2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
29 #include "compression.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
40 #define LOG_OTHER_INODE 2
43 * directory trouble cases
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
52 * rename foo/some_dir foo2/some_dir
54 * fsync foo/some_dir/some_file
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
86 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
94 #define LOG_WALK_PIN_ONLY 0
95 #define LOG_WALK_REPLAY_INODES 1
96 #define LOG_WALK_REPLAY_DIR_INDEX 2
97 #define LOG_WALK_REPLAY_ALL 3
99 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
100 struct btrfs_root
*root
, struct inode
*inode
,
104 struct btrfs_log_ctx
*ctx
);
105 static int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
106 struct btrfs_root
*root
,
107 struct btrfs_path
*path
, u64 objectid
);
108 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
109 struct btrfs_root
*root
,
110 struct btrfs_root
*log
,
111 struct btrfs_path
*path
,
112 u64 dirid
, int del_all
);
115 * tree logging is a special write ahead log used to make sure that
116 * fsyncs and O_SYNCs can happen without doing full tree commits.
118 * Full tree commits are expensive because they require commonly
119 * modified blocks to be recowed, creating many dirty pages in the
120 * extent tree an 4x-6x higher write load than ext3.
122 * Instead of doing a tree commit on every fsync, we use the
123 * key ranges and transaction ids to find items for a given file or directory
124 * that have changed in this transaction. Those items are copied into
125 * a special tree (one per subvolume root), that tree is written to disk
126 * and then the fsync is considered complete.
128 * After a crash, items are copied out of the log-tree back into the
129 * subvolume tree. Any file data extents found are recorded in the extent
130 * allocation tree, and the log-tree freed.
132 * The log tree is read three times, once to pin down all the extents it is
133 * using in ram and once, once to create all the inodes logged in the tree
134 * and once to do all the other items.
138 * start a sub transaction and setup the log tree
139 * this increments the log tree writer count to make the people
140 * syncing the tree wait for us to finish
142 static int start_log_trans(struct btrfs_trans_handle
*trans
,
143 struct btrfs_root
*root
,
144 struct btrfs_log_ctx
*ctx
)
146 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
149 mutex_lock(&root
->log_mutex
);
151 if (root
->log_root
) {
152 if (btrfs_need_log_full_commit(fs_info
, trans
)) {
157 if (!root
->log_start_pid
) {
158 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS
, &root
->state
);
159 root
->log_start_pid
= current
->pid
;
160 } else if (root
->log_start_pid
!= current
->pid
) {
161 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS
, &root
->state
);
164 mutex_lock(&fs_info
->tree_log_mutex
);
165 if (!fs_info
->log_root_tree
)
166 ret
= btrfs_init_log_root_tree(trans
, fs_info
);
167 mutex_unlock(&fs_info
->tree_log_mutex
);
171 ret
= btrfs_add_log_tree(trans
, root
);
175 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS
, &root
->state
);
176 root
->log_start_pid
= current
->pid
;
179 atomic_inc(&root
->log_batch
);
180 atomic_inc(&root
->log_writers
);
182 int index
= root
->log_transid
% 2;
183 list_add_tail(&ctx
->list
, &root
->log_ctxs
[index
]);
184 ctx
->log_transid
= root
->log_transid
;
188 mutex_unlock(&root
->log_mutex
);
193 * returns 0 if there was a log transaction running and we were able
194 * to join, or returns -ENOENT if there were not transactions
197 static int join_running_log_trans(struct btrfs_root
*root
)
205 mutex_lock(&root
->log_mutex
);
206 if (root
->log_root
) {
208 atomic_inc(&root
->log_writers
);
210 mutex_unlock(&root
->log_mutex
);
215 * This either makes the current running log transaction wait
216 * until you call btrfs_end_log_trans() or it makes any future
217 * log transactions wait until you call btrfs_end_log_trans()
219 int btrfs_pin_log_trans(struct btrfs_root
*root
)
223 mutex_lock(&root
->log_mutex
);
224 atomic_inc(&root
->log_writers
);
225 mutex_unlock(&root
->log_mutex
);
230 * indicate we're done making changes to the log tree
231 * and wake up anyone waiting to do a sync
233 void btrfs_end_log_trans(struct btrfs_root
*root
)
235 if (atomic_dec_and_test(&root
->log_writers
)) {
237 * Implicit memory barrier after atomic_dec_and_test
239 if (waitqueue_active(&root
->log_writer_wait
))
240 wake_up(&root
->log_writer_wait
);
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
251 struct walk_control
{
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
267 /* pin only walk, we record which extents on disk belong to the
272 /* what stage of the replay code we're currently in */
275 /* the root we are currently replaying */
276 struct btrfs_root
*replay_dest
;
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle
*trans
;
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
286 int (*process_func
)(struct btrfs_root
*log
, struct extent_buffer
*eb
,
287 struct walk_control
*wc
, u64 gen
);
291 * process_func used to pin down extents, write them or wait on them
293 static int process_one_buffer(struct btrfs_root
*log
,
294 struct extent_buffer
*eb
,
295 struct walk_control
*wc
, u64 gen
)
297 struct btrfs_fs_info
*fs_info
= log
->fs_info
;
301 * If this fs is mixed then we need to be able to process the leaves to
302 * pin down any logged extents, so we have to read the block.
304 if (btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
305 ret
= btrfs_read_buffer(eb
, gen
);
311 ret
= btrfs_pin_extent_for_log_replay(fs_info
, eb
->start
,
314 if (!ret
&& btrfs_buffer_uptodate(eb
, gen
, 0)) {
315 if (wc
->pin
&& btrfs_header_level(eb
) == 0)
316 ret
= btrfs_exclude_logged_extents(fs_info
, eb
);
318 btrfs_write_tree_block(eb
);
320 btrfs_wait_tree_block_writeback(eb
);
326 * Item overwrite used by replay and tree logging. eb, slot and key all refer
327 * to the src data we are copying out.
329 * root is the tree we are copying into, and path is a scratch
330 * path for use in this function (it should be released on entry and
331 * will be released on exit).
333 * If the key is already in the destination tree the existing item is
334 * overwritten. If the existing item isn't big enough, it is extended.
335 * If it is too large, it is truncated.
337 * If the key isn't in the destination yet, a new item is inserted.
339 static noinline
int overwrite_item(struct btrfs_trans_handle
*trans
,
340 struct btrfs_root
*root
,
341 struct btrfs_path
*path
,
342 struct extent_buffer
*eb
, int slot
,
343 struct btrfs_key
*key
)
345 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
348 u64 saved_i_size
= 0;
349 int save_old_i_size
= 0;
350 unsigned long src_ptr
;
351 unsigned long dst_ptr
;
352 int overwrite_root
= 0;
353 bool inode_item
= key
->type
== BTRFS_INODE_ITEM_KEY
;
355 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
358 item_size
= btrfs_item_size_nr(eb
, slot
);
359 src_ptr
= btrfs_item_ptr_offset(eb
, slot
);
361 /* look for the key in the destination tree */
362 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
369 u32 dst_size
= btrfs_item_size_nr(path
->nodes
[0],
371 if (dst_size
!= item_size
)
374 if (item_size
== 0) {
375 btrfs_release_path(path
);
378 dst_copy
= kmalloc(item_size
, GFP_NOFS
);
379 src_copy
= kmalloc(item_size
, GFP_NOFS
);
380 if (!dst_copy
|| !src_copy
) {
381 btrfs_release_path(path
);
387 read_extent_buffer(eb
, src_copy
, src_ptr
, item_size
);
389 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
390 read_extent_buffer(path
->nodes
[0], dst_copy
, dst_ptr
,
392 ret
= memcmp(dst_copy
, src_copy
, item_size
);
397 * they have the same contents, just return, this saves
398 * us from cowing blocks in the destination tree and doing
399 * extra writes that may not have been done by a previous
403 btrfs_release_path(path
);
408 * We need to load the old nbytes into the inode so when we
409 * replay the extents we've logged we get the right nbytes.
412 struct btrfs_inode_item
*item
;
416 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
417 struct btrfs_inode_item
);
418 nbytes
= btrfs_inode_nbytes(path
->nodes
[0], item
);
419 item
= btrfs_item_ptr(eb
, slot
,
420 struct btrfs_inode_item
);
421 btrfs_set_inode_nbytes(eb
, item
, nbytes
);
424 * If this is a directory we need to reset the i_size to
425 * 0 so that we can set it up properly when replaying
426 * the rest of the items in this log.
428 mode
= btrfs_inode_mode(eb
, item
);
430 btrfs_set_inode_size(eb
, item
, 0);
432 } else if (inode_item
) {
433 struct btrfs_inode_item
*item
;
437 * New inode, set nbytes to 0 so that the nbytes comes out
438 * properly when we replay the extents.
440 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
441 btrfs_set_inode_nbytes(eb
, item
, 0);
444 * If this is a directory we need to reset the i_size to 0 so
445 * that we can set it up properly when replaying the rest of
446 * the items in this log.
448 mode
= btrfs_inode_mode(eb
, item
);
450 btrfs_set_inode_size(eb
, item
, 0);
453 btrfs_release_path(path
);
454 /* try to insert the key into the destination tree */
455 path
->skip_release_on_error
= 1;
456 ret
= btrfs_insert_empty_item(trans
, root
, path
,
458 path
->skip_release_on_error
= 0;
460 /* make sure any existing item is the correct size */
461 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
) {
463 found_size
= btrfs_item_size_nr(path
->nodes
[0],
465 if (found_size
> item_size
)
466 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
467 else if (found_size
< item_size
)
468 btrfs_extend_item(fs_info
, path
,
469 item_size
- found_size
);
473 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0],
476 /* don't overwrite an existing inode if the generation number
477 * was logged as zero. This is done when the tree logging code
478 * is just logging an inode to make sure it exists after recovery.
480 * Also, don't overwrite i_size on directories during replay.
481 * log replay inserts and removes directory items based on the
482 * state of the tree found in the subvolume, and i_size is modified
485 if (key
->type
== BTRFS_INODE_ITEM_KEY
&& ret
== -EEXIST
) {
486 struct btrfs_inode_item
*src_item
;
487 struct btrfs_inode_item
*dst_item
;
489 src_item
= (struct btrfs_inode_item
*)src_ptr
;
490 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
492 if (btrfs_inode_generation(eb
, src_item
) == 0) {
493 struct extent_buffer
*dst_eb
= path
->nodes
[0];
494 const u64 ino_size
= btrfs_inode_size(eb
, src_item
);
497 * For regular files an ino_size == 0 is used only when
498 * logging that an inode exists, as part of a directory
499 * fsync, and the inode wasn't fsynced before. In this
500 * case don't set the size of the inode in the fs/subvol
501 * tree, otherwise we would be throwing valid data away.
503 if (S_ISREG(btrfs_inode_mode(eb
, src_item
)) &&
504 S_ISREG(btrfs_inode_mode(dst_eb
, dst_item
)) &&
506 struct btrfs_map_token token
;
508 btrfs_init_map_token(&token
);
509 btrfs_set_token_inode_size(dst_eb
, dst_item
,
515 if (overwrite_root
&&
516 S_ISDIR(btrfs_inode_mode(eb
, src_item
)) &&
517 S_ISDIR(btrfs_inode_mode(path
->nodes
[0], dst_item
))) {
519 saved_i_size
= btrfs_inode_size(path
->nodes
[0],
524 copy_extent_buffer(path
->nodes
[0], eb
, dst_ptr
,
527 if (save_old_i_size
) {
528 struct btrfs_inode_item
*dst_item
;
529 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
530 btrfs_set_inode_size(path
->nodes
[0], dst_item
, saved_i_size
);
533 /* make sure the generation is filled in */
534 if (key
->type
== BTRFS_INODE_ITEM_KEY
) {
535 struct btrfs_inode_item
*dst_item
;
536 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
537 if (btrfs_inode_generation(path
->nodes
[0], dst_item
) == 0) {
538 btrfs_set_inode_generation(path
->nodes
[0], dst_item
,
543 btrfs_mark_buffer_dirty(path
->nodes
[0]);
544 btrfs_release_path(path
);
549 * simple helper to read an inode off the disk from a given root
550 * This can only be called for subvolume roots and not for the log
552 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
555 struct btrfs_key key
;
558 key
.objectid
= objectid
;
559 key
.type
= BTRFS_INODE_ITEM_KEY
;
561 inode
= btrfs_iget(root
->fs_info
->sb
, &key
, root
, NULL
);
564 } else if (is_bad_inode(inode
)) {
571 /* replays a single extent in 'eb' at 'slot' with 'key' into the
572 * subvolume 'root'. path is released on entry and should be released
575 * extents in the log tree have not been allocated out of the extent
576 * tree yet. So, this completes the allocation, taking a reference
577 * as required if the extent already exists or creating a new extent
578 * if it isn't in the extent allocation tree yet.
580 * The extent is inserted into the file, dropping any existing extents
581 * from the file that overlap the new one.
583 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
584 struct btrfs_root
*root
,
585 struct btrfs_path
*path
,
586 struct extent_buffer
*eb
, int slot
,
587 struct btrfs_key
*key
)
589 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
592 u64 start
= key
->offset
;
594 struct btrfs_file_extent_item
*item
;
595 struct inode
*inode
= NULL
;
599 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
600 found_type
= btrfs_file_extent_type(eb
, item
);
602 if (found_type
== BTRFS_FILE_EXTENT_REG
||
603 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
604 nbytes
= btrfs_file_extent_num_bytes(eb
, item
);
605 extent_end
= start
+ nbytes
;
608 * We don't add to the inodes nbytes if we are prealloc or a
611 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
613 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
614 size
= btrfs_file_extent_inline_len(eb
, slot
, item
);
615 nbytes
= btrfs_file_extent_ram_bytes(eb
, item
);
616 extent_end
= ALIGN(start
+ size
,
617 fs_info
->sectorsize
);
623 inode
= read_one_inode(root
, key
->objectid
);
630 * first check to see if we already have this extent in the
631 * file. This must be done before the btrfs_drop_extents run
632 * so we don't try to drop this extent.
634 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
638 (found_type
== BTRFS_FILE_EXTENT_REG
||
639 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)) {
640 struct btrfs_file_extent_item cmp1
;
641 struct btrfs_file_extent_item cmp2
;
642 struct btrfs_file_extent_item
*existing
;
643 struct extent_buffer
*leaf
;
645 leaf
= path
->nodes
[0];
646 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
647 struct btrfs_file_extent_item
);
649 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
651 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
655 * we already have a pointer to this exact extent,
656 * we don't have to do anything
658 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
659 btrfs_release_path(path
);
663 btrfs_release_path(path
);
665 /* drop any overlapping extents */
666 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, extent_end
, 1);
670 if (found_type
== BTRFS_FILE_EXTENT_REG
||
671 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
673 unsigned long dest_offset
;
674 struct btrfs_key ins
;
676 ret
= btrfs_insert_empty_item(trans
, root
, path
, key
,
680 dest_offset
= btrfs_item_ptr_offset(path
->nodes
[0],
682 copy_extent_buffer(path
->nodes
[0], eb
, dest_offset
,
683 (unsigned long)item
, sizeof(*item
));
685 ins
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
686 ins
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
687 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
688 offset
= key
->offset
- btrfs_file_extent_offset(eb
, item
);
691 * Manually record dirty extent, as here we did a shallow
692 * file extent item copy and skip normal backref update,
693 * but modifying extent tree all by ourselves.
694 * So need to manually record dirty extent for qgroup,
695 * as the owner of the file extent changed from log tree
696 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
698 ret
= btrfs_qgroup_trace_extent(trans
, fs_info
,
699 btrfs_file_extent_disk_bytenr(eb
, item
),
700 btrfs_file_extent_disk_num_bytes(eb
, item
),
705 if (ins
.objectid
> 0) {
708 LIST_HEAD(ordered_sums
);
710 * is this extent already allocated in the extent
711 * allocation tree? If so, just add a reference
713 ret
= btrfs_lookup_data_extent(fs_info
, ins
.objectid
,
716 ret
= btrfs_inc_extent_ref(trans
, fs_info
,
717 ins
.objectid
, ins
.offset
,
718 0, root
->root_key
.objectid
,
719 key
->objectid
, offset
);
724 * insert the extent pointer in the extent
727 ret
= btrfs_alloc_logged_file_extent(trans
,
729 root
->root_key
.objectid
,
730 key
->objectid
, offset
, &ins
);
734 btrfs_release_path(path
);
736 if (btrfs_file_extent_compression(eb
, item
)) {
737 csum_start
= ins
.objectid
;
738 csum_end
= csum_start
+ ins
.offset
;
740 csum_start
= ins
.objectid
+
741 btrfs_file_extent_offset(eb
, item
);
742 csum_end
= csum_start
+
743 btrfs_file_extent_num_bytes(eb
, item
);
746 ret
= btrfs_lookup_csums_range(root
->log_root
,
747 csum_start
, csum_end
- 1,
752 * Now delete all existing cums in the csum root that
753 * cover our range. We do this because we can have an
754 * extent that is completely referenced by one file
755 * extent item and partially referenced by another
756 * file extent item (like after using the clone or
757 * extent_same ioctls). In this case if we end up doing
758 * the replay of the one that partially references the
759 * extent first, and we do not do the csum deletion
760 * below, we can get 2 csum items in the csum tree that
761 * overlap each other. For example, imagine our log has
762 * the two following file extent items:
764 * key (257 EXTENT_DATA 409600)
765 * extent data disk byte 12845056 nr 102400
766 * extent data offset 20480 nr 20480 ram 102400
768 * key (257 EXTENT_DATA 819200)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 0 nr 102400 ram 102400
772 * Where the second one fully references the 100K extent
773 * that starts at disk byte 12845056, and the log tree
774 * has a single csum item that covers the entire range
777 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
779 * After the first file extent item is replayed, the
780 * csum tree gets the following csum item:
782 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
784 * Which covers the 20K sub-range starting at offset 20K
785 * of our extent. Now when we replay the second file
786 * extent item, if we do not delete existing csum items
787 * that cover any of its blocks, we end up getting two
788 * csum items in our csum tree that overlap each other:
790 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
791 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
793 * Which is a problem, because after this anyone trying
794 * to lookup up for the checksum of any block of our
795 * extent starting at an offset of 40K or higher, will
796 * end up looking at the second csum item only, which
797 * does not contain the checksum for any block starting
798 * at offset 40K or higher of our extent.
