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>
21 #include "transaction.h"
24 #include "print-tree.h"
27 /* magic values for the inode_only field in btrfs_log_inode:
29 * LOG_INODE_ALL means to log everything
30 * LOG_INODE_EXISTS means to log just enough to recreate the inode
33 #define LOG_INODE_ALL 0
34 #define LOG_INODE_EXISTS 1
37 * stages for the tree walking. The first
38 * stage (0) is to only pin down the blocks we find
39 * the second stage (1) is to make sure that all the inodes
40 * we find in the log are created in the subvolume.
42 * The last stage is to deal with directories and links and extents
43 * and all the other fun semantics
45 #define LOG_WALK_PIN_ONLY 0
46 #define LOG_WALK_REPLAY_INODES 1
47 #define LOG_WALK_REPLAY_ALL 2
49 static int __btrfs_log_inode(struct btrfs_trans_handle
*trans
,
50 struct btrfs_root
*root
, struct inode
*inode
,
54 * tree logging is a special write ahead log used to make sure that
55 * fsyncs and O_SYNCs can happen without doing full tree commits.
57 * Full tree commits are expensive because they require commonly
58 * modified blocks to be recowed, creating many dirty pages in the
59 * extent tree an 4x-6x higher write load than ext3.
61 * Instead of doing a tree commit on every fsync, we use the
62 * key ranges and transaction ids to find items for a given file or directory
63 * that have changed in this transaction. Those items are copied into
64 * a special tree (one per subvolume root), that tree is written to disk
65 * and then the fsync is considered complete.
67 * After a crash, items are copied out of the log-tree back into the
68 * subvolume tree. Any file data extents found are recorded in the extent
69 * allocation tree, and the log-tree freed.
71 * The log tree is read three times, once to pin down all the extents it is
72 * using in ram and once, once to create all the inodes logged in the tree
73 * and once to do all the other items.
77 * btrfs_add_log_tree adds a new per-subvolume log tree into the
78 * tree of log tree roots. This must be called with a tree log transaction
79 * running (see start_log_trans).
81 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
82 struct btrfs_root
*root
)
85 struct btrfs_root_item root_item
;
86 struct btrfs_inode_item
*inode_item
;
87 struct extent_buffer
*leaf
;
88 struct btrfs_root
*new_root
= root
;
90 u64 objectid
= root
->root_key
.objectid
;
92 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
93 BTRFS_TREE_LOG_OBJECTID
,
100 btrfs_set_header_nritems(leaf
, 0);
101 btrfs_set_header_level(leaf
, 0);
102 btrfs_set_header_bytenr(leaf
, leaf
->start
);
103 btrfs_set_header_generation(leaf
, trans
->transid
);
104 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
106 write_extent_buffer(leaf
, root
->fs_info
->fsid
,
107 (unsigned long)btrfs_header_fsid(leaf
),
109 btrfs_mark_buffer_dirty(leaf
);
111 inode_item
= &root_item
.inode
;
112 memset(inode_item
, 0, sizeof(*inode_item
));
113 inode_item
->generation
= cpu_to_le64(1);
114 inode_item
->size
= cpu_to_le64(3);
115 inode_item
->nlink
= cpu_to_le32(1);
116 inode_item
->nblocks
= cpu_to_le64(1);
117 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
119 btrfs_set_root_bytenr(&root_item
, leaf
->start
);
120 btrfs_set_root_level(&root_item
, 0);
121 btrfs_set_root_refs(&root_item
, 0);
122 btrfs_set_root_used(&root_item
, 0);
124 memset(&root_item
.drop_progress
, 0, sizeof(root_item
.drop_progress
));
125 root_item
.drop_level
= 0;
127 btrfs_tree_unlock(leaf
);
128 free_extent_buffer(leaf
);
131 btrfs_set_root_dirid(&root_item
, 0);
133 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
134 key
.offset
= objectid
;
135 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
136 ret
= btrfs_insert_root(trans
, root
->fs_info
->log_root_tree
, &key
,
141 new_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->log_root_tree
,
145 WARN_ON(root
->log_root
);
146 root
->log_root
= new_root
;
149 * log trees do not get reference counted because they go away
150 * before a real commit is actually done. They do store pointers
151 * to file data extents, and those reference counts still get
152 * updated (along with back refs to the log tree).
