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>
17 #include "compression.h"
19 #include "block-group.h"
20 #include "space-info.h"
21 #include "inode-item.h"
23 #include "accessors.h"
24 #include "extent-tree.h"
25 #include "root-tree.h"
27 #include "file-item.h"
30 #include "tree-checker.h"
32 #define MAX_CONFLICT_INODES 10
34 /* magic values for the inode_only field in btrfs_log_inode:
36 * LOG_INODE_ALL means to log everything
37 * LOG_INODE_EXISTS means to log just enough to recreate the inode
46 * directory trouble cases
48 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
49 * log, we must force a full commit before doing an fsync of the directory
50 * where the unlink was done.
51 * ---> record transid of last unlink/rename per directory
55 * rename foo/some_dir foo2/some_dir
57 * fsync foo/some_dir/some_file
59 * The fsync above will unlink the original some_dir without recording
60 * it in its new location (foo2). After a crash, some_dir will be gone
61 * unless the fsync of some_file forces a full commit
63 * 2) we must log any new names for any file or dir that is in the fsync
64 * log. ---> check inode while renaming/linking.
66 * 2a) we must log any new names for any file or dir during rename
67 * when the directory they are being removed from was logged.
68 * ---> check inode and old parent dir during rename
70 * 2a is actually the more important variant. With the extra logging
71 * a crash might unlink the old name without recreating the new one
73 * 3) after a crash, we must go through any directories with a link count
74 * of zero and redo the rm -rf
81 * The directory f1 was fully removed from the FS, but fsync was never
82 * called on f1, only its parent dir. After a crash the rm -rf must
83 * be replayed. This must be able to recurse down the entire
84 * directory tree. The inode link count fixup code takes care of the
89 * stages for the tree walking. The first
90 * stage (0) is to only pin down the blocks we find
91 * the second stage (1) is to make sure that all the inodes
92 * we find in the log are created in the subvolume.
94 * The last stage is to deal with directories and links and extents
95 * and all the other fun semantics
99 LOG_WALK_REPLAY_INODES,
100 LOG_WALK_REPLAY_DIR_INDEX,
104 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
105 struct btrfs_inode *inode,
107 struct btrfs_log_ctx *ctx);
108 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root,
110 struct btrfs_path *path, u64 objectid);
111 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root,
113 struct btrfs_root *log,
114 struct btrfs_path *path,
115 u64 dirid, int del_all);
116 static void wait_log_commit(struct btrfs_root *root, int transid);
119 * tree logging is a special write ahead log used to make sure that
120 * fsyncs and O_SYNCs can happen without doing full tree commits.
122 * Full tree commits are expensive because they require commonly
123 * modified blocks to be recowed, creating many dirty pages in the
124 * extent tree an 4x-6x higher write load than ext3.
126 * Instead of doing a tree commit on every fsync, we use the
127 * key ranges and transaction ids to find items for a given file or directory
128 * that have changed in this transaction. Those items are copied into
129 * a special tree (one per subvolume root), that tree is written to disk
130 * and then the fsync is considered complete.
132 * After a crash, items are copied out of the log-tree back into the
133 * subvolume tree. Any file data extents found are recorded in the extent
134 * allocation tree, and the log-tree freed.
136 * The log tree is read three times, once to pin down all the extents it is
137 * using in ram and once, once to create all the inodes logged in the tree
138 * and once to do all the other items.
142 * start a sub transaction and setup the log tree
143 * this increments the log tree writer count to make the people
144 * syncing the tree wait for us to finish
146 static int start_log_trans(struct btrfs_trans_handle *trans,
147 struct btrfs_root *root,
148 struct btrfs_log_ctx *ctx)
150 struct btrfs_fs_info *fs_info = root->fs_info;
151 struct btrfs_root *tree_root = fs_info->tree_root;
152 const bool zoned = btrfs_is_zoned(fs_info);
154 bool created = false;
157 * First check if the log root tree was already created. If not, create
158 * it before locking the root's log_mutex, just to keep lockdep happy.
160 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
161 mutex_lock(&tree_root->log_mutex);
162 if (!fs_info->log_root_tree) {
163 ret = btrfs_init_log_root_tree(trans, fs_info);
165 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
169 mutex_unlock(&tree_root->log_mutex);
174 mutex_lock(&root->log_mutex);
177 if (root->log_root) {
178 int index = (root->log_transid + 1) % 2;
180 if (btrfs_need_log_full_commit(trans)) {
181 ret = BTRFS_LOG_FORCE_COMMIT;
185 if (zoned && atomic_read(&root->log_commit[index])) {
186 wait_log_commit(root, root->log_transid - 1);
190 if (!root->log_start_pid) {
191 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
192 root->log_start_pid = current->pid;
193 } else if (root->log_start_pid != current->pid) {
194 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
198 * This means fs_info->log_root_tree was already created
199 * for some other FS trees. Do the full commit not to mix
200 * nodes from multiple log transactions to do sequential
203 if (zoned && !created) {
204 ret = BTRFS_LOG_FORCE_COMMIT;
208 ret = btrfs_add_log_tree(trans, root);
212 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
213 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
214 root->log_start_pid = current->pid;
217 atomic_inc(&root->log_writers);
218 if (!ctx->logging_new_name) {
219 int index = root->log_transid % 2;
220 list_add_tail(&ctx->list, &root->log_ctxs[index]);
221 ctx->log_transid = root->log_transid;
225 mutex_unlock(&root->log_mutex);
230 * returns 0 if there was a log transaction running and we were able
231 * to join, or returns -ENOENT if there were not transactions
234 static int join_running_log_trans(struct btrfs_root *root)
236 const bool zoned = btrfs_is_zoned(root->fs_info);
239 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
242 mutex_lock(&root->log_mutex);
244 if (root->log_root) {
245 int index = (root->log_transid + 1) % 2;
248 if (zoned && atomic_read(&root->log_commit[index])) {
249 wait_log_commit(root, root->log_transid - 1);
252 atomic_inc(&root->log_writers);
254 mutex_unlock(&root->log_mutex);
259 * This either makes the current running log transaction wait
260 * until you call btrfs_end_log_trans() or it makes any future
261 * log transactions wait until you call btrfs_end_log_trans()
263 void btrfs_pin_log_trans(struct btrfs_root *root)
265 atomic_inc(&root->log_writers);
269 * indicate we're done making changes to the log tree
270 * and wake up anyone waiting to do a sync
272 void btrfs_end_log_trans(struct btrfs_root *root)
274 if (atomic_dec_and_test(&root->log_writers)) {
275 /* atomic_dec_and_test implies a barrier */
276 cond_wake_up_nomb(&root->log_writer_wait);
281 * the walk control struct is used to pass state down the chain when
282 * processing the log tree. The stage field tells us which part
283 * of the log tree processing we are currently doing. The others
284 * are state fields used for that specific part
286 struct walk_control {
287 /* should we free the extent on disk when done? This is used
288 * at transaction commit time while freeing a log tree
292 /* pin only walk, we record which extents on disk belong to the
297 /* what stage of the replay code we're currently in */
301 * Ignore any items from the inode currently being processed. Needs
302 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
303 * the LOG_WALK_REPLAY_INODES stage.
305 bool ignore_cur_inode;
307 /* the root we are currently replaying */
308 struct btrfs_root *replay_dest;
310 /* the trans handle for the current replay */
311 struct btrfs_trans_handle *trans;
313 /* the function that gets used to process blocks we find in the
314 * tree. Note the extent_buffer might not be up to date when it is
315 * passed in, and it must be checked or read if you need the data
318 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
319 struct walk_control *wc, u64 gen, int level);
323 * process_func used to pin down extents, write them or wait on them
325 static int process_one_buffer(struct btrfs_root *log,
326 struct extent_buffer *eb,
327 struct walk_control *wc, u64 gen, int level)
329 struct btrfs_fs_info *fs_info = log->fs_info;
333 * If this fs is mixed then we need to be able to process the leaves to
334 * pin down any logged extents, so we have to read the block.
336 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
337 struct btrfs_tree_parent_check check = {
342 ret = btrfs_read_extent_buffer(eb, &check);
348 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb);
352 if (btrfs_buffer_uptodate(eb, gen, 0) &&
353 btrfs_header_level(eb) == 0)
354 ret = btrfs_exclude_logged_extents(eb);
360 * Item overwrite used by replay and tree logging. eb, slot and key all refer
361 * to the src data we are copying out.
363 * root is the tree we are copying into, and path is a scratch
364 * path for use in this function (it should be released on entry and
365 * will be released on exit).
367 * If the key is already in the destination tree the existing item is
368 * overwritten. If the existing item isn't big enough, it is extended.
369 * If it is too large, it is truncated.
371 * If the key isn't in the destination yet, a new item is inserted.
373 static int overwrite_item(struct btrfs_trans_handle *trans,
374 struct btrfs_root *root,
375 struct btrfs_path *path,
376 struct extent_buffer *eb, int slot,
377 struct btrfs_key *key)
381 u64 saved_i_size = 0;
382 int save_old_i_size = 0;
383 unsigned long src_ptr;
384 unsigned long dst_ptr;
385 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
388 * This is only used during log replay, so the root is always from a
389 * fs/subvolume tree. In case we ever need to support a log root, then
390 * we'll have to clone the leaf in the path, release the path and use
391 * the leaf before writing into the log tree. See the comments at
392 * copy_items() for more details.
394 ASSERT(root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
396 item_size = btrfs_item_size(eb, slot);
397 src_ptr = btrfs_item_ptr_offset(eb, slot);
399 /* Look for the key in the destination tree. */
400 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
407 u32 dst_size = btrfs_item_size(path->nodes[0],
409 if (dst_size != item_size)
412 if (item_size == 0) {
413 btrfs_release_path(path);
416 dst_copy = kmalloc(item_size, GFP_NOFS);
417 src_copy = kmalloc(item_size, GFP_NOFS);
418 if (!dst_copy || !src_copy) {
419 btrfs_release_path(path);
425 read_extent_buffer(eb, src_copy, src_ptr, item_size);
427 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
428 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
430 ret = memcmp(dst_copy, src_copy, item_size);
435 * they have the same contents, just return, this saves
436 * us from cowing blocks in the destination tree and doing
437 * extra writes that may not have been done by a previous
441 btrfs_release_path(path);
446 * We need to load the old nbytes into the inode so when we
447 * replay the extents we've logged we get the right nbytes.
450 struct btrfs_inode_item *item;
454 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
455 struct btrfs_inode_item);
456 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
457 item = btrfs_item_ptr(eb, slot,
458 struct btrfs_inode_item);
459 btrfs_set_inode_nbytes(eb, item, nbytes);
462 * If this is a directory we need to reset the i_size to
463 * 0 so that we can set it up properly when replaying
464 * the rest of the items in this log.
466 mode = btrfs_inode_mode(eb, item);
468 btrfs_set_inode_size(eb, item, 0);
470 } else if (inode_item) {
471 struct btrfs_inode_item *item;
475 * New inode, set nbytes to 0 so that the nbytes comes out
476 * properly when we replay the extents.
478 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
479 btrfs_set_inode_nbytes(eb, item, 0);
482 * If this is a directory we need to reset the i_size to 0 so
483 * that we can set it up properly when replaying the rest of
484 * the items in this log.
486 mode = btrfs_inode_mode(eb, item);
488 btrfs_set_inode_size(eb, item, 0);
491 btrfs_release_path(path);
492 /* try to insert the key into the destination tree */
493 path->skip_release_on_error = 1;
494 ret = btrfs_insert_empty_item(trans, root, path,
496 path->skip_release_on_error = 0;
498 /* make sure any existing item is the correct size */
499 if (ret == -EEXIST || ret == -EOVERFLOW) {
501 found_size = btrfs_item_size(path->nodes[0],
503 if (found_size > item_size)
504 btrfs_truncate_item(trans, path, item_size, 1);
505 else if (found_size < item_size)
506 btrfs_extend_item(trans, path, item_size - found_size);
510 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
513 /* don't overwrite an existing inode if the generation number
514 * was logged as zero. This is done when the tree logging code
515 * is just logging an inode to make sure it exists after recovery.
517 * Also, don't overwrite i_size on directories during replay.
518 * log replay inserts and removes directory items based on the
519 * state of the tree found in the subvolume, and i_size is modified
522 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
523 struct btrfs_inode_item *src_item;
524 struct btrfs_inode_item *dst_item;
526 src_item = (struct btrfs_inode_item *)src_ptr;
527 dst_item = (struct btrfs_inode_item *)dst_ptr;
529 if (btrfs_inode_generation(eb, src_item) == 0) {
530 struct extent_buffer *dst_eb = path->nodes[0];
531 const u64 ino_size = btrfs_inode_size(eb, src_item);
534 * For regular files an ino_size == 0 is used only when
535 * logging that an inode exists, as part of a directory
536 * fsync, and the inode wasn't fsynced before. In this
537 * case don't set the size of the inode in the fs/subvol
538 * tree, otherwise we would be throwing valid data away.
540 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
541 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
543 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
547 if (S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
548 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
550 saved_i_size = btrfs_inode_size(path->nodes[0],
555 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
558 if (save_old_i_size) {
559 struct btrfs_inode_item *dst_item;
560 dst_item = (struct btrfs_inode_item *)dst_ptr;
561 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
564 /* make sure the generation is filled in */
565 if (key->type == BTRFS_INODE_ITEM_KEY) {
566 struct btrfs_inode_item *dst_item;
567 dst_item = (struct btrfs_inode_item *)dst_ptr;
568 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
569 btrfs_set_inode_generation(path->nodes[0], dst_item,
574 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
575 btrfs_release_path(path);
579 static int read_alloc_one_name(struct extent_buffer *eb, void *start, int len,
580 struct fscrypt_str *name)
584 buf = kmalloc(len, GFP_NOFS);
588 read_extent_buffer(eb, buf, (unsigned long)start, len);
595 * simple helper to read an inode off the disk from a given root
596 * This can only be called for subvolume roots and not for the log
598 static noinline struct inode *read_one_inode(struct btrfs_root *root,
603 inode = btrfs_iget(root->fs_info->sb, objectid, root);
609 /* replays a single extent in 'eb' at 'slot' with 'key' into the
610 * subvolume 'root'. path is released on entry and should be released
613 * extents in the log tree have not been allocated out of the extent
614 * tree yet. So, this completes the allocation, taking a reference
615 * as required if the extent already exists or creating a new extent
616 * if it isn't in the extent allocation tree yet.
618 * The extent is inserted into the file, dropping any existing extents
619 * from the file that overlap the new one.
621 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root,
623 struct btrfs_path *path,
624 struct extent_buffer *eb, int slot,
625 struct btrfs_key *key)
627 struct btrfs_drop_extents_args drop_args = { 0 };
628 struct btrfs_fs_info *fs_info = root->fs_info;
631 u64 start = key->offset;
633 struct btrfs_file_extent_item *item;
634 struct inode *inode = NULL;
638 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
639 found_type = btrfs_file_extent_type(eb, item);
641 if (found_type == BTRFS_FILE_EXTENT_REG ||
642 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
643 nbytes = btrfs_file_extent_num_bytes(eb, item);
644 extent_end = start + nbytes;
647 * We don't add to the inodes nbytes if we are prealloc or a
650 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
652 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
653 size = btrfs_file_extent_ram_bytes(eb, item);
654 nbytes = btrfs_file_extent_ram_bytes(eb, item);
655 extent_end = ALIGN(start + size,
656 fs_info->sectorsize);
662 inode = read_one_inode(root, key->objectid);
669 * first check to see if we already have this extent in the
670 * file. This must be done before the btrfs_drop_extents run
671 * so we don't try to drop this extent.
673 ret = btrfs_lookup_file_extent(trans, root, path,
674 btrfs_ino(BTRFS_I(inode)), start, 0);
677 (found_type == BTRFS_FILE_EXTENT_REG ||
678 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
679 struct btrfs_file_extent_item cmp1;
680 struct btrfs_file_extent_item cmp2;
681 struct btrfs_file_extent_item *existing;
682 struct extent_buffer *leaf;
684 leaf = path->nodes[0];
685 existing = btrfs_item_ptr(leaf, path->slots[0],
686 struct btrfs_file_extent_item);
688 read_extent_buffer(eb, &cmp1, (unsigned long)item,
690 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
694 * we already have a pointer to this exact extent,
695 * we don't have to do anything
697 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
698 btrfs_release_path(path);
702 btrfs_release_path(path);
704 /* drop any overlapping extents */
705 drop_args.start = start;
706 drop_args.end = extent_end;
707 drop_args.drop_cache = true;
708 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
712 if (found_type == BTRFS_FILE_EXTENT_REG ||
713 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
715 unsigned long dest_offset;
716 struct btrfs_key ins;
718 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
719 btrfs_fs_incompat(fs_info, NO_HOLES))
722 ret = btrfs_insert_empty_item(trans, root, path, key,
726 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
728 copy_extent_buffer(path->nodes[0], eb, dest_offset,
729 (unsigned long)item, sizeof(*item));
731 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
732 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
733 ins.type = BTRFS_EXTENT_ITEM_KEY;
734 offset = key->offset - btrfs_file_extent_offset(eb, item);
737 * Manually record dirty extent, as here we did a shallow
738 * file extent item copy and skip normal backref update,
739 * but modifying extent tree all by ourselves.
740 * So need to manually record dirty extent for qgroup,
741 * as the owner of the file extent changed from log tree
742 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
744 ret = btrfs_qgroup_trace_extent(trans,
745 btrfs_file_extent_disk_bytenr(eb, item),
746 btrfs_file_extent_disk_num_bytes(eb, item));
750 if (ins.objectid > 0) {
751 struct btrfs_ref ref = { 0 };
754 LIST_HEAD(ordered_sums);
757 * is this extent already allocated in the extent
758 * allocation tree? If so, just add a reference
760 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
764 } else if (ret == 0) {
765 btrfs_init_generic_ref(&ref,
766 BTRFS_ADD_DELAYED_REF,
767 ins.objectid, ins.offset, 0,
768 root->root_key.objectid);
769 btrfs_init_data_ref(&ref,
770 root->root_key.objectid,
771 key->objectid, offset, 0, false);
772 ret = btrfs_inc_extent_ref(trans, &ref);
777 * insert the extent pointer in the extent
780 ret = btrfs_alloc_logged_file_extent(trans,
781 root->root_key.objectid,
782 key->objectid, offset, &ins);
786 btrfs_release_path(path);
788 if (btrfs_file_extent_compression(eb, item)) {
789 csum_start = ins.objectid;
790 csum_end = csum_start + ins.offset;
792 csum_start = ins.objectid +
793 btrfs_file_extent_offset(eb, item);
794 csum_end = csum_start +
795 btrfs_file_extent_num_bytes(eb, item);
798 ret = btrfs_lookup_csums_list(root->log_root,
799 csum_start, csum_end - 1,
800 &ordered_sums, 0, false);
804 * Now delete all existing cums in the csum root that
805 * cover our range. We do this because we can have an
806 * extent that is completely referenced by one file
807 * extent item and partially referenced by another
808 * file extent item (like after using the clone or
809 * extent_same ioctls). In this case if we end up doing
810 * the replay of the one that partially references the
811 * extent first, and we do not do the csum deletion
812 * below, we can get 2 csum items in the csum tree that
813 * overlap each other. For example, imagine our log has
814 * the two following file extent items:
816 * key (257 EXTENT_DATA 409600)
817 * extent data disk byte 12845056 nr 102400
818 * extent data offset 20480 nr 20480 ram 102400
820 * key (257 EXTENT_DATA 819200)
821 * extent data disk byte 12845056 nr 102400
822 * extent data offset 0 nr 102400 ram 102400
824 * Where the second one fully references the 100K extent
825 * that starts at disk byte 12845056, and the log tree
826 * has a single csum item that covers the entire range
829 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
831 * After the first file extent item is replayed, the
832 * csum tree gets the following csum item:
834 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
836 * Which covers the 20K sub-range starting at offset 20K
837 * of our extent. Now when we replay the second file
838 * extent item, if we do not delete existing csum items
839 * that cover any of its blocks, we end up getting two
840 * csum items in our csum tree that overlap each other:
842 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
843 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
845 * Which is a problem, because after this anyone trying
846 * to lookup up for the checksum of any block of our
847 * extent starting at an offset of 40K or higher, will
848 * end up looking at the second csum item only, which
849 * does not contain the checksum for any block starting
850 * at offset 40K or higher of our extent.
852 while (!list_empty(&ordered_sums)) {
853 struct btrfs_ordered_sum *sums;
854 struct btrfs_root *csum_root;
856 sums = list_entry(ordered_sums.next,
857 struct btrfs_ordered_sum,
859 csum_root = btrfs_csum_root(fs_info,
862 ret = btrfs_del_csums(trans, csum_root,
866 ret = btrfs_csum_file_blocks(trans,
869 list_del(&sums->list);
875 btrfs_release_path(path);
877 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
878 /* inline extents are easy, we just overwrite them */
879 ret = overwrite_item(trans, root, path, eb, slot, key);
884 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
890 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
891 ret = btrfs_update_inode(trans, BTRFS_I(inode));
897 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
898 struct btrfs_inode *dir,
899 struct btrfs_inode *inode,
900 const struct fscrypt_str *name)
904 ret = btrfs_unlink_inode(trans, dir, inode, name);
908 * Whenever we need to check if a name exists or not, we check the
909 * fs/subvolume tree. So after an unlink we must run delayed items, so
910 * that future checks for a name during log replay see that the name
911 * does not exists anymore.
913 return btrfs_run_delayed_items(trans);
917 * when cleaning up conflicts between the directory names in the
918 * subvolume, directory names in the log and directory names in the
919 * inode back references, we may have to unlink inodes from directories.
