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);
282 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
284 filemap_fdatawait_range(buf->pages[0]->mapping,
285 buf->start, buf->start + buf->len - 1);
289 * the walk control struct is used to pass state down the chain when
290 * processing the log tree. The stage field tells us which part
291 * of the log tree processing we are currently doing. The others
292 * are state fields used for that specific part
294 struct walk_control {
295 /* should we free the extent on disk when done? This is used
296 * at transaction commit time while freeing a log tree
300 /* pin only walk, we record which extents on disk belong to the
305 /* what stage of the replay code we're currently in */
309 * Ignore any items from the inode currently being processed. Needs
310 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
311 * the LOG_WALK_REPLAY_INODES stage.
313 bool ignore_cur_inode;
315 /* the root we are currently replaying */
316 struct btrfs_root *replay_dest;
318 /* the trans handle for the current replay */
319 struct btrfs_trans_handle *trans;
321 /* the function that gets used to process blocks we find in the
322 * tree. Note the extent_buffer might not be up to date when it is
323 * passed in, and it must be checked or read if you need the data
326 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
327 struct walk_control *wc, u64 gen, int level);
331 * process_func used to pin down extents, write them or wait on them
333 static int process_one_buffer(struct btrfs_root *log,
334 struct extent_buffer *eb,
335 struct walk_control *wc, u64 gen, int level)
337 struct btrfs_fs_info *fs_info = log->fs_info;
341 * If this fs is mixed then we need to be able to process the leaves to
342 * pin down any logged extents, so we have to read the block.
344 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
345 struct btrfs_tree_parent_check check = {
350 ret = btrfs_read_extent_buffer(eb, &check);
356 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
361 if (btrfs_buffer_uptodate(eb, gen, 0) &&
362 btrfs_header_level(eb) == 0)
363 ret = btrfs_exclude_logged_extents(eb);
369 * Item overwrite used by replay and tree logging. eb, slot and key all refer
370 * to the src data we are copying out.
372 * root is the tree we are copying into, and path is a scratch
373 * path for use in this function (it should be released on entry and
374 * will be released on exit).
376 * If the key is already in the destination tree the existing item is
377 * overwritten. If the existing item isn't big enough, it is extended.
378 * If it is too large, it is truncated.
380 * If the key isn't in the destination yet, a new item is inserted.
382 static int overwrite_item(struct btrfs_trans_handle *trans,
383 struct btrfs_root *root,
384 struct btrfs_path *path,
385 struct extent_buffer *eb, int slot,
386 struct btrfs_key *key)
390 u64 saved_i_size = 0;
391 int save_old_i_size = 0;
392 unsigned long src_ptr;
393 unsigned long dst_ptr;
394 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
397 * This is only used during log replay, so the root is always from a
398 * fs/subvolume tree. In case we ever need to support a log root, then
399 * we'll have to clone the leaf in the path, release the path and use
400 * the leaf before writing into the log tree. See the comments at
401 * copy_items() for more details.
403 ASSERT(root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
405 item_size = btrfs_item_size(eb, slot);
406 src_ptr = btrfs_item_ptr_offset(eb, slot);
408 /* Look for the key in the destination tree. */
409 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
416 u32 dst_size = btrfs_item_size(path->nodes[0],
418 if (dst_size != item_size)
421 if (item_size == 0) {
422 btrfs_release_path(path);
425 dst_copy = kmalloc(item_size, GFP_NOFS);
426 src_copy = kmalloc(item_size, GFP_NOFS);
427 if (!dst_copy || !src_copy) {
428 btrfs_release_path(path);
434 read_extent_buffer(eb, src_copy, src_ptr, item_size);
436 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
437 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
439 ret = memcmp(dst_copy, src_copy, item_size);
444 * they have the same contents, just return, this saves
445 * us from cowing blocks in the destination tree and doing
446 * extra writes that may not have been done by a previous
450 btrfs_release_path(path);
455 * We need to load the old nbytes into the inode so when we
456 * replay the extents we've logged we get the right nbytes.
459 struct btrfs_inode_item *item;
463 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
464 struct btrfs_inode_item);
465 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
466 item = btrfs_item_ptr(eb, slot,
467 struct btrfs_inode_item);
468 btrfs_set_inode_nbytes(eb, item, nbytes);
471 * If this is a directory we need to reset the i_size to
472 * 0 so that we can set it up properly when replaying
473 * the rest of the items in this log.
475 mode = btrfs_inode_mode(eb, item);
477 btrfs_set_inode_size(eb, item, 0);
479 } else if (inode_item) {
480 struct btrfs_inode_item *item;
484 * New inode, set nbytes to 0 so that the nbytes comes out
485 * properly when we replay the extents.
487 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
488 btrfs_set_inode_nbytes(eb, item, 0);
491 * If this is a directory we need to reset the i_size to 0 so
492 * that we can set it up properly when replaying the rest of
493 * the items in this log.
495 mode = btrfs_inode_mode(eb, item);
497 btrfs_set_inode_size(eb, item, 0);
500 btrfs_release_path(path);
501 /* try to insert the key into the destination tree */
502 path->skip_release_on_error = 1;
503 ret = btrfs_insert_empty_item(trans, root, path,
505 path->skip_release_on_error = 0;
507 /* make sure any existing item is the correct size */
508 if (ret == -EEXIST || ret == -EOVERFLOW) {
510 found_size = btrfs_item_size(path->nodes[0],
512 if (found_size > item_size)
513 btrfs_truncate_item(path, item_size, 1);
514 else if (found_size < item_size)
515 btrfs_extend_item(path, item_size - found_size);
519 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
522 /* don't overwrite an existing inode if the generation number
523 * was logged as zero. This is done when the tree logging code
524 * is just logging an inode to make sure it exists after recovery.
526 * Also, don't overwrite i_size on directories during replay.
527 * log replay inserts and removes directory items based on the
528 * state of the tree found in the subvolume, and i_size is modified
531 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
532 struct btrfs_inode_item *src_item;
533 struct btrfs_inode_item *dst_item;
535 src_item = (struct btrfs_inode_item *)src_ptr;
536 dst_item = (struct btrfs_inode_item *)dst_ptr;
538 if (btrfs_inode_generation(eb, src_item) == 0) {
539 struct extent_buffer *dst_eb = path->nodes[0];
540 const u64 ino_size = btrfs_inode_size(eb, src_item);
543 * For regular files an ino_size == 0 is used only when
544 * logging that an inode exists, as part of a directory
545 * fsync, and the inode wasn't fsynced before. In this
546 * case don't set the size of the inode in the fs/subvol
547 * tree, otherwise we would be throwing valid data away.
549 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
550 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
552 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
556 if (S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
557 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
559 saved_i_size = btrfs_inode_size(path->nodes[0],
564 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
567 if (save_old_i_size) {
568 struct btrfs_inode_item *dst_item;
569 dst_item = (struct btrfs_inode_item *)dst_ptr;
570 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
573 /* make sure the generation is filled in */
574 if (key->type == BTRFS_INODE_ITEM_KEY) {
575 struct btrfs_inode_item *dst_item;
576 dst_item = (struct btrfs_inode_item *)dst_ptr;
577 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
578 btrfs_set_inode_generation(path->nodes[0], dst_item,
583 btrfs_mark_buffer_dirty(path->nodes[0]);
584 btrfs_release_path(path);
588 static int read_alloc_one_name(struct extent_buffer *eb, void *start, int len,
589 struct fscrypt_str *name)
593 buf = kmalloc(len, GFP_NOFS);
597 read_extent_buffer(eb, buf, (unsigned long)start, len);
604 * simple helper to read an inode off the disk from a given root
605 * This can only be called for subvolume roots and not for the log
607 static noinline struct inode *read_one_inode(struct btrfs_root *root,
612 inode = btrfs_iget(root->fs_info->sb, objectid, root);
618 /* replays a single extent in 'eb' at 'slot' with 'key' into the
619 * subvolume 'root'. path is released on entry and should be released
622 * extents in the log tree have not been allocated out of the extent
623 * tree yet. So, this completes the allocation, taking a reference
624 * as required if the extent already exists or creating a new extent
625 * if it isn't in the extent allocation tree yet.
627 * The extent is inserted into the file, dropping any existing extents
628 * from the file that overlap the new one.
630 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
631 struct btrfs_root *root,
632 struct btrfs_path *path,
633 struct extent_buffer *eb, int slot,
634 struct btrfs_key *key)
636 struct btrfs_drop_extents_args drop_args = { 0 };
637 struct btrfs_fs_info *fs_info = root->fs_info;
640 u64 start = key->offset;
642 struct btrfs_file_extent_item *item;
643 struct inode *inode = NULL;
647 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
648 found_type = btrfs_file_extent_type(eb, item);
650 if (found_type == BTRFS_FILE_EXTENT_REG ||
651 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
652 nbytes = btrfs_file_extent_num_bytes(eb, item);
653 extent_end = start + nbytes;
656 * We don't add to the inodes nbytes if we are prealloc or a
659 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
661 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
662 size = btrfs_file_extent_ram_bytes(eb, item);
663 nbytes = btrfs_file_extent_ram_bytes(eb, item);
664 extent_end = ALIGN(start + size,
665 fs_info->sectorsize);
671 inode = read_one_inode(root, key->objectid);
678 * first check to see if we already have this extent in the
679 * file. This must be done before the btrfs_drop_extents run
680 * so we don't try to drop this extent.
682 ret = btrfs_lookup_file_extent(trans, root, path,
683 btrfs_ino(BTRFS_I(inode)), start, 0);
686 (found_type == BTRFS_FILE_EXTENT_REG ||
687 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
688 struct btrfs_file_extent_item cmp1;
689 struct btrfs_file_extent_item cmp2;
690 struct btrfs_file_extent_item *existing;
691 struct extent_buffer *leaf;
693 leaf = path->nodes[0];
694 existing = btrfs_item_ptr(leaf, path->slots[0],
695 struct btrfs_file_extent_item);
697 read_extent_buffer(eb, &cmp1, (unsigned long)item,
699 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
703 * we already have a pointer to this exact extent,
704 * we don't have to do anything
706 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
707 btrfs_release_path(path);
711 btrfs_release_path(path);
713 /* drop any overlapping extents */
714 drop_args.start = start;
715 drop_args.end = extent_end;
716 drop_args.drop_cache = true;
717 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
721 if (found_type == BTRFS_FILE_EXTENT_REG ||
722 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
724 unsigned long dest_offset;
725 struct btrfs_key ins;
727 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
728 btrfs_fs_incompat(fs_info, NO_HOLES))
731 ret = btrfs_insert_empty_item(trans, root, path, key,
735 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
737 copy_extent_buffer(path->nodes[0], eb, dest_offset,
738 (unsigned long)item, sizeof(*item));
740 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
741 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
742 ins.type = BTRFS_EXTENT_ITEM_KEY;
743 offset = key->offset - btrfs_file_extent_offset(eb, item);
746 * Manually record dirty extent, as here we did a shallow
747 * file extent item copy and skip normal backref update,
748 * but modifying extent tree all by ourselves.
749 * So need to manually record dirty extent for qgroup,
750 * as the owner of the file extent changed from log tree
751 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
753 ret = btrfs_qgroup_trace_extent(trans,
754 btrfs_file_extent_disk_bytenr(eb, item),
755 btrfs_file_extent_disk_num_bytes(eb, item));
759 if (ins.objectid > 0) {
760 struct btrfs_ref ref = { 0 };
763 LIST_HEAD(ordered_sums);
766 * is this extent already allocated in the extent
767 * allocation tree? If so, just add a reference
769 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
773 } else if (ret == 0) {
774 btrfs_init_generic_ref(&ref,
775 BTRFS_ADD_DELAYED_REF,
776 ins.objectid, ins.offset, 0);
777 btrfs_init_data_ref(&ref,
778 root->root_key.objectid,
779 key->objectid, offset, 0, false);
780 ret = btrfs_inc_extent_ref(trans, &ref);
785 * insert the extent pointer in the extent
788 ret = btrfs_alloc_logged_file_extent(trans,
789 root->root_key.objectid,
790 key->objectid, offset, &ins);
794 btrfs_release_path(path);
796 if (btrfs_file_extent_compression(eb, item)) {
797 csum_start = ins.objectid;
798 csum_end = csum_start + ins.offset;
800 csum_start = ins.objectid +
801 btrfs_file_extent_offset(eb, item);
802 csum_end = csum_start +
803 btrfs_file_extent_num_bytes(eb, item);
806 ret = btrfs_lookup_csums_list(root->log_root,
807 csum_start, csum_end - 1,
808 &ordered_sums, 0, false);
812 * Now delete all existing cums in the csum root that
813 * cover our range. We do this because we can have an
814 * extent that is completely referenced by one file
815 * extent item and partially referenced by another
816 * file extent item (like after using the clone or
817 * extent_same ioctls). In this case if we end up doing
818 * the replay of the one that partially references the
819 * extent first, and we do not do the csum deletion
820 * below, we can get 2 csum items in the csum tree that
821 * overlap each other. For example, imagine our log has
822 * the two following file extent items:
824 * key (257 EXTENT_DATA 409600)
825 * extent data disk byte 12845056 nr 102400
826 * extent data offset 20480 nr 20480 ram 102400
828 * key (257 EXTENT_DATA 819200)
829 * extent data disk byte 12845056 nr 102400
830 * extent data offset 0 nr 102400 ram 102400
832 * Where the second one fully references the 100K extent
833 * that starts at disk byte 12845056, and the log tree
834 * has a single csum item that covers the entire range
837 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
839 * After the first file extent item is replayed, the
840 * csum tree gets the following csum item:
842 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
844 * Which covers the 20K sub-range starting at offset 20K
845 * of our extent. Now when we replay the second file
846 * extent item, if we do not delete existing csum items
847 * that cover any of its blocks, we end up getting two
848 * csum items in our csum tree that overlap each other:
850 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
851 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
853 * Which is a problem, because after this anyone trying
854 * to lookup up for the checksum of any block of our
855 * extent starting at an offset of 40K or higher, will
856 * end up looking at the second csum item only, which
857 * does not contain the checksum for any block starting
858 * at offset 40K or higher of our extent.
860 while (!list_empty(&ordered_sums)) {
861 struct btrfs_ordered_sum *sums;
862 struct btrfs_root *csum_root;
864 sums = list_entry(ordered_sums.next,
865 struct btrfs_ordered_sum,
867 csum_root = btrfs_csum_root(fs_info,
870 ret = btrfs_del_csums(trans, csum_root,
874 ret = btrfs_csum_file_blocks(trans,
877 list_del(&sums->list);
883 btrfs_release_path(path);
885 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
886 /* inline extents are easy, we just overwrite them */
887 ret = overwrite_item(trans, root, path, eb, slot, key);
892 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
898 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
899 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
905 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
906 struct btrfs_inode *dir,
907 struct btrfs_inode *inode,
908 const struct fscrypt_str *name)
912 ret = btrfs_unlink_inode(trans, dir, inode, name);
916 * Whenever we need to check if a name exists or not, we check the
917 * fs/subvolume tree. So after an unlink we must run delayed items, so
918 * that future checks for a name during log replay see that the name
919 * does not exists anymore.
921 return btrfs_run_delayed_items(trans);
925 * when cleaning up conflicts between the directory names in the
926 * subvolume, directory names in the log and directory names in the
927 * inode back references, we may have to unlink inodes from directories.
929 * This is a helper function to do the unlink of a specific directory
932 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
933 struct btrfs_path *path,
934 struct btrfs_inode *dir,
935 struct btrfs_dir_item *di)
937 struct btrfs_root *root = dir->root;
939 struct fscrypt_str name;
940 struct extent_buffer *leaf;
941 struct btrfs_key location;
944 leaf = path->nodes[0];
946 btrfs_dir_item_key_to_cpu(leaf, di, &location);
947 ret = read_alloc_one_name(leaf, di + 1, btrfs_dir_name_len(leaf, di), &name);
951 btrfs_release_path(path);
953 inode = read_one_inode(root, location.objectid);
959 ret = link_to_fixup_dir(trans, root, path, location.objectid);
963 ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), &name);
971 * See if a given name and sequence number found in an inode back reference are
972 * already in a directory and correctly point to this inode.
974 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
977 static noinline int inode_in_dir(struct btrfs_root *root,
978 struct btrfs_path *path,
979 u64 dirid, u64 objectid, u64 index,
980 struct fscrypt_str *name)
982 struct btrfs_dir_item *di;
983 struct btrfs_key location;
986 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
992 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
993 if (location.objectid != objectid)
999 btrfs_release_path(path);
1000 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, 0);
1005 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1006 if (location.objectid == objectid)
1010 btrfs_release_path(path);
1015 * helper function to check a log tree for a named back reference in
1016 * an inode. This is used to decide if a back reference that is
1017 * found in the subvolume conflicts with what we find in the log.
1019 * inode backreferences may have multiple refs in a single item,
1020 * during replay we process one reference at a time, and we don't
1021 * want to delete valid links to a file from the subvolume if that
1022 * link is also in the log.
1024 static noinline int backref_in_log(struct btrfs_root *log,
1025 struct btrfs_key *key,
1027 const struct fscrypt_str *name)
1029 struct btrfs_path *path;
1032 path = btrfs_alloc_path();
1036 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1039 } else if (ret == 1) {
1044 if (key->type == BTRFS_INODE_EXTREF_KEY)
1045 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1047 ref_objectid, name);
1049 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1050 path->slots[0], name);
1052 btrfs_free_path(path);
1056 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root,
1058 struct btrfs_path *path,
1059 struct btrfs_root *log_root,
1060 struct btrfs_inode *dir,
1061 struct btrfs_inode *inode,
1062 u64 inode_objectid, u64 parent_objectid,
1063 u64 ref_index, struct fscrypt_str *name)
1066 struct extent_buffer *leaf;
1067 struct btrfs_dir_item *di;
1068 struct btrfs_key search_key;
1069 struct btrfs_inode_extref *extref;
1072 /* Search old style refs */
1073 search_key.objectid = inode_objectid;
1074 search_key.type = BTRFS_INODE_REF_KEY;
1075 search_key.offset = parent_objectid;
1076 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1078 struct btrfs_inode_ref *victim_ref;
1080 unsigned long ptr_end;
1082 leaf = path->nodes[0];
1084 /* are we trying to overwrite a back ref for the root directory
1085 * if so, just jump out, we're done
1087 if (search_key.objectid == search_key.offset)
1090 /* check all the names in this back reference to see
1091 * if they are in the log. if so, we allow them to stay
1092 * otherwise they must be unlinked as a conflict
1094 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1095 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1096 while (ptr < ptr_end) {
1097 struct fscrypt_str victim_name;
1099 victim_ref = (struct btrfs_inode_ref *)ptr;
1100 ret = read_alloc_one_name(leaf, (victim_ref + 1),
1101 btrfs_inode_ref_name_len(leaf, victim_ref),
1106 ret = backref_in_log(log_root, &search_key,
1107 parent_objectid, &victim_name);
1109 kfree(victim_name.name);
1112 inc_nlink(&inode->vfs_inode);
1113 btrfs_release_path(path);
1115 ret = unlink_inode_for_log_replay(trans, dir, inode,
1117 kfree(victim_name.name);
1122 kfree(victim_name.name);
1124 ptr = (unsigned long)(victim_ref + 1) + victim_name.len;
1127 btrfs_release_path(path);
1129 /* Same search but for extended refs */
1130 extref = btrfs_lookup_inode_extref(NULL, root, path, name,
1131 inode_objectid, parent_objectid, 0,
1133 if (IS_ERR(extref)) {
1134 return PTR_ERR(extref);
1135 } else if (extref) {
1139 struct inode *victim_parent;
1141 leaf = path->nodes[0];
1143 item_size = btrfs_item_size(leaf, path->slots[0]);
1144 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1146 while (cur_offset < item_size) {
1147 struct fscrypt_str victim_name;
1149 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1151 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1154 ret = read_alloc_one_name(leaf, &extref->name,
1155 btrfs_inode_extref_name_len(leaf, extref),
1160 search_key.objectid = inode_objectid;
1161 search_key.type = BTRFS_INODE_EXTREF_KEY;
1162 search_key.offset = btrfs_extref_hash(parent_objectid,
1165 ret = backref_in_log(log_root, &search_key,
1166 parent_objectid, &victim_name);
1168 kfree(victim_name.name);
1172 victim_parent = read_one_inode(root,
1174 if (victim_parent) {
1175 inc_nlink(&inode->vfs_inode);
1176 btrfs_release_path(path);
1178 ret = unlink_inode_for_log_replay(trans,
1179 BTRFS_I(victim_parent),
1180 inode, &victim_name);
1182 iput(victim_parent);
1183 kfree(victim_name.name);
1188 kfree(victim_name.name);
1190 cur_offset += victim_name.len + sizeof(*extref);
1193 btrfs_release_path(path);
1195 /* look for a conflicting sequence number */
1196 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1197 ref_index, name, 0);
1201 ret = drop_one_dir_item(trans, path, dir, di);
1205 btrfs_release_path(path);
1207 /* look for a conflicting name */
1208 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), name, 0);
1212 ret = drop_one_dir_item(trans, path, dir, di);
1216 btrfs_release_path(path);
1221 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1222 struct fscrypt_str *name, u64 *index,
1223 u64 *parent_objectid)
1225 struct btrfs_inode_extref *extref;
1228 extref = (struct btrfs_inode_extref *)ref_ptr;
1230 ret = read_alloc_one_name(eb, &extref->name,
1231 btrfs_inode_extref_name_len(eb, extref), name);
1236 *index = btrfs_inode_extref_index(eb, extref);
1237 if (parent_objectid)
1238 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1243 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1244 struct fscrypt_str *name, u64 *index)
1246 struct btrfs_inode_ref *ref;
1249 ref = (struct btrfs_inode_ref *)ref_ptr;
1251 ret = read_alloc_one_name(eb, ref + 1, btrfs_inode_ref_name_len(eb, ref),
1257 *index = btrfs_inode_ref_index(eb, ref);
1263 * Take an inode reference item from the log tree and iterate all names from the
1264 * inode reference item in the subvolume tree with the same key (if it exists).
