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 #define MAX_CONFLICT_INODES 10
27 /* magic values for the inode_only field in btrfs_log_inode:
29 * LOG_INODE_ALL means to log everything
30 * LOG_INODE_EXISTS means to log just enough to recreate the inode
39 * directory trouble cases
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
48 * rename foo/some_dir foo2/some_dir
50 * fsync foo/some_dir/some_file
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
82 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
92 LOG_WALK_REPLAY_INODES,
93 LOG_WALK_REPLAY_DIR_INDEX,
97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
98 struct btrfs_inode *inode,
100 struct btrfs_log_ctx *ctx);
101 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
102 struct btrfs_root *root,
103 struct btrfs_path *path, u64 objectid);
104 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root,
106 struct btrfs_root *log,
107 struct btrfs_path *path,
108 u64 dirid, int del_all);
109 static void wait_log_commit(struct btrfs_root *root, int transid);
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
139 static int start_log_trans(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root,
141 struct btrfs_log_ctx *ctx)
143 struct btrfs_fs_info *fs_info = root->fs_info;
144 struct btrfs_root *tree_root = fs_info->tree_root;
145 const bool zoned = btrfs_is_zoned(fs_info);
147 bool created = false;
150 * First check if the log root tree was already created. If not, create
151 * it before locking the root's log_mutex, just to keep lockdep happy.
153 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
154 mutex_lock(&tree_root->log_mutex);
155 if (!fs_info->log_root_tree) {
156 ret = btrfs_init_log_root_tree(trans, fs_info);
158 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
162 mutex_unlock(&tree_root->log_mutex);
167 mutex_lock(&root->log_mutex);
170 if (root->log_root) {
171 int index = (root->log_transid + 1) % 2;
173 if (btrfs_need_log_full_commit(trans)) {
174 ret = BTRFS_LOG_FORCE_COMMIT;
178 if (zoned && atomic_read(&root->log_commit[index])) {
179 wait_log_commit(root, root->log_transid - 1);
183 if (!root->log_start_pid) {
184 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
185 root->log_start_pid = current->pid;
186 } else if (root->log_start_pid != current->pid) {
187 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
191 * This means fs_info->log_root_tree was already created
192 * for some other FS trees. Do the full commit not to mix
193 * nodes from multiple log transactions to do sequential
196 if (zoned && !created) {
197 ret = BTRFS_LOG_FORCE_COMMIT;
201 ret = btrfs_add_log_tree(trans, root);
205 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
206 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
207 root->log_start_pid = current->pid;
210 atomic_inc(&root->log_writers);
211 if (!ctx->logging_new_name) {
212 int index = root->log_transid % 2;
213 list_add_tail(&ctx->list, &root->log_ctxs[index]);
214 ctx->log_transid = root->log_transid;
218 mutex_unlock(&root->log_mutex);
223 * returns 0 if there was a log transaction running and we were able
224 * to join, or returns -ENOENT if there were not transactions
227 static int join_running_log_trans(struct btrfs_root *root)
229 const bool zoned = btrfs_is_zoned(root->fs_info);
232 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
235 mutex_lock(&root->log_mutex);
237 if (root->log_root) {
238 int index = (root->log_transid + 1) % 2;
241 if (zoned && atomic_read(&root->log_commit[index])) {
242 wait_log_commit(root, root->log_transid - 1);
245 atomic_inc(&root->log_writers);
247 mutex_unlock(&root->log_mutex);
252 * This either makes the current running log transaction wait
253 * until you call btrfs_end_log_trans() or it makes any future
254 * log transactions wait until you call btrfs_end_log_trans()
256 void btrfs_pin_log_trans(struct btrfs_root *root)
258 atomic_inc(&root->log_writers);
262 * indicate we're done making changes to the log tree
263 * and wake up anyone waiting to do a sync
265 void btrfs_end_log_trans(struct btrfs_root *root)
267 if (atomic_dec_and_test(&root->log_writers)) {
268 /* atomic_dec_and_test implies a barrier */
269 cond_wake_up_nomb(&root->log_writer_wait);
273 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
275 filemap_fdatawait_range(buf->pages[0]->mapping,
276 buf->start, buf->start + buf->len - 1);
280 * the walk control struct is used to pass state down the chain when
281 * processing the log tree. The stage field tells us which part
282 * of the log tree processing we are currently doing. The others
283 * are state fields used for that specific part
285 struct walk_control {
286 /* should we free the extent on disk when done? This is used
287 * at transaction commit time while freeing a log tree
291 /* pin only walk, we record which extents on disk belong to the
296 /* what stage of the replay code we're currently in */
300 * Ignore any items from the inode currently being processed. Needs
301 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
302 * the LOG_WALK_REPLAY_INODES stage.
304 bool ignore_cur_inode;
306 /* the root we are currently replaying */
307 struct btrfs_root *replay_dest;
309 /* the trans handle for the current replay */
310 struct btrfs_trans_handle *trans;
312 /* the function that gets used to process blocks we find in the
313 * tree. Note the extent_buffer might not be up to date when it is
314 * passed in, and it must be checked or read if you need the data
317 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
318 struct walk_control *wc, u64 gen, int level);
322 * process_func used to pin down extents, write them or wait on them
324 static int process_one_buffer(struct btrfs_root *log,
325 struct extent_buffer *eb,
326 struct walk_control *wc, u64 gen, int level)
328 struct btrfs_fs_info *fs_info = log->fs_info;
332 * If this fs is mixed then we need to be able to process the leaves to
333 * pin down any logged extents, so we have to read the block.
335 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
336 ret = btrfs_read_extent_buffer(eb, gen, level, NULL);
342 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
347 if (btrfs_buffer_uptodate(eb, gen, 0) &&
348 btrfs_header_level(eb) == 0)
349 ret = btrfs_exclude_logged_extents(eb);
354 static int do_overwrite_item(struct btrfs_trans_handle *trans,
355 struct btrfs_root *root,
356 struct btrfs_path *path,
357 struct extent_buffer *eb, int slot,
358 struct btrfs_key *key)
362 u64 saved_i_size = 0;
363 int save_old_i_size = 0;
364 unsigned long src_ptr;
365 unsigned long dst_ptr;
366 int overwrite_root = 0;
367 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
369 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
372 item_size = btrfs_item_size(eb, slot);
373 src_ptr = btrfs_item_ptr_offset(eb, slot);
375 /* Our caller must have done a search for the key for us. */
376 ASSERT(path->nodes[0] != NULL);
379 * And the slot must point to the exact key or the slot where the key
380 * should be at (the first item with a key greater than 'key')
382 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
383 struct btrfs_key found_key;
385 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
386 ret = btrfs_comp_cpu_keys(&found_key, key);
395 u32 dst_size = btrfs_item_size(path->nodes[0],
397 if (dst_size != item_size)
400 if (item_size == 0) {
401 btrfs_release_path(path);
404 dst_copy = kmalloc(item_size, GFP_NOFS);
405 src_copy = kmalloc(item_size, GFP_NOFS);
406 if (!dst_copy || !src_copy) {
407 btrfs_release_path(path);
413 read_extent_buffer(eb, src_copy, src_ptr, item_size);
415 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
416 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
418 ret = memcmp(dst_copy, src_copy, item_size);
423 * they have the same contents, just return, this saves
424 * us from cowing blocks in the destination tree and doing
425 * extra writes that may not have been done by a previous
429 btrfs_release_path(path);
434 * We need to load the old nbytes into the inode so when we
435 * replay the extents we've logged we get the right nbytes.
438 struct btrfs_inode_item *item;
442 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
443 struct btrfs_inode_item);
444 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
445 item = btrfs_item_ptr(eb, slot,
446 struct btrfs_inode_item);
447 btrfs_set_inode_nbytes(eb, item, nbytes);
450 * If this is a directory we need to reset the i_size to
451 * 0 so that we can set it up properly when replaying
452 * the rest of the items in this log.
454 mode = btrfs_inode_mode(eb, item);
456 btrfs_set_inode_size(eb, item, 0);
458 } else if (inode_item) {
459 struct btrfs_inode_item *item;
463 * New inode, set nbytes to 0 so that the nbytes comes out
464 * properly when we replay the extents.
466 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
467 btrfs_set_inode_nbytes(eb, item, 0);
470 * If this is a directory we need to reset the i_size to 0 so
471 * that we can set it up properly when replaying the rest of
472 * the items in this log.
474 mode = btrfs_inode_mode(eb, item);
476 btrfs_set_inode_size(eb, item, 0);
479 btrfs_release_path(path);
480 /* try to insert the key into the destination tree */
481 path->skip_release_on_error = 1;
482 ret = btrfs_insert_empty_item(trans, root, path,
484 path->skip_release_on_error = 0;
486 /* make sure any existing item is the correct size */
487 if (ret == -EEXIST || ret == -EOVERFLOW) {
489 found_size = btrfs_item_size(path->nodes[0],
491 if (found_size > item_size)
492 btrfs_truncate_item(path, item_size, 1);
493 else if (found_size < item_size)
494 btrfs_extend_item(path, item_size - found_size);
498 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
501 /* don't overwrite an existing inode if the generation number
502 * was logged as zero. This is done when the tree logging code
503 * is just logging an inode to make sure it exists after recovery.
505 * Also, don't overwrite i_size on directories during replay.
506 * log replay inserts and removes directory items based on the
507 * state of the tree found in the subvolume, and i_size is modified
510 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
511 struct btrfs_inode_item *src_item;
512 struct btrfs_inode_item *dst_item;
514 src_item = (struct btrfs_inode_item *)src_ptr;
515 dst_item = (struct btrfs_inode_item *)dst_ptr;
517 if (btrfs_inode_generation(eb, src_item) == 0) {
518 struct extent_buffer *dst_eb = path->nodes[0];
519 const u64 ino_size = btrfs_inode_size(eb, src_item);
522 * For regular files an ino_size == 0 is used only when
523 * logging that an inode exists, as part of a directory
524 * fsync, and the inode wasn't fsynced before. In this
525 * case don't set the size of the inode in the fs/subvol
526 * tree, otherwise we would be throwing valid data away.
528 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
529 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
531 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
535 if (overwrite_root &&
536 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
537 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
539 saved_i_size = btrfs_inode_size(path->nodes[0],
544 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
547 if (save_old_i_size) {
548 struct btrfs_inode_item *dst_item;
549 dst_item = (struct btrfs_inode_item *)dst_ptr;
550 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
553 /* make sure the generation is filled in */
554 if (key->type == BTRFS_INODE_ITEM_KEY) {
555 struct btrfs_inode_item *dst_item;
556 dst_item = (struct btrfs_inode_item *)dst_ptr;
557 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
558 btrfs_set_inode_generation(path->nodes[0], dst_item,
563 btrfs_mark_buffer_dirty(path->nodes[0]);
564 btrfs_release_path(path);
569 * Item overwrite used by replay and tree logging. eb, slot and key all refer
570 * to the src data we are copying out.
572 * root is the tree we are copying into, and path is a scratch
573 * path for use in this function (it should be released on entry and
574 * will be released on exit).
576 * If the key is already in the destination tree the existing item is
577 * overwritten. If the existing item isn't big enough, it is extended.
578 * If it is too large, it is truncated.
580 * If the key isn't in the destination yet, a new item is inserted.
582 static int overwrite_item(struct btrfs_trans_handle *trans,
583 struct btrfs_root *root,
584 struct btrfs_path *path,
585 struct extent_buffer *eb, int slot,
586 struct btrfs_key *key)
590 /* Look for the key in the destination tree. */
591 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
595 return do_overwrite_item(trans, root, path, eb, slot, key);
599 * simple helper to read an inode off the disk from a given root
600 * This can only be called for subvolume roots and not for the log
602 static noinline struct inode *read_one_inode(struct btrfs_root *root,
607 inode = btrfs_iget(root->fs_info->sb, objectid, root);
613 /* replays a single extent in 'eb' at 'slot' with 'key' into the
614 * subvolume 'root'. path is released on entry and should be released
617 * extents in the log tree have not been allocated out of the extent
618 * tree yet. So, this completes the allocation, taking a reference
619 * as required if the extent already exists or creating a new extent
620 * if it isn't in the extent allocation tree yet.
622 * The extent is inserted into the file, dropping any existing extents
623 * from the file that overlap the new one.
625 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
626 struct btrfs_root *root,
627 struct btrfs_path *path,
628 struct extent_buffer *eb, int slot,
629 struct btrfs_key *key)
631 struct btrfs_drop_extents_args drop_args = { 0 };
632 struct btrfs_fs_info *fs_info = root->fs_info;
635 u64 start = key->offset;
637 struct btrfs_file_extent_item *item;
638 struct inode *inode = NULL;
642 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
643 found_type = btrfs_file_extent_type(eb, item);
645 if (found_type == BTRFS_FILE_EXTENT_REG ||
646 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
647 nbytes = btrfs_file_extent_num_bytes(eb, item);
648 extent_end = start + nbytes;
651 * We don't add to the inodes nbytes if we are prealloc or a
654 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
656 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
657 size = btrfs_file_extent_ram_bytes(eb, item);
658 nbytes = btrfs_file_extent_ram_bytes(eb, item);
659 extent_end = ALIGN(start + size,
660 fs_info->sectorsize);
666 inode = read_one_inode(root, key->objectid);
673 * first check to see if we already have this extent in the
674 * file. This must be done before the btrfs_drop_extents run
675 * so we don't try to drop this extent.
677 ret = btrfs_lookup_file_extent(trans, root, path,
678 btrfs_ino(BTRFS_I(inode)), start, 0);
681 (found_type == BTRFS_FILE_EXTENT_REG ||
682 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
683 struct btrfs_file_extent_item cmp1;
684 struct btrfs_file_extent_item cmp2;
685 struct btrfs_file_extent_item *existing;
686 struct extent_buffer *leaf;
688 leaf = path->nodes[0];
689 existing = btrfs_item_ptr(leaf, path->slots[0],
690 struct btrfs_file_extent_item);
692 read_extent_buffer(eb, &cmp1, (unsigned long)item,
694 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
698 * we already have a pointer to this exact extent,
699 * we don't have to do anything
701 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
702 btrfs_release_path(path);
706 btrfs_release_path(path);
708 /* drop any overlapping extents */
709 drop_args.start = start;
710 drop_args.end = extent_end;
711 drop_args.drop_cache = true;
712 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
716 if (found_type == BTRFS_FILE_EXTENT_REG ||
717 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
719 unsigned long dest_offset;
720 struct btrfs_key ins;
722 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
723 btrfs_fs_incompat(fs_info, NO_HOLES))
726 ret = btrfs_insert_empty_item(trans, root, path, key,
730 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
732 copy_extent_buffer(path->nodes[0], eb, dest_offset,
733 (unsigned long)item, sizeof(*item));
735 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
736 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
737 ins.type = BTRFS_EXTENT_ITEM_KEY;
738 offset = key->offset - btrfs_file_extent_offset(eb, item);
741 * Manually record dirty extent, as here we did a shallow
742 * file extent item copy and skip normal backref update,
743 * but modifying extent tree all by ourselves.
744 * So need to manually record dirty extent for qgroup,
745 * as the owner of the file extent changed from log tree
746 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
748 ret = btrfs_qgroup_trace_extent(trans,
749 btrfs_file_extent_disk_bytenr(eb, item),
750 btrfs_file_extent_disk_num_bytes(eb, item),
755 if (ins.objectid > 0) {
756 struct btrfs_ref ref = { 0 };
759 LIST_HEAD(ordered_sums);
762 * is this extent already allocated in the extent
763 * allocation tree? If so, just add a reference
765 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
769 } else if (ret == 0) {
770 btrfs_init_generic_ref(&ref,
771 BTRFS_ADD_DELAYED_REF,
772 ins.objectid, ins.offset, 0);
773 btrfs_init_data_ref(&ref,
774 root->root_key.objectid,
775 key->objectid, offset, 0, false);
776 ret = btrfs_inc_extent_ref(trans, &ref);
781 * insert the extent pointer in the extent
784 ret = btrfs_alloc_logged_file_extent(trans,
785 root->root_key.objectid,
786 key->objectid, offset, &ins);
790 btrfs_release_path(path);
792 if (btrfs_file_extent_compression(eb, item)) {
793 csum_start = ins.objectid;
794 csum_end = csum_start + ins.offset;
796 csum_start = ins.objectid +
797 btrfs_file_extent_offset(eb, item);
798 csum_end = csum_start +
799 btrfs_file_extent_num_bytes(eb, item);
802 ret = btrfs_lookup_csums_range(root->log_root,
803 csum_start, csum_end - 1,
804 &ordered_sums, 0, false);
808 * Now delete all existing cums in the csum root that
809 * cover our range. We do this because we can have an
810 * extent that is completely referenced by one file
811 * extent item and partially referenced by another
812 * file extent item (like after using the clone or
813 * extent_same ioctls). In this case if we end up doing
814 * the replay of the one that partially references the
815 * extent first, and we do not do the csum deletion
816 * below, we can get 2 csum items in the csum tree that
817 * overlap each other. For example, imagine our log has
818 * the two following file extent items:
820 * key (257 EXTENT_DATA 409600)
821 * extent data disk byte 12845056 nr 102400
822 * extent data offset 20480 nr 20480 ram 102400
824 * key (257 EXTENT_DATA 819200)
825 * extent data disk byte 12845056 nr 102400
826 * extent data offset 0 nr 102400 ram 102400
828 * Where the second one fully references the 100K extent
829 * that starts at disk byte 12845056, and the log tree
830 * has a single csum item that covers the entire range
833 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
835 * After the first file extent item is replayed, the
836 * csum tree gets the following csum item:
838 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
840 * Which covers the 20K sub-range starting at offset 20K
841 * of our extent. Now when we replay the second file
842 * extent item, if we do not delete existing csum items
843 * that cover any of its blocks, we end up getting two
844 * csum items in our csum tree that overlap each other:
846 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
847 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
849 * Which is a problem, because after this anyone trying
850 * to lookup up for the checksum of any block of our
851 * extent starting at an offset of 40K or higher, will
852 * end up looking at the second csum item only, which
853 * does not contain the checksum for any block starting
854 * at offset 40K or higher of our extent.
856 while (!list_empty(&ordered_sums)) {
857 struct btrfs_ordered_sum *sums;
858 struct btrfs_root *csum_root;
860 sums = list_entry(ordered_sums.next,
861 struct btrfs_ordered_sum,
863 csum_root = btrfs_csum_root(fs_info,
866 ret = btrfs_del_csums(trans, csum_root,
870 ret = btrfs_csum_file_blocks(trans,
873 list_del(&sums->list);
879 btrfs_release_path(path);
881 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
882 /* inline extents are easy, we just overwrite them */
883 ret = overwrite_item(trans, root, path, eb, slot, key);
888 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
894 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
895 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
901 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
902 struct btrfs_inode *dir,
903 struct btrfs_inode *inode,
909 ret = btrfs_unlink_inode(trans, dir, inode, name, name_len);
913 * Whenever we need to check if a name exists or not, we check the
914 * fs/subvolume tree. So after an unlink we must run delayed items, so
915 * that future checks for a name during log replay see that the name
916 * does not exists anymore.
918 return btrfs_run_delayed_items(trans);
922 * when cleaning up conflicts between the directory names in the
923 * subvolume, directory names in the log and directory names in the
924 * inode back references, we may have to unlink inodes from directories.
926 * This is a helper function to do the unlink of a specific directory
929 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
930 struct btrfs_path *path,
931 struct btrfs_inode *dir,
932 struct btrfs_dir_item *di)
934 struct btrfs_root *root = dir->root;
938 struct extent_buffer *leaf;
939 struct btrfs_key location;
942 leaf = path->nodes[0];
944 btrfs_dir_item_key_to_cpu(leaf, di, &location);
945 name_len = btrfs_dir_name_len(leaf, di);
946 name = kmalloc(name_len, GFP_NOFS);
950 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
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,
972 * See if a given name and sequence number found in an inode back reference are
973 * already in a directory and correctly point to this inode.
975 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
978 static noinline int inode_in_dir(struct btrfs_root *root,
979 struct btrfs_path *path,
980 u64 dirid, u64 objectid, u64 index,
981 const char *name, int name_len)
983 struct btrfs_dir_item *di;
984 struct btrfs_key location;
987 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
988 index, name, name_len, 0);
993 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
994 if (location.objectid != objectid)
1000 btrfs_release_path(path);
1001 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
1006 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1007 if (location.objectid == objectid)
1011 btrfs_release_path(path);
1016 * helper function to check a log tree for a named back reference in
1017 * an inode. This is used to decide if a back reference that is
1018 * found in the subvolume conflicts with what we find in the log.
1020 * inode backreferences may have multiple refs in a single item,
1021 * during replay we process one reference at a time, and we don't
1022 * want to delete valid links to a file from the subvolume if that
1023 * link is also in the log.