921 * This is a helper function to do the unlink of a specific directory
924 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
925 struct btrfs_path *path,
926 struct btrfs_inode *dir,
927 struct btrfs_dir_item *di)
929 struct btrfs_root *root = dir->root;
931 struct fscrypt_str name;
932 struct extent_buffer *leaf;
933 struct btrfs_key location;
936 leaf = path->nodes[0];
938 btrfs_dir_item_key_to_cpu(leaf, di, &location);
939 ret = read_alloc_one_name(leaf, di + 1, btrfs_dir_name_len(leaf, di), &name);
943 btrfs_release_path(path);
945 inode = read_one_inode(root, location.objectid);
951 ret = link_to_fixup_dir(trans, root, path, location.objectid);
955 ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), &name);
963 * See if a given name and sequence number found in an inode back reference are
964 * already in a directory and correctly point to this inode.
966 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
969 static noinline int inode_in_dir(struct btrfs_root *root,
970 struct btrfs_path *path,
971 u64 dirid, u64 objectid, u64 index,
972 struct fscrypt_str *name)
974 struct btrfs_dir_item *di;
975 struct btrfs_key location;
978 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
984 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
985 if (location.objectid != objectid)
991 btrfs_release_path(path);
992 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, 0);
997 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
998 if (location.objectid == objectid)
1002 btrfs_release_path(path);
1007 * helper function to check a log tree for a named back reference in
1008 * an inode. This is used to decide if a back reference that is
1009 * found in the subvolume conflicts with what we find in the log.
1011 * inode backreferences may have multiple refs in a single item,
1012 * during replay we process one reference at a time, and we don't
1013 * want to delete valid links to a file from the subvolume if that
1014 * link is also in the log.
1016 static noinline int backref_in_log(struct btrfs_root *log,
1017 struct btrfs_key *key,
1019 const struct fscrypt_str *name)
1021 struct btrfs_path *path;
1024 path = btrfs_alloc_path();
1028 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1031 } else if (ret == 1) {
1036 if (key->type == BTRFS_INODE_EXTREF_KEY)
1037 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1039 ref_objectid, name);
1041 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1042 path->slots[0], name);
1044 btrfs_free_path(path);
1048 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1049 struct btrfs_root *root,
1050 struct btrfs_path *path,
1051 struct btrfs_root *log_root,
1052 struct btrfs_inode *dir,
1053 struct btrfs_inode *inode,
1054 u64 inode_objectid, u64 parent_objectid,
1055 u64 ref_index, struct fscrypt_str *name)
1058 struct extent_buffer *leaf;
1059 struct btrfs_dir_item *di;
1060 struct btrfs_key search_key;
1061 struct btrfs_inode_extref *extref;
1064 /* Search old style refs */
1065 search_key.objectid = inode_objectid;
1066 search_key.type = BTRFS_INODE_REF_KEY;
1067 search_key.offset = parent_objectid;
1068 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1070 struct btrfs_inode_ref *victim_ref;
1072 unsigned long ptr_end;
1074 leaf = path->nodes[0];
1076 /* are we trying to overwrite a back ref for the root directory
1077 * if so, just jump out, we're done
1079 if (search_key.objectid == search_key.offset)
1082 /* check all the names in this back reference to see
1083 * if they are in the log. if so, we allow them to stay
1084 * otherwise they must be unlinked as a conflict
1086 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1087 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1088 while (ptr < ptr_end) {
1089 struct fscrypt_str victim_name;
1091 victim_ref = (struct btrfs_inode_ref *)ptr;
1092 ret = read_alloc_one_name(leaf, (victim_ref + 1),
1093 btrfs_inode_ref_name_len(leaf, victim_ref),
1098 ret = backref_in_log(log_root, &search_key,
1099 parent_objectid, &victim_name);
1101 kfree(victim_name.name);
1104 inc_nlink(&inode->vfs_inode);
1105 btrfs_release_path(path);
1107 ret = unlink_inode_for_log_replay(trans, dir, inode,
1109 kfree(victim_name.name);
1114 kfree(victim_name.name);
1116 ptr = (unsigned long)(victim_ref + 1) + victim_name.len;
1119 btrfs_release_path(path);
1121 /* Same search but for extended refs */
1122 extref = btrfs_lookup_inode_extref(NULL, root, path, name,
1123 inode_objectid, parent_objectid, 0,
1125 if (IS_ERR(extref)) {
1126 return PTR_ERR(extref);
1127 } else if (extref) {
1131 struct inode *victim_parent;
1133 leaf = path->nodes[0];
1135 item_size = btrfs_item_size(leaf, path->slots[0]);
1136 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1138 while (cur_offset < item_size) {
1139 struct fscrypt_str victim_name;
1141 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1143 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1146 ret = read_alloc_one_name(leaf, &extref->name,
1147 btrfs_inode_extref_name_len(leaf, extref),
1152 search_key.objectid = inode_objectid;
1153 search_key.type = BTRFS_INODE_EXTREF_KEY;
1154 search_key.offset = btrfs_extref_hash(parent_objectid,
1157 ret = backref_in_log(log_root, &search_key,
1158 parent_objectid, &victim_name);
1160 kfree(victim_name.name);
1164 victim_parent = read_one_inode(root,
1166 if (victim_parent) {
1167 inc_nlink(&inode->vfs_inode);
1168 btrfs_release_path(path);
1170 ret = unlink_inode_for_log_replay(trans,
1171 BTRFS_I(victim_parent),
1172 inode, &victim_name);
1174 iput(victim_parent);
1175 kfree(victim_name.name);
1180 kfree(victim_name.name);
1182 cur_offset += victim_name.len + sizeof(*extref);
1185 btrfs_release_path(path);
1187 /* look for a conflicting sequence number */
1188 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1189 ref_index, name, 0);
1193 ret = drop_one_dir_item(trans, path, dir, di);
1197 btrfs_release_path(path);
1199 /* look for a conflicting name */
1200 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), name, 0);
1204 ret = drop_one_dir_item(trans, path, dir, di);
1208 btrfs_release_path(path);
1213 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1214 struct fscrypt_str *name, u64 *index,
1215 u64 *parent_objectid)
1217 struct btrfs_inode_extref *extref;
1220 extref = (struct btrfs_inode_extref *)ref_ptr;
1222 ret = read_alloc_one_name(eb, &extref->name,
1223 btrfs_inode_extref_name_len(eb, extref), name);
1228 *index = btrfs_inode_extref_index(eb, extref);
1229 if (parent_objectid)
1230 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1235 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1236 struct fscrypt_str *name, u64 *index)
1238 struct btrfs_inode_ref *ref;
1241 ref = (struct btrfs_inode_ref *)ref_ptr;
1243 ret = read_alloc_one_name(eb, ref + 1, btrfs_inode_ref_name_len(eb, ref),
1249 *index = btrfs_inode_ref_index(eb, ref);
1255 * Take an inode reference item from the log tree and iterate all names from the
1256 * inode reference item in the subvolume tree with the same key (if it exists).
1257 * For any name that is not in the inode reference item from the log tree, do a
1258 * proper unlink of that name (that is, remove its entry from the inode
1259 * reference item and both dir index keys).
1261 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 struct btrfs_inode *inode,
1265 struct extent_buffer *log_eb,
1267 struct btrfs_key *key)
1270 unsigned long ref_ptr;
1271 unsigned long ref_end;
1272 struct extent_buffer *eb;
1275 btrfs_release_path(path);
1276 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1284 eb = path->nodes[0];
1285 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1286 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1287 while (ref_ptr < ref_end) {
1288 struct fscrypt_str name;
1291 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1292 ret = extref_get_fields(eb, ref_ptr, &name,
1295 parent_id = key->offset;
1296 ret = ref_get_fields(eb, ref_ptr, &name, NULL);
1301 if (key->type == BTRFS_INODE_EXTREF_KEY)
1302 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1305 ret = !!btrfs_find_name_in_backref(log_eb, log_slot, &name);
1310 btrfs_release_path(path);
1311 dir = read_one_inode(root, parent_id);
1317 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1327 ref_ptr += name.len;
1328 if (key->type == BTRFS_INODE_EXTREF_KEY)
1329 ref_ptr += sizeof(struct btrfs_inode_extref);
1331 ref_ptr += sizeof(struct btrfs_inode_ref);
1335 btrfs_release_path(path);
1340 * replay one inode back reference item found in the log tree.
1341 * eb, slot and key refer to the buffer and key found in the log tree.
1342 * root is the destination we are replaying into, and path is for temp
1343 * use by this function. (it should be released on return).
1345 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1346 struct btrfs_root *root,
1347 struct btrfs_root *log,
1348 struct btrfs_path *path,
1349 struct extent_buffer *eb, int slot,
1350 struct btrfs_key *key)
1352 struct inode *dir = NULL;
1353 struct inode *inode = NULL;
1354 unsigned long ref_ptr;
1355 unsigned long ref_end;
1356 struct fscrypt_str name;
1358 int log_ref_ver = 0;
1359 u64 parent_objectid;
1362 int ref_struct_size;
1364 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1365 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1367 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1368 struct btrfs_inode_extref *r;
1370 ref_struct_size = sizeof(struct btrfs_inode_extref);
1372 r = (struct btrfs_inode_extref *)ref_ptr;
1373 parent_objectid = btrfs_inode_extref_parent(eb, r);
1375 ref_struct_size = sizeof(struct btrfs_inode_ref);
1376 parent_objectid = key->offset;
1378 inode_objectid = key->objectid;
1381 * it is possible that we didn't log all the parent directories
1382 * for a given inode. If we don't find the dir, just don't
1383 * copy the back ref in. The link count fixup code will take
1386 dir = read_one_inode(root, parent_objectid);
1392 inode = read_one_inode(root, inode_objectid);
1398 while (ref_ptr < ref_end) {
1400 ret = extref_get_fields(eb, ref_ptr, &name,
1401 &ref_index, &parent_objectid);
1403 * parent object can change from one array
1407 dir = read_one_inode(root, parent_objectid);
1413 ret = ref_get_fields(eb, ref_ptr, &name, &ref_index);
1418 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1419 btrfs_ino(BTRFS_I(inode)), ref_index, &name);
1422 } else if (ret == 0) {
1424 * look for a conflicting back reference in the
1425 * metadata. if we find one we have to unlink that name
1426 * of the file before we add our new link. Later on, we
1427 * overwrite any existing back reference, and we don't
1428 * want to create dangling pointers in the directory.
1430 ret = __add_inode_ref(trans, root, path, log,
1431 BTRFS_I(dir), BTRFS_I(inode),
1432 inode_objectid, parent_objectid,
1440 /* insert our name */
1441 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1442 &name, 0, ref_index);
1446 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1450 /* Else, ret == 1, we already have a perfect match, we're done. */
1452 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + name.len;
1462 * Before we overwrite the inode reference item in the subvolume tree
1463 * with the item from the log tree, we must unlink all names from the
1464 * parent directory that are in the subvolume's tree inode reference
1465 * item, otherwise we end up with an inconsistent subvolume tree where
1466 * dir index entries exist for a name but there is no inode reference
1467 * item with the same name.
1469 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1474 /* finally write the back reference in the inode */
1475 ret = overwrite_item(trans, root, path, eb, slot, key);
1477 btrfs_release_path(path);
1484 static int count_inode_extrefs(struct btrfs_inode *inode, struct btrfs_path *path)
1488 unsigned int nlink = 0;
1491 u64 inode_objectid = btrfs_ino(inode);
1494 struct btrfs_inode_extref *extref;
1495 struct extent_buffer *leaf;
1498 ret = btrfs_find_one_extref(inode->root, inode_objectid, offset,
1499 path, &extref, &offset);
1503 leaf = path->nodes[0];
1504 item_size = btrfs_item_size(leaf, path->slots[0]);
1505 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1508 while (cur_offset < item_size) {
1509 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1510 name_len = btrfs_inode_extref_name_len(leaf, extref);
1514 cur_offset += name_len + sizeof(*extref);
1518 btrfs_release_path(path);
1520 btrfs_release_path(path);
1522 if (ret < 0 && ret != -ENOENT)
1527 static int count_inode_refs(struct btrfs_inode *inode, struct btrfs_path *path)
1530 struct btrfs_key key;
1531 unsigned int nlink = 0;
1533 unsigned long ptr_end;
1535 u64 ino = btrfs_ino(inode);
1538 key.type = BTRFS_INODE_REF_KEY;
1539 key.offset = (u64)-1;
1542 ret = btrfs_search_slot(NULL, inode->root, &key, path, 0, 0);
1546 if (path->slots[0] == 0)
1551 btrfs_item_key_to_cpu(path->nodes[0], &key,
1553 if (key.objectid != ino ||
1554 key.type != BTRFS_INODE_REF_KEY)
1556 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1557 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1559 while (ptr < ptr_end) {
1560 struct btrfs_inode_ref *ref;
1562 ref = (struct btrfs_inode_ref *)ptr;
1563 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1565 ptr = (unsigned long)(ref + 1) + name_len;
1569 if (key.offset == 0)
1571 if (path->slots[0] > 0) {
1576 btrfs_release_path(path);
1578 btrfs_release_path(path);
1584 * There are a few corners where the link count of the file can't
1585 * be properly maintained during replay. So, instead of adding
1586 * lots of complexity to the log code, we just scan the backrefs
1587 * for any file that has been through replay.
1589 * The scan will update the link count on the inode to reflect the
1590 * number of back refs found. If it goes down to zero, the iput
1591 * will free the inode.
1593 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1594 struct inode *inode)
1596 struct btrfs_root *root = BTRFS_I(inode)->root;
1597 struct btrfs_path *path;
1600 u64 ino = btrfs_ino(BTRFS_I(inode));
1602 path = btrfs_alloc_path();
1606 ret = count_inode_refs(BTRFS_I(inode), path);
1612 ret = count_inode_extrefs(BTRFS_I(inode), path);
1620 if (nlink != inode->i_nlink) {
1621 set_nlink(inode, nlink);
1622 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1626 BTRFS_I(inode)->index_cnt = (u64)-1;
1628 if (inode->i_nlink == 0) {
1629 if (S_ISDIR(inode->i_mode)) {
1630 ret = replay_dir_deletes(trans, root, NULL, path,
1635 ret = btrfs_insert_orphan_item(trans, root, ino);
1641 btrfs_free_path(path);
1645 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1646 struct btrfs_root *root,
1647 struct btrfs_path *path)
1650 struct btrfs_key key;
1651 struct inode *inode;
1653 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1654 key.type = BTRFS_ORPHAN_ITEM_KEY;
1655 key.offset = (u64)-1;
1657 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1663 if (path->slots[0] == 0)
1668 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1669 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1670 key.type != BTRFS_ORPHAN_ITEM_KEY)
1673 ret = btrfs_del_item(trans, root, path);
1677 btrfs_release_path(path);
1678 inode = read_one_inode(root, key.offset);
1684 ret = fixup_inode_link_count(trans, inode);
1690 * fixup on a directory may create new entries,
1691 * make sure we always look for the highset possible
1694 key.offset = (u64)-1;
1696 btrfs_release_path(path);
1702 * record a given inode in the fixup dir so we can check its link
1703 * count when replay is done. The link count is incremented here
1704 * so the inode won't go away until we check it
1706 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1707 struct btrfs_root *root,
1708 struct btrfs_path *path,
1711 struct btrfs_key key;
1713 struct inode *inode;
1715 inode = read_one_inode(root, objectid);
1719 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1720 key.type = BTRFS_ORPHAN_ITEM_KEY;
1721 key.offset = objectid;
1723 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1725 btrfs_release_path(path);
1727 if (!inode->i_nlink)
1728 set_nlink(inode, 1);
1731 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1732 } else if (ret == -EEXIST) {
1741 * when replaying the log for a directory, we only insert names
1742 * for inodes that actually exist. This means an fsync on a directory
1743 * does not implicitly fsync all the new files in it
1745 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1746 struct btrfs_root *root,
1747 u64 dirid, u64 index,
1748 const struct fscrypt_str *name,
1749 struct btrfs_key *location)
1751 struct inode *inode;
1755 inode = read_one_inode(root, location->objectid);
1759 dir = read_one_inode(root, dirid);
1765 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1768 /* FIXME, put inode into FIXUP list */
1775 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1776 struct btrfs_inode *dir,
1777 struct btrfs_path *path,
1778 struct btrfs_dir_item *dst_di,
1779 const struct btrfs_key *log_key,
1783 struct btrfs_key found_key;
1785 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1786 /* The existing dentry points to the same inode, don't delete it. */
1787 if (found_key.objectid == log_key->objectid &&
1788 found_key.type == log_key->type &&
1789 found_key.offset == log_key->offset &&
1790 btrfs_dir_flags(path->nodes[0], dst_di) == log_flags)
1794 * Don't drop the conflicting directory entry if the inode for the new
1795 * entry doesn't exist.
1800 return drop_one_dir_item(trans, path, dir, dst_di);
1804 * take a single entry in a log directory item and replay it into
1807 * if a conflicting item exists in the subdirectory already,
1808 * the inode it points to is unlinked and put into the link count
1811 * If a name from the log points to a file or directory that does
1812 * not exist in the FS, it is skipped. fsyncs on directories
1813 * do not force down inodes inside that directory, just changes to the
1814 * names or unlinks in a directory.
1816 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1817 * non-existing inode) and 1 if the name was replayed.
1819 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1820 struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 struct extent_buffer *eb,
1823 struct btrfs_dir_item *di,
1824 struct btrfs_key *key)
1826 struct fscrypt_str name;
1827 struct btrfs_dir_item *dir_dst_di;
1828 struct btrfs_dir_item *index_dst_di;
1829 bool dir_dst_matches = false;
1830 bool index_dst_matches = false;
1831 struct btrfs_key log_key;
1832 struct btrfs_key search_key;
1837 bool update_size = true;
1838 bool name_added = false;
1840 dir = read_one_inode(root, key->objectid);
1844 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
1848 log_flags = btrfs_dir_flags(eb, di);
1849 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1850 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1851 btrfs_release_path(path);
1854 exists = (ret == 0);
1857 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1859 if (IS_ERR(dir_dst_di)) {
1860 ret = PTR_ERR(dir_dst_di);
1862 } else if (dir_dst_di) {
1863 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1864 dir_dst_di, &log_key,
1868 dir_dst_matches = (ret == 1);
1871 btrfs_release_path(path);
1873 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1874 key->objectid, key->offset,
1876 if (IS_ERR(index_dst_di)) {
1877 ret = PTR_ERR(index_dst_di);
1879 } else if (index_dst_di) {
1880 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1881 index_dst_di, &log_key,
1885 index_dst_matches = (ret == 1);
1888 btrfs_release_path(path);
1890 if (dir_dst_matches && index_dst_matches) {
1892 update_size = false;
1897 * Check if the inode reference exists in the log for the given name,
1898 * inode and parent inode
1900 search_key.objectid = log_key.objectid;
1901 search_key.type = BTRFS_INODE_REF_KEY;
1902 search_key.offset = key->objectid;
1903 ret = backref_in_log(root->log_root, &search_key, 0, &name);
1907 /* The dentry will be added later. */
1909 update_size = false;
1913 search_key.objectid = log_key.objectid;
1914 search_key.type = BTRFS_INODE_EXTREF_KEY;
1915 search_key.offset = key->objectid;
1916 ret = backref_in_log(root->log_root, &search_key, key->objectid, &name);
1920 /* The dentry will be added later. */
1922 update_size = false;
1925 btrfs_release_path(path);
1926 ret = insert_one_name(trans, root, key->objectid, key->offset,
1928 if (ret && ret != -ENOENT && ret != -EEXIST)
1932 update_size = false;
1936 if (!ret && update_size) {
1937 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2);
1938 ret = btrfs_update_inode(trans, BTRFS_I(dir));
1942 if (!ret && name_added)
1947 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1948 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct btrfs_path *path,
1951 struct extent_buffer *eb, int slot,
1952 struct btrfs_key *key)
1955 struct btrfs_dir_item *di;
1957 /* We only log dir index keys, which only contain a single dir item. */
1958 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
1960 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1961 ret = replay_one_name(trans, root, path, eb, di, key);
1966 * If this entry refers to a non-directory (directories can not have a
1967 * link count > 1) and it was added in the transaction that was not
1968 * committed, make sure we fixup the link count of the inode the entry
1969 * points to. Otherwise something like the following would result in a
1970 * directory pointing to an inode with a wrong link that does not account
1971 * for this dir entry:
1978 * ln testdir/bar testdir/bar_link
1979 * ln testdir/foo testdir/foo_link
1980 * xfs_io -c "fsync" testdir/bar
1984 * mount fs, log replay happens
1986 * File foo would remain with a link count of 1 when it has two entries
1987 * pointing to it in the directory testdir. This would make it impossible
1988 * to ever delete the parent directory has it would result in stale
1989 * dentries that can never be deleted.
1991 if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) {
1992 struct btrfs_path *fixup_path;
1993 struct btrfs_key di_key;
1995 fixup_path = btrfs_alloc_path();
1999 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2000 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2001 btrfs_free_path(fixup_path);
2008 * directory replay has two parts. There are the standard directory
2009 * items in the log copied from the subvolume, and range items
2010 * created in the log while the subvolume was logged.
2012 * The range items tell us which parts of the key space the log
2013 * is authoritative for. During replay, if a key in the subvolume
2014 * directory is in a logged range item, but not actually in the log
2015 * that means it was deleted from the directory before the fsync
2016 * and should be removed.