1265 * For any name that is not in the inode reference item from the log tree, do a
1266 * proper unlink of that name (that is, remove its entry from the inode
1267 * reference item and both dir index keys).
1269 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root,
1271 struct btrfs_path *path,
1272 struct btrfs_inode *inode,
1273 struct extent_buffer *log_eb,
1275 struct btrfs_key *key)
1278 unsigned long ref_ptr;
1279 unsigned long ref_end;
1280 struct extent_buffer *eb;
1283 btrfs_release_path(path);
1284 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1292 eb = path->nodes[0];
1293 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1294 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1295 while (ref_ptr < ref_end) {
1296 struct fscrypt_str name;
1299 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1300 ret = extref_get_fields(eb, ref_ptr, &name,
1303 parent_id = key->offset;
1304 ret = ref_get_fields(eb, ref_ptr, &name, NULL);
1309 if (key->type == BTRFS_INODE_EXTREF_KEY)
1310 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1313 ret = !!btrfs_find_name_in_backref(log_eb, log_slot, &name);
1318 btrfs_release_path(path);
1319 dir = read_one_inode(root, parent_id);
1325 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1335 ref_ptr += name.len;
1336 if (key->type == BTRFS_INODE_EXTREF_KEY)
1337 ref_ptr += sizeof(struct btrfs_inode_extref);
1339 ref_ptr += sizeof(struct btrfs_inode_ref);
1343 btrfs_release_path(path);
1348 * replay one inode back reference item found in the log tree.
1349 * eb, slot and key refer to the buffer and key found in the log tree.
1350 * root is the destination we are replaying into, and path is for temp
1351 * use by this function. (it should be released on return).
1353 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1354 struct btrfs_root *root,
1355 struct btrfs_root *log,
1356 struct btrfs_path *path,
1357 struct extent_buffer *eb, int slot,
1358 struct btrfs_key *key)
1360 struct inode *dir = NULL;
1361 struct inode *inode = NULL;
1362 unsigned long ref_ptr;
1363 unsigned long ref_end;
1364 struct fscrypt_str name;
1366 int log_ref_ver = 0;
1367 u64 parent_objectid;
1370 int ref_struct_size;
1372 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1373 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1375 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1376 struct btrfs_inode_extref *r;
1378 ref_struct_size = sizeof(struct btrfs_inode_extref);
1380 r = (struct btrfs_inode_extref *)ref_ptr;
1381 parent_objectid = btrfs_inode_extref_parent(eb, r);
1383 ref_struct_size = sizeof(struct btrfs_inode_ref);
1384 parent_objectid = key->offset;
1386 inode_objectid = key->objectid;
1389 * it is possible that we didn't log all the parent directories
1390 * for a given inode. If we don't find the dir, just don't
1391 * copy the back ref in. The link count fixup code will take
1394 dir = read_one_inode(root, parent_objectid);
1400 inode = read_one_inode(root, inode_objectid);
1406 while (ref_ptr < ref_end) {
1408 ret = extref_get_fields(eb, ref_ptr, &name,
1409 &ref_index, &parent_objectid);
1411 * parent object can change from one array
1415 dir = read_one_inode(root, parent_objectid);
1421 ret = ref_get_fields(eb, ref_ptr, &name, &ref_index);
1426 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1427 btrfs_ino(BTRFS_I(inode)), ref_index, &name);
1430 } else if (ret == 0) {
1432 * look for a conflicting back reference in the
1433 * metadata. if we find one we have to unlink that name
1434 * of the file before we add our new link. Later on, we
1435 * overwrite any existing back reference, and we don't
1436 * want to create dangling pointers in the directory.
1438 ret = __add_inode_ref(trans, root, path, log,
1439 BTRFS_I(dir), BTRFS_I(inode),
1440 inode_objectid, parent_objectid,
1448 /* insert our name */
1449 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1450 &name, 0, ref_index);
1454 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1458 /* Else, ret == 1, we already have a perfect match, we're done. */
1460 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + name.len;
1470 * Before we overwrite the inode reference item in the subvolume tree
1471 * with the item from the log tree, we must unlink all names from the
1472 * parent directory that are in the subvolume's tree inode reference
1473 * item, otherwise we end up with an inconsistent subvolume tree where
1474 * dir index entries exist for a name but there is no inode reference
1475 * item with the same name.
1477 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1482 /* finally write the back reference in the inode */
1483 ret = overwrite_item(trans, root, path, eb, slot, key);
1485 btrfs_release_path(path);
1492 static int count_inode_extrefs(struct btrfs_root *root,
1493 struct btrfs_inode *inode, struct btrfs_path *path)
1497 unsigned int nlink = 0;
1500 u64 inode_objectid = btrfs_ino(inode);
1503 struct btrfs_inode_extref *extref;
1504 struct extent_buffer *leaf;
1507 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1512 leaf = path->nodes[0];
1513 item_size = btrfs_item_size(leaf, path->slots[0]);
1514 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1517 while (cur_offset < item_size) {
1518 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1519 name_len = btrfs_inode_extref_name_len(leaf, extref);
1523 cur_offset += name_len + sizeof(*extref);
1527 btrfs_release_path(path);
1529 btrfs_release_path(path);
1531 if (ret < 0 && ret != -ENOENT)
1536 static int count_inode_refs(struct btrfs_root *root,
1537 struct btrfs_inode *inode, struct btrfs_path *path)
1540 struct btrfs_key key;
1541 unsigned int nlink = 0;
1543 unsigned long ptr_end;
1545 u64 ino = btrfs_ino(inode);
1548 key.type = BTRFS_INODE_REF_KEY;
1549 key.offset = (u64)-1;
1552 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1556 if (path->slots[0] == 0)
1561 btrfs_item_key_to_cpu(path->nodes[0], &key,
1563 if (key.objectid != ino ||
1564 key.type != BTRFS_INODE_REF_KEY)
1566 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1567 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1569 while (ptr < ptr_end) {
1570 struct btrfs_inode_ref *ref;
1572 ref = (struct btrfs_inode_ref *)ptr;
1573 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1575 ptr = (unsigned long)(ref + 1) + name_len;
1579 if (key.offset == 0)
1581 if (path->slots[0] > 0) {
1586 btrfs_release_path(path);
1588 btrfs_release_path(path);
1594 * There are a few corners where the link count of the file can't
1595 * be properly maintained during replay. So, instead of adding
1596 * lots of complexity to the log code, we just scan the backrefs
1597 * for any file that has been through replay.
1599 * The scan will update the link count on the inode to reflect the
1600 * number of back refs found. If it goes down to zero, the iput
1601 * will free the inode.
1603 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1604 struct btrfs_root *root,
1605 struct inode *inode)
1607 struct btrfs_path *path;
1610 u64 ino = btrfs_ino(BTRFS_I(inode));
1612 path = btrfs_alloc_path();
1616 ret = count_inode_refs(root, BTRFS_I(inode), path);
1622 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1630 if (nlink != inode->i_nlink) {
1631 set_nlink(inode, nlink);
1632 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1636 BTRFS_I(inode)->index_cnt = (u64)-1;
1638 if (inode->i_nlink == 0) {
1639 if (S_ISDIR(inode->i_mode)) {
1640 ret = replay_dir_deletes(trans, root, NULL, path,
1645 ret = btrfs_insert_orphan_item(trans, root, ino);
1651 btrfs_free_path(path);
1655 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1656 struct btrfs_root *root,
1657 struct btrfs_path *path)
1660 struct btrfs_key key;
1661 struct inode *inode;
1663 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1664 key.type = BTRFS_ORPHAN_ITEM_KEY;
1665 key.offset = (u64)-1;
1667 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1673 if (path->slots[0] == 0)
1678 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1679 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1680 key.type != BTRFS_ORPHAN_ITEM_KEY)
1683 ret = btrfs_del_item(trans, root, path);
1687 btrfs_release_path(path);
1688 inode = read_one_inode(root, key.offset);
1694 ret = fixup_inode_link_count(trans, root, inode);
1700 * fixup on a directory may create new entries,
1701 * make sure we always look for the highset possible
1704 key.offset = (u64)-1;
1706 btrfs_release_path(path);
1712 * record a given inode in the fixup dir so we can check its link
1713 * count when replay is done. The link count is incremented here
1714 * so the inode won't go away until we check it
1716 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1717 struct btrfs_root *root,
1718 struct btrfs_path *path,
1721 struct btrfs_key key;
1723 struct inode *inode;
1725 inode = read_one_inode(root, objectid);
1729 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1730 key.type = BTRFS_ORPHAN_ITEM_KEY;
1731 key.offset = objectid;
1733 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1735 btrfs_release_path(path);
1737 if (!inode->i_nlink)
1738 set_nlink(inode, 1);
1741 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1742 } else if (ret == -EEXIST) {
1751 * when replaying the log for a directory, we only insert names
1752 * for inodes that actually exist. This means an fsync on a directory
1753 * does not implicitly fsync all the new files in it
1755 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 u64 dirid, u64 index,
1758 const struct fscrypt_str *name,
1759 struct btrfs_key *location)
1761 struct inode *inode;
1765 inode = read_one_inode(root, location->objectid);
1769 dir = read_one_inode(root, dirid);
1775 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1778 /* FIXME, put inode into FIXUP list */
1785 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1786 struct btrfs_inode *dir,
1787 struct btrfs_path *path,
1788 struct btrfs_dir_item *dst_di,
1789 const struct btrfs_key *log_key,
1793 struct btrfs_key found_key;
1795 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1796 /* The existing dentry points to the same inode, don't delete it. */
1797 if (found_key.objectid == log_key->objectid &&
1798 found_key.type == log_key->type &&
1799 found_key.offset == log_key->offset &&
1800 btrfs_dir_flags(path->nodes[0], dst_di) == log_flags)
1804 * Don't drop the conflicting directory entry if the inode for the new
1805 * entry doesn't exist.
1810 return drop_one_dir_item(trans, path, dir, dst_di);
1814 * take a single entry in a log directory item and replay it into
1817 * if a conflicting item exists in the subdirectory already,
1818 * the inode it points to is unlinked and put into the link count
1821 * If a name from the log points to a file or directory that does
1822 * not exist in the FS, it is skipped. fsyncs on directories
1823 * do not force down inodes inside that directory, just changes to the
1824 * names or unlinks in a directory.
1826 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1827 * non-existing inode) and 1 if the name was replayed.
1829 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1830 struct btrfs_root *root,
1831 struct btrfs_path *path,
1832 struct extent_buffer *eb,
1833 struct btrfs_dir_item *di,
1834 struct btrfs_key *key)
1836 struct fscrypt_str name;
1837 struct btrfs_dir_item *dir_dst_di;
1838 struct btrfs_dir_item *index_dst_di;
1839 bool dir_dst_matches = false;
1840 bool index_dst_matches = false;
1841 struct btrfs_key log_key;
1842 struct btrfs_key search_key;
1847 bool update_size = true;
1848 bool name_added = false;
1850 dir = read_one_inode(root, key->objectid);
1854 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
1858 log_flags = btrfs_dir_flags(eb, di);
1859 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1860 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1861 btrfs_release_path(path);
1864 exists = (ret == 0);
1867 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1869 if (IS_ERR(dir_dst_di)) {
1870 ret = PTR_ERR(dir_dst_di);
1872 } else if (dir_dst_di) {
1873 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1874 dir_dst_di, &log_key,
1878 dir_dst_matches = (ret == 1);
1881 btrfs_release_path(path);
1883 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1884 key->objectid, key->offset,
1886 if (IS_ERR(index_dst_di)) {
1887 ret = PTR_ERR(index_dst_di);
1889 } else if (index_dst_di) {
1890 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1891 index_dst_di, &log_key,
1895 index_dst_matches = (ret == 1);
1898 btrfs_release_path(path);
1900 if (dir_dst_matches && index_dst_matches) {
1902 update_size = false;
1907 * Check if the inode reference exists in the log for the given name,
1908 * inode and parent inode
1910 search_key.objectid = log_key.objectid;
1911 search_key.type = BTRFS_INODE_REF_KEY;
1912 search_key.offset = key->objectid;
1913 ret = backref_in_log(root->log_root, &search_key, 0, &name);
1917 /* The dentry will be added later. */
1919 update_size = false;
1923 search_key.objectid = log_key.objectid;
1924 search_key.type = BTRFS_INODE_EXTREF_KEY;
1925 search_key.offset = key->objectid;
1926 ret = backref_in_log(root->log_root, &search_key, key->objectid, &name);
1930 /* The dentry will be added later. */
1932 update_size = false;
1935 btrfs_release_path(path);
1936 ret = insert_one_name(trans, root, key->objectid, key->offset,
1938 if (ret && ret != -ENOENT && ret != -EEXIST)
1942 update_size = false;
1946 if (!ret && update_size) {
1947 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2);
1948 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
1952 if (!ret && name_added)
1957 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1958 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1959 struct btrfs_root *root,
1960 struct btrfs_path *path,
1961 struct extent_buffer *eb, int slot,
1962 struct btrfs_key *key)
1965 struct btrfs_dir_item *di;
1967 /* We only log dir index keys, which only contain a single dir item. */
1968 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
1970 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1971 ret = replay_one_name(trans, root, path, eb, di, key);
1976 * If this entry refers to a non-directory (directories can not have a
1977 * link count > 1) and it was added in the transaction that was not
1978 * committed, make sure we fixup the link count of the inode the entry
1979 * points to. Otherwise something like the following would result in a
1980 * directory pointing to an inode with a wrong link that does not account
1981 * for this dir entry:
1988 * ln testdir/bar testdir/bar_link
1989 * ln testdir/foo testdir/foo_link
1990 * xfs_io -c "fsync" testdir/bar
1994 * mount fs, log replay happens
1996 * File foo would remain with a link count of 1 when it has two entries
1997 * pointing to it in the directory testdir. This would make it impossible
1998 * to ever delete the parent directory has it would result in stale
1999 * dentries that can never be deleted.
2001 if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) {
2002 struct btrfs_path *fixup_path;
2003 struct btrfs_key di_key;
2005 fixup_path = btrfs_alloc_path();
2009 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2010 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2011 btrfs_free_path(fixup_path);
2018 * directory replay has two parts. There are the standard directory
2019 * items in the log copied from the subvolume, and range items
2020 * created in the log while the subvolume was logged.
2022 * The range items tell us which parts of the key space the log
2023 * is authoritative for. During replay, if a key in the subvolume
2024 * directory is in a logged range item, but not actually in the log
2025 * that means it was deleted from the directory before the fsync
2026 * and should be removed.
2028 static noinline int find_dir_range(struct btrfs_root *root,
2029 struct btrfs_path *path,
2031 u64 *start_ret, u64 *end_ret)
2033 struct btrfs_key key;
2035 struct btrfs_dir_log_item *item;
2039 if (*start_ret == (u64)-1)
2042 key.objectid = dirid;
2043 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2044 key.offset = *start_ret;
2046 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2050 if (path->slots[0] == 0)
2055 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2057 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2061 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2062 struct btrfs_dir_log_item);
2063 found_end = btrfs_dir_log_end(path->nodes[0], item);
2065 if (*start_ret >= key.offset && *start_ret <= found_end) {
2067 *start_ret = key.offset;
2068 *end_ret = found_end;
2073 /* check the next slot in the tree to see if it is a valid item */
2074 nritems = btrfs_header_nritems(path->nodes[0]);
2076 if (path->slots[0] >= nritems) {
2077 ret = btrfs_next_leaf(root, path);
2082 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2084 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2088 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2089 struct btrfs_dir_log_item);
2090 found_end = btrfs_dir_log_end(path->nodes[0], item);
2091 *start_ret = key.offset;
2092 *end_ret = found_end;
2095 btrfs_release_path(path);
2100 * this looks for a given directory item in the log. If the directory
2101 * item is not in the log, the item is removed and the inode it points
2104 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *log,
2106 struct btrfs_path *path,
2107 struct btrfs_path *log_path,
2109 struct btrfs_key *dir_key)
2111 struct btrfs_root *root = BTRFS_I(dir)->root;
2113 struct extent_buffer *eb;
2115 struct btrfs_dir_item *di;
2116 struct fscrypt_str name;
2117 struct inode *inode = NULL;
2118 struct btrfs_key location;
2121 * Currently we only log dir index keys. Even if we replay a log created
2122 * by an older kernel that logged both dir index and dir item keys, all
2123 * we need to do is process the dir index keys, we (and our caller) can
2124 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2126 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2128 eb = path->nodes[0];
2129 slot = path->slots[0];
2130 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2131 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
2136 struct btrfs_dir_item *log_di;
2138 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2140 dir_key->offset, &name, 0);
2141 if (IS_ERR(log_di)) {
2142 ret = PTR_ERR(log_di);
2144 } else if (log_di) {
2145 /* The dentry exists in the log, we have nothing to do. */
2151 btrfs_dir_item_key_to_cpu(eb, di, &location);
2152 btrfs_release_path(path);
2153 btrfs_release_path(log_path);
2154 inode = read_one_inode(root, location.objectid);
2160 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2165 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2168 * Unlike dir item keys, dir index keys can only have one name (entry) in
2169 * them, as there are no key collisions since each key has a unique offset
2170 * (an index number), so we're done.
2173 btrfs_release_path(path);
2174 btrfs_release_path(log_path);
2180 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *root,
2182 struct btrfs_root *log,
2183 struct btrfs_path *path,
2186 struct btrfs_key search_key;
2187 struct btrfs_path *log_path;
2192 log_path = btrfs_alloc_path();
2196 search_key.objectid = ino;
2197 search_key.type = BTRFS_XATTR_ITEM_KEY;
2198 search_key.offset = 0;
2200 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2204 nritems = btrfs_header_nritems(path->nodes[0]);
2205 for (i = path->slots[0]; i < nritems; i++) {
2206 struct btrfs_key key;
2207 struct btrfs_dir_item *di;
2208 struct btrfs_dir_item *log_di;
2212 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2213 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2218 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2219 total_size = btrfs_item_size(path->nodes[0], i);
2221 while (cur < total_size) {
2222 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2223 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2224 u32 this_len = sizeof(*di) + name_len + data_len;
2227 name = kmalloc(name_len, GFP_NOFS);
2232 read_extent_buffer(path->nodes[0], name,
2233 (unsigned long)(di + 1), name_len);
2235 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2237 btrfs_release_path(log_path);
2239 /* Doesn't exist in log tree, so delete it. */
2240 btrfs_release_path(path);
2241 di = btrfs_lookup_xattr(trans, root, path, ino,
2242 name, name_len, -1);
2249 ret = btrfs_delete_one_dir_name(trans, root,
2253 btrfs_release_path(path);
2258 if (IS_ERR(log_di)) {
2259 ret = PTR_ERR(log_di);
2263 di = (struct btrfs_dir_item *)((char *)di + this_len);
2266 ret = btrfs_next_leaf(root, path);
2272 btrfs_free_path(log_path);
2273 btrfs_release_path(path);
2279 * deletion replay happens before we copy any new directory items
2280 * out of the log or out of backreferences from inodes. It
2281 * scans the log to find ranges of keys that log is authoritative for,
2282 * and then scans the directory to find items in those ranges that are
2283 * not present in the log.
