1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
16 #include "print-tree.h"
18 #include "compression.h"
20 #include "inode-map.h"
21 #include "block-group.h"
22 #include "space-info.h"
24 /* magic values for the inode_only field in btrfs_log_inode:
26 * LOG_INODE_ALL means to log everything
27 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 * directory trouble cases
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
47 * rename foo/some_dir foo2/some_dir
49 * fsync foo/some_dir/some_file
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
81 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
91 LOG_WALK_REPLAY_INODES,
92 LOG_WALK_REPLAY_DIR_INDEX,
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct btrfs_inode *inode,
99 struct btrfs_log_ctx *ctx);
100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_path *path, u64 objectid);
103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_root *log,
106 struct btrfs_path *path,
107 u64 dirid, int del_all);
110 * tree logging is a special write ahead log used to make sure that
111 * fsyncs and O_SYNCs can happen without doing full tree commits.
113 * Full tree commits are expensive because they require commonly
114 * modified blocks to be recowed, creating many dirty pages in the
115 * extent tree an 4x-6x higher write load than ext3.
117 * Instead of doing a tree commit on every fsync, we use the
118 * key ranges and transaction ids to find items for a given file or directory
119 * that have changed in this transaction. Those items are copied into
120 * a special tree (one per subvolume root), that tree is written to disk
121 * and then the fsync is considered complete.
123 * After a crash, items are copied out of the log-tree back into the
124 * subvolume tree. Any file data extents found are recorded in the extent
125 * allocation tree, and the log-tree freed.
127 * The log tree is read three times, once to pin down all the extents it is
128 * using in ram and once, once to create all the inodes logged in the tree
129 * and once to do all the other items.
133 * start a sub transaction and setup the log tree
134 * this increments the log tree writer count to make the people
135 * syncing the tree wait for us to finish
137 static int start_log_trans(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root,
139 struct btrfs_log_ctx *ctx)
141 struct btrfs_fs_info *fs_info = root->fs_info;
144 mutex_lock(&root->log_mutex);
146 if (root->log_root) {
147 if (btrfs_need_log_full_commit(trans)) {
152 if (!root->log_start_pid) {
153 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
154 root->log_start_pid = current->pid;
155 } else if (root->log_start_pid != current->pid) {
156 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 mutex_lock(&fs_info->tree_log_mutex);
160 if (!fs_info->log_root_tree)
161 ret = btrfs_init_log_root_tree(trans, fs_info);
162 mutex_unlock(&fs_info->tree_log_mutex);
166 ret = btrfs_add_log_tree(trans, root);
170 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
171 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
172 root->log_start_pid = current->pid;
175 atomic_inc(&root->log_batch);
176 atomic_inc(&root->log_writers);
177 if (ctx && !ctx->logging_new_name) {
178 int index = root->log_transid % 2;
179 list_add_tail(&ctx->list, &root->log_ctxs[index]);
180 ctx->log_transid = root->log_transid;
184 mutex_unlock(&root->log_mutex);
189 * returns 0 if there was a log transaction running and we were able
190 * to join, or returns -ENOENT if there were not transactions
193 static int join_running_log_trans(struct btrfs_root *root)
197 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
200 mutex_lock(&root->log_mutex);
201 if (root->log_root) {
203 atomic_inc(&root->log_writers);
205 mutex_unlock(&root->log_mutex);
210 * This either makes the current running log transaction wait
211 * until you call btrfs_end_log_trans() or it makes any future
212 * log transactions wait until you call btrfs_end_log_trans()
214 void btrfs_pin_log_trans(struct btrfs_root *root)
216 atomic_inc(&root->log_writers);
220 * indicate we're done making changes to the log tree
221 * and wake up anyone waiting to do a sync
223 void btrfs_end_log_trans(struct btrfs_root *root)
225 if (atomic_dec_and_test(&root->log_writers)) {
226 /* atomic_dec_and_test implies a barrier */
227 cond_wake_up_nomb(&root->log_writer_wait);
231 static int btrfs_write_tree_block(struct extent_buffer *buf)
233 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
234 buf->start + buf->len - 1);
237 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
239 filemap_fdatawait_range(buf->pages[0]->mapping,
240 buf->start, buf->start + buf->len - 1);
244 * the walk control struct is used to pass state down the chain when
245 * processing the log tree. The stage field tells us which part
246 * of the log tree processing we are currently doing. The others
247 * are state fields used for that specific part
249 struct walk_control {
250 /* should we free the extent on disk when done? This is used
251 * at transaction commit time while freeing a log tree
255 /* should we write out the extent buffer? This is used
256 * while flushing the log tree to disk during a sync
260 /* should we wait for the extent buffer io to finish? Also used
261 * while flushing the log tree to disk for a sync
265 /* pin only walk, we record which extents on disk belong to the
270 /* what stage of the replay code we're currently in */
274 * Ignore any items from the inode currently being processed. Needs
275 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
276 * the LOG_WALK_REPLAY_INODES stage.
278 bool ignore_cur_inode;
280 /* the root we are currently replaying */
281 struct btrfs_root *replay_dest;
283 /* the trans handle for the current replay */
284 struct btrfs_trans_handle *trans;
286 /* the function that gets used to process blocks we find in the
287 * tree. Note the extent_buffer might not be up to date when it is
288 * passed in, and it must be checked or read if you need the data
291 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
292 struct walk_control *wc, u64 gen, int level);
296 * process_func used to pin down extents, write them or wait on them
298 static int process_one_buffer(struct btrfs_root *log,
299 struct extent_buffer *eb,
300 struct walk_control *wc, u64 gen, int level)
302 struct btrfs_fs_info *fs_info = log->fs_info;
306 * If this fs is mixed then we need to be able to process the leaves to
307 * pin down any logged extents, so we have to read the block.
309 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
310 ret = btrfs_read_buffer(eb, gen, level, NULL);
316 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
319 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
320 if (wc->pin && btrfs_header_level(eb) == 0)
321 ret = btrfs_exclude_logged_extents(eb);
323 btrfs_write_tree_block(eb);
325 btrfs_wait_tree_block_writeback(eb);
331 * Item overwrite used by replay and tree logging. eb, slot and key all refer
332 * to the src data we are copying out.
334 * root is the tree we are copying into, and path is a scratch
335 * path for use in this function (it should be released on entry and
336 * will be released on exit).
338 * If the key is already in the destination tree the existing item is
339 * overwritten. If the existing item isn't big enough, it is extended.
340 * If it is too large, it is truncated.
342 * If the key isn't in the destination yet, a new item is inserted.
344 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
345 struct btrfs_root *root,
346 struct btrfs_path *path,
347 struct extent_buffer *eb, int slot,
348 struct btrfs_key *key)
352 u64 saved_i_size = 0;
353 int save_old_i_size = 0;
354 unsigned long src_ptr;
355 unsigned long dst_ptr;
356 int overwrite_root = 0;
357 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
359 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
362 item_size = btrfs_item_size_nr(eb, slot);
363 src_ptr = btrfs_item_ptr_offset(eb, slot);
365 /* look for the key in the destination tree */
366 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
373 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
375 if (dst_size != item_size)
378 if (item_size == 0) {
379 btrfs_release_path(path);
382 dst_copy = kmalloc(item_size, GFP_NOFS);
383 src_copy = kmalloc(item_size, GFP_NOFS);
384 if (!dst_copy || !src_copy) {
385 btrfs_release_path(path);
391 read_extent_buffer(eb, src_copy, src_ptr, item_size);
393 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
394 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
396 ret = memcmp(dst_copy, src_copy, item_size);
401 * they have the same contents, just return, this saves
402 * us from cowing blocks in the destination tree and doing
403 * extra writes that may not have been done by a previous
407 btrfs_release_path(path);
412 * We need to load the old nbytes into the inode so when we
413 * replay the extents we've logged we get the right nbytes.
416 struct btrfs_inode_item *item;
420 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
421 struct btrfs_inode_item);
422 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
423 item = btrfs_item_ptr(eb, slot,
424 struct btrfs_inode_item);
425 btrfs_set_inode_nbytes(eb, item, nbytes);
428 * If this is a directory we need to reset the i_size to
429 * 0 so that we can set it up properly when replaying
430 * the rest of the items in this log.
432 mode = btrfs_inode_mode(eb, item);
434 btrfs_set_inode_size(eb, item, 0);
436 } else if (inode_item) {
437 struct btrfs_inode_item *item;
441 * New inode, set nbytes to 0 so that the nbytes comes out
442 * properly when we replay the extents.
444 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
445 btrfs_set_inode_nbytes(eb, item, 0);
448 * If this is a directory we need to reset the i_size to 0 so
449 * that we can set it up properly when replaying the rest of
450 * the items in this log.
452 mode = btrfs_inode_mode(eb, item);
454 btrfs_set_inode_size(eb, item, 0);
457 btrfs_release_path(path);
458 /* try to insert the key into the destination tree */
459 path->skip_release_on_error = 1;
460 ret = btrfs_insert_empty_item(trans, root, path,
462 path->skip_release_on_error = 0;
464 /* make sure any existing item is the correct size */
465 if (ret == -EEXIST || ret == -EOVERFLOW) {
467 found_size = btrfs_item_size_nr(path->nodes[0],
469 if (found_size > item_size)
470 btrfs_truncate_item(path, item_size, 1);
471 else if (found_size < item_size)
472 btrfs_extend_item(path, item_size - found_size);
476 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
479 /* don't overwrite an existing inode if the generation number
480 * was logged as zero. This is done when the tree logging code
481 * is just logging an inode to make sure it exists after recovery.
483 * Also, don't overwrite i_size on directories during replay.
484 * log replay inserts and removes directory items based on the
485 * state of the tree found in the subvolume, and i_size is modified
488 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
489 struct btrfs_inode_item *src_item;
490 struct btrfs_inode_item *dst_item;
492 src_item = (struct btrfs_inode_item *)src_ptr;
493 dst_item = (struct btrfs_inode_item *)dst_ptr;
495 if (btrfs_inode_generation(eb, src_item) == 0) {
496 struct extent_buffer *dst_eb = path->nodes[0];
497 const u64 ino_size = btrfs_inode_size(eb, src_item);
500 * For regular files an ino_size == 0 is used only when
501 * logging that an inode exists, as part of a directory
502 * fsync, and the inode wasn't fsynced before. In this
503 * case don't set the size of the inode in the fs/subvol
504 * tree, otherwise we would be throwing valid data away.
506 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
507 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
509 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
513 if (overwrite_root &&
514 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
515 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
517 saved_i_size = btrfs_inode_size(path->nodes[0],
522 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
525 if (save_old_i_size) {
526 struct btrfs_inode_item *dst_item;
527 dst_item = (struct btrfs_inode_item *)dst_ptr;
528 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
531 /* make sure the generation is filled in */
532 if (key->type == BTRFS_INODE_ITEM_KEY) {
533 struct btrfs_inode_item *dst_item;
534 dst_item = (struct btrfs_inode_item *)dst_ptr;
535 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
536 btrfs_set_inode_generation(path->nodes[0], dst_item,
541 btrfs_mark_buffer_dirty(path->nodes[0]);
542 btrfs_release_path(path);
547 * simple helper to read an inode off the disk from a given root
548 * This can only be called for subvolume roots and not for the log
550 static noinline struct inode *read_one_inode(struct btrfs_root *root,
555 inode = btrfs_iget(root->fs_info->sb, objectid, root);
561 /* replays a single extent in 'eb' at 'slot' with 'key' into the
562 * subvolume 'root'. path is released on entry and should be released
565 * extents in the log tree have not been allocated out of the extent
566 * tree yet. So, this completes the allocation, taking a reference
567 * as required if the extent already exists or creating a new extent
568 * if it isn't in the extent allocation tree yet.
570 * The extent is inserted into the file, dropping any existing extents
571 * from the file that overlap the new one.
573 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
574 struct btrfs_root *root,
575 struct btrfs_path *path,
576 struct extent_buffer *eb, int slot,
577 struct btrfs_key *key)
579 struct btrfs_fs_info *fs_info = root->fs_info;
582 u64 start = key->offset;
584 struct btrfs_file_extent_item *item;
585 struct inode *inode = NULL;
589 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
590 found_type = btrfs_file_extent_type(eb, item);
592 if (found_type == BTRFS_FILE_EXTENT_REG ||
593 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
594 nbytes = btrfs_file_extent_num_bytes(eb, item);
595 extent_end = start + nbytes;
598 * We don't add to the inodes nbytes if we are prealloc or a
601 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
603 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
604 size = btrfs_file_extent_ram_bytes(eb, item);
605 nbytes = btrfs_file_extent_ram_bytes(eb, item);
606 extent_end = ALIGN(start + size,
607 fs_info->sectorsize);
613 inode = read_one_inode(root, key->objectid);
620 * first check to see if we already have this extent in the
621 * file. This must be done before the btrfs_drop_extents run
622 * so we don't try to drop this extent.
624 ret = btrfs_lookup_file_extent(trans, root, path,
625 btrfs_ino(BTRFS_I(inode)), start, 0);
628 (found_type == BTRFS_FILE_EXTENT_REG ||
629 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
630 struct btrfs_file_extent_item cmp1;
631 struct btrfs_file_extent_item cmp2;
632 struct btrfs_file_extent_item *existing;
633 struct extent_buffer *leaf;
635 leaf = path->nodes[0];
636 existing = btrfs_item_ptr(leaf, path->slots[0],
637 struct btrfs_file_extent_item);
639 read_extent_buffer(eb, &cmp1, (unsigned long)item,
641 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
645 * we already have a pointer to this exact extent,
646 * we don't have to do anything
648 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
649 btrfs_release_path(path);
653 btrfs_release_path(path);
655 /* drop any overlapping extents */
656 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
660 if (found_type == BTRFS_FILE_EXTENT_REG ||
661 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
663 unsigned long dest_offset;
664 struct btrfs_key ins;
666 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
667 btrfs_fs_incompat(fs_info, NO_HOLES))
670 ret = btrfs_insert_empty_item(trans, root, path, key,
674 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
676 copy_extent_buffer(path->nodes[0], eb, dest_offset,
677 (unsigned long)item, sizeof(*item));
679 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
680 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
681 ins.type = BTRFS_EXTENT_ITEM_KEY;
682 offset = key->offset - btrfs_file_extent_offset(eb, item);
685 * Manually record dirty extent, as here we did a shallow
686 * file extent item copy and skip normal backref update,
687 * but modifying extent tree all by ourselves.
688 * So need to manually record dirty extent for qgroup,
689 * as the owner of the file extent changed from log tree
690 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
692 ret = btrfs_qgroup_trace_extent(trans,
693 btrfs_file_extent_disk_bytenr(eb, item),
694 btrfs_file_extent_disk_num_bytes(eb, item),
699 if (ins.objectid > 0) {
700 struct btrfs_ref ref = { 0 };
703 LIST_HEAD(ordered_sums);
706 * is this extent already allocated in the extent
707 * allocation tree? If so, just add a reference
709 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
713 } else if (ret == 0) {
714 btrfs_init_generic_ref(&ref,
715 BTRFS_ADD_DELAYED_REF,
716 ins.objectid, ins.offset, 0);
717 btrfs_init_data_ref(&ref,
718 root->root_key.objectid,
719 key->objectid, offset);
720 ret = btrfs_inc_extent_ref(trans, &ref);
725 * insert the extent pointer in the extent
728 ret = btrfs_alloc_logged_file_extent(trans,
729 root->root_key.objectid,
730 key->objectid, offset, &ins);
734 btrfs_release_path(path);
736 if (btrfs_file_extent_compression(eb, item)) {
737 csum_start = ins.objectid;
738 csum_end = csum_start + ins.offset;
740 csum_start = ins.objectid +
741 btrfs_file_extent_offset(eb, item);
742 csum_end = csum_start +
743 btrfs_file_extent_num_bytes(eb, item);
746 ret = btrfs_lookup_csums_range(root->log_root,
747 csum_start, csum_end - 1,
752 * Now delete all existing cums in the csum root that
753 * cover our range. We do this because we can have an
754 * extent that is completely referenced by one file
755 * extent item and partially referenced by another
756 * file extent item (like after using the clone or
757 * extent_same ioctls). In this case if we end up doing
758 * the replay of the one that partially references the
759 * extent first, and we do not do the csum deletion
760 * below, we can get 2 csum items in the csum tree that
761 * overlap each other. For example, imagine our log has
762 * the two following file extent items:
764 * key (257 EXTENT_DATA 409600)
765 * extent data disk byte 12845056 nr 102400
766 * extent data offset 20480 nr 20480 ram 102400
768 * key (257 EXTENT_DATA 819200)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 0 nr 102400 ram 102400
772 * Where the second one fully references the 100K extent
773 * that starts at disk byte 12845056, and the log tree
774 * has a single csum item that covers the entire range
777 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
779 * After the first file extent item is replayed, the
780 * csum tree gets the following csum item:
782 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
784 * Which covers the 20K sub-range starting at offset 20K
785 * of our extent. Now when we replay the second file
786 * extent item, if we do not delete existing csum items
787 * that cover any of its blocks, we end up getting two
788 * csum items in our csum tree that overlap each other:
790 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
791 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
793 * Which is a problem, because after this anyone trying
794 * to lookup up for the checksum of any block of our
795 * extent starting at an offset of 40K or higher, will
796 * end up looking at the second csum item only, which
797 * does not contain the checksum for any block starting
798 * at offset 40K or higher of our extent.