154 new_root
->ref_cows
= 0;
155 new_root
->last_trans
= trans
->transid
;
161 * start a sub transaction and setup the log tree
162 * this increments the log tree writer count to make the people
163 * syncing the tree wait for us to finish
165 static int start_log_trans(struct btrfs_trans_handle
*trans
,
166 struct btrfs_root
*root
)
169 mutex_lock(&root
->fs_info
->tree_log_mutex
);
170 if (!root
->fs_info
->log_root_tree
) {
171 ret
= btrfs_init_log_root_tree(trans
, root
->fs_info
);
174 if (!root
->log_root
) {
175 ret
= btrfs_add_log_tree(trans
, root
);
178 atomic_inc(&root
->fs_info
->tree_log_writers
);
179 root
->fs_info
->tree_log_batch
++;
180 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
185 * returns 0 if there was a log transaction running and we were able
186 * to join, or returns -ENOENT if there were not transactions
189 static int join_running_log_trans(struct btrfs_root
*root
)
197 mutex_lock(&root
->fs_info
->tree_log_mutex
);
198 if (root
->log_root
) {
200 atomic_inc(&root
->fs_info
->tree_log_writers
);
201 root
->fs_info
->tree_log_batch
++;
203 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
208 * indicate we're done making changes to the log tree
209 * and wake up anyone waiting to do a sync
211 static int end_log_trans(struct btrfs_root
*root
)
213 atomic_dec(&root
->fs_info
->tree_log_writers
);
215 if (waitqueue_active(&root
->fs_info
->tree_log_wait
))
216 wake_up(&root
->fs_info
->tree_log_wait
);
222 * the walk control struct is used to pass state down the chain when
223 * processing the log tree. The stage field tells us which part
224 * of the log tree processing we are currently doing. The others
225 * are state fields used for that specific part
227 struct walk_control
{
228 /* should we free the extent on disk when done? This is used
229 * at transaction commit time while freeing a log tree
233 /* should we write out the extent buffer? This is used
234 * while flushing the log tree to disk during a sync
238 /* should we wait for the extent buffer io to finish? Also used
239 * while flushing the log tree to disk for a sync
243 /* pin only walk, we record which extents on disk belong to the
248 /* what stage of the replay code we're currently in */
251 /* the root we are currently replaying */
252 struct btrfs_root
*replay_dest
;
254 /* the trans handle for the current replay */
255 struct btrfs_trans_handle
*trans
;
257 /* the function that gets used to process blocks we find in the
258 * tree. Note the extent_buffer might not be up to date when it is
259 * passed in, and it must be checked or read if you need the data
262 int (*process_func
)(struct btrfs_root
*log
, struct extent_buffer
*eb
,
263 struct walk_control
*wc
, u64 gen
);
267 * process_func used to pin down extents, write them or wait on them
269 static int process_one_buffer(struct btrfs_root
*log
,
270 struct extent_buffer
*eb
,
271 struct walk_control
*wc
, u64 gen
)
274 mutex_lock(&log
->fs_info
->alloc_mutex
);
275 btrfs_update_pinned_extents(log
->fs_info
->extent_root
,
276 eb
->start
, eb
->len
, 1);
277 mutex_unlock(&log
->fs_info
->alloc_mutex
);
280 if (btrfs_buffer_uptodate(eb
, gen
)) {
282 btrfs_write_tree_block(eb
);
284 btrfs_wait_tree_block_writeback(eb
);
290 * Item overwrite used by replay and tree logging. eb, slot and key all refer
291 * to the src data we are copying out.
293 * root is the tree we are copying into, and path is a scratch
294 * path for use in this function (it should be released on entry and
295 * will be released on exit).
297 * If the key is already in the destination tree the existing item is
298 * overwritten. If the existing item isn't big enough, it is extended.
299 * If it is too large, it is truncated.
301 * If the key isn't in the destination yet, a new item is inserted.