2018 static noinline int find_dir_range(struct btrfs_root *root,
2019 struct btrfs_path *path,
2021 u64 *start_ret, u64 *end_ret)
2023 struct btrfs_key key;
2025 struct btrfs_dir_log_item *item;
2029 if (*start_ret == (u64)-1)
2032 key.objectid = dirid;
2033 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2034 key.offset = *start_ret;
2036 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2040 if (path->slots[0] == 0)
2045 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2047 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2051 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2052 struct btrfs_dir_log_item);
2053 found_end = btrfs_dir_log_end(path->nodes[0], item);
2055 if (*start_ret >= key.offset && *start_ret <= found_end) {
2057 *start_ret = key.offset;
2058 *end_ret = found_end;
2063 /* check the next slot in the tree to see if it is a valid item */
2064 nritems = btrfs_header_nritems(path->nodes[0]);
2066 if (path->slots[0] >= nritems) {
2067 ret = btrfs_next_leaf(root, path);
2072 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2074 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2078 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2079 struct btrfs_dir_log_item);
2080 found_end = btrfs_dir_log_end(path->nodes[0], item);
2081 *start_ret = key.offset;
2082 *end_ret = found_end;
2085 btrfs_release_path(path);
2090 * this looks for a given directory item in the log. If the directory
2091 * item is not in the log, the item is removed and the inode it points
2094 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2095 struct btrfs_root *log,
2096 struct btrfs_path *path,
2097 struct btrfs_path *log_path,
2099 struct btrfs_key *dir_key)
2101 struct btrfs_root *root = BTRFS_I(dir)->root;
2103 struct extent_buffer *eb;
2105 struct btrfs_dir_item *di;
2106 struct fscrypt_str name;
2107 struct inode *inode = NULL;
2108 struct btrfs_key location;
2111 * Currently we only log dir index keys. Even if we replay a log created
2112 * by an older kernel that logged both dir index and dir item keys, all
2113 * we need to do is process the dir index keys, we (and our caller) can
2114 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2116 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2118 eb = path->nodes[0];
2119 slot = path->slots[0];
2120 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2121 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
2126 struct btrfs_dir_item *log_di;
2128 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2130 dir_key->offset, &name, 0);
2131 if (IS_ERR(log_di)) {
2132 ret = PTR_ERR(log_di);
2134 } else if (log_di) {
2135 /* The dentry exists in the log, we have nothing to do. */
2141 btrfs_dir_item_key_to_cpu(eb, di, &location);
2142 btrfs_release_path(path);
2143 btrfs_release_path(log_path);
2144 inode = read_one_inode(root, location.objectid);
2150 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2155 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2158 * Unlike dir item keys, dir index keys can only have one name (entry) in
2159 * them, as there are no key collisions since each key has a unique offset
2160 * (an index number), so we're done.
2163 btrfs_release_path(path);
2164 btrfs_release_path(log_path);
2170 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *root,
2172 struct btrfs_root *log,
2173 struct btrfs_path *path,
2176 struct btrfs_key search_key;
2177 struct btrfs_path *log_path;
2182 log_path = btrfs_alloc_path();
2186 search_key.objectid = ino;
2187 search_key.type = BTRFS_XATTR_ITEM_KEY;
2188 search_key.offset = 0;
2190 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2194 nritems = btrfs_header_nritems(path->nodes[0]);
2195 for (i = path->slots[0]; i < nritems; i++) {
2196 struct btrfs_key key;
2197 struct btrfs_dir_item *di;
2198 struct btrfs_dir_item *log_di;
2202 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2203 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2208 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2209 total_size = btrfs_item_size(path->nodes[0], i);
2211 while (cur < total_size) {
2212 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2213 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2214 u32 this_len = sizeof(*di) + name_len + data_len;
2217 name = kmalloc(name_len, GFP_NOFS);
2222 read_extent_buffer(path->nodes[0], name,
2223 (unsigned long)(di + 1), name_len);
2225 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2227 btrfs_release_path(log_path);
2229 /* Doesn't exist in log tree, so delete it. */
2230 btrfs_release_path(path);
2231 di = btrfs_lookup_xattr(trans, root, path, ino,
2232 name, name_len, -1);
2239 ret = btrfs_delete_one_dir_name(trans, root,
2243 btrfs_release_path(path);
2248 if (IS_ERR(log_di)) {
2249 ret = PTR_ERR(log_di);
2253 di = (struct btrfs_dir_item *)((char *)di + this_len);
2256 ret = btrfs_next_leaf(root, path);
2262 btrfs_free_path(log_path);
2263 btrfs_release_path(path);
2269 * deletion replay happens before we copy any new directory items
2270 * out of the log or out of backreferences from inodes. It
2271 * scans the log to find ranges of keys that log is authoritative for,
2272 * and then scans the directory to find items in those ranges that are
2273 * not present in the log.
2275 * Anything we don't find in the log is unlinked and removed from the
2278 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2279 struct btrfs_root *root,
2280 struct btrfs_root *log,
2281 struct btrfs_path *path,
2282 u64 dirid, int del_all)
2287 struct btrfs_key dir_key;
2288 struct btrfs_key found_key;
2289 struct btrfs_path *log_path;
2292 dir_key.objectid = dirid;
2293 dir_key.type = BTRFS_DIR_INDEX_KEY;
2294 log_path = btrfs_alloc_path();
2298 dir = read_one_inode(root, dirid);
2299 /* it isn't an error if the inode isn't there, that can happen
2300 * because we replay the deletes before we copy in the inode item
2304 btrfs_free_path(log_path);
2312 range_end = (u64)-1;
2314 ret = find_dir_range(log, path, dirid,
2315 &range_start, &range_end);
2322 dir_key.offset = range_start;
2325 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2330 nritems = btrfs_header_nritems(path->nodes[0]);
2331 if (path->slots[0] >= nritems) {
2332 ret = btrfs_next_leaf(root, path);
2338 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2340 if (found_key.objectid != dirid ||
2341 found_key.type != dir_key.type) {
2346 if (found_key.offset > range_end)
2349 ret = check_item_in_log(trans, log, path,
2354 if (found_key.offset == (u64)-1)
2356 dir_key.offset = found_key.offset + 1;
2358 btrfs_release_path(path);
2359 if (range_end == (u64)-1)
2361 range_start = range_end + 1;
2365 btrfs_release_path(path);
2366 btrfs_free_path(log_path);
2372 * the process_func used to replay items from the log tree. This
2373 * gets called in two different stages. The first stage just looks
2374 * for inodes and makes sure they are all copied into the subvolume.
2376 * The second stage copies all the other item types from the log into
2377 * the subvolume. The two stage approach is slower, but gets rid of
2378 * lots of complexity around inodes referencing other inodes that exist
2379 * only in the log (references come from either directory items or inode
2382 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2383 struct walk_control *wc, u64 gen, int level)
2386 struct btrfs_tree_parent_check check = {
2390 struct btrfs_path *path;
2391 struct btrfs_root *root = wc->replay_dest;
2392 struct btrfs_key key;
2396 ret = btrfs_read_extent_buffer(eb, &check);
2400 level = btrfs_header_level(eb);
2405 path = btrfs_alloc_path();
2409 nritems = btrfs_header_nritems(eb);
2410 for (i = 0; i < nritems; i++) {
2411 btrfs_item_key_to_cpu(eb, &key, i);
2413 /* inode keys are done during the first stage */
2414 if (key.type == BTRFS_INODE_ITEM_KEY &&
2415 wc->stage == LOG_WALK_REPLAY_INODES) {
2416 struct btrfs_inode_item *inode_item;
2419 inode_item = btrfs_item_ptr(eb, i,
2420 struct btrfs_inode_item);
2422 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2423 * and never got linked before the fsync, skip it, as
2424 * replaying it is pointless since it would be deleted
2425 * later. We skip logging tmpfiles, but it's always
2426 * possible we are replaying a log created with a kernel
2427 * that used to log tmpfiles.
2429 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2430 wc->ignore_cur_inode = true;
2433 wc->ignore_cur_inode = false;
2435 ret = replay_xattr_deletes(wc->trans, root, log,
2436 path, key.objectid);
2439 mode = btrfs_inode_mode(eb, inode_item);
2440 if (S_ISDIR(mode)) {
2441 ret = replay_dir_deletes(wc->trans,
2442 root, log, path, key.objectid, 0);
2446 ret = overwrite_item(wc->trans, root, path,
2452 * Before replaying extents, truncate the inode to its
2453 * size. We need to do it now and not after log replay
2454 * because before an fsync we can have prealloc extents
2455 * added beyond the inode's i_size. If we did it after,
2456 * through orphan cleanup for example, we would drop
2457 * those prealloc extents just after replaying them.
2459 if (S_ISREG(mode)) {
2460 struct btrfs_drop_extents_args drop_args = { 0 };
2461 struct inode *inode;
2464 inode = read_one_inode(root, key.objectid);
2469 from = ALIGN(i_size_read(inode),
2470 root->fs_info->sectorsize);
2471 drop_args.start = from;
2472 drop_args.end = (u64)-1;
2473 drop_args.drop_cache = true;
2474 ret = btrfs_drop_extents(wc->trans, root,
2478 inode_sub_bytes(inode,
2479 drop_args.bytes_found);
2480 /* Update the inode's nbytes. */
2481 ret = btrfs_update_inode(wc->trans,
2489 ret = link_to_fixup_dir(wc->trans, root,
2490 path, key.objectid);
2495 if (wc->ignore_cur_inode)
2498 if (key.type == BTRFS_DIR_INDEX_KEY &&
2499 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2500 ret = replay_one_dir_item(wc->trans, root, path,
2506 if (wc->stage < LOG_WALK_REPLAY_ALL)
2509 /* these keys are simply copied */
2510 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2511 ret = overwrite_item(wc->trans, root, path,
2515 } else if (key.type == BTRFS_INODE_REF_KEY ||
2516 key.type == BTRFS_INODE_EXTREF_KEY) {
2517 ret = add_inode_ref(wc->trans, root, log, path,
2519 if (ret && ret != -ENOENT)
2522 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2523 ret = replay_one_extent(wc->trans, root, path,
2529 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2530 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2531 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2532 * older kernel with such keys, ignore them.
2535 btrfs_free_path(path);
2540 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2542 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2544 struct btrfs_block_group *cache;
2546 cache = btrfs_lookup_block_group(fs_info, start);
2548 btrfs_err(fs_info, "unable to find block group for %llu", start);
2552 spin_lock(&cache->space_info->lock);
2553 spin_lock(&cache->lock);
2554 cache->reserved -= fs_info->nodesize;
2555 cache->space_info->bytes_reserved -= fs_info->nodesize;
2556 spin_unlock(&cache->lock);
2557 spin_unlock(&cache->space_info->lock);
2559 btrfs_put_block_group(cache);
2562 static int clean_log_buffer(struct btrfs_trans_handle *trans,
2563 struct extent_buffer *eb)
2567 btrfs_tree_lock(eb);
2568 btrfs_clear_buffer_dirty(trans, eb);
2569 wait_on_extent_buffer_writeback(eb);
2570 btrfs_tree_unlock(eb);
2573 ret = btrfs_pin_reserved_extent(trans, eb);
2577 unaccount_log_buffer(eb->fs_info, eb->start);
2583 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2584 struct btrfs_root *root,
2585 struct btrfs_path *path, int *level,
2586 struct walk_control *wc)
2588 struct btrfs_fs_info *fs_info = root->fs_info;
2591 struct extent_buffer *next;
2592 struct extent_buffer *cur;
2595 while (*level > 0) {
2596 struct btrfs_tree_parent_check check = { 0 };
2598 cur = path->nodes[*level];
2600 WARN_ON(btrfs_header_level(cur) != *level);
2602 if (path->slots[*level] >=
2603 btrfs_header_nritems(cur))
2606 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2607 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2608 check.transid = ptr_gen;
2609 check.level = *level - 1;
2610 check.has_first_key = true;
2611 btrfs_node_key_to_cpu(cur, &check.first_key, path->slots[*level]);
2613 next = btrfs_find_create_tree_block(fs_info, bytenr,
2614 btrfs_header_owner(cur),
2617 return PTR_ERR(next);
2620 ret = wc->process_func(root, next, wc, ptr_gen,
2623 free_extent_buffer(next);
2627 path->slots[*level]++;
2629 ret = btrfs_read_extent_buffer(next, &check);
2631 free_extent_buffer(next);
2635 ret = clean_log_buffer(trans, next);
2637 free_extent_buffer(next);
2641 free_extent_buffer(next);
2644 ret = btrfs_read_extent_buffer(next, &check);
2646 free_extent_buffer(next);
2650 if (path->nodes[*level-1])
2651 free_extent_buffer(path->nodes[*level-1]);
2652 path->nodes[*level-1] = next;
2653 *level = btrfs_header_level(next);
2654 path->slots[*level] = 0;
2657 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2663 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2664 struct btrfs_root *root,
2665 struct btrfs_path *path, int *level,
2666 struct walk_control *wc)
2672 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2673 slot = path->slots[i];
2674 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2677 WARN_ON(*level == 0);
2680 ret = wc->process_func(root, path->nodes[*level], wc,
2681 btrfs_header_generation(path->nodes[*level]),
2687 ret = clean_log_buffer(trans, path->nodes[*level]);
2691 free_extent_buffer(path->nodes[*level]);
2692 path->nodes[*level] = NULL;
2700 * drop the reference count on the tree rooted at 'snap'. This traverses
2701 * the tree freeing any blocks that have a ref count of zero after being
2704 static int walk_log_tree(struct btrfs_trans_handle *trans,
2705 struct btrfs_root *log, struct walk_control *wc)
2710 struct btrfs_path *path;
2713 path = btrfs_alloc_path();
2717 level = btrfs_header_level(log->node);
2719 path->nodes[level] = log->node;
2720 atomic_inc(&log->node->refs);
2721 path->slots[level] = 0;
2724 wret = walk_down_log_tree(trans, log, path, &level, wc);
2732 wret = walk_up_log_tree(trans, log, path, &level, wc);
2741 /* was the root node processed? if not, catch it here */
2742 if (path->nodes[orig_level]) {
2743 ret = wc->process_func(log, path->nodes[orig_level], wc,
2744 btrfs_header_generation(path->nodes[orig_level]),
2749 ret = clean_log_buffer(trans, path->nodes[orig_level]);
2753 btrfs_free_path(path);
2758 * helper function to update the item for a given subvolumes log root
2759 * in the tree of log roots
2761 static int update_log_root(struct btrfs_trans_handle *trans,
2762 struct btrfs_root *log,
2763 struct btrfs_root_item *root_item)
2765 struct btrfs_fs_info *fs_info = log->fs_info;
2768 if (log->log_transid == 1) {
2769 /* insert root item on the first sync */
2770 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2771 &log->root_key, root_item);
2773 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2774 &log->root_key, root_item);
2779 static void wait_log_commit(struct btrfs_root *root, int transid)
2782 int index = transid % 2;
2785 * we only allow two pending log transactions at a time,
2786 * so we know that if ours is more than 2 older than the
2787 * current transaction, we're done
2790 prepare_to_wait(&root->log_commit_wait[index],
2791 &wait, TASK_UNINTERRUPTIBLE);
2793 if (!(root->log_transid_committed < transid &&
2794 atomic_read(&root->log_commit[index])))
2797 mutex_unlock(&root->log_mutex);
2799 mutex_lock(&root->log_mutex);
2801 finish_wait(&root->log_commit_wait[index], &wait);
2804 static void wait_for_writer(struct btrfs_root *root)
2809 prepare_to_wait(&root->log_writer_wait, &wait,
2810 TASK_UNINTERRUPTIBLE);
2811 if (!atomic_read(&root->log_writers))
2814 mutex_unlock(&root->log_mutex);
2816 mutex_lock(&root->log_mutex);
2818 finish_wait(&root->log_writer_wait, &wait);
2821 void btrfs_init_log_ctx(struct btrfs_log_ctx *ctx, struct inode *inode)
2824 ctx->log_transid = 0;
2825 ctx->log_new_dentries = false;
2826 ctx->logging_new_name = false;
2827 ctx->logging_new_delayed_dentries = false;
2828 ctx->logged_before = false;
2830 INIT_LIST_HEAD(&ctx->list);
2831 INIT_LIST_HEAD(&ctx->ordered_extents);
2832 INIT_LIST_HEAD(&ctx->conflict_inodes);
2833 ctx->num_conflict_inodes = 0;
2834 ctx->logging_conflict_inodes = false;
2835 ctx->scratch_eb = NULL;
2838 void btrfs_init_log_ctx_scratch_eb(struct btrfs_log_ctx *ctx)
2840 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
2842 if (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
2843 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
2847 * Don't care about allocation failure. This is just for optimization,
2848 * if we fail to allocate here, we will try again later if needed.
2850 ctx->scratch_eb = alloc_dummy_extent_buffer(inode->root->fs_info, 0);
2853 void btrfs_release_log_ctx_extents(struct btrfs_log_ctx *ctx)
2855 struct btrfs_ordered_extent *ordered;
2856 struct btrfs_ordered_extent *tmp;
2858 ASSERT(inode_is_locked(ctx->inode));
2860 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
2861 list_del_init(&ordered->log_list);
2862 btrfs_put_ordered_extent(ordered);
2867 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2868 struct btrfs_log_ctx *ctx)
2870 mutex_lock(&root->log_mutex);
2871 list_del_init(&ctx->list);
2872 mutex_unlock(&root->log_mutex);
2876 * Invoked in log mutex context, or be sure there is no other task which
2877 * can access the list.
2879 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2880 int index, int error)
2882 struct btrfs_log_ctx *ctx;
2883 struct btrfs_log_ctx *safe;
2885 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2886 list_del_init(&ctx->list);
2887 ctx->log_ret = error;
2892 * Sends a given tree log down to the disk and updates the super blocks to
2893 * record it. When this call is done, you know that any inodes previously
2894 * logged are safely on disk only if it returns 0.
2896 * Any other return value means you need to call btrfs_commit_transaction.
2897 * Some of the edge cases for fsyncing directories that have had unlinks
2898 * or renames done in the past mean that sometimes the only safe
2899 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2900 * that has happened.
2902 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2903 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2909 struct btrfs_fs_info *fs_info = root->fs_info;
2910 struct btrfs_root *log = root->log_root;
2911 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2912 struct btrfs_root_item new_root_item;
2913 int log_transid = 0;
2914 struct btrfs_log_ctx root_log_ctx;
2915 struct blk_plug plug;
2919 mutex_lock(&root->log_mutex);
2920 log_transid = ctx->log_transid;
2921 if (root->log_transid_committed >= log_transid) {
2922 mutex_unlock(&root->log_mutex);
2923 return ctx->log_ret;
2926 index1 = log_transid % 2;
2927 if (atomic_read(&root->log_commit[index1])) {
2928 wait_log_commit(root, log_transid);
2929 mutex_unlock(&root->log_mutex);
2930 return ctx->log_ret;
2932 ASSERT(log_transid == root->log_transid);
2933 atomic_set(&root->log_commit[index1], 1);
2935 /* wait for previous tree log sync to complete */
2936 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2937 wait_log_commit(root, log_transid - 1);
2940 int batch = atomic_read(&root->log_batch);
2941 /* when we're on an ssd, just kick the log commit out */
2942 if (!btrfs_test_opt(fs_info, SSD) &&
2943 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2944 mutex_unlock(&root->log_mutex);
2945 schedule_timeout_uninterruptible(1);
2946 mutex_lock(&root->log_mutex);
2948 wait_for_writer(root);
2949 if (batch == atomic_read(&root->log_batch))
2953 /* bail out if we need to do a full commit */
2954 if (btrfs_need_log_full_commit(trans)) {
2955 ret = BTRFS_LOG_FORCE_COMMIT;
2956 mutex_unlock(&root->log_mutex);
2960 if (log_transid % 2 == 0)
2961 mark = EXTENT_DIRTY;
2965 /* we start IO on all the marked extents here, but we don't actually
2966 * wait for them until later.
2968 blk_start_plug(&plug);
2969 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2971 * -EAGAIN happens when someone, e.g., a concurrent transaction
2972 * commit, writes a dirty extent in this tree-log commit. This
2973 * concurrent write will create a hole writing out the extents,
2974 * and we cannot proceed on a zoned filesystem, requiring
2975 * sequential writing. While we can bail out to a full commit
2976 * here, but we can continue hoping the concurrent writing fills
2979 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
2982 blk_finish_plug(&plug);
2983 btrfs_set_log_full_commit(trans);
2984 mutex_unlock(&root->log_mutex);
2989 * We _must_ update under the root->log_mutex in order to make sure we
2990 * have a consistent view of the log root we are trying to commit at
2993 * We _must_ copy this into a local copy, because we are not holding the
2994 * log_root_tree->log_mutex yet. This is important because when we
2995 * commit the log_root_tree we must have a consistent view of the
2996 * log_root_tree when we update the super block to point at the
2997 * log_root_tree bytenr. If we update the log_root_tree here we'll race
2998 * with the commit and possibly point at the new block which we may not
3001 btrfs_set_root_node(&log->root_item, log->node);
3002 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3004 btrfs_set_root_log_transid(root, root->log_transid + 1);
3005 log->log_transid = root->log_transid;
3006 root->log_start_pid = 0;
3008 * IO has been started, blocks of the log tree have WRITTEN flag set
3009 * in their headers. new modifications of the log will be written to
3010 * new positions. so it's safe to allow log writers to go in.
3012 mutex_unlock(&root->log_mutex);
3014 if (btrfs_is_zoned(fs_info)) {
3015 mutex_lock(&fs_info->tree_root->log_mutex);
3016 if (!log_root_tree->node) {
3017 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3019 mutex_unlock(&fs_info->tree_root->log_mutex);
3020 blk_finish_plug(&plug);
3024 mutex_unlock(&fs_info->tree_root->log_mutex);
3027 btrfs_init_log_ctx(&root_log_ctx, NULL);
3029 mutex_lock(&log_root_tree->log_mutex);
3031 index2 = log_root_tree->log_transid % 2;
3032 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3033 root_log_ctx.log_transid = log_root_tree->log_transid;
3036 * Now we are safe to update the log_root_tree because we're under the
3037 * log_mutex, and we're a current writer so we're holding the commit
3038 * open until we drop the log_mutex.