800 while (!list_empty(&ordered_sums)) {
801 struct btrfs_ordered_sum *sums;
802 sums = list_entry(ordered_sums.next,
803 struct btrfs_ordered_sum,
806 ret = btrfs_del_csums(trans,
811 ret = btrfs_csum_file_blocks(trans,
812 fs_info->csum_root, sums);
813 list_del(&sums->list);
819 btrfs_release_path(path);
821 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
822 /* inline extents are easy, we just overwrite them */
823 ret = overwrite_item(trans, root, path, eb, slot, key);
828 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
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 * See if a given name and sequence number found in an inode back reference are
898 * already in a directory and correctly point to this inode.
900 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
903 static noinline int inode_in_dir(struct btrfs_root *root,
904 struct btrfs_path *path,
905 u64 dirid, u64 objectid, u64 index,
906 const char *name, int name_len)
908 struct btrfs_dir_item *di;
909 struct btrfs_key location;
912 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
913 index, name, name_len, 0);
918 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
919 if (location.objectid != objectid)
925 btrfs_release_path(path);
926 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
931 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
932 if (location.objectid == objectid)
936 btrfs_release_path(path);
941 * helper function to check a log tree for a named back reference in
942 * an inode. This is used to decide if a back reference that is
943 * found in the subvolume conflicts with what we find in the log.
945 * inode backreferences may have multiple refs in a single item,
946 * during replay we process one reference at a time, and we don't
947 * want to delete valid links to a file from the subvolume if that
948 * link is also in the log.
950 static noinline int backref_in_log(struct btrfs_root *log,
951 struct btrfs_key *key,
953 const char *name, int namelen)
955 struct btrfs_path *path;
958 path = btrfs_alloc_path();
962 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
965 } else if (ret == 1) {
970 if (key->type == BTRFS_INODE_EXTREF_KEY)
971 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
976 ret = !!btrfs_find_name_in_backref(path->nodes[0],
980 btrfs_free_path(path);
984 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
985 struct btrfs_root *root,
986 struct btrfs_path *path,
987 struct btrfs_root *log_root,
988 struct btrfs_inode *dir,
989 struct btrfs_inode *inode,
990 u64 inode_objectid, u64 parent_objectid,
991 u64 ref_index, char *name, int namelen,
997 struct extent_buffer *leaf;
998 struct btrfs_dir_item *di;
999 struct btrfs_key search_key;
1000 struct btrfs_inode_extref *extref;
1003 /* Search old style refs */
1004 search_key.objectid = inode_objectid;
1005 search_key.type = BTRFS_INODE_REF_KEY;
1006 search_key.offset = parent_objectid;
1007 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1009 struct btrfs_inode_ref *victim_ref;
1011 unsigned long ptr_end;
1013 leaf = path->nodes[0];
1015 /* are we trying to overwrite a back ref for the root directory
1016 * if so, just jump out, we're done
1018 if (search_key.objectid == search_key.offset)
1021 /* check all the names in this back reference to see
1022 * if they are in the log. if so, we allow them to stay
1023 * otherwise they must be unlinked as a conflict
1025 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1026 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1027 while (ptr < ptr_end) {
1028 victim_ref = (struct btrfs_inode_ref *)ptr;
1029 victim_name_len = btrfs_inode_ref_name_len(leaf,
1031 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1035 read_extent_buffer(leaf, victim_name,
1036 (unsigned long)(victim_ref + 1),
1039 ret = backref_in_log(log_root, &search_key,
1040 parent_objectid, victim_name,
1046 inc_nlink(&inode->vfs_inode);
1047 btrfs_release_path(path);
1049 ret = btrfs_unlink_inode(trans, root, dir, inode,
1050 victim_name, victim_name_len);
1054 ret = btrfs_run_delayed_items(trans);
1062 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1066 * NOTE: we have searched root tree and checked the
1067 * corresponding ref, it does not need to check again.
1071 btrfs_release_path(path);
1073 /* Same search but for extended refs */
1074 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1075 inode_objectid, parent_objectid, 0,
1077 if (IS_ERR(extref)) {
1078 return PTR_ERR(extref);
1079 } else if (extref) {
1083 struct inode *victim_parent;
1085 leaf = path->nodes[0];
1087 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1088 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1090 while (cur_offset < item_size) {
1091 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1093 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1095 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1098 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1101 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1104 search_key.objectid = inode_objectid;
1105 search_key.type = BTRFS_INODE_EXTREF_KEY;
1106 search_key.offset = btrfs_extref_hash(parent_objectid,
1109 ret = backref_in_log(log_root, &search_key,
1110 parent_objectid, victim_name,
1117 victim_parent = read_one_inode(root,
1119 if (victim_parent) {
1120 inc_nlink(&inode->vfs_inode);
1121 btrfs_release_path(path);
1123 ret = btrfs_unlink_inode(trans, root,
1124 BTRFS_I(victim_parent),
1129 ret = btrfs_run_delayed_items(
1132 iput(victim_parent);
1141 cur_offset += victim_name_len + sizeof(*extref);
1145 btrfs_release_path(path);
1147 /* look for a conflicting sequence number */
1148 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1149 ref_index, name, namelen, 0);
1153 ret = drop_one_dir_item(trans, root, path, dir, di);
1157 btrfs_release_path(path);
1159 /* look for a conflicting name */
1160 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1165 ret = drop_one_dir_item(trans, root, path, dir, di);
1169 btrfs_release_path(path);
1174 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1175 u32 *namelen, char **name, u64 *index,
1176 u64 *parent_objectid)
1178 struct btrfs_inode_extref *extref;
1180 extref = (struct btrfs_inode_extref *)ref_ptr;
1182 *namelen = btrfs_inode_extref_name_len(eb, extref);
1183 *name = kmalloc(*namelen, GFP_NOFS);
1187 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1191 *index = btrfs_inode_extref_index(eb, extref);
1192 if (parent_objectid)
1193 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1198 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1199 u32 *namelen, char **name, u64 *index)
1201 struct btrfs_inode_ref *ref;
1203 ref = (struct btrfs_inode_ref *)ref_ptr;
1205 *namelen = btrfs_inode_ref_name_len(eb, ref);
1206 *name = kmalloc(*namelen, GFP_NOFS);
1210 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1213 *index = btrfs_inode_ref_index(eb, ref);
1219 * Take an inode reference item from the log tree and iterate all names from the
1220 * inode reference item in the subvolume tree with the same key (if it exists).
1221 * For any name that is not in the inode reference item from the log tree, do a
1222 * proper unlink of that name (that is, remove its entry from the inode
1223 * reference item and both dir index keys).
1225 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1226 struct btrfs_root *root,
1227 struct btrfs_path *path,
1228 struct btrfs_inode *inode,
1229 struct extent_buffer *log_eb,
1231 struct btrfs_key *key)
1234 unsigned long ref_ptr;
1235 unsigned long ref_end;
1236 struct extent_buffer *eb;
1239 btrfs_release_path(path);
1240 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1248 eb = path->nodes[0];
1249 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1250 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1251 while (ref_ptr < ref_end) {
1256 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1257 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1260 parent_id = key->offset;
1261 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1267 if (key->type == BTRFS_INODE_EXTREF_KEY)
1268 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1272 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1278 btrfs_release_path(path);
1279 dir = read_one_inode(root, parent_id);
1285 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1286 inode, name, namelen);
1290 * Whenever we need to check if a name exists or not, we
1291 * check the subvolume tree. So after an unlink we must
1292 * run delayed items, so that future checks for a name
1293 * during log replay see that the name does not exists
1297 ret = btrfs_run_delayed_items(trans);
1305 if (key->type == BTRFS_INODE_EXTREF_KEY)
1306 ref_ptr += sizeof(struct btrfs_inode_extref);
1308 ref_ptr += sizeof(struct btrfs_inode_ref);
1312 btrfs_release_path(path);
1316 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1317 const u8 ref_type, const char *name,
1320 struct btrfs_key key;
1321 struct btrfs_path *path;
1322 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1325 path = btrfs_alloc_path();
1329 key.objectid = btrfs_ino(BTRFS_I(inode));
1330 key.type = ref_type;
1331 if (key.type == BTRFS_INODE_REF_KEY)
1332 key.offset = parent_id;
1334 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1336 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1343 if (key.type == BTRFS_INODE_EXTREF_KEY)
1344 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1345 path->slots[0], parent_id, name, namelen);
1347 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1351 btrfs_free_path(path);
1355 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1356 struct inode *dir, struct inode *inode, const char *name,
1357 int namelen, u64 ref_index)
1359 struct btrfs_dir_item *dir_item;
1360 struct btrfs_key key;
1361 struct btrfs_path *path;
1362 struct inode *other_inode = NULL;
1365 path = btrfs_alloc_path();
1369 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1370 btrfs_ino(BTRFS_I(dir)),
1373 btrfs_release_path(path);
1375 } else if (IS_ERR(dir_item)) {
1376 ret = PTR_ERR(dir_item);
1381 * Our inode's dentry collides with the dentry of another inode which is
1382 * in the log but not yet processed since it has a higher inode number.
1383 * So delete that other dentry.
1385 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1386 btrfs_release_path(path);
1387 other_inode = read_one_inode(root, key.objectid);
1392 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1397 * If we dropped the link count to 0, bump it so that later the iput()
1398 * on the inode will not free it. We will fixup the link count later.
1400 if (other_inode->i_nlink == 0)
1401 inc_nlink(other_inode);
1403 ret = btrfs_run_delayed_items(trans);
1407 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1408 name, namelen, 0, ref_index);
1411 btrfs_free_path(path);
1417 * replay one inode back reference item found in the log tree.
1418 * eb, slot and key refer to the buffer and key found in the log tree.
1419 * root is the destination we are replaying into, and path is for temp
1420 * use by this function. (it should be released on return).
1422 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1423 struct btrfs_root *root,
1424 struct btrfs_root *log,
1425 struct btrfs_path *path,
1426 struct extent_buffer *eb, int slot,
1427 struct btrfs_key *key)
1429 struct inode *dir = NULL;
1430 struct inode *inode = NULL;
1431 unsigned long ref_ptr;
1432 unsigned long ref_end;
1436 int search_done = 0;
1437 int log_ref_ver = 0;
1438 u64 parent_objectid;
1441 int ref_struct_size;
1443 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1444 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1446 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1447 struct btrfs_inode_extref *r;
1449 ref_struct_size = sizeof(struct btrfs_inode_extref);
1451 r = (struct btrfs_inode_extref *)ref_ptr;
1452 parent_objectid = btrfs_inode_extref_parent(eb, r);
1454 ref_struct_size = sizeof(struct btrfs_inode_ref);
1455 parent_objectid = key->offset;
1457 inode_objectid = key->objectid;
1460 * it is possible that we didn't log all the parent directories
1461 * for a given inode. If we don't find the dir, just don't
1462 * copy the back ref in. The link count fixup code will take
1465 dir = read_one_inode(root, parent_objectid);
1471 inode = read_one_inode(root, inode_objectid);
1477 while (ref_ptr < ref_end) {
1479 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1480 &ref_index, &parent_objectid);
1482 * parent object can change from one array
1486 dir = read_one_inode(root, parent_objectid);
1492 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1498 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1499 btrfs_ino(BTRFS_I(inode)), ref_index,
1503 } else if (ret == 0) {
1505 * look for a conflicting back reference in the
1506 * metadata. if we find one we have to unlink that name
1507 * of the file before we add our new link. Later on, we
1508 * overwrite any existing back reference, and we don't
1509 * want to create dangling pointers in the directory.
1513 ret = __add_inode_ref(trans, root, path, log,
1518 ref_index, name, namelen,
1528 * If a reference item already exists for this inode
1529 * with the same parent and name, but different index,
1530 * drop it and the corresponding directory index entries
1531 * from the parent before adding the new reference item
1532 * and dir index entries, otherwise we would fail with
1533 * -EEXIST returned from btrfs_add_link() below.
1535 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1538 ret = btrfs_unlink_inode(trans, root,
1543 * If we dropped the link count to 0, bump it so
1544 * that later the iput() on the inode will not
1545 * free it. We will fixup the link count later.
1547 if (!ret && inode->i_nlink == 0)
1550 * Whenever we need to check if a name exists or
1551 * not, we check the subvolume tree. So after an
1552 * unlink we must run delayed items, so that future
1553 * checks for a name during log replay see that the
1554 * name does not exists anymore.
1557 ret = btrfs_run_delayed_items(trans);
1562 /* insert our name */
1563 ret = add_link(trans, root, dir, inode, name, namelen,
1568 btrfs_update_inode(trans, root, inode);
1570 /* Else, ret == 1, we already have a perfect match, we're done. */
1572 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1582 * Before we overwrite the inode reference item in the subvolume tree
1583 * with the item from the log tree, we must unlink all names from the
1584 * parent directory that are in the subvolume's tree inode reference
1585 * item, otherwise we end up with an inconsistent subvolume tree where
1586 * dir index entries exist for a name but there is no inode reference
1587 * item with the same name.
1589 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1594 /* finally write the back reference in the inode */
1595 ret = overwrite_item(trans, root, path, eb, slot, key);
1597 btrfs_release_path(path);
1604 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1605 struct btrfs_root *root, u64 ino)
1609 ret = btrfs_insert_orphan_item(trans, root, ino);
1616 static int count_inode_extrefs(struct btrfs_root *root,
1617 struct btrfs_inode *inode, struct btrfs_path *path)
1621 unsigned int nlink = 0;
1624 u64 inode_objectid = btrfs_ino(inode);
1627 struct btrfs_inode_extref *extref;
1628 struct extent_buffer *leaf;
1631 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1636 leaf = path->nodes[0];
1637 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1638 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1641 while (cur_offset < item_size) {
1642 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1643 name_len = btrfs_inode_extref_name_len(leaf, extref);
1647 cur_offset += name_len + sizeof(*extref);
1651 btrfs_release_path(path);
1653 btrfs_release_path(path);
1655 if (ret < 0 && ret != -ENOENT)
1660 static int count_inode_refs(struct btrfs_root *root,
1661 struct btrfs_inode *inode, struct btrfs_path *path)
1664 struct btrfs_key key;
1665 unsigned int nlink = 0;
1667 unsigned long ptr_end;
1669 u64 ino = btrfs_ino(inode);
1672 key.type = BTRFS_INODE_REF_KEY;
1673 key.offset = (u64)-1;
1676 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1680 if (path->slots[0] == 0)
1685 btrfs_item_key_to_cpu(path->nodes[0], &key,
1687 if (key.objectid != ino ||
1688 key.type != BTRFS_INODE_REF_KEY)
1690 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1691 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1693 while (ptr < ptr_end) {
1694 struct btrfs_inode_ref *ref;
1696 ref = (struct btrfs_inode_ref *)ptr;
1697 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1699 ptr = (unsigned long)(ref + 1) + name_len;
1703 if (key.offset == 0)
1705 if (path->slots[0] > 0) {
1710 btrfs_release_path(path);
1712 btrfs_release_path(path);
1718 * There are a few corners where the link count of the file can't
1719 * be properly maintained during replay. So, instead of adding
1720 * lots of complexity to the log code, we just scan the backrefs
1721 * for any file that has been through replay.
1723 * The scan will update the link count on the inode to reflect the
1724 * number of back refs found. If it goes down to zero, the iput
1725 * will free the inode.
1727 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1728 struct btrfs_root *root,
1729 struct inode *inode)
1731 struct btrfs_path *path;
1734 u64 ino = btrfs_ino(BTRFS_I(inode));
1736 path = btrfs_alloc_path();
1740 ret = count_inode_refs(root, BTRFS_I(inode), path);
1746 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1754 if (nlink != inode->i_nlink) {
1755 set_nlink(inode, nlink);
1756 btrfs_update_inode(trans, root, inode);
1758 BTRFS_I(inode)->index_cnt = (u64)-1;
1760 if (inode->i_nlink == 0) {
1761 if (S_ISDIR(inode->i_mode)) {
1762 ret = replay_dir_deletes(trans, root, NULL, path,
1767 ret = insert_orphan_item(trans, root, ino);
1771 btrfs_free_path(path);
1775 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1776 struct btrfs_root *root,
1777 struct btrfs_path *path)
1780 struct btrfs_key key;
1781 struct inode *inode;
1783 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1784 key.type = BTRFS_ORPHAN_ITEM_KEY;
1785 key.offset = (u64)-1;
1787 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1793 if (path->slots[0] == 0)
1798 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1799 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1800 key.type != BTRFS_ORPHAN_ITEM_KEY)
1803 ret = btrfs_del_item(trans, root, path);
1807 btrfs_release_path(path);
1808 inode = read_one_inode(root, key.offset);
1814 ret = fixup_inode_link_count(trans, root, inode);
1820 * fixup on a directory may create new entries,
1821 * make sure we always look for the highset possible
1824 key.offset = (u64)-1;
1826 btrfs_release_path(path);
1832 * record a given inode in the fixup dir so we can check its link
1833 * count when replay is done. The link count is incremented here
1834 * so the inode won't go away until we check it
1836 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1837 struct btrfs_root *root,
1838 struct btrfs_path *path,
1841 struct btrfs_key key;
1843 struct inode *inode;
1845 inode = read_one_inode(root, objectid);
1849 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1850 key.type = BTRFS_ORPHAN_ITEM_KEY;
1851 key.offset = objectid;
1853 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1855 btrfs_release_path(path);
1857 if (!inode->i_nlink)
1858 set_nlink(inode, 1);
1861 ret = btrfs_update_inode(trans, root, inode);
1862 } else if (ret == -EEXIST) {
1871 * when replaying the log for a directory, we only insert names
1872 * for inodes that actually exist. This means an fsync on a directory
1873 * does not implicitly fsync all the new files in it
1875 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 dirid, u64 index,
1878 char *name, int name_len,
1879 struct btrfs_key *location)
1881 struct inode *inode;
1885 inode = read_one_inode(root, location->objectid);
1889 dir = read_one_inode(root, dirid);
1895 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1896 name_len, 1, index);
1898 /* FIXME, put inode into FIXUP list */
1906 * take a single entry in a log directory item and replay it into
1909 * if a conflicting item exists in the subdirectory already,
1910 * the inode it points to is unlinked and put into the link count
1913 * If a name from the log points to a file or directory that does
1914 * not exist in the FS, it is skipped. fsyncs on directories
1915 * do not force down inodes inside that directory, just changes to the
1916 * names or unlinks in a directory.
