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(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) {
708 struct btrfs_ref ref = { 0 };
711 LIST_HEAD(ordered_sums);
714 * is this extent already allocated in the extent
715 * allocation tree? If so, just add a reference
717 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
720 btrfs_init_generic_ref(&ref,
721 BTRFS_ADD_DELAYED_REF,
722 ins.objectid, ins.offset, 0);
723 btrfs_init_data_ref(&ref,
724 root->root_key.objectid,
725 key->objectid, offset);
726 ret = btrfs_inc_extent_ref(trans, &ref);
731 * insert the extent pointer in the extent
734 ret = btrfs_alloc_logged_file_extent(trans,
735 root->root_key.objectid,
736 key->objectid, offset, &ins);
740 btrfs_release_path(path);
742 if (btrfs_file_extent_compression(eb, item)) {
743 csum_start = ins.objectid;
744 csum_end = csum_start + ins.offset;
746 csum_start = ins.objectid +
747 btrfs_file_extent_offset(eb, item);
748 csum_end = csum_start +
749 btrfs_file_extent_num_bytes(eb, item);
752 ret = btrfs_lookup_csums_range(root->log_root,
753 csum_start, csum_end - 1,
758 * Now delete all existing cums in the csum root that
759 * cover our range. We do this because we can have an
760 * extent that is completely referenced by one file
761 * extent item and partially referenced by another
762 * file extent item (like after using the clone or
763 * extent_same ioctls). In this case if we end up doing
764 * the replay of the one that partially references the
765 * extent first, and we do not do the csum deletion
766 * below, we can get 2 csum items in the csum tree that
767 * overlap each other. For example, imagine our log has
768 * the two following file extent items:
770 * key (257 EXTENT_DATA 409600)
771 * extent data disk byte 12845056 nr 102400
772 * extent data offset 20480 nr 20480 ram 102400
774 * key (257 EXTENT_DATA 819200)
775 * extent data disk byte 12845056 nr 102400
776 * extent data offset 0 nr 102400 ram 102400
778 * Where the second one fully references the 100K extent
779 * that starts at disk byte 12845056, and the log tree
780 * has a single csum item that covers the entire range
783 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
785 * After the first file extent item is replayed, the
786 * csum tree gets the following csum item:
788 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
790 * Which covers the 20K sub-range starting at offset 20K
791 * of our extent. Now when we replay the second file
792 * extent item, if we do not delete existing csum items
793 * that cover any of its blocks, we end up getting two
794 * csum items in our csum tree that overlap each other:
796 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
797 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
799 * Which is a problem, because after this anyone trying
800 * to lookup up for the checksum of any block of our
801 * extent starting at an offset of 40K or higher, will
802 * end up looking at the second csum item only, which
803 * does not contain the checksum for any block starting
804 * at offset 40K or higher of our extent.
806 while (!list_empty(&ordered_sums)) {
807 struct btrfs_ordered_sum *sums;
808 sums = list_entry(ordered_sums.next,
809 struct btrfs_ordered_sum,
812 ret = btrfs_del_csums(trans, fs_info,
816 ret = btrfs_csum_file_blocks(trans,
817 fs_info->csum_root, sums);
818 list_del(&sums->list);
824 btrfs_release_path(path);
826 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
827 /* inline extents are easy, we just overwrite them */
828 ret = overwrite_item(trans, root, path, eb, slot, key);
833 inode_add_bytes(inode, nbytes);
835 ret = btrfs_update_inode(trans, root, inode);
843 * when cleaning up conflicts between the directory names in the
844 * subvolume, directory names in the log and directory names in the
845 * inode back references, we may have to unlink inodes from directories.
847 * This is a helper function to do the unlink of a specific directory
850 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
851 struct btrfs_root *root,
852 struct btrfs_path *path,
853 struct btrfs_inode *dir,
854 struct btrfs_dir_item *di)
859 struct extent_buffer *leaf;
860 struct btrfs_key location;
863 leaf = path->nodes[0];
865 btrfs_dir_item_key_to_cpu(leaf, di, &location);
866 name_len = btrfs_dir_name_len(leaf, di);
867 name = kmalloc(name_len, GFP_NOFS);
871 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
872 btrfs_release_path(path);
874 inode = read_one_inode(root, location.objectid);
880 ret = link_to_fixup_dir(trans, root, path, location.objectid);
884 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
889 ret = btrfs_run_delayed_items(trans);
897 * helper function to see if a given name and sequence number found
898 * in an inode back reference are already in a directory and correctly
899 * point to this inode
901 static noinline int inode_in_dir(struct btrfs_root *root,
902 struct btrfs_path *path,
903 u64 dirid, u64 objectid, u64 index,
904 const char *name, int name_len)
906 struct btrfs_dir_item *di;
907 struct btrfs_key location;
910 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
911 index, name, name_len, 0);
912 if (di && !IS_ERR(di)) {
913 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
914 if (location.objectid != objectid)
918 btrfs_release_path(path);
920 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
921 if (di && !IS_ERR(di)) {
922 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
923 if (location.objectid != objectid)
929 btrfs_release_path(path);
934 * helper function to check a log tree for a named back reference in
935 * an inode. This is used to decide if a back reference that is
936 * found in the subvolume conflicts with what we find in the log.
938 * inode backreferences may have multiple refs in a single item,
939 * during replay we process one reference at a time, and we don't
940 * want to delete valid links to a file from the subvolume if that
941 * link is also in the log.
943 static noinline int backref_in_log(struct btrfs_root *log,
944 struct btrfs_key *key,
946 const char *name, int namelen)
948 struct btrfs_path *path;
949 struct btrfs_inode_ref *ref;
951 unsigned long ptr_end;
952 unsigned long name_ptr;
958 path = btrfs_alloc_path();
962 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
966 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
968 if (key->type == BTRFS_INODE_EXTREF_KEY) {
969 if (btrfs_find_name_in_ext_backref(path->nodes[0],
972 name, namelen, NULL))
978 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
979 ptr_end = ptr + item_size;
980 while (ptr < ptr_end) {
981 ref = (struct btrfs_inode_ref *)ptr;
982 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
983 if (found_name_len == namelen) {
984 name_ptr = (unsigned long)(ref + 1);
985 ret = memcmp_extent_buffer(path->nodes[0], name,
992 ptr = (unsigned long)(ref + 1) + found_name_len;
995 btrfs_free_path(path);
999 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1000 struct btrfs_root *root,
1001 struct btrfs_path *path,
1002 struct btrfs_root *log_root,
1003 struct btrfs_inode *dir,
1004 struct btrfs_inode *inode,
1005 u64 inode_objectid, u64 parent_objectid,
1006 u64 ref_index, char *name, int namelen,
1011 int victim_name_len;
1012 struct extent_buffer *leaf;
1013 struct btrfs_dir_item *di;
1014 struct btrfs_key search_key;
1015 struct btrfs_inode_extref *extref;
1018 /* Search old style refs */
1019 search_key.objectid = inode_objectid;
1020 search_key.type = BTRFS_INODE_REF_KEY;
1021 search_key.offset = parent_objectid;
1022 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1024 struct btrfs_inode_ref *victim_ref;
1026 unsigned long ptr_end;
1028 leaf = path->nodes[0];
1030 /* are we trying to overwrite a back ref for the root directory
1031 * if so, just jump out, we're done
1033 if (search_key.objectid == search_key.offset)
1036 /* check all the names in this back reference to see
1037 * if they are in the log. if so, we allow them to stay
1038 * otherwise they must be unlinked as a conflict
1040 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1041 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1042 while (ptr < ptr_end) {
1043 victim_ref = (struct btrfs_inode_ref *)ptr;
1044 victim_name_len = btrfs_inode_ref_name_len(leaf,
1046 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1050 read_extent_buffer(leaf, victim_name,
1051 (unsigned long)(victim_ref + 1),
1054 if (!backref_in_log(log_root, &search_key,
1058 inc_nlink(&inode->vfs_inode);
1059 btrfs_release_path(path);
1061 ret = btrfs_unlink_inode(trans, root, dir, inode,
1062 victim_name, victim_name_len);
1066 ret = btrfs_run_delayed_items(trans);
1074 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1078 * NOTE: we have searched root tree and checked the
1079 * corresponding ref, it does not need to check again.
1083 btrfs_release_path(path);
1085 /* Same search but for extended refs */
1086 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1087 inode_objectid, parent_objectid, 0,
1089 if (!IS_ERR_OR_NULL(extref)) {
1093 struct inode *victim_parent;
1095 leaf = path->nodes[0];
1097 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1098 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1100 while (cur_offset < item_size) {
1101 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1103 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1105 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1108 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1111 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1114 search_key.objectid = inode_objectid;
1115 search_key.type = BTRFS_INODE_EXTREF_KEY;
1116 search_key.offset = btrfs_extref_hash(parent_objectid,
1120 if (!backref_in_log(log_root, &search_key,
1121 parent_objectid, victim_name,
1124 victim_parent = read_one_inode(root,
1126 if (victim_parent) {
1127 inc_nlink(&inode->vfs_inode);
1128 btrfs_release_path(path);
1130 ret = btrfs_unlink_inode(trans, root,
1131 BTRFS_I(victim_parent),
1136 ret = btrfs_run_delayed_items(
1139 iput(victim_parent);
1148 cur_offset += victim_name_len + sizeof(*extref);
1152 btrfs_release_path(path);
1154 /* look for a conflicting sequence number */
1155 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1156 ref_index, name, namelen, 0);
1157 if (di && !IS_ERR(di)) {
1158 ret = drop_one_dir_item(trans, root, path, dir, di);
1162 btrfs_release_path(path);
1164 /* look for a conflicting name */
1165 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1167 if (di && !IS_ERR(di)) {
1168 ret = drop_one_dir_item(trans, root, path, dir, di);
1172 btrfs_release_path(path);
1177 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1178 u32 *namelen, char **name, u64 *index,
1179 u64 *parent_objectid)
1181 struct btrfs_inode_extref *extref;
1183 extref = (struct btrfs_inode_extref *)ref_ptr;
1185 *namelen = btrfs_inode_extref_name_len(eb, extref);
1186 *name = kmalloc(*namelen, GFP_NOFS);
1190 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1194 *index = btrfs_inode_extref_index(eb, extref);
1195 if (parent_objectid)
1196 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1201 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1202 u32 *namelen, char **name, u64 *index)
1204 struct btrfs_inode_ref *ref;
1206 ref = (struct btrfs_inode_ref *)ref_ptr;
1208 *namelen = btrfs_inode_ref_name_len(eb, ref);
1209 *name = kmalloc(*namelen, GFP_NOFS);
1213 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1216 *index = btrfs_inode_ref_index(eb, ref);
1222 * Take an inode reference item from the log tree and iterate all names from the
1223 * inode reference item in the subvolume tree with the same key (if it exists).
1224 * For any name that is not in the inode reference item from the log tree, do a
1225 * proper unlink of that name (that is, remove its entry from the inode
1226 * reference item and both dir index keys).
1228 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1229 struct btrfs_root *root,
1230 struct btrfs_path *path,
1231 struct btrfs_inode *inode,
1232 struct extent_buffer *log_eb,
1234 struct btrfs_key *key)
1237 unsigned long ref_ptr;
1238 unsigned long ref_end;
1239 struct extent_buffer *eb;
1242 btrfs_release_path(path);
1243 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1251 eb = path->nodes[0];
1252 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1253 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1254 while (ref_ptr < ref_end) {
1259 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1260 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1263 parent_id = key->offset;
1264 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1270 if (key->type == BTRFS_INODE_EXTREF_KEY)
1271 ret = btrfs_find_name_in_ext_backref(log_eb, log_slot,
1275 ret = btrfs_find_name_in_backref(log_eb, log_slot, name,
1281 btrfs_release_path(path);
1282 dir = read_one_inode(root, parent_id);
1288 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1289 inode, name, namelen);
1299 if (key->type == BTRFS_INODE_EXTREF_KEY)
1300 ref_ptr += sizeof(struct btrfs_inode_extref);
1302 ref_ptr += sizeof(struct btrfs_inode_ref);
1306 btrfs_release_path(path);
1310 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1311 const u8 ref_type, const char *name,
1314 struct btrfs_key key;
1315 struct btrfs_path *path;
1316 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1319 path = btrfs_alloc_path();
1323 key.objectid = btrfs_ino(BTRFS_I(inode));
1324 key.type = ref_type;
1325 if (key.type == BTRFS_INODE_REF_KEY)
1326 key.offset = parent_id;
1328 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1330 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1337 if (key.type == BTRFS_INODE_EXTREF_KEY)
1338 ret = btrfs_find_name_in_ext_backref(path->nodes[0],
1339 path->slots[0], parent_id,
1340 name, namelen, NULL);
1342 ret = btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1343 name, namelen, NULL);
1346 btrfs_free_path(path);
1350 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1351 struct inode *dir, struct inode *inode, const char *name,
1352 int namelen, u64 ref_index)
1354 struct btrfs_dir_item *dir_item;
1355 struct btrfs_key key;
1356 struct btrfs_path *path;
1357 struct inode *other_inode = NULL;
1360 path = btrfs_alloc_path();
1364 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1365 btrfs_ino(BTRFS_I(dir)),
1368 btrfs_release_path(path);
1370 } else if (IS_ERR(dir_item)) {
1371 ret = PTR_ERR(dir_item);
1376 * Our inode's dentry collides with the dentry of another inode which is
1377 * in the log but not yet processed since it has a higher inode number.
1378 * So delete that other dentry.
1380 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1381 btrfs_release_path(path);
1382 other_inode = read_one_inode(root, key.objectid);
1387 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1392 * If we dropped the link count to 0, bump it so that later the iput()
1393 * on the inode will not free it. We will fixup the link count later.
1395 if (other_inode->i_nlink == 0)
1396 inc_nlink(other_inode);
1398 ret = btrfs_run_delayed_items(trans);
1402 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1403 name, namelen, 0, ref_index);
1406 btrfs_free_path(path);
1412 * replay one inode back reference item found in the log tree.
1413 * eb, slot and key refer to the buffer and key found in the log tree.
1414 * root is the destination we are replaying into, and path is for temp
1415 * use by this function. (it should be released on return).
1417 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1418 struct btrfs_root *root,
1419 struct btrfs_root *log,
1420 struct btrfs_path *path,
1421 struct extent_buffer *eb, int slot,
1422 struct btrfs_key *key)
1424 struct inode *dir = NULL;
1425 struct inode *inode = NULL;
1426 unsigned long ref_ptr;
1427 unsigned long ref_end;
1431 int search_done = 0;
1432 int log_ref_ver = 0;
1433 u64 parent_objectid;
1436 int ref_struct_size;
1438 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1439 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1441 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1442 struct btrfs_inode_extref *r;
1444 ref_struct_size = sizeof(struct btrfs_inode_extref);
1446 r = (struct btrfs_inode_extref *)ref_ptr;
1447 parent_objectid = btrfs_inode_extref_parent(eb, r);
1449 ref_struct_size = sizeof(struct btrfs_inode_ref);
1450 parent_objectid = key->offset;
1452 inode_objectid = key->objectid;
1455 * it is possible that we didn't log all the parent directories
1456 * for a given inode. If we don't find the dir, just don't
1457 * copy the back ref in. The link count fixup code will take
1460 dir = read_one_inode(root, parent_objectid);
1466 inode = read_one_inode(root, inode_objectid);
1472 while (ref_ptr < ref_end) {
1474 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1475 &ref_index, &parent_objectid);
1477 * parent object can change from one array
1481 dir = read_one_inode(root, parent_objectid);
1487 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1493 /* if we already have a perfect match, we're done */
1494 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1495 btrfs_ino(BTRFS_I(inode)), ref_index,
1498 * look for a conflicting back reference in the
1499 * metadata. if we find one we have to unlink that name
1500 * of the file before we add our new link. Later on, we
1501 * overwrite any existing back reference, and we don't
1502 * want to create dangling pointers in the directory.