3040 ret = update_log_root(trans, log, &new_root_item);
3042 list_del_init(&root_log_ctx.list);
3043 blk_finish_plug(&plug);
3044 btrfs_set_log_full_commit(trans);
3047 "failed to update log for root %llu ret %d",
3048 root->root_key.objectid, ret);
3049 btrfs_wait_tree_log_extents(log, mark);
3050 mutex_unlock(&log_root_tree->log_mutex);
3054 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3055 blk_finish_plug(&plug);
3056 list_del_init(&root_log_ctx.list);
3057 mutex_unlock(&log_root_tree->log_mutex);
3058 ret = root_log_ctx.log_ret;
3062 if (atomic_read(&log_root_tree->log_commit[index2])) {
3063 blk_finish_plug(&plug);
3064 ret = btrfs_wait_tree_log_extents(log, mark);
3065 wait_log_commit(log_root_tree,
3066 root_log_ctx.log_transid);
3067 mutex_unlock(&log_root_tree->log_mutex);
3069 ret = root_log_ctx.log_ret;
3072 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3073 atomic_set(&log_root_tree->log_commit[index2], 1);
3075 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3076 wait_log_commit(log_root_tree,
3077 root_log_ctx.log_transid - 1);
3081 * now that we've moved on to the tree of log tree roots,
3082 * check the full commit flag again
3084 if (btrfs_need_log_full_commit(trans)) {
3085 blk_finish_plug(&plug);
3086 btrfs_wait_tree_log_extents(log, mark);
3087 mutex_unlock(&log_root_tree->log_mutex);
3088 ret = BTRFS_LOG_FORCE_COMMIT;
3089 goto out_wake_log_root;
3092 ret = btrfs_write_marked_extents(fs_info,
3093 &log_root_tree->dirty_log_pages,
3094 EXTENT_DIRTY | EXTENT_NEW);
3095 blk_finish_plug(&plug);
3097 * As described above, -EAGAIN indicates a hole in the extents. We
3098 * cannot wait for these write outs since the waiting cause a
3099 * deadlock. Bail out to the full commit instead.
3101 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3102 btrfs_set_log_full_commit(trans);
3103 btrfs_wait_tree_log_extents(log, mark);
3104 mutex_unlock(&log_root_tree->log_mutex);
3105 goto out_wake_log_root;
3107 btrfs_set_log_full_commit(trans);
3108 mutex_unlock(&log_root_tree->log_mutex);
3109 goto out_wake_log_root;
3111 ret = btrfs_wait_tree_log_extents(log, mark);
3113 ret = btrfs_wait_tree_log_extents(log_root_tree,
3114 EXTENT_NEW | EXTENT_DIRTY);
3116 btrfs_set_log_full_commit(trans);
3117 mutex_unlock(&log_root_tree->log_mutex);
3118 goto out_wake_log_root;
3121 log_root_start = log_root_tree->node->start;
3122 log_root_level = btrfs_header_level(log_root_tree->node);
3123 log_root_tree->log_transid++;
3124 mutex_unlock(&log_root_tree->log_mutex);
3127 * Here we are guaranteed that nobody is going to write the superblock
3128 * for the current transaction before us and that neither we do write
3129 * our superblock before the previous transaction finishes its commit
3130 * and writes its superblock, because:
3132 * 1) We are holding a handle on the current transaction, so no body
3133 * can commit it until we release the handle;
3135 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3136 * if the previous transaction is still committing, and hasn't yet
3137 * written its superblock, we wait for it to do it, because a
3138 * transaction commit acquires the tree_log_mutex when the commit
3139 * begins and releases it only after writing its superblock.
3141 mutex_lock(&fs_info->tree_log_mutex);
3144 * The previous transaction writeout phase could have failed, and thus
3145 * marked the fs in an error state. We must not commit here, as we
3146 * could have updated our generation in the super_for_commit and
3147 * writing the super here would result in transid mismatches. If there
3148 * is an error here just bail.
3150 if (BTRFS_FS_ERROR(fs_info)) {
3152 btrfs_set_log_full_commit(trans);
3153 btrfs_abort_transaction(trans, ret);
3154 mutex_unlock(&fs_info->tree_log_mutex);
3155 goto out_wake_log_root;
3158 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3159 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3160 ret = write_all_supers(fs_info, 1);
3161 mutex_unlock(&fs_info->tree_log_mutex);
3163 btrfs_set_log_full_commit(trans);
3164 btrfs_abort_transaction(trans, ret);
3165 goto out_wake_log_root;
3169 * We know there can only be one task here, since we have not yet set
3170 * root->log_commit[index1] to 0 and any task attempting to sync the
3171 * log must wait for the previous log transaction to commit if it's
3172 * still in progress or wait for the current log transaction commit if
3173 * someone else already started it. We use <= and not < because the
3174 * first log transaction has an ID of 0.
3176 ASSERT(btrfs_get_root_last_log_commit(root) <= log_transid);
3177 btrfs_set_root_last_log_commit(root, log_transid);
3180 mutex_lock(&log_root_tree->log_mutex);
3181 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3183 log_root_tree->log_transid_committed++;
3184 atomic_set(&log_root_tree->log_commit[index2], 0);
3185 mutex_unlock(&log_root_tree->log_mutex);
3188 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3189 * all the updates above are seen by the woken threads. It might not be
3190 * necessary, but proving that seems to be hard.
3192 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3194 mutex_lock(&root->log_mutex);
3195 btrfs_remove_all_log_ctxs(root, index1, ret);
3196 root->log_transid_committed++;
3197 atomic_set(&root->log_commit[index1], 0);
3198 mutex_unlock(&root->log_mutex);
3201 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3202 * all the updates above are seen by the woken threads. It might not be
3203 * necessary, but proving that seems to be hard.
3205 cond_wake_up(&root->log_commit_wait[index1]);
3209 static void free_log_tree(struct btrfs_trans_handle *trans,
3210 struct btrfs_root *log)
3213 struct walk_control wc = {
3215 .process_func = process_one_buffer
3219 ret = walk_log_tree(trans, log, &wc);
3222 * We weren't able to traverse the entire log tree, the
3223 * typical scenario is getting an -EIO when reading an
3224 * extent buffer of the tree, due to a previous writeback
3227 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3228 &log->fs_info->fs_state);
3231 * Some extent buffers of the log tree may still be dirty
3232 * and not yet written back to storage, because we may
3233 * have updates to a log tree without syncing a log tree,
3234 * such as during rename and link operations. So flush
3235 * them out and wait for their writeback to complete, so
3236 * that we properly cleanup their state and pages.
3238 btrfs_write_marked_extents(log->fs_info,
3239 &log->dirty_log_pages,
3240 EXTENT_DIRTY | EXTENT_NEW);
3241 btrfs_wait_tree_log_extents(log,
3242 EXTENT_DIRTY | EXTENT_NEW);
3245 btrfs_abort_transaction(trans, ret);
3247 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3251 extent_io_tree_release(&log->dirty_log_pages);
3252 extent_io_tree_release(&log->log_csum_range);
3254 btrfs_put_root(log);
3258 * free all the extents used by the tree log. This should be called
3259 * at commit time of the full transaction
3261 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3263 if (root->log_root) {
3264 free_log_tree(trans, root->log_root);
3265 root->log_root = NULL;
3266 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3271 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3272 struct btrfs_fs_info *fs_info)
3274 if (fs_info->log_root_tree) {
3275 free_log_tree(trans, fs_info->log_root_tree);
3276 fs_info->log_root_tree = NULL;
3277 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3283 * Check if an inode was logged in the current transaction. This correctly deals
3284 * with the case where the inode was logged but has a logged_trans of 0, which
3285 * happens if the inode is evicted and loaded again, as logged_trans is an in
3286 * memory only field (not persisted).
3288 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3291 static int inode_logged(const struct btrfs_trans_handle *trans,
3292 struct btrfs_inode *inode,
3293 struct btrfs_path *path_in)
3295 struct btrfs_path *path = path_in;
3296 struct btrfs_key key;
3299 if (inode->logged_trans == trans->transid)
3303 * If logged_trans is not 0, then we know the inode logged was not logged
3304 * in this transaction, so we can return false right away.
3306 if (inode->logged_trans > 0)
3310 * If no log tree was created for this root in this transaction, then
3311 * the inode can not have been logged in this transaction. In that case
3312 * set logged_trans to anything greater than 0 and less than the current
3313 * transaction's ID, to avoid the search below in a future call in case
3314 * a log tree gets created after this.
3316 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3317 inode->logged_trans = trans->transid - 1;
3322 * We have a log tree and the inode's logged_trans is 0. We can't tell
3323 * for sure if the inode was logged before in this transaction by looking
3324 * only at logged_trans. We could be pessimistic and assume it was, but
3325 * that can lead to unnecessarily logging an inode during rename and link
3326 * operations, and then further updating the log in followup rename and
3327 * link operations, specially if it's a directory, which adds latency
3328 * visible to applications doing a series of rename or link operations.
3330 * A logged_trans of 0 here can mean several things:
3332 * 1) The inode was never logged since the filesystem was mounted, and may
3333 * or may have not been evicted and loaded again;
3335 * 2) The inode was logged in a previous transaction, then evicted and
3336 * then loaded again;
3338 * 3) The inode was logged in the current transaction, then evicted and
3339 * then loaded again.
3341 * For cases 1) and 2) we don't want to return true, but we need to detect
3342 * case 3) and return true. So we do a search in the log root for the inode
3345 key.objectid = btrfs_ino(inode);
3346 key.type = BTRFS_INODE_ITEM_KEY;
3350 path = btrfs_alloc_path();
3355 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3358 btrfs_release_path(path);
3360 btrfs_free_path(path);
3363 * Logging an inode always results in logging its inode item. So if we
3364 * did not find the item we know the inode was not logged for sure.
3368 } else if (ret > 0) {
3370 * Set logged_trans to a value greater than 0 and less then the
3371 * current transaction to avoid doing the search in future calls.
3373 inode->logged_trans = trans->transid - 1;
3378 * The inode was previously logged and then evicted, set logged_trans to
3379 * the current transacion's ID, to avoid future tree searches as long as
3380 * the inode is not evicted again.
3382 inode->logged_trans = trans->transid;
3385 * If it's a directory, then we must set last_dir_index_offset to the
3386 * maximum possible value, so that the next attempt to log the inode does
3387 * not skip checking if dir index keys found in modified subvolume tree
3388 * leaves have been logged before, otherwise it would result in attempts
3389 * to insert duplicate dir index keys in the log tree. This must be done
3390 * because last_dir_index_offset is an in-memory only field, not persisted
3391 * in the inode item or any other on-disk structure, so its value is lost
3392 * once the inode is evicted.
3394 if (S_ISDIR(inode->vfs_inode.i_mode))
3395 inode->last_dir_index_offset = (u64)-1;
3401 * Delete a directory entry from the log if it exists.
3403 * Returns < 0 on error
3404 * 1 if the entry does not exists
3405 * 0 if the entry existed and was successfully deleted
3407 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3408 struct btrfs_root *log,
3409 struct btrfs_path *path,
3411 const struct fscrypt_str *name,
3414 struct btrfs_dir_item *di;
3417 * We only log dir index items of a directory, so we don't need to look
3418 * for dir item keys.
3420 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3428 * We do not need to update the size field of the directory's
3429 * inode item because on log replay we update the field to reflect
3430 * all existing entries in the directory (see overwrite_item()).
3432 return btrfs_delete_one_dir_name(trans, log, path, di);
3436 * If both a file and directory are logged, and unlinks or renames are
3437 * mixed in, we have a few interesting corners:
3439 * create file X in dir Y
3440 * link file X to X.link in dir Y
3442 * unlink file X but leave X.link
3445 * After a crash we would expect only X.link to exist. But file X
3446 * didn't get fsync'd again so the log has back refs for X and X.link.
3448 * We solve this by removing directory entries and inode backrefs from the
3449 * log when a file that was logged in the current transaction is
3450 * unlinked. Any later fsync will include the updated log entries, and
3451 * we'll be able to reconstruct the proper directory items from backrefs.
3453 * This optimizations allows us to avoid relogging the entire inode
3454 * or the entire directory.
3456 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3457 struct btrfs_root *root,
3458 const struct fscrypt_str *name,
3459 struct btrfs_inode *dir, u64 index)
3461 struct btrfs_path *path;
3464 ret = inode_logged(trans, dir, NULL);
3468 btrfs_set_log_full_commit(trans);
3472 ret = join_running_log_trans(root);
3476 mutex_lock(&dir->log_mutex);
3478 path = btrfs_alloc_path();
3484 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3486 btrfs_free_path(path);
3488 mutex_unlock(&dir->log_mutex);
3490 btrfs_set_log_full_commit(trans);
3491 btrfs_end_log_trans(root);
3494 /* see comments for btrfs_del_dir_entries_in_log */
3495 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3496 struct btrfs_root *root,
3497 const struct fscrypt_str *name,
3498 struct btrfs_inode *inode, u64 dirid)
3500 struct btrfs_root *log;
3504 ret = inode_logged(trans, inode, NULL);
3508 btrfs_set_log_full_commit(trans);
3512 ret = join_running_log_trans(root);
3515 log = root->log_root;
3516 mutex_lock(&inode->log_mutex);
3518 ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode),
3520 mutex_unlock(&inode->log_mutex);
3521 if (ret < 0 && ret != -ENOENT)
3522 btrfs_set_log_full_commit(trans);
3523 btrfs_end_log_trans(root);
3527 * creates a range item in the log for 'dirid'. first_offset and
3528 * last_offset tell us which parts of the key space the log should
3529 * be considered authoritative for.
3531 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3532 struct btrfs_root *log,
3533 struct btrfs_path *path,
3535 u64 first_offset, u64 last_offset)
3538 struct btrfs_key key;
3539 struct btrfs_dir_log_item *item;
3541 key.objectid = dirid;
3542 key.offset = first_offset;
3543 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3544 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3546 * -EEXIST is fine and can happen sporadically when we are logging a
3547 * directory and have concurrent insertions in the subvolume's tree for
3548 * items from other inodes and that result in pushing off some dir items
3549 * from one leaf to another in order to accommodate for the new items.
3550 * This results in logging the same dir index range key.
3552 if (ret && ret != -EEXIST)
3555 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3556 struct btrfs_dir_log_item);
3557 if (ret == -EEXIST) {
3558 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3561 * btrfs_del_dir_entries_in_log() might have been called during
3562 * an unlink between the initial insertion of this key and the
3563 * current update, or we might be logging a single entry deletion
3564 * during a rename, so set the new last_offset to the max value.
3566 last_offset = max(last_offset, curr_end);
3568 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3569 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
3570 btrfs_release_path(path);
3574 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3575 struct btrfs_inode *inode,
3576 struct extent_buffer *src,
3577 struct btrfs_path *dst_path,
3581 struct btrfs_root *log = inode->root->log_root;
3582 char *ins_data = NULL;
3583 struct btrfs_item_batch batch;
3584 struct extent_buffer *dst;
3585 unsigned long src_offset;
3586 unsigned long dst_offset;
3588 struct btrfs_key key;
3597 btrfs_item_key_to_cpu(src, &key, start_slot);
3598 item_size = btrfs_item_size(src, start_slot);
3600 batch.data_sizes = &item_size;
3601 batch.total_data_size = item_size;
3603 struct btrfs_key *ins_keys;
3606 ins_data = kmalloc(count * sizeof(u32) +
3607 count * sizeof(struct btrfs_key), GFP_NOFS);
3611 ins_sizes = (u32 *)ins_data;
3612 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3613 batch.keys = ins_keys;
3614 batch.data_sizes = ins_sizes;
3615 batch.total_data_size = 0;
3617 for (i = 0; i < count; i++) {
3618 const int slot = start_slot + i;
3620 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3621 ins_sizes[i] = btrfs_item_size(src, slot);
3622 batch.total_data_size += ins_sizes[i];
3626 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3630 dst = dst_path->nodes[0];
3632 * Copy all the items in bulk, in a single copy operation. Item data is
3633 * organized such that it's placed at the end of a leaf and from right
3634 * to left. For example, the data for the second item ends at an offset
3635 * that matches the offset where the data for the first item starts, the
3636 * data for the third item ends at an offset that matches the offset
3637 * where the data of the second items starts, and so on.
3638 * Therefore our source and destination start offsets for copy match the
3639 * offsets of the last items (highest slots).
3641 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3642 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3643 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3644 btrfs_release_path(dst_path);
3646 last_index = batch.keys[count - 1].offset;
3647 ASSERT(last_index > inode->last_dir_index_offset);
3650 * If for some unexpected reason the last item's index is not greater
3651 * than the last index we logged, warn and force a transaction commit.
3653 if (WARN_ON(last_index <= inode->last_dir_index_offset))
3654 ret = BTRFS_LOG_FORCE_COMMIT;
3656 inode->last_dir_index_offset = last_index;
3658 if (btrfs_get_first_dir_index_to_log(inode) == 0)
3659 btrfs_set_first_dir_index_to_log(inode, batch.keys[0].offset);
3666 static int clone_leaf(struct btrfs_path *path, struct btrfs_log_ctx *ctx)
3668 const int slot = path->slots[0];
3670 if (ctx->scratch_eb) {
3671 copy_extent_buffer_full(ctx->scratch_eb, path->nodes[0]);
3673 ctx->scratch_eb = btrfs_clone_extent_buffer(path->nodes[0]);
3674 if (!ctx->scratch_eb)
3678 btrfs_release_path(path);
3679 path->nodes[0] = ctx->scratch_eb;
3680 path->slots[0] = slot;
3682 * Add extra ref to scratch eb so that it is not freed when callers
3683 * release the path, so we can reuse it later if needed.
3685 atomic_inc(&ctx->scratch_eb->refs);
3690 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3691 struct btrfs_inode *inode,
3692 struct btrfs_path *path,
3693 struct btrfs_path *dst_path,
3694 struct btrfs_log_ctx *ctx,
3695 u64 *last_old_dentry_offset)
3697 struct btrfs_root *log = inode->root->log_root;
3698 struct extent_buffer *src;
3699 const int nritems = btrfs_header_nritems(path->nodes[0]);
3700 const u64 ino = btrfs_ino(inode);
3701 bool last_found = false;
3702 int batch_start = 0;
3707 * We need to clone the leaf, release the read lock on it, and use the
3708 * clone before modifying the log tree. See the comment at copy_items()
3709 * about why we need to do this.
3711 ret = clone_leaf(path, ctx);
3715 src = path->nodes[0];
3717 for (int i = path->slots[0]; i < nritems; i++) {
3718 struct btrfs_dir_item *di;
3719 struct btrfs_key key;
3722 btrfs_item_key_to_cpu(src, &key, i);
3724 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3729 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3732 * Skip ranges of items that consist only of dir item keys created
3733 * in past transactions. However if we find a gap, we must log a
3734 * dir index range item for that gap, so that index keys in that
3735 * gap are deleted during log replay.
3737 if (btrfs_dir_transid(src, di) < trans->transid) {
3738 if (key.offset > *last_old_dentry_offset + 1) {
3739 ret = insert_dir_log_key(trans, log, dst_path,
3740 ino, *last_old_dentry_offset + 1,
3746 *last_old_dentry_offset = key.offset;
3750 /* If we logged this dir index item before, we can skip it. */
3751 if (key.offset <= inode->last_dir_index_offset)
3755 * We must make sure that when we log a directory entry, the
3756 * corresponding inode, after log replay, has a matching link
3757 * count. For example:
3763 * xfs_io -c "fsync" mydir
3765 * <mount fs and log replay>
3767 * Would result in a fsync log that when replayed, our file inode
3768 * would have a link count of 1, but we get two directory entries
3769 * pointing to the same inode. After removing one of the names,
3770 * it would not be possible to remove the other name, which
3771 * resulted always in stale file handle errors, and would not be
3772 * possible to rmdir the parent directory, since its i_size could
3773 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3774 * resulting in -ENOTEMPTY errors.
3776 if (!ctx->log_new_dentries) {
3777 struct btrfs_key di_key;
3779 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3780 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3781 ctx->log_new_dentries = true;
3784 if (batch_size == 0)
3789 if (batch_size > 0) {
3792 ret = flush_dir_items_batch(trans, inode, src, dst_path,
3793 batch_start, batch_size);
3798 return last_found ? 1 : 0;
3802 * log all the items included in the current transaction for a given
3803 * directory. This also creates the range items in the log tree required
3804 * to replay anything deleted before the fsync
3806 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3807 struct btrfs_inode *inode,
3808 struct btrfs_path *path,
3809 struct btrfs_path *dst_path,
3810 struct btrfs_log_ctx *ctx,
3811 u64 min_offset, u64 *last_offset_ret)
3813 struct btrfs_key min_key;
3814 struct btrfs_root *root = inode->root;
3815 struct btrfs_root *log = root->log_root;
3817 u64 last_old_dentry_offset = min_offset - 1;
3818 u64 last_offset = (u64)-1;
3819 u64 ino = btrfs_ino(inode);
3821 min_key.objectid = ino;
3822 min_key.type = BTRFS_DIR_INDEX_KEY;
3823 min_key.offset = min_offset;
3825 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3828 * we didn't find anything from this transaction, see if there
3829 * is anything at all
3831 if (ret != 0 || min_key.objectid != ino ||
3832 min_key.type != BTRFS_DIR_INDEX_KEY) {
3833 min_key.objectid = ino;
3834 min_key.type = BTRFS_DIR_INDEX_KEY;
3835 min_key.offset = (u64)-1;
3836 btrfs_release_path(path);
3837 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3839 btrfs_release_path(path);
3842 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3844 /* if ret == 0 there are items for this type,
3845 * create a range to tell us the last key of this type.
3846 * otherwise, there are no items in this directory after
3847 * *min_offset, and we create a range to indicate that.
3850 struct btrfs_key tmp;
3852 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3854 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3855 last_old_dentry_offset = tmp.offset;
3856 } else if (ret > 0) {
3863 /* go backward to find any previous key */
3864 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3866 struct btrfs_key tmp;
3868 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3870 * The dir index key before the first one we found that needs to
3871 * be logged might be in a previous leaf, and there might be a
3872 * gap between these keys, meaning that we had deletions that
3873 * happened. So the key range item we log (key type
3874 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3875 * previous key's offset plus 1, so that those deletes are replayed.
3877 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3878 last_old_dentry_offset = tmp.offset;
3879 } else if (ret < 0) {
3883 btrfs_release_path(path);
3886 * Find the first key from this transaction again or the one we were at
3887 * in the loop below in case we had to reschedule. We may be logging the
3888 * directory without holding its VFS lock, which happen when logging new
3889 * dentries (through log_new_dir_dentries()) or in some cases when we
3890 * need to log the parent directory of an inode. This means a dir index
3891 * key might be deleted from the inode's root, and therefore we may not
3892 * find it anymore. If we can't find it, just move to the next key. We
3893 * can not bail out and ignore, because if we do that we will simply
3894 * not log dir index keys that come after the one that was just deleted
3895 * and we can end up logging a dir index range that ends at (u64)-1
3896 * (@last_offset is initialized to that), resulting in removing dir
3897 * entries we should not remove at log replay time.
