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>
15 #include "print-tree.h"
17 #include "compression.h"
19 #include "inode-map.h"
21 /* magic values for the inode_only field in btrfs_log_inode:
23 * LOG_INODE_ALL means to log everything
24 * LOG_INODE_EXISTS means to log just enough to recreate the inode
27 #define LOG_INODE_ALL 0
28 #define LOG_INODE_EXISTS 1
29 #define LOG_OTHER_INODE 2
30 #define LOG_OTHER_INODE_ALL 3
33 * directory trouble cases
35 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
36 * log, we must force a full commit before doing an fsync of the directory
37 * where the unlink was done.
38 * ---> record transid of last unlink/rename per directory
42 * rename foo/some_dir foo2/some_dir
44 * fsync foo/some_dir/some_file
46 * The fsync above will unlink the original some_dir without recording
47 * it in its new location (foo2). After a crash, some_dir will be gone
48 * unless the fsync of some_file forces a full commit
50 * 2) we must log any new names for any file or dir that is in the fsync
51 * log. ---> check inode while renaming/linking.
53 * 2a) we must log any new names for any file or dir during rename
54 * when the directory they are being removed from was logged.
55 * ---> check inode and old parent dir during rename
57 * 2a is actually the more important variant. With the extra logging
58 * a crash might unlink the old name without recreating the new one
60 * 3) after a crash, we must go through any directories with a link count
61 * of zero and redo the rm -rf
68 * The directory f1 was fully removed from the FS, but fsync was never
69 * called on f1, only its parent dir. After a crash the rm -rf must
70 * be replayed. This must be able to recurse down the entire
71 * directory tree. The inode link count fixup code takes care of the
76 * stages for the tree walking. The first
77 * stage (0) is to only pin down the blocks we find
78 * the second stage (1) is to make sure that all the inodes
79 * we find in the log are created in the subvolume.
81 * The last stage is to deal with directories and links and extents
82 * and all the other fun semantics
84 #define LOG_WALK_PIN_ONLY 0
85 #define LOG_WALK_REPLAY_INODES 1
86 #define LOG_WALK_REPLAY_DIR_INDEX 2
87 #define LOG_WALK_REPLAY_ALL 3
89 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root, struct btrfs_inode *inode,
94 struct btrfs_log_ctx *ctx);
95 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 struct btrfs_path *path, u64 objectid);
98 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
99 struct btrfs_root *root,
100 struct btrfs_root *log,
101 struct btrfs_path *path,
102 u64 dirid, int del_all);
105 * tree logging is a special write ahead log used to make sure that
106 * fsyncs and O_SYNCs can happen without doing full tree commits.
108 * Full tree commits are expensive because they require commonly
109 * modified blocks to be recowed, creating many dirty pages in the
110 * extent tree an 4x-6x higher write load than ext3.
112 * Instead of doing a tree commit on every fsync, we use the
113 * key ranges and transaction ids to find items for a given file or directory
114 * that have changed in this transaction. Those items are copied into
115 * a special tree (one per subvolume root), that tree is written to disk
116 * and then the fsync is considered complete.
118 * After a crash, items are copied out of the log-tree back into the
119 * subvolume tree. Any file data extents found are recorded in the extent
120 * allocation tree, and the log-tree freed.
122 * The log tree is read three times, once to pin down all the extents it is
123 * using in ram and once, once to create all the inodes logged in the tree
124 * and once to do all the other items.
128 * start a sub transaction and setup the log tree
129 * this increments the log tree writer count to make the people
130 * syncing the tree wait for us to finish
132 static int start_log_trans(struct btrfs_trans_handle *trans,
133 struct btrfs_root *root,
134 struct btrfs_log_ctx *ctx)
136 struct btrfs_fs_info *fs_info = root->fs_info;
139 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
142 if (btrfs_need_log_full_commit(trans)) {
147 if (!root->log_start_pid) {
148 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
149 root->log_start_pid = current->pid;
150 } else if (root->log_start_pid != current->pid) {
151 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
154 mutex_lock(&fs_info->tree_log_mutex);
155 if (!fs_info->log_root_tree)
156 ret = btrfs_init_log_root_tree(trans, fs_info);
157 mutex_unlock(&fs_info->tree_log_mutex);
161 ret = btrfs_add_log_tree(trans, root);
165 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
166 root->log_start_pid = current->pid;
169 atomic_inc(&root->log_batch);
170 atomic_inc(&root->log_writers);
172 int index = root->log_transid % 2;
173 list_add_tail(&ctx->list, &root->log_ctxs[index]);
174 ctx->log_transid = root->log_transid;
178 mutex_unlock(&root->log_mutex);
183 * returns 0 if there was a log transaction running and we were able
184 * to join, or returns -ENOENT if there were not transactions
187 static int join_running_log_trans(struct btrfs_root *root)
195 mutex_lock(&root->log_mutex);
196 if (root->log_root) {
198 atomic_inc(&root->log_writers);
200 mutex_unlock(&root->log_mutex);
205 * This either makes the current running log transaction wait
206 * until you call btrfs_end_log_trans() or it makes any future
207 * log transactions wait until you call btrfs_end_log_trans()
209 void btrfs_pin_log_trans(struct btrfs_root *root)
211 mutex_lock(&root->log_mutex);
212 atomic_inc(&root->log_writers);
213 mutex_unlock(&root->log_mutex);
217 * indicate we're done making changes to the log tree
218 * and wake up anyone waiting to do a sync
220 void btrfs_end_log_trans(struct btrfs_root *root)
222 if (atomic_dec_and_test(&root->log_writers)) {
223 /* atomic_dec_and_test implies a barrier */
224 cond_wake_up_nomb(&root->log_writer_wait);
228 static int btrfs_write_tree_block(struct extent_buffer *buf)
230 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
231 buf->start + buf->len - 1);
234 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
236 filemap_fdatawait_range(buf->pages[0]->mapping,
237 buf->start, buf->start + buf->len - 1);
241 * the walk control struct is used to pass state down the chain when
242 * processing the log tree. The stage field tells us which part
243 * of the log tree processing we are currently doing. The others
244 * are state fields used for that specific part
246 struct walk_control {
247 /* should we free the extent on disk when done? This is used
248 * at transaction commit time while freeing a log tree
252 /* should we write out the extent buffer? This is used
253 * while flushing the log tree to disk during a sync
257 /* should we wait for the extent buffer io to finish? Also used
258 * while flushing the log tree to disk for a sync
262 /* pin only walk, we record which extents on disk belong to the
267 /* what stage of the replay code we're currently in */
271 * Ignore any items from the inode currently being processed. Needs
272 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
273 * the LOG_WALK_REPLAY_INODES stage.
275 bool ignore_cur_inode;
277 /* the root we are currently replaying */
278 struct btrfs_root *replay_dest;
280 /* the trans handle for the current replay */
281 struct btrfs_trans_handle *trans;
283 /* the function that gets used to process blocks we find in the
284 * tree. Note the extent_buffer might not be up to date when it is
285 * passed in, and it must be checked or read if you need the data
288 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
289 struct walk_control *wc, u64 gen, int level);
293 * process_func used to pin down extents, write them or wait on them
295 static int process_one_buffer(struct btrfs_root *log,
296 struct extent_buffer *eb,
297 struct walk_control *wc, u64 gen, int level)
299 struct btrfs_fs_info *fs_info = log->fs_info;
303 * If this fs is mixed then we need to be able to process the leaves to
304 * pin down any logged extents, so we have to read the block.
306 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
307 ret = btrfs_read_buffer(eb, gen, level, NULL);
313 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
316 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
317 if (wc->pin && btrfs_header_level(eb) == 0)
318 ret = btrfs_exclude_logged_extents(eb);
320 btrfs_write_tree_block(eb);
322 btrfs_wait_tree_block_writeback(eb);
328 * Item overwrite used by replay and tree logging. eb, slot and key all refer
329 * to the src data we are copying out.
331 * root is the tree we are copying into, and path is a scratch
332 * path for use in this function (it should be released on entry and
333 * will be released on exit).
335 * If the key is already in the destination tree the existing item is
336 * overwritten. If the existing item isn't big enough, it is extended.
337 * If it is too large, it is truncated.
339 * If the key isn't in the destination yet, a new item is inserted.
341 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
342 struct btrfs_root *root,
343 struct btrfs_path *path,
344 struct extent_buffer *eb, int slot,
345 struct btrfs_key *key)
347 struct btrfs_fs_info *fs_info = root->fs_info;
350 u64 saved_i_size = 0;
351 int save_old_i_size = 0;
352 unsigned long src_ptr;
353 unsigned long dst_ptr;
354 int overwrite_root = 0;
355 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
357 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
360 item_size = btrfs_item_size_nr(eb, slot);
361 src_ptr = btrfs_item_ptr_offset(eb, slot);
363 /* look for the key in the destination tree */
364 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
371 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
373 if (dst_size != item_size)
376 if (item_size == 0) {
377 btrfs_release_path(path);
380 dst_copy = kmalloc(item_size, GFP_NOFS);
381 src_copy = kmalloc(item_size, GFP_NOFS);
382 if (!dst_copy || !src_copy) {
383 btrfs_release_path(path);
389 read_extent_buffer(eb, src_copy, src_ptr, item_size);
391 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
392 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
394 ret = memcmp(dst_copy, src_copy, item_size);
399 * they have the same contents, just return, this saves
400 * us from cowing blocks in the destination tree and doing
401 * extra writes that may not have been done by a previous
405 btrfs_release_path(path);
410 * We need to load the old nbytes into the inode so when we
411 * replay the extents we've logged we get the right nbytes.
414 struct btrfs_inode_item *item;
418 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
419 struct btrfs_inode_item);
420 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
421 item = btrfs_item_ptr(eb, slot,
422 struct btrfs_inode_item);
423 btrfs_set_inode_nbytes(eb, item, nbytes);
426 * If this is a directory we need to reset the i_size to
427 * 0 so that we can set it up properly when replaying
428 * the rest of the items in this log.
430 mode = btrfs_inode_mode(eb, item);
432 btrfs_set_inode_size(eb, item, 0);
434 } else if (inode_item) {
435 struct btrfs_inode_item *item;
439 * New inode, set nbytes to 0 so that the nbytes comes out
440 * properly when we replay the extents.
442 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
443 btrfs_set_inode_nbytes(eb, item, 0);
446 * If this is a directory we need to reset the i_size to 0 so
447 * that we can set it up properly when replaying the rest of
448 * the items in this log.
450 mode = btrfs_inode_mode(eb, item);
452 btrfs_set_inode_size(eb, item, 0);
455 btrfs_release_path(path);
456 /* try to insert the key into the destination tree */
457 path->skip_release_on_error = 1;
458 ret = btrfs_insert_empty_item(trans, root, path,
460 path->skip_release_on_error = 0;
462 /* make sure any existing item is the correct size */
463 if (ret == -EEXIST || ret == -EOVERFLOW) {
465 found_size = btrfs_item_size_nr(path->nodes[0],
467 if (found_size > item_size)
468 btrfs_truncate_item(fs_info, path, item_size, 1);
469 else if (found_size < item_size)
470 btrfs_extend_item(fs_info, path,
471 item_size - found_size);
475 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
478 /* don't overwrite an existing inode if the generation number
479 * was logged as zero. This is done when the tree logging code
480 * is just logging an inode to make sure it exists after recovery.
482 * Also, don't overwrite i_size on directories during replay.
483 * log replay inserts and removes directory items based on the
484 * state of the tree found in the subvolume, and i_size is modified
487 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
488 struct btrfs_inode_item *src_item;
489 struct btrfs_inode_item *dst_item;
491 src_item = (struct btrfs_inode_item *)src_ptr;
492 dst_item = (struct btrfs_inode_item *)dst_ptr;
494 if (btrfs_inode_generation(eb, src_item) == 0) {
495 struct extent_buffer *dst_eb = path->nodes[0];
496 const u64 ino_size = btrfs_inode_size(eb, src_item);
499 * For regular files an ino_size == 0 is used only when
500 * logging that an inode exists, as part of a directory
501 * fsync, and the inode wasn't fsynced before. In this
502 * case don't set the size of the inode in the fs/subvol
503 * tree, otherwise we would be throwing valid data away.
505 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
506 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
508 struct btrfs_map_token token;
510 btrfs_init_map_token(&token);
511 btrfs_set_token_inode_size(dst_eb, dst_item,
517 if (overwrite_root &&
518 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
519 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
521 saved_i_size = btrfs_inode_size(path->nodes[0],
526 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
529 if (save_old_i_size) {
530 struct btrfs_inode_item *dst_item;
531 dst_item = (struct btrfs_inode_item *)dst_ptr;
532 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
535 /* make sure the generation is filled in */
536 if (key->type == BTRFS_INODE_ITEM_KEY) {
537 struct btrfs_inode_item *dst_item;
538 dst_item = (struct btrfs_inode_item *)dst_ptr;
539 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
540 btrfs_set_inode_generation(path->nodes[0], dst_item,
545 btrfs_mark_buffer_dirty(path->nodes[0]);
546 btrfs_release_path(path);
551 * simple helper to read an inode off the disk from a given root
552 * This can only be called for subvolume roots and not for the log
554 static noinline struct inode *read_one_inode(struct btrfs_root *root,
557 struct btrfs_key key;
560 key.objectid = objectid;
561 key.type = BTRFS_INODE_ITEM_KEY;
563 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
569 /* replays a single extent in 'eb' at 'slot' with 'key' into the
570 * subvolume 'root'. path is released on entry and should be released
573 * extents in the log tree have not been allocated out of the extent
574 * tree yet. So, this completes the allocation, taking a reference
575 * as required if the extent already exists or creating a new extent
576 * if it isn't in the extent allocation tree yet.
578 * The extent is inserted into the file, dropping any existing extents
579 * from the file that overlap the new one.
581 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
582 struct btrfs_root *root,
583 struct btrfs_path *path,
584 struct extent_buffer *eb, int slot,
585 struct btrfs_key *key)
587 struct btrfs_fs_info *fs_info = root->fs_info;
590 u64 start = key->offset;
592 struct btrfs_file_extent_item *item;
593 struct inode *inode = NULL;
597 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
598 found_type = btrfs_file_extent_type(eb, item);
600 if (found_type == BTRFS_FILE_EXTENT_REG ||
601 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
602 nbytes = btrfs_file_extent_num_bytes(eb, item);
603 extent_end = start + nbytes;
606 * We don't add to the inodes nbytes if we are prealloc or a
609 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
611 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
612 size = btrfs_file_extent_ram_bytes(eb, item);
613 nbytes = btrfs_file_extent_ram_bytes(eb, item);
614 extent_end = ALIGN(start + size,
615 fs_info->sectorsize);
621 inode = read_one_inode(root, key->objectid);
628 * first check to see if we already have this extent in the
629 * file. This must be done before the btrfs_drop_extents run
630 * so we don't try to drop this extent.
632 ret = btrfs_lookup_file_extent(trans, root, path,
633 btrfs_ino(BTRFS_I(inode)), start, 0);
636 (found_type == BTRFS_FILE_EXTENT_REG ||
637 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
638 struct btrfs_file_extent_item cmp1;
639 struct btrfs_file_extent_item cmp2;
640 struct btrfs_file_extent_item *existing;
641 struct extent_buffer *leaf;
643 leaf = path->nodes[0];
644 existing = btrfs_item_ptr(leaf, path->slots[0],
645 struct btrfs_file_extent_item);
647 read_extent_buffer(eb, &cmp1, (unsigned long)item,
649 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
653 * we already have a pointer to this exact extent,
654 * we don't have to do anything
656 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
657 btrfs_release_path(path);
661 btrfs_release_path(path);
663 /* drop any overlapping extents */
664 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
668 if (found_type == BTRFS_FILE_EXTENT_REG ||
669 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
671 unsigned long dest_offset;
672 struct btrfs_key ins;
674 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
675 btrfs_fs_incompat(fs_info, NO_HOLES))
678 ret = btrfs_insert_empty_item(trans, root, path, key,
682 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
684 copy_extent_buffer(path->nodes[0], eb, dest_offset,
685 (unsigned long)item, sizeof(*item));
687 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
688 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
689 ins.type = BTRFS_EXTENT_ITEM_KEY;
690 offset = key->offset - btrfs_file_extent_offset(eb, item);
693 * Manually record dirty extent, as here we did a shallow
694 * file extent item copy and skip normal backref update,
695 * but modifying extent tree all by ourselves.