1918 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1919 * non-existing inode) and 1 if the name was replayed.
1921 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1922 struct btrfs_root *root,
1923 struct btrfs_path *path,
1924 struct extent_buffer *eb,
1925 struct btrfs_dir_item *di,
1926 struct btrfs_key *key)
1930 struct btrfs_dir_item *dst_di;
1931 struct btrfs_key found_key;
1932 struct btrfs_key log_key;
1937 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1938 bool name_added = false;
1940 dir = read_one_inode(root, key->objectid);
1944 name_len = btrfs_dir_name_len(eb, di);
1945 name = kmalloc(name_len, GFP_NOFS);
1951 log_type = btrfs_dir_type(eb, di);
1952 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1955 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1956 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1957 btrfs_release_path(path);
1960 exists = (ret == 0);
1963 if (key->type == BTRFS_DIR_ITEM_KEY) {
1964 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1966 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1967 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1977 if (IS_ERR(dst_di)) {
1978 ret = PTR_ERR(dst_di);
1980 } else if (!dst_di) {
1981 /* we need a sequence number to insert, so we only
1982 * do inserts for the BTRFS_DIR_INDEX_KEY types
1984 if (key->type != BTRFS_DIR_INDEX_KEY)
1989 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1990 /* the existing item matches the logged item */
1991 if (found_key.objectid == log_key.objectid &&
1992 found_key.type == log_key.type &&
1993 found_key.offset == log_key.offset &&
1994 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1995 update_size = false;
2000 * don't drop the conflicting directory entry if the inode
2001 * for the new entry doesn't exist
2006 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2010 if (key->type == BTRFS_DIR_INDEX_KEY)
2013 btrfs_release_path(path);
2014 if (!ret && update_size) {
2015 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2016 ret = btrfs_update_inode(trans, root, dir);
2020 if (!ret && name_added)
2026 * Check if the inode reference exists in the log for the given name,
2027 * inode and parent inode
2029 found_key.objectid = log_key.objectid;
2030 found_key.type = BTRFS_INODE_REF_KEY;
2031 found_key.offset = key->objectid;
2032 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2036 /* The dentry will be added later. */
2038 update_size = false;
2042 found_key.objectid = log_key.objectid;
2043 found_key.type = BTRFS_INODE_EXTREF_KEY;
2044 found_key.offset = key->objectid;
2045 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2050 /* The dentry will be added later. */
2052 update_size = false;
2055 btrfs_release_path(path);
2056 ret = insert_one_name(trans, root, key->objectid, key->offset,
2057 name, name_len, &log_key);
2058 if (ret && ret != -ENOENT && ret != -EEXIST)
2062 update_size = false;
2068 * find all the names in a directory item and reconcile them into
2069 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2070 * one name in a directory item, but the same code gets used for
2071 * both directory index types
2073 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2074 struct btrfs_root *root,
2075 struct btrfs_path *path,
2076 struct extent_buffer *eb, int slot,
2077 struct btrfs_key *key)
2080 u32 item_size = btrfs_item_size_nr(eb, slot);
2081 struct btrfs_dir_item *di;
2084 unsigned long ptr_end;
2085 struct btrfs_path *fixup_path = NULL;
2087 ptr = btrfs_item_ptr_offset(eb, slot);
2088 ptr_end = ptr + item_size;
2089 while (ptr < ptr_end) {
2090 di = (struct btrfs_dir_item *)ptr;
2091 name_len = btrfs_dir_name_len(eb, di);
2092 ret = replay_one_name(trans, root, path, eb, di, key);
2095 ptr = (unsigned long)(di + 1);
2099 * If this entry refers to a non-directory (directories can not
2100 * have a link count > 1) and it was added in the transaction
2101 * that was not committed, make sure we fixup the link count of
2102 * the inode it the entry points to. Otherwise something like
2103 * the following would result in a directory pointing to an
2104 * inode with a wrong link that does not account for this dir
2112 * ln testdir/bar testdir/bar_link
2113 * ln testdir/foo testdir/foo_link
2114 * xfs_io -c "fsync" testdir/bar
2118 * mount fs, log replay happens
2120 * File foo would remain with a link count of 1 when it has two
2121 * entries pointing to it in the directory testdir. This would
2122 * make it impossible to ever delete the parent directory has
2123 * it would result in stale dentries that can never be deleted.
2125 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2126 struct btrfs_key di_key;
2129 fixup_path = btrfs_alloc_path();
2136 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2137 ret = link_to_fixup_dir(trans, root, fixup_path,
2144 btrfs_free_path(fixup_path);
2149 * directory replay has two parts. There are the standard directory
2150 * items in the log copied from the subvolume, and range items
2151 * created in the log while the subvolume was logged.
2153 * The range items tell us which parts of the key space the log
2154 * is authoritative for. During replay, if a key in the subvolume
2155 * directory is in a logged range item, but not actually in the log
2156 * that means it was deleted from the directory before the fsync
2157 * and should be removed.
2159 static noinline int find_dir_range(struct btrfs_root *root,
2160 struct btrfs_path *path,
2161 u64 dirid, int key_type,
2162 u64 *start_ret, u64 *end_ret)
2164 struct btrfs_key key;
2166 struct btrfs_dir_log_item *item;
2170 if (*start_ret == (u64)-1)
2173 key.objectid = dirid;
2174 key.type = key_type;
2175 key.offset = *start_ret;
2177 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2181 if (path->slots[0] == 0)
2186 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2188 if (key.type != key_type || key.objectid != dirid) {
2192 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2193 struct btrfs_dir_log_item);
2194 found_end = btrfs_dir_log_end(path->nodes[0], item);
2196 if (*start_ret >= key.offset && *start_ret <= found_end) {
2198 *start_ret = key.offset;
2199 *end_ret = found_end;
2204 /* check the next slot in the tree to see if it is a valid item */
2205 nritems = btrfs_header_nritems(path->nodes[0]);
2207 if (path->slots[0] >= nritems) {
2208 ret = btrfs_next_leaf(root, path);
2213 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2215 if (key.type != key_type || key.objectid != dirid) {
2219 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2220 struct btrfs_dir_log_item);
2221 found_end = btrfs_dir_log_end(path->nodes[0], item);
2222 *start_ret = key.offset;
2223 *end_ret = found_end;
2226 btrfs_release_path(path);
2231 * this looks for a given directory item in the log. If the directory
2232 * item is not in the log, the item is removed and the inode it points
2235 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2236 struct btrfs_root *root,
2237 struct btrfs_root *log,
2238 struct btrfs_path *path,
2239 struct btrfs_path *log_path,
2241 struct btrfs_key *dir_key)
2244 struct extent_buffer *eb;
2247 struct btrfs_dir_item *di;
2248 struct btrfs_dir_item *log_di;
2251 unsigned long ptr_end;
2253 struct inode *inode;
2254 struct btrfs_key location;
2257 eb = path->nodes[0];
2258 slot = path->slots[0];
2259 item_size = btrfs_item_size_nr(eb, slot);
2260 ptr = btrfs_item_ptr_offset(eb, slot);
2261 ptr_end = ptr + item_size;
2262 while (ptr < ptr_end) {
2263 di = (struct btrfs_dir_item *)ptr;
2264 name_len = btrfs_dir_name_len(eb, di);
2265 name = kmalloc(name_len, GFP_NOFS);
2270 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2273 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2274 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2277 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2278 log_di = btrfs_lookup_dir_index_item(trans, log,
2285 btrfs_dir_item_key_to_cpu(eb, di, &location);
2286 btrfs_release_path(path);
2287 btrfs_release_path(log_path);
2288 inode = read_one_inode(root, location.objectid);
2294 ret = link_to_fixup_dir(trans, root,
2295 path, location.objectid);
2303 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2304 BTRFS_I(inode), name, name_len);
2306 ret = btrfs_run_delayed_items(trans);
2312 /* there might still be more names under this key
2313 * check and repeat if required
2315 ret = btrfs_search_slot(NULL, root, dir_key, path,
2321 } else if (IS_ERR(log_di)) {
2323 return PTR_ERR(log_di);
2325 btrfs_release_path(log_path);
2328 ptr = (unsigned long)(di + 1);
2333 btrfs_release_path(path);
2334 btrfs_release_path(log_path);
2338 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2339 struct btrfs_root *root,
2340 struct btrfs_root *log,
2341 struct btrfs_path *path,
2344 struct btrfs_key search_key;
2345 struct btrfs_path *log_path;
2350 log_path = btrfs_alloc_path();
2354 search_key.objectid = ino;
2355 search_key.type = BTRFS_XATTR_ITEM_KEY;
2356 search_key.offset = 0;
2358 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2362 nritems = btrfs_header_nritems(path->nodes[0]);
2363 for (i = path->slots[0]; i < nritems; i++) {
2364 struct btrfs_key key;
2365 struct btrfs_dir_item *di;
2366 struct btrfs_dir_item *log_di;
2370 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2371 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2376 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2377 total_size = btrfs_item_size_nr(path->nodes[0], i);
2379 while (cur < total_size) {
2380 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2381 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2382 u32 this_len = sizeof(*di) + name_len + data_len;
2385 name = kmalloc(name_len, GFP_NOFS);
2390 read_extent_buffer(path->nodes[0], name,
2391 (unsigned long)(di + 1), name_len);
2393 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2395 btrfs_release_path(log_path);
2397 /* Doesn't exist in log tree, so delete it. */
2398 btrfs_release_path(path);
2399 di = btrfs_lookup_xattr(trans, root, path, ino,
2400 name, name_len, -1);
2407 ret = btrfs_delete_one_dir_name(trans, root,
2411 btrfs_release_path(path);
2416 if (IS_ERR(log_di)) {
2417 ret = PTR_ERR(log_di);
2421 di = (struct btrfs_dir_item *)((char *)di + this_len);
2424 ret = btrfs_next_leaf(root, path);
2430 btrfs_free_path(log_path);
2431 btrfs_release_path(path);
2437 * deletion replay happens before we copy any new directory items
2438 * out of the log or out of backreferences from inodes. It
2439 * scans the log to find ranges of keys that log is authoritative for,
2440 * and then scans the directory to find items in those ranges that are
2441 * not present in the log.
2443 * Anything we don't find in the log is unlinked and removed from the
2446 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2447 struct btrfs_root *root,
2448 struct btrfs_root *log,
2449 struct btrfs_path *path,
2450 u64 dirid, int del_all)
2454 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2456 struct btrfs_key dir_key;
2457 struct btrfs_key found_key;
2458 struct btrfs_path *log_path;
2461 dir_key.objectid = dirid;
2462 dir_key.type = BTRFS_DIR_ITEM_KEY;
2463 log_path = btrfs_alloc_path();
2467 dir = read_one_inode(root, dirid);
2468 /* it isn't an error if the inode isn't there, that can happen
2469 * because we replay the deletes before we copy in the inode item
2473 btrfs_free_path(log_path);
2481 range_end = (u64)-1;
2483 ret = find_dir_range(log, path, dirid, key_type,
2484 &range_start, &range_end);
2491 dir_key.offset = range_start;
2494 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2499 nritems = btrfs_header_nritems(path->nodes[0]);
2500 if (path->slots[0] >= nritems) {
2501 ret = btrfs_next_leaf(root, path);
2507 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2509 if (found_key.objectid != dirid ||
2510 found_key.type != dir_key.type)
2513 if (found_key.offset > range_end)
2516 ret = check_item_in_log(trans, root, log, path,
2521 if (found_key.offset == (u64)-1)
2523 dir_key.offset = found_key.offset + 1;
2525 btrfs_release_path(path);
2526 if (range_end == (u64)-1)
2528 range_start = range_end + 1;
2533 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2534 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2535 dir_key.type = BTRFS_DIR_INDEX_KEY;
2536 btrfs_release_path(path);
2540 btrfs_release_path(path);
2541 btrfs_free_path(log_path);
2547 * the process_func used to replay items from the log tree. This
2548 * gets called in two different stages. The first stage just looks
2549 * for inodes and makes sure they are all copied into the subvolume.
2551 * The second stage copies all the other item types from the log into
2552 * the subvolume. The two stage approach is slower, but gets rid of
2553 * lots of complexity around inodes referencing other inodes that exist
2554 * only in the log (references come from either directory items or inode
2557 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2558 struct walk_control *wc, u64 gen, int level)
2561 struct btrfs_path *path;
2562 struct btrfs_root *root = wc->replay_dest;
2563 struct btrfs_key key;
2567 ret = btrfs_read_buffer(eb, gen, level, NULL);
2571 level = btrfs_header_level(eb);
2576 path = btrfs_alloc_path();
2580 nritems = btrfs_header_nritems(eb);
2581 for (i = 0; i < nritems; i++) {
2582 btrfs_item_key_to_cpu(eb, &key, i);
2584 /* inode keys are done during the first stage */
2585 if (key.type == BTRFS_INODE_ITEM_KEY &&
2586 wc->stage == LOG_WALK_REPLAY_INODES) {
2587 struct btrfs_inode_item *inode_item;
2590 inode_item = btrfs_item_ptr(eb, i,
2591 struct btrfs_inode_item);
2593 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2594 * and never got linked before the fsync, skip it, as
2595 * replaying it is pointless since it would be deleted
2596 * later. We skip logging tmpfiles, but it's always
2597 * possible we are replaying a log created with a kernel
2598 * that used to log tmpfiles.
2600 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2601 wc->ignore_cur_inode = true;
2604 wc->ignore_cur_inode = false;
2606 ret = replay_xattr_deletes(wc->trans, root, log,
2607 path, key.objectid);
2610 mode = btrfs_inode_mode(eb, inode_item);
2611 if (S_ISDIR(mode)) {
2612 ret = replay_dir_deletes(wc->trans,
2613 root, log, path, key.objectid, 0);
2617 ret = overwrite_item(wc->trans, root, path,
2623 * Before replaying extents, truncate the inode to its
2624 * size. We need to do it now and not after log replay
2625 * because before an fsync we can have prealloc extents
2626 * added beyond the inode's i_size. If we did it after,
2627 * through orphan cleanup for example, we would drop
2628 * those prealloc extents just after replaying them.
