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->start,
355 if (btrfs_buffer_uptodate(eb, gen, 0) &&
356 btrfs_header_level(eb) == 0)
357 ret = btrfs_exclude_logged_extents(eb);
363 * Item overwrite used by replay and tree logging. eb, slot and key all refer
364 * to the src data we are copying out.
366 * root is the tree we are copying into, and path is a scratch
367 * path for use in this function (it should be released on entry and
368 * will be released on exit).
370 * If the key is already in the destination tree the existing item is
371 * overwritten. If the existing item isn't big enough, it is extended.
372 * If it is too large, it is truncated.
374 * If the key isn't in the destination yet, a new item is inserted.
376 static int overwrite_item(struct btrfs_trans_handle *trans,
377 struct btrfs_root *root,
378 struct btrfs_path *path,
379 struct extent_buffer *eb, int slot,
380 struct btrfs_key *key)
384 u64 saved_i_size = 0;
385 int save_old_i_size = 0;
386 unsigned long src_ptr;
387 unsigned long dst_ptr;
388 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
391 * This is only used during log replay, so the root is always from a
392 * fs/subvolume tree. In case we ever need to support a log root, then
393 * we'll have to clone the leaf in the path, release the path and use
394 * the leaf before writing into the log tree. See the comments at
395 * copy_items() for more details.
397 ASSERT(root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
399 item_size = btrfs_item_size(eb, slot);
400 src_ptr = btrfs_item_ptr_offset(eb, slot);
402 /* Look for the key in the destination tree. */
403 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
410 u32 dst_size = btrfs_item_size(path->nodes[0],
412 if (dst_size != item_size)
415 if (item_size == 0) {
416 btrfs_release_path(path);
419 dst_copy = kmalloc(item_size, GFP_NOFS);
420 src_copy = kmalloc(item_size, GFP_NOFS);
421 if (!dst_copy || !src_copy) {
422 btrfs_release_path(path);
428 read_extent_buffer(eb, src_copy, src_ptr, item_size);
430 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
431 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
433 ret = memcmp(dst_copy, src_copy, item_size);
438 * they have the same contents, just return, this saves
439 * us from cowing blocks in the destination tree and doing
440 * extra writes that may not have been done by a previous
444 btrfs_release_path(path);
449 * We need to load the old nbytes into the inode so when we
450 * replay the extents we've logged we get the right nbytes.
453 struct btrfs_inode_item *item;
457 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
458 struct btrfs_inode_item);
459 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
460 item = btrfs_item_ptr(eb, slot,
461 struct btrfs_inode_item);
462 btrfs_set_inode_nbytes(eb, item, nbytes);
465 * If this is a directory we need to reset the i_size to
466 * 0 so that we can set it up properly when replaying
467 * the rest of the items in this log.
469 mode = btrfs_inode_mode(eb, item);
471 btrfs_set_inode_size(eb, item, 0);
473 } else if (inode_item) {
474 struct btrfs_inode_item *item;
478 * New inode, set nbytes to 0 so that the nbytes comes out
479 * properly when we replay the extents.
481 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
482 btrfs_set_inode_nbytes(eb, item, 0);
485 * If this is a directory we need to reset the i_size to 0 so
486 * that we can set it up properly when replaying the rest of
487 * the items in this log.
489 mode = btrfs_inode_mode(eb, item);
491 btrfs_set_inode_size(eb, item, 0);
494 btrfs_release_path(path);
495 /* try to insert the key into the destination tree */
496 path->skip_release_on_error = 1;
497 ret = btrfs_insert_empty_item(trans, root, path,
499 path->skip_release_on_error = 0;
501 /* make sure any existing item is the correct size */
502 if (ret == -EEXIST || ret == -EOVERFLOW) {
504 found_size = btrfs_item_size(path->nodes[0],
506 if (found_size > item_size)
507 btrfs_truncate_item(path, item_size, 1);
508 else if (found_size < item_size)
509 btrfs_extend_item(path, item_size - found_size);
513 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
516 /* don't overwrite an existing inode if the generation number
517 * was logged as zero. This is done when the tree logging code
518 * is just logging an inode to make sure it exists after recovery.
520 * Also, don't overwrite i_size on directories during replay.
521 * log replay inserts and removes directory items based on the
522 * state of the tree found in the subvolume, and i_size is modified
525 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
526 struct btrfs_inode_item *src_item;
527 struct btrfs_inode_item *dst_item;
529 src_item = (struct btrfs_inode_item *)src_ptr;
530 dst_item = (struct btrfs_inode_item *)dst_ptr;
532 if (btrfs_inode_generation(eb, src_item) == 0) {
533 struct extent_buffer *dst_eb = path->nodes[0];
534 const u64 ino_size = btrfs_inode_size(eb, src_item);
537 * For regular files an ino_size == 0 is used only when
538 * logging that an inode exists, as part of a directory
539 * fsync, and the inode wasn't fsynced before. In this
540 * case don't set the size of the inode in the fs/subvol
541 * tree, otherwise we would be throwing valid data away.
543 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
544 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
546 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
550 if (S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
551 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
553 saved_i_size = btrfs_inode_size(path->nodes[0],
558 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
561 if (save_old_i_size) {
562 struct btrfs_inode_item *dst_item;
563 dst_item = (struct btrfs_inode_item *)dst_ptr;
564 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
567 /* make sure the generation is filled in */
568 if (key->type == BTRFS_INODE_ITEM_KEY) {
569 struct btrfs_inode_item *dst_item;
570 dst_item = (struct btrfs_inode_item *)dst_ptr;
571 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
572 btrfs_set_inode_generation(path->nodes[0], dst_item,
577 btrfs_mark_buffer_dirty(path->nodes[0]);
578 btrfs_release_path(path);
582 static int read_alloc_one_name(struct extent_buffer *eb, void *start, int len,
583 struct fscrypt_str *name)
587 buf = kmalloc(len, GFP_NOFS);
591 read_extent_buffer(eb, buf, (unsigned long)start, len);
598 * simple helper to read an inode off the disk from a given root
599 * This can only be called for subvolume roots and not for the log
601 static noinline struct inode *read_one_inode(struct btrfs_root *root,
606 inode = btrfs_iget(root->fs_info->sb, objectid, root);
612 /* replays a single extent in 'eb' at 'slot' with 'key' into the
613 * subvolume 'root'. path is released on entry and should be released
616 * extents in the log tree have not been allocated out of the extent
617 * tree yet. So, this completes the allocation, taking a reference
618 * as required if the extent already exists or creating a new extent
619 * if it isn't in the extent allocation tree yet.
621 * The extent is inserted into the file, dropping any existing extents
622 * from the file that overlap the new one.
624 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
625 struct btrfs_root *root,
626 struct btrfs_path *path,
627 struct extent_buffer *eb, int slot,
628 struct btrfs_key *key)
630 struct btrfs_drop_extents_args drop_args = { 0 };
631 struct btrfs_fs_info *fs_info = root->fs_info;
634 u64 start = key->offset;
636 struct btrfs_file_extent_item *item;
637 struct inode *inode = NULL;
641 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
642 found_type = btrfs_file_extent_type(eb, item);
644 if (found_type == BTRFS_FILE_EXTENT_REG ||
645 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
646 nbytes = btrfs_file_extent_num_bytes(eb, item);
647 extent_end = start + nbytes;
650 * We don't add to the inodes nbytes if we are prealloc or a
653 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
655 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
656 size = btrfs_file_extent_ram_bytes(eb, item);
657 nbytes = btrfs_file_extent_ram_bytes(eb, item);
658 extent_end = ALIGN(start + size,
659 fs_info->sectorsize);
665 inode = read_one_inode(root, key->objectid);
672 * first check to see if we already have this extent in the
673 * file. This must be done before the btrfs_drop_extents run
674 * so we don't try to drop this extent.
676 ret = btrfs_lookup_file_extent(trans, root, path,
677 btrfs_ino(BTRFS_I(inode)), start, 0);
680 (found_type == BTRFS_FILE_EXTENT_REG ||
681 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
682 struct btrfs_file_extent_item cmp1;
683 struct btrfs_file_extent_item cmp2;
684 struct btrfs_file_extent_item *existing;
685 struct extent_buffer *leaf;
687 leaf = path->nodes[0];
688 existing = btrfs_item_ptr(leaf, path->slots[0],
689 struct btrfs_file_extent_item);
691 read_extent_buffer(eb, &cmp1, (unsigned long)item,
693 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
697 * we already have a pointer to this exact extent,
698 * we don't have to do anything
700 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
701 btrfs_release_path(path);
705 btrfs_release_path(path);
707 /* drop any overlapping extents */
708 drop_args.start = start;
709 drop_args.end = extent_end;
710 drop_args.drop_cache = true;
711 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
715 if (found_type == BTRFS_FILE_EXTENT_REG ||
716 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
718 unsigned long dest_offset;
719 struct btrfs_key ins;
721 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
722 btrfs_fs_incompat(fs_info, NO_HOLES))
725 ret = btrfs_insert_empty_item(trans, root, path, key,
729 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
731 copy_extent_buffer(path->nodes[0], eb, dest_offset,
732 (unsigned long)item, sizeof(*item));
734 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
735 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
736 ins.type = BTRFS_EXTENT_ITEM_KEY;
737 offset = key->offset - btrfs_file_extent_offset(eb, item);
740 * Manually record dirty extent, as here we did a shallow
741 * file extent item copy and skip normal backref update,
742 * but modifying extent tree all by ourselves.
743 * So need to manually record dirty extent for qgroup,
744 * as the owner of the file extent changed from log tree
745 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
747 ret = btrfs_qgroup_trace_extent(trans,
748 btrfs_file_extent_disk_bytenr(eb, item),
749 btrfs_file_extent_disk_num_bytes(eb, item));
753 if (ins.objectid > 0) {
754 struct btrfs_ref ref = { 0 };
757 LIST_HEAD(ordered_sums);
760 * is this extent already allocated in the extent
761 * allocation tree? If so, just add a reference
763 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
767 } else if (ret == 0) {
768 btrfs_init_generic_ref(&ref,
769 BTRFS_ADD_DELAYED_REF,
770 ins.objectid, ins.offset, 0);
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, root, 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, root, 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_root *root,
1487 struct btrfs_inode *inode, struct btrfs_path *path)
1491 unsigned int nlink = 0;
1494 u64 inode_objectid = btrfs_ino(inode);
1497 struct btrfs_inode_extref *extref;
1498 struct extent_buffer *leaf;
1501 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1506 leaf = path->nodes[0];
1507 item_size = btrfs_item_size(leaf, path->slots[0]);
1508 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1511 while (cur_offset < item_size) {
1512 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1513 name_len = btrfs_inode_extref_name_len(leaf, extref);
1517 cur_offset += name_len + sizeof(*extref);
1521 btrfs_release_path(path);
1523 btrfs_release_path(path);
1525 if (ret < 0 && ret != -ENOENT)
1530 static int count_inode_refs(struct btrfs_root *root,
1531 struct btrfs_inode *inode, struct btrfs_path *path)
1534 struct btrfs_key key;
1535 unsigned int nlink = 0;
1537 unsigned long ptr_end;
1539 u64 ino = btrfs_ino(inode);
1542 key.type = BTRFS_INODE_REF_KEY;
1543 key.offset = (u64)-1;
1546 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1550 if (path->slots[0] == 0)
1555 btrfs_item_key_to_cpu(path->nodes[0], &key,
1557 if (key.objectid != ino ||
1558 key.type != BTRFS_INODE_REF_KEY)
1560 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1561 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1563 while (ptr < ptr_end) {
1564 struct btrfs_inode_ref *ref;
1566 ref = (struct btrfs_inode_ref *)ptr;
1567 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1569 ptr = (unsigned long)(ref + 1) + name_len;
1573 if (key.offset == 0)
1575 if (path->slots[0] > 0) {
1580 btrfs_release_path(path);
1582 btrfs_release_path(path);
1588 * There are a few corners where the link count of the file can't
1589 * be properly maintained during replay. So, instead of adding
1590 * lots of complexity to the log code, we just scan the backrefs
1591 * for any file that has been through replay.
1593 * The scan will update the link count on the inode to reflect the
1594 * number of back refs found. If it goes down to zero, the iput
1595 * will free the inode.
1597 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 struct inode *inode)
1601 struct btrfs_path *path;
1604 u64 ino = btrfs_ino(BTRFS_I(inode));
1606 path = btrfs_alloc_path();
1610 ret = count_inode_refs(root, BTRFS_I(inode), path);
1616 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1624 if (nlink != inode->i_nlink) {
1625 set_nlink(inode, nlink);
1626 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1630 BTRFS_I(inode)->index_cnt = (u64)-1;
1632 if (inode->i_nlink == 0) {
1633 if (S_ISDIR(inode->i_mode)) {
1634 ret = replay_dir_deletes(trans, root, NULL, path,
1639 ret = btrfs_insert_orphan_item(trans, root, ino);
1645 btrfs_free_path(path);
1649 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1650 struct btrfs_root *root,
1651 struct btrfs_path *path)
1654 struct btrfs_key key;
1655 struct inode *inode;
1657 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1658 key.type = BTRFS_ORPHAN_ITEM_KEY;
1659 key.offset = (u64)-1;
1661 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1667 if (path->slots[0] == 0)
1672 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1673 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1674 key.type != BTRFS_ORPHAN_ITEM_KEY)
1677 ret = btrfs_del_item(trans, root, path);
1681 btrfs_release_path(path);
1682 inode = read_one_inode(root, key.offset);
1688 ret = fixup_inode_link_count(trans, root, inode);
1694 * fixup on a directory may create new entries,
1695 * make sure we always look for the highset possible
1698 key.offset = (u64)-1;
1700 btrfs_release_path(path);
1706 * record a given inode in the fixup dir so we can check its link
1707 * count when replay is done. The link count is incremented here
1708 * so the inode won't go away until we check it
1710 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1711 struct btrfs_root *root,
1712 struct btrfs_path *path,
1715 struct btrfs_key key;
1717 struct inode *inode;
1719 inode = read_one_inode(root, objectid);
1723 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1724 key.type = BTRFS_ORPHAN_ITEM_KEY;
1725 key.offset = objectid;
1727 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1729 btrfs_release_path(path);
1731 if (!inode->i_nlink)
1732 set_nlink(inode, 1);
1735 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1736 } else if (ret == -EEXIST) {
1745 * when replaying the log for a directory, we only insert names
1746 * for inodes that actually exist. This means an fsync on a directory
1747 * does not implicitly fsync all the new files in it
1749 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1750 struct btrfs_root *root,
1751 u64 dirid, u64 index,
1752 const struct fscrypt_str *name,
1753 struct btrfs_key *location)
1755 struct inode *inode;
1759 inode = read_one_inode(root, location->objectid);
1763 dir = read_one_inode(root, dirid);
1769 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1772 /* FIXME, put inode into FIXUP list */
1779 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1780 struct btrfs_inode *dir,
1781 struct btrfs_path *path,
1782 struct btrfs_dir_item *dst_di,
1783 const struct btrfs_key *log_key,
1787 struct btrfs_key found_key;
1789 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1790 /* The existing dentry points to the same inode, don't delete it. */
1791 if (found_key.objectid == log_key->objectid &&
1792 found_key.type == log_key->type &&
1793 found_key.offset == log_key->offset &&
1794 btrfs_dir_flags(path->nodes[0], dst_di) == log_flags)
1798 * Don't drop the conflicting directory entry if the inode for the new
1799 * entry doesn't exist.
1804 return drop_one_dir_item(trans, path, dir, dst_di);
1808 * take a single entry in a log directory item and replay it into
1811 * if a conflicting item exists in the subdirectory already,
1812 * the inode it points to is unlinked and put into the link count
1815 * If a name from the log points to a file or directory that does
1816 * not exist in the FS, it is skipped. fsyncs on directories
1817 * do not force down inodes inside that directory, just changes to the
1818 * names or unlinks in a directory.
1820 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1821 * non-existing inode) and 1 if the name was replayed.
1823 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *root,
1825 struct btrfs_path *path,
1826 struct extent_buffer *eb,
1827 struct btrfs_dir_item *di,
1828 struct btrfs_key *key)
1830 struct fscrypt_str name;
1831 struct btrfs_dir_item *dir_dst_di;
1832 struct btrfs_dir_item *index_dst_di;
1833 bool dir_dst_matches = false;
1834 bool index_dst_matches = false;
1835 struct btrfs_key log_key;
1836 struct btrfs_key search_key;
1841 bool update_size = true;
1842 bool name_added = false;
1844 dir = read_one_inode(root, key->objectid);
1848 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
1852 log_flags = btrfs_dir_flags(eb, di);
1853 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1854 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1855 btrfs_release_path(path);
1858 exists = (ret == 0);
1861 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1863 if (IS_ERR(dir_dst_di)) {
1864 ret = PTR_ERR(dir_dst_di);
1866 } else if (dir_dst_di) {
1867 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1868 dir_dst_di, &log_key,
1872 dir_dst_matches = (ret == 1);
1875 btrfs_release_path(path);
1877 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1878 key->objectid, key->offset,
1880 if (IS_ERR(index_dst_di)) {
1881 ret = PTR_ERR(index_dst_di);
1883 } else if (index_dst_di) {
1884 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1885 index_dst_di, &log_key,
1889 index_dst_matches = (ret == 1);
1892 btrfs_release_path(path);
1894 if (dir_dst_matches && index_dst_matches) {
1896 update_size = false;
1901 * Check if the inode reference exists in the log for the given name,
1902 * inode and parent inode
1904 search_key.objectid = log_key.objectid;
1905 search_key.type = BTRFS_INODE_REF_KEY;
1906 search_key.offset = key->objectid;
1907 ret = backref_in_log(root->log_root, &search_key, 0, &name);
1911 /* The dentry will be added later. */
1913 update_size = false;
1917 search_key.objectid = log_key.objectid;
1918 search_key.type = BTRFS_INODE_EXTREF_KEY;
1919 search_key.offset = key->objectid;
1920 ret = backref_in_log(root->log_root, &search_key, key->objectid, &name);
1924 /* The dentry will be added later. */
1926 update_size = false;
1929 btrfs_release_path(path);
1930 ret = insert_one_name(trans, root, key->objectid, key->offset,
1932 if (ret && ret != -ENOENT && ret != -EEXIST)
1936 update_size = false;
1940 if (!ret && update_size) {
1941 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2);
1942 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
1946 if (!ret && name_added)
1951 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1952 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1953 struct btrfs_root *root,
1954 struct btrfs_path *path,
1955 struct extent_buffer *eb, int slot,
1956 struct btrfs_key *key)
1959 struct btrfs_dir_item *di;
1961 /* We only log dir index keys, which only contain a single dir item. */
1962 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
1964 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1965 ret = replay_one_name(trans, root, path, eb, di, key);
1970 * If this entry refers to a non-directory (directories can not have a
1971 * link count > 1) and it was added in the transaction that was not
1972 * committed, make sure we fixup the link count of the inode the entry
1973 * points to. Otherwise something like the following would result in a
1974 * directory pointing to an inode with a wrong link that does not account
1975 * for this dir entry:
1982 * ln testdir/bar testdir/bar_link
1983 * ln testdir/foo testdir/foo_link
1984 * xfs_io -c "fsync" testdir/bar
1988 * mount fs, log replay happens
1990 * File foo would remain with a link count of 1 when it has two entries
1991 * pointing to it in the directory testdir. This would make it impossible
1992 * to ever delete the parent directory has it would result in stale
1993 * dentries that can never be deleted.