696 * So need to manually record dirty extent for qgroup,
697 * as the owner of the file extent changed from log tree
698 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
700 ret = btrfs_qgroup_trace_extent(trans,
701 btrfs_file_extent_disk_bytenr(eb, item),
702 btrfs_file_extent_disk_num_bytes(eb, item),
707 if (ins.objectid > 0) {
710 LIST_HEAD(ordered_sums);
712 * is this extent already allocated in the extent
713 * allocation tree? If so, just add a reference
715 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
718 ret = btrfs_inc_extent_ref(trans, root,
719 ins.objectid, ins.offset,
720 0, root->root_key.objectid,
721 key->objectid, offset);
726 * insert the extent pointer in the extent
729 ret = btrfs_alloc_logged_file_extent(trans,
730 root->root_key.objectid,
731 key->objectid, offset, &ins);
735 btrfs_release_path(path);
737 if (btrfs_file_extent_compression(eb, item)) {
738 csum_start = ins.objectid;
739 csum_end = csum_start + ins.offset;
741 csum_start = ins.objectid +
742 btrfs_file_extent_offset(eb, item);
743 csum_end = csum_start +
744 btrfs_file_extent_num_bytes(eb, item);
747 ret = btrfs_lookup_csums_range(root->log_root,
748 csum_start, csum_end - 1,
753 * Now delete all existing cums in the csum root that
754 * cover our range. We do this because we can have an
755 * extent that is completely referenced by one file
756 * extent item and partially referenced by another
757 * file extent item (like after using the clone or
758 * extent_same ioctls). In this case if we end up doing
759 * the replay of the one that partially references the
760 * extent first, and we do not do the csum deletion
761 * below, we can get 2 csum items in the csum tree that
762 * overlap each other. For example, imagine our log has
763 * the two following file extent items:
765 * key (257 EXTENT_DATA 409600)
766 * extent data disk byte 12845056 nr 102400
767 * extent data offset 20480 nr 20480 ram 102400
769 * key (257 EXTENT_DATA 819200)
770 * extent data disk byte 12845056 nr 102400
771 * extent data offset 0 nr 102400 ram 102400
773 * Where the second one fully references the 100K extent
774 * that starts at disk byte 12845056, and the log tree
775 * has a single csum item that covers the entire range
778 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
780 * After the first file extent item is replayed, the
781 * csum tree gets the following csum item:
783 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
785 * Which covers the 20K sub-range starting at offset 20K
786 * of our extent. Now when we replay the second file
787 * extent item, if we do not delete existing csum items
788 * that cover any of its blocks, we end up getting two
789 * csum items in our csum tree that overlap each other:
791 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
792 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
794 * Which is a problem, because after this anyone trying
795 * to lookup up for the checksum of any block of our
796 * extent starting at an offset of 40K or higher, will
797 * end up looking at the second csum item only, which
798 * does not contain the checksum for any block starting
799 * at offset 40K or higher of our extent.
801 while (!list_empty(&ordered_sums)) {
802 struct btrfs_ordered_sum *sums;
803 sums = list_entry(ordered_sums.next,
804 struct btrfs_ordered_sum,
807 ret = btrfs_del_csums(trans, fs_info,
811 ret = btrfs_csum_file_blocks(trans,
812 fs_info->csum_root, sums);
813 list_del(&sums->list);
819 btrfs_release_path(path);
821 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
822 /* inline extents are easy, we just overwrite them */
823 ret = overwrite_item(trans, root, path, eb, slot, key);
828 inode_add_bytes(inode, nbytes);
830 ret = btrfs_update_inode(trans, root, inode);
838 * when cleaning up conflicts between the directory names in the
839 * subvolume, directory names in the log and directory names in the
840 * inode back references, we may have to unlink inodes from directories.
842 * This is a helper function to do the unlink of a specific directory
845 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
846 struct btrfs_root *root,
847 struct btrfs_path *path,
848 struct btrfs_inode *dir,
849 struct btrfs_dir_item *di)
854 struct extent_buffer *leaf;
855 struct btrfs_key location;
858 leaf = path->nodes[0];
860 btrfs_dir_item_key_to_cpu(leaf, di, &location);
861 name_len = btrfs_dir_name_len(leaf, di);
862 name = kmalloc(name_len, GFP_NOFS);
866 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
867 btrfs_release_path(path);
869 inode = read_one_inode(root, location.objectid);
875 ret = link_to_fixup_dir(trans, root, path, location.objectid);
879 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
884 ret = btrfs_run_delayed_items(trans);
892 * helper function to see if a given name and sequence number found
893 * in an inode back reference are already in a directory and correctly
894 * point to this inode
896 static noinline int inode_in_dir(struct btrfs_root *root,
897 struct btrfs_path *path,
898 u64 dirid, u64 objectid, u64 index,
899 const char *name, int name_len)
901 struct btrfs_dir_item *di;
902 struct btrfs_key location;
905 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
906 index, name, name_len, 0);
907 if (di && !IS_ERR(di)) {
908 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
909 if (location.objectid != objectid)
913 btrfs_release_path(path);
915 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
916 if (di && !IS_ERR(di)) {
917 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
918 if (location.objectid != objectid)
924 btrfs_release_path(path);
929 * helper function to check a log tree for a named back reference in
930 * an inode. This is used to decide if a back reference that is
931 * found in the subvolume conflicts with what we find in the log.
933 * inode backreferences may have multiple refs in a single item,
934 * during replay we process one reference at a time, and we don't
935 * want to delete valid links to a file from the subvolume if that
936 * link is also in the log.
938 static noinline int backref_in_log(struct btrfs_root *log,
939 struct btrfs_key *key,
941 const char *name, int namelen)
943 struct btrfs_path *path;
944 struct btrfs_inode_ref *ref;
946 unsigned long ptr_end;
947 unsigned long name_ptr;
953 path = btrfs_alloc_path();
957 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
961 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
963 if (key->type == BTRFS_INODE_EXTREF_KEY) {
964 if (btrfs_find_name_in_ext_backref(path->nodes[0],
967 name, namelen, NULL))
973 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
974 ptr_end = ptr + item_size;
975 while (ptr < ptr_end) {
976 ref = (struct btrfs_inode_ref *)ptr;
977 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
978 if (found_name_len == namelen) {
979 name_ptr = (unsigned long)(ref + 1);
980 ret = memcmp_extent_buffer(path->nodes[0], name,
987 ptr = (unsigned long)(ref + 1) + found_name_len;
990 btrfs_free_path(path);
994 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
995 struct btrfs_root *root,
996 struct btrfs_path *path,
997 struct btrfs_root *log_root,
998 struct btrfs_inode *dir,
999 struct btrfs_inode *inode,
1000 u64 inode_objectid, u64 parent_objectid,
1001 u64 ref_index, char *name, int namelen,
1006 int victim_name_len;
1007 struct extent_buffer *leaf;
1008 struct btrfs_dir_item *di;
1009 struct btrfs_key search_key;
1010 struct btrfs_inode_extref *extref;
1013 /* Search old style refs */
1014 search_key.objectid = inode_objectid;
1015 search_key.type = BTRFS_INODE_REF_KEY;
1016 search_key.offset = parent_objectid;
1017 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1019 struct btrfs_inode_ref *victim_ref;
1021 unsigned long ptr_end;
1023 leaf = path->nodes[0];
1025 /* are we trying to overwrite a back ref for the root directory
1026 * if so, just jump out, we're done
1028 if (search_key.objectid == search_key.offset)
1031 /* check all the names in this back reference to see
1032 * if they are in the log. if so, we allow them to stay
1033 * otherwise they must be unlinked as a conflict
1035 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1036 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1037 while (ptr < ptr_end) {
1038 victim_ref = (struct btrfs_inode_ref *)ptr;
1039 victim_name_len = btrfs_inode_ref_name_len(leaf,
1041 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1045 read_extent_buffer(leaf, victim_name,
1046 (unsigned long)(victim_ref + 1),
1049 if (!backref_in_log(log_root, &search_key,
1053 inc_nlink(&inode->vfs_inode);
1054 btrfs_release_path(path);
1056 ret = btrfs_unlink_inode(trans, root, dir, inode,
1057 victim_name, victim_name_len);
1061 ret = btrfs_run_delayed_items(trans);
1069 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1073 * NOTE: we have searched root tree and checked the
1074 * corresponding ref, it does not need to check again.
1078 btrfs_release_path(path);
1080 /* Same search but for extended refs */
1081 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1082 inode_objectid, parent_objectid, 0,
1084 if (!IS_ERR_OR_NULL(extref)) {
1088 struct inode *victim_parent;
1090 leaf = path->nodes[0];
1092 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1093 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1095 while (cur_offset < item_size) {
1096 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1098 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1100 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1103 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1106 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1109 search_key.objectid = inode_objectid;
1110 search_key.type = BTRFS_INODE_EXTREF_KEY;
1111 search_key.offset = btrfs_extref_hash(parent_objectid,
1115 if (!backref_in_log(log_root, &search_key,
1116 parent_objectid, victim_name,
1119 victim_parent = read_one_inode(root,
1121 if (victim_parent) {
1122 inc_nlink(&inode->vfs_inode);
1123 btrfs_release_path(path);
1125 ret = btrfs_unlink_inode(trans, root,
1126 BTRFS_I(victim_parent),
1131 ret = btrfs_run_delayed_items(
1134 iput(victim_parent);
1143 cur_offset += victim_name_len + sizeof(*extref);
1147 btrfs_release_path(path);
1149 /* look for a conflicting sequence number */
1150 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1151 ref_index, name, namelen, 0);
1152 if (di && !IS_ERR(di)) {
1153 ret = drop_one_dir_item(trans, root, path, dir, di);
1157 btrfs_release_path(path);
1159 /* look for a conflicting name */
1160 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1162 if (di && !IS_ERR(di)) {
1163 ret = drop_one_dir_item(trans, root, path, dir, di);
1167 btrfs_release_path(path);
1172 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1173 u32 *namelen, char **name, u64 *index,
1174 u64 *parent_objectid)
1176 struct btrfs_inode_extref *extref;
1178 extref = (struct btrfs_inode_extref *)ref_ptr;
1180 *namelen = btrfs_inode_extref_name_len(eb, extref);
1181 *name = kmalloc(*namelen, GFP_NOFS);
1185 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1189 *index = btrfs_inode_extref_index(eb, extref);
1190 if (parent_objectid)
1191 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1196 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1197 u32 *namelen, char **name, u64 *index)
1199 struct btrfs_inode_ref *ref;
1201 ref = (struct btrfs_inode_ref *)ref_ptr;
1203 *namelen = btrfs_inode_ref_name_len(eb, ref);
1204 *name = kmalloc(*namelen, GFP_NOFS);
1208 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1211 *index = btrfs_inode_ref_index(eb, ref);
1217 * Take an inode reference item from the log tree and iterate all names from the
1218 * inode reference item in the subvolume tree with the same key (if it exists).
1219 * For any name that is not in the inode reference item from the log tree, do a
1220 * proper unlink of that name (that is, remove its entry from the inode
1221 * reference item and both dir index keys).
1223 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1224 struct btrfs_root *root,
1225 struct btrfs_path *path,
1226 struct btrfs_inode *inode,
1227 struct extent_buffer *log_eb,
1229 struct btrfs_key *key)
1232 unsigned long ref_ptr;
1233 unsigned long ref_end;
1234 struct extent_buffer *eb;
1237 btrfs_release_path(path);
1238 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1246 eb = path->nodes[0];
1247 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1248 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1249 while (ref_ptr < ref_end) {
1254 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1255 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1258 parent_id = key->offset;
1259 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1265 if (key->type == BTRFS_INODE_EXTREF_KEY)
1266 ret = btrfs_find_name_in_ext_backref(log_eb, log_slot,
1270 ret = btrfs_find_name_in_backref(log_eb, log_slot, name,
1276 btrfs_release_path(path);
1277 dir = read_one_inode(root, parent_id);
1283 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1284 inode, name, namelen);
1294 if (key->type == BTRFS_INODE_EXTREF_KEY)
1295 ref_ptr += sizeof(struct btrfs_inode_extref);
1297 ref_ptr += sizeof(struct btrfs_inode_ref);
1301 btrfs_release_path(path);
1305 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1306 const u8 ref_type, const char *name,
1309 struct btrfs_key key;
1310 struct btrfs_path *path;
1311 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1314 path = btrfs_alloc_path();
1318 key.objectid = btrfs_ino(BTRFS_I(inode));
1319 key.type = ref_type;
1320 if (key.type == BTRFS_INODE_REF_KEY)
1321 key.offset = parent_id;
1323 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1325 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1332 if (key.type == BTRFS_INODE_EXTREF_KEY)
1333 ret = btrfs_find_name_in_ext_backref(path->nodes[0],
1334 path->slots[0], parent_id,
1335 name, namelen, NULL);
1337 ret = btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1338 name, namelen, NULL);
1341 btrfs_free_path(path);
1345 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1346 struct inode *dir, struct inode *inode, const char *name,
1347 int namelen, u64 ref_index)
1349 struct btrfs_dir_item *dir_item;
1350 struct btrfs_key key;
1351 struct btrfs_path *path;
1352 struct inode *other_inode = NULL;
1355 path = btrfs_alloc_path();
1359 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1360 btrfs_ino(BTRFS_I(dir)),
1363 btrfs_release_path(path);
1365 } else if (IS_ERR(dir_item)) {
1366 ret = PTR_ERR(dir_item);
1371 * Our inode's dentry collides with the dentry of another inode which is
1372 * in the log but not yet processed since it has a higher inode number.
1373 * So delete that other dentry.
1375 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1376 btrfs_release_path(path);
1377 other_inode = read_one_inode(root, key.objectid);
1382 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1387 * If we dropped the link count to 0, bump it so that later the iput()
1388 * on the inode will not free it. We will fixup the link count later.
1390 if (other_inode->i_nlink == 0)
1391 inc_nlink(other_inode);
1393 ret = btrfs_run_delayed_items(trans);
1397 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1398 name, namelen, 0, ref_index);
1401 btrfs_free_path(path);
1407 * replay one inode back reference item found in the log tree.
1408 * eb, slot and key refer to the buffer and key found in the log tree.
1409 * root is the destination we are replaying into, and path is for temp
1410 * use by this function. (it should be released on return).
1412 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1413 struct btrfs_root *root,
1414 struct btrfs_root *log,
1415 struct btrfs_path *path,
1416 struct extent_buffer *eb, int slot,
1417 struct btrfs_key *key)
1419 struct inode *dir = NULL;
1420 struct inode *inode = NULL;
1421 unsigned long ref_ptr;
1422 unsigned long ref_end;
1426 int search_done = 0;
1427 int log_ref_ver = 0;
1428 u64 parent_objectid;
1431 int ref_struct_size;
1433 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1434 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1436 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1437 struct btrfs_inode_extref *r;
1439 ref_struct_size = sizeof(struct btrfs_inode_extref);
1441 r = (struct btrfs_inode_extref *)ref_ptr;
1442 parent_objectid = btrfs_inode_extref_parent(eb, r);
1444 ref_struct_size = sizeof(struct btrfs_inode_ref);
1445 parent_objectid = key->offset;
1447 inode_objectid = key->objectid;
1450 * it is possible that we didn't log all the parent directories
1451 * for a given inode. If we don't find the dir, just don't
1452 * copy the back ref in. The link count fixup code will take
1455 dir = read_one_inode(root, parent_objectid);
1461 inode = read_one_inode(root, inode_objectid);
1467 while (ref_ptr < ref_end) {
1469 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1470 &ref_index, &parent_objectid);
1472 * parent object can change from one array
1476 dir = read_one_inode(root, parent_objectid);
1482 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1488 /* if we already have a perfect match, we're done */
1489 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1490 btrfs_ino(BTRFS_I(inode)), ref_index,
1493 * look for a conflicting back reference in the
1494 * metadata. if we find one we have to unlink that name
1495 * of the file before we add our new link. Later on, we
1496 * overwrite any existing back reference, and we don't
1497 * want to create dangling pointers in the directory.
1501 ret = __add_inode_ref(trans, root, path, log,
1506 ref_index, name, namelen,
1516 * If a reference item already exists for this inode
1517 * with the same parent and name, but different index,
1518 * drop it and the corresponding directory index entries
1519 * from the parent before adding the new reference item
1520 * and dir index entries, otherwise we would fail with
1521 * -EEXIST returned from btrfs_add_link() below.
1523 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1526 ret = btrfs_unlink_inode(trans, root,
1531 * If we dropped the link count to 0, bump it so
1532 * that later the iput() on the inode will not
1533 * free it. We will fixup the link count later.
1535 if (!ret && inode->i_nlink == 0)
1541 /* insert our name */
1542 ret = add_link(trans, root, dir, inode, name, namelen,
1547 btrfs_update_inode(trans, root, inode);
1550 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1560 * Before we overwrite the inode reference item in the subvolume tree
1561 * with the item from the log tree, we must unlink all names from the
1562 * parent directory that are in the subvolume's tree inode reference
1563 * item, otherwise we end up with an inconsistent subvolume tree where
1564 * dir index entries exist for a name but there is no inode reference
1565 * item with the same name.
