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(btrfs_root_id(root) != 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) {
753 LIST_HEAD(ordered_sums);
756 * is this extent already allocated in the extent
757 * allocation tree? If so, just add a reference
759 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
763 } else if (ret == 0) {
764 struct btrfs_ref ref = {
765 .action = BTRFS_ADD_DELAYED_REF,
766 .bytenr = ins.objectid,
767 .num_bytes = ins.offset,
768 .owning_root = btrfs_root_id(root),
769 .ref_root = btrfs_root_id(root),
771 btrfs_init_data_ref(&ref, key->objectid, offset,
773 ret = btrfs_inc_extent_ref(trans, &ref);
778 * insert the extent pointer in the extent
781 ret = btrfs_alloc_logged_file_extent(trans,
783 key->objectid, offset, &ins);
787 btrfs_release_path(path);
789 if (btrfs_file_extent_compression(eb, item)) {
790 csum_start = ins.objectid;
791 csum_end = csum_start + ins.offset;
793 csum_start = ins.objectid +
794 btrfs_file_extent_offset(eb, item);
795 csum_end = csum_start +
796 btrfs_file_extent_num_bytes(eb, item);
799 ret = btrfs_lookup_csums_list(root->log_root,
800 csum_start, csum_end - 1,
801 &ordered_sums, false);
806 * Now delete all existing cums in the csum root that
807 * cover our range. We do this because we can have an
808 * extent that is completely referenced by one file
809 * extent item and partially referenced by another
810 * file extent item (like after using the clone or
811 * extent_same ioctls). In this case if we end up doing
812 * the replay of the one that partially references the
813 * extent first, and we do not do the csum deletion
814 * below, we can get 2 csum items in the csum tree that
815 * overlap each other. For example, imagine our log has
816 * the two following file extent items:
818 * key (257 EXTENT_DATA 409600)
819 * extent data disk byte 12845056 nr 102400
820 * extent data offset 20480 nr 20480 ram 102400
822 * key (257 EXTENT_DATA 819200)
823 * extent data disk byte 12845056 nr 102400
824 * extent data offset 0 nr 102400 ram 102400
826 * Where the second one fully references the 100K extent
827 * that starts at disk byte 12845056, and the log tree
828 * has a single csum item that covers the entire range
831 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
833 * After the first file extent item is replayed, the
834 * csum tree gets the following csum item:
836 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
838 * Which covers the 20K sub-range starting at offset 20K
839 * of our extent. Now when we replay the second file
840 * extent item, if we do not delete existing csum items
841 * that cover any of its blocks, we end up getting two
842 * csum items in our csum tree that overlap each other:
844 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
845 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
847 * Which is a problem, because after this anyone trying
848 * to lookup up for the checksum of any block of our
849 * extent starting at an offset of 40K or higher, will
850 * end up looking at the second csum item only, which
851 * does not contain the checksum for any block starting
852 * at offset 40K or higher of our extent.
854 while (!list_empty(&ordered_sums)) {
855 struct btrfs_ordered_sum *sums;
856 struct btrfs_root *csum_root;
858 sums = list_entry(ordered_sums.next,
859 struct btrfs_ordered_sum,
861 csum_root = btrfs_csum_root(fs_info,
864 ret = btrfs_del_csums(trans, csum_root,
868 ret = btrfs_csum_file_blocks(trans,
871 list_del(&sums->list);
877 btrfs_release_path(path);
879 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
880 /* inline extents are easy, we just overwrite them */
881 ret = overwrite_item(trans, root, path, eb, slot, key);
886 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
892 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
893 ret = btrfs_update_inode(trans, BTRFS_I(inode));
899 static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans,
900 struct btrfs_inode *dir,
901 struct btrfs_inode *inode,
902 const struct fscrypt_str *name)
906 ret = btrfs_unlink_inode(trans, dir, inode, name);
910 * Whenever we need to check if a name exists or not, we check the
911 * fs/subvolume tree. So after an unlink we must run delayed items, so
912 * that future checks for a name during log replay see that the name
913 * does not exists anymore.
915 return btrfs_run_delayed_items(trans);
919 * when cleaning up conflicts between the directory names in the
920 * subvolume, directory names in the log and directory names in the
921 * inode back references, we may have to unlink inodes from directories.
923 * This is a helper function to do the unlink of a specific directory
926 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
927 struct btrfs_path *path,
928 struct btrfs_inode *dir,
929 struct btrfs_dir_item *di)
931 struct btrfs_root *root = dir->root;
933 struct fscrypt_str name;
934 struct extent_buffer *leaf;
935 struct btrfs_key location;
938 leaf = path->nodes[0];
940 btrfs_dir_item_key_to_cpu(leaf, di, &location);
941 ret = read_alloc_one_name(leaf, di + 1, btrfs_dir_name_len(leaf, di), &name);
945 btrfs_release_path(path);
947 inode = read_one_inode(root, location.objectid);
953 ret = link_to_fixup_dir(trans, root, path, location.objectid);
957 ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), &name);
965 * See if a given name and sequence number found in an inode back reference are
966 * already in a directory and correctly point to this inode.
968 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
971 static noinline int inode_in_dir(struct btrfs_root *root,
972 struct btrfs_path *path,
973 u64 dirid, u64 objectid, u64 index,
974 struct fscrypt_str *name)
976 struct btrfs_dir_item *di;
977 struct btrfs_key location;
980 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
986 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
987 if (location.objectid != objectid)
993 btrfs_release_path(path);
994 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, 0);
999 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1000 if (location.objectid == objectid)
1004 btrfs_release_path(path);
1009 * helper function to check a log tree for a named back reference in
1010 * an inode. This is used to decide if a back reference that is
1011 * found in the subvolume conflicts with what we find in the log.
1013 * inode backreferences may have multiple refs in a single item,
1014 * during replay we process one reference at a time, and we don't
1015 * want to delete valid links to a file from the subvolume if that
1016 * link is also in the log.
1018 static noinline int backref_in_log(struct btrfs_root *log,
1019 struct btrfs_key *key,
1021 const struct fscrypt_str *name)
1023 struct btrfs_path *path;
1026 path = btrfs_alloc_path();
1030 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1033 } else if (ret == 1) {
1038 if (key->type == BTRFS_INODE_EXTREF_KEY)
1039 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1041 ref_objectid, name);
1043 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1044 path->slots[0], name);
1046 btrfs_free_path(path);
1050 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1051 struct btrfs_root *root,
1052 struct btrfs_path *path,
1053 struct btrfs_root *log_root,
1054 struct btrfs_inode *dir,
1055 struct btrfs_inode *inode,
1056 u64 inode_objectid, u64 parent_objectid,
1057 u64 ref_index, struct fscrypt_str *name)
1060 struct extent_buffer *leaf;
1061 struct btrfs_dir_item *di;
1062 struct btrfs_key search_key;
1063 struct btrfs_inode_extref *extref;
1066 /* Search old style refs */
1067 search_key.objectid = inode_objectid;
1068 search_key.type = BTRFS_INODE_REF_KEY;
1069 search_key.offset = parent_objectid;
1070 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1072 struct btrfs_inode_ref *victim_ref;
1074 unsigned long ptr_end;
1076 leaf = path->nodes[0];
1078 /* are we trying to overwrite a back ref for the root directory
1079 * if so, just jump out, we're done
1081 if (search_key.objectid == search_key.offset)
1084 /* check all the names in this back reference to see
1085 * if they are in the log. if so, we allow them to stay
1086 * otherwise they must be unlinked as a conflict
1088 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1089 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1090 while (ptr < ptr_end) {
1091 struct fscrypt_str victim_name;
1093 victim_ref = (struct btrfs_inode_ref *)ptr;
1094 ret = read_alloc_one_name(leaf, (victim_ref + 1),
1095 btrfs_inode_ref_name_len(leaf, victim_ref),
1100 ret = backref_in_log(log_root, &search_key,
1101 parent_objectid, &victim_name);
1103 kfree(victim_name.name);
1106 inc_nlink(&inode->vfs_inode);
1107 btrfs_release_path(path);
1109 ret = unlink_inode_for_log_replay(trans, dir, inode,
1111 kfree(victim_name.name);
1116 kfree(victim_name.name);
1118 ptr = (unsigned long)(victim_ref + 1) + victim_name.len;
1121 btrfs_release_path(path);
1123 /* Same search but for extended refs */
1124 extref = btrfs_lookup_inode_extref(NULL, root, path, name,
1125 inode_objectid, parent_objectid, 0,
1127 if (IS_ERR(extref)) {
1128 return PTR_ERR(extref);
1129 } else if (extref) {
1133 struct inode *victim_parent;
1135 leaf = path->nodes[0];
1137 item_size = btrfs_item_size(leaf, path->slots[0]);
1138 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1140 while (cur_offset < item_size) {
1141 struct fscrypt_str victim_name;
1143 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1145 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1148 ret = read_alloc_one_name(leaf, &extref->name,
1149 btrfs_inode_extref_name_len(leaf, extref),
1154 search_key.objectid = inode_objectid;
1155 search_key.type = BTRFS_INODE_EXTREF_KEY;
1156 search_key.offset = btrfs_extref_hash(parent_objectid,
1159 ret = backref_in_log(log_root, &search_key,
1160 parent_objectid, &victim_name);
1162 kfree(victim_name.name);
1166 victim_parent = read_one_inode(root,
1168 if (victim_parent) {
1169 inc_nlink(&inode->vfs_inode);
1170 btrfs_release_path(path);
1172 ret = unlink_inode_for_log_replay(trans,
1173 BTRFS_I(victim_parent),
1174 inode, &victim_name);
1176 iput(victim_parent);
1177 kfree(victim_name.name);
1182 kfree(victim_name.name);
1184 cur_offset += victim_name.len + sizeof(*extref);
1187 btrfs_release_path(path);
1189 /* look for a conflicting sequence number */
1190 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1191 ref_index, name, 0);
1195 ret = drop_one_dir_item(trans, path, dir, di);
1199 btrfs_release_path(path);
1201 /* look for a conflicting name */
1202 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), name, 0);
1206 ret = drop_one_dir_item(trans, path, dir, di);
1210 btrfs_release_path(path);
1215 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1216 struct fscrypt_str *name, u64 *index,
1217 u64 *parent_objectid)
1219 struct btrfs_inode_extref *extref;
1222 extref = (struct btrfs_inode_extref *)ref_ptr;
1224 ret = read_alloc_one_name(eb, &extref->name,
1225 btrfs_inode_extref_name_len(eb, extref), name);
1230 *index = btrfs_inode_extref_index(eb, extref);
1231 if (parent_objectid)
1232 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1237 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1238 struct fscrypt_str *name, u64 *index)
1240 struct btrfs_inode_ref *ref;
1243 ref = (struct btrfs_inode_ref *)ref_ptr;
1245 ret = read_alloc_one_name(eb, ref + 1, btrfs_inode_ref_name_len(eb, ref),
1251 *index = btrfs_inode_ref_index(eb, ref);
1257 * Take an inode reference item from the log tree and iterate all names from the
1258 * inode reference item in the subvolume tree with the same key (if it exists).
1259 * For any name that is not in the inode reference item from the log tree, do a
1260 * proper unlink of that name (that is, remove its entry from the inode
1261 * reference item and both dir index keys).
1263 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1264 struct btrfs_root *root,
1265 struct btrfs_path *path,
1266 struct btrfs_inode *inode,
1267 struct extent_buffer *log_eb,
1269 struct btrfs_key *key)
1272 unsigned long ref_ptr;
1273 unsigned long ref_end;
1274 struct extent_buffer *eb;
1277 btrfs_release_path(path);
1278 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1286 eb = path->nodes[0];
1287 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1288 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1289 while (ref_ptr < ref_end) {
1290 struct fscrypt_str name;
1293 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1294 ret = extref_get_fields(eb, ref_ptr, &name,
1297 parent_id = key->offset;
1298 ret = ref_get_fields(eb, ref_ptr, &name, NULL);
1303 if (key->type == BTRFS_INODE_EXTREF_KEY)
1304 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1307 ret = !!btrfs_find_name_in_backref(log_eb, log_slot, &name);
1312 btrfs_release_path(path);
1313 dir = read_one_inode(root, parent_id);
1319 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir),
1329 ref_ptr += name.len;
1330 if (key->type == BTRFS_INODE_EXTREF_KEY)
1331 ref_ptr += sizeof(struct btrfs_inode_extref);
1333 ref_ptr += sizeof(struct btrfs_inode_ref);
1337 btrfs_release_path(path);
1342 * replay one inode back reference item found in the log tree.
1343 * eb, slot and key refer to the buffer and key found in the log tree.
1344 * root is the destination we are replaying into, and path is for temp
1345 * use by this function. (it should be released on return).
1347 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1348 struct btrfs_root *root,
1349 struct btrfs_root *log,
1350 struct btrfs_path *path,
1351 struct extent_buffer *eb, int slot,
1352 struct btrfs_key *key)
1354 struct inode *dir = NULL;
1355 struct inode *inode = NULL;
1356 unsigned long ref_ptr;
1357 unsigned long ref_end;
1358 struct fscrypt_str name;
1360 int log_ref_ver = 0;
1361 u64 parent_objectid;
1364 int ref_struct_size;
1366 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1367 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1369 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1370 struct btrfs_inode_extref *r;
1372 ref_struct_size = sizeof(struct btrfs_inode_extref);
1374 r = (struct btrfs_inode_extref *)ref_ptr;
1375 parent_objectid = btrfs_inode_extref_parent(eb, r);
1377 ref_struct_size = sizeof(struct btrfs_inode_ref);
1378 parent_objectid = key->offset;
1380 inode_objectid = key->objectid;
1383 * it is possible that we didn't log all the parent directories
1384 * for a given inode. If we don't find the dir, just don't
1385 * copy the back ref in. The link count fixup code will take
1388 dir = read_one_inode(root, parent_objectid);
1394 inode = read_one_inode(root, inode_objectid);
1400 while (ref_ptr < ref_end) {
1402 ret = extref_get_fields(eb, ref_ptr, &name,
1403 &ref_index, &parent_objectid);
1405 * parent object can change from one array
1409 dir = read_one_inode(root, parent_objectid);
1415 ret = ref_get_fields(eb, ref_ptr, &name, &ref_index);
1420 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1421 btrfs_ino(BTRFS_I(inode)), ref_index, &name);
1424 } else if (ret == 0) {
1426 * look for a conflicting back reference in the
1427 * metadata. if we find one we have to unlink that name
1428 * of the file before we add our new link. Later on, we
1429 * overwrite any existing back reference, and we don't
1430 * want to create dangling pointers in the directory.
1432 ret = __add_inode_ref(trans, root, path, log,
1433 BTRFS_I(dir), BTRFS_I(inode),
1434 inode_objectid, parent_objectid,
1442 /* insert our name */
1443 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1444 &name, 0, ref_index);
1448 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1452 /* Else, ret == 1, we already have a perfect match, we're done. */
1454 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + name.len;
1464 * Before we overwrite the inode reference item in the subvolume tree
1465 * with the item from the log tree, we must unlink all names from the
1466 * parent directory that are in the subvolume's tree inode reference
1467 * item, otherwise we end up with an inconsistent subvolume tree where
1468 * dir index entries exist for a name but there is no inode reference
1469 * item with the same name.
1471 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1476 /* finally write the back reference in the inode */
1477 ret = overwrite_item(trans, root, path, eb, slot, key);
1479 btrfs_release_path(path);
1486 static int count_inode_extrefs(struct btrfs_inode *inode, struct btrfs_path *path)
1490 unsigned int nlink = 0;
1493 u64 inode_objectid = btrfs_ino(inode);
1496 struct btrfs_inode_extref *extref;
1497 struct extent_buffer *leaf;
1500 ret = btrfs_find_one_extref(inode->root, inode_objectid, offset,
1501 path, &extref, &offset);
1505 leaf = path->nodes[0];
1506 item_size = btrfs_item_size(leaf, path->slots[0]);
1507 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1510 while (cur_offset < item_size) {
1511 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1512 name_len = btrfs_inode_extref_name_len(leaf, extref);
1516 cur_offset += name_len + sizeof(*extref);
1520 btrfs_release_path(path);
1522 btrfs_release_path(path);
1524 if (ret < 0 && ret != -ENOENT)
1529 static int count_inode_refs(struct btrfs_inode *inode, struct btrfs_path *path)
1532 struct btrfs_key key;
1533 unsigned int nlink = 0;
1535 unsigned long ptr_end;
1537 u64 ino = btrfs_ino(inode);
1540 key.type = BTRFS_INODE_REF_KEY;
1541 key.offset = (u64)-1;
1544 ret = btrfs_search_slot(NULL, inode->root, &key, path, 0, 0);
1548 if (path->slots[0] == 0)
1553 btrfs_item_key_to_cpu(path->nodes[0], &key,
1555 if (key.objectid != ino ||
1556 key.type != BTRFS_INODE_REF_KEY)
1558 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1559 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1561 while (ptr < ptr_end) {
1562 struct btrfs_inode_ref *ref;
1564 ref = (struct btrfs_inode_ref *)ptr;
1565 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1567 ptr = (unsigned long)(ref + 1) + name_len;
1571 if (key.offset == 0)
1573 if (path->slots[0] > 0) {
1578 btrfs_release_path(path);
1580 btrfs_release_path(path);
1586 * There are a few corners where the link count of the file can't
1587 * be properly maintained during replay. So, instead of adding
1588 * lots of complexity to the log code, we just scan the backrefs
1589 * for any file that has been through replay.
1591 * The scan will update the link count on the inode to reflect the
1592 * number of back refs found. If it goes down to zero, the iput
1593 * will free the inode.
1595 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1596 struct inode *inode)
1598 struct btrfs_root *root = BTRFS_I(inode)->root;
1599 struct btrfs_path *path;
1602 u64 ino = btrfs_ino(BTRFS_I(inode));
1604 path = btrfs_alloc_path();
1608 ret = count_inode_refs(BTRFS_I(inode), path);
1614 ret = count_inode_extrefs(BTRFS_I(inode), path);
1622 if (nlink != inode->i_nlink) {
1623 set_nlink(inode, nlink);
1624 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1628 BTRFS_I(inode)->index_cnt = (u64)-1;
1630 if (inode->i_nlink == 0) {
1631 if (S_ISDIR(inode->i_mode)) {
1632 ret = replay_dir_deletes(trans, root, NULL, path,
1637 ret = btrfs_insert_orphan_item(trans, root, ino);
1643 btrfs_free_path(path);
1647 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1648 struct btrfs_root *root,
1649 struct btrfs_path *path)
1652 struct btrfs_key key;
1653 struct inode *inode;
1655 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1656 key.type = BTRFS_ORPHAN_ITEM_KEY;
1657 key.offset = (u64)-1;
1659 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1665 if (path->slots[0] == 0)
1670 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1671 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1672 key.type != BTRFS_ORPHAN_ITEM_KEY)
1675 ret = btrfs_del_item(trans, root, path);
1679 btrfs_release_path(path);
1680 inode = read_one_inode(root, key.offset);
1686 ret = fixup_inode_link_count(trans, inode);
1692 * fixup on a directory may create new entries,
1693 * make sure we always look for the highset possible
1696 key.offset = (u64)-1;
1698 btrfs_release_path(path);
1704 * record a given inode in the fixup dir so we can check its link
1705 * count when replay is done. The link count is incremented here
1706 * so the inode won't go away until we check it
1708 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root,
1710 struct btrfs_path *path,
1713 struct btrfs_key key;
1715 struct inode *inode;
1717 inode = read_one_inode(root, objectid);
1721 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1722 key.type = BTRFS_ORPHAN_ITEM_KEY;
1723 key.offset = objectid;
1725 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1727 btrfs_release_path(path);
1729 if (!inode->i_nlink)
1730 set_nlink(inode, 1);
1733 ret = btrfs_update_inode(trans, BTRFS_I(inode));
1734 } else if (ret == -EEXIST) {
1743 * when replaying the log for a directory, we only insert names
1744 * for inodes that actually exist. This means an fsync on a directory
1745 * does not implicitly fsync all the new files in it
1747 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1748 struct btrfs_root *root,
1749 u64 dirid, u64 index,
1750 const struct fscrypt_str *name,
1751 struct btrfs_key *location)
1753 struct inode *inode;
1757 inode = read_one_inode(root, location->objectid);
1761 dir = read_one_inode(root, dirid);
1767 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1770 /* FIXME, put inode into FIXUP list */
1777 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1778 struct btrfs_inode *dir,
1779 struct btrfs_path *path,
1780 struct btrfs_dir_item *dst_di,
1781 const struct btrfs_key *log_key,
1785 struct btrfs_key found_key;
1787 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1788 /* The existing dentry points to the same inode, don't delete it. */
1789 if (found_key.objectid == log_key->objectid &&
1790 found_key.type == log_key->type &&
1791 found_key.offset == log_key->offset &&
1792 btrfs_dir_flags(path->nodes[0], dst_di) == log_flags)
1796 * Don't drop the conflicting directory entry if the inode for the new
1797 * entry doesn't exist.
