1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
16 #include "print-tree.h"
18 #include "compression.h"
20 #include "block-group.h"
21 #include "space-info.h"
23 #include "inode-item.h"
25 #include "accessors.h"
26 #include "extent-tree.h"
27 #include "root-tree.h"
29 #include "file-item.h"
33 #define MAX_CONFLICT_INODES 10
35 /* magic values for the inode_only field in btrfs_log_inode:
37 * LOG_INODE_ALL means to log everything
38 * LOG_INODE_EXISTS means to log just enough to recreate the inode
47 * directory trouble cases
49 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
50 * log, we must force a full commit before doing an fsync of the directory
51 * where the unlink was done.
52 * ---> record transid of last unlink/rename per directory
56 * rename foo/some_dir foo2/some_dir
58 * fsync foo/some_dir/some_file
60 * The fsync above will unlink the original some_dir without recording
61 * it in its new location (foo2). After a crash, some_dir will be gone
62 * unless the fsync of some_file forces a full commit
64 * 2) we must log any new names for any file or dir that is in the fsync
65 * log. ---> check inode while renaming/linking.
67 * 2a) we must log any new names for any file or dir during rename
68 * when the directory they are being removed from was logged.
69 * ---> check inode and old parent dir during rename
71 * 2a is actually the more important variant. With the extra logging
72 * a crash might unlink the old name without recreating the new one
74 * 3) after a crash, we must go through any directories with a link count
75 * of zero and redo the rm -rf
82 * The directory f1 was fully removed from the FS, but fsync was never
83 * called on f1, only its parent dir. After a crash the rm -rf must
84 * be replayed. This must be able to recurse down the entire
85 * directory tree. The inode link count fixup code takes care of the
90 * stages for the tree walking. The first
91 * stage (0) is to only pin down the blocks we find
92 * the second stage (1) is to make sure that all the inodes
93 * we find in the log are created in the subvolume.
95 * The last stage is to deal with directories and links and extents
96 * and all the other fun semantics
100 LOG_WALK_REPLAY_INODES,
101 LOG_WALK_REPLAY_DIR_INDEX,
105 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
106 struct btrfs_inode *inode,
108 struct btrfs_log_ctx *ctx);
109 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
110 struct btrfs_root *root,
111 struct btrfs_path *path, u64 objectid);
112 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
113 struct btrfs_root *root,
114 struct btrfs_root *log,
115 struct btrfs_path *path,
116 u64 dirid, int del_all);
117 static void wait_log_commit(struct btrfs_root *root, int transid);
120 * tree logging is a special write ahead log used to make sure that
121 * fsyncs and O_SYNCs can happen without doing full tree commits.
123 * Full tree commits are expensive because they require commonly
124 * modified blocks to be recowed, creating many dirty pages in the
125 * extent tree an 4x-6x higher write load than ext3.
127 * Instead of doing a tree commit on every fsync, we use the
128 * key ranges and transaction ids to find items for a given file or directory
129 * that have changed in this transaction. Those items are copied into
130 * a special tree (one per subvolume root), that tree is written to disk
131 * and then the fsync is considered complete.
133 * After a crash, items are copied out of the log-tree back into the
134 * subvolume tree. Any file data extents found are recorded in the extent
135 * allocation tree, and the log-tree freed.
137 * The log tree is read three times, once to pin down all the extents it is
138 * using in ram and once, once to create all the inodes logged in the tree
139 * and once to do all the other items.
143 * start a sub transaction and setup the log tree
144 * this increments the log tree writer count to make the people
145 * syncing the tree wait for us to finish
147 static int start_log_trans(struct btrfs_trans_handle *trans,
148 struct btrfs_root *root,
149 struct btrfs_log_ctx *ctx)
151 struct btrfs_fs_info *fs_info = root->fs_info;
152 struct btrfs_root *tree_root = fs_info->tree_root;
153 const bool zoned = btrfs_is_zoned(fs_info);
155 bool created = false;
158 * First check if the log root tree was already created. If not, create
159 * it before locking the root's log_mutex, just to keep lockdep happy.
161 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
162 mutex_lock(&tree_root->log_mutex);
163 if (!fs_info->log_root_tree) {
164 ret = btrfs_init_log_root_tree(trans, fs_info);
166 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
170 mutex_unlock(&tree_root->log_mutex);
175 mutex_lock(&root->log_mutex);
178 if (root->log_root) {
179 int index = (root->log_transid + 1) % 2;
181 if (btrfs_need_log_full_commit(trans)) {
182 ret = BTRFS_LOG_FORCE_COMMIT;
186 if (zoned && atomic_read(&root->log_commit[index])) {
187 wait_log_commit(root, root->log_transid - 1);
191 if (!root->log_start_pid) {
192 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
193 root->log_start_pid = current->pid;
194 } else if (root->log_start_pid != current->pid) {
195 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
199 * This means fs_info->log_root_tree was already created
200 * for some other FS trees. Do the full commit not to mix
201 * nodes from multiple log transactions to do sequential
204 if (zoned && !created) {
205 ret = BTRFS_LOG_FORCE_COMMIT;
209 ret = btrfs_add_log_tree(trans, root);
213 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
214 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
215 root->log_start_pid = current->pid;
218 atomic_inc(&root->log_writers);
219 if (!ctx->logging_new_name) {
220 int index = root->log_transid % 2;
221 list_add_tail(&ctx->list, &root->log_ctxs[index]);
222 ctx->log_transid = root->log_transid;
226 mutex_unlock(&root->log_mutex);
231 * returns 0 if there was a log transaction running and we were able
232 * to join, or returns -ENOENT if there were not transactions
235 static int join_running_log_trans(struct btrfs_root *root)
237 const bool zoned = btrfs_is_zoned(root->fs_info);
240 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
243 mutex_lock(&root->log_mutex);
245 if (root->log_root) {
246 int index = (root->log_transid + 1) % 2;
249 if (zoned && atomic_read(&root->log_commit[index])) {
250 wait_log_commit(root, root->log_transid - 1);
253 atomic_inc(&root->log_writers);
255 mutex_unlock(&root->log_mutex);
260 * This either makes the current running log transaction wait
261 * until you call btrfs_end_log_trans() or it makes any future
262 * log transactions wait until you call btrfs_end_log_trans()
264 void btrfs_pin_log_trans(struct btrfs_root *root)
266 atomic_inc(&root->log_writers);
270 * indicate we're done making changes to the log tree
271 * and wake up anyone waiting to do a sync
273 void btrfs_end_log_trans(struct btrfs_root *root)
275 if (atomic_dec_and_test(&root->log_writers)) {
276 /* atomic_dec_and_test implies a barrier */
277 cond_wake_up_nomb(&root->log_writer_wait);
281 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
283 filemap_fdatawait_range(buf->pages[0]->mapping,
284 buf->start, buf->start + buf->len - 1);
288 * the walk control struct is used to pass state down the chain when
289 * processing the log tree. The stage field tells us which part
290 * of the log tree processing we are currently doing. The others
291 * are state fields used for that specific part
293 struct walk_control {
294 /* should we free the extent on disk when done? This is used
295 * at transaction commit time while freeing a log tree
299 /* pin only walk, we record which extents on disk belong to the
304 /* what stage of the replay code we're currently in */
308 * Ignore any items from the inode currently being processed. Needs
309 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
310 * the LOG_WALK_REPLAY_INODES stage.
312 bool ignore_cur_inode;
314 /* the root we are currently replaying */
315 struct btrfs_root *replay_dest;
317 /* the trans handle for the current replay */
318 struct btrfs_trans_handle *trans;
320 /* the function that gets used to process blocks we find in the
321 * tree. Note the extent_buffer might not be up to date when it is
322 * passed in, and it must be checked or read if you need the data
325 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
326 struct walk_control *wc, u64 gen, int level);
330 * process_func used to pin down extents, write them or wait on them
332 static int process_one_buffer(struct btrfs_root *log,
333 struct extent_buffer *eb,
334 struct walk_control *wc, u64 gen, int level)
336 struct btrfs_fs_info *fs_info = log->fs_info;
340 * If this fs is mixed then we need to be able to process the leaves to
341 * pin down any logged extents, so we have to read the block.
343 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
344 struct btrfs_tree_parent_check check = {
349 ret = btrfs_read_extent_buffer(eb, &check);
355 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
360 if (btrfs_buffer_uptodate(eb, gen, 0) &&
361 btrfs_header_level(eb) == 0)
362 ret = btrfs_exclude_logged_extents(eb);
367 static int do_overwrite_item(struct btrfs_trans_handle *trans,
368 struct btrfs_root *root,
369 struct btrfs_path *path,
370 struct extent_buffer *eb, int slot,
371 struct btrfs_key *key)
375 u64 saved_i_size = 0;
376 int save_old_i_size = 0;
377 unsigned long src_ptr;
378 unsigned long dst_ptr;
379 int overwrite_root = 0;
380 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
382 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
385 item_size = btrfs_item_size(eb, slot);
386 src_ptr = btrfs_item_ptr_offset(eb, slot);
388 /* Our caller must have done a search for the key for us. */
389 ASSERT(path->nodes[0] != NULL);
392 * And the slot must point to the exact key or the slot where the key
393 * should be at (the first item with a key greater than 'key')
395 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
396 struct btrfs_key found_key;
398 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
399 ret = btrfs_comp_cpu_keys(&found_key, key);
408 u32 dst_size = btrfs_item_size(path->nodes[0],
410 if (dst_size != item_size)
413 if (item_size == 0) {
414 btrfs_release_path(path);
417 dst_copy = kmalloc(item_size, GFP_NOFS);
418 src_copy = kmalloc(item_size, GFP_NOFS);
419 if (!dst_copy || !src_copy) {
420 btrfs_release_path(path);
426 read_extent_buffer(eb, src_copy, src_ptr, item_size);
428 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
429 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
431 ret = memcmp(dst_copy, src_copy, item_size);
436 * they have the same contents, just return, this saves
437 * us from cowing blocks in the destination tree and doing
438 * extra writes that may not have been done by a previous
442 btrfs_release_path(path);
447 * We need to load the old nbytes into the inode so when we
448 * replay the extents we've logged we get the right nbytes.
451 struct btrfs_inode_item *item;
455 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
456 struct btrfs_inode_item);
457 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
458 item = btrfs_item_ptr(eb, slot,
459 struct btrfs_inode_item);
460 btrfs_set_inode_nbytes(eb, item, nbytes);
463 * If this is a directory we need to reset the i_size to
464 * 0 so that we can set it up properly when replaying
465 * the rest of the items in this log.
467 mode = btrfs_inode_mode(eb, item);
469 btrfs_set_inode_size(eb, item, 0);
471 } else if (inode_item) {
472 struct btrfs_inode_item *item;
476 * New inode, set nbytes to 0 so that the nbytes comes out
477 * properly when we replay the extents.
479 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
480 btrfs_set_inode_nbytes(eb, item, 0);
483 * If this is a directory we need to reset the i_size to 0 so
484 * that we can set it up properly when replaying the rest of
485 * the items in this log.
487 mode = btrfs_inode_mode(eb, item);
489 btrfs_set_inode_size(eb, item, 0);
492 btrfs_release_path(path);
493 /* try to insert the key into the destination tree */
494 path->skip_release_on_error = 1;
495 ret = btrfs_insert_empty_item(trans, root, path,
497 path->skip_release_on_error = 0;
499 /* make sure any existing item is the correct size */
500 if (ret == -EEXIST || ret == -EOVERFLOW) {
502 found_size = btrfs_item_size(path->nodes[0],
504 if (found_size > item_size)
505 btrfs_truncate_item(path, item_size, 1);
506 else if (found_size < item_size)
507 btrfs_extend_item(path, item_size - found_size);
511 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
514 /* don't overwrite an existing inode if the generation number
515 * was logged as zero. This is done when the tree logging code
516 * is just logging an inode to make sure it exists after recovery.
518 * Also, don't overwrite i_size on directories during replay.
519 * log replay inserts and removes directory items based on the
520 * state of the tree found in the subvolume, and i_size is modified
523 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
524 struct btrfs_inode_item *src_item;
525 struct btrfs_inode_item *dst_item;
527 src_item = (struct btrfs_inode_item *)src_ptr;
528 dst_item = (struct btrfs_inode_item *)dst_ptr;
530 if (btrfs_inode_generation(eb, src_item) == 0) {
531 struct extent_buffer *dst_eb = path->nodes[0];
532 const u64 ino_size = btrfs_inode_size(eb, src_item);
535 * For regular files an ino_size == 0 is used only when
536 * logging that an inode exists, as part of a directory
537 * fsync, and the inode wasn't fsynced before. In this
538 * case don't set the size of the inode in the fs/subvol
539 * tree, otherwise we would be throwing valid data away.
541 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
542 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
544 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
548 if (overwrite_root &&
549 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
550 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
552 saved_i_size = btrfs_inode_size(path->nodes[0],
557 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
560 if (save_old_i_size) {
561 struct btrfs_inode_item *dst_item;
562 dst_item = (struct btrfs_inode_item *)dst_ptr;
563 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
566 /* make sure the generation is filled in */
567 if (key->type == BTRFS_INODE_ITEM_KEY) {
568 struct btrfs_inode_item *dst_item;
569 dst_item = (struct btrfs_inode_item *)dst_ptr;
570 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
571 btrfs_set_inode_generation(path->nodes[0], dst_item,
576 btrfs_mark_buffer_dirty(path->nodes[0]);
577 btrfs_release_path(path);
582 * Item overwrite used by replay and tree logging. eb, slot and key all refer
583 * to the src data we are copying out.
585 * root is the tree we are copying into, and path is a scratch
586 * path for use in this function (it should be released on entry and
587 * will be released on exit).
589 * If the key is already in the destination tree the existing item is
590 * overwritten. If the existing item isn't big enough, it is extended.
591 * If it is too large, it is truncated.
593 * If the key isn't in the destination yet, a new item is inserted.
595 static int overwrite_item(struct btrfs_trans_handle *trans,
596 struct btrfs_root *root,
597 struct btrfs_path *path,
598 struct extent_buffer *eb, int slot,
599 struct btrfs_key *key)
603 /* Look for the key in the destination tree. */
604 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
608 return do_overwrite_item(trans, root, path, eb, slot, key);
611 static int read_alloc_one_name(struct extent_buffer *eb, void *start, int len,
612 struct fscrypt_str *name)
616 buf = kmalloc(len, GFP_NOFS);
620 read_extent_buffer(eb, buf, (unsigned long)start, len);
627 * simple helper to read an inode off the disk from a given root
628 * This can only be called for subvolume roots and not for the log
630 static noinline struct inode *read_one_inode(struct btrfs_root *root,
635 inode = btrfs_iget(root->fs_info->sb, objectid, root);
641 /* replays a single extent in 'eb' at 'slot' with 'key' into the
642 * subvolume 'root'. path is released on entry and should be released
645 * extents in the log tree have not been allocated out of the extent
646 * tree yet. So, this completes the allocation, taking a reference
647 * as required if the extent already exists or creating a new extent
648 * if it isn't in the extent allocation tree yet.
650 * The extent is inserted into the file, dropping any existing extents
651 * from the file that overlap the new one.
653 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
654 struct btrfs_root *root,
655 struct btrfs_path *path,
656 struct extent_buffer *eb, int slot,
657 struct btrfs_key *key)
659 struct btrfs_drop_extents_args drop_args = { 0 };
660 struct btrfs_fs_info *fs_info = root->fs_info;
663 u64 start = key->offset;
665 struct btrfs_file_extent_item *item;
666 struct inode *inode = NULL;
670 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
671 found_type = btrfs_file_extent_type(eb, item);
673 if (found_type == BTRFS_FILE_EXTENT_REG ||
674 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
675 nbytes = btrfs_file_extent_num_bytes(eb, item);
676 extent_end = start + nbytes;
679 * We don't add to the inodes nbytes if we are prealloc or a
682 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
684 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
685 size = btrfs_file_extent_ram_bytes(eb, item);
686 nbytes = btrfs_file_extent_ram_bytes(eb, item);
687 extent_end = ALIGN(start + size,
688 fs_info->sectorsize);
694 inode = read_one_inode(root, key->objectid);
701 * first check to see if we already have this extent in the
702 * file. This must be done before the btrfs_drop_extents run
703 * so we don't try to drop this extent.
705 ret = btrfs_lookup_file_extent(trans, root, path,
706 btrfs_ino(BTRFS_I(inode)), start, 0);
709 (found_type == BTRFS_FILE_EXTENT_REG ||
710 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
711 struct btrfs_file_extent_item cmp1;
712 struct btrfs_file_extent_item cmp2;
713 struct btrfs_file_extent_item *existing;
714 struct extent_buffer *leaf;
716 leaf = path->nodes[0];
717 existing = btrfs_item_ptr(leaf, path->slots[0],
718 struct btrfs_file_extent_item);
720 read_extent_buffer(eb, &cmp1, (unsigned long)item,
722 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
726 * we already have a pointer to this exact extent,
727 * we don't have to do anything
729 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
730 btrfs_release_path(path);
734 btrfs_release_path(path);
736 /* drop any overlapping extents */
737 drop_args.start = start;
738 drop_args.end = extent_end;
739 drop_args.drop_cache = true;
740 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
744 if (found_type == BTRFS_FILE_EXTENT_REG ||
745 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
747 unsigned long dest_offset;
748 struct btrfs_key ins;
750 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
751 btrfs_fs_incompat(fs_info, NO_HOLES))
754 ret = btrfs_insert_empty_item(trans, root, path, key,
758 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
760 copy_extent_buffer(path->nodes[0], eb, dest_offset,
761 (unsigned long)item, sizeof(*item));
763 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
764 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
765 ins.type = BTRFS_EXTENT_ITEM_KEY;
766 offset = key->offset - btrfs_file_extent_offset(eb, item);
769 * Manually record dirty extent, as here we did a shallow
770 * file extent item copy and skip normal backref update,
771 * but modifying extent tree all by ourselves.
772 * So need to manually record dirty extent for qgroup,
773 * as the owner of the file extent changed from log tree
774 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
776 ret = btrfs_qgroup_trace_extent(trans,
777 btrfs_file_extent_disk_bytenr(eb, item),
778 btrfs_file_extent_disk_num_bytes(eb, item));
782 if (ins.objectid > 0) {
783 struct btrfs_ref ref = { 0 };
786 LIST_HEAD(ordered_sums);
789 * is this extent already allocated in the extent
790 * allocation tree? If so, just add a reference
792 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
796 } else if (ret == 0) {
797 btrfs_init_generic_ref(&ref,
798 BTRFS_ADD_DELAYED_REF,
799 ins.objectid, ins.offset, 0);
800 btrfs_init_data_ref(&ref,
801 root->root_key.objectid,
802 key->objectid, offset, 0, false);
803 ret = btrfs_inc_extent_ref(trans, &ref);
808 * insert the extent pointer in the extent
811 ret = btrfs_alloc_logged_file_extent(trans,
812 root->root_key.objectid,
813 key->objectid, offset, &ins);
817 btrfs_release_path(path);
819 if (btrfs_file_extent_compression(eb, item)) {
820 csum_start = ins.objectid;
821 csum_end = csum_start + ins.offset;
823 csum_start = ins.objectid +
824 btrfs_file_extent_offset(eb, item);
825 csum_end = csum_start +
826 btrfs_file_extent_num_bytes(eb, item);
829 ret = btrfs_lookup_csums_list(root->log_root,
830 csum_start, csum_end - 1,
831 &ordered_sums, 0, false);
835 * Now delete all existing cums in the csum root that
836 * cover our range. We do this because we can have an
837 * extent that is completely referenced by one file
838 * extent item and partially referenced by another
839 * file extent item (like after using the clone or
840 * extent_same ioctls). In this case if we end up doing
841 * the replay of the one that partially references the
842 * extent first, and we do not do the csum deletion
843 * below, we can get 2 csum items in the csum tree that
844 * overlap each other. For example, imagine our log has
845 * the two following file extent items:
847 * key (257 EXTENT_DATA 409600)
848 * extent data disk byte 12845056 nr 102400
849 * extent data offset 20480 nr 20480 ram 102400
851 * key (257 EXTENT_DATA 819200)
852 * extent data disk byte 12845056 nr 102400
853 * extent data offset 0 nr 102400 ram 102400
855 * Where the second one fully references the 100K extent
856 * that starts at disk byte 12845056, and the log tree
857 * has a single csum item that covers the entire range
860 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
862 * After the first file extent item is replayed, the
863 * csum tree gets the following csum item:
865 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
867 * Which covers the 20K sub-range starting at offset 20K
868 * of our extent. Now when we replay the second file
869 * extent item, if we do not delete existing csum items
870 * that cover any of its blocks, we end up getting two
871 * csum items in our csum tree that overlap each other:
873 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
874 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
876 * Which is a problem, because after this anyone trying
877 * to lookup up for the checksum of any block of our
878 * extent starting at an offset of 40K or higher, will
879 * end up looking at the second csum item only, which
880 * does not contain the checksum for any block starting
881 * at offset 40K or higher of our extent.
883 while (!list_empty(&ordered_sums)) {
884 struct btrfs_ordered_sum *sums;
885 struct btrfs_root *csum_root;
887 sums = list_entry(ordered_sums.next,
888 struct btrfs_ordered_sum,
890 csum_root = btrfs_csum_root(fs_info,
893 ret = btrfs_del_csums(trans, csum_root,
897 ret = btrfs_csum_file_blocks(trans,
900 list_del(&sums->list);
906 btrfs_release_path(path);
908 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
909 /* inline extents are easy, we just overwrite them */
910 ret = overwrite_item(trans, root, path, eb, slot, key);
915 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
921 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
922 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
928 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
929 struct btrfs_inode *dir,
930 struct btrfs_inode *inode,
931 const struct fscrypt_str *name)
935 ret = btrfs_unlink_inode(trans, dir, inode, name);
939 * Whenever we need to check if a name exists or not, we check the
940 * fs/subvolume tree. So after an unlink we must run delayed items, so
941 * that future checks for a name during log replay see that the name
942 * does not exists anymore.