1567 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1572 /* finally write the back reference in the inode */
1573 ret = overwrite_item(trans, root, path, eb, slot, key);
1575 btrfs_release_path(path);
1582 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1583 struct btrfs_root *root, u64 ino)
1587 ret = btrfs_insert_orphan_item(trans, root, ino);
1594 static int count_inode_extrefs(struct btrfs_root *root,
1595 struct btrfs_inode *inode, struct btrfs_path *path)
1599 unsigned int nlink = 0;
1602 u64 inode_objectid = btrfs_ino(inode);
1605 struct btrfs_inode_extref *extref;
1606 struct extent_buffer *leaf;
1609 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1614 leaf = path->nodes[0];
1615 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1616 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1619 while (cur_offset < item_size) {
1620 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1621 name_len = btrfs_inode_extref_name_len(leaf, extref);
1625 cur_offset += name_len + sizeof(*extref);
1629 btrfs_release_path(path);
1631 btrfs_release_path(path);
1633 if (ret < 0 && ret != -ENOENT)
1638 static int count_inode_refs(struct btrfs_root *root,
1639 struct btrfs_inode *inode, struct btrfs_path *path)
1642 struct btrfs_key key;
1643 unsigned int nlink = 0;
1645 unsigned long ptr_end;
1647 u64 ino = btrfs_ino(inode);
1650 key.type = BTRFS_INODE_REF_KEY;
1651 key.offset = (u64)-1;
1654 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1658 if (path->slots[0] == 0)
1663 btrfs_item_key_to_cpu(path->nodes[0], &key,
1665 if (key.objectid != ino ||
1666 key.type != BTRFS_INODE_REF_KEY)
1668 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1669 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1671 while (ptr < ptr_end) {
1672 struct btrfs_inode_ref *ref;
1674 ref = (struct btrfs_inode_ref *)ptr;
1675 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1677 ptr = (unsigned long)(ref + 1) + name_len;
1681 if (key.offset == 0)
1683 if (path->slots[0] > 0) {
1688 btrfs_release_path(path);
1690 btrfs_release_path(path);
1696 * There are a few corners where the link count of the file can't
1697 * be properly maintained during replay. So, instead of adding
1698 * lots of complexity to the log code, we just scan the backrefs
1699 * for any file that has been through replay.
1701 * The scan will update the link count on the inode to reflect the
1702 * number of back refs found. If it goes down to zero, the iput
1703 * will free the inode.
1705 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1706 struct btrfs_root *root,
1707 struct inode *inode)
1709 struct btrfs_path *path;
1712 u64 ino = btrfs_ino(BTRFS_I(inode));
1714 path = btrfs_alloc_path();
1718 ret = count_inode_refs(root, BTRFS_I(inode), path);
1724 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1732 if (nlink != inode->i_nlink) {
1733 set_nlink(inode, nlink);
1734 btrfs_update_inode(trans, root, inode);
1736 BTRFS_I(inode)->index_cnt = (u64)-1;
1738 if (inode->i_nlink == 0) {
1739 if (S_ISDIR(inode->i_mode)) {
1740 ret = replay_dir_deletes(trans, root, NULL, path,
1745 ret = insert_orphan_item(trans, root, ino);
1749 btrfs_free_path(path);
1753 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1754 struct btrfs_root *root,
1755 struct btrfs_path *path)
1758 struct btrfs_key key;
1759 struct inode *inode;
1761 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1762 key.type = BTRFS_ORPHAN_ITEM_KEY;
1763 key.offset = (u64)-1;
1765 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1770 if (path->slots[0] == 0)
1775 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1776 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1777 key.type != BTRFS_ORPHAN_ITEM_KEY)
1780 ret = btrfs_del_item(trans, root, path);
1784 btrfs_release_path(path);
1785 inode = read_one_inode(root, key.offset);
1789 ret = fixup_inode_link_count(trans, root, inode);
1795 * fixup on a directory may create new entries,
1796 * make sure we always look for the highset possible
1799 key.offset = (u64)-1;
1803 btrfs_release_path(path);
1809 * record a given inode in the fixup dir so we can check its link
1810 * count when replay is done. The link count is incremented here
1811 * so the inode won't go away until we check it
1813 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1814 struct btrfs_root *root,
1815 struct btrfs_path *path,
1818 struct btrfs_key key;
1820 struct inode *inode;
1822 inode = read_one_inode(root, objectid);
1826 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1827 key.type = BTRFS_ORPHAN_ITEM_KEY;
1828 key.offset = objectid;
1830 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1832 btrfs_release_path(path);
1834 if (!inode->i_nlink)
1835 set_nlink(inode, 1);
1838 ret = btrfs_update_inode(trans, root, inode);
1839 } else if (ret == -EEXIST) {
1842 BUG(); /* Logic Error */
1850 * when replaying the log for a directory, we only insert names
1851 * for inodes that actually exist. This means an fsync on a directory
1852 * does not implicitly fsync all the new files in it
1854 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1855 struct btrfs_root *root,
1856 u64 dirid, u64 index,
1857 char *name, int name_len,
1858 struct btrfs_key *location)
1860 struct inode *inode;
1864 inode = read_one_inode(root, location->objectid);
1868 dir = read_one_inode(root, dirid);
1874 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1875 name_len, 1, index);
1877 /* FIXME, put inode into FIXUP list */
1885 * Return true if an inode reference exists in the log for the given name,
1886 * inode and parent inode.
1888 static bool name_in_log_ref(struct btrfs_root *log_root,
1889 const char *name, const int name_len,
1890 const u64 dirid, const u64 ino)
1892 struct btrfs_key search_key;
1894 search_key.objectid = ino;
1895 search_key.type = BTRFS_INODE_REF_KEY;
1896 search_key.offset = dirid;
1897 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1900 search_key.type = BTRFS_INODE_EXTREF_KEY;
1901 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1902 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1909 * take a single entry in a log directory item and replay it into
1912 * if a conflicting item exists in the subdirectory already,
1913 * the inode it points to is unlinked and put into the link count
1916 * If a name from the log points to a file or directory that does
1917 * not exist in the FS, it is skipped. fsyncs on directories
1918 * do not force down inodes inside that directory, just changes to the
1919 * names or unlinks in a directory.
1921 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1922 * non-existing inode) and 1 if the name was replayed.
1924 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1925 struct btrfs_root *root,
1926 struct btrfs_path *path,
1927 struct extent_buffer *eb,
1928 struct btrfs_dir_item *di,
1929 struct btrfs_key *key)
1933 struct btrfs_dir_item *dst_di;
1934 struct btrfs_key found_key;
1935 struct btrfs_key log_key;
1940 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1941 bool name_added = false;
1943 dir = read_one_inode(root, key->objectid);
1947 name_len = btrfs_dir_name_len(eb, di);
1948 name = kmalloc(name_len, GFP_NOFS);
1954 log_type = btrfs_dir_type(eb, di);
1955 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1958 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1959 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1964 btrfs_release_path(path);
1966 if (key->type == BTRFS_DIR_ITEM_KEY) {
1967 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1969 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1970 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1979 if (IS_ERR_OR_NULL(dst_di)) {
1980 /* we need a sequence number to insert, so we only
1981 * do inserts for the BTRFS_DIR_INDEX_KEY types
1983 if (key->type != BTRFS_DIR_INDEX_KEY)
1988 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1989 /* the existing item matches the logged item */
1990 if (found_key.objectid == log_key.objectid &&
1991 found_key.type == log_key.type &&
1992 found_key.offset == log_key.offset &&
1993 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1994 update_size = false;
1999 * don't drop the conflicting directory entry if the inode
2000 * for the new entry doesn't exist
2005 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2009 if (key->type == BTRFS_DIR_INDEX_KEY)
2012 btrfs_release_path(path);
2013 if (!ret && update_size) {
2014 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2015 ret = btrfs_update_inode(trans, root, dir);
2019 if (!ret && name_added)
2024 if (name_in_log_ref(root->log_root, name, name_len,
2025 key->objectid, log_key.objectid)) {
2026 /* The dentry will be added later. */
2028 update_size = false;
2031 btrfs_release_path(path);
2032 ret = insert_one_name(trans, root, key->objectid, key->offset,
2033 name, name_len, &log_key);
2034 if (ret && ret != -ENOENT && ret != -EEXIST)
2038 update_size = false;
2044 * find all the names in a directory item and reconcile them into
2045 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2046 * one name in a directory item, but the same code gets used for
2047 * both directory index types
2049 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2050 struct btrfs_root *root,
2051 struct btrfs_path *path,
2052 struct extent_buffer *eb, int slot,
2053 struct btrfs_key *key)
2056 u32 item_size = btrfs_item_size_nr(eb, slot);
2057 struct btrfs_dir_item *di;
2060 unsigned long ptr_end;
2061 struct btrfs_path *fixup_path = NULL;
2063 ptr = btrfs_item_ptr_offset(eb, slot);
2064 ptr_end = ptr + item_size;
2065 while (ptr < ptr_end) {
2066 di = (struct btrfs_dir_item *)ptr;
2067 name_len = btrfs_dir_name_len(eb, di);
2068 ret = replay_one_name(trans, root, path, eb, di, key);
2071 ptr = (unsigned long)(di + 1);
2075 * If this entry refers to a non-directory (directories can not
2076 * have a link count > 1) and it was added in the transaction
2077 * that was not committed, make sure we fixup the link count of
2078 * the inode it the entry points to. Otherwise something like
2079 * the following would result in a directory pointing to an
2080 * inode with a wrong link that does not account for this dir
2088 * ln testdir/bar testdir/bar_link
2089 * ln testdir/foo testdir/foo_link
2090 * xfs_io -c "fsync" testdir/bar
2094 * mount fs, log replay happens
2096 * File foo would remain with a link count of 1 when it has two
2097 * entries pointing to it in the directory testdir. This would
2098 * make it impossible to ever delete the parent directory has
2099 * it would result in stale dentries that can never be deleted.
2101 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2102 struct btrfs_key di_key;
2105 fixup_path = btrfs_alloc_path();
2112 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2113 ret = link_to_fixup_dir(trans, root, fixup_path,
2120 btrfs_free_path(fixup_path);
2125 * directory replay has two parts. There are the standard directory
2126 * items in the log copied from the subvolume, and range items
2127 * created in the log while the subvolume was logged.
2129 * The range items tell us which parts of the key space the log
2130 * is authoritative for. During replay, if a key in the subvolume
2131 * directory is in a logged range item, but not actually in the log
2132 * that means it was deleted from the directory before the fsync
2133 * and should be removed.
2135 static noinline int find_dir_range(struct btrfs_root *root,
2136 struct btrfs_path *path,
2137 u64 dirid, int key_type,
2138 u64 *start_ret, u64 *end_ret)
2140 struct btrfs_key key;
2142 struct btrfs_dir_log_item *item;
2146 if (*start_ret == (u64)-1)
2149 key.objectid = dirid;
2150 key.type = key_type;
2151 key.offset = *start_ret;
2153 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2157 if (path->slots[0] == 0)
2162 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2164 if (key.type != key_type || key.objectid != dirid) {
2168 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2169 struct btrfs_dir_log_item);
2170 found_end = btrfs_dir_log_end(path->nodes[0], item);
2172 if (*start_ret >= key.offset && *start_ret <= found_end) {
2174 *start_ret = key.offset;
2175 *end_ret = found_end;
2180 /* check the next slot in the tree to see if it is a valid item */
2181 nritems = btrfs_header_nritems(path->nodes[0]);
2183 if (path->slots[0] >= nritems) {
2184 ret = btrfs_next_leaf(root, path);
2189 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2191 if (key.type != key_type || key.objectid != dirid) {
2195 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2196 struct btrfs_dir_log_item);
2197 found_end = btrfs_dir_log_end(path->nodes[0], item);
2198 *start_ret = key.offset;
2199 *end_ret = found_end;
2202 btrfs_release_path(path);
2207 * this looks for a given directory item in the log. If the directory
2208 * item is not in the log, the item is removed and the inode it points
2211 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2212 struct btrfs_root *root,
2213 struct btrfs_root *log,
2214 struct btrfs_path *path,
2215 struct btrfs_path *log_path,
2217 struct btrfs_key *dir_key)
2220 struct extent_buffer *eb;
2223 struct btrfs_dir_item *di;
2224 struct btrfs_dir_item *log_di;
2227 unsigned long ptr_end;
2229 struct inode *inode;
2230 struct btrfs_key location;
2233 eb = path->nodes[0];
2234 slot = path->slots[0];
2235 item_size = btrfs_item_size_nr(eb, slot);
2236 ptr = btrfs_item_ptr_offset(eb, slot);
2237 ptr_end = ptr + item_size;
2238 while (ptr < ptr_end) {
2239 di = (struct btrfs_dir_item *)ptr;
2240 name_len = btrfs_dir_name_len(eb, di);
2241 name = kmalloc(name_len, GFP_NOFS);
2246 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2249 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2250 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2253 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2254 log_di = btrfs_lookup_dir_index_item(trans, log,
2260 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2261 btrfs_dir_item_key_to_cpu(eb, di, &location);
2262 btrfs_release_path(path);
2263 btrfs_release_path(log_path);
2264 inode = read_one_inode(root, location.objectid);
2270 ret = link_to_fixup_dir(trans, root,
2271 path, location.objectid);
2279 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2280 BTRFS_I(inode), name, name_len);
2282 ret = btrfs_run_delayed_items(trans);
2288 /* there might still be more names under this key
2289 * check and repeat if required
2291 ret = btrfs_search_slot(NULL, root, dir_key, path,
2297 } else if (IS_ERR(log_di)) {
2299 return PTR_ERR(log_di);
2301 btrfs_release_path(log_path);
2304 ptr = (unsigned long)(di + 1);
2309 btrfs_release_path(path);
2310 btrfs_release_path(log_path);
2314 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2315 struct btrfs_root *root,
2316 struct btrfs_root *log,
2317 struct btrfs_path *path,
2320 struct btrfs_key search_key;
2321 struct btrfs_path *log_path;
2326 log_path = btrfs_alloc_path();
2330 search_key.objectid = ino;
2331 search_key.type = BTRFS_XATTR_ITEM_KEY;
2332 search_key.offset = 0;
2334 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2338 nritems = btrfs_header_nritems(path->nodes[0]);
2339 for (i = path->slots[0]; i < nritems; i++) {
2340 struct btrfs_key key;
2341 struct btrfs_dir_item *di;
2342 struct btrfs_dir_item *log_di;
2346 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2347 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2352 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2353 total_size = btrfs_item_size_nr(path->nodes[0], i);
2355 while (cur < total_size) {
2356 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2357 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2358 u32 this_len = sizeof(*di) + name_len + data_len;
2361 name = kmalloc(name_len, GFP_NOFS);
2366 read_extent_buffer(path->nodes[0], name,
2367 (unsigned long)(di + 1), name_len);
2369 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2371 btrfs_release_path(log_path);
2373 /* Doesn't exist in log tree, so delete it. */
2374 btrfs_release_path(path);
2375 di = btrfs_lookup_xattr(trans, root, path, ino,
2376 name, name_len, -1);
2383 ret = btrfs_delete_one_dir_name(trans, root,
2387 btrfs_release_path(path);
2392 if (IS_ERR(log_di)) {
2393 ret = PTR_ERR(log_di);
2397 di = (struct btrfs_dir_item *)((char *)di + this_len);
2400 ret = btrfs_next_leaf(root, path);
2406 btrfs_free_path(log_path);
2407 btrfs_release_path(path);
2413 * deletion replay happens before we copy any new directory items
2414 * out of the log or out of backreferences from inodes. It
2415 * scans the log to find ranges of keys that log is authoritative for,
2416 * and then scans the directory to find items in those ranges that are
2417 * not present in the log.
2419 * Anything we don't find in the log is unlinked and removed from the
2422 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2423 struct btrfs_root *root,
2424 struct btrfs_root *log,
2425 struct btrfs_path *path,
2426 u64 dirid, int del_all)
2430 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2432 struct btrfs_key dir_key;
2433 struct btrfs_key found_key;
2434 struct btrfs_path *log_path;
2437 dir_key.objectid = dirid;
2438 dir_key.type = BTRFS_DIR_ITEM_KEY;
2439 log_path = btrfs_alloc_path();
2443 dir = read_one_inode(root, dirid);
2444 /* it isn't an error if the inode isn't there, that can happen
2445 * because we replay the deletes before we copy in the inode item
2449 btrfs_free_path(log_path);
2457 range_end = (u64)-1;
2459 ret = find_dir_range(log, path, dirid, key_type,
2460 &range_start, &range_end);
2465 dir_key.offset = range_start;
2468 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2473 nritems = btrfs_header_nritems(path->nodes[0]);
2474 if (path->slots[0] >= nritems) {
2475 ret = btrfs_next_leaf(root, path);
2481 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2483 if (found_key.objectid != dirid ||
2484 found_key.type != dir_key.type)
2487 if (found_key.offset > range_end)
2490 ret = check_item_in_log(trans, root, log, path,
2495 if (found_key.offset == (u64)-1)
2497 dir_key.offset = found_key.offset + 1;
2499 btrfs_release_path(path);
2500 if (range_end == (u64)-1)
2502 range_start = range_end + 1;
2507 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2508 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2509 dir_key.type = BTRFS_DIR_INDEX_KEY;
2510 btrfs_release_path(path);
2514 btrfs_release_path(path);
2515 btrfs_free_path(log_path);
2521 * the process_func used to replay items from the log tree. This
2522 * gets called in two different stages. The first stage just looks
2523 * for inodes and makes sure they are all copied into the subvolume.
