[i];
3583 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3587 dst = dst_path->nodes[0];
3589 * Copy all the items in bulk, in a single copy operation. Item data is
3590 * organized such that it's placed at the end of a leaf and from right
3591 * to left. For example, the data for the second item ends at an offset
3592 * that matches the offset where the data for the first item starts, the
3593 * data for the third item ends at an offset that matches the offset
3594 * where the data of the second items starts, and so on.
3595 * Therefore our source and destination start offsets for copy match the
3596 * offsets of the last items (highest slots).
3598 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3599 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3600 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3601 btrfs_release_path(dst_path);
3603 last_index = batch.keys[count - 1].offset;
3604 ASSERT(last_index > inode->last_dir_index_offset);
3607 * If for some unexpected reason the last item's index is not greater
3608 * than the last index we logged, warn and force a transaction commit.
3610 if (WARN_ON(last_index <= inode->last_dir_index_offset))
3611 ret = BTRFS_LOG_FORCE_COMMIT;
3613 inode->last_dir_index_offset = last_index;
3615 if (btrfs_get_first_dir_index_to_log(inode) == 0)
3616 btrfs_set_first_dir_index_to_log(inode, batch.keys[0].offset);
3623 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3624 struct btrfs_inode *inode,
3625 struct btrfs_path *path,
3626 struct btrfs_path *dst_path,
3627 struct btrfs_log_ctx *ctx,
3628 u64 *last_old_dentry_offset)
3630 struct btrfs_root *log = inode->root->log_root;
3631 struct extent_buffer *src;
3632 const int nritems = btrfs_header_nritems(path->nodes[0]);
3633 const u64 ino = btrfs_ino(inode);
3634 bool last_found = false;
3635 int batch_start = 0;
3640 * We need to clone the leaf, release the read lock on it, and use the
3641 * clone before modifying the log tree. See the comment at copy_items()
3642 * about why we need to do this.
3644 src = btrfs_clone_extent_buffer(path->nodes[0]);
3649 btrfs_release_path(path);
3650 path->nodes[0] = src;
3653 for (; i < nritems; i++) {
3654 struct btrfs_dir_item *di;
3655 struct btrfs_key key;
3658 btrfs_item_key_to_cpu(src, &key, i);
3660 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3665 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3668 * Skip ranges of items that consist only of dir item keys created
3669 * in past transactions. However if we find a gap, we must log a
3670 * dir index range item for that gap, so that index keys in that
3671 * gap are deleted during log replay.
3673 if (btrfs_dir_transid(src, di) < trans->transid) {
3674 if (key.offset > *last_old_dentry_offset + 1) {
3675 ret = insert_dir_log_key(trans, log, dst_path,
3676 ino, *last_old_dentry_offset + 1,
3682 *last_old_dentry_offset = key.offset;
3686 /* If we logged this dir index item before, we can skip it. */
3687 if (key.offset <= inode->last_dir_index_offset)
3691 * We must make sure that when we log a directory entry, the
3692 * corresponding inode, after log replay, has a matching link
3693 * count. For example:
3699 * xfs_io -c "fsync" mydir
3701 * <mount fs and log replay>
3703 * Would result in a fsync log that when replayed, our file inode
3704 * would have a link count of 1, but we get two directory entries
3705 * pointing to the same inode. After removing one of the names,
3706 * it would not be possible to remove the other name, which
3707 * resulted always in stale file handle errors, and would not be
3708 * possible to rmdir the parent directory, since its i_size could
3709 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3710 * resulting in -ENOTEMPTY errors.
3712 if (!ctx->log_new_dentries) {
3713 struct btrfs_key di_key;
3715 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3716 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3717 ctx->log_new_dentries = true;
3720 if (batch_size == 0)
3725 if (batch_size > 0) {
3728 ret = flush_dir_items_batch(trans, inode, src, dst_path,
3729 batch_start, batch_size);
3734 return last_found ? 1 : 0;
3738 * log all the items included in the current transaction for a given
3739 * directory. This also creates the range items in the log tree required
3740 * to replay anything deleted before the fsync
3742 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3743 struct btrfs_inode *inode,
3744 struct btrfs_path *path,
3745 struct btrfs_path *dst_path,
3746 struct btrfs_log_ctx *ctx,
3747 u64 min_offset, u64 *last_offset_ret)
3749 struct btrfs_key min_key;
3750 struct btrfs_root *root = inode->root;
3751 struct btrfs_root *log = root->log_root;
3753 u64 last_old_dentry_offset = min_offset - 1;
3754 u64 last_offset = (u64)-1;
3755 u64 ino = btrfs_ino(inode);
3757 min_key.objectid = ino;
3758 min_key.type = BTRFS_DIR_INDEX_KEY;
3759 min_key.offset = min_offset;
3761 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3764 * we didn't find anything from this transaction, see if there
3765 * is anything at all
3767 if (ret != 0 || min_key.objectid != ino ||
3768 min_key.type != BTRFS_DIR_INDEX_KEY) {
3769 min_key.objectid = ino;
3770 min_key.type = BTRFS_DIR_INDEX_KEY;
3771 min_key.offset = (u64)-1;
3772 btrfs_release_path(path);
3773 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3775 btrfs_release_path(path);
3778 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3780 /* if ret == 0 there are items for this type,
3781 * create a range to tell us the last key of this type.
3782 * otherwise, there are no items in this directory after
3783 * *min_offset, and we create a range to indicate that.
3786 struct btrfs_key tmp;
3788 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3790 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3791 last_old_dentry_offset = tmp.offset;
3792 } else if (ret > 0) {
3799 /* go backward to find any previous key */
3800 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3802 struct btrfs_key tmp;
3804 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3806 * The dir index key before the first one we found that needs to
3807 * be logged might be in a previous leaf, and there might be a
3808 * gap between these keys, meaning that we had deletions that
3809 * happened. So the key range item we log (key type
3810 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3811 * previous key's offset plus 1, so that those deletes are replayed.
3813 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3814 last_old_dentry_offset = tmp.offset;
3815 } else if (ret < 0) {
3819 btrfs_release_path(path);
3822 * Find the first key from this transaction again or the one we were at
3823 * in the loop below in case we had to reschedule. We may be logging the
3824 * directory without holding its VFS lock, which happen when logging new
3825 * dentries (through log_new_dir_dentries()) or in some cases when we
3826 * need to log the parent directory of an inode. This means a dir index
3827 * key might be deleted from the inode's root, and therefore we may not
3828 * find it anymore. If we can't find it, just move to the next key. We
3829 * can not bail out and ignore, because if we do that we will simply
3830 * not log dir index keys that come after the one that was just deleted
3831 * and we can end up logging a dir index range that ends at (u64)-1
3832 * (@last_offset is initialized to that), resulting in removing dir
3833 * entries we should not remove at log replay time.
3836 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3838 ret = btrfs_next_item(root, path);
3840 /* There are no more keys in the inode's root. */
3849 * we have a block from this transaction, log every item in it
3850 * from our directory
3853 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3854 &last_old_dentry_offset);
3860 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3863 * look ahead to the next item and see if it is also
3864 * from this directory and from this transaction
3866 ret = btrfs_next_leaf(root, path);
3869 last_offset = (u64)-1;
3874 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3875 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3876 last_offset = (u64)-1;
3879 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3881 * The next leaf was not changed in the current transaction
3882 * and has at least one dir index key.
3883 * We check for the next key because there might have been
3884 * one or more deletions between the last key we logged and
3885 * that next key. So the key range item we log (key type
3886 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3887 * offset minus 1, so that those deletes are replayed.
3889 last_offset = min_key.offset - 1;
3892 if (need_resched()) {
3893 btrfs_release_path(path);
3899 btrfs_release_path(path);
3900 btrfs_release_path(dst_path);
3903 *last_offset_ret = last_offset;
3905 * In case the leaf was changed in the current transaction but
3906 * all its dir items are from a past transaction, the last item
3907 * in the leaf is a dir item and there's no gap between that last
3908 * dir item and the first one on the next leaf (which did not
3909 * change in the current transaction), then we don't need to log
3910 * a range, last_old_dentry_offset is == to last_offset.
3912 ASSERT(last_old_dentry_offset <= last_offset);
3913 if (last_old_dentry_offset < last_offset)
3914 ret = insert_dir_log_key(trans, log, path, ino,
3915 last_old_dentry_offset + 1,
3923 * If the inode was logged before and it was evicted, then its
3924 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3925 * key offset. If that's the case, search for it and update the inode. This
3926 * is to avoid lookups in the log tree every time we try to insert a dir index
3927 * key from a leaf changed in the current transaction, and to allow us to always
3928 * do batch insertions of dir index keys.
3930 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3931 struct btrfs_path *path,
3932 const struct btrfs_log_ctx *ctx)
3934 const u64 ino = btrfs_ino(inode);
3935 struct btrfs_key key;
3938 lockdep_assert_held(&inode->log_mutex);
3940 if (inode->last_dir_index_offset != (u64)-1)
3943 if (!ctx->logged_before) {
3944 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3949 key.type = BTRFS_DIR_INDEX_KEY;
3950 key.offset = (u64)-1;
3952 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3954 * An error happened or we actually have an index key with an offset
3955 * value of (u64)-1. Bail out, we're done.
3961 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3964 * No dir index items, bail out and leave last_dir_index_offset with
3965 * the value right before the first valid index value.
3967 if (path->slots[0] == 0)
3971 * btrfs_search_slot() left us at one slot beyond the slot with the last
3972 * index key, or beyond the last key of the directory that is not an
3973 * index key. If we have an index key before, set last_dir_index_offset
3974 * to its offset value, otherwise leave it with a value right before the
3975 * first valid index value, as it means we have an empty directory.
3977 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3978 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
3979 inode->last_dir_index_offset = key.offset;
3982 btrfs_release_path(path);
3988 * logging directories is very similar to logging inodes, We find all the items
3989 * from the current transaction and write them to the log.
3991 * The recovery code scans the directory in the subvolume, and if it finds a
3992 * key in the range logged that is not present in the log tree, then it means
3993 * that dir entry was unlinked during the transaction.
3995 * In order for that scan to work, we must include one key smaller than
3996 * the smallest logged by this transaction and one key larger than the largest
3997 * key logged by this transaction.
3999 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4000 struct btrfs_inode *inode,
4001 struct btrfs_path *path,
4002 struct btrfs_path *dst_path,
4003 struct btrfs_log_ctx *ctx)
4009 ret = update_last_dir_index_offset(inode, path, ctx);
4013 min_key = BTRFS_DIR_START_INDEX;
4017 ret = log_dir_items(trans, inode, path, dst_path,
4018 ctx, min_key, &max_key);
4021 if (max_key == (u64)-1)
4023 min_key = max_key + 1;
4030 * a helper function to drop items from the log before we relog an
4031 * inode. max_key_type indicates the highest item type to remove.
4032 * This cannot be run for file data extents because it does not
4033 * free the extents they point to.
4035 static int drop_inode_items(struct btrfs_trans_handle *trans,
4036 struct btrfs_root *log,
4037 struct btrfs_path *path,
4038 struct btrfs_inode *inode,
4042 struct btrfs_key key;
4043 struct btrfs_key found_key;
4046 key.objectid = btrfs_ino(inode);
4047 key.type = max_key_type;
4048 key.offset = (u64)-1;
4051 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4054 } else if (ret > 0) {
4055 if (path->slots[0] == 0)
4060 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4063 if (found_key.objectid != key.objectid)
4066 found_key.offset = 0;
4068 ret = btrfs_bin_search(path->nodes[0], 0, &found_key, &start_slot);
4072 ret = btrfs_del_items(trans, log, path, start_slot,
4073 path->slots[0] - start_slot + 1);
4075 * If start slot isn't 0 then we don't need to re-search, we've
4076 * found the last guy with the objectid in this tree.