1025 static noinline int backref_in_log(struct btrfs_root *log,
1026 struct btrfs_key *key,
1028 const char *name, int namelen)
1030 struct btrfs_path *path;
1033 path = btrfs_alloc_path();
1037 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1040 } else if (ret == 1) {
1045 if (key->type == BTRFS_INODE_EXTREF_KEY)
1046 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1051 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1055 btrfs_free_path(path);
1059 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1060 struct btrfs_root *root,
1061 struct btrfs_path *path,
1062 struct btrfs_root *log_root,
1063 struct btrfs_inode *dir,
1064 struct btrfs_inode *inode,
1065 u64 inode_objectid, u64 parent_objectid,
1066 u64 ref_index, char *name, int namelen)
1070 int victim_name_len;
1071 struct extent_buffer *leaf;
1072 struct btrfs_dir_item *di;
1073 struct btrfs_key search_key;
1074 struct btrfs_inode_extref *extref;
1077 /* Search old style refs */
1078 search_key.objectid = inode_objectid;
1079 search_key.type = BTRFS_INODE_REF_KEY;
1080 search_key.offset = parent_objectid;
1081 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1083 struct btrfs_inode_ref *victim_ref;
1085 unsigned long ptr_end;
1087 leaf = path->nodes[0];
1089 /* are we trying to overwrite a back ref for the root directory
1090 * if so, just jump out, we're done
1092 if (search_key.objectid == search_key.offset)
1095 /* check all the names in this back reference to see
1096 * if they are in the log. if so, we allow them to stay
1097 * otherwise they must be unlinked as a conflict
1099 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1100 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1101 while (ptr < ptr_end) {
1102 victim_ref = (struct btrfs_inode_ref *)ptr;
1103 victim_name_len = btrfs_inode_ref_name_len(leaf,
1105 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1109 read_extent_buffer(leaf, victim_name,
1110 (unsigned long)(victim_ref + 1),
1113 ret = backref_in_log(log_root, &search_key,
1114 parent_objectid, victim_name,
1120 inc_nlink(&inode->vfs_inode);
1121 btrfs_release_path(path);
1123 ret = unlink_inode_for_log_replay(trans, dir, inode,
1124 victim_name, victim_name_len);
1132 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1135 btrfs_release_path(path);
1137 /* Same search but for extended refs */
1138 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1139 inode_objectid, parent_objectid, 0,
1141 if (IS_ERR(extref)) {
1142 return PTR_ERR(extref);
1143 } else if (extref) {
1147 struct inode *victim_parent;
1149 leaf = path->nodes[0];
1151 item_size = btrfs_item_size(leaf, path->slots[0]);
1152 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1154 while (cur_offset < item_size) {
1155 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1157 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1159 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1162 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1165 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1168 search_key.objectid = inode_objectid;
1169 search_key.type = BTRFS_INODE_EXTREF_KEY;
1170 search_key.offset = btrfs_extref_hash(parent_objectid,
1173 ret = backref_in_log(log_root, &search_key,
1174 parent_objectid, victim_name,
1181 victim_parent = read_one_inode(root,
1183 if (victim_parent) {
1184 inc_nlink(&inode->vfs_inode);
1185 btrfs_release_path(path);
1187 ret = unlink_inode_for_log_replay(trans,
1188 BTRFS_I(victim_parent),
1193 iput(victim_parent);
1201 cur_offset += victim_name_len + sizeof(*extref);
1204 btrfs_release_path(path);
1206 /* look for a conflicting sequence number */
1207 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1208 ref_index, name, namelen, 0);
1212 ret = drop_one_dir_item(trans, path, dir, di);
1216 btrfs_release_path(path);
1218 /* look for a conflicting name */
1219 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1224 ret = drop_one_dir_item(trans, path, dir, di);
1228 btrfs_release_path(path);
1233 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1234 u32 *namelen, char **name, u64 *index,
1235 u64 *parent_objectid)
1237 struct btrfs_inode_extref *extref;
1239 extref = (struct btrfs_inode_extref *)ref_ptr;
1241 *namelen = btrfs_inode_extref_name_len(eb, extref);
1242 *name = kmalloc(*namelen, GFP_NOFS);
1246 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1250 *index = btrfs_inode_extref_index(eb, extref);
1251 if (parent_objectid)
1252 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1257 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1258 u32 *namelen, char **name, u64 *index)
1260 struct btrfs_inode_ref *ref;
1262 ref = (struct btrfs_inode_ref *)ref_ptr;
1264 *namelen = btrfs_inode_ref_name_len(eb, ref);
1265 *name = kmalloc(*namelen, GFP_NOFS);
1269 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1272 *index = btrfs_inode_ref_index(eb, ref);
1278 * Take an inode reference item from the log tree and iterate all names from the
1279 * inode reference item in the subvolume tree with the same key (if it exists).
1280 * For any name that is not in the inode reference item from the log tree, do a
1281 * proper unlink of that name (that is, remove its entry from the inode
1282 * reference item and both dir index keys).
1284 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1285 struct btrfs_root *root,
1286 struct btrfs_path *path,
1287 struct btrfs_inode *inode,
1288 struct extent_buffer *log_eb,
1290 struct btrfs_key *key)
1293 unsigned long ref_ptr;
1294 unsigned long ref_end;
1295 struct extent_buffer *eb;
1298 btrfs_release_path(path);
1299 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1307 eb = path->nodes[0];
1308 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1309 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1310 while (ref_ptr < ref_end) {
1315 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1316 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1319 parent_id = key->offset;
1320 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1326 if (key->type == BTRFS_INODE_EXTREF_KEY)
1327 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1331 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1337 btrfs_release_path(path);
1338 dir = read_one_inode(root, parent_id);
1344 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1345 inode, name, namelen);
1355 if (key->type == BTRFS_INODE_EXTREF_KEY)
1356 ref_ptr += sizeof(struct btrfs_inode_extref);
1358 ref_ptr += sizeof(struct btrfs_inode_ref);
1362 btrfs_release_path(path);
1367 * replay one inode back reference item found in the log tree.
1368 * eb, slot and key refer to the buffer and key found in the log tree.
1369 * root is the destination we are replaying into, and path is for temp
1370 * use by this function. (it should be released on return).
1372 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1373 struct btrfs_root *root,
1374 struct btrfs_root *log,
1375 struct btrfs_path *path,
1376 struct extent_buffer *eb, int slot,
1377 struct btrfs_key *key)
1379 struct inode *dir = NULL;
1380 struct inode *inode = NULL;
1381 unsigned long ref_ptr;
1382 unsigned long ref_end;
1386 int log_ref_ver = 0;
1387 u64 parent_objectid;
1390 int ref_struct_size;
1392 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1393 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1395 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1396 struct btrfs_inode_extref *r;
1398 ref_struct_size = sizeof(struct btrfs_inode_extref);
1400 r = (struct btrfs_inode_extref *)ref_ptr;
1401 parent_objectid = btrfs_inode_extref_parent(eb, r);
1403 ref_struct_size = sizeof(struct btrfs_inode_ref);
1404 parent_objectid = key->offset;
1406 inode_objectid = key->objectid;
1409 * it is possible that we didn't log all the parent directories
1410 * for a given inode. If we don't find the dir, just don't
1411 * copy the back ref in. The link count fixup code will take
1414 dir = read_one_inode(root, parent_objectid);
1420 inode = read_one_inode(root, inode_objectid);
1426 while (ref_ptr < ref_end) {
1428 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1429 &ref_index, &parent_objectid);
1431 * parent object can change from one array
1435 dir = read_one_inode(root, parent_objectid);
1441 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1447 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1448 btrfs_ino(BTRFS_I(inode)), ref_index,
1452 } else if (ret == 0) {
1454 * look for a conflicting back reference in the
1455 * metadata. if we find one we have to unlink that name
1456 * of the file before we add our new link. Later on, we
1457 * overwrite any existing back reference, and we don't
1458 * want to create dangling pointers in the directory.
1460 ret = __add_inode_ref(trans, root, path, log,
1461 BTRFS_I(dir), BTRFS_I(inode),
1462 inode_objectid, parent_objectid,
1463 ref_index, name, namelen);
1470 /* insert our name */
1471 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1472 name, namelen, 0, ref_index);
1476 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1480 /* Else, ret == 1, we already have a perfect match, we're done. */
1482 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1492 * Before we overwrite the inode reference item in the subvolume tree
1493 * with the item from the log tree, we must unlink all names from the
1494 * parent directory that are in the subvolume's tree inode reference
1495 * item, otherwise we end up with an inconsistent subvolume tree where
1496 * dir index entries exist for a name but there is no inode reference
1497 * item with the same name.
1499 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1504 /* finally write the back reference in the inode */
1505 ret = overwrite_item(trans, root, path, eb, slot, key);
1507 btrfs_release_path(path);
1514 static int count_inode_extrefs(struct btrfs_root *root,
1515 struct btrfs_inode *inode, struct btrfs_path *path)
1519 unsigned int nlink = 0;
1522 u64 inode_objectid = btrfs_ino(inode);
1525 struct btrfs_inode_extref *extref;
1526 struct extent_buffer *leaf;
1529 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1534 leaf = path->nodes[0];
1535 item_size = btrfs_item_size(leaf, path->slots[0]);
1536 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1539 while (cur_offset < item_size) {
1540 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1541 name_len = btrfs_inode_extref_name_len(leaf, extref);
1545 cur_offset += name_len + sizeof(*extref);
1549 btrfs_release_path(path);
1551 btrfs_release_path(path);
1553 if (ret < 0 && ret != -ENOENT)
1558 static int count_inode_refs(struct btrfs_root *root,
1559 struct btrfs_inode *inode, struct btrfs_path *path)
1562 struct btrfs_key key;
1563 unsigned int nlink = 0;
1565 unsigned long ptr_end;
1567 u64 ino = btrfs_ino(inode);
1570 key.type = BTRFS_INODE_REF_KEY;
1571 key.offset = (u64)-1;
1574 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1578 if (path->slots[0] == 0)
1583 btrfs_item_key_to_cpu(path->nodes[0], &key,
1585 if (key.objectid != ino ||
1586 key.type != BTRFS_INODE_REF_KEY)
1588 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1589 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1591 while (ptr < ptr_end) {
1592 struct btrfs_inode_ref *ref;
1594 ref = (struct btrfs_inode_ref *)ptr;
1595 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1597 ptr = (unsigned long)(ref + 1) + name_len;
1601 if (key.offset == 0)
1603 if (path->slots[0] > 0) {
1608 btrfs_release_path(path);
1610 btrfs_release_path(path);
1616 * There are a few corners where the link count of the file can't
1617 * be properly maintained during replay. So, instead of adding
1618 * lots of complexity to the log code, we just scan the backrefs
1619 * for any file that has been through replay.
1621 * The scan will update the link count on the inode to reflect the
1622 * number of back refs found. If it goes down to zero, the iput
1623 * will free the inode.
1625 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1626 struct btrfs_root *root,
1627 struct inode *inode)
1629 struct btrfs_path *path;
1632 u64 ino = btrfs_ino(BTRFS_I(inode));
1634 path = btrfs_alloc_path();
1638 ret = count_inode_refs(root, BTRFS_I(inode), path);
1644 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1652 if (nlink != inode->i_nlink) {
1653 set_nlink(inode, nlink);
1654 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1658 BTRFS_I(inode)->index_cnt = (u64)-1;
1660 if (inode->i_nlink == 0) {
1661 if (S_ISDIR(inode->i_mode)) {
1662 ret = replay_dir_deletes(trans, root, NULL, path,
1667 ret = btrfs_insert_orphan_item(trans, root, ino);
1673 btrfs_free_path(path);
1677 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1678 struct btrfs_root *root,
1679 struct btrfs_path *path)
1682 struct btrfs_key key;
1683 struct inode *inode;
1685 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1686 key.type = BTRFS_ORPHAN_ITEM_KEY;
1687 key.offset = (u64)-1;
1689 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1695 if (path->slots[0] == 0)
1700 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1701 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1702 key.type != BTRFS_ORPHAN_ITEM_KEY)
1705 ret = btrfs_del_item(trans, root, path);
1709 btrfs_release_path(path);
1710 inode = read_one_inode(root, key.offset);
1716 ret = fixup_inode_link_count(trans, root, inode);
1722 * fixup on a directory may create new entries,
1723 * make sure we always look for the highset possible
1726 key.offset = (u64)-1;
1728 btrfs_release_path(path);
1734 * record a given inode in the fixup dir so we can check its link
1735 * count when replay is done. The link count is incremented here
1736 * so the inode won't go away until we check it
1738 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1739 struct btrfs_root *root,
1740 struct btrfs_path *path,
1743 struct btrfs_key key;
1745 struct inode *inode;
1747 inode = read_one_inode(root, objectid);
1751 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1752 key.type = BTRFS_ORPHAN_ITEM_KEY;
1753 key.offset = objectid;
1755 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1757 btrfs_release_path(path);
1759 if (!inode->i_nlink)
1760 set_nlink(inode, 1);
1763 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1764 } else if (ret == -EEXIST) {
1773 * when replaying the log for a directory, we only insert names
1774 * for inodes that actually exist. This means an fsync on a directory
1775 * does not implicitly fsync all the new files in it
1777 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1778 struct btrfs_root *root,
1779 u64 dirid, u64 index,
1780 char *name, int name_len,
1781 struct btrfs_key *location)
1783 struct inode *inode;
1787 inode = read_one_inode(root, location->objectid);
1791 dir = read_one_inode(root, dirid);
1797 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1798 name_len, 1, index);
1800 /* FIXME, put inode into FIXUP list */
1807 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1808 struct btrfs_inode *dir,
1809 struct btrfs_path *path,
1810 struct btrfs_dir_item *dst_di,
1811 const struct btrfs_key *log_key,
1815 struct btrfs_key found_key;
1817 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1818 /* The existing dentry points to the same inode, don't delete it. */
1819 if (found_key.objectid == log_key->objectid &&
1820 found_key.type == log_key->type &&
1821 found_key.offset == log_key->offset &&
1822 btrfs_dir_type(path->nodes[0], dst_di) == log_type)
1826 * Don't drop the conflicting directory entry if the inode for the new
1827 * entry doesn't exist.
1832 return drop_one_dir_item(trans, path, dir, dst_di);
1836 * take a single entry in a log directory item and replay it into
1839 * if a conflicting item exists in the subdirectory already,
1840 * the inode it points to is unlinked and put into the link count
1843 * If a name from the log points to a file or directory that does
1844 * not exist in the FS, it is skipped. fsyncs on directories
1845 * do not force down inodes inside that directory, just changes to the
1846 * names or unlinks in a directory.
1848 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1849 * non-existing inode) and 1 if the name was replayed.
1851 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 struct extent_buffer *eb,
1855 struct btrfs_dir_item *di,
1856 struct btrfs_key *key)
1860 struct btrfs_dir_item *dir_dst_di;
1861 struct btrfs_dir_item *index_dst_di;
1862 bool dir_dst_matches = false;
1863 bool index_dst_matches = false;
1864 struct btrfs_key log_key;
1865 struct btrfs_key search_key;
1870 bool update_size = true;
1871 bool name_added = false;
1873 dir = read_one_inode(root, key->objectid);
1877 name_len = btrfs_dir_name_len(eb, di);
1878 name = kmalloc(name_len, GFP_NOFS);
1884 log_type = btrfs_dir_type(eb, di);
1885 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1888 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1889 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1890 btrfs_release_path(path);
1893 exists = (ret == 0);
1896 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1898 if (IS_ERR(dir_dst_di)) {
1899 ret = PTR_ERR(dir_dst_di);
1901 } else if (dir_dst_di) {
1902 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1903 dir_dst_di, &log_key, log_type,
1907 dir_dst_matches = (ret == 1);
1910 btrfs_release_path(path);
1912 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1913 key->objectid, key->offset,
1915 if (IS_ERR(index_dst_di)) {
1916 ret = PTR_ERR(index_dst_di);
1918 } else if (index_dst_di) {
1919 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1920 index_dst_di, &log_key,
1924 index_dst_matches = (ret == 1);
1927 btrfs_release_path(path);
1929 if (dir_dst_matches && index_dst_matches) {
1931 update_size = false;
1936 * Check if the inode reference exists in the log for the given name,
1937 * inode and parent inode
1939 search_key.objectid = log_key.objectid;
1940 search_key.type = BTRFS_INODE_REF_KEY;
1941 search_key.offset = key->objectid;
1942 ret = backref_in_log(root->log_root, &search_key, 0, name, name_len);
1946 /* The dentry will be added later. */
1948 update_size = false;
1952 search_key.objectid = log_key.objectid;
1953 search_key.type = BTRFS_INODE_EXTREF_KEY;
1954 search_key.offset = key->objectid;
1955 ret = backref_in_log(root->log_root, &search_key, key->objectid, name,
1960 /* The dentry will be added later. */
1962 update_size = false;
1965 btrfs_release_path(path);
1966 ret = insert_one_name(trans, root, key->objectid, key->offset,
1967 name, name_len, &log_key);
1968 if (ret && ret != -ENOENT && ret != -EEXIST)
1972 update_size = false;
1976 if (!ret && update_size) {
1977 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1978 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
1982 if (!ret && name_added)
1987 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1988 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1989 struct btrfs_root *root,
1990 struct btrfs_path *path,
1991 struct extent_buffer *eb, int slot,
1992 struct btrfs_key *key)
1995 struct btrfs_dir_item *di;
1997 /* We only log dir index keys, which only contain a single dir item. */
1998 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
2000 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2001 ret = replay_one_name(trans, root, path, eb, di, key);
2006 * If this entry refers to a non-directory (directories can not have a
2007 * link count > 1) and it was added in the transaction that was not
2008 * committed, make sure we fixup the link count of the inode the entry
2009 * points to. Otherwise something like the following would result in a
2010 * directory pointing to an inode with a wrong link that does not account
2011 * for this dir entry:
2018 * ln testdir/bar testdir/bar_link
2019 * ln testdir/foo testdir/foo_link
2020 * xfs_io -c "fsync" testdir/bar
2024 * mount fs, log replay happens
2026 * File foo would remain with a link count of 1 when it has two entries
2027 * pointing to it in the directory testdir. This would make it impossible
2028 * to ever delete the parent directory has it would result in stale
2029 * dentries that can never be deleted.
2031 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2032 struct btrfs_path *fixup_path;
2033 struct btrfs_key di_key;
2035 fixup_path = btrfs_alloc_path();
2039 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2040 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2041 btrfs_free_path(fixup_path);
2048 * directory replay has two parts. There are the standard directory
2049 * items in the log copied from the subvolume, and range items
2050 * created in the log while the subvolume was logged.
2052 * The range items tell us which parts of the key space the log
2053 * is authoritative for. During replay, if a key in the subvolume
2054 * directory is in a logged range item, but not actually in the log
2055 * that means it was deleted from the directory before the fsync
2056 * and should be removed.
2058 static noinline int find_dir_range(struct btrfs_root *root,
2059 struct btrfs_path *path,
2061 u64 *start_ret, u64 *end_ret)
2063 struct btrfs_key key;
2065 struct btrfs_dir_log_item *item;
2069 if (*start_ret == (u64)-1)
2072 key.objectid = dirid;
2073 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2074 key.offset = *start_ret;
2076 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2080 if (path->slots[0] == 0)
2085 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2087 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2091 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2092 struct btrfs_dir_log_item);
2093 found_end = btrfs_dir_log_end(path->nodes[0], item);
2095 if (*start_ret >= key.offset && *start_ret <= found_end) {
2097 *start_ret = key.offset;
2098 *end_ret = found_end;
2103 /* check the next slot in the tree to see if it is a valid item */
2104 nritems = btrfs_header_nritems(path->nodes[0]);
2106 if (path->slots[0] >= nritems) {
2107 ret = btrfs_next_leaf(root, path);
2112 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2114 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2118 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2119 struct btrfs_dir_log_item);
2120 found_end = btrfs_dir_log_end(path->nodes[0], item);
2121 *start_ret = key.offset;
2122 *end_ret = found_end;
2125 btrfs_release_path(path);
2130 * this looks for a given directory item in the log. If the directory
2131 * item is not in the log, the item is removed and the inode it points
2134 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2135 struct btrfs_root *log,
2136 struct btrfs_path *path,
2137 struct btrfs_path *log_path,
2139 struct btrfs_key *dir_key)
2141 struct btrfs_root *root = BTRFS_I(dir)->root;
2143 struct extent_buffer *eb;
2145 struct btrfs_dir_item *di;
2148 struct inode *inode = NULL;
2149 struct btrfs_key location;
2152 * Currently we only log dir index keys. Even if we replay a log created
2153 * by an older kernel that logged both dir index and dir item keys, all
2154 * we need to do is process the dir index keys, we (and our caller) can
2155 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2157 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2159 eb = path->nodes[0];
2160 slot = path->slots[0];
2161 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2162 name_len = btrfs_dir_name_len(eb, di);
2163 name = kmalloc(name_len, GFP_NOFS);
2169 read_extent_buffer(eb, name, (unsigned long)(di + 1), name_len);
2172 struct btrfs_dir_item *log_di;
2174 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2178 if (IS_ERR(log_di)) {
2179 ret = PTR_ERR(log_di);
2181 } else if (log_di) {
2182 /* The dentry exists in the log, we have nothing to do. */
2188 btrfs_dir_item_key_to_cpu(eb, di, &location);
2189 btrfs_release_path(path);
2190 btrfs_release_path(log_path);
2191 inode = read_one_inode(root, location.objectid);
2197 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2202 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2205 * Unlike dir item keys, dir index keys can only have one name (entry) in
2206 * them, as there are no key collisions since each key has a unique offset
2207 * (an index number), so we're done.