2525 * The second stage copies all the other item types from the log into
2526 * the subvolume. The two stage approach is slower, but gets rid of
2527 * lots of complexity around inodes referencing other inodes that exist
2528 * only in the log (references come from either directory items or inode
2531 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2532 struct walk_control *wc, u64 gen, int level)
2535 struct btrfs_path *path;
2536 struct btrfs_root *root = wc->replay_dest;
2537 struct btrfs_key key;
2541 ret = btrfs_read_buffer(eb, gen, level, NULL);
2545 level = btrfs_header_level(eb);
2550 path = btrfs_alloc_path();
2554 nritems = btrfs_header_nritems(eb);
2555 for (i = 0; i < nritems; i++) {
2556 btrfs_item_key_to_cpu(eb, &key, i);
2558 /* inode keys are done during the first stage */
2559 if (key.type == BTRFS_INODE_ITEM_KEY &&
2560 wc->stage == LOG_WALK_REPLAY_INODES) {
2561 struct btrfs_inode_item *inode_item;
2564 inode_item = btrfs_item_ptr(eb, i,
2565 struct btrfs_inode_item);
2567 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2568 * and never got linked before the fsync, skip it, as
2569 * replaying it is pointless since it would be deleted
2570 * later. We skip logging tmpfiles, but it's always
2571 * possible we are replaying a log created with a kernel
2572 * that used to log tmpfiles.
2574 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2575 wc->ignore_cur_inode = true;
2578 wc->ignore_cur_inode = false;
2580 ret = replay_xattr_deletes(wc->trans, root, log,
2581 path, key.objectid);
2584 mode = btrfs_inode_mode(eb, inode_item);
2585 if (S_ISDIR(mode)) {
2586 ret = replay_dir_deletes(wc->trans,
2587 root, log, path, key.objectid, 0);
2591 ret = overwrite_item(wc->trans, root, path,
2597 * Before replaying extents, truncate the inode to its
2598 * size. We need to do it now and not after log replay
2599 * because before an fsync we can have prealloc extents
2600 * added beyond the inode's i_size. If we did it after,
2601 * through orphan cleanup for example, we would drop
2602 * those prealloc extents just after replaying them.
2604 if (S_ISREG(mode)) {
2605 struct inode *inode;
2608 inode = read_one_inode(root, key.objectid);
2613 from = ALIGN(i_size_read(inode),
2614 root->fs_info->sectorsize);
2615 ret = btrfs_drop_extents(wc->trans, root, inode,
2618 /* Update the inode's nbytes. */
2619 ret = btrfs_update_inode(wc->trans,
2627 ret = link_to_fixup_dir(wc->trans, root,
2628 path, key.objectid);
2633 if (wc->ignore_cur_inode)
2636 if (key.type == BTRFS_DIR_INDEX_KEY &&
2637 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2638 ret = replay_one_dir_item(wc->trans, root, path,
2644 if (wc->stage < LOG_WALK_REPLAY_ALL)
2647 /* these keys are simply copied */
2648 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2649 ret = overwrite_item(wc->trans, root, path,
2653 } else if (key.type == BTRFS_INODE_REF_KEY ||
2654 key.type == BTRFS_INODE_EXTREF_KEY) {
2655 ret = add_inode_ref(wc->trans, root, log, path,
2657 if (ret && ret != -ENOENT)
2660 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2661 ret = replay_one_extent(wc->trans, root, path,
2665 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2666 ret = replay_one_dir_item(wc->trans, root, path,
2672 btrfs_free_path(path);
2676 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2677 struct btrfs_root *root,
2678 struct btrfs_path *path, int *level,
2679 struct walk_control *wc)
2681 struct btrfs_fs_info *fs_info = root->fs_info;
2685 struct extent_buffer *next;
2686 struct extent_buffer *cur;
2687 struct extent_buffer *parent;
2691 WARN_ON(*level < 0);
2692 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2694 while (*level > 0) {
2695 struct btrfs_key first_key;
2697 WARN_ON(*level < 0);
2698 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2699 cur = path->nodes[*level];
2701 WARN_ON(btrfs_header_level(cur) != *level);
2703 if (path->slots[*level] >=
2704 btrfs_header_nritems(cur))
2707 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2708 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2709 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2710 blocksize = fs_info->nodesize;
2712 parent = path->nodes[*level];
2713 root_owner = btrfs_header_owner(parent);
2715 next = btrfs_find_create_tree_block(fs_info, bytenr);
2717 return PTR_ERR(next);
2720 ret = wc->process_func(root, next, wc, ptr_gen,
2723 free_extent_buffer(next);
2727 path->slots[*level]++;
2729 ret = btrfs_read_buffer(next, ptr_gen,
2730 *level - 1, &first_key);
2732 free_extent_buffer(next);
2737 btrfs_tree_lock(next);
2738 btrfs_set_lock_blocking_write(next);
2739 btrfs_clean_tree_block(next);
2740 btrfs_wait_tree_block_writeback(next);
2741 btrfs_tree_unlock(next);
2743 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2744 clear_extent_buffer_dirty(next);
2747 WARN_ON(root_owner !=
2748 BTRFS_TREE_LOG_OBJECTID);
2749 ret = btrfs_free_and_pin_reserved_extent(
2753 free_extent_buffer(next);
2757 free_extent_buffer(next);
2760 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2762 free_extent_buffer(next);
2766 WARN_ON(*level <= 0);
2767 if (path->nodes[*level-1])
2768 free_extent_buffer(path->nodes[*level-1]);
2769 path->nodes[*level-1] = next;
2770 *level = btrfs_header_level(next);
2771 path->slots[*level] = 0;
2774 WARN_ON(*level < 0);
2775 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2777 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2783 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root,
2785 struct btrfs_path *path, int *level,
2786 struct walk_control *wc)
2788 struct btrfs_fs_info *fs_info = root->fs_info;
2794 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2795 slot = path->slots[i];
2796 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2799 WARN_ON(*level == 0);
2802 struct extent_buffer *parent;
2803 if (path->nodes[*level] == root->node)
2804 parent = path->nodes[*level];
2806 parent = path->nodes[*level + 1];
2808 root_owner = btrfs_header_owner(parent);
2809 ret = wc->process_func(root, path->nodes[*level], wc,
2810 btrfs_header_generation(path->nodes[*level]),
2816 struct extent_buffer *next;
2818 next = path->nodes[*level];
2821 btrfs_tree_lock(next);
2822 btrfs_set_lock_blocking_write(next);
2823 btrfs_clean_tree_block(next);
2824 btrfs_wait_tree_block_writeback(next);
2825 btrfs_tree_unlock(next);
2827 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2828 clear_extent_buffer_dirty(next);
2831 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2832 ret = btrfs_free_and_pin_reserved_extent(
2834 path->nodes[*level]->start,
2835 path->nodes[*level]->len);
2839 free_extent_buffer(path->nodes[*level]);
2840 path->nodes[*level] = NULL;
2848 * drop the reference count on the tree rooted at 'snap'. This traverses
2849 * the tree freeing any blocks that have a ref count of zero after being
2852 static int walk_log_tree(struct btrfs_trans_handle *trans,
2853 struct btrfs_root *log, struct walk_control *wc)
2855 struct btrfs_fs_info *fs_info = log->fs_info;
2859 struct btrfs_path *path;
2862 path = btrfs_alloc_path();
2866 level = btrfs_header_level(log->node);
2868 path->nodes[level] = log->node;
2869 extent_buffer_get(log->node);
2870 path->slots[level] = 0;
2873 wret = walk_down_log_tree(trans, log, path, &level, wc);
2881 wret = walk_up_log_tree(trans, log, path, &level, wc);
2890 /* was the root node processed? if not, catch it here */
2891 if (path->nodes[orig_level]) {
2892 ret = wc->process_func(log, path->nodes[orig_level], wc,
2893 btrfs_header_generation(path->nodes[orig_level]),
2898 struct extent_buffer *next;
2900 next = path->nodes[orig_level];
2903 btrfs_tree_lock(next);
2904 btrfs_set_lock_blocking_write(next);
2905 btrfs_clean_tree_block(next);
2906 btrfs_wait_tree_block_writeback(next);
2907 btrfs_tree_unlock(next);
2909 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2910 clear_extent_buffer_dirty(next);
2913 WARN_ON(log->root_key.objectid !=
2914 BTRFS_TREE_LOG_OBJECTID);
2915 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2916 next->start, next->len);
2923 btrfs_free_path(path);
2928 * helper function to update the item for a given subvolumes log root
2929 * in the tree of log roots
2931 static int update_log_root(struct btrfs_trans_handle *trans,
2932 struct btrfs_root *log)
2934 struct btrfs_fs_info *fs_info = log->fs_info;
2937 if (log->log_transid == 1) {
2938 /* insert root item on the first sync */
2939 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2940 &log->root_key, &log->root_item);
2942 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2943 &log->root_key, &log->root_item);
2948 static void wait_log_commit(struct btrfs_root *root, int transid)
2951 int index = transid % 2;
2954 * we only allow two pending log transactions at a time,
2955 * so we know that if ours is more than 2 older than the
2956 * current transaction, we're done
2959 prepare_to_wait(&root->log_commit_wait[index],
2960 &wait, TASK_UNINTERRUPTIBLE);
2962 if (!(root->log_transid_committed < transid &&
2963 atomic_read(&root->log_commit[index])))
2966 mutex_unlock(&root->log_mutex);
2968 mutex_lock(&root->log_mutex);
2970 finish_wait(&root->log_commit_wait[index], &wait);
2973 static void wait_for_writer(struct btrfs_root *root)
2978 prepare_to_wait(&root->log_writer_wait, &wait,
2979 TASK_UNINTERRUPTIBLE);
2980 if (!atomic_read(&root->log_writers))
2983 mutex_unlock(&root->log_mutex);
2985 mutex_lock(&root->log_mutex);
2987 finish_wait(&root->log_writer_wait, &wait);
2990 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2991 struct btrfs_log_ctx *ctx)
2996 mutex_lock(&root->log_mutex);
2997 list_del_init(&ctx->list);
2998 mutex_unlock(&root->log_mutex);
3002 * Invoked in log mutex context, or be sure there is no other task which
3003 * can access the list.
3005 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3006 int index, int error)
3008 struct btrfs_log_ctx *ctx;
3009 struct btrfs_log_ctx *safe;
3011 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3012 list_del_init(&ctx->list);
3013 ctx->log_ret = error;
3016 INIT_LIST_HEAD(&root->log_ctxs[index]);
3020 * btrfs_sync_log does sends a given tree log down to the disk and
3021 * updates the super blocks to record it. When this call is done,
3022 * you know that any inodes previously logged are safely on disk only
3025 * Any other return value means you need to call btrfs_commit_transaction.
3026 * Some of the edge cases for fsyncing directories that have had unlinks
3027 * or renames done in the past mean that sometimes the only safe
3028 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3029 * that has happened.
3031 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3032 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3038 struct btrfs_fs_info *fs_info = root->fs_info;
3039 struct btrfs_root *log = root->log_root;
3040 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3041 int log_transid = 0;
3042 struct btrfs_log_ctx root_log_ctx;
3043 struct blk_plug plug;
3045 mutex_lock(&root->log_mutex);
3046 log_transid = ctx->log_transid;
3047 if (root->log_transid_committed >= log_transid) {
3048 mutex_unlock(&root->log_mutex);
3049 return ctx->log_ret;
3052 index1 = log_transid % 2;
3053 if (atomic_read(&root->log_commit[index1])) {
3054 wait_log_commit(root, log_transid);
3055 mutex_unlock(&root->log_mutex);
3056 return ctx->log_ret;
3058 ASSERT(log_transid == root->log_transid);
3059 atomic_set(&root->log_commit[index1], 1);
3061 /* wait for previous tree log sync to complete */
3062 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3063 wait_log_commit(root, log_transid - 1);
3066 int batch = atomic_read(&root->log_batch);
3067 /* when we're on an ssd, just kick the log commit out */
3068 if (!btrfs_test_opt(fs_info, SSD) &&
3069 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3070 mutex_unlock(&root->log_mutex);
3071 schedule_timeout_uninterruptible(1);
3072 mutex_lock(&root->log_mutex);
3074 wait_for_writer(root);
3075 if (batch == atomic_read(&root->log_batch))
3079 /* bail out if we need to do a full commit */
3080 if (btrfs_need_log_full_commit(trans)) {
3082 mutex_unlock(&root->log_mutex);
3086 if (log_transid % 2 == 0)
3087 mark = EXTENT_DIRTY;
3091 /* we start IO on all the marked extents here, but we don't actually
3092 * wait for them until later.
3094 blk_start_plug(&plug);
3095 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3097 blk_finish_plug(&plug);
3098 btrfs_abort_transaction(trans, ret);
3099 btrfs_set_log_full_commit(trans);
3100 mutex_unlock(&root->log_mutex);
3104 btrfs_set_root_node(&log->root_item, log->node);
3106 root->log_transid++;
3107 log->log_transid = root->log_transid;
3108 root->log_start_pid = 0;
3110 * IO has been started, blocks of the log tree have WRITTEN flag set
3111 * in their headers. new modifications of the log will be written to
3112 * new positions. so it's safe to allow log writers to go in.
3114 mutex_unlock(&root->log_mutex);
3116 btrfs_init_log_ctx(&root_log_ctx, NULL);
3118 mutex_lock(&log_root_tree->log_mutex);
3119 atomic_inc(&log_root_tree->log_batch);
3120 atomic_inc(&log_root_tree->log_writers);
3122 index2 = log_root_tree->log_transid % 2;
3123 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3124 root_log_ctx.log_transid = log_root_tree->log_transid;
3126 mutex_unlock(&log_root_tree->log_mutex);
3128 ret = update_log_root(trans, log);
3130 mutex_lock(&log_root_tree->log_mutex);
3131 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3132 /* atomic_dec_and_test implies a barrier */
3133 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3137 if (!list_empty(&root_log_ctx.list))
3138 list_del_init(&root_log_ctx.list);
3140 blk_finish_plug(&plug);
3141 btrfs_set_log_full_commit(trans);
3143 if (ret != -ENOSPC) {
3144 btrfs_abort_transaction(trans, ret);
3145 mutex_unlock(&log_root_tree->log_mutex);
3148 btrfs_wait_tree_log_extents(log, mark);
3149 mutex_unlock(&log_root_tree->log_mutex);
3154 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3155 blk_finish_plug(&plug);
3156 list_del_init(&root_log_ctx.list);
3157 mutex_unlock(&log_root_tree->log_mutex);
3158 ret = root_log_ctx.log_ret;
3162 index2 = root_log_ctx.log_transid % 2;
3163 if (atomic_read(&log_root_tree->log_commit[index2])) {
3164 blk_finish_plug(&plug);
3165 ret = btrfs_wait_tree_log_extents(log, mark);
3166 wait_log_commit(log_root_tree,
3167 root_log_ctx.log_transid);
3168 mutex_unlock(&log_root_tree->log_mutex);
3170 ret = root_log_ctx.log_ret;
3173 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3174 atomic_set(&log_root_tree->log_commit[index2], 1);
3176 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3177 wait_log_commit(log_root_tree,
3178 root_log_ctx.log_transid - 1);
3181 wait_for_writer(log_root_tree);
3184 * now that we've moved on to the tree of log tree roots,
3185 * check the full commit flag again
3187 if (btrfs_need_log_full_commit(trans)) {
3188 blk_finish_plug(&plug);
3189 btrfs_wait_tree_log_extents(log, mark);
3190 mutex_unlock(&log_root_tree->log_mutex);
3192 goto out_wake_log_root;
3195 ret = btrfs_write_marked_extents(fs_info,
3196 &log_root_tree->dirty_log_pages,
3197 EXTENT_DIRTY | EXTENT_NEW);
3198 blk_finish_plug(&plug);
3200 btrfs_set_log_full_commit(trans);
3201 btrfs_abort_transaction(trans, ret);
3202 mutex_unlock(&log_root_tree->log_mutex);
3203 goto out_wake_log_root;
3205 ret = btrfs_wait_tree_log_extents(log, mark);
3207 ret = btrfs_wait_tree_log_extents(log_root_tree,
3208 EXTENT_NEW | EXTENT_DIRTY);
3210 btrfs_set_log_full_commit(trans);
3211 mutex_unlock(&log_root_tree->log_mutex);
3212 goto out_wake_log_root;
3215 btrfs_set_super_log_root(fs_info->super_for_commit,
3216 log_root_tree->node->start);
3217 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3218 btrfs_header_level(log_root_tree->node));
3220 log_root_tree->log_transid++;
3221 mutex_unlock(&log_root_tree->log_mutex);
3224 * Nobody else is going to jump in and write the ctree
3225 * super here because the log_commit atomic below is protecting
3226 * us. We must be called with a transaction handle pinning
3227 * the running transaction open, so a full commit can't hop
3228 * in and cause problems either.
3230 ret = write_all_supers(fs_info, 1);
3232 btrfs_set_log_full_commit(trans);
3233 btrfs_abort_transaction(trans, ret);
3234 goto out_wake_log_root;
3237 mutex_lock(&root->log_mutex);
3238 if (root->last_log_commit < log_transid)
3239 root->last_log_commit = log_transid;
3240 mutex_unlock(&root->log_mutex);
3243 mutex_lock(&log_root_tree->log_mutex);
3244 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3246 log_root_tree->log_transid_committed++;
3247 atomic_set(&log_root_tree->log_commit[index2], 0);
3248 mutex_unlock(&log_root_tree->log_mutex);
3251 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3252 * all the updates above are seen by the woken threads. It might not be
3253 * necessary, but proving that seems to be hard.