2630 if (S_ISREG(mode)) {
2631 struct inode *inode;
2634 inode = read_one_inode(root, key.objectid);
2639 from = ALIGN(i_size_read(inode),
2640 root->fs_info->sectorsize);
2641 ret = btrfs_drop_extents(wc->trans, root, inode,
2644 /* Update the inode's nbytes. */
2645 ret = btrfs_update_inode(wc->trans,
2653 ret = link_to_fixup_dir(wc->trans, root,
2654 path, key.objectid);
2659 if (wc->ignore_cur_inode)
2662 if (key.type == BTRFS_DIR_INDEX_KEY &&
2663 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2664 ret = replay_one_dir_item(wc->trans, root, path,
2670 if (wc->stage < LOG_WALK_REPLAY_ALL)
2673 /* these keys are simply copied */
2674 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2675 ret = overwrite_item(wc->trans, root, path,
2679 } else if (key.type == BTRFS_INODE_REF_KEY ||
2680 key.type == BTRFS_INODE_EXTREF_KEY) {
2681 ret = add_inode_ref(wc->trans, root, log, path,
2683 if (ret && ret != -ENOENT)
2686 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2687 ret = replay_one_extent(wc->trans, root, path,
2691 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2692 ret = replay_one_dir_item(wc->trans, root, path,
2698 btrfs_free_path(path);
2703 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2705 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2707 struct btrfs_block_group *cache;
2709 cache = btrfs_lookup_block_group(fs_info, start);
2711 btrfs_err(fs_info, "unable to find block group for %llu", start);
2715 spin_lock(&cache->space_info->lock);
2716 spin_lock(&cache->lock);
2717 cache->reserved -= fs_info->nodesize;
2718 cache->space_info->bytes_reserved -= fs_info->nodesize;
2719 spin_unlock(&cache->lock);
2720 spin_unlock(&cache->space_info->lock);
2722 btrfs_put_block_group(cache);
2725 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2726 struct btrfs_root *root,
2727 struct btrfs_path *path, int *level,
2728 struct walk_control *wc)
2730 struct btrfs_fs_info *fs_info = root->fs_info;
2733 struct extent_buffer *next;
2734 struct extent_buffer *cur;
2738 while (*level > 0) {
2739 struct btrfs_key first_key;
2741 cur = path->nodes[*level];
2743 WARN_ON(btrfs_header_level(cur) != *level);
2745 if (path->slots[*level] >=
2746 btrfs_header_nritems(cur))
2749 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2750 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2751 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2752 blocksize = fs_info->nodesize;
2754 next = btrfs_find_create_tree_block(fs_info, bytenr);
2756 return PTR_ERR(next);
2759 ret = wc->process_func(root, next, wc, ptr_gen,
2762 free_extent_buffer(next);
2766 path->slots[*level]++;
2768 ret = btrfs_read_buffer(next, ptr_gen,
2769 *level - 1, &first_key);
2771 free_extent_buffer(next);
2776 btrfs_tree_lock(next);
2777 btrfs_set_lock_blocking_write(next);
2778 btrfs_clean_tree_block(next);
2779 btrfs_wait_tree_block_writeback(next);
2780 btrfs_tree_unlock(next);
2781 ret = btrfs_pin_reserved_extent(trans,
2784 free_extent_buffer(next);
2788 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2789 clear_extent_buffer_dirty(next);
2790 unaccount_log_buffer(fs_info, bytenr);
2793 free_extent_buffer(next);
2796 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2798 free_extent_buffer(next);
2802 if (path->nodes[*level-1])
2803 free_extent_buffer(path->nodes[*level-1]);
2804 path->nodes[*level-1] = next;
2805 *level = btrfs_header_level(next);
2806 path->slots[*level] = 0;
2809 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2815 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2816 struct btrfs_root *root,
2817 struct btrfs_path *path, int *level,
2818 struct walk_control *wc)
2820 struct btrfs_fs_info *fs_info = root->fs_info;
2825 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2826 slot = path->slots[i];
2827 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2830 WARN_ON(*level == 0);
2833 ret = wc->process_func(root, path->nodes[*level], wc,
2834 btrfs_header_generation(path->nodes[*level]),
2840 struct extent_buffer *next;
2842 next = path->nodes[*level];
2845 btrfs_tree_lock(next);
2846 btrfs_set_lock_blocking_write(next);
2847 btrfs_clean_tree_block(next);
2848 btrfs_wait_tree_block_writeback(next);
2849 btrfs_tree_unlock(next);
2850 ret = btrfs_pin_reserved_extent(trans,
2851 path->nodes[*level]->start,
2852 path->nodes[*level]->len);
2856 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2857 clear_extent_buffer_dirty(next);
2859 unaccount_log_buffer(fs_info,
2860 path->nodes[*level]->start);
2863 free_extent_buffer(path->nodes[*level]);
2864 path->nodes[*level] = NULL;
2872 * drop the reference count on the tree rooted at 'snap'. This traverses
2873 * the tree freeing any blocks that have a ref count of zero after being
2876 static int walk_log_tree(struct btrfs_trans_handle *trans,
2877 struct btrfs_root *log, struct walk_control *wc)
2879 struct btrfs_fs_info *fs_info = log->fs_info;
2883 struct btrfs_path *path;
2886 path = btrfs_alloc_path();
2890 level = btrfs_header_level(log->node);
2892 path->nodes[level] = log->node;
2893 atomic_inc(&log->node->refs);
2894 path->slots[level] = 0;
2897 wret = walk_down_log_tree(trans, log, path, &level, wc);
2905 wret = walk_up_log_tree(trans, log, path, &level, wc);
2914 /* was the root node processed? if not, catch it here */
2915 if (path->nodes[orig_level]) {
2916 ret = wc->process_func(log, path->nodes[orig_level], wc,
2917 btrfs_header_generation(path->nodes[orig_level]),
2922 struct extent_buffer *next;
2924 next = path->nodes[orig_level];
2927 btrfs_tree_lock(next);
2928 btrfs_set_lock_blocking_write(next);
2929 btrfs_clean_tree_block(next);
2930 btrfs_wait_tree_block_writeback(next);
2931 btrfs_tree_unlock(next);
2932 ret = btrfs_pin_reserved_extent(trans,
2933 next->start, next->len);
2937 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2938 clear_extent_buffer_dirty(next);
2939 unaccount_log_buffer(fs_info, next->start);
2945 btrfs_free_path(path);
2950 * helper function to update the item for a given subvolumes log root
2951 * in the tree of log roots
2953 static int update_log_root(struct btrfs_trans_handle *trans,
2954 struct btrfs_root *log,
2955 struct btrfs_root_item *root_item)
2957 struct btrfs_fs_info *fs_info = log->fs_info;
2960 if (log->log_transid == 1) {
2961 /* insert root item on the first sync */
2962 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2963 &log->root_key, root_item);
2965 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2966 &log->root_key, root_item);
2971 static void wait_log_commit(struct btrfs_root *root, int transid)
2974 int index = transid % 2;
2977 * we only allow two pending log transactions at a time,
2978 * so we know that if ours is more than 2 older than the
2979 * current transaction, we're done
2982 prepare_to_wait(&root->log_commit_wait[index],
2983 &wait, TASK_UNINTERRUPTIBLE);
2985 if (!(root->log_transid_committed < transid &&
2986 atomic_read(&root->log_commit[index])))
2989 mutex_unlock(&root->log_mutex);
2991 mutex_lock(&root->log_mutex);
2993 finish_wait(&root->log_commit_wait[index], &wait);
2996 static void wait_for_writer(struct btrfs_root *root)
3001 prepare_to_wait(&root->log_writer_wait, &wait,
3002 TASK_UNINTERRUPTIBLE);
3003 if (!atomic_read(&root->log_writers))
3006 mutex_unlock(&root->log_mutex);
3008 mutex_lock(&root->log_mutex);
3010 finish_wait(&root->log_writer_wait, &wait);
3013 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3014 struct btrfs_log_ctx *ctx)
3019 mutex_lock(&root->log_mutex);
3020 list_del_init(&ctx->list);
3021 mutex_unlock(&root->log_mutex);
3025 * Invoked in log mutex context, or be sure there is no other task which
3026 * can access the list.
3028 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3029 int index, int error)
3031 struct btrfs_log_ctx *ctx;
3032 struct btrfs_log_ctx *safe;
3034 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3035 list_del_init(&ctx->list);
3036 ctx->log_ret = error;
3039 INIT_LIST_HEAD(&root->log_ctxs[index]);
3043 * btrfs_sync_log does sends a given tree log down to the disk and
3044 * updates the super blocks to record it. When this call is done,
3045 * you know that any inodes previously logged are safely on disk only
3048 * Any other return value means you need to call btrfs_commit_transaction.
3049 * Some of the edge cases for fsyncing directories that have had unlinks
3050 * or renames done in the past mean that sometimes the only safe
3051 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3052 * that has happened.
3054 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3061 struct btrfs_fs_info *fs_info = root->fs_info;
3062 struct btrfs_root *log = root->log_root;
3063 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3064 struct btrfs_root_item new_root_item;
3065 int log_transid = 0;
3066 struct btrfs_log_ctx root_log_ctx;
3067 struct blk_plug plug;
3069 mutex_lock(&root->log_mutex);
3070 log_transid = ctx->log_transid;
3071 if (root->log_transid_committed >= log_transid) {
3072 mutex_unlock(&root->log_mutex);
3073 return ctx->log_ret;
3076 index1 = log_transid % 2;
3077 if (atomic_read(&root->log_commit[index1])) {
3078 wait_log_commit(root, log_transid);
3079 mutex_unlock(&root->log_mutex);
3080 return ctx->log_ret;
3082 ASSERT(log_transid == root->log_transid);
3083 atomic_set(&root->log_commit[index1], 1);
3085 /* wait for previous tree log sync to complete */
3086 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3087 wait_log_commit(root, log_transid - 1);
3090 int batch = atomic_read(&root->log_batch);
3091 /* when we're on an ssd, just kick the log commit out */
3092 if (!btrfs_test_opt(fs_info, SSD) &&
3093 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3094 mutex_unlock(&root->log_mutex);
3095 schedule_timeout_uninterruptible(1);
3096 mutex_lock(&root->log_mutex);
3098 wait_for_writer(root);
3099 if (batch == atomic_read(&root->log_batch))
3103 /* bail out if we need to do a full commit */
3104 if (btrfs_need_log_full_commit(trans)) {
3106 mutex_unlock(&root->log_mutex);
3110 if (log_transid % 2 == 0)
3111 mark = EXTENT_DIRTY;
3115 /* we start IO on all the marked extents here, but we don't actually
3116 * wait for them until later.
3118 blk_start_plug(&plug);
3119 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3121 blk_finish_plug(&plug);
3122 btrfs_abort_transaction(trans, ret);
3123 btrfs_set_log_full_commit(trans);
3124 mutex_unlock(&root->log_mutex);
3129 * We _must_ update under the root->log_mutex in order to make sure we
3130 * have a consistent view of the log root we are trying to commit at
3133 * We _must_ copy this into a local copy, because we are not holding the
3134 * log_root_tree->log_mutex yet. This is important because when we
3135 * commit the log_root_tree we must have a consistent view of the
3136 * log_root_tree when we update the super block to point at the
3137 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3138 * with the commit and possibly point at the new block which we may not
3141 btrfs_set_root_node(&log->root_item, log->node);
3142 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3144 root->log_transid++;
3145 log->log_transid = root->log_transid;
3146 root->log_start_pid = 0;
3148 * IO has been started, blocks of the log tree have WRITTEN flag set
3149 * in their headers. new modifications of the log will be written to
3150 * new positions. so it's safe to allow log writers to go in.
3152 mutex_unlock(&root->log_mutex);
3154 btrfs_init_log_ctx(&root_log_ctx, NULL);
3156 mutex_lock(&log_root_tree->log_mutex);
3158 index2 = log_root_tree->log_transid % 2;
3159 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3160 root_log_ctx.log_transid = log_root_tree->log_transid;
3163 * Now we are safe to update the log_root_tree because we're under the
3164 * log_mutex, and we're a current writer so we're holding the commit
3165 * open until we drop the log_mutex.
3167 ret = update_log_root(trans, log, &new_root_item);
3169 if (!list_empty(&root_log_ctx.list))
3170 list_del_init(&root_log_ctx.list);
3172 blk_finish_plug(&plug);
3173 btrfs_set_log_full_commit(trans);
3175 if (ret != -ENOSPC) {
3176 btrfs_abort_transaction(trans, ret);
3177 mutex_unlock(&log_root_tree->log_mutex);
3180 btrfs_wait_tree_log_extents(log, mark);
3181 mutex_unlock(&log_root_tree->log_mutex);
3186 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3187 blk_finish_plug(&plug);
3188 list_del_init(&root_log_ctx.list);
3189 mutex_unlock(&log_root_tree->log_mutex);
3190 ret = root_log_ctx.log_ret;
3194 index2 = root_log_ctx.log_transid % 2;
3195 if (atomic_read(&log_root_tree->log_commit[index2])) {
3196 blk_finish_plug(&plug);
3197 ret = btrfs_wait_tree_log_extents(log, mark);
3198 wait_log_commit(log_root_tree,
3199 root_log_ctx.log_transid);
3200 mutex_unlock(&log_root_tree->log_mutex);
3202 ret = root_log_ctx.log_ret;
3205 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3206 atomic_set(&log_root_tree->log_commit[index2], 1);
3208 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3209 wait_log_commit(log_root_tree,
3210 root_log_ctx.log_transid - 1);
3214 * now that we've moved on to the tree of log tree roots,
3215 * check the full commit flag again
3217 if (btrfs_need_log_full_commit(trans)) {
3218 blk_finish_plug(&plug);
3219 btrfs_wait_tree_log_extents(log, mark);
3220 mutex_unlock(&log_root_tree->log_mutex);
3222 goto out_wake_log_root;
3225 ret = btrfs_write_marked_extents(fs_info,
3226 &log_root_tree->dirty_log_pages,
3227 EXTENT_DIRTY | EXTENT_NEW);
3228 blk_finish_plug(&plug);
3230 btrfs_set_log_full_commit(trans);
3231 btrfs_abort_transaction(trans, ret);
3232 mutex_unlock(&log_root_tree->log_mutex);
3233 goto out_wake_log_root;
3235 ret = btrfs_wait_tree_log_extents(log, mark);
3237 ret = btrfs_wait_tree_log_extents(log_root_tree,
3238 EXTENT_NEW | EXTENT_DIRTY);
3240 btrfs_set_log_full_commit(trans);
3241 mutex_unlock(&log_root_tree->log_mutex);
3242 goto out_wake_log_root;
3245 btrfs_set_super_log_root(fs_info->super_for_commit,
3246 log_root_tree->node->start);
3247 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3248 btrfs_header_level(log_root_tree->node));
3250 log_root_tree->log_transid++;
3251 mutex_unlock(&log_root_tree->log_mutex);
3254 * Nobody else is going to jump in and write the ctree
3255 * super here because the log_commit atomic below is protecting
3256 * us. We must be called with a transaction handle pinning
3257 * the running transaction open, so a full commit can't hop
3258 * in and cause problems either.
3260 ret = write_all_supers(fs_info, 1);
3262 btrfs_set_log_full_commit(trans);
3263 btrfs_abort_transaction(trans, ret);
3264 goto out_wake_log_root;
3267 mutex_lock(&root->log_mutex);
3268 if (root->last_log_commit < log_transid)
3269 root->last_log_commit = log_transid;
3270 mutex_unlock(&root->log_mutex);
3273 mutex_lock(&log_root_tree->log_mutex);
3274 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3276 log_root_tree->log_transid_committed++;
3277 atomic_set(&log_root_tree->log_commit[index2], 0);
3278 mutex_unlock(&log_root_tree->log_mutex);
3281 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3282 * all the updates above are seen by the woken threads. It might not be
3283 * necessary, but proving that seems to be hard.
3285 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3287 mutex_lock(&root->log_mutex);
3288 btrfs_remove_all_log_ctxs(root, index1, ret);
3289 root->log_transid_committed++;
3290 atomic_set(&root->log_commit[index1], 0);
3291 mutex_unlock(&root->log_mutex);
3294 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3295 * all the updates above are seen by the woken threads. It might not be
3296 * necessary, but proving that seems to be hard.
3298 cond_wake_up(&root->log_commit_wait[index1]);
3302 static void free_log_tree(struct btrfs_trans_handle *trans,
3303 struct btrfs_root *log)
3306 struct walk_control wc = {
3308 .process_func = process_one_buffer
3311 ret = walk_log_tree(trans, log, &wc);
3314 btrfs_abort_transaction(trans, ret);
3316 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3319 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3320 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3321 extent_io_tree_release(&log->log_csum_range);
3322 btrfs_put_root(log);
3326 * free all the extents used by the tree log. This should be called
3327 * at commit time of the full transaction
3329 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3331 if (root->log_root) {
3332 free_log_tree(trans, root->log_root);
3333 root->log_root = NULL;
3334 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3339 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3340 struct btrfs_fs_info *fs_info)
3342 if (fs_info->log_root_tree) {
3343 free_log_tree(trans, fs_info->log_root_tree);
3344 fs_info->log_root_tree = NULL;
3350 * Check if an inode was logged in the current transaction. We can't always rely
3351 * on an inode's logged_trans value, because it's an in-memory only field and
3352 * therefore not persisted. This means that its value is lost if the inode gets
3353 * evicted and loaded again from disk (in which case it has a value of 0, and
3354 * certainly it is smaller then any possible transaction ID), when that happens
3355 * the full_sync flag is set in the inode's runtime flags, so on that case we
3356 * assume eviction happened and ignore the logged_trans value, assuming the
3357 * worst case, that the inode was logged before in the current transaction.
3359 static bool inode_logged(struct btrfs_trans_handle *trans,
3360 struct btrfs_inode *inode)
3362 if (inode->logged_trans == trans->transid)
3365 if (inode->last_trans == trans->transid &&
3366 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3367 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3374 * If both a file and directory are logged, and unlinks or renames are
3375 * mixed in, we have a few interesting corners:
3377 * create file X in dir Y
3378 * link file X to X.link in dir Y
3380 * unlink file X but leave X.link
3383 * After a crash we would expect only X.link to exist. But file X
3384 * didn't get fsync'd again so the log has back refs for X and X.link.
3386 * We solve this by removing directory entries and inode backrefs from the
3387 * log when a file that was logged in the current transaction is
3388 * unlinked. Any later fsync will include the updated log entries, and
3389 * we'll be able to reconstruct the proper directory items from backrefs.
3391 * This optimizations allows us to avoid relogging the entire inode
3392 * or the entire directory.
3394 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3395 struct btrfs_root *root,
3396 const char *name, int name_len,
3397 struct btrfs_inode *dir, u64 index)
3399 struct btrfs_root *log;
3400 struct btrfs_dir_item *di;
3401 struct btrfs_path *path;
3405 u64 dir_ino = btrfs_ino(dir);
3407 if (!inode_logged(trans, dir))
3410 ret = join_running_log_trans(root);
3414 mutex_lock(&dir->log_mutex);
3416 log = root->log_root;
3417 path = btrfs_alloc_path();
3423 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3424 name, name_len, -1);
3430 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3431 bytes_del += name_len;
3437 btrfs_release_path(path);
3438 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3439 index, name, name_len, -1);
3445 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3446 bytes_del += name_len;
3453 /* update the directory size in the log to reflect the names
3457 struct btrfs_key key;
3459 key.objectid = dir_ino;
3461 key.type = BTRFS_INODE_ITEM_KEY;
3462 btrfs_release_path(path);
3464 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3470 struct btrfs_inode_item *item;
3473 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3474 struct btrfs_inode_item);
3475 i_size = btrfs_inode_size(path->nodes[0], item);
3476 if (i_size > bytes_del)
3477 i_size -= bytes_del;
3480 btrfs_set_inode_size(path->nodes[0], item, i_size);
3481 btrfs_mark_buffer_dirty(path->nodes[0]);
3484 btrfs_release_path(path);
3487 btrfs_free_path(path);
3489 mutex_unlock(&dir->log_mutex);
3490 if (err == -ENOSPC) {
3491 btrfs_set_log_full_commit(trans);
3493 } else if (err < 0) {
3494 btrfs_abort_transaction(trans, err);
3497 btrfs_end_log_trans(root);
3502 /* see comments for btrfs_del_dir_entries_in_log */
3503 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3504 struct btrfs_root *root,
3505 const char *name, int name_len,
3506 struct btrfs_inode *inode, u64 dirid)
3508 struct btrfs_root *log;
3512 if (!inode_logged(trans, inode))
3515 ret = join_running_log_trans(root);
3518 log = root->log_root;
3519 mutex_lock(&inode->log_mutex);
3521 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3523 mutex_unlock(&inode->log_mutex);
3524 if (ret == -ENOSPC) {
3525 btrfs_set_log_full_commit(trans);
3527 } else if (ret < 0 && ret != -ENOENT)
3528 btrfs_abort_transaction(trans, ret);
3529 btrfs_end_log_trans(root);
3535 * creates a range item in the log for 'dirid'. first_offset and
3536 * last_offset tell us which parts of the key space the log should
3537 * be considered authoritative for.