1802 return drop_one_dir_item(trans, path, dir, dst_di);
1806 * take a single entry in a log directory item and replay it into
1809 * if a conflicting item exists in the subdirectory already,
1810 * the inode it points to is unlinked and put into the link count
1813 * If a name from the log points to a file or directory that does
1814 * not exist in the FS, it is skipped. fsyncs on directories
1815 * do not force down inodes inside that directory, just changes to the
1816 * names or unlinks in a directory.
1818 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1819 * non-existing inode) and 1 if the name was replayed.
1821 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1822 struct btrfs_root *root,
1823 struct btrfs_path *path,
1824 struct extent_buffer *eb,
1825 struct btrfs_dir_item *di,
1826 struct btrfs_key *key)
1828 struct fscrypt_str name;
1829 struct btrfs_dir_item *dir_dst_di;
1830 struct btrfs_dir_item *index_dst_di;
1831 bool dir_dst_matches = false;
1832 bool index_dst_matches = false;
1833 struct btrfs_key log_key;
1834 struct btrfs_key search_key;
1839 bool update_size = true;
1840 bool name_added = false;
1842 dir = read_one_inode(root, key->objectid);
1846 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
1850 log_flags = btrfs_dir_flags(eb, di);
1851 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1852 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1853 btrfs_release_path(path);
1856 exists = (ret == 0);
1859 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1861 if (IS_ERR(dir_dst_di)) {
1862 ret = PTR_ERR(dir_dst_di);
1864 } else if (dir_dst_di) {
1865 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1866 dir_dst_di, &log_key,
1870 dir_dst_matches = (ret == 1);
1873 btrfs_release_path(path);
1875 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1876 key->objectid, key->offset,
1878 if (IS_ERR(index_dst_di)) {
1879 ret = PTR_ERR(index_dst_di);
1881 } else if (index_dst_di) {
1882 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
1883 index_dst_di, &log_key,
1887 index_dst_matches = (ret == 1);
1890 btrfs_release_path(path);
1892 if (dir_dst_matches && index_dst_matches) {
1894 update_size = false;
1899 * Check if the inode reference exists in the log for the given name,
1900 * inode and parent inode
1902 search_key.objectid = log_key.objectid;
1903 search_key.type = BTRFS_INODE_REF_KEY;
1904 search_key.offset = key->objectid;
1905 ret = backref_in_log(root->log_root, &search_key, 0, &name);
1909 /* The dentry will be added later. */
1911 update_size = false;
1915 search_key.objectid = log_key.objectid;
1916 search_key.type = BTRFS_INODE_EXTREF_KEY;
1917 search_key.offset = key->objectid;
1918 ret = backref_in_log(root->log_root, &search_key, key->objectid, &name);
1922 /* The dentry will be added later. */
1924 update_size = false;
1927 btrfs_release_path(path);
1928 ret = insert_one_name(trans, root, key->objectid, key->offset,
1930 if (ret && ret != -ENOENT && ret != -EEXIST)
1934 update_size = false;
1938 if (!ret && update_size) {
1939 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2);
1940 ret = btrfs_update_inode(trans, BTRFS_I(dir));
1944 if (!ret && name_added)
1949 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
1950 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1951 struct btrfs_root *root,
1952 struct btrfs_path *path,
1953 struct extent_buffer *eb, int slot,
1954 struct btrfs_key *key)
1957 struct btrfs_dir_item *di;
1959 /* We only log dir index keys, which only contain a single dir item. */
1960 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
1962 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1963 ret = replay_one_name(trans, root, path, eb, di, key);
1968 * If this entry refers to a non-directory (directories can not have a
1969 * link count > 1) and it was added in the transaction that was not
1970 * committed, make sure we fixup the link count of the inode the entry
1971 * points to. Otherwise something like the following would result in a
1972 * directory pointing to an inode with a wrong link that does not account
1973 * for this dir entry:
1980 * ln testdir/bar testdir/bar_link
1981 * ln testdir/foo testdir/foo_link
1982 * xfs_io -c "fsync" testdir/bar
1986 * mount fs, log replay happens
1988 * File foo would remain with a link count of 1 when it has two entries
1989 * pointing to it in the directory testdir. This would make it impossible
1990 * to ever delete the parent directory has it would result in stale
1991 * dentries that can never be deleted.
1993 if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) {
1994 struct btrfs_path *fixup_path;
1995 struct btrfs_key di_key;
1997 fixup_path = btrfs_alloc_path();
2001 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2002 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2003 btrfs_free_path(fixup_path);
2010 * directory replay has two parts. There are the standard directory
2011 * items in the log copied from the subvolume, and range items
2012 * created in the log while the subvolume was logged.
2014 * The range items tell us which parts of the key space the log
2015 * is authoritative for. During replay, if a key in the subvolume
2016 * directory is in a logged range item, but not actually in the log
2017 * that means it was deleted from the directory before the fsync
2018 * and should be removed.
2020 static noinline int find_dir_range(struct btrfs_root *root,
2021 struct btrfs_path *path,
2023 u64 *start_ret, u64 *end_ret)
2025 struct btrfs_key key;
2027 struct btrfs_dir_log_item *item;
2031 if (*start_ret == (u64)-1)
2034 key.objectid = dirid;
2035 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2036 key.offset = *start_ret;
2038 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2042 if (path->slots[0] == 0)
2047 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2049 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2053 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2054 struct btrfs_dir_log_item);
2055 found_end = btrfs_dir_log_end(path->nodes[0], item);
2057 if (*start_ret >= key.offset && *start_ret <= found_end) {
2059 *start_ret = key.offset;
2060 *end_ret = found_end;
2065 /* check the next slot in the tree to see if it is a valid item */
2066 nritems = btrfs_header_nritems(path->nodes[0]);
2068 if (path->slots[0] >= nritems) {
2069 ret = btrfs_next_leaf(root, path);
2074 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2076 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2080 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2081 struct btrfs_dir_log_item);
2082 found_end = btrfs_dir_log_end(path->nodes[0], item);
2083 *start_ret = key.offset;
2084 *end_ret = found_end;
2087 btrfs_release_path(path);
2092 * this looks for a given directory item in the log. If the directory
2093 * item is not in the log, the item is removed and the inode it points
2096 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *log,
2098 struct btrfs_path *path,
2099 struct btrfs_path *log_path,
2101 struct btrfs_key *dir_key)
2103 struct btrfs_root *root = BTRFS_I(dir)->root;
2105 struct extent_buffer *eb;
2107 struct btrfs_dir_item *di;
2108 struct fscrypt_str name;
2109 struct inode *inode = NULL;
2110 struct btrfs_key location;
2113 * Currently we only log dir index keys. Even if we replay a log created
2114 * by an older kernel that logged both dir index and dir item keys, all
2115 * we need to do is process the dir index keys, we (and our caller) can
2116 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2118 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2120 eb = path->nodes[0];
2121 slot = path->slots[0];
2122 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2123 ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name);
2128 struct btrfs_dir_item *log_di;
2130 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2132 dir_key->offset, &name, 0);
2133 if (IS_ERR(log_di)) {
2134 ret = PTR_ERR(log_di);
2136 } else if (log_di) {
2137 /* The dentry exists in the log, we have nothing to do. */
2143 btrfs_dir_item_key_to_cpu(eb, di, &location);
2144 btrfs_release_path(path);
2145 btrfs_release_path(log_path);
2146 inode = read_one_inode(root, location.objectid);
2152 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2157 ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode),
2160 * Unlike dir item keys, dir index keys can only have one name (entry) in
2161 * them, as there are no key collisions since each key has a unique offset
2162 * (an index number), so we're done.
2165 btrfs_release_path(path);
2166 btrfs_release_path(log_path);
2172 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2173 struct btrfs_root *root,
2174 struct btrfs_root *log,
2175 struct btrfs_path *path,
2178 struct btrfs_key search_key;
2179 struct btrfs_path *log_path;
2184 log_path = btrfs_alloc_path();
2188 search_key.objectid = ino;
2189 search_key.type = BTRFS_XATTR_ITEM_KEY;
2190 search_key.offset = 0;
2192 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2196 nritems = btrfs_header_nritems(path->nodes[0]);
2197 for (i = path->slots[0]; i < nritems; i++) {
2198 struct btrfs_key key;
2199 struct btrfs_dir_item *di;
2200 struct btrfs_dir_item *log_di;
2204 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2205 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2210 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2211 total_size = btrfs_item_size(path->nodes[0], i);
2213 while (cur < total_size) {
2214 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2215 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2216 u32 this_len = sizeof(*di) + name_len + data_len;
2219 name = kmalloc(name_len, GFP_NOFS);
2224 read_extent_buffer(path->nodes[0], name,
2225 (unsigned long)(di + 1), name_len);
2227 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2229 btrfs_release_path(log_path);
2231 /* Doesn't exist in log tree, so delete it. */
2232 btrfs_release_path(path);
2233 di = btrfs_lookup_xattr(trans, root, path, ino,
2234 name, name_len, -1);
2241 ret = btrfs_delete_one_dir_name(trans, root,
2245 btrfs_release_path(path);
2250 if (IS_ERR(log_di)) {
2251 ret = PTR_ERR(log_di);
2255 di = (struct btrfs_dir_item *)((char *)di + this_len);
2258 ret = btrfs_next_leaf(root, path);
2264 btrfs_free_path(log_path);
2265 btrfs_release_path(path);
2271 * deletion replay happens before we copy any new directory items
2272 * out of the log or out of backreferences from inodes. It
2273 * scans the log to find ranges of keys that log is authoritative for,
2274 * and then scans the directory to find items in those ranges that are
2275 * not present in the log.
2277 * Anything we don't find in the log is unlinked and removed from the
2280 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2281 struct btrfs_root *root,
2282 struct btrfs_root *log,
2283 struct btrfs_path *path,
2284 u64 dirid, int del_all)
2289 struct btrfs_key dir_key;
2290 struct btrfs_key found_key;
2291 struct btrfs_path *log_path;
2294 dir_key.objectid = dirid;
2295 dir_key.type = BTRFS_DIR_INDEX_KEY;
2296 log_path = btrfs_alloc_path();
2300 dir = read_one_inode(root, dirid);
2301 /* it isn't an error if the inode isn't there, that can happen
2302 * because we replay the deletes before we copy in the inode item
2306 btrfs_free_path(log_path);
2314 range_end = (u64)-1;
2316 ret = find_dir_range(log, path, dirid,
2317 &range_start, &range_end);
2324 dir_key.offset = range_start;
2327 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2332 nritems = btrfs_header_nritems(path->nodes[0]);
2333 if (path->slots[0] >= nritems) {
2334 ret = btrfs_next_leaf(root, path);
2340 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2342 if (found_key.objectid != dirid ||
2343 found_key.type != dir_key.type) {
2348 if (found_key.offset > range_end)
2351 ret = check_item_in_log(trans, log, path,
2356 if (found_key.offset == (u64)-1)
2358 dir_key.offset = found_key.offset + 1;
2360 btrfs_release_path(path);
2361 if (range_end == (u64)-1)
2363 range_start = range_end + 1;
2367 btrfs_release_path(path);
2368 btrfs_free_path(log_path);
2374 * the process_func used to replay items from the log tree. This
2375 * gets called in two different stages. The first stage just looks
2376 * for inodes and makes sure they are all copied into the subvolume.
2378 * The second stage copies all the other item types from the log into
2379 * the subvolume. The two stage approach is slower, but gets rid of
2380 * lots of complexity around inodes referencing other inodes that exist
2381 * only in the log (references come from either directory items or inode
2384 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2385 struct walk_control *wc, u64 gen, int level)
2388 struct btrfs_tree_parent_check check = {
2392 struct btrfs_path *path;
2393 struct btrfs_root *root = wc->replay_dest;
2394 struct btrfs_key key;
2398 ret = btrfs_read_extent_buffer(eb, &check);
2402 level = btrfs_header_level(eb);
2407 path = btrfs_alloc_path();
2411 nritems = btrfs_header_nritems(eb);
2412 for (i = 0; i < nritems; i++) {
2413 btrfs_item_key_to_cpu(eb, &key, i);
2415 /* inode keys are done during the first stage */
2416 if (key.type == BTRFS_INODE_ITEM_KEY &&
2417 wc->stage == LOG_WALK_REPLAY_INODES) {
2418 struct btrfs_inode_item *inode_item;
2421 inode_item = btrfs_item_ptr(eb, i,
2422 struct btrfs_inode_item);
2424 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2425 * and never got linked before the fsync, skip it, as
2426 * replaying it is pointless since it would be deleted
2427 * later. We skip logging tmpfiles, but it's always
2428 * possible we are replaying a log created with a kernel
2429 * that used to log tmpfiles.
2431 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2432 wc->ignore_cur_inode = true;
2435 wc->ignore_cur_inode = false;
2437 ret = replay_xattr_deletes(wc->trans, root, log,
2438 path, key.objectid);
2441 mode = btrfs_inode_mode(eb, inode_item);
2442 if (S_ISDIR(mode)) {
2443 ret = replay_dir_deletes(wc->trans,
2444 root, log, path, key.objectid, 0);
2448 ret = overwrite_item(wc->trans, root, path,
2454 * Before replaying extents, truncate the inode to its
2455 * size. We need to do it now and not after log replay
2456 * because before an fsync we can have prealloc extents
2457 * added beyond the inode's i_size. If we did it after,
2458 * through orphan cleanup for example, we would drop
2459 * those prealloc extents just after replaying them.
2461 if (S_ISREG(mode)) {
2462 struct btrfs_drop_extents_args drop_args = { 0 };
2463 struct inode *inode;
2466 inode = read_one_inode(root, key.objectid);
2471 from = ALIGN(i_size_read(inode),
2472 root->fs_info->sectorsize);
2473 drop_args.start = from;
2474 drop_args.end = (u64)-1;
2475 drop_args.drop_cache = true;
2476 ret = btrfs_drop_extents(wc->trans, root,
2480 inode_sub_bytes(inode,
2481 drop_args.bytes_found);
2482 /* Update the inode's nbytes. */
2483 ret = btrfs_update_inode(wc->trans,
2491 ret = link_to_fixup_dir(wc->trans, root,
2492 path, key.objectid);
2497 if (wc->ignore_cur_inode)
2500 if (key.type == BTRFS_DIR_INDEX_KEY &&
2501 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2502 ret = replay_one_dir_item(wc->trans, root, path,
2508 if (wc->stage < LOG_WALK_REPLAY_ALL)
2511 /* these keys are simply copied */
2512 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2513 ret = overwrite_item(wc->trans, root, path,
2517 } else if (key.type == BTRFS_INODE_REF_KEY ||
2518 key.type == BTRFS_INODE_EXTREF_KEY) {
2519 ret = add_inode_ref(wc->trans, root, log, path,
2521 if (ret && ret != -ENOENT)
2524 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2525 ret = replay_one_extent(wc->trans, root, path,
2531 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2532 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2533 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2534 * older kernel with such keys, ignore them.
2537 btrfs_free_path(path);
2542 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2544 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2546 struct btrfs_block_group *cache;
2548 cache = btrfs_lookup_block_group(fs_info, start);
2550 btrfs_err(fs_info, "unable to find block group for %llu", start);
2554 spin_lock(&cache->space_info->lock);
2555 spin_lock(&cache->lock);
2556 cache->reserved -= fs_info->nodesize;
2557 cache->space_info->bytes_reserved -= fs_info->nodesize;
2558 spin_unlock(&cache->lock);
2559 spin_unlock(&cache->space_info->lock);
2561 btrfs_put_block_group(cache);
2564 static int clean_log_buffer(struct btrfs_trans_handle *trans,
2565 struct extent_buffer *eb)
2569 btrfs_tree_lock(eb);
2570 btrfs_clear_buffer_dirty(trans, eb);
2571 wait_on_extent_buffer_writeback(eb);
2572 btrfs_tree_unlock(eb);
2575 ret = btrfs_pin_reserved_extent(trans, eb);
2579 unaccount_log_buffer(eb->fs_info, eb->start);
2585 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2586 struct btrfs_root *root,
2587 struct btrfs_path *path, int *level,
2588 struct walk_control *wc)
2590 struct btrfs_fs_info *fs_info = root->fs_info;
2593 struct extent_buffer *next;
2594 struct extent_buffer *cur;
2597 while (*level > 0) {
2598 struct btrfs_tree_parent_check check = { 0 };
2600 cur = path->nodes[*level];
2602 WARN_ON(btrfs_header_level(cur) != *level);
2604 if (path->slots[*level] >=
2605 btrfs_header_nritems(cur))
2608 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2609 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2610 check.transid = ptr_gen;
2611 check.level = *level - 1;
2612 check.has_first_key = true;
2613 btrfs_node_key_to_cpu(cur, &check.first_key, path->slots[*level]);
2615 next = btrfs_find_create_tree_block(fs_info, bytenr,
2616 btrfs_header_owner(cur),
2619 return PTR_ERR(next);
2622 ret = wc->process_func(root, next, wc, ptr_gen,
2625 free_extent_buffer(next);
2629 path->slots[*level]++;
2631 ret = btrfs_read_extent_buffer(next, &check);
2633 free_extent_buffer(next);
2637 ret = clean_log_buffer(trans, next);
2639 free_extent_buffer(next);
2643 free_extent_buffer(next);
2646 ret = btrfs_read_extent_buffer(next, &check);
2648 free_extent_buffer(next);
2652 if (path->nodes[*level-1])
2653 free_extent_buffer(path->nodes[*level-1]);
2654 path->nodes[*level-1] = next;
2655 *level = btrfs_header_level(next);
2656 path->slots[*level] = 0;
2659 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2665 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *root,
2667 struct btrfs_path *path, int *level,
2668 struct walk_control *wc)
2674 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2675 slot = path->slots[i];
2676 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2679 WARN_ON(*level == 0);
2682 ret = wc->process_func(root, path->nodes[*level], wc,
2683 btrfs_header_generation(path->nodes[*level]),
2689 ret = clean_log_buffer(trans, path->nodes[*level]);
2693 free_extent_buffer(path->nodes[*level]);
2694 path->nodes[*level] = NULL;
2702 * drop the reference count on the tree rooted at 'snap'. This traverses
2703 * the tree freeing any blocks that have a ref count of zero after being
2706 static int walk_log_tree(struct btrfs_trans_handle *trans,
2707 struct btrfs_root *log, struct walk_control *wc)
2712 struct btrfs_path *path;
2715 path = btrfs_alloc_path();
2719 level = btrfs_header_level(log->node);
2721 path->nodes[level] = log->node;
2722 atomic_inc(&log->node->refs);
2723 path->slots[level] = 0;
2726 wret = walk_down_log_tree(trans, log, path, &level, wc);
2734 wret = walk_up_log_tree(trans, log, path, &level, wc);
2743 /* was the root node processed? if not, catch it here */
2744 if (path->nodes[orig_level]) {
2745 ret = wc->process_func(log, path->nodes[orig_level], wc,
2746 btrfs_header_generation(path->nodes[orig_level]),
2751 ret = clean_log_buffer(trans, path->nodes[orig_level]);
2755 btrfs_free_path(path);
2760 * helper function to update the item for a given subvolumes log root
2761 * in the tree of log roots
2763 static int update_log_root(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *log,
2765 struct btrfs_root_item *root_item)
2767 struct btrfs_fs_info *fs_info = log->fs_info;
2770 if (log->log_transid == 1) {
2771 /* insert root item on the first sync */
2772 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2773 &log->root_key, root_item);
2775 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2776 &log->root_key, root_item);
2781 static void wait_log_commit(struct btrfs_root *root, int transid)
2784 int index = transid % 2;
2787 * we only allow two pending log transactions at a time,
2788 * so we know that if ours is more than 2 older than the
2789 * current transaction, we're done
2792 prepare_to_wait(&root->log_commit_wait[index],
2793 &wait, TASK_UNINTERRUPTIBLE);
2795 if (!(root->log_transid_committed < transid &&
2796 atomic_read(&root->log_commit[index])))
2799 mutex_unlock(&root->log_mutex);
2801 mutex_lock(&root->log_mutex);
2803 finish_wait(&root->log_commit_wait[index], &wait);
2806 static void wait_for_writer(struct btrfs_root *root)
2811 prepare_to_wait(&root->log_writer_wait, &wait,
2812 TASK_UNINTERRUPTIBLE);
2813 if (!atomic_read(&root->log_writers))
2816 mutex_unlock(&root->log_mutex);
2818 mutex_lock(&root->log_mutex);
2820 finish_wait(&root->log_writer_wait, &wait);
2823 void btrfs_init_log_ctx(struct btrfs_log_ctx *ctx, struct inode *inode)
2826 ctx->log_transid = 0;
2827 ctx->log_new_dentries = false;
2828 ctx->logging_new_name = false;
2829 ctx->logging_new_delayed_dentries = false;
2830 ctx->logged_before = false;
2832 INIT_LIST_HEAD(&ctx->list);
2833 INIT_LIST_HEAD(&ctx->ordered_extents);
2834 INIT_LIST_HEAD(&ctx->conflict_inodes);
2835 ctx->num_conflict_inodes = 0;
2836 ctx->logging_conflict_inodes = false;
2837 ctx->scratch_eb = NULL;
2840 void btrfs_init_log_ctx_scratch_eb(struct btrfs_log_ctx *ctx)
2842 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
2844 if (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
2845 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
2849 * Don't care about allocation failure. This is just for optimization,
2850 * if we fail to allocate here, we will try again later if needed.