3255 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3257 mutex_lock(&root->log_mutex);
3258 btrfs_remove_all_log_ctxs(root, index1, ret);
3259 root->log_transid_committed++;
3260 atomic_set(&root->log_commit[index1], 0);
3261 mutex_unlock(&root->log_mutex);
3264 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3265 * all the updates above are seen by the woken threads. It might not be
3266 * necessary, but proving that seems to be hard.
3268 cond_wake_up(&root->log_commit_wait[index1]);
3272 static void free_log_tree(struct btrfs_trans_handle *trans,
3273 struct btrfs_root *log)
3276 struct walk_control wc = {
3278 .process_func = process_one_buffer
3281 ret = walk_log_tree(trans, log, &wc);
3284 btrfs_abort_transaction(trans, ret);
3286 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3289 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3290 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3291 free_extent_buffer(log->node);
3296 * free all the extents used by the tree log. This should be called
3297 * at commit time of the full transaction
3299 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3301 if (root->log_root) {
3302 free_log_tree(trans, root->log_root);
3303 root->log_root = NULL;
3308 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3309 struct btrfs_fs_info *fs_info)
3311 if (fs_info->log_root_tree) {
3312 free_log_tree(trans, fs_info->log_root_tree);
3313 fs_info->log_root_tree = NULL;
3319 * If both a file and directory are logged, and unlinks or renames are
3320 * mixed in, we have a few interesting corners:
3322 * create file X in dir Y
3323 * link file X to X.link in dir Y
3325 * unlink file X but leave X.link
3328 * After a crash we would expect only X.link to exist. But file X
3329 * didn't get fsync'd again so the log has back refs for X and X.link.
3331 * We solve this by removing directory entries and inode backrefs from the
3332 * log when a file that was logged in the current transaction is
3333 * unlinked. Any later fsync will include the updated log entries, and
3334 * we'll be able to reconstruct the proper directory items from backrefs.
3336 * This optimizations allows us to avoid relogging the entire inode
3337 * or the entire directory.
3339 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *root,
3341 const char *name, int name_len,
3342 struct btrfs_inode *dir, u64 index)
3344 struct btrfs_root *log;
3345 struct btrfs_dir_item *di;
3346 struct btrfs_path *path;
3350 u64 dir_ino = btrfs_ino(dir);
3352 if (dir->logged_trans < trans->transid)
3355 ret = join_running_log_trans(root);
3359 mutex_lock(&dir->log_mutex);
3361 log = root->log_root;
3362 path = btrfs_alloc_path();
3368 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3369 name, name_len, -1);
3375 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3376 bytes_del += name_len;
3382 btrfs_release_path(path);
3383 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3384 index, name, name_len, -1);
3390 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3391 bytes_del += name_len;
3398 /* update the directory size in the log to reflect the names
3402 struct btrfs_key key;
3404 key.objectid = dir_ino;
3406 key.type = BTRFS_INODE_ITEM_KEY;
3407 btrfs_release_path(path);
3409 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3415 struct btrfs_inode_item *item;
3418 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3419 struct btrfs_inode_item);
3420 i_size = btrfs_inode_size(path->nodes[0], item);
3421 if (i_size > bytes_del)
3422 i_size -= bytes_del;
3425 btrfs_set_inode_size(path->nodes[0], item, i_size);
3426 btrfs_mark_buffer_dirty(path->nodes[0]);
3429 btrfs_release_path(path);
3432 btrfs_free_path(path);
3434 mutex_unlock(&dir->log_mutex);
3435 if (ret == -ENOSPC) {
3436 btrfs_set_log_full_commit(trans);
3439 btrfs_abort_transaction(trans, ret);
3441 btrfs_end_log_trans(root);
3446 /* see comments for btrfs_del_dir_entries_in_log */
3447 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3448 struct btrfs_root *root,
3449 const char *name, int name_len,
3450 struct btrfs_inode *inode, u64 dirid)
3452 struct btrfs_root *log;
3456 if (inode->logged_trans < trans->transid)
3459 ret = join_running_log_trans(root);
3462 log = root->log_root;
3463 mutex_lock(&inode->log_mutex);
3465 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3467 mutex_unlock(&inode->log_mutex);
3468 if (ret == -ENOSPC) {
3469 btrfs_set_log_full_commit(trans);
3471 } else if (ret < 0 && ret != -ENOENT)
3472 btrfs_abort_transaction(trans, ret);
3473 btrfs_end_log_trans(root);
3479 * creates a range item in the log for 'dirid'. first_offset and
3480 * last_offset tell us which parts of the key space the log should
3481 * be considered authoritative for.
3483 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3484 struct btrfs_root *log,
3485 struct btrfs_path *path,
3486 int key_type, u64 dirid,
3487 u64 first_offset, u64 last_offset)
3490 struct btrfs_key key;
3491 struct btrfs_dir_log_item *item;
3493 key.objectid = dirid;
3494 key.offset = first_offset;
3495 if (key_type == BTRFS_DIR_ITEM_KEY)
3496 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3498 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3499 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3503 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3504 struct btrfs_dir_log_item);
3505 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3506 btrfs_mark_buffer_dirty(path->nodes[0]);
3507 btrfs_release_path(path);
3512 * log all the items included in the current transaction for a given
3513 * directory. This also creates the range items in the log tree required
3514 * to replay anything deleted before the fsync
3516 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3517 struct btrfs_root *root, struct btrfs_inode *inode,
3518 struct btrfs_path *path,
3519 struct btrfs_path *dst_path, int key_type,
3520 struct btrfs_log_ctx *ctx,
3521 u64 min_offset, u64 *last_offset_ret)
3523 struct btrfs_key min_key;
3524 struct btrfs_root *log = root->log_root;
3525 struct extent_buffer *src;
3530 u64 first_offset = min_offset;
3531 u64 last_offset = (u64)-1;
3532 u64 ino = btrfs_ino(inode);
3534 log = root->log_root;
3536 min_key.objectid = ino;
3537 min_key.type = key_type;
3538 min_key.offset = min_offset;
3540 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3543 * we didn't find anything from this transaction, see if there
3544 * is anything at all
3546 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3547 min_key.objectid = ino;
3548 min_key.type = key_type;
3549 min_key.offset = (u64)-1;
3550 btrfs_release_path(path);
3551 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3553 btrfs_release_path(path);
3556 ret = btrfs_previous_item(root, path, ino, key_type);
3558 /* if ret == 0 there are items for this type,
3559 * create a range to tell us the last key of this type.
3560 * otherwise, there are no items in this directory after
3561 * *min_offset, and we create a range to indicate that.
3564 struct btrfs_key tmp;
3565 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3567 if (key_type == tmp.type)
3568 first_offset = max(min_offset, tmp.offset) + 1;
3573 /* go backward to find any previous key */
3574 ret = btrfs_previous_item(root, path, ino, key_type);
3576 struct btrfs_key tmp;
3577 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3578 if (key_type == tmp.type) {
3579 first_offset = tmp.offset;
3580 ret = overwrite_item(trans, log, dst_path,
3581 path->nodes[0], path->slots[0],
3589 btrfs_release_path(path);
3592 * Find the first key from this transaction again. See the note for
3593 * log_new_dir_dentries, if we're logging a directory recursively we
3594 * won't be holding its i_mutex, which means we can modify the directory
3595 * while we're logging it. If we remove an entry between our first
3596 * search and this search we'll not find the key again and can just
3599 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3604 * we have a block from this transaction, log every item in it
3605 * from our directory
3608 struct btrfs_key tmp;
3609 src = path->nodes[0];
3610 nritems = btrfs_header_nritems(src);
3611 for (i = path->slots[0]; i < nritems; i++) {
3612 struct btrfs_dir_item *di;
3614 btrfs_item_key_to_cpu(src, &min_key, i);
3616 if (min_key.objectid != ino || min_key.type != key_type)
3618 ret = overwrite_item(trans, log, dst_path, src, i,
3626 * We must make sure that when we log a directory entry,
3627 * the corresponding inode, after log replay, has a
3628 * matching link count. For example:
3634 * xfs_io -c "fsync" mydir
3636 * <mount fs and log replay>
3638 * Would result in a fsync log that when replayed, our
3639 * file inode would have a link count of 1, but we get
3640 * two directory entries pointing to the same inode.
3641 * After removing one of the names, it would not be
3642 * possible to remove the other name, which resulted
3643 * always in stale file handle errors, and would not
3644 * be possible to rmdir the parent directory, since
3645 * its i_size could never decrement to the value
3646 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3648 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3649 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3651 (btrfs_dir_transid(src, di) == trans->transid ||
3652 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3653 tmp.type != BTRFS_ROOT_ITEM_KEY)
3654 ctx->log_new_dentries = true;
3656 path->slots[0] = nritems;
3659 * look ahead to the next item and see if it is also
3660 * from this directory and from this transaction
3662 ret = btrfs_next_leaf(root, path);
3665 last_offset = (u64)-1;
3670 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3671 if (tmp.objectid != ino || tmp.type != key_type) {
3672 last_offset = (u64)-1;
3675 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3676 ret = overwrite_item(trans, log, dst_path,
3677 path->nodes[0], path->slots[0],
3682 last_offset = tmp.offset;
3687 btrfs_release_path(path);
3688 btrfs_release_path(dst_path);
3691 *last_offset_ret = last_offset;
3693 * insert the log range keys to indicate where the log
3696 ret = insert_dir_log_key(trans, log, path, key_type,
3697 ino, first_offset, last_offset);
3705 * logging directories is very similar to logging inodes, We find all the items
3706 * from the current transaction and write them to the log.
3708 * The recovery code scans the directory in the subvolume, and if it finds a
3709 * key in the range logged that is not present in the log tree, then it means
3710 * that dir entry was unlinked during the transaction.
3712 * In order for that scan to work, we must include one key smaller than
3713 * the smallest logged by this transaction and one key larger than the largest
3714 * key logged by this transaction.
3716 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3717 struct btrfs_root *root, struct btrfs_inode *inode,
3718 struct btrfs_path *path,
3719 struct btrfs_path *dst_path,
3720 struct btrfs_log_ctx *ctx)
3725 int key_type = BTRFS_DIR_ITEM_KEY;
3731 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3732 ctx, min_key, &max_key);
3735 if (max_key == (u64)-1)
3737 min_key = max_key + 1;
3740 if (key_type == BTRFS_DIR_ITEM_KEY) {
3741 key_type = BTRFS_DIR_INDEX_KEY;
3748 * a helper function to drop items from the log before we relog an
3749 * inode. max_key_type indicates the highest item type to remove.
3750 * This cannot be run for file data extents because it does not
3751 * free the extents they point to.
3753 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3754 struct btrfs_root *log,
3755 struct btrfs_path *path,
3756 u64 objectid, int max_key_type)
3759 struct btrfs_key key;
3760 struct btrfs_key found_key;
3763 key.objectid = objectid;
3764 key.type = max_key_type;
3765 key.offset = (u64)-1;
3768 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3769 BUG_ON(ret == 0); /* Logic error */
3773 if (path->slots[0] == 0)
3777 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3780 if (found_key.objectid != objectid)
3783 found_key.offset = 0;
3785 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3790 ret = btrfs_del_items(trans, log, path, start_slot,
3791 path->slots[0] - start_slot + 1);
3793 * If start slot isn't 0 then we don't need to re-search, we've
3794 * found the last guy with the objectid in this tree.
3796 if (ret || start_slot != 0)
3798 btrfs_release_path(path);
3800 btrfs_release_path(path);
3806 static void fill_inode_item(struct btrfs_trans_handle *trans,
3807 struct extent_buffer *leaf,
3808 struct btrfs_inode_item *item,
3809 struct inode *inode, int log_inode_only,
3812 struct btrfs_map_token token;
3814 btrfs_init_map_token(&token);
3816 if (log_inode_only) {
3817 /* set the generation to zero so the recover code
3818 * can tell the difference between an logging
3819 * just to say 'this inode exists' and a logging
3820 * to say 'update this inode with these values'
3822 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3823 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3825 btrfs_set_token_inode_generation(leaf, item,
3826 BTRFS_I(inode)->generation,
3828 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3831 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3832 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3833 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3834 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3836 btrfs_set_token_timespec_sec(leaf, &item->atime,
3837 inode->i_atime.tv_sec, &token);
3838 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3839 inode->i_atime.tv_nsec, &token);
3841 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3842 inode->i_mtime.tv_sec, &token);
3843 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3844 inode->i_mtime.tv_nsec, &token);
3846 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3847 inode->i_ctime.tv_sec, &token);
3848 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3849 inode->i_ctime.tv_nsec, &token);
3851 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3854 btrfs_set_token_inode_sequence(leaf, item,
3855 inode_peek_iversion(inode), &token);
3856 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3857 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3858 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3859 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3862 static int log_inode_item(struct btrfs_trans_handle *trans,
3863 struct btrfs_root *log, struct btrfs_path *path,
3864 struct btrfs_inode *inode)
3866 struct btrfs_inode_item *inode_item;
3869 ret = btrfs_insert_empty_item(trans, log, path,
3870 &inode->location, sizeof(*inode_item));
3871 if (ret && ret != -EEXIST)
3873 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3874 struct btrfs_inode_item);
3875 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3877 btrfs_release_path(path);
3881 static noinline int copy_items(struct btrfs_trans_handle *trans,
3882 struct btrfs_inode *inode,
3883 struct btrfs_path *dst_path,
3884 struct btrfs_path *src_path, u64 *last_extent,
3885 int start_slot, int nr, int inode_only,
3888 struct btrfs_fs_info *fs_info = trans->fs_info;
3889 unsigned long src_offset;
3890 unsigned long dst_offset;
3891 struct btrfs_root *log = inode->root->log_root;
3892 struct btrfs_file_extent_item *extent;
3893 struct btrfs_inode_item *inode_item;
3894 struct extent_buffer *src = src_path->nodes[0];
3895 struct btrfs_key first_key, last_key, key;
3897 struct btrfs_key *ins_keys;
3901 struct list_head ordered_sums;
3902 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3903 bool has_extents = false;
3904 bool need_find_last_extent = true;
3907 INIT_LIST_HEAD(&ordered_sums);
3909 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3910 nr * sizeof(u32), GFP_NOFS);
3914 first_key.objectid = (u64)-1;
3916 ins_sizes = (u32 *)ins_data;
3917 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3919 for (i = 0; i < nr; i++) {
3920 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3921 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3923 ret = btrfs_insert_empty_items(trans, log, dst_path,
3924 ins_keys, ins_sizes, nr);
3930 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3931 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3932 dst_path->slots[0]);
3934 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3937 last_key = ins_keys[i];
3939 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3940 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3942 struct btrfs_inode_item);
3943 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3945 inode_only == LOG_INODE_EXISTS,
3948 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3949 src_offset, ins_sizes[i]);
3953 * We set need_find_last_extent here in case we know we were
3954 * processing other items and then walk into the first extent in
3955 * the inode. If we don't hit an extent then nothing changes,
3956 * we'll do the last search the next time around.
3958 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3960 if (first_key.objectid == (u64)-1)
3961 first_key = ins_keys[i];
3963 need_find_last_extent = false;
3966 /* take a reference on file data extents so that truncates
3967 * or deletes of this inode don't have to relog the inode
3970 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3973 extent = btrfs_item_ptr(src, start_slot + i,
3974 struct btrfs_file_extent_item);
3976 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3979 found_type = btrfs_file_extent_type(src, extent);
3980 if (found_type == BTRFS_FILE_EXTENT_REG) {
3982 ds = btrfs_file_extent_disk_bytenr(src,
3984 /* ds == 0 is a hole */
3988 dl = btrfs_file_extent_disk_num_bytes(src,
3990 cs = btrfs_file_extent_offset(src, extent);
3991 cl = btrfs_file_extent_num_bytes(src,
3993 if (btrfs_file_extent_compression(src,
3999 ret = btrfs_lookup_csums_range(
4001 ds + cs, ds + cs + cl - 1,
4004 btrfs_release_path(dst_path);
4012 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4013 btrfs_release_path(dst_path);
4017 * we have to do this after the loop above to avoid changing the
4018 * log tree while trying to change the log tree.
4021 while (!list_empty(&ordered_sums)) {
4022 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4023 struct btrfs_ordered_sum,
4026 ret = btrfs_csum_file_blocks(trans, log, sums);
4027 list_del(&sums->list);
4034 if (need_find_last_extent && *last_extent == first_key.offset) {
4036 * We don't have any leafs between our current one and the one
4037 * we processed before that can have file extent items for our
4038 * inode (and have a generation number smaller than our current
4041 need_find_last_extent = false;
4045 * Because we use btrfs_search_forward we could skip leaves that were
4046 * not modified and then assume *last_extent is valid when it really
4047 * isn't. So back up to the previous leaf and read the end of the last
4048 * extent before we go and fill in holes.