800 while (!list_empty(&ordered_sums
)) {
801 struct btrfs_ordered_sum
*sums
;
802 sums
= list_entry(ordered_sums
.next
,
803 struct btrfs_ordered_sum
,
806 ret
= btrfs_del_csums(trans
, fs_info
,
810 ret
= btrfs_csum_file_blocks(trans
,
811 fs_info
->csum_root
, sums
);
812 list_del(&sums
->list
);
818 btrfs_release_path(path
);
820 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
821 /* inline extents are easy, we just overwrite them */
822 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
827 inode_add_bytes(inode
, nbytes
);
828 ret
= btrfs_update_inode(trans
, root
, inode
);
836 * when cleaning up conflicts between the directory names in the
837 * subvolume, directory names in the log and directory names in the
838 * inode back references, we may have to unlink inodes from directories.
840 * This is a helper function to do the unlink of a specific directory
843 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
844 struct btrfs_root
*root
,
845 struct btrfs_path
*path
,
847 struct btrfs_dir_item
*di
)
849 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
853 struct extent_buffer
*leaf
;
854 struct btrfs_key location
;
857 leaf
= path
->nodes
[0];
859 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
860 name_len
= btrfs_dir_name_len(leaf
, di
);
861 name
= kmalloc(name_len
, GFP_NOFS
);
865 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
866 btrfs_release_path(path
);
868 inode
= read_one_inode(root
, location
.objectid
);
874 ret
= link_to_fixup_dir(trans
, root
, path
, location
.objectid
);
878 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
882 ret
= btrfs_run_delayed_items(trans
, fs_info
);
890 * helper function to see if a given name and sequence number found
891 * in an inode back reference are already in a directory and correctly
892 * point to this inode
894 static noinline
int inode_in_dir(struct btrfs_root
*root
,
895 struct btrfs_path
*path
,
896 u64 dirid
, u64 objectid
, u64 index
,
897 const char *name
, int name_len
)
899 struct btrfs_dir_item
*di
;
900 struct btrfs_key location
;
903 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
904 index
, 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
)
911 btrfs_release_path(path
);
913 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
914 if (di
&& !IS_ERR(di
)) {
915 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
916 if (location
.objectid
!= objectid
)
922 btrfs_release_path(path
);
927 * helper function to check a log tree for a named back reference in
928 * an inode. This is used to decide if a back reference that is
929 * found in the subvolume conflicts with what we find in the log.
931 * inode backreferences may have multiple refs in a single item,
932 * during replay we process one reference at a time, and we don't
933 * want to delete valid links to a file from the subvolume if that
934 * link is also in the log.
936 static noinline
int backref_in_log(struct btrfs_root
*log
,
937 struct btrfs_key
*key
,
939 const char *name
, int namelen
)
941 struct btrfs_path
*path
;
942 struct btrfs_inode_ref
*ref
;
944 unsigned long ptr_end
;
945 unsigned long name_ptr
;
951 path
= btrfs_alloc_path();
955 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
959 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
961 if (key
->type
== BTRFS_INODE_EXTREF_KEY
) {
962 if (btrfs_find_name_in_ext_backref(path
, ref_objectid
,
963 name
, namelen
, NULL
))
969 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
970 ptr_end
= ptr
+ item_size
;
971 while (ptr
< ptr_end
) {
972 ref
= (struct btrfs_inode_ref
*)ptr
;
973 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
974 if (found_name_len
== namelen
) {
975 name_ptr
= (unsigned long)(ref
+ 1);
976 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
983 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
986 btrfs_free_path(path
);
990 static inline int __add_inode_ref(struct btrfs_trans_handle
*trans
,
991 struct btrfs_root
*root
,
992 struct btrfs_path
*path
,
993 struct btrfs_root
*log_root
,
994 struct inode
*dir
, struct inode
*inode
,
995 struct extent_buffer
*eb
,
996 u64 inode_objectid
, u64 parent_objectid
,
997 u64 ref_index
, char *name
, int namelen
,
1000 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1003 int victim_name_len
;
1004 struct extent_buffer
*leaf
;
1005 struct btrfs_dir_item
*di
;
1006 struct btrfs_key search_key
;
1007 struct btrfs_inode_extref
*extref
;
1010 /* Search old style refs */
1011 search_key
.objectid
= inode_objectid
;
1012 search_key
.type
= BTRFS_INODE_REF_KEY
;
1013 search_key
.offset
= parent_objectid
;
1014 ret
= btrfs_search_slot(NULL
, root
, &search_key
, path
, 0, 0);
1016 struct btrfs_inode_ref
*victim_ref
;
1018 unsigned long ptr_end
;
1020 leaf
= path
->nodes
[0];
1022 /* are we trying to overwrite a back ref for the root directory
1023 * if so, just jump out, we're done
1025 if (search_key
.objectid
== search_key
.offset
)
1028 /* check all the names in this back reference to see
1029 * if they are in the log. if so, we allow them to stay
1030 * otherwise they must be unlinked as a conflict
1032 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1033 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1034 while (ptr
< ptr_end
) {
1035 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
1036 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
1038 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
1042 read_extent_buffer(leaf
, victim_name
,
1043 (unsigned long)(victim_ref
+ 1),
1046 if (!backref_in_log(log_root
, &search_key
,
1051 btrfs_release_path(path
);
1053 ret
= btrfs_unlink_inode(trans
, root
, dir
,
1059 ret
= btrfs_run_delayed_items(trans
, fs_info
);
1067 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
1071 * NOTE: we have searched root tree and checked the
1072 * corresponding ref, it does not need to check again.
1076 btrfs_release_path(path
);
1078 /* Same search but for extended refs */
1079 extref
= btrfs_lookup_inode_extref(NULL
, root
, path
, name
, namelen
,
1080 inode_objectid
, parent_objectid
, 0,
1082 if (!IS_ERR_OR_NULL(extref
)) {
1086 struct inode
*victim_parent
;
1088 leaf
= path
->nodes
[0];
1090 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1091 base
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1093 while (cur_offset
< item_size
) {
1094 extref
= (struct btrfs_inode_extref
*)(base
+ cur_offset
);
1096 victim_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
1098 if (btrfs_inode_extref_parent(leaf
, extref
) != parent_objectid
)
1101 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
1104 read_extent_buffer(leaf
, victim_name
, (unsigned long)&extref
->name
,
1107 search_key
.objectid
= inode_objectid
;
1108 search_key
.type
= BTRFS_INODE_EXTREF_KEY
;
1109 search_key
.offset
= btrfs_extref_hash(parent_objectid
,
1113 if (!backref_in_log(log_root
, &search_key
,
1114 parent_objectid
, victim_name
,
1117 victim_parent
= read_one_inode(root
,
1119 if (victim_parent
) {
1121 btrfs_release_path(path
);
1123 ret
= btrfs_unlink_inode(trans
, root
,
1129 ret
= btrfs_run_delayed_items(
1133 iput(victim_parent
);
1144 cur_offset
+= victim_name_len
+ sizeof(*extref
);
1148 btrfs_release_path(path
);
1150 /* look for a conflicting sequence number */
1151 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, btrfs_ino(dir
),
1152 ref_index
, name
, namelen
, 0);
1153 if (di
&& !IS_ERR(di
)) {
1154 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
1158 btrfs_release_path(path
);
1160 /* look for a conflicing name */
1161 di
= btrfs_lookup_dir_item(trans
, root
, path
, btrfs_ino(dir
),
1163 if (di
&& !IS_ERR(di
)) {
1164 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
1168 btrfs_release_path(path
);
1173 static int extref_get_fields(struct extent_buffer
*eb
, unsigned long ref_ptr
,
1174 u32
*namelen
, char **name
, u64
*index
,
1175 u64
*parent_objectid
)
1177 struct btrfs_inode_extref
*extref
;
1179 extref
= (struct btrfs_inode_extref
*)ref_ptr
;
1181 *namelen
= btrfs_inode_extref_name_len(eb
, extref
);
1182 *name
= kmalloc(*namelen
, GFP_NOFS
);
1186 read_extent_buffer(eb
, *name
, (unsigned long)&extref
->name
,
1189 *index
= btrfs_inode_extref_index(eb
, extref
);
1190 if (parent_objectid
)
1191 *parent_objectid
= btrfs_inode_extref_parent(eb
, extref
);
1196 static int ref_get_fields(struct extent_buffer
*eb
, unsigned long ref_ptr
,
1197 u32
*namelen
, char **name
, u64
*index
)
1199 struct btrfs_inode_ref
*ref
;
1201 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
1203 *namelen
= btrfs_inode_ref_name_len(eb
, ref
);
1204 *name
= kmalloc(*namelen
, GFP_NOFS
);
1208 read_extent_buffer(eb
, *name
, (unsigned long)(ref
+ 1), *namelen
);
1210 *index
= btrfs_inode_ref_index(eb
, ref
);
1216 * replay one inode back reference item found in the log tree.
1217 * eb, slot and key refer to the buffer and key found in the log tree.
1218 * root is the destination we are replaying into, and path is for temp
1219 * use by this function. (it should be released on return).
1221 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
1222 struct btrfs_root
*root
,
1223 struct btrfs_root
*log
,
1224 struct btrfs_path
*path
,
1225 struct extent_buffer
*eb
, int slot
,
1226 struct btrfs_key
*key
)
1228 struct inode
*dir
= NULL
;
1229 struct inode
*inode
= NULL
;
1230 unsigned long ref_ptr
;
1231 unsigned long ref_end
;
1235 int search_done
= 0;
1236 int log_ref_ver
= 0;
1237 u64 parent_objectid
;
1240 int ref_struct_size
;
1242 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
1243 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
1245 if (key
->type
== BTRFS_INODE_EXTREF_KEY
) {
1246 struct btrfs_inode_extref
*r
;
1248 ref_struct_size
= sizeof(struct btrfs_inode_extref
);
1250 r
= (struct btrfs_inode_extref
*)ref_ptr
;
1251 parent_objectid
= btrfs_inode_extref_parent(eb
, r
);
1253 ref_struct_size
= sizeof(struct btrfs_inode_ref
);
1254 parent_objectid
= key
->offset
;
1256 inode_objectid
= key
->objectid
;
1259 * it is possible that we didn't log all the parent directories
1260 * for a given inode. If we don't find the dir, just don't
1261 * copy the back ref in. The link count fixup code will take
1264 dir
= read_one_inode(root
, parent_objectid
);
1270 inode
= read_one_inode(root
, inode_objectid
);
1276 while (ref_ptr
< ref_end
) {
1278 ret
= extref_get_fields(eb
, ref_ptr
, &namelen
, &name
,
1279 &ref_index
, &parent_objectid
);
1281 * parent object can change from one array
1285 dir
= read_one_inode(root
, parent_objectid
);
1291 ret
= ref_get_fields(eb
, ref_ptr
, &namelen
, &name
,
1297 /* if we already have a perfect match, we're done */
1298 if (!inode_in_dir(root
, path
, btrfs_ino(dir
), btrfs_ino(inode
),
1299 ref_index
, name
, namelen
)) {
1301 * look for a conflicting back reference in the
1302 * metadata. if we find one we have to unlink that name
1303 * of the file before we add our new link. Later on, we
1304 * overwrite any existing back reference, and we don't
1305 * want to create dangling pointers in the directory.
1309 ret
= __add_inode_ref(trans
, root
, path
, log
,
1313 ref_index
, name
, namelen
,
1322 /* insert our name */
1323 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
,
1328 btrfs_update_inode(trans
, root
, inode
);
1331 ref_ptr
= (unsigned long)(ref_ptr
+ ref_struct_size
) + namelen
;
1340 /* finally write the back reference in the inode */
1341 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
1343 btrfs_release_path(path
);
1350 static int insert_orphan_item(struct btrfs_trans_handle
*trans
,
1351 struct btrfs_root
*root
, u64 ino
)
1355 ret
= btrfs_insert_orphan_item(trans
, root
, ino
);
1362 static int count_inode_extrefs(struct btrfs_root
*root
,
1363 struct inode
*inode
, struct btrfs_path
*path
)
1367 unsigned int nlink
= 0;
1370 u64 inode_objectid
= btrfs_ino(inode
);
1373 struct btrfs_inode_extref
*extref
;
1374 struct extent_buffer
*leaf
;
1377 ret
= btrfs_find_one_extref(root
, inode_objectid
, offset
, path
,
1382 leaf
= path
->nodes
[0];
1383 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1384 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1387 while (cur_offset
< item_size
) {
1388 extref
= (struct btrfs_inode_extref
*) (ptr
+ cur_offset
);
1389 name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
1393 cur_offset
+= name_len
+ sizeof(*extref
);
1397 btrfs_release_path(path
);
1399 btrfs_release_path(path
);
1401 if (ret
< 0 && ret
!= -ENOENT
)
1406 static int count_inode_refs(struct btrfs_root
*root
,
1407 struct inode
*inode
, struct btrfs_path
*path
)
1410 struct btrfs_key key
;
1411 unsigned int nlink
= 0;
1413 unsigned long ptr_end
;
1415 u64 ino
= btrfs_ino(inode
);
1418 key
.type
= BTRFS_INODE_REF_KEY
;
1419 key
.offset
= (u64
)-1;
1422 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1426 if (path
->slots
[0] == 0)
1431 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1433 if (key
.objectid
!= ino
||
1434 key
.type
!= BTRFS_INODE_REF_KEY
)
1436 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1437 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1439 while (ptr
< ptr_end
) {
1440 struct btrfs_inode_ref
*ref
;
1442 ref
= (struct btrfs_inode_ref
*)ptr
;
1443 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1445 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1449 if (key
.offset
== 0)
1451 if (path
->slots
[0] > 0) {
1456 btrfs_release_path(path
);
1458 btrfs_release_path(path
);
1464 * There are a few corners where the link count of the file can't
1465 * be properly maintained during replay. So, instead of adding
1466 * lots of complexity to the log code, we just scan the backrefs
1467 * for any file that has been through replay.
1469 * The scan will update the link count on the inode to reflect the
1470 * number of back refs found. If it goes down to zero, the iput
1471 * will free the inode.
1473 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
1474 struct btrfs_root
*root
,
1475 struct inode
*inode
)
1477 struct btrfs_path
*path
;
1480 u64 ino
= btrfs_ino(inode
);
1482 path
= btrfs_alloc_path();
1486 ret
= count_inode_refs(root
, inode
, path
);
1492 ret
= count_inode_extrefs(root
, inode
, path
);
1500 if (nlink
!= inode
->i_nlink
) {
1501 set_nlink(inode
, nlink
);
1502 btrfs_update_inode(trans
, root
, inode
);
1504 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1506 if (inode
->i_nlink
== 0) {
1507 if (S_ISDIR(inode
->i_mode
)) {
1508 ret
= replay_dir_deletes(trans
, root
, NULL
, path
,
1513 ret
= insert_orphan_item(trans
, root
, ino
);
1517 btrfs_free_path(path
);
1521 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1522 struct btrfs_root
*root
,
1523 struct btrfs_path
*path
)
1526 struct btrfs_key key
;
1527 struct inode
*inode
;
1529 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1530 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1531 key
.offset
= (u64
)-1;
1533 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1538 if (path
->slots
[0] == 0)
1543 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1544 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1545 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1548 ret
= btrfs_del_item(trans
, root
, path
);
1552 btrfs_release_path(path
);
1553 inode
= read_one_inode(root
, key
.offset
);
1557 ret
= fixup_inode_link_count(trans
, root
, inode
);
1563 * fixup on a directory may create new entries,
1564 * make sure we always look for the highset possible
1567 key
.offset
= (u64
)-1;
1571 btrfs_release_path(path
);
1577 * record a given inode in the fixup dir so we can check its link
1578 * count when replay is done. The link count is incremented here
1579 * so the inode won't go away until we check it
1581 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1582 struct btrfs_root
*root
,
1583 struct btrfs_path
*path
,
1586 struct btrfs_key key
;
1588 struct inode
*inode
;
1590 inode
= read_one_inode(root
, objectid
);
1594 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1595 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1596 key
.offset
= objectid
;
1598 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1600 btrfs_release_path(path
);
1602 if (!inode
->i_nlink
)
1603 set_nlink(inode
, 1);
1606 ret
= btrfs_update_inode(trans
, root
, inode
);
1607 } else if (ret
== -EEXIST
) {
1610 BUG(); /* Logic Error */
1618 * when replaying the log for a directory, we only insert names
1619 * for inodes that actually exist. This means an fsync on a directory
1620 * does not implicitly fsync all the new files in it
1622 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1623 struct btrfs_root
*root
,
1624 u64 dirid
, u64 index
,
1625 char *name
, int name_len
,
1626 struct btrfs_key
*location
)
1628 struct inode
*inode
;
1632 inode
= read_one_inode(root
, location
->objectid
);
1636 dir
= read_one_inode(root
, dirid
);
1642 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1644 /* FIXME, put inode into FIXUP list */
1652 * Return true if an inode reference exists in the log for the given name,
1653 * inode and parent inode.