944 return btrfs_run_delayed_items(trans);
948 * when cleaning up conflicts between the directory names in the
949 * subvolume, directory names in the log and directory names in the
950 * inode back references, we may have to unlink inodes from directories.
952 * This is a helper function to do the unlink of a specific directory
955 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
956 struct btrfs_path *path,
957 struct btrfs_inode *dir,
958 struct btrfs_dir_item *di)
960 struct btrfs_root *root = dir->root;
962 struct fscrypt_str name;
963 struct extent_buffer *leaf;
964 struct btrfs_key location;
967 leaf = path->nodes[0];
969 btrfs_dir_item_key_to_cpu(leaf, di, &location);
970 ret = read_alloc_one_name(leaf, di + 1, btrfs_dir_name_len(leaf, di), &name);
974 btrfs_release_path(path);
976 inode = read_one_inode(root, location.objectid);
982 ret = link_to_fixup_dir(trans, root, path, location.objectid);
986 ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), &name);
994 * See if a given name and sequence number found in an inode back reference are
995 * already in a directory and correctly point to this inode.
997 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
1000 static noinline int inode_in_dir(struct btrfs_root *root,
1001 struct btrfs_path *path,
1002 u64 dirid, u64 objectid, u64 index,
1003 struct fscrypt_str *name)
1005 struct btrfs_dir_item *di;
1006 struct btrfs_key location;
1009 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
1015 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1016 if (location.objectid != objectid)
1022 btrfs_release_path(path);
1023 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, 0);
1028 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1029 if (location.objectid == objectid)
1033 btrfs_release_path(path);
1038 * helper function to check a log tree for a named back reference in
1039 * an inode. This is used to decide if a back reference that is
1040 * found in the subvolume conflicts with what we find in the log.
1042 * inode backreferences may have multiple refs in a single item,
1043 * during replay we process one reference at a time, and we don't
1044 * want to delete valid links to a file from the subvolume if that
1045 * link is also in the log.
1047 static noinline int backref_in_log(struct btrfs_root *log,
1048 struct btrfs_key *key,
1050 const struct fscrypt_str *name)
1052 struct btrfs_path *path;
1055 path = btrfs_alloc_path();
1059 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1062 } else if (ret == 1) {
1067 if (key->type == BTRFS_INODE_EXTREF_KEY)
1068 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1070 ref_objectid, name);
1072 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1073 path->slots[0], name);
1075 btrfs_free_path(path);
1079 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1080 struct btrfs_root *root,
1081 struct btrfs_path *path,
1082 struct btrfs_root *log_root,
1083 struct btrfs_inode *dir,
1084 struct btrfs_inode *inode,
1085 u64 inode_objectid, u64 parent_objectid,
1086 u64 ref_index, struct fscrypt_str *name)
1089 struct extent_buffer *leaf;
1090 struct btrfs_dir_item *di;
1091 struct btrfs_key search_key;
1092 struct btrfs_inode_extref *extref;
1095 /* Search old style refs */
1096 search_key.objectid = inode_objectid;
1097 search_key.type = BTRFS_INODE_REF_KEY;
1098 search_key.offset = parent_objectid;
1099 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1101 struct btrfs_inode_ref *victim_ref;
1103 unsigned long ptr_end;
1105 leaf = path->nodes[0];
1107 /* are we trying to overwrite a back ref for the root directory
1108 * if so, just jump out, we're done
1110 if (search_key.objectid == search_key.offset)
1113 /* check all the names in this back reference to see
1114 * if they are in the log. if so, we allow them to stay
1115 * otherwise they must be unlinked as a conflict
1117 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1118 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1119 while (ptr < ptr_end) {
1120 struct fscrypt_str victim_name;
1122 victim_ref = (struct btrfs_inode_ref *)ptr;
1123 ret = read_alloc_one_name(leaf, (victim_ref + 1),
1124 btrfs_inode_ref_name_len(leaf, victim_ref),
1129 ret = backref_in_log(log_root, &search_key,
1130 parent_objectid, &victim_name);
1132 kfree(victim_name.name);
1135 inc_nlink(&inode->vfs_inode);
1136 btrfs_release_path(path);
1138 ret = unlink_inode_for_log_replay(trans, dir, inode,
1140 kfree(victim_name.name);
1145 kfree(victim_name.name);
1147 ptr = (unsigned long)(victim_ref + 1) + victim_name.len;
1150 btrfs_release_path(path);
1152 /* Same search but for extended refs */
1153 extref = btrfs_lookup_inode_extref(NULL, root, path, name,
1154 inode_objectid, parent_objectid, 0,
1156 if (IS_ERR(extref)) {
1157 return PTR_ERR(extref);
1158 } else if (extref) {
1162 struct inode *victim_parent;
1164 leaf = path->nodes[0];
1166 item_size = btrfs_item_size(leaf, path->slots[0]);
1167 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1169 while (cur_offset < item_size) {
1170 struct fscrypt_str victim_name;
1172 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1174 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1177 ret = read_alloc_one_name(leaf, &extref->name,
1178 btrfs_inode_extref_name_len(leaf, extref),
1183 search_key.objectid = inode_objectid;
1184 search_key.type = BTRFS_INODE_EXTREF_KEY;
1185 search_key.offset = btrfs_extref_hash(parent_objectid,
1188 ret = backref_in_log(log_root, &search_key,
1189 parent_objectid, &victim_name);
1191 kfree(victim_name.name);
1195 victim_parent = read_one_inode(root,
1197 if (victim_parent) {
1198 inc_nlink(&inode->vfs_inode);
1199 btrfs_release_path(path);
1201 ret = unlink_inode_for_log_replay(trans,
1202 BTRFS_I(victim_parent),
1203 inode, &victim_name);
1205 iput(victim_parent);
1206 kfree(victim_name.name);
1211 kfree(victim_name.name);
1213 cur_offset += victim_name.len + sizeof(*extref);
1216 btrfs_release_path(path);
1218 /* look for a conflicting sequence number */
1219 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1220 ref_index, name, 0);
1224 ret = drop_one_dir_item(trans, path, dir, di);
1228 btrfs_release_path(path);
1230 /* look for a conflicting name */
1231 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), name, 0);
1235 ret = drop_one_dir_item(trans, path, dir, di);
1239 btrfs_release_path(path);
1244 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1245 struct fscrypt_str *name, u64 *index,
1246 u64 *parent_objectid)
1248 struct btrfs_inode_extref *extref;
1251 extref = (struct btrfs_inode_extref *)ref_ptr;
1253 ret = read_alloc_one_name(eb, &extref->name,
1254 btrfs_inode_extref_name_len(eb, extref), name);
1259 *index = btrfs_inode_extref_index(eb, extref);
1260 if (parent_objectid)
1261 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1266 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1267 struct fscrypt_str *name, u64 *index)
1269 struct btrfs_inode_ref *ref;
1272 ref = (struct btrfs_inode_ref *)ref_ptr;
1274 ret = read_alloc_one_name(eb, ref + 1, btrfs_inode_ref_name_len(eb, ref),
1280 *index = btrfs_inode_ref_index(eb, ref);
1286 * Take an inode reference item from the log tree and iterate all names from the
1287 * inode reference item in the subvolume tree with the same key (if it exists).
1288 * For any name that is not in the inode reference item from the log tree, do a
1289 * proper unlink of that name (that is, remove its entry from the inode
1290 * reference item and both dir index keys).
1292 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1293 struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 struct btrfs_inode *inode,
1296 struct extent_buffer *log_eb,
1298 struct btrfs_key *key)
1301 unsigned long ref_ptr;
1302 unsigned long ref_end;
1303 struct extent_buffer *eb;
1306 btrfs_release_path(path);
1307 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1315 eb = path->nodes[0];
1316 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1317 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1318 while (ref_ptr < ref_end) {
1319 struct fscrypt_str name;
1322 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1323 ret = extref_get_fields(eb, ref_ptr, &name,
1326 parent_id = key->offset;
1327 ret = ref_get_fields(eb, ref_ptr, &name, NULL);
1332 if (key->type == BTRFS_INODE_EXTREF_KEY)
1333 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1336 ret = !!btrfs_find_name_in_backref(log_eb, log_slot, &name);
1341 btrfs_release_path(path);
1342 dir = read_one_inode(root, parent_id);
1348 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1358 ref_ptr += name.len;
1359 if (key->type == BTRFS_INODE_EXTREF_KEY)
1360 ref_ptr += sizeof(struct btrfs_inode_extref);
1362 ref_ptr += sizeof(struct btrfs_inode_ref);
1366 btrfs_release_path(path);
1371 * replay one inode back reference item found in the log tree.
1372 * eb, slot and key refer to the buffer and key found in the log tree.
1373 * root is the destination we are replaying into, and path is for temp
1374 * use by this function. (it should be released on return).
1376 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1377 struct btrfs_root *root,
1378 struct btrfs_root *log,
1379 struct btrfs_path *path,
1380 struct extent_buffer *eb, int slot,
1381 struct btrfs_key *key)
1383 struct inode *dir = NULL;
1384 struct inode *inode = NULL;
1385 unsigned long ref_ptr;
1386 unsigned long ref_end;
1387 struct fscrypt_str name;
1389 int log_ref_ver = 0;
1390 u64 parent_objectid;
1393 int ref_struct_size;
1395 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1396 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1398 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1399 struct btrfs_inode_extref *r;
1401 ref_struct_size = sizeof(struct btrfs_inode_extref);
1403 r = (struct btrfs_inode_extref *)ref_ptr;
1404 parent_objectid = btrfs_inode_extref_parent(eb, r);
1406 ref_struct_size = sizeof(struct btrfs_inode_ref);
1407 parent_objectid = key->offset;
1409 inode_objectid = key->objectid;
1412 * it is possible that we didn't log all the parent directories
1413 * for a given inode. If we don't find the dir, just don't
1414 * copy the back ref in. The link count fixup code will take
1417 dir = read_one_inode(root, parent_objectid);
1423 inode = read_one_inode(root, inode_objectid);
1429 while (ref_ptr < ref_end) {
1431 ret = extref_get_fields(eb, ref_ptr, &name,
1432 &ref_index, &parent_objectid);
1434 * parent object can change from one array
1438 dir = read_one_inode(root, parent_objectid);
1444 ret = ref_get_fields(eb, ref_ptr, &name, &ref_index);
1449 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1450 btrfs_ino(BTRFS_I(inode)), ref_index, &name);
1453 } else if (ret == 0) {
1455 * look for a conflicting back reference in the
1456 * metadata. if we find one we have to unlink that name
1457 * of the file before we add our new link. Later on, we
1458 * overwrite any existing back reference, and we don't
1459 * want to create dangling pointers in the directory.
1461 ret = __add_inode_ref(trans, root, path, log,
1462 BTRFS_I(dir), BTRFS_I(inode),
1463 inode_objectid, parent_objectid,
1471 /* insert our name */
1472 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1473 &name, 0, ref_index);
1477 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1481 /* Else, ret == 1, we already have a perfect match, we're done. */
1483 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + name.len;
1493 * Before we overwrite the inode reference item in the subvolume tree
1494 * with the item from the log tree, we must unlink all names from the
1495 * parent directory that are in the subvolume's tree inode reference
1496 * item, otherwise we end up with an inconsistent subvolume tree where
1497 * dir index entries exist for a name but there is no inode reference
1498 * item with the same name.
1500 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1505 /* finally write the back reference in the inode */
1506 ret = overwrite_item(trans, root, path, eb, slot, key);
1508 btrfs_release_path(path);
1515 static int count_inode_extrefs(struct btrfs_root *root,
1516 struct btrfs_inode *inode, struct btrfs_path *path)
1520 unsigned int nlink = 0;
1523 u64 inode_objectid = btrfs_ino(inode);
1526 struct btrfs_inode_extref *extref;
1527 struct extent_buffer *leaf;
1530 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1535 leaf = path->nodes[0];
1536 item_size = btrfs_item_size(leaf, path->slots[0]);
1537 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1540 while (cur_offset < item_size) {
1541 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1542 name_len = btrfs_inode_extref_name_len(leaf, extref);
1546 cur_offset += name_len + sizeof(*extref);
1550 btrfs_release_path(path);
1552 btrfs_release_path(path);
1554 if (ret < 0 && ret != -ENOENT)
1559 static int count_inode_refs(struct btrfs_root *root,
1560 struct btrfs_inode *inode, struct btrfs_path *path)
1563 struct btrfs_key key;
1564 unsigned int nlink = 0;
1566 unsigned long ptr_end;
1568 u64 ino = btrfs_ino(inode);
1571 key.type = BTRFS_INODE_REF_KEY;
1572 key.offset = (u64)-1;
1575 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1579 if (path->slots[0] == 0)
1584 btrfs_item_key_to_cpu(path->nodes[0], &key,
1586 if (key.objectid != ino ||
1587 key.type != BTRFS_INODE_REF_KEY)
1589 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1590 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1592 while (ptr < ptr_end) {
1593 struct btrfs_inode_ref *ref;
1595 ref = (struct btrfs_inode_ref *)ptr;
1596 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1598 ptr = (unsigned long)(ref + 1) + name_len;
1602 if (key.offset == 0)
1604 if (path->slots[0] > 0) {
1609 btrfs_release_path(path);
1611 btrfs_release_path(path);
1617 * There are a few corners where the link count of the file can't
1618 * be properly maintained during replay. So, instead of adding
1619 * lots of complexity to the log code, we just scan the backrefs
1620 * for any file that has been through replay.
1622 * The scan will update the link count on the inode to reflect the
1623 * number of back refs found. If it goes down to zero, the iput
1624 * will free the inode.
1626 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1627 struct btrfs_root *root,
1628 struct inode *inode)
1630 struct btrfs_path *path;
1633 u64 ino = btrfs_ino(BTRFS_I(inode));
1635 path = btrfs_alloc_path();
1639 ret = count_inode_refs(root, BTRFS_I(inode), path);
1645 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1653 if (nlink != inode->i_nlink) {
1654 set_nlink(inode, nlink);
1655 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1659 BTRFS_I(inode)->index_cnt = (u64)-1;
1661 if (inode->i_nlink == 0) {
1662 if (S_ISDIR(inode->i_mode)) {
1663 ret = replay_dir_deletes(trans, root, NULL, path,
1668 ret = btrfs_insert_orphan_item(trans, root, ino);
1674 btrfs_free_path(path);
1678 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1679 struct btrfs_root *root,
1680 struct btrfs_path *path)
1683 struct btrfs_key key;
1684 struct inode *inode;
1686 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1687 key.type = BTRFS_ORPHAN_ITEM_KEY;
1688 key.offset = (u64)-1;
1690 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1696 if (path->slots[0] == 0)
1701 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1702 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1703 key.type != BTRFS_ORPHAN_ITEM_KEY)
1706 ret = btrfs_del_item(trans, root, path);
1710 btrfs_release_path(path);
1711 inode = read_one_inode(root, key.offset);
1717 ret = fixup_inode_link_count(trans, root, inode);
1723 * fixup on a directory may create new entries,
1724 * make sure we always look for the highset possible
1727 key.offset = (u64)-1;
1729 btrfs_release_path(path);
1735 * record a given inode in the fixup dir so we can check its link
1736 * count when replay is done. The link count is incremented here
1737 * so the inode won't go away until we check it
1739 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1740 struct btrfs_root *root,
1741 struct btrfs_path *path,
1744 struct btrfs_key key;
1746 struct inode *inode;
1748 inode = read_one_inode(root, objectid);
1752 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1753 key.type = BTRFS_ORPHAN_ITEM_KEY;
1754 key.offset = objectid;
1756 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1758 btrfs_release_path(path);
1760 if (!inode->i_nlink)
1761 set_nlink(inode, 1);
1764 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1765 } else if (ret == -EEXIST) {
1774 * when replaying the log for a directory, we only insert names
1775 * for inodes that actually exist. This means an fsync on a directory
1776 * does not implicitly fsync all the new files in it
1778 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1779 struct btrfs_root *root,
1780 u64 dirid, u64 index,
1781 const struct fscrypt_str *name,
1782 struct btrfs_key *location)
1784 struct inode *inode;
1788 inode = read_one_inode(root, location->objectid);
1792 dir = read_one_inode(root, dirid);
1798 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1801 /* FIXME, put inode into FIXUP list */
1808 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1809 struct btrfs_inode *dir,
1810 struct btrfs_path *path,
1811 struct btrfs_dir_item *dst_di,
1812 const struct btrfs_key *log_key,
1816 struct btrfs_key found_key;
1818 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1819 /* The existing dentry points to the same inode, don't delete it. */
1820 if (found_key.objectid == log_key->objectid &&
1821 found_key.type == log_key->type &&
1822 found_key.offset == log_key->offset &&
1823 btrfs_dir_flags(path->nodes[0], dst_di) == log_flags)
1827 * Don't drop the conflicting directory entry if the inode for the new
1828 * entry doesn't exist.
1833 return drop_one_dir_item(trans, path, dir, dst_di);
1837 * take a single entry in a log directory item and replay it into
1840 * if a conflicting item exists in the subdirectory already,
1841 * the inode it points to is unlinked and put into the link count
1844 * If a name from the log points to a file or directory that does
1845 * not exist in the FS, it is skipped. fsyncs on directories
1846 * do not force down inodes inside that directory, just changes to the
1847 * names or unlinks in a directory.
1849 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1850 * non-existing inode) and 1 if the name was replayed.
1852 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1853 struct btrfs_root *root,
1854 struct btrfs_path *path,
1855 struct extent_buffer *eb,
1856 struct btrfs_dir_item *di,
1857 struct btrfs_key *key)
1859 struct fscrypt_str name;
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 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
1881 log_flags = btrfs_dir_flags(eb, di);
1882 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1883 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1884 btrfs_release_path(path);
1887 exists = (ret == 0);
1890 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1892 if (IS_ERR(dir_dst_di)) {
1893 ret = PTR_ERR(dir_dst_di);
1895 } else if (dir_dst_di) {
1896 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1897 dir_dst_di, &log_key,
1901 dir_dst_matches = (ret == 1);
1904 btrfs_release_path(path);
1906 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1907 key->objectid, key->offset,
1909 if (IS_ERR(index_dst_di)) {
1910 ret = PTR_ERR(index_dst_di);
1912 } else if (index_dst_di) {
1913 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1914 index_dst_di, &log_key,
1918 index_dst_matches = (ret == 1);
1921 btrfs_release_path(path);
1923 if (dir_dst_matches && index_dst_matches) {
1925 update_size = false;
1930 * Check if the inode reference exists in the log for the given name,
1931 * inode and parent inode
1933 search_key.objectid = log_key.objectid;
1934 search_key.type = BTRFS_INODE_REF_KEY;
1935 search_key.offset = key->objectid;
1936 ret = backref_in_log(root->log_root, &search_key, 0, &name);
1940 /* The dentry will be added later. */
1942 update_size = false;
1946 search_key.objectid = log_key.objectid;
1947 search_key.type = BTRFS_INODE_EXTREF_KEY;
1948 search_key.offset = key->objectid;
1949 ret = backref_in_log(root->log_root, &search_key, key->objectid, &name);
1953 /* The dentry will be added later. */
1955 update_size = false;
1958 btrfs_release_path(path);
1959 ret = insert_one_name(trans, root, key->objectid, key->offset,
1961 if (ret && ret != -ENOENT && ret != -EEXIST)
1965 update_size = false;
1969 if (!ret && update_size) {
1970 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2);
1971 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
1975 if (!ret && name_added)
1980 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1981 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1982 struct btrfs_root *root,
1983 struct btrfs_path *path,
1984 struct extent_buffer *eb, int slot,
1985 struct btrfs_key *key)
1988 struct btrfs_dir_item *di;
1990 /* We only log dir index keys, which only contain a single dir item. */
1991 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
1993 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1994 ret = replay_one_name(trans, root, path, eb, di, key);
1999 * If this entry refers to a non-directory (directories can not have a
2000 * link count > 1) and it was added in the transaction that was not
2001 * committed, make sure we fixup the link count of the inode the entry
2002 * points to. Otherwise something like the following would result in a
2003 * directory pointing to an inode with a wrong link that does not account
2004 * for this dir entry:
2011 * ln testdir/bar testdir/bar_link
2012 * ln testdir/foo testdir/foo_link
2013 * xfs_io -c "fsync" testdir/bar
2017 * mount fs, log replay happens
2019 * File foo would remain with a link count of 1 when it has two entries
2020 * pointing to it in the directory testdir. This would make it impossible
2021 * to ever delete the parent directory has it would result in stale
2022 * dentries that can never be deleted.
2024 if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) {
2025 struct btrfs_path *fixup_path;
2026 struct btrfs_key di_key;
2028 fixup_path = btrfs_alloc_path();
2032 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2033 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2034 btrfs_free_path(fixup_path);
2041 * directory replay has two parts. There are the standard directory
2042 * items in the log copied from the subvolume, and range items
2043 * created in the log while the subvolume was logged.
2045 * The range items tell us which parts of the key space the log
2046 * is authoritative for. During replay, if a key in the subvolume
2047 * directory is in a logged range item, but not actually in the log
2048 * that means it was deleted from the directory before the fsync
2049 * and should be removed.