4050 if (need_find_last_extent) {
4053 ret = btrfs_prev_leaf(inode->root, src_path);
4058 if (src_path->slots[0])
4059 src_path->slots[0]--;
4060 src = src_path->nodes[0];
4061 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
4062 if (key.objectid != btrfs_ino(inode) ||
4063 key.type != BTRFS_EXTENT_DATA_KEY)
4065 extent = btrfs_item_ptr(src, src_path->slots[0],
4066 struct btrfs_file_extent_item);
4067 if (btrfs_file_extent_type(src, extent) ==
4068 BTRFS_FILE_EXTENT_INLINE) {
4069 len = btrfs_file_extent_ram_bytes(src, extent);
4070 *last_extent = ALIGN(key.offset + len,
4071 fs_info->sectorsize);
4073 len = btrfs_file_extent_num_bytes(src, extent);
4074 *last_extent = key.offset + len;
4078 /* So we did prev_leaf, now we need to move to the next leaf, but a few
4079 * things could have happened
4081 * 1) A merge could have happened, so we could currently be on a leaf
4082 * that holds what we were copying in the first place.
4083 * 2) A split could have happened, and now not all of the items we want
4084 * are on the same leaf.
4086 * So we need to adjust how we search for holes, we need to drop the
4087 * path and re-search for the first extent key we found, and then walk
4088 * forward until we hit the last one we copied.
4090 if (need_find_last_extent) {
4091 /* btrfs_prev_leaf could return 1 without releasing the path */
4092 btrfs_release_path(src_path);
4093 ret = btrfs_search_slot(NULL, inode->root, &first_key,
4098 src = src_path->nodes[0];
4099 i = src_path->slots[0];
4105 * Ok so here we need to go through and fill in any holes we may have
4106 * to make sure that holes are punched for those areas in case they had
4107 * extents previously.
4113 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
4114 ret = btrfs_next_leaf(inode->root, src_path);
4118 src = src_path->nodes[0];
4120 need_find_last_extent = true;
4123 btrfs_item_key_to_cpu(src, &key, i);
4124 if (!btrfs_comp_cpu_keys(&key, &last_key))
4126 if (key.objectid != btrfs_ino(inode) ||
4127 key.type != BTRFS_EXTENT_DATA_KEY) {
4131 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
4132 if (btrfs_file_extent_type(src, extent) ==
4133 BTRFS_FILE_EXTENT_INLINE) {
4134 len = btrfs_file_extent_ram_bytes(src, extent);
4135 extent_end = ALIGN(key.offset + len,
4136 fs_info->sectorsize);
4138 len = btrfs_file_extent_num_bytes(src, extent);
4139 extent_end = key.offset + len;
4143 if (*last_extent == key.offset) {
4144 *last_extent = extent_end;
4147 offset = *last_extent;
4148 len = key.offset - *last_extent;
4149 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4150 offset, 0, 0, len, 0, len, 0, 0, 0);
4153 *last_extent = extent_end;
4157 * Check if there is a hole between the last extent found in our leaf
4158 * and the first extent in the next leaf. If there is one, we need to
4159 * log an explicit hole so that at replay time we can punch the hole.
4162 key.objectid == btrfs_ino(inode) &&
4163 key.type == BTRFS_EXTENT_DATA_KEY &&
4164 i == btrfs_header_nritems(src_path->nodes[0])) {
4165 ret = btrfs_next_leaf(inode->root, src_path);
4166 need_find_last_extent = true;
4169 } else if (ret == 0) {
4170 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4171 src_path->slots[0]);
4172 if (key.objectid == btrfs_ino(inode) &&
4173 key.type == BTRFS_EXTENT_DATA_KEY &&
4174 *last_extent < key.offset) {
4175 const u64 len = key.offset - *last_extent;
4177 ret = btrfs_insert_file_extent(trans, log,
4186 * Need to let the callers know we dropped the path so they should
4189 if (!ret && need_find_last_extent)
4194 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4196 struct extent_map *em1, *em2;
4198 em1 = list_entry(a, struct extent_map, list);
4199 em2 = list_entry(b, struct extent_map, list);
4201 if (em1->start < em2->start)
4203 else if (em1->start > em2->start)
4208 static int log_extent_csums(struct btrfs_trans_handle *trans,
4209 struct btrfs_inode *inode,
4210 struct btrfs_root *log_root,
4211 const struct extent_map *em)
4215 LIST_HEAD(ordered_sums);
4218 if (inode->flags & BTRFS_INODE_NODATASUM ||
4219 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4220 em->block_start == EXTENT_MAP_HOLE)
4223 /* If we're compressed we have to save the entire range of csums. */
4224 if (em->compress_type) {
4226 csum_len = max(em->block_len, em->orig_block_len);
4228 csum_offset = em->mod_start - em->start;
4229 csum_len = em->mod_len;
4232 /* block start is already adjusted for the file extent offset. */
4233 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4234 em->block_start + csum_offset,
4235 em->block_start + csum_offset +
4236 csum_len - 1, &ordered_sums, 0);
4240 while (!list_empty(&ordered_sums)) {
4241 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4242 struct btrfs_ordered_sum,
4245 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4246 list_del(&sums->list);
4253 static int log_one_extent(struct btrfs_trans_handle *trans,
4254 struct btrfs_inode *inode, struct btrfs_root *root,
4255 const struct extent_map *em,
4256 struct btrfs_path *path,
4257 struct btrfs_log_ctx *ctx)
4259 struct btrfs_root *log = root->log_root;
4260 struct btrfs_file_extent_item *fi;
4261 struct extent_buffer *leaf;
4262 struct btrfs_map_token token;
4263 struct btrfs_key key;
4264 u64 extent_offset = em->start - em->orig_start;
4267 int extent_inserted = 0;
4269 ret = log_extent_csums(trans, inode, log, em);
4273 btrfs_init_map_token(&token);
4275 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4276 em->start + em->len, NULL, 0, 1,
4277 sizeof(*fi), &extent_inserted);
4281 if (!extent_inserted) {
4282 key.objectid = btrfs_ino(inode);
4283 key.type = BTRFS_EXTENT_DATA_KEY;
4284 key.offset = em->start;
4286 ret = btrfs_insert_empty_item(trans, log, path, &key,
4291 leaf = path->nodes[0];
4292 fi = btrfs_item_ptr(leaf, path->slots[0],
4293 struct btrfs_file_extent_item);
4295 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4297 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4298 btrfs_set_token_file_extent_type(leaf, fi,
4299 BTRFS_FILE_EXTENT_PREALLOC,
4302 btrfs_set_token_file_extent_type(leaf, fi,
4303 BTRFS_FILE_EXTENT_REG,
4306 block_len = max(em->block_len, em->orig_block_len);
4307 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4308 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4311 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4313 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4314 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4316 extent_offset, &token);
4317 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4320 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4321 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4325 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4326 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4327 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4328 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4330 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4331 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4332 btrfs_mark_buffer_dirty(leaf);
4334 btrfs_release_path(path);
4340 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4341 * lose them after doing a fast fsync and replaying the log. We scan the
4342 * subvolume's root instead of iterating the inode's extent map tree because
4343 * otherwise we can log incorrect extent items based on extent map conversion.
4344 * That can happen due to the fact that extent maps are merged when they
4345 * are not in the extent map tree's list of modified extents.
4347 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4348 struct btrfs_inode *inode,
4349 struct btrfs_path *path)
4351 struct btrfs_root *root = inode->root;
4352 struct btrfs_key key;
4353 const u64 i_size = i_size_read(&inode->vfs_inode);
4354 const u64 ino = btrfs_ino(inode);
4355 struct btrfs_path *dst_path = NULL;
4356 u64 last_extent = (u64)-1;
4361 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4365 key.type = BTRFS_EXTENT_DATA_KEY;
4366 key.offset = i_size;
4367 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4372 struct extent_buffer *leaf = path->nodes[0];
4373 int slot = path->slots[0];
4375 if (slot >= btrfs_header_nritems(leaf)) {
4377 ret = copy_items(trans, inode, dst_path, path,
4378 &last_extent, start_slot,
4384 ret = btrfs_next_leaf(root, path);
4394 btrfs_item_key_to_cpu(leaf, &key, slot);
4395 if (key.objectid > ino)
4397 if (WARN_ON_ONCE(key.objectid < ino) ||
4398 key.type < BTRFS_EXTENT_DATA_KEY ||
4399 key.offset < i_size) {
4403 if (last_extent == (u64)-1) {
4404 last_extent = key.offset;
4406 * Avoid logging extent items logged in past fsync calls
4407 * and leading to duplicate keys in the log tree.
4410 ret = btrfs_truncate_inode_items(trans,
4414 BTRFS_EXTENT_DATA_KEY);
4415 } while (ret == -EAGAIN);
4424 dst_path = btrfs_alloc_path();
4432 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4433 start_slot, ins_nr, 1, 0);
4438 btrfs_release_path(path);
4439 btrfs_free_path(dst_path);
4443 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4444 struct btrfs_root *root,
4445 struct btrfs_inode *inode,
4446 struct btrfs_path *path,
4447 struct btrfs_log_ctx *ctx,
4451 struct extent_map *em, *n;
4452 struct list_head extents;
4453 struct extent_map_tree *tree = &inode->extent_tree;
4458 INIT_LIST_HEAD(&extents);
4460 write_lock(&tree->lock);
4461 test_gen = root->fs_info->last_trans_committed;
4463 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4465 * Skip extents outside our logging range. It's important to do
4466 * it for correctness because if we don't ignore them, we may
4467 * log them before their ordered extent completes, and therefore
4468 * we could log them without logging their respective checksums
4469 * (the checksum items are added to the csum tree at the very
4470 * end of btrfs_finish_ordered_io()). Also leave such extents
4471 * outside of our range in the list, since we may have another
4472 * ranged fsync in the near future that needs them. If an extent
4473 * outside our range corresponds to a hole, log it to avoid
4474 * leaving gaps between extents (fsck will complain when we are
4475 * not using the NO_HOLES feature).
4477 if ((em->start > end || em->start + em->len <= start) &&
4478 em->block_start != EXTENT_MAP_HOLE)
4481 list_del_init(&em->list);
4483 * Just an arbitrary number, this can be really CPU intensive
4484 * once we start getting a lot of extents, and really once we
4485 * have a bunch of extents we just want to commit since it will
4488 if (++num > 32768) {
4489 list_del_init(&tree->modified_extents);
4494 if (em->generation <= test_gen)
4497 /* We log prealloc extents beyond eof later. */
4498 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4499 em->start >= i_size_read(&inode->vfs_inode))
4502 /* Need a ref to keep it from getting evicted from cache */
4503 refcount_inc(&em->refs);
4504 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4505 list_add_tail(&em->list, &extents);
4509 list_sort(NULL, &extents, extent_cmp);
4511 while (!list_empty(&extents)) {
4512 em = list_entry(extents.next, struct extent_map, list);
4514 list_del_init(&em->list);
4517 * If we had an error we just need to delete everybody from our
4521 clear_em_logging(tree, em);
4522 free_extent_map(em);
4526 write_unlock(&tree->lock);
4528 ret = log_one_extent(trans, inode, root, em, path, ctx);
4529 write_lock(&tree->lock);
4530 clear_em_logging(tree, em);
4531 free_extent_map(em);
4533 WARN_ON(!list_empty(&extents));
4534 write_unlock(&tree->lock);
4536 btrfs_release_path(path);
4538 ret = btrfs_log_prealloc_extents(trans, inode, path);
4543 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4544 struct btrfs_path *path, u64 *size_ret)
4546 struct btrfs_key key;
4549 key.objectid = btrfs_ino(inode);
4550 key.type = BTRFS_INODE_ITEM_KEY;
4553 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4556 } else if (ret > 0) {
4559 struct btrfs_inode_item *item;
4561 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4562 struct btrfs_inode_item);
4563 *size_ret = btrfs_inode_size(path->nodes[0], item);
4565 * If the in-memory inode's i_size is smaller then the inode
4566 * size stored in the btree, return the inode's i_size, so
4567 * that we get a correct inode size after replaying the log
4568 * when before a power failure we had a shrinking truncate
4569 * followed by addition of a new name (rename / new hard link).
4570 * Otherwise return the inode size from the btree, to avoid
4571 * data loss when replaying a log due to previously doing a
4572 * write that expands the inode's size and logging a new name
4573 * immediately after.
4575 if (*size_ret > inode->vfs_inode.i_size)
4576 *size_ret = inode->vfs_inode.i_size;
4579 btrfs_release_path(path);
4584 * At the moment we always log all xattrs. This is to figure out at log replay
4585 * time which xattrs must have their deletion replayed. If a xattr is missing
4586 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4587 * because if a xattr is deleted, the inode is fsynced and a power failure
4588 * happens, causing the log to be replayed the next time the fs is mounted,
4589 * we want the xattr to not exist anymore (same behaviour as other filesystems
4590 * with a journal, ext3/4, xfs, f2fs, etc).
4592 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4593 struct btrfs_root *root,
4594 struct btrfs_inode *inode,
4595 struct btrfs_path *path,
4596 struct btrfs_path *dst_path)
4599 struct btrfs_key key;
4600 const u64 ino = btrfs_ino(inode);
4605 key.type = BTRFS_XATTR_ITEM_KEY;
4608 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4613 int slot = path->slots[0];
4614 struct extent_buffer *leaf = path->nodes[0];
4615 int nritems = btrfs_header_nritems(leaf);
4617 if (slot >= nritems) {
4619 u64 last_extent = 0;
4621 ret = copy_items(trans, inode, dst_path, path,
4622 &last_extent, start_slot,
4624 /* can't be 1, extent items aren't processed */
4630 ret = btrfs_next_leaf(root, path);
4638 btrfs_item_key_to_cpu(leaf, &key, slot);
4639 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4649 u64 last_extent = 0;
4651 ret = copy_items(trans, inode, dst_path, path,
4652 &last_extent, start_slot,
4654 /* can't be 1, extent items aren't processed */
4664 * If the no holes feature is enabled we need to make sure any hole between the
4665 * last extent and the i_size of our inode is explicitly marked in the log. This
4666 * is to make sure that doing something like:
4668 * 1) create file with 128Kb of data
4669 * 2) truncate file to 64Kb
4670 * 3) truncate file to 256Kb
4672 * 5) <crash/power failure>
4673 * 6) mount fs and trigger log replay
4675 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4676 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4677 * file correspond to a hole. The presence of explicit holes in a log tree is
4678 * what guarantees that log replay will remove/adjust file extent items in the
4681 * Here we do not need to care about holes between extents, that is already done
4682 * by copy_items(). We also only need to do this in the full sync path, where we
4683 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4684 * lookup the list of modified extent maps and if any represents a hole, we
4685 * insert a corresponding extent representing a hole in the log tree.
4687 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4688 struct btrfs_root *root,
4689 struct btrfs_inode *inode,
4690 struct btrfs_path *path)
4692 struct btrfs_fs_info *fs_info = root->fs_info;
4694 struct btrfs_key key;
4697 struct extent_buffer *leaf;
4698 struct btrfs_root *log = root->log_root;
4699 const u64 ino = btrfs_ino(inode);
4700 const u64 i_size = i_size_read(&inode->vfs_inode);
4702 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4706 key.type = BTRFS_EXTENT_DATA_KEY;
4707 key.offset = (u64)-1;
4709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4714 ASSERT(path->slots[0] > 0);
4716 leaf = path->nodes[0];
4717 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4719 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4720 /* inode does not have any extents */
4724 struct btrfs_file_extent_item *extent;
4728 * If there's an extent beyond i_size, an explicit hole was
4729 * already inserted by copy_items().
4731 if (key.offset >= i_size)
4734 extent = btrfs_item_ptr(leaf, path->slots[0],
4735 struct btrfs_file_extent_item);
4737 if (btrfs_file_extent_type(leaf, extent) ==
4738 BTRFS_FILE_EXTENT_INLINE)
4741 len = btrfs_file_extent_num_bytes(leaf, extent);
4742 /* Last extent goes beyond i_size, no need to log a hole. */
4743 if (key.offset + len > i_size)
4745 hole_start = key.offset + len;
4746 hole_size = i_size - hole_start;
4748 btrfs_release_path(path);
4750 /* Last extent ends at i_size. */
4754 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4755 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4756 hole_size, 0, hole_size, 0, 0, 0);
4761 * When we are logging a new inode X, check if it doesn't have a reference that
4762 * matches the reference from some other inode Y created in a past transaction
4763 * and that was renamed in the current transaction. If we don't do this, then at
4764 * log replay time we can lose inode Y (and all its files if it's a directory):
4767 * echo "hello world" > /mnt/x/foobar
4770 * mkdir /mnt/x # or touch /mnt/x
4771 * xfs_io -c fsync /mnt/x
4773 * mount fs, trigger log replay
4775 * After the log replay procedure, we would lose the first directory and all its
4776 * files (file foobar).
4777 * For the case where inode Y is not a directory we simply end up losing it:
4779 * echo "123" > /mnt/foo
4781 * mv /mnt/foo /mnt/bar
4782 * echo "abc" > /mnt/foo
4783 * xfs_io -c fsync /mnt/foo
4786 * We also need this for cases where a snapshot entry is replaced by some other
4787 * entry (file or directory) otherwise we end up with an unreplayable log due to
4788 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4789 * if it were a regular entry:
4792 * btrfs subvolume snapshot /mnt /mnt/x/snap
4793 * btrfs subvolume delete /mnt/x/snap
4796 * fsync /mnt/x or fsync some new file inside it
4799 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4800 * the same transaction.