303 static noinline
int overwrite_item(struct btrfs_trans_handle
*trans
,
304 struct btrfs_root
*root
,
305 struct btrfs_path
*path
,
306 struct extent_buffer
*eb
, int slot
,
307 struct btrfs_key
*key
)
311 u64 saved_i_size
= 0;
312 int save_old_i_size
= 0;
313 unsigned long src_ptr
;
314 unsigned long dst_ptr
;
315 int overwrite_root
= 0;
317 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
320 item_size
= btrfs_item_size_nr(eb
, slot
);
321 src_ptr
= btrfs_item_ptr_offset(eb
, slot
);
323 /* look for the key in the destination tree */
324 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
328 u32 dst_size
= btrfs_item_size_nr(path
->nodes
[0],
330 if (dst_size
!= item_size
)
333 if (item_size
== 0) {
334 btrfs_release_path(root
, path
);
337 dst_copy
= kmalloc(item_size
, GFP_NOFS
);
338 src_copy
= kmalloc(item_size
, GFP_NOFS
);
340 read_extent_buffer(eb
, src_copy
, src_ptr
, item_size
);
342 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
343 read_extent_buffer(path
->nodes
[0], dst_copy
, dst_ptr
,
345 ret
= memcmp(dst_copy
, src_copy
, item_size
);
350 * they have the same contents, just return, this saves
351 * us from cowing blocks in the destination tree and doing
352 * extra writes that may not have been done by a previous
356 btrfs_release_path(root
, path
);
362 btrfs_release_path(root
, path
);
363 /* try to insert the key into the destination tree */
364 ret
= btrfs_insert_empty_item(trans
, root
, path
,
367 /* make sure any existing item is the correct size */
368 if (ret
== -EEXIST
) {
370 found_size
= btrfs_item_size_nr(path
->nodes
[0],
372 if (found_size
> item_size
) {
373 btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
374 } else if (found_size
< item_size
) {
375 ret
= btrfs_del_item(trans
, root
,
379 btrfs_release_path(root
, path
);
380 ret
= btrfs_insert_empty_item(trans
,
381 root
, path
, key
, item_size
);
387 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0],
390 /* don't overwrite an existing inode if the generation number
391 * was logged as zero. This is done when the tree logging code
392 * is just logging an inode to make sure it exists after recovery.
394 * Also, don't overwrite i_size on directories during replay.
395 * log replay inserts and removes directory items based on the
396 * state of the tree found in the subvolume, and i_size is modified
399 if (key
->type
== BTRFS_INODE_ITEM_KEY
&& ret
== -EEXIST
) {
400 struct btrfs_inode_item
*src_item
;
401 struct btrfs_inode_item
*dst_item
;
403 src_item
= (struct btrfs_inode_item
*)src_ptr
;
404 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
406 if (btrfs_inode_generation(eb
, src_item
) == 0)
409 if (overwrite_root
&&
410 S_ISDIR(btrfs_inode_mode(eb
, src_item
)) &&
411 S_ISDIR(btrfs_inode_mode(path
->nodes
[0], dst_item
))) {
413 saved_i_size
= btrfs_inode_size(path
->nodes
[0],
418 copy_extent_buffer(path
->nodes
[0], eb
, dst_ptr
,
421 if (save_old_i_size
) {
422 struct btrfs_inode_item
*dst_item
;
423 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
424 btrfs_set_inode_size(path
->nodes
[0], dst_item
, saved_i_size
);
427 /* make sure the generation is filled in */
428 if (key
->type
== BTRFS_INODE_ITEM_KEY
) {
429 struct btrfs_inode_item
*dst_item
;
430 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
431 if (btrfs_inode_generation(path
->nodes
[0], dst_item
) == 0) {
432 btrfs_set_inode_generation(path
->nodes
[0], dst_item
,
437 btrfs_mark_buffer_dirty(path
->nodes
[0]);
438 btrfs_release_path(root
, path
);
443 * simple helper to read an inode off the disk from a given root
444 * This can only be called for subvolume roots and not for the log
446 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
450 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
451 if (inode
->i_state
& I_NEW
) {
452 BTRFS_I(inode
)->root
= root
;
453 BTRFS_I(inode
)->location
.objectid
= objectid
;
454 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
455 BTRFS_I(inode
)->location
.offset
= 0;
456 btrfs_read_locked_inode(inode
);
457 unlock_new_inode(inode
);
460 if (is_bad_inode(inode
)) {
467 /* replays a single extent in 'eb' at 'slot' with 'key' into the
468 * subvolume 'root'. path is released on entry and should be released
471 * extents in the log tree have not been allocated out of the extent
472 * tree yet. So, this completes the allocation, taking a reference
473 * as required if the extent already exists or creating a new extent
474 * if it isn't in the extent allocation tree yet.
