1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
15 #include "print-tree.h"
17 #include "compression.h"
19 #include "inode-map.h"
21 /* magic values for the inode_only field in btrfs_log_inode:
23 * LOG_INODE_ALL means to log everything
24 * LOG_INODE_EXISTS means to log just enough to recreate the inode
27 #define LOG_INODE_ALL 0
28 #define LOG_INODE_EXISTS 1
29 #define LOG_OTHER_INODE 2
30 #define LOG_OTHER_INODE_ALL 3
33 * directory trouble cases
35 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
36 * log, we must force a full commit before doing an fsync of the directory
37 * where the unlink was done.
38 * ---> record transid of last unlink/rename per directory
42 * rename foo/some_dir foo2/some_dir
44 * fsync foo/some_dir/some_file
46 * The fsync above will unlink the original some_dir without recording
47 * it in its new location (foo2). After a crash, some_dir will be gone
48 * unless the fsync of some_file forces a full commit
50 * 2) we must log any new names for any file or dir that is in the fsync
51 * log. ---> check inode while renaming/linking.
53 * 2a) we must log any new names for any file or dir during rename
54 * when the directory they are being removed from was logged.
55 * ---> check inode and old parent dir during rename
57 * 2a is actually the more important variant. With the extra logging
58 * a crash might unlink the old name without recreating the new one
60 * 3) after a crash, we must go through any directories with a link count
61 * of zero and redo the rm -rf
68 * The directory f1 was fully removed from the FS, but fsync was never
69 * called on f1, only its parent dir. After a crash the rm -rf must
70 * be replayed. This must be able to recurse down the entire
71 * directory tree. The inode link count fixup code takes care of the
76 * stages for the tree walking. The first
77 * stage (0) is to only pin down the blocks we find
78 * the second stage (1) is to make sure that all the inodes
79 * we find in the log are created in the subvolume.
81 * The last stage is to deal with directories and links and extents
82 * and all the other fun semantics
84 #define LOG_WALK_PIN_ONLY 0
85 #define LOG_WALK_REPLAY_INODES 1
86 #define LOG_WALK_REPLAY_DIR_INDEX 2
87 #define LOG_WALK_REPLAY_ALL 3
89 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root, struct btrfs_inode *inode,
94 struct btrfs_log_ctx *ctx);
95 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 struct btrfs_path *path, u64 objectid);
98 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
99 struct btrfs_root *root,
100 struct btrfs_root *log,
101 struct btrfs_path *path,
102 u64 dirid, int del_all);
105 * tree logging is a special write ahead log used to make sure that
106 * fsyncs and O_SYNCs can happen without doing full tree commits.
108 * Full tree commits are expensive because they require commonly
109 * modified blocks to be recowed, creating many dirty pages in the
110 * extent tree an 4x-6x higher write load than ext3.
112 * Instead of doing a tree commit on every fsync, we use the
113 * key ranges and transaction ids to find items for a given file or directory
114 * that have changed in this transaction. Those items are copied into
115 * a special tree (one per subvolume root), that tree is written to disk
116 * and then the fsync is considered complete.
118 * After a crash, items are copied out of the log-tree back into the
119 * subvolume tree. Any file data extents found are recorded in the extent
120 * allocation tree, and the log-tree freed.
122 * The log tree is read three times, once to pin down all the extents it is
123 * using in ram and once, once to create all the inodes logged in the tree
124 * and once to do all the other items.
128 * start a sub transaction and setup the log tree
129 * this increments the log tree writer count to make the people
130 * syncing the tree wait for us to finish
132 static int start_log_trans(struct btrfs_trans_handle *trans,
133 struct btrfs_root *root,
134 struct btrfs_log_ctx *ctx)
136 struct btrfs_fs_info *fs_info = root->fs_info;
139 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
142 if (btrfs_need_log_full_commit(fs_info, trans)) {
147 if (!root->log_start_pid) {
148 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
149 root->log_start_pid = current->pid;
150 } else if (root->log_start_pid != current->pid) {
151 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
154 mutex_lock(&fs_info->tree_log_mutex);
155 if (!fs_info->log_root_tree)
156 ret = btrfs_init_log_root_tree(trans, fs_info);
157 mutex_unlock(&fs_info->tree_log_mutex);
161 ret = btrfs_add_log_tree(trans, root);
165 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
166 root->log_start_pid = current->pid;
169 atomic_inc(&root->log_batch);
170 atomic_inc(&root->log_writers);
172 int index = root->log_transid % 2;
173 list_add_tail(&ctx->list, &root->log_ctxs[index]);
174 ctx->log_transid = root->log_transid;
178 mutex_unlock(&root->log_mutex);
183 * returns 0 if there was a log transaction running and we were able
184 * to join, or returns -ENOENT if there were not transactions
187 static int join_running_log_trans(struct btrfs_root *root)
195 mutex_lock(&root->log_mutex);
196 if (root->log_root) {
198 atomic_inc(&root->log_writers);
200 mutex_unlock(&root->log_mutex);
205 * This either makes the current running log transaction wait
206 * until you call btrfs_end_log_trans() or it makes any future
207 * log transactions wait until you call btrfs_end_log_trans()
209 void btrfs_pin_log_trans(struct btrfs_root *root)
211 mutex_lock(&root->log_mutex);
212 atomic_inc(&root->log_writers);
213 mutex_unlock(&root->log_mutex);
217 * indicate we're done making changes to the log tree
218 * and wake up anyone waiting to do a sync
220 void btrfs_end_log_trans(struct btrfs_root *root)
222 if (atomic_dec_and_test(&root->log_writers)) {
223 /* atomic_dec_and_test implies a barrier */
224 cond_wake_up_nomb(&root->log_writer_wait);
230 * the walk control struct is used to pass state down the chain when
231 * processing the log tree. The stage field tells us which part
232 * of the log tree processing we are currently doing. The others
233 * are state fields used for that specific part
235 struct walk_control {
236 /* should we free the extent on disk when done? This is used
237 * at transaction commit time while freeing a log tree
241 /* should we write out the extent buffer? This is used
242 * while flushing the log tree to disk during a sync
246 /* should we wait for the extent buffer io to finish? Also used
247 * while flushing the log tree to disk for a sync
251 /* pin only walk, we record which extents on disk belong to the
256 /* what stage of the replay code we're currently in */
260 * Ignore any items from the inode currently being processed. Needs
261 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
262 * the LOG_WALK_REPLAY_INODES stage.
264 bool ignore_cur_inode;
266 /* the root we are currently replaying */
267 struct btrfs_root *replay_dest;
269 /* the trans handle for the current replay */
270 struct btrfs_trans_handle *trans;
272 /* the function that gets used to process blocks we find in the
273 * tree. Note the extent_buffer might not be up to date when it is
274 * passed in, and it must be checked or read if you need the data
277 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
278 struct walk_control *wc, u64 gen, int level);
282 * process_func used to pin down extents, write them or wait on them
284 static int process_one_buffer(struct btrfs_root *log,
285 struct extent_buffer *eb,
286 struct walk_control *wc, u64 gen, int level)
288 struct btrfs_fs_info *fs_info = log->fs_info;
292 * If this fs is mixed then we need to be able to process the leaves to
293 * pin down any logged extents, so we have to read the block.
295 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
296 ret = btrfs_read_buffer(eb, gen, level, NULL);
302 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
305 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
306 if (wc->pin && btrfs_header_level(eb) == 0)
307 ret = btrfs_exclude_logged_extents(fs_info, eb);
309 btrfs_write_tree_block(eb);
311 btrfs_wait_tree_block_writeback(eb);
317 * Item overwrite used by replay and tree logging. eb, slot and key all refer
318 * to the src data we are copying out.
320 * root is the tree we are copying into, and path is a scratch
321 * path for use in this function (it should be released on entry and
322 * will be released on exit).
324 * If the key is already in the destination tree the existing item is
325 * overwritten. If the existing item isn't big enough, it is extended.
326 * If it is too large, it is truncated.
328 * If the key isn't in the destination yet, a new item is inserted.
330 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
331 struct btrfs_root *root,
332 struct btrfs_path *path,
333 struct extent_buffer *eb, int slot,
334 struct btrfs_key *key)
336 struct btrfs_fs_info *fs_info = root->fs_info;
339 u64 saved_i_size = 0;
340 int save_old_i_size = 0;
341 unsigned long src_ptr;
342 unsigned long dst_ptr;
343 int overwrite_root = 0;
344 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
346 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
349 item_size = btrfs_item_size_nr(eb, slot);
350 src_ptr = btrfs_item_ptr_offset(eb, slot);
352 /* look for the key in the destination tree */
353 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
360 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
362 if (dst_size != item_size)
365 if (item_size == 0) {
366 btrfs_release_path(path);
369 dst_copy = kmalloc(item_size, GFP_NOFS);
370 src_copy = kmalloc(item_size, GFP_NOFS);
371 if (!dst_copy || !src_copy) {
372 btrfs_release_path(path);
378 read_extent_buffer(eb, src_copy, src_ptr, item_size);
380 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
381 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
383 ret = memcmp(dst_copy, src_copy, item_size);
388 * they have the same contents, just return, this saves
389 * us from cowing blocks in the destination tree and doing
390 * extra writes that may not have been done by a previous
394 btrfs_release_path(path);
399 * We need to load the old nbytes into the inode so when we
400 * replay the extents we've logged we get the right nbytes.
403 struct btrfs_inode_item *item;
407 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
408 struct btrfs_inode_item);
409 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
410 item = btrfs_item_ptr(eb, slot,
411 struct btrfs_inode_item);
412 btrfs_set_inode_nbytes(eb, item, nbytes);
415 * If this is a directory we need to reset the i_size to
416 * 0 so that we can set it up properly when replaying
417 * the rest of the items in this log.
419 mode = btrfs_inode_mode(eb, item);
421 btrfs_set_inode_size(eb, item, 0);
423 } else if (inode_item) {
424 struct btrfs_inode_item *item;
428 * New inode, set nbytes to 0 so that the nbytes comes out
429 * properly when we replay the extents.
431 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
432 btrfs_set_inode_nbytes(eb, item, 0);
435 * If this is a directory we need to reset the i_size to 0 so
436 * that we can set it up properly when replaying the rest of
437 * the items in this log.
439 mode = btrfs_inode_mode(eb, item);
441 btrfs_set_inode_size(eb, item, 0);
444 btrfs_release_path(path);
445 /* try to insert the key into the destination tree */
446 path->skip_release_on_error = 1;
447 ret = btrfs_insert_empty_item(trans, root, path,
449 path->skip_release_on_error = 0;
451 /* make sure any existing item is the correct size */
452 if (ret == -EEXIST || ret == -EOVERFLOW) {
454 found_size = btrfs_item_size_nr(path->nodes[0],
456 if (found_size > item_size)
457 btrfs_truncate_item(fs_info, path, item_size, 1);
458 else if (found_size < item_size)
459 btrfs_extend_item(fs_info, path,
460 item_size - found_size);
464 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
467 /* don't overwrite an existing inode if the generation number
468 * was logged as zero. This is done when the tree logging code
469 * is just logging an inode to make sure it exists after recovery.
471 * Also, don't overwrite i_size on directories during replay.
472 * log replay inserts and removes directory items based on the
473 * state of the tree found in the subvolume, and i_size is modified
476 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
477 struct btrfs_inode_item *src_item;
478 struct btrfs_inode_item *dst_item;
480 src_item = (struct btrfs_inode_item *)src_ptr;
481 dst_item = (struct btrfs_inode_item *)dst_ptr;
483 if (btrfs_inode_generation(eb, src_item) == 0) {
484 struct extent_buffer *dst_eb = path->nodes[0];
485 const u64 ino_size = btrfs_inode_size(eb, src_item);
488 * For regular files an ino_size == 0 is used only when
489 * logging that an inode exists, as part of a directory
490 * fsync, and the inode wasn't fsynced before. In this
491 * case don't set the size of the inode in the fs/subvol
492 * tree, otherwise we would be throwing valid data away.
494 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
495 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
497 struct btrfs_map_token token;
499 btrfs_init_map_token(&token);
500 btrfs_set_token_inode_size(dst_eb, dst_item,
506 if (overwrite_root &&
507 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
508 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
510 saved_i_size = btrfs_inode_size(path->nodes[0],
515 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
518 if (save_old_i_size) {
519 struct btrfs_inode_item *dst_item;
520 dst_item = (struct btrfs_inode_item *)dst_ptr;
521 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
524 /* make sure the generation is filled in */
525 if (key->type == BTRFS_INODE_ITEM_KEY) {
526 struct btrfs_inode_item *dst_item;
527 dst_item = (struct btrfs_inode_item *)dst_ptr;
528 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
529 btrfs_set_inode_generation(path->nodes[0], dst_item,
534 btrfs_mark_buffer_dirty(path->nodes[0]);
535 btrfs_release_path(path);
540 * simple helper to read an inode off the disk from a given root
541 * This can only be called for subvolume roots and not for the log
543 static noinline struct inode *read_one_inode(struct btrfs_root *root,
546 struct btrfs_key key;
549 key.objectid = objectid;
550 key.type = BTRFS_INODE_ITEM_KEY;
552 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
558 /* replays a single extent in 'eb' at 'slot' with 'key' into the
559 * subvolume 'root'. path is released on entry and should be released
562 * extents in the log tree have not been allocated out of the extent
563 * tree yet. So, this completes the allocation, taking a reference
564 * as required if the extent already exists or creating a new extent
565 * if it isn't in the extent allocation tree yet.
567 * The extent is inserted into the file, dropping any existing extents
568 * from the file that overlap the new one.
570 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
571 struct btrfs_root *root,
572 struct btrfs_path *path,
573 struct extent_buffer *eb, int slot,
574 struct btrfs_key *key)
576 struct btrfs_fs_info *fs_info = root->fs_info;
579 u64 start = key->offset;
581 struct btrfs_file_extent_item *item;
582 struct inode *inode = NULL;
586 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
587 found_type = btrfs_file_extent_type(eb, item);
589 if (found_type == BTRFS_FILE_EXTENT_REG ||
590 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
591 nbytes = btrfs_file_extent_num_bytes(eb, item);
592 extent_end = start + nbytes;
595 * We don't add to the inodes nbytes if we are prealloc or a
598 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
600 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
601 size = btrfs_file_extent_ram_bytes(eb, item);
602 nbytes = btrfs_file_extent_ram_bytes(eb, item);
603 extent_end = ALIGN(start + size,
604 fs_info->sectorsize);
610 inode = read_one_inode(root, key->objectid);
617 * first check to see if we already have this extent in the
618 * file. This must be done before the btrfs_drop_extents run
619 * so we don't try to drop this extent.
621 ret = btrfs_lookup_file_extent(trans, root, path,
622 btrfs_ino(BTRFS_I(inode)), start, 0);
625 (found_type == BTRFS_FILE_EXTENT_REG ||
626 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
627 struct btrfs_file_extent_item cmp1;
628 struct btrfs_file_extent_item cmp2;
629 struct btrfs_file_extent_item *existing;
630 struct extent_buffer *leaf;
632 leaf = path->nodes[0];
633 existing = btrfs_item_ptr(leaf, path->slots[0],
634 struct btrfs_file_extent_item);
636 read_extent_buffer(eb, &cmp1, (unsigned long)item,
638 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
642 * we already have a pointer to this exact extent,
643 * we don't have to do anything
645 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
646 btrfs_release_path(path);
650 btrfs_release_path(path);
652 /* drop any overlapping extents */
653 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
657 if (found_type == BTRFS_FILE_EXTENT_REG ||
658 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
660 unsigned long dest_offset;
661 struct btrfs_key ins;
663 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
664 btrfs_fs_incompat(fs_info, NO_HOLES))
667 ret = btrfs_insert_empty_item(trans, root, path, key,
671 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
673 copy_extent_buffer(path->nodes[0], eb, dest_offset,
674 (unsigned long)item, sizeof(*item));
676 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
677 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
678 ins.type = BTRFS_EXTENT_ITEM_KEY;
679 offset = key->offset - btrfs_file_extent_offset(eb, item);
682 * Manually record dirty extent, as here we did a shallow
683 * file extent item copy and skip normal backref update,
684 * but modifying extent tree all by ourselves.