4802 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4804 const struct btrfs_key *key,
4805 struct btrfs_inode *inode,
4806 u64 *other_ino, u64 *other_parent)
4809 struct btrfs_path *search_path;
4812 u32 item_size = btrfs_item_size_nr(eb, slot);
4814 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4816 search_path = btrfs_alloc_path();
4819 search_path->search_commit_root = 1;
4820 search_path->skip_locking = 1;
4822 while (cur_offset < item_size) {
4826 unsigned long name_ptr;
4827 struct btrfs_dir_item *di;
4829 if (key->type == BTRFS_INODE_REF_KEY) {
4830 struct btrfs_inode_ref *iref;
4832 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4833 parent = key->offset;
4834 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4835 name_ptr = (unsigned long)(iref + 1);
4836 this_len = sizeof(*iref) + this_name_len;
4838 struct btrfs_inode_extref *extref;
4840 extref = (struct btrfs_inode_extref *)(ptr +
4842 parent = btrfs_inode_extref_parent(eb, extref);
4843 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4844 name_ptr = (unsigned long)&extref->name;
4845 this_len = sizeof(*extref) + this_name_len;
4848 if (this_name_len > name_len) {
4851 new_name = krealloc(name, this_name_len, GFP_NOFS);
4856 name_len = this_name_len;
4860 read_extent_buffer(eb, name, name_ptr, this_name_len);
4861 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4862 parent, name, this_name_len, 0);
4863 if (di && !IS_ERR(di)) {
4864 struct btrfs_key di_key;
4866 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4868 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4869 if (di_key.objectid != key->objectid) {
4871 *other_ino = di_key.objectid;
4872 *other_parent = parent;
4880 } else if (IS_ERR(di)) {
4884 btrfs_release_path(search_path);
4886 cur_offset += this_len;
4890 btrfs_free_path(search_path);
4895 struct btrfs_ino_list {
4898 struct list_head list;
4901 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4902 struct btrfs_root *root,
4903 struct btrfs_path *path,
4904 struct btrfs_log_ctx *ctx,
4905 u64 ino, u64 parent)
4907 struct btrfs_ino_list *ino_elem;
4908 LIST_HEAD(inode_list);
4911 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4914 ino_elem->ino = ino;
4915 ino_elem->parent = parent;
4916 list_add_tail(&ino_elem->list, &inode_list);
4918 while (!list_empty(&inode_list)) {
4919 struct btrfs_fs_info *fs_info = root->fs_info;
4920 struct btrfs_key key;
4921 struct inode *inode;
4923 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4925 ino = ino_elem->ino;
4926 parent = ino_elem->parent;
4927 list_del(&ino_elem->list);
4932 btrfs_release_path(path);
4935 key.type = BTRFS_INODE_ITEM_KEY;
4937 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4939 * If the other inode that had a conflicting dir entry was
4940 * deleted in the current transaction, we need to log its parent
4943 if (IS_ERR(inode)) {
4944 ret = PTR_ERR(inode);
4945 if (ret == -ENOENT) {
4946 key.objectid = parent;
4947 inode = btrfs_iget(fs_info->sb, &key, root,
4949 if (IS_ERR(inode)) {
4950 ret = PTR_ERR(inode);
4952 ret = btrfs_log_inode(trans, root,
4954 LOG_OTHER_INODE_ALL,
4962 * We are safe logging the other inode without acquiring its
4963 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4964 * are safe against concurrent renames of the other inode as
4965 * well because during a rename we pin the log and update the
4966 * log with the new name before we unpin it.
4968 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4969 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4976 key.type = BTRFS_INODE_REF_KEY;
4978 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4985 struct extent_buffer *leaf = path->nodes[0];
4986 int slot = path->slots[0];
4988 u64 other_parent = 0;
4990 if (slot >= btrfs_header_nritems(leaf)) {
4991 ret = btrfs_next_leaf(root, path);
4994 } else if (ret > 0) {
5001 btrfs_item_key_to_cpu(leaf, &key, slot);
5002 if (key.objectid != ino ||
5003 (key.type != BTRFS_INODE_REF_KEY &&
5004 key.type != BTRFS_INODE_EXTREF_KEY)) {
5009 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5010 BTRFS_I(inode), &other_ino,
5015 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5020 ino_elem->ino = other_ino;
5021 ino_elem->parent = other_parent;
5022 list_add_tail(&ino_elem->list, &inode_list);
5033 /* log a single inode in the tree log.
5034 * At least one parent directory for this inode must exist in the tree
5035 * or be logged already.
5037 * Any items from this inode changed by the current transaction are copied
5038 * to the log tree. An extra reference is taken on any extents in this
5039 * file, allowing us to avoid a whole pile of corner cases around logging
5040 * blocks that have been removed from the tree.
5042 * See LOG_INODE_ALL and related defines for a description of what inode_only
5045 * This handles both files and directories.
5047 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5048 struct btrfs_root *root, struct btrfs_inode *inode,
5052 struct btrfs_log_ctx *ctx)
5054 struct btrfs_fs_info *fs_info = root->fs_info;
5055 struct btrfs_path *path;
5056 struct btrfs_path *dst_path;
5057 struct btrfs_key min_key;
5058 struct btrfs_key max_key;
5059 struct btrfs_root *log = root->log_root;
5060 u64 last_extent = 0;
5064 int ins_start_slot = 0;
5066 bool fast_search = false;
5067 u64 ino = btrfs_ino(inode);
5068 struct extent_map_tree *em_tree = &inode->extent_tree;
5069 u64 logged_isize = 0;
5070 bool need_log_inode_item = true;
5071 bool xattrs_logged = false;
5072 bool recursive_logging = false;
5074 path = btrfs_alloc_path();
5077 dst_path = btrfs_alloc_path();
5079 btrfs_free_path(path);
5083 min_key.objectid = ino;
5084 min_key.type = BTRFS_INODE_ITEM_KEY;
5087 max_key.objectid = ino;
5090 /* today the code can only do partial logging of directories */
5091 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5092 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5093 &inode->runtime_flags) &&
5094 inode_only >= LOG_INODE_EXISTS))
5095 max_key.type = BTRFS_XATTR_ITEM_KEY;
5097 max_key.type = (u8)-1;
5098 max_key.offset = (u64)-1;
5101 * Only run delayed items if we are a dir or a new file.
5102 * Otherwise commit the delayed inode only, which is needed in
5103 * order for the log replay code to mark inodes for link count
5104 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5106 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5107 inode->generation > fs_info->last_trans_committed)
5108 ret = btrfs_commit_inode_delayed_items(trans, inode);
5110 ret = btrfs_commit_inode_delayed_inode(inode);
5113 btrfs_free_path(path);
5114 btrfs_free_path(dst_path);
5118 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5119 recursive_logging = true;
5120 if (inode_only == LOG_OTHER_INODE)
5121 inode_only = LOG_INODE_EXISTS;
5123 inode_only = LOG_INODE_ALL;
5124 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5126 mutex_lock(&inode->log_mutex);
5130 * a brute force approach to making sure we get the most uptodate
5131 * copies of everything.
5133 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5134 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5136 if (inode_only == LOG_INODE_EXISTS)
5137 max_key_type = BTRFS_XATTR_ITEM_KEY;
5138 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5140 if (inode_only == LOG_INODE_EXISTS) {
5142 * Make sure the new inode item we write to the log has
5143 * the same isize as the current one (if it exists).
5144 * This is necessary to prevent data loss after log
5145 * replay, and also to prevent doing a wrong expanding
5146 * truncate - for e.g. create file, write 4K into offset
5147 * 0, fsync, write 4K into offset 4096, add hard link,
5148 * fsync some other file (to sync log), power fail - if
5149 * we use the inode's current i_size, after log replay
5150 * we get a 8Kb file, with the last 4Kb extent as a hole
5151 * (zeroes), as if an expanding truncate happened,
5152 * instead of getting a file of 4Kb only.
5154 err = logged_inode_size(log, inode, path, &logged_isize);
5158 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5159 &inode->runtime_flags)) {
5160 if (inode_only == LOG_INODE_EXISTS) {
5161 max_key.type = BTRFS_XATTR_ITEM_KEY;
5162 ret = drop_objectid_items(trans, log, path, ino,
5165 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5166 &inode->runtime_flags);
5167 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5168 &inode->runtime_flags);
5170 ret = btrfs_truncate_inode_items(trans,
5171 log, &inode->vfs_inode, 0, 0);
5176 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5177 &inode->runtime_flags) ||
5178 inode_only == LOG_INODE_EXISTS) {
5179 if (inode_only == LOG_INODE_ALL)
5181 max_key.type = BTRFS_XATTR_ITEM_KEY;
5182 ret = drop_objectid_items(trans, log, path, ino,
5185 if (inode_only == LOG_INODE_ALL)
5198 ret = btrfs_search_forward(root, &min_key,
5199 path, trans->transid);
5207 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5208 if (min_key.objectid != ino)
5210 if (min_key.type > max_key.type)
5213 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5214 need_log_inode_item = false;
5216 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5217 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5218 inode->generation == trans->transid &&
5219 !recursive_logging) {
5221 u64 other_parent = 0;
5223 ret = btrfs_check_ref_name_override(path->nodes[0],
5224 path->slots[0], &min_key, inode,
5225 &other_ino, &other_parent);
5229 } else if (ret > 0 && ctx &&
5230 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5235 ins_start_slot = path->slots[0];
5237 ret = copy_items(trans, inode, dst_path, path,
5238 &last_extent, ins_start_slot,
5247 err = log_conflicting_inodes(trans, root, path,
5248 ctx, other_ino, other_parent);
5251 btrfs_release_path(path);
5256 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5257 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5260 ret = copy_items(trans, inode, dst_path, path,
5261 &last_extent, ins_start_slot,
5262 ins_nr, inode_only, logged_isize);
5269 btrfs_release_path(path);
5275 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5278 } else if (!ins_nr) {
5279 ins_start_slot = path->slots[0];
5284 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5285 ins_start_slot, ins_nr, inode_only,
5293 btrfs_release_path(path);
5297 ins_start_slot = path->slots[0];
5300 nritems = btrfs_header_nritems(path->nodes[0]);
5302 if (path->slots[0] < nritems) {
5303 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5308 ret = copy_items(trans, inode, dst_path, path,
5309 &last_extent, ins_start_slot,
5310 ins_nr, inode_only, logged_isize);
5318 btrfs_release_path(path);
5320 if (min_key.offset < (u64)-1) {
5322 } else if (min_key.type < max_key.type) {
5330 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5331 ins_start_slot, ins_nr, inode_only,
5341 btrfs_release_path(path);
5342 btrfs_release_path(dst_path);
5343 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5346 xattrs_logged = true;
5347 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5348 btrfs_release_path(path);
5349 btrfs_release_path(dst_path);
5350 err = btrfs_log_trailing_hole(trans, root, inode, path);
5355 btrfs_release_path(path);
5356 btrfs_release_path(dst_path);
5357 if (need_log_inode_item) {
5358 err = log_inode_item(trans, log, dst_path, inode);
5359 if (!err && !xattrs_logged) {
5360 err = btrfs_log_all_xattrs(trans, root, inode, path,
5362 btrfs_release_path(path);
5368 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5374 } else if (inode_only == LOG_INODE_ALL) {
5375 struct extent_map *em, *n;
5377 write_lock(&em_tree->lock);
5379 * We can't just remove every em if we're called for a ranged
5380 * fsync - that is, one that doesn't cover the whole possible
5381 * file range (0 to LLONG_MAX). This is because we can have
5382 * em's that fall outside the range we're logging and therefore
5383 * their ordered operations haven't completed yet
5384 * (btrfs_finish_ordered_io() not invoked yet). This means we
5385 * didn't get their respective file extent item in the fs/subvol
5386 * tree yet, and need to let the next fast fsync (one which
5387 * consults the list of modified extent maps) find the em so
5388 * that it logs a matching file extent item and waits for the
5389 * respective ordered operation to complete (if it's still
5392 * Removing every em outside the range we're logging would make
5393 * the next fast fsync not log their matching file extent items,
5394 * therefore making us lose data after a log replay.
5396 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5398 const u64 mod_end = em->mod_start + em->mod_len - 1;
5400 if (em->mod_start >= start && mod_end <= end)
5401 list_del_init(&em->list);
5403 write_unlock(&em_tree->lock);
5406 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5407 ret = log_directory_changes(trans, root, inode, path, dst_path,
5415 spin_lock(&inode->lock);
5416 inode->logged_trans = trans->transid;
5417 inode->last_log_commit = inode->last_sub_trans;
5418 spin_unlock(&inode->lock);
5420 mutex_unlock(&inode->log_mutex);
5422 btrfs_free_path(path);
5423 btrfs_free_path(dst_path);
5428 * Check if we must fallback to a transaction commit when logging an inode.
5429 * This must be called after logging the inode and is used only in the context
5430 * when fsyncing an inode requires the need to log some other inode - in which
5431 * case we can't lock the i_mutex of each other inode we need to log as that
5432 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5433 * log inodes up or down in the hierarchy) or rename operations for example. So
5434 * we take the log_mutex of the inode after we have logged it and then check for
5435 * its last_unlink_trans value - this is safe because any task setting
5436 * last_unlink_trans must take the log_mutex and it must do this before it does
5437 * the actual unlink operation, so if we do this check before a concurrent task
5438 * sets last_unlink_trans it means we've logged a consistent version/state of
5439 * all the inode items, otherwise we are not sure and must do a transaction
5440 * commit (the concurrent task might have only updated last_unlink_trans before
5441 * we logged the inode or it might have also done the unlink).
5443 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5444 struct btrfs_inode *inode)
5446 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5449 mutex_lock(&inode->log_mutex);
5450 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5452 * Make sure any commits to the log are forced to be full
5455 btrfs_set_log_full_commit(trans);
5458 mutex_unlock(&inode->log_mutex);
5464 * follow the dentry parent pointers up the chain and see if any
5465 * of the directories in it require a full commit before they can
5466 * be logged. Returns zero if nothing special needs to be done or 1 if
5467 * a full commit is required.
5469 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5470 struct btrfs_inode *inode,
5471 struct dentry *parent,
5472 struct super_block *sb,
5476 struct dentry *old_parent = NULL;
5477 struct btrfs_inode *orig_inode = inode;
5480 * for regular files, if its inode is already on disk, we don't
5481 * have to worry about the parents at all. This is because
5482 * we can use the last_unlink_trans field to record renames
5483 * and other fun in this file.
5485 if (S_ISREG(inode->vfs_inode.i_mode) &&
5486 inode->generation <= last_committed &&
5487 inode->last_unlink_trans <= last_committed)
5490 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5491 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5493 inode = BTRFS_I(d_inode(parent));
5498 * If we are logging a directory then we start with our inode,
5499 * not our parent's inode, so we need to skip setting the
5500 * logged_trans so that further down in the log code we don't
5501 * think this inode has already been logged.
5503 if (inode != orig_inode)
5504 inode->logged_trans = trans->transid;
5507 if (btrfs_must_commit_transaction(trans, inode)) {
5512 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5515 if (IS_ROOT(parent)) {
5516 inode = BTRFS_I(d_inode(parent));
5517 if (btrfs_must_commit_transaction(trans, inode))
5522 parent = dget_parent(parent);
5524 old_parent = parent;
5525 inode = BTRFS_I(d_inode(parent));
5533 struct btrfs_dir_list {
5535 struct list_head list;
5539 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5540 * details about the why it is needed.
5541 * This is a recursive operation - if an existing dentry corresponds to a
5542 * directory, that directory's new entries are logged too (same behaviour as
5543 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5544 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5545 * complains about the following circular lock dependency / possible deadlock:
5549 * lock(&type->i_mutex_dir_key#3/2);
5550 * lock(sb_internal#2);
5551 * lock(&type->i_mutex_dir_key#3/2);
5552 * lock(&sb->s_type->i_mutex_key#14);
5554 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5555 * sb_start_intwrite() in btrfs_start_transaction().
5556 * Not locking i_mutex of the inodes is still safe because:
5558 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5559 * that while logging the inode new references (names) are added or removed
5560 * from the inode, leaving the logged inode item with a link count that does
5561 * not match the number of logged inode reference items. This is fine because
5562 * at log replay time we compute the real number of links and correct the
5563 * link count in the inode item (see replay_one_buffer() and
5564 * link_to_fixup_dir());
5566 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5567 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5568 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5569 * has a size that doesn't match the sum of the lengths of all the logged
5570 * names. This does not result in a problem because if a dir_item key is
5571 * logged but its matching dir_index key is not logged, at log replay time we
5572 * don't use it to replay the respective name (see replay_one_name()). On the
5573 * other hand if only the dir_index key ends up being logged, the respective
5574 * name is added to the fs/subvol tree with both the dir_item and dir_index
5575 * keys created (see replay_one_name()).