3539 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3540 struct btrfs_root *log,
3541 struct btrfs_path *path,
3542 int key_type, u64 dirid,
3543 u64 first_offset, u64 last_offset)
3546 struct btrfs_key key;
3547 struct btrfs_dir_log_item *item;
3549 key.objectid = dirid;
3550 key.offset = first_offset;
3551 if (key_type == BTRFS_DIR_ITEM_KEY)
3552 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3554 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3555 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3559 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3560 struct btrfs_dir_log_item);
3561 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3562 btrfs_mark_buffer_dirty(path->nodes[0]);
3563 btrfs_release_path(path);
3568 * log all the items included in the current transaction for a given
3569 * directory. This also creates the range items in the log tree required
3570 * to replay anything deleted before the fsync
3572 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3573 struct btrfs_root *root, struct btrfs_inode *inode,
3574 struct btrfs_path *path,
3575 struct btrfs_path *dst_path, int key_type,
3576 struct btrfs_log_ctx *ctx,
3577 u64 min_offset, u64 *last_offset_ret)
3579 struct btrfs_key min_key;
3580 struct btrfs_root *log = root->log_root;
3581 struct extent_buffer *src;
3586 u64 first_offset = min_offset;
3587 u64 last_offset = (u64)-1;
3588 u64 ino = btrfs_ino(inode);
3590 log = root->log_root;
3592 min_key.objectid = ino;
3593 min_key.type = key_type;
3594 min_key.offset = min_offset;
3596 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3599 * we didn't find anything from this transaction, see if there
3600 * is anything at all
3602 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3603 min_key.objectid = ino;
3604 min_key.type = key_type;
3605 min_key.offset = (u64)-1;
3606 btrfs_release_path(path);
3607 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3609 btrfs_release_path(path);
3612 ret = btrfs_previous_item(root, path, ino, key_type);
3614 /* if ret == 0 there are items for this type,
3615 * create a range to tell us the last key of this type.
3616 * otherwise, there are no items in this directory after
3617 * *min_offset, and we create a range to indicate that.
3620 struct btrfs_key tmp;
3621 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3623 if (key_type == tmp.type)
3624 first_offset = max(min_offset, tmp.offset) + 1;
3629 /* go backward to find any previous key */
3630 ret = btrfs_previous_item(root, path, ino, key_type);
3632 struct btrfs_key tmp;
3633 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3634 if (key_type == tmp.type) {
3635 first_offset = tmp.offset;
3636 ret = overwrite_item(trans, log, dst_path,
3637 path->nodes[0], path->slots[0],
3645 btrfs_release_path(path);
3648 * Find the first key from this transaction again. See the note for
3649 * log_new_dir_dentries, if we're logging a directory recursively we
3650 * won't be holding its i_mutex, which means we can modify the directory
3651 * while we're logging it. If we remove an entry between our first
3652 * search and this search we'll not find the key again and can just
3656 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3661 * we have a block from this transaction, log every item in it
3662 * from our directory
3665 struct btrfs_key tmp;
3666 src = path->nodes[0];
3667 nritems = btrfs_header_nritems(src);
3668 for (i = path->slots[0]; i < nritems; i++) {
3669 struct btrfs_dir_item *di;
3671 btrfs_item_key_to_cpu(src, &min_key, i);
3673 if (min_key.objectid != ino || min_key.type != key_type)
3676 if (need_resched()) {
3677 btrfs_release_path(path);
3682 ret = overwrite_item(trans, log, dst_path, src, i,
3690 * We must make sure that when we log a directory entry,
3691 * the corresponding inode, after log replay, has a
3692 * matching link count. For example:
3698 * xfs_io -c "fsync" mydir
3700 * <mount fs and log replay>
3702 * Would result in a fsync log that when replayed, our
3703 * file inode would have a link count of 1, but we get
3704 * two directory entries pointing to the same inode.
3705 * After removing one of the names, it would not be
3706 * possible to remove the other name, which resulted
3707 * always in stale file handle errors, and would not
3708 * be possible to rmdir the parent directory, since
3709 * its i_size could never decrement to the value
3710 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3712 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3713 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3715 (btrfs_dir_transid(src, di) == trans->transid ||
3716 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3717 tmp.type != BTRFS_ROOT_ITEM_KEY)
3718 ctx->log_new_dentries = true;
3720 path->slots[0] = nritems;
3723 * look ahead to the next item and see if it is also
3724 * from this directory and from this transaction
3726 ret = btrfs_next_leaf(root, path);
3729 last_offset = (u64)-1;
3734 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3735 if (tmp.objectid != ino || tmp.type != key_type) {
3736 last_offset = (u64)-1;
3739 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3740 ret = overwrite_item(trans, log, dst_path,
3741 path->nodes[0], path->slots[0],
3746 last_offset = tmp.offset;
3751 btrfs_release_path(path);
3752 btrfs_release_path(dst_path);
3755 *last_offset_ret = last_offset;
3757 * insert the log range keys to indicate where the log
3760 ret = insert_dir_log_key(trans, log, path, key_type,
3761 ino, first_offset, last_offset);
3769 * logging directories is very similar to logging inodes, We find all the items
3770 * from the current transaction and write them to the log.
3772 * The recovery code scans the directory in the subvolume, and if it finds a
3773 * key in the range logged that is not present in the log tree, then it means
3774 * that dir entry was unlinked during the transaction.
3776 * In order for that scan to work, we must include one key smaller than
3777 * the smallest logged by this transaction and one key larger than the largest
3778 * key logged by this transaction.
3780 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3781 struct btrfs_root *root, struct btrfs_inode *inode,
3782 struct btrfs_path *path,
3783 struct btrfs_path *dst_path,
3784 struct btrfs_log_ctx *ctx)
3789 int key_type = BTRFS_DIR_ITEM_KEY;
3795 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3796 ctx, min_key, &max_key);
3799 if (max_key == (u64)-1)
3801 min_key = max_key + 1;
3804 if (key_type == BTRFS_DIR_ITEM_KEY) {
3805 key_type = BTRFS_DIR_INDEX_KEY;
3812 * a helper function to drop items from the log before we relog an
3813 * inode. max_key_type indicates the highest item type to remove.
3814 * This cannot be run for file data extents because it does not
3815 * free the extents they point to.
3817 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3818 struct btrfs_root *log,
3819 struct btrfs_path *path,
3820 u64 objectid, int max_key_type)
3823 struct btrfs_key key;
3824 struct btrfs_key found_key;
3827 key.objectid = objectid;
3828 key.type = max_key_type;
3829 key.offset = (u64)-1;
3832 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3833 BUG_ON(ret == 0); /* Logic error */
3837 if (path->slots[0] == 0)
3841 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3844 if (found_key.objectid != objectid)
3847 found_key.offset = 0;
3849 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3853 ret = btrfs_del_items(trans, log, path, start_slot,
3854 path->slots[0] - start_slot + 1);
3856 * If start slot isn't 0 then we don't need to re-search, we've
3857 * found the last guy with the objectid in this tree.
3859 if (ret || start_slot != 0)
3861 btrfs_release_path(path);
3863 btrfs_release_path(path);
3869 static void fill_inode_item(struct btrfs_trans_handle *trans,
3870 struct extent_buffer *leaf,
3871 struct btrfs_inode_item *item,
3872 struct inode *inode, int log_inode_only,
3875 struct btrfs_map_token token;
3877 btrfs_init_map_token(&token, leaf);
3879 if (log_inode_only) {
3880 /* set the generation to zero so the recover code
3881 * can tell the difference between an logging
3882 * just to say 'this inode exists' and a logging
3883 * to say 'update this inode with these values'
3885 btrfs_set_token_inode_generation(&token, item, 0);
3886 btrfs_set_token_inode_size(&token, item, logged_isize);
3888 btrfs_set_token_inode_generation(&token, item,
3889 BTRFS_I(inode)->generation);
3890 btrfs_set_token_inode_size(&token, item, inode->i_size);
3893 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3894 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3895 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3896 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3898 btrfs_set_token_timespec_sec(&token, &item->atime,
3899 inode->i_atime.tv_sec);
3900 btrfs_set_token_timespec_nsec(&token, &item->atime,
3901 inode->i_atime.tv_nsec);
3903 btrfs_set_token_timespec_sec(&token, &item->mtime,
3904 inode->i_mtime.tv_sec);
3905 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3906 inode->i_mtime.tv_nsec);
3908 btrfs_set_token_timespec_sec(&token, &item->ctime,
3909 inode->i_ctime.tv_sec);
3910 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3911 inode->i_ctime.tv_nsec);
3913 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3915 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3916 btrfs_set_token_inode_transid(&token, item, trans->transid);
3917 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3918 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3919 btrfs_set_token_inode_block_group(&token, item, 0);
3922 static int log_inode_item(struct btrfs_trans_handle *trans,
3923 struct btrfs_root *log, struct btrfs_path *path,
3924 struct btrfs_inode *inode)
3926 struct btrfs_inode_item *inode_item;
3929 ret = btrfs_insert_empty_item(trans, log, path,
3930 &inode->location, sizeof(*inode_item));
3931 if (ret && ret != -EEXIST)
3933 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3934 struct btrfs_inode_item);
3935 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3937 btrfs_release_path(path);
3941 static int log_csums(struct btrfs_trans_handle *trans,
3942 struct btrfs_inode *inode,
3943 struct btrfs_root *log_root,
3944 struct btrfs_ordered_sum *sums)
3946 const u64 lock_end = sums->bytenr + sums->len - 1;
3947 struct extent_state *cached_state = NULL;
3951 * If this inode was not used for reflink operations in the current
3952 * transaction with new extents, then do the fast path, no need to
3953 * worry about logging checksum items with overlapping ranges.
3955 if (inode->last_reflink_trans < trans->transid)
3956 return btrfs_csum_file_blocks(trans, log_root, sums);
3959 * Serialize logging for checksums. This is to avoid racing with the
3960 * same checksum being logged by another task that is logging another
3961 * file which happens to refer to the same extent as well. Such races
3962 * can leave checksum items in the log with overlapping ranges.
3964 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
3965 lock_end, &cached_state);
3969 * Due to extent cloning, we might have logged a csum item that covers a
3970 * subrange of a cloned extent, and later we can end up logging a csum
3971 * item for a larger subrange of the same extent or the entire range.
3972 * This would leave csum items in the log tree that cover the same range
3973 * and break the searches for checksums in the log tree, resulting in
3974 * some checksums missing in the fs/subvolume tree. So just delete (or
3975 * trim and adjust) any existing csum items in the log for this range.
3977 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3979 ret = btrfs_csum_file_blocks(trans, log_root, sums);
3981 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
3987 static noinline int copy_items(struct btrfs_trans_handle *trans,
3988 struct btrfs_inode *inode,
3989 struct btrfs_path *dst_path,
3990 struct btrfs_path *src_path,
3991 int start_slot, int nr, int inode_only,
3994 struct btrfs_fs_info *fs_info = trans->fs_info;
3995 unsigned long src_offset;
3996 unsigned long dst_offset;
3997 struct btrfs_root *log = inode->root->log_root;
3998 struct btrfs_file_extent_item *extent;
3999 struct btrfs_inode_item *inode_item;
4000 struct extent_buffer *src = src_path->nodes[0];
4002 struct btrfs_key *ins_keys;
4006 struct list_head ordered_sums;
4007 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4009 INIT_LIST_HEAD(&ordered_sums);
4011 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4012 nr * sizeof(u32), GFP_NOFS);
4016 ins_sizes = (u32 *)ins_data;
4017 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4019 for (i = 0; i < nr; i++) {
4020 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4021 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4023 ret = btrfs_insert_empty_items(trans, log, dst_path,
4024 ins_keys, ins_sizes, nr);
4030 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4031 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4032 dst_path->slots[0]);
4034 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4036 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4037 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4039 struct btrfs_inode_item);
4040 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4042 inode_only == LOG_INODE_EXISTS,
4045 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4046 src_offset, ins_sizes[i]);
4049 /* take a reference on file data extents so that truncates
4050 * or deletes of this inode don't have to relog the inode
4053 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4056 extent = btrfs_item_ptr(src, start_slot + i,
4057 struct btrfs_file_extent_item);
4059 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4062 found_type = btrfs_file_extent_type(src, extent);
4063 if (found_type == BTRFS_FILE_EXTENT_REG) {
4065 ds = btrfs_file_extent_disk_bytenr(src,
4067 /* ds == 0 is a hole */
4071 dl = btrfs_file_extent_disk_num_bytes(src,
4073 cs = btrfs_file_extent_offset(src, extent);
4074 cl = btrfs_file_extent_num_bytes(src,
4076 if (btrfs_file_extent_compression(src,
4082 ret = btrfs_lookup_csums_range(
4084 ds + cs, ds + cs + cl - 1,
4092 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4093 btrfs_release_path(dst_path);
4097 * we have to do this after the loop above to avoid changing the
4098 * log tree while trying to change the log tree.
4100 while (!list_empty(&ordered_sums)) {
4101 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4102 struct btrfs_ordered_sum,
4105 ret = log_csums(trans, inode, log, sums);
4106 list_del(&sums->list);
4113 static int extent_cmp(void *priv, const struct list_head *a,
4114 const struct list_head *b)
4116 struct extent_map *em1, *em2;
4118 em1 = list_entry(a, struct extent_map, list);
4119 em2 = list_entry(b, struct extent_map, list);
4121 if (em1->start < em2->start)
4123 else if (em1->start > em2->start)
4128 static int log_extent_csums(struct btrfs_trans_handle *trans,
4129 struct btrfs_inode *inode,
4130 struct btrfs_root *log_root,
4131 const struct extent_map *em,
4132 struct btrfs_log_ctx *ctx)
4134 struct btrfs_ordered_extent *ordered;
4137 u64 mod_start = em->mod_start;
4138 u64 mod_len = em->mod_len;
4139 LIST_HEAD(ordered_sums);
4142 if (inode->flags & BTRFS_INODE_NODATASUM ||
4143 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4144 em->block_start == EXTENT_MAP_HOLE)
4147 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4148 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4149 const u64 mod_end = mod_start + mod_len;
4150 struct btrfs_ordered_sum *sums;
4155 if (ordered_end <= mod_start)
4157 if (mod_end <= ordered->file_offset)
4161 * We are going to copy all the csums on this ordered extent, so
4162 * go ahead and adjust mod_start and mod_len in case this ordered
4163 * extent has already been logged.
4165 if (ordered->file_offset > mod_start) {
4166 if (ordered_end >= mod_end)
4167 mod_len = ordered->file_offset - mod_start;
4169 * If we have this case
4171 * |--------- logged extent ---------|
4172 * |----- ordered extent ----|
4174 * Just don't mess with mod_start and mod_len, we'll
4175 * just end up logging more csums than we need and it
4179 if (ordered_end < mod_end) {
4180 mod_len = mod_end - ordered_end;
4181 mod_start = ordered_end;
4188 * To keep us from looping for the above case of an ordered
4189 * extent that falls inside of the logged extent.