2285 * Anything we don't find in the log is unlinked and removed from the
2288 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2289 struct btrfs_root *root,
2290 struct btrfs_root *log,
2291 struct btrfs_path *path,
2292 u64 dirid, int del_all)
2297 struct btrfs_key dir_key;
2298 struct btrfs_key found_key;
2299 struct btrfs_path *log_path;
2302 dir_key.objectid = dirid;
2303 dir_key.type = BTRFS_DIR_INDEX_KEY;
2304 log_path = btrfs_alloc_path();
2308 dir = read_one_inode(root, dirid);
2309 /* it isn't an error if the inode isn't there, that can happen
2310 * because we replay the deletes before we copy in the inode item
2314 btrfs_free_path(log_path);
2322 range_end = (u64)-1;
2324 ret = find_dir_range(log, path, dirid,
2325 &range_start, &range_end);
2332 dir_key.offset = range_start;
2335 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2340 nritems = btrfs_header_nritems(path->nodes[0]);
2341 if (path->slots[0] >= nritems) {
2342 ret = btrfs_next_leaf(root, path);
2348 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2350 if (found_key.objectid != dirid ||
2351 found_key.type != dir_key.type) {
2356 if (found_key.offset > range_end)
2359 ret = check_item_in_log(trans, log, path,
2364 if (found_key.offset == (u64)-1)
2366 dir_key.offset = found_key.offset + 1;
2368 btrfs_release_path(path);
2369 if (range_end == (u64)-1)
2371 range_start = range_end + 1;
2375 btrfs_release_path(path);
2376 btrfs_free_path(log_path);
2382 * the process_func used to replay items from the log tree. This
2383 * gets called in two different stages. The first stage just looks
2384 * for inodes and makes sure they are all copied into the subvolume.
2386 * The second stage copies all the other item types from the log into
2387 * the subvolume. The two stage approach is slower, but gets rid of
2388 * lots of complexity around inodes referencing other inodes that exist
2389 * only in the log (references come from either directory items or inode
2392 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2393 struct walk_control *wc, u64 gen, int level)
2396 struct btrfs_tree_parent_check check = {
2400 struct btrfs_path *path;
2401 struct btrfs_root *root = wc->replay_dest;
2402 struct btrfs_key key;
2406 ret = btrfs_read_extent_buffer(eb, &check);
2410 level = btrfs_header_level(eb);
2415 path = btrfs_alloc_path();
2419 nritems = btrfs_header_nritems(eb);
2420 for (i = 0; i < nritems; i++) {
2421 btrfs_item_key_to_cpu(eb, &key, i);
2423 /* inode keys are done during the first stage */
2424 if (key.type == BTRFS_INODE_ITEM_KEY &&
2425 wc->stage == LOG_WALK_REPLAY_INODES) {
2426 struct btrfs_inode_item *inode_item;
2429 inode_item = btrfs_item_ptr(eb, i,
2430 struct btrfs_inode_item);
2432 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2433 * and never got linked before the fsync, skip it, as
2434 * replaying it is pointless since it would be deleted
2435 * later. We skip logging tmpfiles, but it's always
2436 * possible we are replaying a log created with a kernel
2437 * that used to log tmpfiles.
2439 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2440 wc->ignore_cur_inode = true;
2443 wc->ignore_cur_inode = false;
2445 ret = replay_xattr_deletes(wc->trans, root, log,
2446 path, key.objectid);
2449 mode = btrfs_inode_mode(eb, inode_item);
2450 if (S_ISDIR(mode)) {
2451 ret = replay_dir_deletes(wc->trans,
2452 root, log, path, key.objectid, 0);
2456 ret = overwrite_item(wc->trans, root, path,
2462 * Before replaying extents, truncate the inode to its
2463 * size. We need to do it now and not after log replay
2464 * because before an fsync we can have prealloc extents
2465 * added beyond the inode's i_size. If we did it after,
2466 * through orphan cleanup for example, we would drop
2467 * those prealloc extents just after replaying them.
2469 if (S_ISREG(mode)) {
2470 struct btrfs_drop_extents_args drop_args = { 0 };
2471 struct inode *inode;
2474 inode = read_one_inode(root, key.objectid);
2479 from = ALIGN(i_size_read(inode),
2480 root->fs_info->sectorsize);
2481 drop_args.start = from;
2482 drop_args.end = (u64)-1;
2483 drop_args.drop_cache = true;
2484 ret = btrfs_drop_extents(wc->trans, root,
2488 inode_sub_bytes(inode,
2489 drop_args.bytes_found);
2490 /* Update the inode's nbytes. */
2491 ret = btrfs_update_inode(wc->trans,
2492 root, BTRFS_I(inode));
2499 ret = link_to_fixup_dir(wc->trans, root,
2500 path, key.objectid);
2505 if (wc->ignore_cur_inode)
2508 if (key.type == BTRFS_DIR_INDEX_KEY &&
2509 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2510 ret = replay_one_dir_item(wc->trans, root, path,
2516 if (wc->stage < LOG_WALK_REPLAY_ALL)
2519 /* these keys are simply copied */
2520 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2521 ret = overwrite_item(wc->trans, root, path,
2525 } else if (key.type == BTRFS_INODE_REF_KEY ||
2526 key.type == BTRFS_INODE_EXTREF_KEY) {
2527 ret = add_inode_ref(wc->trans, root, log, path,
2529 if (ret && ret != -ENOENT)
2532 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2533 ret = replay_one_extent(wc->trans, root, path,
2539 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2540 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2541 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2542 * older kernel with such keys, ignore them.
2545 btrfs_free_path(path);
2550 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2552 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2554 struct btrfs_block_group *cache;
2556 cache = btrfs_lookup_block_group(fs_info, start);
2558 btrfs_err(fs_info, "unable to find block group for %llu", start);
2562 spin_lock(&cache->space_info->lock);
2563 spin_lock(&cache->lock);
2564 cache->reserved -= fs_info->nodesize;
2565 cache->space_info->bytes_reserved -= fs_info->nodesize;
2566 spin_unlock(&cache->lock);
2567 spin_unlock(&cache->space_info->lock);
2569 btrfs_put_block_group(cache);
2572 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2573 struct btrfs_root *root,
2574 struct btrfs_path *path, int *level,
2575 struct walk_control *wc)
2577 struct btrfs_fs_info *fs_info = root->fs_info;
2580 struct extent_buffer *next;
2581 struct extent_buffer *cur;
2585 while (*level > 0) {
2586 struct btrfs_tree_parent_check check = { 0 };
2588 cur = path->nodes[*level];
2590 WARN_ON(btrfs_header_level(cur) != *level);
2592 if (path->slots[*level] >=
2593 btrfs_header_nritems(cur))
2596 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2597 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2598 check.transid = ptr_gen;
2599 check.level = *level - 1;
2600 check.has_first_key = true;
2601 btrfs_node_key_to_cpu(cur, &check.first_key, path->slots[*level]);
2602 blocksize = fs_info->nodesize;
2604 next = btrfs_find_create_tree_block(fs_info, bytenr,
2605 btrfs_header_owner(cur),
2608 return PTR_ERR(next);
2611 ret = wc->process_func(root, next, wc, ptr_gen,
2614 free_extent_buffer(next);
2618 path->slots[*level]++;
2620 ret = btrfs_read_extent_buffer(next, &check);
2622 free_extent_buffer(next);
2627 btrfs_tree_lock(next);
2628 btrfs_clean_tree_block(next);
2629 btrfs_wait_tree_block_writeback(next);
2630 btrfs_tree_unlock(next);
2631 ret = btrfs_pin_reserved_extent(trans,
2634 free_extent_buffer(next);
2637 btrfs_redirty_list_add(
2638 trans->transaction, next);
2640 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2641 clear_extent_buffer_dirty(next);
2642 unaccount_log_buffer(fs_info, bytenr);
2645 free_extent_buffer(next);
2648 ret = btrfs_read_extent_buffer(next, &check);
2650 free_extent_buffer(next);
2654 if (path->nodes[*level-1])
2655 free_extent_buffer(path->nodes[*level-1]);
2656 path->nodes[*level-1] = next;
2657 *level = btrfs_header_level(next);
2658 path->slots[*level] = 0;
2661 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2667 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2668 struct btrfs_root *root,
2669 struct btrfs_path *path, int *level,
2670 struct walk_control *wc)
2672 struct btrfs_fs_info *fs_info = root->fs_info;
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 struct extent_buffer *next;
2694 next = path->nodes[*level];
2697 btrfs_tree_lock(next);
2698 btrfs_clean_tree_block(next);
2699 btrfs_wait_tree_block_writeback(next);
2700 btrfs_tree_unlock(next);
2701 ret = btrfs_pin_reserved_extent(trans,
2702 path->nodes[*level]->start,
2703 path->nodes[*level]->len);
2706 btrfs_redirty_list_add(trans->transaction,
2709 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2710 clear_extent_buffer_dirty(next);
2712 unaccount_log_buffer(fs_info,
2713 path->nodes[*level]->start);
2716 free_extent_buffer(path->nodes[*level]);
2717 path->nodes[*level] = NULL;
2725 * drop the reference count on the tree rooted at 'snap'. This traverses
2726 * the tree freeing any blocks that have a ref count of zero after being
2729 static int walk_log_tree(struct btrfs_trans_handle *trans,
2730 struct btrfs_root *log, struct walk_control *wc)
2732 struct btrfs_fs_info *fs_info = log->fs_info;
2736 struct btrfs_path *path;
2739 path = btrfs_alloc_path();
2743 level = btrfs_header_level(log->node);
2745 path->nodes[level] = log->node;
2746 atomic_inc(&log->node->refs);
2747 path->slots[level] = 0;
2750 wret = walk_down_log_tree(trans, log, path, &level, wc);
2758 wret = walk_up_log_tree(trans, log, path, &level, wc);
2767 /* was the root node processed? if not, catch it here */
2768 if (path->nodes[orig_level]) {
2769 ret = wc->process_func(log, path->nodes[orig_level], wc,
2770 btrfs_header_generation(path->nodes[orig_level]),
2775 struct extent_buffer *next;
2777 next = path->nodes[orig_level];
2780 btrfs_tree_lock(next);
2781 btrfs_clean_tree_block(next);
2782 btrfs_wait_tree_block_writeback(next);
2783 btrfs_tree_unlock(next);
2784 ret = btrfs_pin_reserved_extent(trans,
2785 next->start, next->len);
2788 btrfs_redirty_list_add(trans->transaction, next);
2790 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2791 clear_extent_buffer_dirty(next);
2792 unaccount_log_buffer(fs_info, next->start);
2798 btrfs_free_path(path);
2803 * helper function to update the item for a given subvolumes log root
2804 * in the tree of log roots
2806 static int update_log_root(struct btrfs_trans_handle *trans,
2807 struct btrfs_root *log,
2808 struct btrfs_root_item *root_item)
2810 struct btrfs_fs_info *fs_info = log->fs_info;
2813 if (log->log_transid == 1) {
2814 /* insert root item on the first sync */
2815 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2816 &log->root_key, root_item);
2818 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2819 &log->root_key, root_item);
2824 static void wait_log_commit(struct btrfs_root *root, int transid)
2827 int index = transid % 2;
2830 * we only allow two pending log transactions at a time,
2831 * so we know that if ours is more than 2 older than the
2832 * current transaction, we're done
2835 prepare_to_wait(&root->log_commit_wait[index],
2836 &wait, TASK_UNINTERRUPTIBLE);
2838 if (!(root->log_transid_committed < transid &&
2839 atomic_read(&root->log_commit[index])))
2842 mutex_unlock(&root->log_mutex);
2844 mutex_lock(&root->log_mutex);
2846 finish_wait(&root->log_commit_wait[index], &wait);
2849 static void wait_for_writer(struct btrfs_root *root)
2854 prepare_to_wait(&root->log_writer_wait, &wait,
2855 TASK_UNINTERRUPTIBLE);
2856 if (!atomic_read(&root->log_writers))
2859 mutex_unlock(&root->log_mutex);
2861 mutex_lock(&root->log_mutex);
2863 finish_wait(&root->log_writer_wait, &wait);
2866 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2867 struct btrfs_log_ctx *ctx)
2869 mutex_lock(&root->log_mutex);
2870 list_del_init(&ctx->list);
2871 mutex_unlock(&root->log_mutex);
2875 * Invoked in log mutex context, or be sure there is no other task which
2876 * can access the list.
2878 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2879 int index, int error)
2881 struct btrfs_log_ctx *ctx;
2882 struct btrfs_log_ctx *safe;
2884 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2885 list_del_init(&ctx->list);
2886 ctx->log_ret = error;
2891 * btrfs_sync_log does sends a given tree log down to the disk and
2892 * updates the super blocks to record it. When this call is done,
2893 * you know that any inodes previously logged are safely on disk only
2896 * Any other return value means you need to call btrfs_commit_transaction.
2897 * Some of the edge cases for fsyncing directories that have had unlinks
2898 * or renames done in the past mean that sometimes the only safe
2899 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2900 * that has happened.
2902 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2903 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2909 struct btrfs_fs_info *fs_info = root->fs_info;
2910 struct btrfs_root *log = root->log_root;
2911 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2912 struct btrfs_root_item new_root_item;
2913 int log_transid = 0;
2914 struct btrfs_log_ctx root_log_ctx;
2915 struct blk_plug plug;
2919 mutex_lock(&root->log_mutex);
2920 log_transid = ctx->log_transid;
2921 if (root->log_transid_committed >= log_transid) {
2922 mutex_unlock(&root->log_mutex);
2923 return ctx->log_ret;
2926 index1 = log_transid % 2;
2927 if (atomic_read(&root->log_commit[index1])) {
2928 wait_log_commit(root, log_transid);
2929 mutex_unlock(&root->log_mutex);
2930 return ctx->log_ret;
2932 ASSERT(log_transid == root->log_transid);
2933 atomic_set(&root->log_commit[index1], 1);
2935 /* wait for previous tree log sync to complete */
2936 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2937 wait_log_commit(root, log_transid - 1);
2940 int batch = atomic_read(&root->log_batch);
2941 /* when we're on an ssd, just kick the log commit out */
2942 if (!btrfs_test_opt(fs_info, SSD) &&
2943 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2944 mutex_unlock(&root->log_mutex);
2945 schedule_timeout_uninterruptible(1);
2946 mutex_lock(&root->log_mutex);
2948 wait_for_writer(root);
2949 if (batch == atomic_read(&root->log_batch))
2953 /* bail out if we need to do a full commit */
2954 if (btrfs_need_log_full_commit(trans)) {
2955 ret = BTRFS_LOG_FORCE_COMMIT;
2956 mutex_unlock(&root->log_mutex);
2960 if (log_transid % 2 == 0)
2961 mark = EXTENT_DIRTY;
2965 /* we start IO on all the marked extents here, but we don't actually
2966 * wait for them until later.
2968 blk_start_plug(&plug);
2969 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2971 * -EAGAIN happens when someone, e.g., a concurrent transaction
2972 * commit, writes a dirty extent in this tree-log commit. This
2973 * concurrent write will create a hole writing out the extents,
2974 * and we cannot proceed on a zoned filesystem, requiring
2975 * sequential writing. While we can bail out to a full commit
2976 * here, but we can continue hoping the concurrent writing fills
2979 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
2982 blk_finish_plug(&plug);
2983 btrfs_set_log_full_commit(trans);
2984 mutex_unlock(&root->log_mutex);
2989 * We _must_ update under the root->log_mutex in order to make sure we
2990 * have a consistent view of the log root we are trying to commit at
2993 * We _must_ copy this into a local copy, because we are not holding the
2994 * log_root_tree->log_mutex yet. This is important because when we
2995 * commit the log_root_tree we must have a consistent view of the
2996 * log_root_tree when we update the super block to point at the
2997 * log_root_tree bytenr. If we update the log_root_tree here we'll race
2998 * with the commit and possibly point at the new block which we may not
3001 btrfs_set_root_node(&log->root_item, log->node);
3002 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3004 root->log_transid++;
3005 log->log_transid = root->log_transid;
3006 root->log_start_pid = 0;
3008 * IO has been started, blocks of the log tree have WRITTEN flag set
3009 * in their headers. new modifications of the log will be written to
3010 * new positions. so it's safe to allow log writers to go in.
3012 mutex_unlock(&root->log_mutex);
3014 if (btrfs_is_zoned(fs_info)) {
3015 mutex_lock(&fs_info->tree_root->log_mutex);
3016 if (!log_root_tree->node) {
3017 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3019 mutex_unlock(&fs_info->tree_root->log_mutex);
3020 blk_finish_plug(&plug);
3024 mutex_unlock(&fs_info->tree_root->log_mutex);
3027 btrfs_init_log_ctx(&root_log_ctx, NULL);
3029 mutex_lock(&log_root_tree->log_mutex);
3031 index2 = log_root_tree->log_transid % 2;
3032 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3033 root_log_ctx.log_transid = log_root_tree->log_transid;
3036 * Now we are safe to update the log_root_tree because we're under the
3037 * log_mutex, and we're a current writer so we're holding the commit
3038 * open until we drop the log_mutex.
3040 ret = update_log_root(trans, log, &new_root_item);
3042 if (!list_empty(&root_log_ctx.list))
3043 list_del_init(&root_log_ctx.list);
3045 blk_finish_plug(&plug);
3046 btrfs_set_log_full_commit(trans);
3049 "failed to update log for root %llu ret %d",
3050 root->root_key.objectid, ret);
3051 btrfs_wait_tree_log_extents(log, mark);
3052 mutex_unlock(&log_root_tree->log_mutex);
3056 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3057 blk_finish_plug(&plug);
3058 list_del_init(&root_log_ctx.list);
3059 mutex_unlock(&log_root_tree->log_mutex);
3060 ret = root_log_ctx.log_ret;
3064 index2 = root_log_ctx.log_transid % 2;
3065 if (atomic_read(&log_root_tree->log_commit[index2])) {
3066 blk_finish_plug(&plug);
3067 ret = btrfs_wait_tree_log_extents(log, mark);
3068 wait_log_commit(log_root_tree,
3069 root_log_ctx.log_transid);
3070 mutex_unlock(&log_root_tree->log_mutex);
3072 ret = root_log_ctx.log_ret;
3075 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3076 atomic_set(&log_root_tree->log_commit[index2], 1);
3078 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3079 wait_log_commit(log_root_tree,
3080 root_log_ctx.log_transid - 1);
3084 * now that we've moved on to the tree of log tree roots,
3085 * check the full commit flag again
3087 if (btrfs_need_log_full_commit(trans)) {
3088 blk_finish_plug(&plug);
3089 btrfs_wait_tree_log_extents(log, mark);
3090 mutex_unlock(&log_root_tree->log_mutex);
3091 ret = BTRFS_LOG_FORCE_COMMIT;
3092 goto out_wake_log_root;
3095 ret = btrfs_write_marked_extents(fs_info,
3096 &log_root_tree->dirty_log_pages,
3097 EXTENT_DIRTY | EXTENT_NEW);
3098 blk_finish_plug(&plug);
3100 * As described above, -EAGAIN indicates a hole in the extents. We
3101 * cannot wait for these write outs since the waiting cause a
3102 * deadlock. Bail out to the full commit instead.
3104 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3105 btrfs_set_log_full_commit(trans);
3106 btrfs_wait_tree_log_extents(log, mark);
3107 mutex_unlock(&log_root_tree->log_mutex);
3108 goto out_wake_log_root;
3110 btrfs_set_log_full_commit(trans);
3111 mutex_unlock(&log_root_tree->log_mutex);
3112 goto out_wake_log_root;
3114 ret = btrfs_wait_tree_log_extents(log, mark);
3116 ret = btrfs_wait_tree_log_extents(log_root_tree,
3117 EXTENT_NEW | EXTENT_DIRTY);
3119 btrfs_set_log_full_commit(trans);
3120 mutex_unlock(&log_root_tree->log_mutex);
3121 goto out_wake_log_root;
3124 log_root_start = log_root_tree->node->start;
3125 log_root_level = btrfs_header_level(log_root_tree->node);
3126 log_root_tree->log_transid++;
3127 mutex_unlock(&log_root_tree->log_mutex);
3130 * Here we are guaranteed that nobody is going to write the superblock
3131 * for the current transaction before us and that neither we do write
3132 * our superblock before the previous transaction finishes its commit
3133 * and writes its superblock, because:
3135 * 1) We are holding a handle on the current transaction, so no body
3136 * can commit it until we release the handle;
3138 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3139 * if the previous transaction is still committing, and hasn't yet
3140 * written its superblock, we wait for it to do it, because a
3141 * transaction commit acquires the tree_log_mutex when the commit
3142 * begins and releases it only after writing its superblock.
3144 mutex_lock(&fs_info->tree_log_mutex);
3147 * The previous transaction writeout phase could have failed, and thus
3148 * marked the fs in an error state. We must not commit here, as we
3149 * could have updated our generation in the super_for_commit and
3150 * writing the super here would result in transid mismatches. If there
3151 * is an error here just bail.