2852 ctx->scratch_eb = alloc_dummy_extent_buffer(inode->root->fs_info, 0);
2855 void btrfs_release_log_ctx_extents(struct btrfs_log_ctx *ctx)
2857 struct btrfs_ordered_extent *ordered;
2858 struct btrfs_ordered_extent *tmp;
2860 ASSERT(inode_is_locked(ctx->inode));
2862 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
2863 list_del_init(&ordered->log_list);
2864 btrfs_put_ordered_extent(ordered);
2869 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2870 struct btrfs_log_ctx *ctx)
2872 mutex_lock(&root->log_mutex);
2873 list_del_init(&ctx->list);
2874 mutex_unlock(&root->log_mutex);
2878 * Invoked in log mutex context, or be sure there is no other task which
2879 * can access the list.
2881 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2882 int index, int error)
2884 struct btrfs_log_ctx *ctx;
2885 struct btrfs_log_ctx *safe;
2887 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2888 list_del_init(&ctx->list);
2889 ctx->log_ret = error;
2894 * Sends a given tree log down to the disk and updates the super blocks to
2895 * record it. When this call is done, you know that any inodes previously
2896 * logged are safely on disk only if it returns 0.
2898 * Any other return value means you need to call btrfs_commit_transaction.
2899 * Some of the edge cases for fsyncing directories that have had unlinks
2900 * or renames done in the past mean that sometimes the only safe
2901 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2902 * that has happened.
2904 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2905 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2911 struct btrfs_fs_info *fs_info = root->fs_info;
2912 struct btrfs_root *log = root->log_root;
2913 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2914 struct btrfs_root_item new_root_item;
2915 int log_transid = 0;
2916 struct btrfs_log_ctx root_log_ctx;
2917 struct blk_plug plug;
2921 mutex_lock(&root->log_mutex);
2922 log_transid = ctx->log_transid;
2923 if (root->log_transid_committed >= log_transid) {
2924 mutex_unlock(&root->log_mutex);
2925 return ctx->log_ret;
2928 index1 = log_transid % 2;
2929 if (atomic_read(&root->log_commit[index1])) {
2930 wait_log_commit(root, log_transid);
2931 mutex_unlock(&root->log_mutex);
2932 return ctx->log_ret;
2934 ASSERT(log_transid == root->log_transid);
2935 atomic_set(&root->log_commit[index1], 1);
2937 /* wait for previous tree log sync to complete */
2938 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2939 wait_log_commit(root, log_transid - 1);
2942 int batch = atomic_read(&root->log_batch);
2943 /* when we're on an ssd, just kick the log commit out */
2944 if (!btrfs_test_opt(fs_info, SSD) &&
2945 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2946 mutex_unlock(&root->log_mutex);
2947 schedule_timeout_uninterruptible(1);
2948 mutex_lock(&root->log_mutex);
2950 wait_for_writer(root);
2951 if (batch == atomic_read(&root->log_batch))
2955 /* bail out if we need to do a full commit */
2956 if (btrfs_need_log_full_commit(trans)) {
2957 ret = BTRFS_LOG_FORCE_COMMIT;
2958 mutex_unlock(&root->log_mutex);
2962 if (log_transid % 2 == 0)
2963 mark = EXTENT_DIRTY;
2967 /* we start IO on all the marked extents here, but we don't actually
2968 * wait for them until later.
2970 blk_start_plug(&plug);
2971 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2973 * -EAGAIN happens when someone, e.g., a concurrent transaction
2974 * commit, writes a dirty extent in this tree-log commit. This
2975 * concurrent write will create a hole writing out the extents,
2976 * and we cannot proceed on a zoned filesystem, requiring
2977 * sequential writing. While we can bail out to a full commit
2978 * here, but we can continue hoping the concurrent writing fills
2981 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
2984 blk_finish_plug(&plug);
2985 btrfs_set_log_full_commit(trans);
2986 mutex_unlock(&root->log_mutex);
2991 * We _must_ update under the root->log_mutex in order to make sure we
2992 * have a consistent view of the log root we are trying to commit at
2995 * We _must_ copy this into a local copy, because we are not holding the
2996 * log_root_tree->log_mutex yet. This is important because when we
2997 * commit the log_root_tree we must have a consistent view of the
2998 * log_root_tree when we update the super block to point at the
2999 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3000 * with the commit and possibly point at the new block which we may not
3003 btrfs_set_root_node(&log->root_item, log->node);
3004 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3006 btrfs_set_root_log_transid(root, root->log_transid + 1);
3007 log->log_transid = root->log_transid;
3008 root->log_start_pid = 0;
3010 * IO has been started, blocks of the log tree have WRITTEN flag set
3011 * in their headers. new modifications of the log will be written to
3012 * new positions. so it's safe to allow log writers to go in.
3014 mutex_unlock(&root->log_mutex);
3016 if (btrfs_is_zoned(fs_info)) {
3017 mutex_lock(&fs_info->tree_root->log_mutex);
3018 if (!log_root_tree->node) {
3019 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3021 mutex_unlock(&fs_info->tree_root->log_mutex);
3022 blk_finish_plug(&plug);
3026 mutex_unlock(&fs_info->tree_root->log_mutex);
3029 btrfs_init_log_ctx(&root_log_ctx, NULL);
3031 mutex_lock(&log_root_tree->log_mutex);
3033 index2 = log_root_tree->log_transid % 2;
3034 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3035 root_log_ctx.log_transid = log_root_tree->log_transid;
3038 * Now we are safe to update the log_root_tree because we're under the
3039 * log_mutex, and we're a current writer so we're holding the commit
3040 * open until we drop the log_mutex.
3042 ret = update_log_root(trans, log, &new_root_item);
3044 list_del_init(&root_log_ctx.list);
3045 blk_finish_plug(&plug);
3046 btrfs_set_log_full_commit(trans);
3049 "failed to update log for root %llu ret %d",
3050 btrfs_root_id(root), ret);
3051 btrfs_wait_tree_log_extents(log, mark);
3052 mutex_unlock(&log_root_tree->log_mutex);
3056 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3057 blk_finish_plug(&plug);
3058 list_del_init(&root_log_ctx.list);
3059 mutex_unlock(&log_root_tree->log_mutex);
3060 ret = root_log_ctx.log_ret;
3064 if (atomic_read(&log_root_tree->log_commit[index2])) {
3065 blk_finish_plug(&plug);
3066 ret = btrfs_wait_tree_log_extents(log, mark);
3067 wait_log_commit(log_root_tree,
3068 root_log_ctx.log_transid);
3069 mutex_unlock(&log_root_tree->log_mutex);
3071 ret = root_log_ctx.log_ret;
3074 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3075 atomic_set(&log_root_tree->log_commit[index2], 1);
3077 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3078 wait_log_commit(log_root_tree,
3079 root_log_ctx.log_transid - 1);
3083 * now that we've moved on to the tree of log tree roots,
3084 * check the full commit flag again
3086 if (btrfs_need_log_full_commit(trans)) {
3087 blk_finish_plug(&plug);
3088 btrfs_wait_tree_log_extents(log, mark);
3089 mutex_unlock(&log_root_tree->log_mutex);
3090 ret = BTRFS_LOG_FORCE_COMMIT;
3091 goto out_wake_log_root;
3094 ret = btrfs_write_marked_extents(fs_info,
3095 &log_root_tree->dirty_log_pages,
3096 EXTENT_DIRTY | EXTENT_NEW);
3097 blk_finish_plug(&plug);
3099 * As described above, -EAGAIN indicates a hole in the extents. We
3100 * cannot wait for these write outs since the waiting cause a
3101 * deadlock. Bail out to the full commit instead.
3103 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3104 btrfs_set_log_full_commit(trans);
3105 btrfs_wait_tree_log_extents(log, mark);
3106 mutex_unlock(&log_root_tree->log_mutex);
3107 goto out_wake_log_root;
3109 btrfs_set_log_full_commit(trans);
3110 mutex_unlock(&log_root_tree->log_mutex);
3111 goto out_wake_log_root;
3113 ret = btrfs_wait_tree_log_extents(log, mark);
3115 ret = btrfs_wait_tree_log_extents(log_root_tree,
3116 EXTENT_NEW | EXTENT_DIRTY);
3118 btrfs_set_log_full_commit(trans);
3119 mutex_unlock(&log_root_tree->log_mutex);
3120 goto out_wake_log_root;
3123 log_root_start = log_root_tree->node->start;
3124 log_root_level = btrfs_header_level(log_root_tree->node);
3125 log_root_tree->log_transid++;
3126 mutex_unlock(&log_root_tree->log_mutex);
3129 * Here we are guaranteed that nobody is going to write the superblock
3130 * for the current transaction before us and that neither we do write
3131 * our superblock before the previous transaction finishes its commit
3132 * and writes its superblock, because:
3134 * 1) We are holding a handle on the current transaction, so no body
3135 * can commit it until we release the handle;
3137 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3138 * if the previous transaction is still committing, and hasn't yet
3139 * written its superblock, we wait for it to do it, because a
3140 * transaction commit acquires the tree_log_mutex when the commit
3141 * begins and releases it only after writing its superblock.
3143 mutex_lock(&fs_info->tree_log_mutex);
3146 * The previous transaction writeout phase could have failed, and thus
3147 * marked the fs in an error state. We must not commit here, as we
3148 * could have updated our generation in the super_for_commit and
3149 * writing the super here would result in transid mismatches. If there
3150 * is an error here just bail.
3152 if (BTRFS_FS_ERROR(fs_info)) {
3154 btrfs_set_log_full_commit(trans);
3155 btrfs_abort_transaction(trans, ret);
3156 mutex_unlock(&fs_info->tree_log_mutex);
3157 goto out_wake_log_root;
3160 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3161 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3162 ret = write_all_supers(fs_info, 1);
3163 mutex_unlock(&fs_info->tree_log_mutex);
3165 btrfs_set_log_full_commit(trans);
3166 btrfs_abort_transaction(trans, ret);
3167 goto out_wake_log_root;
3171 * We know there can only be one task here, since we have not yet set
3172 * root->log_commit[index1] to 0 and any task attempting to sync the
3173 * log must wait for the previous log transaction to commit if it's
3174 * still in progress or wait for the current log transaction commit if
3175 * someone else already started it. We use <= and not < because the
3176 * first log transaction has an ID of 0.
3178 ASSERT(btrfs_get_root_last_log_commit(root) <= log_transid);
3179 btrfs_set_root_last_log_commit(root, log_transid);
3182 mutex_lock(&log_root_tree->log_mutex);
3183 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3185 log_root_tree->log_transid_committed++;
3186 atomic_set(&log_root_tree->log_commit[index2], 0);
3187 mutex_unlock(&log_root_tree->log_mutex);
3190 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3191 * all the updates above are seen by the woken threads. It might not be
3192 * necessary, but proving that seems to be hard.
3194 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3196 mutex_lock(&root->log_mutex);
3197 btrfs_remove_all_log_ctxs(root, index1, ret);
3198 root->log_transid_committed++;
3199 atomic_set(&root->log_commit[index1], 0);
3200 mutex_unlock(&root->log_mutex);
3203 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3204 * all the updates above are seen by the woken threads. It might not be
3205 * necessary, but proving that seems to be hard.
3207 cond_wake_up(&root->log_commit_wait[index1]);
3211 static void free_log_tree(struct btrfs_trans_handle *trans,
3212 struct btrfs_root *log)
3215 struct walk_control wc = {
3217 .process_func = process_one_buffer
3221 ret = walk_log_tree(trans, log, &wc);
3224 * We weren't able to traverse the entire log tree, the
3225 * typical scenario is getting an -EIO when reading an
3226 * extent buffer of the tree, due to a previous writeback
3229 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3230 &log->fs_info->fs_state);
3233 * Some extent buffers of the log tree may still be dirty
3234 * and not yet written back to storage, because we may
3235 * have updates to a log tree without syncing a log tree,
3236 * such as during rename and link operations. So flush
3237 * them out and wait for their writeback to complete, so
3238 * that we properly cleanup their state and pages.
3240 btrfs_write_marked_extents(log->fs_info,
3241 &log->dirty_log_pages,
3242 EXTENT_DIRTY | EXTENT_NEW);
3243 btrfs_wait_tree_log_extents(log,
3244 EXTENT_DIRTY | EXTENT_NEW);
3247 btrfs_abort_transaction(trans, ret);
3249 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3253 extent_io_tree_release(&log->dirty_log_pages);
3254 extent_io_tree_release(&log->log_csum_range);
3256 btrfs_put_root(log);
3260 * free all the extents used by the tree log. This should be called
3261 * at commit time of the full transaction
3263 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3265 if (root->log_root) {
3266 free_log_tree(trans, root->log_root);
3267 root->log_root = NULL;
3268 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3273 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3274 struct btrfs_fs_info *fs_info)
3276 if (fs_info->log_root_tree) {
3277 free_log_tree(trans, fs_info->log_root_tree);
3278 fs_info->log_root_tree = NULL;
3279 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3285 * Check if an inode was logged in the current transaction. This correctly deals
3286 * with the case where the inode was logged but has a logged_trans of 0, which
3287 * happens if the inode is evicted and loaded again, as logged_trans is an in
3288 * memory only field (not persisted).
3290 * Returns 1 if the inode was logged before in the transaction, 0 if it was not,
3293 static int inode_logged(const struct btrfs_trans_handle *trans,
3294 struct btrfs_inode *inode,
3295 struct btrfs_path *path_in)
3297 struct btrfs_path *path = path_in;
3298 struct btrfs_key key;
3301 if (inode->logged_trans == trans->transid)
3305 * If logged_trans is not 0, then we know the inode logged was not logged
3306 * in this transaction, so we can return false right away.
3308 if (inode->logged_trans > 0)
3312 * If no log tree was created for this root in this transaction, then
3313 * the inode can not have been logged in this transaction. In that case
3314 * set logged_trans to anything greater than 0 and less than the current
3315 * transaction's ID, to avoid the search below in a future call in case
3316 * a log tree gets created after this.
3318 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) {
3319 inode->logged_trans = trans->transid - 1;
3324 * We have a log tree and the inode's logged_trans is 0. We can't tell
3325 * for sure if the inode was logged before in this transaction by looking
3326 * only at logged_trans. We could be pessimistic and assume it was, but
3327 * that can lead to unnecessarily logging an inode during rename and link
3328 * operations, and then further updating the log in followup rename and
3329 * link operations, specially if it's a directory, which adds latency
3330 * visible to applications doing a series of rename or link operations.
3332 * A logged_trans of 0 here can mean several things:
3334 * 1) The inode was never logged since the filesystem was mounted, and may
3335 * or may have not been evicted and loaded again;
3337 * 2) The inode was logged in a previous transaction, then evicted and
3338 * then loaded again;
3340 * 3) The inode was logged in the current transaction, then evicted and
3341 * then loaded again.
3343 * For cases 1) and 2) we don't want to return true, but we need to detect
3344 * case 3) and return true. So we do a search in the log root for the inode
3347 key.objectid = btrfs_ino(inode);
3348 key.type = BTRFS_INODE_ITEM_KEY;
3352 path = btrfs_alloc_path();
3357 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
3360 btrfs_release_path(path);
3362 btrfs_free_path(path);
3365 * Logging an inode always results in logging its inode item. So if we
3366 * did not find the item we know the inode was not logged for sure.
3370 } else if (ret > 0) {
3372 * Set logged_trans to a value greater than 0 and less then the
3373 * current transaction to avoid doing the search in future calls.
3375 inode->logged_trans = trans->transid - 1;
3380 * The inode was previously logged and then evicted, set logged_trans to
3381 * the current transacion's ID, to avoid future tree searches as long as
3382 * the inode is not evicted again.
3384 inode->logged_trans = trans->transid;
3387 * If it's a directory, then we must set last_dir_index_offset to the
3388 * maximum possible value, so that the next attempt to log the inode does
3389 * not skip checking if dir index keys found in modified subvolume tree
3390 * leaves have been logged before, otherwise it would result in attempts
3391 * to insert duplicate dir index keys in the log tree. This must be done
3392 * because last_dir_index_offset is an in-memory only field, not persisted
3393 * in the inode item or any other on-disk structure, so its value is lost
3394 * once the inode is evicted.
3396 if (S_ISDIR(inode->vfs_inode.i_mode))
3397 inode->last_dir_index_offset = (u64)-1;
3403 * Delete a directory entry from the log if it exists.
3405 * Returns < 0 on error
3406 * 1 if the entry does not exists
3407 * 0 if the entry existed and was successfully deleted
3409 static int del_logged_dentry(struct btrfs_trans_handle *trans,
3410 struct btrfs_root *log,
3411 struct btrfs_path *path,
3413 const struct fscrypt_str *name,
3416 struct btrfs_dir_item *di;
3419 * We only log dir index items of a directory, so we don't need to look
3420 * for dir item keys.
3422 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3430 * We do not need to update the size field of the directory's
3431 * inode item because on log replay we update the field to reflect
3432 * all existing entries in the directory (see overwrite_item()).
3434 return btrfs_delete_one_dir_name(trans, log, path, di);
3438 * If both a file and directory are logged, and unlinks or renames are
3439 * mixed in, we have a few interesting corners:
3441 * create file X in dir Y
3442 * link file X to X.link in dir Y
3444 * unlink file X but leave X.link
3447 * After a crash we would expect only X.link to exist. But file X
3448 * didn't get fsync'd again so the log has back refs for X and X.link.
3450 * We solve this by removing directory entries and inode backrefs from the
3451 * log when a file that was logged in the current transaction is
3452 * unlinked. Any later fsync will include the updated log entries, and
3453 * we'll be able to reconstruct the proper directory items from backrefs.
3455 * This optimizations allows us to avoid relogging the entire inode
3456 * or the entire directory.
3458 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3459 struct btrfs_root *root,
3460 const struct fscrypt_str *name,
3461 struct btrfs_inode *dir, u64 index)
3463 struct btrfs_path *path;
3466 ret = inode_logged(trans, dir, NULL);
3470 btrfs_set_log_full_commit(trans);
3474 ret = join_running_log_trans(root);
3478 mutex_lock(&dir->log_mutex);
3480 path = btrfs_alloc_path();
3486 ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir),
3488 btrfs_free_path(path);
3490 mutex_unlock(&dir->log_mutex);
3492 btrfs_set_log_full_commit(trans);
3493 btrfs_end_log_trans(root);
3496 /* see comments for btrfs_del_dir_entries_in_log */
3497 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3498 struct btrfs_root *root,
3499 const struct fscrypt_str *name,
3500 struct btrfs_inode *inode, u64 dirid)
3502 struct btrfs_root *log;
3506 ret = inode_logged(trans, inode, NULL);
3510 btrfs_set_log_full_commit(trans);
3514 ret = join_running_log_trans(root);
3517 log = root->log_root;
3518 mutex_lock(&inode->log_mutex);
3520 ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode),
3522 mutex_unlock(&inode->log_mutex);
3523 if (ret < 0 && ret != -ENOENT)
3524 btrfs_set_log_full_commit(trans);
3525 btrfs_end_log_trans(root);
3529 * creates a range item in the log for 'dirid'. first_offset and
3530 * last_offset tell us which parts of the key space the log should
3531 * be considered authoritative for.