1506 ret = __add_inode_ref(trans, root, path, log,
1511 ref_index, name, namelen,
1521 * If a reference item already exists for this inode
1522 * with the same parent and name, but different index,
1523 * drop it and the corresponding directory index entries
1524 * from the parent before adding the new reference item
1525 * and dir index entries, otherwise we would fail with
1526 * -EEXIST returned from btrfs_add_link() below.
1528 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1531 ret = btrfs_unlink_inode(trans, root,
1536 * If we dropped the link count to 0, bump it so
1537 * that later the iput() on the inode will not
1538 * free it. We will fixup the link count later.
1540 if (!ret && inode->i_nlink == 0)
1546 /* insert our name */
1547 ret = add_link(trans, root, dir, inode, name, namelen,
1552 btrfs_update_inode(trans, root, inode);
1555 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1565 * Before we overwrite the inode reference item in the subvolume tree
1566 * with the item from the log tree, we must unlink all names from the
1567 * parent directory that are in the subvolume's tree inode reference
1568 * item, otherwise we end up with an inconsistent subvolume tree where
1569 * dir index entries exist for a name but there is no inode reference
1570 * item with the same name.
1572 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1577 /* finally write the back reference in the inode */
1578 ret = overwrite_item(trans, root, path, eb, slot, key);
1580 btrfs_release_path(path);
1587 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1588 struct btrfs_root *root, u64 ino)
1592 ret = btrfs_insert_orphan_item(trans, root, ino);
1599 static int count_inode_extrefs(struct btrfs_root *root,
1600 struct btrfs_inode *inode, struct btrfs_path *path)
1604 unsigned int nlink = 0;
1607 u64 inode_objectid = btrfs_ino(inode);
1610 struct btrfs_inode_extref *extref;
1611 struct extent_buffer *leaf;
1614 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1619 leaf = path->nodes[0];
1620 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1621 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1624 while (cur_offset < item_size) {
1625 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1626 name_len = btrfs_inode_extref_name_len(leaf, extref);
1630 cur_offset += name_len + sizeof(*extref);
1634 btrfs_release_path(path);
1636 btrfs_release_path(path);
1638 if (ret < 0 && ret != -ENOENT)
1643 static int count_inode_refs(struct btrfs_root *root,
1644 struct btrfs_inode *inode, struct btrfs_path *path)
1647 struct btrfs_key key;
1648 unsigned int nlink = 0;
1650 unsigned long ptr_end;
1652 u64 ino = btrfs_ino(inode);
1655 key.type = BTRFS_INODE_REF_KEY;
1656 key.offset = (u64)-1;
1659 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1663 if (path->slots[0] == 0)
1668 btrfs_item_key_to_cpu(path->nodes[0], &key,
1670 if (key.objectid != ino ||
1671 key.type != BTRFS_INODE_REF_KEY)
1673 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1674 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1676 while (ptr < ptr_end) {
1677 struct btrfs_inode_ref *ref;
1679 ref = (struct btrfs_inode_ref *)ptr;
1680 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1682 ptr = (unsigned long)(ref + 1) + name_len;
1686 if (key.offset == 0)
1688 if (path->slots[0] > 0) {
1693 btrfs_release_path(path);
1695 btrfs_release_path(path);
1701 * There are a few corners where the link count of the file can't
1702 * be properly maintained during replay. So, instead of adding
1703 * lots of complexity to the log code, we just scan the backrefs
1704 * for any file that has been through replay.
1706 * The scan will update the link count on the inode to reflect the
1707 * number of back refs found. If it goes down to zero, the iput
1708 * will free the inode.
1710 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1711 struct btrfs_root *root,
1712 struct inode *inode)
1714 struct btrfs_path *path;
1717 u64 ino = btrfs_ino(BTRFS_I(inode));
1719 path = btrfs_alloc_path();
1723 ret = count_inode_refs(root, BTRFS_I(inode), path);
1729 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1737 if (nlink != inode->i_nlink) {
1738 set_nlink(inode, nlink);
1739 btrfs_update_inode(trans, root, inode);
1741 BTRFS_I(inode)->index_cnt = (u64)-1;
1743 if (inode->i_nlink == 0) {
1744 if (S_ISDIR(inode->i_mode)) {
1745 ret = replay_dir_deletes(trans, root, NULL, path,
1750 ret = insert_orphan_item(trans, root, ino);
1754 btrfs_free_path(path);
1758 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1759 struct btrfs_root *root,
1760 struct btrfs_path *path)
1763 struct btrfs_key key;
1764 struct inode *inode;
1766 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1767 key.type = BTRFS_ORPHAN_ITEM_KEY;
1768 key.offset = (u64)-1;
1770 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1775 if (path->slots[0] == 0)
1780 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1781 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1782 key.type != BTRFS_ORPHAN_ITEM_KEY)
1785 ret = btrfs_del_item(trans, root, path);
1789 btrfs_release_path(path);
1790 inode = read_one_inode(root, key.offset);
1794 ret = fixup_inode_link_count(trans, root, inode);
1800 * fixup on a directory may create new entries,
1801 * make sure we always look for the highset possible
1804 key.offset = (u64)-1;
1808 btrfs_release_path(path);
1814 * record a given inode in the fixup dir so we can check its link
1815 * count when replay is done. The link count is incremented here
1816 * so the inode won't go away until we check it
1818 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1823 struct btrfs_key key;
1825 struct inode *inode;
1827 inode = read_one_inode(root, objectid);
1831 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1832 key.type = BTRFS_ORPHAN_ITEM_KEY;
1833 key.offset = objectid;
1835 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1837 btrfs_release_path(path);
1839 if (!inode->i_nlink)
1840 set_nlink(inode, 1);
1843 ret = btrfs_update_inode(trans, root, inode);
1844 } else if (ret == -EEXIST) {
1847 BUG(); /* Logic Error */
1855 * when replaying the log for a directory, we only insert names
1856 * for inodes that actually exist. This means an fsync on a directory
1857 * does not implicitly fsync all the new files in it
1859 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root,
1861 u64 dirid, u64 index,
1862 char *name, int name_len,
1863 struct btrfs_key *location)
1865 struct inode *inode;
1869 inode = read_one_inode(root, location->objectid);
1873 dir = read_one_inode(root, dirid);
1879 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1880 name_len, 1, index);
1882 /* FIXME, put inode into FIXUP list */
1890 * Return true if an inode reference exists in the log for the given name,
1891 * inode and parent inode.
1893 static bool name_in_log_ref(struct btrfs_root *log_root,
1894 const char *name, const int name_len,
1895 const u64 dirid, const u64 ino)
1897 struct btrfs_key search_key;
1899 search_key.objectid = ino;
1900 search_key.type = BTRFS_INODE_REF_KEY;
1901 search_key.offset = dirid;
1902 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1905 search_key.type = BTRFS_INODE_EXTREF_KEY;
1906 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1907 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1914 * take a single entry in a log directory item and replay it into
1917 * if a conflicting item exists in the subdirectory already,
1918 * the inode it points to is unlinked and put into the link count
1921 * If a name from the log points to a file or directory that does
1922 * not exist in the FS, it is skipped. fsyncs on directories
1923 * do not force down inodes inside that directory, just changes to the
1924 * names or unlinks in a directory.
1926 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1927 * non-existing inode) and 1 if the name was replayed.
1929 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1930 struct btrfs_root *root,
1931 struct btrfs_path *path,
1932 struct extent_buffer *eb,
1933 struct btrfs_dir_item *di,
1934 struct btrfs_key *key)
1938 struct btrfs_dir_item *dst_di;
1939 struct btrfs_key found_key;
1940 struct btrfs_key log_key;
1945 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1946 bool name_added = false;
1948 dir = read_one_inode(root, key->objectid);
1952 name_len = btrfs_dir_name_len(eb, di);
1953 name = kmalloc(name_len, GFP_NOFS);
1959 log_type = btrfs_dir_type(eb, di);
1960 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1963 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1964 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1969 btrfs_release_path(path);
1971 if (key->type == BTRFS_DIR_ITEM_KEY) {
1972 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1974 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1975 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1984 if (IS_ERR_OR_NULL(dst_di)) {
1985 /* we need a sequence number to insert, so we only
1986 * do inserts for the BTRFS_DIR_INDEX_KEY types
1988 if (key->type != BTRFS_DIR_INDEX_KEY)
1993 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1994 /* the existing item matches the logged item */
1995 if (found_key.objectid == log_key.objectid &&
1996 found_key.type == log_key.type &&
1997 found_key.offset == log_key.offset &&
1998 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1999 update_size = false;
2004 * don't drop the conflicting directory entry if the inode
2005 * for the new entry doesn't exist
2010 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2014 if (key->type == BTRFS_DIR_INDEX_KEY)
2017 btrfs_release_path(path);
2018 if (!ret && update_size) {
2019 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2020 ret = btrfs_update_inode(trans, root, dir);
2024 if (!ret && name_added)
2029 if (name_in_log_ref(root->log_root, name, name_len,
2030 key->objectid, log_key.objectid)) {
2031 /* The dentry will be added later. */
2033 update_size = false;
2036 btrfs_release_path(path);
2037 ret = insert_one_name(trans, root, key->objectid, key->offset,
2038 name, name_len, &log_key);
2039 if (ret && ret != -ENOENT && ret != -EEXIST)
2043 update_size = false;
2049 * find all the names in a directory item and reconcile them into
2050 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2051 * one name in a directory item, but the same code gets used for
2052 * both directory index types
2054 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2055 struct btrfs_root *root,
2056 struct btrfs_path *path,
2057 struct extent_buffer *eb, int slot,
2058 struct btrfs_key *key)
2061 u32 item_size = btrfs_item_size_nr(eb, slot);
2062 struct btrfs_dir_item *di;
2065 unsigned long ptr_end;
2066 struct btrfs_path *fixup_path = NULL;
2068 ptr = btrfs_item_ptr_offset(eb, slot);
2069 ptr_end = ptr + item_size;
2070 while (ptr < ptr_end) {
2071 di = (struct btrfs_dir_item *)ptr;
2072 name_len = btrfs_dir_name_len(eb, di);
2073 ret = replay_one_name(trans, root, path, eb, di, key);
2076 ptr = (unsigned long)(di + 1);
2080 * If this entry refers to a non-directory (directories can not
2081 * have a link count > 1) and it was added in the transaction
2082 * that was not committed, make sure we fixup the link count of
2083 * the inode it the entry points to. Otherwise something like
2084 * the following would result in a directory pointing to an
2085 * inode with a wrong link that does not account for this dir
2093 * ln testdir/bar testdir/bar_link
2094 * ln testdir/foo testdir/foo_link
2095 * xfs_io -c "fsync" testdir/bar
2099 * mount fs, log replay happens
2101 * File foo would remain with a link count of 1 when it has two
2102 * entries pointing to it in the directory testdir. This would
2103 * make it impossible to ever delete the parent directory has
2104 * it would result in stale dentries that can never be deleted.
2106 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2107 struct btrfs_key di_key;
2110 fixup_path = btrfs_alloc_path();
2117 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2118 ret = link_to_fixup_dir(trans, root, fixup_path,
2125 btrfs_free_path(fixup_path);
2130 * directory replay has two parts. There are the standard directory
2131 * items in the log copied from the subvolume, and range items
2132 * created in the log while the subvolume was logged.
2134 * The range items tell us which parts of the key space the log
2135 * is authoritative for. During replay, if a key in the subvolume
2136 * directory is in a logged range item, but not actually in the log
2137 * that means it was deleted from the directory before the fsync
2138 * and should be removed.
2140 static noinline int find_dir_range(struct btrfs_root *root,
2141 struct btrfs_path *path,
2142 u64 dirid, int key_type,
2143 u64 *start_ret, u64 *end_ret)
2145 struct btrfs_key key;
2147 struct btrfs_dir_log_item *item;
2151 if (*start_ret == (u64)-1)
2154 key.objectid = dirid;
2155 key.type = key_type;
2156 key.offset = *start_ret;
2158 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2162 if (path->slots[0] == 0)
2167 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2169 if (key.type != key_type || key.objectid != dirid) {
2173 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2174 struct btrfs_dir_log_item);
2175 found_end = btrfs_dir_log_end(path->nodes[0], item);
2177 if (*start_ret >= key.offset && *start_ret <= found_end) {
2179 *start_ret = key.offset;
2180 *end_ret = found_end;
2185 /* check the next slot in the tree to see if it is a valid item */
2186 nritems = btrfs_header_nritems(path->nodes[0]);
2188 if (path->slots[0] >= nritems) {
2189 ret = btrfs_next_leaf(root, path);
2194 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2196 if (key.type != key_type || key.objectid != dirid) {
2200 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2201 struct btrfs_dir_log_item);
2202 found_end = btrfs_dir_log_end(path->nodes[0], item);
2203 *start_ret = key.offset;
2204 *end_ret = found_end;
2207 btrfs_release_path(path);
2212 * this looks for a given directory item in the log. If the directory
2213 * item is not in the log, the item is removed and the inode it points
2216 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2217 struct btrfs_root *root,
2218 struct btrfs_root *log,
2219 struct btrfs_path *path,
2220 struct btrfs_path *log_path,
2222 struct btrfs_key *dir_key)
2225 struct extent_buffer *eb;
2228 struct btrfs_dir_item *di;
2229 struct btrfs_dir_item *log_di;
2232 unsigned long ptr_end;
2234 struct inode *inode;
2235 struct btrfs_key location;
2238 eb = path->nodes[0];
2239 slot = path->slots[0];
2240 item_size = btrfs_item_size_nr(eb, slot);
2241 ptr = btrfs_item_ptr_offset(eb, slot);
2242 ptr_end = ptr + item_size;
2243 while (ptr < ptr_end) {
2244 di = (struct btrfs_dir_item *)ptr;
2245 name_len = btrfs_dir_name_len(eb, di);
2246 name = kmalloc(name_len, GFP_NOFS);
2251 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2254 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2255 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2258 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2259 log_di = btrfs_lookup_dir_index_item(trans, log,
2265 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2266 btrfs_dir_item_key_to_cpu(eb, di, &location);
2267 btrfs_release_path(path);
2268 btrfs_release_path(log_path);
2269 inode = read_one_inode(root, location.objectid);
2275 ret = link_to_fixup_dir(trans, root,
2276 path, location.objectid);
2284 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2285 BTRFS_I(inode), name, name_len);
2287 ret = btrfs_run_delayed_items(trans);
2293 /* there might still be more names under this key
2294 * check and repeat if required
2296 ret = btrfs_search_slot(NULL, root, dir_key, path,
2302 } else if (IS_ERR(log_di)) {
2304 return PTR_ERR(log_di);
2306 btrfs_release_path(log_path);
2309 ptr = (unsigned long)(di + 1);
2314 btrfs_release_path(path);
2315 btrfs_release_path(log_path);
2319 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2320 struct btrfs_root *root,
2321 struct btrfs_root *log,
2322 struct btrfs_path *path,
2325 struct btrfs_key search_key;
2326 struct btrfs_path *log_path;
2331 log_path = btrfs_alloc_path();
2335 search_key.objectid = ino;
2336 search_key.type = BTRFS_XATTR_ITEM_KEY;
2337 search_key.offset = 0;
2339 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2343 nritems = btrfs_header_nritems(path->nodes[0]);
2344 for (i = path->slots[0]; i < nritems; i++) {
2345 struct btrfs_key key;
2346 struct btrfs_dir_item *di;
2347 struct btrfs_dir_item *log_di;
2351 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2352 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2357 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2358 total_size = btrfs_item_size_nr(path->nodes[0], i);
2360 while (cur < total_size) {
2361 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2362 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2363 u32 this_len = sizeof(*di) + name_len + data_len;
2366 name = kmalloc(name_len, GFP_NOFS);
2371 read_extent_buffer(path->nodes[0], name,
2372 (unsigned long)(di + 1), name_len);
2374 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2376 btrfs_release_path(log_path);
2378 /* Doesn't exist in log tree, so delete it. */
2379 btrfs_release_path(path);
2380 di = btrfs_lookup_xattr(trans, root, path, ino,
2381 name, name_len, -1);
2388 ret = btrfs_delete_one_dir_name(trans, root,
2392 btrfs_release_path(path);
2397 if (IS_ERR(log_di)) {
2398 ret = PTR_ERR(log_di);
2402 di = (struct btrfs_dir_item *)((char *)di + this_len);
2405 ret = btrfs_next_leaf(root, path);
2411 btrfs_free_path(log_path);
2412 btrfs_release_path(path);
2418 * deletion replay happens before we copy any new directory items
2419 * out of the log or out of backreferences from inodes. It
2420 * scans the log to find ranges of keys that log is authoritative for,
2421 * and then scans the directory to find items in those ranges that are
2422 * not present in the log.