2210 btrfs_release_path(path);
2211 btrfs_release_path(log_path);
2217 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2218 struct btrfs_root *root,
2219 struct btrfs_root *log,
2220 struct btrfs_path *path,
2223 struct btrfs_key search_key;
2224 struct btrfs_path *log_path;
2229 log_path = btrfs_alloc_path();
2233 search_key.objectid = ino;
2234 search_key.type = BTRFS_XATTR_ITEM_KEY;
2235 search_key.offset = 0;
2237 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2241 nritems = btrfs_header_nritems(path->nodes[0]);
2242 for (i = path->slots[0]; i < nritems; i++) {
2243 struct btrfs_key key;
2244 struct btrfs_dir_item *di;
2245 struct btrfs_dir_item *log_di;
2249 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2250 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2255 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2256 total_size = btrfs_item_size(path->nodes[0], i);
2258 while (cur < total_size) {
2259 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2260 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2261 u32 this_len = sizeof(*di) + name_len + data_len;
2264 name = kmalloc(name_len, GFP_NOFS);
2269 read_extent_buffer(path->nodes[0], name,
2270 (unsigned long)(di + 1), name_len);
2272 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2274 btrfs_release_path(log_path);
2276 /* Doesn't exist in log tree, so delete it. */
2277 btrfs_release_path(path);
2278 di = btrfs_lookup_xattr(trans, root, path, ino,
2279 name, name_len, -1);
2286 ret = btrfs_delete_one_dir_name(trans, root,
2290 btrfs_release_path(path);
2295 if (IS_ERR(log_di)) {
2296 ret = PTR_ERR(log_di);
2300 di = (struct btrfs_dir_item *)((char *)di + this_len);
2303 ret = btrfs_next_leaf(root, path);
2309 btrfs_free_path(log_path);
2310 btrfs_release_path(path);
2316 * deletion replay happens before we copy any new directory items
2317 * out of the log or out of backreferences from inodes. It
2318 * scans the log to find ranges of keys that log is authoritative for,
2319 * and then scans the directory to find items in those ranges that are
2320 * not present in the log.
2322 * Anything we don't find in the log is unlinked and removed from the
2325 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2326 struct btrfs_root *root,
2327 struct btrfs_root *log,
2328 struct btrfs_path *path,
2329 u64 dirid, int del_all)
2334 struct btrfs_key dir_key;
2335 struct btrfs_key found_key;
2336 struct btrfs_path *log_path;
2339 dir_key.objectid = dirid;
2340 dir_key.type = BTRFS_DIR_INDEX_KEY;
2341 log_path = btrfs_alloc_path();
2345 dir = read_one_inode(root, dirid);
2346 /* it isn't an error if the inode isn't there, that can happen
2347 * because we replay the deletes before we copy in the inode item
2351 btrfs_free_path(log_path);
2359 range_end = (u64)-1;
2361 ret = find_dir_range(log, path, dirid,
2362 &range_start, &range_end);
2369 dir_key.offset = range_start;
2372 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2377 nritems = btrfs_header_nritems(path->nodes[0]);
2378 if (path->slots[0] >= nritems) {
2379 ret = btrfs_next_leaf(root, path);
2385 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2387 if (found_key.objectid != dirid ||
2388 found_key.type != dir_key.type) {
2393 if (found_key.offset > range_end)
2396 ret = check_item_in_log(trans, log, path,
2401 if (found_key.offset == (u64)-1)
2403 dir_key.offset = found_key.offset + 1;
2405 btrfs_release_path(path);
2406 if (range_end == (u64)-1)
2408 range_start = range_end + 1;
2412 btrfs_release_path(path);
2413 btrfs_free_path(log_path);
2419 * the process_func used to replay items from the log tree. This
2420 * gets called in two different stages. The first stage just looks
2421 * for inodes and makes sure they are all copied into the subvolume.
2423 * The second stage copies all the other item types from the log into
2424 * the subvolume. The two stage approach is slower, but gets rid of
2425 * lots of complexity around inodes referencing other inodes that exist
2426 * only in the log (references come from either directory items or inode
2429 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2430 struct walk_control *wc, u64 gen, int level)
2433 struct btrfs_path *path;
2434 struct btrfs_root *root = wc->replay_dest;
2435 struct btrfs_key key;
2439 ret = btrfs_read_extent_buffer(eb, gen, level, NULL);
2443 level = btrfs_header_level(eb);
2448 path = btrfs_alloc_path();
2452 nritems = btrfs_header_nritems(eb);
2453 for (i = 0; i < nritems; i++) {
2454 btrfs_item_key_to_cpu(eb, &key, i);
2456 /* inode keys are done during the first stage */
2457 if (key.type == BTRFS_INODE_ITEM_KEY &&
2458 wc->stage == LOG_WALK_REPLAY_INODES) {
2459 struct btrfs_inode_item *inode_item;
2462 inode_item = btrfs_item_ptr(eb, i,
2463 struct btrfs_inode_item);
2465 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2466 * and never got linked before the fsync, skip it, as
2467 * replaying it is pointless since it would be deleted
2468 * later. We skip logging tmpfiles, but it's always
2469 * possible we are replaying a log created with a kernel
2470 * that used to log tmpfiles.
2472 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2473 wc->ignore_cur_inode = true;
2476 wc->ignore_cur_inode = false;
2478 ret = replay_xattr_deletes(wc->trans, root, log,
2479 path, key.objectid);
2482 mode = btrfs_inode_mode(eb, inode_item);
2483 if (S_ISDIR(mode)) {
2484 ret = replay_dir_deletes(wc->trans,
2485 root, log, path, key.objectid, 0);
2489 ret = overwrite_item(wc->trans, root, path,
2495 * Before replaying extents, truncate the inode to its
2496 * size. We need to do it now and not after log replay
2497 * because before an fsync we can have prealloc extents
2498 * added beyond the inode's i_size. If we did it after,
2499 * through orphan cleanup for example, we would drop
2500 * those prealloc extents just after replaying them.
2502 if (S_ISREG(mode)) {
2503 struct btrfs_drop_extents_args drop_args = { 0 };
2504 struct inode *inode;
2507 inode = read_one_inode(root, key.objectid);
2512 from = ALIGN(i_size_read(inode),
2513 root->fs_info->sectorsize);
2514 drop_args.start = from;
2515 drop_args.end = (u64)-1;
2516 drop_args.drop_cache = true;
2517 ret = btrfs_drop_extents(wc->trans, root,
2521 inode_sub_bytes(inode,
2522 drop_args.bytes_found);
2523 /* Update the inode's nbytes. */
2524 ret = btrfs_update_inode(wc->trans,
2525 root, BTRFS_I(inode));
2532 ret = link_to_fixup_dir(wc->trans, root,
2533 path, key.objectid);
2538 if (wc->ignore_cur_inode)
2541 if (key.type == BTRFS_DIR_INDEX_KEY &&
2542 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2543 ret = replay_one_dir_item(wc->trans, root, path,
2549 if (wc->stage < LOG_WALK_REPLAY_ALL)
2552 /* these keys are simply copied */
2553 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2554 ret = overwrite_item(wc->trans, root, path,
2558 } else if (key.type == BTRFS_INODE_REF_KEY ||
2559 key.type == BTRFS_INODE_EXTREF_KEY) {
2560 ret = add_inode_ref(wc->trans, root, log, path,
2562 if (ret && ret != -ENOENT)
2565 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2566 ret = replay_one_extent(wc->trans, root, path,
2572 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2573 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2574 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2575 * older kernel with such keys, ignore them.
2578 btrfs_free_path(path);
2583 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2585 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2587 struct btrfs_block_group *cache;
2589 cache = btrfs_lookup_block_group(fs_info, start);
2591 btrfs_err(fs_info, "unable to find block group for %llu", start);
2595 spin_lock(&cache->space_info->lock);
2596 spin_lock(&cache->lock);
2597 cache->reserved -= fs_info->nodesize;
2598 cache->space_info->bytes_reserved -= fs_info->nodesize;
2599 spin_unlock(&cache->lock);
2600 spin_unlock(&cache->space_info->lock);
2602 btrfs_put_block_group(cache);
2605 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2606 struct btrfs_root *root,
2607 struct btrfs_path *path, int *level,
2608 struct walk_control *wc)
2610 struct btrfs_fs_info *fs_info = root->fs_info;
2613 struct extent_buffer *next;
2614 struct extent_buffer *cur;
2618 while (*level > 0) {
2619 struct btrfs_key first_key;
2621 cur = path->nodes[*level];
2623 WARN_ON(btrfs_header_level(cur) != *level);
2625 if (path->slots[*level] >=
2626 btrfs_header_nritems(cur))
2629 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2630 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2631 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2632 blocksize = fs_info->nodesize;
2634 next = btrfs_find_create_tree_block(fs_info, bytenr,
2635 btrfs_header_owner(cur),
2638 return PTR_ERR(next);
2641 ret = wc->process_func(root, next, wc, ptr_gen,
2644 free_extent_buffer(next);
2648 path->slots[*level]++;
2650 ret = btrfs_read_extent_buffer(next, ptr_gen,
2651 *level - 1, &first_key);
2653 free_extent_buffer(next);
2658 btrfs_tree_lock(next);
2659 btrfs_clean_tree_block(next);
2660 btrfs_wait_tree_block_writeback(next);
2661 btrfs_tree_unlock(next);
2662 ret = btrfs_pin_reserved_extent(trans,
2665 free_extent_buffer(next);
2668 btrfs_redirty_list_add(
2669 trans->transaction, next);
2671 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2672 clear_extent_buffer_dirty(next);
2673 unaccount_log_buffer(fs_info, bytenr);
2676 free_extent_buffer(next);
2679 ret = btrfs_read_extent_buffer(next, ptr_gen, *level - 1, &first_key);
2681 free_extent_buffer(next);
2685 if (path->nodes[*level-1])
2686 free_extent_buffer(path->nodes[*level-1]);
2687 path->nodes[*level-1] = next;
2688 *level = btrfs_header_level(next);
2689 path->slots[*level] = 0;
2692 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2698 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2699 struct btrfs_root *root,
2700 struct btrfs_path *path, int *level,
2701 struct walk_control *wc)
2703 struct btrfs_fs_info *fs_info = root->fs_info;
2708 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2709 slot = path->slots[i];
2710 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2713 WARN_ON(*level == 0);
2716 ret = wc->process_func(root, path->nodes[*level], wc,
2717 btrfs_header_generation(path->nodes[*level]),
2723 struct extent_buffer *next;
2725 next = path->nodes[*level];
2728 btrfs_tree_lock(next);
2729 btrfs_clean_tree_block(next);
2730 btrfs_wait_tree_block_writeback(next);
2731 btrfs_tree_unlock(next);
2732 ret = btrfs_pin_reserved_extent(trans,
2733 path->nodes[*level]->start,
2734 path->nodes[*level]->len);
2737 btrfs_redirty_list_add(trans->transaction,
2740 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2741 clear_extent_buffer_dirty(next);
2743 unaccount_log_buffer(fs_info,
2744 path->nodes[*level]->start);
2747 free_extent_buffer(path->nodes[*level]);
2748 path->nodes[*level] = NULL;
2756 * drop the reference count on the tree rooted at 'snap'. This traverses
2757 * the tree freeing any blocks that have a ref count of zero after being
2760 static int walk_log_tree(struct btrfs_trans_handle *trans,
2761 struct btrfs_root *log, struct walk_control *wc)
2763 struct btrfs_fs_info *fs_info = log->fs_info;
2767 struct btrfs_path *path;
2770 path = btrfs_alloc_path();
2774 level = btrfs_header_level(log->node);
2776 path->nodes[level] = log->node;
2777 atomic_inc(&log->node->refs);
2778 path->slots[level] = 0;
2781 wret = walk_down_log_tree(trans, log, path, &level, wc);
2789 wret = walk_up_log_tree(trans, log, path, &level, wc);
2798 /* was the root node processed? if not, catch it here */
2799 if (path->nodes[orig_level]) {
2800 ret = wc->process_func(log, path->nodes[orig_level], wc,
2801 btrfs_header_generation(path->nodes[orig_level]),
2806 struct extent_buffer *next;
2808 next = path->nodes[orig_level];
2811 btrfs_tree_lock(next);
2812 btrfs_clean_tree_block(next);
2813 btrfs_wait_tree_block_writeback(next);
2814 btrfs_tree_unlock(next);
2815 ret = btrfs_pin_reserved_extent(trans,
2816 next->start, next->len);
2819 btrfs_redirty_list_add(trans->transaction, next);
2821 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2822 clear_extent_buffer_dirty(next);
2823 unaccount_log_buffer(fs_info, next->start);
2829 btrfs_free_path(path);
2834 * helper function to update the item for a given subvolumes log root
2835 * in the tree of log roots
2837 static int update_log_root(struct btrfs_trans_handle *trans,
2838 struct btrfs_root *log,
2839 struct btrfs_root_item *root_item)
2841 struct btrfs_fs_info *fs_info = log->fs_info;
2844 if (log->log_transid == 1) {
2845 /* insert root item on the first sync */
2846 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2847 &log->root_key, root_item);
2849 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2850 &log->root_key, root_item);
2855 static void wait_log_commit(struct btrfs_root *root, int transid)
2858 int index = transid % 2;
2861 * we only allow two pending log transactions at a time,
2862 * so we know that if ours is more than 2 older than the
2863 * current transaction, we're done
2866 prepare_to_wait(&root->log_commit_wait[index],
2867 &wait, TASK_UNINTERRUPTIBLE);
2869 if (!(root->log_transid_committed < transid &&
2870 atomic_read(&root->log_commit[index])))
2873 mutex_unlock(&root->log_mutex);
2875 mutex_lock(&root->log_mutex);
2877 finish_wait(&root->log_commit_wait[index], &wait);
2880 static void wait_for_writer(struct btrfs_root *root)
2885 prepare_to_wait(&root->log_writer_wait, &wait,
2886 TASK_UNINTERRUPTIBLE);
2887 if (!atomic_read(&root->log_writers))
2890 mutex_unlock(&root->log_mutex);
2892 mutex_lock(&root->log_mutex);
2894 finish_wait(&root->log_writer_wait, &wait);
2897 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2898 struct btrfs_log_ctx *ctx)
2900 mutex_lock(&root->log_mutex);
2901 list_del_init(&ctx->list);
2902 mutex_unlock(&root->log_mutex);
2906 * Invoked in log mutex context, or be sure there is no other task which
2907 * can access the list.
2909 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2910 int index, int error)
2912 struct btrfs_log_ctx *ctx;
2913 struct btrfs_log_ctx *safe;
2915 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2916 list_del_init(&ctx->list);
2917 ctx->log_ret = error;
2922 * btrfs_sync_log does sends a given tree log down to the disk and
2923 * updates the super blocks to record it. When this call is done,
2924 * you know that any inodes previously logged are safely on disk only
2927 * Any other return value means you need to call btrfs_commit_transaction.
2928 * Some of the edge cases for fsyncing directories that have had unlinks
2929 * or renames done in the past mean that sometimes the only safe
2930 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2931 * that has happened.
2933 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2934 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2940 struct btrfs_fs_info *fs_info = root->fs_info;
2941 struct btrfs_root *log = root->log_root;
2942 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2943 struct btrfs_root_item new_root_item;
2944 int log_transid = 0;
2945 struct btrfs_log_ctx root_log_ctx;
2946 struct blk_plug plug;
2950 mutex_lock(&root->log_mutex);
2951 log_transid = ctx->log_transid;
2952 if (root->log_transid_committed >= log_transid) {
2953 mutex_unlock(&root->log_mutex);
2954 return ctx->log_ret;
2957 index1 = log_transid % 2;
2958 if (atomic_read(&root->log_commit[index1])) {
2959 wait_log_commit(root, log_transid);
2960 mutex_unlock(&root->log_mutex);
2961 return ctx->log_ret;
2963 ASSERT(log_transid == root->log_transid);
2964 atomic_set(&root->log_commit[index1], 1);
2966 /* wait for previous tree log sync to complete */
2967 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2968 wait_log_commit(root, log_transid - 1);
2971 int batch = atomic_read(&root->log_batch);
2972 /* when we're on an ssd, just kick the log commit out */
2973 if (!btrfs_test_opt(fs_info, SSD) &&
2974 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2975 mutex_unlock(&root->log_mutex);
2976 schedule_timeout_uninterruptible(1);
2977 mutex_lock(&root->log_mutex);
2979 wait_for_writer(root);
2980 if (batch == atomic_read(&root->log_batch))
2984 /* bail out if we need to do a full commit */
2985 if (btrfs_need_log_full_commit(trans)) {
2986 ret = BTRFS_LOG_FORCE_COMMIT;
2987 mutex_unlock(&root->log_mutex);
2991 if (log_transid % 2 == 0)
2992 mark = EXTENT_DIRTY;
2996 /* we start IO on all the marked extents here, but we don't actually
2997 * wait for them until later.
2999 blk_start_plug(&plug);
3000 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3002 * -EAGAIN happens when someone, e.g., a concurrent transaction
3003 * commit, writes a dirty extent in this tree-log commit. This
3004 * concurrent write will create a hole writing out the extents,
3005 * and we cannot proceed on a zoned filesystem, requiring
3006 * sequential writing. While we can bail out to a full commit
3007 * here, but we can continue hoping the concurrent writing fills
3010 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3013 blk_finish_plug(&plug);
3014 btrfs_abort_transaction(trans, ret);
3015 btrfs_set_log_full_commit(trans);
3016 mutex_unlock(&root->log_mutex);
3021 * We _must_ update under the root->log_mutex in order to make sure we
3022 * have a consistent view of the log root we are trying to commit at
3025 * We _must_ copy this into a local copy, because we are not holding the
3026 * log_root_tree->log_mutex yet. This is important because when we
3027 * commit the log_root_tree we must have a consistent view of the
3028 * log_root_tree when we update the super block to point at the
3029 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3030 * with the commit and possibly point at the new block which we may not
3033 btrfs_set_root_node(&log->root_item, log->node);
3034 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3036 root->log_transid++;
3037 log->log_transid = root->log_transid;
3038 root->log_start_pid = 0;
3040 * IO has been started, blocks of the log tree have WRITTEN flag set
3041 * in their headers. new modifications of the log will be written to
3042 * new positions. so it's safe to allow log writers to go in.
3044 mutex_unlock(&root->log_mutex);
3046 if (btrfs_is_zoned(fs_info)) {
3047 mutex_lock(&fs_info->tree_root->log_mutex);
3048 if (!log_root_tree->node) {
3049 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3051 mutex_unlock(&fs_info->tree_root->log_mutex);
3052 blk_finish_plug(&plug);
3056 mutex_unlock(&fs_info->tree_root->log_mutex);
3059 btrfs_init_log_ctx(&root_log_ctx, NULL);
3061 mutex_lock(&log_root_tree->log_mutex);
3063 index2 = log_root_tree->log_transid % 2;
3064 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3065 root_log_ctx.log_transid = log_root_tree->log_transid;
3068 * Now we are safe to update the log_root_tree because we're under the
3069 * log_mutex, and we're a current writer so we're holding the commit
3070 * open until we drop the log_mutex.
3072 ret = update_log_root(trans, log, &new_root_item);
3074 if (!list_empty(&root_log_ctx.list))
3075 list_del_init(&root_log_ctx.list);
3077 blk_finish_plug(&plug);
3078 btrfs_set_log_full_commit(trans);
3080 if (ret != -ENOSPC) {
3081 btrfs_abort_transaction(trans, ret);
3082 mutex_unlock(&log_root_tree->log_mutex);
3085 btrfs_wait_tree_log_extents(log, mark);
3086 mutex_unlock(&log_root_tree->log_mutex);
3087 ret = BTRFS_LOG_FORCE_COMMIT;
3091 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3092 blk_finish_plug(&plug);
3093 list_del_init(&root_log_ctx.list);
3094 mutex_unlock(&log_root_tree->log_mutex);
3095 ret = root_log_ctx.log_ret;
3099 index2 = root_log_ctx.log_transid % 2;
3100 if (atomic_read(&log_root_tree->log_commit[index2])) {
3101 blk_finish_plug(&plug);
3102 ret = btrfs_wait_tree_log_extents(log, mark);
3103 wait_log_commit(log_root_tree,
3104 root_log_ctx.log_transid);
3105 mutex_unlock(&log_root_tree->log_mutex);
3107 ret = root_log_ctx.log_ret;
3110 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3111 atomic_set(&log_root_tree->log_commit[index2], 1);
3113 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3114 wait_log_commit(log_root_tree,
3115 root_log_ctx.log_transid - 1);
3119 * now that we've moved on to the tree of log tree roots,
3120 * check the full commit flag again
3122 if (btrfs_need_log_full_commit(trans)) {
3123 blk_finish_plug(&plug);
3124 btrfs_wait_tree_log_extents(log, mark);
3125 mutex_unlock(&log_root_tree->log_mutex);
3126 ret = BTRFS_LOG_FORCE_COMMIT;
3127 goto out_wake_log_root;
3130 ret = btrfs_write_marked_extents(fs_info,
3131 &log_root_tree->dirty_log_pages,
3132 EXTENT_DIRTY | EXTENT_NEW);
3133 blk_finish_plug(&plug);
3135 * As described above, -EAGAIN indicates a hole in the extents. We
3136 * cannot wait for these write outs since the waiting cause a
3137 * deadlock. Bail out to the full commit instead.