3533 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3534 struct btrfs_root *log,
3535 struct btrfs_path *path,
3537 u64 first_offset, u64 last_offset)
3540 struct btrfs_key key;
3541 struct btrfs_dir_log_item *item;
3543 key.objectid = dirid;
3544 key.offset = first_offset;
3545 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3546 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3548 * -EEXIST is fine and can happen sporadically when we are logging a
3549 * directory and have concurrent insertions in the subvolume's tree for
3550 * items from other inodes and that result in pushing off some dir items
3551 * from one leaf to another in order to accommodate for the new items.
3552 * This results in logging the same dir index range key.
3554 if (ret && ret != -EEXIST)
3557 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3558 struct btrfs_dir_log_item);
3559 if (ret == -EEXIST) {
3560 const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item);
3563 * btrfs_del_dir_entries_in_log() might have been called during
3564 * an unlink between the initial insertion of this key and the
3565 * current update, or we might be logging a single entry deletion
3566 * during a rename, so set the new last_offset to the max value.
3568 last_offset = max(last_offset, curr_end);
3570 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3571 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
3572 btrfs_release_path(path);
3576 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3577 struct btrfs_inode *inode,
3578 struct extent_buffer *src,
3579 struct btrfs_path *dst_path,
3583 struct btrfs_root *log = inode->root->log_root;
3584 char *ins_data = NULL;
3585 struct btrfs_item_batch batch;
3586 struct extent_buffer *dst;
3587 unsigned long src_offset;
3588 unsigned long dst_offset;
3590 struct btrfs_key key;
3599 btrfs_item_key_to_cpu(src, &key, start_slot);
3600 item_size = btrfs_item_size(src, start_slot);
3602 batch.data_sizes = &item_size;
3603 batch.total_data_size = item_size;
3605 struct btrfs_key *ins_keys;
3608 ins_data = kmalloc(count * sizeof(u32) +
3609 count * sizeof(struct btrfs_key), GFP_NOFS);
3613 ins_sizes = (u32 *)ins_data;
3614 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3615 batch.keys = ins_keys;
3616 batch.data_sizes = ins_sizes;
3617 batch.total_data_size = 0;
3619 for (i = 0; i < count; i++) {
3620 const int slot = start_slot + i;
3622 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3623 ins_sizes[i] = btrfs_item_size(src, slot);
3624 batch.total_data_size += ins_sizes[i];
3628 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3632 dst = dst_path->nodes[0];
3634 * Copy all the items in bulk, in a single copy operation. Item data is
3635 * organized such that it's placed at the end of a leaf and from right
3636 * to left. For example, the data for the second item ends at an offset
3637 * that matches the offset where the data for the first item starts, the
3638 * data for the third item ends at an offset that matches the offset
3639 * where the data of the second items starts, and so on.
3640 * Therefore our source and destination start offsets for copy match the
3641 * offsets of the last items (highest slots).
3643 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3644 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3645 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3646 btrfs_release_path(dst_path);
3648 last_index = batch.keys[count - 1].offset;
3649 ASSERT(last_index > inode->last_dir_index_offset);
3652 * If for some unexpected reason the last item's index is not greater
3653 * than the last index we logged, warn and force a transaction commit.
3655 if (WARN_ON(last_index <= inode->last_dir_index_offset))
3656 ret = BTRFS_LOG_FORCE_COMMIT;
3658 inode->last_dir_index_offset = last_index;
3660 if (btrfs_get_first_dir_index_to_log(inode) == 0)
3661 btrfs_set_first_dir_index_to_log(inode, batch.keys[0].offset);
3668 static int clone_leaf(struct btrfs_path *path, struct btrfs_log_ctx *ctx)
3670 const int slot = path->slots[0];
3672 if (ctx->scratch_eb) {
3673 copy_extent_buffer_full(ctx->scratch_eb, path->nodes[0]);
3675 ctx->scratch_eb = btrfs_clone_extent_buffer(path->nodes[0]);
3676 if (!ctx->scratch_eb)
3680 btrfs_release_path(path);
3681 path->nodes[0] = ctx->scratch_eb;
3682 path->slots[0] = slot;
3684 * Add extra ref to scratch eb so that it is not freed when callers
3685 * release the path, so we can reuse it later if needed.
3687 atomic_inc(&ctx->scratch_eb->refs);
3692 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3693 struct btrfs_inode *inode,
3694 struct btrfs_path *path,
3695 struct btrfs_path *dst_path,
3696 struct btrfs_log_ctx *ctx,
3697 u64 *last_old_dentry_offset)
3699 struct btrfs_root *log = inode->root->log_root;
3700 struct extent_buffer *src;
3701 const int nritems = btrfs_header_nritems(path->nodes[0]);
3702 const u64 ino = btrfs_ino(inode);
3703 bool last_found = false;
3704 int batch_start = 0;
3709 * We need to clone the leaf, release the read lock on it, and use the
3710 * clone before modifying the log tree. See the comment at copy_items()
3711 * about why we need to do this.
3713 ret = clone_leaf(path, ctx);
3717 src = path->nodes[0];
3719 for (int i = path->slots[0]; i < nritems; i++) {
3720 struct btrfs_dir_item *di;
3721 struct btrfs_key key;
3724 btrfs_item_key_to_cpu(src, &key, i);
3726 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3731 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3734 * Skip ranges of items that consist only of dir item keys created
3735 * in past transactions. However if we find a gap, we must log a
3736 * dir index range item for that gap, so that index keys in that
3737 * gap are deleted during log replay.
3739 if (btrfs_dir_transid(src, di) < trans->transid) {
3740 if (key.offset > *last_old_dentry_offset + 1) {
3741 ret = insert_dir_log_key(trans, log, dst_path,
3742 ino, *last_old_dentry_offset + 1,
3748 *last_old_dentry_offset = key.offset;
3752 /* If we logged this dir index item before, we can skip it. */
3753 if (key.offset <= inode->last_dir_index_offset)
3757 * We must make sure that when we log a directory entry, the
3758 * corresponding inode, after log replay, has a matching link
3759 * count. For example:
3765 * xfs_io -c "fsync" mydir
3767 * <mount fs and log replay>
3769 * Would result in a fsync log that when replayed, our file inode
3770 * would have a link count of 1, but we get two directory entries
3771 * pointing to the same inode. After removing one of the names,
3772 * it would not be possible to remove the other name, which
3773 * resulted always in stale file handle errors, and would not be
3774 * possible to rmdir the parent directory, since its i_size could
3775 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3776 * resulting in -ENOTEMPTY errors.
3778 if (!ctx->log_new_dentries) {
3779 struct btrfs_key di_key;
3781 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3782 if (di_key.type != BTRFS_ROOT_ITEM_KEY)
3783 ctx->log_new_dentries = true;
3786 if (batch_size == 0)
3791 if (batch_size > 0) {
3794 ret = flush_dir_items_batch(trans, inode, src, dst_path,
3795 batch_start, batch_size);
3800 return last_found ? 1 : 0;
3804 * log all the items included in the current transaction for a given
3805 * directory. This also creates the range items in the log tree required
3806 * to replay anything deleted before the fsync
3808 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3809 struct btrfs_inode *inode,
3810 struct btrfs_path *path,
3811 struct btrfs_path *dst_path,
3812 struct btrfs_log_ctx *ctx,
3813 u64 min_offset, u64 *last_offset_ret)
3815 struct btrfs_key min_key;
3816 struct btrfs_root *root = inode->root;
3817 struct btrfs_root *log = root->log_root;
3819 u64 last_old_dentry_offset = min_offset - 1;
3820 u64 last_offset = (u64)-1;
3821 u64 ino = btrfs_ino(inode);
3823 min_key.objectid = ino;
3824 min_key.type = BTRFS_DIR_INDEX_KEY;
3825 min_key.offset = min_offset;
3827 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3830 * we didn't find anything from this transaction, see if there
3831 * is anything at all
3833 if (ret != 0 || min_key.objectid != ino ||
3834 min_key.type != BTRFS_DIR_INDEX_KEY) {
3835 min_key.objectid = ino;
3836 min_key.type = BTRFS_DIR_INDEX_KEY;
3837 min_key.offset = (u64)-1;
3838 btrfs_release_path(path);
3839 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3841 btrfs_release_path(path);
3844 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3846 /* if ret == 0 there are items for this type,
3847 * create a range to tell us the last key of this type.
3848 * otherwise, there are no items in this directory after
3849 * *min_offset, and we create a range to indicate that.
3852 struct btrfs_key tmp;
3854 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3856 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3857 last_old_dentry_offset = tmp.offset;
3858 } else if (ret > 0) {
3865 /* go backward to find any previous key */
3866 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3868 struct btrfs_key tmp;
3870 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3872 * The dir index key before the first one we found that needs to
3873 * be logged might be in a previous leaf, and there might be a
3874 * gap between these keys, meaning that we had deletions that
3875 * happened. So the key range item we log (key type
3876 * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the
3877 * previous key's offset plus 1, so that those deletes are replayed.
3879 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3880 last_old_dentry_offset = tmp.offset;
3881 } else if (ret < 0) {
3885 btrfs_release_path(path);
3888 * Find the first key from this transaction again or the one we were at
3889 * in the loop below in case we had to reschedule. We may be logging the
3890 * directory without holding its VFS lock, which happen when logging new
3891 * dentries (through log_new_dir_dentries()) or in some cases when we
3892 * need to log the parent directory of an inode. This means a dir index
3893 * key might be deleted from the inode's root, and therefore we may not
3894 * find it anymore. If we can't find it, just move to the next key. We
3895 * can not bail out and ignore, because if we do that we will simply
3896 * not log dir index keys that come after the one that was just deleted
3897 * and we can end up logging a dir index range that ends at (u64)-1
3898 * (@last_offset is initialized to that), resulting in removing dir
3899 * entries we should not remove at log replay time.
3902 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3904 ret = btrfs_next_item(root, path);
3906 /* There are no more keys in the inode's root. */
3915 * we have a block from this transaction, log every item in it
3916 * from our directory
3919 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx,
3920 &last_old_dentry_offset);
3926 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3929 * look ahead to the next item and see if it is also
3930 * from this directory and from this transaction
3932 ret = btrfs_next_leaf(root, path);
3935 last_offset = (u64)-1;
3940 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3941 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3942 last_offset = (u64)-1;
3945 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3947 * The next leaf was not changed in the current transaction
3948 * and has at least one dir index key.
3949 * We check for the next key because there might have been
3950 * one or more deletions between the last key we logged and
3951 * that next key. So the key range item we log (key type
3952 * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's
3953 * offset minus 1, so that those deletes are replayed.
3955 last_offset = min_key.offset - 1;
3958 if (need_resched()) {
3959 btrfs_release_path(path);
3965 btrfs_release_path(path);
3966 btrfs_release_path(dst_path);
3969 *last_offset_ret = last_offset;
3971 * In case the leaf was changed in the current transaction but
3972 * all its dir items are from a past transaction, the last item
3973 * in the leaf is a dir item and there's no gap between that last
3974 * dir item and the first one on the next leaf (which did not
3975 * change in the current transaction), then we don't need to log
3976 * a range, last_old_dentry_offset is == to last_offset.
3978 ASSERT(last_old_dentry_offset <= last_offset);
3979 if (last_old_dentry_offset < last_offset)
3980 ret = insert_dir_log_key(trans, log, path, ino,
3981 last_old_dentry_offset + 1,
3989 * If the inode was logged before and it was evicted, then its
3990 * last_dir_index_offset is (u64)-1, so we don't the value of the last index
3991 * key offset. If that's the case, search for it and update the inode. This
3992 * is to avoid lookups in the log tree every time we try to insert a dir index
3993 * key from a leaf changed in the current transaction, and to allow us to always
3994 * do batch insertions of dir index keys.
3996 static int update_last_dir_index_offset(struct btrfs_inode *inode,
3997 struct btrfs_path *path,
3998 const struct btrfs_log_ctx *ctx)
4000 const u64 ino = btrfs_ino(inode);
4001 struct btrfs_key key;
4004 lockdep_assert_held(&inode->log_mutex);
4006 if (inode->last_dir_index_offset != (u64)-1)
4009 if (!ctx->logged_before) {
4010 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
4015 key.type = BTRFS_DIR_INDEX_KEY;
4016 key.offset = (u64)-1;
4018 ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0);
4020 * An error happened or we actually have an index key with an offset
4021 * value of (u64)-1. Bail out, we're done.
4027 inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1;
4030 * No dir index items, bail out and leave last_dir_index_offset with
4031 * the value right before the first valid index value.
4033 if (path->slots[0] == 0)
4037 * btrfs_search_slot() left us at one slot beyond the slot with the last
4038 * index key, or beyond the last key of the directory that is not an
4039 * index key. If we have an index key before, set last_dir_index_offset
4040 * to its offset value, otherwise leave it with a value right before the
4041 * first valid index value, as it means we have an empty directory.
4043 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4044 if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY)
4045 inode->last_dir_index_offset = key.offset;
4048 btrfs_release_path(path);
4054 * logging directories is very similar to logging inodes, We find all the items
4055 * from the current transaction and write them to the log.
4057 * The recovery code scans the directory in the subvolume, and if it finds a
4058 * key in the range logged that is not present in the log tree, then it means
4059 * that dir entry was unlinked during the transaction.
4061 * In order for that scan to work, we must include one key smaller than
4062 * the smallest logged by this transaction and one key larger than the largest
4063 * key logged by this transaction.
4065 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4066 struct btrfs_inode *inode,
4067 struct btrfs_path *path,
4068 struct btrfs_path *dst_path,
4069 struct btrfs_log_ctx *ctx)
4075 ret = update_last_dir_index_offset(inode, path, ctx);
4079 min_key = BTRFS_DIR_START_INDEX;
4083 ret = log_dir_items(trans, inode, path, dst_path,
4084 ctx, min_key, &max_key);
4087 if (max_key == (u64)-1)
4089 min_key = max_key + 1;
4096 * a helper function to drop items from the log before we relog an
4097 * inode. max_key_type indicates the highest item type to remove.
4098 * This cannot be run for file data extents because it does not
4099 * free the extents they point to.
4101 static int drop_inode_items(struct btrfs_trans_handle *trans,
4102 struct btrfs_root *log,
4103 struct btrfs_path *path,
4104 struct btrfs_inode *inode,
4108 struct btrfs_key key;
4109 struct btrfs_key found_key;
4112 key.objectid = btrfs_ino(inode);
4113 key.type = max_key_type;
4114 key.offset = (u64)-1;
4117 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4120 } else if (ret > 0) {
4121 if (path->slots[0] == 0)
4126 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4129 if (found_key.objectid != key.objectid)
4132 found_key.offset = 0;
4134 ret = btrfs_bin_search(path->nodes[0], 0, &found_key, &start_slot);
4138 ret = btrfs_del_items(trans, log, path, start_slot,
4139 path->slots[0] - start_slot + 1);
4141 * If start slot isn't 0 then we don't need to re-search, we've
4142 * found the last guy with the objectid in this tree.
4144 if (ret || start_slot != 0)
4146 btrfs_release_path(path);
4148 btrfs_release_path(path);
4154 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4155 struct btrfs_root *log_root,
4156 struct btrfs_inode *inode,
4157 u64 new_size, u32 min_type)
4159 struct btrfs_truncate_control control = {
4160 .new_size = new_size,
4161 .ino = btrfs_ino(inode),
4162 .min_type = min_type,
4163 .skip_ref_updates = true,
4166 return btrfs_truncate_inode_items(trans, log_root, &control);
4169 static void fill_inode_item(struct btrfs_trans_handle *trans,
4170 struct extent_buffer *leaf,
4171 struct btrfs_inode_item *item,
4172 struct inode *inode, int log_inode_only,
4175 struct btrfs_map_token token;
4178 btrfs_init_map_token(&token, leaf);
4180 if (log_inode_only) {
4181 /* set the generation to zero so the recover code
4182 * can tell the difference between an logging
4183 * just to say 'this inode exists' and a logging
4184 * to say 'update this inode with these values'
4186 btrfs_set_token_inode_generation(&token, item, 0);
4187 btrfs_set_token_inode_size(&token, item, logged_isize);
4189 btrfs_set_token_inode_generation(&token, item,
4190 BTRFS_I(inode)->generation);
4191 btrfs_set_token_inode_size(&token, item, inode->i_size);
4194 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4195 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4196 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4197 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4199 btrfs_set_token_timespec_sec(&token, &item->atime,
4200 inode_get_atime_sec(inode));
4201 btrfs_set_token_timespec_nsec(&token, &item->atime,
4202 inode_get_atime_nsec(inode));
4204 btrfs_set_token_timespec_sec(&token, &item->mtime,
4205 inode_get_mtime_sec(inode));
4206 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4207 inode_get_mtime_nsec(inode));
4209 btrfs_set_token_timespec_sec(&token, &item->ctime,
4210 inode_get_ctime_sec(inode));
4211 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4212 inode_get_ctime_nsec(inode));
4215 * We do not need to set the nbytes field, in fact during a fast fsync
4216 * its value may not even be correct, since a fast fsync does not wait
4217 * for ordered extent completion, which is where we update nbytes, it
4218 * only waits for writeback to complete. During log replay as we find
4219 * file extent items and replay them, we adjust the nbytes field of the
4220 * inode item in subvolume tree as needed (see overwrite_item()).
4223 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4224 btrfs_set_token_inode_transid(&token, item, trans->transid);
4225 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4226 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4227 BTRFS_I(inode)->ro_flags);
4228 btrfs_set_token_inode_flags(&token, item, flags);
4229 btrfs_set_token_inode_block_group(&token, item, 0);
4232 static int log_inode_item(struct btrfs_trans_handle *trans,
4233 struct btrfs_root *log, struct btrfs_path *path,
4234 struct btrfs_inode *inode, bool inode_item_dropped)
4236 struct btrfs_inode_item *inode_item;
4240 * If we are doing a fast fsync and the inode was logged before in the
4241 * current transaction, then we know the inode was previously logged and
4242 * it exists in the log tree. For performance reasons, in this case use
4243 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4244 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4245 * contention in case there are concurrent fsyncs for other inodes of the
4246 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4247 * already exists can also result in unnecessarily splitting a leaf.
4249 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4250 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4256 * This means it is the first fsync in the current transaction,
4257 * so the inode item is not in the log and we need to insert it.
4258 * We can never get -EEXIST because we are only called for a fast
4259 * fsync and in case an inode eviction happens after the inode was
4260 * logged before in the current transaction, when we load again
4261 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4262 * flags and set ->logged_trans to 0.
4264 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4265 sizeof(*inode_item));
4266 ASSERT(ret != -EEXIST);
4270 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4271 struct btrfs_inode_item);
4272 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4274 btrfs_release_path(path);
4278 static int log_csums(struct btrfs_trans_handle *trans,
4279 struct btrfs_inode *inode,
4280 struct btrfs_root *log_root,
4281 struct btrfs_ordered_sum *sums)
4283 const u64 lock_end = sums->logical + sums->len - 1;
4284 struct extent_state *cached_state = NULL;
4288 * If this inode was not used for reflink operations in the current
4289 * transaction with new extents, then do the fast path, no need to
4290 * worry about logging checksum items with overlapping ranges.
4292 if (inode->last_reflink_trans < trans->transid)
4293 return btrfs_csum_file_blocks(trans, log_root, sums);
4296 * Serialize logging for checksums. This is to avoid racing with the
4297 * same checksum being logged by another task that is logging another
4298 * file which happens to refer to the same extent as well. Such races
4299 * can leave checksum items in the log with overlapping ranges.
4301 ret = lock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4306 * Due to extent cloning, we might have logged a csum item that covers a
4307 * subrange of a cloned extent, and later we can end up logging a csum
4308 * item for a larger subrange of the same extent or the entire range.
4309 * This would leave csum items in the log tree that cover the same range
4310 * and break the searches for checksums in the log tree, resulting in
4311 * some checksums missing in the fs/subvolume tree. So just delete (or
4312 * trim and adjust) any existing csum items in the log for this range.