2424 * Anything we don't find in the log is unlinked and removed from the
2427 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2428 struct btrfs_root *root,
2429 struct btrfs_root *log,
2430 struct btrfs_path *path,
2431 u64 dirid, int del_all)
2435 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2437 struct btrfs_key dir_key;
2438 struct btrfs_key found_key;
2439 struct btrfs_path *log_path;
2442 dir_key.objectid = dirid;
2443 dir_key.type = BTRFS_DIR_ITEM_KEY;
2444 log_path = btrfs_alloc_path();
2448 dir = read_one_inode(root, dirid);
2449 /* it isn't an error if the inode isn't there, that can happen
2450 * because we replay the deletes before we copy in the inode item
2454 btrfs_free_path(log_path);
2462 range_end = (u64)-1;
2464 ret = find_dir_range(log, path, dirid, key_type,
2465 &range_start, &range_end);
2470 dir_key.offset = range_start;
2473 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2478 nritems = btrfs_header_nritems(path->nodes[0]);
2479 if (path->slots[0] >= nritems) {
2480 ret = btrfs_next_leaf(root, path);
2486 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2488 if (found_key.objectid != dirid ||
2489 found_key.type != dir_key.type)
2492 if (found_key.offset > range_end)
2495 ret = check_item_in_log(trans, root, log, path,
2500 if (found_key.offset == (u64)-1)
2502 dir_key.offset = found_key.offset + 1;
2504 btrfs_release_path(path);
2505 if (range_end == (u64)-1)
2507 range_start = range_end + 1;
2512 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2513 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2514 dir_key.type = BTRFS_DIR_INDEX_KEY;
2515 btrfs_release_path(path);
2519 btrfs_release_path(path);
2520 btrfs_free_path(log_path);
2526 * the process_func used to replay items from the log tree. This
2527 * gets called in two different stages. The first stage just looks
2528 * for inodes and makes sure they are all copied into the subvolume.
2530 * The second stage copies all the other item types from the log into
2531 * the subvolume. The two stage approach is slower, but gets rid of
2532 * lots of complexity around inodes referencing other inodes that exist
2533 * only in the log (references come from either directory items or inode
2536 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2537 struct walk_control *wc, u64 gen, int level)
2540 struct btrfs_path *path;
2541 struct btrfs_root *root = wc->replay_dest;
2542 struct btrfs_key key;
2546 ret = btrfs_read_buffer(eb, gen, level, NULL);
2550 level = btrfs_header_level(eb);
2555 path = btrfs_alloc_path();
2559 nritems = btrfs_header_nritems(eb);
2560 for (i = 0; i < nritems; i++) {
2561 btrfs_item_key_to_cpu(eb, &key, i);
2563 /* inode keys are done during the first stage */
2564 if (key.type == BTRFS_INODE_ITEM_KEY &&
2565 wc->stage == LOG_WALK_REPLAY_INODES) {
2566 struct btrfs_inode_item *inode_item;
2569 inode_item = btrfs_item_ptr(eb, i,
2570 struct btrfs_inode_item);
2572 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2573 * and never got linked before the fsync, skip it, as
2574 * replaying it is pointless since it would be deleted
2575 * later. We skip logging tmpfiles, but it's always
2576 * possible we are replaying a log created with a kernel
2577 * that used to log tmpfiles.
2579 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2580 wc->ignore_cur_inode = true;
2583 wc->ignore_cur_inode = false;
2585 ret = replay_xattr_deletes(wc->trans, root, log,
2586 path, key.objectid);
2589 mode = btrfs_inode_mode(eb, inode_item);
2590 if (S_ISDIR(mode)) {
2591 ret = replay_dir_deletes(wc->trans,
2592 root, log, path, key.objectid, 0);
2596 ret = overwrite_item(wc->trans, root, path,
2602 * Before replaying extents, truncate the inode to its
2603 * size. We need to do it now and not after log replay
2604 * because before an fsync we can have prealloc extents
2605 * added beyond the inode's i_size. If we did it after,
2606 * through orphan cleanup for example, we would drop
2607 * those prealloc extents just after replaying them.
2609 if (S_ISREG(mode)) {
2610 struct inode *inode;
2613 inode = read_one_inode(root, key.objectid);
2618 from = ALIGN(i_size_read(inode),
2619 root->fs_info->sectorsize);
2620 ret = btrfs_drop_extents(wc->trans, root, inode,
2623 /* Update the inode's nbytes. */
2624 ret = btrfs_update_inode(wc->trans,
2632 ret = link_to_fixup_dir(wc->trans, root,
2633 path, key.objectid);
2638 if (wc->ignore_cur_inode)
2641 if (key.type == BTRFS_DIR_INDEX_KEY &&
2642 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2643 ret = replay_one_dir_item(wc->trans, root, path,
2649 if (wc->stage < LOG_WALK_REPLAY_ALL)
2652 /* these keys are simply copied */
2653 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2654 ret = overwrite_item(wc->trans, root, path,
2658 } else if (key.type == BTRFS_INODE_REF_KEY ||
2659 key.type == BTRFS_INODE_EXTREF_KEY) {
2660 ret = add_inode_ref(wc->trans, root, log, path,
2662 if (ret && ret != -ENOENT)
2665 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2666 ret = replay_one_extent(wc->trans, root, path,
2670 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2671 ret = replay_one_dir_item(wc->trans, root, path,
2677 btrfs_free_path(path);
2681 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2682 struct btrfs_root *root,
2683 struct btrfs_path *path, int *level,
2684 struct walk_control *wc)
2686 struct btrfs_fs_info *fs_info = root->fs_info;
2690 struct extent_buffer *next;
2691 struct extent_buffer *cur;
2692 struct extent_buffer *parent;
2696 WARN_ON(*level < 0);
2697 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2699 while (*level > 0) {
2700 struct btrfs_key first_key;
2702 WARN_ON(*level < 0);
2703 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2704 cur = path->nodes[*level];
2706 WARN_ON(btrfs_header_level(cur) != *level);
2708 if (path->slots[*level] >=
2709 btrfs_header_nritems(cur))
2712 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2713 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2714 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2715 blocksize = fs_info->nodesize;
2717 parent = path->nodes[*level];
2718 root_owner = btrfs_header_owner(parent);
2720 next = btrfs_find_create_tree_block(fs_info, bytenr);
2722 return PTR_ERR(next);
2725 ret = wc->process_func(root, next, wc, ptr_gen,
2728 free_extent_buffer(next);
2732 path->slots[*level]++;
2734 ret = btrfs_read_buffer(next, ptr_gen,
2735 *level - 1, &first_key);
2737 free_extent_buffer(next);
2742 btrfs_tree_lock(next);
2743 btrfs_set_lock_blocking_write(next);
2744 btrfs_clean_tree_block(next);
2745 btrfs_wait_tree_block_writeback(next);
2746 btrfs_tree_unlock(next);
2748 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2749 clear_extent_buffer_dirty(next);
2752 WARN_ON(root_owner !=
2753 BTRFS_TREE_LOG_OBJECTID);
2754 ret = btrfs_free_and_pin_reserved_extent(
2758 free_extent_buffer(next);
2762 free_extent_buffer(next);
2765 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2767 free_extent_buffer(next);
2771 WARN_ON(*level <= 0);
2772 if (path->nodes[*level-1])
2773 free_extent_buffer(path->nodes[*level-1]);
2774 path->nodes[*level-1] = next;
2775 *level = btrfs_header_level(next);
2776 path->slots[*level] = 0;
2779 WARN_ON(*level < 0);
2780 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2782 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2788 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2789 struct btrfs_root *root,
2790 struct btrfs_path *path, int *level,
2791 struct walk_control *wc)
2793 struct btrfs_fs_info *fs_info = root->fs_info;
2799 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2800 slot = path->slots[i];
2801 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2804 WARN_ON(*level == 0);
2807 struct extent_buffer *parent;
2808 if (path->nodes[*level] == root->node)
2809 parent = path->nodes[*level];
2811 parent = path->nodes[*level + 1];
2813 root_owner = btrfs_header_owner(parent);
2814 ret = wc->process_func(root, path->nodes[*level], wc,
2815 btrfs_header_generation(path->nodes[*level]),
2821 struct extent_buffer *next;
2823 next = path->nodes[*level];
2826 btrfs_tree_lock(next);
2827 btrfs_set_lock_blocking_write(next);
2828 btrfs_clean_tree_block(next);
2829 btrfs_wait_tree_block_writeback(next);
2830 btrfs_tree_unlock(next);
2832 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2833 clear_extent_buffer_dirty(next);
2836 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2837 ret = btrfs_free_and_pin_reserved_extent(
2839 path->nodes[*level]->start,
2840 path->nodes[*level]->len);
2844 free_extent_buffer(path->nodes[*level]);
2845 path->nodes[*level] = NULL;
2853 * drop the reference count on the tree rooted at 'snap'. This traverses
2854 * the tree freeing any blocks that have a ref count of zero after being
2857 static int walk_log_tree(struct btrfs_trans_handle *trans,
2858 struct btrfs_root *log, struct walk_control *wc)
2860 struct btrfs_fs_info *fs_info = log->fs_info;
2864 struct btrfs_path *path;
2867 path = btrfs_alloc_path();
2871 level = btrfs_header_level(log->node);
2873 path->nodes[level] = log->node;
2874 extent_buffer_get(log->node);
2875 path->slots[level] = 0;
2878 wret = walk_down_log_tree(trans, log, path, &level, wc);
2886 wret = walk_up_log_tree(trans, log, path, &level, wc);
2895 /* was the root node processed? if not, catch it here */
2896 if (path->nodes[orig_level]) {
2897 ret = wc->process_func(log, path->nodes[orig_level], wc,
2898 btrfs_header_generation(path->nodes[orig_level]),
2903 struct extent_buffer *next;
2905 next = path->nodes[orig_level];
2908 btrfs_tree_lock(next);
2909 btrfs_set_lock_blocking_write(next);
2910 btrfs_clean_tree_block(next);
2911 btrfs_wait_tree_block_writeback(next);
2912 btrfs_tree_unlock(next);
2914 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2915 clear_extent_buffer_dirty(next);
2918 WARN_ON(log->root_key.objectid !=
2919 BTRFS_TREE_LOG_OBJECTID);
2920 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2921 next->start, next->len);
2928 btrfs_free_path(path);
2933 * helper function to update the item for a given subvolumes log root
2934 * in the tree of log roots
2936 static int update_log_root(struct btrfs_trans_handle *trans,
2937 struct btrfs_root *log)
2939 struct btrfs_fs_info *fs_info = log->fs_info;
2942 if (log->log_transid == 1) {
2943 /* insert root item on the first sync */
2944 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2945 &log->root_key, &log->root_item);
2947 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2948 &log->root_key, &log->root_item);
2953 static void wait_log_commit(struct btrfs_root *root, int transid)
2956 int index = transid % 2;
2959 * we only allow two pending log transactions at a time,
2960 * so we know that if ours is more than 2 older than the
2961 * current transaction, we're done
2964 prepare_to_wait(&root->log_commit_wait[index],
2965 &wait, TASK_UNINTERRUPTIBLE);
2967 if (!(root->log_transid_committed < transid &&
2968 atomic_read(&root->log_commit[index])))
2971 mutex_unlock(&root->log_mutex);
2973 mutex_lock(&root->log_mutex);
2975 finish_wait(&root->log_commit_wait[index], &wait);
2978 static void wait_for_writer(struct btrfs_root *root)
2983 prepare_to_wait(&root->log_writer_wait, &wait,
2984 TASK_UNINTERRUPTIBLE);
2985 if (!atomic_read(&root->log_writers))
2988 mutex_unlock(&root->log_mutex);
2990 mutex_lock(&root->log_mutex);
2992 finish_wait(&root->log_writer_wait, &wait);
2995 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2996 struct btrfs_log_ctx *ctx)
3001 mutex_lock(&root->log_mutex);
3002 list_del_init(&ctx->list);
3003 mutex_unlock(&root->log_mutex);
3007 * Invoked in log mutex context, or be sure there is no other task which
3008 * can access the list.
3010 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3011 int index, int error)
3013 struct btrfs_log_ctx *ctx;
3014 struct btrfs_log_ctx *safe;
3016 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3017 list_del_init(&ctx->list);
3018 ctx->log_ret = error;
3021 INIT_LIST_HEAD(&root->log_ctxs[index]);
3025 * btrfs_sync_log does sends a given tree log down to the disk and
3026 * updates the super blocks to record it. When this call is done,
3027 * you know that any inodes previously logged are safely on disk only
3030 * Any other return value means you need to call btrfs_commit_transaction.