3139 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3140 btrfs_set_log_full_commit(trans);
3141 btrfs_wait_tree_log_extents(log, mark);
3142 mutex_unlock(&log_root_tree->log_mutex);
3143 goto out_wake_log_root;
3145 btrfs_set_log_full_commit(trans);
3146 btrfs_abort_transaction(trans, ret);
3147 mutex_unlock(&log_root_tree->log_mutex);
3148 goto out_wake_log_root;
3150 ret = btrfs_wait_tree_log_extents(log, mark);
3152 ret = btrfs_wait_tree_log_extents(log_root_tree,
3153 EXTENT_NEW | EXTENT_DIRTY);
3155 btrfs_set_log_full_commit(trans);
3156 mutex_unlock(&log_root_tree->log_mutex);
3157 goto out_wake_log_root;
3160 log_root_start = log_root_tree->node->start;
3161 log_root_level = btrfs_header_level(log_root_tree->node);
3162 log_root_tree->log_transid++;
3163 mutex_unlock(&log_root_tree->log_mutex);
3166 * Here we are guaranteed that nobody is going to write the superblock
3167 * for the current transaction before us and that neither we do write
3168 * our superblock before the previous transaction finishes its commit
3169 * and writes its superblock, because:
3171 * 1) We are holding a handle on the current transaction, so no body
3172 * can commit it until we release the handle;
3174 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3175 * if the previous transaction is still committing, and hasn't yet
3176 * written its superblock, we wait for it to do it, because a
3177 * transaction commit acquires the tree_log_mutex when the commit
3178 * begins and releases it only after writing its superblock.
3180 mutex_lock(&fs_info->tree_log_mutex);
3183 * The previous transaction writeout phase could have failed, and thus
3184 * marked the fs in an error state. We must not commit here, as we
3185 * could have updated our generation in the super_for_commit and
3186 * writing the super here would result in transid mismatches. If there
3187 * is an error here just bail.
3189 if (BTRFS_FS_ERROR(fs_info)) {
3191 btrfs_set_log_full_commit(trans);
3192 btrfs_abort_transaction(trans, ret);
3193 mutex_unlock(&fs_info->tree_log_mutex);
3194 goto out_wake_log_root;
3197 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3198 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3199 ret = write_all_supers(fs_info, 1);
3200 mutex_unlock(&fs_info->tree_log_mutex);
3202 btrfs_set_log_full_commit(trans);
3203 btrfs_abort_transaction(trans, ret);
3204 goto out_wake_log_root;
3208 * We know there can only be one task here, since we have not yet set
3209 * root->log_commit[index1] to 0 and any task attempting to sync the
3210 * log must wait for the previous log transaction to commit if it's
3211 * still in progress or wait for the current log transaction commit if
3212 * someone else already started it. We use <= and not < because the
3213 * first log transaction has an ID of 0.
3215 ASSERT(root->last_log_commit <= log_transid);
3216 root->last_log_commit = log_transid;
3219 mutex_lock(&log_root_tree->log_mutex);
3220 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3222 log_root_tree->log_transid_committed++;
3223 atomic_set(&log_root_tree->log_commit[index2], 0);
3224 mutex_unlock(&log_root_tree->log_mutex);
3227 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3228 * all the updates above are seen by the woken threads. It might not be
3229 * necessary, but proving that seems to be hard.
3231 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3233 mutex_lock(&root->log_mutex);
3234 btrfs_remove_all_log_ctxs(root, index1, ret);
3235 root->log_transid_committed++;
3236 atomic_set(&root->log_commit[index1], 0);
3237 mutex_unlock(&root->log_mutex);
3240 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3241 * all the updates above are seen by the woken threads. It might not be
3242 * necessary, but proving that seems to be hard.
3244 cond_wake_up(&root->log_commit_wait[index1]);
3248 static void free_log_tree(struct btrfs_trans_handle *trans,
3249 struct btrfs_root *log)
3252 struct walk_control wc = {
3254 .process_func = process_one_buffer
3258 ret = walk_log_tree(trans, log, &wc);
3261 * We weren't able to traverse the entire log tree, the
3262 * typical scenario is getting an -EIO when reading an
3263 * extent buffer of the tree, due to a previous writeback
3266 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3267 &log->fs_info->fs_state);
3270 * Some extent buffers of the log tree may still be dirty
3271 * and not yet written back to storage, because we may
3272 * have updates to a log tree without syncing a log tree,
3273 * such as during rename and link operations. So flush
3274 * them out and wait for their writeback to complete, so
3275 * that we properly cleanup their state and pages.
3277 btrfs_write_marked_extents(log->fs_info,
3278 &log->dirty_log_pages,
3279 EXTENT_DIRTY | EXTENT_NEW);
3280 btrfs_wait_tree_log_extents(log,
3281 EXTENT_DIRTY | EXTENT_NEW);
3284 btrfs_abort_transaction(trans, ret);
3286 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3290 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3291 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3292 extent_io_tree_release(&log->log_csum_range);
3294 btrfs_put_root(log);
3298 * free all the extents used by the tree log. This should be called
3299 * at commit time of the full transaction
3301 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3303 if (root->log_root) {
3304 free_log_tree(trans, root->log_root);
3305 root->log_root = NULL;
3306 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3311 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3312 struct btrfs_fs_info *fs_info)
3314 if (fs_info->log_root_tree) {
3315 free_log_tree(trans, fs_info->log_root_tree);
3316 fs_info->log_root_tree = NULL;
3317 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3323 * Check if an inode was logged in the current transaction. This correctly deals
3324 * with the case where the inode was logged but has a logged_trans of 0, which
3325 * happens if the inode is evicted and loaded again, as logged_trans is an in
3326 * memory only field (not persisted).
3328 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3331 static int inode_logged(struct btrfs_trans_handle *trans,
3332 struct btrfs_inode *inode,
3333 struct btrfs_path *path_in)
3335 struct btrfs_path *path = path_in;
3336 struct btrfs_key key;
3339 if (inode->logged_trans == trans->transid)
3343 * If logged_trans is not 0, then we know the inode logged was not logged
3344 * in this transaction, so we can return false right away.
3346 if (inode->logged_trans > 0)
3350 * If no log tree was created for this root in this transaction, then
3351 * the inode can not have been logged in this transaction. In that case
3352 * set logged_trans to anything greater than 0 and less than the current
3353 * transaction's ID, to avoid the search below in a future call in case
3354 * a log tree gets created after this.
3356 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3357 inode->logged_trans = trans->transid - 1;
3362 * We have a log tree and the inode's logged_trans is 0. We can't tell
3363 * for sure if the inode was logged before in this transaction by looking
3364 * only at logged_trans. We could be pessimistic and assume it was, but
3365 * that can lead to unnecessarily logging an inode during rename and link
3366 * operations, and then further updating the log in followup rename and
3367 * link operations, specially if it's a directory, which adds latency
3368 * visible to applications doing a series of rename or link operations.
3370 * A logged_trans of 0 here can mean several things:
3372 * 1) The inode was never logged since the filesystem was mounted, and may
3373 * or may have not been evicted and loaded again;
3375 * 2) The inode was logged in a previous transaction, then evicted and
3376 * then loaded again;
3378 * 3) The inode was logged in the current transaction, then evicted and
3379 * then loaded again.
3381 * For cases 1) and 2) we don't want to return true, but we need to detect
3382 * case 3) and return true. So we do a search in the log root for the inode
3385 key.objectid = btrfs_ino(inode);
3386 key.type = BTRFS_INODE_ITEM_KEY;
3390 path = btrfs_alloc_path();
3395 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3398 btrfs_release_path(path);
3400 btrfs_free_path(path);
3403 * Logging an inode always results in logging its inode item. So if we
3404 * did not find the item we know the inode was not logged for sure.
3408 } else if (ret > 0) {
3410 * Set logged_trans to a value greater than 0 and less then the
3411 * current transaction to avoid doing the search in future calls.
3413 inode->logged_trans = trans->transid - 1;
3418 * The inode was previously logged and then evicted, set logged_trans to
3419 * the current transacion's ID, to avoid future tree searches as long as
3420 * the inode is not evicted again.
3422 inode->logged_trans = trans->transid;
3425 * If it's a directory, then we must set last_dir_index_offset to the
3426 * maximum possible value, so that the next attempt to log the inode does
3427 * not skip checking if dir index keys found in modified subvolume tree
3428 * leaves have been logged before, otherwise it would result in attempts
3429 * to insert duplicate dir index keys in the log tree. This must be done
3430 * because last_dir_index_offset is an in-memory only field, not persisted
3431 * in the inode item or any other on-disk structure, so its value is lost
3432 * once the inode is evicted.
3434 if (S_ISDIR(inode->vfs_inode.i_mode))
3435 inode->last_dir_index_offset = (u64)-1;
3441 * Delete a directory entry from the log if it exists.
3443 * Returns < 0 on error
3444 * 1 if the entry does not exists
3445 * 0 if the entry existed and was successfully deleted
3447 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3448 struct btrfs_root *log,
3449 struct btrfs_path *path,
3451 const char *name, int name_len,
3454 struct btrfs_dir_item *di;
3457 * We only log dir index items of a directory, so we don't need to look
3458 * for dir item keys.
3460 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3461 index, name, name_len, -1);
3468 * We do not need to update the size field of the directory's
3469 * inode item because on log replay we update the field to reflect
3470 * all existing entries in the directory (see overwrite_item()).
3472 return btrfs_delete_one_dir_name(trans, log, path, di);
3476 * If both a file and directory are logged, and unlinks or renames are
3477 * mixed in, we have a few interesting corners:
3479 * create file X in dir Y
3480 * link file X to X.link in dir Y
3482 * unlink file X but leave X.link
3485 * After a crash we would expect only X.link to exist. But file X
3486 * didn't get fsync'd again so the log has back refs for X and X.link.
3488 * We solve this by removing directory entries and inode backrefs from the
3489 * log when a file that was logged in the current transaction is
3490 * unlinked. Any later fsync will include the updated log entries, and
3491 * we'll be able to reconstruct the proper directory items from backrefs.
3493 * This optimizations allows us to avoid relogging the entire inode
3494 * or the entire directory.
3496 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3497 struct btrfs_root *root,
3498 const char *name, int name_len,
3499 struct btrfs_inode *dir, u64 index)
3501 struct btrfs_path *path;
3504 ret = inode_logged(trans, dir, NULL);
3508 btrfs_set_log_full_commit(trans);
3512 ret = join_running_log_trans(root);
3516 mutex_lock(&dir->log_mutex);
3518 path = btrfs_alloc_path();
3524 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3525 name, name_len, index);
3526 btrfs_free_path(path);
3528 mutex_unlock(&dir->log_mutex);
3530 btrfs_set_log_full_commit(trans);
3531 btrfs_end_log_trans(root);
3534 /* see comments for btrfs_del_dir_entries_in_log */
3535 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3536 struct btrfs_root *root,
3537 const char *name, int name_len,
3538 struct btrfs_inode *inode, u64 dirid)
3540 struct btrfs_root *log;
3544 ret = inode_logged(trans, inode, NULL);
3548 btrfs_set_log_full_commit(trans);
3552 ret = join_running_log_trans(root);
3555 log = root->log_root;
3556 mutex_lock(&inode->log_mutex);
3558 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3560 mutex_unlock(&inode->log_mutex);
3561 if (ret < 0 && ret != -ENOENT)
3562 btrfs_set_log_full_commit(trans);
3563 btrfs_end_log_trans(root);
3567 * creates a range item in the log for 'dirid'. first_offset and
3568 * last_offset tell us which parts of the key space the log should
3569 * be considered authoritative for.
3571 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3572 struct btrfs_root *log,
3573 struct btrfs_path *path,
3575 u64 first_offset, u64 last_offset)
3578 struct btrfs_key key;
3579 struct btrfs_dir_log_item *item;
3581 key.objectid = dirid;
3582 key.offset = first_offset;
3583 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3584 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3586 * -EEXIST is fine and can happen sporadically when we are logging a
3587 * directory and have concurrent insertions in the subvolume's tree for
3588 * items from other inodes and that result in pushing off some dir items
3589 * from one leaf to another in order to accommodate for the new items.
3590 * This results in logging the same dir index range key.
3592 if (ret && ret != -EEXIST)
3595 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3596 struct btrfs_dir_log_item);
3597 if (ret == -EEXIST) {
3598 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3601 * btrfs_del_dir_entries_in_log() might have been called during
3602 * an unlink between the initial insertion of this key and the
3603 * current update, or we might be logging a single entry deletion
3604 * during a rename, so set the new last_offset to the max value.
3606 last_offset = max(last_offset, curr_end);
3608 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3609 btrfs_mark_buffer_dirty(path->nodes[0]);
3610 btrfs_release_path(path);
3614 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3615 struct btrfs_root *log,
3616 struct extent_buffer *src,
3617 struct btrfs_path *dst_path,
3621 char *ins_data = NULL;
3622 struct btrfs_item_batch batch;
3623 struct extent_buffer *dst;
3624 unsigned long src_offset;
3625 unsigned long dst_offset;
3626 struct btrfs_key key;
3635 btrfs_item_key_to_cpu(src, &key, start_slot);
3636 item_size = btrfs_item_size(src, start_slot);
3638 batch.data_sizes = &item_size;
3639 batch.total_data_size = item_size;
3641 struct btrfs_key *ins_keys;
3644 ins_data = kmalloc(count * sizeof(u32) +
3645 count * sizeof(struct btrfs_key), GFP_NOFS);
3649 ins_sizes = (u32 *)ins_data;
3650 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3651 batch.keys = ins_keys;
3652 batch.data_sizes = ins_sizes;
3653 batch.total_data_size = 0;
3655 for (i = 0; i < count; i++) {
3656 const int slot = start_slot + i;
3658 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3659 ins_sizes[i] = btrfs_item_size(src, slot);
3660 batch.total_data_size += ins_sizes[i];
3664 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3668 dst = dst_path->nodes[0];
3670 * Copy all the items in bulk, in a single copy operation. Item data is
3671 * organized such that it's placed at the end of a leaf and from right
3672 * to left. For example, the data for the second item ends at an offset
3673 * that matches the offset where the data for the first item starts, the
3674 * data for the third item ends at an offset that matches the offset
3675 * where the data of the second items starts, and so on.
3676 * Therefore our source and destination start offsets for copy match the
3677 * offsets of the last items (highest slots).
3679 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3680 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3681 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3682 btrfs_release_path(dst_path);
3689 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3690 struct btrfs_inode *inode,
3691 struct btrfs_path *path,
3692 struct btrfs_path *dst_path,
3693 struct btrfs_log_ctx *ctx,
3694 u64 *last_old_dentry_offset)
3696 struct btrfs_root *log = inode->root->log_root;
3697 struct extent_buffer *src = path->nodes[0];
3698 const int nritems = btrfs_header_nritems(src);
3699 const u64 ino = btrfs_ino(inode);
3700 bool last_found = false;
3701 int batch_start = 0;
3705 for (i = path->slots[0]; i < nritems; i++) {
3706 struct btrfs_dir_item *di;
3707 struct btrfs_key key;
3710 btrfs_item_key_to_cpu(src, &key, i);
3712 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3717 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3718 ctx->last_dir_item_offset = key.offset;
3721 * Skip ranges of items that consist only of dir item keys created
3722 * in past transactions. However if we find a gap, we must log a
3723 * dir index range item for that gap, so that index keys in that
3724 * gap are deleted during log replay.
3726 if (btrfs_dir_transid(src, di) < trans->transid) {
3727 if (key.offset > *last_old_dentry_offset + 1) {
3728 ret = insert_dir_log_key(trans, log, dst_path,
3729 ino, *last_old_dentry_offset + 1,
3735 *last_old_dentry_offset = key.offset;
3739 /* If we logged this dir index item before, we can skip it. */
3740 if (key.offset <= inode->last_dir_index_offset)
3744 * We must make sure that when we log a directory entry, the
3745 * corresponding inode, after log replay, has a matching link
3746 * count. For example:
3752 * xfs_io -c "fsync" mydir
3754 * <mount fs and log replay>
3756 * Would result in a fsync log that when replayed, our file inode
3757 * would have a link count of 1, but we get two directory entries
3758 * pointing to the same inode. After removing one of the names,
3759 * it would not be possible to remove the other name, which
3760 * resulted always in stale file handle errors, and would not be
3761 * possible to rmdir the parent directory, since its i_size could
3762 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3763 * resulting in -ENOTEMPTY errors.
3765 if (!ctx->log_new_dentries) {
3766 struct btrfs_key di_key;
3768 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3769 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3770 ctx->log_new_dentries = true;
3773 if (batch_size == 0)
3778 if (batch_size > 0) {
3781 ret = flush_dir_items_batch(trans, log, src, dst_path,
3782 batch_start, batch_size);
3787 return last_found ? 1 : 0;
3791 * log all the items included in the current transaction for a given
3792 * directory. This also creates the range items in the log tree required
3793 * to replay anything deleted before the fsync
3795 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3796 struct btrfs_inode *inode,
3797 struct btrfs_path *path,
3798 struct btrfs_path *dst_path,
3799 struct btrfs_log_ctx *ctx,
3800 u64 min_offset, u64 *last_offset_ret)
3802 struct btrfs_key min_key;
3803 struct btrfs_root *root = inode->root;
3804 struct btrfs_root *log = root->log_root;
3807 u64 last_old_dentry_offset = min_offset - 1;
3808 u64 last_offset = (u64)-1;
3809 u64 ino = btrfs_ino(inode);
3811 min_key.objectid = ino;
3812 min_key.type = BTRFS_DIR_INDEX_KEY;
3813 min_key.offset = min_offset;
3815 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3818 * we didn't find anything from this transaction, see if there
3819 * is anything at all
3821 if (ret != 0 || min_key.objectid != ino ||
3822 min_key.type != BTRFS_DIR_INDEX_KEY) {
3823 min_key.objectid = ino;
3824 min_key.type = BTRFS_DIR_INDEX_KEY;
3825 min_key.offset = (u64)-1;
3826 btrfs_release_path(path);
3827 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3829 btrfs_release_path(path);
3832 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3834 /* if ret == 0 there are items for this type,
3835 * create a range to tell us the last key of this type.
3836 * otherwise, there are no items in this directory after
3837 * *min_offset, and we create a range to indicate that.
3840 struct btrfs_key tmp;
3842 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3844 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3845 last_old_dentry_offset = tmp.offset;
3850 /* go backward to find any previous key */
3851 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3853 struct btrfs_key tmp;
3855 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3857 * The dir index key before the first one we found that needs to
3858 * be logged might be in a previous leaf, and there might be a
3859 * gap between these keys, meaning that we had deletions that
3860 * happened. So the key range item we log (key type
3861 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3862 * previous key's offset plus 1, so that those deletes are replayed.
3864 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3865 last_old_dentry_offset = tmp.offset;
3867 btrfs_release_path(path);
3870 * Find the first key from this transaction again. See the note for
3871 * log_new_dir_dentries, if we're logging a directory recursively we
3872 * won't be holding its i_mutex, which means we can modify the directory
3873 * while we're logging it. If we remove an entry between our first
3874 * search and this search we'll not find the key again and can just
3878 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3883 * we have a block from this transaction, log every item in it
3884 * from our directory
3887 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3888 &last_old_dentry_offset);
3894 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3897 * look ahead to the next item and see if it is also
3898 * from this directory and from this transaction
3900 ret = btrfs_next_leaf(root, path);
3903 last_offset = (u64)-1;
3908 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3909 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3910 last_offset = (u64)-1;
3913 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3915 * The next leaf was not changed in the current transaction
3916 * and has at least one dir index key.
3917 * We check for the next key because there might have been
3918 * one or more deletions between the last key we logged and
3919 * that next key. So the key range item we log (key type
3920 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3921 * offset minus 1, so that those deletes are replayed.