1995 if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) {
1996 struct btrfs_path *fixup_path;
1997 struct btrfs_key di_key;
1999 fixup_path = btrfs_alloc_path();
2003 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2004 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2005 btrfs_free_path(fixup_path);
2012 * directory replay has two parts. There are the standard directory
2013 * items in the log copied from the subvolume, and range items
2014 * created in the log while the subvolume was logged.
2016 * The range items tell us which parts of the key space the log
2017 * is authoritative for. During replay, if a key in the subvolume
2018 * directory is in a logged range item, but not actually in the log
2019 * that means it was deleted from the directory before the fsync
2020 * and should be removed.
2022 static noinline int find_dir_range(struct btrfs_root *root,
2023 struct btrfs_path *path,
2025 u64 *start_ret, u64 *end_ret)
2027 struct btrfs_key key;
2029 struct btrfs_dir_log_item *item;
2033 if (*start_ret == (u64)-1)
2036 key.objectid = dirid;
2037 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2038 key.offset = *start_ret;
2040 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2044 if (path->slots[0] == 0)
2049 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2051 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2055 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2056 struct btrfs_dir_log_item);
2057 found_end = btrfs_dir_log_end(path->nodes[0], item);
2059 if (*start_ret >= key.offset && *start_ret <= found_end) {
2061 *start_ret = key.offset;
2062 *end_ret = found_end;
2067 /* check the next slot in the tree to see if it is a valid item */
2068 nritems = btrfs_header_nritems(path->nodes[0]);
2070 if (path->slots[0] >= nritems) {
2071 ret = btrfs_next_leaf(root, path);
2076 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2078 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2082 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2083 struct btrfs_dir_log_item);
2084 found_end = btrfs_dir_log_end(path->nodes[0], item);
2085 *start_ret = key.offset;
2086 *end_ret = found_end;
2089 btrfs_release_path(path);
2094 * this looks for a given directory item in the log. If the directory
2095 * item is not in the log, the item is removed and the inode it points
2098 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2099 struct btrfs_root *log,
2100 struct btrfs_path *path,
2101 struct btrfs_path *log_path,
2103 struct btrfs_key *dir_key)
2105 struct btrfs_root *root = BTRFS_I(dir)->root;
2107 struct extent_buffer *eb;
2109 struct btrfs_dir_item *di;
2110 struct fscrypt_str name;
2111 struct inode *inode = NULL;
2112 struct btrfs_key location;
2115 * Currently we only log dir index keys. Even if we replay a log created
2116 * by an older kernel that logged both dir index and dir item keys, all
2117 * we need to do is process the dir index keys, we (and our caller) can
2118 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2120 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2122 eb = path->nodes[0];
2123 slot = path->slots[0];
2124 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2125 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
2130 struct btrfs_dir_item *log_di;
2132 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2134 dir_key->offset, &name, 0);
2135 if (IS_ERR(log_di)) {
2136 ret = PTR_ERR(log_di);
2138 } else if (log_di) {
2139 /* The dentry exists in the log, we have nothing to do. */
2145 btrfs_dir_item_key_to_cpu(eb, di, &location);
2146 btrfs_release_path(path);
2147 btrfs_release_path(log_path);
2148 inode = read_one_inode(root, location.objectid);
2154 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2159 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2162 * Unlike dir item keys, dir index keys can only have one name (entry) in
2163 * them, as there are no key collisions since each key has a unique offset
2164 * (an index number), so we're done.
2167 btrfs_release_path(path);
2168 btrfs_release_path(log_path);
2174 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2175 struct btrfs_root *root,
2176 struct btrfs_root *log,
2177 struct btrfs_path *path,
2180 struct btrfs_key search_key;
2181 struct btrfs_path *log_path;
2186 log_path = btrfs_alloc_path();
2190 search_key.objectid = ino;
2191 search_key.type = BTRFS_XATTR_ITEM_KEY;
2192 search_key.offset = 0;
2194 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2198 nritems = btrfs_header_nritems(path->nodes[0]);
2199 for (i = path->slots[0]; i < nritems; i++) {
2200 struct btrfs_key key;
2201 struct btrfs_dir_item *di;
2202 struct btrfs_dir_item *log_di;
2206 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2207 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2212 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2213 total_size = btrfs_item_size(path->nodes[0], i);
2215 while (cur < total_size) {
2216 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2217 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2218 u32 this_len = sizeof(*di) + name_len + data_len;
2221 name = kmalloc(name_len, GFP_NOFS);
2226 read_extent_buffer(path->nodes[0], name,
2227 (unsigned long)(di + 1), name_len);
2229 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2231 btrfs_release_path(log_path);
2233 /* Doesn't exist in log tree, so delete it. */
2234 btrfs_release_path(path);
2235 di = btrfs_lookup_xattr(trans, root, path, ino,
2236 name, name_len, -1);
2243 ret = btrfs_delete_one_dir_name(trans, root,
2247 btrfs_release_path(path);
2252 if (IS_ERR(log_di)) {
2253 ret = PTR_ERR(log_di);
2257 di = (struct btrfs_dir_item *)((char *)di + this_len);
2260 ret = btrfs_next_leaf(root, path);
2266 btrfs_free_path(log_path);
2267 btrfs_release_path(path);
2273 * deletion replay happens before we copy any new directory items
2274 * out of the log or out of backreferences from inodes. It
2275 * scans the log to find ranges of keys that log is authoritative for,
2276 * and then scans the directory to find items in those ranges that are
2277 * not present in the log.
2279 * Anything we don't find in the log is unlinked and removed from the
2282 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2283 struct btrfs_root *root,
2284 struct btrfs_root *log,
2285 struct btrfs_path *path,
2286 u64 dirid, int del_all)
2291 struct btrfs_key dir_key;
2292 struct btrfs_key found_key;
2293 struct btrfs_path *log_path;
2296 dir_key.objectid = dirid;
2297 dir_key.type = BTRFS_DIR_INDEX_KEY;
2298 log_path = btrfs_alloc_path();
2302 dir = read_one_inode(root, dirid);
2303 /* it isn't an error if the inode isn't there, that can happen
2304 * because we replay the deletes before we copy in the inode item
2308 btrfs_free_path(log_path);
2316 range_end = (u64)-1;
2318 ret = find_dir_range(log, path, dirid,
2319 &range_start, &range_end);
2326 dir_key.offset = range_start;
2329 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2334 nritems = btrfs_header_nritems(path->nodes[0]);
2335 if (path->slots[0] >= nritems) {
2336 ret = btrfs_next_leaf(root, path);
2342 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2344 if (found_key.objectid != dirid ||
2345 found_key.type != dir_key.type) {
2350 if (found_key.offset > range_end)
2353 ret = check_item_in_log(trans, log, path,
2358 if (found_key.offset == (u64)-1)
2360 dir_key.offset = found_key.offset + 1;
2362 btrfs_release_path(path);
2363 if (range_end == (u64)-1)
2365 range_start = range_end + 1;
2369 btrfs_release_path(path);
2370 btrfs_free_path(log_path);
2376 * the process_func used to replay items from the log tree. This
2377 * gets called in two different stages. The first stage just looks
2378 * for inodes and makes sure they are all copied into the subvolume.
2380 * The second stage copies all the other item types from the log into
2381 * the subvolume. The two stage approach is slower, but gets rid of
2382 * lots of complexity around inodes referencing other inodes that exist
2383 * only in the log (references come from either directory items or inode
2386 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2387 struct walk_control *wc, u64 gen, int level)
2390 struct btrfs_tree_parent_check check = {
2394 struct btrfs_path *path;
2395 struct btrfs_root *root = wc->replay_dest;
2396 struct btrfs_key key;
2400 ret = btrfs_read_extent_buffer(eb, &check);
2404 level = btrfs_header_level(eb);
2409 path = btrfs_alloc_path();
2413 nritems = btrfs_header_nritems(eb);
2414 for (i = 0; i < nritems; i++) {
2415 btrfs_item_key_to_cpu(eb, &key, i);
2417 /* inode keys are done during the first stage */
2418 if (key.type == BTRFS_INODE_ITEM_KEY &&
2419 wc->stage == LOG_WALK_REPLAY_INODES) {
2420 struct btrfs_inode_item *inode_item;
2423 inode_item = btrfs_item_ptr(eb, i,
2424 struct btrfs_inode_item);
2426 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2427 * and never got linked before the fsync, skip it, as
2428 * replaying it is pointless since it would be deleted
2429 * later. We skip logging tmpfiles, but it's always
2430 * possible we are replaying a log created with a kernel
2431 * that used to log tmpfiles.
2433 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2434 wc->ignore_cur_inode = true;
2437 wc->ignore_cur_inode = false;
2439 ret = replay_xattr_deletes(wc->trans, root, log,
2440 path, key.objectid);
2443 mode = btrfs_inode_mode(eb, inode_item);
2444 if (S_ISDIR(mode)) {
2445 ret = replay_dir_deletes(wc->trans,
2446 root, log, path, key.objectid, 0);
2450 ret = overwrite_item(wc->trans, root, path,
2456 * Before replaying extents, truncate the inode to its
2457 * size. We need to do it now and not after log replay
2458 * because before an fsync we can have prealloc extents
2459 * added beyond the inode's i_size. If we did it after,
2460 * through orphan cleanup for example, we would drop
2461 * those prealloc extents just after replaying them.
2463 if (S_ISREG(mode)) {
2464 struct btrfs_drop_extents_args drop_args = { 0 };
2465 struct inode *inode;
2468 inode = read_one_inode(root, key.objectid);
2473 from = ALIGN(i_size_read(inode),
2474 root->fs_info->sectorsize);
2475 drop_args.start = from;
2476 drop_args.end = (u64)-1;
2477 drop_args.drop_cache = true;
2478 ret = btrfs_drop_extents(wc->trans, root,
2482 inode_sub_bytes(inode,
2483 drop_args.bytes_found);
2484 /* Update the inode's nbytes. */
2485 ret = btrfs_update_inode(wc->trans,
2486 root, BTRFS_I(inode));
2493 ret = link_to_fixup_dir(wc->trans, root,
2494 path, key.objectid);
2499 if (wc->ignore_cur_inode)
2502 if (key.type == BTRFS_DIR_INDEX_KEY &&
2503 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2504 ret = replay_one_dir_item(wc->trans, root, path,
2510 if (wc->stage < LOG_WALK_REPLAY_ALL)
2513 /* these keys are simply copied */
2514 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2515 ret = overwrite_item(wc->trans, root, path,
2519 } else if (key.type == BTRFS_INODE_REF_KEY ||
2520 key.type == BTRFS_INODE_EXTREF_KEY) {
2521 ret = add_inode_ref(wc->trans, root, log, path,
2523 if (ret && ret != -ENOENT)
2526 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2527 ret = replay_one_extent(wc->trans, root, path,
2533 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2534 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2535 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2536 * older kernel with such keys, ignore them.
2539 btrfs_free_path(path);
2544 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2546 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2548 struct btrfs_block_group *cache;
2550 cache = btrfs_lookup_block_group(fs_info, start);
2552 btrfs_err(fs_info, "unable to find block group for %llu", start);
2556 spin_lock(&cache->space_info->lock);
2557 spin_lock(&cache->lock);
2558 cache->reserved -= fs_info->nodesize;
2559 cache->space_info->bytes_reserved -= fs_info->nodesize;
2560 spin_unlock(&cache->lock);
2561 spin_unlock(&cache->space_info->lock);
2563 btrfs_put_block_group(cache);
2566 static int clean_log_buffer(struct btrfs_trans_handle *trans,
2567 struct extent_buffer *eb)
2571 btrfs_tree_lock(eb);
2572 btrfs_clear_buffer_dirty(trans, eb);
2573 wait_on_extent_buffer_writeback(eb);
2574 btrfs_tree_unlock(eb);
2577 ret = btrfs_pin_reserved_extent(trans, eb->start, eb->len);
2580 btrfs_redirty_list_add(trans->transaction, eb);
2582 unaccount_log_buffer(eb->fs_info, eb->start);
2588 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2589 struct btrfs_root *root,
2590 struct btrfs_path *path, int *level,
2591 struct walk_control *wc)
2593 struct btrfs_fs_info *fs_info = root->fs_info;
2596 struct extent_buffer *next;
2597 struct extent_buffer *cur;
2600 while (*level > 0) {
2601 struct btrfs_tree_parent_check check = { 0 };
2603 cur = path->nodes[*level];
2605 WARN_ON(btrfs_header_level(cur) != *level);
2607 if (path->slots[*level] >=
2608 btrfs_header_nritems(cur))
2611 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2612 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2613 check.transid = ptr_gen;
2614 check.level = *level - 1;
2615 check.has_first_key = true;
2616 btrfs_node_key_to_cpu(cur, &check.first_key, path->slots[*level]);
2618 next = btrfs_find_create_tree_block(fs_info, bytenr,
2619 btrfs_header_owner(cur),
2622 return PTR_ERR(next);
2625 ret = wc->process_func(root, next, wc, ptr_gen,
2628 free_extent_buffer(next);
2632 path->slots[*level]++;
2634 ret = btrfs_read_extent_buffer(next, &check);
2636 free_extent_buffer(next);
2640 ret = clean_log_buffer(trans, next);
2642 free_extent_buffer(next);
2646 free_extent_buffer(next);
2649 ret = btrfs_read_extent_buffer(next, &check);
2651 free_extent_buffer(next);
2655 if (path->nodes[*level-1])
2656 free_extent_buffer(path->nodes[*level-1]);
2657 path->nodes[*level-1] = next;
2658 *level = btrfs_header_level(next);
2659 path->slots[*level] = 0;
2662 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2668 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2669 struct btrfs_root *root,
2670 struct btrfs_path *path, int *level,
2671 struct walk_control *wc)
2677 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2678 slot = path->slots[i];
2679 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2682 WARN_ON(*level == 0);
2685 ret = wc->process_func(root, path->nodes[*level], wc,
2686 btrfs_header_generation(path->nodes[*level]),
2692 ret = clean_log_buffer(trans, path->nodes[*level]);
2696 free_extent_buffer(path->nodes[*level]);
2697 path->nodes[*level] = NULL;
2705 * drop the reference count on the tree rooted at 'snap'. This traverses
2706 * the tree freeing any blocks that have a ref count of zero after being
2709 static int walk_log_tree(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *log, struct walk_control *wc)
2715 struct btrfs_path *path;
2718 path = btrfs_alloc_path();
2722 level = btrfs_header_level(log->node);
2724 path->nodes[level] = log->node;
2725 atomic_inc(&log->node->refs);
2726 path->slots[level] = 0;
2729 wret = walk_down_log_tree(trans, log, path, &level, wc);
2737 wret = walk_up_log_tree(trans, log, path, &level, wc);
2746 /* was the root node processed? if not, catch it here */
2747 if (path->nodes[orig_level]) {
2748 ret = wc->process_func(log, path->nodes[orig_level], wc,
2749 btrfs_header_generation(path->nodes[orig_level]),
2754 ret = clean_log_buffer(trans, path->nodes[orig_level]);
2758 btrfs_free_path(path);
2763 * helper function to update the item for a given subvolumes log root
2764 * in the tree of log roots
2766 static int update_log_root(struct btrfs_trans_handle *trans,
2767 struct btrfs_root *log,
2768 struct btrfs_root_item *root_item)
2770 struct btrfs_fs_info *fs_info = log->fs_info;
2773 if (log->log_transid == 1) {
2774 /* insert root item on the first sync */
2775 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2776 &log->root_key, root_item);
2778 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2779 &log->root_key, root_item);
2784 static void wait_log_commit(struct btrfs_root *root, int transid)
2787 int index = transid % 2;
2790 * we only allow two pending log transactions at a time,
2791 * so we know that if ours is more than 2 older than the
2792 * current transaction, we're done
2795 prepare_to_wait(&root->log_commit_wait[index],
2796 &wait, TASK_UNINTERRUPTIBLE);
2798 if (!(root->log_transid_committed < transid &&
2799 atomic_read(&root->log_commit[index])))
2802 mutex_unlock(&root->log_mutex);
2804 mutex_lock(&root->log_mutex);
2806 finish_wait(&root->log_commit_wait[index], &wait);
2809 static void wait_for_writer(struct btrfs_root *root)
2814 prepare_to_wait(&root->log_writer_wait, &wait,
2815 TASK_UNINTERRUPTIBLE);
2816 if (!atomic_read(&root->log_writers))
2819 mutex_unlock(&root->log_mutex);
2821 mutex_lock(&root->log_mutex);
2823 finish_wait(&root->log_writer_wait, &wait);
2826 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2827 struct btrfs_log_ctx *ctx)
2829 mutex_lock(&root->log_mutex);
2830 list_del_init(&ctx->list);
2831 mutex_unlock(&root->log_mutex);
2835 * Invoked in log mutex context, or be sure there is no other task which
2836 * can access the list.
2838 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2839 int index, int error)
2841 struct btrfs_log_ctx *ctx;
2842 struct btrfs_log_ctx *safe;
2844 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2845 list_del_init(&ctx->list);
2846 ctx->log_ret = error;
2851 * btrfs_sync_log does sends a given tree log down to the disk and
2852 * updates the super blocks to record it. When this call is done,
2853 * you know that any inodes previously logged are safely on disk only
2856 * Any other return value means you need to call btrfs_commit_transaction.
2857 * Some of the edge cases for fsyncing directories that have had unlinks
2858 * or renames done in the past mean that sometimes the only safe
2859 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2860 * that has happened.