3031 * Some of the edge cases for fsyncing directories that have had unlinks
3032 * or renames done in the past mean that sometimes the only safe
3033 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3034 * that has happened.
3036 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3037 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3043 struct btrfs_fs_info *fs_info = root->fs_info;
3044 struct btrfs_root *log = root->log_root;
3045 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3046 int log_transid = 0;
3047 struct btrfs_log_ctx root_log_ctx;
3048 struct blk_plug plug;
3050 mutex_lock(&root->log_mutex);
3051 log_transid = ctx->log_transid;
3052 if (root->log_transid_committed >= log_transid) {
3053 mutex_unlock(&root->log_mutex);
3054 return ctx->log_ret;
3057 index1 = log_transid % 2;
3058 if (atomic_read(&root->log_commit[index1])) {
3059 wait_log_commit(root, log_transid);
3060 mutex_unlock(&root->log_mutex);
3061 return ctx->log_ret;
3063 ASSERT(log_transid == root->log_transid);
3064 atomic_set(&root->log_commit[index1], 1);
3066 /* wait for previous tree log sync to complete */
3067 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3068 wait_log_commit(root, log_transid - 1);
3071 int batch = atomic_read(&root->log_batch);
3072 /* when we're on an ssd, just kick the log commit out */
3073 if (!btrfs_test_opt(fs_info, SSD) &&
3074 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3075 mutex_unlock(&root->log_mutex);
3076 schedule_timeout_uninterruptible(1);
3077 mutex_lock(&root->log_mutex);
3079 wait_for_writer(root);
3080 if (batch == atomic_read(&root->log_batch))
3084 /* bail out if we need to do a full commit */
3085 if (btrfs_need_log_full_commit(trans)) {
3087 mutex_unlock(&root->log_mutex);
3091 if (log_transid % 2 == 0)
3092 mark = EXTENT_DIRTY;
3096 /* we start IO on all the marked extents here, but we don't actually
3097 * wait for them until later.
3099 blk_start_plug(&plug);
3100 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3102 blk_finish_plug(&plug);
3103 btrfs_abort_transaction(trans, ret);
3104 btrfs_set_log_full_commit(trans);
3105 mutex_unlock(&root->log_mutex);
3109 btrfs_set_root_node(&log->root_item, log->node);
3111 root->log_transid++;
3112 log->log_transid = root->log_transid;
3113 root->log_start_pid = 0;
3115 * IO has been started, blocks of the log tree have WRITTEN flag set
3116 * in their headers. new modifications of the log will be written to
3117 * new positions. so it's safe to allow log writers to go in.
3119 mutex_unlock(&root->log_mutex);
3121 btrfs_init_log_ctx(&root_log_ctx, NULL);
3123 mutex_lock(&log_root_tree->log_mutex);
3124 atomic_inc(&log_root_tree->log_batch);
3125 atomic_inc(&log_root_tree->log_writers);
3127 index2 = log_root_tree->log_transid % 2;
3128 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3129 root_log_ctx.log_transid = log_root_tree->log_transid;
3131 mutex_unlock(&log_root_tree->log_mutex);
3133 ret = update_log_root(trans, log);
3135 mutex_lock(&log_root_tree->log_mutex);
3136 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3137 /* atomic_dec_and_test implies a barrier */
3138 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3142 if (!list_empty(&root_log_ctx.list))
3143 list_del_init(&root_log_ctx.list);
3145 blk_finish_plug(&plug);
3146 btrfs_set_log_full_commit(trans);
3148 if (ret != -ENOSPC) {
3149 btrfs_abort_transaction(trans, ret);
3150 mutex_unlock(&log_root_tree->log_mutex);
3153 btrfs_wait_tree_log_extents(log, mark);
3154 mutex_unlock(&log_root_tree->log_mutex);
3159 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3160 blk_finish_plug(&plug);
3161 list_del_init(&root_log_ctx.list);
3162 mutex_unlock(&log_root_tree->log_mutex);
3163 ret = root_log_ctx.log_ret;
3167 index2 = root_log_ctx.log_transid % 2;
3168 if (atomic_read(&log_root_tree->log_commit[index2])) {
3169 blk_finish_plug(&plug);
3170 ret = btrfs_wait_tree_log_extents(log, mark);
3171 wait_log_commit(log_root_tree,
3172 root_log_ctx.log_transid);
3173 mutex_unlock(&log_root_tree->log_mutex);
3175 ret = root_log_ctx.log_ret;
3178 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3179 atomic_set(&log_root_tree->log_commit[index2], 1);
3181 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3182 wait_log_commit(log_root_tree,
3183 root_log_ctx.log_transid - 1);
3186 wait_for_writer(log_root_tree);
3189 * now that we've moved on to the tree of log tree roots,
3190 * check the full commit flag again
3192 if (btrfs_need_log_full_commit(trans)) {
3193 blk_finish_plug(&plug);
3194 btrfs_wait_tree_log_extents(log, mark);
3195 mutex_unlock(&log_root_tree->log_mutex);
3197 goto out_wake_log_root;
3200 ret = btrfs_write_marked_extents(fs_info,
3201 &log_root_tree->dirty_log_pages,
3202 EXTENT_DIRTY | EXTENT_NEW);
3203 blk_finish_plug(&plug);
3205 btrfs_set_log_full_commit(trans);
3206 btrfs_abort_transaction(trans, ret);
3207 mutex_unlock(&log_root_tree->log_mutex);
3208 goto out_wake_log_root;
3210 ret = btrfs_wait_tree_log_extents(log, mark);
3212 ret = btrfs_wait_tree_log_extents(log_root_tree,
3213 EXTENT_NEW | EXTENT_DIRTY);
3215 btrfs_set_log_full_commit(trans);
3216 mutex_unlock(&log_root_tree->log_mutex);
3217 goto out_wake_log_root;
3220 btrfs_set_super_log_root(fs_info->super_for_commit,
3221 log_root_tree->node->start);
3222 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3223 btrfs_header_level(log_root_tree->node));
3225 log_root_tree->log_transid++;
3226 mutex_unlock(&log_root_tree->log_mutex);
3229 * Nobody else is going to jump in and write the ctree
3230 * super here because the log_commit atomic below is protecting
3231 * us. We must be called with a transaction handle pinning
3232 * the running transaction open, so a full commit can't hop
3233 * in and cause problems either.
3235 ret = write_all_supers(fs_info, 1);
3237 btrfs_set_log_full_commit(trans);
3238 btrfs_abort_transaction(trans, ret);
3239 goto out_wake_log_root;
3242 mutex_lock(&root->log_mutex);
3243 if (root->last_log_commit < log_transid)
3244 root->last_log_commit = log_transid;
3245 mutex_unlock(&root->log_mutex);
3248 mutex_lock(&log_root_tree->log_mutex);
3249 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3251 log_root_tree->log_transid_committed++;
3252 atomic_set(&log_root_tree->log_commit[index2], 0);
3253 mutex_unlock(&log_root_tree->log_mutex);
3256 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3257 * all the updates above are seen by the woken threads. It might not be
3258 * necessary, but proving that seems to be hard.
3260 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3262 mutex_lock(&root->log_mutex);
3263 btrfs_remove_all_log_ctxs(root, index1, ret);
3264 root->log_transid_committed++;
3265 atomic_set(&root->log_commit[index1], 0);
3266 mutex_unlock(&root->log_mutex);
3269 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3270 * all the updates above are seen by the woken threads. It might not be
3271 * necessary, but proving that seems to be hard.
3273 cond_wake_up(&root->log_commit_wait[index1]);
3277 static void free_log_tree(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *log)
3281 struct walk_control wc = {
3283 .process_func = process_one_buffer
3286 ret = walk_log_tree(trans, log, &wc);
3289 btrfs_abort_transaction(trans, ret);
3291 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3294 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3295 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3296 free_extent_buffer(log->node);
3301 * free all the extents used by the tree log. This should be called
3302 * at commit time of the full transaction
3304 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3306 if (root->log_root) {
3307 free_log_tree(trans, root->log_root);
3308 root->log_root = NULL;
3313 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3314 struct btrfs_fs_info *fs_info)
3316 if (fs_info->log_root_tree) {
3317 free_log_tree(trans, fs_info->log_root_tree);
3318 fs_info->log_root_tree = NULL;
3324 * If both a file and directory are logged, and unlinks or renames are
3325 * mixed in, we have a few interesting corners:
3327 * create file X in dir Y
3328 * link file X to X.link in dir Y
3330 * unlink file X but leave X.link
3333 * After a crash we would expect only X.link to exist. But file X
3334 * didn't get fsync'd again so the log has back refs for X and X.link.
3336 * We solve this by removing directory entries and inode backrefs from the
3337 * log when a file that was logged in the current transaction is
3338 * unlinked. Any later fsync will include the updated log entries, and
3339 * we'll be able to reconstruct the proper directory items from backrefs.
3341 * This optimizations allows us to avoid relogging the entire inode
3342 * or the entire directory.
3344 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3345 struct btrfs_root *root,
3346 const char *name, int name_len,
3347 struct btrfs_inode *dir, u64 index)
3349 struct btrfs_root *log;
3350 struct btrfs_dir_item *di;
3351 struct btrfs_path *path;
3355 u64 dir_ino = btrfs_ino(dir);
3357 if (dir->logged_trans < trans->transid)
3360 ret = join_running_log_trans(root);
3364 mutex_lock(&dir->log_mutex);
3366 log = root->log_root;
3367 path = btrfs_alloc_path();
3373 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3374 name, name_len, -1);
3380 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3381 bytes_del += name_len;
3387 btrfs_release_path(path);
3388 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3389 index, name, name_len, -1);
3395 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3396 bytes_del += name_len;
3403 /* update the directory size in the log to reflect the names
3407 struct btrfs_key key;
3409 key.objectid = dir_ino;
3411 key.type = BTRFS_INODE_ITEM_KEY;
3412 btrfs_release_path(path);
3414 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3420 struct btrfs_inode_item *item;
3423 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3424 struct btrfs_inode_item);
3425 i_size = btrfs_inode_size(path->nodes[0], item);
3426 if (i_size > bytes_del)
3427 i_size -= bytes_del;
3430 btrfs_set_inode_size(path->nodes[0], item, i_size);
3431 btrfs_mark_buffer_dirty(path->nodes[0]);
3434 btrfs_release_path(path);
3437 btrfs_free_path(path);
3439 mutex_unlock(&dir->log_mutex);
3440 if (ret == -ENOSPC) {
3441 btrfs_set_log_full_commit(trans);
3444 btrfs_abort_transaction(trans, ret);
3446 btrfs_end_log_trans(root);
3451 /* see comments for btrfs_del_dir_entries_in_log */
3452 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3453 struct btrfs_root *root,
3454 const char *name, int name_len,
3455 struct btrfs_inode *inode, u64 dirid)
3457 struct btrfs_root *log;
3461 if (inode->logged_trans < trans->transid)
3464 ret = join_running_log_trans(root);
3467 log = root->log_root;
3468 mutex_lock(&inode->log_mutex);
3470 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3472 mutex_unlock(&inode->log_mutex);
3473 if (ret == -ENOSPC) {
3474 btrfs_set_log_full_commit(trans);
3476 } else if (ret < 0 && ret != -ENOENT)
3477 btrfs_abort_transaction(trans, ret);
3478 btrfs_end_log_trans(root);
3484 * creates a range item in the log for 'dirid'. first_offset and
3485 * last_offset tell us which parts of the key space the log should
3486 * be considered authoritative for.
3488 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *log,
3490 struct btrfs_path *path,
3491 int key_type, u64 dirid,
3492 u64 first_offset, u64 last_offset)
3495 struct btrfs_key key;
3496 struct btrfs_dir_log_item *item;
3498 key.objectid = dirid;
3499 key.offset = first_offset;
3500 if (key_type == BTRFS_DIR_ITEM_KEY)
3501 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3503 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3504 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3508 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3509 struct btrfs_dir_log_item);
3510 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3511 btrfs_mark_buffer_dirty(path->nodes[0]);
3512 btrfs_release_path(path);
3517 * log all the items included in the current transaction for a given
3518 * directory. This also creates the range items in the log tree required
3519 * to replay anything deleted before the fsync
3521 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3522 struct btrfs_root *root, struct btrfs_inode *inode,
3523 struct btrfs_path *path,
3524 struct btrfs_path *dst_path, int key_type,
3525 struct btrfs_log_ctx *ctx,
3526 u64 min_offset, u64 *last_offset_ret)
3528 struct btrfs_key min_key;
3529 struct btrfs_root *log = root->log_root;
3530 struct extent_buffer *src;
3535 u64 first_offset = min_offset;
3536 u64 last_offset = (u64)-1;
3537 u64 ino = btrfs_ino(inode);
3539 log = root->log_root;
3541 min_key.objectid = ino;
3542 min_key.type = key_type;
3543 min_key.offset = min_offset;
3545 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3548 * we didn't find anything from this transaction, see if there
3549 * is anything at all
3551 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3552 min_key.objectid = ino;
3553 min_key.type = key_type;
3554 min_key.offset = (u64)-1;
3555 btrfs_release_path(path);
3556 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3558 btrfs_release_path(path);
3561 ret = btrfs_previous_item(root, path, ino, key_type);
3563 /* if ret == 0 there are items for this type,
3564 * create a range to tell us the last key of this type.
3565 * otherwise, there are no items in this directory after
3566 * *min_offset, and we create a range to indicate that.
3569 struct btrfs_key tmp;
3570 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3572 if (key_type == tmp.type)
3573 first_offset = max(min_offset, tmp.offset) + 1;
3578 /* go backward to find any previous key */
3579 ret = btrfs_previous_item(root, path, ino, key_type);
3581 struct btrfs_key tmp;
3582 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3583 if (key_type == tmp.type) {
3584 first_offset = tmp.offset;
3585 ret = overwrite_item(trans, log, dst_path,
3586 path->nodes[0], path->slots[0],
3594 btrfs_release_path(path);
3597 * Find the first key from this transaction again. See the note for
3598 * log_new_dir_dentries, if we're logging a directory recursively we
3599 * won't be holding its i_mutex, which means we can modify the directory
3600 * while we're logging it. If we remove an entry between our first
3601 * search and this search we'll not find the key again and can just
3604 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3609 * we have a block from this transaction, log every item in it
3610 * from our directory
3613 struct btrfs_key tmp;
3614 src = path->nodes[0];
3615 nritems = btrfs_header_nritems(src);
3616 for (i = path->slots[0]; i < nritems; i++) {
3617 struct btrfs_dir_item *di;
3619 btrfs_item_key_to_cpu(src, &min_key, i);
3621 if (min_key.objectid != ino || min_key.type != key_type)
3623 ret = overwrite_item(trans, log, dst_path, src, i,
3631 * We must make sure that when we log a directory entry,
3632 * the corresponding inode, after log replay, has a
3633 * matching link count. For example:
3639 * xfs_io -c "fsync" mydir
3641 * <mount fs and log replay>
3643 * Would result in a fsync log that when replayed, our
3644 * file inode would have a link count of 1, but we get
3645 * two directory entries pointing to the same inode.