3153 if (BTRFS_FS_ERROR(fs_info)) {
3155 btrfs_set_log_full_commit(trans);
3156 btrfs_abort_transaction(trans, ret);
3157 mutex_unlock(&fs_info->tree_log_mutex);
3158 goto out_wake_log_root;
3161 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3162 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3163 ret = write_all_supers(fs_info, 1);
3164 mutex_unlock(&fs_info->tree_log_mutex);
3166 btrfs_set_log_full_commit(trans);
3167 btrfs_abort_transaction(trans, ret);
3168 goto out_wake_log_root;
3172 * We know there can only be one task here, since we have not yet set
3173 * root->log_commit[index1] to 0 and any task attempting to sync the
3174 * log must wait for the previous log transaction to commit if it's
3175 * still in progress or wait for the current log transaction commit if
3176 * someone else already started it. We use <= and not < because the
3177 * first log transaction has an ID of 0.
3179 ASSERT(root->last_log_commit <= log_transid);
3180 root->last_log_commit = log_transid;
3183 mutex_lock(&log_root_tree->log_mutex);
3184 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3186 log_root_tree->log_transid_committed++;
3187 atomic_set(&log_root_tree->log_commit[index2], 0);
3188 mutex_unlock(&log_root_tree->log_mutex);
3191 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3192 * all the updates above are seen by the woken threads. It might not be
3193 * necessary, but proving that seems to be hard.
3195 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3197 mutex_lock(&root->log_mutex);
3198 btrfs_remove_all_log_ctxs(root, index1, ret);
3199 root->log_transid_committed++;
3200 atomic_set(&root->log_commit[index1], 0);
3201 mutex_unlock(&root->log_mutex);
3204 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3205 * all the updates above are seen by the woken threads. It might not be
3206 * necessary, but proving that seems to be hard.
3208 cond_wake_up(&root->log_commit_wait[index1]);
3212 static void free_log_tree(struct btrfs_trans_handle *trans,
3213 struct btrfs_root *log)
3216 struct walk_control wc = {
3218 .process_func = process_one_buffer
3222 ret = walk_log_tree(trans, log, &wc);
3225 * We weren't able to traverse the entire log tree, the
3226 * typical scenario is getting an -EIO when reading an
3227 * extent buffer of the tree, due to a previous writeback
3230 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3231 &log->fs_info->fs_state);
3234 * Some extent buffers of the log tree may still be dirty
3235 * and not yet written back to storage, because we may
3236 * have updates to a log tree without syncing a log tree,
3237 * such as during rename and link operations. So flush
3238 * them out and wait for their writeback to complete, so
3239 * that we properly cleanup their state and pages.
3241 btrfs_write_marked_extents(log->fs_info,
3242 &log->dirty_log_pages,
3243 EXTENT_DIRTY | EXTENT_NEW);
3244 btrfs_wait_tree_log_extents(log,
3245 EXTENT_DIRTY | EXTENT_NEW);
3248 btrfs_abort_transaction(trans, ret);
3250 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3254 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3255 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3256 extent_io_tree_release(&log->log_csum_range);
3258 btrfs_put_root(log);
3262 * free all the extents used by the tree log. This should be called
3263 * at commit time of the full transaction
3265 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3267 if (root->log_root) {
3268 free_log_tree(trans, root->log_root);
3269 root->log_root = NULL;
3270 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3275 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3276 struct btrfs_fs_info *fs_info)
3278 if (fs_info->log_root_tree) {
3279 free_log_tree(trans, fs_info->log_root_tree);
3280 fs_info->log_root_tree = NULL;
3281 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3287 * Check if an inode was logged in the current transaction. This correctly deals
3288 * with the case where the inode was logged but has a logged_trans of 0, which
3289 * happens if the inode is evicted and loaded again, as logged_trans is an in
3290 * memory only field (not persisted).
3292 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3295 static int inode_logged(struct btrfs_trans_handle *trans,
3296 struct btrfs_inode *inode,
3297 struct btrfs_path *path_in)
3299 struct btrfs_path *path = path_in;
3300 struct btrfs_key key;
3303 if (inode->logged_trans == trans->transid)
3307 * If logged_trans is not 0, then we know the inode logged was not logged
3308 * in this transaction, so we can return false right away.
3310 if (inode->logged_trans > 0)
3314 * If no log tree was created for this root in this transaction, then
3315 * the inode can not have been logged in this transaction. In that case
3316 * set logged_trans to anything greater than 0 and less than the current
3317 * transaction's ID, to avoid the search below in a future call in case
3318 * a log tree gets created after this.
3320 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3321 inode->logged_trans = trans->transid - 1;
3326 * We have a log tree and the inode's logged_trans is 0. We can't tell
3327 * for sure if the inode was logged before in this transaction by looking
3328 * only at logged_trans. We could be pessimistic and assume it was, but
3329 * that can lead to unnecessarily logging an inode during rename and link
3330 * operations, and then further updating the log in followup rename and
3331 * link operations, specially if it's a directory, which adds latency
3332 * visible to applications doing a series of rename or link operations.
3334 * A logged_trans of 0 here can mean several things:
3336 * 1) The inode was never logged since the filesystem was mounted, and may
3337 * or may have not been evicted and loaded again;
3339 * 2) The inode was logged in a previous transaction, then evicted and
3340 * then loaded again;
3342 * 3) The inode was logged in the current transaction, then evicted and
3343 * then loaded again.
3345 * For cases 1) and 2) we don't want to return true, but we need to detect
3346 * case 3) and return true. So we do a search in the log root for the inode
3349 key.objectid = btrfs_ino(inode);
3350 key.type = BTRFS_INODE_ITEM_KEY;
3354 path = btrfs_alloc_path();
3359 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3362 btrfs_release_path(path);
3364 btrfs_free_path(path);
3367 * Logging an inode always results in logging its inode item. So if we
3368 * did not find the item we know the inode was not logged for sure.
3372 } else if (ret > 0) {
3374 * Set logged_trans to a value greater than 0 and less then the
3375 * current transaction to avoid doing the search in future calls.
3377 inode->logged_trans = trans->transid - 1;
3382 * The inode was previously logged and then evicted, set logged_trans to
3383 * the current transacion's ID, to avoid future tree searches as long as
3384 * the inode is not evicted again.
3386 inode->logged_trans = trans->transid;
3389 * If it's a directory, then we must set last_dir_index_offset to the
3390 * maximum possible value, so that the next attempt to log the inode does
3391 * not skip checking if dir index keys found in modified subvolume tree
3392 * leaves have been logged before, otherwise it would result in attempts
3393 * to insert duplicate dir index keys in the log tree. This must be done
3394 * because last_dir_index_offset is an in-memory only field, not persisted
3395 * in the inode item or any other on-disk structure, so its value is lost
3396 * once the inode is evicted.
3398 if (S_ISDIR(inode->vfs_inode.i_mode))
3399 inode->last_dir_index_offset = (u64)-1;
3405 * Delete a directory entry from the log if it exists.
3407 * Returns < 0 on error
3408 * 1 if the entry does not exists
3409 * 0 if the entry existed and was successfully deleted
3411 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3412 struct btrfs_root *log,
3413 struct btrfs_path *path,
3415 const struct fscrypt_str *name,
3418 struct btrfs_dir_item *di;
3421 * We only log dir index items of a directory, so we don't need to look
3422 * for dir item keys.
3424 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3432 * We do not need to update the size field of the directory's
3433 * inode item because on log replay we update the field to reflect
3434 * all existing entries in the directory (see overwrite_item()).
3436 return btrfs_delete_one_dir_name(trans, log, path, di);
3440 * If both a file and directory are logged, and unlinks or renames are
3441 * mixed in, we have a few interesting corners:
3443 * create file X in dir Y
3444 * link file X to X.link in dir Y
3446 * unlink file X but leave X.link
3449 * After a crash we would expect only X.link to exist. But file X
3450 * didn't get fsync'd again so the log has back refs for X and X.link.
3452 * We solve this by removing directory entries and inode backrefs from the
3453 * log when a file that was logged in the current transaction is
3454 * unlinked. Any later fsync will include the updated log entries, and
3455 * we'll be able to reconstruct the proper directory items from backrefs.
3457 * This optimizations allows us to avoid relogging the entire inode
3458 * or the entire directory.
3460 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3461 struct btrfs_root *root,
3462 const struct fscrypt_str *name,
3463 struct btrfs_inode *dir, u64 index)
3465 struct btrfs_path *path;
3468 ret = inode_logged(trans, dir, NULL);
3472 btrfs_set_log_full_commit(trans);
3476 ret = join_running_log_trans(root);
3480 mutex_lock(&dir->log_mutex);
3482 path = btrfs_alloc_path();
3488 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3490 btrfs_free_path(path);
3492 mutex_unlock(&dir->log_mutex);
3494 btrfs_set_log_full_commit(trans);
3495 btrfs_end_log_trans(root);
3498 /* see comments for btrfs_del_dir_entries_in_log */
3499 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3500 struct btrfs_root *root,
3501 const struct fscrypt_str *name,
3502 struct btrfs_inode *inode, u64 dirid)
3504 struct btrfs_root *log;
3508 ret = inode_logged(trans, inode, NULL);
3512 btrfs_set_log_full_commit(trans);
3516 ret = join_running_log_trans(root);
3519 log = root->log_root;
3520 mutex_lock(&inode->log_mutex);
3522 ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode),
3524 mutex_unlock(&inode->log_mutex);
3525 if (ret < 0 && ret != -ENOENT)
3526 btrfs_set_log_full_commit(trans);
3527 btrfs_end_log_trans(root);
3531 * creates a range item in the log for 'dirid'. first_offset and
3532 * last_offset tell us which parts of the key space the log should
3533 * be considered authoritative for.
3535 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3536 struct btrfs_root *log,
3537 struct btrfs_path *path,
3539 u64 first_offset, u64 last_offset)
3542 struct btrfs_key key;
3543 struct btrfs_dir_log_item *item;
3545 key.objectid = dirid;
3546 key.offset = first_offset;
3547 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3548 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3550 * -EEXIST is fine and can happen sporadically when we are logging a
3551 * directory and have concurrent insertions in the subvolume's tree for
3552 * items from other inodes and that result in pushing off some dir items
3553 * from one leaf to another in order to accommodate for the new items.
3554 * This results in logging the same dir index range key.
3556 if (ret && ret != -EEXIST)
3559 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3560 struct btrfs_dir_log_item);
3561 if (ret == -EEXIST) {
3562 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3565 * btrfs_del_dir_entries_in_log() might have been called during
3566 * an unlink between the initial insertion of this key and the
3567 * current update, or we might be logging a single entry deletion
3568 * during a rename, so set the new last_offset to the max value.
3570 last_offset = max(last_offset, curr_end);
3572 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3573 btrfs_mark_buffer_dirty(path->nodes[0]);
3574 btrfs_release_path(path);
3578 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3579 struct btrfs_root *log,
3580 struct extent_buffer *src,
3581 struct btrfs_path *dst_path,
3585 char *ins_data = NULL;
3586 struct btrfs_item_batch batch;
3587 struct extent_buffer *dst;
3588 unsigned long src_offset;
3589 unsigned long dst_offset;
3590 struct btrfs_key key;
3599 btrfs_item_key_to_cpu(src, &key, start_slot);
3600 item_size = btrfs_item_size(src, start_slot);
3602 batch.data_sizes = &item_size;
3603 batch.total_data_size = item_size;
3605 struct btrfs_key *ins_keys;
3608 ins_data = kmalloc(count * sizeof(u32) +
3609 count * sizeof(struct btrfs_key), GFP_NOFS);
3613 ins_sizes = (u32 *)ins_data;
3614 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3615 batch.keys = ins_keys;
3616 batch.data_sizes = ins_sizes;
3617 batch.total_data_size = 0;
3619 for (i = 0; i < count; i++) {
3620 const int slot = start_slot + i;
3622 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3623 ins_sizes[i] = btrfs_item_size(src, slot);
3624 batch.total_data_size += ins_sizes[i];
3628 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3632 dst = dst_path->nodes[0];
3634 * Copy all the items in bulk, in a single copy operation. Item data is
3635 * organized such that it's placed at the end of a leaf and from right
3636 * to left. For example, the data for the second item ends at an offset
3637 * that matches the offset where the data for the first item starts, the
3638 * data for the third item ends at an offset that matches the offset
3639 * where the data of the second items starts, and so on.
3640 * Therefore our source and destination start offsets for copy match the
3641 * offsets of the last items (highest slots).
3643 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3644 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3645 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3646 btrfs_release_path(dst_path);
3653 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3654 struct btrfs_inode *inode,
3655 struct btrfs_path *path,
3656 struct btrfs_path *dst_path,
3657 struct btrfs_log_ctx *ctx,
3658 u64 *last_old_dentry_offset)
3660 struct btrfs_root *log = inode->root->log_root;
3661 struct extent_buffer *src;
3662 const int nritems = btrfs_header_nritems(path->nodes[0]);
3663 const u64 ino = btrfs_ino(inode);
3664 bool last_found = false;
3665 int batch_start = 0;
3670 * We need to clone the leaf, release the read lock on it, and use the
3671 * clone before modifying the log tree. See the comment at copy_items()
3672 * about why we need to do this.
3674 src = btrfs_clone_extent_buffer(path->nodes[0]);
3679 btrfs_release_path(path);
3680 path->nodes[0] = src;
3683 for (; i < nritems; i++) {
3684 struct btrfs_dir_item *di;
3685 struct btrfs_key key;
3688 btrfs_item_key_to_cpu(src, &key, i);
3690 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3695 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3696 ctx->last_dir_item_offset = key.offset;
3699 * Skip ranges of items that consist only of dir item keys created
3700 * in past transactions. However if we find a gap, we must log a
3701 * dir index range item for that gap, so that index keys in that
3702 * gap are deleted during log replay.
3704 if (btrfs_dir_transid(src, di) < trans->transid) {
3705 if (key.offset > *last_old_dentry_offset + 1) {
3706 ret = insert_dir_log_key(trans, log, dst_path,
3707 ino, *last_old_dentry_offset + 1,
3713 *last_old_dentry_offset = key.offset;
3717 /* If we logged this dir index item before, we can skip it. */
3718 if (key.offset <= inode->last_dir_index_offset)
3722 * We must make sure that when we log a directory entry, the
3723 * corresponding inode, after log replay, has a matching link
3724 * count. For example:
3730 * xfs_io -c "fsync" mydir
3732 * <mount fs and log replay>
3734 * Would result in a fsync log that when replayed, our file inode
3735 * would have a link count of 1, but we get two directory entries
3736 * pointing to the same inode. After removing one of the names,
3737 * it would not be possible to remove the other name, which
3738 * resulted always in stale file handle errors, and would not be
3739 * possible to rmdir the parent directory, since its i_size could
3740 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3741 * resulting in -ENOTEMPTY errors.
3743 if (!ctx->log_new_dentries) {
3744 struct btrfs_key di_key;
3746 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3747 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3748 ctx->log_new_dentries = true;
3751 if (batch_size == 0)
3756 if (batch_size > 0) {
3759 ret = flush_dir_items_batch(trans, log, src, dst_path,
3760 batch_start, batch_size);
3765 return last_found ? 1 : 0;
3769 * log all the items included in the current transaction for a given
3770 * directory. This also creates the range items in the log tree required
3771 * to replay anything deleted before the fsync
3773 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3774 struct btrfs_inode *inode,
3775 struct btrfs_path *path,
3776 struct btrfs_path *dst_path,
3777 struct btrfs_log_ctx *ctx,
3778 u64 min_offset, u64 *last_offset_ret)
3780 struct btrfs_key min_key;
3781 struct btrfs_root *root = inode->root;
3782 struct btrfs_root *log = root->log_root;
3785 u64 last_old_dentry_offset = min_offset - 1;
3786 u64 last_offset = (u64)-1;
3787 u64 ino = btrfs_ino(inode);
3789 min_key.objectid = ino;
3790 min_key.type = BTRFS_DIR_INDEX_KEY;
3791 min_key.offset = min_offset;
3793 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3796 * we didn't find anything from this transaction, see if there
3797 * is anything at all
3799 if (ret != 0 || min_key.objectid != ino ||
3800 min_key.type != BTRFS_DIR_INDEX_KEY) {
3801 min_key.objectid = ino;
3802 min_key.type = BTRFS_DIR_INDEX_KEY;
3803 min_key.offset = (u64)-1;
3804 btrfs_release_path(path);
3805 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3807 btrfs_release_path(path);
3810 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3812 /* if ret == 0 there are items for this type,
3813 * create a range to tell us the last key of this type.
3814 * otherwise, there are no items in this directory after
3815 * *min_offset, and we create a range to indicate that.
3818 struct btrfs_key tmp;
3820 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3822 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3823 last_old_dentry_offset = tmp.offset;
3824 } else if (ret < 0) {
3831 /* go backward to find any previous key */
3832 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3834 struct btrfs_key tmp;
3836 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3838 * The dir index key before the first one we found that needs to
3839 * be logged might be in a previous leaf, and there might be a
3840 * gap between these keys, meaning that we had deletions that
3841 * happened. So the key range item we log (key type
3842 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3843 * previous key's offset plus 1, so that those deletes are replayed.
3845 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3846 last_old_dentry_offset = tmp.offset;
3847 } else if (ret < 0) {
3852 btrfs_release_path(path);
3855 * Find the first key from this transaction again or the one we were at
3856 * in the loop below in case we had to reschedule. We may be logging the
3857 * directory without holding its VFS lock, which happen when logging new
3858 * dentries (through log_new_dir_dentries()) or in some cases when we
3859 * need to log the parent directory of an inode. This means a dir index
3860 * key might be deleted from the inode's root, and therefore we may not
3861 * find it anymore. If we can't find it, just move to the next key. We
3862 * can not bail out and ignore, because if we do that we will simply
3863 * not log dir index keys that come after the one that was just deleted
3864 * and we can end up logging a dir index range that ends at (u64)-1
3865 * (@last_offset is initialized to that), resulting in removing dir
3866 * entries we should not remove at log replay time.
3869 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3871 ret = btrfs_next_item(root, path);
3874 /* If ret is 1, there are no more keys in the inode's root. */
3879 * we have a block from this transaction, log every item in it
3880 * from our directory
3883 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3884 &last_old_dentry_offset);
3890 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3893 * look ahead to the next item and see if it is also
3894 * from this directory and from this transaction
3896 ret = btrfs_next_leaf(root, path);
3899 last_offset = (u64)-1;
3904 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3905 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3906 last_offset = (u64)-1;
3909 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3911 * The next leaf was not changed in the current transaction
3912 * and has at least one dir index key.
3913 * We check for the next key because there might have been
3914 * one or more deletions between the last key we logged and
3915 * that next key. So the key range item we log (key type
3916 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3917 * offset minus 1, so that those deletes are replayed.
3919 last_offset = min_key.offset - 1;
3922 if (need_resched()) {
3923 btrfs_release_path(path);
3929 btrfs_release_path(path);
3930 btrfs_release_path(dst_path);
3933 *last_offset_ret = last_offset;
3935 * In case the leaf was changed in the current transaction but
3936 * all its dir items are from a past transaction, the last item
3937 * in the leaf is a dir item and there's no gap between that last
3938 * dir item and the first one on the next leaf (which did not
3939 * change in the current transaction), then we don't need to log
3940 * a range, last_old_dentry_offset is == to last_offset.
3942 ASSERT(last_old_dentry_offset <= last_offset);
3943 if (last_old_dentry_offset < last_offset) {
3944 ret = insert_dir_log_key(trans, log, path, ino,
3945 last_old_dentry_offset + 1,
3955 * If the inode was logged before and it was evicted, then its
3956 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3957 * key offset. If that's the case, search for it and update the inode. This
3958 * is to avoid lookups in the log tree every time we try to insert a dir index
3959 * key from a leaf changed in the current transaction, and to allow us to always
3960 * do batch insertions of dir index keys.
3962 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3963 struct btrfs_path *path,
3964 const struct btrfs_log_ctx *ctx)
3966 const u64 ino = btrfs_ino(inode);
3967 struct btrfs_key key;
3970 lockdep_assert_held(&inode->log_mutex);
3972 if (inode->last_dir_index_offset != (u64)-1)
3975 if (!ctx->logged_before) {
3976 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3981 key.type = BTRFS_DIR_INDEX_KEY;
3982 key.offset = (u64)-1;
3984 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3986 * An error happened or we actually have an index key with an offset
3987 * value of (u64)-1. Bail out, we're done.
3993 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3996 * No dir index items, bail out and leave last_dir_index_offset with
3997 * the value right before the first valid index value.
3999 if (path->slots[0] == 0)
4003 * btrfs_search_slot() left us at one slot beyond the slot with the last
4004 * index key, or beyond the last key of the directory that is not an
4005 * index key. If we have an index key before, set last_dir_index_offset
4006 * to its offset value, otherwise leave it with a value right before the
4007 * first valid index value, as it means we have an empty directory.
4009 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4010 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
4011 inode->last_dir_index_offset = key.offset;
4014 btrfs_release_path(path);
4020 * logging directories is very similar to logging inodes, We find all the items
4021 * from the current transaction and write them to the log.
4023 * The recovery code scans the directory in the subvolume, and if it finds a
4024 * key in the range logged that is not present in the log tree, then it means
4025 * that dir entry was unlinked during the transaction.