1655 static bool name_in_log_ref(struct btrfs_root
*log_root
,
1656 const char *name
, const int name_len
,
1657 const u64 dirid
, const u64 ino
)
1659 struct btrfs_key search_key
;
1661 search_key
.objectid
= ino
;
1662 search_key
.type
= BTRFS_INODE_REF_KEY
;
1663 search_key
.offset
= dirid
;
1664 if (backref_in_log(log_root
, &search_key
, dirid
, name
, name_len
))
1667 search_key
.type
= BTRFS_INODE_EXTREF_KEY
;
1668 search_key
.offset
= btrfs_extref_hash(dirid
, name
, name_len
);
1669 if (backref_in_log(log_root
, &search_key
, dirid
, name
, name_len
))
1676 * take a single entry in a log directory item and replay it into
1679 * if a conflicting item exists in the subdirectory already,
1680 * the inode it points to is unlinked and put into the link count
1683 * If a name from the log points to a file or directory that does
1684 * not exist in the FS, it is skipped. fsyncs on directories
1685 * do not force down inodes inside that directory, just changes to the
1686 * names or unlinks in a directory.
1688 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1689 * non-existing inode) and 1 if the name was replayed.
1691 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1692 struct btrfs_root
*root
,
1693 struct btrfs_path
*path
,
1694 struct extent_buffer
*eb
,
1695 struct btrfs_dir_item
*di
,
1696 struct btrfs_key
*key
)
1700 struct btrfs_dir_item
*dst_di
;
1701 struct btrfs_key found_key
;
1702 struct btrfs_key log_key
;
1707 bool update_size
= (key
->type
== BTRFS_DIR_INDEX_KEY
);
1708 bool name_added
= false;
1710 dir
= read_one_inode(root
, key
->objectid
);
1714 name_len
= btrfs_dir_name_len(eb
, di
);
1715 name
= kmalloc(name_len
, GFP_NOFS
);
1721 log_type
= btrfs_dir_type(eb
, di
);
1722 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1725 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1726 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1731 btrfs_release_path(path
);
1733 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1734 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1736 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1737 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1746 if (IS_ERR_OR_NULL(dst_di
)) {
1747 /* we need a sequence number to insert, so we only
1748 * do inserts for the BTRFS_DIR_INDEX_KEY types
1750 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1755 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1756 /* the existing item matches the logged item */
1757 if (found_key
.objectid
== log_key
.objectid
&&
1758 found_key
.type
== log_key
.type
&&
1759 found_key
.offset
== log_key
.offset
&&
1760 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1761 update_size
= false;
1766 * don't drop the conflicting directory entry if the inode
1767 * for the new entry doesn't exist
1772 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1776 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1779 btrfs_release_path(path
);
1780 if (!ret
&& update_size
) {
1781 btrfs_i_size_write(dir
, dir
->i_size
+ name_len
* 2);
1782 ret
= btrfs_update_inode(trans
, root
, dir
);
1786 if (!ret
&& name_added
)
1791 if (name_in_log_ref(root
->log_root
, name
, name_len
,
1792 key
->objectid
, log_key
.objectid
)) {
1793 /* The dentry will be added later. */
1795 update_size
= false;
1798 btrfs_release_path(path
);
1799 ret
= insert_one_name(trans
, root
, key
->objectid
, key
->offset
,
1800 name
, name_len
, &log_key
);
1801 if (ret
&& ret
!= -ENOENT
&& ret
!= -EEXIST
)
1805 update_size
= false;
1811 * find all the names in a directory item and reconcile them into
1812 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1813 * one name in a directory item, but the same code gets used for
1814 * both directory index types
1816 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1817 struct btrfs_root
*root
,
1818 struct btrfs_path
*path
,
1819 struct extent_buffer
*eb
, int slot
,
1820 struct btrfs_key
*key
)
1822 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1824 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1825 struct btrfs_dir_item
*di
;
1828 unsigned long ptr_end
;
1829 struct btrfs_path
*fixup_path
= NULL
;
1831 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1832 ptr_end
= ptr
+ item_size
;
1833 while (ptr
< ptr_end
) {
1834 di
= (struct btrfs_dir_item
*)ptr
;
1835 if (verify_dir_item(fs_info
, eb
, di
))
1837 name_len
= btrfs_dir_name_len(eb
, di
);
1838 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1841 ptr
= (unsigned long)(di
+ 1);
1845 * If this entry refers to a non-directory (directories can not
1846 * have a link count > 1) and it was added in the transaction
1847 * that was not committed, make sure we fixup the link count of
1848 * the inode it the entry points to. Otherwise something like
1849 * the following would result in a directory pointing to an
1850 * inode with a wrong link that does not account for this dir
1858 * ln testdir/bar testdir/bar_link
1859 * ln testdir/foo testdir/foo_link
1860 * xfs_io -c "fsync" testdir/bar
1864 * mount fs, log replay happens
1866 * File foo would remain with a link count of 1 when it has two
1867 * entries pointing to it in the directory testdir. This would
1868 * make it impossible to ever delete the parent directory has
1869 * it would result in stale dentries that can never be deleted.
1871 if (ret
== 1 && btrfs_dir_type(eb
, di
) != BTRFS_FT_DIR
) {
1872 struct btrfs_key di_key
;
1875 fixup_path
= btrfs_alloc_path();
1882 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
1883 ret
= link_to_fixup_dir(trans
, root
, fixup_path
,
1890 btrfs_free_path(fixup_path
);
1895 * directory replay has two parts. There are the standard directory
1896 * items in the log copied from the subvolume, and range items
1897 * created in the log while the subvolume was logged.
1899 * The range items tell us which parts of the key space the log
1900 * is authoritative for. During replay, if a key in the subvolume
1901 * directory is in a logged range item, but not actually in the log
1902 * that means it was deleted from the directory before the fsync
1903 * and should be removed.
1905 static noinline
int find_dir_range(struct btrfs_root
*root
,
1906 struct btrfs_path
*path
,
1907 u64 dirid
, int key_type
,
1908 u64
*start_ret
, u64
*end_ret
)
1910 struct btrfs_key key
;
1912 struct btrfs_dir_log_item
*item
;
1916 if (*start_ret
== (u64
)-1)
1919 key
.objectid
= dirid
;
1920 key
.type
= key_type
;
1921 key
.offset
= *start_ret
;
1923 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1927 if (path
->slots
[0] == 0)
1932 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1934 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1938 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1939 struct btrfs_dir_log_item
);
1940 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1942 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1944 *start_ret
= key
.offset
;
1945 *end_ret
= found_end
;
1950 /* check the next slot in the tree to see if it is a valid item */
1951 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1953 if (path
->slots
[0] >= nritems
) {
1954 ret
= btrfs_next_leaf(root
, path
);
1959 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1961 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1965 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1966 struct btrfs_dir_log_item
);
1967 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1968 *start_ret
= key
.offset
;
1969 *end_ret
= found_end
;
1972 btrfs_release_path(path
);
1977 * this looks for a given directory item in the log. If the directory
1978 * item is not in the log, the item is removed and the inode it points
1981 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1982 struct btrfs_root
*root
,
1983 struct btrfs_root
*log
,
1984 struct btrfs_path
*path
,
1985 struct btrfs_path
*log_path
,
1987 struct btrfs_key
*dir_key
)
1989 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1991 struct extent_buffer
*eb
;
1994 struct btrfs_dir_item
*di
;
1995 struct btrfs_dir_item
*log_di
;
1998 unsigned long ptr_end
;
2000 struct inode
*inode
;
2001 struct btrfs_key location
;
2004 eb
= path
->nodes
[0];
2005 slot
= path
->slots
[0];
2006 item_size
= btrfs_item_size_nr(eb
, slot
);
2007 ptr
= btrfs_item_ptr_offset(eb
, slot
);
2008 ptr_end
= ptr
+ item_size
;
2009 while (ptr
< ptr_end
) {
2010 di
= (struct btrfs_dir_item
*)ptr
;
2011 if (verify_dir_item(fs_info
, eb
, di
)) {
2016 name_len
= btrfs_dir_name_len(eb
, di
);
2017 name
= kmalloc(name_len
, GFP_NOFS
);
2022 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
2025 if (log
&& dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
2026 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
2029 } else if (log
&& dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
2030 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
2036 if (!log_di
|| (IS_ERR(log_di
) && PTR_ERR(log_di
) == -ENOENT
)) {
2037 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
2038 btrfs_release_path(path
);
2039 btrfs_release_path(log_path
);
2040 inode
= read_one_inode(root
, location
.objectid
);
2046 ret
= link_to_fixup_dir(trans
, root
,
2047 path
, location
.objectid
);
2055 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
2058 ret
= btrfs_run_delayed_items(trans
, fs_info
);
2064 /* there might still be more names under this key
2065 * check and repeat if required
2067 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
2073 } else if (IS_ERR(log_di
)) {
2075 return PTR_ERR(log_di
);
2077 btrfs_release_path(log_path
);
2080 ptr
= (unsigned long)(di
+ 1);
2085 btrfs_release_path(path
);
2086 btrfs_release_path(log_path
);
2090 static int replay_xattr_deletes(struct btrfs_trans_handle
*trans
,
2091 struct btrfs_root
*root
,
2092 struct btrfs_root
*log
,
2093 struct btrfs_path
*path
,
2096 struct btrfs_key search_key
;
2097 struct btrfs_path
*log_path
;
2102 log_path
= btrfs_alloc_path();
2106 search_key
.objectid
= ino
;
2107 search_key
.type
= BTRFS_XATTR_ITEM_KEY
;
2108 search_key
.offset
= 0;
2110 ret
= btrfs_search_slot(NULL
, root
, &search_key
, path
, 0, 0);
2114 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2115 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2116 struct btrfs_key key
;
2117 struct btrfs_dir_item
*di
;
2118 struct btrfs_dir_item
*log_di
;
2122 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, i
);
2123 if (key
.objectid
!= ino
|| key
.type
!= BTRFS_XATTR_ITEM_KEY
) {
2128 di
= btrfs_item_ptr(path
->nodes
[0], i
, struct btrfs_dir_item
);
2129 total_size
= btrfs_item_size_nr(path
->nodes
[0], i
);
2131 while (cur
< total_size
) {
2132 u16 name_len
= btrfs_dir_name_len(path
->nodes
[0], di
);
2133 u16 data_len
= btrfs_dir_data_len(path
->nodes
[0], di
);
2134 u32 this_len
= sizeof(*di
) + name_len
+ data_len
;
2137 name
= kmalloc(name_len
, GFP_NOFS
);
2142 read_extent_buffer(path
->nodes
[0], name
,
2143 (unsigned long)(di
+ 1), name_len
);
2145 log_di
= btrfs_lookup_xattr(NULL
, log
, log_path
, ino
,
2147 btrfs_release_path(log_path
);
2149 /* Doesn't exist in log tree, so delete it. */
2150 btrfs_release_path(path
);
2151 di
= btrfs_lookup_xattr(trans
, root
, path
, ino
,
2152 name
, name_len
, -1);
2159 ret
= btrfs_delete_one_dir_name(trans
, root
,
2163 btrfs_release_path(path
);
2168 if (IS_ERR(log_di
)) {
2169 ret
= PTR_ERR(log_di
);
2173 di
= (struct btrfs_dir_item
*)((char *)di
+ this_len
);
2176 ret
= btrfs_next_leaf(root
, path
);
2182 btrfs_free_path(log_path
);
2183 btrfs_release_path(path
);
2189 * deletion replay happens before we copy any new directory items
2190 * out of the log or out of backreferences from inodes. It
2191 * scans the log to find ranges of keys that log is authoritative for,
2192 * and then scans the directory to find items in those ranges that are
2193 * not present in the log.
2195 * Anything we don't find in the log is unlinked and removed from the
2198 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
2199 struct btrfs_root
*root
,
2200 struct btrfs_root
*log
,
2201 struct btrfs_path
*path
,
2202 u64 dirid
, int del_all
)
2206 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
2208 struct btrfs_key dir_key
;
2209 struct btrfs_key found_key
;
2210 struct btrfs_path
*log_path
;
2213 dir_key
.objectid
= dirid
;
2214 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
2215 log_path
= btrfs_alloc_path();
2219 dir
= read_one_inode(root
, dirid
);
2220 /* it isn't an error if the inode isn't there, that can happen
2221 * because we replay the deletes before we copy in the inode item
2225 btrfs_free_path(log_path
);
2233 range_end
= (u64
)-1;
2235 ret
= find_dir_range(log
, path
, dirid
, key_type
,
2236 &range_start
, &range_end
);
2241 dir_key
.offset
= range_start
;
2244 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
2249 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2250 if (path
->slots
[0] >= nritems
) {
2251 ret
= btrfs_next_leaf(root
, path
);
2255 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2257 if (found_key
.objectid
!= dirid
||
2258 found_key
.type
!= dir_key
.type
)
2261 if (found_key
.offset
> range_end
)
2264 ret
= check_item_in_log(trans
, root
, log
, path
,
2269 if (found_key
.offset
== (u64
)-1)
2271 dir_key
.offset
= found_key
.offset
+ 1;
2273 btrfs_release_path(path
);
2274 if (range_end
== (u64
)-1)
2276 range_start
= range_end
+ 1;
2281 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
2282 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
2283 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
2284 btrfs_release_path(path
);
2288 btrfs_release_path(path
);
2289 btrfs_free_path(log_path
);
2295 * the process_func used to replay items from the log tree. This
2296 * gets called in two different stages. The first stage just looks
2297 * for inodes and makes sure they are all copied into the subvolume.
2299 * The second stage copies all the other item types from the log into
2300 * the subvolume. The two stage approach is slower, but gets rid of
2301 * lots of complexity around inodes referencing other inodes that exist
2302 * only in the log (references come from either directory items or inode
2305 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
2306 struct walk_control
*wc
, u64 gen
)
2309 struct btrfs_path
*path
;
2310 struct btrfs_root
*root
= wc
->replay_dest
;
2311 struct btrfs_key key
;
2316 ret
= btrfs_read_buffer(eb
, gen
);
2320 level
= btrfs_header_level(eb
);
2325 path
= btrfs_alloc_path();
2329 nritems
= btrfs_header_nritems(eb
);
2330 for (i
= 0; i
< nritems
; i
++) {
2331 btrfs_item_key_to_cpu(eb
, &key
, i
);
2333 /* inode keys are done during the first stage */
2334 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
2335 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
2336 struct btrfs_inode_item
*inode_item
;
2339 inode_item
= btrfs_item_ptr(eb
, i
,
2340 struct btrfs_inode_item
);
2341 ret
= replay_xattr_deletes(wc
->trans
, root
, log
,
2342 path
, key
.objectid
);
2345 mode
= btrfs_inode_mode(eb
, inode_item
);
2346 if (S_ISDIR(mode
)) {
2347 ret
= replay_dir_deletes(wc
->trans
,
2348 root
, log
, path
, key
.objectid
, 0);
2352 ret
= overwrite_item(wc
->trans
, root
, path
,
2357 /* for regular files, make sure corresponding
2358 * orphan item exist. extents past the new EOF
2359 * will be truncated later by orphan cleanup.