4078 if (ret || start_slot != 0)
4080 btrfs_release_path(path);
4082 btrfs_release_path(path);
4088 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4089 struct btrfs_root *log_root,
4090 struct btrfs_inode *inode,
4091 u64 new_size, u32 min_type)
4093 struct btrfs_truncate_control control = {
4094 .new_size = new_size,
4095 .ino = btrfs_ino(inode),
4096 .min_type = min_type,
4097 .skip_ref_updates = true,
4100 return btrfs_truncate_inode_items(trans, log_root, &control);
4103 static void fill_inode_item(struct btrfs_trans_handle *trans,
4104 struct extent_buffer *leaf,
4105 struct btrfs_inode_item *item,
4106 struct inode *inode, int log_inode_only,
4109 struct btrfs_map_token token;
4112 btrfs_init_map_token(&token, leaf);
4114 if (log_inode_only) {
4115 /* set the generation to zero so the recover code
4116 * can tell the difference between an logging
4117 * just to say 'this inode exists' and a logging
4118 * to say 'update this inode with these values'
4120 btrfs_set_token_inode_generation(&token, item, 0);
4121 btrfs_set_token_inode_size(&token, item, logged_isize);
4123 btrfs_set_token_inode_generation(&token, item,
4124 BTRFS_I(inode)->generation);
4125 btrfs_set_token_inode_size(&token, item, inode->i_size);
4128 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4129 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4130 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4131 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4133 btrfs_set_token_timespec_sec(&token, &item->atime,
4134 inode_get_atime_sec(inode));
4135 btrfs_set_token_timespec_nsec(&token, &item->atime,
4136 inode_get_atime_nsec(inode));
4138 btrfs_set_token_timespec_sec(&token, &item->mtime,
4139 inode_get_mtime_sec(inode));
4140 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4141 inode_get_mtime_nsec(inode));
4143 btrfs_set_token_timespec_sec(&token, &item->ctime,
4144 inode_get_ctime_sec(inode));
4145 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4146 inode_get_ctime_nsec(inode));
4149 * We do not need to set the nbytes field, in fact during a fast fsync
4150 * its value may not even be correct, since a fast fsync does not wait
4151 * for ordered extent completion, which is where we update nbytes, it
4152 * only waits for writeback to complete. During log replay as we find
4153 * file extent items and replay them, we adjust the nbytes field of the
4154 * inode item in subvolume tree as needed (see overwrite_item()).
4157 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4158 btrfs_set_token_inode_transid(&token, item, trans->transid);
4159 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4160 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4161 BTRFS_I(inode)->ro_flags);
4162 btrfs_set_token_inode_flags(&token, item, flags);
4163 btrfs_set_token_inode_block_group(&token, item, 0);
4166 static int log_inode_item(struct btrfs_trans_handle *trans,
4167 struct btrfs_root *log, struct btrfs_path *path,
4168 struct btrfs_inode *inode, bool inode_item_dropped)
4170 struct btrfs_inode_item *inode_item;
4174 * If we are doing a fast fsync and the inode was logged before in the
4175 * current transaction, then we know the inode was previously logged and
4176 * it exists in the log tree. For performance reasons, in this case use
4177 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4178 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4179 * contention in case there are concurrent fsyncs for other inodes of the
4180 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4181 * already exists can also result in unnecessarily splitting a leaf.
4183 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4184 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4190 * This means it is the first fsync in the current transaction,
4191 * so the inode item is not in the log and we need to insert it.
4192 * We can never get -EEXIST because we are only called for a fast
4193 * fsync and in case an inode eviction happens after the inode was
4194 * logged before in the current transaction, when we load again
4195 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4196 * flags and set ->logged_trans to 0.
4198 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4199 sizeof(*inode_item));
4200 ASSERT(ret != -EEXIST);
4204 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4205 struct btrfs_inode_item);
4206 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4208 btrfs_release_path(path);
4212 static int log_csums(struct btrfs_trans_handle *trans,
4213 struct btrfs_inode *inode,
4214 struct btrfs_root *log_root,
4215 struct btrfs_ordered_sum *sums)
4217 const u64 lock_end = sums->logical + sums->len - 1;
4218 struct extent_state *cached_state = NULL;
4222 * If this inode was not used for reflink operations in the current
4223 * transaction with new extents, then do the fast path, no need to
4224 * worry about logging checksum items with overlapping ranges.
4226 if (inode->last_reflink_trans < trans->transid)
4227 return btrfs_csum_file_blocks(trans, log_root, sums);
4230 * Serialize logging for checksums. This is to avoid racing with the
4231 * same checksum being logged by another task that is logging another
4232 * file which happens to refer to the same extent as well. Such races
4233 * can leave checksum items in the log with overlapping ranges.
4235 ret = lock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4240 * Due to extent cloning, we might have logged a csum item that covers a
4241 * subrange of a cloned extent, and later we can end up logging a csum
4242 * item for a larger subrange of the same extent or the entire range.
4243 * This would leave csum items in the log tree that cover the same range
4244 * and break the searches for checksums in the log tree, resulting in
4245 * some checksums missing in the fs/subvolume tree. So just delete (or
4246 * trim and adjust) any existing csum items in the log for this range.
4248 ret = btrfs_del_csums(trans, log_root, sums->logical, sums->len);
4250 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4252 unlock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4258 static noinline int copy_items(struct btrfs_trans_handle *trans,
4259 struct btrfs_inode *inode,
4260 struct btrfs_path *dst_path,
4261 struct btrfs_path *src_path,
4262 int start_slot, int nr, int inode_only,
4265 struct btrfs_root *log = inode->root->log_root;
4266 struct btrfs_file_extent_item *extent;
4267 struct extent_buffer *src;
4269 struct btrfs_key *ins_keys;
4271 struct btrfs_item_batch batch;
4275 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4276 const u64 i_size = i_size_read(&inode->vfs_inode);
4279 * To keep lockdep happy and avoid deadlocks, clone the source leaf and
4280 * use the clone. This is because otherwise we would be changing the log
4281 * tree, to insert items from the subvolume tree or insert csum items,
4282 * while holding a read lock on a leaf from the subvolume tree, which
4283 * creates a nasty lock dependency when COWing log tree nodes/leaves:
4285 * 1) Modifying the log tree triggers an extent buffer allocation while
4286 * holding a write lock on a parent extent buffer from the log tree.
4287 * Allocating the pages for an extent buffer, or the extent buffer
4288 * struct, can trigger inode eviction and finally the inode eviction
4289 * will trigger a release/remove of a delayed node, which requires
4290 * taking the delayed node's mutex;
4292 * 2) Allocating a metadata extent for a log tree can trigger the async
4293 * reclaim thread and make us wait for it to release enough space and
4294 * unblock our reservation ticket. The reclaim thread can start
4295 * flushing delayed items, and that in turn results in the need to
4296 * lock delayed node mutexes and in the need to write lock extent
4297 * buffers of a subvolume tree - all this while holding a write lock
4298 * on the parent extent buffer in the log tree.
4300 * So one task in scenario 1) running in parallel with another task in
4301 * scenario 2) could lead to a deadlock, one wanting to lock a delayed
4302 * node mutex while having a read lock on a leaf from the subvolume,
4303 * while the other is holding the delayed node's mutex and wants to
4304 * write lock the same subvolume leaf for flushing delayed items.
4306 src = btrfs_clone_extent_buffer(src_path->nodes[0]);
4310 i = src_path->slots[0];
4311 btrfs_release_path(src_path);
4312 src_path->nodes[0] = src;
4313 src_path->slots[0] = i;
4315 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4316 nr * sizeof(u32), GFP_NOFS);
4320 ins_sizes = (u32 *)ins_data;
4321 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4322 batch.keys = ins_keys;
4323 batch.data_sizes = ins_sizes;
4324 batch.total_data_size = 0;
4328 for (i = 0; i < nr; i++) {
4329 const int src_slot = start_slot + i;
4330 struct btrfs_root *csum_root;
4331 struct btrfs_ordered_sum *sums;
4332 struct btrfs_ordered_sum *sums_next;
4333 LIST_HEAD(ordered_sums);
4337 u64 extent_num_bytes;
4340 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4342 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4345 extent = btrfs_item_ptr(src, src_slot,
4346 struct btrfs_file_extent_item);
4348 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4352 * Don't copy extents from past generations. That would make us
4353 * log a lot more metadata for common cases like doing only a
4354 * few random writes into a file and then fsync it for the first
4355 * time or after the full sync flag is set on the inode. We can
4356 * get leaves full of extent items, most of which are from past
4357 * generations, so we can skip them - as long as the inode has
4358 * not been the target of a reflink operation in this transaction,
4359 * as in that case it might have had file extent items with old
4360 * generations copied into it. We also must always log prealloc
4361 * extents that start at or beyond eof, otherwise we would lose
4362 * them on log replay.
4364 if (is_old_extent &&
4365 ins_keys[dst_index].offset < i_size &&
4366 inode->last_reflink_trans < trans->transid)
4372 /* Only regular extents have checksums. */
4373 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4377 * If it's an extent created in a past transaction, then its
4378 * checksums are already accessible from the committed csum tree,
4379 * no need to log them.
4384 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4385 /* If it's an explicit hole, there are no checksums. */
4386 if (disk_bytenr == 0)
4389 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4391 if (btrfs_file_extent_compression(src, extent)) {
4393 extent_num_bytes = disk_num_bytes;
4395 extent_offset = btrfs_file_extent_offset(src, extent);
4396 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4399 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4400 disk_bytenr += extent_offset;
4401 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4402 disk_bytenr + extent_num_bytes - 1,
4403 &ordered_sums, 0, false);
4407 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4409 ret = log_csums(trans, inode, log, sums);
4410 list_del(&sums->list);
4417 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4418 batch.total_data_size += ins_sizes[dst_index];
4424 * We have a leaf full of old extent items that don't need to be logged,
4425 * so we don't need to do anything.
4430 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4435 for (i = 0; i < nr; i++) {
4436 const int src_slot = start_slot + i;
4437 const int dst_slot = dst_path->slots[0] + dst_index;
4438 struct btrfs_key key;
4439 unsigned long src_offset;
4440 unsigned long dst_offset;
4443 * We're done, all the remaining items in the source leaf
4444 * correspond to old file extent items.
4446 if (dst_index >= batch.nr)
4449 btrfs_item_key_to_cpu(src, &key, src_slot);
4451 if (key.type != BTRFS_EXTENT_DATA_KEY)
4454 extent = btrfs_item_ptr(src, src_slot,
4455 struct btrfs_file_extent_item);
4457 /* See the comment in the previous loop, same logic. */
4458 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4459 key.offset < i_size &&
4460 inode->last_reflink_trans < trans->transid)
4464 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4465 src_offset = btrfs_item_ptr_offset(src, src_slot);
4467 if (key.type == BTRFS_INODE_ITEM_KEY) {
4468 struct btrfs_inode_item *inode_item;
4470 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4471 struct btrfs_inode_item);
4472 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4474 inode_only == LOG_INODE_EXISTS,
4477 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4478 src_offset, ins_sizes[dst_index]);
4484 btrfs_mark_buffer_dirty(trans, dst_path->nodes[0]);
4485 btrfs_release_path(dst_path);
4492 static int extent_cmp(void *priv, const struct list_head *a,
4493 const struct list_head *b)
4495 const struct extent_map *em1, *em2;
4497 em1 = list_entry(a, struct extent_map, list);
4498 em2 = list_entry(b, struct extent_map, list);
4500 if (em1->start < em2->start)
4502 else if (em1->start > em2->start)
4507 static int log_extent_csums(struct btrfs_trans_handle *trans,
4508 struct btrfs_inode *inode,
4509 struct btrfs_root *log_root,
4510 const struct extent_map *em,
4511 struct btrfs_log_ctx *ctx)
4513 struct btrfs_ordered_extent *ordered;
4514 struct btrfs_root *csum_root;
4517 u64 mod_start = em->mod_start;
4518 u64 mod_len = em->mod_len;
4519 LIST_HEAD(ordered_sums);
4522 if (inode->flags & BTRFS_INODE_NODATASUM ||
4523 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4524 em->block_start == EXTENT_MAP_HOLE)
4527 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4528 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4529 const u64 mod_end = mod_start + mod_len;
4530 struct btrfs_ordered_sum *sums;
4535 if (ordered_end <= mod_start)
4537 if (mod_end <= ordered->file_offset)
4541 * We are going to copy all the csums on this ordered extent, so
4542 * go ahead and adjust mod_start and mod_len in case this ordered
4543 * extent has already been logged.