685 * So need to manually record dirty extent for qgroup,
686 * as the owner of the file extent changed from log tree
687 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
689 ret = btrfs_qgroup_trace_extent(trans,
690 btrfs_file_extent_disk_bytenr(eb, item),
691 btrfs_file_extent_disk_num_bytes(eb, item),
696 if (ins.objectid > 0) {
699 LIST_HEAD(ordered_sums);
701 * is this extent already allocated in the extent
702 * allocation tree? If so, just add a reference
704 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
707 ret = btrfs_inc_extent_ref(trans, root,
708 ins.objectid, ins.offset,
709 0, root->root_key.objectid,
710 key->objectid, offset);
715 * insert the extent pointer in the extent
718 ret = btrfs_alloc_logged_file_extent(trans,
719 root->root_key.objectid,
720 key->objectid, offset, &ins);
724 btrfs_release_path(path);
726 if (btrfs_file_extent_compression(eb, item)) {
727 csum_start = ins.objectid;
728 csum_end = csum_start + ins.offset;
730 csum_start = ins.objectid +
731 btrfs_file_extent_offset(eb, item);
732 csum_end = csum_start +
733 btrfs_file_extent_num_bytes(eb, item);
736 ret = btrfs_lookup_csums_range(root->log_root,
737 csum_start, csum_end - 1,
742 * Now delete all existing cums in the csum root that
743 * cover our range. We do this because we can have an
744 * extent that is completely referenced by one file
745 * extent item and partially referenced by another
746 * file extent item (like after using the clone or
747 * extent_same ioctls). In this case if we end up doing
748 * the replay of the one that partially references the
749 * extent first, and we do not do the csum deletion
750 * below, we can get 2 csum items in the csum tree that
751 * overlap each other. For example, imagine our log has
752 * the two following file extent items:
754 * key (257 EXTENT_DATA 409600)
755 * extent data disk byte 12845056 nr 102400
756 * extent data offset 20480 nr 20480 ram 102400
758 * key (257 EXTENT_DATA 819200)
759 * extent data disk byte 12845056 nr 102400
760 * extent data offset 0 nr 102400 ram 102400
762 * Where the second one fully references the 100K extent
763 * that starts at disk byte 12845056, and the log tree
764 * has a single csum item that covers the entire range
767 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
769 * After the first file extent item is replayed, the
770 * csum tree gets the following csum item:
772 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
774 * Which covers the 20K sub-range starting at offset 20K
775 * of our extent. Now when we replay the second file
776 * extent item, if we do not delete existing csum items
777 * that cover any of its blocks, we end up getting two
778 * csum items in our csum tree that overlap each other:
780 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
781 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
783 * Which is a problem, because after this anyone trying
784 * to lookup up for the checksum of any block of our
785 * extent starting at an offset of 40K or higher, will
786 * end up looking at the second csum item only, which
787 * does not contain the checksum for any block starting
788 * at offset 40K or higher of our extent.
790 while (!list_empty(&ordered_sums)) {
791 struct btrfs_ordered_sum *sums;
792 sums = list_entry(ordered_sums.next,
793 struct btrfs_ordered_sum,
796 ret = btrfs_del_csums(trans, fs_info,
800 ret = btrfs_csum_file_blocks(trans,
801 fs_info->csum_root, sums);
802 list_del(&sums->list);
808 btrfs_release_path(path);
810 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
811 /* inline extents are easy, we just overwrite them */
812 ret = overwrite_item(trans, root, path, eb, slot, key);
817 inode_add_bytes(inode, nbytes);
819 ret = btrfs_update_inode(trans, root, inode);
827 * when cleaning up conflicts between the directory names in the
828 * subvolume, directory names in the log and directory names in the
829 * inode back references, we may have to unlink inodes from directories.
831 * This is a helper function to do the unlink of a specific directory
834 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root,
836 struct btrfs_path *path,
837 struct btrfs_inode *dir,
838 struct btrfs_dir_item *di)
843 struct extent_buffer *leaf;
844 struct btrfs_key location;
847 leaf = path->nodes[0];
849 btrfs_dir_item_key_to_cpu(leaf, di, &location);
850 name_len = btrfs_dir_name_len(leaf, di);
851 name = kmalloc(name_len, GFP_NOFS);
855 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
856 btrfs_release_path(path);
858 inode = read_one_inode(root, location.objectid);
864 ret = link_to_fixup_dir(trans, root, path, location.objectid);
868 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
873 ret = btrfs_run_delayed_items(trans);
881 * helper function to see if a given name and sequence number found
882 * in an inode back reference are already in a directory and correctly
883 * point to this inode
885 static noinline int inode_in_dir(struct btrfs_root *root,
886 struct btrfs_path *path,
887 u64 dirid, u64 objectid, u64 index,
888 const char *name, int name_len)
890 struct btrfs_dir_item *di;
891 struct btrfs_key location;
894 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
895 index, name, name_len, 0);
896 if (di && !IS_ERR(di)) {
897 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
898 if (location.objectid != objectid)
902 btrfs_release_path(path);
904 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
905 if (di && !IS_ERR(di)) {
906 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
907 if (location.objectid != objectid)
913 btrfs_release_path(path);
918 * helper function to check a log tree for a named back reference in
919 * an inode. This is used to decide if a back reference that is
920 * found in the subvolume conflicts with what we find in the log.
922 * inode backreferences may have multiple refs in a single item,
923 * during replay we process one reference at a time, and we don't
924 * want to delete valid links to a file from the subvolume if that
925 * link is also in the log.
927 static noinline int backref_in_log(struct btrfs_root *log,
928 struct btrfs_key *key,
930 const char *name, int namelen)
932 struct btrfs_path *path;
933 struct btrfs_inode_ref *ref;
935 unsigned long ptr_end;
936 unsigned long name_ptr;
942 path = btrfs_alloc_path();
946 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
950 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
952 if (key->type == BTRFS_INODE_EXTREF_KEY) {
953 if (btrfs_find_name_in_ext_backref(path->nodes[0],
956 name, namelen, NULL))
962 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
963 ptr_end = ptr + item_size;
964 while (ptr < ptr_end) {
965 ref = (struct btrfs_inode_ref *)ptr;
966 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
967 if (found_name_len == namelen) {
968 name_ptr = (unsigned long)(ref + 1);
969 ret = memcmp_extent_buffer(path->nodes[0], name,
976 ptr = (unsigned long)(ref + 1) + found_name_len;
979 btrfs_free_path(path);
983 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
984 struct btrfs_root *root,
985 struct btrfs_path *path,
986 struct btrfs_root *log_root,
987 struct btrfs_inode *dir,
988 struct btrfs_inode *inode,
989 u64 inode_objectid, u64 parent_objectid,
990 u64 ref_index, char *name, int namelen,
996 struct extent_buffer *leaf;
997 struct btrfs_dir_item *di;
998 struct btrfs_key search_key;
999 struct btrfs_inode_extref *extref;
1002 /* Search old style refs */
1003 search_key.objectid = inode_objectid;
1004 search_key.type = BTRFS_INODE_REF_KEY;
1005 search_key.offset = parent_objectid;
1006 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1008 struct btrfs_inode_ref *victim_ref;
1010 unsigned long ptr_end;
1012 leaf = path->nodes[0];
1014 /* are we trying to overwrite a back ref for the root directory
1015 * if so, just jump out, we're done
1017 if (search_key.objectid == search_key.offset)
1020 /* check all the names in this back reference to see
1021 * if they are in the log. if so, we allow them to stay
1022 * otherwise they must be unlinked as a conflict
1024 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1025 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1026 while (ptr < ptr_end) {
1027 victim_ref = (struct btrfs_inode_ref *)ptr;
1028 victim_name_len = btrfs_inode_ref_name_len(leaf,
1030 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1034 read_extent_buffer(leaf, victim_name,
1035 (unsigned long)(victim_ref + 1),
1038 if (!backref_in_log(log_root, &search_key,
1042 inc_nlink(&inode->vfs_inode);
1043 btrfs_release_path(path);
1045 ret = btrfs_unlink_inode(trans, root, dir, inode,
1046 victim_name, victim_name_len);
1050 ret = btrfs_run_delayed_items(trans);
1058 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1062 * NOTE: we have searched root tree and checked the
1063 * corresponding ref, it does not need to check again.
1067 btrfs_release_path(path);
1069 /* Same search but for extended refs */
1070 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1071 inode_objectid, parent_objectid, 0,
1073 if (!IS_ERR_OR_NULL(extref)) {
1077 struct inode *victim_parent;
1079 leaf = path->nodes[0];
1081 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1082 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1084 while (cur_offset < item_size) {
1085 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1087 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1089 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1092 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1095 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1098 search_key.objectid = inode_objectid;
1099 search_key.type = BTRFS_INODE_EXTREF_KEY;
1100 search_key.offset = btrfs_extref_hash(parent_objectid,
1104 if (!backref_in_log(log_root, &search_key,
1105 parent_objectid, victim_name,
1108 victim_parent = read_one_inode(root,
1110 if (victim_parent) {
1111 inc_nlink(&inode->vfs_inode);
1112 btrfs_release_path(path);
1114 ret = btrfs_unlink_inode(trans, root,
1115 BTRFS_I(victim_parent),
1120 ret = btrfs_run_delayed_items(
1123 iput(victim_parent);
1132 cur_offset += victim_name_len + sizeof(*extref);
1136 btrfs_release_path(path);
1138 /* look for a conflicting sequence number */
1139 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1140 ref_index, name, namelen, 0);
1141 if (di && !IS_ERR(di)) {
1142 ret = drop_one_dir_item(trans, root, path, dir, di);
1146 btrfs_release_path(path);
1148 /* look for a conflicting name */
1149 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1151 if (di && !IS_ERR(di)) {
1152 ret = drop_one_dir_item(trans, root, path, dir, di);
1156 btrfs_release_path(path);
1161 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1162 u32 *namelen, char **name, u64 *index,
1163 u64 *parent_objectid)
1165 struct btrfs_inode_extref *extref;
1167 extref = (struct btrfs_inode_extref *)ref_ptr;
1169 *namelen = btrfs_inode_extref_name_len(eb, extref);
1170 *name = kmalloc(*namelen, GFP_NOFS);
1174 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1178 *index = btrfs_inode_extref_index(eb, extref);
1179 if (parent_objectid)
1180 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1185 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1186 u32 *namelen, char **name, u64 *index)
1188 struct btrfs_inode_ref *ref;
1190 ref = (struct btrfs_inode_ref *)ref_ptr;
1192 *namelen = btrfs_inode_ref_name_len(eb, ref);
1193 *name = kmalloc(*namelen, GFP_NOFS);
1197 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1200 *index = btrfs_inode_ref_index(eb, ref);
1206 * Take an inode reference item from the log tree and iterate all names from the
1207 * inode reference item in the subvolume tree with the same key (if it exists).
1208 * For any name that is not in the inode reference item from the log tree, do a
1209 * proper unlink of that name (that is, remove its entry from the inode
1210 * reference item and both dir index keys).
1212 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1213 struct btrfs_root *root,
1214 struct btrfs_path *path,
1215 struct btrfs_inode *inode,
1216 struct extent_buffer *log_eb,
1218 struct btrfs_key *key)
1221 unsigned long ref_ptr;
1222 unsigned long ref_end;
1223 struct extent_buffer *eb;
1226 btrfs_release_path(path);
1227 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1235 eb = path->nodes[0];
1236 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1237 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1238 while (ref_ptr < ref_end) {
1243 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1244 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1247 parent_id = key->offset;
1248 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1254 if (key->type == BTRFS_INODE_EXTREF_KEY)
1255 ret = btrfs_find_name_in_ext_backref(log_eb, log_slot,
1259 ret = btrfs_find_name_in_backref(log_eb, log_slot, name,
1265 btrfs_release_path(path);
1266 dir = read_one_inode(root, parent_id);
1272 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1273 inode, name, namelen);
1283 if (key->type == BTRFS_INODE_EXTREF_KEY)
1284 ref_ptr += sizeof(struct btrfs_inode_extref);
1286 ref_ptr += sizeof(struct btrfs_inode_ref);
1290 btrfs_release_path(path);
1294 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1295 const u8 ref_type, const char *name,
1298 struct btrfs_key key;
1299 struct btrfs_path *path;
1300 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1303 path = btrfs_alloc_path();
1307 key.objectid = btrfs_ino(BTRFS_I(inode));
1308 key.type = ref_type;
1309 if (key.type == BTRFS_INODE_REF_KEY)
1310 key.offset = parent_id;
1312 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1314 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1321 if (key.type == BTRFS_INODE_EXTREF_KEY)
1322 ret = btrfs_find_name_in_ext_backref(path->nodes[0],
1323 path->slots[0], parent_id,
1324 name, namelen, NULL);
1326 ret = btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1327 name, namelen, NULL);
1330 btrfs_free_path(path);
1334 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1335 struct inode *dir, struct inode *inode, const char *name,
1336 int namelen, u64 ref_index)
1338 struct btrfs_dir_item *dir_item;
1339 struct btrfs_key key;
1340 struct btrfs_path *path;
1341 struct inode *other_inode = NULL;
1344 path = btrfs_alloc_path();
1348 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1349 btrfs_ino(BTRFS_I(dir)),
1352 btrfs_release_path(path);
1354 } else if (IS_ERR(dir_item)) {
1355 ret = PTR_ERR(dir_item);
1360 * Our inode's dentry collides with the dentry of another inode which is
1361 * in the log but not yet processed since it has a higher inode number.
1362 * So delete that other dentry.
1364 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1365 btrfs_release_path(path);
1366 other_inode = read_one_inode(root, key.objectid);
1371 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1376 * If we dropped the link count to 0, bump it so that later the iput()
1377 * on the inode will not free it. We will fixup the link count later.
1379 if (other_inode->i_nlink == 0)
1380 inc_nlink(other_inode);
1382 ret = btrfs_run_delayed_items(trans);
1386 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1387 name, namelen, 0, ref_index);
1390 btrfs_free_path(path);
1396 * replay one inode back reference item found in the log tree.
1397 * eb, slot and key refer to the buffer and key found in the log tree.
1398 * root is the destination we are replaying into, and path is for temp
1399 * use by this function. (it should be released on return).
1401 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_root *log,
1404 struct btrfs_path *path,
1405 struct extent_buffer *eb, int slot,
1406 struct btrfs_key *key)
1408 struct inode *dir = NULL;
1409 struct inode *inode = NULL;
1410 unsigned long ref_ptr;
1411 unsigned long ref_end;
1415 int search_done = 0;
1416 int log_ref_ver = 0;
1417 u64 parent_objectid;
1420 int ref_struct_size;
1422 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1423 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1425 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1426 struct btrfs_inode_extref *r;
1428 ref_struct_size = sizeof(struct btrfs_inode_extref);
1430 r = (struct btrfs_inode_extref *)ref_ptr;
1431 parent_objectid = btrfs_inode_extref_parent(eb, r);
1433 ref_struct_size = sizeof(struct btrfs_inode_ref);
1434 parent_objectid = key->offset;
1436 inode_objectid = key->objectid;
1439 * it is possible that we didn't log all the parent directories
1440 * for a given inode. If we don't find the dir, just don't
1441 * copy the back ref in. The link count fixup code will take
1444 dir = read_one_inode(root, parent_objectid);
1450 inode = read_one_inode(root, inode_objectid);
1456 while (ref_ptr < ref_end) {
1458 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1459 &ref_index, &parent_objectid);
1461 * parent object can change from one array
1465 dir = read_one_inode(root, parent_objectid);
1471 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1477 /* if we already have a perfect match, we're done */
1478 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1479 btrfs_ino(BTRFS_I(inode)), ref_index,
1482 * look for a conflicting back reference in the
1483 * metadata. if we find one we have to unlink that name
1484 * of the file before we add our new link. Later on, we
1485 * overwrite any existing back reference, and we don't
1486 * want to create dangling pointers in the directory.
1490 ret = __add_inode_ref(trans, root, path, log,
1495 ref_index, name, namelen,
1505 * If a reference item already exists for this inode
1506 * with the same parent and name, but different index,
1507 * drop it and the corresponding directory index entries
1508 * from the parent before adding the new reference item
1509 * and dir index entries, otherwise we would fail with
1510 * -EEXIST returned from btrfs_add_link() below.
1512 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1515 ret = btrfs_unlink_inode(trans, root,
1520 * If we dropped the link count to 0, bump it so
1521 * that later the iput() on the inode will not
1522 * free it. We will fixup the link count later.
1524 if (!ret && inode->i_nlink == 0)
1530 /* insert our name */
1531 ret = add_link(trans, root, dir, inode, name, namelen,
1536 btrfs_update_inode(trans, root, inode);
1539 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1549 * Before we overwrite the inode reference item in the subvolume tree
1550 * with the item from the log tree, we must unlink all names from the
1551 * parent directory that are in the subvolume's tree inode reference
1552 * item, otherwise we end up with an inconsistent subvolume tree where
1553 * dir index entries exist for a name but there is no inode reference
1554 * item with the same name.