3646 * After removing one of the names, it would not be
3647 * possible to remove the other name, which resulted
3648 * always in stale file handle errors, and would not
3649 * be possible to rmdir the parent directory, since
3650 * its i_size could never decrement to the value
3651 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3653 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3654 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3656 (btrfs_dir_transid(src, di) == trans->transid ||
3657 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3658 tmp.type != BTRFS_ROOT_ITEM_KEY)
3659 ctx->log_new_dentries = true;
3661 path->slots[0] = nritems;
3664 * look ahead to the next item and see if it is also
3665 * from this directory and from this transaction
3667 ret = btrfs_next_leaf(root, path);
3670 last_offset = (u64)-1;
3675 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3676 if (tmp.objectid != ino || tmp.type != key_type) {
3677 last_offset = (u64)-1;
3680 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3681 ret = overwrite_item(trans, log, dst_path,
3682 path->nodes[0], path->slots[0],
3687 last_offset = tmp.offset;
3692 btrfs_release_path(path);
3693 btrfs_release_path(dst_path);
3696 *last_offset_ret = last_offset;
3698 * insert the log range keys to indicate where the log
3701 ret = insert_dir_log_key(trans, log, path, key_type,
3702 ino, first_offset, last_offset);
3710 * logging directories is very similar to logging inodes, We find all the items
3711 * from the current transaction and write them to the log.
3713 * The recovery code scans the directory in the subvolume, and if it finds a
3714 * key in the range logged that is not present in the log tree, then it means
3715 * that dir entry was unlinked during the transaction.
3717 * In order for that scan to work, we must include one key smaller than
3718 * the smallest logged by this transaction and one key larger than the largest
3719 * key logged by this transaction.
3721 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3722 struct btrfs_root *root, struct btrfs_inode *inode,
3723 struct btrfs_path *path,
3724 struct btrfs_path *dst_path,
3725 struct btrfs_log_ctx *ctx)
3730 int key_type = BTRFS_DIR_ITEM_KEY;
3736 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3737 ctx, min_key, &max_key);
3740 if (max_key == (u64)-1)
3742 min_key = max_key + 1;
3745 if (key_type == BTRFS_DIR_ITEM_KEY) {
3746 key_type = BTRFS_DIR_INDEX_KEY;
3753 * a helper function to drop items from the log before we relog an
3754 * inode. max_key_type indicates the highest item type to remove.
3755 * This cannot be run for file data extents because it does not
3756 * free the extents they point to.
3758 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3759 struct btrfs_root *log,
3760 struct btrfs_path *path,
3761 u64 objectid, int max_key_type)
3764 struct btrfs_key key;
3765 struct btrfs_key found_key;
3768 key.objectid = objectid;
3769 key.type = max_key_type;
3770 key.offset = (u64)-1;
3773 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3774 BUG_ON(ret == 0); /* Logic error */
3778 if (path->slots[0] == 0)
3782 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3785 if (found_key.objectid != objectid)
3788 found_key.offset = 0;
3790 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3795 ret = btrfs_del_items(trans, log, path, start_slot,
3796 path->slots[0] - start_slot + 1);
3798 * If start slot isn't 0 then we don't need to re-search, we've
3799 * found the last guy with the objectid in this tree.
3801 if (ret || start_slot != 0)
3803 btrfs_release_path(path);
3805 btrfs_release_path(path);
3811 static void fill_inode_item(struct btrfs_trans_handle *trans,
3812 struct extent_buffer *leaf,
3813 struct btrfs_inode_item *item,
3814 struct inode *inode, int log_inode_only,
3817 struct btrfs_map_token token;
3819 btrfs_init_map_token(&token);
3821 if (log_inode_only) {
3822 /* set the generation to zero so the recover code
3823 * can tell the difference between an logging
3824 * just to say 'this inode exists' and a logging
3825 * to say 'update this inode with these values'
3827 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3828 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3830 btrfs_set_token_inode_generation(leaf, item,
3831 BTRFS_I(inode)->generation,
3833 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3836 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3837 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3838 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3839 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3841 btrfs_set_token_timespec_sec(leaf, &item->atime,
3842 inode->i_atime.tv_sec, &token);
3843 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3844 inode->i_atime.tv_nsec, &token);
3846 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3847 inode->i_mtime.tv_sec, &token);
3848 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3849 inode->i_mtime.tv_nsec, &token);
3851 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3852 inode->i_ctime.tv_sec, &token);
3853 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3854 inode->i_ctime.tv_nsec, &token);
3856 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3859 btrfs_set_token_inode_sequence(leaf, item,
3860 inode_peek_iversion(inode), &token);
3861 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3862 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3863 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3864 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3867 static int log_inode_item(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *log, struct btrfs_path *path,
3869 struct btrfs_inode *inode)
3871 struct btrfs_inode_item *inode_item;
3874 ret = btrfs_insert_empty_item(trans, log, path,
3875 &inode->location, sizeof(*inode_item));
3876 if (ret && ret != -EEXIST)
3878 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3879 struct btrfs_inode_item);
3880 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3882 btrfs_release_path(path);
3886 static noinline int copy_items(struct btrfs_trans_handle *trans,
3887 struct btrfs_inode *inode,
3888 struct btrfs_path *dst_path,
3889 struct btrfs_path *src_path, u64 *last_extent,
3890 int start_slot, int nr, int inode_only,
3893 struct btrfs_fs_info *fs_info = trans->fs_info;
3894 unsigned long src_offset;
3895 unsigned long dst_offset;
3896 struct btrfs_root *log = inode->root->log_root;
3897 struct btrfs_file_extent_item *extent;
3898 struct btrfs_inode_item *inode_item;
3899 struct extent_buffer *src = src_path->nodes[0];
3900 struct btrfs_key first_key, last_key, key;
3902 struct btrfs_key *ins_keys;
3906 struct list_head ordered_sums;
3907 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3908 bool has_extents = false;
3909 bool need_find_last_extent = true;
3912 INIT_LIST_HEAD(&ordered_sums);
3914 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3915 nr * sizeof(u32), GFP_NOFS);
3919 first_key.objectid = (u64)-1;
3921 ins_sizes = (u32 *)ins_data;
3922 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3924 for (i = 0; i < nr; i++) {
3925 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3926 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3928 ret = btrfs_insert_empty_items(trans, log, dst_path,
3929 ins_keys, ins_sizes, nr);
3935 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3936 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3937 dst_path->slots[0]);
3939 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3942 last_key = ins_keys[i];
3944 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3945 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3947 struct btrfs_inode_item);
3948 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3950 inode_only == LOG_INODE_EXISTS,
3953 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3954 src_offset, ins_sizes[i]);
3958 * We set need_find_last_extent here in case we know we were
3959 * processing other items and then walk into the first extent in
3960 * the inode. If we don't hit an extent then nothing changes,
3961 * we'll do the last search the next time around.
3963 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3965 if (first_key.objectid == (u64)-1)
3966 first_key = ins_keys[i];
3968 need_find_last_extent = false;
3971 /* take a reference on file data extents so that truncates
3972 * or deletes of this inode don't have to relog the inode
3975 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3978 extent = btrfs_item_ptr(src, start_slot + i,
3979 struct btrfs_file_extent_item);
3981 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3984 found_type = btrfs_file_extent_type(src, extent);
3985 if (found_type == BTRFS_FILE_EXTENT_REG) {
3987 ds = btrfs_file_extent_disk_bytenr(src,
3989 /* ds == 0 is a hole */
3993 dl = btrfs_file_extent_disk_num_bytes(src,
3995 cs = btrfs_file_extent_offset(src, extent);
3996 cl = btrfs_file_extent_num_bytes(src,
3998 if (btrfs_file_extent_compression(src,
4004 ret = btrfs_lookup_csums_range(
4006 ds + cs, ds + cs + cl - 1,
4009 btrfs_release_path(dst_path);
4017 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4018 btrfs_release_path(dst_path);
4022 * we have to do this after the loop above to avoid changing the
4023 * log tree while trying to change the log tree.
4026 while (!list_empty(&ordered_sums)) {
4027 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4028 struct btrfs_ordered_sum,
4031 ret = btrfs_csum_file_blocks(trans, log, sums);
4032 list_del(&sums->list);
4039 if (need_find_last_extent && *last_extent == first_key.offset) {
4041 * We don't have any leafs between our current one and the one
4042 * we processed before that can have file extent items for our
4043 * inode (and have a generation number smaller than our current
4046 need_find_last_extent = false;
4050 * Because we use btrfs_search_forward we could skip leaves that were
4051 * not modified and then assume *last_extent is valid when it really
4052 * isn't. So back up to the previous leaf and read the end of the last
4053 * extent before we go and fill in holes.
4055 if (need_find_last_extent) {
4058 ret = btrfs_prev_leaf(inode->root, src_path);
4063 if (src_path->slots[0])
4064 src_path->slots[0]--;
4065 src = src_path->nodes[0];
4066 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
4067 if (key.objectid != btrfs_ino(inode) ||
4068 key.type != BTRFS_EXTENT_DATA_KEY)
4070 extent = btrfs_item_ptr(src, src_path->slots[0],
4071 struct btrfs_file_extent_item);
4072 if (btrfs_file_extent_type(src, extent) ==
4073 BTRFS_FILE_EXTENT_INLINE) {
4074 len = btrfs_file_extent_ram_bytes(src, extent);
4075 *last_extent = ALIGN(key.offset + len,
4076 fs_info->sectorsize);
4078 len = btrfs_file_extent_num_bytes(src, extent);
4079 *last_extent = key.offset + len;
4083 /* So we did prev_leaf, now we need to move to the next leaf, but a few
4084 * things could have happened
4086 * 1) A merge could have happened, so we could currently be on a leaf
4087 * that holds what we were copying in the first place.
4088 * 2) A split could have happened, and now not all of the items we want
4089 * are on the same leaf.
4091 * So we need to adjust how we search for holes, we need to drop the
4092 * path and re-search for the first extent key we found, and then walk
4093 * forward until we hit the last one we copied.
4095 if (need_find_last_extent) {
4096 /* btrfs_prev_leaf could return 1 without releasing the path */
4097 btrfs_release_path(src_path);
4098 ret = btrfs_search_slot(NULL, inode->root, &first_key,
4103 src = src_path->nodes[0];
4104 i = src_path->slots[0];
4110 * Ok so here we need to go through and fill in any holes we may have
4111 * to make sure that holes are punched for those areas in case they had
4112 * extents previously.
4118 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
4119 ret = btrfs_next_leaf(inode->root, src_path);
4123 src = src_path->nodes[0];
4125 need_find_last_extent = true;
4128 btrfs_item_key_to_cpu(src, &key, i);
4129 if (!btrfs_comp_cpu_keys(&key, &last_key))
4131 if (key.objectid != btrfs_ino(inode) ||
4132 key.type != BTRFS_EXTENT_DATA_KEY) {
4136 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
4137 if (btrfs_file_extent_type(src, extent) ==
4138 BTRFS_FILE_EXTENT_INLINE) {
4139 len = btrfs_file_extent_ram_bytes(src, extent);
4140 extent_end = ALIGN(key.offset + len,
4141 fs_info->sectorsize);
4143 len = btrfs_file_extent_num_bytes(src, extent);
4144 extent_end = key.offset + len;
4148 if (*last_extent == key.offset) {
4149 *last_extent = extent_end;
4152 offset = *last_extent;
4153 len = key.offset - *last_extent;
4154 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4155 offset, 0, 0, len, 0, len, 0, 0, 0);
4158 *last_extent = extent_end;
4162 * Check if there is a hole between the last extent found in our leaf
4163 * and the first extent in the next leaf. If there is one, we need to
4164 * log an explicit hole so that at replay time we can punch the hole.
4167 key.objectid == btrfs_ino(inode) &&
4168 key.type == BTRFS_EXTENT_DATA_KEY &&
4169 i == btrfs_header_nritems(src_path->nodes[0])) {
4170 ret = btrfs_next_leaf(inode->root, src_path);
4171 need_find_last_extent = true;
4174 } else if (ret == 0) {
4175 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4176 src_path->slots[0]);
4177 if (key.objectid == btrfs_ino(inode) &&
4178 key.type == BTRFS_EXTENT_DATA_KEY &&
4179 *last_extent < key.offset) {
4180 const u64 len = key.offset - *last_extent;
4182 ret = btrfs_insert_file_extent(trans, log,
4191 * Need to let the callers know we dropped the path so they should
4194 if (!ret && need_find_last_extent)
4199 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4201 struct extent_map *em1, *em2;
4203 em1 = list_entry(a, struct extent_map, list);
4204 em2 = list_entry(b, struct extent_map, list);
4206 if (em1->start < em2->start)
4208 else if (em1->start > em2->start)
4213 static int log_extent_csums(struct btrfs_trans_handle *trans,
4214 struct btrfs_inode *inode,
4215 struct btrfs_root *log_root,
4216 const struct extent_map *em)
4220 LIST_HEAD(ordered_sums);
4223 if (inode->flags & BTRFS_INODE_NODATASUM ||
4224 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4225 em->block_start == EXTENT_MAP_HOLE)
4228 /* If we're compressed we have to save the entire range of csums. */
4229 if (em->compress_type) {
4231 csum_len = max(em->block_len, em->orig_block_len);
4233 csum_offset = em->mod_start - em->start;
4234 csum_len = em->mod_len;
4237 /* block start is already adjusted for the file extent offset. */
4238 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4239 em->block_start + csum_offset,
4240 em->block_start + csum_offset +
4241 csum_len - 1, &ordered_sums, 0);
4245 while (!list_empty(&ordered_sums)) {
4246 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4247 struct btrfs_ordered_sum,
4250 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4251 list_del(&sums->list);
4258 static int log_one_extent(struct btrfs_trans_handle *trans,
4259 struct btrfs_inode *inode, struct btrfs_root *root,
4260 const struct extent_map *em,
4261 struct btrfs_path *path,
4262 struct btrfs_log_ctx *ctx)
4264 struct btrfs_root *log = root->log_root;
4265 struct btrfs_file_extent_item *fi;
4266 struct extent_buffer *leaf;
4267 struct btrfs_map_token token;
4268 struct btrfs_key key;
4269 u64 extent_offset = em->start - em->orig_start;
4272 int extent_inserted = 0;
4274 ret = log_extent_csums(trans, inode, log, em);
4278 btrfs_init_map_token(&token);
4280 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4281 em->start + em->len, NULL, 0, 1,
4282 sizeof(*fi), &extent_inserted);
4286 if (!extent_inserted) {
4287 key.objectid = btrfs_ino(inode);
4288 key.type = BTRFS_EXTENT_DATA_KEY;
4289 key.offset = em->start;
4291 ret = btrfs_insert_empty_item(trans, log, path, &key,
4296 leaf = path->nodes[0];
4297 fi = btrfs_item_ptr(leaf, path->slots[0],
4298 struct btrfs_file_extent_item);
4300 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4302 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4303 btrfs_set_token_file_extent_type(leaf, fi,
4304 BTRFS_FILE_EXTENT_PREALLOC,
4307 btrfs_set_token_file_extent_type(leaf, fi,
4308 BTRFS_FILE_EXTENT_REG,
4311 block_len = max(em->block_len, em->orig_block_len);
4312 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4313 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4316 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4318 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4319 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4321 extent_offset, &token);
4322 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4325 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4326 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4330 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4331 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4332 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4333 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4335 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4336 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4337 btrfs_mark_buffer_dirty(leaf);
4339 btrfs_release_path(path);
4345 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4346 * lose them after doing a fast fsync and replaying the log. We scan the
4347 * subvolume's root instead of iterating the inode's extent map tree because
4348 * otherwise we can log incorrect extent items based on extent map conversion.