4314 ret = btrfs_del_csums(trans, log_root, sums->logical, sums->len);
4316 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4318 unlock_extent(&log_root->log_csum_range, sums->logical, lock_end,
4324 static noinline int copy_items(struct btrfs_trans_handle *trans,
4325 struct btrfs_inode *inode,
4326 struct btrfs_path *dst_path,
4327 struct btrfs_path *src_path,
4328 int start_slot, int nr, int inode_only,
4329 u64 logged_isize, struct btrfs_log_ctx *ctx)
4331 struct btrfs_root *log = inode->root->log_root;
4332 struct btrfs_file_extent_item *extent;
4333 struct extent_buffer *src;
4335 struct btrfs_key *ins_keys;
4337 struct btrfs_item_batch batch;
4340 const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM);
4341 const u64 i_size = i_size_read(&inode->vfs_inode);
4344 * To keep lockdep happy and avoid deadlocks, clone the source leaf and
4345 * use the clone. This is because otherwise we would be changing the log
4346 * tree, to insert items from the subvolume tree or insert csum items,
4347 * while holding a read lock on a leaf from the subvolume tree, which
4348 * creates a nasty lock dependency when COWing log tree nodes/leaves:
4350 * 1) Modifying the log tree triggers an extent buffer allocation while
4351 * holding a write lock on a parent extent buffer from the log tree.
4352 * Allocating the pages for an extent buffer, or the extent buffer
4353 * struct, can trigger inode eviction and finally the inode eviction
4354 * will trigger a release/remove of a delayed node, which requires
4355 * taking the delayed node's mutex;
4357 * 2) Allocating a metadata extent for a log tree can trigger the async
4358 * reclaim thread and make us wait for it to release enough space and
4359 * unblock our reservation ticket. The reclaim thread can start
4360 * flushing delayed items, and that in turn results in the need to
4361 * lock delayed node mutexes and in the need to write lock extent
4362 * buffers of a subvolume tree - all this while holding a write lock
4363 * on the parent extent buffer in the log tree.
4365 * So one task in scenario 1) running in parallel with another task in
4366 * scenario 2) could lead to a deadlock, one wanting to lock a delayed
4367 * node mutex while having a read lock on a leaf from the subvolume,
4368 * while the other is holding the delayed node's mutex and wants to
4369 * write lock the same subvolume leaf for flushing delayed items.
4371 ret = clone_leaf(src_path, ctx);
4375 src = src_path->nodes[0];
4377 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4378 nr * sizeof(u32), GFP_NOFS);
4382 ins_sizes = (u32 *)ins_data;
4383 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4384 batch.keys = ins_keys;
4385 batch.data_sizes = ins_sizes;
4386 batch.total_data_size = 0;
4390 for (int i = 0; i < nr; i++) {
4391 const int src_slot = start_slot + i;
4392 struct btrfs_root *csum_root;
4393 struct btrfs_ordered_sum *sums;
4394 struct btrfs_ordered_sum *sums_next;
4395 LIST_HEAD(ordered_sums);
4399 u64 extent_num_bytes;
4402 btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot);
4404 if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY)
4407 extent = btrfs_item_ptr(src, src_slot,
4408 struct btrfs_file_extent_item);
4410 is_old_extent = (btrfs_file_extent_generation(src, extent) <
4414 * Don't copy extents from past generations. That would make us
4415 * log a lot more metadata for common cases like doing only a
4416 * few random writes into a file and then fsync it for the first
4417 * time or after the full sync flag is set on the inode. We can
4418 * get leaves full of extent items, most of which are from past
4419 * generations, so we can skip them - as long as the inode has
4420 * not been the target of a reflink operation in this transaction,
4421 * as in that case it might have had file extent items with old
4422 * generations copied into it. We also must always log prealloc
4423 * extents that start at or beyond eof, otherwise we would lose
4424 * them on log replay.
4426 if (is_old_extent &&
4427 ins_keys[dst_index].offset < i_size &&
4428 inode->last_reflink_trans < trans->transid)
4434 /* Only regular extents have checksums. */
4435 if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG)
4439 * If it's an extent created in a past transaction, then its
4440 * checksums are already accessible from the committed csum tree,
4441 * no need to log them.
4446 disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent);
4447 /* If it's an explicit hole, there are no checksums. */
4448 if (disk_bytenr == 0)
4451 disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent);
4453 if (btrfs_file_extent_compression(src, extent)) {
4455 extent_num_bytes = disk_num_bytes;
4457 extent_offset = btrfs_file_extent_offset(src, extent);
4458 extent_num_bytes = btrfs_file_extent_num_bytes(src, extent);
4461 csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr);
4462 disk_bytenr += extent_offset;
4463 ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4464 disk_bytenr + extent_num_bytes - 1,
4465 &ordered_sums, false);
4470 list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) {
4472 ret = log_csums(trans, inode, log, sums);
4473 list_del(&sums->list);
4480 ins_sizes[dst_index] = btrfs_item_size(src, src_slot);
4481 batch.total_data_size += ins_sizes[dst_index];
4487 * We have a leaf full of old extent items that don't need to be logged,
4488 * so we don't need to do anything.
4493 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4498 for (int i = 0; i < nr; i++) {
4499 const int src_slot = start_slot + i;
4500 const int dst_slot = dst_path->slots[0] + dst_index;
4501 struct btrfs_key key;
4502 unsigned long src_offset;
4503 unsigned long dst_offset;
4506 * We're done, all the remaining items in the source leaf
4507 * correspond to old file extent items.
4509 if (dst_index >= batch.nr)
4512 btrfs_item_key_to_cpu(src, &key, src_slot);
4514 if (key.type != BTRFS_EXTENT_DATA_KEY)
4517 extent = btrfs_item_ptr(src, src_slot,
4518 struct btrfs_file_extent_item);
4520 /* See the comment in the previous loop, same logic. */
4521 if (btrfs_file_extent_generation(src, extent) < trans->transid &&
4522 key.offset < i_size &&
4523 inode->last_reflink_trans < trans->transid)
4527 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot);
4528 src_offset = btrfs_item_ptr_offset(src, src_slot);
4530 if (key.type == BTRFS_INODE_ITEM_KEY) {
4531 struct btrfs_inode_item *inode_item;
4533 inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot,
4534 struct btrfs_inode_item);
4535 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4537 inode_only == LOG_INODE_EXISTS,
4540 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4541 src_offset, ins_sizes[dst_index]);
4547 btrfs_mark_buffer_dirty(trans, dst_path->nodes[0]);
4548 btrfs_release_path(dst_path);
4555 static int extent_cmp(void *priv, const struct list_head *a,
4556 const struct list_head *b)
4558 const struct extent_map *em1, *em2;
4560 em1 = list_entry(a, struct extent_map, list);
4561 em2 = list_entry(b, struct extent_map, list);
4563 if (em1->start < em2->start)
4565 else if (em1->start > em2->start)
4570 static int log_extent_csums(struct btrfs_trans_handle *trans,
4571 struct btrfs_inode *inode,
4572 struct btrfs_root *log_root,
4573 const struct extent_map *em,
4574 struct btrfs_log_ctx *ctx)
4576 struct btrfs_ordered_extent *ordered;
4577 struct btrfs_root *csum_root;
4580 u64 mod_start = em->start;
4581 u64 mod_len = em->len;
4582 LIST_HEAD(ordered_sums);
4585 if (inode->flags & BTRFS_INODE_NODATASUM ||
4586 (em->flags & EXTENT_FLAG_PREALLOC) ||
4587 em->block_start == EXTENT_MAP_HOLE)
4590 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4591 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4592 const u64 mod_end = mod_start + mod_len;
4593 struct btrfs_ordered_sum *sums;
4598 if (ordered_end <= mod_start)
4600 if (mod_end <= ordered->file_offset)
4604 * We are going to copy all the csums on this ordered extent, so
4605 * go ahead and adjust mod_start and mod_len in case this ordered
4606 * extent has already been logged.
4608 if (ordered->file_offset > mod_start) {
4609 if (ordered_end >= mod_end)
4610 mod_len = ordered->file_offset - mod_start;
4612 * If we have this case
4614 * |--------- logged extent ---------|
4615 * |----- ordered extent ----|
4617 * Just don't mess with mod_start and mod_len, we'll
4618 * just end up logging more csums than we need and it
4622 if (ordered_end < mod_end) {
4623 mod_len = mod_end - ordered_end;
4624 mod_start = ordered_end;
4631 * To keep us from looping for the above case of an ordered
4632 * extent that falls inside of the logged extent.
4634 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4637 list_for_each_entry(sums, &ordered->list, list) {
4638 ret = log_csums(trans, inode, log_root, sums);
4644 /* We're done, found all csums in the ordered extents. */
4648 /* If we're compressed we have to save the entire range of csums. */
4649 if (extent_map_is_compressed(em)) {
4651 csum_len = max(em->block_len, em->orig_block_len);
4653 csum_offset = mod_start - em->start;
4657 /* block start is already adjusted for the file extent offset. */
4658 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4659 ret = btrfs_lookup_csums_list(csum_root, em->block_start + csum_offset,
4660 em->block_start + csum_offset +
4661 csum_len - 1, &ordered_sums, false);
4666 while (!list_empty(&ordered_sums)) {
4667 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4668 struct btrfs_ordered_sum,
4671 ret = log_csums(trans, inode, log_root, sums);
4672 list_del(&sums->list);
4679 static int log_one_extent(struct btrfs_trans_handle *trans,
4680 struct btrfs_inode *inode,
4681 const struct extent_map *em,
4682 struct btrfs_path *path,
4683 struct btrfs_log_ctx *ctx)
4685 struct btrfs_drop_extents_args drop_args = { 0 };
4686 struct btrfs_root *log = inode->root->log_root;
4687 struct btrfs_file_extent_item fi = { 0 };
4688 struct extent_buffer *leaf;
4689 struct btrfs_key key;
4690 enum btrfs_compression_type compress_type;
4691 u64 extent_offset = em->start - em->orig_start;
4695 btrfs_set_stack_file_extent_generation(&fi, trans->transid);
4696 if (em->flags & EXTENT_FLAG_PREALLOC)
4697 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC);
4699 btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG);
4701 block_len = max(em->block_len, em->orig_block_len);
4702 compress_type = extent_map_compression(em);
4703 if (compress_type != BTRFS_COMPRESS_NONE) {
4704 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start);
4705 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4706 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4707 btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start -
4709 btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len);
4712 btrfs_set_stack_file_extent_offset(&fi, extent_offset);
4713 btrfs_set_stack_file_extent_num_bytes(&fi, em->len);
4714 btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes);
4715 btrfs_set_stack_file_extent_compression(&fi, compress_type);
4717 ret = log_extent_csums(trans, inode, log, em, ctx);
4722 * If this is the first time we are logging the inode in the current
4723 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4724 * because it does a deletion search, which always acquires write locks
4725 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4726 * but also adds significant contention in a log tree, since log trees
4727 * are small, with a root at level 2 or 3 at most, due to their short
4730 if (ctx->logged_before) {
4731 drop_args.path = path;
4732 drop_args.start = em->start;
4733 drop_args.end = em->start + em->len;
4734 drop_args.replace_extent = true;
4735 drop_args.extent_item_size = sizeof(fi);
4736 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4741 if (!drop_args.extent_inserted) {
4742 key.objectid = btrfs_ino(inode);
4743 key.type = BTRFS_EXTENT_DATA_KEY;
4744 key.offset = em->start;
4746 ret = btrfs_insert_empty_item(trans, log, path, &key,
4751 leaf = path->nodes[0];
4752 write_extent_buffer(leaf, &fi,
4753 btrfs_item_ptr_offset(leaf, path->slots[0]),
4755 btrfs_mark_buffer_dirty(trans, leaf);
4757 btrfs_release_path(path);
4763 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4764 * lose them after doing a full/fast fsync and replaying the log. We scan the
4765 * subvolume's root instead of iterating the inode's extent map tree because
4766 * otherwise we can log incorrect extent items based on extent map conversion.
4767 * That can happen due to the fact that extent maps are merged when they
4768 * are not in the extent map tree's list of modified extents.
4770 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4771 struct btrfs_inode *inode,
4772 struct btrfs_path *path,
4773 struct btrfs_log_ctx *ctx)
4775 struct btrfs_root *root = inode->root;
4776 struct btrfs_key key;
4777 const u64 i_size = i_size_read(&inode->vfs_inode);
4778 const u64 ino = btrfs_ino(inode);
4779 struct btrfs_path *dst_path = NULL;
4780 bool dropped_extents = false;
4781 u64 truncate_offset = i_size;
4782 struct extent_buffer *leaf;
4788 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4792 key.type = BTRFS_EXTENT_DATA_KEY;
4793 key.offset = i_size;
4794 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4799 * We must check if there is a prealloc extent that starts before the
4800 * i_size and crosses the i_size boundary. This is to ensure later we
4801 * truncate down to the end of that extent and not to the i_size, as
4802 * otherwise we end up losing part of the prealloc extent after a log
4803 * replay and with an implicit hole if there is another prealloc extent
4804 * that starts at an offset beyond i_size.
4806 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4811 struct btrfs_file_extent_item *ei;
4813 leaf = path->nodes[0];
4814 slot = path->slots[0];
4815 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4817 if (btrfs_file_extent_type(leaf, ei) ==
4818 BTRFS_FILE_EXTENT_PREALLOC) {
4821 btrfs_item_key_to_cpu(leaf, &key, slot);
4822 extent_end = key.offset +
4823 btrfs_file_extent_num_bytes(leaf, ei);
4825 if (extent_end > i_size)
4826 truncate_offset = extent_end;
4833 leaf = path->nodes[0];
4834 slot = path->slots[0];
4836 if (slot >= btrfs_header_nritems(leaf)) {
4838 ret = copy_items(trans, inode, dst_path, path,
4839 start_slot, ins_nr, 1, 0, ctx);
4844 ret = btrfs_next_leaf(root, path);
4854 btrfs_item_key_to_cpu(leaf, &key, slot);
4855 if (key.objectid > ino)
4857 if (WARN_ON_ONCE(key.objectid < ino) ||
4858 key.type < BTRFS_EXTENT_DATA_KEY ||
4859 key.offset < i_size) {
4863 if (!dropped_extents) {
4865 * Avoid logging extent items logged in past fsync calls
4866 * and leading to duplicate keys in the log tree.
4868 ret = truncate_inode_items(trans, root->log_root, inode,
4870 BTRFS_EXTENT_DATA_KEY);
4873 dropped_extents = true;
4880 dst_path = btrfs_alloc_path();
4888 ret = copy_items(trans, inode, dst_path, path,
4889 start_slot, ins_nr, 1, 0, ctx);
4891 btrfs_release_path(path);
4892 btrfs_free_path(dst_path);
4896 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4897 struct btrfs_inode *inode,
4898 struct btrfs_path *path,
4899 struct btrfs_log_ctx *ctx)
4901 struct btrfs_ordered_extent *ordered;
4902 struct btrfs_ordered_extent *tmp;
4903 struct extent_map *em, *n;
4905 struct extent_map_tree *tree = &inode->extent_tree;
4909 write_lock(&tree->lock);
4911 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4912 list_del_init(&em->list);
4914 * Just an arbitrary number, this can be really CPU intensive
4915 * once we start getting a lot of extents, and really once we
4916 * have a bunch of extents we just want to commit since it will
4919 if (++num > 32768) {
4920 list_del_init(&tree->modified_extents);
4925 if (em->generation < trans->transid)
4928 /* We log prealloc extents beyond eof later. */
4929 if ((em->flags & EXTENT_FLAG_PREALLOC) &&
4930 em->start >= i_size_read(&inode->vfs_inode))
4933 /* Need a ref to keep it from getting evicted from cache */
4934 refcount_inc(&em->refs);
4935 em->flags |= EXTENT_FLAG_LOGGING;
4936 list_add_tail(&em->list, &extents);
4940 list_sort(NULL, &extents, extent_cmp);
4942 while (!list_empty(&extents)) {
4943 em = list_entry(extents.next, struct extent_map, list);
4945 list_del_init(&em->list);
4948 * If we had an error we just need to delete everybody from our
4952 clear_em_logging(inode, em);
4953 free_extent_map(em);
4957 write_unlock(&tree->lock);
4959 ret = log_one_extent(trans, inode, em, path, ctx);
4960 write_lock(&tree->lock);
4961 clear_em_logging(inode, em);
4962 free_extent_map(em);
4964 WARN_ON(!list_empty(&extents));
4965 write_unlock(&tree->lock);
4968 ret = btrfs_log_prealloc_extents(trans, inode, path, ctx);
4973 * We have logged all extents successfully, now make sure the commit of
4974 * the current transaction waits for the ordered extents to complete
4975 * before it commits and wipes out the log trees, otherwise we would
4976 * lose data if an ordered extents completes after the transaction
4977 * commits and a power failure happens after the transaction commit.
4979 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4980 list_del_init(&ordered->log_list);
4981 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4983 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4984 spin_lock_irq(&inode->ordered_tree_lock);
4985 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4986 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4987 atomic_inc(&trans->transaction->pending_ordered);
4989 spin_unlock_irq(&inode->ordered_tree_lock);
4991 btrfs_put_ordered_extent(ordered);
4997 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4998 struct btrfs_path *path, u64 *size_ret)
5000 struct btrfs_key key;
5003 key.objectid = btrfs_ino(inode);
5004 key.type = BTRFS_INODE_ITEM_KEY;
5007 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
5010 } else if (ret > 0) {
5013 struct btrfs_inode_item *item;
5015 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5016 struct btrfs_inode_item);
5017 *size_ret = btrfs_inode_size(path->nodes[0], item);
5019 * If the in-memory inode's i_size is smaller then the inode
5020 * size stored in the btree, return the inode's i_size, so
5021 * that we get a correct inode size after replaying the log
5022 * when before a power failure we had a shrinking truncate
5023 * followed by addition of a new name (rename / new hard link).
5024 * Otherwise return the inode size from the btree, to avoid
5025 * data loss when replaying a log due to previously doing a
5026 * write that expands the inode's size and logging a new name
5027 * immediately after.
5029 if (*size_ret > inode->vfs_inode.i_size)
5030 *size_ret = inode->vfs_inode.i_size;
5033 btrfs_release_path(path);
5038 * At the moment we always log all xattrs. This is to figure out at log replay
5039 * time which xattrs must have their deletion replayed. If a xattr is missing
5040 * in the log tree and exists in the fs/subvol tree, we delete it. This is
5041 * because if a xattr is deleted, the inode is fsynced and a power failure
5042 * happens, causing the log to be replayed the next time the fs is mounted,
5043 * we want the xattr to not exist anymore (same behaviour as other filesystems
5044 * with a journal, ext3/4, xfs, f2fs, etc).
5046 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
5047 struct btrfs_inode *inode,
5048 struct btrfs_path *path,
5049 struct btrfs_path *dst_path,
5050 struct btrfs_log_ctx *ctx)
5052 struct btrfs_root *root = inode->root;
5054 struct btrfs_key key;
5055 const u64 ino = btrfs_ino(inode);
5058 bool found_xattrs = false;
5060 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
5064 key.type = BTRFS_XATTR_ITEM_KEY;
5067 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5072 int slot = path->slots[0];
5073 struct extent_buffer *leaf = path->nodes[0];
5074 int nritems = btrfs_header_nritems(leaf);
5076 if (slot >= nritems) {
5078 ret = copy_items(trans, inode, dst_path, path,
5079 start_slot, ins_nr, 1, 0, ctx);
5084 ret = btrfs_next_leaf(root, path);
5092 btrfs_item_key_to_cpu(leaf, &key, slot);
5093 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
5100 found_xattrs = true;
5104 ret = copy_items(trans, inode, dst_path, path,
5105 start_slot, ins_nr, 1, 0, ctx);
5111 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
5117 * When using the NO_HOLES feature if we punched a hole that causes the
5118 * deletion of entire leafs or all the extent items of the first leaf (the one
5119 * that contains the inode item and references) we may end up not processing
5120 * any extents, because there are no leafs with a generation matching the
5121 * current transaction that have extent items for our inode. So we need to find
5122 * if any holes exist and then log them. We also need to log holes after any
5123 * truncate operation that changes the inode's size.
5125 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5126 struct btrfs_inode *inode,
5127 struct btrfs_path *path)
5129 struct btrfs_root *root = inode->root;
5130 struct btrfs_fs_info *fs_info = root->fs_info;
5131 struct btrfs_key key;
5132 const u64 ino = btrfs_ino(inode);
5133 const u64 i_size = i_size_read(&inode->vfs_inode);
5134 u64 prev_extent_end = 0;
5137 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5141 key.type = BTRFS_EXTENT_DATA_KEY;
5144 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5149 struct extent_buffer *leaf = path->nodes[0];
5151 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5152 ret = btrfs_next_leaf(root, path);
5159 leaf = path->nodes[0];
5162 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5163 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5166 /* We have a hole, log it. */
5167 if (prev_extent_end < key.offset) {
5168 const u64 hole_len = key.offset - prev_extent_end;
5171 * Release the path to avoid deadlocks with other code
5172 * paths that search the root while holding locks on
5173 * leafs from the log root.