1556 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1561 /* finally write the back reference in the inode */
1562 ret = overwrite_item(trans, root, path, eb, slot, key);
1564 btrfs_release_path(path);
1571 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1572 struct btrfs_root *root, u64 ino)
1576 ret = btrfs_insert_orphan_item(trans, root, ino);
1583 static int count_inode_extrefs(struct btrfs_root *root,
1584 struct btrfs_inode *inode, struct btrfs_path *path)
1588 unsigned int nlink = 0;
1591 u64 inode_objectid = btrfs_ino(inode);
1594 struct btrfs_inode_extref *extref;
1595 struct extent_buffer *leaf;
1598 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1603 leaf = path->nodes[0];
1604 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1605 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1608 while (cur_offset < item_size) {
1609 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1610 name_len = btrfs_inode_extref_name_len(leaf, extref);
1614 cur_offset += name_len + sizeof(*extref);
1618 btrfs_release_path(path);
1620 btrfs_release_path(path);
1622 if (ret < 0 && ret != -ENOENT)
1627 static int count_inode_refs(struct btrfs_root *root,
1628 struct btrfs_inode *inode, struct btrfs_path *path)
1631 struct btrfs_key key;
1632 unsigned int nlink = 0;
1634 unsigned long ptr_end;
1636 u64 ino = btrfs_ino(inode);
1639 key.type = BTRFS_INODE_REF_KEY;
1640 key.offset = (u64)-1;
1643 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1647 if (path->slots[0] == 0)
1652 btrfs_item_key_to_cpu(path->nodes[0], &key,
1654 if (key.objectid != ino ||
1655 key.type != BTRFS_INODE_REF_KEY)
1657 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1658 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1660 while (ptr < ptr_end) {
1661 struct btrfs_inode_ref *ref;
1663 ref = (struct btrfs_inode_ref *)ptr;
1664 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1666 ptr = (unsigned long)(ref + 1) + name_len;
1670 if (key.offset == 0)
1672 if (path->slots[0] > 0) {
1677 btrfs_release_path(path);
1679 btrfs_release_path(path);
1685 * There are a few corners where the link count of the file can't
1686 * be properly maintained during replay. So, instead of adding
1687 * lots of complexity to the log code, we just scan the backrefs
1688 * for any file that has been through replay.
1690 * The scan will update the link count on the inode to reflect the
1691 * number of back refs found. If it goes down to zero, the iput
1692 * will free the inode.
1694 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct inode *inode)
1698 struct btrfs_path *path;
1701 u64 ino = btrfs_ino(BTRFS_I(inode));
1703 path = btrfs_alloc_path();
1707 ret = count_inode_refs(root, BTRFS_I(inode), path);
1713 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1721 if (nlink != inode->i_nlink) {
1722 set_nlink(inode, nlink);
1723 btrfs_update_inode(trans, root, inode);
1725 BTRFS_I(inode)->index_cnt = (u64)-1;
1727 if (inode->i_nlink == 0) {
1728 if (S_ISDIR(inode->i_mode)) {
1729 ret = replay_dir_deletes(trans, root, NULL, path,
1734 ret = insert_orphan_item(trans, root, ino);
1738 btrfs_free_path(path);
1742 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1743 struct btrfs_root *root,
1744 struct btrfs_path *path)
1747 struct btrfs_key key;
1748 struct inode *inode;
1750 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1751 key.type = BTRFS_ORPHAN_ITEM_KEY;
1752 key.offset = (u64)-1;
1754 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1759 if (path->slots[0] == 0)
1764 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1765 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1766 key.type != BTRFS_ORPHAN_ITEM_KEY)
1769 ret = btrfs_del_item(trans, root, path);
1773 btrfs_release_path(path);
1774 inode = read_one_inode(root, key.offset);
1778 ret = fixup_inode_link_count(trans, root, inode);
1784 * fixup on a directory may create new entries,
1785 * make sure we always look for the highset possible
1788 key.offset = (u64)-1;
1792 btrfs_release_path(path);
1798 * record a given inode in the fixup dir so we can check its link
1799 * count when replay is done. The link count is incremented here
1800 * so the inode won't go away until we check it
1802 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1807 struct btrfs_key key;
1809 struct inode *inode;
1811 inode = read_one_inode(root, objectid);
1815 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1816 key.type = BTRFS_ORPHAN_ITEM_KEY;
1817 key.offset = objectid;
1819 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1821 btrfs_release_path(path);
1823 if (!inode->i_nlink)
1824 set_nlink(inode, 1);
1827 ret = btrfs_update_inode(trans, root, inode);
1828 } else if (ret == -EEXIST) {
1831 BUG(); /* Logic Error */
1839 * when replaying the log for a directory, we only insert names
1840 * for inodes that actually exist. This means an fsync on a directory
1841 * does not implicitly fsync all the new files in it
1843 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1844 struct btrfs_root *root,
1845 u64 dirid, u64 index,
1846 char *name, int name_len,
1847 struct btrfs_key *location)
1849 struct inode *inode;
1853 inode = read_one_inode(root, location->objectid);
1857 dir = read_one_inode(root, dirid);
1863 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1864 name_len, 1, index);
1866 /* FIXME, put inode into FIXUP list */
1874 * Return true if an inode reference exists in the log for the given name,
1875 * inode and parent inode.
1877 static bool name_in_log_ref(struct btrfs_root *log_root,
1878 const char *name, const int name_len,
1879 const u64 dirid, const u64 ino)
1881 struct btrfs_key search_key;
1883 search_key.objectid = ino;
1884 search_key.type = BTRFS_INODE_REF_KEY;
1885 search_key.offset = dirid;
1886 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1889 search_key.type = BTRFS_INODE_EXTREF_KEY;
1890 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1891 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1898 * take a single entry in a log directory item and replay it into
1901 * if a conflicting item exists in the subdirectory already,
1902 * the inode it points to is unlinked and put into the link count
1905 * If a name from the log points to a file or directory that does
1906 * not exist in the FS, it is skipped. fsyncs on directories
1907 * do not force down inodes inside that directory, just changes to the
1908 * names or unlinks in a directory.
1910 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1911 * non-existing inode) and 1 if the name was replayed.
1913 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1914 struct btrfs_root *root,
1915 struct btrfs_path *path,
1916 struct extent_buffer *eb,
1917 struct btrfs_dir_item *di,
1918 struct btrfs_key *key)
1922 struct btrfs_dir_item *dst_di;
1923 struct btrfs_key found_key;
1924 struct btrfs_key log_key;
1929 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1930 bool name_added = false;
1932 dir = read_one_inode(root, key->objectid);
1936 name_len = btrfs_dir_name_len(eb, di);
1937 name = kmalloc(name_len, GFP_NOFS);
1943 log_type = btrfs_dir_type(eb, di);
1944 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1947 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1948 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1953 btrfs_release_path(path);
1955 if (key->type == BTRFS_DIR_ITEM_KEY) {
1956 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1958 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1959 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1968 if (IS_ERR_OR_NULL(dst_di)) {
1969 /* we need a sequence number to insert, so we only
1970 * do inserts for the BTRFS_DIR_INDEX_KEY types
1972 if (key->type != BTRFS_DIR_INDEX_KEY)
1977 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1978 /* the existing item matches the logged item */
1979 if (found_key.objectid == log_key.objectid &&
1980 found_key.type == log_key.type &&
1981 found_key.offset == log_key.offset &&
1982 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1983 update_size = false;
1988 * don't drop the conflicting directory entry if the inode
1989 * for the new entry doesn't exist
1994 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1998 if (key->type == BTRFS_DIR_INDEX_KEY)
2001 btrfs_release_path(path);
2002 if (!ret && update_size) {
2003 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2004 ret = btrfs_update_inode(trans, root, dir);
2008 if (!ret && name_added)
2013 if (name_in_log_ref(root->log_root, name, name_len,
2014 key->objectid, log_key.objectid)) {
2015 /* The dentry will be added later. */
2017 update_size = false;
2020 btrfs_release_path(path);
2021 ret = insert_one_name(trans, root, key->objectid, key->offset,
2022 name, name_len, &log_key);
2023 if (ret && ret != -ENOENT && ret != -EEXIST)
2027 update_size = false;
2033 * find all the names in a directory item and reconcile them into
2034 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2035 * one name in a directory item, but the same code gets used for
2036 * both directory index types
2038 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2039 struct btrfs_root *root,
2040 struct btrfs_path *path,
2041 struct extent_buffer *eb, int slot,
2042 struct btrfs_key *key)
2045 u32 item_size = btrfs_item_size_nr(eb, slot);
2046 struct btrfs_dir_item *di;
2049 unsigned long ptr_end;
2050 struct btrfs_path *fixup_path = NULL;
2052 ptr = btrfs_item_ptr_offset(eb, slot);
2053 ptr_end = ptr + item_size;
2054 while (ptr < ptr_end) {
2055 di = (struct btrfs_dir_item *)ptr;
2056 name_len = btrfs_dir_name_len(eb, di);
2057 ret = replay_one_name(trans, root, path, eb, di, key);
2060 ptr = (unsigned long)(di + 1);
2064 * If this entry refers to a non-directory (directories can not
2065 * have a link count > 1) and it was added in the transaction
2066 * that was not committed, make sure we fixup the link count of
2067 * the inode it the entry points to. Otherwise something like
2068 * the following would result in a directory pointing to an
2069 * inode with a wrong link that does not account for this dir
2077 * ln testdir/bar testdir/bar_link
2078 * ln testdir/foo testdir/foo_link
2079 * xfs_io -c "fsync" testdir/bar
2083 * mount fs, log replay happens
2085 * File foo would remain with a link count of 1 when it has two
2086 * entries pointing to it in the directory testdir. This would
2087 * make it impossible to ever delete the parent directory has
2088 * it would result in stale dentries that can never be deleted.
2090 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2091 struct btrfs_key di_key;
2094 fixup_path = btrfs_alloc_path();
2101 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2102 ret = link_to_fixup_dir(trans, root, fixup_path,
2109 btrfs_free_path(fixup_path);
2114 * directory replay has two parts. There are the standard directory
2115 * items in the log copied from the subvolume, and range items
2116 * created in the log while the subvolume was logged.
2118 * The range items tell us which parts of the key space the log
2119 * is authoritative for. During replay, if a key in the subvolume
2120 * directory is in a logged range item, but not actually in the log
2121 * that means it was deleted from the directory before the fsync
2122 * and should be removed.
2124 static noinline int find_dir_range(struct btrfs_root *root,
2125 struct btrfs_path *path,
2126 u64 dirid, int key_type,
2127 u64 *start_ret, u64 *end_ret)
2129 struct btrfs_key key;
2131 struct btrfs_dir_log_item *item;
2135 if (*start_ret == (u64)-1)
2138 key.objectid = dirid;
2139 key.type = key_type;
2140 key.offset = *start_ret;
2142 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2146 if (path->slots[0] == 0)
2151 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2153 if (key.type != key_type || key.objectid != dirid) {
2157 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2158 struct btrfs_dir_log_item);
2159 found_end = btrfs_dir_log_end(path->nodes[0], item);
2161 if (*start_ret >= key.offset && *start_ret <= found_end) {
2163 *start_ret = key.offset;
2164 *end_ret = found_end;
2169 /* check the next slot in the tree to see if it is a valid item */
2170 nritems = btrfs_header_nritems(path->nodes[0]);
2172 if (path->slots[0] >= nritems) {
2173 ret = btrfs_next_leaf(root, path);
2178 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2180 if (key.type != key_type || key.objectid != dirid) {
2184 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2185 struct btrfs_dir_log_item);
2186 found_end = btrfs_dir_log_end(path->nodes[0], item);
2187 *start_ret = key.offset;
2188 *end_ret = found_end;
2191 btrfs_release_path(path);
2196 * this looks for a given directory item in the log. If the directory
2197 * item is not in the log, the item is removed and the inode it points
2200 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2201 struct btrfs_root *root,
2202 struct btrfs_root *log,
2203 struct btrfs_path *path,
2204 struct btrfs_path *log_path,
2206 struct btrfs_key *dir_key)
2209 struct extent_buffer *eb;
2212 struct btrfs_dir_item *di;
2213 struct btrfs_dir_item *log_di;
2216 unsigned long ptr_end;
2218 struct inode *inode;
2219 struct btrfs_key location;
2222 eb = path->nodes[0];
2223 slot = path->slots[0];
2224 item_size = btrfs_item_size_nr(eb, slot);
2225 ptr = btrfs_item_ptr_offset(eb, slot);
2226 ptr_end = ptr + item_size;
2227 while (ptr < ptr_end) {
2228 di = (struct btrfs_dir_item *)ptr;
2229 name_len = btrfs_dir_name_len(eb, di);
2230 name = kmalloc(name_len, GFP_NOFS);
2235 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2238 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2239 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2242 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2243 log_di = btrfs_lookup_dir_index_item(trans, log,
2249 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2250 btrfs_dir_item_key_to_cpu(eb, di, &location);
2251 btrfs_release_path(path);
2252 btrfs_release_path(log_path);
2253 inode = read_one_inode(root, location.objectid);
2259 ret = link_to_fixup_dir(trans, root,
2260 path, location.objectid);
2268 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2269 BTRFS_I(inode), name, name_len);
2271 ret = btrfs_run_delayed_items(trans);
2277 /* there might still be more names under this key
2278 * check and repeat if required
2280 ret = btrfs_search_slot(NULL, root, dir_key, path,
2286 } else if (IS_ERR(log_di)) {
2288 return PTR_ERR(log_di);
2290 btrfs_release_path(log_path);
2293 ptr = (unsigned long)(di + 1);
2298 btrfs_release_path(path);
2299 btrfs_release_path(log_path);
2303 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2304 struct btrfs_root *root,
2305 struct btrfs_root *log,
2306 struct btrfs_path *path,
2309 struct btrfs_key search_key;
2310 struct btrfs_path *log_path;
2315 log_path = btrfs_alloc_path();
2319 search_key.objectid = ino;
2320 search_key.type = BTRFS_XATTR_ITEM_KEY;
2321 search_key.offset = 0;
2323 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2327 nritems = btrfs_header_nritems(path->nodes[0]);
2328 for (i = path->slots[0]; i < nritems; i++) {
2329 struct btrfs_key key;
2330 struct btrfs_dir_item *di;
2331 struct btrfs_dir_item *log_di;
2335 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2336 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2341 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2342 total_size = btrfs_item_size_nr(path->nodes[0], i);
2344 while (cur < total_size) {
2345 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2346 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2347 u32 this_len = sizeof(*di) + name_len + data_len;
2350 name = kmalloc(name_len, GFP_NOFS);
2355 read_extent_buffer(path->nodes[0], name,
2356 (unsigned long)(di + 1), name_len);
2358 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2360 btrfs_release_path(log_path);
2362 /* Doesn't exist in log tree, so delete it. */
2363 btrfs_release_path(path);
2364 di = btrfs_lookup_xattr(trans, root, path, ino,
2365 name, name_len, -1);
2372 ret = btrfs_delete_one_dir_name(trans, root,
2376 btrfs_release_path(path);
2381 if (IS_ERR(log_di)) {
2382 ret = PTR_ERR(log_di);
2386 di = (struct btrfs_dir_item *)((char *)di + this_len);
2389 ret = btrfs_next_leaf(root, path);
2395 btrfs_free_path(log_path);
2396 btrfs_release_path(path);
2402 * deletion replay happens before we copy any new directory items
2403 * out of the log or out of backreferences from inodes. It
2404 * scans the log to find ranges of keys that log is authoritative for,
2405 * and then scans the directory to find items in those ranges that are
2406 * not present in the log.
2408 * Anything we don't find in the log is unlinked and removed from the
2411 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2412 struct btrfs_root *root,
2413 struct btrfs_root *log,
2414 struct btrfs_path *path,
2415 u64 dirid, int del_all)
2419 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2421 struct btrfs_key dir_key;
2422 struct btrfs_key found_key;
2423 struct btrfs_path *log_path;
2426 dir_key.objectid = dirid;
2427 dir_key.type = BTRFS_DIR_ITEM_KEY;
2428 log_path = btrfs_alloc_path();
2432 dir = read_one_inode(root, dirid);
2433 /* it isn't an error if the inode isn't there, that can happen
2434 * because we replay the deletes before we copy in the inode item
2438 btrfs_free_path(log_path);
2446 range_end = (u64)-1;
2448 ret = find_dir_range(log, path, dirid, key_type,
2449 &range_start, &range_end);
2454 dir_key.offset = range_start;
2457 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2462 nritems = btrfs_header_nritems(path->nodes[0]);
2463 if (path->slots[0] >= nritems) {
2464 ret = btrfs_next_leaf(root, path);
2470 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2472 if (found_key.objectid != dirid ||
2473 found_key.type != dir_key.type)
2476 if (found_key.offset > range_end)
2479 ret = check_item_in_log(trans, root, log, path,
2484 if (found_key.offset == (u64)-1)
2486 dir_key.offset = found_key.offset + 1;
2488 btrfs_release_path(path);
2489 if (range_end == (u64)-1)
2491 range_start = range_end + 1;
2496 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2497 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2498 dir_key.type = BTRFS_DIR_INDEX_KEY;
2499 btrfs_release_path(path);
2503 btrfs_release_path(path);
2504 btrfs_free_path(log_path);
2510 * the process_func used to replay items from the log tree. This
2511 * gets called in two different stages. The first stage just looks
2512 * for inodes and makes sure they are all copied into the subvolume.