3923 last_offset = min_key.offset - 1;
3926 if (need_resched()) {
3927 btrfs_release_path(path);
3933 btrfs_release_path(path);
3934 btrfs_release_path(dst_path);
3937 *last_offset_ret = last_offset;
3939 * In case the leaf was changed in the current transaction but
3940 * all its dir items are from a past transaction, the last item
3941 * in the leaf is a dir item and there's no gap between that last
3942 * dir item and the first one on the next leaf (which did not
3943 * change in the current transaction), then we don't need to log
3944 * a range, last_old_dentry_offset is == to last_offset.
3946 ASSERT(last_old_dentry_offset <= last_offset);
3947 if (last_old_dentry_offset < last_offset) {
3948 ret = insert_dir_log_key(trans, log, path, ino,
3949 last_old_dentry_offset + 1,
3959 * If the inode was logged before and it was evicted, then its
3960 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3961 * key offset. If that's the case, search for it and update the inode. This
3962 * is to avoid lookups in the log tree every time we try to insert a dir index
3963 * key from a leaf changed in the current transaction, and to allow us to always
3964 * do batch insertions of dir index keys.
3966 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3967 struct btrfs_path *path,
3968 const struct btrfs_log_ctx *ctx)
3970 const u64 ino = btrfs_ino(inode);
3971 struct btrfs_key key;
3974 lockdep_assert_held(&inode->log_mutex);
3976 if (inode->last_dir_index_offset != (u64)-1)
3979 if (!ctx->logged_before) {
3980 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3985 key.type = BTRFS_DIR_INDEX_KEY;
3986 key.offset = (u64)-1;
3988 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3990 * An error happened or we actually have an index key with an offset
3991 * value of (u64)-1. Bail out, we're done.
3997 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
4000 * No dir index items, bail out and leave last_dir_index_offset with
4001 * the value right before the first valid index value.
4003 if (path->slots[0] == 0)
4007 * btrfs_search_slot() left us at one slot beyond the slot with the last
4008 * index key, or beyond the last key of the directory that is not an
4009 * index key. If we have an index key before, set last_dir_index_offset
4010 * to its offset value, otherwise leave it with a value right before the
4011 * first valid index value, as it means we have an empty directory.
4013 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4014 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
4015 inode->last_dir_index_offset = key.offset;
4018 btrfs_release_path(path);
4024 * logging directories is very similar to logging inodes, We find all the items
4025 * from the current transaction and write them to the log.
4027 * The recovery code scans the directory in the subvolume, and if it finds a
4028 * key in the range logged that is not present in the log tree, then it means
4029 * that dir entry was unlinked during the transaction.
4031 * In order for that scan to work, we must include one key smaller than
4032 * the smallest logged by this transaction and one key larger than the largest
4033 * key logged by this transaction.
4035 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4036 struct btrfs_inode *inode,
4037 struct btrfs_path *path,
4038 struct btrfs_path *dst_path,
4039 struct btrfs_log_ctx *ctx)
4045 ret = update_last_dir_index_offset(inode, path, ctx);
4049 min_key = BTRFS_DIR_START_INDEX;
4051 ctx->last_dir_item_offset = inode->last_dir_index_offset;
4054 ret = log_dir_items(trans, inode, path, dst_path,
4055 ctx, min_key, &max_key);
4058 if (max_key == (u64)-1)
4060 min_key = max_key + 1;
4063 inode->last_dir_index_offset = ctx->last_dir_item_offset;
4069 * a helper function to drop items from the log before we relog an
4070 * inode. max_key_type indicates the highest item type to remove.
4071 * This cannot be run for file data extents because it does not
4072 * free the extents they point to.
4074 static int drop_inode_items(struct btrfs_trans_handle *trans,
4075 struct btrfs_root *log,
4076 struct btrfs_path *path,
4077 struct btrfs_inode *inode,
4081 struct btrfs_key key;
4082 struct btrfs_key found_key;
4085 key.objectid = btrfs_ino(inode);
4086 key.type = max_key_type;
4087 key.offset = (u64)-1;
4090 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4091 BUG_ON(ret == 0); /* Logic error */
4095 if (path->slots[0] == 0)
4099 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4102 if (found_key.objectid != key.objectid)
4105 found_key.offset = 0;
4107 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
4111 ret = btrfs_del_items(trans, log, path, start_slot,
4112 path->slots[0] - start_slot + 1);
4114 * If start slot isn't 0 then we don't need to re-search, we've
4115 * found the last guy with the objectid in this tree.
4117 if (ret || start_slot != 0)
4119 btrfs_release_path(path);
4121 btrfs_release_path(path);
4127 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4128 struct btrfs_root *log_root,
4129 struct btrfs_inode *inode,
4130 u64 new_size, u32 min_type)
4132 struct btrfs_truncate_control control = {
4133 .new_size = new_size,
4134 .ino = btrfs_ino(inode),
4135 .min_type = min_type,
4136 .skip_ref_updates = true,
4139 return btrfs_truncate_inode_items(trans, log_root, &control);
4142 static void fill_inode_item(struct btrfs_trans_handle *trans,
4143 struct extent_buffer *leaf,
4144 struct btrfs_inode_item *item,
4145 struct inode *inode, int log_inode_only,
4148 struct btrfs_map_token token;
4151 btrfs_init_map_token(&token, leaf);
4153 if (log_inode_only) {
4154 /* set the generation to zero so the recover code
4155 * can tell the difference between an logging
4156 * just to say 'this inode exists' and a logging
4157 * to say 'update this inode with these values'
4159 btrfs_set_token_inode_generation(&token, item, 0);
4160 btrfs_set_token_inode_size(&token, item, logged_isize);
4162 btrfs_set_token_inode_generation(&token, item,
4163 BTRFS_I(inode)->generation);
4164 btrfs_set_token_inode_size(&token, item, inode->i_size);
4167 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4168 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4169 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4170 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4172 btrfs_set_token_timespec_sec(&token, &item->atime,
4173 inode->i_atime.tv_sec);
4174 btrfs_set_token_timespec_nsec(&token, &item->atime,
4175 inode->i_atime.tv_nsec);
4177 btrfs_set_token_timespec_sec(&token, &item->mtime,
4178 inode->i_mtime.tv_sec);
4179 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4180 inode->i_mtime.tv_nsec);
4182 btrfs_set_token_timespec_sec(&token, &item->ctime,
4183 inode->i_ctime.tv_sec);
4184 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4185 inode->i_ctime.tv_nsec);
4188 * We do not need to set the nbytes field, in fact during a fast fsync
4189 * its value may not even be correct, since a fast fsync does not wait
4190 * for ordered extent completion, which is where we update nbytes, it
4191 * only waits for writeback to complete. During log replay as we find
4192 * file extent items and replay them, we adjust the nbytes field of the
4193 * inode item in subvolume tree as needed (see overwrite_item()).
4196 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4197 btrfs_set_token_inode_transid(&token, item, trans->transid);
4198 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4199 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4200 BTRFS_I(inode)->ro_flags);
4201 btrfs_set_token_inode_flags(&token, item, flags);
4202 btrfs_set_token_inode_block_group(&token, item, 0);
4205 static int log_inode_item(struct btrfs_trans_handle *trans,
4206 struct btrfs_root *log, struct btrfs_path *path,
4207 struct btrfs_inode *inode, bool inode_item_dropped)
4209 struct btrfs_inode_item *inode_item;
4213 * If we are doing a fast fsync and the inode was logged before in the
4214 * current transaction, then we know the inode was previously logged and
4215 * it exists in the log tree. For performance reasons, in this case use
4216 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4217 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4218 * contention in case there are concurrent fsyncs for other inodes of the
4219 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4220 * already exists can also result in unnecessarily splitting a leaf.
4222 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4223 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4229 * This means it is the first fsync in the current transaction,
4230 * so the inode item is not in the log and we need to insert it.
4231 * We can never get -EEXIST because we are only called for a fast
4232 * fsync and in case an inode eviction happens after the inode was
4233 * logged before in the current transaction, when we load again
4234 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4235 * flags and set ->logged_trans to 0.
4237 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4238 sizeof(*inode_item));
4239 ASSERT(ret != -EEXIST);
4243 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4244 struct btrfs_inode_item);
4245 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4247 btrfs_release_path(path);
4251 static int log_csums(struct btrfs_trans_handle *trans,
4252 struct btrfs_inode *inode,
4253 struct btrfs_root *log_root,
4254 struct btrfs_ordered_sum *sums)
4256 const u64 lock_end = sums->bytenr + sums->len - 1;
4257 struct extent_state *cached_state = NULL;
4261 * If this inode was not used for reflink operations in the current
4262 * transaction with new extents, then do the fast path, no need to
4263 * worry about logging checksum items with overlapping ranges.
4265 if (inode->last_reflink_trans < trans->transid)
4266 return btrfs_csum_file_blocks(trans, log_root, sums);
4269 * Serialize logging for checksums. This is to avoid racing with the
4270 * same checksum being logged by another task that is logging another
4271 * file which happens to refer to the same extent as well. Such races
4272 * can leave checksum items in the log with overlapping ranges.
4274 ret = lock_extent(&log_root->log_csum_range, sums->bytenr, lock_end,
4279 * Due to extent cloning, we might have logged a csum item that covers a
4280 * subrange of a cloned extent, and later we can end up logging a csum
4281 * item for a larger subrange of the same extent or the entire range.
4282 * This would leave csum items in the log tree that cover the same range
4283 * and break the searches for checksums in the log tree, resulting in
4284 * some checksums missing in the fs/subvolume tree. So just delete (or
4285 * trim and adjust) any existing csum items in the log for this range.
4287 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4289 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4291 unlock_extent(&log_root->log_csum_range, sums->bytenr, lock_end,
4297 static noinline int copy_items(struct btrfs_trans_handle *trans,
4298 struct btrfs_inode *inode,
4299 struct btrfs_path *dst_path,
4300 struct btrfs_path *src_path,
4301 int start_slot, int nr, int inode_only,
4304 struct btrfs_root *log = inode->root->log_root;
4305 struct btrfs_file_extent_item *extent;
4306 struct extent_buffer *src = src_path->nodes[0];
4308 struct btrfs_key *ins_keys;
4310 struct btrfs_item_batch batch;
4314 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4315 const u64 i_size = i_size_read(&inode->vfs_inode);
4317 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4318 nr * sizeof(u32), GFP_NOFS);
4322 ins_sizes = (u32 *)ins_data;
4323 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4324 batch.keys = ins_keys;
4325 batch.data_sizes = ins_sizes;
4326 batch.total_data_size = 0;
4330 for (i = 0; i < nr; i++) {
4331 const int src_slot = start_slot + i;
4332 struct btrfs_root *csum_root;
4333 struct btrfs_ordered_sum *sums;
4334 struct btrfs_ordered_sum *sums_next;
4335 LIST_HEAD(ordered_sums);
4339 u64 extent_num_bytes;
4342 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4344 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4347 extent = btrfs_item_ptr(src, src_slot,
4348 struct btrfs_file_extent_item);
4350 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4354 * Don't copy extents from past generations. That would make us
4355 * log a lot more metadata for common cases like doing only a
4356 * few random writes into a file and then fsync it for the first
4357 * time or after the full sync flag is set on the inode. We can
4358 * get leaves full of extent items, most of which are from past
4359 * generations, so we can skip them - as long as the inode has
4360 * not been the target of a reflink operation in this transaction,
4361 * as in that case it might have had file extent items with old
4362 * generations copied into it. We also must always log prealloc
4363 * extents that start at or beyond eof, otherwise we would lose
4364 * them on log replay.
4366 if (is_old_extent &&
4367 ins_keys[dst_index].offset < i_size &&
4368 inode->last_reflink_trans < trans->transid)
4374 /* Only regular extents have checksums. */
4375 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4379 * If it's an extent created in a past transaction, then its
4380 * checksums are already accessible from the committed csum tree,
4381 * no need to log them.
4386 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4387 /* If it's an explicit hole, there are no checksums. */
4388 if (disk_bytenr == 0)
4391 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4393 if (btrfs_file_extent_compression(src, extent)) {
4395 extent_num_bytes = disk_num_bytes;
4397 extent_offset = btrfs_file_extent_offset(src, extent);
4398 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4401 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4402 disk_bytenr += extent_offset;
4403 ret = btrfs_lookup_csums_range(csum_root, disk_bytenr,
4404 disk_bytenr + extent_num_bytes - 1,
4405 &ordered_sums, 0, false);
4409 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4411 ret = log_csums(trans, inode, log, sums);
4412 list_del(&sums->list);
4419 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4420 batch.total_data_size += ins_sizes[dst_index];
4426 * We have a leaf full of old extent items that don't need to be logged,
4427 * so we don't need to do anything.
4432 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4437 for (i = 0; i < nr; i++) {
4438 const int src_slot = start_slot + i;
4439 const int dst_slot = dst_path->slots[0] + dst_index;
4440 struct btrfs_key key;
4441 unsigned long src_offset;
4442 unsigned long dst_offset;
4445 * We're done, all the remaining items in the source leaf
4446 * correspond to old file extent items.
4448 if (dst_index >= batch.nr)
4451 btrfs_item_key_to_cpu(src, &key, src_slot);
4453 if (key.type != BTRFS_EXTENT_DATA_KEY)
4456 extent = btrfs_item_ptr(src, src_slot,
4457 struct btrfs_file_extent_item);
4459 /* See the comment in the previous loop, same logic. */
4460 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4461 key.offset < i_size &&
4462 inode->last_reflink_trans < trans->transid)
4466 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4467 src_offset = btrfs_item_ptr_offset(src, src_slot);
4469 if (key.type == BTRFS_INODE_ITEM_KEY) {
4470 struct btrfs_inode_item *inode_item;
4472 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4473 struct btrfs_inode_item);
4474 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4476 inode_only == LOG_INODE_EXISTS,
4479 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4480 src_offset, ins_sizes[dst_index]);
4486 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4487 btrfs_release_path(dst_path);
4494 static int extent_cmp(void *priv, const struct list_head *a,
4495 const struct list_head *b)
4497 const struct extent_map *em1, *em2;
4499 em1 = list_entry(a, struct extent_map, list);
4500 em2 = list_entry(b, struct extent_map, list);
4502 if (em1->start < em2->start)
4504 else if (em1->start > em2->start)
4509 static int log_extent_csums(struct btrfs_trans_handle *trans,
4510 struct btrfs_inode *inode,
4511 struct btrfs_root *log_root,
4512 const struct extent_map *em,
4513 struct btrfs_log_ctx *ctx)
4515 struct btrfs_ordered_extent *ordered;
4516 struct btrfs_root *csum_root;
4519 u64 mod_start = em->mod_start;
4520 u64 mod_len = em->mod_len;
4521 LIST_HEAD(ordered_sums);
4524 if (inode->flags & BTRFS_INODE_NODATASUM ||
4525 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4526 em->block_start == EXTENT_MAP_HOLE)
4529 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4530 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4531 const u64 mod_end = mod_start + mod_len;
4532 struct btrfs_ordered_sum *sums;
4537 if (ordered_end <= mod_start)
4539 if (mod_end <= ordered->file_offset)
4543 * We are going to copy all the csums on this ordered extent, so
4544 * go ahead and adjust mod_start and mod_len in case this ordered
4545 * extent has already been logged.
4547 if (ordered->file_offset > mod_start) {
4548 if (ordered_end >= mod_end)
4549 mod_len = ordered->file_offset - mod_start;
4551 * If we have this case
4553 * |--------- logged extent ---------|
4554 * |----- ordered extent ----|
4556 * Just don't mess with mod_start and mod_len, we'll
4557 * just end up logging more csums than we need and it
4561 if (ordered_end < mod_end) {
4562 mod_len = mod_end - ordered_end;
4563 mod_start = ordered_end;
4570 * To keep us from looping for the above case of an ordered
4571 * extent that falls inside of the logged extent.
4573 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4576 list_for_each_entry(sums, &ordered->list, list) {
4577 ret = log_csums(trans, inode, log_root, sums);
4583 /* We're done, found all csums in the ordered extents. */
4587 /* If we're compressed we have to save the entire range of csums. */
4588 if (em->compress_type) {
4590 csum_len = max(em->block_len, em->orig_block_len);
4592 csum_offset = mod_start - em->start;
4596 /* block start is already adjusted for the file extent offset. */
4597 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4598 ret = btrfs_lookup_csums_range(csum_root,
4599 em->block_start + csum_offset,
4600 em->block_start + csum_offset +
4601 csum_len - 1, &ordered_sums, 0, false);
4605 while (!list_empty(&ordered_sums)) {
4606 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4607 struct btrfs_ordered_sum,
4610 ret = log_csums(trans, inode, log_root, sums);
4611 list_del(&sums->list);
4618 static int log_one_extent(struct btrfs_trans_handle *trans,
4619 struct btrfs_inode *inode,
4620 const struct extent_map *em,
4621 struct btrfs_path *path,
4622 struct btrfs_log_ctx *ctx)
4624 struct btrfs_drop_extents_args drop_args = { 0 };
4625 struct btrfs_root *log = inode->root->log_root;
4626 struct btrfs_file_extent_item fi = { 0 };
4627 struct extent_buffer *leaf;
4628 struct btrfs_key key;
4629 u64 extent_offset = em->start - em->orig_start;
4633 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4634 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4635 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4637 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4639 block_len = max(em->block_len, em->orig_block_len);
4640 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4641 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4642 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4643 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4644 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4646 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4649 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4650 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4651 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4652 btrfs_set_stack_file_extent_compression(&fi, em->compress_type);
4654 ret = log_extent_csums(trans, inode, log, em, ctx);
4659 * If this is the first time we are logging the inode in the current
4660 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4661 * because it does a deletion search, which always acquires write locks
4662 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4663 * but also adds significant contention in a log tree, since log trees
4664 * are small, with a root at level 2 or 3 at most, due to their short
4667 if (ctx->logged_before) {
4668 drop_args.path = path;
4669 drop_args.start = em->start;
4670 drop_args.end = em->start + em->len;
4671 drop_args.replace_extent = true;
4672 drop_args.extent_item_size = sizeof(fi);
4673 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4678 if (!drop_args.extent_inserted) {
4679 key.objectid = btrfs_ino(inode);
4680 key.type = BTRFS_EXTENT_DATA_KEY;
4681 key.offset = em->start;
4683 ret = btrfs_insert_empty_item(trans, log, path, &key,
4688 leaf = path->nodes[0];
4689 write_extent_buffer(leaf, &fi,
4690 btrfs_item_ptr_offset(leaf, path->slots[0]),
4692 btrfs_mark_buffer_dirty(leaf);
4694 btrfs_release_path(path);
4700 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4701 * lose them after doing a full/fast fsync and replaying the log. We scan the
4702 * subvolume's root instead of iterating the inode's extent map tree because
4703 * otherwise we can log incorrect extent items based on extent map conversion.
4704 * That can happen due to the fact that extent maps are merged when they
4705 * are not in the extent map tree's list of modified extents.
4707 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4708 struct btrfs_inode *inode,
4709 struct btrfs_path *path)
4711 struct btrfs_root *root = inode->root;
4712 struct btrfs_key key;
4713 const u64 i_size = i_size_read(&inode->vfs_inode);
4714 const u64 ino = btrfs_ino(inode);
4715 struct btrfs_path *dst_path = NULL;
4716 bool dropped_extents = false;
4717 u64 truncate_offset = i_size;
4718 struct extent_buffer *leaf;
4724 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4728 key.type = BTRFS_EXTENT_DATA_KEY;
4729 key.offset = i_size;
4730 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4735 * We must check if there is a prealloc extent that starts before the
4736 * i_size and crosses the i_size boundary. This is to ensure later we
4737 * truncate down to the end of that extent and not to the i_size, as
4738 * otherwise we end up losing part of the prealloc extent after a log
4739 * replay and with an implicit hole if there is another prealloc extent
4740 * that starts at an offset beyond i_size.
4742 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4747 struct btrfs_file_extent_item *ei;
4749 leaf = path->nodes[0];
4750 slot = path->slots[0];
4751 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4753 if (btrfs_file_extent_type(leaf, ei) ==
4754 BTRFS_FILE_EXTENT_PREALLOC) {
4757 btrfs_item_key_to_cpu(leaf, &key, slot);
4758 extent_end = key.offset +
4759 btrfs_file_extent_num_bytes(leaf, ei);
4761 if (extent_end > i_size)
4762 truncate_offset = extent_end;
4769 leaf = path->nodes[0];
4770 slot = path->slots[0];
4772 if (slot >= btrfs_header_nritems(leaf)) {
4774 ret = copy_items(trans, inode, dst_path, path,
4775 start_slot, ins_nr, 1, 0);
4780 ret = btrfs_next_leaf(root, path);
4790 btrfs_item_key_to_cpu(leaf, &key, slot);
4791 if (key.objectid > ino)
4793 if (WARN_ON_ONCE(key.objectid < ino) ||
4794 key.type < BTRFS_EXTENT_DATA_KEY ||
4795 key.offset < i_size) {
4799 if (!dropped_extents) {
4801 * Avoid logging extent items logged in past fsync calls
4802 * and leading to duplicate keys in the log tree.