3900 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3902 ret = btrfs_next_item(root, path);
3904 /* There are no more keys in the inode's root. */
3913 * we have a block from this transaction, log every item in it
3914 * from our directory
3917 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3918 &last_old_dentry_offset);
3924 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3927 * look ahead to the next item and see if it is also
3928 * from this directory and from this transaction
3930 ret = btrfs_next_leaf(root, path);
3933 last_offset = (u64)-1;
3938 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3939 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3940 last_offset = (u64)-1;
3943 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3945 * The next leaf was not changed in the current transaction
3946 * and has at least one dir index key.
3947 * We check for the next key because there might have been
3948 * one or more deletions between the last key we logged and
3949 * that next key. So the key range item we log (key type
3950 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3951 * offset minus 1, so that those deletes are replayed.
3953 last_offset = min_key.offset - 1;
3956 if (need_resched()) {
3957 btrfs_release_path(path);
3963 btrfs_release_path(path);
3964 btrfs_release_path(dst_path);
3967 *last_offset_ret = last_offset;
3969 * In case the leaf was changed in the current transaction but
3970 * all its dir items are from a past transaction, the last item
3971 * in the leaf is a dir item and there's no gap between that last
3972 * dir item and the first one on the next leaf (which did not
3973 * change in the current transaction), then we don't need to log
3974 * a range, last_old_dentry_offset is == to last_offset.
3976 ASSERT(last_old_dentry_offset <= last_offset);
3977 if (last_old_dentry_offset < last_offset)
3978 ret = insert_dir_log_key(trans, log, path, ino,
3979 last_old_dentry_offset + 1,
3987 * If the inode was logged before and it was evicted, then its
3988 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3989 * key offset. If that's the case, search for it and update the inode. This
3990 * is to avoid lookups in the log tree every time we try to insert a dir index
3991 * key from a leaf changed in the current transaction, and to allow us to always
3992 * do batch insertions of dir index keys.
3994 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3995 struct btrfs_path *path,
3996 const struct btrfs_log_ctx *ctx)
3998 const u64 ino = btrfs_ino(inode);
3999 struct btrfs_key key;
4002 lockdep_assert_held(&inode->log_mutex);
4004 if (inode->last_dir_index_offset != (u64)-1)
4007 if (!ctx->logged_before) {
4008 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
4013 key.type = BTRFS_DIR_INDEX_KEY;
4014 key.offset = (u64)-1;
4016 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
4018 * An error happened or we actually have an index key with an offset
4019 * value of (u64)-1. Bail out, we're done.
4025 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
4028 * No dir index items, bail out and leave last_dir_index_offset with
4029 * the value right before the first valid index value.
4031 if (path->slots[0] == 0)
4035 * btrfs_search_slot() left us at one slot beyond the slot with the last
4036 * index key, or beyond the last key of the directory that is not an
4037 * index key. If we have an index key before, set last_dir_index_offset
4038 * to its offset value, otherwise leave it with a value right before the
4039 * first valid index value, as it means we have an empty directory.
4041 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4042 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
4043 inode->last_dir_index_offset = key.offset;
4046 btrfs_release_path(path);
4052 * logging directories is very similar to logging inodes, We find all the items
4053 * from the current transaction and write them to the log.
4055 * The recovery code scans the directory in the subvolume, and if it finds a
4056 * key in the range logged that is not present in the log tree, then it means
4057 * that dir entry was unlinked during the transaction.
4059 * In order for that scan to work, we must include one key smaller than
4060 * the smallest logged by this transaction and one key larger than the largest
4061 * key logged by this transaction.
4063 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4064 struct btrfs_inode *inode,
4065 struct btrfs_path *path,
4066 struct btrfs_path *dst_path,
4067 struct btrfs_log_ctx *ctx)
4073 ret = update_last_dir_index_offset(inode, path, ctx);
4077 min_key = BTRFS_DIR_START_INDEX;
4081 ret = log_dir_items(trans, inode, path, dst_path,
4082 ctx, min_key, &max_key);
4085 if (max_key == (u64)-1)
4087 min_key = max_key + 1;
4094 * a helper function to drop items from the log before we relog an
4095 * inode. max_key_type indicates the highest item type to remove.
4096 * This cannot be run for file data extents because it does not
4097 * free the extents they point to.
4099 static int drop_inode_items(struct btrfs_trans_handle *trans,
4100 struct btrfs_root *log,
4101 struct btrfs_path *path,
4102 struct btrfs_inode *inode,
4106 struct btrfs_key key;
4107 struct btrfs_key found_key;
4110 key.objectid = btrfs_ino(inode);
4111 key.type = max_key_type;
4112 key.offset = (u64)-1;
4115 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4118 } else if (ret > 0) {
4119 if (path->slots[0] == 0)
4124 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4127 if (found_key.objectid != key.objectid)
4130 found_key.offset = 0;
4132 ret = btrfs_bin_search(path->nodes[0], 0, &found_key, &start_slot);
4136 ret = btrfs_del_items(trans, log, path, start_slot,
4137 path->slots[0] - start_slot + 1);
4139 * If start slot isn't 0 then we don't need to re-search, we've
4140 * found the last guy with the objectid in this tree.
4142 if (ret || start_slot != 0)
4144 btrfs_release_path(path);
4146 btrfs_release_path(path);
4152 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4153 struct btrfs_root *log_root,
4154 struct btrfs_inode *inode,
4155 u64 new_size, u32 min_type)
4157 struct btrfs_truncate_control control = {
4158 .new_size = new_size,
4159 .ino = btrfs_ino(inode),
4160 .min_type = min_type,
4161 .skip_ref_updates = true,
4164 return btrfs_truncate_inode_items(trans, log_root, &control);
4167 static void fill_inode_item(struct btrfs_trans_handle *trans,
4168 struct extent_buffer *leaf,
4169 struct btrfs_inode_item *item,
4170 struct inode *inode, int log_inode_only,
4173 struct btrfs_map_token token;
4176 btrfs_init_map_token(&token, leaf);
4178 if (log_inode_only) {
4179 /* set the generation to zero so the recover code
4180 * can tell the difference between an logging
4181 * just to say 'this inode exists' and a logging
4182 * to say 'update this inode with these values'
4184 btrfs_set_token_inode_generation(&token, item, 0);
4185 btrfs_set_token_inode_size(&token, item, logged_isize);
4187 btrfs_set_token_inode_generation(&token, item,
4188 BTRFS_I(inode)->generation);
4189 btrfs_set_token_inode_size(&token, item, inode->i_size);
4192 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4193 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4194 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4195 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4197 btrfs_set_token_timespec_sec(&token, &item->atime,
4198 inode_get_atime_sec(inode));
4199 btrfs_set_token_timespec_nsec(&token, &item->atime,
4200 inode_get_atime_nsec(inode));
4202 btrfs_set_token_timespec_sec(&token, &item->mtime,
4203 inode_get_mtime_sec(inode));
4204 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4205 inode_get_mtime_nsec(inode));
4207 btrfs_set_token_timespec_sec(&token, &item->ctime,
4208 inode_get_ctime_sec(inode));
4209 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4210 inode_get_ctime_nsec(inode));
4213 * We do not need to set the nbytes field, in fact during a fast fsync
4214 * its value may not even be correct, since a fast fsync does not wait
4215 * for ordered extent completion, which is where we update nbytes, it
4216 * only waits for writeback to complete. During log replay as we find
4217 * file extent items and replay them, we adjust the nbytes field of the
4218 * inode item in subvolume tree as needed (see overwrite_item()).
4221 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4222 btrfs_set_token_inode_transid(&token, item, trans->transid);
4223 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4224 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4225 BTRFS_I(inode)->ro_flags);
4226 btrfs_set_token_inode_flags(&token, item, flags);
4227 btrfs_set_token_inode_block_group(&token, item, 0);
4230 static int log_inode_item(struct btrfs_trans_handle *trans,
4231 struct btrfs_root *log, struct btrfs_path *path,
4232 struct btrfs_inode *inode, bool inode_item_dropped)
4234 struct btrfs_inode_item *inode_item;
4238 * If we are doing a fast fsync and the inode was logged before in the
4239 * current transaction, then we know the inode was previously logged and
4240 * it exists in the log tree. For performance reasons, in this case use
4241 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4242 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4243 * contention in case there are concurrent fsyncs for other inodes of the
4244 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4245 * already exists can also result in unnecessarily splitting a leaf.
4247 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4248 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4254 * This means it is the first fsync in the current transaction,
4255 * so the inode item is not in the log and we need to insert it.
4256 * We can never get -EEXIST because we are only called for a fast
4257 * fsync and in case an inode eviction happens after the inode was
4258 * logged before in the current transaction, when we load again
4259 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4260 * flags and set ->logged_trans to 0.
4262 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4263 sizeof(*inode_item));
4264 ASSERT(ret != -EEXIST);
4268 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4269 struct btrfs_inode_item);
4270 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4272 btrfs_release_path(path);
4276 static int log_csums(struct btrfs_trans_handle *trans,
4277 struct btrfs_inode *inode,
4278 struct btrfs_root *log_root,
4279 struct btrfs_ordered_sum *sums)
4281 const u64 lock_end = sums->logical + sums->len - 1;
4282 struct extent_state *cached_state = NULL;
4286 * If this inode was not used for reflink operations in the current
4287 * transaction with new extents, then do the fast path, no need to
4288 * worry about logging checksum items with overlapping ranges.
4290 if (inode->last_reflink_trans < trans->transid)
4291 return btrfs_csum_file_blocks(trans, log_root, sums);
4294 * Serialize logging for checksums. This is to avoid racing with the
4295 * same checksum being logged by another task that is logging another
4296 * file which happens to refer to the same extent as well. Such races
4297 * can leave checksum items in the log with overlapping ranges.
4299 ret = lock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4304 * Due to extent cloning, we might have logged a csum item that covers a
4305 * subrange of a cloned extent, and later we can end up logging a csum
4306 * item for a larger subrange of the same extent or the entire range.
4307 * This would leave csum items in the log tree that cover the same range
4308 * and break the searches for checksums in the log tree, resulting in
4309 * some checksums missing in the fs/subvolume tree. So just delete (or
4310 * trim and adjust) any existing csum items in the log for this range.
4312 ret = btrfs_del_csums(trans, log_root, sums->logical, sums->len);
4314 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4316 unlock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4322 static noinline int copy_items(struct btrfs_trans_handle *trans,
4323 struct btrfs_inode *inode,
4324 struct btrfs_path *dst_path,
4325 struct btrfs_path *src_path,
4326 int start_slot, int nr, int inode_only,
4327 u64 logged_isize, struct btrfs_log_ctx *ctx)
4329 struct btrfs_root *log = inode->root->log_root;
4330 struct btrfs_file_extent_item *extent;
4331 struct extent_buffer *src;
4333 struct btrfs_key *ins_keys;
4335 struct btrfs_item_batch batch;
4338 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4339 const u64 i_size = i_size_read(&inode->vfs_inode);
4342 * To keep lockdep happy and avoid deadlocks, clone the source leaf and
4343 * use the clone. This is because otherwise we would be changing the log
4344 * tree, to insert items from the subvolume tree or insert csum items,
4345 * while holding a read lock on a leaf from the subvolume tree, which
4346 * creates a nasty lock dependency when COWing log tree nodes/leaves:
4348 * 1) Modifying the log tree triggers an extent buffer allocation while
4349 * holding a write lock on a parent extent buffer from the log tree.
4350 * Allocating the pages for an extent buffer, or the extent buffer
4351 * struct, can trigger inode eviction and finally the inode eviction
4352 * will trigger a release/remove of a delayed node, which requires
4353 * taking the delayed node's mutex;
4355 * 2) Allocating a metadata extent for a log tree can trigger the async
4356 * reclaim thread and make us wait for it to release enough space and
4357 * unblock our reservation ticket. The reclaim thread can start
4358 * flushing delayed items, and that in turn results in the need to
4359 * lock delayed node mutexes and in the need to write lock extent
4360 * buffers of a subvolume tree - all this while holding a write lock
4361 * on the parent extent buffer in the log tree.
4363 * So one task in scenario 1) running in parallel with another task in
4364 * scenario 2) could lead to a deadlock, one wanting to lock a delayed
4365 * node mutex while having a read lock on a leaf from the subvolume,
4366 * while the other is holding the delayed node's mutex and wants to
4367 * write lock the same subvolume leaf for flushing delayed items.
4369 ret = clone_leaf(src_path, ctx);
4373 src = src_path->nodes[0];
4375 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4376 nr * sizeof(u32), GFP_NOFS);
4380 ins_sizes = (u32 *)ins_data;
4381 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4382 batch.keys = ins_keys;
4383 batch.data_sizes = ins_sizes;
4384 batch.total_data_size = 0;
4388 for (int i = 0; i < nr; i++) {
4389 const int src_slot = start_slot + i;
4390 struct btrfs_root *csum_root;
4391 struct btrfs_ordered_sum *sums;
4392 struct btrfs_ordered_sum *sums_next;
4393 LIST_HEAD(ordered_sums);
4397 u64 extent_num_bytes;
4400 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4402 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4405 extent = btrfs_item_ptr(src, src_slot,
4406 struct btrfs_file_extent_item);
4408 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4412 * Don't copy extents from past generations. That would make us
4413 * log a lot more metadata for common cases like doing only a
4414 * few random writes into a file and then fsync it for the first
4415 * time or after the full sync flag is set on the inode. We can
4416 * get leaves full of extent items, most of which are from past
4417 * generations, so we can skip them - as long as the inode has
4418 * not been the target of a reflink operation in this transaction,
4419 * as in that case it might have had file extent items with old
4420 * generations copied into it. We also must always log prealloc
4421 * extents that start at or beyond eof, otherwise we would lose
4422 * them on log replay.
4424 if (is_old_extent &&
4425 ins_keys[dst_index].offset < i_size &&
4426 inode->last_reflink_trans < trans->transid)
4432 /* Only regular extents have checksums. */
4433 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4437 * If it's an extent created in a past transaction, then its
4438 * checksums are already accessible from the committed csum tree,
4439 * no need to log them.
4444 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4445 /* If it's an explicit hole, there are no checksums. */
4446 if (disk_bytenr == 0)
4449 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4451 if (btrfs_file_extent_compression(src, extent)) {
4453 extent_num_bytes = disk_num_bytes;
4455 extent_offset = btrfs_file_extent_offset(src, extent);
4456 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4459 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4460 disk_bytenr += extent_offset;
4461 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4462 disk_bytenr + extent_num_bytes - 1,
4463 &ordered_sums, 0, false);
4467 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4469 ret = log_csums(trans, inode, log, sums);
4470 list_del(&sums->list);
4477 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4478 batch.total_data_size += ins_sizes[dst_index];
4484 * We have a leaf full of old extent items that don't need to be logged,
4485 * so we don't need to do anything.
4490 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4495 for (int i = 0; i < nr; i++) {
4496 const int src_slot = start_slot + i;
4497 const int dst_slot = dst_path->slots[0] + dst_index;
4498 struct btrfs_key key;
4499 unsigned long src_offset;
4500 unsigned long dst_offset;
4503 * We're done, all the remaining items in the source leaf
4504 * correspond to old file extent items.
4506 if (dst_index >= batch.nr)
4509 btrfs_item_key_to_cpu(src, &key, src_slot);
4511 if (key.type != BTRFS_EXTENT_DATA_KEY)
4514 extent = btrfs_item_ptr(src, src_slot,
4515 struct btrfs_file_extent_item);
4517 /* See the comment in the previous loop, same logic. */
4518 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4519 key.offset < i_size &&
4520 inode->last_reflink_trans < trans->transid)
4524 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4525 src_offset = btrfs_item_ptr_offset(src, src_slot);
4527 if (key.type == BTRFS_INODE_ITEM_KEY) {
4528 struct btrfs_inode_item *inode_item;
4530 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4531 struct btrfs_inode_item);
4532 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4534 inode_only == LOG_INODE_EXISTS,
4537 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4538 src_offset, ins_sizes[dst_index]);
4544 btrfs_mark_buffer_dirty(trans, dst_path->nodes[0]);
4545 btrfs_release_path(dst_path);
4552 static int extent_cmp(void *priv, const struct list_head *a,
4553 const struct list_head *b)
4555 const struct extent_map *em1, *em2;
4557 em1 = list_entry(a, struct extent_map, list);
4558 em2 = list_entry(b, struct extent_map, list);
4560 if (em1->start < em2->start)
4562 else if (em1->start > em2->start)
4567 static int log_extent_csums(struct btrfs_trans_handle *trans,
4568 struct btrfs_inode *inode,
4569 struct btrfs_root *log_root,
4570 const struct extent_map *em,
4571 struct btrfs_log_ctx *ctx)
4573 struct btrfs_ordered_extent *ordered;
4574 struct btrfs_root *csum_root;
4577 u64 mod_start = em->start;
4578 u64 mod_len = em->len;
4579 LIST_HEAD(ordered_sums);
4582 if (inode->flags & BTRFS_INODE_NODATASUM ||
4583 (em->flags & EXTENT_FLAG_PREALLOC) ||
4584 em->block_start == EXTENT_MAP_HOLE)
4587 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4588 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4589 const u64 mod_end = mod_start + mod_len;
4590 struct btrfs_ordered_sum *sums;
4595 if (ordered_end <= mod_start)
4597 if (mod_end <= ordered->file_offset)
4601 * We are going to copy all the csums on this ordered extent, so
4602 * go ahead and adjust mod_start and mod_len in case this ordered
4603 * extent has already been logged.
4605 if (ordered->file_offset > mod_start) {
4606 if (ordered_end >= mod_end)
4607 mod_len = ordered->file_offset - mod_start;
4609 * If we have this case
4611 * |--------- logged extent ---------|
4612 * |----- ordered extent ----|
4614 * Just don't mess with mod_start and mod_len, we'll
4615 * just end up logging more csums than we need and it
4619 if (ordered_end < mod_end) {
4620 mod_len = mod_end - ordered_end;
4621 mod_start = ordered_end;
4628 * To keep us from looping for the above case of an ordered
4629 * extent that falls inside of the logged extent.
4631 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4634 list_for_each_entry(sums, &ordered->list, list) {
4635 ret = log_csums(trans, inode, log_root, sums);
4641 /* We're done, found all csums in the ordered extents. */
4645 /* If we're compressed we have to save the entire range of csums. */
4646 if (extent_map_is_compressed(em)) {
4648 csum_len = max(em->block_len, em->orig_block_len);
4650 csum_offset = mod_start - em->start;
4654 /* block start is already adjusted for the file extent offset. */
4655 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4656 ret = btrfs_lookup_csums_list(csum_root, em->block_start + csum_offset,
4657 em->block_start + csum_offset +
4658 csum_len - 1, &ordered_sums, 0, false);
4662 while (!list_empty(&ordered_sums)) {
4663 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4664 struct btrfs_ordered_sum,
4667 ret = log_csums(trans, inode, log_root, sums);
4668 list_del(&sums->list);
4675 static int log_one_extent(struct btrfs_trans_handle *trans,
4676 struct btrfs_inode *inode,
4677 const struct extent_map *em,
4678 struct btrfs_path *path,
4679 struct btrfs_log_ctx *ctx)
4681 struct btrfs_drop_extents_args drop_args = { 0 };
4682 struct btrfs_root *log = inode->root->log_root;
4683 struct btrfs_file_extent_item fi = { 0 };
4684 struct extent_buffer *leaf;
4685 struct btrfs_key key;
4686 enum btrfs_compression_type compress_type;
4687 u64 extent_offset = em->start - em->orig_start;
4691 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4692 if (em->flags & EXTENT_FLAG_PREALLOC)
4693 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4695 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4697 block_len = max(em->block_len, em->orig_block_len);
4698 compress_type = extent_map_compression(em);
4699 if (compress_type != BTRFS_COMPRESS_NONE) {
4700 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4701 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4702 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4703 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4705 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4708 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4709 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4710 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4711 btrfs_set_stack_file_extent_compression(&fi, compress_type);
4713 ret = log_extent_csums(trans, inode, log, em, ctx);
4718 * If this is the first time we are logging the inode in the current
4719 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4720 * because it does a deletion search, which always acquires write locks
4721 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4722 * but also adds significant contention in a log tree, since log trees
4723 * are small, with a root at level 2 or 3 at most, due to their short
4726 if (ctx->logged_before) {
4727 drop_args.path = path;
4728 drop_args.start = em->start;
4729 drop_args.end = em->start + em->len;
4730 drop_args.replace_extent = true;
4731 drop_args.extent_item_size = sizeof(fi);
4732 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4737 if (!drop_args.extent_inserted) {
4738 key.objectid = btrfs_ino(inode);
4739 key.type = BTRFS_EXTENT_DATA_KEY;
4740 key.offset = em->start;
4742 ret = btrfs_insert_empty_item(trans, log, path, &key,
4747 leaf = path->nodes[0];
4748 write_extent_buffer(leaf, &fi,
4749 btrfs_item_ptr_offset(leaf, path->slots[0]),
4751 btrfs_mark_buffer_dirty(trans, leaf);
4753 btrfs_release_path(path);
4759 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4760 * lose them after doing a full/fast fsync and replaying the log. We scan the
4761 * subvolume's root instead of iterating the inode's extent map tree because
4762 * otherwise we can log incorrect extent items based on extent map conversion.
4763 * That can happen due to the fact that extent maps are merged when they
4764 * are not in the extent map tree's list of modified extents.
4766 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4767 struct btrfs_inode *inode,
4768 struct btrfs_path *path,
4769 struct btrfs_log_ctx *ctx)
4771 struct btrfs_root *root = inode->root;
4772 struct btrfs_key key;
4773 const u64 i_size = i_size_read(&inode->vfs_inode);
4774 const u64 ino = btrfs_ino(inode);
4775 struct btrfs_path *dst_path = NULL;
4776 bool dropped_extents = false;
4777 u64 truncate_offset = i_size;
4778 struct extent_buffer *leaf;
4784 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4788 key.type = BTRFS_EXTENT_DATA_KEY;
4789 key.offset = i_size;
4790 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4795 * We must check if there is a prealloc extent that starts before the
4796 * i_size and crosses the i_size boundary. This is to ensure later we
4797 * truncate down to the end of that extent and not to the i_size, as
4798 * otherwise we end up losing part of the prealloc extent after a log
4799 * replay and with an implicit hole if there is another prealloc extent
4800 * that starts at an offset beyond i_size.