2514 * The second stage copies all the other item types from the log into
2515 * the subvolume. The two stage approach is slower, but gets rid of
2516 * lots of complexity around inodes referencing other inodes that exist
2517 * only in the log (references come from either directory items or inode
2520 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2521 struct walk_control *wc, u64 gen, int level)
2524 struct btrfs_path *path;
2525 struct btrfs_root *root = wc->replay_dest;
2526 struct btrfs_key key;
2530 ret = btrfs_read_buffer(eb, gen, level, NULL);
2534 level = btrfs_header_level(eb);
2539 path = btrfs_alloc_path();
2543 nritems = btrfs_header_nritems(eb);
2544 for (i = 0; i < nritems; i++) {
2545 btrfs_item_key_to_cpu(eb, &key, i);
2547 /* inode keys are done during the first stage */
2548 if (key.type == BTRFS_INODE_ITEM_KEY &&
2549 wc->stage == LOG_WALK_REPLAY_INODES) {
2550 struct btrfs_inode_item *inode_item;
2553 inode_item = btrfs_item_ptr(eb, i,
2554 struct btrfs_inode_item);
2556 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2557 * and never got linked before the fsync, skip it, as
2558 * replaying it is pointless since it would be deleted
2559 * later. We skip logging tmpfiles, but it's always
2560 * possible we are replaying a log created with a kernel
2561 * that used to log tmpfiles.
2563 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2564 wc->ignore_cur_inode = true;
2567 wc->ignore_cur_inode = false;
2569 ret = replay_xattr_deletes(wc->trans, root, log,
2570 path, key.objectid);
2573 mode = btrfs_inode_mode(eb, inode_item);
2574 if (S_ISDIR(mode)) {
2575 ret = replay_dir_deletes(wc->trans,
2576 root, log, path, key.objectid, 0);
2580 ret = overwrite_item(wc->trans, root, path,
2586 * Before replaying extents, truncate the inode to its
2587 * size. We need to do it now and not after log replay
2588 * because before an fsync we can have prealloc extents
2589 * added beyond the inode's i_size. If we did it after,
2590 * through orphan cleanup for example, we would drop
2591 * those prealloc extents just after replaying them.
2593 if (S_ISREG(mode)) {
2594 struct inode *inode;
2597 inode = read_one_inode(root, key.objectid);
2602 from = ALIGN(i_size_read(inode),
2603 root->fs_info->sectorsize);
2604 ret = btrfs_drop_extents(wc->trans, root, inode,
2607 /* Update the inode's nbytes. */
2608 ret = btrfs_update_inode(wc->trans,
2616 ret = link_to_fixup_dir(wc->trans, root,
2617 path, key.objectid);
2622 if (wc->ignore_cur_inode)
2625 if (key.type == BTRFS_DIR_INDEX_KEY &&
2626 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2627 ret = replay_one_dir_item(wc->trans, root, path,
2633 if (wc->stage < LOG_WALK_REPLAY_ALL)
2636 /* these keys are simply copied */
2637 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2638 ret = overwrite_item(wc->trans, root, path,
2642 } else if (key.type == BTRFS_INODE_REF_KEY ||
2643 key.type == BTRFS_INODE_EXTREF_KEY) {
2644 ret = add_inode_ref(wc->trans, root, log, path,
2646 if (ret && ret != -ENOENT)
2649 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2650 ret = replay_one_extent(wc->trans, root, path,
2654 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2655 ret = replay_one_dir_item(wc->trans, root, path,
2661 btrfs_free_path(path);
2665 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *root,
2667 struct btrfs_path *path, int *level,
2668 struct walk_control *wc)
2670 struct btrfs_fs_info *fs_info = root->fs_info;
2674 struct extent_buffer *next;
2675 struct extent_buffer *cur;
2676 struct extent_buffer *parent;
2680 WARN_ON(*level < 0);
2681 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2683 while (*level > 0) {
2684 struct btrfs_key first_key;
2686 WARN_ON(*level < 0);
2687 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2688 cur = path->nodes[*level];
2690 WARN_ON(btrfs_header_level(cur) != *level);
2692 if (path->slots[*level] >=
2693 btrfs_header_nritems(cur))
2696 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2697 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2698 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2699 blocksize = fs_info->nodesize;
2701 parent = path->nodes[*level];
2702 root_owner = btrfs_header_owner(parent);
2704 next = btrfs_find_create_tree_block(fs_info, bytenr);
2706 return PTR_ERR(next);
2709 ret = wc->process_func(root, next, wc, ptr_gen,
2712 free_extent_buffer(next);
2716 path->slots[*level]++;
2718 ret = btrfs_read_buffer(next, ptr_gen,
2719 *level - 1, &first_key);
2721 free_extent_buffer(next);
2726 btrfs_tree_lock(next);
2727 btrfs_set_lock_blocking_write(next);
2728 clean_tree_block(fs_info, next);
2729 btrfs_wait_tree_block_writeback(next);
2730 btrfs_tree_unlock(next);
2732 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2733 clear_extent_buffer_dirty(next);
2736 WARN_ON(root_owner !=
2737 BTRFS_TREE_LOG_OBJECTID);
2738 ret = btrfs_free_and_pin_reserved_extent(
2742 free_extent_buffer(next);
2746 free_extent_buffer(next);
2749 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2751 free_extent_buffer(next);
2755 WARN_ON(*level <= 0);
2756 if (path->nodes[*level-1])
2757 free_extent_buffer(path->nodes[*level-1]);
2758 path->nodes[*level-1] = next;
2759 *level = btrfs_header_level(next);
2760 path->slots[*level] = 0;
2763 WARN_ON(*level < 0);
2764 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2766 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2772 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2773 struct btrfs_root *root,
2774 struct btrfs_path *path, int *level,
2775 struct walk_control *wc)
2777 struct btrfs_fs_info *fs_info = root->fs_info;
2783 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2784 slot = path->slots[i];
2785 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2788 WARN_ON(*level == 0);
2791 struct extent_buffer *parent;
2792 if (path->nodes[*level] == root->node)
2793 parent = path->nodes[*level];
2795 parent = path->nodes[*level + 1];
2797 root_owner = btrfs_header_owner(parent);
2798 ret = wc->process_func(root, path->nodes[*level], wc,
2799 btrfs_header_generation(path->nodes[*level]),
2805 struct extent_buffer *next;
2807 next = path->nodes[*level];
2810 btrfs_tree_lock(next);
2811 btrfs_set_lock_blocking_write(next);
2812 clean_tree_block(fs_info, next);
2813 btrfs_wait_tree_block_writeback(next);
2814 btrfs_tree_unlock(next);
2816 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2817 clear_extent_buffer_dirty(next);
2820 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2821 ret = btrfs_free_and_pin_reserved_extent(
2823 path->nodes[*level]->start,
2824 path->nodes[*level]->len);
2828 free_extent_buffer(path->nodes[*level]);
2829 path->nodes[*level] = NULL;
2837 * drop the reference count on the tree rooted at 'snap'. This traverses
2838 * the tree freeing any blocks that have a ref count of zero after being
2841 static int walk_log_tree(struct btrfs_trans_handle *trans,
2842 struct btrfs_root *log, struct walk_control *wc)
2844 struct btrfs_fs_info *fs_info = log->fs_info;
2848 struct btrfs_path *path;
2851 path = btrfs_alloc_path();
2855 level = btrfs_header_level(log->node);
2857 path->nodes[level] = log->node;
2858 extent_buffer_get(log->node);
2859 path->slots[level] = 0;
2862 wret = walk_down_log_tree(trans, log, path, &level, wc);
2870 wret = walk_up_log_tree(trans, log, path, &level, wc);
2879 /* was the root node processed? if not, catch it here */
2880 if (path->nodes[orig_level]) {
2881 ret = wc->process_func(log, path->nodes[orig_level], wc,
2882 btrfs_header_generation(path->nodes[orig_level]),
2887 struct extent_buffer *next;
2889 next = path->nodes[orig_level];
2892 btrfs_tree_lock(next);
2893 btrfs_set_lock_blocking_write(next);
2894 clean_tree_block(fs_info, next);
2895 btrfs_wait_tree_block_writeback(next);
2896 btrfs_tree_unlock(next);
2898 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2899 clear_extent_buffer_dirty(next);
2902 WARN_ON(log->root_key.objectid !=
2903 BTRFS_TREE_LOG_OBJECTID);
2904 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2905 next->start, next->len);
2912 btrfs_free_path(path);
2917 * helper function to update the item for a given subvolumes log root
2918 * in the tree of log roots
2920 static int update_log_root(struct btrfs_trans_handle *trans,
2921 struct btrfs_root *log)
2923 struct btrfs_fs_info *fs_info = log->fs_info;
2926 if (log->log_transid == 1) {
2927 /* insert root item on the first sync */
2928 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2929 &log->root_key, &log->root_item);
2931 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2932 &log->root_key, &log->root_item);
2937 static void wait_log_commit(struct btrfs_root *root, int transid)
2940 int index = transid % 2;
2943 * we only allow two pending log transactions at a time,
2944 * so we know that if ours is more than 2 older than the
2945 * current transaction, we're done
2948 prepare_to_wait(&root->log_commit_wait[index],
2949 &wait, TASK_UNINTERRUPTIBLE);
2951 if (!(root->log_transid_committed < transid &&
2952 atomic_read(&root->log_commit[index])))
2955 mutex_unlock(&root->log_mutex);
2957 mutex_lock(&root->log_mutex);
2959 finish_wait(&root->log_commit_wait[index], &wait);
2962 static void wait_for_writer(struct btrfs_root *root)
2967 prepare_to_wait(&root->log_writer_wait, &wait,
2968 TASK_UNINTERRUPTIBLE);
2969 if (!atomic_read(&root->log_writers))
2972 mutex_unlock(&root->log_mutex);
2974 mutex_lock(&root->log_mutex);
2976 finish_wait(&root->log_writer_wait, &wait);
2979 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2980 struct btrfs_log_ctx *ctx)
2985 mutex_lock(&root->log_mutex);
2986 list_del_init(&ctx->list);
2987 mutex_unlock(&root->log_mutex);
2991 * Invoked in log mutex context, or be sure there is no other task which
2992 * can access the list.
2994 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2995 int index, int error)
2997 struct btrfs_log_ctx *ctx;
2998 struct btrfs_log_ctx *safe;
3000 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3001 list_del_init(&ctx->list);
3002 ctx->log_ret = error;
3005 INIT_LIST_HEAD(&root->log_ctxs[index]);
3009 * btrfs_sync_log does sends a given tree log down to the disk and
3010 * updates the super blocks to record it. When this call is done,
3011 * you know that any inodes previously logged are safely on disk only
3014 * Any other return value means you need to call btrfs_commit_transaction.
3015 * Some of the edge cases for fsyncing directories that have had unlinks
3016 * or renames done in the past mean that sometimes the only safe
3017 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3018 * that has happened.
3020 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3021 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3027 struct btrfs_fs_info *fs_info = root->fs_info;
3028 struct btrfs_root *log = root->log_root;
3029 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3030 int log_transid = 0;
3031 struct btrfs_log_ctx root_log_ctx;
3032 struct blk_plug plug;
3034 mutex_lock(&root->log_mutex);
3035 log_transid = ctx->log_transid;
3036 if (root->log_transid_committed >= log_transid) {
3037 mutex_unlock(&root->log_mutex);
3038 return ctx->log_ret;
3041 index1 = log_transid % 2;
3042 if (atomic_read(&root->log_commit[index1])) {
3043 wait_log_commit(root, log_transid);
3044 mutex_unlock(&root->log_mutex);
3045 return ctx->log_ret;
3047 ASSERT(log_transid == root->log_transid);
3048 atomic_set(&root->log_commit[index1], 1);
3050 /* wait for previous tree log sync to complete */
3051 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3052 wait_log_commit(root, log_transid - 1);
3055 int batch = atomic_read(&root->log_batch);
3056 /* when we're on an ssd, just kick the log commit out */
3057 if (!btrfs_test_opt(fs_info, SSD) &&
3058 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3059 mutex_unlock(&root->log_mutex);
3060 schedule_timeout_uninterruptible(1);
3061 mutex_lock(&root->log_mutex);
3063 wait_for_writer(root);
3064 if (batch == atomic_read(&root->log_batch))
3068 /* bail out if we need to do a full commit */
3069 if (btrfs_need_log_full_commit(fs_info, trans)) {
3071 mutex_unlock(&root->log_mutex);
3075 if (log_transid % 2 == 0)
3076 mark = EXTENT_DIRTY;
3080 /* we start IO on all the marked extents here, but we don't actually
3081 * wait for them until later.
3083 blk_start_plug(&plug);
3084 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3086 blk_finish_plug(&plug);
3087 btrfs_abort_transaction(trans, ret);
3088 btrfs_set_log_full_commit(fs_info, trans);
3089 mutex_unlock(&root->log_mutex);
3093 btrfs_set_root_node(&log->root_item, log->node);
3095 root->log_transid++;
3096 log->log_transid = root->log_transid;
3097 root->log_start_pid = 0;
3099 * IO has been started, blocks of the log tree have WRITTEN flag set
3100 * in their headers. new modifications of the log will be written to
3101 * new positions. so it's safe to allow log writers to go in.
3103 mutex_unlock(&root->log_mutex);
3105 btrfs_init_log_ctx(&root_log_ctx, NULL);
3107 mutex_lock(&log_root_tree->log_mutex);
3108 atomic_inc(&log_root_tree->log_batch);
3109 atomic_inc(&log_root_tree->log_writers);
3111 index2 = log_root_tree->log_transid % 2;
3112 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3113 root_log_ctx.log_transid = log_root_tree->log_transid;
3115 mutex_unlock(&log_root_tree->log_mutex);
3117 ret = update_log_root(trans, log);
3119 mutex_lock(&log_root_tree->log_mutex);
3120 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3121 /* atomic_dec_and_test implies a barrier */
3122 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3126 if (!list_empty(&root_log_ctx.list))
3127 list_del_init(&root_log_ctx.list);
3129 blk_finish_plug(&plug);
3130 btrfs_set_log_full_commit(fs_info, trans);
3132 if (ret != -ENOSPC) {
3133 btrfs_abort_transaction(trans, ret);
3134 mutex_unlock(&log_root_tree->log_mutex);
3137 btrfs_wait_tree_log_extents(log, mark);
3138 mutex_unlock(&log_root_tree->log_mutex);
3143 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3144 blk_finish_plug(&plug);
3145 list_del_init(&root_log_ctx.list);
3146 mutex_unlock(&log_root_tree->log_mutex);
3147 ret = root_log_ctx.log_ret;
3151 index2 = root_log_ctx.log_transid % 2;
3152 if (atomic_read(&log_root_tree->log_commit[index2])) {
3153 blk_finish_plug(&plug);
3154 ret = btrfs_wait_tree_log_extents(log, mark);
3155 wait_log_commit(log_root_tree,
3156 root_log_ctx.log_transid);
3157 mutex_unlock(&log_root_tree->log_mutex);
3159 ret = root_log_ctx.log_ret;
3162 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3163 atomic_set(&log_root_tree->log_commit[index2], 1);
3165 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3166 wait_log_commit(log_root_tree,
3167 root_log_ctx.log_transid - 1);
3170 wait_for_writer(log_root_tree);
3173 * now that we've moved on to the tree of log tree roots,
3174 * check the full commit flag again
3176 if (btrfs_need_log_full_commit(fs_info, trans)) {
3177 blk_finish_plug(&plug);
3178 btrfs_wait_tree_log_extents(log, mark);
3179 mutex_unlock(&log_root_tree->log_mutex);
3181 goto out_wake_log_root;
3184 ret = btrfs_write_marked_extents(fs_info,
3185 &log_root_tree->dirty_log_pages,
3186 EXTENT_DIRTY | EXTENT_NEW);
3187 blk_finish_plug(&plug);
3189 btrfs_set_log_full_commit(fs_info, trans);
3190 btrfs_abort_transaction(trans, ret);
3191 mutex_unlock(&log_root_tree->log_mutex);
3192 goto out_wake_log_root;
3194 ret = btrfs_wait_tree_log_extents(log, mark);
3196 ret = btrfs_wait_tree_log_extents(log_root_tree,
3197 EXTENT_NEW | EXTENT_DIRTY);
3199 btrfs_set_log_full_commit(fs_info, trans);
3200 mutex_unlock(&log_root_tree->log_mutex);
3201 goto out_wake_log_root;
3204 btrfs_set_super_log_root(fs_info->super_for_commit,
3205 log_root_tree->node->start);
3206 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3207 btrfs_header_level(log_root_tree->node));
3209 log_root_tree->log_transid++;
3210 mutex_unlock(&log_root_tree->log_mutex);
3213 * Nobody else is going to jump in and write the ctree
3214 * super here because the log_commit atomic below is protecting
3215 * us. We must be called with a transaction handle pinning
3216 * the running transaction open, so a full commit can't hop
3217 * in and cause problems either.
3219 ret = write_all_supers(fs_info, 1);
3221 btrfs_set_log_full_commit(fs_info, trans);
3222 btrfs_abort_transaction(trans, ret);
3223 goto out_wake_log_root;
3226 mutex_lock(&root->log_mutex);
3227 if (root->last_log_commit < log_transid)
3228 root->last_log_commit = log_transid;
3229 mutex_unlock(&root->log_mutex);
3232 mutex_lock(&log_root_tree->log_mutex);
3233 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3235 log_root_tree->log_transid_committed++;
3236 atomic_set(&log_root_tree->log_commit[index2], 0);
3237 mutex_unlock(&log_root_tree->log_mutex);
3240 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3241 * all the updates above are seen by the woken threads. It might not be
3242 * necessary, but proving that seems to be hard.