4804 ret = truncate_inode_items(trans, root->log_root, inode,
4806 BTRFS_EXTENT_DATA_KEY);
4809 dropped_extents = true;
4816 dst_path = btrfs_alloc_path();
4824 ret = copy_items(trans, inode, dst_path, path,
4825 start_slot, ins_nr, 1, 0);
4827 btrfs_release_path(path);
4828 btrfs_free_path(dst_path);
4832 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4833 struct btrfs_inode *inode,
4834 struct btrfs_path *path,
4835 struct btrfs_log_ctx *ctx)
4837 struct btrfs_ordered_extent *ordered;
4838 struct btrfs_ordered_extent *tmp;
4839 struct extent_map *em, *n;
4840 struct list_head extents;
4841 struct extent_map_tree *tree = &inode->extent_tree;
4845 INIT_LIST_HEAD(&extents);
4847 write_lock(&tree->lock);
4849 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4850 list_del_init(&em->list);
4852 * Just an arbitrary number, this can be really CPU intensive
4853 * once we start getting a lot of extents, and really once we
4854 * have a bunch of extents we just want to commit since it will
4857 if (++num > 32768) {
4858 list_del_init(&tree->modified_extents);
4863 if (em->generation < trans->transid)
4866 /* We log prealloc extents beyond eof later. */
4867 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4868 em->start >= i_size_read(&inode->vfs_inode))
4871 /* Need a ref to keep it from getting evicted from cache */
4872 refcount_inc(&em->refs);
4873 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4874 list_add_tail(&em->list, &extents);
4878 list_sort(NULL, &extents, extent_cmp);
4880 while (!list_empty(&extents)) {
4881 em = list_entry(extents.next, struct extent_map, list);
4883 list_del_init(&em->list);
4886 * If we had an error we just need to delete everybody from our
4890 clear_em_logging(tree, em);
4891 free_extent_map(em);
4895 write_unlock(&tree->lock);
4897 ret = log_one_extent(trans, inode, em, path, ctx);
4898 write_lock(&tree->lock);
4899 clear_em_logging(tree, em);
4900 free_extent_map(em);
4902 WARN_ON(!list_empty(&extents));
4903 write_unlock(&tree->lock);
4906 ret = btrfs_log_prealloc_extents(trans, inode, path);
4911 * We have logged all extents successfully, now make sure the commit of
4912 * the current transaction waits for the ordered extents to complete
4913 * before it commits and wipes out the log trees, otherwise we would
4914 * lose data if an ordered extents completes after the transaction
4915 * commits and a power failure happens after the transaction commit.
4917 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4918 list_del_init(&ordered->log_list);
4919 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4921 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4922 spin_lock_irq(&inode->ordered_tree.lock);
4923 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4924 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4925 atomic_inc(&trans->transaction->pending_ordered);
4927 spin_unlock_irq(&inode->ordered_tree.lock);
4929 btrfs_put_ordered_extent(ordered);
4935 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4936 struct btrfs_path *path, u64 *size_ret)
4938 struct btrfs_key key;
4941 key.objectid = btrfs_ino(inode);
4942 key.type = BTRFS_INODE_ITEM_KEY;
4945 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4948 } else if (ret > 0) {
4951 struct btrfs_inode_item *item;
4953 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4954 struct btrfs_inode_item);
4955 *size_ret = btrfs_inode_size(path->nodes[0], item);
4957 * If the in-memory inode's i_size is smaller then the inode
4958 * size stored in the btree, return the inode's i_size, so
4959 * that we get a correct inode size after replaying the log
4960 * when before a power failure we had a shrinking truncate
4961 * followed by addition of a new name (rename / new hard link).
4962 * Otherwise return the inode size from the btree, to avoid
4963 * data loss when replaying a log due to previously doing a
4964 * write that expands the inode's size and logging a new name
4965 * immediately after.
4967 if (*size_ret > inode->vfs_inode.i_size)
4968 *size_ret = inode->vfs_inode.i_size;
4971 btrfs_release_path(path);
4976 * At the moment we always log all xattrs. This is to figure out at log replay
4977 * time which xattrs must have their deletion replayed. If a xattr is missing
4978 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4979 * because if a xattr is deleted, the inode is fsynced and a power failure
4980 * happens, causing the log to be replayed the next time the fs is mounted,
4981 * we want the xattr to not exist anymore (same behaviour as other filesystems
4982 * with a journal, ext3/4, xfs, f2fs, etc).
4984 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4985 struct btrfs_inode *inode,
4986 struct btrfs_path *path,
4987 struct btrfs_path *dst_path)
4989 struct btrfs_root *root = inode->root;
4991 struct btrfs_key key;
4992 const u64 ino = btrfs_ino(inode);
4995 bool found_xattrs = false;
4997 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
5001 key.type = BTRFS_XATTR_ITEM_KEY;
5004 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5009 int slot = path->slots[0];
5010 struct extent_buffer *leaf = path->nodes[0];
5011 int nritems = btrfs_header_nritems(leaf);
5013 if (slot >= nritems) {
5015 ret = copy_items(trans, inode, dst_path, path,
5016 start_slot, ins_nr, 1, 0);
5021 ret = btrfs_next_leaf(root, path);
5029 btrfs_item_key_to_cpu(leaf, &key, slot);
5030 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5037 found_xattrs = true;
5041 ret = copy_items(trans, inode, dst_path, path,
5042 start_slot, ins_nr, 1, 0);
5048 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5054 * When using the NO_HOLES feature if we punched a hole that causes the
5055 * deletion of entire leafs or all the extent items of the first leaf (the one
5056 * that contains the inode item and references) we may end up not processing
5057 * any extents, because there are no leafs with a generation matching the
5058 * current transaction that have extent items for our inode. So we need to find
5059 * if any holes exist and then log them. We also need to log holes after any
5060 * truncate operation that changes the inode's size.
5062 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5063 struct btrfs_inode *inode,
5064 struct btrfs_path *path)
5066 struct btrfs_root *root = inode->root;
5067 struct btrfs_fs_info *fs_info = root->fs_info;
5068 struct btrfs_key key;
5069 const u64 ino = btrfs_ino(inode);
5070 const u64 i_size = i_size_read(&inode->vfs_inode);
5071 u64 prev_extent_end = 0;
5074 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5078 key.type = BTRFS_EXTENT_DATA_KEY;
5081 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5086 struct extent_buffer *leaf = path->nodes[0];
5088 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5089 ret = btrfs_next_leaf(root, path);
5096 leaf = path->nodes[0];
5099 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5100 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5103 /* We have a hole, log it. */
5104 if (prev_extent_end < key.offset) {
5105 const u64 hole_len = key.offset - prev_extent_end;
5108 * Release the path to avoid deadlocks with other code
5109 * paths that search the root while holding locks on
5110 * leafs from the log root.
5112 btrfs_release_path(path);
5113 ret = btrfs_insert_hole_extent(trans, root->log_root,
5114 ino, prev_extent_end,
5120 * Search for the same key again in the root. Since it's
5121 * an extent item and we are holding the inode lock, the
5122 * key must still exist. If it doesn't just emit warning
5123 * and return an error to fall back to a transaction
5126 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5129 if (WARN_ON(ret > 0))
5131 leaf = path->nodes[0];
5134 prev_extent_end = btrfs_file_extent_end(path);
5139 if (prev_extent_end < i_size) {
5142 btrfs_release_path(path);
5143 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5144 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5145 prev_extent_end, hole_len);
5154 * When we are logging a new inode X, check if it doesn't have a reference that
5155 * matches the reference from some other inode Y created in a past transaction
5156 * and that was renamed in the current transaction. If we don't do this, then at
5157 * log replay time we can lose inode Y (and all its files if it's a directory):
5160 * echo "hello world" > /mnt/x/foobar
5163 * mkdir /mnt/x # or touch /mnt/x
5164 * xfs_io -c fsync /mnt/x
5166 * mount fs, trigger log replay
5168 * After the log replay procedure, we would lose the first directory and all its
5169 * files (file foobar).
5170 * For the case where inode Y is not a directory we simply end up losing it:
5172 * echo "123" > /mnt/foo
5174 * mv /mnt/foo /mnt/bar
5175 * echo "abc" > /mnt/foo
5176 * xfs_io -c fsync /mnt/foo
5179 * We also need this for cases where a snapshot entry is replaced by some other
5180 * entry (file or directory) otherwise we end up with an unreplayable log due to
5181 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5182 * if it were a regular entry:
5185 * btrfs subvolume snapshot /mnt /mnt/x/snap
5186 * btrfs subvolume delete /mnt/x/snap
5189 * fsync /mnt/x or fsync some new file inside it
5192 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5193 * the same transaction.
5195 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5197 const struct btrfs_key *key,
5198 struct btrfs_inode *inode,
5199 u64 *other_ino, u64 *other_parent)
5202 struct btrfs_path *search_path;
5205 u32 item_size = btrfs_item_size(eb, slot);
5207 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5209 search_path = btrfs_alloc_path();
5212 search_path->search_commit_root = 1;
5213 search_path->skip_locking = 1;
5215 while (cur_offset < item_size) {
5219 unsigned long name_ptr;
5220 struct btrfs_dir_item *di;
5222 if (key->type == BTRFS_INODE_REF_KEY) {
5223 struct btrfs_inode_ref *iref;
5225 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5226 parent = key->offset;
5227 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5228 name_ptr = (unsigned long)(iref + 1);
5229 this_len = sizeof(*iref) + this_name_len;
5231 struct btrfs_inode_extref *extref;
5233 extref = (struct btrfs_inode_extref *)(ptr +
5235 parent = btrfs_inode_extref_parent(eb, extref);
5236 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5237 name_ptr = (unsigned long)&extref->name;
5238 this_len = sizeof(*extref) + this_name_len;
5241 if (this_name_len > name_len) {
5244 new_name = krealloc(name, this_name_len, GFP_NOFS);
5249 name_len = this_name_len;
5253 read_extent_buffer(eb, name, name_ptr, this_name_len);
5254 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5255 parent, name, this_name_len, 0);
5256 if (di && !IS_ERR(di)) {
5257 struct btrfs_key di_key;
5259 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5261 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5262 if (di_key.objectid != key->objectid) {
5264 *other_ino = di_key.objectid;
5265 *other_parent = parent;
5273 } else if (IS_ERR(di)) {
5277 btrfs_release_path(search_path);
5279 cur_offset += this_len;
5283 btrfs_free_path(search_path);
5289 * Check if we need to log an inode. This is used in contexts where while
5290 * logging an inode we need to log another inode (either that it exists or in
5291 * full mode). This is used instead of btrfs_inode_in_log() because the later
5292 * requires the inode to be in the log and have the log transaction committed,
5293 * while here we do not care if the log transaction was already committed - our
5294 * caller will commit the log later - and we want to avoid logging an inode
5295 * multiple times when multiple tasks have joined the same log transaction.
5297 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5298 const struct btrfs_inode *inode)
5301 * If a directory was not modified, no dentries added or removed, we can
5302 * and should avoid logging it.
5304 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5308 * If this inode does not have new/updated/deleted xattrs since the last
5309 * time it was logged and is flagged as logged in the current transaction,
5310 * we can skip logging it. As for new/deleted names, those are updated in
5311 * the log by link/unlink/rename operations.
5312 * In case the inode was logged and then evicted and reloaded, its
5313 * logged_trans will be 0, in which case we have to fully log it since
5314 * logged_trans is a transient field, not persisted.
5316 if (inode->logged_trans == trans->transid &&
5317 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5323 struct btrfs_dir_list {
5325 struct list_head list;
5329 * Log the inodes of the new dentries of a directory.
5330 * See process_dir_items_leaf() for details about why it is needed.
5331 * This is a recursive operation - if an existing dentry corresponds to a
5332 * directory, that directory's new entries are logged too (same behaviour as
5333 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5334 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5335 * complains about the following circular lock dependency / possible deadlock:
5339 * lock(&type->i_mutex_dir_key#3/2);
5340 * lock(sb_internal#2);
5341 * lock(&type->i_mutex_dir_key#3/2);
5342 * lock(&sb->s_type->i_mutex_key#14);
5344 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5345 * sb_start_intwrite() in btrfs_start_transaction().
5346 * Not acquiring the VFS lock of the inodes is still safe because:
5348 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5349 * that while logging the inode new references (names) are added or removed
5350 * from the inode, leaving the logged inode item with a link count that does
5351 * not match the number of logged inode reference items. This is fine because
5352 * at log replay time we compute the real number of links and correct the
5353 * link count in the inode item (see replay_one_buffer() and
5354 * link_to_fixup_dir());
5356 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5357 * while logging the inode's items new index items (key type
5358 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5359 * has a size that doesn't match the sum of the lengths of all the logged
5360 * names - this is ok, not a problem, because at log replay time we set the
5361 * directory's i_size to the correct value (see replay_one_name() and
5362 * do_overwrite_item()).
5364 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5365 struct btrfs_inode *start_inode,
5366 struct btrfs_log_ctx *ctx)
5368 struct btrfs_root *root = start_inode->root;
5369 struct btrfs_fs_info *fs_info = root->fs_info;
5370 struct btrfs_path *path;
5371 LIST_HEAD(dir_list);
5372 struct btrfs_dir_list *dir_elem;
5373 u64 ino = btrfs_ino(start_inode);
5377 * If we are logging a new name, as part of a link or rename operation,
5378 * don't bother logging new dentries, as we just want to log the names
5379 * of an inode and that any new parents exist.
5381 if (ctx->logging_new_name)
5384 path = btrfs_alloc_path();
5389 struct extent_buffer *leaf;
5390 struct btrfs_key min_key;
5391 bool continue_curr_inode = true;
5395 min_key.objectid = ino;
5396 min_key.type = BTRFS_DIR_INDEX_KEY;
5399 btrfs_release_path(path);
5400 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
5403 } else if (ret > 0) {
5408 leaf = path->nodes[0];
5409 nritems = btrfs_header_nritems(leaf);
5410 for (i = path->slots[0]; i < nritems; i++) {
5411 struct btrfs_dir_item *di;
5412 struct btrfs_key di_key;
5413 struct inode *di_inode;
5414 int log_mode = LOG_INODE_EXISTS;
5417 btrfs_item_key_to_cpu(leaf, &min_key, i);
5418 if (min_key.objectid != ino ||
5419 min_key.type != BTRFS_DIR_INDEX_KEY) {
5420 continue_curr_inode = false;
5424 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5425 type = btrfs_dir_type(leaf, di);
5426 if (btrfs_dir_transid(leaf, di) < trans->transid)
5428 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5429 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5432 btrfs_release_path(path);
5433 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5434 if (IS_ERR(di_inode)) {
5435 ret = PTR_ERR(di_inode);
5439 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5440 btrfs_add_delayed_iput(di_inode);
5444 ctx->log_new_dentries = false;
5445 if (type == BTRFS_FT_DIR)
5446 log_mode = LOG_INODE_ALL;
5447 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5449 btrfs_add_delayed_iput(di_inode);
5452 if (ctx->log_new_dentries) {
5453 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5458 dir_elem->ino = di_key.objectid;
5459 list_add_tail(&dir_elem->list, &dir_list);
5464 if (continue_curr_inode && min_key.offset < (u64)-1) {
5470 if (list_empty(&dir_list))
5473 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5474 ino = dir_elem->ino;
5475 list_del(&dir_elem->list);
5479 btrfs_free_path(path);
5481 struct btrfs_dir_list *next;
5483 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5490 struct btrfs_ino_list {
5493 struct list_head list;
5496 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5498 struct btrfs_ino_list *curr;
5499 struct btrfs_ino_list *next;
5501 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5502 list_del(&curr->list);
5507 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5508 struct btrfs_path *path)
5510 struct btrfs_key key;
5514 key.type = BTRFS_INODE_ITEM_KEY;
5517 path->search_commit_root = 1;
5518 path->skip_locking = 1;
5520 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5521 if (WARN_ON_ONCE(ret > 0)) {
5523 * We have previously found the inode through the commit root
5524 * so this should not happen. If it does, just error out and
5525 * fallback to a transaction commit.
5528 } else if (ret == 0) {
5529 struct btrfs_inode_item *item;
5531 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5532 struct btrfs_inode_item);
5533 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5537 btrfs_release_path(path);
5538 path->search_commit_root = 0;
5539 path->skip_locking = 0;
5544 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5545 struct btrfs_root *root,
5546 struct btrfs_path *path,
5547 u64 ino, u64 parent,
5548 struct btrfs_log_ctx *ctx)
5550 struct btrfs_ino_list *ino_elem;
5551 struct inode *inode;
5554 * It's rare to have a lot of conflicting inodes, in practice it is not
5555 * common to have more than 1 or 2. We don't want to collect too many,
5556 * as we could end up logging too many inodes (even if only in
5557 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5560 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES)
5561 return BTRFS_LOG_FORCE_COMMIT;
5563 inode = btrfs_iget(root->fs_info->sb, ino, root);
5565 * If the other inode that had a conflicting dir entry was deleted in
5566 * the current transaction then we either:
5568 * 1) Log the parent directory (later after adding it to the list) if
5569 * the inode is a directory. This is because it may be a deleted
5570 * subvolume/snapshot or it may be a regular directory that had
5571 * deleted subvolumes/snapshots (or subdirectories that had them),
5572 * and at the moment we can't deal with dropping subvolumes/snapshots
5573 * during log replay. So we just log the parent, which will result in
5574 * a fallback to a transaction commit if we are dealing with those
5575 * cases (last_unlink_trans will match the current transaction);
5577 * 2) Do nothing if it's not a directory. During log replay we simply
5578 * unlink the conflicting dentry from the parent directory and then
5579 * add the dentry for our inode. Like this we can avoid logging the
5580 * parent directory (and maybe fallback to a transaction commit in
5581 * case it has a last_unlink_trans == trans->transid, due to moving
5582 * some inode from it to some other directory).
5584 if (IS_ERR(inode)) {
5585 int ret = PTR_ERR(inode);
5590 ret = conflicting_inode_is_dir(root, ino, path);
5591 /* Not a directory or we got an error. */
5595 /* Conflicting inode is a directory, so we'll log its parent. */
5596 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5599 ino_elem->ino = ino;
5600 ino_elem->parent = parent;
5601 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5602 ctx->num_conflict_inodes++;
5608 * If the inode was already logged skip it - otherwise we can hit an
5609 * infinite loop. Example:
5611 * From the commit root (previous transaction) we have the following
5614 * inode 257 a directory
5615 * inode 258 with references "zz" and "zz_link" on inode 257
5616 * inode 259 with reference "a" on inode 257
5618 * And in the current (uncommitted) transaction we have:
5620 * inode 257 a directory, unchanged
5621 * inode 258 with references "a" and "a2" on inode 257
5622 * inode 259 with reference "zz_link" on inode 257
5623 * inode 261 with reference "zz" on inode 257
5625 * When logging inode 261 the following infinite loop could
5626 * happen if we don't skip already logged inodes:
5628 * - we detect inode 258 as a conflicting inode, with inode 261
5629 * on reference "zz", and log it;
5631 * - we detect inode 259 as a conflicting inode, with inode 258
5632 * on reference "a", and log it;
5634 * - we detect inode 258 as a conflicting inode, with inode 259
5635 * on reference "zz_link", and log it - again! After this we
5636 * repeat the above steps forever.
5638 * Here we can use need_log_inode() because we only need to log the
5639 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5640 * so that the log ends up with the new name and without the old name.
5642 if (!need_log_inode(trans, BTRFS_I(inode))) {
5643 btrfs_add_delayed_iput(inode);
5647 btrfs_add_delayed_iput(inode);
5649 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5652 ino_elem->ino = ino;
5653 ino_elem->parent = parent;
5654 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5655 ctx->num_conflict_inodes++;
5660 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5661 struct btrfs_root *root,
5662 struct btrfs_log_ctx *ctx)
5664 struct btrfs_fs_info *fs_info = root->fs_info;
5668 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5669 * otherwise we could have unbounded recursion of btrfs_log_inode()
5670 * calls. This check guarantees we can have only 1 level of recursion.
5672 if (ctx->logging_conflict_inodes)
5675 ctx->logging_conflict_inodes = true;
5678 * New conflicting inodes may be found and added to the list while we
5679 * are logging a conflicting inode, so keep iterating while the list is
5682 while (!list_empty(&ctx->conflict_inodes)) {
5683 struct btrfs_ino_list *curr;
5684 struct inode *inode;
5688 curr = list_first_entry(&ctx->conflict_inodes,
5689 struct btrfs_ino_list, list);
5691 parent = curr->parent;
5692 list_del(&curr->list);
5695 inode = btrfs_iget(fs_info->sb, ino, root);
5697 * If the other inode that had a conflicting dir entry was
5698 * deleted in the current transaction, we need to log its parent
5699 * directory. See the comment at add_conflicting_inode().