4191 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4194 list_for_each_entry(sums, &ordered->list, list) {
4195 ret = log_csums(trans, inode, log_root, sums);
4201 /* We're done, found all csums in the ordered extents. */
4205 /* If we're compressed we have to save the entire range of csums. */
4206 if (em->compress_type) {
4208 csum_len = max(em->block_len, em->orig_block_len);
4210 csum_offset = mod_start - em->start;
4214 /* block start is already adjusted for the file extent offset. */
4215 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4216 em->block_start + csum_offset,
4217 em->block_start + csum_offset +
4218 csum_len - 1, &ordered_sums, 0);
4222 while (!list_empty(&ordered_sums)) {
4223 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4224 struct btrfs_ordered_sum,
4227 ret = log_csums(trans, inode, log_root, sums);
4228 list_del(&sums->list);
4235 static int log_one_extent(struct btrfs_trans_handle *trans,
4236 struct btrfs_inode *inode, struct btrfs_root *root,
4237 const struct extent_map *em,
4238 struct btrfs_path *path,
4239 struct btrfs_log_ctx *ctx)
4241 struct btrfs_root *log = root->log_root;
4242 struct btrfs_file_extent_item *fi;
4243 struct extent_buffer *leaf;
4244 struct btrfs_map_token token;
4245 struct btrfs_key key;
4246 u64 extent_offset = em->start - em->orig_start;
4249 int extent_inserted = 0;
4251 ret = log_extent_csums(trans, inode, log, em, ctx);
4255 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4256 em->start + em->len, NULL, 0, 1,
4257 sizeof(*fi), &extent_inserted);
4261 if (!extent_inserted) {
4262 key.objectid = btrfs_ino(inode);
4263 key.type = BTRFS_EXTENT_DATA_KEY;
4264 key.offset = em->start;
4266 ret = btrfs_insert_empty_item(trans, log, path, &key,
4271 leaf = path->nodes[0];
4272 btrfs_init_map_token(&token, leaf);
4273 fi = btrfs_item_ptr(leaf, path->slots[0],
4274 struct btrfs_file_extent_item);
4276 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4277 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4278 btrfs_set_token_file_extent_type(&token, fi,
4279 BTRFS_FILE_EXTENT_PREALLOC);
4281 btrfs_set_token_file_extent_type(&token, fi,
4282 BTRFS_FILE_EXTENT_REG);
4284 block_len = max(em->block_len, em->orig_block_len);
4285 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4286 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4288 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4289 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4290 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4293 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4295 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4296 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4299 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4300 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4301 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4302 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4303 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4304 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4305 btrfs_mark_buffer_dirty(leaf);
4307 btrfs_release_path(path);
4313 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4314 * lose them after doing a full/fast fsync and replaying the log. We scan the
4315 * subvolume's root instead of iterating the inode's extent map tree because
4316 * otherwise we can log incorrect extent items based on extent map conversion.
4317 * That can happen due to the fact that extent maps are merged when they
4318 * are not in the extent map tree's list of modified extents.
4320 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4321 struct btrfs_inode *inode,
4322 struct btrfs_path *path)
4324 struct btrfs_root *root = inode->root;
4325 struct btrfs_key key;
4326 const u64 i_size = i_size_read(&inode->vfs_inode);
4327 const u64 ino = btrfs_ino(inode);
4328 struct btrfs_path *dst_path = NULL;
4329 bool dropped_extents = false;
4330 u64 truncate_offset = i_size;
4331 struct extent_buffer *leaf;
4337 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4341 key.type = BTRFS_EXTENT_DATA_KEY;
4342 key.offset = i_size;
4343 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4348 * We must check if there is a prealloc extent that starts before the
4349 * i_size and crosses the i_size boundary. This is to ensure later we
4350 * truncate down to the end of that extent and not to the i_size, as
4351 * otherwise we end up losing part of the prealloc extent after a log
4352 * replay and with an implicit hole if there is another prealloc extent
4353 * that starts at an offset beyond i_size.
4355 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4360 struct btrfs_file_extent_item *ei;
4362 leaf = path->nodes[0];
4363 slot = path->slots[0];
4364 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4366 if (btrfs_file_extent_type(leaf, ei) ==
4367 BTRFS_FILE_EXTENT_PREALLOC) {
4370 btrfs_item_key_to_cpu(leaf, &key, slot);
4371 extent_end = key.offset +
4372 btrfs_file_extent_num_bytes(leaf, ei);
4374 if (extent_end > i_size)
4375 truncate_offset = extent_end;
4382 leaf = path->nodes[0];
4383 slot = path->slots[0];
4385 if (slot >= btrfs_header_nritems(leaf)) {
4387 ret = copy_items(trans, inode, dst_path, path,
4388 start_slot, ins_nr, 1, 0);
4393 ret = btrfs_next_leaf(root, path);
4403 btrfs_item_key_to_cpu(leaf, &key, slot);
4404 if (key.objectid > ino)
4406 if (WARN_ON_ONCE(key.objectid < ino) ||
4407 key.type < BTRFS_EXTENT_DATA_KEY ||
4408 key.offset < i_size) {
4412 if (!dropped_extents) {
4414 * Avoid logging extent items logged in past fsync calls
4415 * and leading to duplicate keys in the log tree.
4418 ret = btrfs_truncate_inode_items(trans,
4422 BTRFS_EXTENT_DATA_KEY);
4423 } while (ret == -EAGAIN);
4426 dropped_extents = true;
4433 dst_path = btrfs_alloc_path();
4441 ret = copy_items(trans, inode, dst_path, path,
4442 start_slot, ins_nr, 1, 0);
4444 btrfs_release_path(path);
4445 btrfs_free_path(dst_path);
4449 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4450 struct btrfs_root *root,
4451 struct btrfs_inode *inode,
4452 struct btrfs_path *path,
4453 struct btrfs_log_ctx *ctx)
4455 struct btrfs_ordered_extent *ordered;
4456 struct btrfs_ordered_extent *tmp;
4457 struct extent_map *em, *n;
4458 struct list_head extents;
4459 struct extent_map_tree *tree = &inode->extent_tree;
4464 INIT_LIST_HEAD(&extents);
4466 write_lock(&tree->lock);
4467 test_gen = root->fs_info->last_trans_committed;
4469 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4470 list_del_init(&em->list);
4472 * Just an arbitrary number, this can be really CPU intensive
4473 * once we start getting a lot of extents, and really once we
4474 * have a bunch of extents we just want to commit since it will
4477 if (++num > 32768) {
4478 list_del_init(&tree->modified_extents);
4483 if (em->generation <= test_gen)
4486 /* We log prealloc extents beyond eof later. */
4487 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4488 em->start >= i_size_read(&inode->vfs_inode))
4491 /* Need a ref to keep it from getting evicted from cache */
4492 refcount_inc(&em->refs);
4493 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4494 list_add_tail(&em->list, &extents);
4498 list_sort(NULL, &extents, extent_cmp);
4500 while (!list_empty(&extents)) {
4501 em = list_entry(extents.next, struct extent_map, list);
4503 list_del_init(&em->list);
4506 * If we had an error we just need to delete everybody from our
4510 clear_em_logging(tree, em);
4511 free_extent_map(em);
4515 write_unlock(&tree->lock);
4517 ret = log_one_extent(trans, inode, root, em, path, ctx);
4518 write_lock(&tree->lock);
4519 clear_em_logging(tree, em);
4520 free_extent_map(em);
4522 WARN_ON(!list_empty(&extents));
4523 write_unlock(&tree->lock);
4525 btrfs_release_path(path);
4527 ret = btrfs_log_prealloc_extents(trans, inode, path);
4532 * We have logged all extents successfully, now make sure the commit of
4533 * the current transaction waits for the ordered extents to complete
4534 * before it commits and wipes out the log trees, otherwise we would
4535 * lose data if an ordered extents completes after the transaction
4536 * commits and a power failure happens after the transaction commit.
4538 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4539 list_del_init(&ordered->log_list);
4540 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4542 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4543 spin_lock_irq(&inode->ordered_tree.lock);
4544 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4545 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4546 atomic_inc(&trans->transaction->pending_ordered);
4548 spin_unlock_irq(&inode->ordered_tree.lock);
4550 btrfs_put_ordered_extent(ordered);
4556 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4557 struct btrfs_path *path, u64 *size_ret)
4559 struct btrfs_key key;
4562 key.objectid = btrfs_ino(inode);
4563 key.type = BTRFS_INODE_ITEM_KEY;
4566 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4569 } else if (ret > 0) {
4572 struct btrfs_inode_item *item;
4574 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4575 struct btrfs_inode_item);
4576 *size_ret = btrfs_inode_size(path->nodes[0], item);
4578 * If the in-memory inode's i_size is smaller then the inode
4579 * size stored in the btree, return the inode's i_size, so
4580 * that we get a correct inode size after replaying the log
4581 * when before a power failure we had a shrinking truncate
4582 * followed by addition of a new name (rename / new hard link).
4583 * Otherwise return the inode size from the btree, to avoid
4584 * data loss when replaying a log due to previously doing a
4585 * write that expands the inode's size and logging a new name
4586 * immediately after.
4588 if (*size_ret > inode->vfs_inode.i_size)
4589 *size_ret = inode->vfs_inode.i_size;
4592 btrfs_release_path(path);
4597 * At the moment we always log all xattrs. This is to figure out at log replay
4598 * time which xattrs must have their deletion replayed. If a xattr is missing
4599 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4600 * because if a xattr is deleted, the inode is fsynced and a power failure
4601 * happens, causing the log to be replayed the next time the fs is mounted,
4602 * we want the xattr to not exist anymore (same behaviour as other filesystems
4603 * with a journal, ext3/4, xfs, f2fs, etc).
4605 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4606 struct btrfs_root *root,
4607 struct btrfs_inode *inode,
4608 struct btrfs_path *path,
4609 struct btrfs_path *dst_path)
4612 struct btrfs_key key;
4613 const u64 ino = btrfs_ino(inode);
4616 bool found_xattrs = false;
4618 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4622 key.type = BTRFS_XATTR_ITEM_KEY;
4625 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4630 int slot = path->slots[0];
4631 struct extent_buffer *leaf = path->nodes[0];
4632 int nritems = btrfs_header_nritems(leaf);
4634 if (slot >= nritems) {
4636 ret = copy_items(trans, inode, dst_path, path,
4637 start_slot, ins_nr, 1, 0);
4642 ret = btrfs_next_leaf(root, path);
4650 btrfs_item_key_to_cpu(leaf, &key, slot);
4651 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4658 found_xattrs = true;
4662 ret = copy_items(trans, inode, dst_path, path,
4663 start_slot, ins_nr, 1, 0);
4669 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4675 * When using the NO_HOLES feature if we punched a hole that causes the
4676 * deletion of entire leafs or all the extent items of the first leaf (the one
4677 * that contains the inode item and references) we may end up not processing
4678 * any extents, because there are no leafs with a generation matching the
4679 * current transaction that have extent items for our inode. So we need to find
4680 * if any holes exist and then log them. We also need to log holes after any
4681 * truncate operation that changes the inode's size.
4683 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4684 struct btrfs_root *root,
4685 struct btrfs_inode *inode,
4686 struct btrfs_path *path)
4688 struct btrfs_fs_info *fs_info = root->fs_info;
4689 struct btrfs_key key;
4690 const u64 ino = btrfs_ino(inode);
4691 const u64 i_size = i_size_read(&inode->vfs_inode);
4692 u64 prev_extent_end = 0;
4695 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4699 key.type = BTRFS_EXTENT_DATA_KEY;
4702 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4707 struct extent_buffer *leaf = path->nodes[0];
4709 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4710 ret = btrfs_next_leaf(root, path);
4717 leaf = path->nodes[0];
4720 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4721 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4724 /* We have a hole, log it. */
4725 if (prev_extent_end < key.offset) {
4726 const u64 hole_len = key.offset - prev_extent_end;
4729 * Release the path to avoid deadlocks with other code
4730 * paths that search the root while holding locks on
4731 * leafs from the log root.
4733 btrfs_release_path(path);
4734 ret = btrfs_insert_file_extent(trans, root->log_root,
4735 ino, prev_extent_end, 0,
4736 0, hole_len, 0, hole_len,
4742 * Search for the same key again in the root. Since it's
4743 * an extent item and we are holding the inode lock, the
4744 * key must still exist. If it doesn't just emit warning
4745 * and return an error to fall back to a transaction
4748 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4751 if (WARN_ON(ret > 0))
4753 leaf = path->nodes[0];
4756 prev_extent_end = btrfs_file_extent_end(path);
4761 if (prev_extent_end < i_size) {
4764 btrfs_release_path(path);
4765 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4766 ret = btrfs_insert_file_extent(trans, root->log_root,
4767 ino, prev_extent_end, 0, 0,
4768 hole_len, 0, hole_len,
4778 * When we are logging a new inode X, check if it doesn't have a reference that
4779 * matches the reference from some other inode Y created in a past transaction
4780 * and that was renamed in the current transaction. If we don't do this, then at
4781 * log replay time we can lose inode Y (and all its files if it's a directory):
4784 * echo "hello world" > /mnt/x/foobar
4787 * mkdir /mnt/x # or touch /mnt/x
4788 * xfs_io -c fsync /mnt/x
4790 * mount fs, trigger log replay
4792 * After the log replay procedure, we would lose the first directory and all its
4793 * files (file foobar).
4794 * For the case where inode Y is not a directory we simply end up losing it:
4796 * echo "123" > /mnt/foo
4798 * mv /mnt/foo /mnt/bar
4799 * echo "abc" > /mnt/foo
4800 * xfs_io -c fsync /mnt/foo
4803 * We also need this for cases where a snapshot entry is replaced by some other
4804 * entry (file or directory) otherwise we end up with an unreplayable log due to
4805 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4806 * if it were a regular entry:
4809 * btrfs subvolume snapshot /mnt /mnt/x/snap
4810 * btrfs subvolume delete /mnt/x/snap
4813 * fsync /mnt/x or fsync some new file inside it
4816 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4817 * the same transaction.
4819 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4821 const struct btrfs_key *key,
4822 struct btrfs_inode *inode,
4823 u64 *other_ino, u64 *other_parent)
4826 struct btrfs_path *search_path;
4829 u32 item_size = btrfs_item_size_nr(eb, slot);
4831 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4833 search_path = btrfs_alloc_path();
4836 search_path->search_commit_root = 1;
4837 search_path->skip_locking = 1;
4839 while (cur_offset < item_size) {
4843 unsigned long name_ptr;
4844 struct btrfs_dir_item *di;
4846 if (key->type == BTRFS_INODE_REF_KEY) {
4847 struct btrfs_inode_ref *iref;
4849 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4850 parent = key->offset;
4851 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4852 name_ptr = (unsigned long)(iref + 1);
4853 this_len = sizeof(*iref) + this_name_len;
4855 struct btrfs_inode_extref *extref;
4857 extref = (struct btrfs_inode_extref *)(ptr +
4859 parent = btrfs_inode_extref_parent(eb, extref);
4860 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4861 name_ptr = (unsigned long)&extref->name;
4862 this_len = sizeof(*extref) + this_name_len;
4865 if (this_name_len > name_len) {
4868 new_name = krealloc(name, this_name_len, GFP_NOFS);
4873 name_len = this_name_len;
4877 read_extent_buffer(eb, name, name_ptr, this_name_len);
4878 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4879 parent, name, this_name_len, 0);
4880 if (di && !IS_ERR(di)) {
4881 struct btrfs_key di_key;
4883 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4885 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4886 if (di_key.objectid != key->objectid) {
4888 *other_ino = di_key.objectid;
4889 *other_parent = parent;
4897 } else if (IS_ERR(di)) {
4901 btrfs_release_path(search_path);
4903 cur_offset += this_len;
4907 btrfs_free_path(search_path);
4912 struct btrfs_ino_list {
4915 struct list_head list;
4918 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4919 struct btrfs_root *root,
4920 struct btrfs_path *path,
4921 struct btrfs_log_ctx *ctx,
4922 u64 ino, u64 parent)
4924 struct btrfs_ino_list *ino_elem;
4925 LIST_HEAD(inode_list);
4928 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4931 ino_elem->ino = ino;
4932 ino_elem->parent = parent;
4933 list_add_tail(&ino_elem->list, &inode_list);
4935 while (!list_empty(&inode_list)) {
4936 struct btrfs_fs_info *fs_info = root->fs_info;
4937 struct btrfs_key key;
4938 struct inode *inode;
4940 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4942 ino = ino_elem->ino;
4943 parent = ino_elem->parent;
4944 list_del(&ino_elem->list);
4949 btrfs_release_path(path);
4951 inode = btrfs_iget(fs_info->sb, ino, root);
4953 * If the other inode that had a conflicting dir entry was
4954 * deleted in the current transaction, we need to log its parent
4957 if (IS_ERR(inode)) {
4958 ret = PTR_ERR(inode);
4959 if (ret == -ENOENT) {
4960 inode = btrfs_iget(fs_info->sb, parent, root);
4961 if (IS_ERR(inode)) {
4962 ret = PTR_ERR(inode);
4964 ret = btrfs_log_inode(trans, root,
4966 LOG_OTHER_INODE_ALL,
4968 btrfs_add_delayed_iput(inode);
4974 * If the inode was already logged skip it - otherwise we can
4975 * hit an infinite loop. Example:
4977 * From the commit root (previous transaction) we have the
4980 * inode 257 a directory
4981 * inode 258 with references "zz" and "zz_link" on inode 257
4982 * inode 259 with reference "a" on inode 257
4984 * And in the current (uncommitted) transaction we have:
4986 * inode 257 a directory, unchanged
4987 * inode 258 with references "a" and "a2" on inode 257
4988 * inode 259 with reference "zz_link" on inode 257
4989 * inode 261 with reference "zz" on inode 257
4991 * When logging inode 261 the following infinite loop could
4992 * happen if we don't skip already logged inodes:
4994 * - we detect inode 258 as a conflicting inode, with inode 261
4995 * on reference "zz", and log it;
4997 * - we detect inode 259 as a conflicting inode, with inode 258
4998 * on reference "a", and log it;
5000 * - we detect inode 258 as a conflicting inode, with inode 259
5001 * on reference "zz_link", and log it - again! After this we
5002 * repeat the above steps forever.
5004 spin_lock(&BTRFS_I(inode)->lock);
5006 * Check the inode's logged_trans only instead of
5007 * btrfs_inode_in_log(). This is because the last_log_commit of
5008 * the inode is not updated when we only log that it exists and
5009 * it has the full sync bit set (see btrfs_log_inode()).