4545 if (ordered->file_offset > mod_start) {
4546 if (ordered_end >= mod_end)
4547 mod_len = ordered->file_offset - mod_start;
4549 * If we have this case
4551 * |--------- logged extent ---------|
4552 * |----- ordered extent ----|
4554 * Just don't mess with mod_start and mod_len, we'll
4555 * just end up logging more csums than we need and it
4559 if (ordered_end < mod_end) {
4560 mod_len = mod_end - ordered_end;
4561 mod_start = ordered_end;
4568 * To keep us from looping for the above case of an ordered
4569 * extent that falls inside of the logged extent.
4571 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4574 list_for_each_entry(sums, &ordered->list, list) {
4575 ret = log_csums(trans, inode, log_root, sums);
4581 /* We're done, found all csums in the ordered extents. */
4585 /* If we're compressed we have to save the entire range of csums. */
4586 if (em->compress_type) {
4588 csum_len = max(em->block_len, em->orig_block_len);
4590 csum_offset = mod_start - em->start;
4594 /* block start is already adjusted for the file extent offset. */
4595 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4596 ret = btrfs_lookup_csums_list(csum_root, em->block_start + csum_offset,
4597 em->block_start + csum_offset +
4598 csum_len - 1, &ordered_sums, 0, false);
4602 while (!list_empty(&ordered_sums)) {
4603 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4604 struct btrfs_ordered_sum,
4607 ret = log_csums(trans, inode, log_root, sums);
4608 list_del(&sums->list);
4615 static int log_one_extent(struct btrfs_trans_handle *trans,
4616 struct btrfs_inode *inode,
4617 const struct extent_map *em,
4618 struct btrfs_path *path,
4619 struct btrfs_log_ctx *ctx)
4621 struct btrfs_drop_extents_args drop_args = { 0 };
4622 struct btrfs_root *log = inode->root->log_root;
4623 struct btrfs_file_extent_item fi = { 0 };
4624 struct extent_buffer *leaf;
4625 struct btrfs_key key;
4626 u64 extent_offset = em->start - em->orig_start;
4630 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4631 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4632 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4634 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4636 block_len = max(em->block_len, em->orig_block_len);
4637 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4638 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4639 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4640 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4641 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4643 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4646 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4647 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4648 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4649 btrfs_set_stack_file_extent_compression(&fi, em->compress_type);
4651 ret = log_extent_csums(trans, inode, log, em, ctx);
4656 * If this is the first time we are logging the inode in the current
4657 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4658 * because it does a deletion search, which always acquires write locks
4659 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4660 * but also adds significant contention in a log tree, since log trees
4661 * are small, with a root at level 2 or 3 at most, due to their short
4664 if (ctx->logged_before) {
4665 drop_args.path = path;
4666 drop_args.start = em->start;
4667 drop_args.end = em->start + em->len;
4668 drop_args.replace_extent = true;
4669 drop_args.extent_item_size = sizeof(fi);
4670 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4675 if (!drop_args.extent_inserted) {
4676 key.objectid = btrfs_ino(inode);
4677 key.type = BTRFS_EXTENT_DATA_KEY;
4678 key.offset = em->start;
4680 ret = btrfs_insert_empty_item(trans, log, path, &key,
4685 leaf = path->nodes[0];
4686 write_extent_buffer(leaf, &fi,
4687 btrfs_item_ptr_offset(leaf, path->slots[0]),
4689 btrfs_mark_buffer_dirty(trans, leaf);
4691 btrfs_release_path(path);
4697 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4698 * lose them after doing a full/fast fsync and replaying the log. We scan the
4699 * subvolume's root instead of iterating the inode's extent map tree because
4700 * otherwise we can log incorrect extent items based on extent map conversion.
4701 * That can happen due to the fact that extent maps are merged when they
4702 * are not in the extent map tree's list of modified extents.
4704 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4705 struct btrfs_inode *inode,
4706 struct btrfs_path *path)
4708 struct btrfs_root *root = inode->root;
4709 struct btrfs_key key;
4710 const u64 i_size = i_size_read(&inode->vfs_inode);
4711 const u64 ino = btrfs_ino(inode);
4712 struct btrfs_path *dst_path = NULL;
4713 bool dropped_extents = false;
4714 u64 truncate_offset = i_size;
4715 struct extent_buffer *leaf;
4721 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4725 key.type = BTRFS_EXTENT_DATA_KEY;
4726 key.offset = i_size;
4727 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4732 * We must check if there is a prealloc extent that starts before the
4733 * i_size and crosses the i_size boundary. This is to ensure later we
4734 * truncate down to the end of that extent and not to the i_size, as
4735 * otherwise we end up losing part of the prealloc extent after a log
4736 * replay and with an implicit hole if there is another prealloc extent
4737 * that starts at an offset beyond i_size.
4739 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4744 struct btrfs_file_extent_item *ei;
4746 leaf = path->nodes[0];
4747 slot = path->slots[0];
4748 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4750 if (btrfs_file_extent_type(leaf, ei) ==
4751 BTRFS_FILE_EXTENT_PREALLOC) {
4754 btrfs_item_key_to_cpu(leaf, &key, slot);
4755 extent_end = key.offset +
4756 btrfs_file_extent_num_bytes(leaf, ei);
4758 if (extent_end > i_size)
4759 truncate_offset = extent_end;
4766 leaf = path->nodes[0];
4767 slot = path->slots[0];
4769 if (slot >= btrfs_header_nritems(leaf)) {
4771 ret = copy_items(trans, inode, dst_path, path,
4772 start_slot, ins_nr, 1, 0);
4777 ret = btrfs_next_leaf(root, path);
4787 btrfs_item_key_to_cpu(leaf, &key, slot);
4788 if (key.objectid > ino)
4790 if (WARN_ON_ONCE(key.objectid < ino) ||
4791 key.type < BTRFS_EXTENT_DATA_KEY ||
4792 key.offset < i_size) {
4796 if (!dropped_extents) {
4798 * Avoid logging extent items logged in past fsync calls
4799 * and leading to duplicate keys in the log tree.
4801 ret = truncate_inode_items(trans, root->log_root, inode,
4803 BTRFS_EXTENT_DATA_KEY);
4806 dropped_extents = true;
4813 dst_path = btrfs_alloc_path();
4821 ret = copy_items(trans, inode, dst_path, path,
4822 start_slot, ins_nr, 1, 0);
4824 btrfs_release_path(path);
4825 btrfs_free_path(dst_path);
4829 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4830 struct btrfs_inode *inode,
4831 struct btrfs_path *path,
4832 struct btrfs_log_ctx *ctx)
4834 struct btrfs_ordered_extent *ordered;
4835 struct btrfs_ordered_extent *tmp;
4836 struct extent_map *em, *n;
4838 struct extent_map_tree *tree = &inode->extent_tree;
4842 write_lock(&tree->lock);
4844 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4845 list_del_init(&em->list);
4847 * Just an arbitrary number, this can be really CPU intensive
4848 * once we start getting a lot of extents, and really once we
4849 * have a bunch of extents we just want to commit since it will
4852 if (++num > 32768) {
4853 list_del_init(&tree->modified_extents);
4858 if (em->generation < trans->transid)
4861 /* We log prealloc extents beyond eof later. */
4862 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4863 em->start >= i_size_read(&inode->vfs_inode))
4866 /* Need a ref to keep it from getting evicted from cache */
4867 refcount_inc(&em->refs);
4868 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4869 list_add_tail(&em->list, &extents);
4873 list_sort(NULL, &extents, extent_cmp);
4875 while (!list_empty(&extents)) {
4876 em = list_entry(extents.next, struct extent_map, list);
4878 list_del_init(&em->list);
4881 * If we had an error we just need to delete everybody from our
4885 clear_em_logging(tree, em);
4886 free_extent_map(em);
4890 write_unlock(&tree->lock);
4892 ret = log_one_extent(trans, inode, em, path, ctx);
4893 write_lock(&tree->lock);
4894 clear_em_logging(tree, em);
4895 free_extent_map(em);
4897 WARN_ON(!list_empty(&extents));
4898 write_unlock(&tree->lock);
4901 ret = btrfs_log_prealloc_extents(trans, inode, path);
4906 * We have logged all extents successfully, now make sure the commit of
4907 * the current transaction waits for the ordered extents to complete
4908 * before it commits and wipes out the log trees, otherwise we would
4909 * lose data if an ordered extents completes after the transaction
4910 * commits and a power failure happens after the transaction commit.
4912 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4913 list_del_init(&ordered->log_list);
4914 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4916 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4917 spin_lock_irq(&inode->ordered_tree_lock);
4918 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4919 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4920 atomic_inc(&trans->transaction->pending_ordered);
4922 spin_unlock_irq(&inode->ordered_tree_lock);
4924 btrfs_put_ordered_extent(ordered);
4930 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4931 struct btrfs_path *path, u64 *size_ret)
4933 struct btrfs_key key;
4936 key.objectid = btrfs_ino(inode);
4937 key.type = BTRFS_INODE_ITEM_KEY;
4940 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4943 } else if (ret > 0) {
4946 struct btrfs_inode_item *item;
4948 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4949 struct btrfs_inode_item);
4950 *size_ret = btrfs_inode_size(path->nodes[0], item);
4952 * If the in-memory inode's i_size is smaller then the inode
4953 * size stored in the btree, return the inode's i_size, so
4954 * that we get a correct inode size after replaying the log
4955 * when before a power failure we had a shrinking truncate
4956 * followed by addition of a new name (rename / new hard link).
4957 * Otherwise return the inode size from the btree, to avoid
4958 * data loss when replaying a log due to previously doing a
4959 * write that expands the inode's size and logging a new name
4960 * immediately after.
4962 if (*size_ret > inode->vfs_inode.i_size)
4963 *size_ret = inode->vfs_inode.i_size;
4966 btrfs_release_path(path);
4971 * At the moment we always log all xattrs. This is to figure out at log replay
4972 * time which xattrs must have their deletion replayed. If a xattr is missing
4973 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4974 * because if a xattr is deleted, the inode is fsynced and a power failure
4975 * happens, causing the log to be replayed the next time the fs is mounted,
4976 * we want the xattr to not exist anymore (same behaviour as other filesystems
4977 * with a journal, ext3/4, xfs, f2fs, etc).
4979 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4980 struct btrfs_inode *inode,
4981 struct btrfs_path *path,
4982 struct btrfs_path *dst_path)
4984 struct btrfs_root *root = inode->root;
4986 struct btrfs_key key;
4987 const u64 ino = btrfs_ino(inode);
4990 bool found_xattrs = false;
4992 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4996 key.type = BTRFS_XATTR_ITEM_KEY;
4999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5004 int slot = path->slots[0];
5005 struct extent_buffer *leaf = path->nodes[0];
5006 int nritems = btrfs_header_nritems(leaf);
5008 if (slot >= nritems) {
5010 ret = copy_items(trans, inode, dst_path, path,
5011 start_slot, ins_nr, 1, 0);
5016 ret = btrfs_next_leaf(root, path);
5024 btrfs_item_key_to_cpu(leaf, &key, slot);
5025 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5032 found_xattrs = true;
5036 ret = copy_items(trans, inode, dst_path, path,
5037 start_slot, ins_nr, 1, 0);
5043 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5049 * When using the NO_HOLES feature if we punched a hole that causes the
5050 * deletion of entire leafs or all the extent items of the first leaf (the one
5051 * that contains the inode item and references) we may end up not processing
5052 * any extents, because there are no leafs with a generation matching the
5053 * current transaction that have extent items for our inode. So we need to find
5054 * if any holes exist and then log them. We also need to log holes after any
5055 * truncate operation that changes the inode's size.