4027 * In order for that scan to work, we must include one key smaller than
4028 * the smallest logged by this transaction and one key larger than the largest
4029 * key logged by this transaction.
4031 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4032 struct btrfs_inode *inode,
4033 struct btrfs_path *path,
4034 struct btrfs_path *dst_path,
4035 struct btrfs_log_ctx *ctx)
4041 ret = update_last_dir_index_offset(inode, path, ctx);
4045 min_key = BTRFS_DIR_START_INDEX;
4047 ctx->last_dir_item_offset = inode->last_dir_index_offset;
4050 ret = log_dir_items(trans, inode, path, dst_path,
4051 ctx, min_key, &max_key);
4054 if (max_key == (u64)-1)
4056 min_key = max_key + 1;
4059 inode->last_dir_index_offset = ctx->last_dir_item_offset;
4065 * a helper function to drop items from the log before we relog an
4066 * inode. max_key_type indicates the highest item type to remove.
4067 * This cannot be run for file data extents because it does not
4068 * free the extents they point to.
4070 static int drop_inode_items(struct btrfs_trans_handle *trans,
4071 struct btrfs_root *log,
4072 struct btrfs_path *path,
4073 struct btrfs_inode *inode,
4077 struct btrfs_key key;
4078 struct btrfs_key found_key;
4081 key.objectid = btrfs_ino(inode);
4082 key.type = max_key_type;
4083 key.offset = (u64)-1;
4086 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4087 BUG_ON(ret == 0); /* Logic error */
4091 if (path->slots[0] == 0)
4095 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4098 if (found_key.objectid != key.objectid)
4101 found_key.offset = 0;
4103 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
4107 ret = btrfs_del_items(trans, log, path, start_slot,
4108 path->slots[0] - start_slot + 1);
4110 * If start slot isn't 0 then we don't need to re-search, we've
4111 * found the last guy with the objectid in this tree.
4113 if (ret || start_slot != 0)
4115 btrfs_release_path(path);
4117 btrfs_release_path(path);
4123 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4124 struct btrfs_root *log_root,
4125 struct btrfs_inode *inode,
4126 u64 new_size, u32 min_type)
4128 struct btrfs_truncate_control control = {
4129 .new_size = new_size,
4130 .ino = btrfs_ino(inode),
4131 .min_type = min_type,
4132 .skip_ref_updates = true,
4135 return btrfs_truncate_inode_items(trans, log_root, &control);
4138 static void fill_inode_item(struct btrfs_trans_handle *trans,
4139 struct extent_buffer *leaf,
4140 struct btrfs_inode_item *item,
4141 struct inode *inode, int log_inode_only,
4144 struct btrfs_map_token token;
4147 btrfs_init_map_token(&token, leaf);
4149 if (log_inode_only) {
4150 /* set the generation to zero so the recover code
4151 * can tell the difference between an logging
4152 * just to say 'this inode exists' and a logging
4153 * to say 'update this inode with these values'
4155 btrfs_set_token_inode_generation(&token, item, 0);
4156 btrfs_set_token_inode_size(&token, item, logged_isize);
4158 btrfs_set_token_inode_generation(&token, item,
4159 BTRFS_I(inode)->generation);
4160 btrfs_set_token_inode_size(&token, item, inode->i_size);
4163 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4164 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4165 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4166 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4168 btrfs_set_token_timespec_sec(&token, &item->atime,
4169 inode->i_atime.tv_sec);
4170 btrfs_set_token_timespec_nsec(&token, &item->atime,
4171 inode->i_atime.tv_nsec);
4173 btrfs_set_token_timespec_sec(&token, &item->mtime,
4174 inode->i_mtime.tv_sec);
4175 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4176 inode->i_mtime.tv_nsec);
4178 btrfs_set_token_timespec_sec(&token, &item->ctime,
4179 inode->i_ctime.tv_sec);
4180 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4181 inode->i_ctime.tv_nsec);
4184 * We do not need to set the nbytes field, in fact during a fast fsync
4185 * its value may not even be correct, since a fast fsync does not wait
4186 * for ordered extent completion, which is where we update nbytes, it
4187 * only waits for writeback to complete. During log replay as we find
4188 * file extent items and replay them, we adjust the nbytes field of the
4189 * inode item in subvolume tree as needed (see overwrite_item()).
4192 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4193 btrfs_set_token_inode_transid(&token, item, trans->transid);
4194 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4195 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4196 BTRFS_I(inode)->ro_flags);
4197 btrfs_set_token_inode_flags(&token, item, flags);
4198 btrfs_set_token_inode_block_group(&token, item, 0);
4201 static int log_inode_item(struct btrfs_trans_handle *trans,
4202 struct btrfs_root *log, struct btrfs_path *path,
4203 struct btrfs_inode *inode, bool inode_item_dropped)
4205 struct btrfs_inode_item *inode_item;
4209 * If we are doing a fast fsync and the inode was logged before in the
4210 * current transaction, then we know the inode was previously logged and
4211 * it exists in the log tree. For performance reasons, in this case use
4212 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4213 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4214 * contention in case there are concurrent fsyncs for other inodes of the
4215 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4216 * already exists can also result in unnecessarily splitting a leaf.
4218 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4219 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4225 * This means it is the first fsync in the current transaction,
4226 * so the inode item is not in the log and we need to insert it.
4227 * We can never get -EEXIST because we are only called for a fast
4228 * fsync and in case an inode eviction happens after the inode was
4229 * logged before in the current transaction, when we load again
4230 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4231 * flags and set ->logged_trans to 0.
4233 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4234 sizeof(*inode_item));
4235 ASSERT(ret != -EEXIST);
4239 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4240 struct btrfs_inode_item);
4241 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4243 btrfs_release_path(path);
4247 static int log_csums(struct btrfs_trans_handle *trans,
4248 struct btrfs_inode *inode,
4249 struct btrfs_root *log_root,
4250 struct btrfs_ordered_sum *sums)
4252 const u64 lock_end = sums->bytenr + sums->len - 1;
4253 struct extent_state *cached_state = NULL;
4257 * If this inode was not used for reflink operations in the current
4258 * transaction with new extents, then do the fast path, no need to
4259 * worry about logging checksum items with overlapping ranges.
4261 if (inode->last_reflink_trans < trans->transid)
4262 return btrfs_csum_file_blocks(trans, log_root, sums);
4265 * Serialize logging for checksums. This is to avoid racing with the
4266 * same checksum being logged by another task that is logging another
4267 * file which happens to refer to the same extent as well. Such races
4268 * can leave checksum items in the log with overlapping ranges.
4270 ret = lock_extent(&log_root->log_csum_range, sums->bytenr, lock_end,
4275 * Due to extent cloning, we might have logged a csum item that covers a
4276 * subrange of a cloned extent, and later we can end up logging a csum
4277 * item for a larger subrange of the same extent or the entire range.
4278 * This would leave csum items in the log tree that cover the same range
4279 * and break the searches for checksums in the log tree, resulting in
4280 * some checksums missing in the fs/subvolume tree. So just delete (or
4281 * trim and adjust) any existing csum items in the log for this range.
4283 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4285 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4287 unlock_extent(&log_root->log_csum_range, sums->bytenr, lock_end,
4293 static noinline int copy_items(struct btrfs_trans_handle *trans,
4294 struct btrfs_inode *inode,
4295 struct btrfs_path *dst_path,
4296 struct btrfs_path *src_path,
4297 int start_slot, int nr, int inode_only,
4300 struct btrfs_root *log = inode->root->log_root;
4301 struct btrfs_file_extent_item *extent;
4302 struct extent_buffer *src;
4304 struct btrfs_key *ins_keys;
4306 struct btrfs_item_batch batch;
4310 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4311 const u64 i_size = i_size_read(&inode->vfs_inode);
4314 * To keep lockdep happy and avoid deadlocks, clone the source leaf and
4315 * use the clone. This is because otherwise we would be changing the log
4316 * tree, to insert items from the subvolume tree or insert csum items,
4317 * while holding a read lock on a leaf from the subvolume tree, which
4318 * creates a nasty lock dependency when COWing log tree nodes/leaves:
4320 * 1) Modifying the log tree triggers an extent buffer allocation while
4321 * holding a write lock on a parent extent buffer from the log tree.
4322 * Allocating the pages for an extent buffer, or the extent buffer
4323 * struct, can trigger inode eviction and finally the inode eviction
4324 * will trigger a release/remove of a delayed node, which requires
4325 * taking the delayed node's mutex;
4327 * 2) Allocating a metadata extent for a log tree can trigger the async
4328 * reclaim thread and make us wait for it to release enough space and
4329 * unblock our reservation ticket. The reclaim thread can start
4330 * flushing delayed items, and that in turn results in the need to
4331 * lock delayed node mutexes and in the need to write lock extent
4332 * buffers of a subvolume tree - all this while holding a write lock
4333 * on the parent extent buffer in the log tree.
4335 * So one task in scenario 1) running in parallel with another task in
4336 * scenario 2) could lead to a deadlock, one wanting to lock a delayed
4337 * node mutex while having a read lock on a leaf from the subvolume,
4338 * while the other is holding the delayed node's mutex and wants to
4339 * write lock the same subvolume leaf for flushing delayed items.
4341 src = btrfs_clone_extent_buffer(src_path->nodes[0]);
4345 i = src_path->slots[0];
4346 btrfs_release_path(src_path);
4347 src_path->nodes[0] = src;
4348 src_path->slots[0] = i;
4350 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4351 nr * sizeof(u32), GFP_NOFS);
4355 ins_sizes = (u32 *)ins_data;
4356 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4357 batch.keys = ins_keys;
4358 batch.data_sizes = ins_sizes;
4359 batch.total_data_size = 0;
4363 for (i = 0; i < nr; i++) {
4364 const int src_slot = start_slot + i;
4365 struct btrfs_root *csum_root;
4366 struct btrfs_ordered_sum *sums;
4367 struct btrfs_ordered_sum *sums_next;
4368 LIST_HEAD(ordered_sums);
4372 u64 extent_num_bytes;
4375 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4377 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4380 extent = btrfs_item_ptr(src, src_slot,
4381 struct btrfs_file_extent_item);
4383 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4387 * Don't copy extents from past generations. That would make us
4388 * log a lot more metadata for common cases like doing only a
4389 * few random writes into a file and then fsync it for the first
4390 * time or after the full sync flag is set on the inode. We can
4391 * get leaves full of extent items, most of which are from past
4392 * generations, so we can skip them - as long as the inode has
4393 * not been the target of a reflink operation in this transaction,
4394 * as in that case it might have had file extent items with old
4395 * generations copied into it. We also must always log prealloc
4396 * extents that start at or beyond eof, otherwise we would lose
4397 * them on log replay.
4399 if (is_old_extent &&
4400 ins_keys[dst_index].offset < i_size &&
4401 inode->last_reflink_trans < trans->transid)
4407 /* Only regular extents have checksums. */
4408 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4412 * If it's an extent created in a past transaction, then its
4413 * checksums are already accessible from the committed csum tree,
4414 * no need to log them.
4419 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4420 /* If it's an explicit hole, there are no checksums. */
4421 if (disk_bytenr == 0)
4424 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4426 if (btrfs_file_extent_compression(src, extent)) {
4428 extent_num_bytes = disk_num_bytes;
4430 extent_offset = btrfs_file_extent_offset(src, extent);
4431 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4434 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4435 disk_bytenr += extent_offset;
4436 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4437 disk_bytenr + extent_num_bytes - 1,
4438 &ordered_sums, 0, false);
4442 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4444 ret = log_csums(trans, inode, log, sums);
4445 list_del(&sums->list);
4452 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4453 batch.total_data_size += ins_sizes[dst_index];
4459 * We have a leaf full of old extent items that don't need to be logged,
4460 * so we don't need to do anything.
4465 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4470 for (i = 0; i < nr; i++) {
4471 const int src_slot = start_slot + i;
4472 const int dst_slot = dst_path->slots[0] + dst_index;
4473 struct btrfs_key key;
4474 unsigned long src_offset;
4475 unsigned long dst_offset;
4478 * We're done, all the remaining items in the source leaf
4479 * correspond to old file extent items.
4481 if (dst_index >= batch.nr)
4484 btrfs_item_key_to_cpu(src, &key, src_slot);
4486 if (key.type != BTRFS_EXTENT_DATA_KEY)
4489 extent = btrfs_item_ptr(src, src_slot,
4490 struct btrfs_file_extent_item);
4492 /* See the comment in the previous loop, same logic. */
4493 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4494 key.offset < i_size &&
4495 inode->last_reflink_trans < trans->transid)
4499 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4500 src_offset = btrfs_item_ptr_offset(src, src_slot);
4502 if (key.type == BTRFS_INODE_ITEM_KEY) {
4503 struct btrfs_inode_item *inode_item;
4505 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4506 struct btrfs_inode_item);
4507 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4509 inode_only == LOG_INODE_EXISTS,
4512 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4513 src_offset, ins_sizes[dst_index]);
4519 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4520 btrfs_release_path(dst_path);
4527 static int extent_cmp(void *priv, const struct list_head *a,
4528 const struct list_head *b)
4530 const struct extent_map *em1, *em2;
4532 em1 = list_entry(a, struct extent_map, list);
4533 em2 = list_entry(b, struct extent_map, list);
4535 if (em1->start < em2->start)
4537 else if (em1->start > em2->start)
4542 static int log_extent_csums(struct btrfs_trans_handle *trans,
4543 struct btrfs_inode *inode,
4544 struct btrfs_root *log_root,
4545 const struct extent_map *em,
4546 struct btrfs_log_ctx *ctx)
4548 struct btrfs_ordered_extent *ordered;
4549 struct btrfs_root *csum_root;
4552 u64 mod_start = em->mod_start;
4553 u64 mod_len = em->mod_len;
4554 LIST_HEAD(ordered_sums);
4557 if (inode->flags & BTRFS_INODE_NODATASUM ||
4558 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4559 em->block_start == EXTENT_MAP_HOLE)
4562 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4563 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4564 const u64 mod_end = mod_start + mod_len;
4565 struct btrfs_ordered_sum *sums;
4570 if (ordered_end <= mod_start)
4572 if (mod_end <= ordered->file_offset)
4576 * We are going to copy all the csums on this ordered extent, so
4577 * go ahead and adjust mod_start and mod_len in case this ordered
4578 * extent has already been logged.
4580 if (ordered->file_offset > mod_start) {
4581 if (ordered_end >= mod_end)
4582 mod_len = ordered->file_offset - mod_start;
4584 * If we have this case
4586 * |--------- logged extent ---------|
4587 * |----- ordered extent ----|
4589 * Just don't mess with mod_start and mod_len, we'll
4590 * just end up logging more csums than we need and it
4594 if (ordered_end < mod_end) {
4595 mod_len = mod_end - ordered_end;
4596 mod_start = ordered_end;
4603 * To keep us from looping for the above case of an ordered
4604 * extent that falls inside of the logged extent.
4606 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4609 list_for_each_entry(sums, &ordered->list, list) {
4610 ret = log_csums(trans, inode, log_root, sums);
4616 /* We're done, found all csums in the ordered extents. */
4620 /* If we're compressed we have to save the entire range of csums. */
4621 if (em->compress_type) {
4623 csum_len = max(em->block_len, em->orig_block_len);
4625 csum_offset = mod_start - em->start;
4629 /* block start is already adjusted for the file extent offset. */
4630 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4631 ret = btrfs_lookup_csums_list(csum_root, em->block_start + csum_offset,
4632 em->block_start + csum_offset +
4633 csum_len - 1, &ordered_sums, 0, false);
4637 while (!list_empty(&ordered_sums)) {
4638 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4639 struct btrfs_ordered_sum,
4642 ret = log_csums(trans, inode, log_root, sums);
4643 list_del(&sums->list);
4650 static int log_one_extent(struct btrfs_trans_handle *trans,
4651 struct btrfs_inode *inode,
4652 const struct extent_map *em,
4653 struct btrfs_path *path,
4654 struct btrfs_log_ctx *ctx)
4656 struct btrfs_drop_extents_args drop_args = { 0 };
4657 struct btrfs_root *log = inode->root->log_root;
4658 struct btrfs_file_extent_item fi = { 0 };
4659 struct extent_buffer *leaf;
4660 struct btrfs_key key;
4661 u64 extent_offset = em->start - em->orig_start;
4665 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4666 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4667 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4669 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4671 block_len = max(em->block_len, em->orig_block_len);
4672 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4673 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4674 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4675 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4676 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4678 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4681 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4682 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4683 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4684 btrfs_set_stack_file_extent_compression(&fi, em->compress_type);
4686 ret = log_extent_csums(trans, inode, log, em, ctx);
4691 * If this is the first time we are logging the inode in the current
4692 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4693 * because it does a deletion search, which always acquires write locks
4694 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4695 * but also adds significant contention in a log tree, since log trees
4696 * are small, with a root at level 2 or 3 at most, due to their short
4699 if (ctx->logged_before) {
4700 drop_args.path = path;
4701 drop_args.start = em->start;
4702 drop_args.end = em->start + em->len;
4703 drop_args.replace_extent = true;
4704 drop_args.extent_item_size = sizeof(fi);
4705 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4710 if (!drop_args.extent_inserted) {
4711 key.objectid = btrfs_ino(inode);
4712 key.type = BTRFS_EXTENT_DATA_KEY;
4713 key.offset = em->start;
4715 ret = btrfs_insert_empty_item(trans, log, path, &key,
4720 leaf = path->nodes[0];
4721 write_extent_buffer(leaf, &fi,
4722 btrfs_item_ptr_offset(leaf, path->slots[0]),
4724 btrfs_mark_buffer_dirty(leaf);
4726 btrfs_release_path(path);
4732 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4733 * lose them after doing a full/fast fsync and replaying the log. We scan the
4734 * subvolume's root instead of iterating the inode's extent map tree because
4735 * otherwise we can log incorrect extent items based on extent map conversion.
4736 * That can happen due to the fact that extent maps are merged when they
4737 * are not in the extent map tree's list of modified extents.
4739 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4740 struct btrfs_inode *inode,
4741 struct btrfs_path *path)
4743 struct btrfs_root *root = inode->root;
4744 struct btrfs_key key;
4745 const u64 i_size = i_size_read(&inode->vfs_inode);
4746 const u64 ino = btrfs_ino(inode);
4747 struct btrfs_path *dst_path = NULL;
4748 bool dropped_extents = false;
4749 u64 truncate_offset = i_size;
4750 struct extent_buffer *leaf;
4756 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4760 key.type = BTRFS_EXTENT_DATA_KEY;
4761 key.offset = i_size;
4762 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4767 * We must check if there is a prealloc extent that starts before the
4768 * i_size and crosses the i_size boundary. This is to ensure later we
4769 * truncate down to the end of that extent and not to the i_size, as
4770 * otherwise we end up losing part of the prealloc extent after a log
4771 * replay and with an implicit hole if there is another prealloc extent
4772 * that starts at an offset beyond i_size.
4774 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4779 struct btrfs_file_extent_item *ei;
4781 leaf = path->nodes[0];
4782 slot = path->slots[0];
4783 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4785 if (btrfs_file_extent_type(leaf, ei) ==
4786 BTRFS_FILE_EXTENT_PREALLOC) {
4789 btrfs_item_key_to_cpu(leaf, &key, slot);
4790 extent_end = key.offset +
4791 btrfs_file_extent_num_bytes(leaf, ei);
4793 if (extent_end > i_size)
4794 truncate_offset = extent_end;
4801 leaf = path->nodes[0];
4802 slot = path->slots[0];
4804 if (slot >= btrfs_header_nritems(leaf)) {
4806 ret = copy_items(trans, inode, dst_path, path,
4807 start_slot, ins_nr, 1, 0);
4812 ret = btrfs_next_leaf(root, path);
4822 btrfs_item_key_to_cpu(leaf, &key, slot);
4823 if (key.objectid > ino)
4825 if (WARN_ON_ONCE(key.objectid < ino) ||
4826 key.type < BTRFS_EXTENT_DATA_KEY ||
4827 key.offset < i_size) {
4831 if (!dropped_extents) {
4833 * Avoid logging extent items logged in past fsync calls
4834 * and leading to duplicate keys in the log tree.