4802 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4807 struct btrfs_file_extent_item *ei;
4809 leaf = path->nodes[0];
4810 slot = path->slots[0];
4811 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4813 if (btrfs_file_extent_type(leaf, ei) ==
4814 BTRFS_FILE_EXTENT_PREALLOC) {
4817 btrfs_item_key_to_cpu(leaf, &key, slot);
4818 extent_end = key.offset +
4819 btrfs_file_extent_num_bytes(leaf, ei);
4821 if (extent_end > i_size)
4822 truncate_offset = extent_end;
4829 leaf = path->nodes[0];
4830 slot = path->slots[0];
4832 if (slot >= btrfs_header_nritems(leaf)) {
4834 ret = copy_items(trans, inode, dst_path, path,
4835 start_slot, ins_nr, 1, 0, ctx);
4840 ret = btrfs_next_leaf(root, path);
4850 btrfs_item_key_to_cpu(leaf, &key, slot);
4851 if (key.objectid > ino)
4853 if (WARN_ON_ONCE(key.objectid < ino) ||
4854 key.type < BTRFS_EXTENT_DATA_KEY ||
4855 key.offset < i_size) {
4859 if (!dropped_extents) {
4861 * Avoid logging extent items logged in past fsync calls
4862 * and leading to duplicate keys in the log tree.
4864 ret = truncate_inode_items(trans, root->log_root, inode,
4866 BTRFS_EXTENT_DATA_KEY);
4869 dropped_extents = true;
4876 dst_path = btrfs_alloc_path();
4884 ret = copy_items(trans, inode, dst_path, path,
4885 start_slot, ins_nr, 1, 0, ctx);
4887 btrfs_release_path(path);
4888 btrfs_free_path(dst_path);
4892 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4893 struct btrfs_inode *inode,
4894 struct btrfs_path *path,
4895 struct btrfs_log_ctx *ctx)
4897 struct btrfs_ordered_extent *ordered;
4898 struct btrfs_ordered_extent *tmp;
4899 struct extent_map *em, *n;
4901 struct extent_map_tree *tree = &inode->extent_tree;
4905 write_lock(&tree->lock);
4907 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4908 list_del_init(&em->list);
4910 * Just an arbitrary number, this can be really CPU intensive
4911 * once we start getting a lot of extents, and really once we
4912 * have a bunch of extents we just want to commit since it will
4915 if (++num > 32768) {
4916 list_del_init(&tree->modified_extents);
4921 if (em->generation < trans->transid)
4924 /* We log prealloc extents beyond eof later. */
4925 if ((em->flags & EXTENT_FLAG_PREALLOC) &&
4926 em->start >= i_size_read(&inode->vfs_inode))
4929 /* Need a ref to keep it from getting evicted from cache */
4930 refcount_inc(&em->refs);
4931 em->flags |= EXTENT_FLAG_LOGGING;
4932 list_add_tail(&em->list, &extents);
4936 list_sort(NULL, &extents, extent_cmp);
4938 while (!list_empty(&extents)) {
4939 em = list_entry(extents.next, struct extent_map, list);
4941 list_del_init(&em->list);
4944 * If we had an error we just need to delete everybody from our
4948 clear_em_logging(tree, em);
4949 free_extent_map(em);
4953 write_unlock(&tree->lock);
4955 ret = log_one_extent(trans, inode, em, path, ctx);
4956 write_lock(&tree->lock);
4957 clear_em_logging(tree, em);
4958 free_extent_map(em);
4960 WARN_ON(!list_empty(&extents));
4961 write_unlock(&tree->lock);
4964 ret = btrfs_log_prealloc_extents(trans, inode, path, ctx);
4969 * We have logged all extents successfully, now make sure the commit of
4970 * the current transaction waits for the ordered extents to complete
4971 * before it commits and wipes out the log trees, otherwise we would
4972 * lose data if an ordered extents completes after the transaction
4973 * commits and a power failure happens after the transaction commit.
4975 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4976 list_del_init(&ordered->log_list);
4977 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4979 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4980 spin_lock_irq(&inode->ordered_tree_lock);
4981 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4982 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4983 atomic_inc(&trans->transaction->pending_ordered);
4985 spin_unlock_irq(&inode->ordered_tree_lock);
4987 btrfs_put_ordered_extent(ordered);
4993 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4994 struct btrfs_path *path, u64 *size_ret)
4996 struct btrfs_key key;
4999 key.objectid = btrfs_ino(inode);
5000 key.type = BTRFS_INODE_ITEM_KEY;
5003 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
5006 } else if (ret > 0) {
5009 struct btrfs_inode_item *item;
5011 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5012 struct btrfs_inode_item);
5013 *size_ret = btrfs_inode_size(path->nodes[0], item);
5015 * If the in-memory inode's i_size is smaller then the inode
5016 * size stored in the btree, return the inode's i_size, so
5017 * that we get a correct inode size after replaying the log
5018 * when before a power failure we had a shrinking truncate
5019 * followed by addition of a new name (rename / new hard link).
5020 * Otherwise return the inode size from the btree, to avoid
5021 * data loss when replaying a log due to previously doing a
5022 * write that expands the inode's size and logging a new name
5023 * immediately after.
5025 if (*size_ret > inode->vfs_inode.i_size)
5026 *size_ret = inode->vfs_inode.i_size;
5029 btrfs_release_path(path);
5034 * At the moment we always log all xattrs. This is to figure out at log replay
5035 * time which xattrs must have their deletion replayed. If a xattr is missing
5036 * in the log tree and exists in the fs/subvol tree, we delete it. This is
5037 * because if a xattr is deleted, the inode is fsynced and a power failure
5038 * happens, causing the log to be replayed the next time the fs is mounted,
5039 * we want the xattr to not exist anymore (same behaviour as other filesystems
5040 * with a journal, ext3/4, xfs, f2fs, etc).
5042 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
5043 struct btrfs_inode *inode,
5044 struct btrfs_path *path,
5045 struct btrfs_path *dst_path,
5046 struct btrfs_log_ctx *ctx)
5048 struct btrfs_root *root = inode->root;
5050 struct btrfs_key key;
5051 const u64 ino = btrfs_ino(inode);
5054 bool found_xattrs = false;
5056 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
5060 key.type = BTRFS_XATTR_ITEM_KEY;
5063 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5068 int slot = path->slots[0];
5069 struct extent_buffer *leaf = path->nodes[0];
5070 int nritems = btrfs_header_nritems(leaf);
5072 if (slot >= nritems) {
5074 ret = copy_items(trans, inode, dst_path, path,
5075 start_slot, ins_nr, 1, 0, ctx);
5080 ret = btrfs_next_leaf(root, path);
5088 btrfs_item_key_to_cpu(leaf, &key, slot);
5089 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5096 found_xattrs = true;
5100 ret = copy_items(trans, inode, dst_path, path,
5101 start_slot, ins_nr, 1, 0, ctx);
5107 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5113 * When using the NO_HOLES feature if we punched a hole that causes the
5114 * deletion of entire leafs or all the extent items of the first leaf (the one
5115 * that contains the inode item and references) we may end up not processing
5116 * any extents, because there are no leafs with a generation matching the
5117 * current transaction that have extent items for our inode. So we need to find
5118 * if any holes exist and then log them. We also need to log holes after any
5119 * truncate operation that changes the inode's size.
5121 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5122 struct btrfs_inode *inode,
5123 struct btrfs_path *path)
5125 struct btrfs_root *root = inode->root;
5126 struct btrfs_fs_info *fs_info = root->fs_info;
5127 struct btrfs_key key;
5128 const u64 ino = btrfs_ino(inode);
5129 const u64 i_size = i_size_read(&inode->vfs_inode);
5130 u64 prev_extent_end = 0;
5133 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5137 key.type = BTRFS_EXTENT_DATA_KEY;
5140 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5145 struct extent_buffer *leaf = path->nodes[0];
5147 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5148 ret = btrfs_next_leaf(root, path);
5155 leaf = path->nodes[0];
5158 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5159 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5162 /* We have a hole, log it. */
5163 if (prev_extent_end < key.offset) {
5164 const u64 hole_len = key.offset - prev_extent_end;
5167 * Release the path to avoid deadlocks with other code
5168 * paths that search the root while holding locks on
5169 * leafs from the log root.
5171 btrfs_release_path(path);
5172 ret = btrfs_insert_hole_extent(trans, root->log_root,
5173 ino, prev_extent_end,
5179 * Search for the same key again in the root. Since it's
5180 * an extent item and we are holding the inode lock, the
5181 * key must still exist. If it doesn't just emit warning
5182 * and return an error to fall back to a transaction
5185 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5188 if (WARN_ON(ret > 0))
5190 leaf = path->nodes[0];
5193 prev_extent_end = btrfs_file_extent_end(path);
5198 if (prev_extent_end < i_size) {
5201 btrfs_release_path(path);
5202 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5203 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5204 prev_extent_end, hole_len);
5213 * When we are logging a new inode X, check if it doesn't have a reference that
5214 * matches the reference from some other inode Y created in a past transaction
5215 * and that was renamed in the current transaction. If we don't do this, then at
5216 * log replay time we can lose inode Y (and all its files if it's a directory):
5219 * echo "hello world" > /mnt/x/foobar
5222 * mkdir /mnt/x # or touch /mnt/x
5223 * xfs_io -c fsync /mnt/x
5225 * mount fs, trigger log replay
5227 * After the log replay procedure, we would lose the first directory and all its
5228 * files (file foobar).
5229 * For the case where inode Y is not a directory we simply end up losing it:
5231 * echo "123" > /mnt/foo
5233 * mv /mnt/foo /mnt/bar
5234 * echo "abc" > /mnt/foo
5235 * xfs_io -c fsync /mnt/foo
5238 * We also need this for cases where a snapshot entry is replaced by some other
5239 * entry (file or directory) otherwise we end up with an unreplayable log due to
5240 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5241 * if it were a regular entry:
5244 * btrfs subvolume snapshot /mnt /mnt/x/snap
5245 * btrfs subvolume delete /mnt/x/snap
5248 * fsync /mnt/x or fsync some new file inside it
5251 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5252 * the same transaction.
5254 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5256 const struct btrfs_key *key,
5257 struct btrfs_inode *inode,
5258 u64 *other_ino, u64 *other_parent)
5261 struct btrfs_path *search_path;
5264 u32 item_size = btrfs_item_size(eb, slot);
5266 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5268 search_path = btrfs_alloc_path();
5271 search_path->search_commit_root = 1;
5272 search_path->skip_locking = 1;
5274 while (cur_offset < item_size) {
5278 unsigned long name_ptr;
5279 struct btrfs_dir_item *di;
5280 struct fscrypt_str name_str;
5282 if (key->type == BTRFS_INODE_REF_KEY) {
5283 struct btrfs_inode_ref *iref;
5285 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5286 parent = key->offset;
5287 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5288 name_ptr = (unsigned long)(iref + 1);
5289 this_len = sizeof(*iref) + this_name_len;
5291 struct btrfs_inode_extref *extref;
5293 extref = (struct btrfs_inode_extref *)(ptr +
5295 parent = btrfs_inode_extref_parent(eb, extref);
5296 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5297 name_ptr = (unsigned long)&extref->name;
5298 this_len = sizeof(*extref) + this_name_len;
5301 if (this_name_len > name_len) {
5304 new_name = krealloc(name, this_name_len, GFP_NOFS);
5309 name_len = this_name_len;
5313 read_extent_buffer(eb, name, name_ptr, this_name_len);
5315 name_str.name = name;
5316 name_str.len = this_name_len;
5317 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5318 parent, &name_str, 0);
5319 if (di && !IS_ERR(di)) {
5320 struct btrfs_key di_key;
5322 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5324 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5325 if (di_key.objectid != key->objectid) {
5327 *other_ino = di_key.objectid;
5328 *other_parent = parent;
5336 } else if (IS_ERR(di)) {
5340 btrfs_release_path(search_path);
5342 cur_offset += this_len;
5346 btrfs_free_path(search_path);
5352 * Check if we need to log an inode. This is used in contexts where while
5353 * logging an inode we need to log another inode (either that it exists or in
5354 * full mode). This is used instead of btrfs_inode_in_log() because the later
5355 * requires the inode to be in the log and have the log transaction committed,
5356 * while here we do not care if the log transaction was already committed - our
5357 * caller will commit the log later - and we want to avoid logging an inode
5358 * multiple times when multiple tasks have joined the same log transaction.
5360 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5361 struct btrfs_inode *inode)
5364 * If a directory was not modified, no dentries added or removed, we can
5365 * and should avoid logging it.
5367 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5371 * If this inode does not have new/updated/deleted xattrs since the last
5372 * time it was logged and is flagged as logged in the current transaction,
5373 * we can skip logging it. As for new/deleted names, those are updated in
5374 * the log by link/unlink/rename operations.
5375 * In case the inode was logged and then evicted and reloaded, its
5376 * logged_trans will be 0, in which case we have to fully log it since
5377 * logged_trans is a transient field, not persisted.
5379 if (inode_logged(trans, inode, NULL) == 1 &&
5380 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5386 struct btrfs_dir_list {
5388 struct list_head list;
5392 * Log the inodes of the new dentries of a directory.
5393 * See process_dir_items_leaf() for details about why it is needed.
5394 * This is a recursive operation - if an existing dentry corresponds to a
5395 * directory, that directory's new entries are logged too (same behaviour as
5396 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5397 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5398 * complains about the following circular lock dependency / possible deadlock:
5402 * lock(&type->i_mutex_dir_key#3/2);
5403 * lock(sb_internal#2);
5404 * lock(&type->i_mutex_dir_key#3/2);
5405 * lock(&sb->s_type->i_mutex_key#14);
5407 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5408 * sb_start_intwrite() in btrfs_start_transaction().
5409 * Not acquiring the VFS lock of the inodes is still safe because:
5411 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5412 * that while logging the inode new references (names) are added or removed
5413 * from the inode, leaving the logged inode item with a link count that does
5414 * not match the number of logged inode reference items. This is fine because
5415 * at log replay time we compute the real number of links and correct the
5416 * link count in the inode item (see replay_one_buffer() and
5417 * link_to_fixup_dir());
5419 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5420 * while logging the inode's items new index items (key type
5421 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5422 * has a size that doesn't match the sum of the lengths of all the logged
5423 * names - this is ok, not a problem, because at log replay time we set the
5424 * directory's i_size to the correct value (see replay_one_name() and
5425 * overwrite_item()).
5427 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5428 struct btrfs_inode *start_inode,
5429 struct btrfs_log_ctx *ctx)
5431 struct btrfs_root *root = start_inode->root;
5432 struct btrfs_fs_info *fs_info = root->fs_info;
5433 struct btrfs_path *path;
5434 LIST_HEAD(dir_list);
5435 struct btrfs_dir_list *dir_elem;
5436 u64 ino = btrfs_ino(start_inode);
5437 struct btrfs_inode *curr_inode = start_inode;
5441 * If we are logging a new name, as part of a link or rename operation,
5442 * don't bother logging new dentries, as we just want to log the names
5443 * of an inode and that any new parents exist.
5445 if (ctx->logging_new_name)
5448 path = btrfs_alloc_path();
5452 /* Pairs with btrfs_add_delayed_iput below. */
5453 ihold(&curr_inode->vfs_inode);
5456 struct inode *vfs_inode;
5457 struct btrfs_key key;
5458 struct btrfs_key found_key;
5460 bool continue_curr_inode = true;
5464 key.type = BTRFS_DIR_INDEX_KEY;
5465 key.offset = btrfs_get_first_dir_index_to_log(curr_inode);
5466 next_index = key.offset;
5468 btrfs_for_each_slot(root->log_root, &key, &found_key, path, iter_ret) {
5469 struct extent_buffer *leaf = path->nodes[0];
5470 struct btrfs_dir_item *di;
5471 struct btrfs_key di_key;
5472 struct inode *di_inode;
5473 int log_mode = LOG_INODE_EXISTS;
5476 if (found_key.objectid != ino ||
5477 found_key.type != BTRFS_DIR_INDEX_KEY) {
5478 continue_curr_inode = false;
5482 next_index = found_key.offset + 1;
5484 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5485 type = btrfs_dir_ftype(leaf, di);
5486 if (btrfs_dir_transid(leaf, di) < trans->transid)
5488 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5489 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5492 btrfs_release_path(path);
5493 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5494 if (IS_ERR(di_inode)) {
5495 ret = PTR_ERR(di_inode);
5499 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5500 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5504 ctx->log_new_dentries = false;
5505 if (type == BTRFS_FT_DIR)
5506 log_mode = LOG_INODE_ALL;
5507 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5509 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5512 if (ctx->log_new_dentries) {
5513 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5518 dir_elem->ino = di_key.objectid;
5519 list_add_tail(&dir_elem->list, &dir_list);
5524 btrfs_release_path(path);
5529 } else if (iter_ret > 0) {
5530 continue_curr_inode = false;
5535 if (continue_curr_inode && key.offset < (u64)-1) {
5540 btrfs_set_first_dir_index_to_log(curr_inode, next_index);
5542 if (list_empty(&dir_list))
5545 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5546 ino = dir_elem->ino;
5547 list_del(&dir_elem->list);
5550 btrfs_add_delayed_iput(curr_inode);
5553 vfs_inode = btrfs_iget(fs_info->sb, ino, root);
5554 if (IS_ERR(vfs_inode)) {
5555 ret = PTR_ERR(vfs_inode);
5558 curr_inode = BTRFS_I(vfs_inode);
5561 btrfs_free_path(path);
5563 btrfs_add_delayed_iput(curr_inode);
5566 struct btrfs_dir_list *next;
5568 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5575 struct btrfs_ino_list {
5578 struct list_head list;
5581 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5583 struct btrfs_ino_list *curr;
5584 struct btrfs_ino_list *next;
5586 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5587 list_del(&curr->list);
5592 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5593 struct btrfs_path *path)
5595 struct btrfs_key key;
5599 key.type = BTRFS_INODE_ITEM_KEY;
5602 path->search_commit_root = 1;
5603 path->skip_locking = 1;
5605 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5606 if (WARN_ON_ONCE(ret > 0)) {
5608 * We have previously found the inode through the commit root
5609 * so this should not happen. If it does, just error out and
5610 * fallback to a transaction commit.
5613 } else if (ret == 0) {
5614 struct btrfs_inode_item *item;
5616 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5617 struct btrfs_inode_item);
5618 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5622 btrfs_release_path(path);
5623 path->search_commit_root = 0;
5624 path->skip_locking = 0;
5629 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5630 struct btrfs_root *root,
5631 struct btrfs_path *path,
5632 u64 ino, u64 parent,
5633 struct btrfs_log_ctx *ctx)
5635 struct btrfs_ino_list *ino_elem;
5636 struct inode *inode;
5639 * It's rare to have a lot of conflicting inodes, in practice it is not
5640 * common to have more than 1 or 2. We don't want to collect too many,
5641 * as we could end up logging too many inodes (even if only in
5642 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5645 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES)
5646 return BTRFS_LOG_FORCE_COMMIT;
5648 inode = btrfs_iget(root->fs_info->sb, ino, root);
5650 * If the other inode that had a conflicting dir entry was deleted in
5651 * the current transaction then we either:
5653 * 1) Log the parent directory (later after adding it to the list) if
5654 * the inode is a directory. This is because it may be a deleted
5655 * subvolume/snapshot or it may be a regular directory that had
5656 * deleted subvolumes/snapshots (or subdirectories that had them),
5657 * and at the moment we can't deal with dropping subvolumes/snapshots
5658 * during log replay. So we just log the parent, which will result in
5659 * a fallback to a transaction commit if we are dealing with those
5660 * cases (last_unlink_trans will match the current transaction);
5662 * 2) Do nothing if it's not a directory. During log replay we simply
5663 * unlink the conflicting dentry from the parent directory and then
5664 * add the dentry for our inode. Like this we can avoid logging the
5665 * parent directory (and maybe fallback to a transaction commit in
5666 * case it has a last_unlink_trans == trans->transid, due to moving
5667 * some inode from it to some other directory).
5669 if (IS_ERR(inode)) {
5670 int ret = PTR_ERR(inode);
5675 ret = conflicting_inode_is_dir(root, ino, path);
5676 /* Not a directory or we got an error. */
5680 /* Conflicting inode is a directory, so we'll log its parent. */
5681 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5684 ino_elem->ino = ino;
5685 ino_elem->parent = parent;
5686 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5687 ctx->num_conflict_inodes++;
5693 * If the inode was already logged skip it - otherwise we can hit an
5694 * infinite loop. Example:
5696 * From the commit root (previous transaction) we have the following
5699 * inode 257 a directory
5700 * inode 258 with references "zz" and "zz_link" on inode 257
5701 * inode 259 with reference "a" on inode 257
5703 * And in the current (uncommitted) transaction we have:
5705 * inode 257 a directory, unchanged
5706 * inode 258 with references "a" and "a2" on inode 257
5707 * inode 259 with reference "zz_link" on inode 257
5708 * inode 261 with reference "zz" on inode 257
5710 * When logging inode 261 the following infinite loop could
5711 * happen if we don't skip already logged inodes:
5713 * - we detect inode 258 as a conflicting inode, with inode 261
5714 * on reference "zz", and log it;
5716 * - we detect inode 259 as a conflicting inode, with inode 258
5717 * on reference "a", and log it;
5719 * - we detect inode 258 as a conflicting inode, with inode 259
5720 * on reference "zz_link", and log it - again! After this we
5721 * repeat the above steps forever.
5723 * Here we can use need_log_inode() because we only need to log the
5724 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5725 * so that the log ends up with the new name and without the old name.