2051 static noinline int find_dir_range(struct btrfs_root *root,
2052 struct btrfs_path *path,
2054 u64 *start_ret, u64 *end_ret)
2056 struct btrfs_key key;
2058 struct btrfs_dir_log_item *item;
2062 if (*start_ret == (u64)-1)
2065 key.objectid = dirid;
2066 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2067 key.offset = *start_ret;
2069 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2073 if (path->slots[0] == 0)
2078 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2080 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2084 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2085 struct btrfs_dir_log_item);
2086 found_end = btrfs_dir_log_end(path->nodes[0], item);
2088 if (*start_ret >= key.offset && *start_ret <= found_end) {
2090 *start_ret = key.offset;
2091 *end_ret = found_end;
2096 /* check the next slot in the tree to see if it is a valid item */
2097 nritems = btrfs_header_nritems(path->nodes[0]);
2099 if (path->slots[0] >= nritems) {
2100 ret = btrfs_next_leaf(root, path);
2105 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2107 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2111 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2112 struct btrfs_dir_log_item);
2113 found_end = btrfs_dir_log_end(path->nodes[0], item);
2114 *start_ret = key.offset;
2115 *end_ret = found_end;
2118 btrfs_release_path(path);
2123 * this looks for a given directory item in the log. If the directory
2124 * item is not in the log, the item is removed and the inode it points
2127 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2128 struct btrfs_root *log,
2129 struct btrfs_path *path,
2130 struct btrfs_path *log_path,
2132 struct btrfs_key *dir_key)
2134 struct btrfs_root *root = BTRFS_I(dir)->root;
2136 struct extent_buffer *eb;
2138 struct btrfs_dir_item *di;
2139 struct fscrypt_str name;
2140 struct inode *inode = NULL;
2141 struct btrfs_key location;
2144 * Currently we only log dir index keys. Even if we replay a log created
2145 * by an older kernel that logged both dir index and dir item keys, all
2146 * we need to do is process the dir index keys, we (and our caller) can
2147 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2149 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2151 eb = path->nodes[0];
2152 slot = path->slots[0];
2153 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2154 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
2159 struct btrfs_dir_item *log_di;
2161 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2163 dir_key->offset, &name, 0);
2164 if (IS_ERR(log_di)) {
2165 ret = PTR_ERR(log_di);
2167 } else if (log_di) {
2168 /* The dentry exists in the log, we have nothing to do. */
2174 btrfs_dir_item_key_to_cpu(eb, di, &location);
2175 btrfs_release_path(path);
2176 btrfs_release_path(log_path);
2177 inode = read_one_inode(root, location.objectid);
2183 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2188 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2191 * Unlike dir item keys, dir index keys can only have one name (entry) in
2192 * them, as there are no key collisions since each key has a unique offset
2193 * (an index number), so we're done.
2196 btrfs_release_path(path);
2197 btrfs_release_path(log_path);
2203 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2204 struct btrfs_root *root,
2205 struct btrfs_root *log,
2206 struct btrfs_path *path,
2209 struct btrfs_key search_key;
2210 struct btrfs_path *log_path;
2215 log_path = btrfs_alloc_path();
2219 search_key.objectid = ino;
2220 search_key.type = BTRFS_XATTR_ITEM_KEY;
2221 search_key.offset = 0;
2223 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2227 nritems = btrfs_header_nritems(path->nodes[0]);
2228 for (i = path->slots[0]; i < nritems; i++) {
2229 struct btrfs_key key;
2230 struct btrfs_dir_item *di;
2231 struct btrfs_dir_item *log_di;
2235 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2236 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2241 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2242 total_size = btrfs_item_size(path->nodes[0], i);
2244 while (cur < total_size) {
2245 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2246 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2247 u32 this_len = sizeof(*di) + name_len + data_len;
2250 name = kmalloc(name_len, GFP_NOFS);
2255 read_extent_buffer(path->nodes[0], name,
2256 (unsigned long)(di + 1), name_len);
2258 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2260 btrfs_release_path(log_path);
2262 /* Doesn't exist in log tree, so delete it. */
2263 btrfs_release_path(path);
2264 di = btrfs_lookup_xattr(trans, root, path, ino,
2265 name, name_len, -1);
2272 ret = btrfs_delete_one_dir_name(trans, root,
2276 btrfs_release_path(path);
2281 if (IS_ERR(log_di)) {
2282 ret = PTR_ERR(log_di);
2286 di = (struct btrfs_dir_item *)((char *)di + this_len);
2289 ret = btrfs_next_leaf(root, path);
2295 btrfs_free_path(log_path);
2296 btrfs_release_path(path);
2302 * deletion replay happens before we copy any new directory items
2303 * out of the log or out of backreferences from inodes. It
2304 * scans the log to find ranges of keys that log is authoritative for,
2305 * and then scans the directory to find items in those ranges that are
2306 * not present in the log.
2308 * Anything we don't find in the log is unlinked and removed from the
2311 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2312 struct btrfs_root *root,
2313 struct btrfs_root *log,
2314 struct btrfs_path *path,
2315 u64 dirid, int del_all)
2320 struct btrfs_key dir_key;
2321 struct btrfs_key found_key;
2322 struct btrfs_path *log_path;
2325 dir_key.objectid = dirid;
2326 dir_key.type = BTRFS_DIR_INDEX_KEY;
2327 log_path = btrfs_alloc_path();
2331 dir = read_one_inode(root, dirid);
2332 /* it isn't an error if the inode isn't there, that can happen
2333 * because we replay the deletes before we copy in the inode item
2337 btrfs_free_path(log_path);
2345 range_end = (u64)-1;
2347 ret = find_dir_range(log, path, dirid,
2348 &range_start, &range_end);
2355 dir_key.offset = range_start;
2358 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2363 nritems = btrfs_header_nritems(path->nodes[0]);
2364 if (path->slots[0] >= nritems) {
2365 ret = btrfs_next_leaf(root, path);
2371 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2373 if (found_key.objectid != dirid ||
2374 found_key.type != dir_key.type) {
2379 if (found_key.offset > range_end)
2382 ret = check_item_in_log(trans, log, path,
2387 if (found_key.offset == (u64)-1)
2389 dir_key.offset = found_key.offset + 1;
2391 btrfs_release_path(path);
2392 if (range_end == (u64)-1)
2394 range_start = range_end + 1;
2398 btrfs_release_path(path);
2399 btrfs_free_path(log_path);
2405 * the process_func used to replay items from the log tree. This
2406 * gets called in two different stages. The first stage just looks
2407 * for inodes and makes sure they are all copied into the subvolume.
2409 * The second stage copies all the other item types from the log into
2410 * the subvolume. The two stage approach is slower, but gets rid of
2411 * lots of complexity around inodes referencing other inodes that exist
2412 * only in the log (references come from either directory items or inode
2415 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2416 struct walk_control *wc, u64 gen, int level)
2419 struct btrfs_tree_parent_check check = {
2423 struct btrfs_path *path;
2424 struct btrfs_root *root = wc->replay_dest;
2425 struct btrfs_key key;
2429 ret = btrfs_read_extent_buffer(eb, &check);
2433 level = btrfs_header_level(eb);
2438 path = btrfs_alloc_path();
2442 nritems = btrfs_header_nritems(eb);
2443 for (i = 0; i < nritems; i++) {
2444 btrfs_item_key_to_cpu(eb, &key, i);
2446 /* inode keys are done during the first stage */
2447 if (key.type == BTRFS_INODE_ITEM_KEY &&
2448 wc->stage == LOG_WALK_REPLAY_INODES) {
2449 struct btrfs_inode_item *inode_item;
2452 inode_item = btrfs_item_ptr(eb, i,
2453 struct btrfs_inode_item);
2455 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2456 * and never got linked before the fsync, skip it, as
2457 * replaying it is pointless since it would be deleted
2458 * later. We skip logging tmpfiles, but it's always
2459 * possible we are replaying a log created with a kernel
2460 * that used to log tmpfiles.
2462 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2463 wc->ignore_cur_inode = true;
2466 wc->ignore_cur_inode = false;
2468 ret = replay_xattr_deletes(wc->trans, root, log,
2469 path, key.objectid);
2472 mode = btrfs_inode_mode(eb, inode_item);
2473 if (S_ISDIR(mode)) {
2474 ret = replay_dir_deletes(wc->trans,
2475 root, log, path, key.objectid, 0);
2479 ret = overwrite_item(wc->trans, root, path,
2485 * Before replaying extents, truncate the inode to its
2486 * size. We need to do it now and not after log replay
2487 * because before an fsync we can have prealloc extents
2488 * added beyond the inode's i_size. If we did it after,
2489 * through orphan cleanup for example, we would drop
2490 * those prealloc extents just after replaying them.
2492 if (S_ISREG(mode)) {
2493 struct btrfs_drop_extents_args drop_args = { 0 };
2494 struct inode *inode;
2497 inode = read_one_inode(root, key.objectid);
2502 from = ALIGN(i_size_read(inode),
2503 root->fs_info->sectorsize);
2504 drop_args.start = from;
2505 drop_args.end = (u64)-1;
2506 drop_args.drop_cache = true;
2507 ret = btrfs_drop_extents(wc->trans, root,
2511 inode_sub_bytes(inode,
2512 drop_args.bytes_found);
2513 /* Update the inode's nbytes. */
2514 ret = btrfs_update_inode(wc->trans,
2515 root, BTRFS_I(inode));
2522 ret = link_to_fixup_dir(wc->trans, root,
2523 path, key.objectid);
2528 if (wc->ignore_cur_inode)
2531 if (key.type == BTRFS_DIR_INDEX_KEY &&
2532 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2533 ret = replay_one_dir_item(wc->trans, root, path,
2539 if (wc->stage < LOG_WALK_REPLAY_ALL)
2542 /* these keys are simply copied */
2543 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2544 ret = overwrite_item(wc->trans, root, path,
2548 } else if (key.type == BTRFS_INODE_REF_KEY ||
2549 key.type == BTRFS_INODE_EXTREF_KEY) {
2550 ret = add_inode_ref(wc->trans, root, log, path,
2552 if (ret && ret != -ENOENT)
2555 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2556 ret = replay_one_extent(wc->trans, root, path,
2562 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2563 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2564 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2565 * older kernel with such keys, ignore them.
2568 btrfs_free_path(path);
2573 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2575 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2577 struct btrfs_block_group *cache;
2579 cache = btrfs_lookup_block_group(fs_info, start);
2581 btrfs_err(fs_info, "unable to find block group for %llu", start);
2585 spin_lock(&cache->space_info->lock);
2586 spin_lock(&cache->lock);
2587 cache->reserved -= fs_info->nodesize;
2588 cache->space_info->bytes_reserved -= fs_info->nodesize;
2589 spin_unlock(&cache->lock);
2590 spin_unlock(&cache->space_info->lock);
2592 btrfs_put_block_group(cache);
2595 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2596 struct btrfs_root *root,
2597 struct btrfs_path *path, int *level,
2598 struct walk_control *wc)
2600 struct btrfs_fs_info *fs_info = root->fs_info;
2603 struct extent_buffer *next;
2604 struct extent_buffer *cur;
2608 while (*level > 0) {
2609 struct btrfs_tree_parent_check check = { 0 };
2611 cur = path->nodes[*level];
2613 WARN_ON(btrfs_header_level(cur) != *level);
2615 if (path->slots[*level] >=
2616 btrfs_header_nritems(cur))
2619 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2620 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2621 check.transid = ptr_gen;
2622 check.level = *level - 1;
2623 check.has_first_key = true;
2624 btrfs_node_key_to_cpu(cur, &check.first_key, path->slots[*level]);
2625 blocksize = fs_info->nodesize;
2627 next = btrfs_find_create_tree_block(fs_info, bytenr,
2628 btrfs_header_owner(cur),
2631 return PTR_ERR(next);
2634 ret = wc->process_func(root, next, wc, ptr_gen,
2637 free_extent_buffer(next);
2641 path->slots[*level]++;
2643 ret = btrfs_read_extent_buffer(next, &check);
2645 free_extent_buffer(next);
2650 btrfs_tree_lock(next);
2651 btrfs_clean_tree_block(next);
2652 btrfs_wait_tree_block_writeback(next);
2653 btrfs_tree_unlock(next);
2654 ret = btrfs_pin_reserved_extent(trans,
2657 free_extent_buffer(next);
2660 btrfs_redirty_list_add(
2661 trans->transaction, next);
2663 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2664 clear_extent_buffer_dirty(next);
2665 unaccount_log_buffer(fs_info, bytenr);
2668 free_extent_buffer(next);
2671 ret = btrfs_read_extent_buffer(next, &check);
2673 free_extent_buffer(next);
2677 if (path->nodes[*level-1])
2678 free_extent_buffer(path->nodes[*level-1]);
2679 path->nodes[*level-1] = next;
2680 *level = btrfs_header_level(next);
2681 path->slots[*level] = 0;
2684 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2690 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *root,
2692 struct btrfs_path *path, int *level,
2693 struct walk_control *wc)
2695 struct btrfs_fs_info *fs_info = root->fs_info;
2700 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2701 slot = path->slots[i];
2702 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2705 WARN_ON(*level == 0);
2708 ret = wc->process_func(root, path->nodes[*level], wc,
2709 btrfs_header_generation(path->nodes[*level]),
2715 struct extent_buffer *next;
2717 next = path->nodes[*level];
2720 btrfs_tree_lock(next);
2721 btrfs_clean_tree_block(next);
2722 btrfs_wait_tree_block_writeback(next);
2723 btrfs_tree_unlock(next);
2724 ret = btrfs_pin_reserved_extent(trans,
2725 path->nodes[*level]->start,
2726 path->nodes[*level]->len);
2729 btrfs_redirty_list_add(trans->transaction,
2732 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2733 clear_extent_buffer_dirty(next);
2735 unaccount_log_buffer(fs_info,
2736 path->nodes[*level]->start);
2739 free_extent_buffer(path->nodes[*level]);
2740 path->nodes[*level] = NULL;
2748 * drop the reference count on the tree rooted at 'snap'. This traverses
2749 * the tree freeing any blocks that have a ref count of zero after being
2752 static int walk_log_tree(struct btrfs_trans_handle *trans,
2753 struct btrfs_root *log, struct walk_control *wc)
2755 struct btrfs_fs_info *fs_info = log->fs_info;
2759 struct btrfs_path *path;
2762 path = btrfs_alloc_path();
2766 level = btrfs_header_level(log->node);
2768 path->nodes[level] = log->node;
2769 atomic_inc(&log->node->refs);
2770 path->slots[level] = 0;
2773 wret = walk_down_log_tree(trans, log, path, &level, wc);
2781 wret = walk_up_log_tree(trans, log, path, &level, wc);
2790 /* was the root node processed? if not, catch it here */
2791 if (path->nodes[orig_level]) {
2792 ret = wc->process_func(log, path->nodes[orig_level], wc,
2793 btrfs_header_generation(path->nodes[orig_level]),
2798 struct extent_buffer *next;
2800 next = path->nodes[orig_level];
2803 btrfs_tree_lock(next);
2804 btrfs_clean_tree_block(next);
2805 btrfs_wait_tree_block_writeback(next);
2806 btrfs_tree_unlock(next);
2807 ret = btrfs_pin_reserved_extent(trans,
2808 next->start, next->len);
2811 btrfs_redirty_list_add(trans->transaction, next);
2813 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2814 clear_extent_buffer_dirty(next);
2815 unaccount_log_buffer(fs_info, next->start);
2821 btrfs_free_path(path);
2826 * helper function to update the item for a given subvolumes log root
2827 * in the tree of log roots
2829 static int update_log_root(struct btrfs_trans_handle *trans,
2830 struct btrfs_root *log,
2831 struct btrfs_root_item *root_item)
2833 struct btrfs_fs_info *fs_info = log->fs_info;
2836 if (log->log_transid == 1) {
2837 /* insert root item on the first sync */
2838 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2839 &log->root_key, root_item);
2841 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2842 &log->root_key, root_item);
2847 static void wait_log_commit(struct btrfs_root *root, int transid)
2850 int index = transid % 2;
2853 * we only allow two pending log transactions at a time,
2854 * so we know that if ours is more than 2 older than the
2855 * current transaction, we're done
2858 prepare_to_wait(&root->log_commit_wait[index],
2859 &wait, TASK_UNINTERRUPTIBLE);
2861 if (!(root->log_transid_committed < transid &&
2862 atomic_read(&root->log_commit[index])))
2865 mutex_unlock(&root->log_mutex);
2867 mutex_lock(&root->log_mutex);
2869 finish_wait(&root->log_commit_wait[index], &wait);
2872 static void wait_for_writer(struct btrfs_root *root)
2877 prepare_to_wait(&root->log_writer_wait, &wait,
2878 TASK_UNINTERRUPTIBLE);
2879 if (!atomic_read(&root->log_writers))
2882 mutex_unlock(&root->log_mutex);
2884 mutex_lock(&root->log_mutex);
2886 finish_wait(&root->log_writer_wait, &wait);
2889 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2890 struct btrfs_log_ctx *ctx)
2892 mutex_lock(&root->log_mutex);
2893 list_del_init(&ctx->list);
2894 mutex_unlock(&root->log_mutex);
2898 * Invoked in log mutex context, or be sure there is no other task which
2899 * can access the list.
2901 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2902 int index, int error)
2904 struct btrfs_log_ctx *ctx;
2905 struct btrfs_log_ctx *safe;
2907 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2908 list_del_init(&ctx->list);
2909 ctx->log_ret = error;
2914 * btrfs_sync_log does sends a given tree log down to the disk and
2915 * updates the super blocks to record it. When this call is done,
2916 * you know that any inodes previously logged are safely on disk only
2919 * Any other return value means you need to call btrfs_commit_transaction.
2920 * Some of the edge cases for fsyncing directories that have had unlinks
2921 * or renames done in the past mean that sometimes the only safe
2922 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2923 * that has happened.
2925 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2926 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2932 struct btrfs_fs_info *fs_info = root->fs_info;
2933 struct btrfs_root *log = root->log_root;
2934 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2935 struct btrfs_root_item new_root_item;
2936 int log_transid = 0;
2937 struct btrfs_log_ctx root_log_ctx;
2938 struct blk_plug plug;
2942 mutex_lock(&root->log_mutex);
2943 log_transid = ctx->log_transid;
2944 if (root->log_transid_committed >= log_transid) {
2945 mutex_unlock(&root->log_mutex);
2946 return ctx->log_ret;
2949 index1 = log_transid % 2;
2950 if (atomic_read(&root->log_commit[index1])) {
2951 wait_log_commit(root, log_transid);
2952 mutex_unlock(&root->log_mutex);
2953 return ctx->log_ret;
2955 ASSERT(log_transid == root->log_transid);
2956 atomic_set(&root->log_commit[index1], 1);
2958 /* wait for previous tree log sync to complete */
2959 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2960 wait_log_commit(root, log_transid - 1);
2963 int batch = atomic_read(&root->log_batch);
2964 /* when we're on an ssd, just kick the log commit out */
2965 if (!btrfs_test_opt(fs_info, SSD) &&
2966 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2967 mutex_unlock(&root->log_mutex);
2968 schedule_timeout_uninterruptible(1);
2969 mutex_lock(&root->log_mutex);
2971 wait_for_writer(root);
2972 if (batch == atomic_read(&root->log_batch))
2976 /* bail out if we need to do a full commit */
2977 if (btrfs_need_log_full_commit(trans)) {
2978 ret = BTRFS_LOG_FORCE_COMMIT;
2979 mutex_unlock(&root->log_mutex);
2983 if (log_transid % 2 == 0)
2984 mark = EXTENT_DIRTY;
2988 /* we start IO on all the marked extents here, but we don't actually
2989 * wait for them until later.
2991 blk_start_plug(&plug);
2992 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2994 * -EAGAIN happens when someone, e.g., a concurrent transaction
2995 * commit, writes a dirty extent in this tree-log commit. This
2996 * concurrent write will create a hole writing out the extents,
2997 * and we cannot proceed on a zoned filesystem, requiring
2998 * sequential writing. While we can bail out to a full commit
2999 * here, but we can continue hoping the concurrent writing fills
3002 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3005 blk_finish_plug(&plug);
3006 btrfs_abort_transaction(trans, ret);
3007 btrfs_set_log_full_commit(trans);
3008 mutex_unlock(&root->log_mutex);
3013 * We _must_ update under the root->log_mutex in order to make sure we
3014 * have a consistent view of the log root we are trying to commit at
3017 * We _must_ copy this into a local copy, because we are not holding the
3018 * log_root_tree->log_mutex yet. This is important because when we
3019 * commit the log_root_tree we must have a consistent view of the
3020 * log_root_tree when we update the super block to point at the
3021 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3022 * with the commit and possibly point at the new block which we may not
3025 btrfs_set_root_node(&log->root_item, log->node);
3026 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3028 root->log_transid++;
3029 log->log_transid = root->log_transid;
3030 root->log_start_pid = 0;
3032 * IO has been started, blocks of the log tree have WRITTEN flag set
3033 * in their headers. new modifications of the log will be written to
3034 * new positions. so it's safe to allow log writers to go in.
3036 mutex_unlock(&root->log_mutex);
3038 if (btrfs_is_zoned(fs_info)) {
3039 mutex_lock(&fs_info->tree_root->log_mutex);
3040 if (!log_root_tree->node) {
3041 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3043 mutex_unlock(&fs_info->tree_root->log_mutex);
3044 blk_finish_plug(&plug);
3048 mutex_unlock(&fs_info->tree_root->log_mutex);
3051 btrfs_init_log_ctx(&root_log_ctx, NULL);
3053 mutex_lock(&log_root_tree->log_mutex);
3055 index2 = log_root_tree->log_transid % 2;
3056 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3057 root_log_ctx.log_transid = log_root_tree->log_transid;
3060 * Now we are safe to update the log_root_tree because we're under the
3061 * log_mutex, and we're a current writer so we're holding the commit
3062 * open until we drop the log_mutex.