4836 ret = truncate_inode_items(trans, root->log_root, inode,
4838 BTRFS_EXTENT_DATA_KEY);
4841 dropped_extents = true;
4848 dst_path = btrfs_alloc_path();
4856 ret = copy_items(trans, inode, dst_path, path,
4857 start_slot, ins_nr, 1, 0);
4859 btrfs_release_path(path);
4860 btrfs_free_path(dst_path);
4864 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4865 struct btrfs_inode *inode,
4866 struct btrfs_path *path,
4867 struct btrfs_log_ctx *ctx)
4869 struct btrfs_ordered_extent *ordered;
4870 struct btrfs_ordered_extent *tmp;
4871 struct extent_map *em, *n;
4872 struct list_head extents;
4873 struct extent_map_tree *tree = &inode->extent_tree;
4877 INIT_LIST_HEAD(&extents);
4879 write_lock(&tree->lock);
4881 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4882 list_del_init(&em->list);
4884 * Just an arbitrary number, this can be really CPU intensive
4885 * once we start getting a lot of extents, and really once we
4886 * have a bunch of extents we just want to commit since it will
4889 if (++num > 32768) {
4890 list_del_init(&tree->modified_extents);
4895 if (em->generation < trans->transid)
4898 /* We log prealloc extents beyond eof later. */
4899 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4900 em->start >= i_size_read(&inode->vfs_inode))
4903 /* Need a ref to keep it from getting evicted from cache */
4904 refcount_inc(&em->refs);
4905 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4906 list_add_tail(&em->list, &extents);
4910 list_sort(NULL, &extents, extent_cmp);
4912 while (!list_empty(&extents)) {
4913 em = list_entry(extents.next, struct extent_map, list);
4915 list_del_init(&em->list);
4918 * If we had an error we just need to delete everybody from our
4922 clear_em_logging(tree, em);
4923 free_extent_map(em);
4927 write_unlock(&tree->lock);
4929 ret = log_one_extent(trans, inode, em, path, ctx);
4930 write_lock(&tree->lock);
4931 clear_em_logging(tree, em);
4932 free_extent_map(em);
4934 WARN_ON(!list_empty(&extents));
4935 write_unlock(&tree->lock);
4938 ret = btrfs_log_prealloc_extents(trans, inode, path);
4943 * We have logged all extents successfully, now make sure the commit of
4944 * the current transaction waits for the ordered extents to complete
4945 * before it commits and wipes out the log trees, otherwise we would
4946 * lose data if an ordered extents completes after the transaction
4947 * commits and a power failure happens after the transaction commit.
4949 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4950 list_del_init(&ordered->log_list);
4951 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4953 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4954 spin_lock_irq(&inode->ordered_tree.lock);
4955 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4956 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4957 atomic_inc(&trans->transaction->pending_ordered);
4959 spin_unlock_irq(&inode->ordered_tree.lock);
4961 btrfs_put_ordered_extent(ordered);
4967 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4968 struct btrfs_path *path, u64 *size_ret)
4970 struct btrfs_key key;
4973 key.objectid = btrfs_ino(inode);
4974 key.type = BTRFS_INODE_ITEM_KEY;
4977 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4980 } else if (ret > 0) {
4983 struct btrfs_inode_item *item;
4985 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4986 struct btrfs_inode_item);
4987 *size_ret = btrfs_inode_size(path->nodes[0], item);
4989 * If the in-memory inode's i_size is smaller then the inode
4990 * size stored in the btree, return the inode's i_size, so
4991 * that we get a correct inode size after replaying the log
4992 * when before a power failure we had a shrinking truncate
4993 * followed by addition of a new name (rename / new hard link).
4994 * Otherwise return the inode size from the btree, to avoid
4995 * data loss when replaying a log due to previously doing a
4996 * write that expands the inode's size and logging a new name
4997 * immediately after.
4999 if (*size_ret > inode->vfs_inode.i_size)
5000 *size_ret = inode->vfs_inode.i_size;
5003 btrfs_release_path(path);
5008 * At the moment we always log all xattrs. This is to figure out at log replay
5009 * time which xattrs must have their deletion replayed. If a xattr is missing
5010 * in the log tree and exists in the fs/subvol tree, we delete it. This is
5011 * because if a xattr is deleted, the inode is fsynced and a power failure
5012 * happens, causing the log to be replayed the next time the fs is mounted,
5013 * we want the xattr to not exist anymore (same behaviour as other filesystems
5014 * with a journal, ext3/4, xfs, f2fs, etc).
5016 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
5017 struct btrfs_inode *inode,
5018 struct btrfs_path *path,
5019 struct btrfs_path *dst_path)
5021 struct btrfs_root *root = inode->root;
5023 struct btrfs_key key;
5024 const u64 ino = btrfs_ino(inode);
5027 bool found_xattrs = false;
5029 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
5033 key.type = BTRFS_XATTR_ITEM_KEY;
5036 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5041 int slot = path->slots[0];
5042 struct extent_buffer *leaf = path->nodes[0];
5043 int nritems = btrfs_header_nritems(leaf);
5045 if (slot >= nritems) {
5047 ret = copy_items(trans, inode, dst_path, path,
5048 start_slot, ins_nr, 1, 0);
5053 ret = btrfs_next_leaf(root, path);
5061 btrfs_item_key_to_cpu(leaf, &key, slot);
5062 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5069 found_xattrs = true;
5073 ret = copy_items(trans, inode, dst_path, path,
5074 start_slot, ins_nr, 1, 0);
5080 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5086 * When using the NO_HOLES feature if we punched a hole that causes the
5087 * deletion of entire leafs or all the extent items of the first leaf (the one
5088 * that contains the inode item and references) we may end up not processing
5089 * any extents, because there are no leafs with a generation matching the
5090 * current transaction that have extent items for our inode. So we need to find
5091 * if any holes exist and then log them. We also need to log holes after any
5092 * truncate operation that changes the inode's size.
5094 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5095 struct btrfs_inode *inode,
5096 struct btrfs_path *path)
5098 struct btrfs_root *root = inode->root;
5099 struct btrfs_fs_info *fs_info = root->fs_info;
5100 struct btrfs_key key;
5101 const u64 ino = btrfs_ino(inode);
5102 const u64 i_size = i_size_read(&inode->vfs_inode);
5103 u64 prev_extent_end = 0;
5106 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5110 key.type = BTRFS_EXTENT_DATA_KEY;
5113 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5118 struct extent_buffer *leaf = path->nodes[0];
5120 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5121 ret = btrfs_next_leaf(root, path);
5128 leaf = path->nodes[0];
5131 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5132 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5135 /* We have a hole, log it. */
5136 if (prev_extent_end < key.offset) {
5137 const u64 hole_len = key.offset - prev_extent_end;
5140 * Release the path to avoid deadlocks with other code
5141 * paths that search the root while holding locks on
5142 * leafs from the log root.
5144 btrfs_release_path(path);
5145 ret = btrfs_insert_hole_extent(trans, root->log_root,
5146 ino, prev_extent_end,
5152 * Search for the same key again in the root. Since it's
5153 * an extent item and we are holding the inode lock, the
5154 * key must still exist. If it doesn't just emit warning
5155 * and return an error to fall back to a transaction
5158 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5161 if (WARN_ON(ret > 0))
5163 leaf = path->nodes[0];
5166 prev_extent_end = btrfs_file_extent_end(path);
5171 if (prev_extent_end < i_size) {
5174 btrfs_release_path(path);
5175 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5176 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5177 prev_extent_end, hole_len);
5186 * When we are logging a new inode X, check if it doesn't have a reference that
5187 * matches the reference from some other inode Y created in a past transaction
5188 * and that was renamed in the current transaction. If we don't do this, then at
5189 * log replay time we can lose inode Y (and all its files if it's a directory):
5192 * echo "hello world" > /mnt/x/foobar
5195 * mkdir /mnt/x # or touch /mnt/x
5196 * xfs_io -c fsync /mnt/x
5198 * mount fs, trigger log replay
5200 * After the log replay procedure, we would lose the first directory and all its
5201 * files (file foobar).
5202 * For the case where inode Y is not a directory we simply end up losing it:
5204 * echo "123" > /mnt/foo
5206 * mv /mnt/foo /mnt/bar
5207 * echo "abc" > /mnt/foo
5208 * xfs_io -c fsync /mnt/foo
5211 * We also need this for cases where a snapshot entry is replaced by some other
5212 * entry (file or directory) otherwise we end up with an unreplayable log due to
5213 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5214 * if it were a regular entry:
5217 * btrfs subvolume snapshot /mnt /mnt/x/snap
5218 * btrfs subvolume delete /mnt/x/snap
5221 * fsync /mnt/x or fsync some new file inside it
5224 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5225 * the same transaction.
5227 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5229 const struct btrfs_key *key,
5230 struct btrfs_inode *inode,
5231 u64 *other_ino, u64 *other_parent)
5234 struct btrfs_path *search_path;
5237 u32 item_size = btrfs_item_size(eb, slot);
5239 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5241 search_path = btrfs_alloc_path();
5244 search_path->search_commit_root = 1;
5245 search_path->skip_locking = 1;
5247 while (cur_offset < item_size) {
5251 unsigned long name_ptr;
5252 struct btrfs_dir_item *di;
5253 struct fscrypt_str name_str;
5255 if (key->type == BTRFS_INODE_REF_KEY) {
5256 struct btrfs_inode_ref *iref;
5258 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5259 parent = key->offset;
5260 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5261 name_ptr = (unsigned long)(iref + 1);
5262 this_len = sizeof(*iref) + this_name_len;
5264 struct btrfs_inode_extref *extref;
5266 extref = (struct btrfs_inode_extref *)(ptr +
5268 parent = btrfs_inode_extref_parent(eb, extref);
5269 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5270 name_ptr = (unsigned long)&extref->name;
5271 this_len = sizeof(*extref) + this_name_len;
5274 if (this_name_len > name_len) {
5277 new_name = krealloc(name, this_name_len, GFP_NOFS);
5282 name_len = this_name_len;
5286 read_extent_buffer(eb, name, name_ptr, this_name_len);
5288 name_str.name = name;
5289 name_str.len = this_name_len;
5290 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5291 parent, &name_str, 0);
5292 if (di && !IS_ERR(di)) {
5293 struct btrfs_key di_key;
5295 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5297 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5298 if (di_key.objectid != key->objectid) {
5300 *other_ino = di_key.objectid;
5301 *other_parent = parent;
5309 } else if (IS_ERR(di)) {
5313 btrfs_release_path(search_path);
5315 cur_offset += this_len;
5319 btrfs_free_path(search_path);
5325 * Check if we need to log an inode. This is used in contexts where while
5326 * logging an inode we need to log another inode (either that it exists or in
5327 * full mode). This is used instead of btrfs_inode_in_log() because the later
5328 * requires the inode to be in the log and have the log transaction committed,
5329 * while here we do not care if the log transaction was already committed - our
5330 * caller will commit the log later - and we want to avoid logging an inode
5331 * multiple times when multiple tasks have joined the same log transaction.
5333 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5334 const struct btrfs_inode *inode)
5337 * If a directory was not modified, no dentries added or removed, we can
5338 * and should avoid logging it.
5340 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5344 * If this inode does not have new/updated/deleted xattrs since the last
5345 * time it was logged and is flagged as logged in the current transaction,
5346 * we can skip logging it. As for new/deleted names, those are updated in
5347 * the log by link/unlink/rename operations.
5348 * In case the inode was logged and then evicted and reloaded, its
5349 * logged_trans will be 0, in which case we have to fully log it since
5350 * logged_trans is a transient field, not persisted.
5352 if (inode->logged_trans == trans->transid &&
5353 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5359 struct btrfs_dir_list {
5361 struct list_head list;
5365 * Log the inodes of the new dentries of a directory.
5366 * See process_dir_items_leaf() for details about why it is needed.
5367 * This is a recursive operation - if an existing dentry corresponds to a
5368 * directory, that directory's new entries are logged too (same behaviour as
5369 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5370 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5371 * complains about the following circular lock dependency / possible deadlock:
5375 * lock(&type->i_mutex_dir_key#3/2);
5376 * lock(sb_internal#2);
5377 * lock(&type->i_mutex_dir_key#3/2);
5378 * lock(&sb->s_type->i_mutex_key#14);
5380 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5381 * sb_start_intwrite() in btrfs_start_transaction().
5382 * Not acquiring the VFS lock of the inodes is still safe because:
5384 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5385 * that while logging the inode new references (names) are added or removed
5386 * from the inode, leaving the logged inode item with a link count that does
5387 * not match the number of logged inode reference items. This is fine because
5388 * at log replay time we compute the real number of links and correct the
5389 * link count in the inode item (see replay_one_buffer() and
5390 * link_to_fixup_dir());
5392 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5393 * while logging the inode's items new index items (key type
5394 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5395 * has a size that doesn't match the sum of the lengths of all the logged
5396 * names - this is ok, not a problem, because at log replay time we set the
5397 * directory's i_size to the correct value (see replay_one_name() and
5398 * overwrite_item()).
5400 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5401 struct btrfs_inode *start_inode,
5402 struct btrfs_log_ctx *ctx)
5404 struct btrfs_root *root = start_inode->root;
5405 struct btrfs_fs_info *fs_info = root->fs_info;
5406 struct btrfs_path *path;
5407 LIST_HEAD(dir_list);
5408 struct btrfs_dir_list *dir_elem;
5409 u64 ino = btrfs_ino(start_inode);
5413 * If we are logging a new name, as part of a link or rename operation,
5414 * don't bother logging new dentries, as we just want to log the names
5415 * of an inode and that any new parents exist.
5417 if (ctx->logging_new_name)
5420 path = btrfs_alloc_path();
5425 struct extent_buffer *leaf;
5426 struct btrfs_key min_key;
5427 bool continue_curr_inode = true;
5431 min_key.objectid = ino;
5432 min_key.type = BTRFS_DIR_INDEX_KEY;
5435 btrfs_release_path(path);
5436 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
5439 } else if (ret > 0) {
5444 leaf = path->nodes[0];
5445 nritems = btrfs_header_nritems(leaf);
5446 for (i = path->slots[0]; i < nritems; i++) {
5447 struct btrfs_dir_item *di;
5448 struct btrfs_key di_key;
5449 struct inode *di_inode;
5450 int log_mode = LOG_INODE_EXISTS;
5453 btrfs_item_key_to_cpu(leaf, &min_key, i);
5454 if (min_key.objectid != ino ||
5455 min_key.type != BTRFS_DIR_INDEX_KEY) {
5456 continue_curr_inode = false;
5460 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5461 type = btrfs_dir_ftype(leaf, di);
5462 if (btrfs_dir_transid(leaf, di) < trans->transid)
5464 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5465 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5468 btrfs_release_path(path);
5469 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5470 if (IS_ERR(di_inode)) {
5471 ret = PTR_ERR(di_inode);
5475 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5476 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5480 ctx->log_new_dentries = false;
5481 if (type == BTRFS_FT_DIR)
5482 log_mode = LOG_INODE_ALL;
5483 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5485 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5488 if (ctx->log_new_dentries) {
5489 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5494 dir_elem->ino = di_key.objectid;
5495 list_add_tail(&dir_elem->list, &dir_list);
5500 if (continue_curr_inode && min_key.offset < (u64)-1) {
5506 if (list_empty(&dir_list))
5509 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5510 ino = dir_elem->ino;
5511 list_del(&dir_elem->list);
5515 btrfs_free_path(path);
5517 struct btrfs_dir_list *next;
5519 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5526 struct btrfs_ino_list {
5529 struct list_head list;
5532 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5534 struct btrfs_ino_list *curr;
5535 struct btrfs_ino_list *next;
5537 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5538 list_del(&curr->list);
5543 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5544 struct btrfs_path *path)
5546 struct btrfs_key key;
5550 key.type = BTRFS_INODE_ITEM_KEY;
5553 path->search_commit_root = 1;
5554 path->skip_locking = 1;
5556 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5557 if (WARN_ON_ONCE(ret > 0)) {
5559 * We have previously found the inode through the commit root
5560 * so this should not happen. If it does, just error out and
5561 * fallback to a transaction commit.
5564 } else if (ret == 0) {
5565 struct btrfs_inode_item *item;
5567 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5568 struct btrfs_inode_item);
5569 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5573 btrfs_release_path(path);
5574 path->search_commit_root = 0;
5575 path->skip_locking = 0;
5580 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5581 struct btrfs_root *root,
5582 struct btrfs_path *path,
5583 u64 ino, u64 parent,
5584 struct btrfs_log_ctx *ctx)
5586 struct btrfs_ino_list *ino_elem;
5587 struct inode *inode;
5590 * It's rare to have a lot of conflicting inodes, in practice it is not
5591 * common to have more than 1 or 2. We don't want to collect too many,
5592 * as we could end up logging too many inodes (even if only in
5593 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5596 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES) {
5597 btrfs_set_log_full_commit(trans);
5598 return BTRFS_LOG_FORCE_COMMIT;
5601 inode = btrfs_iget(root->fs_info->sb, ino, root);
5603 * If the other inode that had a conflicting dir entry was deleted in
5604 * the current transaction then we either:
5606 * 1) Log the parent directory (later after adding it to the list) if
5607 * the inode is a directory. This is because it may be a deleted
5608 * subvolume/snapshot or it may be a regular directory that had
5609 * deleted subvolumes/snapshots (or subdirectories that had them),
5610 * and at the moment we can't deal with dropping subvolumes/snapshots
5611 * during log replay. So we just log the parent, which will result in
5612 * a fallback to a transaction commit if we are dealing with those
5613 * cases (last_unlink_trans will match the current transaction);
5615 * 2) Do nothing if it's not a directory. During log replay we simply
5616 * unlink the conflicting dentry from the parent directory and then
5617 * add the dentry for our inode. Like this we can avoid logging the
5618 * parent directory (and maybe fallback to a transaction commit in
5619 * case it has a last_unlink_trans == trans->transid, due to moving
5620 * some inode from it to some other directory).
5622 if (IS_ERR(inode)) {
5623 int ret = PTR_ERR(inode);
5628 ret = conflicting_inode_is_dir(root, ino, path);
5629 /* Not a directory or we got an error. */
5633 /* Conflicting inode is a directory, so we'll log its parent. */
5634 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5637 ino_elem->ino = ino;
5638 ino_elem->parent = parent;
5639 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5640 ctx->num_conflict_inodes++;
5646 * If the inode was already logged skip it - otherwise we can hit an
5647 * infinite loop. Example:
5649 * From the commit root (previous transaction) we have the following
5652 * inode 257 a directory
5653 * inode 258 with references "zz" and "zz_link" on inode 257
5654 * inode 259 with reference "a" on inode 257
5656 * And in the current (uncommitted) transaction we have:
5658 * inode 257 a directory, unchanged
5659 * inode 258 with references "a" and "a2" on inode 257
5660 * inode 259 with reference "zz_link" on inode 257
5661 * inode 261 with reference "zz" on inode 257
5663 * When logging inode 261 the following infinite loop could
5664 * happen if we don't skip already logged inodes:
5666 * - we detect inode 258 as a conflicting inode, with inode 261
5667 * on reference "zz", and log it;
5669 * - we detect inode 259 as a conflicting inode, with inode 258
5670 * on reference "a", and log it;
5672 * - we detect inode 258 as a conflicting inode, with inode 259
5673 * on reference "zz_link", and log it - again! After this we
5674 * repeat the above steps forever.
5676 * Here we can use need_log_inode() because we only need to log the
5677 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5678 * so that the log ends up with the new name and without the old name.
5680 if (!need_log_inode(trans, BTRFS_I(inode))) {
5681 btrfs_add_delayed_iput(BTRFS_I(inode));
5685 btrfs_add_delayed_iput(BTRFS_I(inode));
5687 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5690 ino_elem->ino = ino;
5691 ino_elem->parent = parent;
5692 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5693 ctx->num_conflict_inodes++;
5698 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5699 struct btrfs_root *root,
5700 struct btrfs_log_ctx *ctx)
5702 struct btrfs_fs_info *fs_info = root->fs_info;
5706 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5707 * otherwise we could have unbounded recursion of btrfs_log_inode()
5708 * calls. This check guarantees we can have only 1 level of recursion.
5710 if (ctx->logging_conflict_inodes)
5713 ctx->logging_conflict_inodes = true;
5716 * New conflicting inodes may be found and added to the list while we
5717 * are logging a conflicting inode, so keep iterating while the list is
5720 while (!list_empty(&ctx->conflict_inodes)) {
5721 struct btrfs_ino_list *curr;
5722 struct inode *inode;
5726 curr = list_first_entry(&ctx->conflict_inodes,
5727 struct btrfs_ino_list, list);
5729 parent = curr->parent;
5730 list_del(&curr->list);
5733 inode = btrfs_iget(fs_info->sb, ino, root);
5735 * If the other inode that had a conflicting dir entry was
5736 * deleted in the current transaction, we need to log its parent
5737 * directory. See the comment at add_conflicting_inode().
5739 if (IS_ERR(inode)) {
5740 ret = PTR_ERR(inode);
5744 inode = btrfs_iget(fs_info->sb, parent, root);
5745 if (IS_ERR(inode)) {
5746 ret = PTR_ERR(inode);
5751 * Always log the directory, we cannot make this
5752 * conditional on need_log_inode() because the directory
5753 * might have been logged in LOG_INODE_EXISTS mode or
5754 * the dir index of the conflicting inode is not in a
5755 * dir index key range logged for the directory. So we
5756 * must make sure the deletion is recorded.