4349 * That can happen due to the fact that extent maps are merged when they
4350 * are not in the extent map tree's list of modified extents.
4352 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4353 struct btrfs_inode *inode,
4354 struct btrfs_path *path)
4356 struct btrfs_root *root = inode->root;
4357 struct btrfs_key key;
4358 const u64 i_size = i_size_read(&inode->vfs_inode);
4359 const u64 ino = btrfs_ino(inode);
4360 struct btrfs_path *dst_path = NULL;
4361 u64 last_extent = (u64)-1;
4366 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4370 key.type = BTRFS_EXTENT_DATA_KEY;
4371 key.offset = i_size;
4372 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4377 struct extent_buffer *leaf = path->nodes[0];
4378 int slot = path->slots[0];
4380 if (slot >= btrfs_header_nritems(leaf)) {
4382 ret = copy_items(trans, inode, dst_path, path,
4383 &last_extent, start_slot,
4389 ret = btrfs_next_leaf(root, path);
4399 btrfs_item_key_to_cpu(leaf, &key, slot);
4400 if (key.objectid > ino)
4402 if (WARN_ON_ONCE(key.objectid < ino) ||
4403 key.type < BTRFS_EXTENT_DATA_KEY ||
4404 key.offset < i_size) {
4408 if (last_extent == (u64)-1) {
4409 last_extent = key.offset;
4411 * Avoid logging extent items logged in past fsync calls
4412 * and leading to duplicate keys in the log tree.
4415 ret = btrfs_truncate_inode_items(trans,
4419 BTRFS_EXTENT_DATA_KEY);
4420 } while (ret == -EAGAIN);
4429 dst_path = btrfs_alloc_path();
4437 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4438 start_slot, ins_nr, 1, 0);
4443 btrfs_release_path(path);
4444 btrfs_free_path(dst_path);
4448 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4449 struct btrfs_root *root,
4450 struct btrfs_inode *inode,
4451 struct btrfs_path *path,
4452 struct btrfs_log_ctx *ctx,
4456 struct extent_map *em, *n;
4457 struct list_head extents;
4458 struct extent_map_tree *tree = &inode->extent_tree;
4463 INIT_LIST_HEAD(&extents);
4465 write_lock(&tree->lock);
4466 test_gen = root->fs_info->last_trans_committed;
4468 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4470 * Skip extents outside our logging range. It's important to do
4471 * it for correctness because if we don't ignore them, we may
4472 * log them before their ordered extent completes, and therefore
4473 * we could log them without logging their respective checksums
4474 * (the checksum items are added to the csum tree at the very
4475 * end of btrfs_finish_ordered_io()). Also leave such extents
4476 * outside of our range in the list, since we may have another
4477 * ranged fsync in the near future that needs them. If an extent
4478 * outside our range corresponds to a hole, log it to avoid
4479 * leaving gaps between extents (fsck will complain when we are
4480 * not using the NO_HOLES feature).
4482 if ((em->start > end || em->start + em->len <= start) &&
4483 em->block_start != EXTENT_MAP_HOLE)
4486 list_del_init(&em->list);
4488 * Just an arbitrary number, this can be really CPU intensive
4489 * once we start getting a lot of extents, and really once we
4490 * have a bunch of extents we just want to commit since it will
4493 if (++num > 32768) {
4494 list_del_init(&tree->modified_extents);
4499 if (em->generation <= test_gen)
4502 /* We log prealloc extents beyond eof later. */
4503 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4504 em->start >= i_size_read(&inode->vfs_inode))
4507 /* Need a ref to keep it from getting evicted from cache */
4508 refcount_inc(&em->refs);
4509 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4510 list_add_tail(&em->list, &extents);
4514 list_sort(NULL, &extents, extent_cmp);
4516 while (!list_empty(&extents)) {
4517 em = list_entry(extents.next, struct extent_map, list);
4519 list_del_init(&em->list);
4522 * If we had an error we just need to delete everybody from our
4526 clear_em_logging(tree, em);
4527 free_extent_map(em);
4531 write_unlock(&tree->lock);
4533 ret = log_one_extent(trans, inode, root, em, path, ctx);
4534 write_lock(&tree->lock);
4535 clear_em_logging(tree, em);
4536 free_extent_map(em);
4538 WARN_ON(!list_empty(&extents));
4539 write_unlock(&tree->lock);
4541 btrfs_release_path(path);
4543 ret = btrfs_log_prealloc_extents(trans, inode, path);
4548 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4549 struct btrfs_path *path, u64 *size_ret)
4551 struct btrfs_key key;
4554 key.objectid = btrfs_ino(inode);
4555 key.type = BTRFS_INODE_ITEM_KEY;
4558 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4561 } else if (ret > 0) {
4564 struct btrfs_inode_item *item;
4566 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4567 struct btrfs_inode_item);
4568 *size_ret = btrfs_inode_size(path->nodes[0], item);
4570 * If the in-memory inode's i_size is smaller then the inode
4571 * size stored in the btree, return the inode's i_size, so
4572 * that we get a correct inode size after replaying the log
4573 * when before a power failure we had a shrinking truncate
4574 * followed by addition of a new name (rename / new hard link).
4575 * Otherwise return the inode size from the btree, to avoid
4576 * data loss when replaying a log due to previously doing a
4577 * write that expands the inode's size and logging a new name
4578 * immediately after.
4580 if (*size_ret > inode->vfs_inode.i_size)
4581 *size_ret = inode->vfs_inode.i_size;
4584 btrfs_release_path(path);
4589 * At the moment we always log all xattrs. This is to figure out at log replay
4590 * time which xattrs must have their deletion replayed. If a xattr is missing
4591 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4592 * because if a xattr is deleted, the inode is fsynced and a power failure
4593 * happens, causing the log to be replayed the next time the fs is mounted,
4594 * we want the xattr to not exist anymore (same behaviour as other filesystems
4595 * with a journal, ext3/4, xfs, f2fs, etc).
4597 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4598 struct btrfs_root *root,
4599 struct btrfs_inode *inode,
4600 struct btrfs_path *path,
4601 struct btrfs_path *dst_path)
4604 struct btrfs_key key;
4605 const u64 ino = btrfs_ino(inode);
4610 key.type = BTRFS_XATTR_ITEM_KEY;
4613 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4618 int slot = path->slots[0];
4619 struct extent_buffer *leaf = path->nodes[0];
4620 int nritems = btrfs_header_nritems(leaf);
4622 if (slot >= nritems) {
4624 u64 last_extent = 0;
4626 ret = copy_items(trans, inode, dst_path, path,
4627 &last_extent, start_slot,
4629 /* can't be 1, extent items aren't processed */
4635 ret = btrfs_next_leaf(root, path);
4643 btrfs_item_key_to_cpu(leaf, &key, slot);
4644 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4654 u64 last_extent = 0;
4656 ret = copy_items(trans, inode, dst_path, path,
4657 &last_extent, start_slot,
4659 /* can't be 1, extent items aren't processed */
4669 * If the no holes feature is enabled we need to make sure any hole between the
4670 * last extent and the i_size of our inode is explicitly marked in the log. This
4671 * is to make sure that doing something like:
4673 * 1) create file with 128Kb of data
4674 * 2) truncate file to 64Kb
4675 * 3) truncate file to 256Kb
4677 * 5) <crash/power failure>
4678 * 6) mount fs and trigger log replay
4680 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4681 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4682 * file correspond to a hole. The presence of explicit holes in a log tree is
4683 * what guarantees that log replay will remove/adjust file extent items in the
4686 * Here we do not need to care about holes between extents, that is already done
4687 * by copy_items(). We also only need to do this in the full sync path, where we
4688 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4689 * lookup the list of modified extent maps and if any represents a hole, we
4690 * insert a corresponding extent representing a hole in the log tree.
4692 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4693 struct btrfs_root *root,
4694 struct btrfs_inode *inode,
4695 struct btrfs_path *path)
4697 struct btrfs_fs_info *fs_info = root->fs_info;
4699 struct btrfs_key key;
4702 struct extent_buffer *leaf;
4703 struct btrfs_root *log = root->log_root;
4704 const u64 ino = btrfs_ino(inode);
4705 const u64 i_size = i_size_read(&inode->vfs_inode);
4707 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4711 key.type = BTRFS_EXTENT_DATA_KEY;
4712 key.offset = (u64)-1;
4714 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4719 ASSERT(path->slots[0] > 0);
4721 leaf = path->nodes[0];
4722 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4724 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4725 /* inode does not have any extents */
4729 struct btrfs_file_extent_item *extent;
4733 * If there's an extent beyond i_size, an explicit hole was
4734 * already inserted by copy_items().
4736 if (key.offset >= i_size)
4739 extent = btrfs_item_ptr(leaf, path->slots[0],
4740 struct btrfs_file_extent_item);
4742 if (btrfs_file_extent_type(leaf, extent) ==
4743 BTRFS_FILE_EXTENT_INLINE)
4746 len = btrfs_file_extent_num_bytes(leaf, extent);
4747 /* Last extent goes beyond i_size, no need to log a hole. */
4748 if (key.offset + len > i_size)
4750 hole_start = key.offset + len;
4751 hole_size = i_size - hole_start;
4753 btrfs_release_path(path);
4755 /* Last extent ends at i_size. */
4759 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4760 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4761 hole_size, 0, hole_size, 0, 0, 0);
4766 * When we are logging a new inode X, check if it doesn't have a reference that
4767 * matches the reference from some other inode Y created in a past transaction
4768 * and that was renamed in the current transaction. If we don't do this, then at
4769 * log replay time we can lose inode Y (and all its files if it's a directory):
4772 * echo "hello world" > /mnt/x/foobar
4775 * mkdir /mnt/x # or touch /mnt/x
4776 * xfs_io -c fsync /mnt/x
4778 * mount fs, trigger log replay
4780 * After the log replay procedure, we would lose the first directory and all its
4781 * files (file foobar).
4782 * For the case where inode Y is not a directory we simply end up losing it:
4784 * echo "123" > /mnt/foo
4786 * mv /mnt/foo /mnt/bar
4787 * echo "abc" > /mnt/foo
4788 * xfs_io -c fsync /mnt/foo
4791 * We also need this for cases where a snapshot entry is replaced by some other
4792 * entry (file or directory) otherwise we end up with an unreplayable log due to
4793 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4794 * if it were a regular entry:
4797 * btrfs subvolume snapshot /mnt /mnt/x/snap
4798 * btrfs subvolume delete /mnt/x/snap
4801 * fsync /mnt/x or fsync some new file inside it
4804 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4805 * the same transaction.
4807 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4809 const struct btrfs_key *key,
4810 struct btrfs_inode *inode,
4811 u64 *other_ino, u64 *other_parent)
4814 struct btrfs_path *search_path;
4817 u32 item_size = btrfs_item_size_nr(eb, slot);
4819 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4821 search_path = btrfs_alloc_path();
4824 search_path->search_commit_root = 1;
4825 search_path->skip_locking = 1;
4827 while (cur_offset < item_size) {
4831 unsigned long name_ptr;
4832 struct btrfs_dir_item *di;
4834 if (key->type == BTRFS_INODE_REF_KEY) {
4835 struct btrfs_inode_ref *iref;
4837 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4838 parent = key->offset;
4839 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4840 name_ptr = (unsigned long)(iref + 1);
4841 this_len = sizeof(*iref) + this_name_len;
4843 struct btrfs_inode_extref *extref;
4845 extref = (struct btrfs_inode_extref *)(ptr +
4847 parent = btrfs_inode_extref_parent(eb, extref);
4848 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4849 name_ptr = (unsigned long)&extref->name;
4850 this_len = sizeof(*extref) + this_name_len;
4853 if (this_name_len > name_len) {
4856 new_name = krealloc(name, this_name_len, GFP_NOFS);
4861 name_len = this_name_len;
4865 read_extent_buffer(eb, name, name_ptr, this_name_len);
4866 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4867 parent, name, this_name_len, 0);
4868 if (di && !IS_ERR(di)) {
4869 struct btrfs_key di_key;
4871 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4873 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4874 if (di_key.objectid != key->objectid) {
4876 *other_ino = di_key.objectid;
4877 *other_parent = parent;
4885 } else if (IS_ERR(di)) {
4889 btrfs_release_path(search_path);
4891 cur_offset += this_len;
4895 btrfs_free_path(search_path);
4900 struct btrfs_ino_list {
4903 struct list_head list;
4906 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4907 struct btrfs_root *root,
4908 struct btrfs_path *path,
4909 struct btrfs_log_ctx *ctx,
4910 u64 ino, u64 parent)
4912 struct btrfs_ino_list *ino_elem;
4913 LIST_HEAD(inode_list);
4916 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4919 ino_elem->ino = ino;
4920 ino_elem->parent = parent;
4921 list_add_tail(&ino_elem->list, &inode_list);
4923 while (!list_empty(&inode_list)) {
4924 struct btrfs_fs_info *fs_info = root->fs_info;
4925 struct btrfs_key key;
4926 struct inode *inode;
4928 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4930 ino = ino_elem->ino;
4931 parent = ino_elem->parent;
4932 list_del(&ino_elem->list);
4937 btrfs_release_path(path);
4940 key.type = BTRFS_INODE_ITEM_KEY;
4942 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4944 * If the other inode that had a conflicting dir entry was
4945 * deleted in the current transaction, we need to log its parent
4948 if (IS_ERR(inode)) {
4949 ret = PTR_ERR(inode);
4950 if (ret == -ENOENT) {
4951 key.objectid = parent;
4952 inode = btrfs_iget(fs_info->sb, &key, root,
4954 if (IS_ERR(inode)) {
4955 ret = PTR_ERR(inode);
4957 ret = btrfs_log_inode(trans, root,
4959 LOG_OTHER_INODE_ALL,
4967 * We are safe logging the other inode without acquiring its
4968 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4969 * are safe against concurrent renames of the other inode as
4970 * well because during a rename we pin the log and update the
4971 * log with the new name before we unpin it.
4973 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4974 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4981 key.type = BTRFS_INODE_REF_KEY;
4983 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4990 struct extent_buffer *leaf = path->nodes[0];
4991 int slot = path->slots[0];
4993 u64 other_parent = 0;
4995 if (slot >= btrfs_header_nritems(leaf)) {
4996 ret = btrfs_next_leaf(root, path);
4999 } else if (ret > 0) {
5006 btrfs_item_key_to_cpu(leaf, &key, slot);
5007 if (key.objectid != ino ||
5008 (key.type != BTRFS_INODE_REF_KEY &&
5009 key.type != BTRFS_INODE_EXTREF_KEY)) {
5014 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5015 BTRFS_I(inode), &other_ino,
5020 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5025 ino_elem->ino = other_ino;
5026 ino_elem->parent = other_parent;
5027 list_add_tail(&ino_elem->list, &inode_list);
5038 /* log a single inode in the tree log.