5175 btrfs_release_path(path);
5176 ret = btrfs_insert_hole_extent(trans, root->log_root,
5177 ino, prev_extent_end,
5183 * Search for the same key again in the root. Since it's
5184 * an extent item and we are holding the inode lock, the
5185 * key must still exist. If it doesn't just emit warning
5186 * and return an error to fall back to a transaction
5189 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5192 if (WARN_ON(ret > 0))
5194 leaf = path->nodes[0];
5197 prev_extent_end = btrfs_file_extent_end(path);
5202 if (prev_extent_end < i_size) {
5205 btrfs_release_path(path);
5206 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5207 ret = btrfs_insert_hole_extent(trans, root->log_root, ino,
5208 prev_extent_end, hole_len);
5217 * When we are logging a new inode X, check if it doesn't have a reference that
5218 * matches the reference from some other inode Y created in a past transaction
5219 * and that was renamed in the current transaction. If we don't do this, then at
5220 * log replay time we can lose inode Y (and all its files if it's a directory):
5223 * echo "hello world" > /mnt/x/foobar
5226 * mkdir /mnt/x # or touch /mnt/x
5227 * xfs_io -c fsync /mnt/x
5229 * mount fs, trigger log replay
5231 * After the log replay procedure, we would lose the first directory and all its
5232 * files (file foobar).
5233 * For the case where inode Y is not a directory we simply end up losing it:
5235 * echo "123" > /mnt/foo
5237 * mv /mnt/foo /mnt/bar
5238 * echo "abc" > /mnt/foo
5239 * xfs_io -c fsync /mnt/foo
5242 * We also need this for cases where a snapshot entry is replaced by some other
5243 * entry (file or directory) otherwise we end up with an unreplayable log due to
5244 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5245 * if it were a regular entry:
5248 * btrfs subvolume snapshot /mnt /mnt/x/snap
5249 * btrfs subvolume delete /mnt/x/snap
5252 * fsync /mnt/x or fsync some new file inside it
5255 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5256 * the same transaction.
5258 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5260 const struct btrfs_key *key,
5261 struct btrfs_inode *inode,
5262 u64 *other_ino, u64 *other_parent)
5265 struct btrfs_path *search_path;
5268 u32 item_size = btrfs_item_size(eb, slot);
5270 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5272 search_path = btrfs_alloc_path();
5275 search_path->search_commit_root = 1;
5276 search_path->skip_locking = 1;
5278 while (cur_offset < item_size) {
5282 unsigned long name_ptr;
5283 struct btrfs_dir_item *di;
5284 struct fscrypt_str name_str;
5286 if (key->type == BTRFS_INODE_REF_KEY) {
5287 struct btrfs_inode_ref *iref;
5289 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5290 parent = key->offset;
5291 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5292 name_ptr = (unsigned long)(iref + 1);
5293 this_len = sizeof(*iref) + this_name_len;
5295 struct btrfs_inode_extref *extref;
5297 extref = (struct btrfs_inode_extref *)(ptr +
5299 parent = btrfs_inode_extref_parent(eb, extref);
5300 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5301 name_ptr = (unsigned long)&extref->name;
5302 this_len = sizeof(*extref) + this_name_len;
5305 if (this_name_len > name_len) {
5308 new_name = krealloc(name, this_name_len, GFP_NOFS);
5313 name_len = this_name_len;
5317 read_extent_buffer(eb, name, name_ptr, this_name_len);
5319 name_str.name = name;
5320 name_str.len = this_name_len;
5321 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5322 parent, &name_str, 0);
5323 if (di && !IS_ERR(di)) {
5324 struct btrfs_key di_key;
5326 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5328 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5329 if (di_key.objectid != key->objectid) {
5331 *other_ino = di_key.objectid;
5332 *other_parent = parent;
5340 } else if (IS_ERR(di)) {
5344 btrfs_release_path(search_path);
5346 cur_offset += this_len;
5350 btrfs_free_path(search_path);
5356 * Check if we need to log an inode. This is used in contexts where while
5357 * logging an inode we need to log another inode (either that it exists or in
5358 * full mode). This is used instead of btrfs_inode_in_log() because the later
5359 * requires the inode to be in the log and have the log transaction committed,
5360 * while here we do not care if the log transaction was already committed - our
5361 * caller will commit the log later - and we want to avoid logging an inode
5362 * multiple times when multiple tasks have joined the same log transaction.
5364 static bool need_log_inode(const struct btrfs_trans_handle *trans,
5365 struct btrfs_inode *inode)
5368 * If a directory was not modified, no dentries added or removed, we can
5369 * and should avoid logging it.
5371 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5375 * If this inode does not have new/updated/deleted xattrs since the last
5376 * time it was logged and is flagged as logged in the current transaction,
5377 * we can skip logging it. As for new/deleted names, those are updated in
5378 * the log by link/unlink/rename operations.
5379 * In case the inode was logged and then evicted and reloaded, its
5380 * logged_trans will be 0, in which case we have to fully log it since
5381 * logged_trans is a transient field, not persisted.
5383 if (inode_logged(trans, inode, NULL) == 1 &&
5384 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5390 struct btrfs_dir_list {
5392 struct list_head list;
5396 * Log the inodes of the new dentries of a directory.
5397 * See process_dir_items_leaf() for details about why it is needed.
5398 * This is a recursive operation - if an existing dentry corresponds to a
5399 * directory, that directory's new entries are logged too (same behaviour as
5400 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5401 * the dentries point to we do not acquire their VFS lock, otherwise lockdep
5402 * complains about the following circular lock dependency / possible deadlock:
5406 * lock(&type->i_mutex_dir_key#3/2);
5407 * lock(sb_internal#2);
5408 * lock(&type->i_mutex_dir_key#3/2);
5409 * lock(&sb->s_type->i_mutex_key#14);
5411 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5412 * sb_start_intwrite() in btrfs_start_transaction().
5413 * Not acquiring the VFS lock of the inodes is still safe because:
5415 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5416 * that while logging the inode new references (names) are added or removed
5417 * from the inode, leaving the logged inode item with a link count that does
5418 * not match the number of logged inode reference items. This is fine because
5419 * at log replay time we compute the real number of links and correct the
5420 * link count in the inode item (see replay_one_buffer() and
5421 * link_to_fixup_dir());
5423 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5424 * while logging the inode's items new index items (key type
5425 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5426 * has a size that doesn't match the sum of the lengths of all the logged
5427 * names - this is ok, not a problem, because at log replay time we set the
5428 * directory's i_size to the correct value (see replay_one_name() and
5429 * overwrite_item()).
5431 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5432 struct btrfs_inode *start_inode,
5433 struct btrfs_log_ctx *ctx)
5435 struct btrfs_root *root = start_inode->root;
5436 struct btrfs_fs_info *fs_info = root->fs_info;
5437 struct btrfs_path *path;
5438 LIST_HEAD(dir_list);
5439 struct btrfs_dir_list *dir_elem;
5440 u64 ino = btrfs_ino(start_inode);
5441 struct btrfs_inode *curr_inode = start_inode;
5445 * If we are logging a new name, as part of a link or rename operation,
5446 * don't bother logging new dentries, as we just want to log the names
5447 * of an inode and that any new parents exist.
5449 if (ctx->logging_new_name)
5452 path = btrfs_alloc_path();
5456 /* Pairs with btrfs_add_delayed_iput below. */
5457 ihold(&curr_inode->vfs_inode);
5460 struct inode *vfs_inode;
5461 struct btrfs_key key;
5462 struct btrfs_key found_key;
5464 bool continue_curr_inode = true;
5468 key.type = BTRFS_DIR_INDEX_KEY;
5469 key.offset = btrfs_get_first_dir_index_to_log(curr_inode);
5470 next_index = key.offset;
5472 btrfs_for_each_slot(root->log_root, &key, &found_key, path, iter_ret) {
5473 struct extent_buffer *leaf = path->nodes[0];
5474 struct btrfs_dir_item *di;
5475 struct btrfs_key di_key;
5476 struct inode *di_inode;
5477 int log_mode = LOG_INODE_EXISTS;
5480 if (found_key.objectid != ino ||
5481 found_key.type != BTRFS_DIR_INDEX_KEY) {
5482 continue_curr_inode = false;
5486 next_index = found_key.offset + 1;
5488 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5489 type = btrfs_dir_ftype(leaf, di);
5490 if (btrfs_dir_transid(leaf, di) < trans->transid)
5492 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5493 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5496 btrfs_release_path(path);
5497 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5498 if (IS_ERR(di_inode)) {
5499 ret = PTR_ERR(di_inode);
5503 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5504 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5508 ctx->log_new_dentries = false;
5509 if (type == BTRFS_FT_DIR)
5510 log_mode = LOG_INODE_ALL;
5511 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5513 btrfs_add_delayed_iput(BTRFS_I(di_inode));
5516 if (ctx->log_new_dentries) {
5517 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5522 dir_elem->ino = di_key.objectid;
5523 list_add_tail(&dir_elem->list, &dir_list);
5528 btrfs_release_path(path);
5533 } else if (iter_ret > 0) {
5534 continue_curr_inode = false;
5539 if (continue_curr_inode && key.offset < (u64)-1) {
5544 btrfs_set_first_dir_index_to_log(curr_inode, next_index);
5546 if (list_empty(&dir_list))
5549 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list);
5550 ino = dir_elem->ino;
5551 list_del(&dir_elem->list);
5554 btrfs_add_delayed_iput(curr_inode);
5557 vfs_inode = btrfs_iget(fs_info->sb, ino, root);
5558 if (IS_ERR(vfs_inode)) {
5559 ret = PTR_ERR(vfs_inode);
5562 curr_inode = BTRFS_I(vfs_inode);
5565 btrfs_free_path(path);
5567 btrfs_add_delayed_iput(curr_inode);
5570 struct btrfs_dir_list *next;
5572 list_for_each_entry_safe(dir_elem, next, &dir_list, list)
5579 struct btrfs_ino_list {
5582 struct list_head list;
5585 static void free_conflicting_inodes(struct btrfs_log_ctx *ctx)
5587 struct btrfs_ino_list *curr;
5588 struct btrfs_ino_list *next;
5590 list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) {
5591 list_del(&curr->list);
5596 static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino,
5597 struct btrfs_path *path)
5599 struct btrfs_key key;
5603 key.type = BTRFS_INODE_ITEM_KEY;
5606 path->search_commit_root = 1;
5607 path->skip_locking = 1;
5609 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5610 if (WARN_ON_ONCE(ret > 0)) {
5612 * We have previously found the inode through the commit root
5613 * so this should not happen. If it does, just error out and
5614 * fallback to a transaction commit.
5617 } else if (ret == 0) {
5618 struct btrfs_inode_item *item;
5620 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5621 struct btrfs_inode_item);
5622 if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item)))
5626 btrfs_release_path(path);
5627 path->search_commit_root = 0;
5628 path->skip_locking = 0;
5633 static int add_conflicting_inode(struct btrfs_trans_handle *trans,
5634 struct btrfs_root *root,
5635 struct btrfs_path *path,
5636 u64 ino, u64 parent,
5637 struct btrfs_log_ctx *ctx)
5639 struct btrfs_ino_list *ino_elem;
5640 struct inode *inode;
5643 * It's rare to have a lot of conflicting inodes, in practice it is not
5644 * common to have more than 1 or 2. We don't want to collect too many,
5645 * as we could end up logging too many inodes (even if only in
5646 * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction
5649 if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES)
5650 return BTRFS_LOG_FORCE_COMMIT;
5652 inode = btrfs_iget(root->fs_info->sb, ino, root);
5654 * If the other inode that had a conflicting dir entry was deleted in
5655 * the current transaction then we either:
5657 * 1) Log the parent directory (later after adding it to the list) if
5658 * the inode is a directory. This is because it may be a deleted
5659 * subvolume/snapshot or it may be a regular directory that had
5660 * deleted subvolumes/snapshots (or subdirectories that had them),
5661 * and at the moment we can't deal with dropping subvolumes/snapshots
5662 * during log replay. So we just log the parent, which will result in
5663 * a fallback to a transaction commit if we are dealing with those
5664 * cases (last_unlink_trans will match the current transaction);
5666 * 2) Do nothing if it's not a directory. During log replay we simply
5667 * unlink the conflicting dentry from the parent directory and then
5668 * add the dentry for our inode. Like this we can avoid logging the
5669 * parent directory (and maybe fallback to a transaction commit in
5670 * case it has a last_unlink_trans == trans->transid, due to moving
5671 * some inode from it to some other directory).
5673 if (IS_ERR(inode)) {
5674 int ret = PTR_ERR(inode);
5679 ret = conflicting_inode_is_dir(root, ino, path);
5680 /* Not a directory or we got an error. */
5684 /* Conflicting inode is a directory, so we'll log its parent. */
5685 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5688 ino_elem->ino = ino;
5689 ino_elem->parent = parent;
5690 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5691 ctx->num_conflict_inodes++;
5697 * If the inode was already logged skip it - otherwise we can hit an
5698 * infinite loop. Example:
5700 * From the commit root (previous transaction) we have the following
5703 * inode 257 a directory
5704 * inode 258 with references "zz" and "zz_link" on inode 257
5705 * inode 259 with reference "a" on inode 257
5707 * And in the current (uncommitted) transaction we have:
5709 * inode 257 a directory, unchanged
5710 * inode 258 with references "a" and "a2" on inode 257
5711 * inode 259 with reference "zz_link" on inode 257
5712 * inode 261 with reference "zz" on inode 257
5714 * When logging inode 261 the following infinite loop could
5715 * happen if we don't skip already logged inodes:
5717 * - we detect inode 258 as a conflicting inode, with inode 261
5718 * on reference "zz", and log it;
5720 * - we detect inode 259 as a conflicting inode, with inode 258
5721 * on reference "a", and log it;
5723 * - we detect inode 258 as a conflicting inode, with inode 259
5724 * on reference "zz_link", and log it - again! After this we
5725 * repeat the above steps forever.
5727 * Here we can use need_log_inode() because we only need to log the
5728 * inode in LOG_INODE_EXISTS mode and rename operations update the log,
5729 * so that the log ends up with the new name and without the old name.
5731 if (!need_log_inode(trans, BTRFS_I(inode))) {
5732 btrfs_add_delayed_iput(BTRFS_I(inode));
5736 btrfs_add_delayed_iput(BTRFS_I(inode));
5738 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5741 ino_elem->ino = ino;
5742 ino_elem->parent = parent;
5743 list_add_tail(&ino_elem->list, &ctx->conflict_inodes);
5744 ctx->num_conflict_inodes++;
5749 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5750 struct btrfs_root *root,
5751 struct btrfs_log_ctx *ctx)
5753 struct btrfs_fs_info *fs_info = root->fs_info;
5757 * Conflicting inodes are logged by the first call to btrfs_log_inode(),
5758 * otherwise we could have unbounded recursion of btrfs_log_inode()
5759 * calls. This check guarantees we can have only 1 level of recursion.
5761 if (ctx->logging_conflict_inodes)
5764 ctx->logging_conflict_inodes = true;
5767 * New conflicting inodes may be found and added to the list while we
5768 * are logging a conflicting inode, so keep iterating while the list is
5771 while (!list_empty(&ctx->conflict_inodes)) {
5772 struct btrfs_ino_list *curr;
5773 struct inode *inode;
5777 curr = list_first_entry(&ctx->conflict_inodes,
5778 struct btrfs_ino_list, list);
5780 parent = curr->parent;
5781 list_del(&curr->list);
5784 inode = btrfs_iget(fs_info->sb, ino, root);
5786 * If the other inode that had a conflicting dir entry was
5787 * deleted in the current transaction, we need to log its parent
5788 * directory. See the comment at add_conflicting_inode().
5790 if (IS_ERR(inode)) {
5791 ret = PTR_ERR(inode);
5795 inode = btrfs_iget(fs_info->sb, parent, root);
5796 if (IS_ERR(inode)) {
5797 ret = PTR_ERR(inode);
5802 * Always log the directory, we cannot make this
5803 * conditional on need_log_inode() because the directory
5804 * might have been logged in LOG_INODE_EXISTS mode or
5805 * the dir index of the conflicting inode is not in a
5806 * dir index key range logged for the directory. So we
5807 * must make sure the deletion is recorded.
5809 ret = btrfs_log_inode(trans, BTRFS_I(inode),
5810 LOG_INODE_ALL, ctx);
5811 btrfs_add_delayed_iput(BTRFS_I(inode));
5818 * Here we can use need_log_inode() because we only need to log
5819 * the inode in LOG_INODE_EXISTS mode and rename operations
5820 * update the log, so that the log ends up with the new name and
5821 * without the old name.
5823 * We did this check at add_conflicting_inode(), but here we do
5824 * it again because if some other task logged the inode after
5825 * that, we can avoid doing it again.
5827 if (!need_log_inode(trans, BTRFS_I(inode))) {
5828 btrfs_add_delayed_iput(BTRFS_I(inode));
5833 * We are safe logging the other inode without acquiring its
5834 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5835 * are safe against concurrent renames of the other inode as
5836 * well because during a rename we pin the log and update the
5837 * log with the new name before we unpin it.
5839 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx);
5840 btrfs_add_delayed_iput(BTRFS_I(inode));
5845 ctx->logging_conflict_inodes = false;
5847 free_conflicting_inodes(ctx);
5852 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5853 struct btrfs_inode *inode,
5854 struct btrfs_key *min_key,
5855 const struct btrfs_key *max_key,
5856 struct btrfs_path *path,
5857 struct btrfs_path *dst_path,
5858 const u64 logged_isize,
5859 const int inode_only,
5860 struct btrfs_log_ctx *ctx,
5861 bool *need_log_inode_item)
5863 const u64 i_size = i_size_read(&inode->vfs_inode);
5864 struct btrfs_root *root = inode->root;
5865 int ins_start_slot = 0;
5870 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5878 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5879 if (min_key->objectid != max_key->objectid)
5881 if (min_key->type > max_key->type)
5884 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5885 *need_log_inode_item = false;
5886 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5887 min_key->offset >= i_size) {
5889 * Extents at and beyond eof are logged with
5890 * btrfs_log_prealloc_extents().
5891 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5892 * and no keys greater than that, so bail out.
5895 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5896 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5897 (inode->generation == trans->transid ||
5898 ctx->logging_conflict_inodes)) {
5900 u64 other_parent = 0;
5902 ret = btrfs_check_ref_name_override(path->nodes[0],
5903 path->slots[0], min_key, inode,
5904 &other_ino, &other_parent);
5907 } else if (ret > 0 &&
5908 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5913 ins_start_slot = path->slots[0];
5915 ret = copy_items(trans, inode, dst_path, path,
5916 ins_start_slot, ins_nr,
5917 inode_only, logged_isize, ctx);
5922 btrfs_release_path(path);
5923 ret = add_conflicting_inode(trans, root, path,
5930 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5931 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5934 ret = copy_items(trans, inode, dst_path, path,
5936 ins_nr, inode_only, logged_isize, ctx);
5943 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5946 } else if (!ins_nr) {
5947 ins_start_slot = path->slots[0];
5952 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5953 ins_nr, inode_only, logged_isize, ctx);
5957 ins_start_slot = path->slots[0];
5960 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5961 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5966 ret = copy_items(trans, inode, dst_path, path,
5967 ins_start_slot, ins_nr, inode_only,
5973 btrfs_release_path(path);
5975 if (min_key->offset < (u64)-1) {
5977 } else if (min_key->type < max_key->type) {
5979 min_key->offset = 0;
5985 * We may process many leaves full of items for our inode, so
5986 * avoid monopolizing a cpu for too long by rescheduling while
5987 * not holding locks on any tree.
5992 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5993 ins_nr, inode_only, logged_isize, ctx);
5998 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
6000 * Release the path because otherwise we might attempt to double
6001 * lock the same leaf with btrfs_log_prealloc_extents() below.