5758 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5759 LOG_INODE_ALL, ctx);
5760 btrfs_add_delayed_iput(BTRFS_I(inode));
5767 * Here we can use need_log_inode() because we only need to log
5768 * the inode in LOG_INODE_EXISTS mode and rename operations
5769 * update the log, so that the log ends up with the new name and
5770 * without the old name.
5772 * We did this check at add_conflicting_inode(), but here we do
5773 * it again because if some other task logged the inode after
5774 * that, we can avoid doing it again.
5776 if (!need_log_inode(trans, BTRFS_I(inode))) {
5777 btrfs_add_delayed_iput(BTRFS_I(inode));
5782 * We are safe logging the other inode without acquiring its
5783 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5784 * are safe against concurrent renames of the other inode as
5785 * well because during a rename we pin the log and update the
5786 * log with the new name before we unpin it.
5788 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5789 btrfs_add_delayed_iput(BTRFS_I(inode));
5794 ctx->logging_conflict_inodes = false;
5796 free_conflicting_inodes(ctx);
5801 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5802 struct btrfs_inode *inode,
5803 struct btrfs_key *min_key,
5804 const struct btrfs_key *max_key,
5805 struct btrfs_path *path,
5806 struct btrfs_path *dst_path,
5807 const u64 logged_isize,
5808 const int inode_only,
5809 struct btrfs_log_ctx *ctx,
5810 bool *need_log_inode_item)
5812 const u64 i_size = i_size_read(&inode->vfs_inode);
5813 struct btrfs_root *root = inode->root;
5814 int ins_start_slot = 0;
5819 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5827 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5828 if (min_key->objectid != max_key->objectid)
5830 if (min_key->type > max_key->type)
5833 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5834 *need_log_inode_item = false;
5835 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5836 min_key->offset >= i_size) {
5838 * Extents at and beyond eof are logged with
5839 * btrfs_log_prealloc_extents().
5840 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5841 * and no keys greater than that, so bail out.
5844 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5845 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5846 (inode->generation == trans->transid ||
5847 ctx->logging_conflict_inodes)) {
5849 u64 other_parent = 0;
5851 ret = btrfs_check_ref_name_override(path->nodes[0],
5852 path->slots[0], min_key, inode,
5853 &other_ino, &other_parent);
5856 } else if (ret > 0 &&
5857 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5862 ins_start_slot = path->slots[0];
5864 ret = copy_items(trans, inode, dst_path, path,
5865 ins_start_slot, ins_nr,
5866 inode_only, logged_isize);
5871 btrfs_release_path(path);
5872 ret = add_conflicting_inode(trans, root, path,
5879 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5880 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5883 ret = copy_items(trans, inode, dst_path, path,
5885 ins_nr, inode_only, logged_isize);
5892 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5895 } else if (!ins_nr) {
5896 ins_start_slot = path->slots[0];
5901 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5902 ins_nr, inode_only, logged_isize);
5906 ins_start_slot = path->slots[0];
5909 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5910 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5915 ret = copy_items(trans, inode, dst_path, path,
5916 ins_start_slot, ins_nr, inode_only,
5922 btrfs_release_path(path);
5924 if (min_key->offset < (u64)-1) {
5926 } else if (min_key->type < max_key->type) {
5928 min_key->offset = 0;
5934 * We may process many leaves full of items for our inode, so
5935 * avoid monopolizing a cpu for too long by rescheduling while
5936 * not holding locks on any tree.
5941 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5942 ins_nr, inode_only, logged_isize);
5947 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5949 * Release the path because otherwise we might attempt to double
5950 * lock the same leaf with btrfs_log_prealloc_extents() below.
5952 btrfs_release_path(path);
5953 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5959 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
5960 struct btrfs_root *log,
5961 struct btrfs_path *path,
5962 const struct btrfs_item_batch *batch,
5963 const struct btrfs_delayed_item *first_item)
5965 const struct btrfs_delayed_item *curr = first_item;
5968 ret = btrfs_insert_empty_items(trans, log, path, batch);
5972 for (int i = 0; i < batch->nr; i++) {
5975 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
5976 write_extent_buffer(path->nodes[0], &curr->data,
5977 (unsigned long)data_ptr, curr->data_len);
5978 curr = list_next_entry(curr, log_list);
5982 btrfs_release_path(path);
5987 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
5988 struct btrfs_inode *inode,
5989 struct btrfs_path *path,
5990 const struct list_head *delayed_ins_list,
5991 struct btrfs_log_ctx *ctx)
5993 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
5994 const int max_batch_size = 195;
5995 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
5996 const u64 ino = btrfs_ino(inode);
5997 struct btrfs_root *log = inode->root->log_root;
5998 struct btrfs_item_batch batch = {
6000 .total_data_size = 0,
6002 const struct btrfs_delayed_item *first = NULL;
6003 const struct btrfs_delayed_item *curr;
6005 struct btrfs_key *ins_keys;
6007 u64 curr_batch_size = 0;
6011 /* We are adding dir index items to the log tree. */
6012 lockdep_assert_held(&inode->log_mutex);
6015 * We collect delayed items before copying index keys from the subvolume
6016 * to the log tree. However just after we collected them, they may have
6017 * been flushed (all of them or just some of them), and therefore we
6018 * could have copied them from the subvolume tree to the log tree.
6019 * So find the first delayed item that was not yet logged (they are
6020 * sorted by index number).
6022 list_for_each_entry(curr, delayed_ins_list, log_list) {
6023 if (curr->index > inode->last_dir_index_offset) {
6029 /* Empty list or all delayed items were already logged. */
6033 ins_data = kmalloc(max_batch_size * sizeof(u32) +
6034 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
6037 ins_sizes = (u32 *)ins_data;
6038 batch.data_sizes = ins_sizes;
6039 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6040 batch.keys = ins_keys;
6043 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6044 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6046 if (curr_batch_size + curr_size > leaf_data_size ||
6047 batch.nr == max_batch_size) {
6048 ret = insert_delayed_items_batch(trans, log, path,
6054 batch.total_data_size = 0;
6055 curr_batch_size = 0;
6059 ins_sizes[batch_idx] = curr->data_len;
6060 ins_keys[batch_idx].objectid = ino;
6061 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6062 ins_keys[batch_idx].offset = curr->index;
6063 curr_batch_size += curr_size;
6064 batch.total_data_size += curr->data_len;
6067 curr = list_next_entry(curr, log_list);
6070 ASSERT(batch.nr >= 1);
6071 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6073 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6075 inode->last_dir_index_offset = curr->index;
6082 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6083 struct btrfs_inode *inode,
6084 struct btrfs_path *path,
6085 const struct list_head *delayed_del_list,
6086 struct btrfs_log_ctx *ctx)
6088 const u64 ino = btrfs_ino(inode);
6089 const struct btrfs_delayed_item *curr;
6091 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6094 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6095 u64 first_dir_index = curr->index;
6097 const struct btrfs_delayed_item *next;
6101 * Find a range of consecutive dir index items to delete. Like
6102 * this we log a single dir range item spanning several contiguous
6103 * dir items instead of logging one range item per dir index item.
6105 next = list_next_entry(curr, log_list);
6106 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6107 if (next->index != curr->index + 1)
6110 next = list_next_entry(next, log_list);
6113 last_dir_index = curr->index;
6114 ASSERT(last_dir_index >= first_dir_index);
6116 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6117 ino, first_dir_index, last_dir_index);
6120 curr = list_next_entry(curr, log_list);
6126 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6127 struct btrfs_inode *inode,
6128 struct btrfs_path *path,
6129 struct btrfs_log_ctx *ctx,
6130 const struct list_head *delayed_del_list,
6131 const struct btrfs_delayed_item *first,
6132 const struct btrfs_delayed_item **last_ret)
6134 const struct btrfs_delayed_item *next;
6135 struct extent_buffer *leaf = path->nodes[0];
6136 const int last_slot = btrfs_header_nritems(leaf) - 1;
6137 int slot = path->slots[0] + 1;
6138 const u64 ino = btrfs_ino(inode);
6140 next = list_next_entry(first, log_list);
6142 while (slot < last_slot &&
6143 !list_entry_is_head(next, delayed_del_list, log_list)) {
6144 struct btrfs_key key;
6146 btrfs_item_key_to_cpu(leaf, &key, slot);
6147 if (key.objectid != ino ||
6148 key.type != BTRFS_DIR_INDEX_KEY ||
6149 key.offset != next->index)
6154 next = list_next_entry(next, log_list);
6157 return btrfs_del_items(trans, inode->root->log_root, path,
6158 path->slots[0], slot - path->slots[0]);
6161 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6162 struct btrfs_inode *inode,
6163 struct btrfs_path *path,
6164 const struct list_head *delayed_del_list,
6165 struct btrfs_log_ctx *ctx)
6167 struct btrfs_root *log = inode->root->log_root;
6168 const struct btrfs_delayed_item *curr;
6169 u64 last_range_start;
6170 u64 last_range_end = 0;
6171 struct btrfs_key key;
6173 key.objectid = btrfs_ino(inode);
6174 key.type = BTRFS_DIR_INDEX_KEY;
6175 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6178 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6179 const struct btrfs_delayed_item *last = curr;
6180 u64 first_dir_index = curr->index;
6182 bool deleted_items = false;
6185 key.offset = curr->index;
6186 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6189 } else if (ret == 0) {
6190 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6191 delayed_del_list, curr,
6195 deleted_items = true;
6198 btrfs_release_path(path);
6201 * If we deleted items from the leaf, it means we have a range
6202 * item logging their range, so no need to add one or update an
6203 * existing one. Otherwise we have to log a dir range item.
6208 last_dir_index = last->index;
6209 ASSERT(last_dir_index >= first_dir_index);
6211 * If this range starts right after where the previous one ends,
6212 * then we want to reuse the previous range item and change its
6213 * end offset to the end of this range. This is just to minimize
6214 * leaf space usage, by avoiding adding a new range item.
6216 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6217 first_dir_index = last_range_start;
6219 ret = insert_dir_log_key(trans, log, path, key.objectid,
6220 first_dir_index, last_dir_index);
6224 last_range_start = first_dir_index;
6225 last_range_end = last_dir_index;
6227 curr = list_next_entry(last, log_list);
6233 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6234 struct btrfs_inode *inode,
6235 struct btrfs_path *path,
6236 const struct list_head *delayed_del_list,
6237 struct btrfs_log_ctx *ctx)
6240 * We are deleting dir index items from the log tree or adding range
6243 lockdep_assert_held(&inode->log_mutex);
6245 if (list_empty(delayed_del_list))
6248 if (ctx->logged_before)
6249 return log_delayed_deletions_incremental(trans, inode, path,
6250 delayed_del_list, ctx);
6252 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6257 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6258 * items instead of the subvolume tree.
6260 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6261 struct btrfs_inode *inode,
6262 const struct list_head *delayed_ins_list,
6263 struct btrfs_log_ctx *ctx)
6265 const bool orig_log_new_dentries = ctx->log_new_dentries;
6266 struct btrfs_fs_info *fs_info = trans->fs_info;
6267 struct btrfs_delayed_item *item;
6271 * No need for the log mutex, plus to avoid potential deadlocks or
6272 * lockdep annotations due to nesting of delayed inode mutexes and log
6275 lockdep_assert_not_held(&inode->log_mutex);
6277 ASSERT(!ctx->logging_new_delayed_dentries);
6278 ctx->logging_new_delayed_dentries = true;
6280 list_for_each_entry(item, delayed_ins_list, log_list) {
6281 struct btrfs_dir_item *dir_item;
6282 struct inode *di_inode;
6283 struct btrfs_key key;
6284 int log_mode = LOG_INODE_EXISTS;
6286 dir_item = (struct btrfs_dir_item *)item->data;
6287 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6289 if (key.type == BTRFS_ROOT_ITEM_KEY)
6292 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6293 if (IS_ERR(di_inode)) {
6294 ret = PTR_ERR(di_inode);
6298 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6299 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6303 if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR)
6304 log_mode = LOG_INODE_ALL;
6306 ctx->log_new_dentries = false;
6307 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6309 if (!ret && ctx->log_new_dentries)
6310 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6312 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6318 ctx->log_new_dentries = orig_log_new_dentries;
6319 ctx->logging_new_delayed_dentries = false;
6324 /* log a single inode in the tree log.
6325 * At least one parent directory for this inode must exist in the tree
6326 * or be logged already.
6328 * Any items from this inode changed by the current transaction are copied
6329 * to the log tree. An extra reference is taken on any extents in this
6330 * file, allowing us to avoid a whole pile of corner cases around logging
6331 * blocks that have been removed from the tree.
6333 * See LOG_INODE_ALL and related defines for a description of what inode_only
6336 * This handles both files and directories.
6338 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6339 struct btrfs_inode *inode,
6341 struct btrfs_log_ctx *ctx)
6343 struct btrfs_path *path;
6344 struct btrfs_path *dst_path;
6345 struct btrfs_key min_key;
6346 struct btrfs_key max_key;
6347 struct btrfs_root *log = inode->root->log_root;
6349 bool fast_search = false;
6350 u64 ino = btrfs_ino(inode);
6351 struct extent_map_tree *em_tree = &inode->extent_tree;
6352 u64 logged_isize = 0;
6353 bool need_log_inode_item = true;
6354 bool xattrs_logged = false;
6355 bool inode_item_dropped = true;
6356 bool full_dir_logging = false;
6357 LIST_HEAD(delayed_ins_list);
6358 LIST_HEAD(delayed_del_list);
6360 path = btrfs_alloc_path();
6363 dst_path = btrfs_alloc_path();
6365 btrfs_free_path(path);
6369 min_key.objectid = ino;
6370 min_key.type = BTRFS_INODE_ITEM_KEY;
6373 max_key.objectid = ino;
6376 /* today the code can only do partial logging of directories */
6377 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6378 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6379 &inode->runtime_flags) &&
6380 inode_only >= LOG_INODE_EXISTS))
6381 max_key.type = BTRFS_XATTR_ITEM_KEY;
6383 max_key.type = (u8)-1;
6384 max_key.offset = (u64)-1;
6386 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6387 full_dir_logging = true;
6390 * If we are logging a directory while we are logging dentries of the
6391 * delayed items of some other inode, then we need to flush the delayed
6392 * items of this directory and not log the delayed items directly. This
6393 * is to prevent more than one level of recursion into btrfs_log_inode()
6394 * by having something like this:
6396 * $ mkdir -p a/b/c/d/e/f/g/h/...
6397 * $ xfs_io -c "fsync" a
6399 * Where all directories in the path did not exist before and are
6400 * created in the current transaction.
6401 * So in such a case we directly log the delayed items of the main
6402 * directory ("a") without flushing them first, while for each of its
6403 * subdirectories we flush their delayed items before logging them.
6404 * This prevents a potential unbounded recursion like this:
6407 * log_new_delayed_dentries()
6409 * log_new_delayed_dentries()
6411 * log_new_delayed_dentries()
6414 * We have thresholds for the maximum number of delayed items to have in
6415 * memory, and once they are hit, the items are flushed asynchronously.
6416 * However the limit is quite high, so lets prevent deep levels of
6417 * recursion to happen by limiting the maximum depth to be 1.
6419 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6420 ret = btrfs_commit_inode_delayed_items(trans, inode);
6425 mutex_lock(&inode->log_mutex);
6428 * For symlinks, we must always log their content, which is stored in an
6429 * inline extent, otherwise we could end up with an empty symlink after
6430 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6431 * one attempts to create an empty symlink).
6432 * We don't need to worry about flushing delalloc, because when we create
6433 * the inline extent when the symlink is created (we never have delalloc
6436 if (S_ISLNK(inode->vfs_inode.i_mode))
6437 inode_only = LOG_INODE_ALL;
6440 * Before logging the inode item, cache the value returned by
6441 * inode_logged(), because after that we have the need to figure out if
6442 * the inode was previously logged in this transaction.
6444 ret = inode_logged(trans, inode, path);
6447 ctx->logged_before = (ret == 1);
6451 * This is for cases where logging a directory could result in losing a
6452 * a file after replaying the log. For example, if we move a file from a
6453 * directory A to a directory B, then fsync directory A, we have no way
6454 * to known the file was moved from A to B, so logging just A would
6455 * result in losing the file after a log replay.
6457 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6458 btrfs_set_log_full_commit(trans);
6459 ret = BTRFS_LOG_FORCE_COMMIT;
6464 * a brute force approach to making sure we get the most uptodate
6465 * copies of everything.
6467 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6468 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6469 if (ctx->logged_before)
6470 ret = drop_inode_items(trans, log, path, inode,
6471 BTRFS_XATTR_ITEM_KEY);
6473 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6475 * Make sure the new inode item we write to the log has
6476 * the same isize as the current one (if it exists).
6477 * This is necessary to prevent data loss after log
6478 * replay, and also to prevent doing a wrong expanding
6479 * truncate - for e.g. create file, write 4K into offset
6480 * 0, fsync, write 4K into offset 4096, add hard link,
6481 * fsync some other file (to sync log), power fail - if
6482 * we use the inode's current i_size, after log replay
6483 * we get a 8Kb file, with the last 4Kb extent as a hole
6484 * (zeroes), as if an expanding truncate happened,
6485 * instead of getting a file of 4Kb only.
6487 ret = logged_inode_size(log, inode, path, &logged_isize);
6491 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6492 &inode->runtime_flags)) {
6493 if (inode_only == LOG_INODE_EXISTS) {
6494 max_key.type = BTRFS_XATTR_ITEM_KEY;
6495 if (ctx->logged_before)
6496 ret = drop_inode_items(trans, log, path,
6497 inode, max_key.type);
6499 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6500 &inode->runtime_flags);
6501 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6502 &inode->runtime_flags);
6503 if (ctx->logged_before)
6504 ret = truncate_inode_items(trans, log,
6507 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6508 &inode->runtime_flags) ||
6509 inode_only == LOG_INODE_EXISTS) {
6510 if (inode_only == LOG_INODE_ALL)
6512 max_key.type = BTRFS_XATTR_ITEM_KEY;
6513 if (ctx->logged_before)
6514 ret = drop_inode_items(trans, log, path, inode,
6517 if (inode_only == LOG_INODE_ALL)
6519 inode_item_dropped = false;
6528 * If we are logging a directory in full mode, collect the delayed items
6529 * before iterating the subvolume tree, so that we don't miss any new
6530 * dir index items in case they get flushed while or right after we are
6531 * iterating the subvolume tree.
6533 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6534 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6537 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6538 path, dst_path, logged_isize,
6540 &need_log_inode_item);
6544 btrfs_release_path(path);
6545 btrfs_release_path(dst_path);
6546 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6549 xattrs_logged = true;
6550 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6551 btrfs_release_path(path);
6552 btrfs_release_path(dst_path);
6553 ret = btrfs_log_holes(trans, inode, path);
6558 btrfs_release_path(path);
6559 btrfs_release_path(dst_path);
6560 if (need_log_inode_item) {
6561 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6565 * If we are doing a fast fsync and the inode was logged before
6566 * in this transaction, we don't need to log the xattrs because
6567 * they were logged before. If xattrs were added, changed or
6568 * deleted since the last time we logged the inode, then we have
6569 * already logged them because the inode had the runtime flag
6570 * BTRFS_INODE_COPY_EVERYTHING set.
6572 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6573 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6576 btrfs_release_path(path);
6580 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6583 } else if (inode_only == LOG_INODE_ALL) {
6584 struct extent_map *em, *n;
6586 write_lock(&em_tree->lock);
6587 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6588 list_del_init(&em->list);
6589 write_unlock(&em_tree->lock);
6592 if (full_dir_logging) {
6593 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6596 ret = log_delayed_insertion_items(trans, inode, path,
6597 &delayed_ins_list, ctx);
6600 ret = log_delayed_deletion_items(trans, inode, path,
6601 &delayed_del_list, ctx);
6606 spin_lock(&inode->lock);
6607 inode->logged_trans = trans->transid;
6609 * Don't update last_log_commit if we logged that an inode exists.
6610 * We do this for three reasons:
6612 * 1) We might have had buffered writes to this inode that were
6613 * flushed and had their ordered extents completed in this
6614 * transaction, but we did not previously log the inode with
6615 * LOG_INODE_ALL. Later the inode was evicted and after that
6616 * it was loaded again and this LOG_INODE_EXISTS log operation
6617 * happened. We must make sure that if an explicit fsync against
6618 * the inode is performed later, it logs the new extents, an
6619 * updated inode item, etc, and syncs the log. The same logic
6620 * applies to direct IO writes instead of buffered writes.
6622 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6623 * is logged with an i_size of 0 or whatever value was logged
6624 * before. If later the i_size of the inode is increased by a
6625 * truncate operation, the log is synced through an fsync of
6626 * some other inode and then finally an explicit fsync against
6627 * this inode is made, we must make sure this fsync logs the
6628 * inode with the new i_size, the hole between old i_size and
6629 * the new i_size, and syncs the log.