5039 * At least one parent directory for this inode must exist in the tree
5040 * or be logged already.
5042 * Any items from this inode changed by the current transaction are copied
5043 * to the log tree. An extra reference is taken on any extents in this
5044 * file, allowing us to avoid a whole pile of corner cases around logging
5045 * blocks that have been removed from the tree.
5047 * See LOG_INODE_ALL and related defines for a description of what inode_only
5050 * This handles both files and directories.
5052 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5053 struct btrfs_root *root, struct btrfs_inode *inode,
5057 struct btrfs_log_ctx *ctx)
5059 struct btrfs_fs_info *fs_info = root->fs_info;
5060 struct btrfs_path *path;
5061 struct btrfs_path *dst_path;
5062 struct btrfs_key min_key;
5063 struct btrfs_key max_key;
5064 struct btrfs_root *log = root->log_root;
5065 u64 last_extent = 0;
5069 int ins_start_slot = 0;
5071 bool fast_search = false;
5072 u64 ino = btrfs_ino(inode);
5073 struct extent_map_tree *em_tree = &inode->extent_tree;
5074 u64 logged_isize = 0;
5075 bool need_log_inode_item = true;
5076 bool xattrs_logged = false;
5077 bool recursive_logging = false;
5079 path = btrfs_alloc_path();
5082 dst_path = btrfs_alloc_path();
5084 btrfs_free_path(path);
5088 min_key.objectid = ino;
5089 min_key.type = BTRFS_INODE_ITEM_KEY;
5092 max_key.objectid = ino;
5095 /* today the code can only do partial logging of directories */
5096 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5097 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5098 &inode->runtime_flags) &&
5099 inode_only >= LOG_INODE_EXISTS))
5100 max_key.type = BTRFS_XATTR_ITEM_KEY;
5102 max_key.type = (u8)-1;
5103 max_key.offset = (u64)-1;
5106 * Only run delayed items if we are a dir or a new file.
5107 * Otherwise commit the delayed inode only, which is needed in
5108 * order for the log replay code to mark inodes for link count
5109 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5111 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5112 inode->generation > fs_info->last_trans_committed)
5113 ret = btrfs_commit_inode_delayed_items(trans, inode);
5115 ret = btrfs_commit_inode_delayed_inode(inode);
5118 btrfs_free_path(path);
5119 btrfs_free_path(dst_path);
5123 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5124 recursive_logging = true;
5125 if (inode_only == LOG_OTHER_INODE)
5126 inode_only = LOG_INODE_EXISTS;
5128 inode_only = LOG_INODE_ALL;
5129 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5131 mutex_lock(&inode->log_mutex);
5135 * a brute force approach to making sure we get the most uptodate
5136 * copies of everything.
5138 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5139 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5141 if (inode_only == LOG_INODE_EXISTS)
5142 max_key_type = BTRFS_XATTR_ITEM_KEY;
5143 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5145 if (inode_only == LOG_INODE_EXISTS) {
5147 * Make sure the new inode item we write to the log has
5148 * the same isize as the current one (if it exists).
5149 * This is necessary to prevent data loss after log
5150 * replay, and also to prevent doing a wrong expanding
5151 * truncate - for e.g. create file, write 4K into offset
5152 * 0, fsync, write 4K into offset 4096, add hard link,
5153 * fsync some other file (to sync log), power fail - if
5154 * we use the inode's current i_size, after log replay
5155 * we get a 8Kb file, with the last 4Kb extent as a hole
5156 * (zeroes), as if an expanding truncate happened,
5157 * instead of getting a file of 4Kb only.
5159 err = logged_inode_size(log, inode, path, &logged_isize);
5163 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5164 &inode->runtime_flags)) {
5165 if (inode_only == LOG_INODE_EXISTS) {
5166 max_key.type = BTRFS_XATTR_ITEM_KEY;
5167 ret = drop_objectid_items(trans, log, path, ino,
5170 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5171 &inode->runtime_flags);
5172 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5173 &inode->runtime_flags);
5175 ret = btrfs_truncate_inode_items(trans,
5176 log, &inode->vfs_inode, 0, 0);
5181 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5182 &inode->runtime_flags) ||
5183 inode_only == LOG_INODE_EXISTS) {
5184 if (inode_only == LOG_INODE_ALL)
5186 max_key.type = BTRFS_XATTR_ITEM_KEY;
5187 ret = drop_objectid_items(trans, log, path, ino,
5190 if (inode_only == LOG_INODE_ALL)
5203 ret = btrfs_search_forward(root, &min_key,
5204 path, trans->transid);
5212 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5213 if (min_key.objectid != ino)
5215 if (min_key.type > max_key.type)
5218 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5219 need_log_inode_item = false;
5221 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5222 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5223 inode->generation == trans->transid &&
5224 !recursive_logging) {
5226 u64 other_parent = 0;
5228 ret = btrfs_check_ref_name_override(path->nodes[0],
5229 path->slots[0], &min_key, inode,
5230 &other_ino, &other_parent);
5234 } else if (ret > 0 && ctx &&
5235 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5240 ins_start_slot = path->slots[0];
5242 ret = copy_items(trans, inode, dst_path, path,
5243 &last_extent, ins_start_slot,
5252 err = log_conflicting_inodes(trans, root, path,
5253 ctx, other_ino, other_parent);
5256 btrfs_release_path(path);
5261 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5262 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5265 ret = copy_items(trans, inode, dst_path, path,
5266 &last_extent, ins_start_slot,
5267 ins_nr, inode_only, logged_isize);
5274 btrfs_release_path(path);
5280 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5283 } else if (!ins_nr) {
5284 ins_start_slot = path->slots[0];
5289 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5290 ins_start_slot, ins_nr, inode_only,
5298 btrfs_release_path(path);
5302 ins_start_slot = path->slots[0];
5305 nritems = btrfs_header_nritems(path->nodes[0]);
5307 if (path->slots[0] < nritems) {
5308 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5313 ret = copy_items(trans, inode, dst_path, path,
5314 &last_extent, ins_start_slot,
5315 ins_nr, inode_only, logged_isize);
5323 btrfs_release_path(path);
5325 if (min_key.offset < (u64)-1) {
5327 } else if (min_key.type < max_key.type) {
5335 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5336 ins_start_slot, ins_nr, inode_only,
5346 btrfs_release_path(path);
5347 btrfs_release_path(dst_path);
5348 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5351 xattrs_logged = true;
5352 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5353 btrfs_release_path(path);
5354 btrfs_release_path(dst_path);
5355 err = btrfs_log_trailing_hole(trans, root, inode, path);
5360 btrfs_release_path(path);
5361 btrfs_release_path(dst_path);
5362 if (need_log_inode_item) {
5363 err = log_inode_item(trans, log, dst_path, inode);
5364 if (!err && !xattrs_logged) {
5365 err = btrfs_log_all_xattrs(trans, root, inode, path,
5367 btrfs_release_path(path);
5373 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5379 } else if (inode_only == LOG_INODE_ALL) {
5380 struct extent_map *em, *n;
5382 write_lock(&em_tree->lock);
5384 * We can't just remove every em if we're called for a ranged
5385 * fsync - that is, one that doesn't cover the whole possible
5386 * file range (0 to LLONG_MAX). This is because we can have
5387 * em's that fall outside the range we're logging and therefore
5388 * their ordered operations haven't completed yet
5389 * (btrfs_finish_ordered_io() not invoked yet). This means we
5390 * didn't get their respective file extent item in the fs/subvol
5391 * tree yet, and need to let the next fast fsync (one which
5392 * consults the list of modified extent maps) find the em so
5393 * that it logs a matching file extent item and waits for the
5394 * respective ordered operation to complete (if it's still
5397 * Removing every em outside the range we're logging would make
5398 * the next fast fsync not log their matching file extent items,
5399 * therefore making us lose data after a log replay.
5401 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5403 const u64 mod_end = em->mod_start + em->mod_len - 1;
5405 if (em->mod_start >= start && mod_end <= end)
5406 list_del_init(&em->list);
5408 write_unlock(&em_tree->lock);
5411 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5412 ret = log_directory_changes(trans, root, inode, path, dst_path,
5420 spin_lock(&inode->lock);
5421 inode->logged_trans = trans->transid;
5422 inode->last_log_commit = inode->last_sub_trans;
5423 spin_unlock(&inode->lock);
5425 mutex_unlock(&inode->log_mutex);
5427 btrfs_free_path(path);
5428 btrfs_free_path(dst_path);
5433 * Check if we must fallback to a transaction commit when logging an inode.
5434 * This must be called after logging the inode and is used only in the context
5435 * when fsyncing an inode requires the need to log some other inode - in which
5436 * case we can't lock the i_mutex of each other inode we need to log as that
5437 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5438 * log inodes up or down in the hierarchy) or rename operations for example. So
5439 * we take the log_mutex of the inode after we have logged it and then check for
5440 * its last_unlink_trans value - this is safe because any task setting
5441 * last_unlink_trans must take the log_mutex and it must do this before it does
5442 * the actual unlink operation, so if we do this check before a concurrent task
5443 * sets last_unlink_trans it means we've logged a consistent version/state of
5444 * all the inode items, otherwise we are not sure and must do a transaction
5445 * commit (the concurrent task might have only updated last_unlink_trans before
5446 * we logged the inode or it might have also done the unlink).
5448 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5449 struct btrfs_inode *inode)
5451 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5454 mutex_lock(&inode->log_mutex);
5455 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5457 * Make sure any commits to the log are forced to be full
5460 btrfs_set_log_full_commit(trans);
5463 mutex_unlock(&inode->log_mutex);
5469 * follow the dentry parent pointers up the chain and see if any
5470 * of the directories in it require a full commit before they can
5471 * be logged. Returns zero if nothing special needs to be done or 1 if
5472 * a full commit is required.
5474 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5475 struct btrfs_inode *inode,
5476 struct dentry *parent,
5477 struct super_block *sb,
5481 struct dentry *old_parent = NULL;
5482 struct btrfs_inode *orig_inode = inode;
5485 * for regular files, if its inode is already on disk, we don't
5486 * have to worry about the parents at all. This is because
5487 * we can use the last_unlink_trans field to record renames
5488 * and other fun in this file.
5490 if (S_ISREG(inode->vfs_inode.i_mode) &&
5491 inode->generation <= last_committed &&
5492 inode->last_unlink_trans <= last_committed)
5495 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5496 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5498 inode = BTRFS_I(d_inode(parent));
5503 * If we are logging a directory then we start with our inode,
5504 * not our parent's inode, so we need to skip setting the
5505 * logged_trans so that further down in the log code we don't
5506 * think this inode has already been logged.
5508 if (inode != orig_inode)
5509 inode->logged_trans = trans->transid;
5512 if (btrfs_must_commit_transaction(trans, inode)) {
5517 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5520 if (IS_ROOT(parent)) {
5521 inode = BTRFS_I(d_inode(parent));
5522 if (btrfs_must_commit_transaction(trans, inode))
5527 parent = dget_parent(parent);
5529 old_parent = parent;
5530 inode = BTRFS_I(d_inode(parent));
5538 struct btrfs_dir_list {
5540 struct list_head list;
5544 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5545 * details about the why it is needed.
5546 * This is a recursive operation - if an existing dentry corresponds to a
5547 * directory, that directory's new entries are logged too (same behaviour as
5548 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5549 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5550 * complains about the following circular lock dependency / possible deadlock:
5554 * lock(&type->i_mutex_dir_key#3/2);
5555 * lock(sb_internal#2);
5556 * lock(&type->i_mutex_dir_key#3/2);
5557 * lock(&sb->s_type->i_mutex_key#14);
5559 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5560 * sb_start_intwrite() in btrfs_start_transaction().
5561 * Not locking i_mutex of the inodes is still safe because:
5563 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5564 * that while logging the inode new references (names) are added or removed
5565 * from the inode, leaving the logged inode item with a link count that does
5566 * not match the number of logged inode reference items. This is fine because
5567 * at log replay time we compute the real number of links and correct the
5568 * link count in the inode item (see replay_one_buffer() and
5569 * link_to_fixup_dir());
5571 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5572 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5573 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5574 * has a size that doesn't match the sum of the lengths of all the logged
5575 * names. This does not result in a problem because if a dir_item key is
5576 * logged but its matching dir_index key is not logged, at log replay time we
5577 * don't use it to replay the respective name (see replay_one_name()). On the
5578 * other hand if only the dir_index key ends up being logged, the respective
5579 * name is added to the fs/subvol tree with both the dir_item and dir_index
5580 * keys created (see replay_one_name()).
5581 * The directory's inode item with a wrong i_size is not a problem as well,
5582 * since we don't use it at log replay time to set the i_size in the inode
5583 * item of the fs/subvol tree (see overwrite_item()).