5057 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5058 struct btrfs_inode *inode,
5059 struct btrfs_path *path)
5061 struct btrfs_root *root = inode->root;
5062 struct btrfs_fs_info *fs_info = root->fs_info;
5063 struct btrfs_key key;
5064 const u64 ino = btrfs_ino(inode);
5065 const u64 i_size = i_size_read(&inode->vfs_inode);
5066 u64 prev_extent_end = 0;
5069 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5073 key.type = BTRFS_EXTENT_DATA_KEY;
5076 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5081 struct extent_buffer *leaf = path->nodes[0];
5083 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5084 ret = btrfs_next_leaf(root, path);
5091 leaf = path->nodes[0];
5094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5095 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5098 /* We have a hole, log it. */
5099 if (prev_extent_end < key.offset) {
5100 const u64 hole_len = key.offset - prev_extent_end;
5103 * Release the path to avoid deadlocks with other code
5104 * paths that search the root while holding locks on
5105 * leafs from the log root.
5107 btrfs_release_path(path);
5108 ret = btrfs_insert_hole_extent(trans, root->log_root,
5109 ino, prev_extent_end,
5115 * Search for the same key again in the root. Since it's
5116 * an extent item and we are holding the inode lock, the
5117 * key must still exist. If it doesn't just emit warning
5118 * and return an error to fall back to a transaction
5121 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5124 if (WARN_ON(ret > 0))
5126 leaf = path->nodes[0];
5129 prev_extent_end = btrfs_file_extent_end(path);
5134 if (prev_extent_end < i_size) {
5137 btrfs_release_path(path);
5138 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5139 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5140 prev_extent_end, hole_len);
5149 * When we are logging a new inode X, check if it doesn't have a reference that
5150 * matches the reference from some other inode Y created in a past transaction
5151 * and that was renamed in the current transaction. If we don't do this, then at
5152 * log replay time we can lose inode Y (and all its files if it's a directory):
5155 * echo "hello world" > /mnt/x/foobar
5158 * mkdir /mnt/x # or touch /mnt/x
5159 * xfs_io -c fsync /mnt/x
5161 * mount fs, trigger log replay
5163 * After the log replay procedure, we would lose the first directory and all its
5164 * files (file foobar).
5165 * For the case where inode Y is not a directory we simply end up losing it:
5167 * echo "123" > /mnt/foo
5169 * mv /mnt/foo /mnt/bar
5170 * echo "abc" > /mnt/foo
5171 * xfs_io -c fsync /mnt/foo
5174 * We also need this for cases where a snapshot entry is replaced by some other
5175 * entry (file or directory) otherwise we end up with an unreplayable log due to
5176 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5177 * if it were a regular entry:
5180 * btrfs subvolume snapshot /mnt /mnt/x/snap
5181 * btrfs subvolume delete /mnt/x/snap
5184 * fsync /mnt/x or fsync some new file inside it
5187 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5188 * the same transaction.
5190 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5192 const struct btrfs_key *key,
5193 struct btrfs_inode *inode,
5194 u64 *other_ino, u64 *other_parent)
5197 struct btrfs_path *search_path;
5200 u32 item_size = btrfs_item_size(eb, slot);
5202 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5204 search_path = btrfs_alloc_path();
5207 search_path->search_commit_root = 1;
5208 search_path->skip_locking = 1;
5210 while (cur_offset < item_size) {
5214 unsigned long name_ptr;
5215 struct btrfs_dir_item *di;
5216 struct fscrypt_str name_str;
5218 if (key->type == BTRFS_INODE_REF_KEY) {
5219 struct btrfs_inode_ref *iref;
5221 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5222 parent = key->offset;
5223 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5224 name_ptr = (unsigned long)(iref + 1);
5225 this_len = sizeof(*iref) + this_name_len;
5227 struct btrfs_inode_extref *extref;
5229 extref = (struct btrfs_inode_extref *)(ptr +
5231 parent = btrfs_inode_extref_parent(eb, extref);
5232 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5233 name_ptr = (unsigned long)&extref->name;
5234 this_len = sizeof(*extref) + this_name_len;
5237 if (this_name_len > name_len) {
5240 new_name = krealloc(name, this_name_len, GFP_NOFS);
5245 name_len = this_name_len;
5249 read_extent_buffer(eb, name, name_ptr, this_name_len);
5251 name_str.name = name;
5252 name_str.len = this_name_len;
5253 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5254 parent, &name_str, 0);
5255 if (di && !IS_ERR(di)) {
5256 struct btrfs_key di_key;
5258 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5260 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5261 if (di_key.objectid != key->objectid) {
5263 *other_ino = di_key.objectid;
5264 *other_parent = parent;
5272 } else if (IS_ERR(di)) {
5276 btrfs_release_path(search_path);
5278 cur_offset += this_len;
5282 btrfs_free_path(search_path);
5288 * Check if we need to log an inode. This is used in contexts where while
5289 * logging an inode we need to log another inode (either that it exists or in
5290 * full mode). This is used instead of btrfs_inode_in_log() because the later
5291 * requires the inode to be in the log and have the log transaction committed,
5292 * while here we do not care if the log transaction was already committed - our
5293 * caller will commit the log later - and we want to avoid logging an inode
5294 * multiple times when multiple tasks have joined the same log transaction.
5296 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5297 struct btrfs_inode *inode)
5300 * If a directory was not modified, no dentries added or removed, we can
5301 * and should avoid logging it.
5303 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5307 * If this inode does not have new/updated/deleted xattrs since the last
5308 * time it was logged and is flagged as logged in the current transaction,
5309 * we can skip logging it. As for new/deleted names, those are updated in
5310 * the log by link/unlink/rename operations.
5311 * In case the inode was logged and then evicted and reloaded, its
5312 * logged_trans will be 0, in which case we have to fully log it since
5313 * logged_trans is a transient field, not persisted.
5315 if (inode_logged(trans, inode, NULL) == 1 &&
5316 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5322 struct btrfs_dir_list {
5324 struct list_head list;
5328 * Log the inodes of the new dentries of a directory.
5329 * See process_dir_items_leaf() for details about why it is needed.
5330 * This is a recursive operation - if an existing dentry corresponds to a
5331 * directory, that directory's new entries are logged too (same behaviour as
5332 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5333 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5334 * complains about the following circular lock dependency / possible deadlock:
5338 * lock(&type->i_mutex_dir_key#3/2);
5339 * lock(sb_internal#2);
5340 * lock(&type->i_mutex_dir_key#3/2);
5341 * lock(&sb->s_type->i_mutex_key#14);
5343 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5344 * sb_start_intwrite() in btrfs_start_transaction().
5345 * Not acquiring the VFS lock of the inodes is still safe because:
5347 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5348 * that while logging the inode new references (names) are added or removed
5349 * from the inode, leaving the logged inode item with a link count that does
5350 * not match the number of logged inode reference items. This is fine because
5351 * at log replay time we compute the real number of links and correct the
5352 * link count in the inode item (see replay_one_buffer() and
5353 * link_to_fixup_dir());
5355 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5356 * while logging the inode's items new index items (key type
5357 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5358 * has a size that doesn't match the sum of the lengths of all the logged
5359 * names - this is ok, not a problem, because at log replay time we set the
5360 * directory's i_size to the correct value (see replay_one_name() and
5361 * overwrite_item()).
5363 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5364 struct btrfs_inode *start_inode,
5365 struct btrfs_log_ctx *ctx)
5367 struct btrfs_root *root = start_inode->root;
5368 struct btrfs_fs_info *fs_info = root->fs_info;
5369 struct btrfs_path *path;
5370 LIST_HEAD(dir_list);
5371 struct btrfs_dir_list *dir_elem;
5372 u64 ino = btrfs_ino(start_inode);
5373 struct btrfs_inode *curr_inode = start_inode;
5377 * If we are logging a new name, as part of a link or rename operation,
5378 * don't bother logging new dentries, as we just want to log the names
5379 * of an inode and that any new parents exist.
5381 if (ctx->logging_new_name)
5384 path = btrfs_alloc_path();
5388 /* Pairs with btrfs_add_delayed_iput below. */
5389 ihold(&curr_inode->vfs_inode);
5392 struct inode *vfs_inode;
5393 struct btrfs_key key;
5394 struct btrfs_key found_key;
5396 bool continue_curr_inode = true;
5400 key.type = BTRFS_DIR_INDEX_KEY;
5401 key.offset = btrfs_get_first_dir_index_to_log(curr_inode);
5402 next_index = key.offset;
5404 btrfs_for_each_slot(root->log_root, &key, &found_key, path, iter_ret) {
5405 struct extent_buffer *leaf = path->nodes[0];
5406 struct btrfs_dir_item *di;
5407 struct btrfs_key di_key;
5408 struct inode *di_inode;
5409 int log_mode = LOG_INODE_EXISTS;
5412 if (found_key.objectid != ino ||
5413 found_key.type != BTRFS_DIR_INDEX_KEY) {
5414 continue_curr_inode = false;
5418 next_index = found_key.offset + 1;
5420 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5421 type = btrfs_dir_ftype(leaf, di);
5422 if (btrfs_dir_transid(leaf, di) < trans->transid)
5424 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5425 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5428 btrfs_release_path(path);
5429 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5430 if (IS_ERR(di_inode)) {
5431 ret = PTR_ERR(di_inode);
5435 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5436 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5440 ctx->log_new_dentries = false;
5441 if (type == BTRFS_FT_DIR)
5442 log_mode = LOG_INODE_ALL;
5443 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5445 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5448 if (ctx->log_new_dentries) {
5449 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5454 dir_elem->ino = di_key.objectid;
5455 list_add_tail(&dir_elem->list, &dir_list);
5460 btrfs_release_path(path);
5465 } else if (iter_ret > 0) {
5466 continue_curr_inode = false;
5471 if (continue_curr_inode && key.offset < (u64)-1) {
5476 btrfs_set_first_dir_index_to_log(curr_inode, next_index);
5478 if (list_empty(&dir_list))
5481 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5482 ino = dir_elem->ino;
5483 list_del(&dir_elem->list);
5486 btrfs_add_delayed_iput(curr_inode);
5489 vfs_inode = btrfs_iget(fs_info->sb, ino, root);
5490 if (IS_ERR(vfs_inode)) {
5491 ret = PTR_ERR(vfs_inode);
5494 curr_inode = BTRFS_I(vfs_inode);
5497 btrfs_free_path(path);
5499 btrfs_add_delayed_iput(curr_inode);
5502 struct btrfs_dir_list *next;
5504 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5511 struct btrfs_ino_list {
5514 struct list_head list;
5517 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5519 struct btrfs_ino_list *curr;
5520 struct btrfs_ino_list *next;
5522 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5523 list_del(&curr->list);
5528 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5529 struct btrfs_path *path)
5531 struct btrfs_key key;
5535 key.type = BTRFS_INODE_ITEM_KEY;
5538 path->search_commit_root = 1;
5539 path->skip_locking = 1;
5541 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5542 if (WARN_ON_ONCE(ret > 0)) {
5544 * We have previously found the inode through the commit root
5545 * so this should not happen. If it does, just error out and
5546 * fallback to a transaction commit.