3244 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3246 mutex_lock(&root->log_mutex);
3247 btrfs_remove_all_log_ctxs(root, index1, ret);
3248 root->log_transid_committed++;
3249 atomic_set(&root->log_commit[index1], 0);
3250 mutex_unlock(&root->log_mutex);
3253 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3254 * all the updates above are seen by the woken threads. It might not be
3255 * necessary, but proving that seems to be hard.
3257 cond_wake_up(&root->log_commit_wait[index1]);
3261 static void free_log_tree(struct btrfs_trans_handle *trans,
3262 struct btrfs_root *log)
3265 struct walk_control wc = {
3267 .process_func = process_one_buffer
3270 ret = walk_log_tree(trans, log, &wc);
3273 btrfs_abort_transaction(trans, ret);
3275 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3278 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3279 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3280 free_extent_buffer(log->node);
3285 * free all the extents used by the tree log. This should be called
3286 * at commit time of the full transaction
3288 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3290 if (root->log_root) {
3291 free_log_tree(trans, root->log_root);
3292 root->log_root = NULL;
3297 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3298 struct btrfs_fs_info *fs_info)
3300 if (fs_info->log_root_tree) {
3301 free_log_tree(trans, fs_info->log_root_tree);
3302 fs_info->log_root_tree = NULL;
3308 * If both a file and directory are logged, and unlinks or renames are
3309 * mixed in, we have a few interesting corners:
3311 * create file X in dir Y
3312 * link file X to X.link in dir Y
3314 * unlink file X but leave X.link
3317 * After a crash we would expect only X.link to exist. But file X
3318 * didn't get fsync'd again so the log has back refs for X and X.link.
3320 * We solve this by removing directory entries and inode backrefs from the
3321 * log when a file that was logged in the current transaction is
3322 * unlinked. Any later fsync will include the updated log entries, and
3323 * we'll be able to reconstruct the proper directory items from backrefs.
3325 * This optimizations allows us to avoid relogging the entire inode
3326 * or the entire directory.
3328 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3329 struct btrfs_root *root,
3330 const char *name, int name_len,
3331 struct btrfs_inode *dir, u64 index)
3333 struct btrfs_root *log;
3334 struct btrfs_dir_item *di;
3335 struct btrfs_path *path;
3339 u64 dir_ino = btrfs_ino(dir);
3341 if (dir->logged_trans < trans->transid)
3344 ret = join_running_log_trans(root);
3348 mutex_lock(&dir->log_mutex);
3350 log = root->log_root;
3351 path = btrfs_alloc_path();
3357 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3358 name, name_len, -1);
3364 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3365 bytes_del += name_len;
3371 btrfs_release_path(path);
3372 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3373 index, name, name_len, -1);
3379 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3380 bytes_del += name_len;
3387 /* update the directory size in the log to reflect the names
3391 struct btrfs_key key;
3393 key.objectid = dir_ino;
3395 key.type = BTRFS_INODE_ITEM_KEY;
3396 btrfs_release_path(path);
3398 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3404 struct btrfs_inode_item *item;
3407 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3408 struct btrfs_inode_item);
3409 i_size = btrfs_inode_size(path->nodes[0], item);
3410 if (i_size > bytes_del)
3411 i_size -= bytes_del;
3414 btrfs_set_inode_size(path->nodes[0], item, i_size);
3415 btrfs_mark_buffer_dirty(path->nodes[0]);
3418 btrfs_release_path(path);
3421 btrfs_free_path(path);
3423 mutex_unlock(&dir->log_mutex);
3424 if (ret == -ENOSPC) {
3425 btrfs_set_log_full_commit(root->fs_info, trans);
3428 btrfs_abort_transaction(trans, ret);
3430 btrfs_end_log_trans(root);
3435 /* see comments for btrfs_del_dir_entries_in_log */
3436 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3437 struct btrfs_root *root,
3438 const char *name, int name_len,
3439 struct btrfs_inode *inode, u64 dirid)
3441 struct btrfs_fs_info *fs_info = root->fs_info;
3442 struct btrfs_root *log;
3446 if (inode->logged_trans < trans->transid)
3449 ret = join_running_log_trans(root);
3452 log = root->log_root;
3453 mutex_lock(&inode->log_mutex);
3455 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3457 mutex_unlock(&inode->log_mutex);
3458 if (ret == -ENOSPC) {
3459 btrfs_set_log_full_commit(fs_info, trans);
3461 } else if (ret < 0 && ret != -ENOENT)
3462 btrfs_abort_transaction(trans, ret);
3463 btrfs_end_log_trans(root);
3469 * creates a range item in the log for 'dirid'. first_offset and
3470 * last_offset tell us which parts of the key space the log should
3471 * be considered authoritative for.
3473 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *log,
3475 struct btrfs_path *path,
3476 int key_type, u64 dirid,
3477 u64 first_offset, u64 last_offset)
3480 struct btrfs_key key;
3481 struct btrfs_dir_log_item *item;
3483 key.objectid = dirid;
3484 key.offset = first_offset;
3485 if (key_type == BTRFS_DIR_ITEM_KEY)
3486 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3488 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3489 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3493 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3494 struct btrfs_dir_log_item);
3495 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3496 btrfs_mark_buffer_dirty(path->nodes[0]);
3497 btrfs_release_path(path);
3502 * log all the items included in the current transaction for a given
3503 * directory. This also creates the range items in the log tree required
3504 * to replay anything deleted before the fsync
3506 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3507 struct btrfs_root *root, struct btrfs_inode *inode,
3508 struct btrfs_path *path,
3509 struct btrfs_path *dst_path, int key_type,
3510 struct btrfs_log_ctx *ctx,
3511 u64 min_offset, u64 *last_offset_ret)
3513 struct btrfs_key min_key;
3514 struct btrfs_root *log = root->log_root;
3515 struct extent_buffer *src;
3520 u64 first_offset = min_offset;
3521 u64 last_offset = (u64)-1;
3522 u64 ino = btrfs_ino(inode);
3524 log = root->log_root;
3526 min_key.objectid = ino;
3527 min_key.type = key_type;
3528 min_key.offset = min_offset;
3530 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3533 * we didn't find anything from this transaction, see if there
3534 * is anything at all
3536 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3537 min_key.objectid = ino;
3538 min_key.type = key_type;
3539 min_key.offset = (u64)-1;
3540 btrfs_release_path(path);
3541 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3543 btrfs_release_path(path);
3546 ret = btrfs_previous_item(root, path, ino, key_type);
3548 /* if ret == 0 there are items for this type,
3549 * create a range to tell us the last key of this type.
3550 * otherwise, there are no items in this directory after
3551 * *min_offset, and we create a range to indicate that.
3554 struct btrfs_key tmp;
3555 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3557 if (key_type == tmp.type)
3558 first_offset = max(min_offset, tmp.offset) + 1;
3563 /* go backward to find any previous key */
3564 ret = btrfs_previous_item(root, path, ino, key_type);
3566 struct btrfs_key tmp;
3567 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3568 if (key_type == tmp.type) {
3569 first_offset = tmp.offset;
3570 ret = overwrite_item(trans, log, dst_path,
3571 path->nodes[0], path->slots[0],
3579 btrfs_release_path(path);
3582 * Find the first key from this transaction again. See the note for
3583 * log_new_dir_dentries, if we're logging a directory recursively we
3584 * won't be holding its i_mutex, which means we can modify the directory
3585 * while we're logging it. If we remove an entry between our first
3586 * search and this search we'll not find the key again and can just
3589 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3594 * we have a block from this transaction, log every item in it
3595 * from our directory
3598 struct btrfs_key tmp;
3599 src = path->nodes[0];
3600 nritems = btrfs_header_nritems(src);
3601 for (i = path->slots[0]; i < nritems; i++) {
3602 struct btrfs_dir_item *di;
3604 btrfs_item_key_to_cpu(src, &min_key, i);
3606 if (min_key.objectid != ino || min_key.type != key_type)
3608 ret = overwrite_item(trans, log, dst_path, src, i,
3616 * We must make sure that when we log a directory entry,
3617 * the corresponding inode, after log replay, has a
3618 * matching link count. For example:
3624 * xfs_io -c "fsync" mydir
3626 * <mount fs and log replay>
3628 * Would result in a fsync log that when replayed, our
3629 * file inode would have a link count of 1, but we get
3630 * two directory entries pointing to the same inode.
3631 * After removing one of the names, it would not be
3632 * possible to remove the other name, which resulted
3633 * always in stale file handle errors, and would not
3634 * be possible to rmdir the parent directory, since
3635 * its i_size could never decrement to the value
3636 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3638 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3639 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3641 (btrfs_dir_transid(src, di) == trans->transid ||
3642 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3643 tmp.type != BTRFS_ROOT_ITEM_KEY)
3644 ctx->log_new_dentries = true;
3646 path->slots[0] = nritems;
3649 * look ahead to the next item and see if it is also
3650 * from this directory and from this transaction
3652 ret = btrfs_next_leaf(root, path);
3655 last_offset = (u64)-1;
3660 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3661 if (tmp.objectid != ino || tmp.type != key_type) {
3662 last_offset = (u64)-1;
3665 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3666 ret = overwrite_item(trans, log, dst_path,
3667 path->nodes[0], path->slots[0],
3672 last_offset = tmp.offset;
3677 btrfs_release_path(path);
3678 btrfs_release_path(dst_path);
3681 *last_offset_ret = last_offset;
3683 * insert the log range keys to indicate where the log
3686 ret = insert_dir_log_key(trans, log, path, key_type,
3687 ino, first_offset, last_offset);
3695 * logging directories is very similar to logging inodes, We find all the items
3696 * from the current transaction and write them to the log.
3698 * The recovery code scans the directory in the subvolume, and if it finds a
3699 * key in the range logged that is not present in the log tree, then it means
3700 * that dir entry was unlinked during the transaction.
3702 * In order for that scan to work, we must include one key smaller than
3703 * the smallest logged by this transaction and one key larger than the largest
3704 * key logged by this transaction.
3706 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3707 struct btrfs_root *root, struct btrfs_inode *inode,
3708 struct btrfs_path *path,
3709 struct btrfs_path *dst_path,
3710 struct btrfs_log_ctx *ctx)
3715 int key_type = BTRFS_DIR_ITEM_KEY;
3721 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3722 ctx, min_key, &max_key);
3725 if (max_key == (u64)-1)
3727 min_key = max_key + 1;
3730 if (key_type == BTRFS_DIR_ITEM_KEY) {
3731 key_type = BTRFS_DIR_INDEX_KEY;
3738 * a helper function to drop items from the log before we relog an
3739 * inode. max_key_type indicates the highest item type to remove.
3740 * This cannot be run for file data extents because it does not
3741 * free the extents they point to.
3743 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3744 struct btrfs_root *log,
3745 struct btrfs_path *path,
3746 u64 objectid, int max_key_type)
3749 struct btrfs_key key;
3750 struct btrfs_key found_key;
3753 key.objectid = objectid;
3754 key.type = max_key_type;
3755 key.offset = (u64)-1;
3758 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3759 BUG_ON(ret == 0); /* Logic error */
3763 if (path->slots[0] == 0)
3767 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3770 if (found_key.objectid != objectid)
3773 found_key.offset = 0;
3775 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3780 ret = btrfs_del_items(trans, log, path, start_slot,
3781 path->slots[0] - start_slot + 1);
3783 * If start slot isn't 0 then we don't need to re-search, we've
3784 * found the last guy with the objectid in this tree.
3786 if (ret || start_slot != 0)
3788 btrfs_release_path(path);
3790 btrfs_release_path(path);
3796 static void fill_inode_item(struct btrfs_trans_handle *trans,
3797 struct extent_buffer *leaf,
3798 struct btrfs_inode_item *item,
3799 struct inode *inode, int log_inode_only,
3802 struct btrfs_map_token token;
3804 btrfs_init_map_token(&token);
3806 if (log_inode_only) {
3807 /* set the generation to zero so the recover code
3808 * can tell the difference between an logging
3809 * just to say 'this inode exists' and a logging
3810 * to say 'update this inode with these values'
3812 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3813 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3815 btrfs_set_token_inode_generation(leaf, item,
3816 BTRFS_I(inode)->generation,
3818 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3821 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3822 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3823 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3824 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3826 btrfs_set_token_timespec_sec(leaf, &item->atime,
3827 inode->i_atime.tv_sec, &token);
3828 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3829 inode->i_atime.tv_nsec, &token);
3831 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3832 inode->i_mtime.tv_sec, &token);
3833 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3834 inode->i_mtime.tv_nsec, &token);
3836 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3837 inode->i_ctime.tv_sec, &token);
3838 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3839 inode->i_ctime.tv_nsec, &token);
3841 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3844 btrfs_set_token_inode_sequence(leaf, item,
3845 inode_peek_iversion(inode), &token);
3846 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3847 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3848 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3849 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3852 static int log_inode_item(struct btrfs_trans_handle *trans,
3853 struct btrfs_root *log, struct btrfs_path *path,
3854 struct btrfs_inode *inode)
3856 struct btrfs_inode_item *inode_item;
3859 ret = btrfs_insert_empty_item(trans, log, path,
3860 &inode->location, sizeof(*inode_item));
3861 if (ret && ret != -EEXIST)
3863 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3864 struct btrfs_inode_item);
3865 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3867 btrfs_release_path(path);
3871 static noinline int copy_items(struct btrfs_trans_handle *trans,
3872 struct btrfs_inode *inode,
3873 struct btrfs_path *dst_path,
3874 struct btrfs_path *src_path, u64 *last_extent,
3875 int start_slot, int nr, int inode_only,
3878 struct btrfs_fs_info *fs_info = trans->fs_info;
3879 unsigned long src_offset;
3880 unsigned long dst_offset;
3881 struct btrfs_root *log = inode->root->log_root;
3882 struct btrfs_file_extent_item *extent;
3883 struct btrfs_inode_item *inode_item;
3884 struct extent_buffer *src = src_path->nodes[0];
3885 struct btrfs_key first_key, last_key, key;
3887 struct btrfs_key *ins_keys;
3891 struct list_head ordered_sums;
3892 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3893 bool has_extents = false;
3894 bool need_find_last_extent = true;
3897 INIT_LIST_HEAD(&ordered_sums);
3899 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3900 nr * sizeof(u32), GFP_NOFS);
3904 first_key.objectid = (u64)-1;
3906 ins_sizes = (u32 *)ins_data;
3907 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3909 for (i = 0; i < nr; i++) {
3910 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3911 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3913 ret = btrfs_insert_empty_items(trans, log, dst_path,
3914 ins_keys, ins_sizes, nr);
3920 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3921 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3922 dst_path->slots[0]);
3924 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3927 last_key = ins_keys[i];
3929 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3930 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3932 struct btrfs_inode_item);
3933 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3935 inode_only == LOG_INODE_EXISTS,
3938 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3939 src_offset, ins_sizes[i]);
3943 * We set need_find_last_extent here in case we know we were
3944 * processing other items and then walk into the first extent in
3945 * the inode. If we don't hit an extent then nothing changes,
3946 * we'll do the last search the next time around.
3948 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3950 if (first_key.objectid == (u64)-1)
3951 first_key = ins_keys[i];
3953 need_find_last_extent = false;
3956 /* take a reference on file data extents so that truncates
3957 * or deletes of this inode don't have to relog the inode
3960 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3963 extent = btrfs_item_ptr(src, start_slot + i,
3964 struct btrfs_file_extent_item);
3966 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3969 found_type = btrfs_file_extent_type(src, extent);
3970 if (found_type == BTRFS_FILE_EXTENT_REG) {
3972 ds = btrfs_file_extent_disk_bytenr(src,
3974 /* ds == 0 is a hole */
3978 dl = btrfs_file_extent_disk_num_bytes(src,
3980 cs = btrfs_file_extent_offset(src, extent);
3981 cl = btrfs_file_extent_num_bytes(src,
3983 if (btrfs_file_extent_compression(src,
3989 ret = btrfs_lookup_csums_range(
3991 ds + cs, ds + cs + cl - 1,
3994 btrfs_release_path(dst_path);
4002 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4003 btrfs_release_path(dst_path);
4007 * we have to do this after the loop above to avoid changing the
4008 * log tree while trying to change the log tree.
4011 while (!list_empty(&ordered_sums)) {
4012 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4013 struct btrfs_ordered_sum,
4016 ret = btrfs_csum_file_blocks(trans, log, sums);
4017 list_del(&sums->list);
4024 if (need_find_last_extent && *last_extent == first_key.offset) {
4026 * We don't have any leafs between our current one and the one
4027 * we processed before that can have file extent items for our
4028 * inode (and have a generation number smaller than our current
4031 need_find_last_extent = false;
4035 * Because we use btrfs_search_forward we could skip leaves that were
4036 * not modified and then assume *last_extent is valid when it really
4037 * isn't. So back up to the previous leaf and read the end of the last
4038 * extent before we go and fill in holes.