2361 if (S_ISREG(mode
)) {
2362 ret
= insert_orphan_item(wc
->trans
, root
,
2368 ret
= link_to_fixup_dir(wc
->trans
, root
,
2369 path
, key
.objectid
);
2374 if (key
.type
== BTRFS_DIR_INDEX_KEY
&&
2375 wc
->stage
== LOG_WALK_REPLAY_DIR_INDEX
) {
2376 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
2382 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
2385 /* these keys are simply copied */
2386 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
2387 ret
= overwrite_item(wc
->trans
, root
, path
,
2391 } else if (key
.type
== BTRFS_INODE_REF_KEY
||
2392 key
.type
== BTRFS_INODE_EXTREF_KEY
) {
2393 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
2395 if (ret
&& ret
!= -ENOENT
)
2398 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
2399 ret
= replay_one_extent(wc
->trans
, root
, path
,
2403 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
) {
2404 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
2410 btrfs_free_path(path
);
2414 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
2415 struct btrfs_root
*root
,
2416 struct btrfs_path
*path
, int *level
,
2417 struct walk_control
*wc
)
2419 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2423 struct extent_buffer
*next
;
2424 struct extent_buffer
*cur
;
2425 struct extent_buffer
*parent
;
2429 WARN_ON(*level
< 0);
2430 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2432 while (*level
> 0) {
2433 WARN_ON(*level
< 0);
2434 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2435 cur
= path
->nodes
[*level
];
2437 WARN_ON(btrfs_header_level(cur
) != *level
);
2439 if (path
->slots
[*level
] >=
2440 btrfs_header_nritems(cur
))
2443 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2444 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2445 blocksize
= fs_info
->nodesize
;
2447 parent
= path
->nodes
[*level
];
2448 root_owner
= btrfs_header_owner(parent
);
2450 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
2452 return PTR_ERR(next
);
2455 ret
= wc
->process_func(root
, next
, wc
, ptr_gen
);
2457 free_extent_buffer(next
);
2461 path
->slots
[*level
]++;
2463 ret
= btrfs_read_buffer(next
, ptr_gen
);
2465 free_extent_buffer(next
);
2470 btrfs_tree_lock(next
);
2471 btrfs_set_lock_blocking(next
);
2472 clean_tree_block(trans
, fs_info
, next
);
2473 btrfs_wait_tree_block_writeback(next
);
2474 btrfs_tree_unlock(next
);
2477 WARN_ON(root_owner
!=
2478 BTRFS_TREE_LOG_OBJECTID
);
2479 ret
= btrfs_free_and_pin_reserved_extent(
2483 free_extent_buffer(next
);
2487 free_extent_buffer(next
);
2490 ret
= btrfs_read_buffer(next
, ptr_gen
);
2492 free_extent_buffer(next
);
2496 WARN_ON(*level
<= 0);
2497 if (path
->nodes
[*level
-1])
2498 free_extent_buffer(path
->nodes
[*level
-1]);
2499 path
->nodes
[*level
-1] = next
;
2500 *level
= btrfs_header_level(next
);
2501 path
->slots
[*level
] = 0;
2504 WARN_ON(*level
< 0);
2505 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2507 path
->slots
[*level
] = btrfs_header_nritems(path
->nodes
[*level
]);
2513 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
2514 struct btrfs_root
*root
,
2515 struct btrfs_path
*path
, int *level
,
2516 struct walk_control
*wc
)
2518 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2524 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2525 slot
= path
->slots
[i
];
2526 if (slot
+ 1 < btrfs_header_nritems(path
->nodes
[i
])) {
2529 WARN_ON(*level
== 0);
2532 struct extent_buffer
*parent
;
2533 if (path
->nodes
[*level
] == root
->node
)
2534 parent
= path
->nodes
[*level
];
2536 parent
= path
->nodes
[*level
+ 1];
2538 root_owner
= btrfs_header_owner(parent
);
2539 ret
= wc
->process_func(root
, path
->nodes
[*level
], wc
,
2540 btrfs_header_generation(path
->nodes
[*level
]));
2545 struct extent_buffer
*next
;
2547 next
= path
->nodes
[*level
];
2550 btrfs_tree_lock(next
);
2551 btrfs_set_lock_blocking(next
);
2552 clean_tree_block(trans
, fs_info
, next
);
2553 btrfs_wait_tree_block_writeback(next
);
2554 btrfs_tree_unlock(next
);
2557 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
2558 ret
= btrfs_free_and_pin_reserved_extent(
2560 path
->nodes
[*level
]->start
,
2561 path
->nodes
[*level
]->len
);
2565 free_extent_buffer(path
->nodes
[*level
]);
2566 path
->nodes
[*level
] = NULL
;
2574 * drop the reference count on the tree rooted at 'snap'. This traverses
2575 * the tree freeing any blocks that have a ref count of zero after being
2578 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
2579 struct btrfs_root
*log
, struct walk_control
*wc
)
2581 struct btrfs_fs_info
*fs_info
= log
->fs_info
;
2585 struct btrfs_path
*path
;
2588 path
= btrfs_alloc_path();
2592 level
= btrfs_header_level(log
->node
);
2594 path
->nodes
[level
] = log
->node
;
2595 extent_buffer_get(log
->node
);
2596 path
->slots
[level
] = 0;
2599 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
2607 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
2616 /* was the root node processed? if not, catch it here */
2617 if (path
->nodes
[orig_level
]) {
2618 ret
= wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
2619 btrfs_header_generation(path
->nodes
[orig_level
]));
2623 struct extent_buffer
*next
;
2625 next
= path
->nodes
[orig_level
];
2628 btrfs_tree_lock(next
);
2629 btrfs_set_lock_blocking(next
);
2630 clean_tree_block(trans
, fs_info
, next
);
2631 btrfs_wait_tree_block_writeback(next
);
2632 btrfs_tree_unlock(next
);
2635 WARN_ON(log
->root_key
.objectid
!=
2636 BTRFS_TREE_LOG_OBJECTID
);
2637 ret
= btrfs_free_and_pin_reserved_extent(fs_info
,
2638 next
->start
, next
->len
);
2645 btrfs_free_path(path
);
2650 * helper function to update the item for a given subvolumes log root
2651 * in the tree of log roots
2653 static int update_log_root(struct btrfs_trans_handle
*trans
,
2654 struct btrfs_root
*log
)
2656 struct btrfs_fs_info
*fs_info
= log
->fs_info
;
2659 if (log
->log_transid
== 1) {
2660 /* insert root item on the first sync */
2661 ret
= btrfs_insert_root(trans
, fs_info
->log_root_tree
,
2662 &log
->root_key
, &log
->root_item
);
2664 ret
= btrfs_update_root(trans
, fs_info
->log_root_tree
,
2665 &log
->root_key
, &log
->root_item
);
2670 static void wait_log_commit(struct btrfs_root
*root
, int transid
)
2673 int index
= transid
% 2;
2676 * we only allow two pending log transactions at a time,
2677 * so we know that if ours is more than 2 older than the
2678 * current transaction, we're done
2681 prepare_to_wait(&root
->log_commit_wait
[index
],
2682 &wait
, TASK_UNINTERRUPTIBLE
);
2683 mutex_unlock(&root
->log_mutex
);
2685 if (root
->log_transid_committed
< transid
&&
2686 atomic_read(&root
->log_commit
[index
]))
2689 finish_wait(&root
->log_commit_wait
[index
], &wait
);
2690 mutex_lock(&root
->log_mutex
);
2691 } while (root
->log_transid_committed
< transid
&&
2692 atomic_read(&root
->log_commit
[index
]));
2695 static void wait_for_writer(struct btrfs_root
*root
)
2699 while (atomic_read(&root
->log_writers
)) {
2700 prepare_to_wait(&root
->log_writer_wait
,
2701 &wait
, TASK_UNINTERRUPTIBLE
);
2702 mutex_unlock(&root
->log_mutex
);
2703 if (atomic_read(&root
->log_writers
))
2705 finish_wait(&root
->log_writer_wait
, &wait
);
2706 mutex_lock(&root
->log_mutex
);
2710 static inline void btrfs_remove_log_ctx(struct btrfs_root
*root
,
2711 struct btrfs_log_ctx
*ctx
)
2716 mutex_lock(&root
->log_mutex
);
2717 list_del_init(&ctx
->list
);
2718 mutex_unlock(&root
->log_mutex
);
2722 * Invoked in log mutex context, or be sure there is no other task which
2723 * can access the list.
2725 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root
*root
,
2726 int index
, int error
)
2728 struct btrfs_log_ctx
*ctx
;
2729 struct btrfs_log_ctx
*safe
;
2731 list_for_each_entry_safe(ctx
, safe
, &root
->log_ctxs
[index
], list
) {
2732 list_del_init(&ctx
->list
);
2733 ctx
->log_ret
= error
;
2736 INIT_LIST_HEAD(&root
->log_ctxs
[index
]);
2740 * btrfs_sync_log does sends a given tree log down to the disk and
2741 * updates the super blocks to record it. When this call is done,
2742 * you know that any inodes previously logged are safely on disk only
2745 * Any other return value means you need to call btrfs_commit_transaction.
2746 * Some of the edge cases for fsyncing directories that have had unlinks
2747 * or renames done in the past mean that sometimes the only safe
2748 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2749 * that has happened.
2751 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
2752 struct btrfs_root
*root
, struct btrfs_log_ctx
*ctx
)
2758 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2759 struct btrfs_root
*log
= root
->log_root
;
2760 struct btrfs_root
*log_root_tree
= fs_info
->log_root_tree
;
2761 int log_transid
= 0;
2762 struct btrfs_log_ctx root_log_ctx
;
2763 struct blk_plug plug
;
2765 mutex_lock(&root
->log_mutex
);
2766 log_transid
= ctx
->log_transid
;
2767 if (root
->log_transid_committed
>= log_transid
) {
2768 mutex_unlock(&root
->log_mutex
);
2769 return ctx
->log_ret
;
2772 index1
= log_transid
% 2;
2773 if (atomic_read(&root
->log_commit
[index1
])) {
2774 wait_log_commit(root
, log_transid
);
2775 mutex_unlock(&root
->log_mutex
);
2776 return ctx
->log_ret
;
2778 ASSERT(log_transid
== root
->log_transid
);
2779 atomic_set(&root
->log_commit
[index1
], 1);
2781 /* wait for previous tree log sync to complete */
2782 if (atomic_read(&root
->log_commit
[(index1
+ 1) % 2]))
2783 wait_log_commit(root
, log_transid
- 1);
2786 int batch
= atomic_read(&root
->log_batch
);
2787 /* when we're on an ssd, just kick the log commit out */
2788 if (!btrfs_test_opt(fs_info
, SSD
) &&
2789 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS
, &root
->state
)) {
2790 mutex_unlock(&root
->log_mutex
);
2791 schedule_timeout_uninterruptible(1);
2792 mutex_lock(&root
->log_mutex
);
2794 wait_for_writer(root
);
2795 if (batch
== atomic_read(&root
->log_batch
))
2799 /* bail out if we need to do a full commit */
2800 if (btrfs_need_log_full_commit(fs_info
, trans
)) {
2802 btrfs_free_logged_extents(log
, log_transid
);
2803 mutex_unlock(&root
->log_mutex
);
2807 if (log_transid
% 2 == 0)
2808 mark
= EXTENT_DIRTY
;
2812 /* we start IO on all the marked extents here, but we don't actually
2813 * wait for them until later.
2815 blk_start_plug(&plug
);
2816 ret
= btrfs_write_marked_extents(fs_info
, &log
->dirty_log_pages
, mark
);
2818 blk_finish_plug(&plug
);
2819 btrfs_abort_transaction(trans
, ret
);
2820 btrfs_free_logged_extents(log
, log_transid
);
2821 btrfs_set_log_full_commit(fs_info
, trans
);
2822 mutex_unlock(&root
->log_mutex
);
2826 btrfs_set_root_node(&log
->root_item
, log
->node
);
2828 root
->log_transid
++;
2829 log
->log_transid
= root
->log_transid
;
2830 root
->log_start_pid
= 0;
2832 * IO has been started, blocks of the log tree have WRITTEN flag set
2833 * in their headers. new modifications of the log will be written to
2834 * new positions. so it's safe to allow log writers to go in.
2836 mutex_unlock(&root
->log_mutex
);
2838 btrfs_init_log_ctx(&root_log_ctx
, NULL
);
2840 mutex_lock(&log_root_tree
->log_mutex
);
2841 atomic_inc(&log_root_tree
->log_batch
);
2842 atomic_inc(&log_root_tree
->log_writers
);
2844 index2
= log_root_tree
->log_transid
% 2;
2845 list_add_tail(&root_log_ctx
.list
, &log_root_tree
->log_ctxs
[index2
]);
2846 root_log_ctx
.log_transid
= log_root_tree
->log_transid
;
2848 mutex_unlock(&log_root_tree
->log_mutex
);
2850 ret
= update_log_root(trans
, log
);
2852 mutex_lock(&log_root_tree
->log_mutex
);
2853 if (atomic_dec_and_test(&log_root_tree
->log_writers
)) {
2855 * Implicit memory barrier after atomic_dec_and_test
2857 if (waitqueue_active(&log_root_tree
->log_writer_wait
))
2858 wake_up(&log_root_tree
->log_writer_wait
);
2862 if (!list_empty(&root_log_ctx
.list
))
2863 list_del_init(&root_log_ctx
.list
);
2865 blk_finish_plug(&plug
);
2866 btrfs_set_log_full_commit(fs_info
, trans
);
2868 if (ret
!= -ENOSPC
) {
2869 btrfs_abort_transaction(trans
, ret
);
2870 mutex_unlock(&log_root_tree
->log_mutex
);
2873 btrfs_wait_tree_log_extents(log
, mark
);
2874 btrfs_free_logged_extents(log
, log_transid
);
2875 mutex_unlock(&log_root_tree
->log_mutex
);
2880 if (log_root_tree
->log_transid_committed
>= root_log_ctx
.log_transid
) {
2881 blk_finish_plug(&plug
);
2882 list_del_init(&root_log_ctx
.list
);
2883 mutex_unlock(&log_root_tree
->log_mutex
);
2884 ret
= root_log_ctx
.log_ret
;
2888 index2
= root_log_ctx
.log_transid
% 2;
2889 if (atomic_read(&log_root_tree
->log_commit
[index2
])) {
2890 blk_finish_plug(&plug
);
2891 ret
= btrfs_wait_tree_log_extents(log
, mark
);
2892 btrfs_wait_logged_extents(trans
, log
, log_transid
);
2893 wait_log_commit(log_root_tree
,
2894 root_log_ctx
.log_transid
);
2895 mutex_unlock(&log_root_tree
->log_mutex
);
2897 ret
= root_log_ctx
.log_ret
;
2900 ASSERT(root_log_ctx
.log_transid
== log_root_tree
->log_transid
);
2901 atomic_set(&log_root_tree
->log_commit
[index2
], 1);
2903 if (atomic_read(&log_root_tree
->log_commit
[(index2
+ 1) % 2])) {
2904 wait_log_commit(log_root_tree
,
2905 root_log_ctx
.log_transid
- 1);
2908 wait_for_writer(log_root_tree
);
2911 * now that we've moved on to the tree of log tree roots,
2912 * check the full commit flag again
2914 if (btrfs_need_log_full_commit(fs_info
, trans
)) {
2915 blk_finish_plug(&plug
);
2916 btrfs_wait_tree_log_extents(log
, mark
);
2917 btrfs_free_logged_extents(log
, log_transid
);
2918 mutex_unlock(&log_root_tree
->log_mutex
);
2920 goto out_wake_log_root
;
2923 ret
= btrfs_write_marked_extents(fs_info
,
2924 &log_root_tree
->dirty_log_pages
,
2925 EXTENT_DIRTY
| EXTENT_NEW
);
2926 blk_finish_plug(&plug
);
2928 btrfs_set_log_full_commit(fs_info
, trans
);
2929 btrfs_abort_transaction(trans
, ret
);
2930 btrfs_free_logged_extents(log
, log_transid
);
2931 mutex_unlock(&log_root_tree
->log_mutex
);
2932 goto out_wake_log_root
;
2934 ret
= btrfs_wait_tree_log_extents(log
, mark
);
2936 ret
= btrfs_wait_tree_log_extents(log_root_tree
,
2937 EXTENT_NEW
| EXTENT_DIRTY
);
2939 btrfs_set_log_full_commit(fs_info
, trans
);
2940 btrfs_free_logged_extents(log
, log_transid
);
2941 mutex_unlock(&log_root_tree
->log_mutex
);
2942 goto out_wake_log_root
;
2944 btrfs_wait_logged_extents(trans
, log
, log_transid
);
2946 btrfs_set_super_log_root(fs_info
->super_for_commit
,
2947 log_root_tree
->node
->start
);
2948 btrfs_set_super_log_root_level(fs_info
->super_for_commit
,
2949 btrfs_header_level(log_root_tree
->node
));
2951 log_root_tree
->log_transid
++;
2952 mutex_unlock(&log_root_tree
->log_mutex
);
2955 * nobody else is going to jump in and write the the ctree
2956 * super here because the log_commit atomic below is protecting
2957 * us. We must be called with a transaction handle pinning
2958 * the running transaction open, so a full commit can't hop
2959 * in and cause problems either.
2961 ret
= write_ctree_super(trans
, fs_info
, 1);
2963 btrfs_set_log_full_commit(fs_info
, trans
);
2964 btrfs_abort_transaction(trans
, ret
);
2965 goto out_wake_log_root
;
2968 mutex_lock(&root
->log_mutex
);
2969 if (root
->last_log_commit
< log_transid
)
2970 root
->last_log_commit
= log_transid
;
2971 mutex_unlock(&root
->log_mutex
);
2974 mutex_lock(&log_root_tree
->log_mutex
);
2975 btrfs_remove_all_log_ctxs(log_root_tree
, index2
, ret
);
2977 log_root_tree
->log_transid_committed
++;
2978 atomic_set(&log_root_tree
->log_commit
[index2
], 0);
2979 mutex_unlock(&log_root_tree
->log_mutex
);
2982 * The barrier before waitqueue_active is implied by mutex_unlock
2984 if (waitqueue_active(&log_root_tree
->log_commit_wait
[index2
]))
2985 wake_up(&log_root_tree
->log_commit_wait
[index2
]);
2987 mutex_lock(&root
->log_mutex
);
2988 btrfs_remove_all_log_ctxs(root
, index1
, ret
);
2989 root
->log_transid_committed
++;
2990 atomic_set(&root
->log_commit
[index1
], 0);
2991 mutex_unlock(&root
->log_mutex
);
2994 * The barrier before waitqueue_active is implied by mutex_unlock
2996 if (waitqueue_active(&root
->log_commit_wait
[index1
]))
2997 wake_up(&root
->log_commit_wait
[index1
]);
3001 static void free_log_tree(struct btrfs_trans_handle
*trans
,
3002 struct btrfs_root
*log
)
3007 struct walk_control wc
= {
3009 .process_func
= process_one_buffer
3012 ret
= walk_log_tree(trans
, log
, &wc
);
3013 /* I don't think this can happen but just in case */
3015 btrfs_abort_transaction(trans
, ret
);
3018 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
3019 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
,
3024 clear_extent_bits(&log
->dirty_log_pages
, start
, end
,
3025 EXTENT_DIRTY
| EXTENT_NEW
);
3029 * We may have short-circuited the log tree with the full commit logic
3030 * and left ordered extents on our list, so clear these out to keep us
3031 * from leaking inodes and memory.
3033 btrfs_free_logged_extents(log
, 0);
3034 btrfs_free_logged_extents(log
, 1);
3036 free_extent_buffer(log
->node
);
3041 * free all the extents used by the tree log. This should be called
3042 * at commit time of the full transaction
3044 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
3046 if (root
->log_root
) {
3047 free_log_tree(trans
, root
->log_root
);
3048 root
->log_root
= NULL
;
3053 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
3054 struct btrfs_fs_info
*fs_info
)
3056 if (fs_info
->log_root_tree
) {
3057 free_log_tree(trans
, fs_info
->log_root_tree
);
3058 fs_info
->log_root_tree
= NULL
;
3064 * If both a file and directory are logged, and unlinks or renames are
3065 * mixed in, we have a few interesting corners:
3067 * create file X in dir Y
3068 * link file X to X.link in dir Y
3070 * unlink file X but leave X.link
3073 * After a crash we would expect only X.link to exist. But file X
3074 * didn't get fsync'd again so the log has back refs for X and X.link.