5549 } else if (ret == 0) {
5550 struct btrfs_inode_item *item;
5552 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5553 struct btrfs_inode_item);
5554 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5558 btrfs_release_path(path);
5559 path->search_commit_root = 0;
5560 path->skip_locking = 0;
5565 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5566 struct btrfs_root *root,
5567 struct btrfs_path *path,
5568 u64 ino, u64 parent,
5569 struct btrfs_log_ctx *ctx)
5571 struct btrfs_ino_list *ino_elem;
5572 struct inode *inode;
5575 * It's rare to have a lot of conflicting inodes, in practice it is not
5576 * common to have more than 1 or 2. We don't want to collect too many,
5577 * as we could end up logging too many inodes (even if only in
5578 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5581 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES)
5582 return BTRFS_LOG_FORCE_COMMIT;
5584 inode = btrfs_iget(root->fs_info->sb, ino, root);
5586 * If the other inode that had a conflicting dir entry was deleted in
5587 * the current transaction then we either:
5589 * 1) Log the parent directory (later after adding it to the list) if
5590 * the inode is a directory. This is because it may be a deleted
5591 * subvolume/snapshot or it may be a regular directory that had
5592 * deleted subvolumes/snapshots (or subdirectories that had them),
5593 * and at the moment we can't deal with dropping subvolumes/snapshots
5594 * during log replay. So we just log the parent, which will result in
5595 * a fallback to a transaction commit if we are dealing with those
5596 * cases (last_unlink_trans will match the current transaction);
5598 * 2) Do nothing if it's not a directory. During log replay we simply
5599 * unlink the conflicting dentry from the parent directory and then
5600 * add the dentry for our inode. Like this we can avoid logging the
5601 * parent directory (and maybe fallback to a transaction commit in
5602 * case it has a last_unlink_trans == trans->transid, due to moving
5603 * some inode from it to some other directory).
5605 if (IS_ERR(inode)) {
5606 int ret = PTR_ERR(inode);
5611 ret = conflicting_inode_is_dir(root, ino, path);
5612 /* Not a directory or we got an error. */
5616 /* Conflicting inode is a directory, so we'll log its parent. */
5617 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5620 ino_elem->ino = ino;
5621 ino_elem->parent = parent;
5622 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5623 ctx->num_conflict_inodes++;
5629 * If the inode was already logged skip it - otherwise we can hit an
5630 * infinite loop. Example:
5632 * From the commit root (previous transaction) we have the following
5635 * inode 257 a directory
5636 * inode 258 with references "zz" and "zz_link" on inode 257
5637 * inode 259 with reference "a" on inode 257
5639 * And in the current (uncommitted) transaction we have:
5641 * inode 257 a directory, unchanged
5642 * inode 258 with references "a" and "a2" on inode 257
5643 * inode 259 with reference "zz_link" on inode 257
5644 * inode 261 with reference "zz" on inode 257
5646 * When logging inode 261 the following infinite loop could
5647 * happen if we don't skip already logged inodes:
5649 * - we detect inode 258 as a conflicting inode, with inode 261
5650 * on reference "zz", and log it;
5652 * - we detect inode 259 as a conflicting inode, with inode 258
5653 * on reference "a", and log it;
5655 * - we detect inode 258 as a conflicting inode, with inode 259
5656 * on reference "zz_link", and log it - again! After this we
5657 * repeat the above steps forever.
5659 * Here we can use need_log_inode() because we only need to log the
5660 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5661 * so that the log ends up with the new name and without the old name.
5663 if (!need_log_inode(trans, BTRFS_I(inode))) {
5664 btrfs_add_delayed_iput(BTRFS_I(inode));
5668 btrfs_add_delayed_iput(BTRFS_I(inode));
5670 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5673 ino_elem->ino = ino;
5674 ino_elem->parent = parent;
5675 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5676 ctx->num_conflict_inodes++;
5681 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5682 struct btrfs_root *root,
5683 struct btrfs_log_ctx *ctx)
5685 struct btrfs_fs_info *fs_info = root->fs_info;
5689 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5690 * otherwise we could have unbounded recursion of btrfs_log_inode()
5691 * calls. This check guarantees we can have only 1 level of recursion.
5693 if (ctx->logging_conflict_inodes)
5696 ctx->logging_conflict_inodes = true;
5699 * New conflicting inodes may be found and added to the list while we
5700 * are logging a conflicting inode, so keep iterating while the list is
5703 while (!list_empty(&ctx->conflict_inodes)) {
5704 struct btrfs_ino_list *curr;
5705 struct inode *inode;
5709 curr = list_first_entry(&ctx->conflict_inodes,
5710 struct btrfs_ino_list, list);
5712 parent = curr->parent;
5713 list_del(&curr->list);
5716 inode = btrfs_iget(fs_info->sb, ino, root);
5718 * If the other inode that had a conflicting dir entry was
5719 * deleted in the current transaction, we need to log its parent
5720 * directory. See the comment at add_conflicting_inode().
5722 if (IS_ERR(inode)) {
5723 ret = PTR_ERR(inode);
5727 inode = btrfs_iget(fs_info->sb, parent, root);
5728 if (IS_ERR(inode)) {
5729 ret = PTR_ERR(inode);
5734 * Always log the directory, we cannot make this
5735 * conditional on need_log_inode() because the directory
5736 * might have been logged in LOG_INODE_EXISTS mode or
5737 * the dir index of the conflicting inode is not in a
5738 * dir index key range logged for the directory. So we
5739 * must make sure the deletion is recorded.
5741 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5742 LOG_INODE_ALL, ctx);
5743 btrfs_add_delayed_iput(BTRFS_I(inode));
5750 * Here we can use need_log_inode() because we only need to log
5751 * the inode in LOG_INODE_EXISTS mode and rename operations
5752 * update the log, so that the log ends up with the new name and
5753 * without the old name.
5755 * We did this check at add_conflicting_inode(), but here we do
5756 * it again because if some other task logged the inode after
5757 * that, we can avoid doing it again.
5759 if (!need_log_inode(trans, BTRFS_I(inode))) {
5760 btrfs_add_delayed_iput(BTRFS_I(inode));
5765 * We are safe logging the other inode without acquiring its
5766 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5767 * are safe against concurrent renames of the other inode as
5768 * well because during a rename we pin the log and update the
5769 * log with the new name before we unpin it.
5771 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5772 btrfs_add_delayed_iput(BTRFS_I(inode));
5777 ctx->logging_conflict_inodes = false;
5779 free_conflicting_inodes(ctx);
5784 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5785 struct btrfs_inode *inode,
5786 struct btrfs_key *min_key,
5787 const struct btrfs_key *max_key,
5788 struct btrfs_path *path,
5789 struct btrfs_path *dst_path,
5790 const u64 logged_isize,
5791 const int inode_only,
5792 struct btrfs_log_ctx *ctx,
5793 bool *need_log_inode_item)
5795 const u64 i_size = i_size_read(&inode->vfs_inode);
5796 struct btrfs_root *root = inode->root;
5797 int ins_start_slot = 0;
5802 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5810 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5811 if (min_key->objectid != max_key->objectid)
5813 if (min_key->type > max_key->type)
5816 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5817 *need_log_inode_item = false;
5818 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5819 min_key->offset >= i_size) {
5821 * Extents at and beyond eof are logged with
5822 * btrfs_log_prealloc_extents().
5823 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5824 * and no keys greater than that, so bail out.
5827 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5828 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5829 (inode->generation == trans->transid ||
5830 ctx->logging_conflict_inodes)) {
5832 u64 other_parent = 0;
5834 ret = btrfs_check_ref_name_override(path->nodes[0],
5835 path->slots[0], min_key, inode,
5836 &other_ino, &other_parent);
5839 } else if (ret > 0 &&
5840 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5845 ins_start_slot = path->slots[0];
5847 ret = copy_items(trans, inode, dst_path, path,
5848 ins_start_slot, ins_nr,
5849 inode_only, logged_isize);
5854 btrfs_release_path(path);
5855 ret = add_conflicting_inode(trans, root, path,
5862 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5863 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5866 ret = copy_items(trans, inode, dst_path, path,
5868 ins_nr, inode_only, logged_isize);
5875 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5878 } else if (!ins_nr) {
5879 ins_start_slot = path->slots[0];
5884 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5885 ins_nr, inode_only, logged_isize);
5889 ins_start_slot = path->slots[0];
5892 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5893 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5898 ret = copy_items(trans, inode, dst_path, path,
5899 ins_start_slot, ins_nr, inode_only,
5905 btrfs_release_path(path);
5907 if (min_key->offset < (u64)-1) {
5909 } else if (min_key->type < max_key->type) {
5911 min_key->offset = 0;
5917 * We may process many leaves full of items for our inode, so
5918 * avoid monopolizing a cpu for too long by rescheduling while
5919 * not holding locks on any tree.
5924 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5925 ins_nr, inode_only, logged_isize);
5930 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5932 * Release the path because otherwise we might attempt to double
5933 * lock the same leaf with btrfs_log_prealloc_extents() below.
5935 btrfs_release_path(path);
5936 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5942 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
5943 struct btrfs_root *log,
5944 struct btrfs_path *path,
5945 const struct btrfs_item_batch *batch,
5946 const struct btrfs_delayed_item *first_item)
5948 const struct btrfs_delayed_item *curr = first_item;
5951 ret = btrfs_insert_empty_items(trans, log, path, batch);
5955 for (int i = 0; i < batch->nr; i++) {
5958 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
5959 write_extent_buffer(path->nodes[0], &curr->data,
5960 (unsigned long)data_ptr, curr->data_len);
5961 curr = list_next_entry(curr, log_list);
5965 btrfs_release_path(path);
5970 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
5971 struct btrfs_inode *inode,
5972 struct btrfs_path *path,
5973 const struct list_head *delayed_ins_list,
5974 struct btrfs_log_ctx *ctx)
5976 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
5977 const int max_batch_size = 195;
5978 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
5979 const u64 ino = btrfs_ino(inode);
5980 struct btrfs_root *log = inode->root->log_root;
5981 struct btrfs_item_batch batch = {
5983 .total_data_size = 0,
5985 const struct btrfs_delayed_item *first = NULL;
5986 const struct btrfs_delayed_item *curr;
5988 struct btrfs_key *ins_keys;
5990 u64 curr_batch_size = 0;
5994 /* We are adding dir index items to the log tree. */
5995 lockdep_assert_held(&inode->log_mutex);
5998 * We collect delayed items before copying index keys from the subvolume
5999 * to the log tree. However just after we collected them, they may have
6000 * been flushed (all of them or just some of them), and therefore we
6001 * could have copied them from the subvolume tree to the log tree.
6002 * So find the first delayed item that was not yet logged (they are
6003 * sorted by index number).
6005 list_for_each_entry(curr, delayed_ins_list, log_list) {
6006 if (curr->index > inode->last_dir_index_offset) {
6012 /* Empty list or all delayed items were already logged. */
6016 ins_data = kmalloc(max_batch_size * sizeof(u32) +
6017 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
6020 ins_sizes = (u32 *)ins_data;
6021 batch.data_sizes = ins_sizes;
6022 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6023 batch.keys = ins_keys;
6026 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6027 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6029 if (curr_batch_size + curr_size > leaf_data_size ||
6030 batch.nr == max_batch_size) {
6031 ret = insert_delayed_items_batch(trans, log, path,
6037 batch.total_data_size = 0;
6038 curr_batch_size = 0;
6042 ins_sizes[batch_idx] = curr->data_len;
6043 ins_keys[batch_idx].objectid = ino;
6044 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6045 ins_keys[batch_idx].offset = curr->index;
6046 curr_batch_size += curr_size;
6047 batch.total_data_size += curr->data_len;
6050 curr = list_next_entry(curr, log_list);
6053 ASSERT(batch.nr >= 1);
6054 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6056 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6058 inode->last_dir_index_offset = curr->index;
6065 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6066 struct btrfs_inode *inode,
6067 struct btrfs_path *path,
6068 const struct list_head *delayed_del_list,
6069 struct btrfs_log_ctx *ctx)
6071 const u64 ino = btrfs_ino(inode);
6072 const struct btrfs_delayed_item *curr;
6074 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6077 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6078 u64 first_dir_index = curr->index;
6080 const struct btrfs_delayed_item *next;
6084 * Find a range of consecutive dir index items to delete. Like
6085 * this we log a single dir range item spanning several contiguous
6086 * dir items instead of logging one range item per dir index item.