5011 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5012 spin_unlock(&BTRFS_I(inode)->lock);
5013 btrfs_add_delayed_iput(inode);
5016 spin_unlock(&BTRFS_I(inode)->lock);
5018 * We are safe logging the other inode without acquiring its
5019 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5020 * are safe against concurrent renames of the other inode as
5021 * well because during a rename we pin the log and update the
5022 * log with the new name before we unpin it.
5024 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5025 LOG_OTHER_INODE, ctx);
5027 btrfs_add_delayed_iput(inode);
5032 key.type = BTRFS_INODE_REF_KEY;
5034 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5036 btrfs_add_delayed_iput(inode);
5041 struct extent_buffer *leaf = path->nodes[0];
5042 int slot = path->slots[0];
5044 u64 other_parent = 0;
5046 if (slot >= btrfs_header_nritems(leaf)) {
5047 ret = btrfs_next_leaf(root, path);
5050 } else if (ret > 0) {
5057 btrfs_item_key_to_cpu(leaf, &key, slot);
5058 if (key.objectid != ino ||
5059 (key.type != BTRFS_INODE_REF_KEY &&
5060 key.type != BTRFS_INODE_EXTREF_KEY)) {
5065 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5066 BTRFS_I(inode), &other_ino,
5071 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5076 ino_elem->ino = other_ino;
5077 ino_elem->parent = other_parent;
5078 list_add_tail(&ino_elem->list, &inode_list);
5083 btrfs_add_delayed_iput(inode);
5089 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5090 struct btrfs_inode *inode,
5091 struct btrfs_key *min_key,
5092 const struct btrfs_key *max_key,
5093 struct btrfs_path *path,
5094 struct btrfs_path *dst_path,
5095 const u64 logged_isize,
5096 const bool recursive_logging,
5097 const int inode_only,
5098 struct btrfs_log_ctx *ctx,
5099 bool *need_log_inode_item)
5101 const u64 i_size = i_size_read(&inode->vfs_inode);
5102 struct btrfs_root *root = inode->root;
5103 int ins_start_slot = 0;
5108 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5116 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5117 if (min_key->objectid != max_key->objectid)
5119 if (min_key->type > max_key->type)
5122 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5123 *need_log_inode_item = false;
5124 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5125 min_key->offset >= i_size) {
5127 * Extents at and beyond eof are logged with
5128 * btrfs_log_prealloc_extents().
5129 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5130 * and no keys greater than that, so bail out.
5133 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5134 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5135 inode->generation == trans->transid &&
5136 !recursive_logging) {
5138 u64 other_parent = 0;
5140 ret = btrfs_check_ref_name_override(path->nodes[0],
5141 path->slots[0], min_key, inode,
5142 &other_ino, &other_parent);
5145 } else if (ret > 0 && ctx &&
5146 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5151 ins_start_slot = path->slots[0];
5153 ret = copy_items(trans, inode, dst_path, path,
5154 ins_start_slot, ins_nr,
5155 inode_only, logged_isize);
5160 ret = log_conflicting_inodes(trans, root, path,
5161 ctx, other_ino, other_parent);
5164 btrfs_release_path(path);
5167 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5168 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5171 ret = copy_items(trans, inode, dst_path, path,
5173 ins_nr, inode_only, logged_isize);
5180 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5183 } else if (!ins_nr) {
5184 ins_start_slot = path->slots[0];
5189 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5190 ins_nr, inode_only, logged_isize);
5194 ins_start_slot = path->slots[0];
5197 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5198 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5203 ret = copy_items(trans, inode, dst_path, path,
5204 ins_start_slot, ins_nr, inode_only,
5210 btrfs_release_path(path);
5212 if (min_key->offset < (u64)-1) {
5214 } else if (min_key->type < max_key->type) {
5216 min_key->offset = 0;
5222 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5223 ins_nr, inode_only, logged_isize);
5228 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5230 * Release the path because otherwise we might attempt to double
5231 * lock the same leaf with btrfs_log_prealloc_extents() below.
5233 btrfs_release_path(path);
5234 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5240 /* log a single inode in the tree log.
5241 * At least one parent directory for this inode must exist in the tree
5242 * or be logged already.
5244 * Any items from this inode changed by the current transaction are copied
5245 * to the log tree. An extra reference is taken on any extents in this
5246 * file, allowing us to avoid a whole pile of corner cases around logging
5247 * blocks that have been removed from the tree.
5249 * See LOG_INODE_ALL and related defines for a description of what inode_only
5252 * This handles both files and directories.
5254 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5255 struct btrfs_root *root, struct btrfs_inode *inode,
5257 struct btrfs_log_ctx *ctx)
5259 struct btrfs_path *path;
5260 struct btrfs_path *dst_path;
5261 struct btrfs_key min_key;
5262 struct btrfs_key max_key;
5263 struct btrfs_root *log = root->log_root;
5266 bool fast_search = false;
5267 u64 ino = btrfs_ino(inode);
5268 struct extent_map_tree *em_tree = &inode->extent_tree;
5269 u64 logged_isize = 0;
5270 bool need_log_inode_item = true;
5271 bool xattrs_logged = false;
5272 bool recursive_logging = false;
5274 path = btrfs_alloc_path();
5277 dst_path = btrfs_alloc_path();
5279 btrfs_free_path(path);
5283 min_key.objectid = ino;
5284 min_key.type = BTRFS_INODE_ITEM_KEY;
5287 max_key.objectid = ino;
5290 /* today the code can only do partial logging of directories */
5291 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5292 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5293 &inode->runtime_flags) &&
5294 inode_only >= LOG_INODE_EXISTS))
5295 max_key.type = BTRFS_XATTR_ITEM_KEY;
5297 max_key.type = (u8)-1;
5298 max_key.offset = (u64)-1;
5301 * Only run delayed items if we are a directory. We want to make sure
5302 * all directory indexes hit the fs/subvolume tree so we can find them
5303 * and figure out which index ranges have to be logged.
5305 * Otherwise commit the delayed inode only if the full sync flag is set,
5306 * as we want to make sure an up to date version is in the subvolume
5307 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5308 * it to the log tree. For a non full sync, we always log the inode item
5309 * based on the in-memory struct btrfs_inode which is always up to date.
5311 if (S_ISDIR(inode->vfs_inode.i_mode))
5312 ret = btrfs_commit_inode_delayed_items(trans, inode);
5313 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5314 ret = btrfs_commit_inode_delayed_inode(inode);
5317 btrfs_free_path(path);
5318 btrfs_free_path(dst_path);
5322 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5323 recursive_logging = true;
5324 if (inode_only == LOG_OTHER_INODE)
5325 inode_only = LOG_INODE_EXISTS;
5327 inode_only = LOG_INODE_ALL;
5328 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5330 mutex_lock(&inode->log_mutex);
5334 * For symlinks, we must always log their content, which is stored in an
5335 * inline extent, otherwise we could end up with an empty symlink after
5336 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
5337 * one attempts to create an empty symlink).
5338 * We don't need to worry about flushing delalloc, because when we create
5339 * the inline extent when the symlink is created (we never have delalloc
5342 if (S_ISLNK(inode->vfs_inode.i_mode))
5343 inode_only = LOG_INODE_ALL;
5346 * a brute force approach to making sure we get the most uptodate
5347 * copies of everything.
5349 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5350 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5352 if (inode_only == LOG_INODE_EXISTS)
5353 max_key_type = BTRFS_XATTR_ITEM_KEY;
5354 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5356 if (inode_only == LOG_INODE_EXISTS) {
5358 * Make sure the new inode item we write to the log has
5359 * the same isize as the current one (if it exists).
5360 * This is necessary to prevent data loss after log
5361 * replay, and also to prevent doing a wrong expanding
5362 * truncate - for e.g. create file, write 4K into offset
5363 * 0, fsync, write 4K into offset 4096, add hard link,
5364 * fsync some other file (to sync log), power fail - if
5365 * we use the inode's current i_size, after log replay
5366 * we get a 8Kb file, with the last 4Kb extent as a hole
5367 * (zeroes), as if an expanding truncate happened,
5368 * instead of getting a file of 4Kb only.
5370 err = logged_inode_size(log, inode, path, &logged_isize);
5374 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5375 &inode->runtime_flags)) {
5376 if (inode_only == LOG_INODE_EXISTS) {
5377 max_key.type = BTRFS_XATTR_ITEM_KEY;
5378 ret = drop_objectid_items(trans, log, path, ino,
5381 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5382 &inode->runtime_flags);
5383 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5384 &inode->runtime_flags);
5386 ret = btrfs_truncate_inode_items(trans,
5387 log, &inode->vfs_inode, 0, 0);
5392 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5393 &inode->runtime_flags) ||
5394 inode_only == LOG_INODE_EXISTS) {
5395 if (inode_only == LOG_INODE_ALL)
5397 max_key.type = BTRFS_XATTR_ITEM_KEY;
5398 ret = drop_objectid_items(trans, log, path, ino,
5401 if (inode_only == LOG_INODE_ALL)
5412 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5413 path, dst_path, logged_isize,
5414 recursive_logging, inode_only, ctx,
5415 &need_log_inode_item);
5419 btrfs_release_path(path);
5420 btrfs_release_path(dst_path);
5421 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5424 xattrs_logged = true;
5425 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5426 btrfs_release_path(path);
5427 btrfs_release_path(dst_path);
5428 err = btrfs_log_holes(trans, root, inode, path);
5433 btrfs_release_path(path);
5434 btrfs_release_path(dst_path);
5435 if (need_log_inode_item) {
5436 err = log_inode_item(trans, log, dst_path, inode);
5437 if (!err && !xattrs_logged) {
5438 err = btrfs_log_all_xattrs(trans, root, inode, path,
5440 btrfs_release_path(path);
5446 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5452 } else if (inode_only == LOG_INODE_ALL) {
5453 struct extent_map *em, *n;
5455 write_lock(&em_tree->lock);
5456 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5457 list_del_init(&em->list);
5458 write_unlock(&em_tree->lock);
5461 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5462 ret = log_directory_changes(trans, root, inode, path, dst_path,
5471 * If we are logging that an ancestor inode exists as part of logging a
5472 * new name from a link or rename operation, don't mark the inode as
5473 * logged - otherwise if an explicit fsync is made against an ancestor,
5474 * the fsync considers the inode in the log and doesn't sync the log,
5475 * resulting in the ancestor missing after a power failure unless the
5476 * log was synced as part of an fsync against any other unrelated inode.
5477 * So keep it simple for this case and just don't flag the ancestors as
5481 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5482 &inode->vfs_inode != ctx->inode)) {
5483 spin_lock(&inode->lock);
5484 inode->logged_trans = trans->transid;
5486 * Don't update last_log_commit if we logged that an inode exists
5487 * after it was loaded to memory (full_sync bit set).
5488 * This is to prevent data loss when we do a write to the inode,
5489 * then the inode gets evicted after all delalloc was flushed,
5490 * then we log it exists (due to a rename for example) and then
5491 * fsync it. This last fsync would do nothing (not logging the
5492 * extents previously written).
5494 if (inode_only != LOG_INODE_EXISTS ||
5495 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5496 inode->last_log_commit = inode->last_sub_trans;
5497 spin_unlock(&inode->lock);
5500 mutex_unlock(&inode->log_mutex);
5502 btrfs_free_path(path);
5503 btrfs_free_path(dst_path);
5508 * Check if we must fallback to a transaction commit when logging an inode.
5509 * This must be called after logging the inode and is used only in the context
5510 * when fsyncing an inode requires the need to log some other inode - in which
5511 * case we can't lock the i_mutex of each other inode we need to log as that
5512 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5513 * log inodes up or down in the hierarchy) or rename operations for example. So
5514 * we take the log_mutex of the inode after we have logged it and then check for
5515 * its last_unlink_trans value - this is safe because any task setting
5516 * last_unlink_trans must take the log_mutex and it must do this before it does
5517 * the actual unlink operation, so if we do this check before a concurrent task
5518 * sets last_unlink_trans it means we've logged a consistent version/state of
5519 * all the inode items, otherwise we are not sure and must do a transaction
5520 * commit (the concurrent task might have only updated last_unlink_trans before
5521 * we logged the inode or it might have also done the unlink).
5523 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5524 struct btrfs_inode *inode)
5526 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5529 mutex_lock(&inode->log_mutex);
5530 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5532 * Make sure any commits to the log are forced to be full
5535 btrfs_set_log_full_commit(trans);
5538 mutex_unlock(&inode->log_mutex);
5544 * follow the dentry parent pointers up the chain and see if any
5545 * of the directories in it require a full commit before they can
5546 * be logged. Returns zero if nothing special needs to be done or 1 if
5547 * a full commit is required.
5549 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5550 struct btrfs_inode *inode,
5551 struct dentry *parent,
5552 struct super_block *sb,
5556 struct dentry *old_parent = NULL;
5559 * for regular files, if its inode is already on disk, we don't
5560 * have to worry about the parents at all. This is because
5561 * we can use the last_unlink_trans field to record renames
5562 * and other fun in this file.
5564 if (S_ISREG(inode->vfs_inode.i_mode) &&
5565 inode->generation <= last_committed &&
5566 inode->last_unlink_trans <= last_committed)
5569 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5570 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5572 inode = BTRFS_I(d_inode(parent));
5576 if (btrfs_must_commit_transaction(trans, inode)) {
5581 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5584 if (IS_ROOT(parent)) {
5585 inode = BTRFS_I(d_inode(parent));
5586 if (btrfs_must_commit_transaction(trans, inode))
5591 parent = dget_parent(parent);
5593 old_parent = parent;
5594 inode = BTRFS_I(d_inode(parent));
5602 struct btrfs_dir_list {
5604 struct list_head list;
5608 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5609 * details about the why it is needed.
5610 * This is a recursive operation - if an existing dentry corresponds to a
5611 * directory, that directory's new entries are logged too (same behaviour as
5612 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5613 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5614 * complains about the following circular lock dependency / possible deadlock:
5618 * lock(&type->i_mutex_dir_key#3/2);
5619 * lock(sb_internal#2);
5620 * lock(&type->i_mutex_dir_key#3/2);
5621 * lock(&sb->s_type->i_mutex_key#14);
5623 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5624 * sb_start_intwrite() in btrfs_start_transaction().
5625 * Not locking i_mutex of the inodes is still safe because:
5627 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5628 * that while logging the inode new references (names) are added or removed
5629 * from the inode, leaving the logged inode item with a link count that does
5630 * not match the number of logged inode reference items. This is fine because
5631 * at log replay time we compute the real number of links and correct the
5632 * link count in the inode item (see replay_one_buffer() and
5633 * link_to_fixup_dir());
5635 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5636 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5637 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5638 * has a size that doesn't match the sum of the lengths of all the logged
5639 * names. This does not result in a problem because if a dir_item key is
5640 * logged but its matching dir_index key is not logged, at log replay time we
5641 * don't use it to replay the respective name (see replay_one_name()). On the
5642 * other hand if only the dir_index key ends up being logged, the respective
5643 * name is added to the fs/subvol tree with both the dir_item and dir_index
5644 * keys created (see replay_one_name()).
5645 * The directory's inode item with a wrong i_size is not a problem as well,
5646 * since we don't use it at log replay time to set the i_size in the inode
5647 * item of the fs/subvol tree (see overwrite_item()).