3076 * We solve this by removing directory entries and inode backrefs from the
3077 * log when a file that was logged in the current transaction is
3078 * unlinked. Any later fsync will include the updated log entries, and
3079 * we'll be able to reconstruct the proper directory items from backrefs.
3081 * This optimizations allows us to avoid relogging the entire inode
3082 * or the entire directory.
3084 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
3085 struct btrfs_root
*root
,
3086 const char *name
, int name_len
,
3087 struct inode
*dir
, u64 index
)
3089 struct btrfs_root
*log
;
3090 struct btrfs_dir_item
*di
;
3091 struct btrfs_path
*path
;
3095 u64 dir_ino
= btrfs_ino(dir
);
3097 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
3100 ret
= join_running_log_trans(root
);
3104 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
3106 log
= root
->log_root
;
3107 path
= btrfs_alloc_path();
3113 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir_ino
,
3114 name
, name_len
, -1);
3120 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
3121 bytes_del
+= name_len
;
3127 btrfs_release_path(path
);
3128 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir_ino
,
3129 index
, name
, name_len
, -1);
3135 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
3136 bytes_del
+= name_len
;
3143 /* update the directory size in the log to reflect the names
3147 struct btrfs_key key
;
3149 key
.objectid
= dir_ino
;
3151 key
.type
= BTRFS_INODE_ITEM_KEY
;
3152 btrfs_release_path(path
);
3154 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
3160 struct btrfs_inode_item
*item
;
3163 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3164 struct btrfs_inode_item
);
3165 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
3166 if (i_size
> bytes_del
)
3167 i_size
-= bytes_del
;
3170 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
3171 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3174 btrfs_release_path(path
);
3177 btrfs_free_path(path
);
3179 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
3180 if (ret
== -ENOSPC
) {
3181 btrfs_set_log_full_commit(root
->fs_info
, trans
);
3184 btrfs_abort_transaction(trans
, ret
);
3186 btrfs_end_log_trans(root
);
3191 /* see comments for btrfs_del_dir_entries_in_log */
3192 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
3193 struct btrfs_root
*root
,
3194 const char *name
, int name_len
,
3195 struct inode
*inode
, u64 dirid
)
3197 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3198 struct btrfs_root
*log
;
3202 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
3205 ret
= join_running_log_trans(root
);
3208 log
= root
->log_root
;
3209 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
3211 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, btrfs_ino(inode
),
3213 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
3214 if (ret
== -ENOSPC
) {
3215 btrfs_set_log_full_commit(fs_info
, trans
);
3217 } else if (ret
< 0 && ret
!= -ENOENT
)
3218 btrfs_abort_transaction(trans
, ret
);
3219 btrfs_end_log_trans(root
);
3225 * creates a range item in the log for 'dirid'. first_offset and
3226 * last_offset tell us which parts of the key space the log should
3227 * be considered authoritative for.
3229 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
3230 struct btrfs_root
*log
,
3231 struct btrfs_path
*path
,
3232 int key_type
, u64 dirid
,
3233 u64 first_offset
, u64 last_offset
)
3236 struct btrfs_key key
;
3237 struct btrfs_dir_log_item
*item
;
3239 key
.objectid
= dirid
;
3240 key
.offset
= first_offset
;
3241 if (key_type
== BTRFS_DIR_ITEM_KEY
)
3242 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
3244 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
3245 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
3249 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3250 struct btrfs_dir_log_item
);
3251 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
3252 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3253 btrfs_release_path(path
);
3258 * log all the items included in the current transaction for a given
3259 * directory. This also creates the range items in the log tree required
3260 * to replay anything deleted before the fsync
3262 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
3263 struct btrfs_root
*root
, struct inode
*inode
,
3264 struct btrfs_path
*path
,
3265 struct btrfs_path
*dst_path
, int key_type
,
3266 struct btrfs_log_ctx
*ctx
,
3267 u64 min_offset
, u64
*last_offset_ret
)
3269 struct btrfs_key min_key
;
3270 struct btrfs_root
*log
= root
->log_root
;
3271 struct extent_buffer
*src
;
3276 u64 first_offset
= min_offset
;
3277 u64 last_offset
= (u64
)-1;
3278 u64 ino
= btrfs_ino(inode
);
3280 log
= root
->log_root
;
3282 min_key
.objectid
= ino
;
3283 min_key
.type
= key_type
;
3284 min_key
.offset
= min_offset
;
3286 ret
= btrfs_search_forward(root
, &min_key
, path
, trans
->transid
);
3289 * we didn't find anything from this transaction, see if there
3290 * is anything at all
3292 if (ret
!= 0 || min_key
.objectid
!= ino
|| min_key
.type
!= key_type
) {
3293 min_key
.objectid
= ino
;
3294 min_key
.type
= key_type
;
3295 min_key
.offset
= (u64
)-1;
3296 btrfs_release_path(path
);
3297 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
3299 btrfs_release_path(path
);
3302 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
3304 /* if ret == 0 there are items for this type,
3305 * create a range to tell us the last key of this type.
3306 * otherwise, there are no items in this directory after
3307 * *min_offset, and we create a range to indicate that.
3310 struct btrfs_key tmp
;
3311 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
3313 if (key_type
== tmp
.type
)
3314 first_offset
= max(min_offset
, tmp
.offset
) + 1;
3319 /* go backward to find any previous key */
3320 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
3322 struct btrfs_key tmp
;
3323 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
3324 if (key_type
== tmp
.type
) {
3325 first_offset
= tmp
.offset
;
3326 ret
= overwrite_item(trans
, log
, dst_path
,
3327 path
->nodes
[0], path
->slots
[0],
3335 btrfs_release_path(path
);
3337 /* find the first key from this transaction again */
3338 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
3339 if (WARN_ON(ret
!= 0))
3343 * we have a block from this transaction, log every item in it
3344 * from our directory
3347 struct btrfs_key tmp
;
3348 src
= path
->nodes
[0];
3349 nritems
= btrfs_header_nritems(src
);
3350 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
3351 struct btrfs_dir_item
*di
;
3353 btrfs_item_key_to_cpu(src
, &min_key
, i
);
3355 if (min_key
.objectid
!= ino
|| min_key
.type
!= key_type
)
3357 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
3365 * We must make sure that when we log a directory entry,
3366 * the corresponding inode, after log replay, has a
3367 * matching link count. For example:
3373 * xfs_io -c "fsync" mydir
3375 * <mount fs and log replay>
3377 * Would result in a fsync log that when replayed, our
3378 * file inode would have a link count of 1, but we get
3379 * two directory entries pointing to the same inode.
3380 * After removing one of the names, it would not be
3381 * possible to remove the other name, which resulted
3382 * always in stale file handle errors, and would not
3383 * be possible to rmdir the parent directory, since
3384 * its i_size could never decrement to the value
3385 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3387 di
= btrfs_item_ptr(src
, i
, struct btrfs_dir_item
);
3388 btrfs_dir_item_key_to_cpu(src
, di
, &tmp
);
3390 (btrfs_dir_transid(src
, di
) == trans
->transid
||
3391 btrfs_dir_type(src
, di
) == BTRFS_FT_DIR
) &&
3392 tmp
.type
!= BTRFS_ROOT_ITEM_KEY
)
3393 ctx
->log_new_dentries
= true;
3395 path
->slots
[0] = nritems
;
3398 * look ahead to the next item and see if it is also
3399 * from this directory and from this transaction
3401 ret
= btrfs_next_leaf(root
, path
);
3403 last_offset
= (u64
)-1;
3406 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
3407 if (tmp
.objectid
!= ino
|| tmp
.type
!= key_type
) {
3408 last_offset
= (u64
)-1;
3411 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
3412 ret
= overwrite_item(trans
, log
, dst_path
,
3413 path
->nodes
[0], path
->slots
[0],
3418 last_offset
= tmp
.offset
;
3423 btrfs_release_path(path
);
3424 btrfs_release_path(dst_path
);
3427 *last_offset_ret
= last_offset
;
3429 * insert the log range keys to indicate where the log
3432 ret
= insert_dir_log_key(trans
, log
, path
, key_type
,
3433 ino
, first_offset
, last_offset
);
3441 * logging directories is very similar to logging inodes, We find all the items
3442 * from the current transaction and write them to the log.
3444 * The recovery code scans the directory in the subvolume, and if it finds a
3445 * key in the range logged that is not present in the log tree, then it means
3446 * that dir entry was unlinked during the transaction.
3448 * In order for that scan to work, we must include one key smaller than
3449 * the smallest logged by this transaction and one key larger than the largest
3450 * key logged by this transaction.
3452 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
3453 struct btrfs_root
*root
, struct inode
*inode
,
3454 struct btrfs_path
*path
,
3455 struct btrfs_path
*dst_path
,
3456 struct btrfs_log_ctx
*ctx
)
3461 int key_type
= BTRFS_DIR_ITEM_KEY
;
3467 ret
= log_dir_items(trans
, root
, inode
, path
,
3468 dst_path
, key_type
, ctx
, min_key
,
3472 if (max_key
== (u64
)-1)
3474 min_key
= max_key
+ 1;
3477 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
3478 key_type
= BTRFS_DIR_INDEX_KEY
;
3485 * a helper function to drop items from the log before we relog an
3486 * inode. max_key_type indicates the highest item type to remove.
3487 * This cannot be run for file data extents because it does not
3488 * free the extents they point to.
3490 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
3491 struct btrfs_root
*log
,
3492 struct btrfs_path
*path
,
3493 u64 objectid
, int max_key_type
)
3496 struct btrfs_key key
;
3497 struct btrfs_key found_key
;
3500 key
.objectid
= objectid
;
3501 key
.type
= max_key_type
;
3502 key
.offset
= (u64
)-1;
3505 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
3506 BUG_ON(ret
== 0); /* Logic error */
3510 if (path
->slots
[0] == 0)
3514 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
3517 if (found_key
.objectid
!= objectid
)
3520 found_key
.offset
= 0;
3522 ret
= btrfs_bin_search(path
->nodes
[0], &found_key
, 0,
3525 ret
= btrfs_del_items(trans
, log
, path
, start_slot
,
3526 path
->slots
[0] - start_slot
+ 1);
3528 * If start slot isn't 0 then we don't need to re-search, we've
3529 * found the last guy with the objectid in this tree.
3531 if (ret
|| start_slot
!= 0)
3533 btrfs_release_path(path
);
3535 btrfs_release_path(path
);
3541 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3542 struct extent_buffer
*leaf
,
3543 struct btrfs_inode_item
*item
,
3544 struct inode
*inode
, int log_inode_only
,
3547 struct btrfs_map_token token
;
3549 btrfs_init_map_token(&token
);
3551 if (log_inode_only
) {
3552 /* set the generation to zero so the recover code
3553 * can tell the difference between an logging
3554 * just to say 'this inode exists' and a logging
3555 * to say 'update this inode with these values'
3557 btrfs_set_token_inode_generation(leaf
, item
, 0, &token
);
3558 btrfs_set_token_inode_size(leaf
, item
, logged_isize
, &token
);
3560 btrfs_set_token_inode_generation(leaf
, item
,
3561 BTRFS_I(inode
)->generation
,
3563 btrfs_set_token_inode_size(leaf
, item
, inode
->i_size
, &token
);
3566 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3567 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3568 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3569 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3571 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3572 inode
->i_atime
.tv_sec
, &token
);
3573 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3574 inode
->i_atime
.tv_nsec
, &token
);
3576 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3577 inode
->i_mtime
.tv_sec
, &token
);
3578 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3579 inode
->i_mtime
.tv_nsec
, &token
);
3581 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3582 inode
->i_ctime
.tv_sec
, &token
);
3583 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3584 inode
->i_ctime
.tv_nsec
, &token
);
3586 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3589 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3590 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3591 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3592 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3593 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3596 static int log_inode_item(struct btrfs_trans_handle
*trans
,
3597 struct btrfs_root
*log
, struct btrfs_path
*path
,
3598 struct inode
*inode
)
3600 struct btrfs_inode_item
*inode_item
;
3603 ret
= btrfs_insert_empty_item(trans
, log
, path
,
3604 &BTRFS_I(inode
)->location
,
3605 sizeof(*inode_item
));
3606 if (ret
&& ret
!= -EEXIST
)
3608 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3609 struct btrfs_inode_item
);
3610 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
, 0, 0);
3611 btrfs_release_path(path
);
3615 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
3616 struct inode
*inode
,
3617 struct btrfs_path
*dst_path
,
3618 struct btrfs_path
*src_path
, u64
*last_extent
,
3619 int start_slot
, int nr
, int inode_only
,
3622 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3623 unsigned long src_offset
;
3624 unsigned long dst_offset
;
3625 struct btrfs_root
*log
= BTRFS_I(inode
)->root
->log_root
;
3626 struct btrfs_file_extent_item
*extent
;
3627 struct btrfs_inode_item
*inode_item
;
3628 struct extent_buffer
*src
= src_path
->nodes
[0];
3629 struct btrfs_key first_key
, last_key
, key
;
3631 struct btrfs_key
*ins_keys
;
3635 struct list_head ordered_sums
;
3636 int skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
3637 bool has_extents
= false;
3638 bool need_find_last_extent
= true;
3641 INIT_LIST_HEAD(&ordered_sums
);
3643 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
3644 nr
* sizeof(u32
), GFP_NOFS
);
3648 first_key
.objectid
= (u64
)-1;
3650 ins_sizes
= (u32
*)ins_data
;
3651 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
3653 for (i
= 0; i
< nr
; i
++) {
3654 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
3655 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
3657 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
3658 ins_keys
, ins_sizes
, nr
);
3664 for (i
= 0; i
< nr
; i
++, dst_path
->slots
[0]++) {
3665 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
3666 dst_path
->slots
[0]);
3668 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
3670 if ((i
== (nr
- 1)))
3671 last_key
= ins_keys
[i
];
3673 if (ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
3674 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
3676 struct btrfs_inode_item
);
3677 fill_inode_item(trans
, dst_path
->nodes
[0], inode_item
,
3678 inode
, inode_only
== LOG_INODE_EXISTS
,
3681 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
3682 src_offset
, ins_sizes
[i
]);
3686 * We set need_find_last_extent here in case we know we were
3687 * processing other items and then walk into the first extent in
3688 * the inode. If we don't hit an extent then nothing changes,
3689 * we'll do the last search the next time around.
3691 if (ins_keys
[i
].type
== BTRFS_EXTENT_DATA_KEY
) {
3693 if (first_key
.objectid
== (u64
)-1)
3694 first_key
= ins_keys
[i
];
3696 need_find_last_extent
= false;
3699 /* take a reference on file data extents so that truncates
3700 * or deletes of this inode don't have to relog the inode
3703 if (ins_keys
[i
].type
== BTRFS_EXTENT_DATA_KEY
&&
3706 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
3707 struct btrfs_file_extent_item
);
3709 if (btrfs_file_extent_generation(src
, extent
) < trans
->transid
)
3712 found_type
= btrfs_file_extent_type(src
, extent
);
3713 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
3715 ds
= btrfs_file_extent_disk_bytenr(src
,
3717 /* ds == 0 is a hole */
3721 dl
= btrfs_file_extent_disk_num_bytes(src
,
3723 cs
= btrfs_file_extent_offset(src
, extent
);
3724 cl
= btrfs_file_extent_num_bytes(src
,
3726 if (btrfs_file_extent_compression(src
,
3732 ret
= btrfs_lookup_csums_range(
3734 ds
+ cs
, ds
+ cs
+ cl
- 1,
3737 btrfs_release_path(dst_path
);
3745 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
3746 btrfs_release_path(dst_path
);
3750 * we have to do this after the loop above to avoid changing the
3751 * log tree while trying to change the log tree.
3754 while (!list_empty(&ordered_sums
)) {
3755 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
3756 struct btrfs_ordered_sum
,
3759 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
3760 list_del(&sums
->list
);
3767 if (need_find_last_extent
&& *last_extent
== first_key
.offset
) {
3769 * We don't have any leafs between our current one and the one
3770 * we processed before that can have file extent items for our
3771 * inode (and have a generation number smaller than our current
3774 need_find_last_extent
= false;
3778 * Because we use btrfs_search_forward we could skip leaves that were
3779 * not modified and then assume *last_extent is valid when it really
3780 * isn't. So back up to the previous leaf and read the end of the last
3781 * extent before we go and fill in holes.
3783 if (need_find_last_extent
) {
3786 ret
= btrfs_prev_leaf(BTRFS_I(inode
)->root
, src_path
);
3791 if (src_path
->slots
[0])
3792 src_path
->slots
[0]--;
3793 src
= src_path
->nodes
[0];
3794 btrfs_item_key_to_cpu(src
, &key
, src_path
->slots
[0]);
3795 if (key
.objectid
!= btrfs_ino(inode
) ||
3796 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3798 extent
= btrfs_item_ptr(src
, src_path
->slots
[0],
3799 struct btrfs_file_extent_item
);
3800 if (btrfs_file_extent_type(src
, extent
) ==
3801 BTRFS_FILE_EXTENT_INLINE
) {
3802 len
= btrfs_file_extent_inline_len(src
,
3805 *last_extent
= ALIGN(key
.offset
+ len
,
3806 fs_info
->sectorsize
);
3808 len
= btrfs_file_extent_num_bytes(src
, extent
);
3809 *last_extent
= key
.offset
+ len
;
3813 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3814 * things could have happened
3816 * 1) A merge could have happened, so we could currently be on a leaf
3817 * that holds what we were copying in the first place.