6088 next = list_next_entry(curr, log_list);
6089 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6090 if (next->index != curr->index + 1)
6093 next = list_next_entry(next, log_list);
6096 last_dir_index = curr->index;
6097 ASSERT(last_dir_index >= first_dir_index);
6099 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6100 ino, first_dir_index, last_dir_index);
6103 curr = list_next_entry(curr, log_list);
6109 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6110 struct btrfs_inode *inode,
6111 struct btrfs_path *path,
6112 struct btrfs_log_ctx *ctx,
6113 const struct list_head *delayed_del_list,
6114 const struct btrfs_delayed_item *first,
6115 const struct btrfs_delayed_item **last_ret)
6117 const struct btrfs_delayed_item *next;
6118 struct extent_buffer *leaf = path->nodes[0];
6119 const int last_slot = btrfs_header_nritems(leaf) - 1;
6120 int slot = path->slots[0] + 1;
6121 const u64 ino = btrfs_ino(inode);
6123 next = list_next_entry(first, log_list);
6125 while (slot < last_slot &&
6126 !list_entry_is_head(next, delayed_del_list, log_list)) {
6127 struct btrfs_key key;
6129 btrfs_item_key_to_cpu(leaf, &key, slot);
6130 if (key.objectid != ino ||
6131 key.type != BTRFS_DIR_INDEX_KEY ||
6132 key.offset != next->index)
6137 next = list_next_entry(next, log_list);
6140 return btrfs_del_items(trans, inode->root->log_root, path,
6141 path->slots[0], slot - path->slots[0]);
6144 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6145 struct btrfs_inode *inode,
6146 struct btrfs_path *path,
6147 const struct list_head *delayed_del_list,
6148 struct btrfs_log_ctx *ctx)
6150 struct btrfs_root *log = inode->root->log_root;
6151 const struct btrfs_delayed_item *curr;
6152 u64 last_range_start = 0;
6153 u64 last_range_end = 0;
6154 struct btrfs_key key;
6156 key.objectid = btrfs_ino(inode);
6157 key.type = BTRFS_DIR_INDEX_KEY;
6158 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6161 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6162 const struct btrfs_delayed_item *last = curr;
6163 u64 first_dir_index = curr->index;
6165 bool deleted_items = false;
6168 key.offset = curr->index;
6169 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6172 } else if (ret == 0) {
6173 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6174 delayed_del_list, curr,
6178 deleted_items = true;
6181 btrfs_release_path(path);
6184 * If we deleted items from the leaf, it means we have a range
6185 * item logging their range, so no need to add one or update an
6186 * existing one. Otherwise we have to log a dir range item.
6191 last_dir_index = last->index;
6192 ASSERT(last_dir_index >= first_dir_index);
6194 * If this range starts right after where the previous one ends,
6195 * then we want to reuse the previous range item and change its
6196 * end offset to the end of this range. This is just to minimize
6197 * leaf space usage, by avoiding adding a new range item.
6199 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6200 first_dir_index = last_range_start;
6202 ret = insert_dir_log_key(trans, log, path, key.objectid,
6203 first_dir_index, last_dir_index);
6207 last_range_start = first_dir_index;
6208 last_range_end = last_dir_index;
6210 curr = list_next_entry(last, log_list);
6216 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6217 struct btrfs_inode *inode,
6218 struct btrfs_path *path,
6219 const struct list_head *delayed_del_list,
6220 struct btrfs_log_ctx *ctx)
6223 * We are deleting dir index items from the log tree or adding range
6226 lockdep_assert_held(&inode->log_mutex);
6228 if (list_empty(delayed_del_list))
6231 if (ctx->logged_before)
6232 return log_delayed_deletions_incremental(trans, inode, path,
6233 delayed_del_list, ctx);
6235 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6240 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6241 * items instead of the subvolume tree.
6243 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6244 struct btrfs_inode *inode,
6245 const struct list_head *delayed_ins_list,
6246 struct btrfs_log_ctx *ctx)
6248 const bool orig_log_new_dentries = ctx->log_new_dentries;
6249 struct btrfs_fs_info *fs_info = trans->fs_info;
6250 struct btrfs_delayed_item *item;
6254 * No need for the log mutex, plus to avoid potential deadlocks or
6255 * lockdep annotations due to nesting of delayed inode mutexes and log
6258 lockdep_assert_not_held(&inode->log_mutex);
6260 ASSERT(!ctx->logging_new_delayed_dentries);
6261 ctx->logging_new_delayed_dentries = true;
6263 list_for_each_entry(item, delayed_ins_list, log_list) {
6264 struct btrfs_dir_item *dir_item;
6265 struct inode *di_inode;
6266 struct btrfs_key key;
6267 int log_mode = LOG_INODE_EXISTS;
6269 dir_item = (struct btrfs_dir_item *)item->data;
6270 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6272 if (key.type == BTRFS_ROOT_ITEM_KEY)
6275 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6276 if (IS_ERR(di_inode)) {
6277 ret = PTR_ERR(di_inode);
6281 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6282 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6286 if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR)
6287 log_mode = LOG_INODE_ALL;
6289 ctx->log_new_dentries = false;
6290 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6292 if (!ret && ctx->log_new_dentries)
6293 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6295 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6301 ctx->log_new_dentries = orig_log_new_dentries;
6302 ctx->logging_new_delayed_dentries = false;
6307 /* log a single inode in the tree log.
6308 * At least one parent directory for this inode must exist in the tree
6309 * or be logged already.
6311 * Any items from this inode changed by the current transaction are copied
6312 * to the log tree. An extra reference is taken on any extents in this
6313 * file, allowing us to avoid a whole pile of corner cases around logging
6314 * blocks that have been removed from the tree.
6316 * See LOG_INODE_ALL and related defines for a description of what inode_only
6319 * This handles both files and directories.
6321 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6322 struct btrfs_inode *inode,
6324 struct btrfs_log_ctx *ctx)
6326 struct btrfs_path *path;
6327 struct btrfs_path *dst_path;
6328 struct btrfs_key min_key;
6329 struct btrfs_key max_key;
6330 struct btrfs_root *log = inode->root->log_root;
6332 bool fast_search = false;
6333 u64 ino = btrfs_ino(inode);
6334 struct extent_map_tree *em_tree = &inode->extent_tree;
6335 u64 logged_isize = 0;
6336 bool need_log_inode_item = true;
6337 bool xattrs_logged = false;
6338 bool inode_item_dropped = true;
6339 bool full_dir_logging = false;
6340 LIST_HEAD(delayed_ins_list);
6341 LIST_HEAD(delayed_del_list);
6343 path = btrfs_alloc_path();
6346 dst_path = btrfs_alloc_path();
6348 btrfs_free_path(path);
6352 min_key.objectid = ino;
6353 min_key.type = BTRFS_INODE_ITEM_KEY;
6356 max_key.objectid = ino;
6359 /* today the code can only do partial logging of directories */
6360 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6361 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6362 &inode->runtime_flags) &&
6363 inode_only >= LOG_INODE_EXISTS))
6364 max_key.type = BTRFS_XATTR_ITEM_KEY;
6366 max_key.type = (u8)-1;
6367 max_key.offset = (u64)-1;
6369 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6370 full_dir_logging = true;
6373 * If we are logging a directory while we are logging dentries of the
6374 * delayed items of some other inode, then we need to flush the delayed
6375 * items of this directory and not log the delayed items directly. This
6376 * is to prevent more than one level of recursion into btrfs_log_inode()
6377 * by having something like this:
6379 * $ mkdir -p a/b/c/d/e/f/g/h/...
6380 * $ xfs_io -c "fsync" a
6382 * Where all directories in the path did not exist before and are
6383 * created in the current transaction.
6384 * So in such a case we directly log the delayed items of the main
6385 * directory ("a") without flushing them first, while for each of its
6386 * subdirectories we flush their delayed items before logging them.
6387 * This prevents a potential unbounded recursion like this:
6390 * log_new_delayed_dentries()
6392 * log_new_delayed_dentries()
6394 * log_new_delayed_dentries()
6397 * We have thresholds for the maximum number of delayed items to have in
6398 * memory, and once they are hit, the items are flushed asynchronously.
6399 * However the limit is quite high, so lets prevent deep levels of
6400 * recursion to happen by limiting the maximum depth to be 1.
6402 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6403 ret = btrfs_commit_inode_delayed_items(trans, inode);
6408 mutex_lock(&inode->log_mutex);
6411 * For symlinks, we must always log their content, which is stored in an
6412 * inline extent, otherwise we could end up with an empty symlink after
6413 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6414 * one attempts to create an empty symlink).
6415 * We don't need to worry about flushing delalloc, because when we create
6416 * the inline extent when the symlink is created (we never have delalloc
6419 if (S_ISLNK(inode->vfs_inode.i_mode))
6420 inode_only = LOG_INODE_ALL;
6423 * Before logging the inode item, cache the value returned by
6424 * inode_logged(), because after that we have the need to figure out if
6425 * the inode was previously logged in this transaction.
6427 ret = inode_logged(trans, inode, path);
6430 ctx->logged_before = (ret == 1);
6434 * This is for cases where logging a directory could result in losing a
6435 * a file after replaying the log. For example, if we move a file from a
6436 * directory A to a directory B, then fsync directory A, we have no way
6437 * to known the file was moved from A to B, so logging just A would
6438 * result in losing the file after a log replay.
6440 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6441 ret = BTRFS_LOG_FORCE_COMMIT;
6446 * a brute force approach to making sure we get the most uptodate
6447 * copies of everything.
6449 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6450 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6451 if (ctx->logged_before)
6452 ret = drop_inode_items(trans, log, path, inode,
6453 BTRFS_XATTR_ITEM_KEY);
6455 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6457 * Make sure the new inode item we write to the log has
6458 * the same isize as the current one (if it exists).
6459 * This is necessary to prevent data loss after log
6460 * replay, and also to prevent doing a wrong expanding
6461 * truncate - for e.g. create file, write 4K into offset
6462 * 0, fsync, write 4K into offset 4096, add hard link,
6463 * fsync some other file (to sync log), power fail - if
6464 * we use the inode's current i_size, after log replay
6465 * we get a 8Kb file, with the last 4Kb extent as a hole
6466 * (zeroes), as if an expanding truncate happened,
6467 * instead of getting a file of 4Kb only.
6469 ret = logged_inode_size(log, inode, path, &logged_isize);
6473 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6474 &inode->runtime_flags)) {
6475 if (inode_only == LOG_INODE_EXISTS) {
6476 max_key.type = BTRFS_XATTR_ITEM_KEY;
6477 if (ctx->logged_before)
6478 ret = drop_inode_items(trans, log, path,
6479 inode, max_key.type);
6481 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6482 &inode->runtime_flags);
6483 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6484 &inode->runtime_flags);
6485 if (ctx->logged_before)
6486 ret = truncate_inode_items(trans, log,
6489 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6490 &inode->runtime_flags) ||
6491 inode_only == LOG_INODE_EXISTS) {
6492 if (inode_only == LOG_INODE_ALL)
6494 max_key.type = BTRFS_XATTR_ITEM_KEY;
6495 if (ctx->logged_before)
6496 ret = drop_inode_items(trans, log, path, inode,
6499 if (inode_only == LOG_INODE_ALL)
6501 inode_item_dropped = false;
6510 * If we are logging a directory in full mode, collect the delayed items
6511 * before iterating the subvolume tree, so that we don't miss any new
6512 * dir index items in case they get flushed while or right after we are
6513 * iterating the subvolume tree.