3064 ret = update_log_root(trans, log, &new_root_item);
3066 if (!list_empty(&root_log_ctx.list))
3067 list_del_init(&root_log_ctx.list);
3069 blk_finish_plug(&plug);
3070 btrfs_set_log_full_commit(trans);
3072 if (ret != -ENOSPC) {
3073 btrfs_abort_transaction(trans, ret);
3074 mutex_unlock(&log_root_tree->log_mutex);
3077 btrfs_wait_tree_log_extents(log, mark);
3078 mutex_unlock(&log_root_tree->log_mutex);
3079 ret = BTRFS_LOG_FORCE_COMMIT;
3083 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3084 blk_finish_plug(&plug);
3085 list_del_init(&root_log_ctx.list);
3086 mutex_unlock(&log_root_tree->log_mutex);
3087 ret = root_log_ctx.log_ret;
3091 index2 = root_log_ctx.log_transid % 2;
3092 if (atomic_read(&log_root_tree->log_commit[index2])) {
3093 blk_finish_plug(&plug);
3094 ret = btrfs_wait_tree_log_extents(log, mark);
3095 wait_log_commit(log_root_tree,
3096 root_log_ctx.log_transid);
3097 mutex_unlock(&log_root_tree->log_mutex);
3099 ret = root_log_ctx.log_ret;
3102 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3103 atomic_set(&log_root_tree->log_commit[index2], 1);
3105 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3106 wait_log_commit(log_root_tree,
3107 root_log_ctx.log_transid - 1);
3111 * now that we've moved on to the tree of log tree roots,
3112 * check the full commit flag again
3114 if (btrfs_need_log_full_commit(trans)) {
3115 blk_finish_plug(&plug);
3116 btrfs_wait_tree_log_extents(log, mark);
3117 mutex_unlock(&log_root_tree->log_mutex);
3118 ret = BTRFS_LOG_FORCE_COMMIT;
3119 goto out_wake_log_root;
3122 ret = btrfs_write_marked_extents(fs_info,
3123 &log_root_tree->dirty_log_pages,
3124 EXTENT_DIRTY | EXTENT_NEW);
3125 blk_finish_plug(&plug);
3127 * As described above, -EAGAIN indicates a hole in the extents. We
3128 * cannot wait for these write outs since the waiting cause a
3129 * deadlock. Bail out to the full commit instead.
3131 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3132 btrfs_set_log_full_commit(trans);
3133 btrfs_wait_tree_log_extents(log, mark);
3134 mutex_unlock(&log_root_tree->log_mutex);
3135 goto out_wake_log_root;
3137 btrfs_set_log_full_commit(trans);
3138 btrfs_abort_transaction(trans, ret);
3139 mutex_unlock(&log_root_tree->log_mutex);
3140 goto out_wake_log_root;
3142 ret = btrfs_wait_tree_log_extents(log, mark);
3144 ret = btrfs_wait_tree_log_extents(log_root_tree,
3145 EXTENT_NEW | EXTENT_DIRTY);
3147 btrfs_set_log_full_commit(trans);
3148 mutex_unlock(&log_root_tree->log_mutex);
3149 goto out_wake_log_root;
3152 log_root_start = log_root_tree->node->start;
3153 log_root_level = btrfs_header_level(log_root_tree->node);
3154 log_root_tree->log_transid++;
3155 mutex_unlock(&log_root_tree->log_mutex);
3158 * Here we are guaranteed that nobody is going to write the superblock
3159 * for the current transaction before us and that neither we do write
3160 * our superblock before the previous transaction finishes its commit
3161 * and writes its superblock, because:
3163 * 1) We are holding a handle on the current transaction, so no body
3164 * can commit it until we release the handle;
3166 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3167 * if the previous transaction is still committing, and hasn't yet
3168 * written its superblock, we wait for it to do it, because a
3169 * transaction commit acquires the tree_log_mutex when the commit
3170 * begins and releases it only after writing its superblock.
3172 mutex_lock(&fs_info->tree_log_mutex);
3175 * The previous transaction writeout phase could have failed, and thus
3176 * marked the fs in an error state. We must not commit here, as we
3177 * could have updated our generation in the super_for_commit and
3178 * writing the super here would result in transid mismatches. If there
3179 * is an error here just bail.
3181 if (BTRFS_FS_ERROR(fs_info)) {
3183 btrfs_set_log_full_commit(trans);
3184 btrfs_abort_transaction(trans, ret);
3185 mutex_unlock(&fs_info->tree_log_mutex);
3186 goto out_wake_log_root;
3189 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3190 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3191 ret = write_all_supers(fs_info, 1);
3192 mutex_unlock(&fs_info->tree_log_mutex);
3194 btrfs_set_log_full_commit(trans);
3195 btrfs_abort_transaction(trans, ret);
3196 goto out_wake_log_root;
3200 * We know there can only be one task here, since we have not yet set
3201 * root->log_commit[index1] to 0 and any task attempting to sync the
3202 * log must wait for the previous log transaction to commit if it's
3203 * still in progress or wait for the current log transaction commit if
3204 * someone else already started it. We use <= and not < because the
3205 * first log transaction has an ID of 0.
3207 ASSERT(root->last_log_commit <= log_transid);
3208 root->last_log_commit = log_transid;
3211 mutex_lock(&log_root_tree->log_mutex);
3212 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3214 log_root_tree->log_transid_committed++;
3215 atomic_set(&log_root_tree->log_commit[index2], 0);
3216 mutex_unlock(&log_root_tree->log_mutex);
3219 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3220 * all the updates above are seen by the woken threads. It might not be
3221 * necessary, but proving that seems to be hard.
3223 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3225 mutex_lock(&root->log_mutex);
3226 btrfs_remove_all_log_ctxs(root, index1, ret);
3227 root->log_transid_committed++;
3228 atomic_set(&root->log_commit[index1], 0);
3229 mutex_unlock(&root->log_mutex);
3232 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3233 * all the updates above are seen by the woken threads. It might not be
3234 * necessary, but proving that seems to be hard.
3236 cond_wake_up(&root->log_commit_wait[index1]);
3240 static void free_log_tree(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *log)
3244 struct walk_control wc = {
3246 .process_func = process_one_buffer
3250 ret = walk_log_tree(trans, log, &wc);
3253 * We weren't able to traverse the entire log tree, the
3254 * typical scenario is getting an -EIO when reading an
3255 * extent buffer of the tree, due to a previous writeback
3258 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3259 &log->fs_info->fs_state);
3262 * Some extent buffers of the log tree may still be dirty
3263 * and not yet written back to storage, because we may
3264 * have updates to a log tree without syncing a log tree,
3265 * such as during rename and link operations. So flush
3266 * them out and wait for their writeback to complete, so
3267 * that we properly cleanup their state and pages.
3269 btrfs_write_marked_extents(log->fs_info,
3270 &log->dirty_log_pages,
3271 EXTENT_DIRTY | EXTENT_NEW);
3272 btrfs_wait_tree_log_extents(log,
3273 EXTENT_DIRTY | EXTENT_NEW);
3276 btrfs_abort_transaction(trans, ret);
3278 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3282 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3283 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3284 extent_io_tree_release(&log->log_csum_range);
3286 btrfs_put_root(log);
3290 * free all the extents used by the tree log. This should be called
3291 * at commit time of the full transaction
3293 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3295 if (root->log_root) {
3296 free_log_tree(trans, root->log_root);
3297 root->log_root = NULL;
3298 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3303 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3304 struct btrfs_fs_info *fs_info)
3306 if (fs_info->log_root_tree) {
3307 free_log_tree(trans, fs_info->log_root_tree);
3308 fs_info->log_root_tree = NULL;
3309 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3315 * Check if an inode was logged in the current transaction. This correctly deals
3316 * with the case where the inode was logged but has a logged_trans of 0, which
3317 * happens if the inode is evicted and loaded again, as logged_trans is an in
3318 * memory only field (not persisted).
3320 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3323 static int inode_logged(struct btrfs_trans_handle *trans,
3324 struct btrfs_inode *inode,
3325 struct btrfs_path *path_in)
3327 struct btrfs_path *path = path_in;
3328 struct btrfs_key key;
3331 if (inode->logged_trans == trans->transid)
3335 * If logged_trans is not 0, then we know the inode logged was not logged
3336 * in this transaction, so we can return false right away.
3338 if (inode->logged_trans > 0)
3342 * If no log tree was created for this root in this transaction, then
3343 * the inode can not have been logged in this transaction. In that case
3344 * set logged_trans to anything greater than 0 and less than the current
3345 * transaction's ID, to avoid the search below in a future call in case
3346 * a log tree gets created after this.
3348 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3349 inode->logged_trans = trans->transid - 1;
3354 * We have a log tree and the inode's logged_trans is 0. We can't tell
3355 * for sure if the inode was logged before in this transaction by looking
3356 * only at logged_trans. We could be pessimistic and assume it was, but
3357 * that can lead to unnecessarily logging an inode during rename and link
3358 * operations, and then further updating the log in followup rename and
3359 * link operations, specially if it's a directory, which adds latency
3360 * visible to applications doing a series of rename or link operations.
3362 * A logged_trans of 0 here can mean several things:
3364 * 1) The inode was never logged since the filesystem was mounted, and may
3365 * or may have not been evicted and loaded again;
3367 * 2) The inode was logged in a previous transaction, then evicted and
3368 * then loaded again;
3370 * 3) The inode was logged in the current transaction, then evicted and
3371 * then loaded again.
3373 * For cases 1) and 2) we don't want to return true, but we need to detect
3374 * case 3) and return true. So we do a search in the log root for the inode
3377 key.objectid = btrfs_ino(inode);
3378 key.type = BTRFS_INODE_ITEM_KEY;
3382 path = btrfs_alloc_path();
3387 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3390 btrfs_release_path(path);
3392 btrfs_free_path(path);
3395 * Logging an inode always results in logging its inode item. So if we
3396 * did not find the item we know the inode was not logged for sure.
3400 } else if (ret > 0) {
3402 * Set logged_trans to a value greater than 0 and less then the
3403 * current transaction to avoid doing the search in future calls.
3405 inode->logged_trans = trans->transid - 1;
3410 * The inode was previously logged and then evicted, set logged_trans to
3411 * the current transacion's ID, to avoid future tree searches as long as
3412 * the inode is not evicted again.
3414 inode->logged_trans = trans->transid;
3417 * If it's a directory, then we must set last_dir_index_offset to the
3418 * maximum possible value, so that the next attempt to log the inode does
3419 * not skip checking if dir index keys found in modified subvolume tree
3420 * leaves have been logged before, otherwise it would result in attempts
3421 * to insert duplicate dir index keys in the log tree. This must be done
3422 * because last_dir_index_offset is an in-memory only field, not persisted
3423 * in the inode item or any other on-disk structure, so its value is lost
3424 * once the inode is evicted.
3426 if (S_ISDIR(inode->vfs_inode.i_mode))
3427 inode->last_dir_index_offset = (u64)-1;
3433 * Delete a directory entry from the log if it exists.
3435 * Returns < 0 on error
3436 * 1 if the entry does not exists
3437 * 0 if the entry existed and was successfully deleted
3439 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3440 struct btrfs_root *log,
3441 struct btrfs_path *path,
3443 const struct fscrypt_str *name,
3446 struct btrfs_dir_item *di;
3449 * We only log dir index items of a directory, so we don't need to look
3450 * for dir item keys.
3452 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3460 * We do not need to update the size field of the directory's
3461 * inode item because on log replay we update the field to reflect
3462 * all existing entries in the directory (see overwrite_item()).
3464 return btrfs_delete_one_dir_name(trans, log, path, di);
3468 * If both a file and directory are logged, and unlinks or renames are
3469 * mixed in, we have a few interesting corners:
3471 * create file X in dir Y
3472 * link file X to X.link in dir Y
3474 * unlink file X but leave X.link
3477 * After a crash we would expect only X.link to exist. But file X
3478 * didn't get fsync'd again so the log has back refs for X and X.link.
3480 * We solve this by removing directory entries and inode backrefs from the
3481 * log when a file that was logged in the current transaction is
3482 * unlinked. Any later fsync will include the updated log entries, and
3483 * we'll be able to reconstruct the proper directory items from backrefs.
3485 * This optimizations allows us to avoid relogging the entire inode
3486 * or the entire directory.
3488 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *root,
3490 const struct fscrypt_str *name,
3491 struct btrfs_inode *dir, u64 index)
3493 struct btrfs_path *path;
3496 ret = inode_logged(trans, dir, NULL);
3500 btrfs_set_log_full_commit(trans);
3504 ret = join_running_log_trans(root);
3508 mutex_lock(&dir->log_mutex);
3510 path = btrfs_alloc_path();
3516 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3518 btrfs_free_path(path);
3520 mutex_unlock(&dir->log_mutex);
3522 btrfs_set_log_full_commit(trans);
3523 btrfs_end_log_trans(root);
3526 /* see comments for btrfs_del_dir_entries_in_log */
3527 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3528 struct btrfs_root *root,
3529 const struct fscrypt_str *name,
3530 struct btrfs_inode *inode, u64 dirid)
3532 struct btrfs_root *log;
3536 ret = inode_logged(trans, inode, NULL);
3540 btrfs_set_log_full_commit(trans);
3544 ret = join_running_log_trans(root);
3547 log = root->log_root;
3548 mutex_lock(&inode->log_mutex);
3550 ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode),
3552 mutex_unlock(&inode->log_mutex);
3553 if (ret < 0 && ret != -ENOENT)
3554 btrfs_set_log_full_commit(trans);
3555 btrfs_end_log_trans(root);
3559 * creates a range item in the log for 'dirid'. first_offset and
3560 * last_offset tell us which parts of the key space the log should
3561 * be considered authoritative for.
3563 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3564 struct btrfs_root *log,
3565 struct btrfs_path *path,
3567 u64 first_offset, u64 last_offset)
3570 struct btrfs_key key;
3571 struct btrfs_dir_log_item *item;
3573 key.objectid = dirid;
3574 key.offset = first_offset;
3575 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3576 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3578 * -EEXIST is fine and can happen sporadically when we are logging a
3579 * directory and have concurrent insertions in the subvolume's tree for
3580 * items from other inodes and that result in pushing off some dir items
3581 * from one leaf to another in order to accommodate for the new items.
3582 * This results in logging the same dir index range key.
3584 if (ret && ret != -EEXIST)
3587 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3588 struct btrfs_dir_log_item);
3589 if (ret == -EEXIST) {
3590 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3593 * btrfs_del_dir_entries_in_log() might have been called during
3594 * an unlink between the initial insertion of this key and the
3595 * current update, or we might be logging a single entry deletion
3596 * during a rename, so set the new last_offset to the max value.
3598 last_offset = max(last_offset, curr_end);
3600 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3601 btrfs_mark_buffer_dirty(path->nodes[0]);
3602 btrfs_release_path(path);
3606 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3607 struct btrfs_root *log,
3608 struct extent_buffer *src,
3609 struct btrfs_path *dst_path,
3613 char *ins_data = NULL;
3614 struct btrfs_item_batch batch;
3615 struct extent_buffer *dst;
3616 unsigned long src_offset;
3617 unsigned long dst_offset;
3618 struct btrfs_key key;
3627 btrfs_item_key_to_cpu(src, &key, start_slot);
3628 item_size = btrfs_item_size(src, start_slot);
3630 batch.data_sizes = &item_size;
3631 batch.total_data_size = item_size;
3633 struct btrfs_key *ins_keys;
3636 ins_data = kmalloc(count * sizeof(u32) +
3637 count * sizeof(struct btrfs_key), GFP_NOFS);
3641 ins_sizes = (u32 *)ins_data;
3642 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3643 batch.keys = ins_keys;
3644 batch.data_sizes = ins_sizes;
3645 batch.total_data_size = 0;
3647 for (i = 0; i < count; i++) {
3648 const int slot = start_slot + i;
3650 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3651 ins_sizes[i] = btrfs_item_size(src, slot);
3652 batch.total_data_size += ins_sizes[i];
3656 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3660 dst = dst_path->nodes[0];
3662 * Copy all the items in bulk, in a single copy operation. Item data is
3663 * organized such that it's placed at the end of a leaf and from right
3664 * to left. For example, the data for the second item ends at an offset
3665 * that matches the offset where the data for the first item starts, the
3666 * data for the third item ends at an offset that matches the offset
3667 * where the data of the second items starts, and so on.
3668 * Therefore our source and destination start offsets for copy match the
3669 * offsets of the last items (highest slots).
3671 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3672 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3673 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3674 btrfs_release_path(dst_path);
3681 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3682 struct btrfs_inode *inode,
3683 struct btrfs_path *path,
3684 struct btrfs_path *dst_path,
3685 struct btrfs_log_ctx *ctx,
3686 u64 *last_old_dentry_offset)
3688 struct btrfs_root *log = inode->root->log_root;
3689 struct extent_buffer *src;
3690 const int nritems = btrfs_header_nritems(path->nodes[0]);
3691 const u64 ino = btrfs_ino(inode);
3692 bool last_found = false;
3693 int batch_start = 0;
3698 * We need to clone the leaf, release the read lock on it, and use the
3699 * clone before modifying the log tree. See the comment at copy_items()
3700 * about why we need to do this.
3702 src = btrfs_clone_extent_buffer(path->nodes[0]);
3707 btrfs_release_path(path);
3708 path->nodes[0] = src;
3711 for (; i < nritems; i++) {
3712 struct btrfs_dir_item *di;
3713 struct btrfs_key key;
3716 btrfs_item_key_to_cpu(src, &key, i);
3718 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3723 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3724 ctx->last_dir_item_offset = key.offset;
3727 * Skip ranges of items that consist only of dir item keys created
3728 * in past transactions. However if we find a gap, we must log a
3729 * dir index range item for that gap, so that index keys in that
3730 * gap are deleted during log replay.
3732 if (btrfs_dir_transid(src, di) < trans->transid) {
3733 if (key.offset > *last_old_dentry_offset + 1) {
3734 ret = insert_dir_log_key(trans, log, dst_path,
3735 ino, *last_old_dentry_offset + 1,
3741 *last_old_dentry_offset = key.offset;
3745 /* If we logged this dir index item before, we can skip it. */
3746 if (key.offset <= inode->last_dir_index_offset)
3750 * We must make sure that when we log a directory entry, the
3751 * corresponding inode, after log replay, has a matching link
3752 * count. For example:
3758 * xfs_io -c "fsync" mydir
3760 * <mount fs and log replay>
3762 * Would result in a fsync log that when replayed, our file inode
3763 * would have a link count of 1, but we get two directory entries
3764 * pointing to the same inode. After removing one of the names,
3765 * it would not be possible to remove the other name, which
3766 * resulted always in stale file handle errors, and would not be
3767 * possible to rmdir the parent directory, since its i_size could
3768 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3769 * resulting in -ENOTEMPTY errors.
3771 if (!ctx->log_new_dentries) {
3772 struct btrfs_key di_key;
3774 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3775 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3776 ctx->log_new_dentries = true;
3779 if (batch_size == 0)
3784 if (batch_size > 0) {
3787 ret = flush_dir_items_batch(trans, log, src, dst_path,
3788 batch_start, batch_size);
3793 return last_found ? 1 : 0;
3797 * log all the items included in the current transaction for a given
3798 * directory. This also creates the range items in the log tree required
3799 * to replay anything deleted before the fsync
3801 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3802 struct btrfs_inode *inode,
3803 struct btrfs_path *path,
3804 struct btrfs_path *dst_path,
3805 struct btrfs_log_ctx *ctx,
3806 u64 min_offset, u64 *last_offset_ret)
3808 struct btrfs_key min_key;
3809 struct btrfs_root *root = inode->root;
3810 struct btrfs_root *log = root->log_root;
3813 u64 last_old_dentry_offset = min_offset - 1;
3814 u64 last_offset = (u64)-1;
3815 u64 ino = btrfs_ino(inode);
3817 min_key.objectid = ino;
3818 min_key.type = BTRFS_DIR_INDEX_KEY;
3819 min_key.offset = min_offset;
3821 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3824 * we didn't find anything from this transaction, see if there
3825 * is anything at all
3827 if (ret != 0 || min_key.objectid != ino ||
3828 min_key.type != BTRFS_DIR_INDEX_KEY) {
3829 min_key.objectid = ino;
3830 min_key.type = BTRFS_DIR_INDEX_KEY;
3831 min_key.offset = (u64)-1;
3832 btrfs_release_path(path);
3833 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3835 btrfs_release_path(path);
3838 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3840 /* if ret == 0 there are items for this type,
3841 * create a range to tell us the last key of this type.
3842 * otherwise, there are no items in this directory after
3843 * *min_offset, and we create a range to indicate that.
3846 struct btrfs_key tmp;
3848 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3850 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3851 last_old_dentry_offset = tmp.offset;
3856 /* go backward to find any previous key */
3857 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3859 struct btrfs_key tmp;
3861 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3863 * The dir index key before the first one we found that needs to
3864 * be logged might be in a previous leaf, and there might be a
3865 * gap between these keys, meaning that we had deletions that
3866 * happened. So the key range item we log (key type
3867 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3868 * previous key's offset plus 1, so that those deletes are replayed.