5649 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5650 struct btrfs_root *root,
5651 struct btrfs_inode *start_inode,
5652 struct btrfs_log_ctx *ctx)
5654 struct btrfs_fs_info *fs_info = root->fs_info;
5655 struct btrfs_root *log = root->log_root;
5656 struct btrfs_path *path;
5657 LIST_HEAD(dir_list);
5658 struct btrfs_dir_list *dir_elem;
5661 path = btrfs_alloc_path();
5665 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5667 btrfs_free_path(path);
5670 dir_elem->ino = btrfs_ino(start_inode);
5671 list_add_tail(&dir_elem->list, &dir_list);
5673 while (!list_empty(&dir_list)) {
5674 struct extent_buffer *leaf;
5675 struct btrfs_key min_key;
5679 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5682 goto next_dir_inode;
5684 min_key.objectid = dir_elem->ino;
5685 min_key.type = BTRFS_DIR_ITEM_KEY;
5688 btrfs_release_path(path);
5689 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5691 goto next_dir_inode;
5692 } else if (ret > 0) {
5694 goto next_dir_inode;
5698 leaf = path->nodes[0];
5699 nritems = btrfs_header_nritems(leaf);
5700 for (i = path->slots[0]; i < nritems; i++) {
5701 struct btrfs_dir_item *di;
5702 struct btrfs_key di_key;
5703 struct inode *di_inode;
5704 struct btrfs_dir_list *new_dir_elem;
5705 int log_mode = LOG_INODE_EXISTS;
5708 btrfs_item_key_to_cpu(leaf, &min_key, i);
5709 if (min_key.objectid != dir_elem->ino ||
5710 min_key.type != BTRFS_DIR_ITEM_KEY)
5711 goto next_dir_inode;
5713 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5714 type = btrfs_dir_type(leaf, di);
5715 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5716 type != BTRFS_FT_DIR)
5718 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5719 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5722 btrfs_release_path(path);
5723 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5724 if (IS_ERR(di_inode)) {
5725 ret = PTR_ERR(di_inode);
5726 goto next_dir_inode;
5729 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5730 btrfs_add_delayed_iput(di_inode);
5734 ctx->log_new_dentries = false;
5735 if (type == BTRFS_FT_DIR)
5736 log_mode = LOG_INODE_ALL;
5737 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5740 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5742 btrfs_add_delayed_iput(di_inode);
5744 goto next_dir_inode;
5745 if (ctx->log_new_dentries) {
5746 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5748 if (!new_dir_elem) {
5750 goto next_dir_inode;
5752 new_dir_elem->ino = di_key.objectid;
5753 list_add_tail(&new_dir_elem->list, &dir_list);
5758 ret = btrfs_next_leaf(log, path);
5760 goto next_dir_inode;
5761 } else if (ret > 0) {
5763 goto next_dir_inode;
5767 if (min_key.offset < (u64)-1) {
5772 list_del(&dir_elem->list);
5776 btrfs_free_path(path);
5780 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5781 struct btrfs_inode *inode,
5782 struct btrfs_log_ctx *ctx)
5784 struct btrfs_fs_info *fs_info = trans->fs_info;
5786 struct btrfs_path *path;
5787 struct btrfs_key key;
5788 struct btrfs_root *root = inode->root;
5789 const u64 ino = btrfs_ino(inode);
5791 path = btrfs_alloc_path();
5794 path->skip_locking = 1;
5795 path->search_commit_root = 1;
5798 key.type = BTRFS_INODE_REF_KEY;
5800 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5805 struct extent_buffer *leaf = path->nodes[0];
5806 int slot = path->slots[0];
5811 if (slot >= btrfs_header_nritems(leaf)) {
5812 ret = btrfs_next_leaf(root, path);
5820 btrfs_item_key_to_cpu(leaf, &key, slot);
5821 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5822 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5825 item_size = btrfs_item_size_nr(leaf, slot);
5826 ptr = btrfs_item_ptr_offset(leaf, slot);
5827 while (cur_offset < item_size) {
5828 struct btrfs_key inode_key;
5829 struct inode *dir_inode;
5831 inode_key.type = BTRFS_INODE_ITEM_KEY;
5832 inode_key.offset = 0;
5834 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5835 struct btrfs_inode_extref *extref;
5837 extref = (struct btrfs_inode_extref *)
5839 inode_key.objectid = btrfs_inode_extref_parent(
5841 cur_offset += sizeof(*extref);
5842 cur_offset += btrfs_inode_extref_name_len(leaf,
5845 inode_key.objectid = key.offset;
5846 cur_offset = item_size;
5849 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5852 * If the parent inode was deleted, return an error to
5853 * fallback to a transaction commit. This is to prevent
5854 * getting an inode that was moved from one parent A to
5855 * a parent B, got its former parent A deleted and then
5856 * it got fsync'ed, from existing at both parents after
5857 * a log replay (and the old parent still existing).
5864 * mv /mnt/B/bar /mnt/A/bar
5865 * mv -T /mnt/A /mnt/B
5869 * If we ignore the old parent B which got deleted,
5870 * after a log replay we would have file bar linked
5871 * at both parents and the old parent B would still
5874 if (IS_ERR(dir_inode)) {
5875 ret = PTR_ERR(dir_inode);
5880 ctx->log_new_dentries = false;
5881 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5882 LOG_INODE_ALL, ctx);
5884 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5886 if (!ret && ctx && ctx->log_new_dentries)
5887 ret = log_new_dir_dentries(trans, root,
5888 BTRFS_I(dir_inode), ctx);
5889 btrfs_add_delayed_iput(dir_inode);
5897 btrfs_free_path(path);
5901 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5902 struct btrfs_root *root,
5903 struct btrfs_path *path,
5904 struct btrfs_log_ctx *ctx)
5906 struct btrfs_key found_key;
5908 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5911 struct btrfs_fs_info *fs_info = root->fs_info;
5912 const u64 last_committed = fs_info->last_trans_committed;
5913 struct extent_buffer *leaf = path->nodes[0];
5914 int slot = path->slots[0];
5915 struct btrfs_key search_key;
5916 struct inode *inode;
5920 btrfs_release_path(path);
5922 ino = found_key.offset;
5924 search_key.objectid = found_key.offset;
5925 search_key.type = BTRFS_INODE_ITEM_KEY;
5926 search_key.offset = 0;
5927 inode = btrfs_iget(fs_info->sb, ino, root);
5929 return PTR_ERR(inode);
5931 if (BTRFS_I(inode)->generation > last_committed)
5932 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5933 LOG_INODE_EXISTS, ctx);
5934 btrfs_add_delayed_iput(inode);
5938 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5941 search_key.type = BTRFS_INODE_REF_KEY;
5942 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5946 leaf = path->nodes[0];
5947 slot = path->slots[0];
5948 if (slot >= btrfs_header_nritems(leaf)) {
5949 ret = btrfs_next_leaf(root, path);
5954 leaf = path->nodes[0];
5955 slot = path->slots[0];
5958 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5959 if (found_key.objectid != search_key.objectid ||
5960 found_key.type != BTRFS_INODE_REF_KEY)
5966 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5967 struct btrfs_inode *inode,
5968 struct dentry *parent,
5969 struct btrfs_log_ctx *ctx)
5971 struct btrfs_root *root = inode->root;
5972 struct btrfs_fs_info *fs_info = root->fs_info;
5973 struct dentry *old_parent = NULL;
5974 struct super_block *sb = inode->vfs_inode.i_sb;
5978 if (!parent || d_really_is_negative(parent) ||
5982 inode = BTRFS_I(d_inode(parent));
5983 if (root != inode->root)
5986 if (inode->generation > fs_info->last_trans_committed) {
5987 ret = btrfs_log_inode(trans, root, inode,
5988 LOG_INODE_EXISTS, ctx);
5992 if (IS_ROOT(parent))
5995 parent = dget_parent(parent);
5997 old_parent = parent;
6004 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6005 struct btrfs_inode *inode,
6006 struct dentry *parent,
6007 struct btrfs_log_ctx *ctx)
6009 struct btrfs_root *root = inode->root;
6010 const u64 ino = btrfs_ino(inode);
6011 struct btrfs_path *path;
6012 struct btrfs_key search_key;
6016 * For a single hard link case, go through a fast path that does not
6017 * need to iterate the fs/subvolume tree.
6019 if (inode->vfs_inode.i_nlink < 2)
6020 return log_new_ancestors_fast(trans, inode, parent, ctx);
6022 path = btrfs_alloc_path();
6026 search_key.objectid = ino;
6027 search_key.type = BTRFS_INODE_REF_KEY;
6028 search_key.offset = 0;
6030 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6037 struct extent_buffer *leaf = path->nodes[0];
6038 int slot = path->slots[0];
6039 struct btrfs_key found_key;
6041 if (slot >= btrfs_header_nritems(leaf)) {
6042 ret = btrfs_next_leaf(root, path);
6050 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6051 if (found_key.objectid != ino ||
6052 found_key.type > BTRFS_INODE_EXTREF_KEY)
6056 * Don't deal with extended references because they are rare
6057 * cases and too complex to deal with (we would need to keep
6058 * track of which subitem we are processing for each item in
6059 * this loop, etc). So just return some error to fallback to
6060 * a transaction commit.
6062 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6068 * Logging ancestors needs to do more searches on the fs/subvol
6069 * tree, so it releases the path as needed to avoid deadlocks.
6070 * Keep track of the last inode ref key and resume from that key
6071 * after logging all new ancestors for the current hard link.
6073 memcpy(&search_key, &found_key, sizeof(search_key));
6075 ret = log_new_ancestors(trans, root, path, ctx);
6078 btrfs_release_path(path);
6083 btrfs_free_path(path);
6088 * helper function around btrfs_log_inode to make sure newly created
6089 * parent directories also end up in the log. A minimal inode and backref
6090 * only logging is done of any parent directories that are older than
6091 * the last committed transaction
6093 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6094 struct btrfs_inode *inode,
6095 struct dentry *parent,
6097 struct btrfs_log_ctx *ctx)
6099 struct btrfs_root *root = inode->root;
6100 struct btrfs_fs_info *fs_info = root->fs_info;
6101 struct super_block *sb;
6103 u64 last_committed = fs_info->last_trans_committed;
6104 bool log_dentries = false;
6106 sb = inode->vfs_inode.i_sb;
6108 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6114 * The prev transaction commit doesn't complete, we need do
6115 * full commit by ourselves.
6117 if (fs_info->last_trans_log_full_commit >
6118 fs_info->last_trans_committed) {
6123 if (btrfs_root_refs(&root->root_item) == 0) {
6128 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
6134 * Skip already logged inodes or inodes corresponding to tmpfiles
6135 * (since logging them is pointless, a link count of 0 means they
6136 * will never be accessible).
6138 if ((btrfs_inode_in_log(inode, trans->transid) &&
6139 list_empty(&ctx->ordered_extents)) ||
6140 inode->vfs_inode.i_nlink == 0) {
6141 ret = BTRFS_NO_LOG_SYNC;
6145 ret = start_log_trans(trans, root, ctx);
6149 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6154 * for regular files, if its inode is already on disk, we don't
6155 * have to worry about the parents at all. This is because
6156 * we can use the last_unlink_trans field to record renames
6157 * and other fun in this file.
6159 if (S_ISREG(inode->vfs_inode.i_mode) &&
6160 inode->generation <= last_committed &&
6161 inode->last_unlink_trans <= last_committed) {
6166 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6167 log_dentries = true;
6170 * On unlink we must make sure all our current and old parent directory
6171 * inodes are fully logged. This is to prevent leaving dangling
6172 * directory index entries in directories that were our parents but are
6173 * not anymore. Not doing this results in old parent directory being
6174 * impossible to delete after log replay (rmdir will always fail with
6175 * error -ENOTEMPTY).
6181 * ln testdir/foo testdir/bar
6183 * unlink testdir/bar
6184 * xfs_io -c fsync testdir/foo
6186 * mount fs, triggers log replay
6188 * If we don't log the parent directory (testdir), after log replay the
6189 * directory still has an entry pointing to the file inode using the bar
6190 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6191 * the file inode has a link count of 1.
6197 * ln foo testdir/foo2
6198 * ln foo testdir/foo3
6200 * unlink testdir/foo3
6201 * xfs_io -c fsync foo
6203 * mount fs, triggers log replay
6205 * Similar as the first example, after log replay the parent directory
6206 * testdir still has an entry pointing to the inode file with name foo3
6207 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6208 * and has a link count of 2.
6210 if (inode->last_unlink_trans > last_committed) {
6211 ret = btrfs_log_all_parents(trans, inode, ctx);
6216 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6221 ret = log_new_dir_dentries(trans, root, inode, ctx);
6226 btrfs_set_log_full_commit(trans);
6231 btrfs_remove_log_ctx(root, ctx);
6232 btrfs_end_log_trans(root);
6238 * it is not safe to log dentry if the chunk root has added new
6239 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6240 * If this returns 1, you must commit the transaction to safely get your
6243 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6244 struct dentry *dentry,
6245 struct btrfs_log_ctx *ctx)
6247 struct dentry *parent = dget_parent(dentry);
6250 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6251 LOG_INODE_ALL, ctx);
6258 * should be called during mount to recover any replay any log trees
6261 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6264 struct btrfs_path *path;
6265 struct btrfs_trans_handle *trans;
6266 struct btrfs_key key;
6267 struct btrfs_key found_key;
6268 struct btrfs_root *log;
6269 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6270 struct walk_control wc = {
6271 .process_func = process_one_buffer,
6272 .stage = LOG_WALK_PIN_ONLY,
6275 path = btrfs_alloc_path();
6279 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6281 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6282 if (IS_ERR(trans)) {
6283 ret = PTR_ERR(trans);
6290 ret = walk_log_tree(trans, log_root_tree, &wc);
6292 btrfs_handle_fs_error(fs_info, ret,
6293 "Failed to pin buffers while recovering log root tree.");
6298 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6299 key.offset = (u64)-1;
6300 key.type = BTRFS_ROOT_ITEM_KEY;
6303 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6306 btrfs_handle_fs_error(fs_info, ret,
6307 "Couldn't find tree log root.");
6311 if (path->slots[0] == 0)
6315 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6317 btrfs_release_path(path);
6318 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6321 log = btrfs_read_tree_root(log_root_tree, &found_key);
6324 btrfs_handle_fs_error(fs_info, ret,
6325 "Couldn't read tree log root.");
6329 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6331 if (IS_ERR(wc.replay_dest)) {
6332 ret = PTR_ERR(wc.replay_dest);
6335 * We didn't find the subvol, likely because it was
6336 * deleted. This is ok, simply skip this log and go to
6339 * We need to exclude the root because we can't have
6340 * other log replays overwriting this log as we'll read
6341 * it back in a few more times. This will keep our
6342 * block from being modified, and we'll just bail for
6343 * each subsequent pass.
6346 ret = btrfs_pin_extent_for_log_replay(trans,
6349 btrfs_put_root(log);
6353 btrfs_handle_fs_error(fs_info, ret,
6354 "Couldn't read target root for tree log recovery.");
6358 wc.replay_dest->log_root = log;
6359 btrfs_record_root_in_trans(trans, wc.replay_dest);
6360 ret = walk_log_tree(trans, log, &wc);
6362 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6363 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6367 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6368 struct btrfs_root *root = wc.replay_dest;
6370 btrfs_release_path(path);
6373 * We have just replayed everything, and the highest
6374 * objectid of fs roots probably has changed in case
6375 * some inode_item's got replayed.
6377 * root->objectid_mutex is not acquired as log replay
6378 * could only happen during mount.
6380 ret = btrfs_find_highest_objectid(root,
6381 &root->highest_objectid);
6384 wc.replay_dest->log_root = NULL;
6385 btrfs_put_root(wc.replay_dest);
6386 btrfs_put_root(log);
6391 if (found_key.offset == 0)
6393 key.offset = found_key.offset - 1;
6395 btrfs_release_path(path);
6397 /* step one is to pin it all, step two is to replay just inodes */
6400 wc.process_func = replay_one_buffer;
6401 wc.stage = LOG_WALK_REPLAY_INODES;
6404 /* step three is to replay everything */
6405 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6410 btrfs_free_path(path);
6412 /* step 4: commit the transaction, which also unpins the blocks */
6413 ret = btrfs_commit_transaction(trans);
6417 log_root_tree->log_root = NULL;
6418 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6419 btrfs_put_root(log_root_tree);
6424 btrfs_end_transaction(wc.trans);
6425 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6426 btrfs_free_path(path);
6431 * there are some corner cases where we want to force a full
6432 * commit instead of allowing a directory to be logged.
6434 * They revolve around files there were unlinked from the directory, and
6435 * this function updates the parent directory so that a full commit is
6436 * properly done if it is fsync'd later after the unlinks are done.
6438 * Must be called before the unlink operations (updates to the subvolume tree,
6439 * inodes, etc) are done.
6441 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6442 struct btrfs_inode *dir, struct btrfs_inode *inode,
6446 * when we're logging a file, if it hasn't been renamed
6447 * or unlinked, and its inode is fully committed on disk,
6448 * we don't have to worry about walking up the directory chain
6449 * to log its parents.
6451 * So, we use the last_unlink_trans field to put this transid
6452 * into the file. When the file is logged we check it and
6453 * don't log the parents if the file is fully on disk.
6455 mutex_lock(&inode->log_mutex);
6456 inode->last_unlink_trans = trans->transid;
6457 mutex_unlock(&inode->log_mutex);
6460 * if this directory was already logged any new
6461 * names for this file/dir will get recorded
6463 if (dir->logged_trans == trans->transid)
6467 * if the inode we're about to unlink was logged,
6468 * the log will be properly updated for any new names
6470 if (inode->logged_trans == trans->transid)
6474 * when renaming files across directories, if the directory
6475 * there we're unlinking from gets fsync'd later on, there's
6476 * no way to find the destination directory later and fsync it
6477 * properly. So, we have to be conservative and force commits
6478 * so the new name gets discovered.
6483 /* we can safely do the unlink without any special recording */
6487 mutex_lock(&dir->log_mutex);
6488 dir->last_unlink_trans = trans->transid;
6489 mutex_unlock(&dir->log_mutex);
6493 * Make sure that if someone attempts to fsync the parent directory of a deleted
6494 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6495 * that after replaying the log tree of the parent directory's root we will not
6496 * see the snapshot anymore and at log replay time we will not see any log tree
6497 * corresponding to the deleted snapshot's root, which could lead to replaying
6498 * it after replaying the log tree of the parent directory (which would replay
6499 * the snapshot delete operation).
6501 * Must be called before the actual snapshot destroy operation (updates to the
6502 * parent root and tree of tree roots trees, etc) are done.
6504 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6505 struct btrfs_inode *dir)
6507 mutex_lock(&dir->log_mutex);
6508 dir->last_unlink_trans = trans->transid;
6509 mutex_unlock(&dir->log_mutex);
6513 * Call this after adding a new name for a file and it will properly
6514 * update the log to reflect the new name.
6516 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6517 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6518 struct dentry *parent)
6520 struct btrfs_log_ctx ctx;
6523 * this will force the logging code to walk the dentry chain
6526 if (!S_ISDIR(inode->vfs_inode.i_mode))
6527 inode->last_unlink_trans = trans->transid;
6530 * if this inode hasn't been logged and directory we're renaming it
6531 * from hasn't been logged, we don't need to log it
6533 if (!inode_logged(trans, inode) &&
6534 (!old_dir || !inode_logged(trans, old_dir)))
6537 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6538 ctx.logging_new_name = true;
6540 * We don't care about the return value. If we fail to log the new name
6541 * then we know the next attempt to sync the log will fallback to a full
6542 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6543 * we don't need to worry about getting a log committed that has an
6544 * inconsistent state after a rename operation.
6546 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);