6515 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6516 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6519 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6520 path, dst_path, logged_isize,
6522 &need_log_inode_item);
6526 btrfs_release_path(path);
6527 btrfs_release_path(dst_path);
6528 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6531 xattrs_logged = true;
6532 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6533 btrfs_release_path(path);
6534 btrfs_release_path(dst_path);
6535 ret = btrfs_log_holes(trans, inode, path);
6540 btrfs_release_path(path);
6541 btrfs_release_path(dst_path);
6542 if (need_log_inode_item) {
6543 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6547 * If we are doing a fast fsync and the inode was logged before
6548 * in this transaction, we don't need to log the xattrs because
6549 * they were logged before. If xattrs were added, changed or
6550 * deleted since the last time we logged the inode, then we have
6551 * already logged them because the inode had the runtime flag
6552 * BTRFS_INODE_COPY_EVERYTHING set.
6554 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6555 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6558 btrfs_release_path(path);
6562 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6565 } else if (inode_only == LOG_INODE_ALL) {
6566 struct extent_map *em, *n;
6568 write_lock(&em_tree->lock);
6569 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6570 list_del_init(&em->list);
6571 write_unlock(&em_tree->lock);
6574 if (full_dir_logging) {
6575 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6578 ret = log_delayed_insertion_items(trans, inode, path,
6579 &delayed_ins_list, ctx);
6582 ret = log_delayed_deletion_items(trans, inode, path,
6583 &delayed_del_list, ctx);
6588 spin_lock(&inode->lock);
6589 inode->logged_trans = trans->transid;
6591 * Don't update last_log_commit if we logged that an inode exists.
6592 * We do this for three reasons:
6594 * 1) We might have had buffered writes to this inode that were
6595 * flushed and had their ordered extents completed in this
6596 * transaction, but we did not previously log the inode with
6597 * LOG_INODE_ALL. Later the inode was evicted and after that
6598 * it was loaded again and this LOG_INODE_EXISTS log operation
6599 * happened. We must make sure that if an explicit fsync against
6600 * the inode is performed later, it logs the new extents, an
6601 * updated inode item, etc, and syncs the log. The same logic
6602 * applies to direct IO writes instead of buffered writes.
6604 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6605 * is logged with an i_size of 0 or whatever value was logged
6606 * before. If later the i_size of the inode is increased by a
6607 * truncate operation, the log is synced through an fsync of
6608 * some other inode and then finally an explicit fsync against
6609 * this inode is made, we must make sure this fsync logs the
6610 * inode with the new i_size, the hole between old i_size and
6611 * the new i_size, and syncs the log.
6613 * 3) If we are logging that an ancestor inode exists as part of
6614 * logging a new name from a link or rename operation, don't update
6615 * its last_log_commit - otherwise if an explicit fsync is made
6616 * against an ancestor, the fsync considers the inode in the log
6617 * and doesn't sync the log, resulting in the ancestor missing after
6618 * a power failure unless the log was synced as part of an fsync
6619 * against any other unrelated inode.
6621 if (inode_only != LOG_INODE_EXISTS)
6622 inode->last_log_commit = inode->last_sub_trans;
6623 spin_unlock(&inode->lock);
6626 * Reset the last_reflink_trans so that the next fsync does not need to
6627 * go through the slower path when logging extents and their checksums.
6629 if (inode_only == LOG_INODE_ALL)
6630 inode->last_reflink_trans = 0;
6633 mutex_unlock(&inode->log_mutex);
6635 btrfs_free_path(path);
6636 btrfs_free_path(dst_path);
6639 free_conflicting_inodes(ctx);
6641 ret = log_conflicting_inodes(trans, inode->root, ctx);
6643 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6645 ret = log_new_delayed_dentries(trans, inode,
6646 &delayed_ins_list, ctx);
6648 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6655 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6656 struct btrfs_inode *inode,
6657 struct btrfs_log_ctx *ctx)
6659 struct btrfs_fs_info *fs_info = trans->fs_info;
6661 struct btrfs_path *path;
6662 struct btrfs_key key;
6663 struct btrfs_root *root = inode->root;
6664 const u64 ino = btrfs_ino(inode);
6666 path = btrfs_alloc_path();
6669 path->skip_locking = 1;
6670 path->search_commit_root = 1;
6673 key.type = BTRFS_INODE_REF_KEY;
6675 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6680 struct extent_buffer *leaf = path->nodes[0];
6681 int slot = path->slots[0];
6686 if (slot >= btrfs_header_nritems(leaf)) {
6687 ret = btrfs_next_leaf(root, path);
6695 btrfs_item_key_to_cpu(leaf, &key, slot);
6696 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6697 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6700 item_size = btrfs_item_size(leaf, slot);
6701 ptr = btrfs_item_ptr_offset(leaf, slot);
6702 while (cur_offset < item_size) {
6703 struct btrfs_key inode_key;
6704 struct inode *dir_inode;
6706 inode_key.type = BTRFS_INODE_ITEM_KEY;
6707 inode_key.offset = 0;
6709 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6710 struct btrfs_inode_extref *extref;
6712 extref = (struct btrfs_inode_extref *)
6714 inode_key.objectid = btrfs_inode_extref_parent(
6716 cur_offset += sizeof(*extref);
6717 cur_offset += btrfs_inode_extref_name_len(leaf,
6720 inode_key.objectid = key.offset;
6721 cur_offset = item_size;
6724 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6727 * If the parent inode was deleted, return an error to
6728 * fallback to a transaction commit. This is to prevent
6729 * getting an inode that was moved from one parent A to
6730 * a parent B, got its former parent A deleted and then
6731 * it got fsync'ed, from existing at both parents after
6732 * a log replay (and the old parent still existing).
6739 * mv /mnt/B/bar /mnt/A/bar
6740 * mv -T /mnt/A /mnt/B
6744 * If we ignore the old parent B which got deleted,
6745 * after a log replay we would have file bar linked
6746 * at both parents and the old parent B would still
6749 if (IS_ERR(dir_inode)) {
6750 ret = PTR_ERR(dir_inode);
6754 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6755 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6759 ctx->log_new_dentries = false;
6760 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6761 LOG_INODE_ALL, ctx);
6762 if (!ret && ctx->log_new_dentries)
6763 ret = log_new_dir_dentries(trans,
6764 BTRFS_I(dir_inode), ctx);
6765 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6773 btrfs_free_path(path);
6777 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6778 struct btrfs_root *root,
6779 struct btrfs_path *path,
6780 struct btrfs_log_ctx *ctx)
6782 struct btrfs_key found_key;
6784 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6787 struct btrfs_fs_info *fs_info = root->fs_info;
6788 struct extent_buffer *leaf;
6790 struct btrfs_key search_key;
6791 struct inode *inode;
6795 btrfs_release_path(path);
6797 ino = found_key.offset;
6799 search_key.objectid = found_key.offset;
6800 search_key.type = BTRFS_INODE_ITEM_KEY;
6801 search_key.offset = 0;
6802 inode = btrfs_iget(fs_info->sb, ino, root);
6804 return PTR_ERR(inode);
6806 if (BTRFS_I(inode)->generation >= trans->transid &&
6807 need_log_inode(trans, BTRFS_I(inode)))
6808 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6809 LOG_INODE_EXISTS, ctx);
6810 btrfs_add_delayed_iput(BTRFS_I(inode));
6814 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6817 search_key.type = BTRFS_INODE_REF_KEY;
6818 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6822 leaf = path->nodes[0];
6823 slot = path->slots[0];
6824 if (slot >= btrfs_header_nritems(leaf)) {
6825 ret = btrfs_next_leaf(root, path);
6830 leaf = path->nodes[0];
6831 slot = path->slots[0];
6834 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6835 if (found_key.objectid != search_key.objectid ||
6836 found_key.type != BTRFS_INODE_REF_KEY)
6842 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6843 struct btrfs_inode *inode,
6844 struct dentry *parent,
6845 struct btrfs_log_ctx *ctx)
6847 struct btrfs_root *root = inode->root;
6848 struct dentry *old_parent = NULL;
6849 struct super_block *sb = inode->vfs_inode.i_sb;
6853 if (!parent || d_really_is_negative(parent) ||
6857 inode = BTRFS_I(d_inode(parent));
6858 if (root != inode->root)
6861 if (inode->generation >= trans->transid &&
6862 need_log_inode(trans, inode)) {
6863 ret = btrfs_log_inode(trans, inode,
6864 LOG_INODE_EXISTS, ctx);
6868 if (IS_ROOT(parent))
6871 parent = dget_parent(parent);
6873 old_parent = parent;
6880 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6881 struct btrfs_inode *inode,
6882 struct dentry *parent,
6883 struct btrfs_log_ctx *ctx)
6885 struct btrfs_root *root = inode->root;
6886 const u64 ino = btrfs_ino(inode);
6887 struct btrfs_path *path;
6888 struct btrfs_key search_key;
6892 * For a single hard link case, go through a fast path that does not
6893 * need to iterate the fs/subvolume tree.
6895 if (inode->vfs_inode.i_nlink < 2)
6896 return log_new_ancestors_fast(trans, inode, parent, ctx);
6898 path = btrfs_alloc_path();
6902 search_key.objectid = ino;
6903 search_key.type = BTRFS_INODE_REF_KEY;
6904 search_key.offset = 0;
6906 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6913 struct extent_buffer *leaf = path->nodes[0];
6914 int slot = path->slots[0];
6915 struct btrfs_key found_key;
6917 if (slot >= btrfs_header_nritems(leaf)) {
6918 ret = btrfs_next_leaf(root, path);
6926 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6927 if (found_key.objectid != ino ||
6928 found_key.type > BTRFS_INODE_EXTREF_KEY)
6932 * Don't deal with extended references because they are rare
6933 * cases and too complex to deal with (we would need to keep
6934 * track of which subitem we are processing for each item in
6935 * this loop, etc). So just return some error to fallback to
6936 * a transaction commit.
6938 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6944 * Logging ancestors needs to do more searches on the fs/subvol
6945 * tree, so it releases the path as needed to avoid deadlocks.
6946 * Keep track of the last inode ref key and resume from that key
6947 * after logging all new ancestors for the current hard link.
6949 memcpy(&search_key, &found_key, sizeof(search_key));
6951 ret = log_new_ancestors(trans, root, path, ctx);
6954 btrfs_release_path(path);
6959 btrfs_free_path(path);
6964 * helper function around btrfs_log_inode to make sure newly created
6965 * parent directories also end up in the log. A minimal inode and backref
6966 * only logging is done of any parent directories that are older than
6967 * the last committed transaction
6969 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6970 struct btrfs_inode *inode,
6971 struct dentry *parent,
6973 struct btrfs_log_ctx *ctx)
6975 struct btrfs_root *root = inode->root;
6976 struct btrfs_fs_info *fs_info = root->fs_info;
6978 bool log_dentries = false;
6980 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6981 ret = BTRFS_LOG_FORCE_COMMIT;
6985 if (btrfs_root_refs(&root->root_item) == 0) {
6986 ret = BTRFS_LOG_FORCE_COMMIT;
6991 * Skip already logged inodes or inodes corresponding to tmpfiles
6992 * (since logging them is pointless, a link count of 0 means they
6993 * will never be accessible).
6995 if ((btrfs_inode_in_log(inode, trans->transid) &&
6996 list_empty(&ctx->ordered_extents)) ||
6997 inode->vfs_inode.i_nlink == 0) {
6998 ret = BTRFS_NO_LOG_SYNC;
7002 ret = start_log_trans(trans, root, ctx);
7006 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
7011 * for regular files, if its inode is already on disk, we don't
7012 * have to worry about the parents at all. This is because
7013 * we can use the last_unlink_trans field to record renames
7014 * and other fun in this file.
7016 if (S_ISREG(inode->vfs_inode.i_mode) &&
7017 inode->generation < trans->transid &&
7018 inode->last_unlink_trans < trans->transid) {
7023 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7024 log_dentries = true;
7027 * On unlink we must make sure all our current and old parent directory
7028 * inodes are fully logged. This is to prevent leaving dangling
7029 * directory index entries in directories that were our parents but are
7030 * not anymore. Not doing this results in old parent directory being
7031 * impossible to delete after log replay (rmdir will always fail with
7032 * error -ENOTEMPTY).