3870 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3871 last_old_dentry_offset = tmp.offset;
3873 btrfs_release_path(path);
3876 * Find the first key from this transaction again. See the note for
3877 * log_new_dir_dentries, if we're logging a directory recursively we
3878 * won't be holding its i_mutex, which means we can modify the directory
3879 * while we're logging it. If we remove an entry between our first
3880 * search and this search we'll not find the key again and can just
3884 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3889 * we have a block from this transaction, log every item in it
3890 * from our directory
3893 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3894 &last_old_dentry_offset);
3900 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3903 * look ahead to the next item and see if it is also
3904 * from this directory and from this transaction
3906 ret = btrfs_next_leaf(root, path);
3909 last_offset = (u64)-1;
3914 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3915 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3916 last_offset = (u64)-1;
3919 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3921 * The next leaf was not changed in the current transaction
3922 * and has at least one dir index key.
3923 * We check for the next key because there might have been
3924 * one or more deletions between the last key we logged and
3925 * that next key. So the key range item we log (key type
3926 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3927 * offset minus 1, so that those deletes are replayed.
3929 last_offset = min_key.offset - 1;
3932 if (need_resched()) {
3933 btrfs_release_path(path);
3939 btrfs_release_path(path);
3940 btrfs_release_path(dst_path);
3943 *last_offset_ret = last_offset;
3945 * In case the leaf was changed in the current transaction but
3946 * all its dir items are from a past transaction, the last item
3947 * in the leaf is a dir item and there's no gap between that last
3948 * dir item and the first one on the next leaf (which did not
3949 * change in the current transaction), then we don't need to log
3950 * a range, last_old_dentry_offset is == to last_offset.
3952 ASSERT(last_old_dentry_offset <= last_offset);
3953 if (last_old_dentry_offset < last_offset) {
3954 ret = insert_dir_log_key(trans, log, path, ino,
3955 last_old_dentry_offset + 1,
3965 * If the inode was logged before and it was evicted, then its
3966 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3967 * key offset. If that's the case, search for it and update the inode. This
3968 * is to avoid lookups in the log tree every time we try to insert a dir index
3969 * key from a leaf changed in the current transaction, and to allow us to always
3970 * do batch insertions of dir index keys.
3972 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3973 struct btrfs_path *path,
3974 const struct btrfs_log_ctx *ctx)
3976 const u64 ino = btrfs_ino(inode);
3977 struct btrfs_key key;
3980 lockdep_assert_held(&inode->log_mutex);
3982 if (inode->last_dir_index_offset != (u64)-1)
3985 if (!ctx->logged_before) {
3986 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
3991 key.type = BTRFS_DIR_INDEX_KEY;
3992 key.offset = (u64)-1;
3994 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3996 * An error happened or we actually have an index key with an offset
3997 * value of (u64)-1. Bail out, we're done.
4003 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
4006 * No dir index items, bail out and leave last_dir_index_offset with
4007 * the value right before the first valid index value.
4009 if (path->slots[0] == 0)
4013 * btrfs_search_slot() left us at one slot beyond the slot with the last
4014 * index key, or beyond the last key of the directory that is not an
4015 * index key. If we have an index key before, set last_dir_index_offset
4016 * to its offset value, otherwise leave it with a value right before the
4017 * first valid index value, as it means we have an empty directory.
4019 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4020 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
4021 inode->last_dir_index_offset = key.offset;
4024 btrfs_release_path(path);
4030 * logging directories is very similar to logging inodes, We find all the items
4031 * from the current transaction and write them to the log.
4033 * The recovery code scans the directory in the subvolume, and if it finds a
4034 * key in the range logged that is not present in the log tree, then it means
4035 * that dir entry was unlinked during the transaction.
4037 * In order for that scan to work, we must include one key smaller than
4038 * the smallest logged by this transaction and one key larger than the largest
4039 * key logged by this transaction.
4041 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4042 struct btrfs_inode *inode,
4043 struct btrfs_path *path,
4044 struct btrfs_path *dst_path,
4045 struct btrfs_log_ctx *ctx)
4051 ret = update_last_dir_index_offset(inode, path, ctx);
4055 min_key = BTRFS_DIR_START_INDEX;
4057 ctx->last_dir_item_offset = inode->last_dir_index_offset;
4060 ret = log_dir_items(trans, inode, path, dst_path,
4061 ctx, min_key, &max_key);
4064 if (max_key == (u64)-1)
4066 min_key = max_key + 1;
4069 inode->last_dir_index_offset = ctx->last_dir_item_offset;
4075 * a helper function to drop items from the log before we relog an
4076 * inode. max_key_type indicates the highest item type to remove.
4077 * This cannot be run for file data extents because it does not
4078 * free the extents they point to.
4080 static int drop_inode_items(struct btrfs_trans_handle *trans,
4081 struct btrfs_root *log,
4082 struct btrfs_path *path,
4083 struct btrfs_inode *inode,
4087 struct btrfs_key key;
4088 struct btrfs_key found_key;
4091 key.objectid = btrfs_ino(inode);
4092 key.type = max_key_type;
4093 key.offset = (u64)-1;
4096 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4097 BUG_ON(ret == 0); /* Logic error */
4101 if (path->slots[0] == 0)
4105 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4108 if (found_key.objectid != key.objectid)
4111 found_key.offset = 0;
4113 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
4117 ret = btrfs_del_items(trans, log, path, start_slot,
4118 path->slots[0] - start_slot + 1);
4120 * If start slot isn't 0 then we don't need to re-search, we've
4121 * found the last guy with the objectid in this tree.
4123 if (ret || start_slot != 0)
4125 btrfs_release_path(path);
4127 btrfs_release_path(path);
4133 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4134 struct btrfs_root *log_root,
4135 struct btrfs_inode *inode,
4136 u64 new_size, u32 min_type)
4138 struct btrfs_truncate_control control = {
4139 .new_size = new_size,
4140 .ino = btrfs_ino(inode),
4141 .min_type = min_type,
4142 .skip_ref_updates = true,
4145 return btrfs_truncate_inode_items(trans, log_root, &control);
4148 static void fill_inode_item(struct btrfs_trans_handle *trans,
4149 struct extent_buffer *leaf,
4150 struct btrfs_inode_item *item,
4151 struct inode *inode, int log_inode_only,
4154 struct btrfs_map_token token;
4157 btrfs_init_map_token(&token, leaf);
4159 if (log_inode_only) {
4160 /* set the generation to zero so the recover code
4161 * can tell the difference between an logging
4162 * just to say 'this inode exists' and a logging
4163 * to say 'update this inode with these values'
4165 btrfs_set_token_inode_generation(&token, item, 0);
4166 btrfs_set_token_inode_size(&token, item, logged_isize);
4168 btrfs_set_token_inode_generation(&token, item,
4169 BTRFS_I(inode)->generation);
4170 btrfs_set_token_inode_size(&token, item, inode->i_size);
4173 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4174 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4175 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4176 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4178 btrfs_set_token_timespec_sec(&token, &item->atime,
4179 inode->i_atime.tv_sec);
4180 btrfs_set_token_timespec_nsec(&token, &item->atime,
4181 inode->i_atime.tv_nsec);
4183 btrfs_set_token_timespec_sec(&token, &item->mtime,
4184 inode->i_mtime.tv_sec);
4185 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4186 inode->i_mtime.tv_nsec);
4188 btrfs_set_token_timespec_sec(&token, &item->ctime,
4189 inode->i_ctime.tv_sec);
4190 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4191 inode->i_ctime.tv_nsec);
4194 * We do not need to set the nbytes field, in fact during a fast fsync
4195 * its value may not even be correct, since a fast fsync does not wait
4196 * for ordered extent completion, which is where we update nbytes, it
4197 * only waits for writeback to complete. During log replay as we find
4198 * file extent items and replay them, we adjust the nbytes field of the
4199 * inode item in subvolume tree as needed (see overwrite_item()).
4202 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4203 btrfs_set_token_inode_transid(&token, item, trans->transid);
4204 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4205 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4206 BTRFS_I(inode)->ro_flags);
4207 btrfs_set_token_inode_flags(&token, item, flags);
4208 btrfs_set_token_inode_block_group(&token, item, 0);
4211 static int log_inode_item(struct btrfs_trans_handle *trans,
4212 struct btrfs_root *log, struct btrfs_path *path,
4213 struct btrfs_inode *inode, bool inode_item_dropped)
4215 struct btrfs_inode_item *inode_item;
4219 * If we are doing a fast fsync and the inode was logged before in the
4220 * current transaction, then we know the inode was previously logged and
4221 * it exists in the log tree. For performance reasons, in this case use
4222 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4223 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4224 * contention in case there are concurrent fsyncs for other inodes of the
4225 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4226 * already exists can also result in unnecessarily splitting a leaf.
4228 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4229 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4235 * This means it is the first fsync in the current transaction,
4236 * so the inode item is not in the log and we need to insert it.
4237 * We can never get -EEXIST because we are only called for a fast
4238 * fsync and in case an inode eviction happens after the inode was
4239 * logged before in the current transaction, when we load again
4240 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4241 * flags and set ->logged_trans to 0.
4243 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4244 sizeof(*inode_item));
4245 ASSERT(ret != -EEXIST);
4249 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4250 struct btrfs_inode_item);
4251 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4253 btrfs_release_path(path);
4257 static int log_csums(struct btrfs_trans_handle *trans,
4258 struct btrfs_inode *inode,
4259 struct btrfs_root *log_root,
4260 struct btrfs_ordered_sum *sums)
4262 const u64 lock_end = sums->bytenr + sums->len - 1;
4263 struct extent_state *cached_state = NULL;
4267 * If this inode was not used for reflink operations in the current
4268 * transaction with new extents, then do the fast path, no need to
4269 * worry about logging checksum items with overlapping ranges.
4271 if (inode->last_reflink_trans < trans->transid)
4272 return btrfs_csum_file_blocks(trans, log_root, sums);
4275 * Serialize logging for checksums. This is to avoid racing with the
4276 * same checksum being logged by another task that is logging another
4277 * file which happens to refer to the same extent as well. Such races
4278 * can leave checksum items in the log with overlapping ranges.
4280 ret = lock_extent(&log_root->log_csum_range, sums->bytenr, lock_end,
4285 * Due to extent cloning, we might have logged a csum item that covers a
4286 * subrange of a cloned extent, and later we can end up logging a csum
4287 * item for a larger subrange of the same extent or the entire range.
4288 * This would leave csum items in the log tree that cover the same range
4289 * and break the searches for checksums in the log tree, resulting in
4290 * some checksums missing in the fs/subvolume tree. So just delete (or
4291 * trim and adjust) any existing csum items in the log for this range.
4293 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4295 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4297 unlock_extent(&log_root->log_csum_range, sums->bytenr, lock_end,
4303 static noinline int copy_items(struct btrfs_trans_handle *trans,
4304 struct btrfs_inode *inode,
4305 struct btrfs_path *dst_path,
4306 struct btrfs_path *src_path,
4307 int start_slot, int nr, int inode_only,
4310 struct btrfs_root *log = inode->root->log_root;
4311 struct btrfs_file_extent_item *extent;
4312 struct extent_buffer *src;
4314 struct btrfs_key *ins_keys;
4316 struct btrfs_item_batch batch;
4320 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4321 const u64 i_size = i_size_read(&inode->vfs_inode);
4324 * To keep lockdep happy and avoid deadlocks, clone the source leaf and
4325 * use the clone. This is because otherwise we would be changing the log
4326 * tree, to insert items from the subvolume tree or insert csum items,
4327 * while holding a read lock on a leaf from the subvolume tree, which
4328 * creates a nasty lock dependency when COWing log tree nodes/leaves:
4330 * 1) Modifying the log tree triggers an extent buffer allocation while
4331 * holding a write lock on a parent extent buffer from the log tree.
4332 * Allocating the pages for an extent buffer, or the extent buffer
4333 * struct, can trigger inode eviction and finally the inode eviction
4334 * will trigger a release/remove of a delayed node, which requires
4335 * taking the delayed node's mutex;
4337 * 2) Allocating a metadata extent for a log tree can trigger the async
4338 * reclaim thread and make us wait for it to release enough space and
4339 * unblock our reservation ticket. The reclaim thread can start
4340 * flushing delayed items, and that in turn results in the need to
4341 * lock delayed node mutexes and in the need to write lock extent
4342 * buffers of a subvolume tree - all this while holding a write lock
4343 * on the parent extent buffer in the log tree.
4345 * So one task in scenario 1) running in parallel with another task in
4346 * scenario 2) could lead to a deadlock, one wanting to lock a delayed
4347 * node mutex while having a read lock on a leaf from the subvolume,
4348 * while the other is holding the delayed node's mutex and wants to
4349 * write lock the same subvolume leaf for flushing delayed items.
4351 src = btrfs_clone_extent_buffer(src_path->nodes[0]);
4355 i = src_path->slots[0];
4356 btrfs_release_path(src_path);
4357 src_path->nodes[0] = src;
4358 src_path->slots[0] = i;
4360 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4361 nr * sizeof(u32), GFP_NOFS);
4365 ins_sizes = (u32 *)ins_data;
4366 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4367 batch.keys = ins_keys;
4368 batch.data_sizes = ins_sizes;
4369 batch.total_data_size = 0;
4373 for (i = 0; i < nr; i++) {
4374 const int src_slot = start_slot + i;
4375 struct btrfs_root *csum_root;
4376 struct btrfs_ordered_sum *sums;
4377 struct btrfs_ordered_sum *sums_next;
4378 LIST_HEAD(ordered_sums);
4382 u64 extent_num_bytes;
4385 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4387 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4390 extent = btrfs_item_ptr(src, src_slot,
4391 struct btrfs_file_extent_item);
4393 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4397 * Don't copy extents from past generations. That would make us
4398 * log a lot more metadata for common cases like doing only a
4399 * few random writes into a file and then fsync it for the first
4400 * time or after the full sync flag is set on the inode. We can
4401 * get leaves full of extent items, most of which are from past
4402 * generations, so we can skip them - as long as the inode has
4403 * not been the target of a reflink operation in this transaction,
4404 * as in that case it might have had file extent items with old
4405 * generations copied into it. We also must always log prealloc
4406 * extents that start at or beyond eof, otherwise we would lose
4407 * them on log replay.
4409 if (is_old_extent &&
4410 ins_keys[dst_index].offset < i_size &&
4411 inode->last_reflink_trans < trans->transid)
4417 /* Only regular extents have checksums. */
4418 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4422 * If it's an extent created in a past transaction, then its
4423 * checksums are already accessible from the committed csum tree,
4424 * no need to log them.
4429 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4430 /* If it's an explicit hole, there are no checksums. */
4431 if (disk_bytenr == 0)
4434 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4436 if (btrfs_file_extent_compression(src, extent)) {
4438 extent_num_bytes = disk_num_bytes;
4440 extent_offset = btrfs_file_extent_offset(src, extent);
4441 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4444 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4445 disk_bytenr += extent_offset;
4446 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4447 disk_bytenr + extent_num_bytes - 1,
4448 &ordered_sums, 0, false);
4452 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4454 ret = log_csums(trans, inode, log, sums);
4455 list_del(&sums->list);
4462 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4463 batch.total_data_size += ins_sizes[dst_index];
4469 * We have a leaf full of old extent items that don't need to be logged,
4470 * so we don't need to do anything.
4475 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4480 for (i = 0; i < nr; i++) {
4481 const int src_slot = start_slot + i;
4482 const int dst_slot = dst_path->slots[0] + dst_index;
4483 struct btrfs_key key;
4484 unsigned long src_offset;
4485 unsigned long dst_offset;
4488 * We're done, all the remaining items in the source leaf
4489 * correspond to old file extent items.
4491 if (dst_index >= batch.nr)
4494 btrfs_item_key_to_cpu(src, &key, src_slot);
4496 if (key.type != BTRFS_EXTENT_DATA_KEY)
4499 extent = btrfs_item_ptr(src, src_slot,
4500 struct btrfs_file_extent_item);
4502 /* See the comment in the previous loop, same logic. */
4503 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4504 key.offset < i_size &&
4505 inode->last_reflink_trans < trans->transid)
4509 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4510 src_offset = btrfs_item_ptr_offset(src, src_slot);
4512 if (key.type == BTRFS_INODE_ITEM_KEY) {
4513 struct btrfs_inode_item *inode_item;
4515 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4516 struct btrfs_inode_item);
4517 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4519 inode_only == LOG_INODE_EXISTS,
4522 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4523 src_offset, ins_sizes[dst_index]);
4529 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4530 btrfs_release_path(dst_path);
4537 static int extent_cmp(void *priv, const struct list_head *a,
4538 const struct list_head *b)
4540 const struct extent_map *em1, *em2;
4542 em1 = list_entry(a, struct extent_map, list);
4543 em2 = list_entry(b, struct extent_map, list);
4545 if (em1->start < em2->start)
4547 else if (em1->start > em2->start)
4552 static int log_extent_csums(struct btrfs_trans_handle *trans,
4553 struct btrfs_inode *inode,
4554 struct btrfs_root *log_root,
4555 const struct extent_map *em,
4556 struct btrfs_log_ctx *ctx)
4558 struct btrfs_ordered_extent *ordered;
4559 struct btrfs_root *csum_root;
4562 u64 mod_start = em->mod_start;
4563 u64 mod_len = em->mod_len;
4564 LIST_HEAD(ordered_sums);
4567 if (inode->flags & BTRFS_INODE_NODATASUM ||
4568 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4569 em->block_start == EXTENT_MAP_HOLE)
4572 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4573 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4574 const u64 mod_end = mod_start + mod_len;
4575 struct btrfs_ordered_sum *sums;
4580 if (ordered_end <= mod_start)
4582 if (mod_end <= ordered->file_offset)
4586 * We are going to copy all the csums on this ordered extent, so
4587 * go ahead and adjust mod_start and mod_len in case this ordered
4588 * extent has already been logged.
4590 if (ordered->file_offset > mod_start) {
4591 if (ordered_end >= mod_end)
4592 mod_len = ordered->file_offset - mod_start;
4594 * If we have this case
4596 * |--------- logged extent ---------|
4597 * |----- ordered extent ----|
4599 * Just don't mess with mod_start and mod_len, we'll
4600 * just end up logging more csums than we need and it
4604 if (ordered_end < mod_end) {
4605 mod_len = mod_end - ordered_end;
4606 mod_start = ordered_end;
4613 * To keep us from looping for the above case of an ordered
4614 * extent that falls inside of the logged extent.
4616 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4619 list_for_each_entry(sums, &ordered->list, list) {
4620 ret = log_csums(trans, inode, log_root, sums);
4626 /* We're done, found all csums in the ordered extents. */
4630 /* If we're compressed we have to save the entire range of csums. */
4631 if (em->compress_type) {
4633 csum_len = max(em->block_len, em->orig_block_len);
4635 csum_offset = mod_start - em->start;
4639 /* block start is already adjusted for the file extent offset. */
4640 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4641 ret = btrfs_lookup_csums_list(csum_root, em->block_start + csum_offset,
4642 em->block_start + csum_offset +
4643 csum_len - 1, &ordered_sums, 0, false);
4647 while (!list_empty(&ordered_sums)) {
4648 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4649 struct btrfs_ordered_sum,
4652 ret = log_csums(trans, inode, log_root, sums);
4653 list_del(&sums->list);
4660 static int log_one_extent(struct btrfs_trans_handle *trans,
4661 struct btrfs_inode *inode,
4662 const struct extent_map *em,
4663 struct btrfs_path *path,
4664 struct btrfs_log_ctx *ctx)
4666 struct btrfs_drop_extents_args drop_args = { 0 };
4667 struct btrfs_root *log = inode->root->log_root;
4668 struct btrfs_file_extent_item fi = { 0 };
4669 struct extent_buffer *leaf;
4670 struct btrfs_key key;
4671 u64 extent_offset = em->start - em->orig_start;
4675 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4676 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4677 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4679 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4681 block_len = max(em->block_len, em->orig_block_len);
4682 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4683 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4684 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4685 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4686 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4688 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4691 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4692 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4693 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4694 btrfs_set_stack_file_extent_compression(&fi, em->compress_type);
4696 ret = log_extent_csums(trans, inode, log, em, ctx);
4701 * If this is the first time we are logging the inode in the current
4702 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4703 * because it does a deletion search, which always acquires write locks
4704 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4705 * but also adds significant contention in a log tree, since log trees
4706 * are small, with a root at level 2 or 3 at most, due to their short
4709 if (ctx->logged_before) {
4710 drop_args.path = path;
4711 drop_args.start = em->start;
4712 drop_args.end = em->start + em->len;
4713 drop_args.replace_extent = true;
4714 drop_args.extent_item_size = sizeof(fi);
4715 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4720 if (!drop_args.extent_inserted) {
4721 key.objectid = btrfs_ino(inode);
4722 key.type = BTRFS_EXTENT_DATA_KEY;
4723 key.offset = em->start;
4725 ret = btrfs_insert_empty_item(trans, log, path, &key,
4730 leaf = path->nodes[0];
4731 write_extent_buffer(leaf, &fi,
4732 btrfs_item_ptr_offset(leaf, path->slots[0]),
4734 btrfs_mark_buffer_dirty(leaf);
4736 btrfs_release_path(path);
4742 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4743 * lose them after doing a full/fast fsync and replaying the log. We scan the
4744 * subvolume's root instead of iterating the inode's extent map tree because
4745 * otherwise we can log incorrect extent items based on extent map conversion.
4746 * That can happen due to the fact that extent maps are merged when they
4747 * are not in the extent map tree's list of modified extents.
4749 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4750 struct btrfs_inode *inode,
4751 struct btrfs_path *path)
4753 struct btrfs_root *root = inode->root;
4754 struct btrfs_key key;
4755 const u64 i_size = i_size_read(&inode->vfs_inode);
4756 const u64 ino = btrfs_ino(inode);
4757 struct btrfs_path *dst_path = NULL;
4758 bool dropped_extents = false;
4759 u64 truncate_offset = i_size;
4760 struct extent_buffer *leaf;
4766 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4770 key.type = BTRFS_EXTENT_DATA_KEY;
4771 key.offset = i_size;
4772 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4777 * We must check if there is a prealloc extent that starts before the
4778 * i_size and crosses the i_size boundary. This is to ensure later we
4779 * truncate down to the end of that extent and not to the i_size, as
4780 * otherwise we end up losing part of the prealloc extent after a log
4781 * replay and with an implicit hole if there is another prealloc extent
4782 * that starts at an offset beyond i_size.