4040 if (need_find_last_extent) {
4043 ret = btrfs_prev_leaf(inode->root, src_path);
4048 if (src_path->slots[0])
4049 src_path->slots[0]--;
4050 src = src_path->nodes[0];
4051 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
4052 if (key.objectid != btrfs_ino(inode) ||
4053 key.type != BTRFS_EXTENT_DATA_KEY)
4055 extent = btrfs_item_ptr(src, src_path->slots[0],
4056 struct btrfs_file_extent_item);
4057 if (btrfs_file_extent_type(src, extent) ==
4058 BTRFS_FILE_EXTENT_INLINE) {
4059 len = btrfs_file_extent_ram_bytes(src, extent);
4060 *last_extent = ALIGN(key.offset + len,
4061 fs_info->sectorsize);
4063 len = btrfs_file_extent_num_bytes(src, extent);
4064 *last_extent = key.offset + len;
4068 /* So we did prev_leaf, now we need to move to the next leaf, but a few
4069 * things could have happened
4071 * 1) A merge could have happened, so we could currently be on a leaf
4072 * that holds what we were copying in the first place.
4073 * 2) A split could have happened, and now not all of the items we want
4074 * are on the same leaf.
4076 * So we need to adjust how we search for holes, we need to drop the
4077 * path and re-search for the first extent key we found, and then walk
4078 * forward until we hit the last one we copied.
4080 if (need_find_last_extent) {
4081 /* btrfs_prev_leaf could return 1 without releasing the path */
4082 btrfs_release_path(src_path);
4083 ret = btrfs_search_slot(NULL, inode->root, &first_key,
4088 src = src_path->nodes[0];
4089 i = src_path->slots[0];
4095 * Ok so here we need to go through and fill in any holes we may have
4096 * to make sure that holes are punched for those areas in case they had
4097 * extents previously.
4103 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
4104 ret = btrfs_next_leaf(inode->root, src_path);
4108 src = src_path->nodes[0];
4110 need_find_last_extent = true;
4113 btrfs_item_key_to_cpu(src, &key, i);
4114 if (!btrfs_comp_cpu_keys(&key, &last_key))
4116 if (key.objectid != btrfs_ino(inode) ||
4117 key.type != BTRFS_EXTENT_DATA_KEY) {
4121 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
4122 if (btrfs_file_extent_type(src, extent) ==
4123 BTRFS_FILE_EXTENT_INLINE) {
4124 len = btrfs_file_extent_ram_bytes(src, extent);
4125 extent_end = ALIGN(key.offset + len,
4126 fs_info->sectorsize);
4128 len = btrfs_file_extent_num_bytes(src, extent);
4129 extent_end = key.offset + len;
4133 if (*last_extent == key.offset) {
4134 *last_extent = extent_end;
4137 offset = *last_extent;
4138 len = key.offset - *last_extent;
4139 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4140 offset, 0, 0, len, 0, len, 0, 0, 0);
4143 *last_extent = extent_end;
4147 * Check if there is a hole between the last extent found in our leaf
4148 * and the first extent in the next leaf. If there is one, we need to
4149 * log an explicit hole so that at replay time we can punch the hole.
4152 key.objectid == btrfs_ino(inode) &&
4153 key.type == BTRFS_EXTENT_DATA_KEY &&
4154 i == btrfs_header_nritems(src_path->nodes[0])) {
4155 ret = btrfs_next_leaf(inode->root, src_path);
4156 need_find_last_extent = true;
4159 } else if (ret == 0) {
4160 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4161 src_path->slots[0]);
4162 if (key.objectid == btrfs_ino(inode) &&
4163 key.type == BTRFS_EXTENT_DATA_KEY &&
4164 *last_extent < key.offset) {
4165 const u64 len = key.offset - *last_extent;
4167 ret = btrfs_insert_file_extent(trans, log,
4176 * Need to let the callers know we dropped the path so they should
4179 if (!ret && need_find_last_extent)
4184 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4186 struct extent_map *em1, *em2;
4188 em1 = list_entry(a, struct extent_map, list);
4189 em2 = list_entry(b, struct extent_map, list);
4191 if (em1->start < em2->start)
4193 else if (em1->start > em2->start)
4198 static int log_extent_csums(struct btrfs_trans_handle *trans,
4199 struct btrfs_inode *inode,
4200 struct btrfs_root *log_root,
4201 const struct extent_map *em)
4205 LIST_HEAD(ordered_sums);
4208 if (inode->flags & BTRFS_INODE_NODATASUM ||
4209 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4210 em->block_start == EXTENT_MAP_HOLE)
4213 /* If we're compressed we have to save the entire range of csums. */
4214 if (em->compress_type) {
4216 csum_len = max(em->block_len, em->orig_block_len);
4218 csum_offset = em->mod_start - em->start;
4219 csum_len = em->mod_len;
4222 /* block start is already adjusted for the file extent offset. */
4223 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4224 em->block_start + csum_offset,
4225 em->block_start + csum_offset +
4226 csum_len - 1, &ordered_sums, 0);
4230 while (!list_empty(&ordered_sums)) {
4231 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4232 struct btrfs_ordered_sum,
4235 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4236 list_del(&sums->list);
4243 static int log_one_extent(struct btrfs_trans_handle *trans,
4244 struct btrfs_inode *inode, struct btrfs_root *root,
4245 const struct extent_map *em,
4246 struct btrfs_path *path,
4247 struct btrfs_log_ctx *ctx)
4249 struct btrfs_root *log = root->log_root;
4250 struct btrfs_file_extent_item *fi;
4251 struct extent_buffer *leaf;
4252 struct btrfs_map_token token;
4253 struct btrfs_key key;
4254 u64 extent_offset = em->start - em->orig_start;
4257 int extent_inserted = 0;
4259 ret = log_extent_csums(trans, inode, log, em);
4263 btrfs_init_map_token(&token);
4265 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4266 em->start + em->len, NULL, 0, 1,
4267 sizeof(*fi), &extent_inserted);
4271 if (!extent_inserted) {
4272 key.objectid = btrfs_ino(inode);
4273 key.type = BTRFS_EXTENT_DATA_KEY;
4274 key.offset = em->start;
4276 ret = btrfs_insert_empty_item(trans, log, path, &key,
4281 leaf = path->nodes[0];
4282 fi = btrfs_item_ptr(leaf, path->slots[0],
4283 struct btrfs_file_extent_item);
4285 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4287 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4288 btrfs_set_token_file_extent_type(leaf, fi,
4289 BTRFS_FILE_EXTENT_PREALLOC,
4292 btrfs_set_token_file_extent_type(leaf, fi,
4293 BTRFS_FILE_EXTENT_REG,
4296 block_len = max(em->block_len, em->orig_block_len);
4297 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4298 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4301 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4303 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4304 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4306 extent_offset, &token);
4307 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4310 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4311 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4315 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4316 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4317 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4318 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4320 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4321 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4322 btrfs_mark_buffer_dirty(leaf);
4324 btrfs_release_path(path);
4330 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4331 * lose them after doing a fast fsync and replaying the log. We scan the
4332 * subvolume's root instead of iterating the inode's extent map tree because
4333 * otherwise we can log incorrect extent items based on extent map conversion.
4334 * That can happen due to the fact that extent maps are merged when they
4335 * are not in the extent map tree's list of modified extents.
4337 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4338 struct btrfs_inode *inode,
4339 struct btrfs_path *path)
4341 struct btrfs_root *root = inode->root;
4342 struct btrfs_key key;
4343 const u64 i_size = i_size_read(&inode->vfs_inode);
4344 const u64 ino = btrfs_ino(inode);
4345 struct btrfs_path *dst_path = NULL;
4346 u64 last_extent = (u64)-1;
4351 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4355 key.type = BTRFS_EXTENT_DATA_KEY;
4356 key.offset = i_size;
4357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4362 struct extent_buffer *leaf = path->nodes[0];
4363 int slot = path->slots[0];
4365 if (slot >= btrfs_header_nritems(leaf)) {
4367 ret = copy_items(trans, inode, dst_path, path,
4368 &last_extent, start_slot,
4374 ret = btrfs_next_leaf(root, path);
4384 btrfs_item_key_to_cpu(leaf, &key, slot);
4385 if (key.objectid > ino)
4387 if (WARN_ON_ONCE(key.objectid < ino) ||
4388 key.type < BTRFS_EXTENT_DATA_KEY ||
4389 key.offset < i_size) {
4393 if (last_extent == (u64)-1) {
4394 last_extent = key.offset;
4396 * Avoid logging extent items logged in past fsync calls
4397 * and leading to duplicate keys in the log tree.
4400 ret = btrfs_truncate_inode_items(trans,
4404 BTRFS_EXTENT_DATA_KEY);
4405 } while (ret == -EAGAIN);
4414 dst_path = btrfs_alloc_path();
4422 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4423 start_slot, ins_nr, 1, 0);
4428 btrfs_release_path(path);
4429 btrfs_free_path(dst_path);
4433 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4434 struct btrfs_root *root,
4435 struct btrfs_inode *inode,
4436 struct btrfs_path *path,
4437 struct btrfs_log_ctx *ctx,
4441 struct extent_map *em, *n;
4442 struct list_head extents;
4443 struct extent_map_tree *tree = &inode->extent_tree;
4448 INIT_LIST_HEAD(&extents);
4450 write_lock(&tree->lock);
4451 test_gen = root->fs_info->last_trans_committed;
4453 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4455 * Skip extents outside our logging range. It's important to do
4456 * it for correctness because if we don't ignore them, we may
4457 * log them before their ordered extent completes, and therefore
4458 * we could log them without logging their respective checksums
4459 * (the checksum items are added to the csum tree at the very
4460 * end of btrfs_finish_ordered_io()). Also leave such extents
4461 * outside of our range in the list, since we may have another
4462 * ranged fsync in the near future that needs them. If an extent
4463 * outside our range corresponds to a hole, log it to avoid
4464 * leaving gaps between extents (fsck will complain when we are
4465 * not using the NO_HOLES feature).
4467 if ((em->start > end || em->start + em->len <= start) &&
4468 em->block_start != EXTENT_MAP_HOLE)
4471 list_del_init(&em->list);
4473 * Just an arbitrary number, this can be really CPU intensive
4474 * once we start getting a lot of extents, and really once we
4475 * have a bunch of extents we just want to commit since it will
4478 if (++num > 32768) {
4479 list_del_init(&tree->modified_extents);
4484 if (em->generation <= test_gen)
4487 /* We log prealloc extents beyond eof later. */
4488 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4489 em->start >= i_size_read(&inode->vfs_inode))
4492 /* Need a ref to keep it from getting evicted from cache */
4493 refcount_inc(&em->refs);
4494 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4495 list_add_tail(&em->list, &extents);
4499 list_sort(NULL, &extents, extent_cmp);
4501 while (!list_empty(&extents)) {
4502 em = list_entry(extents.next, struct extent_map, list);
4504 list_del_init(&em->list);
4507 * If we had an error we just need to delete everybody from our
4511 clear_em_logging(tree, em);
4512 free_extent_map(em);
4516 write_unlock(&tree->lock);
4518 ret = log_one_extent(trans, inode, root, em, path, ctx);
4519 write_lock(&tree->lock);
4520 clear_em_logging(tree, em);
4521 free_extent_map(em);
4523 WARN_ON(!list_empty(&extents));
4524 write_unlock(&tree->lock);
4526 btrfs_release_path(path);
4528 ret = btrfs_log_prealloc_extents(trans, inode, path);
4533 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4534 struct btrfs_path *path, u64 *size_ret)
4536 struct btrfs_key key;
4539 key.objectid = btrfs_ino(inode);
4540 key.type = BTRFS_INODE_ITEM_KEY;
4543 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4546 } else if (ret > 0) {
4549 struct btrfs_inode_item *item;
4551 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4552 struct btrfs_inode_item);
4553 *size_ret = btrfs_inode_size(path->nodes[0], item);
4555 * If the in-memory inode's i_size is smaller then the inode
4556 * size stored in the btree, return the inode's i_size, so
4557 * that we get a correct inode size after replaying the log
4558 * when before a power failure we had a shrinking truncate
4559 * followed by addition of a new name (rename / new hard link).
4560 * Otherwise return the inode size from the btree, to avoid
4561 * data loss when replaying a log due to previously doing a
4562 * write that expands the inode's size and logging a new name
4563 * immediately after.
4565 if (*size_ret > inode->vfs_inode.i_size)
4566 *size_ret = inode->vfs_inode.i_size;
4569 btrfs_release_path(path);
4574 * At the moment we always log all xattrs. This is to figure out at log replay
4575 * time which xattrs must have their deletion replayed. If a xattr is missing
4576 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4577 * because if a xattr is deleted, the inode is fsynced and a power failure
4578 * happens, causing the log to be replayed the next time the fs is mounted,
4579 * we want the xattr to not exist anymore (same behaviour as other filesystems
4580 * with a journal, ext3/4, xfs, f2fs, etc).
4582 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4583 struct btrfs_root *root,
4584 struct btrfs_inode *inode,
4585 struct btrfs_path *path,
4586 struct btrfs_path *dst_path)
4589 struct btrfs_key key;
4590 const u64 ino = btrfs_ino(inode);
4595 key.type = BTRFS_XATTR_ITEM_KEY;
4598 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4603 int slot = path->slots[0];
4604 struct extent_buffer *leaf = path->nodes[0];
4605 int nritems = btrfs_header_nritems(leaf);
4607 if (slot >= nritems) {
4609 u64 last_extent = 0;
4611 ret = copy_items(trans, inode, dst_path, path,
4612 &last_extent, start_slot,
4614 /* can't be 1, extent items aren't processed */
4620 ret = btrfs_next_leaf(root, path);
4628 btrfs_item_key_to_cpu(leaf, &key, slot);
4629 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4639 u64 last_extent = 0;
4641 ret = copy_items(trans, inode, dst_path, path,
4642 &last_extent, start_slot,
4644 /* can't be 1, extent items aren't processed */
4654 * If the no holes feature is enabled we need to make sure any hole between the
4655 * last extent and the i_size of our inode is explicitly marked in the log. This
4656 * is to make sure that doing something like:
4658 * 1) create file with 128Kb of data
4659 * 2) truncate file to 64Kb
4660 * 3) truncate file to 256Kb
4662 * 5) <crash/power failure>
4663 * 6) mount fs and trigger log replay
4665 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4666 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4667 * file correspond to a hole. The presence of explicit holes in a log tree is
4668 * what guarantees that log replay will remove/adjust file extent items in the
4671 * Here we do not need to care about holes between extents, that is already done
4672 * by copy_items(). We also only need to do this in the full sync path, where we
4673 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4674 * lookup the list of modified extent maps and if any represents a hole, we
4675 * insert a corresponding extent representing a hole in the log tree.
4677 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4678 struct btrfs_root *root,
4679 struct btrfs_inode *inode,
4680 struct btrfs_path *path)
4682 struct btrfs_fs_info *fs_info = root->fs_info;
4684 struct btrfs_key key;
4687 struct extent_buffer *leaf;
4688 struct btrfs_root *log = root->log_root;
4689 const u64 ino = btrfs_ino(inode);
4690 const u64 i_size = i_size_read(&inode->vfs_inode);
4692 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4696 key.type = BTRFS_EXTENT_DATA_KEY;
4697 key.offset = (u64)-1;
4699 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4704 ASSERT(path->slots[0] > 0);
4706 leaf = path->nodes[0];
4707 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4709 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4710 /* inode does not have any extents */
4714 struct btrfs_file_extent_item *extent;
4718 * If there's an extent beyond i_size, an explicit hole was
4719 * already inserted by copy_items().
4721 if (key.offset >= i_size)
4724 extent = btrfs_item_ptr(leaf, path->slots[0],
4725 struct btrfs_file_extent_item);
4727 if (btrfs_file_extent_type(leaf, extent) ==
4728 BTRFS_FILE_EXTENT_INLINE)
4731 len = btrfs_file_extent_num_bytes(leaf, extent);
4732 /* Last extent goes beyond i_size, no need to log a hole. */
4733 if (key.offset + len > i_size)
4735 hole_start = key.offset + len;
4736 hole_size = i_size - hole_start;
4738 btrfs_release_path(path);
4740 /* Last extent ends at i_size. */
4744 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4745 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4746 hole_size, 0, hole_size, 0, 0, 0);
4751 * When we are logging a new inode X, check if it doesn't have a reference that
4752 * matches the reference from some other inode Y created in a past transaction
4753 * and that was renamed in the current transaction. If we don't do this, then at
4754 * log replay time we can lose inode Y (and all its files if it's a directory):
4757 * echo "hello world" > /mnt/x/foobar
4760 * mkdir /mnt/x # or touch /mnt/x
4761 * xfs_io -c fsync /mnt/x
4763 * mount fs, trigger log replay
4765 * After the log replay procedure, we would lose the first directory and all its
4766 * files (file foobar).
4767 * For the case where inode Y is not a directory we simply end up losing it:
4769 * echo "123" > /mnt/foo
4771 * mv /mnt/foo /mnt/bar
4772 * echo "abc" > /mnt/foo
4773 * xfs_io -c fsync /mnt/foo
4776 * We also need this for cases where a snapshot entry is replaced by some other
4777 * entry (file or directory) otherwise we end up with an unreplayable log due to
4778 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4779 * if it were a regular entry:
4782 * btrfs subvolume snapshot /mnt /mnt/x/snap
4783 * btrfs subvolume delete /mnt/x/snap
4786 * fsync /mnt/x or fsync some new file inside it
4789 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4790 * the same transaction.