6003 btrfs_release_path(path);
6004 ret = btrfs_log_prealloc_extents(trans, inode, dst_path, ctx);
6010 static int insert_delayed_items_batch(struct btrfs_trans_handle *trans,
6011 struct btrfs_root *log,
6012 struct btrfs_path *path,
6013 const struct btrfs_item_batch *batch,
6014 const struct btrfs_delayed_item *first_item)
6016 const struct btrfs_delayed_item *curr = first_item;
6019 ret = btrfs_insert_empty_items(trans, log, path, batch);
6023 for (int i = 0; i < batch->nr; i++) {
6026 data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char);
6027 write_extent_buffer(path->nodes[0], &curr->data,
6028 (unsigned long)data_ptr, curr->data_len);
6029 curr = list_next_entry(curr, log_list);
6033 btrfs_release_path(path);
6038 static int log_delayed_insertion_items(struct btrfs_trans_handle *trans,
6039 struct btrfs_inode *inode,
6040 struct btrfs_path *path,
6041 const struct list_head *delayed_ins_list,
6042 struct btrfs_log_ctx *ctx)
6044 /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
6045 const int max_batch_size = 195;
6046 const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info);
6047 const u64 ino = btrfs_ino(inode);
6048 struct btrfs_root *log = inode->root->log_root;
6049 struct btrfs_item_batch batch = {
6051 .total_data_size = 0,
6053 const struct btrfs_delayed_item *first = NULL;
6054 const struct btrfs_delayed_item *curr;
6056 struct btrfs_key *ins_keys;
6058 u64 curr_batch_size = 0;
6062 /* We are adding dir index items to the log tree. */
6063 lockdep_assert_held(&inode->log_mutex);
6066 * We collect delayed items before copying index keys from the subvolume
6067 * to the log tree. However just after we collected them, they may have
6068 * been flushed (all of them or just some of them), and therefore we
6069 * could have copied them from the subvolume tree to the log tree.
6070 * So find the first delayed item that was not yet logged (they are
6071 * sorted by index number).
6073 list_for_each_entry(curr, delayed_ins_list, log_list) {
6074 if (curr->index > inode->last_dir_index_offset) {
6080 /* Empty list or all delayed items were already logged. */
6084 ins_data = kmalloc(max_batch_size * sizeof(u32) +
6085 max_batch_size * sizeof(struct btrfs_key), GFP_NOFS);
6088 ins_sizes = (u32 *)ins_data;
6089 batch.data_sizes = ins_sizes;
6090 ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32));
6091 batch.keys = ins_keys;
6094 while (!list_entry_is_head(curr, delayed_ins_list, log_list)) {
6095 const u32 curr_size = curr->data_len + sizeof(struct btrfs_item);
6097 if (curr_batch_size + curr_size > leaf_data_size ||
6098 batch.nr == max_batch_size) {
6099 ret = insert_delayed_items_batch(trans, log, path,
6105 batch.total_data_size = 0;
6106 curr_batch_size = 0;
6110 ins_sizes[batch_idx] = curr->data_len;
6111 ins_keys[batch_idx].objectid = ino;
6112 ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY;
6113 ins_keys[batch_idx].offset = curr->index;
6114 curr_batch_size += curr_size;
6115 batch.total_data_size += curr->data_len;
6118 curr = list_next_entry(curr, log_list);
6121 ASSERT(batch.nr >= 1);
6122 ret = insert_delayed_items_batch(trans, log, path, &batch, first);
6124 curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item,
6126 inode->last_dir_index_offset = curr->index;
6133 static int log_delayed_deletions_full(struct btrfs_trans_handle *trans,
6134 struct btrfs_inode *inode,
6135 struct btrfs_path *path,
6136 const struct list_head *delayed_del_list,
6137 struct btrfs_log_ctx *ctx)
6139 const u64 ino = btrfs_ino(inode);
6140 const struct btrfs_delayed_item *curr;
6142 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6145 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6146 u64 first_dir_index = curr->index;
6148 const struct btrfs_delayed_item *next;
6152 * Find a range of consecutive dir index items to delete. Like
6153 * this we log a single dir range item spanning several contiguous
6154 * dir items instead of logging one range item per dir index item.
6156 next = list_next_entry(curr, log_list);
6157 while (!list_entry_is_head(next, delayed_del_list, log_list)) {
6158 if (next->index != curr->index + 1)
6161 next = list_next_entry(next, log_list);
6164 last_dir_index = curr->index;
6165 ASSERT(last_dir_index >= first_dir_index);
6167 ret = insert_dir_log_key(trans, inode->root->log_root, path,
6168 ino, first_dir_index, last_dir_index);
6171 curr = list_next_entry(curr, log_list);
6177 static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans,
6178 struct btrfs_inode *inode,
6179 struct btrfs_path *path,
6180 struct btrfs_log_ctx *ctx,
6181 const struct list_head *delayed_del_list,
6182 const struct btrfs_delayed_item *first,
6183 const struct btrfs_delayed_item **last_ret)
6185 const struct btrfs_delayed_item *next;
6186 struct extent_buffer *leaf = path->nodes[0];
6187 const int last_slot = btrfs_header_nritems(leaf) - 1;
6188 int slot = path->slots[0] + 1;
6189 const u64 ino = btrfs_ino(inode);
6191 next = list_next_entry(first, log_list);
6193 while (slot < last_slot &&
6194 !list_entry_is_head(next, delayed_del_list, log_list)) {
6195 struct btrfs_key key;
6197 btrfs_item_key_to_cpu(leaf, &key, slot);
6198 if (key.objectid != ino ||
6199 key.type != BTRFS_DIR_INDEX_KEY ||
6200 key.offset != next->index)
6205 next = list_next_entry(next, log_list);
6208 return btrfs_del_items(trans, inode->root->log_root, path,
6209 path->slots[0], slot - path->slots[0]);
6212 static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans,
6213 struct btrfs_inode *inode,
6214 struct btrfs_path *path,
6215 const struct list_head *delayed_del_list,
6216 struct btrfs_log_ctx *ctx)
6218 struct btrfs_root *log = inode->root->log_root;
6219 const struct btrfs_delayed_item *curr;
6220 u64 last_range_start = 0;
6221 u64 last_range_end = 0;
6222 struct btrfs_key key;
6224 key.objectid = btrfs_ino(inode);
6225 key.type = BTRFS_DIR_INDEX_KEY;
6226 curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item,
6229 while (!list_entry_is_head(curr, delayed_del_list, log_list)) {
6230 const struct btrfs_delayed_item *last = curr;
6231 u64 first_dir_index = curr->index;
6233 bool deleted_items = false;
6236 key.offset = curr->index;
6237 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
6240 } else if (ret == 0) {
6241 ret = batch_delete_dir_index_items(trans, inode, path, ctx,
6242 delayed_del_list, curr,
6246 deleted_items = true;
6249 btrfs_release_path(path);
6252 * If we deleted items from the leaf, it means we have a range
6253 * item logging their range, so no need to add one or update an
6254 * existing one. Otherwise we have to log a dir range item.
6259 last_dir_index = last->index;
6260 ASSERT(last_dir_index >= first_dir_index);
6262 * If this range starts right after where the previous one ends,
6263 * then we want to reuse the previous range item and change its
6264 * end offset to the end of this range. This is just to minimize
6265 * leaf space usage, by avoiding adding a new range item.
6267 if (last_range_end != 0 && first_dir_index == last_range_end + 1)
6268 first_dir_index = last_range_start;
6270 ret = insert_dir_log_key(trans, log, path, key.objectid,
6271 first_dir_index, last_dir_index);
6275 last_range_start = first_dir_index;
6276 last_range_end = last_dir_index;
6278 curr = list_next_entry(last, log_list);
6284 static int log_delayed_deletion_items(struct btrfs_trans_handle *trans,
6285 struct btrfs_inode *inode,
6286 struct btrfs_path *path,
6287 const struct list_head *delayed_del_list,
6288 struct btrfs_log_ctx *ctx)
6291 * We are deleting dir index items from the log tree or adding range
6294 lockdep_assert_held(&inode->log_mutex);
6296 if (list_empty(delayed_del_list))
6299 if (ctx->logged_before)
6300 return log_delayed_deletions_incremental(trans, inode, path,
6301 delayed_del_list, ctx);
6303 return log_delayed_deletions_full(trans, inode, path, delayed_del_list,
6308 * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed
6309 * items instead of the subvolume tree.
6311 static int log_new_delayed_dentries(struct btrfs_trans_handle *trans,
6312 struct btrfs_inode *inode,
6313 const struct list_head *delayed_ins_list,
6314 struct btrfs_log_ctx *ctx)
6316 const bool orig_log_new_dentries = ctx->log_new_dentries;
6317 struct btrfs_fs_info *fs_info = trans->fs_info;
6318 struct btrfs_delayed_item *item;
6322 * No need for the log mutex, plus to avoid potential deadlocks or
6323 * lockdep annotations due to nesting of delayed inode mutexes and log
6326 lockdep_assert_not_held(&inode->log_mutex);
6328 ASSERT(!ctx->logging_new_delayed_dentries);
6329 ctx->logging_new_delayed_dentries = true;
6331 list_for_each_entry(item, delayed_ins_list, log_list) {
6332 struct btrfs_dir_item *dir_item;
6333 struct inode *di_inode;
6334 struct btrfs_key key;
6335 int log_mode = LOG_INODE_EXISTS;
6337 dir_item = (struct btrfs_dir_item *)item->data;
6338 btrfs_disk_key_to_cpu(&key, &dir_item->location);
6340 if (key.type == BTRFS_ROOT_ITEM_KEY)
6343 di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root);
6344 if (IS_ERR(di_inode)) {
6345 ret = PTR_ERR(di_inode);
6349 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
6350 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6354 if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR)
6355 log_mode = LOG_INODE_ALL;
6357 ctx->log_new_dentries = false;
6358 ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx);
6360 if (!ret && ctx->log_new_dentries)
6361 ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx);
6363 btrfs_add_delayed_iput(BTRFS_I(di_inode));
6369 ctx->log_new_dentries = orig_log_new_dentries;
6370 ctx->logging_new_delayed_dentries = false;
6375 /* log a single inode in the tree log.
6376 * At least one parent directory for this inode must exist in the tree
6377 * or be logged already.
6379 * Any items from this inode changed by the current transaction are copied
6380 * to the log tree. An extra reference is taken on any extents in this
6381 * file, allowing us to avoid a whole pile of corner cases around logging
6382 * blocks that have been removed from the tree.
6384 * See LOG_INODE_ALL and related defines for a description of what inode_only
6387 * This handles both files and directories.
6389 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
6390 struct btrfs_inode *inode,
6392 struct btrfs_log_ctx *ctx)
6394 struct btrfs_path *path;
6395 struct btrfs_path *dst_path;
6396 struct btrfs_key min_key;
6397 struct btrfs_key max_key;
6398 struct btrfs_root *log = inode->root->log_root;
6400 bool fast_search = false;
6401 u64 ino = btrfs_ino(inode);
6402 struct extent_map_tree *em_tree = &inode->extent_tree;
6403 u64 logged_isize = 0;
6404 bool need_log_inode_item = true;
6405 bool xattrs_logged = false;
6406 bool inode_item_dropped = true;
6407 bool full_dir_logging = false;
6408 LIST_HEAD(delayed_ins_list);
6409 LIST_HEAD(delayed_del_list);
6411 path = btrfs_alloc_path();
6414 dst_path = btrfs_alloc_path();
6416 btrfs_free_path(path);
6420 min_key.objectid = ino;
6421 min_key.type = BTRFS_INODE_ITEM_KEY;
6424 max_key.objectid = ino;
6427 /* today the code can only do partial logging of directories */
6428 if (S_ISDIR(inode->vfs_inode.i_mode) ||
6429 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6430 &inode->runtime_flags) &&
6431 inode_only >= LOG_INODE_EXISTS))
6432 max_key.type = BTRFS_XATTR_ITEM_KEY;
6434 max_key.type = (u8)-1;
6435 max_key.offset = (u64)-1;
6437 if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL)
6438 full_dir_logging = true;
6441 * If we are logging a directory while we are logging dentries of the
6442 * delayed items of some other inode, then we need to flush the delayed
6443 * items of this directory and not log the delayed items directly. This
6444 * is to prevent more than one level of recursion into btrfs_log_inode()
6445 * by having something like this:
6447 * $ mkdir -p a/b/c/d/e/f/g/h/...
6448 * $ xfs_io -c "fsync" a
6450 * Where all directories in the path did not exist before and are
6451 * created in the current transaction.
6452 * So in such a case we directly log the delayed items of the main
6453 * directory ("a") without flushing them first, while for each of its
6454 * subdirectories we flush their delayed items before logging them.
6455 * This prevents a potential unbounded recursion like this:
6458 * log_new_delayed_dentries()
6460 * log_new_delayed_dentries()
6462 * log_new_delayed_dentries()
6465 * We have thresholds for the maximum number of delayed items to have in
6466 * memory, and once they are hit, the items are flushed asynchronously.
6467 * However the limit is quite high, so lets prevent deep levels of
6468 * recursion to happen by limiting the maximum depth to be 1.
6470 if (full_dir_logging && ctx->logging_new_delayed_dentries) {
6471 ret = btrfs_commit_inode_delayed_items(trans, inode);
6476 mutex_lock(&inode->log_mutex);
6479 * For symlinks, we must always log their content, which is stored in an
6480 * inline extent, otherwise we could end up with an empty symlink after
6481 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
6482 * one attempts to create an empty symlink).
6483 * We don't need to worry about flushing delalloc, because when we create
6484 * the inline extent when the symlink is created (we never have delalloc
6487 if (S_ISLNK(inode->vfs_inode.i_mode))
6488 inode_only = LOG_INODE_ALL;
6491 * Before logging the inode item, cache the value returned by
6492 * inode_logged(), because after that we have the need to figure out if
6493 * the inode was previously logged in this transaction.
6495 ret = inode_logged(trans, inode, path);
6498 ctx->logged_before = (ret == 1);
6502 * This is for cases where logging a directory could result in losing a
6503 * a file after replaying the log. For example, if we move a file from a
6504 * directory A to a directory B, then fsync directory A, we have no way
6505 * to known the file was moved from A to B, so logging just A would
6506 * result in losing the file after a log replay.
6508 if (full_dir_logging && inode->last_unlink_trans >= trans->transid) {
6509 ret = BTRFS_LOG_FORCE_COMMIT;
6514 * a brute force approach to making sure we get the most uptodate
6515 * copies of everything.
6517 if (S_ISDIR(inode->vfs_inode.i_mode)) {
6518 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
6519 if (ctx->logged_before)
6520 ret = drop_inode_items(trans, log, path, inode,
6521 BTRFS_XATTR_ITEM_KEY);
6523 if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) {
6525 * Make sure the new inode item we write to the log has
6526 * the same isize as the current one (if it exists).
6527 * This is necessary to prevent data loss after log
6528 * replay, and also to prevent doing a wrong expanding
6529 * truncate - for e.g. create file, write 4K into offset
6530 * 0, fsync, write 4K into offset 4096, add hard link,
6531 * fsync some other file (to sync log), power fail - if
6532 * we use the inode's current i_size, after log replay
6533 * we get a 8Kb file, with the last 4Kb extent as a hole
6534 * (zeroes), as if an expanding truncate happened,
6535 * instead of getting a file of 4Kb only.
6537 ret = logged_inode_size(log, inode, path, &logged_isize);
6541 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6542 &inode->runtime_flags)) {
6543 if (inode_only == LOG_INODE_EXISTS) {
6544 max_key.type = BTRFS_XATTR_ITEM_KEY;
6545 if (ctx->logged_before)
6546 ret = drop_inode_items(trans, log, path,
6547 inode, max_key.type);
6549 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
6550 &inode->runtime_flags);
6551 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6552 &inode->runtime_flags);
6553 if (ctx->logged_before)
6554 ret = truncate_inode_items(trans, log,
6557 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
6558 &inode->runtime_flags) ||
6559 inode_only == LOG_INODE_EXISTS) {
6560 if (inode_only == LOG_INODE_ALL)
6562 max_key.type = BTRFS_XATTR_ITEM_KEY;
6563 if (ctx->logged_before)
6564 ret = drop_inode_items(trans, log, path, inode,
6567 if (inode_only == LOG_INODE_ALL)
6569 inode_item_dropped = false;
6578 * If we are logging a directory in full mode, collect the delayed items
6579 * before iterating the subvolume tree, so that we don't miss any new
6580 * dir index items in case they get flushed while or right after we are
6581 * iterating the subvolume tree.
6583 if (full_dir_logging && !ctx->logging_new_delayed_dentries)
6584 btrfs_log_get_delayed_items(inode, &delayed_ins_list,
6587 ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
6588 path, dst_path, logged_isize,
6590 &need_log_inode_item);
6594 btrfs_release_path(path);
6595 btrfs_release_path(dst_path);
6596 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path, ctx);
6599 xattrs_logged = true;
6600 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
6601 btrfs_release_path(path);
6602 btrfs_release_path(dst_path);
6603 ret = btrfs_log_holes(trans, inode, path);
6608 btrfs_release_path(path);
6609 btrfs_release_path(dst_path);
6610 if (need_log_inode_item) {
6611 ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
6615 * If we are doing a fast fsync and the inode was logged before
6616 * in this transaction, we don't need to log the xattrs because
6617 * they were logged before. If xattrs were added, changed or
6618 * deleted since the last time we logged the inode, then we have
6619 * already logged them because the inode had the runtime flag
6620 * BTRFS_INODE_COPY_EVERYTHING set.
6622 if (!xattrs_logged && inode->logged_trans < trans->transid) {
6623 ret = btrfs_log_all_xattrs(trans, inode, path, dst_path, ctx);
6626 btrfs_release_path(path);
6630 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
6633 } else if (inode_only == LOG_INODE_ALL) {
6634 struct extent_map *em, *n;
6636 write_lock(&em_tree->lock);
6637 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
6638 list_del_init(&em->list);
6639 write_unlock(&em_tree->lock);
6642 if (full_dir_logging) {
6643 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
6646 ret = log_delayed_insertion_items(trans, inode, path,
6647 &delayed_ins_list, ctx);
6650 ret = log_delayed_deletion_items(trans, inode, path,
6651 &delayed_del_list, ctx);
6656 spin_lock(&inode->lock);
6657 inode->logged_trans = trans->transid;
6659 * Don't update last_log_commit if we logged that an inode exists.
6660 * We do this for three reasons:
6662 * 1) We might have had buffered writes to this inode that were
6663 * flushed and had their ordered extents completed in this
6664 * transaction, but we did not previously log the inode with
6665 * LOG_INODE_ALL. Later the inode was evicted and after that
6666 * it was loaded again and this LOG_INODE_EXISTS log operation
6667 * happened. We must make sure that if an explicit fsync against
6668 * the inode is performed later, it logs the new extents, an
6669 * updated inode item, etc, and syncs the log. The same logic
6670 * applies to direct IO writes instead of buffered writes.
6672 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
6673 * is logged with an i_size of 0 or whatever value was logged
6674 * before. If later the i_size of the inode is increased by a
6675 * truncate operation, the log is synced through an fsync of
6676 * some other inode and then finally an explicit fsync against
6677 * this inode is made, we must make sure this fsync logs the
6678 * inode with the new i_size, the hole between old i_size and
6679 * the new i_size, and syncs the log.
6681 * 3) If we are logging that an ancestor inode exists as part of
6682 * logging a new name from a link or rename operation, don't update
6683 * its last_log_commit - otherwise if an explicit fsync is made
6684 * against an ancestor, the fsync considers the inode in the log
6685 * and doesn't sync the log, resulting in the ancestor missing after
6686 * a power failure unless the log was synced as part of an fsync
6687 * against any other unrelated inode.
6689 if (inode_only != LOG_INODE_EXISTS)
6690 inode->last_log_commit = inode->last_sub_trans;
6691 spin_unlock(&inode->lock);
6694 * Reset the last_reflink_trans so that the next fsync does not need to
6695 * go through the slower path when logging extents and their checksums.