4784 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4789 struct btrfs_file_extent_item *ei;
4791 leaf = path->nodes[0];
4792 slot = path->slots[0];
4793 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4795 if (btrfs_file_extent_type(leaf, ei) ==
4796 BTRFS_FILE_EXTENT_PREALLOC) {
4799 btrfs_item_key_to_cpu(leaf, &key, slot);
4800 extent_end = key.offset +
4801 btrfs_file_extent_num_bytes(leaf, ei);
4803 if (extent_end > i_size)
4804 truncate_offset = extent_end;
4811 leaf = path->nodes[0];
4812 slot = path->slots[0];
4814 if (slot >= btrfs_header_nritems(leaf)) {
4816 ret = copy_items(trans, inode, dst_path, path,
4817 start_slot, ins_nr, 1, 0);
4822 ret = btrfs_next_leaf(root, path);
4832 btrfs_item_key_to_cpu(leaf, &key, slot);
4833 if (key.objectid > ino)
4835 if (WARN_ON_ONCE(key.objectid < ino) ||
4836 key.type < BTRFS_EXTENT_DATA_KEY ||
4837 key.offset < i_size) {
4841 if (!dropped_extents) {
4843 * Avoid logging extent items logged in past fsync calls
4844 * and leading to duplicate keys in the log tree.
4846 ret = truncate_inode_items(trans, root->log_root, inode,
4848 BTRFS_EXTENT_DATA_KEY);
4851 dropped_extents = true;
4858 dst_path = btrfs_alloc_path();
4866 ret = copy_items(trans, inode, dst_path, path,
4867 start_slot, ins_nr, 1, 0);
4869 btrfs_release_path(path);
4870 btrfs_free_path(dst_path);
4874 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4875 struct btrfs_inode *inode,
4876 struct btrfs_path *path,
4877 struct btrfs_log_ctx *ctx)
4879 struct btrfs_ordered_extent *ordered;
4880 struct btrfs_ordered_extent *tmp;
4881 struct extent_map *em, *n;
4882 struct list_head extents;
4883 struct extent_map_tree *tree = &inode->extent_tree;
4887 INIT_LIST_HEAD(&extents);
4889 write_lock(&tree->lock);
4891 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4892 list_del_init(&em->list);
4894 * Just an arbitrary number, this can be really CPU intensive
4895 * once we start getting a lot of extents, and really once we
4896 * have a bunch of extents we just want to commit since it will
4899 if (++num > 32768) {
4900 list_del_init(&tree->modified_extents);
4905 if (em->generation < trans->transid)
4908 /* We log prealloc extents beyond eof later. */
4909 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4910 em->start >= i_size_read(&inode->vfs_inode))
4913 /* Need a ref to keep it from getting evicted from cache */
4914 refcount_inc(&em->refs);
4915 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4916 list_add_tail(&em->list, &extents);
4920 list_sort(NULL, &extents, extent_cmp);
4922 while (!list_empty(&extents)) {
4923 em = list_entry(extents.next, struct extent_map, list);
4925 list_del_init(&em->list);
4928 * If we had an error we just need to delete everybody from our
4932 clear_em_logging(tree, em);
4933 free_extent_map(em);
4937 write_unlock(&tree->lock);
4939 ret = log_one_extent(trans, inode, em, path, ctx);
4940 write_lock(&tree->lock);
4941 clear_em_logging(tree, em);
4942 free_extent_map(em);
4944 WARN_ON(!list_empty(&extents));
4945 write_unlock(&tree->lock);
4948 ret = btrfs_log_prealloc_extents(trans, inode, path);
4953 * We have logged all extents successfully, now make sure the commit of
4954 * the current transaction waits for the ordered extents to complete
4955 * before it commits and wipes out the log trees, otherwise we would
4956 * lose data if an ordered extents completes after the transaction
4957 * commits and a power failure happens after the transaction commit.
4959 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4960 list_del_init(&ordered->log_list);
4961 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4963 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4964 spin_lock_irq(&inode->ordered_tree.lock);
4965 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4966 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4967 atomic_inc(&trans->transaction->pending_ordered);
4969 spin_unlock_irq(&inode->ordered_tree.lock);
4971 btrfs_put_ordered_extent(ordered);
4977 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4978 struct btrfs_path *path, u64 *size_ret)
4980 struct btrfs_key key;
4983 key.objectid = btrfs_ino(inode);
4984 key.type = BTRFS_INODE_ITEM_KEY;
4987 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4990 } else if (ret > 0) {
4993 struct btrfs_inode_item *item;
4995 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4996 struct btrfs_inode_item);
4997 *size_ret = btrfs_inode_size(path->nodes[0], item);
4999 * If the in-memory inode's i_size is smaller then the inode
5000 * size stored in the btree, return the inode's i_size, so
5001 * that we get a correct inode size after replaying the log
5002 * when before a power failure we had a shrinking truncate
5003 * followed by addition of a new name (rename / new hard link).
5004 * Otherwise return the inode size from the btree, to avoid
5005 * data loss when replaying a log due to previously doing a
5006 * write that expands the inode's size and logging a new name
5007 * immediately after.
5009 if (*size_ret > inode->vfs_inode.i_size)
5010 *size_ret = inode->vfs_inode.i_size;
5013 btrfs_release_path(path);
5018 * At the moment we always log all xattrs. This is to figure out at log replay
5019 * time which xattrs must have their deletion replayed. If a xattr is missing
5020 * in the log tree and exists in the fs/subvol tree, we delete it. This is
5021 * because if a xattr is deleted, the inode is fsynced and a power failure
5022 * happens, causing the log to be replayed the next time the fs is mounted,
5023 * we want the xattr to not exist anymore (same behaviour as other filesystems
5024 * with a journal, ext3/4, xfs, f2fs, etc).
5026 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
5027 struct btrfs_inode *inode,
5028 struct btrfs_path *path,
5029 struct btrfs_path *dst_path)
5031 struct btrfs_root *root = inode->root;
5033 struct btrfs_key key;
5034 const u64 ino = btrfs_ino(inode);
5037 bool found_xattrs = false;
5039 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
5043 key.type = BTRFS_XATTR_ITEM_KEY;
5046 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5051 int slot = path->slots[0];
5052 struct extent_buffer *leaf = path->nodes[0];
5053 int nritems = btrfs_header_nritems(leaf);
5055 if (slot >= nritems) {
5057 ret = copy_items(trans, inode, dst_path, path,
5058 start_slot, ins_nr, 1, 0);
5063 ret = btrfs_next_leaf(root, path);
5071 btrfs_item_key_to_cpu(leaf, &key, slot);
5072 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5079 found_xattrs = true;
5083 ret = copy_items(trans, inode, dst_path, path,
5084 start_slot, ins_nr, 1, 0);
5090 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5096 * When using the NO_HOLES feature if we punched a hole that causes the
5097 * deletion of entire leafs or all the extent items of the first leaf (the one
5098 * that contains the inode item and references) we may end up not processing
5099 * any extents, because there are no leafs with a generation matching the
5100 * current transaction that have extent items for our inode. So we need to find
5101 * if any holes exist and then log them. We also need to log holes after any
5102 * truncate operation that changes the inode's size.
5104 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5105 struct btrfs_inode *inode,
5106 struct btrfs_path *path)
5108 struct btrfs_root *root = inode->root;
5109 struct btrfs_fs_info *fs_info = root->fs_info;
5110 struct btrfs_key key;
5111 const u64 ino = btrfs_ino(inode);
5112 const u64 i_size = i_size_read(&inode->vfs_inode);
5113 u64 prev_extent_end = 0;
5116 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5120 key.type = BTRFS_EXTENT_DATA_KEY;
5123 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5128 struct extent_buffer *leaf = path->nodes[0];
5130 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5131 ret = btrfs_next_leaf(root, path);
5138 leaf = path->nodes[0];
5141 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5142 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5145 /* We have a hole, log it. */
5146 if (prev_extent_end < key.offset) {
5147 const u64 hole_len = key.offset - prev_extent_end;
5150 * Release the path to avoid deadlocks with other code
5151 * paths that search the root while holding locks on
5152 * leafs from the log root.
5154 btrfs_release_path(path);
5155 ret = btrfs_insert_hole_extent(trans, root->log_root,
5156 ino, prev_extent_end,
5162 * Search for the same key again in the root. Since it's
5163 * an extent item and we are holding the inode lock, the
5164 * key must still exist. If it doesn't just emit warning
5165 * and return an error to fall back to a transaction
5168 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5171 if (WARN_ON(ret > 0))
5173 leaf = path->nodes[0];
5176 prev_extent_end = btrfs_file_extent_end(path);
5181 if (prev_extent_end < i_size) {
5184 btrfs_release_path(path);
5185 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5186 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5187 prev_extent_end, hole_len);
5196 * When we are logging a new inode X, check if it doesn't have a reference that
5197 * matches the reference from some other inode Y created in a past transaction
5198 * and that was renamed in the current transaction. If we don't do this, then at
5199 * log replay time we can lose inode Y (and all its files if it's a directory):
5202 * echo "hello world" > /mnt/x/foobar
5205 * mkdir /mnt/x # or touch /mnt/x
5206 * xfs_io -c fsync /mnt/x
5208 * mount fs, trigger log replay
5210 * After the log replay procedure, we would lose the first directory and all its
5211 * files (file foobar).
5212 * For the case where inode Y is not a directory we simply end up losing it:
5214 * echo "123" > /mnt/foo
5216 * mv /mnt/foo /mnt/bar
5217 * echo "abc" > /mnt/foo
5218 * xfs_io -c fsync /mnt/foo
5221 * We also need this for cases where a snapshot entry is replaced by some other
5222 * entry (file or directory) otherwise we end up with an unreplayable log due to
5223 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5224 * if it were a regular entry:
5227 * btrfs subvolume snapshot /mnt /mnt/x/snap
5228 * btrfs subvolume delete /mnt/x/snap
5231 * fsync /mnt/x or fsync some new file inside it
5234 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5235 * the same transaction.
5237 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5239 const struct btrfs_key *key,
5240 struct btrfs_inode *inode,
5241 u64 *other_ino, u64 *other_parent)
5244 struct btrfs_path *search_path;
5247 u32 item_size = btrfs_item_size(eb, slot);
5249 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5251 search_path = btrfs_alloc_path();
5254 search_path->search_commit_root = 1;
5255 search_path->skip_locking = 1;
5257 while (cur_offset < item_size) {
5261 unsigned long name_ptr;
5262 struct btrfs_dir_item *di;
5263 struct fscrypt_str name_str;
5265 if (key->type == BTRFS_INODE_REF_KEY) {
5266 struct btrfs_inode_ref *iref;
5268 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5269 parent = key->offset;
5270 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5271 name_ptr = (unsigned long)(iref + 1);
5272 this_len = sizeof(*iref) + this_name_len;
5274 struct btrfs_inode_extref *extref;
5276 extref = (struct btrfs_inode_extref *)(ptr +
5278 parent = btrfs_inode_extref_parent(eb, extref);
5279 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5280 name_ptr = (unsigned long)&extref->name;
5281 this_len = sizeof(*extref) + this_name_len;
5284 if (this_name_len > name_len) {
5287 new_name = krealloc(name, this_name_len, GFP_NOFS);
5292 name_len = this_name_len;
5296 read_extent_buffer(eb, name, name_ptr, this_name_len);
5298 name_str.name = name;
5299 name_str.len = this_name_len;
5300 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5301 parent, &name_str, 0);
5302 if (di && !IS_ERR(di)) {
5303 struct btrfs_key di_key;
5305 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5307 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5308 if (di_key.objectid != key->objectid) {
5310 *other_ino = di_key.objectid;
5311 *other_parent = parent;
5319 } else if (IS_ERR(di)) {
5323 btrfs_release_path(search_path);
5325 cur_offset += this_len;
5329 btrfs_free_path(search_path);
5335 * Check if we need to log an inode. This is used in contexts where while
5336 * logging an inode we need to log another inode (either that it exists or in
5337 * full mode). This is used instead of btrfs_inode_in_log() because the later
5338 * requires the inode to be in the log and have the log transaction committed,
5339 * while here we do not care if the log transaction was already committed - our
5340 * caller will commit the log later - and we want to avoid logging an inode
5341 * multiple times when multiple tasks have joined the same log transaction.
5343 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5344 const struct btrfs_inode *inode)
5347 * If a directory was not modified, no dentries added or removed, we can
5348 * and should avoid logging it.
5350 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5354 * If this inode does not have new/updated/deleted xattrs since the last
5355 * time it was logged and is flagged as logged in the current transaction,
5356 * we can skip logging it. As for new/deleted names, those are updated in
5357 * the log by link/unlink/rename operations.
5358 * In case the inode was logged and then evicted and reloaded, its
5359 * logged_trans will be 0, in which case we have to fully log it since
5360 * logged_trans is a transient field, not persisted.
5362 if (inode->logged_trans == trans->transid &&
5363 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5369 struct btrfs_dir_list {
5371 struct list_head list;
5375 * Log the inodes of the new dentries of a directory.
5376 * See process_dir_items_leaf() for details about why it is needed.
5377 * This is a recursive operation - if an existing dentry corresponds to a
5378 * directory, that directory's new entries are logged too (same behaviour as
5379 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5380 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5381 * complains about the following circular lock dependency / possible deadlock:
5385 * lock(&type->i_mutex_dir_key#3/2);
5386 * lock(sb_internal#2);
5387 * lock(&type->i_mutex_dir_key#3/2);
5388 * lock(&sb->s_type->i_mutex_key#14);
5390 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5391 * sb_start_intwrite() in btrfs_start_transaction().
5392 * Not acquiring the VFS lock of the inodes is still safe because:
5394 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5395 * that while logging the inode new references (names) are added or removed
5396 * from the inode, leaving the logged inode item with a link count that does
5397 * not match the number of logged inode reference items. This is fine because
5398 * at log replay time we compute the real number of links and correct the
5399 * link count in the inode item (see replay_one_buffer() and
5400 * link_to_fixup_dir());
5402 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5403 * while logging the inode's items new index items (key type
5404 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5405 * has a size that doesn't match the sum of the lengths of all the logged
5406 * names - this is ok, not a problem, because at log replay time we set the
5407 * directory's i_size to the correct value (see replay_one_name() and
5408 * do_overwrite_item()).
5410 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5411 struct btrfs_inode *start_inode,
5412 struct btrfs_log_ctx *ctx)
5414 struct btrfs_root *root = start_inode->root;
5415 struct btrfs_fs_info *fs_info = root->fs_info;
5416 struct btrfs_path *path;
5417 LIST_HEAD(dir_list);
5418 struct btrfs_dir_list *dir_elem;
5419 u64 ino = btrfs_ino(start_inode);
5423 * If we are logging a new name, as part of a link or rename operation,
5424 * don't bother logging new dentries, as we just want to log the names
5425 * of an inode and that any new parents exist.
5427 if (ctx->logging_new_name)
5430 path = btrfs_alloc_path();
5435 struct extent_buffer *leaf;
5436 struct btrfs_key min_key;
5437 bool continue_curr_inode = true;
5441 min_key.objectid = ino;
5442 min_key.type = BTRFS_DIR_INDEX_KEY;
5445 btrfs_release_path(path);
5446 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
5449 } else if (ret > 0) {
5454 leaf = path->nodes[0];
5455 nritems = btrfs_header_nritems(leaf);
5456 for (i = path->slots[0]; i < nritems; i++) {
5457 struct btrfs_dir_item *di;
5458 struct btrfs_key di_key;
5459 struct inode *di_inode;
5460 int log_mode = LOG_INODE_EXISTS;
5463 btrfs_item_key_to_cpu(leaf, &min_key, i);
5464 if (min_key.objectid != ino ||
5465 min_key.type != BTRFS_DIR_INDEX_KEY) {
5466 continue_curr_inode = false;
5470 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5471 type = btrfs_dir_ftype(leaf, di);
5472 if (btrfs_dir_transid(leaf, di) < trans->transid)
5474 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5475 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5478 btrfs_release_path(path);
5479 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5480 if (IS_ERR(di_inode)) {
5481 ret = PTR_ERR(di_inode);
5485 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5486 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5490 ctx->log_new_dentries = false;
5491 if (type == BTRFS_FT_DIR)
5492 log_mode = LOG_INODE_ALL;
5493 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5495 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5498 if (ctx->log_new_dentries) {
5499 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5504 dir_elem->ino = di_key.objectid;
5505 list_add_tail(&dir_elem->list, &dir_list);
5510 if (continue_curr_inode && min_key.offset < (u64)-1) {
5516 if (list_empty(&dir_list))
5519 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5520 ino = dir_elem->ino;
5521 list_del(&dir_elem->list);
5525 btrfs_free_path(path);
5527 struct btrfs_dir_list *next;
5529 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5536 struct btrfs_ino_list {
5539 struct list_head list;
5542 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5544 struct btrfs_ino_list *curr;
5545 struct btrfs_ino_list *next;
5547 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5548 list_del(&curr->list);
5553 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5554 struct btrfs_path *path)
5556 struct btrfs_key key;
5560 key.type = BTRFS_INODE_ITEM_KEY;
5563 path->search_commit_root = 1;
5564 path->skip_locking = 1;
5566 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5567 if (WARN_ON_ONCE(ret > 0)) {
5569 * We have previously found the inode through the commit root
5570 * so this should not happen. If it does, just error out and
5571 * fallback to a transaction commit.
5574 } else if (ret == 0) {
5575 struct btrfs_inode_item *item;
5577 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5578 struct btrfs_inode_item);
5579 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5583 btrfs_release_path(path);
5584 path->search_commit_root = 0;
5585 path->skip_locking = 0;
5590 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5591 struct btrfs_root *root,
5592 struct btrfs_path *path,
5593 u64 ino, u64 parent,
5594 struct btrfs_log_ctx *ctx)
5596 struct btrfs_ino_list *ino_elem;
5597 struct inode *inode;
5600 * It's rare to have a lot of conflicting inodes, in practice it is not
5601 * common to have more than 1 or 2. We don't want to collect too many,
5602 * as we could end up logging too many inodes (even if only in
5603 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5606 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES)
5607 return BTRFS_LOG_FORCE_COMMIT;
5609 inode = btrfs_iget(root->fs_info->sb, ino, root);
5611 * If the other inode that had a conflicting dir entry was deleted in
5612 * the current transaction then we either:
5614 * 1) Log the parent directory (later after adding it to the list) if
5615 * the inode is a directory. This is because it may be a deleted
5616 * subvolume/snapshot or it may be a regular directory that had
5617 * deleted subvolumes/snapshots (or subdirectories that had them),
5618 * and at the moment we can't deal with dropping subvolumes/snapshots
5619 * during log replay. So we just log the parent, which will result in
5620 * a fallback to a transaction commit if we are dealing with those
5621 * cases (last_unlink_trans will match the current transaction);
5623 * 2) Do nothing if it's not a directory. During log replay we simply
5624 * unlink the conflicting dentry from the parent directory and then
5625 * add the dentry for our inode. Like this we can avoid logging the
5626 * parent directory (and maybe fallback to a transaction commit in
5627 * case it has a last_unlink_trans == trans->transid, due to moving
5628 * some inode from it to some other directory).
5630 if (IS_ERR(inode)) {
5631 int ret = PTR_ERR(inode);
5636 ret = conflicting_inode_is_dir(root, ino, path);
5637 /* Not a directory or we got an error. */
5641 /* Conflicting inode is a directory, so we'll log its parent. */
5642 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5645 ino_elem->ino = ino;
5646 ino_elem->parent = parent;
5647 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5648 ctx->num_conflict_inodes++;
5654 * If the inode was already logged skip it - otherwise we can hit an
5655 * infinite loop. Example:
5657 * From the commit root (previous transaction) we have the following
5660 * inode 257 a directory
5661 * inode 258 with references "zz" and "zz_link" on inode 257
5662 * inode 259 with reference "a" on inode 257
5664 * And in the current (uncommitted) transaction we have:
5666 * inode 257 a directory, unchanged
5667 * inode 258 with references "a" and "a2" on inode 257
5668 * inode 259 with reference "zz_link" on inode 257
5669 * inode 261 with reference "zz" on inode 257
5671 * When logging inode 261 the following infinite loop could
5672 * happen if we don't skip already logged inodes:
5674 * - we detect inode 258 as a conflicting inode, with inode 261
5675 * on reference "zz", and log it;
5677 * - we detect inode 259 as a conflicting inode, with inode 258
5678 * on reference "a", and log it;
5680 * - we detect inode 258 as a conflicting inode, with inode 259
5681 * on reference "zz_link", and log it - again! After this we
5682 * repeat the above steps forever.
5684 * Here we can use need_log_inode() because we only need to log the
5685 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5686 * so that the log ends up with the new name and without the old name.
5688 if (!need_log_inode(trans, BTRFS_I(inode))) {
5689 btrfs_add_delayed_iput(BTRFS_I(inode));
5693 btrfs_add_delayed_iput(BTRFS_I(inode));
5695 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5698 ino_elem->ino = ino;
5699 ino_elem->parent = parent;
5700 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5701 ctx->num_conflict_inodes++;
5706 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5707 struct btrfs_root *root,
5708 struct btrfs_log_ctx *ctx)
5710 struct btrfs_fs_info *fs_info = root->fs_info;
5714 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5715 * otherwise we could have unbounded recursion of btrfs_log_inode()
5716 * calls. This check guarantees we can have only 1 level of recursion.