4792 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4794 const struct btrfs_key *key,
4795 struct btrfs_inode *inode,
4796 u64 *other_ino, u64 *other_parent)
4799 struct btrfs_path *search_path;
4802 u32 item_size = btrfs_item_size_nr(eb, slot);
4804 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4806 search_path = btrfs_alloc_path();
4809 search_path->search_commit_root = 1;
4810 search_path->skip_locking = 1;
4812 while (cur_offset < item_size) {
4816 unsigned long name_ptr;
4817 struct btrfs_dir_item *di;
4819 if (key->type == BTRFS_INODE_REF_KEY) {
4820 struct btrfs_inode_ref *iref;
4822 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4823 parent = key->offset;
4824 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4825 name_ptr = (unsigned long)(iref + 1);
4826 this_len = sizeof(*iref) + this_name_len;
4828 struct btrfs_inode_extref *extref;
4830 extref = (struct btrfs_inode_extref *)(ptr +
4832 parent = btrfs_inode_extref_parent(eb, extref);
4833 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4834 name_ptr = (unsigned long)&extref->name;
4835 this_len = sizeof(*extref) + this_name_len;
4838 if (this_name_len > name_len) {
4841 new_name = krealloc(name, this_name_len, GFP_NOFS);
4846 name_len = this_name_len;
4850 read_extent_buffer(eb, name, name_ptr, this_name_len);
4851 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4852 parent, name, this_name_len, 0);
4853 if (di && !IS_ERR(di)) {
4854 struct btrfs_key di_key;
4856 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4858 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4859 if (di_key.objectid != key->objectid) {
4861 *other_ino = di_key.objectid;
4862 *other_parent = parent;
4870 } else if (IS_ERR(di)) {
4874 btrfs_release_path(search_path);
4876 cur_offset += this_len;
4880 btrfs_free_path(search_path);
4885 struct btrfs_ino_list {
4888 struct list_head list;
4891 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4892 struct btrfs_root *root,
4893 struct btrfs_path *path,
4894 struct btrfs_log_ctx *ctx,
4895 u64 ino, u64 parent)
4897 struct btrfs_ino_list *ino_elem;
4898 LIST_HEAD(inode_list);
4901 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4904 ino_elem->ino = ino;
4905 ino_elem->parent = parent;
4906 list_add_tail(&ino_elem->list, &inode_list);
4908 while (!list_empty(&inode_list)) {
4909 struct btrfs_fs_info *fs_info = root->fs_info;
4910 struct btrfs_key key;
4911 struct inode *inode;
4913 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4915 ino = ino_elem->ino;
4916 parent = ino_elem->parent;
4917 list_del(&ino_elem->list);
4922 btrfs_release_path(path);
4925 key.type = BTRFS_INODE_ITEM_KEY;
4927 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4929 * If the other inode that had a conflicting dir entry was
4930 * deleted in the current transaction, we need to log its parent
4933 if (IS_ERR(inode)) {
4934 ret = PTR_ERR(inode);
4935 if (ret == -ENOENT) {
4936 key.objectid = parent;
4937 inode = btrfs_iget(fs_info->sb, &key, root,
4939 if (IS_ERR(inode)) {
4940 ret = PTR_ERR(inode);
4942 ret = btrfs_log_inode(trans, root,
4944 LOG_OTHER_INODE_ALL,
4952 * We are safe logging the other inode without acquiring its
4953 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4954 * are safe against concurrent renames of the other inode as
4955 * well because during a rename we pin the log and update the
4956 * log with the new name before we unpin it.
4958 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4959 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4966 key.type = BTRFS_INODE_REF_KEY;
4968 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4975 struct extent_buffer *leaf = path->nodes[0];
4976 int slot = path->slots[0];
4978 u64 other_parent = 0;
4980 if (slot >= btrfs_header_nritems(leaf)) {
4981 ret = btrfs_next_leaf(root, path);
4984 } else if (ret > 0) {
4991 btrfs_item_key_to_cpu(leaf, &key, slot);
4992 if (key.objectid != ino ||
4993 (key.type != BTRFS_INODE_REF_KEY &&
4994 key.type != BTRFS_INODE_EXTREF_KEY)) {
4999 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5000 BTRFS_I(inode), &other_ino,
5005 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5010 ino_elem->ino = other_ino;
5011 ino_elem->parent = other_parent;
5012 list_add_tail(&ino_elem->list, &inode_list);
5023 /* log a single inode in the tree log.
5024 * At least one parent directory for this inode must exist in the tree
5025 * or be logged already.
5027 * Any items from this inode changed by the current transaction are copied
5028 * to the log tree. An extra reference is taken on any extents in this
5029 * file, allowing us to avoid a whole pile of corner cases around logging
5030 * blocks that have been removed from the tree.
5032 * See LOG_INODE_ALL and related defines for a description of what inode_only
5035 * This handles both files and directories.
5037 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5038 struct btrfs_root *root, struct btrfs_inode *inode,
5042 struct btrfs_log_ctx *ctx)
5044 struct btrfs_fs_info *fs_info = root->fs_info;
5045 struct btrfs_path *path;
5046 struct btrfs_path *dst_path;
5047 struct btrfs_key min_key;
5048 struct btrfs_key max_key;
5049 struct btrfs_root *log = root->log_root;
5050 u64 last_extent = 0;
5054 int ins_start_slot = 0;
5056 bool fast_search = false;
5057 u64 ino = btrfs_ino(inode);
5058 struct extent_map_tree *em_tree = &inode->extent_tree;
5059 u64 logged_isize = 0;
5060 bool need_log_inode_item = true;
5061 bool xattrs_logged = false;
5062 bool recursive_logging = false;
5064 path = btrfs_alloc_path();
5067 dst_path = btrfs_alloc_path();
5069 btrfs_free_path(path);
5073 min_key.objectid = ino;
5074 min_key.type = BTRFS_INODE_ITEM_KEY;
5077 max_key.objectid = ino;
5080 /* today the code can only do partial logging of directories */
5081 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5082 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5083 &inode->runtime_flags) &&
5084 inode_only >= LOG_INODE_EXISTS))
5085 max_key.type = BTRFS_XATTR_ITEM_KEY;
5087 max_key.type = (u8)-1;
5088 max_key.offset = (u64)-1;
5091 * Only run delayed items if we are a dir or a new file.
5092 * Otherwise commit the delayed inode only, which is needed in
5093 * order for the log replay code to mark inodes for link count
5094 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5096 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5097 inode->generation > fs_info->last_trans_committed)
5098 ret = btrfs_commit_inode_delayed_items(trans, inode);
5100 ret = btrfs_commit_inode_delayed_inode(inode);
5103 btrfs_free_path(path);
5104 btrfs_free_path(dst_path);
5108 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5109 recursive_logging = true;
5110 if (inode_only == LOG_OTHER_INODE)
5111 inode_only = LOG_INODE_EXISTS;
5113 inode_only = LOG_INODE_ALL;
5114 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5116 mutex_lock(&inode->log_mutex);
5120 * a brute force approach to making sure we get the most uptodate
5121 * copies of everything.
5123 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5124 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5126 if (inode_only == LOG_INODE_EXISTS)
5127 max_key_type = BTRFS_XATTR_ITEM_KEY;
5128 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5130 if (inode_only == LOG_INODE_EXISTS) {
5132 * Make sure the new inode item we write to the log has
5133 * the same isize as the current one (if it exists).
5134 * This is necessary to prevent data loss after log
5135 * replay, and also to prevent doing a wrong expanding
5136 * truncate - for e.g. create file, write 4K into offset
5137 * 0, fsync, write 4K into offset 4096, add hard link,
5138 * fsync some other file (to sync log), power fail - if
5139 * we use the inode's current i_size, after log replay
5140 * we get a 8Kb file, with the last 4Kb extent as a hole
5141 * (zeroes), as if an expanding truncate happened,
5142 * instead of getting a file of 4Kb only.
5144 err = logged_inode_size(log, inode, path, &logged_isize);
5148 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5149 &inode->runtime_flags)) {
5150 if (inode_only == LOG_INODE_EXISTS) {
5151 max_key.type = BTRFS_XATTR_ITEM_KEY;
5152 ret = drop_objectid_items(trans, log, path, ino,
5155 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5156 &inode->runtime_flags);
5157 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5158 &inode->runtime_flags);
5160 ret = btrfs_truncate_inode_items(trans,
5161 log, &inode->vfs_inode, 0, 0);
5166 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5167 &inode->runtime_flags) ||
5168 inode_only == LOG_INODE_EXISTS) {
5169 if (inode_only == LOG_INODE_ALL)
5171 max_key.type = BTRFS_XATTR_ITEM_KEY;
5172 ret = drop_objectid_items(trans, log, path, ino,
5175 if (inode_only == LOG_INODE_ALL)
5188 ret = btrfs_search_forward(root, &min_key,
5189 path, trans->transid);
5197 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5198 if (min_key.objectid != ino)
5200 if (min_key.type > max_key.type)
5203 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5204 need_log_inode_item = false;
5206 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5207 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5208 inode->generation == trans->transid &&
5209 !recursive_logging) {
5211 u64 other_parent = 0;
5213 ret = btrfs_check_ref_name_override(path->nodes[0],
5214 path->slots[0], &min_key, inode,
5215 &other_ino, &other_parent);
5219 } else if (ret > 0 && ctx &&
5220 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5225 ins_start_slot = path->slots[0];
5227 ret = copy_items(trans, inode, dst_path, path,
5228 &last_extent, ins_start_slot,
5237 err = log_conflicting_inodes(trans, root, path,
5238 ctx, other_ino, other_parent);
5241 btrfs_release_path(path);
5246 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5247 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5250 ret = copy_items(trans, inode, dst_path, path,
5251 &last_extent, ins_start_slot,
5252 ins_nr, inode_only, logged_isize);
5259 btrfs_release_path(path);
5265 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5268 } else if (!ins_nr) {
5269 ins_start_slot = path->slots[0];
5274 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5275 ins_start_slot, ins_nr, inode_only,
5283 btrfs_release_path(path);
5287 ins_start_slot = path->slots[0];
5290 nritems = btrfs_header_nritems(path->nodes[0]);
5292 if (path->slots[0] < nritems) {
5293 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5298 ret = copy_items(trans, inode, dst_path, path,
5299 &last_extent, ins_start_slot,
5300 ins_nr, inode_only, logged_isize);
5308 btrfs_release_path(path);
5310 if (min_key.offset < (u64)-1) {
5312 } else if (min_key.type < max_key.type) {
5320 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5321 ins_start_slot, ins_nr, inode_only,
5331 btrfs_release_path(path);
5332 btrfs_release_path(dst_path);
5333 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5336 xattrs_logged = true;
5337 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5338 btrfs_release_path(path);
5339 btrfs_release_path(dst_path);
5340 err = btrfs_log_trailing_hole(trans, root, inode, path);
5345 btrfs_release_path(path);
5346 btrfs_release_path(dst_path);
5347 if (need_log_inode_item) {
5348 err = log_inode_item(trans, log, dst_path, inode);
5349 if (!err && !xattrs_logged) {
5350 err = btrfs_log_all_xattrs(trans, root, inode, path,
5352 btrfs_release_path(path);
5358 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5364 } else if (inode_only == LOG_INODE_ALL) {
5365 struct extent_map *em, *n;
5367 write_lock(&em_tree->lock);
5369 * We can't just remove every em if we're called for a ranged
5370 * fsync - that is, one that doesn't cover the whole possible
5371 * file range (0 to LLONG_MAX). This is because we can have
5372 * em's that fall outside the range we're logging and therefore
5373 * their ordered operations haven't completed yet
5374 * (btrfs_finish_ordered_io() not invoked yet). This means we
5375 * didn't get their respective file extent item in the fs/subvol
5376 * tree yet, and need to let the next fast fsync (one which
5377 * consults the list of modified extent maps) find the em so
5378 * that it logs a matching file extent item and waits for the
5379 * respective ordered operation to complete (if it's still
5382 * Removing every em outside the range we're logging would make
5383 * the next fast fsync not log their matching file extent items,
5384 * therefore making us lose data after a log replay.
5386 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5388 const u64 mod_end = em->mod_start + em->mod_len - 1;
5390 if (em->mod_start >= start && mod_end <= end)
5391 list_del_init(&em->list);
5393 write_unlock(&em_tree->lock);
5396 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5397 ret = log_directory_changes(trans, root, inode, path, dst_path,
5405 spin_lock(&inode->lock);
5406 inode->logged_trans = trans->transid;
5407 inode->last_log_commit = inode->last_sub_trans;
5408 spin_unlock(&inode->lock);
5410 mutex_unlock(&inode->log_mutex);
5412 btrfs_free_path(path);
5413 btrfs_free_path(dst_path);
5418 * Check if we must fallback to a transaction commit when logging an inode.
5419 * This must be called after logging the inode and is used only in the context
5420 * when fsyncing an inode requires the need to log some other inode - in which
5421 * case we can't lock the i_mutex of each other inode we need to log as that
5422 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5423 * log inodes up or down in the hierarchy) or rename operations for example. So
5424 * we take the log_mutex of the inode after we have logged it and then check for
5425 * its last_unlink_trans value - this is safe because any task setting
5426 * last_unlink_trans must take the log_mutex and it must do this before it does
5427 * the actual unlink operation, so if we do this check before a concurrent task
5428 * sets last_unlink_trans it means we've logged a consistent version/state of
5429 * all the inode items, otherwise we are not sure and must do a transaction
5430 * commit (the concurrent task might have only updated last_unlink_trans before
5431 * we logged the inode or it might have also done the unlink).
5433 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5434 struct btrfs_inode *inode)
5436 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5439 mutex_lock(&inode->log_mutex);
5440 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5442 * Make sure any commits to the log are forced to be full
5445 btrfs_set_log_full_commit(fs_info, trans);
5448 mutex_unlock(&inode->log_mutex);
5454 * follow the dentry parent pointers up the chain and see if any
5455 * of the directories in it require a full commit before they can
5456 * be logged. Returns zero if nothing special needs to be done or 1 if
5457 * a full commit is required.
5459 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5460 struct btrfs_inode *inode,
5461 struct dentry *parent,
5462 struct super_block *sb,
5466 struct dentry *old_parent = NULL;
5467 struct btrfs_inode *orig_inode = inode;
5470 * for regular files, if its inode is already on disk, we don't
5471 * have to worry about the parents at all. This is because
5472 * we can use the last_unlink_trans field to record renames
5473 * and other fun in this file.
5475 if (S_ISREG(inode->vfs_inode.i_mode) &&
5476 inode->generation <= last_committed &&
5477 inode->last_unlink_trans <= last_committed)
5480 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5481 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5483 inode = BTRFS_I(d_inode(parent));
5488 * If we are logging a directory then we start with our inode,
5489 * not our parent's inode, so we need to skip setting the
5490 * logged_trans so that further down in the log code we don't
5491 * think this inode has already been logged.
5493 if (inode != orig_inode)
5494 inode->logged_trans = trans->transid;
5497 if (btrfs_must_commit_transaction(trans, inode)) {
5502 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5505 if (IS_ROOT(parent)) {
5506 inode = BTRFS_I(d_inode(parent));
5507 if (btrfs_must_commit_transaction(trans, inode))
5512 parent = dget_parent(parent);
5514 old_parent = parent;
5515 inode = BTRFS_I(d_inode(parent));
5523 struct btrfs_dir_list {
5525 struct list_head list;
5529 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5530 * details about the why it is needed.
5531 * This is a recursive operation - if an existing dentry corresponds to a
5532 * directory, that directory's new entries are logged too (same behaviour as
5533 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5534 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5535 * complains about the following circular lock dependency / possible deadlock:
5539 * lock(&type->i_mutex_dir_key#3/2);
5540 * lock(sb_internal#2);
5541 * lock(&type->i_mutex_dir_key#3/2);
5542 * lock(&sb->s_type->i_mutex_key#14);
5544 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5545 * sb_start_intwrite() in btrfs_start_transaction().
5546 * Not locking i_mutex of the inodes is still safe because:
5548 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5549 * that while logging the inode new references (names) are added or removed
5550 * from the inode, leaving the logged inode item with a link count that does
5551 * not match the number of logged inode reference items. This is fine because
5552 * at log replay time we compute the real number of links and correct the
5553 * link count in the inode item (see replay_one_buffer() and
5554 * link_to_fixup_dir());
5556 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5557 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5558 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5559 * has a size that doesn't match the sum of the lengths of all the logged
5560 * names. This does not result in a problem because if a dir_item key is
5561 * logged but its matching dir_index key is not logged, at log replay time we
5562 * don't use it to replay the respective name (see replay_one_name()). On the
5563 * other hand if only the dir_index key ends up being logged, the respective
5564 * name is added to the fs/subvol tree with both the dir_item and dir_index
5565 * keys created (see replay_one_name()).