3818 * 2) A split could have happened, and now not all of the items we want
3819 * are on the same leaf.
3821 * So we need to adjust how we search for holes, we need to drop the
3822 * path and re-search for the first extent key we found, and then walk
3823 * forward until we hit the last one we copied.
3825 if (need_find_last_extent
) {
3826 /* btrfs_prev_leaf could return 1 without releasing the path */
3827 btrfs_release_path(src_path
);
3828 ret
= btrfs_search_slot(NULL
, BTRFS_I(inode
)->root
, &first_key
,
3833 src
= src_path
->nodes
[0];
3834 i
= src_path
->slots
[0];
3840 * Ok so here we need to go through and fill in any holes we may have
3841 * to make sure that holes are punched for those areas in case they had
3842 * extents previously.
3848 if (i
>= btrfs_header_nritems(src_path
->nodes
[0])) {
3849 ret
= btrfs_next_leaf(BTRFS_I(inode
)->root
, src_path
);
3853 src
= src_path
->nodes
[0];
3857 btrfs_item_key_to_cpu(src
, &key
, i
);
3858 if (!btrfs_comp_cpu_keys(&key
, &last_key
))
3860 if (key
.objectid
!= btrfs_ino(inode
) ||
3861 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
3865 extent
= btrfs_item_ptr(src
, i
, struct btrfs_file_extent_item
);
3866 if (btrfs_file_extent_type(src
, extent
) ==
3867 BTRFS_FILE_EXTENT_INLINE
) {
3868 len
= btrfs_file_extent_inline_len(src
, i
, extent
);
3869 extent_end
= ALIGN(key
.offset
+ len
,
3870 fs_info
->sectorsize
);
3872 len
= btrfs_file_extent_num_bytes(src
, extent
);
3873 extent_end
= key
.offset
+ len
;
3877 if (*last_extent
== key
.offset
) {
3878 *last_extent
= extent_end
;
3881 offset
= *last_extent
;
3882 len
= key
.offset
- *last_extent
;
3883 ret
= btrfs_insert_file_extent(trans
, log
, btrfs_ino(inode
),
3884 offset
, 0, 0, len
, 0, len
, 0,
3888 *last_extent
= extent_end
;
3891 * Need to let the callers know we dropped the path so they should
3894 if (!ret
&& need_find_last_extent
)
3899 static int extent_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3901 struct extent_map
*em1
, *em2
;
3903 em1
= list_entry(a
, struct extent_map
, list
);
3904 em2
= list_entry(b
, struct extent_map
, list
);
3906 if (em1
->start
< em2
->start
)
3908 else if (em1
->start
> em2
->start
)
3913 static int wait_ordered_extents(struct btrfs_trans_handle
*trans
,
3914 struct inode
*inode
,
3915 struct btrfs_root
*root
,
3916 const struct extent_map
*em
,
3917 const struct list_head
*logged_list
,
3918 bool *ordered_io_error
)
3920 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3921 struct btrfs_ordered_extent
*ordered
;
3922 struct btrfs_root
*log
= root
->log_root
;
3923 u64 mod_start
= em
->mod_start
;
3924 u64 mod_len
= em
->mod_len
;
3925 const bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
3928 LIST_HEAD(ordered_sums
);
3931 *ordered_io_error
= false;
3933 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
3934 em
->block_start
== EXTENT_MAP_HOLE
)
3938 * Wait far any ordered extent that covers our extent map. If it
3939 * finishes without an error, first check and see if our csums are on
3940 * our outstanding ordered extents.
3942 list_for_each_entry(ordered
, logged_list
, log_list
) {
3943 struct btrfs_ordered_sum
*sum
;
3948 if (ordered
->file_offset
+ ordered
->len
<= mod_start
||
3949 mod_start
+ mod_len
<= ordered
->file_offset
)
3952 if (!test_bit(BTRFS_ORDERED_IO_DONE
, &ordered
->flags
) &&
3953 !test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
) &&
3954 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
)) {
3955 const u64 start
= ordered
->file_offset
;
3956 const u64 end
= ordered
->file_offset
+ ordered
->len
- 1;
3958 WARN_ON(ordered
->inode
!= inode
);
3959 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
3962 wait_event(ordered
->wait
,
3963 (test_bit(BTRFS_ORDERED_IO_DONE
, &ordered
->flags
) ||
3964 test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
)));
3966 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
)) {
3968 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3969 * i_mapping flags, so that the next fsync won't get
3970 * an outdated io error too.
3972 filemap_check_errors(inode
->i_mapping
);
3973 *ordered_io_error
= true;
3977 * We are going to copy all the csums on this ordered extent, so
3978 * go ahead and adjust mod_start and mod_len in case this
3979 * ordered extent has already been logged.
3981 if (ordered
->file_offset
> mod_start
) {
3982 if (ordered
->file_offset
+ ordered
->len
>=
3983 mod_start
+ mod_len
)
3984 mod_len
= ordered
->file_offset
- mod_start
;
3986 * If we have this case
3988 * |--------- logged extent ---------|
3989 * |----- ordered extent ----|
3991 * Just don't mess with mod_start and mod_len, we'll
3992 * just end up logging more csums than we need and it
3996 if (ordered
->file_offset
+ ordered
->len
<
3997 mod_start
+ mod_len
) {
3998 mod_len
= (mod_start
+ mod_len
) -
3999 (ordered
->file_offset
+ ordered
->len
);
4000 mod_start
= ordered
->file_offset
+
4011 * To keep us from looping for the above case of an ordered
4012 * extent that falls inside of the logged extent.
4014 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM
,
4018 list_for_each_entry(sum
, &ordered
->list
, list
) {
4019 ret
= btrfs_csum_file_blocks(trans
, log
, sum
);
4025 if (*ordered_io_error
|| !mod_len
|| ret
|| skip_csum
)
4028 if (em
->compress_type
) {
4030 csum_len
= max(em
->block_len
, em
->orig_block_len
);
4032 csum_offset
= mod_start
- em
->start
;
4036 /* block start is already adjusted for the file extent offset. */
4037 ret
= btrfs_lookup_csums_range(fs_info
->csum_root
,
4038 em
->block_start
+ csum_offset
,
4039 em
->block_start
+ csum_offset
+
4040 csum_len
- 1, &ordered_sums
, 0);
4044 while (!list_empty(&ordered_sums
)) {
4045 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
4046 struct btrfs_ordered_sum
,
4049 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
4050 list_del(&sums
->list
);
4057 static int log_one_extent(struct btrfs_trans_handle
*trans
,
4058 struct inode
*inode
, struct btrfs_root
*root
,
4059 const struct extent_map
*em
,
4060 struct btrfs_path
*path
,
4061 const struct list_head
*logged_list
,
4062 struct btrfs_log_ctx
*ctx
)
4064 struct btrfs_root
*log
= root
->log_root
;
4065 struct btrfs_file_extent_item
*fi
;
4066 struct extent_buffer
*leaf
;
4067 struct btrfs_map_token token
;
4068 struct btrfs_key key
;
4069 u64 extent_offset
= em
->start
- em
->orig_start
;
4072 int extent_inserted
= 0;
4073 bool ordered_io_err
= false;
4075 ret
= wait_ordered_extents(trans
, inode
, root
, em
, logged_list
,
4080 if (ordered_io_err
) {
4085 btrfs_init_map_token(&token
);
4087 ret
= __btrfs_drop_extents(trans
, log
, inode
, path
, em
->start
,
4088 em
->start
+ em
->len
, NULL
, 0, 1,
4089 sizeof(*fi
), &extent_inserted
);
4093 if (!extent_inserted
) {
4094 key
.objectid
= btrfs_ino(inode
);
4095 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4096 key
.offset
= em
->start
;
4098 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
,
4103 leaf
= path
->nodes
[0];
4104 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4105 struct btrfs_file_extent_item
);
4107 btrfs_set_token_file_extent_generation(leaf
, fi
, trans
->transid
,
4109 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
4110 btrfs_set_token_file_extent_type(leaf
, fi
,
4111 BTRFS_FILE_EXTENT_PREALLOC
,
4114 btrfs_set_token_file_extent_type(leaf
, fi
,
4115 BTRFS_FILE_EXTENT_REG
,
4118 block_len
= max(em
->block_len
, em
->orig_block_len
);
4119 if (em
->compress_type
!= BTRFS_COMPRESS_NONE
) {
4120 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
,
4123 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, block_len
,
4125 } else if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
4126 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
,
4128 extent_offset
, &token
);
4129 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, block_len
,
4132 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
, 0, &token
);
4133 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, 0,
4137 btrfs_set_token_file_extent_offset(leaf
, fi
, extent_offset
, &token
);
4138 btrfs_set_token_file_extent_num_bytes(leaf
, fi
, em
->len
, &token
);
4139 btrfs_set_token_file_extent_ram_bytes(leaf
, fi
, em
->ram_bytes
, &token
);
4140 btrfs_set_token_file_extent_compression(leaf
, fi
, em
->compress_type
,
4142 btrfs_set_token_file_extent_encryption(leaf
, fi
, 0, &token
);
4143 btrfs_set_token_file_extent_other_encoding(leaf
, fi
, 0, &token
);
4144 btrfs_mark_buffer_dirty(leaf
);
4146 btrfs_release_path(path
);
4151 static int btrfs_log_changed_extents(struct btrfs_trans_handle
*trans
,
4152 struct btrfs_root
*root
,
4153 struct inode
*inode
,
4154 struct btrfs_path
*path
,
4155 struct list_head
*logged_list
,
4156 struct btrfs_log_ctx
*ctx
,
4160 struct extent_map
*em
, *n
;
4161 struct list_head extents
;
4162 struct extent_map_tree
*tree
= &BTRFS_I(inode
)->extent_tree
;
4167 INIT_LIST_HEAD(&extents
);
4169 down_write(&BTRFS_I(inode
)->dio_sem
);
4170 write_lock(&tree
->lock
);
4171 test_gen
= root
->fs_info
->last_trans_committed
;
4173 list_for_each_entry_safe(em
, n
, &tree
->modified_extents
, list
) {
4174 list_del_init(&em
->list
);
4177 * Just an arbitrary number, this can be really CPU intensive
4178 * once we start getting a lot of extents, and really once we
4179 * have a bunch of extents we just want to commit since it will
4182 if (++num
> 32768) {
4183 list_del_init(&tree
->modified_extents
);
4188 if (em
->generation
<= test_gen
)
4190 /* Need a ref to keep it from getting evicted from cache */
4191 atomic_inc(&em
->refs
);
4192 set_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
4193 list_add_tail(&em
->list
, &extents
);
4197 list_sort(NULL
, &extents
, extent_cmp
);
4198 btrfs_get_logged_extents(inode
, logged_list
, start
, end
);
4200 * Some ordered extents started by fsync might have completed
4201 * before we could collect them into the list logged_list, which
4202 * means they're gone, not in our logged_list nor in the inode's
4203 * ordered tree. We want the application/user space to know an
4204 * error happened while attempting to persist file data so that
4205 * it can take proper action. If such error happened, we leave
4206 * without writing to the log tree and the fsync must report the
4207 * file data write error and not commit the current transaction.
4209 ret
= filemap_check_errors(inode
->i_mapping
);
4213 while (!list_empty(&extents
)) {
4214 em
= list_entry(extents
.next
, struct extent_map
, list
);
4216 list_del_init(&em
->list
);
4219 * If we had an error we just need to delete everybody from our
4223 clear_em_logging(tree
, em
);
4224 free_extent_map(em
);
4228 write_unlock(&tree
->lock
);
4230 ret
= log_one_extent(trans
, inode
, root
, em
, path
, logged_list
,
4232 write_lock(&tree
->lock
);
4233 clear_em_logging(tree
, em
);
4234 free_extent_map(em
);
4236 WARN_ON(!list_empty(&extents
));
4237 write_unlock(&tree
->lock
);
4238 up_write(&BTRFS_I(inode
)->dio_sem
);
4240 btrfs_release_path(path
);
4244 static int logged_inode_size(struct btrfs_root
*log
, struct inode
*inode
,
4245 struct btrfs_path
*path
, u64
*size_ret
)
4247 struct btrfs_key key
;
4250 key
.objectid
= btrfs_ino(inode
);
4251 key
.type
= BTRFS_INODE_ITEM_KEY
;
4254 ret
= btrfs_search_slot(NULL
, log
, &key
, path
, 0, 0);
4257 } else if (ret
> 0) {
4260 struct btrfs_inode_item
*item
;
4262 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4263 struct btrfs_inode_item
);
4264 *size_ret
= btrfs_inode_size(path
->nodes
[0], item
);
4267 btrfs_release_path(path
);
4272 * At the moment we always log all xattrs. This is to figure out at log replay
4273 * time which xattrs must have their deletion replayed. If a xattr is missing
4274 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4275 * because if a xattr is deleted, the inode is fsynced and a power failure
4276 * happens, causing the log to be replayed the next time the fs is mounted,
4277 * we want the xattr to not exist anymore (same behaviour as other filesystems
4278 * with a journal, ext3/4, xfs, f2fs, etc).
4280 static int btrfs_log_all_xattrs(struct btrfs_trans_handle
*trans
,
4281 struct btrfs_root
*root
,
4282 struct inode
*inode
,
4283 struct btrfs_path
*path
,
4284 struct btrfs_path
*dst_path
)
4287 struct btrfs_key key
;
4288 const u64 ino
= btrfs_ino(inode
);
4293 key
.type
= BTRFS_XATTR_ITEM_KEY
;
4296 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4301 int slot
= path
->slots
[0];
4302 struct extent_buffer
*leaf
= path
->nodes
[0];
4303 int nritems
= btrfs_header_nritems(leaf
);
4305 if (slot
>= nritems
) {
4307 u64 last_extent
= 0;
4309 ret
= copy_items(trans
, inode
, dst_path
, path
,
4310 &last_extent
, start_slot
,
4312 /* can't be 1, extent items aren't processed */
4318 ret
= btrfs_next_leaf(root
, path
);
4326 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
4327 if (key
.objectid
!= ino
|| key
.type
!= BTRFS_XATTR_ITEM_KEY
)
4337 u64 last_extent
= 0;
4339 ret
= copy_items(trans
, inode
, dst_path
, path
,
4340 &last_extent
, start_slot
,
4342 /* can't be 1, extent items aren't processed */
4352 * If the no holes feature is enabled we need to make sure any hole between the
4353 * last extent and the i_size of our inode is explicitly marked in the log. This
4354 * is to make sure that doing something like:
4356 * 1) create file with 128Kb of data
4357 * 2) truncate file to 64Kb
4358 * 3) truncate file to 256Kb
4360 * 5) <crash/power failure>
4361 * 6) mount fs and trigger log replay
4363 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4364 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4365 * file correspond to a hole. The presence of explicit holes in a log tree is
4366 * what guarantees that log replay will remove/adjust file extent items in the
4369 * Here we do not need to care about holes between extents, that is already done
4370 * by copy_items(). We also only need to do this in the full sync path, where we
4371 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4372 * lookup the list of modified extent maps and if any represents a hole, we
4373 * insert a corresponding extent representing a hole in the log tree.
4375 static int btrfs_log_trailing_hole(struct btrfs_trans_handle
*trans
,
4376 struct btrfs_root
*root
,
4377 struct inode
*inode
,
4378 struct btrfs_path
*path
)
4380 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4382 struct btrfs_key key
;
4385 struct extent_buffer
*leaf
;
4386 struct btrfs_root
*log
= root
->log_root
;
4387 const u64 ino
= btrfs_ino(inode
);
4388 const u64 i_size
= i_size_read(inode
);
4390 if (!btrfs_fs_incompat(fs_info
, NO_HOLES
))
4394 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4395 key
.offset
= (u64
)-1;
4397 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4402 ASSERT(path
->slots
[0] > 0);
4404 leaf
= path
->nodes
[0];
4405 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4407 if (key
.objectid
!= ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
4408 /* inode does not have any extents */
4412 struct btrfs_file_extent_item
*extent
;
4416 * If there's an extent beyond i_size, an explicit hole was
4417 * already inserted by copy_items().
4419 if (key
.offset
>= i_size
)
4422 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
4423 struct btrfs_file_extent_item
);
4425 if (btrfs_file_extent_type(leaf
, extent
) ==
4426 BTRFS_FILE_EXTENT_INLINE
) {
4427 len
= btrfs_file_extent_inline_len(leaf
,
4430 ASSERT(len
== i_size
);
4434 len
= btrfs_file_extent_num_bytes(leaf
, extent
);
4435 /* Last extent goes beyond i_size, no need to log a hole. */
4436 if (key
.offset
+ len
> i_size
)
4438 hole_start
= key
.offset
+ len
;
4439 hole_size
= i_size
- hole_start
;
4441 btrfs_release_path(path
);
4443 /* Last extent ends at i_size. */
4447 hole_size
= ALIGN(hole_size
, fs_info
->sectorsize
);
4448 ret
= btrfs_insert_file_extent(trans
, log
, ino
, hole_start
, 0, 0,
4449 hole_size
, 0, hole_size
, 0, 0, 0);
4454 * When we are logging a new inode X, check if it doesn't have a reference that
4455 * matches the reference from some other inode Y created in a past transaction
4456 * and that was renamed in the current transaction. If we don't do this, then at
4457 * log replay time we can lose inode Y (and all its files if it's a directory):
4460 * echo "hello world" > /mnt/x/foobar
4463 * mkdir /mnt/x # or touch /mnt/x
4464 * xfs_io -c fsync /mnt/x
4466 * mount fs, trigger log replay
4468 * After the log replay procedure, we would lose the first directory and all its
4469 * files (file foobar).