5585 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5586 struct btrfs_root *root,
5587 struct btrfs_inode *start_inode,
5588 struct btrfs_log_ctx *ctx)
5590 struct btrfs_fs_info *fs_info = root->fs_info;
5591 struct btrfs_root *log = root->log_root;
5592 struct btrfs_path *path;
5593 LIST_HEAD(dir_list);
5594 struct btrfs_dir_list *dir_elem;
5597 path = btrfs_alloc_path();
5601 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5603 btrfs_free_path(path);
5606 dir_elem->ino = btrfs_ino(start_inode);
5607 list_add_tail(&dir_elem->list, &dir_list);
5609 while (!list_empty(&dir_list)) {
5610 struct extent_buffer *leaf;
5611 struct btrfs_key min_key;
5615 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5618 goto next_dir_inode;
5620 min_key.objectid = dir_elem->ino;
5621 min_key.type = BTRFS_DIR_ITEM_KEY;
5624 btrfs_release_path(path);
5625 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5627 goto next_dir_inode;
5628 } else if (ret > 0) {
5630 goto next_dir_inode;
5634 leaf = path->nodes[0];
5635 nritems = btrfs_header_nritems(leaf);
5636 for (i = path->slots[0]; i < nritems; i++) {
5637 struct btrfs_dir_item *di;
5638 struct btrfs_key di_key;
5639 struct inode *di_inode;
5640 struct btrfs_dir_list *new_dir_elem;
5641 int log_mode = LOG_INODE_EXISTS;
5644 btrfs_item_key_to_cpu(leaf, &min_key, i);
5645 if (min_key.objectid != dir_elem->ino ||
5646 min_key.type != BTRFS_DIR_ITEM_KEY)
5647 goto next_dir_inode;
5649 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5650 type = btrfs_dir_type(leaf, di);
5651 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5652 type != BTRFS_FT_DIR)
5654 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5655 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5658 btrfs_release_path(path);
5659 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5660 if (IS_ERR(di_inode)) {
5661 ret = PTR_ERR(di_inode);
5662 goto next_dir_inode;
5665 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5670 ctx->log_new_dentries = false;
5671 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5672 log_mode = LOG_INODE_ALL;
5673 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5674 log_mode, 0, LLONG_MAX, ctx);
5676 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5680 goto next_dir_inode;
5681 if (ctx->log_new_dentries) {
5682 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5684 if (!new_dir_elem) {
5686 goto next_dir_inode;
5688 new_dir_elem->ino = di_key.objectid;
5689 list_add_tail(&new_dir_elem->list, &dir_list);
5694 ret = btrfs_next_leaf(log, path);
5696 goto next_dir_inode;
5697 } else if (ret > 0) {
5699 goto next_dir_inode;
5703 if (min_key.offset < (u64)-1) {
5708 list_del(&dir_elem->list);
5712 btrfs_free_path(path);
5716 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5717 struct btrfs_inode *inode,
5718 struct btrfs_log_ctx *ctx)
5720 struct btrfs_fs_info *fs_info = trans->fs_info;
5722 struct btrfs_path *path;
5723 struct btrfs_key key;
5724 struct btrfs_root *root = inode->root;
5725 const u64 ino = btrfs_ino(inode);
5727 path = btrfs_alloc_path();
5730 path->skip_locking = 1;
5731 path->search_commit_root = 1;
5734 key.type = BTRFS_INODE_REF_KEY;
5736 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5741 struct extent_buffer *leaf = path->nodes[0];
5742 int slot = path->slots[0];
5747 if (slot >= btrfs_header_nritems(leaf)) {
5748 ret = btrfs_next_leaf(root, path);
5756 btrfs_item_key_to_cpu(leaf, &key, slot);
5757 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5758 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5761 item_size = btrfs_item_size_nr(leaf, slot);
5762 ptr = btrfs_item_ptr_offset(leaf, slot);
5763 while (cur_offset < item_size) {
5764 struct btrfs_key inode_key;
5765 struct inode *dir_inode;
5767 inode_key.type = BTRFS_INODE_ITEM_KEY;
5768 inode_key.offset = 0;
5770 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5771 struct btrfs_inode_extref *extref;
5773 extref = (struct btrfs_inode_extref *)
5775 inode_key.objectid = btrfs_inode_extref_parent(
5777 cur_offset += sizeof(*extref);
5778 cur_offset += btrfs_inode_extref_name_len(leaf,
5781 inode_key.objectid = key.offset;
5782 cur_offset = item_size;
5785 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5788 * If the parent inode was deleted, return an error to
5789 * fallback to a transaction commit. This is to prevent
5790 * getting an inode that was moved from one parent A to
5791 * a parent B, got its former parent A deleted and then
5792 * it got fsync'ed, from existing at both parents after
5793 * a log replay (and the old parent still existing).
5800 * mv /mnt/B/bar /mnt/A/bar
5801 * mv -T /mnt/A /mnt/B
5805 * If we ignore the old parent B which got deleted,
5806 * after a log replay we would have file bar linked
5807 * at both parents and the old parent B would still
5810 if (IS_ERR(dir_inode)) {
5811 ret = PTR_ERR(dir_inode);
5816 ctx->log_new_dentries = false;
5817 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5818 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5820 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5822 if (!ret && ctx && ctx->log_new_dentries)
5823 ret = log_new_dir_dentries(trans, root,
5824 BTRFS_I(dir_inode), ctx);
5833 btrfs_free_path(path);
5838 * helper function around btrfs_log_inode to make sure newly created
5839 * parent directories also end up in the log. A minimal inode and backref
5840 * only logging is done of any parent directories that are older than
5841 * the last committed transaction
5843 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5844 struct btrfs_inode *inode,
5845 struct dentry *parent,
5849 struct btrfs_log_ctx *ctx)
5851 struct btrfs_root *root = inode->root;
5852 struct btrfs_fs_info *fs_info = root->fs_info;
5853 struct super_block *sb;
5854 struct dentry *old_parent = NULL;
5856 u64 last_committed = fs_info->last_trans_committed;
5857 bool log_dentries = false;
5858 struct btrfs_inode *orig_inode = inode;
5860 sb = inode->vfs_inode.i_sb;
5862 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5868 * The prev transaction commit doesn't complete, we need do
5869 * full commit by ourselves.
5871 if (fs_info->last_trans_log_full_commit >
5872 fs_info->last_trans_committed) {
5877 if (btrfs_root_refs(&root->root_item) == 0) {
5882 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5888 * Skip already logged inodes or inodes corresponding to tmpfiles
5889 * (since logging them is pointless, a link count of 0 means they
5890 * will never be accessible).
5892 if (btrfs_inode_in_log(inode, trans->transid) ||
5893 inode->vfs_inode.i_nlink == 0) {
5894 ret = BTRFS_NO_LOG_SYNC;
5898 ret = start_log_trans(trans, root, ctx);
5902 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5907 * for regular files, if its inode is already on disk, we don't
5908 * have to worry about the parents at all. This is because
5909 * we can use the last_unlink_trans field to record renames
5910 * and other fun in this file.
5912 if (S_ISREG(inode->vfs_inode.i_mode) &&
5913 inode->generation <= last_committed &&
5914 inode->last_unlink_trans <= last_committed) {
5919 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5920 log_dentries = true;
5923 * On unlink we must make sure all our current and old parent directory
5924 * inodes are fully logged. This is to prevent leaving dangling
5925 * directory index entries in directories that were our parents but are
5926 * not anymore. Not doing this results in old parent directory being
5927 * impossible to delete after log replay (rmdir will always fail with
5928 * error -ENOTEMPTY).
5934 * ln testdir/foo testdir/bar
5936 * unlink testdir/bar
5937 * xfs_io -c fsync testdir/foo
5939 * mount fs, triggers log replay
5941 * If we don't log the parent directory (testdir), after log replay the
5942 * directory still has an entry pointing to the file inode using the bar
5943 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5944 * the file inode has a link count of 1.
5950 * ln foo testdir/foo2
5951 * ln foo testdir/foo3
5953 * unlink testdir/foo3
5954 * xfs_io -c fsync foo
5956 * mount fs, triggers log replay
5958 * Similar as the first example, after log replay the parent directory
5959 * testdir still has an entry pointing to the inode file with name foo3
5960 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5961 * and has a link count of 2.
5963 if (inode->last_unlink_trans > last_committed) {
5964 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5970 * If a new hard link was added to the inode in the current transaction
5971 * and its link count is now greater than 1, we need to fallback to a
5972 * transaction commit, otherwise we can end up not logging all its new
5973 * parents for all the hard links. Here just from the dentry used to
5974 * fsync, we can not visit the ancestor inodes for all the other hard
5975 * links to figure out if any is new, so we fallback to a transaction
5976 * commit (instead of adding a lot of complexity of scanning a btree,
5977 * since this scenario is not a common use case).
5979 if (inode->vfs_inode.i_nlink > 1 &&
5980 inode->last_link_trans > last_committed) {
5986 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5989 inode = BTRFS_I(d_inode(parent));
5990 if (root != inode->root)
5993 if (inode->generation > last_committed) {
5994 ret = btrfs_log_inode(trans, root, inode,
5995 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5999 if (IS_ROOT(parent))
6002 parent = dget_parent(parent);
6004 old_parent = parent;
6007 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
6013 btrfs_set_log_full_commit(trans);
6018 btrfs_remove_log_ctx(root, ctx);
6019 btrfs_end_log_trans(root);
6025 * it is not safe to log dentry if the chunk root has added new
6026 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6027 * If this returns 1, you must commit the transaction to safely get your
6030 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6031 struct dentry *dentry,
6034 struct btrfs_log_ctx *ctx)
6036 struct dentry *parent = dget_parent(dentry);
6039 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6040 start, end, LOG_INODE_ALL, ctx);
6047 * should be called during mount to recover any replay any log trees
6050 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6053 struct btrfs_path *path;
6054 struct btrfs_trans_handle *trans;
6055 struct btrfs_key key;
6056 struct btrfs_key found_key;
6057 struct btrfs_key tmp_key;
6058 struct btrfs_root *log;
6059 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6060 struct walk_control wc = {
6061 .process_func = process_one_buffer,
6065 path = btrfs_alloc_path();
6069 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6071 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6072 if (IS_ERR(trans)) {
6073 ret = PTR_ERR(trans);
6080 ret = walk_log_tree(trans, log_root_tree, &wc);
6082 btrfs_handle_fs_error(fs_info, ret,
6083 "Failed to pin buffers while recovering log root tree.");
6088 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6089 key.offset = (u64)-1;
6090 key.type = BTRFS_ROOT_ITEM_KEY;
6093 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6096 btrfs_handle_fs_error(fs_info, ret,
6097 "Couldn't find tree log root.");
6101 if (path->slots[0] == 0)
6105 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6107 btrfs_release_path(path);
6108 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6111 log = btrfs_read_fs_root(log_root_tree, &found_key);
6114 btrfs_handle_fs_error(fs_info, ret,
6115 "Couldn't read tree log root.");
6119 tmp_key.objectid = found_key.offset;
6120 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6121 tmp_key.offset = (u64)-1;
6123 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
6124 if (IS_ERR(wc.replay_dest)) {
6125 ret = PTR_ERR(wc.replay_dest);
6126 free_extent_buffer(log->node);
6127 free_extent_buffer(log->commit_root);
6129 btrfs_handle_fs_error(fs_info, ret,
6130 "Couldn't read target root for tree log recovery.");
6134 wc.replay_dest->log_root = log;
6135 btrfs_record_root_in_trans(trans, wc.replay_dest);
6136 ret = walk_log_tree(trans, log, &wc);
6138 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6139 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6143 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6144 struct btrfs_root *root = wc.replay_dest;
6146 btrfs_release_path(path);
6149 * We have just replayed everything, and the highest
6150 * objectid of fs roots probably has changed in case
6151 * some inode_item's got replayed.
6153 * root->objectid_mutex is not acquired as log replay
6154 * could only happen during mount.
6156 ret = btrfs_find_highest_objectid(root,
6157 &root->highest_objectid);
6160 key.offset = found_key.offset - 1;
6161 wc.replay_dest->log_root = NULL;
6162 free_extent_buffer(log->node);
6163 free_extent_buffer(log->commit_root);
6169 if (found_key.offset == 0)
6172 btrfs_release_path(path);
6174 /* step one is to pin it all, step two is to replay just inodes */
6177 wc.process_func = replay_one_buffer;
6178 wc.stage = LOG_WALK_REPLAY_INODES;
6181 /* step three is to replay everything */
6182 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6187 btrfs_free_path(path);
6189 /* step 4: commit the transaction, which also unpins the blocks */
6190 ret = btrfs_commit_transaction(trans);
6194 free_extent_buffer(log_root_tree->node);
6195 log_root_tree->log_root = NULL;
6196 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6197 kfree(log_root_tree);
6202 btrfs_end_transaction(wc.trans);
6203 btrfs_free_path(path);
6208 * there are some corner cases where we want to force a full
6209 * commit instead of allowing a directory to be logged.
6211 * They revolve around files there were unlinked from the directory, and
6212 * this function updates the parent directory so that a full commit is
6213 * properly done if it is fsync'd later after the unlinks are done.
6215 * Must be called before the unlink operations (updates to the subvolume tree,
6216 * inodes, etc) are done.
6218 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6219 struct btrfs_inode *dir, struct btrfs_inode *inode,
6223 * when we're logging a file, if it hasn't been renamed
6224 * or unlinked, and its inode is fully committed on disk,
6225 * we don't have to worry about walking up the directory chain
6226 * to log its parents.
6228 * So, we use the last_unlink_trans field to put this transid
6229 * into the file. When the file is logged we check it and
6230 * don't log the parents if the file is fully on disk.
6232 mutex_lock(&inode->log_mutex);
6233 inode->last_unlink_trans = trans->transid;
6234 mutex_unlock(&inode->log_mutex);
6237 * if this directory was already logged any new
6238 * names for this file/dir will get recorded
6241 if (dir->logged_trans == trans->transid)
6245 * if the inode we're about to unlink was logged,
6246 * the log will be properly updated for any new names
6248 if (inode->logged_trans == trans->transid)
6252 * when renaming files across directories, if the directory
6253 * there we're unlinking from gets fsync'd later on, there's
6254 * no way to find the destination directory later and fsync it
6255 * properly. So, we have to be conservative and force commits
6256 * so the new name gets discovered.
6261 /* we can safely do the unlink without any special recording */
6265 mutex_lock(&dir->log_mutex);
6266 dir->last_unlink_trans = trans->transid;
6267 mutex_unlock(&dir->log_mutex);
6271 * Make sure that if someone attempts to fsync the parent directory of a deleted
6272 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6273 * that after replaying the log tree of the parent directory's root we will not
6274 * see the snapshot anymore and at log replay time we will not see any log tree
6275 * corresponding to the deleted snapshot's root, which could lead to replaying
6276 * it after replaying the log tree of the parent directory (which would replay
6277 * the snapshot delete operation).
6279 * Must be called before the actual snapshot destroy operation (updates to the
6280 * parent root and tree of tree roots trees, etc) are done.
6282 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6283 struct btrfs_inode *dir)
6285 mutex_lock(&dir->log_mutex);
6286 dir->last_unlink_trans = trans->transid;
6287 mutex_unlock(&dir->log_mutex);
6291 * Call this after adding a new name for a file and it will properly
6292 * update the log to reflect the new name.
6294 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6295 * true (because it's not used).
6297 * Return value depends on whether @sync_log is true or false.
6298 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6299 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6301 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6302 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6303 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6304 * committed (without attempting to sync the log).
6306 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6307 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6308 struct dentry *parent,
6309 bool sync_log, struct btrfs_log_ctx *ctx)
6311 struct btrfs_fs_info *fs_info = trans->fs_info;
6315 * this will force the logging code to walk the dentry chain
6318 if (!S_ISDIR(inode->vfs_inode.i_mode))
6319 inode->last_unlink_trans = trans->transid;
6322 * if this inode hasn't been logged and directory we're renaming it
6323 * from hasn't been logged, we don't need to log it
6325 if (inode->logged_trans <= fs_info->last_trans_committed &&
6326 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6327 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6328 BTRFS_DONT_NEED_LOG_SYNC;
6331 struct btrfs_log_ctx ctx2;
6333 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6334 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6335 LOG_INODE_EXISTS, &ctx2);
6336 if (ret == BTRFS_NO_LOG_SYNC)
6337 return BTRFS_DONT_NEED_TRANS_COMMIT;
6339 return BTRFS_NEED_TRANS_COMMIT;
6341 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6343 return BTRFS_NEED_TRANS_COMMIT;
6344 return BTRFS_DONT_NEED_TRANS_COMMIT;
6348 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6349 LOG_INODE_EXISTS, ctx);
6350 if (ret == BTRFS_NO_LOG_SYNC)
6351 return BTRFS_DONT_NEED_LOG_SYNC;
6353 return BTRFS_NEED_TRANS_COMMIT;
6355 return BTRFS_NEED_LOG_SYNC;