6631 * 3) If we are logging that an ancestor inode exists as part of
6632 * logging a new name from a link or rename operation, don't update
6633 * its last_log_commit - otherwise if an explicit fsync is made
6634 * against an ancestor, the fsync considers the inode in the log
6635 * and doesn't sync the log, resulting in the ancestor missing after
6636 * a power failure unless the log was synced as part of an fsync
6637 * against any other unrelated inode.
6639 if (inode_only != LOG_INODE_EXISTS)
6640 inode->last_log_commit = inode->last_sub_trans;
6641 spin_unlock(&inode->lock);
6644 * Reset the last_reflink_trans so that the next fsync does not need to
6645 * go through the slower path when logging extents and their checksums.
6647 if (inode_only == LOG_INODE_ALL)
6648 inode->last_reflink_trans = 0;
6651 mutex_unlock(&inode->log_mutex);
6653 btrfs_free_path(path);
6654 btrfs_free_path(dst_path);
6657 free_conflicting_inodes(ctx);
6659 ret = log_conflicting_inodes(trans, inode->root, ctx);
6661 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6663 ret = log_new_delayed_dentries(trans, inode,
6664 &delayed_ins_list, ctx);
6666 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6673 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6674 struct btrfs_inode *inode,
6675 struct btrfs_log_ctx *ctx)
6677 struct btrfs_fs_info *fs_info = trans->fs_info;
6679 struct btrfs_path *path;
6680 struct btrfs_key key;
6681 struct btrfs_root *root = inode->root;
6682 const u64 ino = btrfs_ino(inode);
6684 path = btrfs_alloc_path();
6687 path->skip_locking = 1;
6688 path->search_commit_root = 1;
6691 key.type = BTRFS_INODE_REF_KEY;
6693 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6698 struct extent_buffer *leaf = path->nodes[0];
6699 int slot = path->slots[0];
6704 if (slot >= btrfs_header_nritems(leaf)) {
6705 ret = btrfs_next_leaf(root, path);
6713 btrfs_item_key_to_cpu(leaf, &key, slot);
6714 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6715 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6718 item_size = btrfs_item_size(leaf, slot);
6719 ptr = btrfs_item_ptr_offset(leaf, slot);
6720 while (cur_offset < item_size) {
6721 struct btrfs_key inode_key;
6722 struct inode *dir_inode;
6724 inode_key.type = BTRFS_INODE_ITEM_KEY;
6725 inode_key.offset = 0;
6727 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6728 struct btrfs_inode_extref *extref;
6730 extref = (struct btrfs_inode_extref *)
6732 inode_key.objectid = btrfs_inode_extref_parent(
6734 cur_offset += sizeof(*extref);
6735 cur_offset += btrfs_inode_extref_name_len(leaf,
6738 inode_key.objectid = key.offset;
6739 cur_offset = item_size;
6742 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6745 * If the parent inode was deleted, return an error to
6746 * fallback to a transaction commit. This is to prevent
6747 * getting an inode that was moved from one parent A to
6748 * a parent B, got its former parent A deleted and then
6749 * it got fsync'ed, from existing at both parents after
6750 * a log replay (and the old parent still existing).
6757 * mv /mnt/B/bar /mnt/A/bar
6758 * mv -T /mnt/A /mnt/B
6762 * If we ignore the old parent B which got deleted,
6763 * after a log replay we would have file bar linked
6764 * at both parents and the old parent B would still
6767 if (IS_ERR(dir_inode)) {
6768 ret = PTR_ERR(dir_inode);
6772 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6773 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6777 ctx->log_new_dentries = false;
6778 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6779 LOG_INODE_ALL, ctx);
6780 if (!ret && ctx->log_new_dentries)
6781 ret = log_new_dir_dentries(trans,
6782 BTRFS_I(dir_inode), ctx);
6783 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6791 btrfs_free_path(path);
6795 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6796 struct btrfs_root *root,
6797 struct btrfs_path *path,
6798 struct btrfs_log_ctx *ctx)
6800 struct btrfs_key found_key;
6802 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6805 struct btrfs_fs_info *fs_info = root->fs_info;
6806 struct extent_buffer *leaf = path->nodes[0];
6807 int slot = path->slots[0];
6808 struct btrfs_key search_key;
6809 struct inode *inode;
6813 btrfs_release_path(path);
6815 ino = found_key.offset;
6817 search_key.objectid = found_key.offset;
6818 search_key.type = BTRFS_INODE_ITEM_KEY;
6819 search_key.offset = 0;
6820 inode = btrfs_iget(fs_info->sb, ino, root);
6822 return PTR_ERR(inode);
6824 if (BTRFS_I(inode)->generation >= trans->transid &&
6825 need_log_inode(trans, BTRFS_I(inode)))
6826 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6827 LOG_INODE_EXISTS, ctx);
6828 btrfs_add_delayed_iput(BTRFS_I(inode));
6832 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6835 search_key.type = BTRFS_INODE_REF_KEY;
6836 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6840 leaf = path->nodes[0];
6841 slot = path->slots[0];
6842 if (slot >= btrfs_header_nritems(leaf)) {
6843 ret = btrfs_next_leaf(root, path);
6848 leaf = path->nodes[0];
6849 slot = path->slots[0];
6852 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6853 if (found_key.objectid != search_key.objectid ||
6854 found_key.type != BTRFS_INODE_REF_KEY)
6860 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6861 struct btrfs_inode *inode,
6862 struct dentry *parent,
6863 struct btrfs_log_ctx *ctx)
6865 struct btrfs_root *root = inode->root;
6866 struct dentry *old_parent = NULL;
6867 struct super_block *sb = inode->vfs_inode.i_sb;
6871 if (!parent || d_really_is_negative(parent) ||
6875 inode = BTRFS_I(d_inode(parent));
6876 if (root != inode->root)
6879 if (inode->generation >= trans->transid &&
6880 need_log_inode(trans, inode)) {
6881 ret = btrfs_log_inode(trans, inode,
6882 LOG_INODE_EXISTS, ctx);
6886 if (IS_ROOT(parent))
6889 parent = dget_parent(parent);
6891 old_parent = parent;
6898 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6899 struct btrfs_inode *inode,
6900 struct dentry *parent,
6901 struct btrfs_log_ctx *ctx)
6903 struct btrfs_root *root = inode->root;
6904 const u64 ino = btrfs_ino(inode);
6905 struct btrfs_path *path;
6906 struct btrfs_key search_key;
6910 * For a single hard link case, go through a fast path that does not
6911 * need to iterate the fs/subvolume tree.
6913 if (inode->vfs_inode.i_nlink < 2)
6914 return log_new_ancestors_fast(trans, inode, parent, ctx);
6916 path = btrfs_alloc_path();
6920 search_key.objectid = ino;
6921 search_key.type = BTRFS_INODE_REF_KEY;
6922 search_key.offset = 0;
6924 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6931 struct extent_buffer *leaf = path->nodes[0];
6932 int slot = path->slots[0];
6933 struct btrfs_key found_key;
6935 if (slot >= btrfs_header_nritems(leaf)) {
6936 ret = btrfs_next_leaf(root, path);
6944 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6945 if (found_key.objectid != ino ||
6946 found_key.type > BTRFS_INODE_EXTREF_KEY)
6950 * Don't deal with extended references because they are rare
6951 * cases and too complex to deal with (we would need to keep
6952 * track of which subitem we are processing for each item in
6953 * this loop, etc). So just return some error to fallback to
6954 * a transaction commit.
6956 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6962 * Logging ancestors needs to do more searches on the fs/subvol
6963 * tree, so it releases the path as needed to avoid deadlocks.
6964 * Keep track of the last inode ref key and resume from that key
6965 * after logging all new ancestors for the current hard link.
6967 memcpy(&search_key, &found_key, sizeof(search_key));
6969 ret = log_new_ancestors(trans, root, path, ctx);
6972 btrfs_release_path(path);
6977 btrfs_free_path(path);
6982 * helper function around btrfs_log_inode to make sure newly created
6983 * parent directories also end up in the log. A minimal inode and backref
6984 * only logging is done of any parent directories that are older than
6985 * the last committed transaction
6987 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6988 struct btrfs_inode *inode,
6989 struct dentry *parent,
6991 struct btrfs_log_ctx *ctx)
6993 struct btrfs_root *root = inode->root;
6994 struct btrfs_fs_info *fs_info = root->fs_info;
6996 bool log_dentries = false;
6998 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6999 ret = BTRFS_LOG_FORCE_COMMIT;
7003 if (btrfs_root_refs(&root->root_item) == 0) {
7004 ret = BTRFS_LOG_FORCE_COMMIT;
7009 * Skip already logged inodes or inodes corresponding to tmpfiles
7010 * (since logging them is pointless, a link count of 0 means they
7011 * will never be accessible).
7013 if ((btrfs_inode_in_log(inode, trans->transid) &&
7014 list_empty(&ctx->ordered_extents)) ||
7015 inode->vfs_inode.i_nlink == 0) {
7016 ret = BTRFS_NO_LOG_SYNC;
7020 ret = start_log_trans(trans, root, ctx);
7024 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
7029 * for regular files, if its inode is already on disk, we don't
7030 * have to worry about the parents at all. This is because
7031 * we can use the last_unlink_trans field to record renames
7032 * and other fun in this file.
7034 if (S_ISREG(inode->vfs_inode.i_mode) &&
7035 inode->generation < trans->transid &&
7036 inode->last_unlink_trans < trans->transid) {
7041 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7042 log_dentries = true;
7045 * On unlink we must make sure all our current and old parent directory
7046 * inodes are fully logged. This is to prevent leaving dangling
7047 * directory index entries in directories that were our parents but are
7048 * not anymore. Not doing this results in old parent directory being
7049 * impossible to delete after log replay (rmdir will always fail with
7050 * error -ENOTEMPTY).
7056 * ln testdir/foo testdir/bar
7058 * unlink testdir/bar
7059 * xfs_io -c fsync testdir/foo
7061 * mount fs, triggers log replay
7063 * If we don't log the parent directory (testdir), after log replay the
7064 * directory still has an entry pointing to the file inode using the bar
7065 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7066 * the file inode has a link count of 1.
7072 * ln foo testdir/foo2
7073 * ln foo testdir/foo3
7075 * unlink testdir/foo3
7076 * xfs_io -c fsync foo
7078 * mount fs, triggers log replay
7080 * Similar as the first example, after log replay the parent directory
7081 * testdir still has an entry pointing to the inode file with name foo3
7082 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7083 * and has a link count of 2.
7085 if (inode->last_unlink_trans >= trans->transid) {
7086 ret = btrfs_log_all_parents(trans, inode, ctx);
7091 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7096 ret = log_new_dir_dentries(trans, inode, ctx);
7101 btrfs_set_log_full_commit(trans);
7102 ret = BTRFS_LOG_FORCE_COMMIT;
7106 btrfs_remove_log_ctx(root, ctx);
7107 btrfs_end_log_trans(root);
7113 * it is not safe to log dentry if the chunk root has added new
7114 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7115 * If this returns 1, you must commit the transaction to safely get your
7118 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7119 struct dentry *dentry,
7120 struct btrfs_log_ctx *ctx)
7122 struct dentry *parent = dget_parent(dentry);
7125 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7126 LOG_INODE_ALL, ctx);
7133 * should be called during mount to recover any replay any log trees
7136 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7139 struct btrfs_path *path;
7140 struct btrfs_trans_handle *trans;
7141 struct btrfs_key key;
7142 struct btrfs_key found_key;
7143 struct btrfs_root *log;
7144 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7145 struct walk_control wc = {
7146 .process_func = process_one_buffer,
7147 .stage = LOG_WALK_PIN_ONLY,
7150 path = btrfs_alloc_path();
7154 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7156 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7157 if (IS_ERR(trans)) {
7158 ret = PTR_ERR(trans);
7165 ret = walk_log_tree(trans, log_root_tree, &wc);
7167 btrfs_abort_transaction(trans, ret);
7172 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7173 key.offset = (u64)-1;
7174 key.type = BTRFS_ROOT_ITEM_KEY;
7177 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7180 btrfs_abort_transaction(trans, ret);
7184 if (path->slots[0] == 0)
7188 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7190 btrfs_release_path(path);
7191 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7194 log = btrfs_read_tree_root(log_root_tree, &found_key);
7197 btrfs_abort_transaction(trans, ret);
7201 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7203 if (IS_ERR(wc.replay_dest)) {
7204 ret = PTR_ERR(wc.replay_dest);
7207 * We didn't find the subvol, likely because it was
7208 * deleted. This is ok, simply skip this log and go to
7211 * We need to exclude the root because we can't have
7212 * other log replays overwriting this log as we'll read
7213 * it back in a few more times. This will keep our
7214 * block from being modified, and we'll just bail for
7215 * each subsequent pass.
7218 ret = btrfs_pin_extent_for_log_replay(trans,
7221 btrfs_put_root(log);
7225 btrfs_abort_transaction(trans, ret);
7229 wc.replay_dest->log_root = log;
7230 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7232 /* The loop needs to continue due to the root refs */
7233 btrfs_abort_transaction(trans, ret);
7235 ret = walk_log_tree(trans, log, &wc);
7237 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7238 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7241 btrfs_abort_transaction(trans, ret);
7244 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7245 struct btrfs_root *root = wc.replay_dest;
7247 btrfs_release_path(path);
7250 * We have just replayed everything, and the highest
7251 * objectid of fs roots probably has changed in case
7252 * some inode_item's got replayed.
7254 * root->objectid_mutex is not acquired as log replay
7255 * could only happen during mount.
7257 ret = btrfs_init_root_free_objectid(root);
7259 btrfs_abort_transaction(trans, ret);
7262 wc.replay_dest->log_root = NULL;
7263 btrfs_put_root(wc.replay_dest);
7264 btrfs_put_root(log);
7269 if (found_key.offset == 0)
7271 key.offset = found_key.offset - 1;
7273 btrfs_release_path(path);
7275 /* step one is to pin it all, step two is to replay just inodes */
7278 wc.process_func = replay_one_buffer;
7279 wc.stage = LOG_WALK_REPLAY_INODES;
7282 /* step three is to replay everything */
7283 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7288 btrfs_free_path(path);
7290 /* step 4: commit the transaction, which also unpins the blocks */
7291 ret = btrfs_commit_transaction(trans);
7295 log_root_tree->log_root = NULL;
7296 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7297 btrfs_put_root(log_root_tree);
7302 btrfs_end_transaction(wc.trans);
7303 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7304 btrfs_free_path(path);
7309 * there are some corner cases where we want to force a full
7310 * commit instead of allowing a directory to be logged.
7312 * They revolve around files there were unlinked from the directory, and
7313 * this function updates the parent directory so that a full commit is
7314 * properly done if it is fsync'd later after the unlinks are done.
7316 * Must be called before the unlink operations (updates to the subvolume tree,
7317 * inodes, etc) are done.
7319 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7320 struct btrfs_inode *dir, struct btrfs_inode *inode,
7324 * when we're logging a file, if it hasn't been renamed
7325 * or unlinked, and its inode is fully committed on disk,
7326 * we don't have to worry about walking up the directory chain
7327 * to log its parents.
7329 * So, we use the last_unlink_trans field to put this transid
7330 * into the file. When the file is logged we check it and
7331 * don't log the parents if the file is fully on disk.
7333 mutex_lock(&inode->log_mutex);
7334 inode->last_unlink_trans = trans->transid;
7335 mutex_unlock(&inode->log_mutex);
7338 * if this directory was already logged any new
7339 * names for this file/dir will get recorded
7341 if (dir->logged_trans == trans->transid)
7345 * if the inode we're about to unlink was logged,
7346 * the log will be properly updated for any new names
7348 if (inode->logged_trans == trans->transid)
7352 * when renaming files across directories, if the directory
7353 * there we're unlinking from gets fsync'd later on, there's
7354 * no way to find the destination directory later and fsync it
7355 * properly. So, we have to be conservative and force commits
7356 * so the new name gets discovered.
7361 /* we can safely do the unlink without any special recording */
7365 mutex_lock(&dir->log_mutex);
7366 dir->last_unlink_trans = trans->transid;
7367 mutex_unlock(&dir->log_mutex);
7371 * Make sure that if someone attempts to fsync the parent directory of a deleted
7372 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7373 * that after replaying the log tree of the parent directory's root we will not
7374 * see the snapshot anymore and at log replay time we will not see any log tree
7375 * corresponding to the deleted snapshot's root, which could lead to replaying
7376 * it after replaying the log tree of the parent directory (which would replay
7377 * the snapshot delete operation).
7379 * Must be called before the actual snapshot destroy operation (updates to the
7380 * parent root and tree of tree roots trees, etc) are done.
7382 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7383 struct btrfs_inode *dir)
7385 mutex_lock(&dir->log_mutex);
7386 dir->last_unlink_trans = trans->transid;
7387 mutex_unlock(&dir->log_mutex);
7391 * Update the log after adding a new name for an inode.
7393 * @trans: Transaction handle.
7394 * @old_dentry: The dentry associated with the old name and the old
7396 * @old_dir: The inode of the previous parent directory for the case
7397 * of a rename. For a link operation, it must be NULL.
7398 * @old_dir_index: The index number associated with the old name, meaningful
7399 * only for rename operations (when @old_dir is not NULL).
7400 * Ignored for link operations.
7401 * @parent: The dentry associated with the directory under which the
7402 * new name is located.
7404 * Call this after adding a new name for an inode, as a result of a link or
7405 * rename operation, and it will properly update the log to reflect the new name.
7407 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7408 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7409 u64 old_dir_index, struct dentry *parent)
7411 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7412 struct btrfs_root *root = inode->root;
7413 struct btrfs_log_ctx ctx;
7414 bool log_pinned = false;
7418 * this will force the logging code to walk the dentry chain
7421 if (!S_ISDIR(inode->vfs_inode.i_mode))
7422 inode->last_unlink_trans = trans->transid;
7425 * if this inode hasn't been logged and directory we're renaming it
7426 * from hasn't been logged, we don't need to log it
7428 ret = inode_logged(trans, inode, NULL);
7431 } else if (ret == 0) {
7435 * If the inode was not logged and we are doing a rename (old_dir is not
7436 * NULL), check if old_dir was logged - if it was not we can return and
7439 ret = inode_logged(trans, old_dir, NULL);
7448 * If we are doing a rename (old_dir is not NULL) from a directory that
7449 * was previously logged, make sure that on log replay we get the old
7450 * dir entry deleted. This is needed because we will also log the new
7451 * name of the renamed inode, so we need to make sure that after log
7452 * replay we don't end up with both the new and old dir entries existing.
7454 if (old_dir && old_dir->logged_trans == trans->transid) {
7455 struct btrfs_root *log = old_dir->root->log_root;
7456 struct btrfs_path *path;
7457 struct fscrypt_name fname;
7459 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7461 ret = fscrypt_setup_filename(&old_dir->vfs_inode,
7462 &old_dentry->d_name, 0, &fname);
7466 * We have two inodes to update in the log, the old directory and
7467 * the inode that got renamed, so we must pin the log to prevent
7468 * anyone from syncing the log until we have updated both inodes
7471 ret = join_running_log_trans(root);
7473 * At least one of the inodes was logged before, so this should
7474 * not fail, but if it does, it's not serious, just bail out and
7475 * mark the log for a full commit.
7477 if (WARN_ON_ONCE(ret < 0)) {
7478 fscrypt_free_filename(&fname);
7484 path = btrfs_alloc_path();
7487 fscrypt_free_filename(&fname);
7492 * Other concurrent task might be logging the old directory,
7493 * as it can be triggered when logging other inode that had or
7494 * still has a dentry in the old directory. We lock the old
7495 * directory's log_mutex to ensure the deletion of the old
7496 * name is persisted, because during directory logging we
7497 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7498 * the old name's dir index item is in the delayed items, so
7499 * it could be missed by an in progress directory logging.
7501 mutex_lock(&old_dir->log_mutex);
7502 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7503 &fname.disk_name, old_dir_index);
7506 * The dentry does not exist in the log, so record its
7509 btrfs_release_path(path);
7510 ret = insert_dir_log_key(trans, log, path,
7512 old_dir_index, old_dir_index);
7514 mutex_unlock(&old_dir->log_mutex);
7516 btrfs_free_path(path);
7517 fscrypt_free_filename(&fname);
7522 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7523 ctx.logging_new_name = true;
7525 * We don't care about the return value. If we fail to log the new name
7526 * then we know the next attempt to sync the log will fallback to a full
7527 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7528 * we don't need to worry about getting a log committed that has an
7529 * inconsistent state after a rename operation.
7531 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7532 ASSERT(list_empty(&ctx.conflict_inodes));
7535 * If an error happened mark the log for a full commit because it's not
7536 * consistent and up to date or we couldn't find out if one of the
7537 * inodes was logged before in this transaction. Do it before unpinning
7538 * the log, to avoid any races with someone else trying to commit it.
7541 btrfs_set_log_full_commit(trans);
7543 btrfs_end_log_trans(root);