5701 if (IS_ERR(inode)) {
5702 ret = PTR_ERR(inode);
5706 inode = btrfs_iget(fs_info->sb, parent, root);
5707 if (IS_ERR(inode)) {
5708 ret = PTR_ERR(inode);
5713 * Always log the directory, we cannot make this
5714 * conditional on need_log_inode() because the directory
5715 * might have been logged in LOG_INODE_EXISTS mode or
5716 * the dir index of the conflicting inode is not in a
5717 * dir index key range logged for the directory. So we
5718 * must make sure the deletion is recorded.
5720 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5721 LOG_INODE_ALL, ctx);
5722 btrfs_add_delayed_iput(inode);
5729 * Here we can use need_log_inode() because we only need to log
5730 * the inode in LOG_INODE_EXISTS mode and rename operations
5731 * update the log, so that the log ends up with the new name and
5732 * without the old name.
5734 * We did this check at add_conflicting_inode(), but here we do
5735 * it again because if some other task logged the inode after
5736 * that, we can avoid doing it again.
5738 if (!need_log_inode(trans, BTRFS_I(inode))) {
5739 btrfs_add_delayed_iput(inode);
5744 * We are safe logging the other inode without acquiring its
5745 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5746 * are safe against concurrent renames of the other inode as
5747 * well because during a rename we pin the log and update the
5748 * log with the new name before we unpin it.
5750 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5751 btrfs_add_delayed_iput(inode);
5756 ctx->logging_conflict_inodes = false;
5758 free_conflicting_inodes(ctx);
5763 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5764 struct btrfs_inode *inode,
5765 struct btrfs_key *min_key,
5766 const struct btrfs_key *max_key,
5767 struct btrfs_path *path,
5768 struct btrfs_path *dst_path,
5769 const u64 logged_isize,
5770 const int inode_only,
5771 struct btrfs_log_ctx *ctx,
5772 bool *need_log_inode_item)
5774 const u64 i_size = i_size_read(&inode->vfs_inode);
5775 struct btrfs_root *root = inode->root;
5776 int ins_start_slot = 0;
5781 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5789 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5790 if (min_key->objectid != max_key->objectid)
5792 if (min_key->type > max_key->type)
5795 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5796 *need_log_inode_item = false;
5797 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5798 min_key->offset >= i_size) {
5800 * Extents at and beyond eof are logged with
5801 * btrfs_log_prealloc_extents().
5802 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5803 * and no keys greater than that, so bail out.
5806 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5807 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5808 (inode->generation == trans->transid ||
5809 ctx->logging_conflict_inodes)) {
5811 u64 other_parent = 0;
5813 ret = btrfs_check_ref_name_override(path->nodes[0],
5814 path->slots[0], min_key, inode,
5815 &other_ino, &other_parent);
5818 } else if (ret > 0 &&
5819 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5824 ins_start_slot = path->slots[0];
5826 ret = copy_items(trans, inode, dst_path, path,
5827 ins_start_slot, ins_nr,
5828 inode_only, logged_isize);
5833 btrfs_release_path(path);
5834 ret = add_conflicting_inode(trans, root, path,
5841 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5842 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5845 ret = copy_items(trans, inode, dst_path, path,
5847 ins_nr, inode_only, logged_isize);
5854 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5857 } else if (!ins_nr) {
5858 ins_start_slot = path->slots[0];
5863 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5864 ins_nr, inode_only, logged_isize);
5868 ins_start_slot = path->slots[0];
5871 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5872 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5877 ret = copy_items(trans, inode, dst_path, path,
5878 ins_start_slot, ins_nr, inode_only,
5884 btrfs_release_path(path);
5886 if (min_key->offset < (u64)-1) {
5888 } else if (min_key->type < max_key->type) {
5890 min_key->offset = 0;
5896 * We may process many leaves full of items for our inode, so
5897 * avoid monopolizing a cpu for too long by rescheduling while
5898 * not holding locks on any tree.
5903 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5904 ins_nr, inode_only, logged_isize);
5909 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5911 * Release the path because otherwise we might attempt to double
5912 * lock the same leaf with btrfs_log_prealloc_extents() below.
5914 btrfs_release_path(path);
5915 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5921 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
5922 struct btrfs_root *log,
5923 struct btrfs_path *path,
5924 const struct btrfs_item_batch *batch,
5925 const struct btrfs_delayed_item *first_item)
5927 const struct btrfs_delayed_item *curr = first_item;
5930 ret = btrfs_insert_empty_items(trans, log, path, batch);
5934 for (int i = 0; i < batch->nr; i++) {
5937 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
5938 write_extent_buffer(path->nodes[0], &curr->data,
5939 (unsigned long)data_ptr, curr->data_len);
5940 curr = list_next_entry(curr, log_list);
5944 btrfs_release_path(path);
5949 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
5950 struct btrfs_inode *inode,
5951 struct btrfs_path *path,
5952 const struct list_head *delayed_ins_list,
5953 struct btrfs_log_ctx *ctx)
5955 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
5956 const int max_batch_size = 195;
5957 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
5958 const u64 ino = btrfs_ino(inode);
5959 struct btrfs_root *log = inode->root->log_root;
5960 struct btrfs_item_batch batch = {
5962 .total_data_size = 0,
5964 const struct btrfs_delayed_item *first = NULL;
5965 const struct btrfs_delayed_item *curr;
5967 struct btrfs_key *ins_keys;
5969 u64 curr_batch_size = 0;
5973 /* We are adding dir index items to the log tree. */
5974 lockdep_assert_held(&inode->log_mutex);
5977 * We collect delayed items before copying index keys from the subvolume
5978 * to the log tree. However just after we collected them, they may have
5979 * been flushed (all of them or just some of them), and therefore we
5980 * could have copied them from the subvolume tree to the log tree.
5981 * So find the first delayed item that was not yet logged (they are
5982 * sorted by index number).
5984 list_for_each_entry(curr, delayed_ins_list, log_list) {
5985 if (curr->index > inode->last_dir_index_offset) {
5991 /* Empty list or all delayed items were already logged. */
5995 ins_data = kmalloc(max_batch_size * sizeof(u32) +
5996 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
5999 ins_sizes = (u32 *)ins_data;
6000 batch.data_sizes = ins_sizes;
6001 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6002 batch.keys = ins_keys;
6005 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6006 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6008 if (curr_batch_size + curr_size > leaf_data_size ||
6009 batch.nr == max_batch_size) {
6010 ret = insert_delayed_items_batch(trans, log, path,
6016 batch.total_data_size = 0;
6017 curr_batch_size = 0;
6021 ins_sizes[batch_idx] = curr->data_len;
6022 ins_keys[batch_idx].objectid = ino;
6023 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6024 ins_keys[batch_idx].offset = curr->index;
6025 curr_batch_size += curr_size;
6026 batch.total_data_size += curr->data_len;
6029 curr = list_next_entry(curr, log_list);
6032 ASSERT(batch.nr >= 1);
6033 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6035 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6037 inode->last_dir_index_offset = curr->index;
6044 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6045 struct btrfs_inode *inode,
6046 struct btrfs_path *path,
6047 const struct list_head *delayed_del_list,
6048 struct btrfs_log_ctx *ctx)
6050 const u64 ino = btrfs_ino(inode);
6051 const struct btrfs_delayed_item *curr;
6053 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6056 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6057 u64 first_dir_index = curr->index;
6059 const struct btrfs_delayed_item *next;
6063 * Find a range of consecutive dir index items to delete. Like
6064 * this we log a single dir range item spanning several contiguous
6065 * dir items instead of logging one range item per dir index item.
6067 next = list_next_entry(curr, log_list);
6068 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6069 if (next->index != curr->index + 1)
6072 next = list_next_entry(next, log_list);
6075 last_dir_index = curr->index;
6076 ASSERT(last_dir_index >= first_dir_index);
6078 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6079 ino, first_dir_index, last_dir_index);
6082 curr = list_next_entry(curr, log_list);
6088 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6089 struct btrfs_inode *inode,
6090 struct btrfs_path *path,
6091 struct btrfs_log_ctx *ctx,
6092 const struct list_head *delayed_del_list,
6093 const struct btrfs_delayed_item *first,
6094 const struct btrfs_delayed_item **last_ret)
6096 const struct btrfs_delayed_item *next;
6097 struct extent_buffer *leaf = path->nodes[0];
6098 const int last_slot = btrfs_header_nritems(leaf) - 1;
6099 int slot = path->slots[0] + 1;
6100 const u64 ino = btrfs_ino(inode);
6102 next = list_next_entry(first, log_list);
6104 while (slot < last_slot &&
6105 !list_entry_is_head(next, delayed_del_list, log_list)) {
6106 struct btrfs_key key;
6108 btrfs_item_key_to_cpu(leaf, &key, slot);
6109 if (key.objectid != ino ||
6110 key.type != BTRFS_DIR_INDEX_KEY ||
6111 key.offset != next->index)
6116 next = list_next_entry(next, log_list);
6119 return btrfs_del_items(trans, inode->root->log_root, path,
6120 path->slots[0], slot - path->slots[0]);
6123 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6124 struct btrfs_inode *inode,
6125 struct btrfs_path *path,
6126 const struct list_head *delayed_del_list,
6127 struct btrfs_log_ctx *ctx)
6129 struct btrfs_root *log = inode->root->log_root;
6130 const struct btrfs_delayed_item *curr;
6131 u64 last_range_start;
6132 u64 last_range_end = 0;
6133 struct btrfs_key key;
6135 key.objectid = btrfs_ino(inode);
6136 key.type = BTRFS_DIR_INDEX_KEY;
6137 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6140 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6141 const struct btrfs_delayed_item *last = curr;
6142 u64 first_dir_index = curr->index;
6144 bool deleted_items = false;
6147 key.offset = curr->index;
6148 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6151 } else if (ret == 0) {
6152 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6153 delayed_del_list, curr,
6157 deleted_items = true;
6160 btrfs_release_path(path);
6163 * If we deleted items from the leaf, it means we have a range
6164 * item logging their range, so no need to add one or update an
6165 * existing one. Otherwise we have to log a dir range item.
6170 last_dir_index = last->index;
6171 ASSERT(last_dir_index >= first_dir_index);
6173 * If this range starts right after where the previous one ends,
6174 * then we want to reuse the previous range item and change its
6175 * end offset to the end of this range. This is just to minimize
6176 * leaf space usage, by avoiding adding a new range item.
6178 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6179 first_dir_index = last_range_start;
6181 ret = insert_dir_log_key(trans, log, path, key.objectid,
6182 first_dir_index, last_dir_index);
6186 last_range_start = first_dir_index;
6187 last_range_end = last_dir_index;
6189 curr = list_next_entry(last, log_list);
6195 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6196 struct btrfs_inode *inode,
6197 struct btrfs_path *path,
6198 const struct list_head *delayed_del_list,
6199 struct btrfs_log_ctx *ctx)
6202 * We are deleting dir index items from the log tree or adding range
6205 lockdep_assert_held(&inode->log_mutex);
6207 if (list_empty(delayed_del_list))
6210 if (ctx->logged_before)
6211 return log_delayed_deletions_incremental(trans, inode, path,
6212 delayed_del_list, ctx);
6214 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6219 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6220 * items instead of the subvolume tree.
6222 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6223 struct btrfs_inode *inode,
6224 const struct list_head *delayed_ins_list,
6225 struct btrfs_log_ctx *ctx)
6227 const bool orig_log_new_dentries = ctx->log_new_dentries;
6228 struct btrfs_fs_info *fs_info = trans->fs_info;
6229 struct btrfs_delayed_item *item;
6233 * No need for the log mutex, plus to avoid potential deadlocks or
6234 * lockdep annotations due to nesting of delayed inode mutexes and log
6237 lockdep_assert_not_held(&inode->log_mutex);
6239 ASSERT(!ctx->logging_new_delayed_dentries);
6240 ctx->logging_new_delayed_dentries = true;
6242 list_for_each_entry(item, delayed_ins_list, log_list) {
6243 struct btrfs_dir_item *dir_item;
6244 struct inode *di_inode;
6245 struct btrfs_key key;
6246 int log_mode = LOG_INODE_EXISTS;
6248 dir_item = (struct btrfs_dir_item *)item->data;
6249 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6251 if (key.type == BTRFS_ROOT_ITEM_KEY)
6254 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6255 if (IS_ERR(di_inode)) {
6256 ret = PTR_ERR(di_inode);
6260 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6261 btrfs_add_delayed_iput(di_inode);
6265 if (btrfs_stack_dir_type(dir_item) == BTRFS_FT_DIR)
6266 log_mode = LOG_INODE_ALL;
6268 ctx->log_new_dentries = false;
6269 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6271 if (!ret && ctx->log_new_dentries)
6272 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6274 btrfs_add_delayed_iput(di_inode);
6280 ctx->log_new_dentries = orig_log_new_dentries;
6281 ctx->logging_new_delayed_dentries = false;
6286 /* log a single inode in the tree log.
6287 * At least one parent directory for this inode must exist in the tree
6288 * or be logged already.
6290 * Any items from this inode changed by the current transaction are copied
6291 * to the log tree. An extra reference is taken on any extents in this
6292 * file, allowing us to avoid a whole pile of corner cases around logging
6293 * blocks that have been removed from the tree.
6295 * See LOG_INODE_ALL and related defines for a description of what inode_only
6298 * This handles both files and directories.
6300 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6301 struct btrfs_inode *inode,
6303 struct btrfs_log_ctx *ctx)
6305 struct btrfs_path *path;
6306 struct btrfs_path *dst_path;
6307 struct btrfs_key min_key;
6308 struct btrfs_key max_key;
6309 struct btrfs_root *log = inode->root->log_root;
6311 bool fast_search = false;
6312 u64 ino = btrfs_ino(inode);
6313 struct extent_map_tree *em_tree = &inode->extent_tree;
6314 u64 logged_isize = 0;
6315 bool need_log_inode_item = true;
6316 bool xattrs_logged = false;
6317 bool inode_item_dropped = true;
6318 bool full_dir_logging = false;
6319 LIST_HEAD(delayed_ins_list);
6320 LIST_HEAD(delayed_del_list);
6322 path = btrfs_alloc_path();
6325 dst_path = btrfs_alloc_path();
6327 btrfs_free_path(path);
6331 min_key.objectid = ino;
6332 min_key.type = BTRFS_INODE_ITEM_KEY;
6335 max_key.objectid = ino;
6338 /* today the code can only do partial logging of directories */
6339 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6340 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6341 &inode->runtime_flags) &&
6342 inode_only >= LOG_INODE_EXISTS))
6343 max_key.type = BTRFS_XATTR_ITEM_KEY;
6345 max_key.type = (u8)-1;
6346 max_key.offset = (u64)-1;
6348 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6349 full_dir_logging = true;
6352 * If we are logging a directory while we are logging dentries of the
6353 * delayed items of some other inode, then we need to flush the delayed
6354 * items of this directory and not log the delayed items directly. This
6355 * is to prevent more than one level of recursion into btrfs_log_inode()
6356 * by having something like this:
6358 * $ mkdir -p a/b/c/d/e/f/g/h/...
6359 * $ xfs_io -c "fsync" a
6361 * Where all directories in the path did not exist before and are
6362 * created in the current transaction.
6363 * So in such a case we directly log the delayed items of the main
6364 * directory ("a") without flushing them first, while for each of its
6365 * subdirectories we flush their delayed items before logging them.
6366 * This prevents a potential unbounded recursion like this:
6369 * log_new_delayed_dentries()
6371 * log_new_delayed_dentries()
6373 * log_new_delayed_dentries()
6376 * We have thresholds for the maximum number of delayed items to have in
6377 * memory, and once they are hit, the items are flushed asynchronously.
6378 * However the limit is quite high, so lets prevent deep levels of
6379 * recursion to happen by limiting the maximum depth to be 1.
6381 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6382 ret = btrfs_commit_inode_delayed_items(trans, inode);
6387 mutex_lock(&inode->log_mutex);
6390 * For symlinks, we must always log their content, which is stored in an
6391 * inline extent, otherwise we could end up with an empty symlink after
6392 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6393 * one attempts to create an empty symlink).
6394 * We don't need to worry about flushing delalloc, because when we create
6395 * the inline extent when the symlink is created (we never have delalloc
6398 if (S_ISLNK(inode->vfs_inode.i_mode))
6399 inode_only = LOG_INODE_ALL;
6402 * Before logging the inode item, cache the value returned by
6403 * inode_logged(), because after that we have the need to figure out if
6404 * the inode was previously logged in this transaction.
6406 ret = inode_logged(trans, inode, path);
6409 ctx->logged_before = (ret == 1);
6413 * This is for cases where logging a directory could result in losing a
6414 * a file after replaying the log. For example, if we move a file from a
6415 * directory A to a directory B, then fsync directory A, we have no way
6416 * to known the file was moved from A to B, so logging just A would
6417 * result in losing the file after a log replay.
6419 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6420 btrfs_set_log_full_commit(trans);
6421 ret = BTRFS_LOG_FORCE_COMMIT;
6426 * a brute force approach to making sure we get the most uptodate
6427 * copies of everything.
6429 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6430 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6431 if (ctx->logged_before)
6432 ret = drop_inode_items(trans, log, path, inode,
6433 BTRFS_XATTR_ITEM_KEY);
6435 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6437 * Make sure the new inode item we write to the log has
6438 * the same isize as the current one (if it exists).
6439 * This is necessary to prevent data loss after log
6440 * replay, and also to prevent doing a wrong expanding
6441 * truncate - for e.g. create file, write 4K into offset
6442 * 0, fsync, write 4K into offset 4096, add hard link,
6443 * fsync some other file (to sync log), power fail - if
6444 * we use the inode's current i_size, after log replay
6445 * we get a 8Kb file, with the last 4Kb extent as a hole
6446 * (zeroes), as if an expanding truncate happened,
6447 * instead of getting a file of 4Kb only.
6449 ret = logged_inode_size(log, inode, path, &logged_isize);
6453 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6454 &inode->runtime_flags)) {
6455 if (inode_only == LOG_INODE_EXISTS) {
6456 max_key.type = BTRFS_XATTR_ITEM_KEY;
6457 if (ctx->logged_before)
6458 ret = drop_inode_items(trans, log, path,
6459 inode, max_key.type);
6461 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6462 &inode->runtime_flags);
6463 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6464 &inode->runtime_flags);
6465 if (ctx->logged_before)
6466 ret = truncate_inode_items(trans, log,
6469 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6470 &inode->runtime_flags) ||
6471 inode_only == LOG_INODE_EXISTS) {
6472 if (inode_only == LOG_INODE_ALL)
6474 max_key.type = BTRFS_XATTR_ITEM_KEY;
6475 if (ctx->logged_before)
6476 ret = drop_inode_items(trans, log, path, inode,
6479 if (inode_only == LOG_INODE_ALL)
6481 inode_item_dropped = false;
6490 * If we are logging a directory in full mode, collect the delayed items
6491 * before iterating the subvolume tree, so that we don't miss any new
6492 * dir index items in case they get flushed while or right after we are
6493 * iterating the subvolume tree.
6495 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6496 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6499 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6500 path, dst_path, logged_isize,
6502 &need_log_inode_item);
6506 btrfs_release_path(path);
6507 btrfs_release_path(dst_path);
6508 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6511 xattrs_logged = true;
6512 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6513 btrfs_release_path(path);
6514 btrfs_release_path(dst_path);
6515 ret = btrfs_log_holes(trans, inode, path);
6520 btrfs_release_path(path);
6521 btrfs_release_path(dst_path);
6522 if (need_log_inode_item) {
6523 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6527 * If we are doing a fast fsync and the inode was logged before
6528 * in this transaction, we don't need to log the xattrs because
6529 * they were logged before. If xattrs were added, changed or
6530 * deleted since the last time we logged the inode, then we have
6531 * already logged them because the inode had the runtime flag
6532 * BTRFS_INODE_COPY_EVERYTHING set.
6534 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6535 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6538 btrfs_release_path(path);
6542 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6545 } else if (inode_only == LOG_INODE_ALL) {
6546 struct extent_map *em, *n;
6548 write_lock(&em_tree->lock);
6549 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6550 list_del_init(&em->list);
6551 write_unlock(&em_tree->lock);
6554 if (full_dir_logging) {
6555 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6558 ret = log_delayed_insertion_items(trans, inode, path,
6559 &delayed_ins_list, ctx);
6562 ret = log_delayed_deletion_items(trans, inode, path,
6563 &delayed_del_list, ctx);
6568 spin_lock(&inode->lock);
6569 inode->logged_trans = trans->transid;
6571 * Don't update last_log_commit if we logged that an inode exists.
6572 * We do this for three reasons:
6574 * 1) We might have had buffered writes to this inode that were
6575 * flushed and had their ordered extents completed in this
6576 * transaction, but we did not previously log the inode with
6577 * LOG_INODE_ALL. Later the inode was evicted and after that
6578 * it was loaded again and this LOG_INODE_EXISTS log operation
6579 * happened. We must make sure that if an explicit fsync against
6580 * the inode is performed later, it logs the new extents, an
6581 * updated inode item, etc, and syncs the log. The same logic
6582 * applies to direct IO writes instead of buffered writes.