2862 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2863 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2869 struct btrfs_fs_info *fs_info = root->fs_info;
2870 struct btrfs_root *log = root->log_root;
2871 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2872 struct btrfs_root_item new_root_item;
2873 int log_transid = 0;
2874 struct btrfs_log_ctx root_log_ctx;
2875 struct blk_plug plug;
2879 mutex_lock(&root->log_mutex);
2880 log_transid = ctx->log_transid;
2881 if (root->log_transid_committed >= log_transid) {
2882 mutex_unlock(&root->log_mutex);
2883 return ctx->log_ret;
2886 index1 = log_transid % 2;
2887 if (atomic_read(&root->log_commit[index1])) {
2888 wait_log_commit(root, log_transid);
2889 mutex_unlock(&root->log_mutex);
2890 return ctx->log_ret;
2892 ASSERT(log_transid == root->log_transid);
2893 atomic_set(&root->log_commit[index1], 1);
2895 /* wait for previous tree log sync to complete */
2896 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2897 wait_log_commit(root, log_transid - 1);
2900 int batch = atomic_read(&root->log_batch);
2901 /* when we're on an ssd, just kick the log commit out */
2902 if (!btrfs_test_opt(fs_info, SSD) &&
2903 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2904 mutex_unlock(&root->log_mutex);
2905 schedule_timeout_uninterruptible(1);
2906 mutex_lock(&root->log_mutex);
2908 wait_for_writer(root);
2909 if (batch == atomic_read(&root->log_batch))
2913 /* bail out if we need to do a full commit */
2914 if (btrfs_need_log_full_commit(trans)) {
2915 ret = BTRFS_LOG_FORCE_COMMIT;
2916 mutex_unlock(&root->log_mutex);
2920 if (log_transid % 2 == 0)
2921 mark = EXTENT_DIRTY;
2925 /* we start IO on all the marked extents here, but we don't actually
2926 * wait for them until later.
2928 blk_start_plug(&plug);
2929 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2931 * -EAGAIN happens when someone, e.g., a concurrent transaction
2932 * commit, writes a dirty extent in this tree-log commit. This
2933 * concurrent write will create a hole writing out the extents,
2934 * and we cannot proceed on a zoned filesystem, requiring
2935 * sequential writing. While we can bail out to a full commit
2936 * here, but we can continue hoping the concurrent writing fills
2939 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
2942 blk_finish_plug(&plug);
2943 btrfs_set_log_full_commit(trans);
2944 mutex_unlock(&root->log_mutex);
2949 * We _must_ update under the root->log_mutex in order to make sure we
2950 * have a consistent view of the log root we are trying to commit at
2953 * We _must_ copy this into a local copy, because we are not holding the
2954 * log_root_tree->log_mutex yet. This is important because when we
2955 * commit the log_root_tree we must have a consistent view of the
2956 * log_root_tree when we update the super block to point at the
2957 * log_root_tree bytenr. If we update the log_root_tree here we'll race
2958 * with the commit and possibly point at the new block which we may not
2961 btrfs_set_root_node(&log->root_item, log->node);
2962 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
2964 root->log_transid++;
2965 log->log_transid = root->log_transid;
2966 root->log_start_pid = 0;
2968 * IO has been started, blocks of the log tree have WRITTEN flag set
2969 * in their headers. new modifications of the log will be written to
2970 * new positions. so it's safe to allow log writers to go in.
2972 mutex_unlock(&root->log_mutex);
2974 if (btrfs_is_zoned(fs_info)) {
2975 mutex_lock(&fs_info->tree_root->log_mutex);
2976 if (!log_root_tree->node) {
2977 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
2979 mutex_unlock(&fs_info->tree_root->log_mutex);
2980 blk_finish_plug(&plug);
2984 mutex_unlock(&fs_info->tree_root->log_mutex);
2987 btrfs_init_log_ctx(&root_log_ctx, NULL);
2989 mutex_lock(&log_root_tree->log_mutex);
2991 index2 = log_root_tree->log_transid % 2;
2992 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2993 root_log_ctx.log_transid = log_root_tree->log_transid;
2996 * Now we are safe to update the log_root_tree because we're under the
2997 * log_mutex, and we're a current writer so we're holding the commit
2998 * open until we drop the log_mutex.
3000 ret = update_log_root(trans, log, &new_root_item);
3002 if (!list_empty(&root_log_ctx.list))
3003 list_del_init(&root_log_ctx.list);
3005 blk_finish_plug(&plug);
3006 btrfs_set_log_full_commit(trans);
3009 "failed to update log for root %llu ret %d",
3010 root->root_key.objectid, ret);
3011 btrfs_wait_tree_log_extents(log, mark);
3012 mutex_unlock(&log_root_tree->log_mutex);
3016 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3017 blk_finish_plug(&plug);
3018 list_del_init(&root_log_ctx.list);
3019 mutex_unlock(&log_root_tree->log_mutex);
3020 ret = root_log_ctx.log_ret;
3024 index2 = root_log_ctx.log_transid % 2;
3025 if (atomic_read(&log_root_tree->log_commit[index2])) {
3026 blk_finish_plug(&plug);
3027 ret = btrfs_wait_tree_log_extents(log, mark);
3028 wait_log_commit(log_root_tree,
3029 root_log_ctx.log_transid);
3030 mutex_unlock(&log_root_tree->log_mutex);
3032 ret = root_log_ctx.log_ret;
3035 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3036 atomic_set(&log_root_tree->log_commit[index2], 1);
3038 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3039 wait_log_commit(log_root_tree,
3040 root_log_ctx.log_transid - 1);
3044 * now that we've moved on to the tree of log tree roots,
3045 * check the full commit flag again
3047 if (btrfs_need_log_full_commit(trans)) {
3048 blk_finish_plug(&plug);
3049 btrfs_wait_tree_log_extents(log, mark);
3050 mutex_unlock(&log_root_tree->log_mutex);
3051 ret = BTRFS_LOG_FORCE_COMMIT;
3052 goto out_wake_log_root;
3055 ret = btrfs_write_marked_extents(fs_info,
3056 &log_root_tree->dirty_log_pages,
3057 EXTENT_DIRTY | EXTENT_NEW);
3058 blk_finish_plug(&plug);
3060 * As described above, -EAGAIN indicates a hole in the extents. We
3061 * cannot wait for these write outs since the waiting cause a
3062 * deadlock. Bail out to the full commit instead.
3064 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3065 btrfs_set_log_full_commit(trans);
3066 btrfs_wait_tree_log_extents(log, mark);
3067 mutex_unlock(&log_root_tree->log_mutex);
3068 goto out_wake_log_root;
3070 btrfs_set_log_full_commit(trans);
3071 mutex_unlock(&log_root_tree->log_mutex);
3072 goto out_wake_log_root;
3074 ret = btrfs_wait_tree_log_extents(log, mark);
3076 ret = btrfs_wait_tree_log_extents(log_root_tree,
3077 EXTENT_NEW | EXTENT_DIRTY);
3079 btrfs_set_log_full_commit(trans);
3080 mutex_unlock(&log_root_tree->log_mutex);
3081 goto out_wake_log_root;
3084 log_root_start = log_root_tree->node->start;
3085 log_root_level = btrfs_header_level(log_root_tree->node);
3086 log_root_tree->log_transid++;
3087 mutex_unlock(&log_root_tree->log_mutex);
3090 * Here we are guaranteed that nobody is going to write the superblock
3091 * for the current transaction before us and that neither we do write
3092 * our superblock before the previous transaction finishes its commit
3093 * and writes its superblock, because:
3095 * 1) We are holding a handle on the current transaction, so no body
3096 * can commit it until we release the handle;
3098 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3099 * if the previous transaction is still committing, and hasn't yet
3100 * written its superblock, we wait for it to do it, because a
3101 * transaction commit acquires the tree_log_mutex when the commit
3102 * begins and releases it only after writing its superblock.
3104 mutex_lock(&fs_info->tree_log_mutex);
3107 * The previous transaction writeout phase could have failed, and thus
3108 * marked the fs in an error state. We must not commit here, as we
3109 * could have updated our generation in the super_for_commit and
3110 * writing the super here would result in transid mismatches. If there
3111 * is an error here just bail.
3113 if (BTRFS_FS_ERROR(fs_info)) {
3115 btrfs_set_log_full_commit(trans);
3116 btrfs_abort_transaction(trans, ret);
3117 mutex_unlock(&fs_info->tree_log_mutex);
3118 goto out_wake_log_root;
3121 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3122 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3123 ret = write_all_supers(fs_info, 1);
3124 mutex_unlock(&fs_info->tree_log_mutex);
3126 btrfs_set_log_full_commit(trans);
3127 btrfs_abort_transaction(trans, ret);
3128 goto out_wake_log_root;
3132 * We know there can only be one task here, since we have not yet set
3133 * root->log_commit[index1] to 0 and any task attempting to sync the
3134 * log must wait for the previous log transaction to commit if it's
3135 * still in progress or wait for the current log transaction commit if
3136 * someone else already started it. We use <= and not < because the
3137 * first log transaction has an ID of 0.
3139 ASSERT(root->last_log_commit <= log_transid);
3140 root->last_log_commit = log_transid;
3143 mutex_lock(&log_root_tree->log_mutex);
3144 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3146 log_root_tree->log_transid_committed++;
3147 atomic_set(&log_root_tree->log_commit[index2], 0);
3148 mutex_unlock(&log_root_tree->log_mutex);
3151 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3152 * all the updates above are seen by the woken threads. It might not be
3153 * necessary, but proving that seems to be hard.
3155 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3157 mutex_lock(&root->log_mutex);
3158 btrfs_remove_all_log_ctxs(root, index1, ret);
3159 root->log_transid_committed++;
3160 atomic_set(&root->log_commit[index1], 0);
3161 mutex_unlock(&root->log_mutex);
3164 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3165 * all the updates above are seen by the woken threads. It might not be
3166 * necessary, but proving that seems to be hard.
3168 cond_wake_up(&root->log_commit_wait[index1]);
3172 static void free_log_tree(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *log)
3176 struct walk_control wc = {
3178 .process_func = process_one_buffer
3182 ret = walk_log_tree(trans, log, &wc);
3185 * We weren't able to traverse the entire log tree, the
3186 * typical scenario is getting an -EIO when reading an
3187 * extent buffer of the tree, due to a previous writeback
3190 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3191 &log->fs_info->fs_state);
3194 * Some extent buffers of the log tree may still be dirty
3195 * and not yet written back to storage, because we may
3196 * have updates to a log tree without syncing a log tree,
3197 * such as during rename and link operations. So flush
3198 * them out and wait for their writeback to complete, so
3199 * that we properly cleanup their state and pages.
3201 btrfs_write_marked_extents(log->fs_info,
3202 &log->dirty_log_pages,
3203 EXTENT_DIRTY | EXTENT_NEW);
3204 btrfs_wait_tree_log_extents(log,
3205 EXTENT_DIRTY | EXTENT_NEW);
3208 btrfs_abort_transaction(trans, ret);
3210 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3214 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3215 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3216 extent_io_tree_release(&log->log_csum_range);
3218 btrfs_put_root(log);
3222 * free all the extents used by the tree log. This should be called
3223 * at commit time of the full transaction
3225 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3227 if (root->log_root) {
3228 free_log_tree(trans, root->log_root);
3229 root->log_root = NULL;
3230 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3235 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3236 struct btrfs_fs_info *fs_info)
3238 if (fs_info->log_root_tree) {
3239 free_log_tree(trans, fs_info->log_root_tree);
3240 fs_info->log_root_tree = NULL;
3241 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3247 * Check if an inode was logged in the current transaction. This correctly deals
3248 * with the case where the inode was logged but has a logged_trans of 0, which
3249 * happens if the inode is evicted and loaded again, as logged_trans is an in
3250 * memory only field (not persisted).
3252 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3255 static int inode_logged(const struct btrfs_trans_handle *trans,
3256 struct btrfs_inode *inode,
3257 struct btrfs_path *path_in)
3259 struct btrfs_path *path = path_in;
3260 struct btrfs_key key;
3263 if (inode->logged_trans == trans->transid)
3267 * If logged_trans is not 0, then we know the inode logged was not logged
3268 * in this transaction, so we can return false right away.
3270 if (inode->logged_trans > 0)
3274 * If no log tree was created for this root in this transaction, then
3275 * the inode can not have been logged in this transaction. In that case
3276 * set logged_trans to anything greater than 0 and less than the current
3277 * transaction's ID, to avoid the search below in a future call in case
3278 * a log tree gets created after this.
3280 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3281 inode->logged_trans = trans->transid - 1;
3286 * We have a log tree and the inode's logged_trans is 0. We can't tell
3287 * for sure if the inode was logged before in this transaction by looking
3288 * only at logged_trans. We could be pessimistic and assume it was, but
3289 * that can lead to unnecessarily logging an inode during rename and link
3290 * operations, and then further updating the log in followup rename and
3291 * link operations, specially if it's a directory, which adds latency
3292 * visible to applications doing a series of rename or link operations.
3294 * A logged_trans of 0 here can mean several things:
3296 * 1) The inode was never logged since the filesystem was mounted, and may
3297 * or may have not been evicted and loaded again;
3299 * 2) The inode was logged in a previous transaction, then evicted and
3300 * then loaded again;
3302 * 3) The inode was logged in the current transaction, then evicted and
3303 * then loaded again.
3305 * For cases 1) and 2) we don't want to return true, but we need to detect
3306 * case 3) and return true. So we do a search in the log root for the inode
3309 key.objectid = btrfs_ino(inode);
3310 key.type = BTRFS_INODE_ITEM_KEY;
3314 path = btrfs_alloc_path();
3319 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3322 btrfs_release_path(path);
3324 btrfs_free_path(path);
3327 * Logging an inode always results in logging its inode item. So if we
3328 * did not find the item we know the inode was not logged for sure.
3332 } else if (ret > 0) {
3334 * Set logged_trans to a value greater than 0 and less then the
3335 * current transaction to avoid doing the search in future calls.
3337 inode->logged_trans = trans->transid - 1;
3342 * The inode was previously logged and then evicted, set logged_trans to
3343 * the current transacion's ID, to avoid future tree searches as long as
3344 * the inode is not evicted again.
3346 inode->logged_trans = trans->transid;
3349 * If it's a directory, then we must set last_dir_index_offset to the
3350 * maximum possible value, so that the next attempt to log the inode does
3351 * not skip checking if dir index keys found in modified subvolume tree
3352 * leaves have been logged before, otherwise it would result in attempts
3353 * to insert duplicate dir index keys in the log tree. This must be done
3354 * because last_dir_index_offset is an in-memory only field, not persisted
3355 * in the inode item or any other on-disk structure, so its value is lost
3356 * once the inode is evicted.
3358 if (S_ISDIR(inode->vfs_inode.i_mode))
3359 inode->last_dir_index_offset = (u64)-1;
3365 * Delete a directory entry from the log if it exists.
3367 * Returns < 0 on error
3368 * 1 if the entry does not exists
3369 * 0 if the entry existed and was successfully deleted
3371 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3372 struct btrfs_root *log,
3373 struct btrfs_path *path,
3375 const struct fscrypt_str *name,
3378 struct btrfs_dir_item *di;
3381 * We only log dir index items of a directory, so we don't need to look
3382 * for dir item keys.
3384 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3392 * We do not need to update the size field of the directory's
3393 * inode item because on log replay we update the field to reflect
3394 * all existing entries in the directory (see overwrite_item()).
3396 return btrfs_delete_one_dir_name(trans, log, path, di);
3400 * If both a file and directory are logged, and unlinks or renames are
3401 * mixed in, we have a few interesting corners:
3403 * create file X in dir Y
3404 * link file X to X.link in dir Y
3406 * unlink file X but leave X.link
3409 * After a crash we would expect only X.link to exist. But file X
3410 * didn't get fsync'd again so the log has back refs for X and X.link.
3412 * We solve this by removing directory entries and inode backrefs from the
3413 * log when a file that was logged in the current transaction is
3414 * unlinked. Any later fsync will include the updated log entries, and
3415 * we'll be able to reconstruct the proper directory items from backrefs.
3417 * This optimizations allows us to avoid relogging the entire inode
3418 * or the entire directory.
3420 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3421 struct btrfs_root *root,
3422 const struct fscrypt_str *name,
3423 struct btrfs_inode *dir, u64 index)
3425 struct btrfs_path *path;
3428 ret = inode_logged(trans, dir, NULL);
3432 btrfs_set_log_full_commit(trans);
3436 ret = join_running_log_trans(root);
3440 mutex_lock(&dir->log_mutex);
3442 path = btrfs_alloc_path();
3448 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3450 btrfs_free_path(path);
3452 mutex_unlock(&dir->log_mutex);
3454 btrfs_set_log_full_commit(trans);
3455 btrfs_end_log_trans(root);
3458 /* see comments for btrfs_del_dir_entries_in_log */
3459 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3460 struct btrfs_root *root,
3461 const struct fscrypt_str *name,
3462 struct btrfs_inode *inode, u64 dirid)
3464 struct btrfs_root *log;
3468 ret = inode_logged(trans, inode, NULL);
3472 btrfs_set_log_full_commit(trans);
3476 ret = join_running_log_trans(root);
3479 log = root->log_root;
3480 mutex_lock(&inode->log_mutex);
3482 ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode),
3484 mutex_unlock(&inode->log_mutex);
3485 if (ret < 0 && ret != -ENOENT)
3486 btrfs_set_log_full_commit(trans);
3487 btrfs_end_log_trans(root);
3491 * creates a range item in the log for 'dirid'. first_offset and
3492 * last_offset tell us which parts of the key space the log should
3493 * be considered authoritative for.
3495 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3496 struct btrfs_root *log,
3497 struct btrfs_path *path,
3499 u64 first_offset, u64 last_offset)
3502 struct btrfs_key key;
3503 struct btrfs_dir_log_item *item;
3505 key.objectid = dirid;
3506 key.offset = first_offset;
3507 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3508 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3510 * -EEXIST is fine and can happen sporadically when we are logging a
3511 * directory and have concurrent insertions in the subvolume's tree for
3512 * items from other inodes and that result in pushing off some dir items
3513 * from one leaf to another in order to accommodate for the new items.
3514 * This results in logging the same dir index range key.
3516 if (ret && ret != -EEXIST)
3519 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3520 struct btrfs_dir_log_item);
3521 if (ret == -EEXIST) {
3522 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3525 * btrfs_del_dir_entries_in_log() might have been called during
3526 * an unlink between the initial insertion of this key and the
3527 * current update, or we might be logging a single entry deletion
3528 * during a rename, so set the new last_offset to the max value.
3530 last_offset = max(last_offset, curr_end);
3532 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3533 btrfs_mark_buffer_dirty(path->nodes[0]);
3534 btrfs_release_path(path);
3538 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3539 struct btrfs_inode *inode,
3540 struct extent_buffer *src,
3541 struct btrfs_path *dst_path,
3545 struct btrfs_root *log = inode->root->log_root;
3546 char *ins_data = NULL;
3547 struct btrfs_item_batch batch;
3548 struct extent_buffer *dst;
3549 unsigned long src_offset;
3550 unsigned long dst_offset;
3552 struct btrfs_key key;
3561 btrfs_item_key_to_cpu(src, &key, start_slot);
3562 item_size = btrfs_item_size(src, start_slot);
3564 batch.data_sizes = &item_size;
3565 batch.total_data_size = item_size;
3567 struct btrfs_key *ins_keys;
3570 ins_data = kmalloc(count * sizeof(u32) +
3571 count * sizeof(struct btrfs_key), GFP_NOFS);
3575 ins_sizes = (u32 *)ins_data;
3576 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3577 batch.keys = ins_keys;
3578 batch.data_sizes = ins_sizes;
3579 batch.total_data_size = 0;
3581 for (i = 0; i < count; i++) {
3582 const int slot = start_slot + i;
3584 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3585 ins_sizes[i] = btrfs_item_size(src, slot);
3586 batch.total_data_size += ins_sizes[i];
3590 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3594 dst = dst_path->nodes[0];
3596 * Copy all the items in bulk, in a single copy operation. Item data is
3597 * organized such that it's placed at the end of a leaf and from right
3598 * to left. For example, the data for the second item ends at an offset
3599 * that matches the offset where the data for the first item starts, the
3600 * data for the third item ends at an offset that matches the offset
3601 * where the data of the second items starts, and so on.