5718 if (ctx->logging_conflict_inodes)
5721 ctx->logging_conflict_inodes = true;
5724 * New conflicting inodes may be found and added to the list while we
5725 * are logging a conflicting inode, so keep iterating while the list is
5728 while (!list_empty(&ctx->conflict_inodes)) {
5729 struct btrfs_ino_list *curr;
5730 struct inode *inode;
5734 curr = list_first_entry(&ctx->conflict_inodes,
5735 struct btrfs_ino_list, list);
5737 parent = curr->parent;
5738 list_del(&curr->list);
5741 inode = btrfs_iget(fs_info->sb, ino, root);
5743 * If the other inode that had a conflicting dir entry was
5744 * deleted in the current transaction, we need to log its parent
5745 * directory. See the comment at add_conflicting_inode().
5747 if (IS_ERR(inode)) {
5748 ret = PTR_ERR(inode);
5752 inode = btrfs_iget(fs_info->sb, parent, root);
5753 if (IS_ERR(inode)) {
5754 ret = PTR_ERR(inode);
5759 * Always log the directory, we cannot make this
5760 * conditional on need_log_inode() because the directory
5761 * might have been logged in LOG_INODE_EXISTS mode or
5762 * the dir index of the conflicting inode is not in a
5763 * dir index key range logged for the directory. So we
5764 * must make sure the deletion is recorded.
5766 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5767 LOG_INODE_ALL, ctx);
5768 btrfs_add_delayed_iput(BTRFS_I(inode));
5775 * Here we can use need_log_inode() because we only need to log
5776 * the inode in LOG_INODE_EXISTS mode and rename operations
5777 * update the log, so that the log ends up with the new name and
5778 * without the old name.
5780 * We did this check at add_conflicting_inode(), but here we do
5781 * it again because if some other task logged the inode after
5782 * that, we can avoid doing it again.
5784 if (!need_log_inode(trans, BTRFS_I(inode))) {
5785 btrfs_add_delayed_iput(BTRFS_I(inode));
5790 * We are safe logging the other inode without acquiring its
5791 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5792 * are safe against concurrent renames of the other inode as
5793 * well because during a rename we pin the log and update the
5794 * log with the new name before we unpin it.
5796 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5797 btrfs_add_delayed_iput(BTRFS_I(inode));
5802 ctx->logging_conflict_inodes = false;
5804 free_conflicting_inodes(ctx);
5809 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5810 struct btrfs_inode *inode,
5811 struct btrfs_key *min_key,
5812 const struct btrfs_key *max_key,
5813 struct btrfs_path *path,
5814 struct btrfs_path *dst_path,
5815 const u64 logged_isize,
5816 const int inode_only,
5817 struct btrfs_log_ctx *ctx,
5818 bool *need_log_inode_item)
5820 const u64 i_size = i_size_read(&inode->vfs_inode);
5821 struct btrfs_root *root = inode->root;
5822 int ins_start_slot = 0;
5827 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5835 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5836 if (min_key->objectid != max_key->objectid)
5838 if (min_key->type > max_key->type)
5841 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5842 *need_log_inode_item = false;
5843 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5844 min_key->offset >= i_size) {
5846 * Extents at and beyond eof are logged with
5847 * btrfs_log_prealloc_extents().
5848 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5849 * and no keys greater than that, so bail out.
5852 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5853 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5854 (inode->generation == trans->transid ||
5855 ctx->logging_conflict_inodes)) {
5857 u64 other_parent = 0;
5859 ret = btrfs_check_ref_name_override(path->nodes[0],
5860 path->slots[0], min_key, inode,
5861 &other_ino, &other_parent);
5864 } else if (ret > 0 &&
5865 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5870 ins_start_slot = path->slots[0];
5872 ret = copy_items(trans, inode, dst_path, path,
5873 ins_start_slot, ins_nr,
5874 inode_only, logged_isize);
5879 btrfs_release_path(path);
5880 ret = add_conflicting_inode(trans, root, path,
5887 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5888 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5891 ret = copy_items(trans, inode, dst_path, path,
5893 ins_nr, inode_only, logged_isize);
5900 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5903 } else if (!ins_nr) {
5904 ins_start_slot = path->slots[0];
5909 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5910 ins_nr, inode_only, logged_isize);
5914 ins_start_slot = path->slots[0];
5917 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5918 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5923 ret = copy_items(trans, inode, dst_path, path,
5924 ins_start_slot, ins_nr, inode_only,
5930 btrfs_release_path(path);
5932 if (min_key->offset < (u64)-1) {
5934 } else if (min_key->type < max_key->type) {
5936 min_key->offset = 0;
5942 * We may process many leaves full of items for our inode, so
5943 * avoid monopolizing a cpu for too long by rescheduling while
5944 * not holding locks on any tree.
5949 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5950 ins_nr, inode_only, logged_isize);
5955 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5957 * Release the path because otherwise we might attempt to double
5958 * lock the same leaf with btrfs_log_prealloc_extents() below.
5960 btrfs_release_path(path);
5961 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5967 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
5968 struct btrfs_root *log,
5969 struct btrfs_path *path,
5970 const struct btrfs_item_batch *batch,
5971 const struct btrfs_delayed_item *first_item)
5973 const struct btrfs_delayed_item *curr = first_item;
5976 ret = btrfs_insert_empty_items(trans, log, path, batch);
5980 for (int i = 0; i < batch->nr; i++) {
5983 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
5984 write_extent_buffer(path->nodes[0], &curr->data,
5985 (unsigned long)data_ptr, curr->data_len);
5986 curr = list_next_entry(curr, log_list);
5990 btrfs_release_path(path);
5995 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
5996 struct btrfs_inode *inode,
5997 struct btrfs_path *path,
5998 const struct list_head *delayed_ins_list,
5999 struct btrfs_log_ctx *ctx)
6001 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
6002 const int max_batch_size = 195;
6003 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
6004 const u64 ino = btrfs_ino(inode);
6005 struct btrfs_root *log = inode->root->log_root;
6006 struct btrfs_item_batch batch = {
6008 .total_data_size = 0,
6010 const struct btrfs_delayed_item *first = NULL;
6011 const struct btrfs_delayed_item *curr;
6013 struct btrfs_key *ins_keys;
6015 u64 curr_batch_size = 0;
6019 /* We are adding dir index items to the log tree. */
6020 lockdep_assert_held(&inode->log_mutex);
6023 * We collect delayed items before copying index keys from the subvolume
6024 * to the log tree. However just after we collected them, they may have
6025 * been flushed (all of them or just some of them), and therefore we
6026 * could have copied them from the subvolume tree to the log tree.
6027 * So find the first delayed item that was not yet logged (they are
6028 * sorted by index number).
6030 list_for_each_entry(curr, delayed_ins_list, log_list) {
6031 if (curr->index > inode->last_dir_index_offset) {
6037 /* Empty list or all delayed items were already logged. */
6041 ins_data = kmalloc(max_batch_size * sizeof(u32) +
6042 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
6045 ins_sizes = (u32 *)ins_data;
6046 batch.data_sizes = ins_sizes;
6047 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6048 batch.keys = ins_keys;
6051 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6052 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6054 if (curr_batch_size + curr_size > leaf_data_size ||
6055 batch.nr == max_batch_size) {
6056 ret = insert_delayed_items_batch(trans, log, path,
6062 batch.total_data_size = 0;
6063 curr_batch_size = 0;
6067 ins_sizes[batch_idx] = curr->data_len;
6068 ins_keys[batch_idx].objectid = ino;
6069 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6070 ins_keys[batch_idx].offset = curr->index;
6071 curr_batch_size += curr_size;
6072 batch.total_data_size += curr->data_len;
6075 curr = list_next_entry(curr, log_list);
6078 ASSERT(batch.nr >= 1);
6079 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6081 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6083 inode->last_dir_index_offset = curr->index;
6090 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6091 struct btrfs_inode *inode,
6092 struct btrfs_path *path,
6093 const struct list_head *delayed_del_list,
6094 struct btrfs_log_ctx *ctx)
6096 const u64 ino = btrfs_ino(inode);
6097 const struct btrfs_delayed_item *curr;
6099 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6102 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6103 u64 first_dir_index = curr->index;
6105 const struct btrfs_delayed_item *next;
6109 * Find a range of consecutive dir index items to delete. Like
6110 * this we log a single dir range item spanning several contiguous
6111 * dir items instead of logging one range item per dir index item.
6113 next = list_next_entry(curr, log_list);
6114 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6115 if (next->index != curr->index + 1)
6118 next = list_next_entry(next, log_list);
6121 last_dir_index = curr->index;
6122 ASSERT(last_dir_index >= first_dir_index);
6124 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6125 ino, first_dir_index, last_dir_index);
6128 curr = list_next_entry(curr, log_list);
6134 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6135 struct btrfs_inode *inode,
6136 struct btrfs_path *path,
6137 struct btrfs_log_ctx *ctx,
6138 const struct list_head *delayed_del_list,
6139 const struct btrfs_delayed_item *first,
6140 const struct btrfs_delayed_item **last_ret)
6142 const struct btrfs_delayed_item *next;
6143 struct extent_buffer *leaf = path->nodes[0];
6144 const int last_slot = btrfs_header_nritems(leaf) - 1;
6145 int slot = path->slots[0] + 1;
6146 const u64 ino = btrfs_ino(inode);
6148 next = list_next_entry(first, log_list);
6150 while (slot < last_slot &&
6151 !list_entry_is_head(next, delayed_del_list, log_list)) {
6152 struct btrfs_key key;
6154 btrfs_item_key_to_cpu(leaf, &key, slot);
6155 if (key.objectid != ino ||
6156 key.type != BTRFS_DIR_INDEX_KEY ||
6157 key.offset != next->index)
6162 next = list_next_entry(next, log_list);
6165 return btrfs_del_items(trans, inode->root->log_root, path,
6166 path->slots[0], slot - path->slots[0]);
6169 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6170 struct btrfs_inode *inode,
6171 struct btrfs_path *path,
6172 const struct list_head *delayed_del_list,
6173 struct btrfs_log_ctx *ctx)
6175 struct btrfs_root *log = inode->root->log_root;
6176 const struct btrfs_delayed_item *curr;
6177 u64 last_range_start;
6178 u64 last_range_end = 0;
6179 struct btrfs_key key;
6181 key.objectid = btrfs_ino(inode);
6182 key.type = BTRFS_DIR_INDEX_KEY;
6183 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6186 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6187 const struct btrfs_delayed_item *last = curr;
6188 u64 first_dir_index = curr->index;
6190 bool deleted_items = false;
6193 key.offset = curr->index;
6194 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6197 } else if (ret == 0) {
6198 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6199 delayed_del_list, curr,
6203 deleted_items = true;
6206 btrfs_release_path(path);
6209 * If we deleted items from the leaf, it means we have a range
6210 * item logging their range, so no need to add one or update an
6211 * existing one. Otherwise we have to log a dir range item.
6216 last_dir_index = last->index;
6217 ASSERT(last_dir_index >= first_dir_index);
6219 * If this range starts right after where the previous one ends,
6220 * then we want to reuse the previous range item and change its
6221 * end offset to the end of this range. This is just to minimize
6222 * leaf space usage, by avoiding adding a new range item.
6224 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6225 first_dir_index = last_range_start;
6227 ret = insert_dir_log_key(trans, log, path, key.objectid,
6228 first_dir_index, last_dir_index);
6232 last_range_start = first_dir_index;
6233 last_range_end = last_dir_index;
6235 curr = list_next_entry(last, log_list);
6241 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6242 struct btrfs_inode *inode,
6243 struct btrfs_path *path,
6244 const struct list_head *delayed_del_list,
6245 struct btrfs_log_ctx *ctx)
6248 * We are deleting dir index items from the log tree or adding range
6251 lockdep_assert_held(&inode->log_mutex);
6253 if (list_empty(delayed_del_list))
6256 if (ctx->logged_before)
6257 return log_delayed_deletions_incremental(trans, inode, path,
6258 delayed_del_list, ctx);
6260 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6265 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6266 * items instead of the subvolume tree.
6268 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6269 struct btrfs_inode *inode,
6270 const struct list_head *delayed_ins_list,
6271 struct btrfs_log_ctx *ctx)
6273 const bool orig_log_new_dentries = ctx->log_new_dentries;
6274 struct btrfs_fs_info *fs_info = trans->fs_info;
6275 struct btrfs_delayed_item *item;
6279 * No need for the log mutex, plus to avoid potential deadlocks or
6280 * lockdep annotations due to nesting of delayed inode mutexes and log
6283 lockdep_assert_not_held(&inode->log_mutex);
6285 ASSERT(!ctx->logging_new_delayed_dentries);
6286 ctx->logging_new_delayed_dentries = true;
6288 list_for_each_entry(item, delayed_ins_list, log_list) {
6289 struct btrfs_dir_item *dir_item;
6290 struct inode *di_inode;
6291 struct btrfs_key key;
6292 int log_mode = LOG_INODE_EXISTS;
6294 dir_item = (struct btrfs_dir_item *)item->data;
6295 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6297 if (key.type == BTRFS_ROOT_ITEM_KEY)
6300 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6301 if (IS_ERR(di_inode)) {
6302 ret = PTR_ERR(di_inode);
6306 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6307 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6311 if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR)
6312 log_mode = LOG_INODE_ALL;
6314 ctx->log_new_dentries = false;
6315 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6317 if (!ret && ctx->log_new_dentries)
6318 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6320 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6326 ctx->log_new_dentries = orig_log_new_dentries;
6327 ctx->logging_new_delayed_dentries = false;
6332 /* log a single inode in the tree log.
6333 * At least one parent directory for this inode must exist in the tree
6334 * or be logged already.
6336 * Any items from this inode changed by the current transaction are copied
6337 * to the log tree. An extra reference is taken on any extents in this
6338 * file, allowing us to avoid a whole pile of corner cases around logging
6339 * blocks that have been removed from the tree.
6341 * See LOG_INODE_ALL and related defines for a description of what inode_only
6344 * This handles both files and directories.
6346 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6347 struct btrfs_inode *inode,
6349 struct btrfs_log_ctx *ctx)
6351 struct btrfs_path *path;
6352 struct btrfs_path *dst_path;
6353 struct btrfs_key min_key;
6354 struct btrfs_key max_key;
6355 struct btrfs_root *log = inode->root->log_root;
6357 bool fast_search = false;
6358 u64 ino = btrfs_ino(inode);
6359 struct extent_map_tree *em_tree = &inode->extent_tree;
6360 u64 logged_isize = 0;
6361 bool need_log_inode_item = true;
6362 bool xattrs_logged = false;
6363 bool inode_item_dropped = true;
6364 bool full_dir_logging = false;
6365 LIST_HEAD(delayed_ins_list);
6366 LIST_HEAD(delayed_del_list);
6368 path = btrfs_alloc_path();
6371 dst_path = btrfs_alloc_path();
6373 btrfs_free_path(path);
6377 min_key.objectid = ino;
6378 min_key.type = BTRFS_INODE_ITEM_KEY;
6381 max_key.objectid = ino;
6384 /* today the code can only do partial logging of directories */
6385 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6386 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6387 &inode->runtime_flags) &&
6388 inode_only >= LOG_INODE_EXISTS))
6389 max_key.type = BTRFS_XATTR_ITEM_KEY;
6391 max_key.type = (u8)-1;
6392 max_key.offset = (u64)-1;
6394 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6395 full_dir_logging = true;
6398 * If we are logging a directory while we are logging dentries of the
6399 * delayed items of some other inode, then we need to flush the delayed
6400 * items of this directory and not log the delayed items directly. This
6401 * is to prevent more than one level of recursion into btrfs_log_inode()
6402 * by having something like this:
6404 * $ mkdir -p a/b/c/d/e/f/g/h/...
6405 * $ xfs_io -c "fsync" a
6407 * Where all directories in the path did not exist before and are
6408 * created in the current transaction.
6409 * So in such a case we directly log the delayed items of the main
6410 * directory ("a") without flushing them first, while for each of its
6411 * subdirectories we flush their delayed items before logging them.
6412 * This prevents a potential unbounded recursion like this:
6415 * log_new_delayed_dentries()
6417 * log_new_delayed_dentries()
6419 * log_new_delayed_dentries()
6422 * We have thresholds for the maximum number of delayed items to have in
6423 * memory, and once they are hit, the items are flushed asynchronously.
6424 * However the limit is quite high, so lets prevent deep levels of
6425 * recursion to happen by limiting the maximum depth to be 1.
6427 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6428 ret = btrfs_commit_inode_delayed_items(trans, inode);
6433 mutex_lock(&inode->log_mutex);
6436 * For symlinks, we must always log their content, which is stored in an
6437 * inline extent, otherwise we could end up with an empty symlink after
6438 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6439 * one attempts to create an empty symlink).
6440 * We don't need to worry about flushing delalloc, because when we create
6441 * the inline extent when the symlink is created (we never have delalloc
6444 if (S_ISLNK(inode->vfs_inode.i_mode))
6445 inode_only = LOG_INODE_ALL;
6448 * Before logging the inode item, cache the value returned by
6449 * inode_logged(), because after that we have the need to figure out if
6450 * the inode was previously logged in this transaction.
6452 ret = inode_logged(trans, inode, path);
6455 ctx->logged_before = (ret == 1);
6459 * This is for cases where logging a directory could result in losing a
6460 * a file after replaying the log. For example, if we move a file from a
6461 * directory A to a directory B, then fsync directory A, we have no way
6462 * to known the file was moved from A to B, so logging just A would
6463 * result in losing the file after a log replay.
6465 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6466 btrfs_set_log_full_commit(trans);
6467 ret = BTRFS_LOG_FORCE_COMMIT;
6472 * a brute force approach to making sure we get the most uptodate
6473 * copies of everything.
6475 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6476 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6477 if (ctx->logged_before)
6478 ret = drop_inode_items(trans, log, path, inode,
6479 BTRFS_XATTR_ITEM_KEY);
6481 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6483 * Make sure the new inode item we write to the log has
6484 * the same isize as the current one (if it exists).
6485 * This is necessary to prevent data loss after log
6486 * replay, and also to prevent doing a wrong expanding
6487 * truncate - for e.g. create file, write 4K into offset
6488 * 0, fsync, write 4K into offset 4096, add hard link,
6489 * fsync some other file (to sync log), power fail - if
6490 * we use the inode's current i_size, after log replay
6491 * we get a 8Kb file, with the last 4Kb extent as a hole
6492 * (zeroes), as if an expanding truncate happened,
6493 * instead of getting a file of 4Kb only.
6495 ret = logged_inode_size(log, inode, path, &logged_isize);
6499 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6500 &inode->runtime_flags)) {
6501 if (inode_only == LOG_INODE_EXISTS) {
6502 max_key.type = BTRFS_XATTR_ITEM_KEY;
6503 if (ctx->logged_before)
6504 ret = drop_inode_items(trans, log, path,
6505 inode, max_key.type);
6507 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6508 &inode->runtime_flags);
6509 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6510 &inode->runtime_flags);
6511 if (ctx->logged_before)
6512 ret = truncate_inode_items(trans, log,
6515 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6516 &inode->runtime_flags) ||
6517 inode_only == LOG_INODE_EXISTS) {
6518 if (inode_only == LOG_INODE_ALL)
6520 max_key.type = BTRFS_XATTR_ITEM_KEY;
6521 if (ctx->logged_before)
6522 ret = drop_inode_items(trans, log, path, inode,
6525 if (inode_only == LOG_INODE_ALL)
6527 inode_item_dropped = false;
6536 * If we are logging a directory in full mode, collect the delayed items
6537 * before iterating the subvolume tree, so that we don't miss any new
6538 * dir index items in case they get flushed while or right after we are
6539 * iterating the subvolume tree.
6541 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6542 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6545 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6546 path, dst_path, logged_isize,
6548 &need_log_inode_item);
6552 btrfs_release_path(path);
6553 btrfs_release_path(dst_path);
6554 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6557 xattrs_logged = true;
6558 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6559 btrfs_release_path(path);
6560 btrfs_release_path(dst_path);
6561 ret = btrfs_log_holes(trans, inode, path);
6566 btrfs_release_path(path);
6567 btrfs_release_path(dst_path);
6568 if (need_log_inode_item) {
6569 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6573 * If we are doing a fast fsync and the inode was logged before
6574 * in this transaction, we don't need to log the xattrs because
6575 * they were logged before. If xattrs were added, changed or
6576 * deleted since the last time we logged the inode, then we have
6577 * already logged them because the inode had the runtime flag
6578 * BTRFS_INODE_COPY_EVERYTHING set.
6580 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6581 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path);
6584 btrfs_release_path(path);
6588 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6591 } else if (inode_only == LOG_INODE_ALL) {
6592 struct extent_map *em, *n;
6594 write_lock(&em_tree->lock);
6595 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6596 list_del_init(&em->list);
6597 write_unlock(&em_tree->lock);
6600 if (full_dir_logging) {
6601 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6604 ret = log_delayed_insertion_items(trans, inode, path,
6605 &delayed_ins_list, ctx);
6608 ret = log_delayed_deletion_items(trans, inode, path,
6609 &delayed_del_list, ctx);
6614 spin_lock(&inode->lock);
6615 inode->logged_trans = trans->transid;
6617 * Don't update last_log_commit if we logged that an inode exists.