5566 * The directory's inode item with a wrong i_size is not a problem as well,
5567 * since we don't use it at log replay time to set the i_size in the inode
5568 * item of the fs/subvol tree (see overwrite_item()).
5570 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5571 struct btrfs_root *root,
5572 struct btrfs_inode *start_inode,
5573 struct btrfs_log_ctx *ctx)
5575 struct btrfs_fs_info *fs_info = root->fs_info;
5576 struct btrfs_root *log = root->log_root;
5577 struct btrfs_path *path;
5578 LIST_HEAD(dir_list);
5579 struct btrfs_dir_list *dir_elem;
5582 path = btrfs_alloc_path();
5586 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5588 btrfs_free_path(path);
5591 dir_elem->ino = btrfs_ino(start_inode);
5592 list_add_tail(&dir_elem->list, &dir_list);
5594 while (!list_empty(&dir_list)) {
5595 struct extent_buffer *leaf;
5596 struct btrfs_key min_key;
5600 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5603 goto next_dir_inode;
5605 min_key.objectid = dir_elem->ino;
5606 min_key.type = BTRFS_DIR_ITEM_KEY;
5609 btrfs_release_path(path);
5610 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5612 goto next_dir_inode;
5613 } else if (ret > 0) {
5615 goto next_dir_inode;
5619 leaf = path->nodes[0];
5620 nritems = btrfs_header_nritems(leaf);
5621 for (i = path->slots[0]; i < nritems; i++) {
5622 struct btrfs_dir_item *di;
5623 struct btrfs_key di_key;
5624 struct inode *di_inode;
5625 struct btrfs_dir_list *new_dir_elem;
5626 int log_mode = LOG_INODE_EXISTS;
5629 btrfs_item_key_to_cpu(leaf, &min_key, i);
5630 if (min_key.objectid != dir_elem->ino ||
5631 min_key.type != BTRFS_DIR_ITEM_KEY)
5632 goto next_dir_inode;
5634 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5635 type = btrfs_dir_type(leaf, di);
5636 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5637 type != BTRFS_FT_DIR)
5639 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5640 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5643 btrfs_release_path(path);
5644 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5645 if (IS_ERR(di_inode)) {
5646 ret = PTR_ERR(di_inode);
5647 goto next_dir_inode;
5650 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5655 ctx->log_new_dentries = false;
5656 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5657 log_mode = LOG_INODE_ALL;
5658 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5659 log_mode, 0, LLONG_MAX, ctx);
5661 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5665 goto next_dir_inode;
5666 if (ctx->log_new_dentries) {
5667 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5669 if (!new_dir_elem) {
5671 goto next_dir_inode;
5673 new_dir_elem->ino = di_key.objectid;
5674 list_add_tail(&new_dir_elem->list, &dir_list);
5679 ret = btrfs_next_leaf(log, path);
5681 goto next_dir_inode;
5682 } else if (ret > 0) {
5684 goto next_dir_inode;
5688 if (min_key.offset < (u64)-1) {
5693 list_del(&dir_elem->list);
5697 btrfs_free_path(path);
5701 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5702 struct btrfs_inode *inode,
5703 struct btrfs_log_ctx *ctx)
5705 struct btrfs_fs_info *fs_info = trans->fs_info;
5707 struct btrfs_path *path;
5708 struct btrfs_key key;
5709 struct btrfs_root *root = inode->root;
5710 const u64 ino = btrfs_ino(inode);
5712 path = btrfs_alloc_path();
5715 path->skip_locking = 1;
5716 path->search_commit_root = 1;
5719 key.type = BTRFS_INODE_REF_KEY;
5721 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5726 struct extent_buffer *leaf = path->nodes[0];
5727 int slot = path->slots[0];
5732 if (slot >= btrfs_header_nritems(leaf)) {
5733 ret = btrfs_next_leaf(root, path);
5741 btrfs_item_key_to_cpu(leaf, &key, slot);
5742 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5743 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5746 item_size = btrfs_item_size_nr(leaf, slot);
5747 ptr = btrfs_item_ptr_offset(leaf, slot);
5748 while (cur_offset < item_size) {
5749 struct btrfs_key inode_key;
5750 struct inode *dir_inode;
5752 inode_key.type = BTRFS_INODE_ITEM_KEY;
5753 inode_key.offset = 0;
5755 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5756 struct btrfs_inode_extref *extref;
5758 extref = (struct btrfs_inode_extref *)
5760 inode_key.objectid = btrfs_inode_extref_parent(
5762 cur_offset += sizeof(*extref);
5763 cur_offset += btrfs_inode_extref_name_len(leaf,
5766 inode_key.objectid = key.offset;
5767 cur_offset = item_size;
5770 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5773 * If the parent inode was deleted, return an error to
5774 * fallback to a transaction commit. This is to prevent
5775 * getting an inode that was moved from one parent A to
5776 * a parent B, got its former parent A deleted and then
5777 * it got fsync'ed, from existing at both parents after
5778 * a log replay (and the old parent still existing).
5785 * mv /mnt/B/bar /mnt/A/bar
5786 * mv -T /mnt/A /mnt/B
5790 * If we ignore the old parent B which got deleted,
5791 * after a log replay we would have file bar linked
5792 * at both parents and the old parent B would still
5795 if (IS_ERR(dir_inode)) {
5796 ret = PTR_ERR(dir_inode);
5801 ctx->log_new_dentries = false;
5802 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5803 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5805 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5807 if (!ret && ctx && ctx->log_new_dentries)
5808 ret = log_new_dir_dentries(trans, root,
5809 BTRFS_I(dir_inode), ctx);
5818 btrfs_free_path(path);
5823 * helper function around btrfs_log_inode to make sure newly created
5824 * parent directories also end up in the log. A minimal inode and backref
5825 * only logging is done of any parent directories that are older than
5826 * the last committed transaction
5828 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5829 struct btrfs_inode *inode,
5830 struct dentry *parent,
5834 struct btrfs_log_ctx *ctx)
5836 struct btrfs_root *root = inode->root;
5837 struct btrfs_fs_info *fs_info = root->fs_info;
5838 struct super_block *sb;
5839 struct dentry *old_parent = NULL;
5841 u64 last_committed = fs_info->last_trans_committed;
5842 bool log_dentries = false;
5843 struct btrfs_inode *orig_inode = inode;
5845 sb = inode->vfs_inode.i_sb;
5847 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5853 * The prev transaction commit doesn't complete, we need do
5854 * full commit by ourselves.
5856 if (fs_info->last_trans_log_full_commit >
5857 fs_info->last_trans_committed) {
5862 if (btrfs_root_refs(&root->root_item) == 0) {
5867 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5873 * Skip already logged inodes or inodes corresponding to tmpfiles
5874 * (since logging them is pointless, a link count of 0 means they
5875 * will never be accessible).
5877 if (btrfs_inode_in_log(inode, trans->transid) ||
5878 inode->vfs_inode.i_nlink == 0) {
5879 ret = BTRFS_NO_LOG_SYNC;
5883 ret = start_log_trans(trans, root, ctx);
5887 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5892 * for regular files, if its inode is already on disk, we don't
5893 * have to worry about the parents at all. This is because
5894 * we can use the last_unlink_trans field to record renames
5895 * and other fun in this file.
5897 if (S_ISREG(inode->vfs_inode.i_mode) &&
5898 inode->generation <= last_committed &&
5899 inode->last_unlink_trans <= last_committed) {
5904 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5905 log_dentries = true;
5908 * On unlink we must make sure all our current and old parent directory
5909 * inodes are fully logged. This is to prevent leaving dangling
5910 * directory index entries in directories that were our parents but are
5911 * not anymore. Not doing this results in old parent directory being
5912 * impossible to delete after log replay (rmdir will always fail with
5913 * error -ENOTEMPTY).
5919 * ln testdir/foo testdir/bar
5921 * unlink testdir/bar
5922 * xfs_io -c fsync testdir/foo
5924 * mount fs, triggers log replay
5926 * If we don't log the parent directory (testdir), after log replay the
5927 * directory still has an entry pointing to the file inode using the bar
5928 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5929 * the file inode has a link count of 1.
5935 * ln foo testdir/foo2
5936 * ln foo testdir/foo3
5938 * unlink testdir/foo3
5939 * xfs_io -c fsync foo
5941 * mount fs, triggers log replay
5943 * Similar as the first example, after log replay the parent directory
5944 * testdir still has an entry pointing to the inode file with name foo3
5945 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5946 * and has a link count of 2.
5948 if (inode->last_unlink_trans > last_committed) {
5949 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5955 * If a new hard link was added to the inode in the current transaction
5956 * and its link count is now greater than 1, we need to fallback to a
5957 * transaction commit, otherwise we can end up not logging all its new
5958 * parents for all the hard links. Here just from the dentry used to
5959 * fsync, we can not visit the ancestor inodes for all the other hard
5960 * links to figure out if any is new, so we fallback to a transaction
5961 * commit (instead of adding a lot of complexity of scanning a btree,
5962 * since this scenario is not a common use case).
5964 if (inode->vfs_inode.i_nlink > 1 &&
5965 inode->last_link_trans > last_committed) {
5971 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5974 inode = BTRFS_I(d_inode(parent));
5975 if (root != inode->root)
5978 if (inode->generation > last_committed) {
5979 ret = btrfs_log_inode(trans, root, inode,
5980 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5984 if (IS_ROOT(parent))
5987 parent = dget_parent(parent);
5989 old_parent = parent;
5992 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5998 btrfs_set_log_full_commit(fs_info, trans);
6003 btrfs_remove_log_ctx(root, ctx);
6004 btrfs_end_log_trans(root);
6010 * it is not safe to log dentry if the chunk root has added new
6011 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6012 * If this returns 1, you must commit the transaction to safely get your
6015 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6016 struct dentry *dentry,
6019 struct btrfs_log_ctx *ctx)
6021 struct dentry *parent = dget_parent(dentry);
6024 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6025 start, end, LOG_INODE_ALL, ctx);
6032 * should be called during mount to recover any replay any log trees
6035 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6038 struct btrfs_path *path;
6039 struct btrfs_trans_handle *trans;
6040 struct btrfs_key key;
6041 struct btrfs_key found_key;
6042 struct btrfs_key tmp_key;
6043 struct btrfs_root *log;
6044 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6045 struct walk_control wc = {
6046 .process_func = process_one_buffer,
6050 path = btrfs_alloc_path();
6054 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6056 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6057 if (IS_ERR(trans)) {
6058 ret = PTR_ERR(trans);
6065 ret = walk_log_tree(trans, log_root_tree, &wc);
6067 btrfs_handle_fs_error(fs_info, ret,
6068 "Failed to pin buffers while recovering log root tree.");
6073 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6074 key.offset = (u64)-1;
6075 key.type = BTRFS_ROOT_ITEM_KEY;
6078 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6081 btrfs_handle_fs_error(fs_info, ret,
6082 "Couldn't find tree log root.");
6086 if (path->slots[0] == 0)
6090 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6092 btrfs_release_path(path);
6093 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6096 log = btrfs_read_fs_root(log_root_tree, &found_key);
6099 btrfs_handle_fs_error(fs_info, ret,
6100 "Couldn't read tree log root.");
6104 tmp_key.objectid = found_key.offset;
6105 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6106 tmp_key.offset = (u64)-1;
6108 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
6109 if (IS_ERR(wc.replay_dest)) {
6110 ret = PTR_ERR(wc.replay_dest);
6111 free_extent_buffer(log->node);
6112 free_extent_buffer(log->commit_root);
6114 btrfs_handle_fs_error(fs_info, ret,
6115 "Couldn't read target root for tree log recovery.");
6119 wc.replay_dest->log_root = log;
6120 btrfs_record_root_in_trans(trans, wc.replay_dest);
6121 ret = walk_log_tree(trans, log, &wc);
6123 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6124 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6128 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6129 struct btrfs_root *root = wc.replay_dest;
6131 btrfs_release_path(path);
6134 * We have just replayed everything, and the highest
6135 * objectid of fs roots probably has changed in case
6136 * some inode_item's got replayed.
6138 * root->objectid_mutex is not acquired as log replay
6139 * could only happen during mount.
6141 ret = btrfs_find_highest_objectid(root,
6142 &root->highest_objectid);
6145 key.offset = found_key.offset - 1;
6146 wc.replay_dest->log_root = NULL;
6147 free_extent_buffer(log->node);
6148 free_extent_buffer(log->commit_root);
6154 if (found_key.offset == 0)
6157 btrfs_release_path(path);
6159 /* step one is to pin it all, step two is to replay just inodes */
6162 wc.process_func = replay_one_buffer;
6163 wc.stage = LOG_WALK_REPLAY_INODES;
6166 /* step three is to replay everything */
6167 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6172 btrfs_free_path(path);
6174 /* step 4: commit the transaction, which also unpins the blocks */
6175 ret = btrfs_commit_transaction(trans);
6179 free_extent_buffer(log_root_tree->node);
6180 log_root_tree->log_root = NULL;
6181 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6182 kfree(log_root_tree);
6187 btrfs_end_transaction(wc.trans);
6188 btrfs_free_path(path);
6193 * there are some corner cases where we want to force a full
6194 * commit instead of allowing a directory to be logged.
6196 * They revolve around files there were unlinked from the directory, and
6197 * this function updates the parent directory so that a full commit is
6198 * properly done if it is fsync'd later after the unlinks are done.
6200 * Must be called before the unlink operations (updates to the subvolume tree,
6201 * inodes, etc) are done.
6203 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6204 struct btrfs_inode *dir, struct btrfs_inode *inode,
6208 * when we're logging a file, if it hasn't been renamed
6209 * or unlinked, and its inode is fully committed on disk,
6210 * we don't have to worry about walking up the directory chain
6211 * to log its parents.
6213 * So, we use the last_unlink_trans field to put this transid
6214 * into the file. When the file is logged we check it and
6215 * don't log the parents if the file is fully on disk.
6217 mutex_lock(&inode->log_mutex);
6218 inode->last_unlink_trans = trans->transid;
6219 mutex_unlock(&inode->log_mutex);
6222 * if this directory was already logged any new
6223 * names for this file/dir will get recorded
6226 if (dir->logged_trans == trans->transid)
6230 * if the inode we're about to unlink was logged,
6231 * the log will be properly updated for any new names
6233 if (inode->logged_trans == trans->transid)
6237 * when renaming files across directories, if the directory
6238 * there we're unlinking from gets fsync'd later on, there's
6239 * no way to find the destination directory later and fsync it
6240 * properly. So, we have to be conservative and force commits
6241 * so the new name gets discovered.
6246 /* we can safely do the unlink without any special recording */
6250 mutex_lock(&dir->log_mutex);
6251 dir->last_unlink_trans = trans->transid;
6252 mutex_unlock(&dir->log_mutex);
6256 * Make sure that if someone attempts to fsync the parent directory of a deleted
6257 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6258 * that after replaying the log tree of the parent directory's root we will not
6259 * see the snapshot anymore and at log replay time we will not see any log tree
6260 * corresponding to the deleted snapshot's root, which could lead to replaying
6261 * it after replaying the log tree of the parent directory (which would replay
6262 * the snapshot delete operation).
6264 * Must be called before the actual snapshot destroy operation (updates to the
6265 * parent root and tree of tree roots trees, etc) are done.
6267 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6268 struct btrfs_inode *dir)
6270 mutex_lock(&dir->log_mutex);
6271 dir->last_unlink_trans = trans->transid;
6272 mutex_unlock(&dir->log_mutex);
6276 * Call this after adding a new name for a file and it will properly
6277 * update the log to reflect the new name.
6279 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6280 * true (because it's not used).
6282 * Return value depends on whether @sync_log is true or false.
6283 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6284 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6286 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6287 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6288 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6289 * committed (without attempting to sync the log).
6291 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6292 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6293 struct dentry *parent,
6294 bool sync_log, struct btrfs_log_ctx *ctx)
6296 struct btrfs_fs_info *fs_info = trans->fs_info;
6300 * this will force the logging code to walk the dentry chain
6303 if (!S_ISDIR(inode->vfs_inode.i_mode))
6304 inode->last_unlink_trans = trans->transid;
6307 * if this inode hasn't been logged and directory we're renaming it
6308 * from hasn't been logged, we don't need to log it
6310 if (inode->logged_trans <= fs_info->last_trans_committed &&
6311 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6312 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6313 BTRFS_DONT_NEED_LOG_SYNC;
6316 struct btrfs_log_ctx ctx2;
6318 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6319 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6320 LOG_INODE_EXISTS, &ctx2);
6321 if (ret == BTRFS_NO_LOG_SYNC)
6322 return BTRFS_DONT_NEED_TRANS_COMMIT;
6324 return BTRFS_NEED_TRANS_COMMIT;
6326 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6328 return BTRFS_NEED_TRANS_COMMIT;
6329 return BTRFS_DONT_NEED_TRANS_COMMIT;
6333 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6334 LOG_INODE_EXISTS, ctx);
6335 if (ret == BTRFS_NO_LOG_SYNC)
6336 return BTRFS_DONT_NEED_LOG_SYNC;
6338 return BTRFS_NEED_TRANS_COMMIT;
6340 return BTRFS_NEED_LOG_SYNC;