3602 * Therefore our source and destination start offsets for copy match the
3603 * offsets of the last items (highest slots).
3605 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3606 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3607 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3608 btrfs_release_path(dst_path);
3610 last_index = batch.keys[count - 1].offset;
3611 ASSERT(last_index > inode->last_dir_index_offset);
3614 * If for some unexpected reason the last item's index is not greater
3615 * than the last index we logged, warn and force a transaction commit.
3617 if (WARN_ON(last_index <= inode->last_dir_index_offset))
3618 ret = BTRFS_LOG_FORCE_COMMIT;
3620 inode->last_dir_index_offset = last_index;
3622 if (btrfs_get_first_dir_index_to_log(inode) == 0)
3623 btrfs_set_first_dir_index_to_log(inode, batch.keys[0].offset);
3630 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3631 struct btrfs_inode *inode,
3632 struct btrfs_path *path,
3633 struct btrfs_path *dst_path,
3634 struct btrfs_log_ctx *ctx,
3635 u64 *last_old_dentry_offset)
3637 struct btrfs_root *log = inode->root->log_root;
3638 struct extent_buffer *src;
3639 const int nritems = btrfs_header_nritems(path->nodes[0]);
3640 const u64 ino = btrfs_ino(inode);
3641 bool last_found = false;
3642 int batch_start = 0;
3647 * We need to clone the leaf, release the read lock on it, and use the
3648 * clone before modifying the log tree. See the comment at copy_items()
3649 * about why we need to do this.
3651 src = btrfs_clone_extent_buffer(path->nodes[0]);
3656 btrfs_release_path(path);
3657 path->nodes[0] = src;
3660 for (; i < nritems; i++) {
3661 struct btrfs_dir_item *di;
3662 struct btrfs_key key;
3665 btrfs_item_key_to_cpu(src, &key, i);
3667 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3672 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3675 * Skip ranges of items that consist only of dir item keys created
3676 * in past transactions. However if we find a gap, we must log a
3677 * dir index range item for that gap, so that index keys in that
3678 * gap are deleted during log replay.
3680 if (btrfs_dir_transid(src, di) < trans->transid) {
3681 if (key.offset > *last_old_dentry_offset + 1) {
3682 ret = insert_dir_log_key(trans, log, dst_path,
3683 ino, *last_old_dentry_offset + 1,
3689 *last_old_dentry_offset = key.offset;
3693 /* If we logged this dir index item before, we can skip it. */
3694 if (key.offset <= inode->last_dir_index_offset)
3698 * We must make sure that when we log a directory entry, the
3699 * corresponding inode, after log replay, has a matching link
3700 * count. For example:
3706 * xfs_io -c "fsync" mydir
3708 * <mount fs and log replay>
3710 * Would result in a fsync log that when replayed, our file inode
3711 * would have a link count of 1, but we get two directory entries
3712 * pointing to the same inode. After removing one of the names,
3713 * it would not be possible to remove the other name, which
3714 * resulted always in stale file handle errors, and would not be
3715 * possible to rmdir the parent directory, since its i_size could
3716 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3717 * resulting in -ENOTEMPTY errors.
3719 if (!ctx->log_new_dentries) {
3720 struct btrfs_key di_key;
3722 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3723 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3724 ctx->log_new_dentries = true;
3727 if (batch_size == 0)
3732 if (batch_size > 0) {
3735 ret = flush_dir_items_batch(trans, inode, src, dst_path,
3736 batch_start, batch_size);
3741 return last_found ? 1 : 0;
3745 * log all the items included in the current transaction for a given
3746 * directory. This also creates the range items in the log tree required
3747 * to replay anything deleted before the fsync
3749 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3750 struct btrfs_inode *inode,
3751 struct btrfs_path *path,
3752 struct btrfs_path *dst_path,
3753 struct btrfs_log_ctx *ctx,
3754 u64 min_offset, u64 *last_offset_ret)
3756 struct btrfs_key min_key;
3757 struct btrfs_root *root = inode->root;
3758 struct btrfs_root *log = root->log_root;
3760 u64 last_old_dentry_offset = min_offset - 1;
3761 u64 last_offset = (u64)-1;
3762 u64 ino = btrfs_ino(inode);
3764 min_key.objectid = ino;
3765 min_key.type = BTRFS_DIR_INDEX_KEY;
3766 min_key.offset = min_offset;
3768 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3771 * we didn't find anything from this transaction, see if there
3772 * is anything at all
3774 if (ret != 0 || min_key.objectid != ino ||
3775 min_key.type != BTRFS_DIR_INDEX_KEY) {
3776 min_key.objectid = ino;
3777 min_key.type = BTRFS_DIR_INDEX_KEY;
3778 min_key.offset = (u64)-1;
3779 btrfs_release_path(path);
3780 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3782 btrfs_release_path(path);
3785 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3787 /* if ret == 0 there are items for this type,
3788 * create a range to tell us the last key of this type.
3789 * otherwise, there are no items in this directory after
3790 * *min_offset, and we create a range to indicate that.
3793 struct btrfs_key tmp;
3795 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3797 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3798 last_old_dentry_offset = tmp.offset;
3799 } else if (ret > 0) {
3806 /* go backward to find any previous key */
3807 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3809 struct btrfs_key tmp;
3811 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3813 * The dir index key before the first one we found that needs to
3814 * be logged might be in a previous leaf, and there might be a
3815 * gap between these keys, meaning that we had deletions that
3816 * happened. So the key range item we log (key type
3817 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3818 * previous key's offset plus 1, so that those deletes are replayed.
3820 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3821 last_old_dentry_offset = tmp.offset;
3822 } else if (ret < 0) {
3826 btrfs_release_path(path);
3829 * Find the first key from this transaction again or the one we were at
3830 * in the loop below in case we had to reschedule. We may be logging the
3831 * directory without holding its VFS lock, which happen when logging new
3832 * dentries (through log_new_dir_dentries()) or in some cases when we
3833 * need to log the parent directory of an inode. This means a dir index
3834 * key might be deleted from the inode's root, and therefore we may not
3835 * find it anymore. If we can't find it, just move to the next key. We
3836 * can not bail out and ignore, because if we do that we will simply
3837 * not log dir index keys that come after the one that was just deleted
3838 * and we can end up logging a dir index range that ends at (u64)-1
3839 * (@last_offset is initialized to that), resulting in removing dir
3840 * entries we should not remove at log replay time.
3843 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3845 ret = btrfs_next_item(root, path);
3847 /* There are no more keys in the inode's root. */
3856 * we have a block from this transaction, log every item in it
3857 * from our directory
3860 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3861 &last_old_dentry_offset);
3867 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3870 * look ahead to the next item and see if it is also
3871 * from this directory and from this transaction
3873 ret = btrfs_next_leaf(root, path);
3876 last_offset = (u64)-1;
3881 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3882 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3883 last_offset = (u64)-1;
3886 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3888 * The next leaf was not changed in the current transaction
3889 * and has at least one dir index key.
3890 * We check for the next key because there might have been
3891 * one or more deletions between the last key we logged and
3892 * that next key. So the key range item we log (key type
3893 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3894 * offset minus 1, so that those deletes are replayed.
3896 last_offset = min_key.offset - 1;
3899 if (need_resched()) {
3900 btrfs_release_path(path);
3906 btrfs_release_path(path);
3907 btrfs_release_path(dst_path);
3910 *last_offset_ret = last_offset;
3912 * In case the leaf was changed in the current transaction but
3913 * all its dir items are from a past transaction, the last item
3914 * in the leaf is a dir item and there's no gap between that last
3915 * dir item and the first one on the next leaf (which did not
3916 * change in the current transaction), then we don't need to log
3917 * a range, last_old_dentry_offset is == to last_offset.
3919 ASSERT(last_old_dentry_offset <= last_offset);
3920 if (last_old_dentry_offset < last_offset)
3921 ret = insert_dir_log_key(trans, log, path, ino,
3922 last_old_dentry_offset + 1,
3930 * If the inode was logged before and it was evicted, then its
3931 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3932 * key offset. If that's the case, search for it and update the inode. This
3933 * is to avoid lookups in the log tree every time we try to insert a dir index
3934 * key from a leaf changed in the current transaction, and to allow us to always
3935 * do batch insertions of dir index keys.
3937 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3938 struct btrfs_path *path,
3939 const struct btrfs_log_ctx *ctx)
3941 const u64 ino = btrfs_ino(inode);
3942 struct btrfs_key key;
3945 lockdep_assert_held(&inode->log_mutex);
3947 if (inode->last_dir_index_offset != (u64)-1)
3950 if (!ctx->logged_before) {
3951 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3956 key.type = BTRFS_DIR_INDEX_KEY;
3957 key.offset = (u64)-1;
3959 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3961 * An error happened or we actually have an index key with an offset
3962 * value of (u64)-1. Bail out, we're done.
3968 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3971 * No dir index items, bail out and leave last_dir_index_offset with
3972 * the value right before the first valid index value.
3974 if (path->slots[0] == 0)
3978 * btrfs_search_slot() left us at one slot beyond the slot with the last
3979 * index key, or beyond the last key of the directory that is not an
3980 * index key. If we have an index key before, set last_dir_index_offset
3981 * to its offset value, otherwise leave it with a value right before the
3982 * first valid index value, as it means we have an empty directory.
3984 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3985 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
3986 inode->last_dir_index_offset = key.offset;
3989 btrfs_release_path(path);
3995 * logging directories is very similar to logging inodes, We find all the items
3996 * from the current transaction and write them to the log.
3998 * The recovery code scans the directory in the subvolume, and if it finds a
3999 * key in the range logged that is not present in the log tree, then it means
4000 * that dir entry was unlinked during the transaction.
4002 * In order for that scan to work, we must include one key smaller than
4003 * the smallest logged by this transaction and one key larger than the largest
4004 * key logged by this transaction.
4006 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4007 struct btrfs_inode *inode,
4008 struct btrfs_path *path,
4009 struct btrfs_path *dst_path,
4010 struct btrfs_log_ctx *ctx)
4016 ret = update_last_dir_index_offset(inode, path, ctx);
4020 min_key = BTRFS_DIR_START_INDEX;
4024 ret = log_dir_items(trans, inode, path, dst_path,
4025 ctx, min_key, &max_key);
4028 if (max_key == (u64)-1)
4030 min_key = max_key + 1;
4037 * a helper function to drop items from the log before we relog an
4038 * inode. max_key_type indicates the highest item type to remove.
4039 * This cannot be run for file data extents because it does not
4040 * free the extents they point to.
4042 static int drop_inode_items(struct btrfs_trans_handle *trans,
4043 struct btrfs_root *log,
4044 struct btrfs_path *path,
4045 struct btrfs_inode *inode,
4049 struct btrfs_key key;
4050 struct btrfs_key found_key;
4053 key.objectid = btrfs_ino(inode);
4054 key.type = max_key_type;
4055 key.offset = (u64)-1;
4058 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4061 } else if (ret > 0) {
4062 if (path->slots[0] == 0)
4067 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4070 if (found_key.objectid != key.objectid)
4073 found_key.offset = 0;
4075 ret = btrfs_bin_search(path->nodes[0], 0, &found_key, &start_slot);
4079 ret = btrfs_del_items(trans, log, path, start_slot,
4080 path->slots[0] - start_slot + 1);
4082 * If start slot isn't 0 then we don't need to re-search, we've
4083 * found the last guy with the objectid in this tree.
4085 if (ret || start_slot != 0)
4087 btrfs_release_path(path);
4089 btrfs_release_path(path);
4095 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4096 struct btrfs_root *log_root,
4097 struct btrfs_inode *inode,
4098 u64 new_size, u32 min_type)
4100 struct btrfs_truncate_control control = {
4101 .new_size = new_size,
4102 .ino = btrfs_ino(inode),
4103 .min_type = min_type,
4104 .skip_ref_updates = true,
4107 return btrfs_truncate_inode_items(trans, log_root, &control);
4110 static void fill_inode_item(struct btrfs_trans_handle *trans,
4111 struct extent_buffer *leaf,
4112 struct btrfs_inode_item *item,
4113 struct inode *inode, int log_inode_only,
4116 struct btrfs_map_token token;
4119 btrfs_init_map_token(&token, leaf);
4121 if (log_inode_only) {
4122 /* set the generation to zero so the recover code
4123 * can tell the difference between an logging
4124 * just to say 'this inode exists' and a logging
4125 * to say 'update this inode with these values'
4127 btrfs_set_token_inode_generation(&token, item, 0);
4128 btrfs_set_token_inode_size(&token, item, logged_isize);
4130 btrfs_set_token_inode_generation(&token, item,
4131 BTRFS_I(inode)->generation);
4132 btrfs_set_token_inode_size(&token, item, inode->i_size);
4135 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4136 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4137 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4138 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4140 btrfs_set_token_timespec_sec(&token, &item->atime,
4141 inode->i_atime.tv_sec);
4142 btrfs_set_token_timespec_nsec(&token, &item->atime,
4143 inode->i_atime.tv_nsec);
4145 btrfs_set_token_timespec_sec(&token, &item->mtime,
4146 inode->i_mtime.tv_sec);
4147 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4148 inode->i_mtime.tv_nsec);
4150 btrfs_set_token_timespec_sec(&token, &item->ctime,
4151 inode_get_ctime(inode).tv_sec);
4152 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4153 inode_get_ctime(inode).tv_nsec);
4156 * We do not need to set the nbytes field, in fact during a fast fsync
4157 * its value may not even be correct, since a fast fsync does not wait
4158 * for ordered extent completion, which is where we update nbytes, it
4159 * only waits for writeback to complete. During log replay as we find
4160 * file extent items and replay them, we adjust the nbytes field of the
4161 * inode item in subvolume tree as needed (see overwrite_item()).
4164 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4165 btrfs_set_token_inode_transid(&token, item, trans->transid);
4166 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4167 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4168 BTRFS_I(inode)->ro_flags);
4169 btrfs_set_token_inode_flags(&token, item, flags);
4170 btrfs_set_token_inode_block_group(&token, item, 0);
4173 static int log_inode_item(struct btrfs_trans_handle *trans,
4174 struct btrfs_root *log, struct btrfs_path *path,
4175 struct btrfs_inode *inode, bool inode_item_dropped)
4177 struct btrfs_inode_item *inode_item;
4181 * If we are doing a fast fsync and the inode was logged before in the
4182 * current transaction, then we know the inode was previously logged and
4183 * it exists in the log tree. For performance reasons, in this case use
4184 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4185 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4186 * contention in case there are concurrent fsyncs for other inodes of the
4187 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4188 * already exists can also result in unnecessarily splitting a leaf.
4190 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4191 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4197 * This means it is the first fsync in the current transaction,
4198 * so the inode item is not in the log and we need to insert it.
4199 * We can never get -EEXIST because we are only called for a fast
4200 * fsync and in case an inode eviction happens after the inode was
4201 * logged before in the current transaction, when we load again
4202 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4203 * flags and set ->logged_trans to 0.
4205 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4206 sizeof(*inode_item));
4207 ASSERT(ret != -EEXIST);
4211 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4212 struct btrfs_inode_item);
4213 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4215 btrfs_release_path(path);
4219 static int log_csums(struct btrfs_trans_handle *trans,
4220 struct btrfs_inode *inode,
4221 struct btrfs_root *log_root,
4222 struct btrfs_ordered_sum *sums)
4224 const u64 lock_end = sums->logical + sums->len - 1;
4225 struct extent_state *cached_state = NULL;
4229 * If this inode was not used for reflink operations in the current
4230 * transaction with new extents, then do the fast path, no need to
4231 * worry about logging checksum items with overlapping ranges.
4233 if (inode->last_reflink_trans < trans->transid)
4234 return btrfs_csum_file_blocks(trans, log_root, sums);
4237 * Serialize logging for checksums. This is to avoid racing with the
4238 * same checksum being logged by another task that is logging another
4239 * file which happens to refer to the same extent as well. Such races
4240 * can leave checksum items in the log with overlapping ranges.
4242 ret = lock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4247 * Due to extent cloning, we might have logged a csum item that covers a
4248 * subrange of a cloned extent, and later we can end up logging a csum
4249 * item for a larger subrange of the same extent or the entire range.
4250 * This would leave csum items in the log tree that cover the same range
4251 * and break the searches for checksums in the log tree, resulting in
4252 * some checksums missing in the fs/subvolume tree. So just delete (or
4253 * trim and adjust) any existing csum items in the log for this range.
4255 ret = btrfs_del_csums(trans, log_root, sums->logical, sums->len);
4257 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4259 unlock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4265 static noinline int copy_items(struct btrfs_trans_handle *trans,
4266 struct btrfs_inode *inode,
4267 struct btrfs_path *dst_path,
4268 struct btrfs_path *src_path,
4269 int start_slot, int nr, int inode_only,
4272 struct btrfs_root *log = inode->root->log_root;
4273 struct btrfs_file_extent_item *extent;
4274 struct extent_buffer *src;
4276 struct btrfs_key *ins_keys;
4278 struct btrfs_item_batch batch;
4282 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4283 const u64 i_size = i_size_read(&inode->vfs_inode);
4286 * To keep lockdep happy and avoid deadlocks, clone the source leaf and
4287 * use the clone. This is because otherwise we would be changing the log
4288 * tree, to insert items from the subvolume tree or insert csum items,
4289 * while holding a read lock on a leaf from the subvolume tree, which
4290 * creates a nasty lock dependency when COWing log tree nodes/leaves:
4292 * 1) Modifying the log tree triggers an extent buffer allocation while
4293 * holding a write lock on a parent extent buffer from the log tree.
4294 * Allocating the pages for an extent buffer, or the extent buffer
4295 * struct, can trigger inode eviction and finally the inode eviction
4296 * will trigger a release/remove of a delayed node, which requires
4297 * taking the delayed node's mutex;
4299 * 2) Allocating a metadata extent for a log tree can trigger the async
4300 * reclaim thread and make us wait for it to release enough space and
4301 * unblock our reservation ticket. The reclaim thread can start
4302 * flushing delayed items, and that in turn results in the need to
4303 * lock delayed node mutexes and in the need to write lock extent
4304 * buffers of a subvolume tree - all this while holding a write lock
4305 * on the parent extent buffer in the log tree.