5727 if (!need_log_inode(trans, BTRFS_I(inode))) {
5728 btrfs_add_delayed_iput(BTRFS_I(inode));
5732 btrfs_add_delayed_iput(BTRFS_I(inode));
5734 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5737 ino_elem->ino = ino;
5738 ino_elem->parent = parent;
5739 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5740 ctx->num_conflict_inodes++;
5745 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5746 struct btrfs_root *root,
5747 struct btrfs_log_ctx *ctx)
5749 struct btrfs_fs_info *fs_info = root->fs_info;
5753 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5754 * otherwise we could have unbounded recursion of btrfs_log_inode()
5755 * calls. This check guarantees we can have only 1 level of recursion.
5757 if (ctx->logging_conflict_inodes)
5760 ctx->logging_conflict_inodes = true;
5763 * New conflicting inodes may be found and added to the list while we
5764 * are logging a conflicting inode, so keep iterating while the list is
5767 while (!list_empty(&ctx->conflict_inodes)) {
5768 struct btrfs_ino_list *curr;
5769 struct inode *inode;
5773 curr = list_first_entry(&ctx->conflict_inodes,
5774 struct btrfs_ino_list, list);
5776 parent = curr->parent;
5777 list_del(&curr->list);
5780 inode = btrfs_iget(fs_info->sb, ino, root);
5782 * If the other inode that had a conflicting dir entry was
5783 * deleted in the current transaction, we need to log its parent
5784 * directory. See the comment at add_conflicting_inode().
5786 if (IS_ERR(inode)) {
5787 ret = PTR_ERR(inode);
5791 inode = btrfs_iget(fs_info->sb, parent, root);
5792 if (IS_ERR(inode)) {
5793 ret = PTR_ERR(inode);
5798 * Always log the directory, we cannot make this
5799 * conditional on need_log_inode() because the directory
5800 * might have been logged in LOG_INODE_EXISTS mode or
5801 * the dir index of the conflicting inode is not in a
5802 * dir index key range logged for the directory. So we
5803 * must make sure the deletion is recorded.
5805 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5806 LOG_INODE_ALL, ctx);
5807 btrfs_add_delayed_iput(BTRFS_I(inode));
5814 * Here we can use need_log_inode() because we only need to log
5815 * the inode in LOG_INODE_EXISTS mode and rename operations
5816 * update the log, so that the log ends up with the new name and
5817 * without the old name.
5819 * We did this check at add_conflicting_inode(), but here we do
5820 * it again because if some other task logged the inode after
5821 * that, we can avoid doing it again.
5823 if (!need_log_inode(trans, BTRFS_I(inode))) {
5824 btrfs_add_delayed_iput(BTRFS_I(inode));
5829 * We are safe logging the other inode without acquiring its
5830 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5831 * are safe against concurrent renames of the other inode as
5832 * well because during a rename we pin the log and update the
5833 * log with the new name before we unpin it.
5835 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5836 btrfs_add_delayed_iput(BTRFS_I(inode));
5841 ctx->logging_conflict_inodes = false;
5843 free_conflicting_inodes(ctx);
5848 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5849 struct btrfs_inode *inode,
5850 struct btrfs_key *min_key,
5851 const struct btrfs_key *max_key,
5852 struct btrfs_path *path,
5853 struct btrfs_path *dst_path,
5854 const u64 logged_isize,
5855 const int inode_only,
5856 struct btrfs_log_ctx *ctx,
5857 bool *need_log_inode_item)
5859 const u64 i_size = i_size_read(&inode->vfs_inode);
5860 struct btrfs_root *root = inode->root;
5861 int ins_start_slot = 0;
5866 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5874 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5875 if (min_key->objectid != max_key->objectid)
5877 if (min_key->type > max_key->type)
5880 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5881 *need_log_inode_item = false;
5882 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5883 min_key->offset >= i_size) {
5885 * Extents at and beyond eof are logged with
5886 * btrfs_log_prealloc_extents().
5887 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5888 * and no keys greater than that, so bail out.
5891 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5892 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5893 (inode->generation == trans->transid ||
5894 ctx->logging_conflict_inodes)) {
5896 u64 other_parent = 0;
5898 ret = btrfs_check_ref_name_override(path->nodes[0],
5899 path->slots[0], min_key, inode,
5900 &other_ino, &other_parent);
5903 } else if (ret > 0 &&
5904 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5909 ins_start_slot = path->slots[0];
5911 ret = copy_items(trans, inode, dst_path, path,
5912 ins_start_slot, ins_nr,
5913 inode_only, logged_isize, ctx);
5918 btrfs_release_path(path);
5919 ret = add_conflicting_inode(trans, root, path,
5926 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5927 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5930 ret = copy_items(trans, inode, dst_path, path,
5932 ins_nr, inode_only, logged_isize, ctx);
5939 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5942 } else if (!ins_nr) {
5943 ins_start_slot = path->slots[0];
5948 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5949 ins_nr, inode_only, logged_isize, ctx);
5953 ins_start_slot = path->slots[0];
5956 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5957 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5962 ret = copy_items(trans, inode, dst_path, path,
5963 ins_start_slot, ins_nr, inode_only,
5969 btrfs_release_path(path);
5971 if (min_key->offset < (u64)-1) {
5973 } else if (min_key->type < max_key->type) {
5975 min_key->offset = 0;
5981 * We may process many leaves full of items for our inode, so
5982 * avoid monopolizing a cpu for too long by rescheduling while
5983 * not holding locks on any tree.
5988 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5989 ins_nr, inode_only, logged_isize, ctx);
5994 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5996 * Release the path because otherwise we might attempt to double
5997 * lock the same leaf with btrfs_log_prealloc_extents() below.
5999 btrfs_release_path(path);
6000 ret = btrfs_log_prealloc_extents(trans, inode, dst_path, ctx);
6006 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
6007 struct btrfs_root *log,
6008 struct btrfs_path *path,
6009 const struct btrfs_item_batch *batch,
6010 const struct btrfs_delayed_item *first_item)
6012 const struct btrfs_delayed_item *curr = first_item;
6015 ret = btrfs_insert_empty_items(trans, log, path, batch);
6019 for (int i = 0; i < batch->nr; i++) {
6022 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
6023 write_extent_buffer(path->nodes[0], &curr->data,
6024 (unsigned long)data_ptr, curr->data_len);
6025 curr = list_next_entry(curr, log_list);
6029 btrfs_release_path(path);
6034 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
6035 struct btrfs_inode *inode,
6036 struct btrfs_path *path,
6037 const struct list_head *delayed_ins_list,
6038 struct btrfs_log_ctx *ctx)
6040 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
6041 const int max_batch_size = 195;
6042 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
6043 const u64 ino = btrfs_ino(inode);
6044 struct btrfs_root *log = inode->root->log_root;
6045 struct btrfs_item_batch batch = {
6047 .total_data_size = 0,
6049 const struct btrfs_delayed_item *first = NULL;
6050 const struct btrfs_delayed_item *curr;
6052 struct btrfs_key *ins_keys;
6054 u64 curr_batch_size = 0;
6058 /* We are adding dir index items to the log tree. */
6059 lockdep_assert_held(&inode->log_mutex);
6062 * We collect delayed items before copying index keys from the subvolume
6063 * to the log tree. However just after we collected them, they may have
6064 * been flushed (all of them or just some of them), and therefore we
6065 * could have copied them from the subvolume tree to the log tree.
6066 * So find the first delayed item that was not yet logged (they are
6067 * sorted by index number).
6069 list_for_each_entry(curr, delayed_ins_list, log_list) {
6070 if (curr->index > inode->last_dir_index_offset) {
6076 /* Empty list or all delayed items were already logged. */
6080 ins_data = kmalloc(max_batch_size * sizeof(u32) +
6081 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
6084 ins_sizes = (u32 *)ins_data;
6085 batch.data_sizes = ins_sizes;
6086 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6087 batch.keys = ins_keys;
6090 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6091 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6093 if (curr_batch_size + curr_size > leaf_data_size ||
6094 batch.nr == max_batch_size) {
6095 ret = insert_delayed_items_batch(trans, log, path,
6101 batch.total_data_size = 0;
6102 curr_batch_size = 0;
6106 ins_sizes[batch_idx] = curr->data_len;
6107 ins_keys[batch_idx].objectid = ino;
6108 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6109 ins_keys[batch_idx].offset = curr->index;
6110 curr_batch_size += curr_size;
6111 batch.total_data_size += curr->data_len;
6114 curr = list_next_entry(curr, log_list);
6117 ASSERT(batch.nr >= 1);
6118 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6120 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6122 inode->last_dir_index_offset = curr->index;
6129 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6130 struct btrfs_inode *inode,
6131 struct btrfs_path *path,
6132 const struct list_head *delayed_del_list,
6133 struct btrfs_log_ctx *ctx)
6135 const u64 ino = btrfs_ino(inode);
6136 const struct btrfs_delayed_item *curr;
6138 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6141 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6142 u64 first_dir_index = curr->index;
6144 const struct btrfs_delayed_item *next;
6148 * Find a range of consecutive dir index items to delete. Like
6149 * this we log a single dir range item spanning several contiguous
6150 * dir items instead of logging one range item per dir index item.
6152 next = list_next_entry(curr, log_list);
6153 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6154 if (next->index != curr->index + 1)
6157 next = list_next_entry(next, log_list);
6160 last_dir_index = curr->index;
6161 ASSERT(last_dir_index >= first_dir_index);
6163 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6164 ino, first_dir_index, last_dir_index);
6167 curr = list_next_entry(curr, log_list);
6173 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6174 struct btrfs_inode *inode,
6175 struct btrfs_path *path,
6176 struct btrfs_log_ctx *ctx,
6177 const struct list_head *delayed_del_list,
6178 const struct btrfs_delayed_item *first,
6179 const struct btrfs_delayed_item **last_ret)
6181 const struct btrfs_delayed_item *next;
6182 struct extent_buffer *leaf = path->nodes[0];
6183 const int last_slot = btrfs_header_nritems(leaf) - 1;
6184 int slot = path->slots[0] + 1;
6185 const u64 ino = btrfs_ino(inode);
6187 next = list_next_entry(first, log_list);
6189 while (slot < last_slot &&
6190 !list_entry_is_head(next, delayed_del_list, log_list)) {
6191 struct btrfs_key key;
6193 btrfs_item_key_to_cpu(leaf, &key, slot);
6194 if (key.objectid != ino ||
6195 key.type != BTRFS_DIR_INDEX_KEY ||
6196 key.offset != next->index)
6201 next = list_next_entry(next, log_list);
6204 return btrfs_del_items(trans, inode->root->log_root, path,
6205 path->slots[0], slot - path->slots[0]);
6208 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6209 struct btrfs_inode *inode,
6210 struct btrfs_path *path,
6211 const struct list_head *delayed_del_list,
6212 struct btrfs_log_ctx *ctx)
6214 struct btrfs_root *log = inode->root->log_root;
6215 const struct btrfs_delayed_item *curr;
6216 u64 last_range_start = 0;
6217 u64 last_range_end = 0;
6218 struct btrfs_key key;
6220 key.objectid = btrfs_ino(inode);
6221 key.type = BTRFS_DIR_INDEX_KEY;
6222 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6225 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6226 const struct btrfs_delayed_item *last = curr;
6227 u64 first_dir_index = curr->index;
6229 bool deleted_items = false;
6232 key.offset = curr->index;
6233 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6236 } else if (ret == 0) {
6237 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6238 delayed_del_list, curr,
6242 deleted_items = true;
6245 btrfs_release_path(path);
6248 * If we deleted items from the leaf, it means we have a range
6249 * item logging their range, so no need to add one or update an
6250 * existing one. Otherwise we have to log a dir range item.
6255 last_dir_index = last->index;
6256 ASSERT(last_dir_index >= first_dir_index);
6258 * If this range starts right after where the previous one ends,
6259 * then we want to reuse the previous range item and change its
6260 * end offset to the end of this range. This is just to minimize
6261 * leaf space usage, by avoiding adding a new range item.
6263 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6264 first_dir_index = last_range_start;
6266 ret = insert_dir_log_key(trans, log, path, key.objectid,
6267 first_dir_index, last_dir_index);
6271 last_range_start = first_dir_index;
6272 last_range_end = last_dir_index;
6274 curr = list_next_entry(last, log_list);
6280 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6281 struct btrfs_inode *inode,
6282 struct btrfs_path *path,
6283 const struct list_head *delayed_del_list,
6284 struct btrfs_log_ctx *ctx)
6287 * We are deleting dir index items from the log tree or adding range
6290 lockdep_assert_held(&inode->log_mutex);
6292 if (list_empty(delayed_del_list))
6295 if (ctx->logged_before)
6296 return log_delayed_deletions_incremental(trans, inode, path,
6297 delayed_del_list, ctx);
6299 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6304 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6305 * items instead of the subvolume tree.
6307 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6308 struct btrfs_inode *inode,
6309 const struct list_head *delayed_ins_list,
6310 struct btrfs_log_ctx *ctx)
6312 const bool orig_log_new_dentries = ctx->log_new_dentries;
6313 struct btrfs_fs_info *fs_info = trans->fs_info;
6314 struct btrfs_delayed_item *item;
6318 * No need for the log mutex, plus to avoid potential deadlocks or
6319 * lockdep annotations due to nesting of delayed inode mutexes and log
6322 lockdep_assert_not_held(&inode->log_mutex);
6324 ASSERT(!ctx->logging_new_delayed_dentries);
6325 ctx->logging_new_delayed_dentries = true;
6327 list_for_each_entry(item, delayed_ins_list, log_list) {
6328 struct btrfs_dir_item *dir_item;
6329 struct inode *di_inode;
6330 struct btrfs_key key;
6331 int log_mode = LOG_INODE_EXISTS;
6333 dir_item = (struct btrfs_dir_item *)item->data;
6334 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6336 if (key.type == BTRFS_ROOT_ITEM_KEY)
6339 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6340 if (IS_ERR(di_inode)) {
6341 ret = PTR_ERR(di_inode);
6345 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6346 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6350 if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR)
6351 log_mode = LOG_INODE_ALL;
6353 ctx->log_new_dentries = false;
6354 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6356 if (!ret && ctx->log_new_dentries)
6357 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6359 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6365 ctx->log_new_dentries = orig_log_new_dentries;
6366 ctx->logging_new_delayed_dentries = false;
6371 /* log a single inode in the tree log.
6372 * At least one parent directory for this inode must exist in the tree
6373 * or be logged already.
6375 * Any items from this inode changed by the current transaction are copied
6376 * to the log tree. An extra reference is taken on any extents in this
6377 * file, allowing us to avoid a whole pile of corner cases around logging
6378 * blocks that have been removed from the tree.
6380 * See LOG_INODE_ALL and related defines for a description of what inode_only
6383 * This handles both files and directories.
6385 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6386 struct btrfs_inode *inode,
6388 struct btrfs_log_ctx *ctx)
6390 struct btrfs_path *path;
6391 struct btrfs_path *dst_path;
6392 struct btrfs_key min_key;
6393 struct btrfs_key max_key;
6394 struct btrfs_root *log = inode->root->log_root;
6396 bool fast_search = false;
6397 u64 ino = btrfs_ino(inode);
6398 struct extent_map_tree *em_tree = &inode->extent_tree;
6399 u64 logged_isize = 0;
6400 bool need_log_inode_item = true;
6401 bool xattrs_logged = false;
6402 bool inode_item_dropped = true;
6403 bool full_dir_logging = false;
6404 LIST_HEAD(delayed_ins_list);
6405 LIST_HEAD(delayed_del_list);
6407 path = btrfs_alloc_path();
6410 dst_path = btrfs_alloc_path();
6412 btrfs_free_path(path);
6416 min_key.objectid = ino;
6417 min_key.type = BTRFS_INODE_ITEM_KEY;
6420 max_key.objectid = ino;
6423 /* today the code can only do partial logging of directories */
6424 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6425 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6426 &inode->runtime_flags) &&
6427 inode_only >= LOG_INODE_EXISTS))
6428 max_key.type = BTRFS_XATTR_ITEM_KEY;
6430 max_key.type = (u8)-1;
6431 max_key.offset = (u64)-1;
6433 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6434 full_dir_logging = true;
6437 * If we are logging a directory while we are logging dentries of the
6438 * delayed items of some other inode, then we need to flush the delayed
6439 * items of this directory and not log the delayed items directly. This
6440 * is to prevent more than one level of recursion into btrfs_log_inode()
6441 * by having something like this:
6443 * $ mkdir -p a/b/c/d/e/f/g/h/...
6444 * $ xfs_io -c "fsync" a
6446 * Where all directories in the path did not exist before and are
6447 * created in the current transaction.
6448 * So in such a case we directly log the delayed items of the main
6449 * directory ("a") without flushing them first, while for each of its
6450 * subdirectories we flush their delayed items before logging them.
6451 * This prevents a potential unbounded recursion like this:
6454 * log_new_delayed_dentries()
6456 * log_new_delayed_dentries()
6458 * log_new_delayed_dentries()
6461 * We have thresholds for the maximum number of delayed items to have in
6462 * memory, and once they are hit, the items are flushed asynchronously.
6463 * However the limit is quite high, so lets prevent deep levels of
6464 * recursion to happen by limiting the maximum depth to be 1.
6466 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6467 ret = btrfs_commit_inode_delayed_items(trans, inode);
6472 mutex_lock(&inode->log_mutex);
6475 * For symlinks, we must always log their content, which is stored in an
6476 * inline extent, otherwise we could end up with an empty symlink after
6477 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6478 * one attempts to create an empty symlink).
6479 * We don't need to worry about flushing delalloc, because when we create
6480 * the inline extent when the symlink is created (we never have delalloc
6483 if (S_ISLNK(inode->vfs_inode.i_mode))
6484 inode_only = LOG_INODE_ALL;
6487 * Before logging the inode item, cache the value returned by
6488 * inode_logged(), because after that we have the need to figure out if
6489 * the inode was previously logged in this transaction.
6491 ret = inode_logged(trans, inode, path);
6494 ctx->logged_before = (ret == 1);
6498 * This is for cases where logging a directory could result in losing a
6499 * a file after replaying the log. For example, if we move a file from a
6500 * directory A to a directory B, then fsync directory A, we have no way
6501 * to known the file was moved from A to B, so logging just A would
6502 * result in losing the file after a log replay.
6504 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6505 ret = BTRFS_LOG_FORCE_COMMIT;
6510 * a brute force approach to making sure we get the most uptodate
6511 * copies of everything.
6513 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6514 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6515 if (ctx->logged_before)
6516 ret = drop_inode_items(trans, log, path, inode,
6517 BTRFS_XATTR_ITEM_KEY);
6519 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6521 * Make sure the new inode item we write to the log has
6522 * the same isize as the current one (if it exists).
6523 * This is necessary to prevent data loss after log
6524 * replay, and also to prevent doing a wrong expanding
6525 * truncate - for e.g. create file, write 4K into offset
6526 * 0, fsync, write 4K into offset 4096, add hard link,
6527 * fsync some other file (to sync log), power fail - if
6528 * we use the inode's current i_size, after log replay
6529 * we get a 8Kb file, with the last 4Kb extent as a hole
6530 * (zeroes), as if an expanding truncate happened,
6531 * instead of getting a file of 4Kb only.
6533 ret = logged_inode_size(log, inode, path, &logged_isize);
6537 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6538 &inode->runtime_flags)) {
6539 if (inode_only == LOG_INODE_EXISTS) {
6540 max_key.type = BTRFS_XATTR_ITEM_KEY;
6541 if (ctx->logged_before)
6542 ret = drop_inode_items(trans, log, path,
6543 inode, max_key.type);
6545 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6546 &inode->runtime_flags);
6547 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6548 &inode->runtime_flags);
6549 if (ctx->logged_before)
6550 ret = truncate_inode_items(trans, log,
6553 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6554 &inode->runtime_flags) ||
6555 inode_only == LOG_INODE_EXISTS) {
6556 if (inode_only == LOG_INODE_ALL)
6558 max_key.type = BTRFS_XATTR_ITEM_KEY;
6559 if (ctx->logged_before)
6560 ret = drop_inode_items(trans, log, path, inode,
6563 if (inode_only == LOG_INODE_ALL)
6565 inode_item_dropped = false;
6574 * If we are logging a directory in full mode, collect the delayed items
6575 * before iterating the subvolume tree, so that we don't miss any new
6576 * dir index items in case they get flushed while or right after we are
6577 * iterating the subvolume tree.
6579 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6580 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6583 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6584 path, dst_path, logged_isize,
6586 &need_log_inode_item);
6590 btrfs_release_path(path);
6591 btrfs_release_path(dst_path);
6592 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path, ctx);
6595 xattrs_logged = true;
6596 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6597 btrfs_release_path(path);
6598 btrfs_release_path(dst_path);
6599 ret = btrfs_log_holes(trans, inode, path);
6604 btrfs_release_path(path);
6605 btrfs_release_path(dst_path);
6606 if (need_log_inode_item) {
6607 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6611 * If we are doing a fast fsync and the inode was logged before
6612 * in this transaction, we don't need to log the xattrs because
6613 * they were logged before. If xattrs were added, changed or
6614 * deleted since the last time we logged the inode, then we have
6615 * already logged them because the inode had the runtime flag
6616 * BTRFS_INODE_COPY_EVERYTHING set.
6618 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6619 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path, ctx);
6622 btrfs_release_path(path);
6626 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6629 } else if (inode_only == LOG_INODE_ALL) {
6630 struct extent_map *em, *n;
6632 write_lock(&em_tree->lock);
6633 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6634 list_del_init(&em->list);
6635 write_unlock(&em_tree->lock);
6638 if (full_dir_logging) {
6639 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6642 ret = log_delayed_insertion_items(trans, inode, path,
6643 &delayed_ins_list, ctx);
6646 ret = log_delayed_deletion_items(trans, inode, path,
6647 &delayed_del_list, ctx);
6652 spin_lock(&inode->lock);
6653 inode->logged_trans = trans->transid;
6655 * Don't update last_log_commit if we logged that an inode exists.