6618 * We do this for three reasons:
6620 * 1) We might have had buffered writes to this inode that were
6621 * flushed and had their ordered extents completed in this
6622 * transaction, but we did not previously log the inode with
6623 * LOG_INODE_ALL. Later the inode was evicted and after that
6624 * it was loaded again and this LOG_INODE_EXISTS log operation
6625 * happened. We must make sure that if an explicit fsync against
6626 * the inode is performed later, it logs the new extents, an
6627 * updated inode item, etc, and syncs the log. The same logic
6628 * applies to direct IO writes instead of buffered writes.
6630 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6631 * is logged with an i_size of 0 or whatever value was logged
6632 * before. If later the i_size of the inode is increased by a
6633 * truncate operation, the log is synced through an fsync of
6634 * some other inode and then finally an explicit fsync against
6635 * this inode is made, we must make sure this fsync logs the
6636 * inode with the new i_size, the hole between old i_size and
6637 * the new i_size, and syncs the log.
6639 * 3) If we are logging that an ancestor inode exists as part of
6640 * logging a new name from a link or rename operation, don't update
6641 * its last_log_commit - otherwise if an explicit fsync is made
6642 * against an ancestor, the fsync considers the inode in the log
6643 * and doesn't sync the log, resulting in the ancestor missing after
6644 * a power failure unless the log was synced as part of an fsync
6645 * against any other unrelated inode.
6647 if (inode_only != LOG_INODE_EXISTS)
6648 inode->last_log_commit = inode->last_sub_trans;
6649 spin_unlock(&inode->lock);
6652 * Reset the last_reflink_trans so that the next fsync does not need to
6653 * go through the slower path when logging extents and their checksums.
6655 if (inode_only == LOG_INODE_ALL)
6656 inode->last_reflink_trans = 0;
6659 mutex_unlock(&inode->log_mutex);
6661 btrfs_free_path(path);
6662 btrfs_free_path(dst_path);
6665 free_conflicting_inodes(ctx);
6667 ret = log_conflicting_inodes(trans, inode->root, ctx);
6669 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6671 ret = log_new_delayed_dentries(trans, inode,
6672 &delayed_ins_list, ctx);
6674 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6681 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6682 struct btrfs_inode *inode,
6683 struct btrfs_log_ctx *ctx)
6685 struct btrfs_fs_info *fs_info = trans->fs_info;
6687 struct btrfs_path *path;
6688 struct btrfs_key key;
6689 struct btrfs_root *root = inode->root;
6690 const u64 ino = btrfs_ino(inode);
6692 path = btrfs_alloc_path();
6695 path->skip_locking = 1;
6696 path->search_commit_root = 1;
6699 key.type = BTRFS_INODE_REF_KEY;
6701 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6706 struct extent_buffer *leaf = path->nodes[0];
6707 int slot = path->slots[0];
6712 if (slot >= btrfs_header_nritems(leaf)) {
6713 ret = btrfs_next_leaf(root, path);
6721 btrfs_item_key_to_cpu(leaf, &key, slot);
6722 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6723 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6726 item_size = btrfs_item_size(leaf, slot);
6727 ptr = btrfs_item_ptr_offset(leaf, slot);
6728 while (cur_offset < item_size) {
6729 struct btrfs_key inode_key;
6730 struct inode *dir_inode;
6732 inode_key.type = BTRFS_INODE_ITEM_KEY;
6733 inode_key.offset = 0;
6735 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6736 struct btrfs_inode_extref *extref;
6738 extref = (struct btrfs_inode_extref *)
6740 inode_key.objectid = btrfs_inode_extref_parent(
6742 cur_offset += sizeof(*extref);
6743 cur_offset += btrfs_inode_extref_name_len(leaf,
6746 inode_key.objectid = key.offset;
6747 cur_offset = item_size;
6750 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6753 * If the parent inode was deleted, return an error to
6754 * fallback to a transaction commit. This is to prevent
6755 * getting an inode that was moved from one parent A to
6756 * a parent B, got its former parent A deleted and then
6757 * it got fsync'ed, from existing at both parents after
6758 * a log replay (and the old parent still existing).
6765 * mv /mnt/B/bar /mnt/A/bar
6766 * mv -T /mnt/A /mnt/B
6770 * If we ignore the old parent B which got deleted,
6771 * after a log replay we would have file bar linked
6772 * at both parents and the old parent B would still
6775 if (IS_ERR(dir_inode)) {
6776 ret = PTR_ERR(dir_inode);
6780 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6781 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6785 ctx->log_new_dentries = false;
6786 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6787 LOG_INODE_ALL, ctx);
6788 if (!ret && ctx->log_new_dentries)
6789 ret = log_new_dir_dentries(trans,
6790 BTRFS_I(dir_inode), ctx);
6791 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6799 btrfs_free_path(path);
6803 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6804 struct btrfs_root *root,
6805 struct btrfs_path *path,
6806 struct btrfs_log_ctx *ctx)
6808 struct btrfs_key found_key;
6810 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6813 struct btrfs_fs_info *fs_info = root->fs_info;
6814 struct extent_buffer *leaf = path->nodes[0];
6815 int slot = path->slots[0];
6816 struct btrfs_key search_key;
6817 struct inode *inode;
6821 btrfs_release_path(path);
6823 ino = found_key.offset;
6825 search_key.objectid = found_key.offset;
6826 search_key.type = BTRFS_INODE_ITEM_KEY;
6827 search_key.offset = 0;
6828 inode = btrfs_iget(fs_info->sb, ino, root);
6830 return PTR_ERR(inode);
6832 if (BTRFS_I(inode)->generation >= trans->transid &&
6833 need_log_inode(trans, BTRFS_I(inode)))
6834 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6835 LOG_INODE_EXISTS, ctx);
6836 btrfs_add_delayed_iput(BTRFS_I(inode));
6840 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6843 search_key.type = BTRFS_INODE_REF_KEY;
6844 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6848 leaf = path->nodes[0];
6849 slot = path->slots[0];
6850 if (slot >= btrfs_header_nritems(leaf)) {
6851 ret = btrfs_next_leaf(root, path);
6856 leaf = path->nodes[0];
6857 slot = path->slots[0];
6860 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6861 if (found_key.objectid != search_key.objectid ||
6862 found_key.type != BTRFS_INODE_REF_KEY)
6868 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6869 struct btrfs_inode *inode,
6870 struct dentry *parent,
6871 struct btrfs_log_ctx *ctx)
6873 struct btrfs_root *root = inode->root;
6874 struct dentry *old_parent = NULL;
6875 struct super_block *sb = inode->vfs_inode.i_sb;
6879 if (!parent || d_really_is_negative(parent) ||
6883 inode = BTRFS_I(d_inode(parent));
6884 if (root != inode->root)
6887 if (inode->generation >= trans->transid &&
6888 need_log_inode(trans, inode)) {
6889 ret = btrfs_log_inode(trans, inode,
6890 LOG_INODE_EXISTS, ctx);
6894 if (IS_ROOT(parent))
6897 parent = dget_parent(parent);
6899 old_parent = parent;
6906 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6907 struct btrfs_inode *inode,
6908 struct dentry *parent,
6909 struct btrfs_log_ctx *ctx)
6911 struct btrfs_root *root = inode->root;
6912 const u64 ino = btrfs_ino(inode);
6913 struct btrfs_path *path;
6914 struct btrfs_key search_key;
6918 * For a single hard link case, go through a fast path that does not
6919 * need to iterate the fs/subvolume tree.
6921 if (inode->vfs_inode.i_nlink < 2)
6922 return log_new_ancestors_fast(trans, inode, parent, ctx);
6924 path = btrfs_alloc_path();
6928 search_key.objectid = ino;
6929 search_key.type = BTRFS_INODE_REF_KEY;
6930 search_key.offset = 0;
6932 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6939 struct extent_buffer *leaf = path->nodes[0];
6940 int slot = path->slots[0];
6941 struct btrfs_key found_key;
6943 if (slot >= btrfs_header_nritems(leaf)) {
6944 ret = btrfs_next_leaf(root, path);
6952 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6953 if (found_key.objectid != ino ||
6954 found_key.type > BTRFS_INODE_EXTREF_KEY)
6958 * Don't deal with extended references because they are rare
6959 * cases and too complex to deal with (we would need to keep
6960 * track of which subitem we are processing for each item in
6961 * this loop, etc). So just return some error to fallback to
6962 * a transaction commit.
6964 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6970 * Logging ancestors needs to do more searches on the fs/subvol
6971 * tree, so it releases the path as needed to avoid deadlocks.
6972 * Keep track of the last inode ref key and resume from that key
6973 * after logging all new ancestors for the current hard link.
6975 memcpy(&search_key, &found_key, sizeof(search_key));
6977 ret = log_new_ancestors(trans, root, path, ctx);
6980 btrfs_release_path(path);
6985 btrfs_free_path(path);
6990 * helper function around btrfs_log_inode to make sure newly created
6991 * parent directories also end up in the log. A minimal inode and backref
6992 * only logging is done of any parent directories that are older than
6993 * the last committed transaction
6995 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6996 struct btrfs_inode *inode,
6997 struct dentry *parent,
6999 struct btrfs_log_ctx *ctx)
7001 struct btrfs_root *root = inode->root;
7002 struct btrfs_fs_info *fs_info = root->fs_info;
7004 bool log_dentries = false;
7006 if (btrfs_test_opt(fs_info, NOTREELOG)) {
7007 ret = BTRFS_LOG_FORCE_COMMIT;
7011 if (btrfs_root_refs(&root->root_item) == 0) {
7012 ret = BTRFS_LOG_FORCE_COMMIT;
7017 * Skip already logged inodes or inodes corresponding to tmpfiles
7018 * (since logging them is pointless, a link count of 0 means they
7019 * will never be accessible).
7021 if ((btrfs_inode_in_log(inode, trans->transid) &&
7022 list_empty(&ctx->ordered_extents)) ||
7023 inode->vfs_inode.i_nlink == 0) {
7024 ret = BTRFS_NO_LOG_SYNC;
7028 ret = start_log_trans(trans, root, ctx);
7032 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
7037 * for regular files, if its inode is already on disk, we don't
7038 * have to worry about the parents at all. This is because
7039 * we can use the last_unlink_trans field to record renames
7040 * and other fun in this file.
7042 if (S_ISREG(inode->vfs_inode.i_mode) &&
7043 inode->generation < trans->transid &&
7044 inode->last_unlink_trans < trans->transid) {
7049 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7050 log_dentries = true;
7053 * On unlink we must make sure all our current and old parent directory
7054 * inodes are fully logged. This is to prevent leaving dangling
7055 * directory index entries in directories that were our parents but are
7056 * not anymore. Not doing this results in old parent directory being
7057 * impossible to delete after log replay (rmdir will always fail with
7058 * error -ENOTEMPTY).
7064 * ln testdir/foo testdir/bar
7066 * unlink testdir/bar
7067 * xfs_io -c fsync testdir/foo
7069 * mount fs, triggers log replay
7071 * If we don't log the parent directory (testdir), after log replay the
7072 * directory still has an entry pointing to the file inode using the bar
7073 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7074 * the file inode has a link count of 1.
7080 * ln foo testdir/foo2
7081 * ln foo testdir/foo3
7083 * unlink testdir/foo3
7084 * xfs_io -c fsync foo
7086 * mount fs, triggers log replay
7088 * Similar as the first example, after log replay the parent directory
7089 * testdir still has an entry pointing to the inode file with name foo3
7090 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7091 * and has a link count of 2.
7093 if (inode->last_unlink_trans >= trans->transid) {
7094 ret = btrfs_log_all_parents(trans, inode, ctx);
7099 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7104 ret = log_new_dir_dentries(trans, inode, ctx);
7109 btrfs_set_log_full_commit(trans);
7110 ret = BTRFS_LOG_FORCE_COMMIT;
7114 btrfs_remove_log_ctx(root, ctx);
7115 btrfs_end_log_trans(root);
7121 * it is not safe to log dentry if the chunk root has added new
7122 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7123 * If this returns 1, you must commit the transaction to safely get your
7126 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7127 struct dentry *dentry,
7128 struct btrfs_log_ctx *ctx)
7130 struct dentry *parent = dget_parent(dentry);
7133 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7134 LOG_INODE_ALL, ctx);
7141 * should be called during mount to recover any replay any log trees
7144 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7147 struct btrfs_path *path;
7148 struct btrfs_trans_handle *trans;
7149 struct btrfs_key key;
7150 struct btrfs_key found_key;
7151 struct btrfs_root *log;
7152 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7153 struct walk_control wc = {
7154 .process_func = process_one_buffer,
7155 .stage = LOG_WALK_PIN_ONLY,
7158 path = btrfs_alloc_path();
7162 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7164 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7165 if (IS_ERR(trans)) {
7166 ret = PTR_ERR(trans);
7173 ret = walk_log_tree(trans, log_root_tree, &wc);
7175 btrfs_abort_transaction(trans, ret);
7180 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7181 key.offset = (u64)-1;
7182 key.type = BTRFS_ROOT_ITEM_KEY;
7185 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7188 btrfs_abort_transaction(trans, ret);
7192 if (path->slots[0] == 0)
7196 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7198 btrfs_release_path(path);
7199 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7202 log = btrfs_read_tree_root(log_root_tree, &found_key);
7205 btrfs_abort_transaction(trans, ret);
7209 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7211 if (IS_ERR(wc.replay_dest)) {
7212 ret = PTR_ERR(wc.replay_dest);
7215 * We didn't find the subvol, likely because it was
7216 * deleted. This is ok, simply skip this log and go to
7219 * We need to exclude the root because we can't have
7220 * other log replays overwriting this log as we'll read
7221 * it back in a few more times. This will keep our
7222 * block from being modified, and we'll just bail for
7223 * each subsequent pass.
7226 ret = btrfs_pin_extent_for_log_replay(trans,
7229 btrfs_put_root(log);
7233 btrfs_abort_transaction(trans, ret);
7237 wc.replay_dest->log_root = log;
7238 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7240 /* The loop needs to continue due to the root refs */
7241 btrfs_abort_transaction(trans, ret);
7243 ret = walk_log_tree(trans, log, &wc);
7245 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7246 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7249 btrfs_abort_transaction(trans, ret);
7252 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7253 struct btrfs_root *root = wc.replay_dest;
7255 btrfs_release_path(path);
7258 * We have just replayed everything, and the highest
7259 * objectid of fs roots probably has changed in case
7260 * some inode_item's got replayed.
7262 * root->objectid_mutex is not acquired as log replay
7263 * could only happen during mount.
7265 ret = btrfs_init_root_free_objectid(root);
7267 btrfs_abort_transaction(trans, ret);
7270 wc.replay_dest->log_root = NULL;
7271 btrfs_put_root(wc.replay_dest);
7272 btrfs_put_root(log);
7277 if (found_key.offset == 0)
7279 key.offset = found_key.offset - 1;
7281 btrfs_release_path(path);
7283 /* step one is to pin it all, step two is to replay just inodes */
7286 wc.process_func = replay_one_buffer;
7287 wc.stage = LOG_WALK_REPLAY_INODES;
7290 /* step three is to replay everything */
7291 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7296 btrfs_free_path(path);
7298 /* step 4: commit the transaction, which also unpins the blocks */
7299 ret = btrfs_commit_transaction(trans);
7303 log_root_tree->log_root = NULL;
7304 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7305 btrfs_put_root(log_root_tree);
7310 btrfs_end_transaction(wc.trans);
7311 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7312 btrfs_free_path(path);
7317 * there are some corner cases where we want to force a full
7318 * commit instead of allowing a directory to be logged.
7320 * They revolve around files there were unlinked from the directory, and
7321 * this function updates the parent directory so that a full commit is
7322 * properly done if it is fsync'd later after the unlinks are done.
7324 * Must be called before the unlink operations (updates to the subvolume tree,
7325 * inodes, etc) are done.
7327 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7328 struct btrfs_inode *dir, struct btrfs_inode *inode,
7332 * when we're logging a file, if it hasn't been renamed
7333 * or unlinked, and its inode is fully committed on disk,
7334 * we don't have to worry about walking up the directory chain
7335 * to log its parents.
7337 * So, we use the last_unlink_trans field to put this transid
7338 * into the file. When the file is logged we check it and
7339 * don't log the parents if the file is fully on disk.
7341 mutex_lock(&inode->log_mutex);
7342 inode->last_unlink_trans = trans->transid;
7343 mutex_unlock(&inode->log_mutex);
7346 * if this directory was already logged any new
7347 * names for this file/dir will get recorded
7349 if (dir->logged_trans == trans->transid)
7353 * if the inode we're about to unlink was logged,
7354 * the log will be properly updated for any new names
7356 if (inode->logged_trans == trans->transid)
7360 * when renaming files across directories, if the directory
7361 * there we're unlinking from gets fsync'd later on, there's
7362 * no way to find the destination directory later and fsync it
7363 * properly. So, we have to be conservative and force commits
7364 * so the new name gets discovered.
7369 /* we can safely do the unlink without any special recording */
7373 mutex_lock(&dir->log_mutex);
7374 dir->last_unlink_trans = trans->transid;
7375 mutex_unlock(&dir->log_mutex);
7379 * Make sure that if someone attempts to fsync the parent directory of a deleted
7380 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7381 * that after replaying the log tree of the parent directory's root we will not
7382 * see the snapshot anymore and at log replay time we will not see any log tree
7383 * corresponding to the deleted snapshot's root, which could lead to replaying
7384 * it after replaying the log tree of the parent directory (which would replay
7385 * the snapshot delete operation).
7387 * Must be called before the actual snapshot destroy operation (updates to the
7388 * parent root and tree of tree roots trees, etc) are done.
7390 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7391 struct btrfs_inode *dir)
7393 mutex_lock(&dir->log_mutex);
7394 dir->last_unlink_trans = trans->transid;
7395 mutex_unlock(&dir->log_mutex);
7399 * Update the log after adding a new name for an inode.
7401 * @trans: Transaction handle.
7402 * @old_dentry: The dentry associated with the old name and the old
7404 * @old_dir: The inode of the previous parent directory for the case
7405 * of a rename. For a link operation, it must be NULL.
7406 * @old_dir_index: The index number associated with the old name, meaningful
7407 * only for rename operations (when @old_dir is not NULL).
7408 * Ignored for link operations.
7409 * @parent: The dentry associated with the directory under which the
7410 * new name is located.
7412 * Call this after adding a new name for an inode, as a result of a link or
7413 * rename operation, and it will properly update the log to reflect the new name.
7415 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7416 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7417 u64 old_dir_index, struct dentry *parent)
7419 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7420 struct btrfs_root *root = inode->root;
7421 struct btrfs_log_ctx ctx;
7422 bool log_pinned = false;
7426 * this will force the logging code to walk the dentry chain
7429 if (!S_ISDIR(inode->vfs_inode.i_mode))
7430 inode->last_unlink_trans = trans->transid;
7433 * if this inode hasn't been logged and directory we're renaming it
7434 * from hasn't been logged, we don't need to log it
7436 ret = inode_logged(trans, inode, NULL);
7439 } else if (ret == 0) {
7443 * If the inode was not logged and we are doing a rename (old_dir is not
7444 * NULL), check if old_dir was logged - if it was not we can return and
7447 ret = inode_logged(trans, old_dir, NULL);
7456 * If we are doing a rename (old_dir is not NULL) from a directory that
7457 * was previously logged, make sure that on log replay we get the old
7458 * dir entry deleted. This is needed because we will also log the new
7459 * name of the renamed inode, so we need to make sure that after log
7460 * replay we don't end up with both the new and old dir entries existing.
7462 if (old_dir && old_dir->logged_trans == trans->transid) {
7463 struct btrfs_root *log = old_dir->root->log_root;
7464 struct btrfs_path *path;
7465 struct fscrypt_name fname;
7467 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7469 ret = fscrypt_setup_filename(&old_dir->vfs_inode,
7470 &old_dentry->d_name, 0, &fname);
7474 * We have two inodes to update in the log, the old directory and
7475 * the inode that got renamed, so we must pin the log to prevent
7476 * anyone from syncing the log until we have updated both inodes
7479 ret = join_running_log_trans(root);
7481 * At least one of the inodes was logged before, so this should
7482 * not fail, but if it does, it's not serious, just bail out and
7483 * mark the log for a full commit.
7485 if (WARN_ON_ONCE(ret < 0))
7489 path = btrfs_alloc_path();
7492 fscrypt_free_filename(&fname);
7497 * Other concurrent task might be logging the old directory,
7498 * as it can be triggered when logging other inode that had or
7499 * still has a dentry in the old directory. We lock the old
7500 * directory's log_mutex to ensure the deletion of the old
7501 * name is persisted, because during directory logging we
7502 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7503 * the old name's dir index item is in the delayed items, so
7504 * it could be missed by an in progress directory logging.
7506 mutex_lock(&old_dir->log_mutex);
7507 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7508 &fname.disk_name, old_dir_index);
7511 * The dentry does not exist in the log, so record its
7514 btrfs_release_path(path);
7515 ret = insert_dir_log_key(trans, log, path,
7517 old_dir_index, old_dir_index);
7519 mutex_unlock(&old_dir->log_mutex);
7521 btrfs_free_path(path);
7522 fscrypt_free_filename(&fname);
7527 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7528 ctx.logging_new_name = true;
7530 * We don't care about the return value. If we fail to log the new name
7531 * then we know the next attempt to sync the log will fallback to a full
7532 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7533 * we don't need to worry about getting a log committed that has an
7534 * inconsistent state after a rename operation.
7536 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7537 ASSERT(list_empty(&ctx.conflict_inodes));
7540 * If an error happened mark the log for a full commit because it's not
7541 * consistent and up to date or we couldn't find out if one of the
7542 * inodes was logged before in this transaction. Do it before unpinning
7543 * the log, to avoid any races with someone else trying to commit it.
7546 btrfs_set_log_full_commit(trans);
7548 btrfs_end_log_trans(root);