4307 * So one task in scenario 1) running in parallel with another task in
4308 * scenario 2) could lead to a deadlock, one wanting to lock a delayed
4309 * node mutex while having a read lock on a leaf from the subvolume,
4310 * while the other is holding the delayed node's mutex and wants to
4311 * write lock the same subvolume leaf for flushing delayed items.
4313 src = btrfs_clone_extent_buffer(src_path->nodes[0]);
4317 i = src_path->slots[0];
4318 btrfs_release_path(src_path);
4319 src_path->nodes[0] = src;
4320 src_path->slots[0] = i;
4322 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4323 nr * sizeof(u32), GFP_NOFS);
4327 ins_sizes = (u32 *)ins_data;
4328 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4329 batch.keys = ins_keys;
4330 batch.data_sizes = ins_sizes;
4331 batch.total_data_size = 0;
4335 for (i = 0; i < nr; i++) {
4336 const int src_slot = start_slot + i;
4337 struct btrfs_root *csum_root;
4338 struct btrfs_ordered_sum *sums;
4339 struct btrfs_ordered_sum *sums_next;
4340 LIST_HEAD(ordered_sums);
4344 u64 extent_num_bytes;
4347 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4349 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4352 extent = btrfs_item_ptr(src, src_slot,
4353 struct btrfs_file_extent_item);
4355 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4359 * Don't copy extents from past generations. That would make us
4360 * log a lot more metadata for common cases like doing only a
4361 * few random writes into a file and then fsync it for the first
4362 * time or after the full sync flag is set on the inode. We can
4363 * get leaves full of extent items, most of which are from past
4364 * generations, so we can skip them - as long as the inode has
4365 * not been the target of a reflink operation in this transaction,
4366 * as in that case it might have had file extent items with old
4367 * generations copied into it. We also must always log prealloc
4368 * extents that start at or beyond eof, otherwise we would lose
4369 * them on log replay.
4371 if (is_old_extent &&
4372 ins_keys[dst_index].offset < i_size &&
4373 inode->last_reflink_trans < trans->transid)
4379 /* Only regular extents have checksums. */
4380 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4384 * If it's an extent created in a past transaction, then its
4385 * checksums are already accessible from the committed csum tree,
4386 * no need to log them.
4391 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4392 /* If it's an explicit hole, there are no checksums. */
4393 if (disk_bytenr == 0)
4396 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4398 if (btrfs_file_extent_compression(src, extent)) {
4400 extent_num_bytes = disk_num_bytes;
4402 extent_offset = btrfs_file_extent_offset(src, extent);
4403 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4406 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4407 disk_bytenr += extent_offset;
4408 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4409 disk_bytenr + extent_num_bytes - 1,
4410 &ordered_sums, 0, false);
4414 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4416 ret = log_csums(trans, inode, log, sums);
4417 list_del(&sums->list);
4424 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4425 batch.total_data_size += ins_sizes[dst_index];
4431 * We have a leaf full of old extent items that don't need to be logged,
4432 * so we don't need to do anything.
4437 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4442 for (i = 0; i < nr; i++) {
4443 const int src_slot = start_slot + i;
4444 const int dst_slot = dst_path->slots[0] + dst_index;
4445 struct btrfs_key key;
4446 unsigned long src_offset;
4447 unsigned long dst_offset;
4450 * We're done, all the remaining items in the source leaf
4451 * correspond to old file extent items.
4453 if (dst_index >= batch.nr)
4456 btrfs_item_key_to_cpu(src, &key, src_slot);
4458 if (key.type != BTRFS_EXTENT_DATA_KEY)
4461 extent = btrfs_item_ptr(src, src_slot,
4462 struct btrfs_file_extent_item);
4464 /* See the comment in the previous loop, same logic. */
4465 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4466 key.offset < i_size &&
4467 inode->last_reflink_trans < trans->transid)
4471 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4472 src_offset = btrfs_item_ptr_offset(src, src_slot);
4474 if (key.type == BTRFS_INODE_ITEM_KEY) {
4475 struct btrfs_inode_item *inode_item;
4477 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4478 struct btrfs_inode_item);
4479 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4481 inode_only == LOG_INODE_EXISTS,
4484 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4485 src_offset, ins_sizes[dst_index]);
4491 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4492 btrfs_release_path(dst_path);
4499 static int extent_cmp(void *priv, const struct list_head *a,
4500 const struct list_head *b)
4502 const struct extent_map *em1, *em2;
4504 em1 = list_entry(a, struct extent_map, list);
4505 em2 = list_entry(b, struct extent_map, list);
4507 if (em1->start < em2->start)
4509 else if (em1->start > em2->start)
4514 static int log_extent_csums(struct btrfs_trans_handle *trans,
4515 struct btrfs_inode *inode,
4516 struct btrfs_root *log_root,
4517 const struct extent_map *em,
4518 struct btrfs_log_ctx *ctx)
4520 struct btrfs_ordered_extent *ordered;
4521 struct btrfs_root *csum_root;
4524 u64 mod_start = em->mod_start;
4525 u64 mod_len = em->mod_len;
4526 LIST_HEAD(ordered_sums);
4529 if (inode->flags & BTRFS_INODE_NODATASUM ||
4530 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4531 em->block_start == EXTENT_MAP_HOLE)
4534 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4535 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4536 const u64 mod_end = mod_start + mod_len;
4537 struct btrfs_ordered_sum *sums;
4542 if (ordered_end <= mod_start)
4544 if (mod_end <= ordered->file_offset)
4548 * We are going to copy all the csums on this ordered extent, so
4549 * go ahead and adjust mod_start and mod_len in case this ordered
4550 * extent has already been logged.
4552 if (ordered->file_offset > mod_start) {
4553 if (ordered_end >= mod_end)
4554 mod_len = ordered->file_offset - mod_start;
4556 * If we have this case
4558 * |--------- logged extent ---------|
4559 * |----- ordered extent ----|
4561 * Just don't mess with mod_start and mod_len, we'll
4562 * just end up logging more csums than we need and it
4566 if (ordered_end < mod_end) {
4567 mod_len = mod_end - ordered_end;
4568 mod_start = ordered_end;
4575 * To keep us from looping for the above case of an ordered
4576 * extent that falls inside of the logged extent.
4578 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4581 list_for_each_entry(sums, &ordered->list, list) {
4582 ret = log_csums(trans, inode, log_root, sums);
4588 /* We're done, found all csums in the ordered extents. */
4592 /* If we're compressed we have to save the entire range of csums. */
4593 if (em->compress_type) {
4595 csum_len = max(em->block_len, em->orig_block_len);
4597 csum_offset = mod_start - em->start;
4601 /* block start is already adjusted for the file extent offset. */
4602 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4603 ret = btrfs_lookup_csums_list(csum_root, em->block_start + csum_offset,
4604 em->block_start + csum_offset +
4605 csum_len - 1, &ordered_sums, 0, false);
4609 while (!list_empty(&ordered_sums)) {
4610 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4611 struct btrfs_ordered_sum,
4614 ret = log_csums(trans, inode, log_root, sums);
4615 list_del(&sums->list);
4622 static int log_one_extent(struct btrfs_trans_handle *trans,
4623 struct btrfs_inode *inode,
4624 const struct extent_map *em,
4625 struct btrfs_path *path,
4626 struct btrfs_log_ctx *ctx)
4628 struct btrfs_drop_extents_args drop_args = { 0 };
4629 struct btrfs_root *log = inode->root->log_root;
4630 struct btrfs_file_extent_item fi = { 0 };
4631 struct extent_buffer *leaf;
4632 struct btrfs_key key;
4633 u64 extent_offset = em->start - em->orig_start;
4637 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4638 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4639 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4641 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4643 block_len = max(em->block_len, em->orig_block_len);
4644 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4645 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4646 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4647 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4648 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4650 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4653 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4654 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4655 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4656 btrfs_set_stack_file_extent_compression(&fi, em->compress_type);
4658 ret = log_extent_csums(trans, inode, log, em, ctx);
4663 * If this is the first time we are logging the inode in the current
4664 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4665 * because it does a deletion search, which always acquires write locks
4666 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4667 * but also adds significant contention in a log tree, since log trees
4668 * are small, with a root at level 2 or 3 at most, due to their short
4671 if (ctx->logged_before) {
4672 drop_args.path = path;
4673 drop_args.start = em->start;
4674 drop_args.end = em->start + em->len;
4675 drop_args.replace_extent = true;
4676 drop_args.extent_item_size = sizeof(fi);
4677 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4682 if (!drop_args.extent_inserted) {
4683 key.objectid = btrfs_ino(inode);
4684 key.type = BTRFS_EXTENT_DATA_KEY;
4685 key.offset = em->start;
4687 ret = btrfs_insert_empty_item(trans, log, path, &key,
4692 leaf = path->nodes[0];
4693 write_extent_buffer(leaf, &fi,
4694 btrfs_item_ptr_offset(leaf, path->slots[0]),
4696 btrfs_mark_buffer_dirty(leaf);
4698 btrfs_release_path(path);
4704 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4705 * lose them after doing a full/fast fsync and replaying the log. We scan the
4706 * subvolume's root instead of iterating the inode's extent map tree because
4707 * otherwise we can log incorrect extent items based on extent map conversion.
4708 * That can happen due to the fact that extent maps are merged when they
4709 * are not in the extent map tree's list of modified extents.
4711 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4712 struct btrfs_inode *inode,
4713 struct btrfs_path *path)
4715 struct btrfs_root *root = inode->root;
4716 struct btrfs_key key;
4717 const u64 i_size = i_size_read(&inode->vfs_inode);
4718 const u64 ino = btrfs_ino(inode);
4719 struct btrfs_path *dst_path = NULL;
4720 bool dropped_extents = false;
4721 u64 truncate_offset = i_size;
4722 struct extent_buffer *leaf;
4728 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4732 key.type = BTRFS_EXTENT_DATA_KEY;
4733 key.offset = i_size;
4734 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4739 * We must check if there is a prealloc extent that starts before the
4740 * i_size and crosses the i_size boundary. This is to ensure later we
4741 * truncate down to the end of that extent and not to the i_size, as
4742 * otherwise we end up losing part of the prealloc extent after a log
4743 * replay and with an implicit hole if there is another prealloc extent
4744 * that starts at an offset beyond i_size.
4746 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4751 struct btrfs_file_extent_item *ei;
4753 leaf = path->nodes[0];
4754 slot = path->slots[0];
4755 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4757 if (btrfs_file_extent_type(leaf, ei) ==
4758 BTRFS_FILE_EXTENT_PREALLOC) {
4761 btrfs_item_key_to_cpu(leaf, &key, slot);
4762 extent_end = key.offset +
4763 btrfs_file_extent_num_bytes(leaf, ei);
4765 if (extent_end > i_size)
4766 truncate_offset = extent_end;
4773 leaf = path->nodes[0];
4774 slot = path->slots[0];
4776 if (slot >= btrfs_header_nritems(leaf)) {
4778 ret = copy_items(trans, inode, dst_path, path,
4779 start_slot, ins_nr, 1, 0);
4784 ret = btrfs_next_leaf(root, path);
4794 btrfs_item_key_to_cpu(leaf, &key, slot);
4795 if (key.objectid > ino)
4797 if (WARN_ON_ONCE(key.objectid < ino) ||
4798 key.type < BTRFS_EXTENT_DATA_KEY ||
4799 key.offset < i_size) {
4803 if (!dropped_extents) {
4805 * Avoid logging extent items logged in past fsync calls
4806 * and leading to duplicate keys in the log tree.
4808 ret = truncate_inode_items(trans, root->log_root, inode,
4810 BTRFS_EXTENT_DATA_KEY);
4813 dropped_extents = true;
4820 dst_path = btrfs_alloc_path();
4828 ret = copy_items(trans, inode, dst_path, path,
4829 start_slot, ins_nr, 1, 0);
4831 btrfs_release_path(path);
4832 btrfs_free_path(dst_path);
4836 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4837 struct btrfs_inode *inode,
4838 struct btrfs_path *path,
4839 struct btrfs_log_ctx *ctx)
4841 struct btrfs_ordered_extent *ordered;
4842 struct btrfs_ordered_extent *tmp;
4843 struct extent_map *em, *n;
4845 struct extent_map_tree *tree = &inode->extent_tree;
4849 write_lock(&tree->lock);
4851 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4852 list_del_init(&em->list);
4854 * Just an arbitrary number, this can be really CPU intensive
4855 * once we start getting a lot of extents, and really once we
4856 * have a bunch of extents we just want to commit since it will
4859 if (++num > 32768) {
4860 list_del_init(&tree->modified_extents);
4865 if (em->generation < trans->transid)
4868 /* We log prealloc extents beyond eof later. */
4869 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4870 em->start >= i_size_read(&inode->vfs_inode))
4873 /* Need a ref to keep it from getting evicted from cache */
4874 refcount_inc(&em->refs);
4875 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4876 list_add_tail(&em->list, &extents);
4880 list_sort(NULL, &extents, extent_cmp);
4882 while (!list_empty(&extents)) {
4883 em = list_entry(extents.next, struct extent_map, list);
4885 list_del_init(&em->list);
4888 * If we had an error we just need to delete everybody from our
4892 clear_em_logging(tree, em);
4893 free_extent_map(em);
4897 write_unlock(&tree->lock);
4899 ret = log_one_extent(trans, inode, em, path, ctx);
4900 write_lock(&tree->lock);
4901 clear_em_logging(tree, em);
4902 free_extent_map(em);
4904 WARN_ON(!list_empty(&extents));
4905 write_unlock(&tree->lock);
4908 ret = btrfs_log_prealloc_extents(trans, inode, path);
4913 * We have logged all extents successfully, now make sure the commit of
4914 * the current transaction waits for the ordered extents to complete
4915 * before it commits and wipes out the log trees, otherwise we would
4916 * lose data if an ordered extents completes after the transaction
4917 * commits and a power failure happens after the transaction commit.
4919 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4920 list_del_init(&ordered->log_list);
4921 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4923 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4924 spin_lock_irq(&inode->ordered_tree.lock);
4925 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4926 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4927 atomic_inc(&trans->transaction->pending_ordered);
4929 spin_unlock_irq(&inode->ordered_tree.lock);
4931 btrfs_put_ordered_extent(ordered);
4937 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4938 struct btrfs_path *path, u64 *size_ret)
4940 struct btrfs_key key;
4943 key.objectid = btrfs_ino(inode);
4944 key.type = BTRFS_INODE_ITEM_KEY;
4947 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4950 } else if (ret > 0) {
4953 struct btrfs_inode_item *item;
4955 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4956 struct btrfs_inode_item);
4957 *size_ret = btrfs_inode_size(path->nodes[0], item);
4959 * If the in-memory inode's i_size is smaller then the inode
4960 * size stored in the btree, return the inode's i_size, so
4961 * that we get a correct inode size after replaying the log
4962 * when before a power failure we had a shrinking truncate
4963 * followed by addition of a new name (rename / new hard link).
4964 * Otherwise return the inode size from the btree, to avoid
4965 * data loss when replaying a log due to previously doing a
4966 * write that expands the inode's size and logging a new name
4967 * immediately after.
4969 if (*size_ret > inode->vfs_inode.i_size)
4970 *size_ret = inode->vfs_inode.i_size;
4973 btrfs_release_path(path);
4978 * At the moment we always log all xattrs. This is to figure out at log replay
4979 * time which xattrs must have their deletion replayed. If a xattr is missing
4980 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4981 * because if a xattr is deleted, the inode is fsynced and a power failure
4982 * happens, causing the log to be replayed the next time the fs is mounted,
4983 * we want the xattr to not exist anymore (same behaviour as other filesystems
4984 * with a journal, ext3/4, xfs, f2fs, etc).
4986 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4987 struct btrfs_inode *inode,
4988 struct btrfs_path *path,
4989 struct btrfs_path *dst_path)
4991 struct btrfs_root *root = inode->root;
4993 struct btrfs_key key;
4994 const u64 ino = btrfs_ino(inode);
4997 bool found_xattrs = false;
4999 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
5003 key.type = BTRFS_XATTR_ITEM_KEY;
5006 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5011 int slot = path->slots[0];
5012 struct extent_buffer *leaf = path->nodes[0];
5013 int nritems = btrfs_header_nritems(leaf);
5015 if (slot >= nritems) {
5017 ret = copy_items(trans, inode, dst_path, path,
5018 start_slot, ins_nr, 1, 0);
5023 ret = btrfs_next_leaf(root, path);
5031 btrfs_item_key_to_cpu(leaf, &key, slot);
5032 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5039 found_xattrs = true;
5043 ret = copy_items(trans, inode, dst_path, path,
5044 start_slot, ins_nr, 1, 0);
5050 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5056 * When using the NO_HOLES feature if we punched a hole that causes the
5057 * deletion of entire leafs or all the extent items of the first leaf (the one
5058 * that contains the inode item and references) we may end up not processing
5059 * any extents, because there are no leafs with a generation matching the
5060 * current transaction that have extent items for our inode. So we need to find
5061 * if any holes exist and then log them. We also need to log holes after any
5062 * truncate operation that changes the inode's size.
5064 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5065 struct btrfs_inode *inode,
5066 struct btrfs_path *path)
5068 struct btrfs_root *root = inode->root;
5069 struct btrfs_fs_info *fs_info = root->fs_info;
5070 struct btrfs_key key;
5071 const u64 ino = btrfs_ino(inode);
5072 const u64 i_size = i_size_read(&inode->vfs_inode);
5073 u64 prev_extent_end = 0;
5076 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5080 key.type = BTRFS_EXTENT_DATA_KEY;
5083 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5088 struct extent_buffer *leaf = path->nodes[0];
5090 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5091 ret = btrfs_next_leaf(root, path);
5098 leaf = path->nodes[0];
5101 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5102 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5105 /* We have a hole, log it. */
5106 if (prev_extent_end < key.offset) {
5107 const u64 hole_len = key.offset - prev_extent_end;
5110 * Release the path to avoid deadlocks with other code
5111 * paths that search the root while holding locks on
5112 * leafs from the log root.
5114 btrfs_release_path(path);
5115 ret = btrfs_insert_hole_extent(trans, root->log_root,
5116 ino, prev_extent_end,
5122 * Search for the same key again in the root. Since it's
5123 * an extent item and we are holding the inode lock, the
5124 * key must still exist. If it doesn't just emit warning
5125 * and return an error to fall back to a transaction
5128 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5131 if (WARN_ON(ret > 0))
5133 leaf = path->nodes[0];
5136 prev_extent_end = btrfs_file_extent_end(path);
5141 if (prev_extent_end < i_size) {
5144 btrfs_release_path(path);
5145 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5146 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5147 prev_extent_end, hole_len);
5156 * When we are logging a new inode X, check if it doesn't have a reference that
5157 * matches the reference from some other inode Y created in a past transaction
5158 * and that was renamed in the current transaction. If we don't do this, then at
5159 * log replay time we can lose inode Y (and all its files if it's a directory):
5162 * echo "hello world" > /mnt/x/foobar
5165 * mkdir /mnt/x # or touch /mnt/x
5166 * xfs_io -c fsync /mnt/x
5168 * mount fs, trigger log replay
5170 * After the log replay procedure, we would lose the first directory and all its
5171 * files (file foobar).