6584 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6585 * is logged with an i_size of 0 or whatever value was logged
6586 * before. If later the i_size of the inode is increased by a
6587 * truncate operation, the log is synced through an fsync of
6588 * some other inode and then finally an explicit fsync against
6589 * this inode is made, we must make sure this fsync logs the
6590 * inode with the new i_size, the hole between old i_size and
6591 * the new i_size, and syncs the log.
6593 * 3) If we are logging that an ancestor inode exists as part of
6594 * logging a new name from a link or rename operation, don't update
6595 * its last_log_commit - otherwise if an explicit fsync is made
6596 * against an ancestor, the fsync considers the inode in the log
6597 * and doesn't sync the log, resulting in the ancestor missing after
6598 * a power failure unless the log was synced as part of an fsync
6599 * against any other unrelated inode.
6601 if (inode_only != LOG_INODE_EXISTS)
6602 inode->last_log_commit = inode->last_sub_trans;
6603 spin_unlock(&inode->lock);
6606 * Reset the last_reflink_trans so that the next fsync does not need to
6607 * go through the slower path when logging extents and their checksums.
6609 if (inode_only == LOG_INODE_ALL)
6610 inode->last_reflink_trans = 0;
6613 mutex_unlock(&inode->log_mutex);
6615 btrfs_free_path(path);
6616 btrfs_free_path(dst_path);
6619 free_conflicting_inodes(ctx);
6621 ret = log_conflicting_inodes(trans, inode->root, ctx);
6623 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6625 ret = log_new_delayed_dentries(trans, inode,
6626 &delayed_ins_list, ctx);
6628 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6635 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6636 struct btrfs_inode *inode,
6637 struct btrfs_log_ctx *ctx)
6639 struct btrfs_fs_info *fs_info = trans->fs_info;
6641 struct btrfs_path *path;
6642 struct btrfs_key key;
6643 struct btrfs_root *root = inode->root;
6644 const u64 ino = btrfs_ino(inode);
6646 path = btrfs_alloc_path();
6649 path->skip_locking = 1;
6650 path->search_commit_root = 1;
6653 key.type = BTRFS_INODE_REF_KEY;
6655 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6660 struct extent_buffer *leaf = path->nodes[0];
6661 int slot = path->slots[0];
6666 if (slot >= btrfs_header_nritems(leaf)) {
6667 ret = btrfs_next_leaf(root, path);
6675 btrfs_item_key_to_cpu(leaf, &key, slot);
6676 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6677 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6680 item_size = btrfs_item_size(leaf, slot);
6681 ptr = btrfs_item_ptr_offset(leaf, slot);
6682 while (cur_offset < item_size) {
6683 struct btrfs_key inode_key;
6684 struct inode *dir_inode;
6686 inode_key.type = BTRFS_INODE_ITEM_KEY;
6687 inode_key.offset = 0;
6689 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6690 struct btrfs_inode_extref *extref;
6692 extref = (struct btrfs_inode_extref *)
6694 inode_key.objectid = btrfs_inode_extref_parent(
6696 cur_offset += sizeof(*extref);
6697 cur_offset += btrfs_inode_extref_name_len(leaf,
6700 inode_key.objectid = key.offset;
6701 cur_offset = item_size;
6704 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6707 * If the parent inode was deleted, return an error to
6708 * fallback to a transaction commit. This is to prevent
6709 * getting an inode that was moved from one parent A to
6710 * a parent B, got its former parent A deleted and then
6711 * it got fsync'ed, from existing at both parents after
6712 * a log replay (and the old parent still existing).
6719 * mv /mnt/B/bar /mnt/A/bar
6720 * mv -T /mnt/A /mnt/B
6724 * If we ignore the old parent B which got deleted,
6725 * after a log replay we would have file bar linked
6726 * at both parents and the old parent B would still
6729 if (IS_ERR(dir_inode)) {
6730 ret = PTR_ERR(dir_inode);
6734 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6735 btrfs_add_delayed_iput(dir_inode);
6739 ctx->log_new_dentries = false;
6740 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6741 LOG_INODE_ALL, ctx);
6742 if (!ret && ctx->log_new_dentries)
6743 ret = log_new_dir_dentries(trans,
6744 BTRFS_I(dir_inode), ctx);
6745 btrfs_add_delayed_iput(dir_inode);
6753 btrfs_free_path(path);
6757 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6758 struct btrfs_root *root,
6759 struct btrfs_path *path,
6760 struct btrfs_log_ctx *ctx)
6762 struct btrfs_key found_key;
6764 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6767 struct btrfs_fs_info *fs_info = root->fs_info;
6768 struct extent_buffer *leaf = path->nodes[0];
6769 int slot = path->slots[0];
6770 struct btrfs_key search_key;
6771 struct inode *inode;
6775 btrfs_release_path(path);
6777 ino = found_key.offset;
6779 search_key.objectid = found_key.offset;
6780 search_key.type = BTRFS_INODE_ITEM_KEY;
6781 search_key.offset = 0;
6782 inode = btrfs_iget(fs_info->sb, ino, root);
6784 return PTR_ERR(inode);
6786 if (BTRFS_I(inode)->generation >= trans->transid &&
6787 need_log_inode(trans, BTRFS_I(inode)))
6788 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6789 LOG_INODE_EXISTS, ctx);
6790 btrfs_add_delayed_iput(inode);
6794 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6797 search_key.type = BTRFS_INODE_REF_KEY;
6798 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6802 leaf = path->nodes[0];
6803 slot = path->slots[0];
6804 if (slot >= btrfs_header_nritems(leaf)) {
6805 ret = btrfs_next_leaf(root, path);
6810 leaf = path->nodes[0];
6811 slot = path->slots[0];
6814 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6815 if (found_key.objectid != search_key.objectid ||
6816 found_key.type != BTRFS_INODE_REF_KEY)
6822 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6823 struct btrfs_inode *inode,
6824 struct dentry *parent,
6825 struct btrfs_log_ctx *ctx)
6827 struct btrfs_root *root = inode->root;
6828 struct dentry *old_parent = NULL;
6829 struct super_block *sb = inode->vfs_inode.i_sb;
6833 if (!parent || d_really_is_negative(parent) ||
6837 inode = BTRFS_I(d_inode(parent));
6838 if (root != inode->root)
6841 if (inode->generation >= trans->transid &&
6842 need_log_inode(trans, inode)) {
6843 ret = btrfs_log_inode(trans, inode,
6844 LOG_INODE_EXISTS, ctx);
6848 if (IS_ROOT(parent))
6851 parent = dget_parent(parent);
6853 old_parent = parent;
6860 static int log_all_new_ancestors(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 const u64 ino = btrfs_ino(inode);
6867 struct btrfs_path *path;
6868 struct btrfs_key search_key;
6872 * For a single hard link case, go through a fast path that does not
6873 * need to iterate the fs/subvolume tree.
6875 if (inode->vfs_inode.i_nlink < 2)
6876 return log_new_ancestors_fast(trans, inode, parent, ctx);
6878 path = btrfs_alloc_path();
6882 search_key.objectid = ino;
6883 search_key.type = BTRFS_INODE_REF_KEY;
6884 search_key.offset = 0;
6886 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6893 struct extent_buffer *leaf = path->nodes[0];
6894 int slot = path->slots[0];
6895 struct btrfs_key found_key;
6897 if (slot >= btrfs_header_nritems(leaf)) {
6898 ret = btrfs_next_leaf(root, path);
6906 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6907 if (found_key.objectid != ino ||
6908 found_key.type > BTRFS_INODE_EXTREF_KEY)
6912 * Don't deal with extended references because they are rare
6913 * cases and too complex to deal with (we would need to keep
6914 * track of which subitem we are processing for each item in
6915 * this loop, etc). So just return some error to fallback to
6916 * a transaction commit.
6918 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6924 * Logging ancestors needs to do more searches on the fs/subvol
6925 * tree, so it releases the path as needed to avoid deadlocks.
6926 * Keep track of the last inode ref key and resume from that key
6927 * after logging all new ancestors for the current hard link.
6929 memcpy(&search_key, &found_key, sizeof(search_key));
6931 ret = log_new_ancestors(trans, root, path, ctx);
6934 btrfs_release_path(path);
6939 btrfs_free_path(path);
6944 * helper function around btrfs_log_inode to make sure newly created
6945 * parent directories also end up in the log. A minimal inode and backref
6946 * only logging is done of any parent directories that are older than
6947 * the last committed transaction
6949 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6950 struct btrfs_inode *inode,
6951 struct dentry *parent,
6953 struct btrfs_log_ctx *ctx)
6955 struct btrfs_root *root = inode->root;
6956 struct btrfs_fs_info *fs_info = root->fs_info;
6958 bool log_dentries = false;
6960 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6961 ret = BTRFS_LOG_FORCE_COMMIT;
6965 if (btrfs_root_refs(&root->root_item) == 0) {
6966 ret = BTRFS_LOG_FORCE_COMMIT;
6971 * Skip already logged inodes or inodes corresponding to tmpfiles
6972 * (since logging them is pointless, a link count of 0 means they
6973 * will never be accessible).
6975 if ((btrfs_inode_in_log(inode, trans->transid) &&
6976 list_empty(&ctx->ordered_extents)) ||
6977 inode->vfs_inode.i_nlink == 0) {
6978 ret = BTRFS_NO_LOG_SYNC;
6982 ret = start_log_trans(trans, root, ctx);
6986 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
6991 * for regular files, if its inode is already on disk, we don't
6992 * have to worry about the parents at all. This is because
6993 * we can use the last_unlink_trans field to record renames
6994 * and other fun in this file.
6996 if (S_ISREG(inode->vfs_inode.i_mode) &&
6997 inode->generation < trans->transid &&
6998 inode->last_unlink_trans < trans->transid) {
7003 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7004 log_dentries = true;
7007 * On unlink we must make sure all our current and old parent directory
7008 * inodes are fully logged. This is to prevent leaving dangling
7009 * directory index entries in directories that were our parents but are
7010 * not anymore. Not doing this results in old parent directory being
7011 * impossible to delete after log replay (rmdir will always fail with
7012 * error -ENOTEMPTY).
7018 * ln testdir/foo testdir/bar
7020 * unlink testdir/bar
7021 * xfs_io -c fsync testdir/foo
7023 * mount fs, triggers log replay
7025 * If we don't log the parent directory (testdir), after log replay the
7026 * directory still has an entry pointing to the file inode using the bar
7027 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7028 * the file inode has a link count of 1.
7034 * ln foo testdir/foo2
7035 * ln foo testdir/foo3
7037 * unlink testdir/foo3
7038 * xfs_io -c fsync foo
7040 * mount fs, triggers log replay
7042 * Similar as the first example, after log replay the parent directory
7043 * testdir still has an entry pointing to the inode file with name foo3
7044 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7045 * and has a link count of 2.
7047 if (inode->last_unlink_trans >= trans->transid) {
7048 ret = btrfs_log_all_parents(trans, inode, ctx);
7053 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7058 ret = log_new_dir_dentries(trans, inode, ctx);
7063 btrfs_set_log_full_commit(trans);
7064 ret = BTRFS_LOG_FORCE_COMMIT;
7068 btrfs_remove_log_ctx(root, ctx);
7069 btrfs_end_log_trans(root);
7075 * it is not safe to log dentry if the chunk root has added new
7076 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7077 * If this returns 1, you must commit the transaction to safely get your
7080 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7081 struct dentry *dentry,
7082 struct btrfs_log_ctx *ctx)
7084 struct dentry *parent = dget_parent(dentry);
7087 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7088 LOG_INODE_ALL, ctx);
7095 * should be called during mount to recover any replay any log trees
7098 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7101 struct btrfs_path *path;
7102 struct btrfs_trans_handle *trans;
7103 struct btrfs_key key;
7104 struct btrfs_key found_key;
7105 struct btrfs_root *log;
7106 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7107 struct walk_control wc = {
7108 .process_func = process_one_buffer,
7109 .stage = LOG_WALK_PIN_ONLY,
7112 path = btrfs_alloc_path();
7116 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7118 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7119 if (IS_ERR(trans)) {
7120 ret = PTR_ERR(trans);
7127 ret = walk_log_tree(trans, log_root_tree, &wc);
7129 btrfs_abort_transaction(trans, ret);
7134 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7135 key.offset = (u64)-1;
7136 key.type = BTRFS_ROOT_ITEM_KEY;
7139 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7142 btrfs_abort_transaction(trans, ret);
7146 if (path->slots[0] == 0)
7150 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7152 btrfs_release_path(path);
7153 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7156 log = btrfs_read_tree_root(log_root_tree, &found_key);
7159 btrfs_abort_transaction(trans, ret);
7163 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7165 if (IS_ERR(wc.replay_dest)) {
7166 ret = PTR_ERR(wc.replay_dest);
7169 * We didn't find the subvol, likely because it was
7170 * deleted. This is ok, simply skip this log and go to
7173 * We need to exclude the root because we can't have
7174 * other log replays overwriting this log as we'll read
7175 * it back in a few more times. This will keep our
7176 * block from being modified, and we'll just bail for
7177 * each subsequent pass.
7180 ret = btrfs_pin_extent_for_log_replay(trans,
7183 btrfs_put_root(log);
7187 btrfs_abort_transaction(trans, ret);
7191 wc.replay_dest->log_root = log;
7192 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7194 /* The loop needs to continue due to the root refs */
7195 btrfs_abort_transaction(trans, ret);
7197 ret = walk_log_tree(trans, log, &wc);
7199 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7200 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7203 btrfs_abort_transaction(trans, ret);
7206 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7207 struct btrfs_root *root = wc.replay_dest;
7209 btrfs_release_path(path);
7212 * We have just replayed everything, and the highest
7213 * objectid of fs roots probably has changed in case
7214 * some inode_item's got replayed.
7216 * root->objectid_mutex is not acquired as log replay
7217 * could only happen during mount.
7219 ret = btrfs_init_root_free_objectid(root);
7221 btrfs_abort_transaction(trans, ret);
7224 wc.replay_dest->log_root = NULL;
7225 btrfs_put_root(wc.replay_dest);
7226 btrfs_put_root(log);
7231 if (found_key.offset == 0)
7233 key.offset = found_key.offset - 1;
7235 btrfs_release_path(path);
7237 /* step one is to pin it all, step two is to replay just inodes */
7240 wc.process_func = replay_one_buffer;
7241 wc.stage = LOG_WALK_REPLAY_INODES;
7244 /* step three is to replay everything */
7245 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7250 btrfs_free_path(path);
7252 /* step 4: commit the transaction, which also unpins the blocks */
7253 ret = btrfs_commit_transaction(trans);
7257 log_root_tree->log_root = NULL;
7258 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7259 btrfs_put_root(log_root_tree);
7264 btrfs_end_transaction(wc.trans);
7265 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7266 btrfs_free_path(path);
7271 * there are some corner cases where we want to force a full
7272 * commit instead of allowing a directory to be logged.
7274 * They revolve around files there were unlinked from the directory, and
7275 * this function updates the parent directory so that a full commit is
7276 * properly done if it is fsync'd later after the unlinks are done.
7278 * Must be called before the unlink operations (updates to the subvolume tree,
7279 * inodes, etc) are done.
7281 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7282 struct btrfs_inode *dir, struct btrfs_inode *inode,
7286 * when we're logging a file, if it hasn't been renamed
7287 * or unlinked, and its inode is fully committed on disk,
7288 * we don't have to worry about walking up the directory chain
7289 * to log its parents.
7291 * So, we use the last_unlink_trans field to put this transid
7292 * into the file. When the file is logged we check it and
7293 * don't log the parents if the file is fully on disk.
7295 mutex_lock(&inode->log_mutex);
7296 inode->last_unlink_trans = trans->transid;
7297 mutex_unlock(&inode->log_mutex);
7300 * if this directory was already logged any new
7301 * names for this file/dir will get recorded
7303 if (dir->logged_trans == trans->transid)
7307 * if the inode we're about to unlink was logged,
7308 * the log will be properly updated for any new names
7310 if (inode->logged_trans == trans->transid)
7314 * when renaming files across directories, if the directory
7315 * there we're unlinking from gets fsync'd later on, there's
7316 * no way to find the destination directory later and fsync it
7317 * properly. So, we have to be conservative and force commits
7318 * so the new name gets discovered.
7323 /* we can safely do the unlink without any special recording */
7327 mutex_lock(&dir->log_mutex);
7328 dir->last_unlink_trans = trans->transid;
7329 mutex_unlock(&dir->log_mutex);
7333 * Make sure that if someone attempts to fsync the parent directory of a deleted
7334 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7335 * that after replaying the log tree of the parent directory's root we will not
7336 * see the snapshot anymore and at log replay time we will not see any log tree
7337 * corresponding to the deleted snapshot's root, which could lead to replaying
7338 * it after replaying the log tree of the parent directory (which would replay
7339 * the snapshot delete operation).
7341 * Must be called before the actual snapshot destroy operation (updates to the
7342 * parent root and tree of tree roots trees, etc) are done.
7344 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7345 struct btrfs_inode *dir)
7347 mutex_lock(&dir->log_mutex);
7348 dir->last_unlink_trans = trans->transid;
7349 mutex_unlock(&dir->log_mutex);
7353 * Update the log after adding a new name for an inode.
7355 * @trans: Transaction handle.
7356 * @old_dentry: The dentry associated with the old name and the old
7358 * @old_dir: The inode of the previous parent directory for the case
7359 * of a rename. For a link operation, it must be NULL.
7360 * @old_dir_index: The index number associated with the old name, meaningful
7361 * only for rename operations (when @old_dir is not NULL).
7362 * Ignored for link operations.
7363 * @parent: The dentry associated with the directory under which the
7364 * new name is located.
7366 * Call this after adding a new name for an inode, as a result of a link or
7367 * rename operation, and it will properly update the log to reflect the new name.
7369 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7370 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7371 u64 old_dir_index, struct dentry *parent)
7373 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7374 struct btrfs_root *root = inode->root;
7375 struct btrfs_log_ctx ctx;
7376 bool log_pinned = false;
7380 * this will force the logging code to walk the dentry chain
7383 if (!S_ISDIR(inode->vfs_inode.i_mode))
7384 inode->last_unlink_trans = trans->transid;
7387 * if this inode hasn't been logged and directory we're renaming it
7388 * from hasn't been logged, we don't need to log it
7390 ret = inode_logged(trans, inode, NULL);
7393 } else if (ret == 0) {
7397 * If the inode was not logged and we are doing a rename (old_dir is not
7398 * NULL), check if old_dir was logged - if it was not we can return and
7401 ret = inode_logged(trans, old_dir, NULL);
7410 * If we are doing a rename (old_dir is not NULL) from a directory that
7411 * was previously logged, make sure that on log replay we get the old
7412 * dir entry deleted. This is needed because we will also log the new
7413 * name of the renamed inode, so we need to make sure that after log
7414 * replay we don't end up with both the new and old dir entries existing.
7416 if (old_dir && old_dir->logged_trans == trans->transid) {
7417 struct btrfs_root *log = old_dir->root->log_root;
7418 struct btrfs_path *path;
7420 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7423 * We have two inodes to update in the log, the old directory and
7424 * the inode that got renamed, so we must pin the log to prevent
7425 * anyone from syncing the log until we have updated both inodes
7428 ret = join_running_log_trans(root);
7430 * At least one of the inodes was logged before, so this should
7431 * not fail, but if it does, it's not serious, just bail out and
7432 * mark the log for a full commit.
7434 if (WARN_ON_ONCE(ret < 0))
7438 path = btrfs_alloc_path();
7445 * Other concurrent task might be logging the old directory,
7446 * as it can be triggered when logging other inode that had or
7447 * still has a dentry in the old directory. We lock the old
7448 * directory's log_mutex to ensure the deletion of the old
7449 * name is persisted, because during directory logging we
7450 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7451 * the old name's dir index item is in the delayed items, so
7452 * it could be missed by an in progress directory logging.
7454 mutex_lock(&old_dir->log_mutex);
7455 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7456 old_dentry->d_name.name,
7457 old_dentry->d_name.len, old_dir_index);
7460 * The dentry does not exist in the log, so record its
7463 btrfs_release_path(path);
7464 ret = insert_dir_log_key(trans, log, path,
7466 old_dir_index, old_dir_index);
7468 mutex_unlock(&old_dir->log_mutex);
7470 btrfs_free_path(path);
7475 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7476 ctx.logging_new_name = true;
7478 * We don't care about the return value. If we fail to log the new name
7479 * then we know the next attempt to sync the log will fallback to a full
7480 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7481 * we don't need to worry about getting a log committed that has an
7482 * inconsistent state after a rename operation.
7484 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7485 ASSERT(list_empty(&ctx.conflict_inodes));
7488 * If an error happened mark the log for a full commit because it's not
7489 * consistent and up to date or we couldn't find out if one of the
7490 * inodes was logged before in this transaction. Do it before unpinning
7491 * the log, to avoid any races with someone else trying to commit it.
7494 btrfs_set_log_full_commit(trans);
7496 btrfs_end_log_trans(root);