4470 * For the case where inode Y is not a directory we simply end up losing it:
4472 * echo "123" > /mnt/foo
4474 * mv /mnt/foo /mnt/bar
4475 * echo "abc" > /mnt/foo
4476 * xfs_io -c fsync /mnt/foo
4479 * We also need this for cases where a snapshot entry is replaced by some other
4480 * entry (file or directory) otherwise we end up with an unreplayable log due to
4481 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4482 * if it were a regular entry:
4485 * btrfs subvolume snapshot /mnt /mnt/x/snap
4486 * btrfs subvolume delete /mnt/x/snap
4489 * fsync /mnt/x or fsync some new file inside it
4492 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4493 * the same transaction.
4495 static int btrfs_check_ref_name_override(struct extent_buffer
*eb
,
4497 const struct btrfs_key
*key
,
4498 struct inode
*inode
,
4502 struct btrfs_path
*search_path
;
4505 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
4507 unsigned long ptr
= btrfs_item_ptr_offset(eb
, slot
);
4509 search_path
= btrfs_alloc_path();
4512 search_path
->search_commit_root
= 1;
4513 search_path
->skip_locking
= 1;
4515 while (cur_offset
< item_size
) {
4519 unsigned long name_ptr
;
4520 struct btrfs_dir_item
*di
;
4522 if (key
->type
== BTRFS_INODE_REF_KEY
) {
4523 struct btrfs_inode_ref
*iref
;
4525 iref
= (struct btrfs_inode_ref
*)(ptr
+ cur_offset
);
4526 parent
= key
->offset
;
4527 this_name_len
= btrfs_inode_ref_name_len(eb
, iref
);
4528 name_ptr
= (unsigned long)(iref
+ 1);
4529 this_len
= sizeof(*iref
) + this_name_len
;
4531 struct btrfs_inode_extref
*extref
;
4533 extref
= (struct btrfs_inode_extref
*)(ptr
+
4535 parent
= btrfs_inode_extref_parent(eb
, extref
);
4536 this_name_len
= btrfs_inode_extref_name_len(eb
, extref
);
4537 name_ptr
= (unsigned long)&extref
->name
;
4538 this_len
= sizeof(*extref
) + this_name_len
;
4541 if (this_name_len
> name_len
) {
4544 new_name
= krealloc(name
, this_name_len
, GFP_NOFS
);
4549 name_len
= this_name_len
;
4553 read_extent_buffer(eb
, name
, name_ptr
, this_name_len
);
4554 di
= btrfs_lookup_dir_item(NULL
, BTRFS_I(inode
)->root
,
4555 search_path
, parent
,
4556 name
, this_name_len
, 0);
4557 if (di
&& !IS_ERR(di
)) {
4558 struct btrfs_key di_key
;
4560 btrfs_dir_item_key_to_cpu(search_path
->nodes
[0],
4562 if (di_key
.type
== BTRFS_INODE_ITEM_KEY
) {
4564 *other_ino
= di_key
.objectid
;
4569 } else if (IS_ERR(di
)) {
4573 btrfs_release_path(search_path
);
4575 cur_offset
+= this_len
;
4579 btrfs_free_path(search_path
);
4584 /* log a single inode in the tree log.
4585 * At least one parent directory for this inode must exist in the tree
4586 * or be logged already.
4588 * Any items from this inode changed by the current transaction are copied
4589 * to the log tree. An extra reference is taken on any extents in this
4590 * file, allowing us to avoid a whole pile of corner cases around logging
4591 * blocks that have been removed from the tree.
4593 * See LOG_INODE_ALL and related defines for a description of what inode_only
4596 * This handles both files and directories.
4598 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
4599 struct btrfs_root
*root
, struct inode
*inode
,
4603 struct btrfs_log_ctx
*ctx
)
4605 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4606 struct btrfs_path
*path
;
4607 struct btrfs_path
*dst_path
;
4608 struct btrfs_key min_key
;
4609 struct btrfs_key max_key
;
4610 struct btrfs_root
*log
= root
->log_root
;
4611 struct extent_buffer
*src
= NULL
;
4612 LIST_HEAD(logged_list
);
4613 u64 last_extent
= 0;
4617 int ins_start_slot
= 0;
4619 bool fast_search
= false;
4620 u64 ino
= btrfs_ino(inode
);
4621 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4622 u64 logged_isize
= 0;
4623 bool need_log_inode_item
= true;
4625 path
= btrfs_alloc_path();
4628 dst_path
= btrfs_alloc_path();
4630 btrfs_free_path(path
);
4634 min_key
.objectid
= ino
;
4635 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
4638 max_key
.objectid
= ino
;
4641 /* today the code can only do partial logging of directories */
4642 if (S_ISDIR(inode
->i_mode
) ||
4643 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4644 &BTRFS_I(inode
)->runtime_flags
) &&
4645 inode_only
>= LOG_INODE_EXISTS
))
4646 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
4648 max_key
.type
= (u8
)-1;
4649 max_key
.offset
= (u64
)-1;
4652 * Only run delayed items if we are a dir or a new file.
4653 * Otherwise commit the delayed inode only, which is needed in
4654 * order for the log replay code to mark inodes for link count
4655 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4657 if (S_ISDIR(inode
->i_mode
) ||
4658 BTRFS_I(inode
)->generation
> fs_info
->last_trans_committed
)
4659 ret
= btrfs_commit_inode_delayed_items(trans
, inode
);
4661 ret
= btrfs_commit_inode_delayed_inode(inode
);
4664 btrfs_free_path(path
);
4665 btrfs_free_path(dst_path
);
4669 if (inode_only
== LOG_OTHER_INODE
) {
4670 inode_only
= LOG_INODE_EXISTS
;
4671 mutex_lock_nested(&BTRFS_I(inode
)->log_mutex
,
4672 SINGLE_DEPTH_NESTING
);
4674 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
4678 * a brute force approach to making sure we get the most uptodate
4679 * copies of everything.
4681 if (S_ISDIR(inode
->i_mode
)) {
4682 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
4684 if (inode_only
== LOG_INODE_EXISTS
)
4685 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
4686 ret
= drop_objectid_items(trans
, log
, path
, ino
, max_key_type
);
4688 if (inode_only
== LOG_INODE_EXISTS
) {
4690 * Make sure the new inode item we write to the log has
4691 * the same isize as the current one (if it exists).
4692 * This is necessary to prevent data loss after log
4693 * replay, and also to prevent doing a wrong expanding
4694 * truncate - for e.g. create file, write 4K into offset
4695 * 0, fsync, write 4K into offset 4096, add hard link,
4696 * fsync some other file (to sync log), power fail - if
4697 * we use the inode's current i_size, after log replay
4698 * we get a 8Kb file, with the last 4Kb extent as a hole
4699 * (zeroes), as if an expanding truncate happened,
4700 * instead of getting a file of 4Kb only.
4702 err
= logged_inode_size(log
, inode
, path
,
4707 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4708 &BTRFS_I(inode
)->runtime_flags
)) {
4709 if (inode_only
== LOG_INODE_EXISTS
) {
4710 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
4711 ret
= drop_objectid_items(trans
, log
, path
, ino
,
4714 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4715 &BTRFS_I(inode
)->runtime_flags
);
4716 clear_bit(BTRFS_INODE_COPY_EVERYTHING
,
4717 &BTRFS_I(inode
)->runtime_flags
);
4719 ret
= btrfs_truncate_inode_items(trans
,
4725 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING
,
4726 &BTRFS_I(inode
)->runtime_flags
) ||
4727 inode_only
== LOG_INODE_EXISTS
) {
4728 if (inode_only
== LOG_INODE_ALL
)
4730 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
4731 ret
= drop_objectid_items(trans
, log
, path
, ino
,
4734 if (inode_only
== LOG_INODE_ALL
)
4747 ret
= btrfs_search_forward(root
, &min_key
,
4748 path
, trans
->transid
);
4756 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4757 if (min_key
.objectid
!= ino
)
4759 if (min_key
.type
> max_key
.type
)
4762 if (min_key
.type
== BTRFS_INODE_ITEM_KEY
)
4763 need_log_inode_item
= false;
4765 if ((min_key
.type
== BTRFS_INODE_REF_KEY
||
4766 min_key
.type
== BTRFS_INODE_EXTREF_KEY
) &&
4767 BTRFS_I(inode
)->generation
== trans
->transid
) {
4770 ret
= btrfs_check_ref_name_override(path
->nodes
[0],
4777 } else if (ret
> 0 && ctx
&&
4778 other_ino
!= btrfs_ino(ctx
->inode
)) {
4779 struct btrfs_key inode_key
;
4780 struct inode
*other_inode
;
4786 ins_start_slot
= path
->slots
[0];
4788 ret
= copy_items(trans
, inode
, dst_path
, path
,
4789 &last_extent
, ins_start_slot
,
4797 btrfs_release_path(path
);
4798 inode_key
.objectid
= other_ino
;
4799 inode_key
.type
= BTRFS_INODE_ITEM_KEY
;
4800 inode_key
.offset
= 0;
4801 other_inode
= btrfs_iget(fs_info
->sb
,
4805 * If the other inode that had a conflicting dir
4806 * entry was deleted in the current transaction,
4807 * we don't need to do more work nor fallback to
4808 * a transaction commit.
4810 if (IS_ERR(other_inode
) &&
4811 PTR_ERR(other_inode
) == -ENOENT
) {
4813 } else if (IS_ERR(other_inode
)) {
4814 err
= PTR_ERR(other_inode
);
4818 * We are safe logging the other inode without
4819 * acquiring its i_mutex as long as we log with
4820 * the LOG_INODE_EXISTS mode. We're safe against
4821 * concurrent renames of the other inode as well
4822 * because during a rename we pin the log and
4823 * update the log with the new name before we
4826 err
= btrfs_log_inode(trans
, root
, other_inode
,
4837 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4838 if (min_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
4841 ret
= copy_items(trans
, inode
, dst_path
, path
,
4842 &last_extent
, ins_start_slot
,
4843 ins_nr
, inode_only
, logged_isize
);
4850 btrfs_release_path(path
);
4856 src
= path
->nodes
[0];
4857 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
4860 } else if (!ins_nr
) {
4861 ins_start_slot
= path
->slots
[0];
4866 ret
= copy_items(trans
, inode
, dst_path
, path
, &last_extent
,
4867 ins_start_slot
, ins_nr
, inode_only
,
4875 btrfs_release_path(path
);
4879 ins_start_slot
= path
->slots
[0];
4882 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4884 if (path
->slots
[0] < nritems
) {
4885 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
4890 ret
= copy_items(trans
, inode
, dst_path
, path
,
4891 &last_extent
, ins_start_slot
,
4892 ins_nr
, inode_only
, logged_isize
);
4900 btrfs_release_path(path
);
4902 if (min_key
.offset
< (u64
)-1) {
4904 } else if (min_key
.type
< max_key
.type
) {
4912 ret
= copy_items(trans
, inode
, dst_path
, path
, &last_extent
,
4913 ins_start_slot
, ins_nr
, inode_only
,
4923 btrfs_release_path(path
);
4924 btrfs_release_path(dst_path
);
4925 err
= btrfs_log_all_xattrs(trans
, root
, inode
, path
, dst_path
);
4928 if (max_key
.type
>= BTRFS_EXTENT_DATA_KEY
&& !fast_search
) {
4929 btrfs_release_path(path
);
4930 btrfs_release_path(dst_path
);
4931 err
= btrfs_log_trailing_hole(trans
, root
, inode
, path
);
4936 btrfs_release_path(path
);
4937 btrfs_release_path(dst_path
);
4938 if (need_log_inode_item
) {
4939 err
= log_inode_item(trans
, log
, dst_path
, inode
);
4944 ret
= btrfs_log_changed_extents(trans
, root
, inode
, dst_path
,
4945 &logged_list
, ctx
, start
, end
);
4950 } else if (inode_only
== LOG_INODE_ALL
) {
4951 struct extent_map
*em
, *n
;
4953 write_lock(&em_tree
->lock
);
4955 * We can't just remove every em if we're called for a ranged
4956 * fsync - that is, one that doesn't cover the whole possible
4957 * file range (0 to LLONG_MAX). This is because we can have
4958 * em's that fall outside the range we're logging and therefore
4959 * their ordered operations haven't completed yet
4960 * (btrfs_finish_ordered_io() not invoked yet). This means we
4961 * didn't get their respective file extent item in the fs/subvol
4962 * tree yet, and need to let the next fast fsync (one which
4963 * consults the list of modified extent maps) find the em so
4964 * that it logs a matching file extent item and waits for the
4965 * respective ordered operation to complete (if it's still
4968 * Removing every em outside the range we're logging would make
4969 * the next fast fsync not log their matching file extent items,
4970 * therefore making us lose data after a log replay.
4972 list_for_each_entry_safe(em
, n
, &em_tree
->modified_extents
,
4974 const u64 mod_end
= em
->mod_start
+ em
->mod_len
- 1;
4976 if (em
->mod_start
>= start
&& mod_end
<= end
)
4977 list_del_init(&em
->list
);
4979 write_unlock(&em_tree
->lock
);
4982 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
4983 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
,
4991 spin_lock(&BTRFS_I(inode
)->lock
);
4992 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
4993 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->last_sub_trans
;
4994 spin_unlock(&BTRFS_I(inode
)->lock
);
4997 btrfs_put_logged_extents(&logged_list
);
4999 btrfs_submit_logged_extents(&logged_list
, log
);
5000 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
5002 btrfs_free_path(path
);
5003 btrfs_free_path(dst_path
);
5008 * Check if we must fallback to a transaction commit when logging an inode.
5009 * This must be called after logging the inode and is used only in the context
5010 * when fsyncing an inode requires the need to log some other inode - in which
5011 * case we can't lock the i_mutex of each other inode we need to log as that
5012 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5013 * log inodes up or down in the hierarchy) or rename operations for example. So
5014 * we take the log_mutex of the inode after we have logged it and then check for
5015 * its last_unlink_trans value - this is safe because any task setting
5016 * last_unlink_trans must take the log_mutex and it must do this before it does
5017 * the actual unlink operation, so if we do this check before a concurrent task
5018 * sets last_unlink_trans it means we've logged a consistent version/state of
5019 * all the inode items, otherwise we are not sure and must do a transaction
5020 * commit (the concurrent task might have only updated last_unlink_trans before
5021 * we logged the inode or it might have also done the unlink).
5023 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle
*trans
,
5024 struct inode
*inode
)
5026 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
5029 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
5030 if (BTRFS_I(inode
)->last_unlink_trans
> fs_info
->last_trans_committed
) {
5032 * Make sure any commits to the log are forced to be full
5035 btrfs_set_log_full_commit(fs_info
, trans
);
5038 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
5044 * follow the dentry parent pointers up the chain and see if any
5045 * of the directories in it require a full commit before they can
5046 * be logged. Returns zero if nothing special needs to be done or 1 if
5047 * a full commit is required.
5049 static noinline
int check_parent_dirs_for_sync(struct btrfs_trans_handle
*trans
,
5050 struct inode
*inode
,
5051 struct dentry
*parent
,
5052 struct super_block
*sb
,
5056 struct dentry
*old_parent
= NULL
;
5057 struct inode
*orig_inode
= inode
;
5060 * for regular files, if its inode is already on disk, we don't
5061 * have to worry about the parents at all. This is because
5062 * we can use the last_unlink_trans field to record renames
5063 * and other fun in this file.
5065 if (S_ISREG(inode
->i_mode
) &&
5066 BTRFS_I(inode
)->generation
<= last_committed
&&
5067 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
)
5070 if (!S_ISDIR(inode
->i_mode
)) {
5071 if (!parent
|| d_really_is_negative(parent
) || sb
!= parent
->d_sb
)
5073 inode
= d_inode(parent
);
5078 * If we are logging a directory then we start with our inode,
5079 * not our parent's inode, so we need to skip setting the
5080 * logged_trans so that further down in the log code we don't
5081 * think this inode has already been logged.
5083 if (inode
!= orig_inode
)
5084 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
5087 if (btrfs_must_commit_transaction(trans
, inode
)) {
5092 if (!parent
|| d_really_is_negative(parent
) || sb
!= parent
->d_sb
)
5095 if (IS_ROOT(parent
)) {
5096 inode
= d_inode(parent
);
5097 if (btrfs_must_commit_transaction(trans
, inode
))
5102 parent
= dget_parent(parent
);
5104 old_parent
= parent
;
5105 inode
= d_inode(parent
);
5113 struct btrfs_dir_list
{
5115 struct list_head list
;
5119 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5120 * details about the why it is needed.
5121 * This is a recursive operation - if an existing dentry corresponds to a
5122 * directory, that directory's new entries are logged too (same behaviour as
5123 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5124 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5125 * complains about the following circular lock dependency / possible deadlock:
5129 * lock(&type->i_mutex_dir_key#3/2);
5130 * lock(sb_internal#2);
5131 * lock(&type->i_mutex_dir_key#3/2);
5132 * lock(&sb->s_type->i_mutex_key#14);
5134 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5135 * sb_start_intwrite() in btrfs_start_transaction().
5136 * Not locking i_mutex of the inodes is still safe because:
5138 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5139 * that while logging the inode new references (names) are added or removed
5140 * from the inode, leaving the logged inode item with a link count that does
5141 * not match the number of logged inode reference items. This is fine because
5142 * at log replay time we compute the real number of links and correct the
5143 * link count in the inode item (see replay_one_buffer() and
5144 * link_to_fixup_dir());
5146 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5147 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5148 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5149 * has a size that doesn't match the sum of the lengths of all the logged
5150 * names. This does not result in a problem because if a dir_item key is
5151 * logged but its matching dir_index key is not logged, at log replay time we
5152 * don't use it to replay the respective name (see replay_one_name()). On the
5153 * other hand if only the dir_index key ends up being logged, the respective
5154 * name is added to the fs/subvol tree with both the dir_item and dir_index
5155 * keys created (see replay_one_name()).