5172 * For the case where inode Y is not a directory we simply end up losing it:
5174 * echo "123" > /mnt/foo
5176 * mv /mnt/foo /mnt/bar
5177 * echo "abc" > /mnt/foo
5178 * xfs_io -c fsync /mnt/foo
5181 * We also need this for cases where a snapshot entry is replaced by some other
5182 * entry (file or directory) otherwise we end up with an unreplayable log due to
5183 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5184 * if it were a regular entry:
5187 * btrfs subvolume snapshot /mnt /mnt/x/snap
5188 * btrfs subvolume delete /mnt/x/snap
5191 * fsync /mnt/x or fsync some new file inside it
5194 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5195 * the same transaction.
5197 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5199 const struct btrfs_key *key,
5200 struct btrfs_inode *inode,
5201 u64 *other_ino, u64 *other_parent)
5204 struct btrfs_path *search_path;
5207 u32 item_size = btrfs_item_size(eb, slot);
5209 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5211 search_path = btrfs_alloc_path();
5214 search_path->search_commit_root = 1;
5215 search_path->skip_locking = 1;
5217 while (cur_offset < item_size) {
5221 unsigned long name_ptr;
5222 struct btrfs_dir_item *di;
5223 struct fscrypt_str name_str;
5225 if (key->type == BTRFS_INODE_REF_KEY) {
5226 struct btrfs_inode_ref *iref;
5228 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5229 parent = key->offset;
5230 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5231 name_ptr = (unsigned long)(iref + 1);
5232 this_len = sizeof(*iref) + this_name_len;
5234 struct btrfs_inode_extref *extref;
5236 extref = (struct btrfs_inode_extref *)(ptr +
5238 parent = btrfs_inode_extref_parent(eb, extref);
5239 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5240 name_ptr = (unsigned long)&extref->name;
5241 this_len = sizeof(*extref) + this_name_len;
5244 if (this_name_len > name_len) {
5247 new_name = krealloc(name, this_name_len, GFP_NOFS);
5252 name_len = this_name_len;
5256 read_extent_buffer(eb, name, name_ptr, this_name_len);
5258 name_str.name = name;
5259 name_str.len = this_name_len;
5260 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5261 parent, &name_str, 0);
5262 if (di && !IS_ERR(di)) {
5263 struct btrfs_key di_key;
5265 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5267 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5268 if (di_key.objectid != key->objectid) {
5270 *other_ino = di_key.objectid;
5271 *other_parent = parent;
5279 } else if (IS_ERR(di)) {
5283 btrfs_release_path(search_path);
5285 cur_offset += this_len;
5289 btrfs_free_path(search_path);
5295 * Check if we need to log an inode. This is used in contexts where while
5296 * logging an inode we need to log another inode (either that it exists or in
5297 * full mode). This is used instead of btrfs_inode_in_log() because the later
5298 * requires the inode to be in the log and have the log transaction committed,
5299 * while here we do not care if the log transaction was already committed - our
5300 * caller will commit the log later - and we want to avoid logging an inode
5301 * multiple times when multiple tasks have joined the same log transaction.
5303 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5304 struct btrfs_inode *inode)
5307 * If a directory was not modified, no dentries added or removed, we can
5308 * and should avoid logging it.
5310 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5314 * If this inode does not have new/updated/deleted xattrs since the last
5315 * time it was logged and is flagged as logged in the current transaction,
5316 * we can skip logging it. As for new/deleted names, those are updated in
5317 * the log by link/unlink/rename operations.
5318 * In case the inode was logged and then evicted and reloaded, its
5319 * logged_trans will be 0, in which case we have to fully log it since
5320 * logged_trans is a transient field, not persisted.
5322 if (inode_logged(trans, inode, NULL) == 1 &&
5323 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5329 struct btrfs_dir_list {
5331 struct list_head list;
5335 * Log the inodes of the new dentries of a directory.
5336 * See process_dir_items_leaf() for details about why it is needed.
5337 * This is a recursive operation - if an existing dentry corresponds to a
5338 * directory, that directory's new entries are logged too (same behaviour as
5339 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5340 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5341 * complains about the following circular lock dependency / possible deadlock:
5345 * lock(&type->i_mutex_dir_key#3/2);
5346 * lock(sb_internal#2);
5347 * lock(&type->i_mutex_dir_key#3/2);
5348 * lock(&sb->s_type->i_mutex_key#14);
5350 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5351 * sb_start_intwrite() in btrfs_start_transaction().
5352 * Not acquiring the VFS lock of the inodes is still safe because:
5354 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5355 * that while logging the inode new references (names) are added or removed
5356 * from the inode, leaving the logged inode item with a link count that does
5357 * not match the number of logged inode reference items. This is fine because
5358 * at log replay time we compute the real number of links and correct the
5359 * link count in the inode item (see replay_one_buffer() and
5360 * link_to_fixup_dir());
5362 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5363 * while logging the inode's items new index items (key type
5364 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5365 * has a size that doesn't match the sum of the lengths of all the logged
5366 * names - this is ok, not a problem, because at log replay time we set the
5367 * directory's i_size to the correct value (see replay_one_name() and
5368 * overwrite_item()).
5370 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5371 struct btrfs_inode *start_inode,
5372 struct btrfs_log_ctx *ctx)
5374 struct btrfs_root *root = start_inode->root;
5375 struct btrfs_fs_info *fs_info = root->fs_info;
5376 struct btrfs_path *path;
5377 LIST_HEAD(dir_list);
5378 struct btrfs_dir_list *dir_elem;
5379 u64 ino = btrfs_ino(start_inode);
5380 struct btrfs_inode *curr_inode = start_inode;
5384 * If we are logging a new name, as part of a link or rename operation,
5385 * don't bother logging new dentries, as we just want to log the names
5386 * of an inode and that any new parents exist.
5388 if (ctx->logging_new_name)
5391 path = btrfs_alloc_path();
5395 /* Pairs with btrfs_add_delayed_iput below. */
5396 ihold(&curr_inode->vfs_inode);
5399 struct inode *vfs_inode;
5400 struct btrfs_key key;
5401 struct btrfs_key found_key;
5403 bool continue_curr_inode = true;
5407 key.type = BTRFS_DIR_INDEX_KEY;
5408 key.offset = btrfs_get_first_dir_index_to_log(curr_inode);
5409 next_index = key.offset;
5411 btrfs_for_each_slot(root->log_root, &key, &found_key, path, iter_ret) {
5412 struct extent_buffer *leaf = path->nodes[0];
5413 struct btrfs_dir_item *di;
5414 struct btrfs_key di_key;
5415 struct inode *di_inode;
5416 int log_mode = LOG_INODE_EXISTS;
5419 if (found_key.objectid != ino ||
5420 found_key.type != BTRFS_DIR_INDEX_KEY) {
5421 continue_curr_inode = false;
5425 next_index = found_key.offset + 1;
5427 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5428 type = btrfs_dir_ftype(leaf, di);
5429 if (btrfs_dir_transid(leaf, di) < trans->transid)
5431 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5432 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5435 btrfs_release_path(path);
5436 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5437 if (IS_ERR(di_inode)) {
5438 ret = PTR_ERR(di_inode);
5442 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5443 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5447 ctx->log_new_dentries = false;
5448 if (type == BTRFS_FT_DIR)
5449 log_mode = LOG_INODE_ALL;
5450 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5452 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5455 if (ctx->log_new_dentries) {
5456 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5461 dir_elem->ino = di_key.objectid;
5462 list_add_tail(&dir_elem->list, &dir_list);
5467 btrfs_release_path(path);
5472 } else if (iter_ret > 0) {
5473 continue_curr_inode = false;
5478 if (continue_curr_inode && key.offset < (u64)-1) {
5483 btrfs_set_first_dir_index_to_log(curr_inode, next_index);
5485 if (list_empty(&dir_list))
5488 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5489 ino = dir_elem->ino;
5490 list_del(&dir_elem->list);
5493 btrfs_add_delayed_iput(curr_inode);
5496 vfs_inode = btrfs_iget(fs_info->sb, ino, root);
5497 if (IS_ERR(vfs_inode)) {
5498 ret = PTR_ERR(vfs_inode);
5501 curr_inode = BTRFS_I(vfs_inode);
5504 btrfs_free_path(path);
5506 btrfs_add_delayed_iput(curr_inode);
5509 struct btrfs_dir_list *next;
5511 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5518 struct btrfs_ino_list {
5521 struct list_head list;
5524 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5526 struct btrfs_ino_list *curr;
5527 struct btrfs_ino_list *next;
5529 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5530 list_del(&curr->list);
5535 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5536 struct btrfs_path *path)
5538 struct btrfs_key key;
5542 key.type = BTRFS_INODE_ITEM_KEY;
5545 path->search_commit_root = 1;
5546 path->skip_locking = 1;
5548 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5549 if (WARN_ON_ONCE(ret > 0)) {
5551 * We have previously found the inode through the commit root
5552 * so this should not happen. If it does, just error out and
5553 * fallback to a transaction commit.
5556 } else if (ret == 0) {
5557 struct btrfs_inode_item *item;
5559 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5560 struct btrfs_inode_item);
5561 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5565 btrfs_release_path(path);
5566 path->search_commit_root = 0;
5567 path->skip_locking = 0;
5572 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5573 struct btrfs_root *root,
5574 struct btrfs_path *path,
5575 u64 ino, u64 parent,
5576 struct btrfs_log_ctx *ctx)
5578 struct btrfs_ino_list *ino_elem;
5579 struct inode *inode;
5582 * It's rare to have a lot of conflicting inodes, in practice it is not
5583 * common to have more than 1 or 2. We don't want to collect too many,
5584 * as we could end up logging too many inodes (even if only in
5585 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5588 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES)
5589 return BTRFS_LOG_FORCE_COMMIT;
5591 inode = btrfs_iget(root->fs_info->sb, ino, root);
5593 * If the other inode that had a conflicting dir entry was deleted in
5594 * the current transaction then we either:
5596 * 1) Log the parent directory (later after adding it to the list) if
5597 * the inode is a directory. This is because it may be a deleted
5598 * subvolume/snapshot or it may be a regular directory that had
5599 * deleted subvolumes/snapshots (or subdirectories that had them),
5600 * and at the moment we can't deal with dropping subvolumes/snapshots
5601 * during log replay. So we just log the parent, which will result in
5602 * a fallback to a transaction commit if we are dealing with those
5603 * cases (last_unlink_trans will match the current transaction);
5605 * 2) Do nothing if it's not a directory. During log replay we simply
5606 * unlink the conflicting dentry from the parent directory and then
5607 * add the dentry for our inode. Like this we can avoid logging the
5608 * parent directory (and maybe fallback to a transaction commit in
5609 * case it has a last_unlink_trans == trans->transid, due to moving
5610 * some inode from it to some other directory).
5612 if (IS_ERR(inode)) {
5613 int ret = PTR_ERR(inode);
5618 ret = conflicting_inode_is_dir(root, ino, path);
5619 /* Not a directory or we got an error. */
5623 /* Conflicting inode is a directory, so we'll log its parent. */
5624 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5627 ino_elem->ino = ino;
5628 ino_elem->parent = parent;
5629 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5630 ctx->num_conflict_inodes++;
5636 * If the inode was already logged skip it - otherwise we can hit an
5637 * infinite loop. Example:
5639 * From the commit root (previous transaction) we have the following
5642 * inode 257 a directory
5643 * inode 258 with references "zz" and "zz_link" on inode 257
5644 * inode 259 with reference "a" on inode 257
5646 * And in the current (uncommitted) transaction we have:
5648 * inode 257 a directory, unchanged
5649 * inode 258 with references "a" and "a2" on inode 257
5650 * inode 259 with reference "zz_link" on inode 257
5651 * inode 261 with reference "zz" on inode 257
5653 * When logging inode 261 the following infinite loop could
5654 * happen if we don't skip already logged inodes:
5656 * - we detect inode 258 as a conflicting inode, with inode 261
5657 * on reference "zz", and log it;
5659 * - we detect inode 259 as a conflicting inode, with inode 258
5660 * on reference "a", and log it;
5662 * - we detect inode 258 as a conflicting inode, with inode 259
5663 * on reference "zz_link", and log it - again! After this we
5664 * repeat the above steps forever.
5666 * Here we can use need_log_inode() because we only need to log the
5667 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5668 * so that the log ends up with the new name and without the old name.
5670 if (!need_log_inode(trans, BTRFS_I(inode))) {
5671 btrfs_add_delayed_iput(BTRFS_I(inode));
5675 btrfs_add_delayed_iput(BTRFS_I(inode));
5677 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5680 ino_elem->ino = ino;
5681 ino_elem->parent = parent;
5682 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5683 ctx->num_conflict_inodes++;
5688 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5689 struct btrfs_root *root,
5690 struct btrfs_log_ctx *ctx)
5692 struct btrfs_fs_info *fs_info = root->fs_info;
5696 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5697 * otherwise we could have unbounded recursion of btrfs_log_inode()
5698 * calls. This check guarantees we can have only 1 level of recursion.
5700 if (ctx->logging_conflict_inodes)
5703 ctx->logging_conflict_inodes = true;
5706 * New conflicting inodes may be found and added to the list while we
5707 * are logging a conflicting inode, so keep iterating while the list is
5710 while (!list_empty(&ctx->conflict_inodes)) {
5711 struct btrfs_ino_list *curr;
5712 struct inode *inode;
5716 curr = list_first_entry(&ctx->conflict_inodes,
5717 struct btrfs_ino_list, list);
5719 parent = curr->parent;
5720 list_del(&curr->list);
5723 inode = btrfs_iget(fs_info->sb, ino, root);
5725 * If the other inode that had a conflicting dir entry was
5726 * deleted in the current transaction, we need to log its parent
5727 * directory. See the comment at add_conflicting_inode().
5729 if (IS_ERR(inode)) {
5730 ret = PTR_ERR(inode);
5734 inode = btrfs_iget(fs_info->sb, parent, root);
5735 if (IS_ERR(inode)) {
5736 ret = PTR_ERR(inode);
5741 * Always log the directory, we cannot make this
5742 * conditional on need_log_inode() because the directory
5743 * might have been logged in LOG_INODE_EXISTS mode or
5744 * the dir index of the conflicting inode is not in a
5745 * dir index key range logged for the directory. So we
5746 * must make sure the deletion is recorded.
5748 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5749 LOG_INODE_ALL, ctx);
5750 btrfs_add_delayed_iput(BTRFS_I(inode));
5757 * Here we can use need_log_inode() because we only need to log
5758 * the inode in LOG_INODE_EXISTS mode and rename operations
5759 * update the log, so that the log ends up with the new name and
5760 * without the old name.
5762 * We did this check at add_conflicting_inode(), but here we do
5763 * it again because if some other task logged the inode after
5764 * that, we can avoid doing it again.
5766 if (!need_log_inode(trans, BTRFS_I(inode))) {
5767 btrfs_add_delayed_iput(BTRFS_I(inode));
5772 * We are safe logging the other inode without acquiring its
5773 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5774 * are safe against concurrent renames of the other inode as
5775 * well because during a rename we pin the log and update the
5776 * log with the new name before we unpin it.
5778 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5779 btrfs_add_delayed_iput(BTRFS_I(inode));
5784 ctx->logging_conflict_inodes = false;
5786 free_conflicting_inodes(ctx);
5791 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5792 struct btrfs_inode *inode,
5793 struct btrfs_key *min_key,
5794 const struct btrfs_key *max_key,
5795 struct btrfs_path *path,
5796 struct btrfs_path *dst_path,
5797 const u64 logged_isize,
5798 const int inode_only,
5799 struct btrfs_log_ctx *ctx,
5800 bool *need_log_inode_item)
5802 const u64 i_size = i_size_read(&inode->vfs_inode);
5803 struct btrfs_root *root = inode->root;
5804 int ins_start_slot = 0;
5809 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5817 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5818 if (min_key->objectid != max_key->objectid)
5820 if (min_key->type > max_key->type)
5823 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5824 *need_log_inode_item = false;
5825 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5826 min_key->offset >= i_size) {
5828 * Extents at and beyond eof are logged with
5829 * btrfs_log_prealloc_extents().
5830 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5831 * and no keys greater than that, so bail out.
5834 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5835 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5836 (inode->generation == trans->transid ||
5837 ctx->logging_conflict_inodes)) {
5839 u64 other_parent = 0;
5841 ret = btrfs_check_ref_name_override(path->nodes[0],
5842 path->slots[0], min_key, inode,
5843 &other_ino, &other_parent);
5846 } else if (ret > 0 &&
5847 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5852 ins_start_slot = path->slots[0];
5854 ret = copy_items(trans, inode, dst_path, path,
5855 ins_start_slot, ins_nr,
5856 inode_only, logged_isize);
5861 btrfs_release_path(path);
5862 ret = add_conflicting_inode(trans, root, path,
5869 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5870 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5873 ret = copy_items(trans, inode, dst_path, path,
5875 ins_nr, inode_only, logged_isize);
5882 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5885 } else if (!ins_nr) {
5886 ins_start_slot = path->slots[0];
5891 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5892 ins_nr, inode_only, logged_isize);
5896 ins_start_slot = path->slots[0];
5899 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5900 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5905 ret = copy_items(trans, inode, dst_path, path,
5906 ins_start_slot, ins_nr, inode_only,
5912 btrfs_release_path(path);
5914 if (min_key->offset < (u64)-1) {
5916 } else if (min_key->type < max_key->type) {
5918 min_key->offset = 0;
5924 * We may process many leaves full of items for our inode, so
5925 * avoid monopolizing a cpu for too long by rescheduling while
5926 * not holding locks on any tree.
5931 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5932 ins_nr, inode_only, logged_isize);
5937 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5939 * Release the path because otherwise we might attempt to double
5940 * lock the same leaf with btrfs_log_prealloc_extents() below.