7038 * ln testdir/foo testdir/bar
7040 * unlink testdir/bar
7041 * xfs_io -c fsync testdir/foo
7043 * mount fs, triggers log replay
7045 * If we don't log the parent directory (testdir), after log replay the
7046 * directory still has an entry pointing to the file inode using the bar
7047 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7048 * the file inode has a link count of 1.
7054 * ln foo testdir/foo2
7055 * ln foo testdir/foo3
7057 * unlink testdir/foo3
7058 * xfs_io -c fsync foo
7060 * mount fs, triggers log replay
7062 * Similar as the first example, after log replay the parent directory
7063 * testdir still has an entry pointing to the inode file with name foo3
7064 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7065 * and has a link count of 2.
7067 if (inode->last_unlink_trans >= trans->transid) {
7068 ret = btrfs_log_all_parents(trans, inode, ctx);
7073 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7078 ret = log_new_dir_dentries(trans, inode, ctx);
7083 btrfs_set_log_full_commit(trans);
7084 ret = BTRFS_LOG_FORCE_COMMIT;
7088 btrfs_remove_log_ctx(root, ctx);
7089 btrfs_end_log_trans(root);
7095 * it is not safe to log dentry if the chunk root has added new
7096 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7097 * If this returns 1, you must commit the transaction to safely get your
7100 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7101 struct dentry *dentry,
7102 struct btrfs_log_ctx *ctx)
7104 struct dentry *parent = dget_parent(dentry);
7107 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7108 LOG_INODE_ALL, ctx);
7115 * should be called during mount to recover any replay any log trees
7118 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7121 struct btrfs_path *path;
7122 struct btrfs_trans_handle *trans;
7123 struct btrfs_key key;
7124 struct btrfs_key found_key;
7125 struct btrfs_root *log;
7126 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7127 struct walk_control wc = {
7128 .process_func = process_one_buffer,
7129 .stage = LOG_WALK_PIN_ONLY,
7132 path = btrfs_alloc_path();
7136 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7138 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7139 if (IS_ERR(trans)) {
7140 ret = PTR_ERR(trans);
7147 ret = walk_log_tree(trans, log_root_tree, &wc);
7149 btrfs_abort_transaction(trans, ret);
7154 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7155 key.offset = (u64)-1;
7156 key.type = BTRFS_ROOT_ITEM_KEY;
7159 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7162 btrfs_abort_transaction(trans, ret);
7166 if (path->slots[0] == 0)
7170 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7172 btrfs_release_path(path);
7173 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7176 log = btrfs_read_tree_root(log_root_tree, &found_key);
7179 btrfs_abort_transaction(trans, ret);
7183 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7185 if (IS_ERR(wc.replay_dest)) {
7186 ret = PTR_ERR(wc.replay_dest);
7189 * We didn't find the subvol, likely because it was
7190 * deleted. This is ok, simply skip this log and go to
7193 * We need to exclude the root because we can't have
7194 * other log replays overwriting this log as we'll read
7195 * it back in a few more times. This will keep our
7196 * block from being modified, and we'll just bail for
7197 * each subsequent pass.
7200 ret = btrfs_pin_extent_for_log_replay(trans, log->node);
7201 btrfs_put_root(log);
7205 btrfs_abort_transaction(trans, ret);
7209 wc.replay_dest->log_root = log;
7210 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7212 /* The loop needs to continue due to the root refs */
7213 btrfs_abort_transaction(trans, ret);
7215 ret = walk_log_tree(trans, log, &wc);
7217 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7218 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7221 btrfs_abort_transaction(trans, ret);
7224 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7225 struct btrfs_root *root = wc.replay_dest;
7227 btrfs_release_path(path);
7230 * We have just replayed everything, and the highest
7231 * objectid of fs roots probably has changed in case
7232 * some inode_item's got replayed.
7234 * root->objectid_mutex is not acquired as log replay
7235 * could only happen during mount.
7237 ret = btrfs_init_root_free_objectid(root);
7239 btrfs_abort_transaction(trans, ret);
7242 wc.replay_dest->log_root = NULL;
7243 btrfs_put_root(wc.replay_dest);
7244 btrfs_put_root(log);
7249 if (found_key.offset == 0)
7251 key.offset = found_key.offset - 1;
7253 btrfs_release_path(path);
7255 /* step one is to pin it all, step two is to replay just inodes */
7258 wc.process_func = replay_one_buffer;
7259 wc.stage = LOG_WALK_REPLAY_INODES;
7262 /* step three is to replay everything */
7263 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7268 btrfs_free_path(path);
7270 /* step 4: commit the transaction, which also unpins the blocks */
7271 ret = btrfs_commit_transaction(trans);
7275 log_root_tree->log_root = NULL;
7276 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7277 btrfs_put_root(log_root_tree);
7282 btrfs_end_transaction(wc.trans);
7283 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7284 btrfs_free_path(path);
7289 * there are some corner cases where we want to force a full
7290 * commit instead of allowing a directory to be logged.
7292 * They revolve around files there were unlinked from the directory, and
7293 * this function updates the parent directory so that a full commit is
7294 * properly done if it is fsync'd later after the unlinks are done.
7296 * Must be called before the unlink operations (updates to the subvolume tree,
7297 * inodes, etc) are done.
7299 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7300 struct btrfs_inode *dir, struct btrfs_inode *inode,
7304 * when we're logging a file, if it hasn't been renamed
7305 * or unlinked, and its inode is fully committed on disk,
7306 * we don't have to worry about walking up the directory chain
7307 * to log its parents.
7309 * So, we use the last_unlink_trans field to put this transid
7310 * into the file. When the file is logged we check it and
7311 * don't log the parents if the file is fully on disk.
7313 mutex_lock(&inode->log_mutex);
7314 inode->last_unlink_trans = trans->transid;
7315 mutex_unlock(&inode->log_mutex);
7321 * If this directory was already logged, any new names will be logged
7322 * with btrfs_log_new_name() and old names will be deleted from the log
7323 * tree with btrfs_del_dir_entries_in_log() or with
7324 * btrfs_del_inode_ref_in_log().
7326 if (inode_logged(trans, dir, NULL) == 1)
7330 * If the inode we're about to unlink was logged before, the log will be
7331 * properly updated with the new name with btrfs_log_new_name() and the
7332 * old name removed with btrfs_del_dir_entries_in_log() or with
7333 * btrfs_del_inode_ref_in_log().
7335 if (inode_logged(trans, inode, NULL) == 1)
7339 * when renaming files across directories, if the directory
7340 * there we're unlinking from gets fsync'd later on, there's
7341 * no way to find the destination directory later and fsync it
7342 * properly. So, we have to be conservative and force commits
7343 * so the new name gets discovered.
7345 mutex_lock(&dir->log_mutex);
7346 dir->last_unlink_trans = trans->transid;
7347 mutex_unlock(&dir->log_mutex);
7351 * Make sure that if someone attempts to fsync the parent directory of a deleted
7352 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7353 * that after replaying the log tree of the parent directory's root we will not
7354 * see the snapshot anymore and at log replay time we will not see any log tree
7355 * corresponding to the deleted snapshot's root, which could lead to replaying
7356 * it after replaying the log tree of the parent directory (which would replay
7357 * the snapshot delete operation).
7359 * Must be called before the actual snapshot destroy operation (updates to the
7360 * parent root and tree of tree roots trees, etc) are done.
7362 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7363 struct btrfs_inode *dir)
7365 mutex_lock(&dir->log_mutex);
7366 dir->last_unlink_trans = trans->transid;
7367 mutex_unlock(&dir->log_mutex);
7371 * Update the log after adding a new name for an inode.
7373 * @trans: Transaction handle.
7374 * @old_dentry: The dentry associated with the old name and the old
7376 * @old_dir: The inode of the previous parent directory for the case
7377 * of a rename. For a link operation, it must be NULL.
7378 * @old_dir_index: The index number associated with the old name, meaningful
7379 * only for rename operations (when @old_dir is not NULL).
7380 * Ignored for link operations.
7381 * @parent: The dentry associated with the directory under which the
7382 * new name is located.
7384 * Call this after adding a new name for an inode, as a result of a link or
7385 * rename operation, and it will properly update the log to reflect the new name.
7387 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7388 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7389 u64 old_dir_index, struct dentry *parent)
7391 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7392 struct btrfs_root *root = inode->root;
7393 struct btrfs_log_ctx ctx;
7394 bool log_pinned = false;
7398 * this will force the logging code to walk the dentry chain
7401 if (!S_ISDIR(inode->vfs_inode.i_mode))
7402 inode->last_unlink_trans = trans->transid;
7405 * if this inode hasn't been logged and directory we're renaming it
7406 * from hasn't been logged, we don't need to log it
7408 ret = inode_logged(trans, inode, NULL);
7411 } else if (ret == 0) {
7415 * If the inode was not logged and we are doing a rename (old_dir is not
7416 * NULL), check if old_dir was logged - if it was not we can return and
7419 ret = inode_logged(trans, old_dir, NULL);
7428 * If we are doing a rename (old_dir is not NULL) from a directory that
7429 * was previously logged, make sure that on log replay we get the old
7430 * dir entry deleted. This is needed because we will also log the new
7431 * name of the renamed inode, so we need to make sure that after log
7432 * replay we don't end up with both the new and old dir entries existing.
7434 if (old_dir && old_dir->logged_trans == trans->transid) {
7435 struct btrfs_root *log = old_dir->root->log_root;
7436 struct btrfs_path *path;
7437 struct fscrypt_name fname;
7439 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7441 ret = fscrypt_setup_filename(&old_dir->vfs_inode,
7442 &old_dentry->d_name, 0, &fname);
7446 * We have two inodes to update in the log, the old directory and
7447 * the inode that got renamed, so we must pin the log to prevent
7448 * anyone from syncing the log until we have updated both inodes
7451 ret = join_running_log_trans(root);
7453 * At least one of the inodes was logged before, so this should
7454 * not fail, but if it does, it's not serious, just bail out and
7455 * mark the log for a full commit.
7457 if (WARN_ON_ONCE(ret < 0)) {
7458 fscrypt_free_filename(&fname);
7464 path = btrfs_alloc_path();
7467 fscrypt_free_filename(&fname);
7472 * Other concurrent task might be logging the old directory,
7473 * as it can be triggered when logging other inode that had or
7474 * still has a dentry in the old directory. We lock the old
7475 * directory's log_mutex to ensure the deletion of the old
7476 * name is persisted, because during directory logging we
7477 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7478 * the old name's dir index item is in the delayed items, so
7479 * it could be missed by an in progress directory logging.
7481 mutex_lock(&old_dir->log_mutex);
7482 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7483 &fname.disk_name, old_dir_index);
7486 * The dentry does not exist in the log, so record its
7489 btrfs_release_path(path);
7490 ret = insert_dir_log_key(trans, log, path,
7492 old_dir_index, old_dir_index);
7494 mutex_unlock(&old_dir->log_mutex);
7496 btrfs_free_path(path);
7497 fscrypt_free_filename(&fname);
7502 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7503 ctx.logging_new_name = true;
7505 * We don't care about the return value. If we fail to log the new name
7506 * then we know the next attempt to sync the log will fallback to a full
7507 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7508 * we don't need to worry about getting a log committed that has an
7509 * inconsistent state after a rename operation.
7511 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7512 ASSERT(list_empty(&ctx.conflict_inodes));
7515 * If an error happened mark the log for a full commit because it's not
7516 * consistent and up to date or we couldn't find out if one of the
7517 * inodes was logged before in this transaction. Do it before unpinning
7518 * the log, to avoid any races with someone else trying to commit it.
7521 btrfs_set_log_full_commit(trans);
7523 btrfs_end_log_trans(root);