5156 * The directory's inode item with a wrong i_size is not a problem as well,
5157 * since we don't use it at log replay time to set the i_size in the inode
5158 * item of the fs/subvol tree (see overwrite_item()).
5160 static int log_new_dir_dentries(struct btrfs_trans_handle
*trans
,
5161 struct btrfs_root
*root
,
5162 struct inode
*start_inode
,
5163 struct btrfs_log_ctx
*ctx
)
5165 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5166 struct btrfs_root
*log
= root
->log_root
;
5167 struct btrfs_path
*path
;
5168 LIST_HEAD(dir_list
);
5169 struct btrfs_dir_list
*dir_elem
;
5172 path
= btrfs_alloc_path();
5176 dir_elem
= kmalloc(sizeof(*dir_elem
), GFP_NOFS
);
5178 btrfs_free_path(path
);
5181 dir_elem
->ino
= btrfs_ino(start_inode
);
5182 list_add_tail(&dir_elem
->list
, &dir_list
);
5184 while (!list_empty(&dir_list
)) {
5185 struct extent_buffer
*leaf
;
5186 struct btrfs_key min_key
;
5190 dir_elem
= list_first_entry(&dir_list
, struct btrfs_dir_list
,
5193 goto next_dir_inode
;
5195 min_key
.objectid
= dir_elem
->ino
;
5196 min_key
.type
= BTRFS_DIR_ITEM_KEY
;
5199 btrfs_release_path(path
);
5200 ret
= btrfs_search_forward(log
, &min_key
, path
, trans
->transid
);
5202 goto next_dir_inode
;
5203 } else if (ret
> 0) {
5205 goto next_dir_inode
;
5209 leaf
= path
->nodes
[0];
5210 nritems
= btrfs_header_nritems(leaf
);
5211 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
5212 struct btrfs_dir_item
*di
;
5213 struct btrfs_key di_key
;
5214 struct inode
*di_inode
;
5215 struct btrfs_dir_list
*new_dir_elem
;
5216 int log_mode
= LOG_INODE_EXISTS
;
5219 btrfs_item_key_to_cpu(leaf
, &min_key
, i
);
5220 if (min_key
.objectid
!= dir_elem
->ino
||
5221 min_key
.type
!= BTRFS_DIR_ITEM_KEY
)
5222 goto next_dir_inode
;
5224 di
= btrfs_item_ptr(leaf
, i
, struct btrfs_dir_item
);
5225 type
= btrfs_dir_type(leaf
, di
);
5226 if (btrfs_dir_transid(leaf
, di
) < trans
->transid
&&
5227 type
!= BTRFS_FT_DIR
)
5229 btrfs_dir_item_key_to_cpu(leaf
, di
, &di_key
);
5230 if (di_key
.type
== BTRFS_ROOT_ITEM_KEY
)
5233 btrfs_release_path(path
);
5234 di_inode
= btrfs_iget(fs_info
->sb
, &di_key
, root
, NULL
);
5235 if (IS_ERR(di_inode
)) {
5236 ret
= PTR_ERR(di_inode
);
5237 goto next_dir_inode
;
5240 if (btrfs_inode_in_log(di_inode
, trans
->transid
)) {
5245 ctx
->log_new_dentries
= false;
5246 if (type
== BTRFS_FT_DIR
|| type
== BTRFS_FT_SYMLINK
)
5247 log_mode
= LOG_INODE_ALL
;
5248 ret
= btrfs_log_inode(trans
, root
, di_inode
,
5249 log_mode
, 0, LLONG_MAX
, ctx
);
5251 btrfs_must_commit_transaction(trans
, di_inode
))
5255 goto next_dir_inode
;
5256 if (ctx
->log_new_dentries
) {
5257 new_dir_elem
= kmalloc(sizeof(*new_dir_elem
),
5259 if (!new_dir_elem
) {
5261 goto next_dir_inode
;
5263 new_dir_elem
->ino
= di_key
.objectid
;
5264 list_add_tail(&new_dir_elem
->list
, &dir_list
);
5269 ret
= btrfs_next_leaf(log
, path
);
5271 goto next_dir_inode
;
5272 } else if (ret
> 0) {
5274 goto next_dir_inode
;
5278 if (min_key
.offset
< (u64
)-1) {
5283 list_del(&dir_elem
->list
);
5287 btrfs_free_path(path
);
5291 static int btrfs_log_all_parents(struct btrfs_trans_handle
*trans
,
5292 struct inode
*inode
,
5293 struct btrfs_log_ctx
*ctx
)
5295 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5297 struct btrfs_path
*path
;
5298 struct btrfs_key key
;
5299 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5300 const u64 ino
= btrfs_ino(inode
);
5302 path
= btrfs_alloc_path();
5305 path
->skip_locking
= 1;
5306 path
->search_commit_root
= 1;
5309 key
.type
= BTRFS_INODE_REF_KEY
;
5311 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5316 struct extent_buffer
*leaf
= path
->nodes
[0];
5317 int slot
= path
->slots
[0];
5322 if (slot
>= btrfs_header_nritems(leaf
)) {
5323 ret
= btrfs_next_leaf(root
, path
);
5331 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5332 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5333 if (key
.objectid
!= ino
|| key
.type
> BTRFS_INODE_EXTREF_KEY
)
5336 item_size
= btrfs_item_size_nr(leaf
, slot
);
5337 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
5338 while (cur_offset
< item_size
) {
5339 struct btrfs_key inode_key
;
5340 struct inode
*dir_inode
;
5342 inode_key
.type
= BTRFS_INODE_ITEM_KEY
;
5343 inode_key
.offset
= 0;
5345 if (key
.type
== BTRFS_INODE_EXTREF_KEY
) {
5346 struct btrfs_inode_extref
*extref
;
5348 extref
= (struct btrfs_inode_extref
*)
5350 inode_key
.objectid
= btrfs_inode_extref_parent(
5352 cur_offset
+= sizeof(*extref
);
5353 cur_offset
+= btrfs_inode_extref_name_len(leaf
,
5356 inode_key
.objectid
= key
.offset
;
5357 cur_offset
= item_size
;
5360 dir_inode
= btrfs_iget(fs_info
->sb
, &inode_key
,
5362 /* If parent inode was deleted, skip it. */
5363 if (IS_ERR(dir_inode
))
5367 ctx
->log_new_dentries
= false;
5368 ret
= btrfs_log_inode(trans
, root
, dir_inode
,
5369 LOG_INODE_ALL
, 0, LLONG_MAX
, ctx
);
5371 btrfs_must_commit_transaction(trans
, dir_inode
))
5373 if (!ret
&& ctx
&& ctx
->log_new_dentries
)
5374 ret
= log_new_dir_dentries(trans
, root
,
5384 btrfs_free_path(path
);
5389 * helper function around btrfs_log_inode to make sure newly created
5390 * parent directories also end up in the log. A minimal inode and backref
5391 * only logging is done of any parent directories that are older than
5392 * the last committed transaction
5394 static int btrfs_log_inode_parent(struct btrfs_trans_handle
*trans
,
5395 struct btrfs_root
*root
, struct inode
*inode
,
5396 struct dentry
*parent
,
5400 struct btrfs_log_ctx
*ctx
)
5402 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5403 int inode_only
= exists_only
? LOG_INODE_EXISTS
: LOG_INODE_ALL
;
5404 struct super_block
*sb
;
5405 struct dentry
*old_parent
= NULL
;
5407 u64 last_committed
= fs_info
->last_trans_committed
;
5408 bool log_dentries
= false;
5409 struct inode
*orig_inode
= inode
;
5413 if (btrfs_test_opt(fs_info
, NOTREELOG
)) {
5419 * The prev transaction commit doesn't complete, we need do
5420 * full commit by ourselves.
5422 if (fs_info
->last_trans_log_full_commit
>
5423 fs_info
->last_trans_committed
) {
5428 if (root
!= BTRFS_I(inode
)->root
||
5429 btrfs_root_refs(&root
->root_item
) == 0) {
5434 ret
= check_parent_dirs_for_sync(trans
, inode
, parent
,
5435 sb
, last_committed
);
5439 if (btrfs_inode_in_log(inode
, trans
->transid
)) {
5440 ret
= BTRFS_NO_LOG_SYNC
;
5444 ret
= start_log_trans(trans
, root
, ctx
);
5448 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
, start
, end
, ctx
);
5453 * for regular files, if its inode is already on disk, we don't
5454 * have to worry about the parents at all. This is because
5455 * we can use the last_unlink_trans field to record renames
5456 * and other fun in this file.
5458 if (S_ISREG(inode
->i_mode
) &&
5459 BTRFS_I(inode
)->generation
<= last_committed
&&
5460 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
) {
5465 if (S_ISDIR(inode
->i_mode
) && ctx
&& ctx
->log_new_dentries
)
5466 log_dentries
= true;
5469 * On unlink we must make sure all our current and old parent directory
5470 * inodes are fully logged. This is to prevent leaving dangling
5471 * directory index entries in directories that were our parents but are
5472 * not anymore. Not doing this results in old parent directory being
5473 * impossible to delete after log replay (rmdir will always fail with
5474 * error -ENOTEMPTY).
5480 * ln testdir/foo testdir/bar
5482 * unlink testdir/bar
5483 * xfs_io -c fsync testdir/foo
5485 * mount fs, triggers log replay
5487 * If we don't log the parent directory (testdir), after log replay the
5488 * directory still has an entry pointing to the file inode using the bar
5489 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5490 * the file inode has a link count of 1.
5496 * ln foo testdir/foo2
5497 * ln foo testdir/foo3
5499 * unlink testdir/foo3
5500 * xfs_io -c fsync foo
5502 * mount fs, triggers log replay
5504 * Similar as the first example, after log replay the parent directory
5505 * testdir still has an entry pointing to the inode file with name foo3
5506 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5507 * and has a link count of 2.
5509 if (BTRFS_I(inode
)->last_unlink_trans
> last_committed
) {
5510 ret
= btrfs_log_all_parents(trans
, orig_inode
, ctx
);
5516 if (!parent
|| d_really_is_negative(parent
) || sb
!= parent
->d_sb
)
5519 inode
= d_inode(parent
);
5520 if (root
!= BTRFS_I(inode
)->root
)
5523 if (BTRFS_I(inode
)->generation
> last_committed
) {
5524 ret
= btrfs_log_inode(trans
, root
, inode
,
5530 if (IS_ROOT(parent
))
5533 parent
= dget_parent(parent
);
5535 old_parent
= parent
;
5538 ret
= log_new_dir_dentries(trans
, root
, orig_inode
, ctx
);
5544 btrfs_set_log_full_commit(fs_info
, trans
);
5549 btrfs_remove_log_ctx(root
, ctx
);
5550 btrfs_end_log_trans(root
);
5556 * it is not safe to log dentry if the chunk root has added new
5557 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5558 * If this returns 1, you must commit the transaction to safely get your
5561 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
5562 struct btrfs_root
*root
, struct dentry
*dentry
,
5565 struct btrfs_log_ctx
*ctx
)
5567 struct dentry
*parent
= dget_parent(dentry
);
5570 ret
= btrfs_log_inode_parent(trans
, root
, d_inode(dentry
), parent
,
5571 start
, end
, 0, ctx
);
5578 * should be called during mount to recover any replay any log trees
5581 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
5584 struct btrfs_path
*path
;
5585 struct btrfs_trans_handle
*trans
;
5586 struct btrfs_key key
;
5587 struct btrfs_key found_key
;
5588 struct btrfs_key tmp_key
;
5589 struct btrfs_root
*log
;
5590 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
5591 struct walk_control wc
= {
5592 .process_func
= process_one_buffer
,
5596 path
= btrfs_alloc_path();
5600 set_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
);
5602 trans
= btrfs_start_transaction(fs_info
->tree_root
, 0);
5603 if (IS_ERR(trans
)) {
5604 ret
= PTR_ERR(trans
);
5611 ret
= walk_log_tree(trans
, log_root_tree
, &wc
);
5613 btrfs_handle_fs_error(fs_info
, ret
,
5614 "Failed to pin buffers while recovering log root tree.");
5619 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
5620 key
.offset
= (u64
)-1;
5621 key
.type
= BTRFS_ROOT_ITEM_KEY
;
5624 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
5627 btrfs_handle_fs_error(fs_info
, ret
,
5628 "Couldn't find tree log root.");
5632 if (path
->slots
[0] == 0)
5636 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
5638 btrfs_release_path(path
);
5639 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
5642 log
= btrfs_read_fs_root(log_root_tree
, &found_key
);
5645 btrfs_handle_fs_error(fs_info
, ret
,
5646 "Couldn't read tree log root.");
5650 tmp_key
.objectid
= found_key
.offset
;
5651 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5652 tmp_key
.offset
= (u64
)-1;
5654 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
5655 if (IS_ERR(wc
.replay_dest
)) {
5656 ret
= PTR_ERR(wc
.replay_dest
);
5657 free_extent_buffer(log
->node
);
5658 free_extent_buffer(log
->commit_root
);
5660 btrfs_handle_fs_error(fs_info
, ret
,
5661 "Couldn't read target root for tree log recovery.");
5665 wc
.replay_dest
->log_root
= log
;
5666 btrfs_record_root_in_trans(trans
, wc
.replay_dest
);
5667 ret
= walk_log_tree(trans
, log
, &wc
);
5669 if (!ret
&& wc
.stage
== LOG_WALK_REPLAY_ALL
) {
5670 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
5674 key
.offset
= found_key
.offset
- 1;
5675 wc
.replay_dest
->log_root
= NULL
;
5676 free_extent_buffer(log
->node
);
5677 free_extent_buffer(log
->commit_root
);
5683 if (found_key
.offset
== 0)
5686 btrfs_release_path(path
);
5688 /* step one is to pin it all, step two is to replay just inodes */
5691 wc
.process_func
= replay_one_buffer
;
5692 wc
.stage
= LOG_WALK_REPLAY_INODES
;
5695 /* step three is to replay everything */
5696 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
5701 btrfs_free_path(path
);
5703 /* step 4: commit the transaction, which also unpins the blocks */
5704 ret
= btrfs_commit_transaction(trans
);
5708 free_extent_buffer(log_root_tree
->node
);
5709 log_root_tree
->log_root
= NULL
;
5710 clear_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
);
5711 kfree(log_root_tree
);
5716 btrfs_end_transaction(wc
.trans
);
5717 btrfs_free_path(path
);
5722 * there are some corner cases where we want to force a full
5723 * commit instead of allowing a directory to be logged.
5725 * They revolve around files there were unlinked from the directory, and
5726 * this function updates the parent directory so that a full commit is
5727 * properly done if it is fsync'd later after the unlinks are done.
5729 * Must be called before the unlink operations (updates to the subvolume tree,
5730 * inodes, etc) are done.
5732 void btrfs_record_unlink_dir(struct btrfs_trans_handle
*trans
,
5733 struct inode
*dir
, struct inode
*inode
,
5737 * when we're logging a file, if it hasn't been renamed
5738 * or unlinked, and its inode is fully committed on disk,
5739 * we don't have to worry about walking up the directory chain
5740 * to log its parents.
5742 * So, we use the last_unlink_trans field to put this transid
5743 * into the file. When the file is logged we check it and
5744 * don't log the parents if the file is fully on disk.
5746 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
5747 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
5748 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
5751 * if this directory was already logged any new
5752 * names for this file/dir will get recorded
5755 if (BTRFS_I(dir
)->logged_trans
== trans
->transid
)
5759 * if the inode we're about to unlink was logged,
5760 * the log will be properly updated for any new names
5762 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
)
5766 * when renaming files across directories, if the directory
5767 * there we're unlinking from gets fsync'd later on, there's
5768 * no way to find the destination directory later and fsync it
5769 * properly. So, we have to be conservative and force commits
5770 * so the new name gets discovered.
5775 /* we can safely do the unlink without any special recording */
5779 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
5780 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
5781 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
5785 * Make sure that if someone attempts to fsync the parent directory of a deleted
5786 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5787 * that after replaying the log tree of the parent directory's root we will not
5788 * see the snapshot anymore and at log replay time we will not see any log tree
5789 * corresponding to the deleted snapshot's root, which could lead to replaying
5790 * it after replaying the log tree of the parent directory (which would replay
5791 * the snapshot delete operation).
5793 * Must be called before the actual snapshot destroy operation (updates to the
5794 * parent root and tree of tree roots trees, etc) are done.
5796 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle
*trans
,
5799 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
5800 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
5801 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
5805 * Call this after adding a new name for a file and it will properly
5806 * update the log to reflect the new name.
5808 * It will return zero if all goes well, and it will return 1 if a
5809 * full transaction commit is required.
5811 int btrfs_log_new_name(struct btrfs_trans_handle
*trans
,
5812 struct inode
*inode
, struct inode
*old_dir
,
5813 struct dentry
*parent
)
5815 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5816 struct btrfs_root
* root
= BTRFS_I(inode
)->root
;
5819 * this will force the logging code to walk the dentry chain
5822 if (S_ISREG(inode
->i_mode
))
5823 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
5826 * if this inode hasn't been logged and directory we're renaming it
5827 * from hasn't been logged, we don't need to log it
5829 if (BTRFS_I(inode
)->logged_trans
<=
5830 fs_info
->last_trans_committed
&&
5831 (!old_dir
|| BTRFS_I(old_dir
)->logged_trans
<=
5832 fs_info
->last_trans_committed
))
5835 return btrfs_log_inode_parent(trans
, root
, inode
, parent
, 0,
5836 LLONG_MAX
, 1, NULL
);