5942 btrfs_release_path(path);
5943 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5949 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
5950 struct btrfs_root *log,
5951 struct btrfs_path *path,
5952 const struct btrfs_item_batch *batch,
5953 const struct btrfs_delayed_item *first_item)
5955 const struct btrfs_delayed_item *curr = first_item;
5958 ret = btrfs_insert_empty_items(trans, log, path, batch);
5962 for (int i = 0; i < batch->nr; i++) {
5965 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
5966 write_extent_buffer(path->nodes[0], &curr->data,
5967 (unsigned long)data_ptr, curr->data_len);
5968 curr = list_next_entry(curr, log_list);
5972 btrfs_release_path(path);
5977 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
5978 struct btrfs_inode *inode,
5979 struct btrfs_path *path,
5980 const struct list_head *delayed_ins_list,
5981 struct btrfs_log_ctx *ctx)
5983 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
5984 const int max_batch_size = 195;
5985 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
5986 const u64 ino = btrfs_ino(inode);
5987 struct btrfs_root *log = inode->root->log_root;
5988 struct btrfs_item_batch batch = {
5990 .total_data_size = 0,
5992 const struct btrfs_delayed_item *first = NULL;
5993 const struct btrfs_delayed_item *curr;
5995 struct btrfs_key *ins_keys;
5997 u64 curr_batch_size = 0;
6001 /* We are adding dir index items to the log tree. */
6002 lockdep_assert_held(&inode->log_mutex);
6005 * We collect delayed items before copying index keys from the subvolume
6006 * to the log tree. However just after we collected them, they may have
6007 * been flushed (all of them or just some of them), and therefore we
6008 * could have copied them from the subvolume tree to the log tree.
6009 * So find the first delayed item that was not yet logged (they are
6010 * sorted by index number).
6012 list_for_each_entry(curr, delayed_ins_list, log_list) {
6013 if (curr->index > inode->last_dir_index_offset) {
6019 /* Empty list or all delayed items were already logged. */
6023 ins_data = kmalloc(max_batch_size * sizeof(u32) +
6024 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
6027 ins_sizes = (u32 *)ins_data;
6028 batch.data_sizes = ins_sizes;
6029 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6030 batch.keys = ins_keys;
6033 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6034 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6036 if (curr_batch_size + curr_size > leaf_data_size ||
6037 batch.nr == max_batch_size) {
6038 ret = insert_delayed_items_batch(trans, log, path,
6044 batch.total_data_size = 0;
6045 curr_batch_size = 0;
6049 ins_sizes[batch_idx] = curr->data_len;
6050 ins_keys[batch_idx].objectid = ino;
6051 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6052 ins_keys[batch_idx].offset = curr->index;
6053 curr_batch_size += curr_size;
6054 batch.total_data_size += curr->data_len;
6057 curr = list_next_entry(curr, log_list);
6060 ASSERT(batch.nr >= 1);
6061 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6063 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6065 inode->last_dir_index_offset = curr->index;
6072 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6073 struct btrfs_inode *inode,
6074 struct btrfs_path *path,
6075 const struct list_head *delayed_del_list,
6076 struct btrfs_log_ctx *ctx)
6078 const u64 ino = btrfs_ino(inode);
6079 const struct btrfs_delayed_item *curr;
6081 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6084 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6085 u64 first_dir_index = curr->index;
6087 const struct btrfs_delayed_item *next;
6091 * Find a range of consecutive dir index items to delete. Like
6092 * this we log a single dir range item spanning several contiguous
6093 * dir items instead of logging one range item per dir index item.
6095 next = list_next_entry(curr, log_list);
6096 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6097 if (next->index != curr->index + 1)
6100 next = list_next_entry(next, log_list);
6103 last_dir_index = curr->index;
6104 ASSERT(last_dir_index >= first_dir_index);
6106 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6107 ino, first_dir_index, last_dir_index);
6110 curr = list_next_entry(curr, log_list);
6116 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6117 struct btrfs_inode *inode,
6118 struct btrfs_path *path,
6119 struct btrfs_log_ctx *ctx,
6120 const struct list_head *delayed_del_list,
6121 const struct btrfs_delayed_item *first,
6122 const struct btrfs_delayed_item **last_ret)
6124 const struct btrfs_delayed_item *next;
6125 struct extent_buffer *leaf = path->nodes[0];
6126 const int last_slot = btrfs_header_nritems(leaf) - 1;
6127 int slot = path->slots[0] + 1;
6128 const u64 ino = btrfs_ino(inode);
6130 next = list_next_entry(first, log_list);
6132 while (slot < last_slot &&
6133 !list_entry_is_head(next, delayed_del_list, log_list)) {
6134 struct btrfs_key key;
6136 btrfs_item_key_to_cpu(leaf, &key, slot);
6137 if (key.objectid != ino ||
6138 key.type != BTRFS_DIR_INDEX_KEY ||
6139 key.offset != next->index)
6144 next = list_next_entry(next, log_list);
6147 return btrfs_del_items(trans, inode->root->log_root, path,
6148 path->slots[0], slot - path->slots[0]);
6151 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6152 struct btrfs_inode *inode,
6153 struct btrfs_path *path,
6154 const struct list_head *delayed_del_list,
6155 struct btrfs_log_ctx *ctx)
6157 struct btrfs_root *log = inode->root->log_root;
6158 const struct btrfs_delayed_item *curr;
6159 u64 last_range_start = 0;
6160 u64 last_range_end = 0;
6161 struct btrfs_key key;
6163 key.objectid = btrfs_ino(inode);
6164 key.type = BTRFS_DIR_INDEX_KEY;
6165 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6168 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6169 const struct btrfs_delayed_item *last = curr;
6170 u64 first_dir_index = curr->index;
6172 bool deleted_items = false;
6175 key.offset = curr->index;
6176 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6179 } else if (ret == 0) {
6180 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6181 delayed_del_list, curr,
6185 deleted_items = true;
6188 btrfs_release_path(path);
6191 * If we deleted items from the leaf, it means we have a range
6192 * item logging their range, so no need to add one or update an
6193 * existing one. Otherwise we have to log a dir range item.
6198 last_dir_index = last->index;
6199 ASSERT(last_dir_index >= first_dir_index);
6201 * If this range starts right after where the previous one ends,
6202 * then we want to reuse the previous range item and change its
6203 * end offset to the end of this range. This is just to minimize
6204 * leaf space usage, by avoiding adding a new range item.
6206 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6207 first_dir_index = last_range_start;
6209 ret = insert_dir_log_key(trans, log, path, key.objectid,
6210 first_dir_index, last_dir_index);
6214 last_range_start = first_dir_index;
6215 last_range_end = last_dir_index;
6217 curr = list_next_entry(last, log_list);
6223 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6224 struct btrfs_inode *inode,
6225 struct btrfs_path *path,
6226 const struct list_head *delayed_del_list,
6227 struct btrfs_log_ctx *ctx)
6230 * We are deleting dir index items from the log tree or adding range
6233 lockdep_assert_held(&inode->log_mutex);
6235 if (list_empty(delayed_del_list))
6238 if (ctx->logged_before)
6239 return log_delayed_deletions_incremental(trans, inode, path,
6240 delayed_del_list, ctx);
6242 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6247 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6248 * items instead of the subvolume tree.
6250 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6251 struct btrfs_inode *inode,
6252 const struct list_head *delayed_ins_list,
6253 struct btrfs_log_ctx *ctx)
6255 const bool orig_log_new_dentries = ctx->log_new_dentries;
6256 struct btrfs_fs_info *fs_info = trans->fs_info;
6257 struct btrfs_delayed_item *item;
6261 * No need for the log mutex, plus to avoid potential deadlocks or
6262 * lockdep annotations due to nesting of delayed inode mutexes and log
6265 lockdep_assert_not_held(&inode->log_mutex);
6267 ASSERT(!ctx->logging_new_delayed_dentries);
6268 ctx->logging_new_delayed_dentries = true;
6270 list_for_each_entry(item, delayed_ins_list, log_list) {
6271 struct btrfs_dir_item *dir_item;
6272 struct inode *di_inode;
6273 struct btrfs_key key;
6274 int log_mode = LOG_INODE_EXISTS;
6276 dir_item = (struct btrfs_dir_item *)item->data;
6277 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6279 if (key.type == BTRFS_ROOT_ITEM_KEY)
6282 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6283 if (IS_ERR(di_inode)) {
6284 ret = PTR_ERR(di_inode);
6288 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6289 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6293 if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR)
6294 log_mode = LOG_INODE_ALL;
6296 ctx->log_new_dentries = false;
6297 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6299 if (!ret && ctx->log_new_dentries)
6300 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6302 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6308 ctx->log_new_dentries = orig_log_new_dentries;
6309 ctx->logging_new_delayed_dentries = false;
6314 /* log a single inode in the tree log.
6315 * At least one parent directory for this inode must exist in the tree
6316 * or be logged already.
6318 * Any items from this inode changed by the current transaction are copied
6319 * to the log tree. An extra reference is taken on any extents in this
6320 * file, allowing us to avoid a whole pile of corner cases around logging
6321 * blocks that have been removed from the tree.
6323 * See LOG_INODE_ALL and related defines for a description of what inode_only
6326 * This handles both files and directories.
6328 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6329 struct btrfs_inode *inode,
6331 struct btrfs_log_ctx *ctx)
6333 struct btrfs_path *path;
6334 struct btrfs_path *dst_path;
6335 struct btrfs_key min_key;
6336 struct btrfs_key max_key;
6337 struct btrfs_root *log = inode->root->log_root;
6339 bool fast_search = false;
6340 u64 ino = btrfs_ino(inode);
6341 struct extent_map_tree *em_tree = &inode->extent_tree;
6342 u64 logged_isize = 0;
6343 bool need_log_inode_item = true;
6344 bool xattrs_logged = false;
6345 bool inode_item_dropped = true;
6346 bool full_dir_logging = false;
6347 LIST_HEAD(delayed_ins_list);
6348 LIST_HEAD(delayed_del_list);
6350 path = btrfs_alloc_path();
6353 dst_path = btrfs_alloc_path();
6355 btrfs_free_path(path);
6359 min_key.objectid = ino;
6360 min_key.type = BTRFS_INODE_ITEM_KEY;
6363 max_key.objectid = ino;
6366 /* today the code can only do partial logging of directories */
6367 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6368 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6369 &inode->runtime_flags) &&
6370 inode_only >= LOG_INODE_EXISTS))
6371 max_key.type = BTRFS_XATTR_ITEM_KEY;
6373 max_key.type = (u8)-1;
6374 max_key.offset = (u64)-1;
6376 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6377 full_dir_logging = true;
6380 * If we are logging a directory while we are logging dentries of the
6381 * delayed items of some other inode, then we need to flush the delayed
6382 * items of this directory and not log the delayed items directly. This
6383 * is to prevent more than one level of recursion into btrfs_log_inode()
6384 * by having something like this:
6386 * $ mkdir -p a/b/c/d/e/f/g/h/...
6387 * $ xfs_io -c "fsync" a
6389 * Where all directories in the path did not exist before and are
6390 * created in the current transaction.
6391 * So in such a case we directly log the delayed items of the main
6392 * directory ("a") without flushing them first, while for each of its
6393 * subdirectories we flush their delayed items before logging them.
6394 * This prevents a potential unbounded recursion like this:
6397 * log_new_delayed_dentries()
6399 * log_new_delayed_dentries()
6401 * log_new_delayed_dentries()
6404 * We have thresholds for the maximum number of delayed items to have in
6405 * memory, and once they are hit, the items are flushed asynchronously.
6406 * However the limit is quite high, so lets prevent deep levels of
6407 * recursion to happen by limiting the maximum depth to be 1.
6409 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6410 ret = btrfs_commit_inode_delayed_items(trans, inode);
6415 mutex_lock(&inode->log_mutex);
6418 * For symlinks, we must always log their content, which is stored in an
6419 * inline extent, otherwise we could end up with an empty symlink after
6420 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6421 * one attempts to create an empty symlink).
6422 * We don't need to worry about flushing delalloc, because when we create
6423 * the inline extent when the symlink is created (we never have delalloc
6426 if (S_ISLNK(inode->vfs_inode.i_mode))
6427 inode_only = LOG_INODE_ALL;
6430 * Before logging the inode item, cache the value returned by
6431 * inode_logged(), because after that we have the need to figure out if
6432 * the inode was previously logged in this transaction.
6434 ret = inode_logged(trans, inode, path);
6437 ctx->logged_before = (ret == 1);
6441 * This is for cases where logging a directory could result in losing a
6442 * a file after replaying the log. For example, if we move a file from a
6443 * directory A to a directory B, then fsync directory A, we have no way
6444 * to known the file was moved from A to B, so logging just A would
6445 * result in losing the file after a log replay.
6447 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6448 ret = BTRFS_LOG_FORCE_COMMIT;
6453 * a brute force approach to making sure we get the most uptodate
6454 * copies of everything.
6456 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6457 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6458 if (ctx->logged_before)
6459 ret = drop_inode_items(trans, log, path, inode,
6460 BTRFS_XATTR_ITEM_KEY);
6462 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6464 * Make sure the new inode item we write to the log has
6465 * the same isize as the current one (if it exists).
6466 * This is necessary to prevent data loss after log
6467 * replay, and also to prevent doing a wrong expanding
6468 * truncate - for e.g. create file, write 4K into offset
6469 * 0, fsync, write 4K into offset 4096, add hard link,
6470 * fsync some other file (to sync log), power fail - if
6471 * we use the inode's current i_size, after log replay
6472 * we get a 8Kb file, with the last 4Kb extent as a hole
6473 * (zeroes), as if an expanding truncate happened,
6474 * instead of getting a file of 4Kb only.
6476 ret = logged_inode_size(log, inode, path, &logged_isize);
6480 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6481 &inode->runtime_flags)) {
6482 if (inode_only == LOG_INODE_EXISTS) {
6483 max_key.type = BTRFS_XATTR_ITEM_KEY;
6484 if (ctx->logged_before)
6485 ret = drop_inode_items(trans, log, path,
6486 inode, max_key.type);
6488 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6489 &inode->runtime_flags);
6490 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6491 &inode->runtime_flags);
6492 if (ctx->logged_before)
6493 ret = truncate_inode_items(trans, log,
6496 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6497 &inode->runtime_flags) ||
6498 inode_only == LOG_INODE_EXISTS) {
6499 if (inode_only == LOG_INODE_ALL)
6501 max_key.type = BTRFS_XATTR_ITEM_KEY;
6502 if (ctx->logged_before)
6503 ret = drop_inode_items(trans, log, path, inode,
6506 if (inode_only == LOG_INODE_ALL)
6508 inode_item_dropped = false;
6517 * If we are logging a directory in full mode, collect the delayed items
6518 * before iterating the subvolume tree, so that we don't miss any new
6519 * dir index items in case they get flushed while or right after we are
6520 * iterating the subvolume tree.
6522 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6523 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6526 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6527 path, dst_path, logged_isize,
6529 &need_log_inode_item);
6533 btrfs_release_path(path);
6534 btrfs_release_path(dst_path);
6535 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6538 xattrs_logged = true;
6539 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6540 btrfs_release_path(path);
6541 btrfs_release_path(dst_path);
6542 ret = btrfs_log_holes(trans, inode, path);
6547 btrfs_release_path(path);
6548 btrfs_release_path(dst_path);
6549 if (need_log_inode_item) {
6550 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6554 * If we are doing a fast fsync and the inode was logged before
6555 * in this transaction, we don't need to log the xattrs because
6556 * they were logged before. If xattrs were added, changed or
6557 * deleted since the last time we logged the inode, then we have
6558 * already logged them because the inode had the runtime flag
6559 * BTRFS_INODE_COPY_EVERYTHING set.
6561 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6562 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6565 btrfs_release_path(path);
6569 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6572 } else if (inode_only == LOG_INODE_ALL) {
6573 struct extent_map *em, *n;
6575 write_lock(&em_tree->lock);
6576 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6577 list_del_init(&em->list);
6578 write_unlock(&em_tree->lock);
6581 if (full_dir_logging) {
6582 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6585 ret = log_delayed_insertion_items(trans, inode, path,
6586 &delayed_ins_list, ctx);
6589 ret = log_delayed_deletion_items(trans, inode, path,
6590 &delayed_del_list, ctx);
6595 spin_lock(&inode->lock);
6596 inode->logged_trans = trans->transid;
6598 * Don't update last_log_commit if we logged that an inode exists.
6599 * We do this for three reasons:
6601 * 1) We might have had buffered writes to this inode that were
6602 * flushed and had their ordered extents completed in this
6603 * transaction, but we did not previously log the inode with
6604 * LOG_INODE_ALL. Later the inode was evicted and after that
6605 * it was loaded again and this LOG_INODE_EXISTS log operation
6606 * happened. We must make sure that if an explicit fsync against
6607 * the inode is performed later, it logs the new extents, an
6608 * updated inode item, etc, and syncs the log. The same logic
6609 * applies to direct IO writes instead of buffered writes.
6611 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6612 * is logged with an i_size of 0 or whatever value was logged
6613 * before. If later the i_size of the inode is increased by a
6614 * truncate operation, the log is synced through an fsync of
6615 * some other inode and then finally an explicit fsync against
6616 * this inode is made, we must make sure this fsync logs the
6617 * inode with the new i_size, the hole between old i_size and
6618 * the new i_size, and syncs the log.
6620 * 3) If we are logging that an ancestor inode exists as part of
6621 * logging a new name from a link or rename operation, don't update
6622 * its last_log_commit - otherwise if an explicit fsync is made
6623 * against an ancestor, the fsync considers the inode in the log
6624 * and doesn't sync the log, resulting in the ancestor missing after
6625 * a power failure unless the log was synced as part of an fsync
6626 * against any other unrelated inode.
6628 if (inode_only != LOG_INODE_EXISTS)
6629 inode->last_log_commit = inode->last_sub_trans;
6630 spin_unlock(&inode->lock);
6633 * Reset the last_reflink_trans so that the next fsync does not need to
6634 * go through the slower path when logging extents and their checksums.
6636 if (inode_only == LOG_INODE_ALL)
6637 inode->last_reflink_trans = 0;
6640 mutex_unlock(&inode->log_mutex);
6642 btrfs_free_path(path);
6643 btrfs_free_path(dst_path);
6646 free_conflicting_inodes(ctx);
6648 ret = log_conflicting_inodes(trans, inode->root, ctx);
6650 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6652 ret = log_new_delayed_dentries(trans, inode,
6653 &delayed_ins_list, ctx);
6655 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6662 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6663 struct btrfs_inode *inode,
6664 struct btrfs_log_ctx *ctx)
6666 struct btrfs_fs_info *fs_info = trans->fs_info;
6668 struct btrfs_path *path;
6669 struct btrfs_key key;
6670 struct btrfs_root *root = inode->root;
6671 const u64 ino = btrfs_ino(inode);
6673 path = btrfs_alloc_path();
6676 path->skip_locking = 1;
6677 path->search_commit_root = 1;
6680 key.type = BTRFS_INODE_REF_KEY;
6682 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6687 struct extent_buffer *leaf = path->nodes[0];
6688 int slot = path->slots[0];
6693 if (slot >= btrfs_header_nritems(leaf)) {
6694 ret = btrfs_next_leaf(root, path);
6702 btrfs_item_key_to_cpu(leaf, &key, slot);
6703 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6704 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6707 item_size = btrfs_item_size(leaf, slot);
6708 ptr = btrfs_item_ptr_offset(leaf, slot);
6709 while (cur_offset < item_size) {
6710 struct btrfs_key inode_key;
6711 struct inode *dir_inode;
6713 inode_key.type = BTRFS_INODE_ITEM_KEY;
6714 inode_key.offset = 0;
6716 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6717 struct btrfs_inode_extref *extref;
6719 extref = (struct btrfs_inode_extref *)
6721 inode_key.objectid = btrfs_inode_extref_parent(
6723 cur_offset += sizeof(*extref);
6724 cur_offset += btrfs_inode_extref_name_len(leaf,
6727 inode_key.objectid = key.offset;
6728 cur_offset = item_size;
6731 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6734 * If the parent inode was deleted, return an error to
6735 * fallback to a transaction commit. This is to prevent
6736 * getting an inode that was moved from one parent A to
6737 * a parent B, got its former parent A deleted and then
6738 * it got fsync'ed, from existing at both parents after
6739 * a log replay (and the old parent still existing).