6697 if (inode_only == LOG_INODE_ALL)
6698 inode->last_reflink_trans = 0;
6701 mutex_unlock(&inode->log_mutex);
6703 btrfs_free_path(path);
6704 btrfs_free_path(dst_path);
6707 free_conflicting_inodes(ctx);
6709 ret = log_conflicting_inodes(trans, inode->root, ctx);
6711 if (full_dir_logging && !ctx->logging_new_delayed_dentries) {
6713 ret = log_new_delayed_dentries(trans, inode,
6714 &delayed_ins_list, ctx);
6716 btrfs_log_put_delayed_items(inode, &delayed_ins_list,
6723 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6724 struct btrfs_inode *inode,
6725 struct btrfs_log_ctx *ctx)
6727 struct btrfs_fs_info *fs_info = trans->fs_info;
6729 struct btrfs_path *path;
6730 struct btrfs_key key;
6731 struct btrfs_root *root = inode->root;
6732 const u64 ino = btrfs_ino(inode);
6734 path = btrfs_alloc_path();
6737 path->skip_locking = 1;
6738 path->search_commit_root = 1;
6741 key.type = BTRFS_INODE_REF_KEY;
6743 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6748 struct extent_buffer *leaf = path->nodes[0];
6749 int slot = path->slots[0];
6754 if (slot >= btrfs_header_nritems(leaf)) {
6755 ret = btrfs_next_leaf(root, path);
6763 btrfs_item_key_to_cpu(leaf, &key, slot);
6764 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6765 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6768 item_size = btrfs_item_size(leaf, slot);
6769 ptr = btrfs_item_ptr_offset(leaf, slot);
6770 while (cur_offset < item_size) {
6771 struct btrfs_key inode_key;
6772 struct inode *dir_inode;
6774 inode_key.type = BTRFS_INODE_ITEM_KEY;
6775 inode_key.offset = 0;
6777 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6778 struct btrfs_inode_extref *extref;
6780 extref = (struct btrfs_inode_extref *)
6782 inode_key.objectid = btrfs_inode_extref_parent(
6784 cur_offset += sizeof(*extref);
6785 cur_offset += btrfs_inode_extref_name_len(leaf,
6788 inode_key.objectid = key.offset;
6789 cur_offset = item_size;
6792 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6795 * If the parent inode was deleted, return an error to
6796 * fallback to a transaction commit. This is to prevent
6797 * getting an inode that was moved from one parent A to
6798 * a parent B, got its former parent A deleted and then
6799 * it got fsync'ed, from existing at both parents after
6800 * a log replay (and the old parent still existing).
6807 * mv /mnt/B/bar /mnt/A/bar
6808 * mv -T /mnt/A /mnt/B
6812 * If we ignore the old parent B which got deleted,
6813 * after a log replay we would have file bar linked
6814 * at both parents and the old parent B would still
6817 if (IS_ERR(dir_inode)) {
6818 ret = PTR_ERR(dir_inode);
6822 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6823 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6827 ctx->log_new_dentries = false;
6828 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6829 LOG_INODE_ALL, ctx);
6830 if (!ret && ctx->log_new_dentries)
6831 ret = log_new_dir_dentries(trans,
6832 BTRFS_I(dir_inode), ctx);
6833 btrfs_add_delayed_iput(BTRFS_I(dir_inode));
6841 btrfs_free_path(path);
6845 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6846 struct btrfs_root *root,
6847 struct btrfs_path *path,
6848 struct btrfs_log_ctx *ctx)
6850 struct btrfs_key found_key;
6852 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6855 struct btrfs_fs_info *fs_info = root->fs_info;
6856 struct extent_buffer *leaf;
6858 struct btrfs_key search_key;
6859 struct inode *inode;
6863 btrfs_release_path(path);
6865 ino = found_key.offset;
6867 search_key.objectid = found_key.offset;
6868 search_key.type = BTRFS_INODE_ITEM_KEY;
6869 search_key.offset = 0;
6870 inode = btrfs_iget(fs_info->sb, ino, root);
6872 return PTR_ERR(inode);
6874 if (BTRFS_I(inode)->generation >= trans->transid &&
6875 need_log_inode(trans, BTRFS_I(inode)))
6876 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6877 LOG_INODE_EXISTS, ctx);
6878 btrfs_add_delayed_iput(BTRFS_I(inode));
6882 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6885 search_key.type = BTRFS_INODE_REF_KEY;
6886 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6890 leaf = path->nodes[0];
6891 slot = path->slots[0];
6892 if (slot >= btrfs_header_nritems(leaf)) {
6893 ret = btrfs_next_leaf(root, path);
6898 leaf = path->nodes[0];
6899 slot = path->slots[0];
6902 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6903 if (found_key.objectid != search_key.objectid ||
6904 found_key.type != BTRFS_INODE_REF_KEY)
6910 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6911 struct btrfs_inode *inode,
6912 struct dentry *parent,
6913 struct btrfs_log_ctx *ctx)
6915 struct btrfs_root *root = inode->root;
6916 struct dentry *old_parent = NULL;
6917 struct super_block *sb = inode->vfs_inode.i_sb;
6921 if (!parent || d_really_is_negative(parent) ||
6925 inode = BTRFS_I(d_inode(parent));
6926 if (root != inode->root)
6929 if (inode->generation >= trans->transid &&
6930 need_log_inode(trans, inode)) {
6931 ret = btrfs_log_inode(trans, inode,
6932 LOG_INODE_EXISTS, ctx);
6936 if (IS_ROOT(parent))
6939 parent = dget_parent(parent);
6941 old_parent = parent;
6948 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6949 struct btrfs_inode *inode,
6950 struct dentry *parent,
6951 struct btrfs_log_ctx *ctx)
6953 struct btrfs_root *root = inode->root;
6954 const u64 ino = btrfs_ino(inode);
6955 struct btrfs_path *path;
6956 struct btrfs_key search_key;
6960 * For a single hard link case, go through a fast path that does not
6961 * need to iterate the fs/subvolume tree.
6963 if (inode->vfs_inode.i_nlink < 2)
6964 return log_new_ancestors_fast(trans, inode, parent, ctx);
6966 path = btrfs_alloc_path();
6970 search_key.objectid = ino;
6971 search_key.type = BTRFS_INODE_REF_KEY;
6972 search_key.offset = 0;
6974 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6981 struct extent_buffer *leaf = path->nodes[0];
6982 int slot = path->slots[0];
6983 struct btrfs_key found_key;
6985 if (slot >= btrfs_header_nritems(leaf)) {
6986 ret = btrfs_next_leaf(root, path);
6994 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6995 if (found_key.objectid != ino ||
6996 found_key.type > BTRFS_INODE_EXTREF_KEY)
7000 * Don't deal with extended references because they are rare
7001 * cases and too complex to deal with (we would need to keep
7002 * track of which subitem we are processing for each item in
7003 * this loop, etc). So just return some error to fallback to
7004 * a transaction commit.
7006 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
7012 * Logging ancestors needs to do more searches on the fs/subvol
7013 * tree, so it releases the path as needed to avoid deadlocks.
7014 * Keep track of the last inode ref key and resume from that key
7015 * after logging all new ancestors for the current hard link.
7017 memcpy(&search_key, &found_key, sizeof(search_key));
7019 ret = log_new_ancestors(trans, root, path, ctx);
7022 btrfs_release_path(path);
7027 btrfs_free_path(path);
7032 * helper function around btrfs_log_inode to make sure newly created
7033 * parent directories also end up in the log. A minimal inode and backref
7034 * only logging is done of any parent directories that are older than
7035 * the last committed transaction
7037 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
7038 struct btrfs_inode *inode,
7039 struct dentry *parent,
7041 struct btrfs_log_ctx *ctx)
7043 struct btrfs_root *root = inode->root;
7044 struct btrfs_fs_info *fs_info = root->fs_info;
7046 bool log_dentries = false;
7048 if (btrfs_test_opt(fs_info, NOTREELOG)) {
7049 ret = BTRFS_LOG_FORCE_COMMIT;
7053 if (btrfs_root_refs(&root->root_item) == 0) {
7054 ret = BTRFS_LOG_FORCE_COMMIT;
7059 * Skip already logged inodes or inodes corresponding to tmpfiles
7060 * (since logging them is pointless, a link count of 0 means they
7061 * will never be accessible).
7063 if ((btrfs_inode_in_log(inode, trans->transid) &&
7064 list_empty(&ctx->ordered_extents)) ||
7065 inode->vfs_inode.i_nlink == 0) {
7066 ret = BTRFS_NO_LOG_SYNC;
7070 ret = start_log_trans(trans, root, ctx);
7074 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
7079 * for regular files, if its inode is already on disk, we don't
7080 * have to worry about the parents at all. This is because
7081 * we can use the last_unlink_trans field to record renames
7082 * and other fun in this file.
7084 if (S_ISREG(inode->vfs_inode.i_mode) &&
7085 inode->generation < trans->transid &&
7086 inode->last_unlink_trans < trans->transid) {
7091 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
7092 log_dentries = true;
7095 * On unlink we must make sure all our current and old parent directory
7096 * inodes are fully logged. This is to prevent leaving dangling
7097 * directory index entries in directories that were our parents but are
7098 * not anymore. Not doing this results in old parent directory being
7099 * impossible to delete after log replay (rmdir will always fail with
7100 * error -ENOTEMPTY).
7106 * ln testdir/foo testdir/bar
7108 * unlink testdir/bar
7109 * xfs_io -c fsync testdir/foo
7111 * mount fs, triggers log replay
7113 * If we don't log the parent directory (testdir), after log replay the
7114 * directory still has an entry pointing to the file inode using the bar
7115 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
7116 * the file inode has a link count of 1.
7122 * ln foo testdir/foo2
7123 * ln foo testdir/foo3
7125 * unlink testdir/foo3
7126 * xfs_io -c fsync foo
7128 * mount fs, triggers log replay
7130 * Similar as the first example, after log replay the parent directory
7131 * testdir still has an entry pointing to the inode file with name foo3
7132 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
7133 * and has a link count of 2.
7135 if (inode->last_unlink_trans >= trans->transid) {
7136 ret = btrfs_log_all_parents(trans, inode, ctx);
7141 ret = log_all_new_ancestors(trans, inode, parent, ctx);
7146 ret = log_new_dir_dentries(trans, inode, ctx);
7151 btrfs_set_log_full_commit(trans);
7152 ret = BTRFS_LOG_FORCE_COMMIT;
7156 btrfs_remove_log_ctx(root, ctx);
7157 btrfs_end_log_trans(root);
7163 * it is not safe to log dentry if the chunk root has added new
7164 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
7165 * If this returns 1, you must commit the transaction to safely get your
7168 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
7169 struct dentry *dentry,
7170 struct btrfs_log_ctx *ctx)
7172 struct dentry *parent = dget_parent(dentry);
7175 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
7176 LOG_INODE_ALL, ctx);
7183 * should be called during mount to recover any replay any log trees
7186 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
7189 struct btrfs_path *path;
7190 struct btrfs_trans_handle *trans;
7191 struct btrfs_key key;
7192 struct btrfs_key found_key;
7193 struct btrfs_root *log;
7194 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
7195 struct walk_control wc = {
7196 .process_func = process_one_buffer,
7197 .stage = LOG_WALK_PIN_ONLY,
7200 path = btrfs_alloc_path();
7204 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7206 trans = btrfs_start_transaction(fs_info->tree_root, 0);
7207 if (IS_ERR(trans)) {
7208 ret = PTR_ERR(trans);
7215 ret = walk_log_tree(trans, log_root_tree, &wc);
7217 btrfs_abort_transaction(trans, ret);
7222 key.objectid = BTRFS_TREE_LOG_OBJECTID;
7223 key.offset = (u64)-1;
7224 key.type = BTRFS_ROOT_ITEM_KEY;
7227 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
7230 btrfs_abort_transaction(trans, ret);
7234 if (path->slots[0] == 0)
7238 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
7240 btrfs_release_path(path);
7241 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7244 log = btrfs_read_tree_root(log_root_tree, &found_key);
7247 btrfs_abort_transaction(trans, ret);
7251 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
7253 if (IS_ERR(wc.replay_dest)) {
7254 ret = PTR_ERR(wc.replay_dest);
7257 * We didn't find the subvol, likely because it was
7258 * deleted. This is ok, simply skip this log and go to
7261 * We need to exclude the root because we can't have
7262 * other log replays overwriting this log as we'll read
7263 * it back in a few more times. This will keep our
7264 * block from being modified, and we'll just bail for
7265 * each subsequent pass.
7268 ret = btrfs_pin_extent_for_log_replay(trans, log->node);
7269 btrfs_put_root(log);
7273 btrfs_abort_transaction(trans, ret);
7277 wc.replay_dest->log_root = log;
7278 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
7280 /* The loop needs to continue due to the root refs */
7281 btrfs_abort_transaction(trans, ret);
7283 ret = walk_log_tree(trans, log, &wc);
7285 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7286 ret = fixup_inode_link_counts(trans, wc.replay_dest,
7289 btrfs_abort_transaction(trans, ret);
7292 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
7293 struct btrfs_root *root = wc.replay_dest;
7295 btrfs_release_path(path);
7298 * We have just replayed everything, and the highest
7299 * objectid of fs roots probably has changed in case
7300 * some inode_item's got replayed.
7302 * root->objectid_mutex is not acquired as log replay
7303 * could only happen during mount.
7305 ret = btrfs_init_root_free_objectid(root);
7307 btrfs_abort_transaction(trans, ret);
7310 wc.replay_dest->log_root = NULL;
7311 btrfs_put_root(wc.replay_dest);
7312 btrfs_put_root(log);
7317 if (found_key.offset == 0)
7319 key.offset = found_key.offset - 1;
7321 btrfs_release_path(path);
7323 /* step one is to pin it all, step two is to replay just inodes */
7326 wc.process_func = replay_one_buffer;
7327 wc.stage = LOG_WALK_REPLAY_INODES;
7330 /* step three is to replay everything */
7331 if (wc.stage < LOG_WALK_REPLAY_ALL) {
7336 btrfs_free_path(path);
7338 /* step 4: commit the transaction, which also unpins the blocks */
7339 ret = btrfs_commit_transaction(trans);
7343 log_root_tree->log_root = NULL;
7344 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7345 btrfs_put_root(log_root_tree);
7350 btrfs_end_transaction(wc.trans);
7351 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
7352 btrfs_free_path(path);
7357 * there are some corner cases where we want to force a full
7358 * commit instead of allowing a directory to be logged.
7360 * They revolve around files there were unlinked from the directory, and
7361 * this function updates the parent directory so that a full commit is
7362 * properly done if it is fsync'd later after the unlinks are done.
7364 * Must be called before the unlink operations (updates to the subvolume tree,
7365 * inodes, etc) are done.
7367 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
7368 struct btrfs_inode *dir, struct btrfs_inode *inode,
7372 * when we're logging a file, if it hasn't been renamed
7373 * or unlinked, and its inode is fully committed on disk,
7374 * we don't have to worry about walking up the directory chain
7375 * to log its parents.
7377 * So, we use the last_unlink_trans field to put this transid
7378 * into the file. When the file is logged we check it and
7379 * don't log the parents if the file is fully on disk.
7381 mutex_lock(&inode->log_mutex);
7382 inode->last_unlink_trans = trans->transid;
7383 mutex_unlock(&inode->log_mutex);
7389 * If this directory was already logged, any new names will be logged
7390 * with btrfs_log_new_name() and old names will be deleted from the log
7391 * tree with btrfs_del_dir_entries_in_log() or with
7392 * btrfs_del_inode_ref_in_log().
7394 if (inode_logged(trans, dir, NULL) == 1)
7398 * If the inode we're about to unlink was logged before, the log will be
7399 * properly updated with the new name with btrfs_log_new_name() and the
7400 * old name removed with btrfs_del_dir_entries_in_log() or with
7401 * btrfs_del_inode_ref_in_log().
7403 if (inode_logged(trans, inode, NULL) == 1)
7407 * when renaming files across directories, if the directory
7408 * there we're unlinking from gets fsync'd later on, there's
7409 * no way to find the destination directory later and fsync it
7410 * properly. So, we have to be conservative and force commits
7411 * so the new name gets discovered.
7413 mutex_lock(&dir->log_mutex);
7414 dir->last_unlink_trans = trans->transid;
7415 mutex_unlock(&dir->log_mutex);
7419 * Make sure that if someone attempts to fsync the parent directory of a deleted
7420 * snapshot, it ends up triggering a transaction commit. This is to guarantee
7421 * that after replaying the log tree of the parent directory's root we will not
7422 * see the snapshot anymore and at log replay time we will not see any log tree
7423 * corresponding to the deleted snapshot's root, which could lead to replaying
7424 * it after replaying the log tree of the parent directory (which would replay
7425 * the snapshot delete operation).
7427 * Must be called before the actual snapshot destroy operation (updates to the
7428 * parent root and tree of tree roots trees, etc) are done.
7430 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
7431 struct btrfs_inode *dir)
7433 mutex_lock(&dir->log_mutex);
7434 dir->last_unlink_trans = trans->transid;
7435 mutex_unlock(&dir->log_mutex);
7439 * Update the log after adding a new name for an inode.
7441 * @trans: Transaction handle.
7442 * @old_dentry: The dentry associated with the old name and the old
7444 * @old_dir: The inode of the previous parent directory for the case
7445 * of a rename. For a link operation, it must be NULL.
7446 * @old_dir_index: The index number associated with the old name, meaningful
7447 * only for rename operations (when @old_dir is not NULL).
7448 * Ignored for link operations.
7449 * @parent: The dentry associated with the directory under which the
7450 * new name is located.
7452 * Call this after adding a new name for an inode, as a result of a link or
7453 * rename operation, and it will properly update the log to reflect the new name.
7455 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
7456 struct dentry *old_dentry, struct btrfs_inode *old_dir,
7457 u64 old_dir_index, struct dentry *parent)
7459 struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry));
7460 struct btrfs_root *root = inode->root;
7461 struct btrfs_log_ctx ctx;
7462 bool log_pinned = false;
7466 * this will force the logging code to walk the dentry chain
7469 if (!S_ISDIR(inode->vfs_inode.i_mode))
7470 inode->last_unlink_trans = trans->transid;
7473 * if this inode hasn't been logged and directory we're renaming it
7474 * from hasn't been logged, we don't need to log it
7476 ret = inode_logged(trans, inode, NULL);
7479 } else if (ret == 0) {
7483 * If the inode was not logged and we are doing a rename (old_dir is not
7484 * NULL), check if old_dir was logged - if it was not we can return and
7487 ret = inode_logged(trans, old_dir, NULL);
7496 * If we are doing a rename (old_dir is not NULL) from a directory that
7497 * was previously logged, make sure that on log replay we get the old
7498 * dir entry deleted. This is needed because we will also log the new
7499 * name of the renamed inode, so we need to make sure that after log
7500 * replay we don't end up with both the new and old dir entries existing.
7502 if (old_dir && old_dir->logged_trans == trans->transid) {
7503 struct btrfs_root *log = old_dir->root->log_root;
7504 struct btrfs_path *path;
7505 struct fscrypt_name fname;
7507 ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX);
7509 ret = fscrypt_setup_filename(&old_dir->vfs_inode,
7510 &old_dentry->d_name, 0, &fname);
7514 * We have two inodes to update in the log, the old directory and
7515 * the inode that got renamed, so we must pin the log to prevent
7516 * anyone from syncing the log until we have updated both inodes
7519 ret = join_running_log_trans(root);
7521 * At least one of the inodes was logged before, so this should
7522 * not fail, but if it does, it's not serious, just bail out and
7523 * mark the log for a full commit.
7525 if (WARN_ON_ONCE(ret < 0)) {
7526 fscrypt_free_filename(&fname);
7532 path = btrfs_alloc_path();
7535 fscrypt_free_filename(&fname);
7540 * Other concurrent task might be logging the old directory,
7541 * as it can be triggered when logging other inode that had or
7542 * still has a dentry in the old directory. We lock the old
7543 * directory's log_mutex to ensure the deletion of the old
7544 * name is persisted, because during directory logging we
7545 * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of
7546 * the old name's dir index item is in the delayed items, so
7547 * it could be missed by an in progress directory logging.
7549 mutex_lock(&old_dir->log_mutex);
7550 ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir),
7551 &fname.disk_name, old_dir_index);
7554 * The dentry does not exist in the log, so record its
7557 btrfs_release_path(path);
7558 ret = insert_dir_log_key(trans, log, path,
7560 old_dir_index, old_dir_index);
7562 mutex_unlock(&old_dir->log_mutex);
7564 btrfs_free_path(path);
7565 fscrypt_free_filename(&fname);
7570 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
7571 ctx.logging_new_name = true;
7572 btrfs_init_log_ctx_scratch_eb(&ctx);
7574 * We don't care about the return value. If we fail to log the new name
7575 * then we know the next attempt to sync the log will fallback to a full
7576 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
7577 * we don't need to worry about getting a log committed that has an
7578 * inconsistent state after a rename operation.
7580 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
7581 free_extent_buffer(ctx.scratch_eb);
7582 ASSERT(list_empty(&ctx.conflict_inodes));
7585 * If an error happened mark the log for a full commit because it's not
7586 * consistent and up to date or we couldn't find out if one of the
7587 * inodes was logged before in this transaction. Do it before unpinning
7588 * the log, to avoid any races with someone else trying to commit it.
7591 btrfs_set_log_full_commit(trans);
7593 btrfs_end_log_trans(root);