5576 * The directory's inode item with a wrong i_size is not a problem as well,
5577 * since we don't use it at log replay time to set the i_size in the inode
5578 * item of the fs/subvol tree (see overwrite_item()).
5580 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5581 struct btrfs_root *root,
5582 struct btrfs_inode *start_inode,
5583 struct btrfs_log_ctx *ctx)
5585 struct btrfs_fs_info *fs_info = root->fs_info;
5586 struct btrfs_root *log = root->log_root;
5587 struct btrfs_path *path;
5588 LIST_HEAD(dir_list);
5589 struct btrfs_dir_list *dir_elem;
5592 path = btrfs_alloc_path();
5596 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5598 btrfs_free_path(path);
5601 dir_elem->ino = btrfs_ino(start_inode);
5602 list_add_tail(&dir_elem->list, &dir_list);
5604 while (!list_empty(&dir_list)) {
5605 struct extent_buffer *leaf;
5606 struct btrfs_key min_key;
5610 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5613 goto next_dir_inode;
5615 min_key.objectid = dir_elem->ino;
5616 min_key.type = BTRFS_DIR_ITEM_KEY;
5619 btrfs_release_path(path);
5620 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5622 goto next_dir_inode;
5623 } else if (ret > 0) {
5625 goto next_dir_inode;
5629 leaf = path->nodes[0];
5630 nritems = btrfs_header_nritems(leaf);
5631 for (i = path->slots[0]; i < nritems; i++) {
5632 struct btrfs_dir_item *di;
5633 struct btrfs_key di_key;
5634 struct inode *di_inode;
5635 struct btrfs_dir_list *new_dir_elem;
5636 int log_mode = LOG_INODE_EXISTS;
5639 btrfs_item_key_to_cpu(leaf, &min_key, i);
5640 if (min_key.objectid != dir_elem->ino ||
5641 min_key.type != BTRFS_DIR_ITEM_KEY)
5642 goto next_dir_inode;
5644 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5645 type = btrfs_dir_type(leaf, di);
5646 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5647 type != BTRFS_FT_DIR)
5649 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5650 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5653 btrfs_release_path(path);
5654 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5655 if (IS_ERR(di_inode)) {
5656 ret = PTR_ERR(di_inode);
5657 goto next_dir_inode;
5660 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5665 ctx->log_new_dentries = false;
5666 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5667 log_mode = LOG_INODE_ALL;
5668 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5669 log_mode, 0, LLONG_MAX, ctx);
5671 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5675 goto next_dir_inode;
5676 if (ctx->log_new_dentries) {
5677 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5679 if (!new_dir_elem) {
5681 goto next_dir_inode;
5683 new_dir_elem->ino = di_key.objectid;
5684 list_add_tail(&new_dir_elem->list, &dir_list);
5689 ret = btrfs_next_leaf(log, path);
5691 goto next_dir_inode;
5692 } else if (ret > 0) {
5694 goto next_dir_inode;
5698 if (min_key.offset < (u64)-1) {
5703 list_del(&dir_elem->list);
5707 btrfs_free_path(path);
5711 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5712 struct btrfs_inode *inode,
5713 struct btrfs_log_ctx *ctx)
5715 struct btrfs_fs_info *fs_info = trans->fs_info;
5717 struct btrfs_path *path;
5718 struct btrfs_key key;
5719 struct btrfs_root *root = inode->root;
5720 const u64 ino = btrfs_ino(inode);
5722 path = btrfs_alloc_path();
5725 path->skip_locking = 1;
5726 path->search_commit_root = 1;
5729 key.type = BTRFS_INODE_REF_KEY;
5731 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5736 struct extent_buffer *leaf = path->nodes[0];
5737 int slot = path->slots[0];
5742 if (slot >= btrfs_header_nritems(leaf)) {
5743 ret = btrfs_next_leaf(root, path);
5751 btrfs_item_key_to_cpu(leaf, &key, slot);
5752 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5753 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5756 item_size = btrfs_item_size_nr(leaf, slot);
5757 ptr = btrfs_item_ptr_offset(leaf, slot);
5758 while (cur_offset < item_size) {
5759 struct btrfs_key inode_key;
5760 struct inode *dir_inode;
5762 inode_key.type = BTRFS_INODE_ITEM_KEY;
5763 inode_key.offset = 0;
5765 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5766 struct btrfs_inode_extref *extref;
5768 extref = (struct btrfs_inode_extref *)
5770 inode_key.objectid = btrfs_inode_extref_parent(
5772 cur_offset += sizeof(*extref);
5773 cur_offset += btrfs_inode_extref_name_len(leaf,
5776 inode_key.objectid = key.offset;
5777 cur_offset = item_size;
5780 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5783 * If the parent inode was deleted, return an error to
5784 * fallback to a transaction commit. This is to prevent
5785 * getting an inode that was moved from one parent A to
5786 * a parent B, got its former parent A deleted and then
5787 * it got fsync'ed, from existing at both parents after
5788 * a log replay (and the old parent still existing).
5795 * mv /mnt/B/bar /mnt/A/bar
5796 * mv -T /mnt/A /mnt/B
5800 * If we ignore the old parent B which got deleted,
5801 * after a log replay we would have file bar linked
5802 * at both parents and the old parent B would still
5805 if (IS_ERR(dir_inode)) {
5806 ret = PTR_ERR(dir_inode);
5811 ctx->log_new_dentries = false;
5812 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5813 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5815 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5817 if (!ret && ctx && ctx->log_new_dentries)
5818 ret = log_new_dir_dentries(trans, root,
5819 BTRFS_I(dir_inode), ctx);
5828 btrfs_free_path(path);
5833 * helper function around btrfs_log_inode to make sure newly created
5834 * parent directories also end up in the log. A minimal inode and backref
5835 * only logging is done of any parent directories that are older than
5836 * the last committed transaction
5838 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5839 struct btrfs_inode *inode,
5840 struct dentry *parent,
5844 struct btrfs_log_ctx *ctx)
5846 struct btrfs_root *root = inode->root;
5847 struct btrfs_fs_info *fs_info = root->fs_info;
5848 struct super_block *sb;
5849 struct dentry *old_parent = NULL;
5851 u64 last_committed = fs_info->last_trans_committed;
5852 bool log_dentries = false;
5853 struct btrfs_inode *orig_inode = inode;
5855 sb = inode->vfs_inode.i_sb;
5857 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5863 * The prev transaction commit doesn't complete, we need do
5864 * full commit by ourselves.
5866 if (fs_info->last_trans_log_full_commit >
5867 fs_info->last_trans_committed) {
5872 if (btrfs_root_refs(&root->root_item) == 0) {
5877 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5883 * Skip already logged inodes or inodes corresponding to tmpfiles
5884 * (since logging them is pointless, a link count of 0 means they
5885 * will never be accessible).
5887 if (btrfs_inode_in_log(inode, trans->transid) ||
5888 inode->vfs_inode.i_nlink == 0) {
5889 ret = BTRFS_NO_LOG_SYNC;
5893 ret = start_log_trans(trans, root, ctx);
5897 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5902 * for regular files, if its inode is already on disk, we don't
5903 * have to worry about the parents at all. This is because
5904 * we can use the last_unlink_trans field to record renames
5905 * and other fun in this file.
5907 if (S_ISREG(inode->vfs_inode.i_mode) &&
5908 inode->generation <= last_committed &&
5909 inode->last_unlink_trans <= last_committed) {
5914 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5915 log_dentries = true;
5918 * On unlink we must make sure all our current and old parent directory
5919 * inodes are fully logged. This is to prevent leaving dangling
5920 * directory index entries in directories that were our parents but are
5921 * not anymore. Not doing this results in old parent directory being
5922 * impossible to delete after log replay (rmdir will always fail with
5923 * error -ENOTEMPTY).
5929 * ln testdir/foo testdir/bar
5931 * unlink testdir/bar
5932 * xfs_io -c fsync testdir/foo
5934 * mount fs, triggers log replay
5936 * If we don't log the parent directory (testdir), after log replay the
5937 * directory still has an entry pointing to the file inode using the bar
5938 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5939 * the file inode has a link count of 1.
5945 * ln foo testdir/foo2
5946 * ln foo testdir/foo3
5948 * unlink testdir/foo3
5949 * xfs_io -c fsync foo
5951 * mount fs, triggers log replay
5953 * Similar as the first example, after log replay the parent directory
5954 * testdir still has an entry pointing to the inode file with name foo3
5955 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5956 * and has a link count of 2.
5958 if (inode->last_unlink_trans > last_committed) {
5959 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5965 * If a new hard link was added to the inode in the current transaction
5966 * and its link count is now greater than 1, we need to fallback to a
5967 * transaction commit, otherwise we can end up not logging all its new
5968 * parents for all the hard links. Here just from the dentry used to
5969 * fsync, we can not visit the ancestor inodes for all the other hard
5970 * links to figure out if any is new, so we fallback to a transaction
5971 * commit (instead of adding a lot of complexity of scanning a btree,
5972 * since this scenario is not a common use case).
5974 if (inode->vfs_inode.i_nlink > 1 &&
5975 inode->last_link_trans > last_committed) {
5981 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5984 inode = BTRFS_I(d_inode(parent));
5985 if (root != inode->root)
5988 if (inode->generation > last_committed) {
5989 ret = btrfs_log_inode(trans, root, inode,
5990 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5994 if (IS_ROOT(parent))
5997 parent = dget_parent(parent);
5999 old_parent = parent;
6002 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
6008 btrfs_set_log_full_commit(trans);
6013 btrfs_remove_log_ctx(root, ctx);
6014 btrfs_end_log_trans(root);
6020 * it is not safe to log dentry if the chunk root has added new
6021 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6022 * If this returns 1, you must commit the transaction to safely get your
6025 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6026 struct dentry *dentry,
6029 struct btrfs_log_ctx *ctx)
6031 struct dentry *parent = dget_parent(dentry);
6034 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6035 start, end, LOG_INODE_ALL, ctx);
6042 * should be called during mount to recover any replay any log trees
6045 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6048 struct btrfs_path *path;
6049 struct btrfs_trans_handle *trans;
6050 struct btrfs_key key;
6051 struct btrfs_key found_key;
6052 struct btrfs_key tmp_key;
6053 struct btrfs_root *log;
6054 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6055 struct walk_control wc = {
6056 .process_func = process_one_buffer,
6060 path = btrfs_alloc_path();
6064 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6066 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6067 if (IS_ERR(trans)) {
6068 ret = PTR_ERR(trans);
6075 ret = walk_log_tree(trans, log_root_tree, &wc);
6077 btrfs_handle_fs_error(fs_info, ret,
6078 "Failed to pin buffers while recovering log root tree.");
6083 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6084 key.offset = (u64)-1;
6085 key.type = BTRFS_ROOT_ITEM_KEY;
6088 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6091 btrfs_handle_fs_error(fs_info, ret,
6092 "Couldn't find tree log root.");
6096 if (path->slots[0] == 0)
6100 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6102 btrfs_release_path(path);
6103 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6106 log = btrfs_read_fs_root(log_root_tree, &found_key);
6109 btrfs_handle_fs_error(fs_info, ret,
6110 "Couldn't read tree log root.");
6114 tmp_key.objectid = found_key.offset;
6115 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6116 tmp_key.offset = (u64)-1;
6118 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
6119 if (IS_ERR(wc.replay_dest)) {
6120 ret = PTR_ERR(wc.replay_dest);
6121 free_extent_buffer(log->node);
6122 free_extent_buffer(log->commit_root);
6124 btrfs_handle_fs_error(fs_info, ret,
6125 "Couldn't read target root for tree log recovery.");
6129 wc.replay_dest->log_root = log;
6130 btrfs_record_root_in_trans(trans, wc.replay_dest);
6131 ret = walk_log_tree(trans, log, &wc);
6133 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6134 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6138 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6139 struct btrfs_root *root = wc.replay_dest;
6141 btrfs_release_path(path);
6144 * We have just replayed everything, and the highest
6145 * objectid of fs roots probably has changed in case
6146 * some inode_item's got replayed.
6148 * root->objectid_mutex is not acquired as log replay
6149 * could only happen during mount.
6151 ret = btrfs_find_highest_objectid(root,
6152 &root->highest_objectid);
6155 key.offset = found_key.offset - 1;
6156 wc.replay_dest->log_root = NULL;
6157 free_extent_buffer(log->node);
6158 free_extent_buffer(log->commit_root);
6164 if (found_key.offset == 0)
6167 btrfs_release_path(path);
6169 /* step one is to pin it all, step two is to replay just inodes */
6172 wc.process_func = replay_one_buffer;
6173 wc.stage = LOG_WALK_REPLAY_INODES;
6176 /* step three is to replay everything */
6177 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6182 btrfs_free_path(path);
6184 /* step 4: commit the transaction, which also unpins the blocks */
6185 ret = btrfs_commit_transaction(trans);
6189 free_extent_buffer(log_root_tree->node);
6190 log_root_tree->log_root = NULL;
6191 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6192 kfree(log_root_tree);
6197 btrfs_end_transaction(wc.trans);
6198 btrfs_free_path(path);
6203 * there are some corner cases where we want to force a full
6204 * commit instead of allowing a directory to be logged.
6206 * They revolve around files there were unlinked from the directory, and
6207 * this function updates the parent directory so that a full commit is
6208 * properly done if it is fsync'd later after the unlinks are done.
6210 * Must be called before the unlink operations (updates to the subvolume tree,
6211 * inodes, etc) are done.
6213 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6214 struct btrfs_inode *dir, struct btrfs_inode *inode,
6218 * when we're logging a file, if it hasn't been renamed
6219 * or unlinked, and its inode is fully committed on disk,
6220 * we don't have to worry about walking up the directory chain
6221 * to log its parents.
6223 * So, we use the last_unlink_trans field to put this transid
6224 * into the file. When the file is logged we check it and
6225 * don't log the parents if the file is fully on disk.
6227 mutex_lock(&inode->log_mutex);
6228 inode->last_unlink_trans = trans->transid;
6229 mutex_unlock(&inode->log_mutex);
6232 * if this directory was already logged any new
6233 * names for this file/dir will get recorded
6236 if (dir->logged_trans == trans->transid)
6240 * if the inode we're about to unlink was logged,
6241 * the log will be properly updated for any new names
6243 if (inode->logged_trans == trans->transid)
6247 * when renaming files across directories, if the directory
6248 * there we're unlinking from gets fsync'd later on, there's
6249 * no way to find the destination directory later and fsync it
6250 * properly. So, we have to be conservative and force commits
6251 * so the new name gets discovered.
6256 /* we can safely do the unlink without any special recording */
6260 mutex_lock(&dir->log_mutex);
6261 dir->last_unlink_trans = trans->transid;
6262 mutex_unlock(&dir->log_mutex);
6266 * Make sure that if someone attempts to fsync the parent directory of a deleted
6267 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6268 * that after replaying the log tree of the parent directory's root we will not
6269 * see the snapshot anymore and at log replay time we will not see any log tree
6270 * corresponding to the deleted snapshot's root, which could lead to replaying
6271 * it after replaying the log tree of the parent directory (which would replay
6272 * the snapshot delete operation).
6274 * Must be called before the actual snapshot destroy operation (updates to the
6275 * parent root and tree of tree roots trees, etc) are done.
6277 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6278 struct btrfs_inode *dir)
6280 mutex_lock(&dir->log_mutex);
6281 dir->last_unlink_trans = trans->transid;
6282 mutex_unlock(&dir->log_mutex);
6286 * Call this after adding a new name for a file and it will properly
6287 * update the log to reflect the new name.
6289 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6290 * true (because it's not used).
6292 * Return value depends on whether @sync_log is true or false.
6293 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6294 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6296 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6297 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6298 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6299 * committed (without attempting to sync the log).
6301 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6302 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6303 struct dentry *parent,
6304 bool sync_log, struct btrfs_log_ctx *ctx)
6306 struct btrfs_fs_info *fs_info = trans->fs_info;
6310 * this will force the logging code to walk the dentry chain
6313 if (!S_ISDIR(inode->vfs_inode.i_mode))
6314 inode->last_unlink_trans = trans->transid;
6317 * if this inode hasn't been logged and directory we're renaming it
6318 * from hasn't been logged, we don't need to log it
6320 if (inode->logged_trans <= fs_info->last_trans_committed &&
6321 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6322 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6323 BTRFS_DONT_NEED_LOG_SYNC;
6326 struct btrfs_log_ctx ctx2;
6328 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6329 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6330 LOG_INODE_EXISTS, &ctx2);
6331 if (ret == BTRFS_NO_LOG_SYNC)
6332 return BTRFS_DONT_NEED_TRANS_COMMIT;
6334 return BTRFS_NEED_TRANS_COMMIT;
6336 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6338 return BTRFS_NEED_TRANS_COMMIT;
6339 return BTRFS_DONT_NEED_TRANS_COMMIT;
6343 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6344 LOG_INODE_EXISTS, ctx);
6345 if (ret == BTRFS_NO_LOG_SYNC)
6346 return BTRFS_DONT_NEED_LOG_SYNC;
6348 return BTRFS_NEED_TRANS_COMMIT;
6350 return BTRFS_NEED_LOG_SYNC;