6746 * mv /mnt/B/bar /mnt/A/bar
6747 * mv -T /mnt/A /mnt/B
6751 * If we ignore the old parent B which got deleted,
6752 * after a log replay we would have file bar linked
6753 * at both parents and the old parent B would still
6756 if (IS_ERR(dir_inode)) {
6757 ret = PTR_ERR(dir_inode);
6761 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6762 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6766 ctx->log_new_dentries = false;
6767 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6768 LOG_INODE_ALL, ctx);
6769 if (!ret && ctx->log_new_dentries)
6770 ret = log_new_dir_dentries(trans,
6771 BTRFS_I(dir_inode), ctx);
6772 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6780 btrfs_free_path(path);
6784 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6785 struct btrfs_root *root,
6786 struct btrfs_path *path,
6787 struct btrfs_log_ctx *ctx)
6789 struct btrfs_key found_key;
6791 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6794 struct btrfs_fs_info *fs_info = root->fs_info;
6795 struct extent_buffer *leaf;
6797 struct btrfs_key search_key;
6798 struct inode *inode;
6802 btrfs_release_path(path);
6804 ino = found_key.offset;
6806 search_key.objectid = found_key.offset;
6807 search_key.type = BTRFS_INODE_ITEM_KEY;
6808 search_key.offset = 0;
6809 inode = btrfs_iget(fs_info->sb, ino, root);
6811 return PTR_ERR(inode);
6813 if (BTRFS_I(inode)->generation >= trans->transid &&
6814 need_log_inode(trans, BTRFS_I(inode)))
6815 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6816 LOG_INODE_EXISTS, ctx);
6817 btrfs_add_delayed_iput(BTRFS_I(inode));
6821 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6824 search_key.type = BTRFS_INODE_REF_KEY;
6825 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6829 leaf = path->nodes[0];
6830 slot = path->slots[0];
6831 if (slot >= btrfs_header_nritems(leaf)) {
6832 ret = btrfs_next_leaf(root, path);
6837 leaf = path->nodes[0];
6838 slot = path->slots[0];
6841 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6842 if (found_key.objectid != search_key.objectid ||
6843 found_key.type != BTRFS_INODE_REF_KEY)
6849 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6850 struct btrfs_inode *inode,
6851 struct dentry *parent,
6852 struct btrfs_log_ctx *ctx)
6854 struct btrfs_root *root = inode->root;
6855 struct dentry *old_parent = NULL;
6856 struct super_block *sb = inode->vfs_inode.i_sb;
6860 if (!parent || d_really_is_negative(parent) ||
6864 inode = BTRFS_I(d_inode(parent));
6865 if (root != inode->root)
6868 if (inode->generation >= trans->transid &&
6869 need_log_inode(trans, inode)) {
6870 ret = btrfs_log_inode(trans, inode,
6871 LOG_INODE_EXISTS, ctx);
6875 if (IS_ROOT(parent))
6878 parent = dget_parent(parent);
6880 old_parent = parent;
6887 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6888 struct btrfs_inode *inode,
6889 struct dentry *parent,
6890 struct btrfs_log_ctx *ctx)
6892 struct btrfs_root *root = inode->root;
6893 const u64 ino = btrfs_ino(inode);
6894 struct btrfs_path *path;
6895 struct btrfs_key search_key;
6899 * For a single hard link case, go through a fast path that does not
6900 * need to iterate the fs/subvolume tree.
6902 if (inode->vfs_inode.i_nlink < 2)
6903 return log_new_ancestors_fast(trans, inode, parent, ctx);
6905 path = btrfs_alloc_path();
6909 search_key.objectid = ino;
6910 search_key.type = BTRFS_INODE_REF_KEY;
6911 search_key.offset = 0;
6913 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6920 struct extent_buffer *leaf = path->nodes[0];
6921 int slot = path->slots[0];
6922 struct btrfs_key found_key;
6924 if (slot >= btrfs_header_nritems(leaf)) {
6925 ret = btrfs_next_leaf(root, path);
6933 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6934 if (found_key.objectid != ino ||
6935 found_key.type > BTRFS_INODE_EXTREF_KEY)
6939 * Don't deal with extended references because they are rare
6940 * cases and too complex to deal with (we would need to keep
6941 * track of which subitem we are processing for each item in
6942 * this loop, etc). So just return some error to fallback to
6943 * a transaction commit.
6945 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6951 * Logging ancestors needs to do more searches on the fs/subvol
6952 * tree, so it releases the path as needed to avoid deadlocks.
6953 * Keep track of the last inode ref key and resume from that key
6954 * after logging all new ancestors for the current hard link.
6956 memcpy(&search_key, &found_key, sizeof(search_key));
6958 ret = log_new_ancestors(trans, root, path, ctx);
6961 btrfs_release_path(path);
6966 btrfs_free_path(path);
6971 * helper function around btrfs_log_inode to make sure newly created
6972 * parent directories also end up in the log. A minimal inode and backref
6973 * only logging is done of any parent directories that are older than
6974 * the last committed transaction
6976 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6977 struct btrfs_inode *inode,
6978 struct dentry *parent,
6980 struct btrfs_log_ctx *ctx)
6982 struct btrfs_root *root = inode->root;
6983 struct btrfs_fs_info *fs_info = root->fs_info;
6985 bool log_dentries = false;
6987 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6988 ret = BTRFS_LOG_FORCE_COMMIT;
6992 if (btrfs_root_refs(&root->root_item) == 0) {
6993 ret = BTRFS_LOG_FORCE_COMMIT;
6998 * Skip already logged inodes or inodes corresponding to tmpfiles
6999 * (since logging them is pointless, a link count of 0 means they
7000 * will never be accessible).
7002 if ((btrfs_inode_in_log(inode, trans->transid) &&
7003 list_empty(&ctx->ordered_extents)) ||
7004 inode->vfs_inode.i_nlink == 0) {
7005 ret = BTRFS_NO_LOG_SYNC;
7009 ret = start_log_trans(trans, root, ctx);
7013 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
7018 * for regular files, if its inode is already on disk, we don't
7019 * have to worry about the parents at all. This is because
7020 * we can use the last_unlink_trans field to record renames
7021 * and other fun in this file.
7023 if (S_ISREG(inode->vfs_inode.i_mode) &&
7024 inode->generation < trans->transid &&
7025 inode->last_unlink_trans < trans->transid) {
7030 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7031 log_dentries = true;
7034 * On unlink we must make sure all our current and old parent directory
7035 * inodes are fully logged. This is to prevent leaving dangling
7036 * directory index entries in directories that were our parents but are
7037 * not anymore. Not doing this results in old parent directory being
7038 * impossible to delete after log replay (rmdir will always fail with
7039 * error -ENOTEMPTY).
7045 * ln testdir/foo testdir/bar
7047 * unlink testdir/bar
7048 * xfs_io -c fsync testdir/foo
7050 * mount fs, triggers log replay
7052 * If we don't log the parent directory (testdir), after log replay the
7053 * directory still has an entry pointing to the file inode using the bar
7054 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7055 * the file inode has a link count of 1.
7061 * ln foo testdir/foo2
7062 * ln foo testdir/foo3
7064 * unlink testdir/foo3
7065 * xfs_io -c fsync foo
7067 * mount fs, triggers log replay
7069 * Similar as the first example, after log replay the parent directory
7070 * testdir still has an entry pointing to the inode file with name foo3
7071 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7072 * and has a link count of 2.
7074 if (inode->last_unlink_trans >= trans->transid) {
7075 ret = btrfs_log_all_parents(trans, inode, ctx);
7080 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7085 ret = log_new_dir_dentries(trans, inode, ctx);
7090 btrfs_set_log_full_commit(trans);
7091 ret = BTRFS_LOG_FORCE_COMMIT;
7095 btrfs_remove_log_ctx(root, ctx);
7096 btrfs_end_log_trans(root);
7102 * it is not safe to log dentry if the chunk root has added new
7103 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7104 * If this returns 1, you must commit the transaction to safely get your
7107 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7108 struct dentry *dentry,
7109 struct btrfs_log_ctx *ctx)
7111 struct dentry *parent = dget_parent(dentry);
7114 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7115 LOG_INODE_ALL, ctx);
7122 * should be called during mount to recover any replay any log trees
7125 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7128 struct btrfs_path *path;
7129 struct btrfs_trans_handle *trans;
7130 struct btrfs_key key;
7131 struct btrfs_key found_key;
7132 struct btrfs_root *log;
7133 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7134 struct walk_control wc = {
7135 .process_func = process_one_buffer,
7136 .stage = LOG_WALK_PIN_ONLY,
7139 path = btrfs_alloc_path();
7143 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7145 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7146 if (IS_ERR(trans)) {
7147 ret = PTR_ERR(trans);
7154 ret = walk_log_tree(trans, log_root_tree, &wc);
7156 btrfs_abort_transaction(trans, ret);
7161 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7162 key.offset = (u64)-1;
7163 key.type = BTRFS_ROOT_ITEM_KEY;
7166 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7169 btrfs_abort_transaction(trans, ret);
7173 if (path->slots[0] == 0)
7177 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7179 btrfs_release_path(path);
7180 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7183 log = btrfs_read_tree_root(log_root_tree, &found_key);
7186 btrfs_abort_transaction(trans, ret);
7190 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7192 if (IS_ERR(wc.replay_dest)) {
7193 ret = PTR_ERR(wc.replay_dest);
7196 * We didn't find the subvol, likely because it was
7197 * deleted. This is ok, simply skip this log and go to
7200 * We need to exclude the root because we can't have
7201 * other log replays overwriting this log as we'll read
7202 * it back in a few more times. This will keep our
7203 * block from being modified, and we'll just bail for
7204 * each subsequent pass.
7207 ret = btrfs_pin_extent_for_log_replay(trans,
7210 btrfs_put_root(log);
7214 btrfs_abort_transaction(trans, ret);
7218 wc.replay_dest->log_root = log;
7219 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7221 /* The loop needs to continue due to the root refs */
7222 btrfs_abort_transaction(trans, ret);
7224 ret = walk_log_tree(trans, log, &wc);
7226 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7227 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7230 btrfs_abort_transaction(trans, ret);
7233 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7234 struct btrfs_root *root = wc.replay_dest;
7236 btrfs_release_path(path);
7239 * We have just replayed everything, and the highest
7240 * objectid of fs roots probably has changed in case
7241 * some inode_item's got replayed.
7243 * root->objectid_mutex is not acquired as log replay
7244 * could only happen during mount.
7246 ret = btrfs_init_root_free_objectid(root);
7248 btrfs_abort_transaction(trans, ret);
7251 wc.replay_dest->log_root = NULL;
7252 btrfs_put_root(wc.replay_dest);
7253 btrfs_put_root(log);
7258 if (found_key.offset == 0)
7260 key.offset = found_key.offset - 1;
7262 btrfs_release_path(path);
7264 /* step one is to pin it all, step two is to replay just inodes */
7267 wc.process_func = replay_one_buffer;
7268 wc.stage = LOG_WALK_REPLAY_INODES;
7271 /* step three is to replay everything */
7272 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7277 btrfs_free_path(path);
7279 /* step 4: commit the transaction, which also unpins the blocks */
7280 ret = btrfs_commit_transaction(trans);
7284 log_root_tree->log_root = NULL;
7285 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7286 btrfs_put_root(log_root_tree);
7291 btrfs_end_transaction(wc.trans);
7292 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7293 btrfs_free_path(path);
7298 * there are some corner cases where we want to force a full
7299 * commit instead of allowing a directory to be logged.
7301 * They revolve around files there were unlinked from the directory, and
7302 * this function updates the parent directory so that a full commit is
7303 * properly done if it is fsync'd later after the unlinks are done.
7305 * Must be called before the unlink operations (updates to the subvolume tree,
7306 * inodes, etc) are done.
7308 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7309 struct btrfs_inode *dir, struct btrfs_inode *inode,
7313 * when we're logging a file, if it hasn't been renamed
7314 * or unlinked, and its inode is fully committed on disk,
7315 * we don't have to worry about walking up the directory chain
7316 * to log its parents.
7318 * So, we use the last_unlink_trans field to put this transid
7319 * into the file. When the file is logged we check it and
7320 * don't log the parents if the file is fully on disk.
7322 mutex_lock(&inode->log_mutex);
7323 inode->last_unlink_trans = trans->transid;
7324 mutex_unlock(&inode->log_mutex);
7330 * If this directory was already logged, any new names will be logged
7331 * with btrfs_log_new_name() and old names will be deleted from the log
7332 * tree with btrfs_del_dir_entries_in_log() or with
7333 * btrfs_del_inode_ref_in_log().
7335 if (inode_logged(trans, dir, NULL) == 1)
7339 * If the inode we're about to unlink was logged before, the log will be
7340 * properly updated with the new name with btrfs_log_new_name() and the
7341 * old name removed with btrfs_del_dir_entries_in_log() or with
7342 * btrfs_del_inode_ref_in_log().
7344 if (inode_logged(trans, inode, NULL) == 1)
7348 * when renaming files across directories, if the directory
7349 * there we're unlinking from gets fsync'd later on, there's
7350 * no way to find the destination directory later and fsync it
7351 * properly. So, we have to be conservative and force commits
7352 * so the new name gets discovered.
7354 mutex_lock(&dir->log_mutex);
7355 dir->last_unlink_trans = trans->transid;
7356 mutex_unlock(&dir->log_mutex);
7360 * Make sure that if someone attempts to fsync the parent directory of a deleted
7361 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7362 * that after replaying the log tree of the parent directory's root we will not
7363 * see the snapshot anymore and at log replay time we will not see any log tree
7364 * corresponding to the deleted snapshot's root, which could lead to replaying
7365 * it after replaying the log tree of the parent directory (which would replay
7366 * the snapshot delete operation).
7368 * Must be called before the actual snapshot destroy operation (updates to the
7369 * parent root and tree of tree roots trees, etc) are done.
7371 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7372 struct btrfs_inode *dir)
7374 mutex_lock(&dir->log_mutex);
7375 dir->last_unlink_trans = trans->transid;
7376 mutex_unlock(&dir->log_mutex);
7380 * Update the log after adding a new name for an inode.
7382 * @trans: Transaction handle.
7383 * @old_dentry: The dentry associated with the old name and the old
7385 * @old_dir: The inode of the previous parent directory for the case
7386 * of a rename. For a link operation, it must be NULL.
7387 * @old_dir_index: The index number associated with the old name, meaningful
7388 * only for rename operations (when @old_dir is not NULL).
7389 * Ignored for link operations.
7390 * @parent: The dentry associated with the directory under which the
7391 * new name is located.
7393 * Call this after adding a new name for an inode, as a result of a link or
7394 * rename operation, and it will properly update the log to reflect the new name.
7396 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7397 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7398 u64 old_dir_index, struct dentry *parent)
7400 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7401 struct btrfs_root *root = inode->root;
7402 struct btrfs_log_ctx ctx;
7403 bool log_pinned = false;
7407 * this will force the logging code to walk the dentry chain
7410 if (!S_ISDIR(inode->vfs_inode.i_mode))
7411 inode->last_unlink_trans = trans->transid;
7414 * if this inode hasn't been logged and directory we're renaming it
7415 * from hasn't been logged, we don't need to log it
7417 ret = inode_logged(trans, inode, NULL);
7420 } else if (ret == 0) {
7424 * If the inode was not logged and we are doing a rename (old_dir is not
7425 * NULL), check if old_dir was logged - if it was not we can return and
7428 ret = inode_logged(trans, old_dir, NULL);
7437 * If we are doing a rename (old_dir is not NULL) from a directory that
7438 * was previously logged, make sure that on log replay we get the old
7439 * dir entry deleted. This is needed because we will also log the new
7440 * name of the renamed inode, so we need to make sure that after log
7441 * replay we don't end up with both the new and old dir entries existing.
7443 if (old_dir && old_dir->logged_trans == trans->transid) {
7444 struct btrfs_root *log = old_dir->root->log_root;
7445 struct btrfs_path *path;
7446 struct fscrypt_name fname;
7448 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7450 ret = fscrypt_setup_filename(&old_dir->vfs_inode,
7451 &old_dentry->d_name, 0, &fname);
7455 * We have two inodes to update in the log, the old directory and
7456 * the inode that got renamed, so we must pin the log to prevent
7457 * anyone from syncing the log until we have updated both inodes
7460 ret = join_running_log_trans(root);
7462 * At least one of the inodes was logged before, so this should
7463 * not fail, but if it does, it's not serious, just bail out and
7464 * mark the log for a full commit.
7466 if (WARN_ON_ONCE(ret < 0)) {
7467 fscrypt_free_filename(&fname);
7473 path = btrfs_alloc_path();
7476 fscrypt_free_filename(&fname);
7481 * Other concurrent task might be logging the old directory,
7482 * as it can be triggered when logging other inode that had or
7483 * still has a dentry in the old directory. We lock the old
7484 * directory's log_mutex to ensure the deletion of the old
7485 * name is persisted, because during directory logging we
7486 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7487 * the old name's dir index item is in the delayed items, so
7488 * it could be missed by an in progress directory logging.
7490 mutex_lock(&old_dir->log_mutex);
7491 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7492 &fname.disk_name, old_dir_index);
7495 * The dentry does not exist in the log, so record its
7498 btrfs_release_path(path);
7499 ret = insert_dir_log_key(trans, log, path,
7501 old_dir_index, old_dir_index);
7503 mutex_unlock(&old_dir->log_mutex);
7505 btrfs_free_path(path);
7506 fscrypt_free_filename(&fname);
7511 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7512 ctx.logging_new_name = true;
7514 * We don't care about the return value. If we fail to log the new name
7515 * then we know the next attempt to sync the log will fallback to a full
7516 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7517 * we don't need to worry about getting a log committed that has an
7518 * inconsistent state after a rename operation.
7520 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7521 ASSERT(list_empty(&ctx.conflict_inodes));
7524 * If an error happened mark the log for a full commit because it's not
7525 * consistent and up to date or we couldn't find out if one of the
7526 * inodes was logged before in this transaction. Do it before unpinning
7527 * the log, to avoid any races with someone else trying to commit it.
7530 btrfs_set_log_full_commit(trans);
7532 btrfs_end_log_trans(root);