6656 * We do this for three reasons:
6658 * 1) We might have had buffered writes to this inode that were
6659 * flushed and had their ordered extents completed in this
6660 * transaction, but we did not previously log the inode with
6661 * LOG_INODE_ALL. Later the inode was evicted and after that
6662 * it was loaded again and this LOG_INODE_EXISTS log operation
6663 * happened. We must make sure that if an explicit fsync against
6664 * the inode is performed later, it logs the new extents, an
6665 * updated inode item, etc, and syncs the log. The same logic
6666 * applies to direct IO writes instead of buffered writes.
6668 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6669 * is logged with an i_size of 0 or whatever value was logged
6670 * before. If later the i_size of the inode is increased by a
6671 * truncate operation, the log is synced through an fsync of
6672 * some other inode and then finally an explicit fsync against
6673 * this inode is made, we must make sure this fsync logs the
6674 * inode with the new i_size, the hole between old i_size and
6675 * the new i_size, and syncs the log.
6677 * 3) If we are logging that an ancestor inode exists as part of
6678 * logging a new name from a link or rename operation, don't update
6679 * its last_log_commit - otherwise if an explicit fsync is made
6680 * against an ancestor, the fsync considers the inode in the log
6681 * and doesn't sync the log, resulting in the ancestor missing after
6682 * a power failure unless the log was synced as part of an fsync
6683 * against any other unrelated inode.
6685 if (inode_only != LOG_INODE_EXISTS)
6686 inode->last_log_commit = inode->last_sub_trans;
6687 spin_unlock(&inode->lock);
6690 * Reset the last_reflink_trans so that the next fsync does not need to
6691 * go through the slower path when logging extents and their checksums.
6693 if (inode_only == LOG_INODE_ALL)
6694 inode->last_reflink_trans = 0;
6697 mutex_unlock(&inode->log_mutex);
6699 btrfs_free_path(path);
6700 btrfs_free_path(dst_path);
6703 free_conflicting_inodes(ctx);
6705 ret = log_conflicting_inodes(trans, inode->root, ctx);
6707 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6709 ret = log_new_delayed_dentries(trans, inode,
6710 &delayed_ins_list, ctx);
6712 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6719 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6720 struct btrfs_inode *inode,
6721 struct btrfs_log_ctx *ctx)
6723 struct btrfs_fs_info *fs_info = trans->fs_info;
6725 struct btrfs_path *path;
6726 struct btrfs_key key;
6727 struct btrfs_root *root = inode->root;
6728 const u64 ino = btrfs_ino(inode);
6730 path = btrfs_alloc_path();
6733 path->skip_locking = 1;
6734 path->search_commit_root = 1;
6737 key.type = BTRFS_INODE_REF_KEY;
6739 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6744 struct extent_buffer *leaf = path->nodes[0];
6745 int slot = path->slots[0];
6750 if (slot >= btrfs_header_nritems(leaf)) {
6751 ret = btrfs_next_leaf(root, path);
6759 btrfs_item_key_to_cpu(leaf, &key, slot);
6760 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6761 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6764 item_size = btrfs_item_size(leaf, slot);
6765 ptr = btrfs_item_ptr_offset(leaf, slot);
6766 while (cur_offset < item_size) {
6767 struct btrfs_key inode_key;
6768 struct inode *dir_inode;
6770 inode_key.type = BTRFS_INODE_ITEM_KEY;
6771 inode_key.offset = 0;
6773 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6774 struct btrfs_inode_extref *extref;
6776 extref = (struct btrfs_inode_extref *)
6778 inode_key.objectid = btrfs_inode_extref_parent(
6780 cur_offset += sizeof(*extref);
6781 cur_offset += btrfs_inode_extref_name_len(leaf,
6784 inode_key.objectid = key.offset;
6785 cur_offset = item_size;
6788 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6791 * If the parent inode was deleted, return an error to
6792 * fallback to a transaction commit. This is to prevent
6793 * getting an inode that was moved from one parent A to
6794 * a parent B, got its former parent A deleted and then
6795 * it got fsync'ed, from existing at both parents after
6796 * a log replay (and the old parent still existing).
6803 * mv /mnt/B/bar /mnt/A/bar
6804 * mv -T /mnt/A /mnt/B
6808 * If we ignore the old parent B which got deleted,
6809 * after a log replay we would have file bar linked
6810 * at both parents and the old parent B would still
6813 if (IS_ERR(dir_inode)) {
6814 ret = PTR_ERR(dir_inode);
6818 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6819 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6823 ctx->log_new_dentries = false;
6824 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6825 LOG_INODE_ALL, ctx);
6826 if (!ret && ctx->log_new_dentries)
6827 ret = log_new_dir_dentries(trans,
6828 BTRFS_I(dir_inode), ctx);
6829 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6837 btrfs_free_path(path);
6841 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6842 struct btrfs_root *root,
6843 struct btrfs_path *path,
6844 struct btrfs_log_ctx *ctx)
6846 struct btrfs_key found_key;
6848 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6851 struct btrfs_fs_info *fs_info = root->fs_info;
6852 struct extent_buffer *leaf;
6854 struct btrfs_key search_key;
6855 struct inode *inode;
6859 btrfs_release_path(path);
6861 ino = found_key.offset;
6863 search_key.objectid = found_key.offset;
6864 search_key.type = BTRFS_INODE_ITEM_KEY;
6865 search_key.offset = 0;
6866 inode = btrfs_iget(fs_info->sb, ino, root);
6868 return PTR_ERR(inode);
6870 if (BTRFS_I(inode)->generation >= trans->transid &&
6871 need_log_inode(trans, BTRFS_I(inode)))
6872 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6873 LOG_INODE_EXISTS, ctx);
6874 btrfs_add_delayed_iput(BTRFS_I(inode));
6878 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6881 search_key.type = BTRFS_INODE_REF_KEY;
6882 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6886 leaf = path->nodes[0];
6887 slot = path->slots[0];
6888 if (slot >= btrfs_header_nritems(leaf)) {
6889 ret = btrfs_next_leaf(root, path);
6894 leaf = path->nodes[0];
6895 slot = path->slots[0];
6898 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6899 if (found_key.objectid != search_key.objectid ||
6900 found_key.type != BTRFS_INODE_REF_KEY)
6906 static int log_new_ancestors_fast(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 struct dentry *old_parent = NULL;
6913 struct super_block *sb = inode->vfs_inode.i_sb;
6917 if (!parent || d_really_is_negative(parent) ||
6921 inode = BTRFS_I(d_inode(parent));
6922 if (root != inode->root)
6925 if (inode->generation >= trans->transid &&
6926 need_log_inode(trans, inode)) {
6927 ret = btrfs_log_inode(trans, inode,
6928 LOG_INODE_EXISTS, ctx);
6932 if (IS_ROOT(parent))
6935 parent = dget_parent(parent);
6937 old_parent = parent;
6944 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6945 struct btrfs_inode *inode,
6946 struct dentry *parent,
6947 struct btrfs_log_ctx *ctx)
6949 struct btrfs_root *root = inode->root;
6950 const u64 ino = btrfs_ino(inode);
6951 struct btrfs_path *path;
6952 struct btrfs_key search_key;
6956 * For a single hard link case, go through a fast path that does not
6957 * need to iterate the fs/subvolume tree.
6959 if (inode->vfs_inode.i_nlink < 2)
6960 return log_new_ancestors_fast(trans, inode, parent, ctx);
6962 path = btrfs_alloc_path();
6966 search_key.objectid = ino;
6967 search_key.type = BTRFS_INODE_REF_KEY;
6968 search_key.offset = 0;
6970 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6977 struct extent_buffer *leaf = path->nodes[0];
6978 int slot = path->slots[0];
6979 struct btrfs_key found_key;
6981 if (slot >= btrfs_header_nritems(leaf)) {
6982 ret = btrfs_next_leaf(root, path);
6990 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6991 if (found_key.objectid != ino ||
6992 found_key.type > BTRFS_INODE_EXTREF_KEY)
6996 * Don't deal with extended references because they are rare
6997 * cases and too complex to deal with (we would need to keep
6998 * track of which subitem we are processing for each item in
6999 * this loop, etc). So just return some error to fallback to
7000 * a transaction commit.
7002 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
7008 * Logging ancestors needs to do more searches on the fs/subvol
7009 * tree, so it releases the path as needed to avoid deadlocks.
7010 * Keep track of the last inode ref key and resume from that key
7011 * after logging all new ancestors for the current hard link.
7013 memcpy(&search_key, &found_key, sizeof(search_key));
7015 ret = log_new_ancestors(trans, root, path, ctx);
7018 btrfs_release_path(path);
7023 btrfs_free_path(path);
7028 * helper function around btrfs_log_inode to make sure newly created
7029 * parent directories also end up in the log. A minimal inode and backref
7030 * only logging is done of any parent directories that are older than
7031 * the last committed transaction
7033 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
7034 struct btrfs_inode *inode,
7035 struct dentry *parent,
7037 struct btrfs_log_ctx *ctx)
7039 struct btrfs_root *root = inode->root;
7040 struct btrfs_fs_info *fs_info = root->fs_info;
7042 bool log_dentries = false;
7044 if (btrfs_test_opt(fs_info, NOTREELOG)) {
7045 ret = BTRFS_LOG_FORCE_COMMIT;
7049 if (btrfs_root_refs(&root->root_item) == 0) {
7050 ret = BTRFS_LOG_FORCE_COMMIT;
7055 * Skip already logged inodes or inodes corresponding to tmpfiles
7056 * (since logging them is pointless, a link count of 0 means they
7057 * will never be accessible).
7059 if ((btrfs_inode_in_log(inode, trans->transid) &&
7060 list_empty(&ctx->ordered_extents)) ||
7061 inode->vfs_inode.i_nlink == 0) {
7062 ret = BTRFS_NO_LOG_SYNC;
7066 ret = start_log_trans(trans, root, ctx);
7070 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
7075 * for regular files, if its inode is already on disk, we don't
7076 * have to worry about the parents at all. This is because
7077 * we can use the last_unlink_trans field to record renames
7078 * and other fun in this file.
7080 if (S_ISREG(inode->vfs_inode.i_mode) &&
7081 inode->generation < trans->transid &&
7082 inode->last_unlink_trans < trans->transid) {
7087 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7088 log_dentries = true;
7091 * On unlink we must make sure all our current and old parent directory
7092 * inodes are fully logged. This is to prevent leaving dangling
7093 * directory index entries in directories that were our parents but are
7094 * not anymore. Not doing this results in old parent directory being
7095 * impossible to delete after log replay (rmdir will always fail with
7096 * error -ENOTEMPTY).
7102 * ln testdir/foo testdir/bar
7104 * unlink testdir/bar
7105 * xfs_io -c fsync testdir/foo
7107 * mount fs, triggers log replay
7109 * If we don't log the parent directory (testdir), after log replay the
7110 * directory still has an entry pointing to the file inode using the bar
7111 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7112 * the file inode has a link count of 1.
7118 * ln foo testdir/foo2
7119 * ln foo testdir/foo3
7121 * unlink testdir/foo3
7122 * xfs_io -c fsync foo
7124 * mount fs, triggers log replay
7126 * Similar as the first example, after log replay the parent directory
7127 * testdir still has an entry pointing to the inode file with name foo3
7128 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7129 * and has a link count of 2.
7131 if (inode->last_unlink_trans >= trans->transid) {
7132 ret = btrfs_log_all_parents(trans, inode, ctx);
7137 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7142 ret = log_new_dir_dentries(trans, inode, ctx);
7147 btrfs_set_log_full_commit(trans);
7148 ret = BTRFS_LOG_FORCE_COMMIT;
7152 btrfs_remove_log_ctx(root, ctx);
7153 btrfs_end_log_trans(root);
7159 * it is not safe to log dentry if the chunk root has added new
7160 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7161 * If this returns 1, you must commit the transaction to safely get your
7164 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7165 struct dentry *dentry,
7166 struct btrfs_log_ctx *ctx)
7168 struct dentry *parent = dget_parent(dentry);
7171 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7172 LOG_INODE_ALL, ctx);
7179 * should be called during mount to recover any replay any log trees
7182 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7185 struct btrfs_path *path;
7186 struct btrfs_trans_handle *trans;
7187 struct btrfs_key key;
7188 struct btrfs_key found_key;
7189 struct btrfs_root *log;
7190 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7191 struct walk_control wc = {
7192 .process_func = process_one_buffer,
7193 .stage = LOG_WALK_PIN_ONLY,
7196 path = btrfs_alloc_path();
7200 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7202 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7203 if (IS_ERR(trans)) {
7204 ret = PTR_ERR(trans);
7211 ret = walk_log_tree(trans, log_root_tree, &wc);
7213 btrfs_abort_transaction(trans, ret);
7218 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7219 key.offset = (u64)-1;
7220 key.type = BTRFS_ROOT_ITEM_KEY;
7223 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7226 btrfs_abort_transaction(trans, ret);
7230 if (path->slots[0] == 0)
7234 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7236 btrfs_release_path(path);
7237 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7240 log = btrfs_read_tree_root(log_root_tree, &found_key);
7243 btrfs_abort_transaction(trans, ret);
7247 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7249 if (IS_ERR(wc.replay_dest)) {
7250 ret = PTR_ERR(wc.replay_dest);
7253 * We didn't find the subvol, likely because it was
7254 * deleted. This is ok, simply skip this log and go to
7257 * We need to exclude the root because we can't have
7258 * other log replays overwriting this log as we'll read
7259 * it back in a few more times. This will keep our
7260 * block from being modified, and we'll just bail for
7261 * each subsequent pass.
7264 ret = btrfs_pin_extent_for_log_replay(trans, log->node);
7265 btrfs_put_root(log);
7269 btrfs_abort_transaction(trans, ret);
7273 wc.replay_dest->log_root = log;
7274 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7276 /* The loop needs to continue due to the root refs */
7277 btrfs_abort_transaction(trans, ret);
7279 ret = walk_log_tree(trans, log, &wc);
7281 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7282 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7285 btrfs_abort_transaction(trans, ret);
7288 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7289 struct btrfs_root *root = wc.replay_dest;
7291 btrfs_release_path(path);
7294 * We have just replayed everything, and the highest
7295 * objectid of fs roots probably has changed in case
7296 * some inode_item's got replayed.
7298 * root->objectid_mutex is not acquired as log replay
7299 * could only happen during mount.
7301 ret = btrfs_init_root_free_objectid(root);
7303 btrfs_abort_transaction(trans, ret);
7306 wc.replay_dest->log_root = NULL;
7307 btrfs_put_root(wc.replay_dest);
7308 btrfs_put_root(log);
7313 if (found_key.offset == 0)
7315 key.offset = found_key.offset - 1;
7317 btrfs_release_path(path);
7319 /* step one is to pin it all, step two is to replay just inodes */
7322 wc.process_func = replay_one_buffer;
7323 wc.stage = LOG_WALK_REPLAY_INODES;
7326 /* step three is to replay everything */
7327 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7332 btrfs_free_path(path);
7334 /* step 4: commit the transaction, which also unpins the blocks */
7335 ret = btrfs_commit_transaction(trans);
7339 log_root_tree->log_root = NULL;
7340 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7341 btrfs_put_root(log_root_tree);
7346 btrfs_end_transaction(wc.trans);
7347 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7348 btrfs_free_path(path);
7353 * there are some corner cases where we want to force a full
7354 * commit instead of allowing a directory to be logged.
7356 * They revolve around files there were unlinked from the directory, and
7357 * this function updates the parent directory so that a full commit is
7358 * properly done if it is fsync'd later after the unlinks are done.
7360 * Must be called before the unlink operations (updates to the subvolume tree,
7361 * inodes, etc) are done.
7363 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7364 struct btrfs_inode *dir, struct btrfs_inode *inode,
7368 * when we're logging a file, if it hasn't been renamed
7369 * or unlinked, and its inode is fully committed on disk,
7370 * we don't have to worry about walking up the directory chain
7371 * to log its parents.
7373 * So, we use the last_unlink_trans field to put this transid
7374 * into the file. When the file is logged we check it and
7375 * don't log the parents if the file is fully on disk.
7377 mutex_lock(&inode->log_mutex);
7378 inode->last_unlink_trans = trans->transid;
7379 mutex_unlock(&inode->log_mutex);
7385 * If this directory was already logged, any new names will be logged
7386 * with btrfs_log_new_name() and old names will be deleted from the log
7387 * tree with btrfs_del_dir_entries_in_log() or with
7388 * btrfs_del_inode_ref_in_log().
7390 if (inode_logged(trans, dir, NULL) == 1)
7394 * If the inode we're about to unlink was logged before, the log will be
7395 * properly updated with the new name with btrfs_log_new_name() and the
7396 * old name removed with btrfs_del_dir_entries_in_log() or with
7397 * btrfs_del_inode_ref_in_log().
7399 if (inode_logged(trans, inode, NULL) == 1)
7403 * when renaming files across directories, if the directory
7404 * there we're unlinking from gets fsync'd later on, there's
7405 * no way to find the destination directory later and fsync it
7406 * properly. So, we have to be conservative and force commits
7407 * so the new name gets discovered.
7409 mutex_lock(&dir->log_mutex);
7410 dir->last_unlink_trans = trans->transid;
7411 mutex_unlock(&dir->log_mutex);
7415 * Make sure that if someone attempts to fsync the parent directory of a deleted
7416 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7417 * that after replaying the log tree of the parent directory's root we will not
7418 * see the snapshot anymore and at log replay time we will not see any log tree
7419 * corresponding to the deleted snapshot's root, which could lead to replaying
7420 * it after replaying the log tree of the parent directory (which would replay
7421 * the snapshot delete operation).
7423 * Must be called before the actual snapshot destroy operation (updates to the
7424 * parent root and tree of tree roots trees, etc) are done.
7426 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7427 struct btrfs_inode *dir)
7429 mutex_lock(&dir->log_mutex);
7430 dir->last_unlink_trans = trans->transid;
7431 mutex_unlock(&dir->log_mutex);
7435 * Update the log after adding a new name for an inode.
7437 * @trans: Transaction handle.
7438 * @old_dentry: The dentry associated with the old name and the old
7440 * @old_dir: The inode of the previous parent directory for the case
7441 * of a rename. For a link operation, it must be NULL.
7442 * @old_dir_index: The index number associated with the old name, meaningful
7443 * only for rename operations (when @old_dir is not NULL).
7444 * Ignored for link operations.
7445 * @parent: The dentry associated with the directory under which the
7446 * new name is located.
7448 * Call this after adding a new name for an inode, as a result of a link or
7449 * rename operation, and it will properly update the log to reflect the new name.
7451 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7452 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7453 u64 old_dir_index, struct dentry *parent)
7455 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7456 struct btrfs_root *root = inode->root;
7457 struct btrfs_log_ctx ctx;
7458 bool log_pinned = false;
7462 * this will force the logging code to walk the dentry chain
7465 if (!S_ISDIR(inode->vfs_inode.i_mode))
7466 inode->last_unlink_trans = trans->transid;
7469 * if this inode hasn't been logged and directory we're renaming it
7470 * from hasn't been logged, we don't need to log it
7472 ret = inode_logged(trans, inode, NULL);
7475 } else if (ret == 0) {
7479 * If the inode was not logged and we are doing a rename (old_dir is not
7480 * NULL), check if old_dir was logged - if it was not we can return and
7483 ret = inode_logged(trans, old_dir, NULL);
7492 * If we are doing a rename (old_dir is not NULL) from a directory that
7493 * was previously logged, make sure that on log replay we get the old
7494 * dir entry deleted. This is needed because we will also log the new
7495 * name of the renamed inode, so we need to make sure that after log
7496 * replay we don't end up with both the new and old dir entries existing.
7498 if (old_dir && old_dir->logged_trans == trans->transid) {
7499 struct btrfs_root *log = old_dir->root->log_root;
7500 struct btrfs_path *path;
7501 struct fscrypt_name fname;
7503 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7505 ret = fscrypt_setup_filename(&old_dir->vfs_inode,
7506 &old_dentry->d_name, 0, &fname);
7510 * We have two inodes to update in the log, the old directory and
7511 * the inode that got renamed, so we must pin the log to prevent
7512 * anyone from syncing the log until we have updated both inodes
7515 ret = join_running_log_trans(root);
7517 * At least one of the inodes was logged before, so this should
7518 * not fail, but if it does, it's not serious, just bail out and
7519 * mark the log for a full commit.
7521 if (WARN_ON_ONCE(ret < 0)) {
7522 fscrypt_free_filename(&fname);
7528 path = btrfs_alloc_path();
7531 fscrypt_free_filename(&fname);
7536 * Other concurrent task might be logging the old directory,
7537 * as it can be triggered when logging other inode that had or
7538 * still has a dentry in the old directory. We lock the old
7539 * directory's log_mutex to ensure the deletion of the old
7540 * name is persisted, because during directory logging we
7541 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7542 * the old name's dir index item is in the delayed items, so
7543 * it could be missed by an in progress directory logging.
7545 mutex_lock(&old_dir->log_mutex);
7546 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7547 &fname.disk_name, old_dir_index);
7550 * The dentry does not exist in the log, so record its
7553 btrfs_release_path(path);
7554 ret = insert_dir_log_key(trans, log, path,
7556 old_dir_index, old_dir_index);
7558 mutex_unlock(&old_dir->log_mutex);
7560 btrfs_free_path(path);
7561 fscrypt_free_filename(&fname);
7566 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7567 ctx.logging_new_name = true;
7568 btrfs_init_log_ctx_scratch_eb(&ctx);
7570 * We don't care about the return value. If we fail to log the new name
7571 * then we know the next attempt to sync the log will fallback to a full
7572 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7573 * we don't need to worry about getting a log committed that has an
7574 * inconsistent state after a rename operation.
7576 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7577 free_extent_buffer(ctx.scratch_eb);
7578 ASSERT(list_empty(&ctx.conflict_inodes));
7581 * If an error happened mark the log for a full commit because it's not
7582 * consistent and up to date or we couldn't find out if one of the
7583 * inodes was logged before in this transaction. Do it before unpinning
7584 * the log, to avoid any races with someone else trying to commit it.
7587 btrfs_set_log_full_commit(trans);
7589 btrfs_end_log_trans(root);