476 * The extent is inserted into the file, dropping any existing extents
477 * from the file that overlap the new one.
479 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
480 struct btrfs_root
*root
,
481 struct btrfs_path
*path
,
482 struct extent_buffer
*eb
, int slot
,
483 struct btrfs_key
*key
)
486 u64 mask
= root
->sectorsize
- 1;
489 u64 start
= key
->offset
;
490 struct btrfs_file_extent_item
*item
;
491 struct inode
*inode
= NULL
;
495 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
496 found_type
= btrfs_file_extent_type(eb
, item
);
498 if (found_type
== BTRFS_FILE_EXTENT_REG
)
499 extent_end
= start
+ btrfs_file_extent_num_bytes(eb
, item
);
500 else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
501 size
= btrfs_file_extent_inline_len(eb
,
502 btrfs_item_nr(eb
, slot
));
503 extent_end
= (start
+ size
+ mask
) & ~mask
;
509 inode
= read_one_inode(root
, key
->objectid
);
516 * first check to see if we already have this extent in the
517 * file. This must be done before the btrfs_drop_extents run
518 * so we don't try to drop this extent.
520 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
523 if (ret
== 0 && found_type
== BTRFS_FILE_EXTENT_REG
) {
524 struct btrfs_file_extent_item cmp1
;
525 struct btrfs_file_extent_item cmp2
;
526 struct btrfs_file_extent_item
*existing
;
527 struct extent_buffer
*leaf
;
529 leaf
= path
->nodes
[0];
530 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
531 struct btrfs_file_extent_item
);
533 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
535 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
539 * we already have a pointer to this exact extent,
540 * we don't have to do anything
542 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
543 btrfs_release_path(root
, path
);
547 btrfs_release_path(root
, path
);
549 /* drop any overlapping extents */
550 ret
= btrfs_drop_extents(trans
, root
, inode
,
551 start
, extent_end
, start
, &alloc_hint
);
555 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
556 struct btrfs_key ins
;
558 ins
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
559 ins
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
560 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
562 /* insert the extent pointer in the file */
563 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
567 * is this extent already allocated in the extent
568 * allocation tree? If so, just add a reference
570 ret
= btrfs_lookup_extent(root
, path
, ins
.objectid
, ins
.offset
);
571 btrfs_release_path(root
, path
);
573 ret
= btrfs_inc_extent_ref(trans
, root
,
574 ins
.objectid
, ins
.offset
,
575 root
->root_key
.objectid
,
576 trans
->transid
, key
->objectid
, start
);
579 * insert the extent pointer in the extent
582 ret
= btrfs_alloc_logged_extent(trans
, root
,
583 root
->root_key
.objectid
,
584 trans
->transid
, key
->objectid
,
588 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
589 /* inline extents are easy, we just overwrite them */
590 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
593 /* btrfs_drop_extents changes i_blocks, update it here */
594 inode
->i_blocks
+= (extent_end
- start
) >> 9;
595 btrfs_update_inode(trans
, root
, inode
);
603 * when cleaning up conflicts between the directory names in the
604 * subvolume, directory names in the log and directory names in the
605 * inode back references, we may have to unlink inodes from directories.
607 * This is a helper function to do the unlink of a specific directory
610 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
611 struct btrfs_root
*root
,
612 struct btrfs_path
*path
,
614 struct btrfs_dir_item
*di
)
619 struct extent_buffer
*leaf
;
620 struct btrfs_key location
;
623 leaf
= path
->nodes
[0];
625 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
626 name_len
= btrfs_dir_name_len(leaf
, di
);
627 name
= kmalloc(name_len
, GFP_NOFS
);
628 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
629 btrfs_release_path(root
, path
);
631 inode
= read_one_inode(root
, location
.objectid
);
634 btrfs_inc_nlink(inode
);
635 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
643 * helper function to see if a given name and sequence number found
644 * in an inode back reference are already in a directory and correctly
645 * point to this inode
647 static noinline
int inode_in_dir(struct btrfs_root
*root
,
648 struct btrfs_path
*path
,
649 u64 dirid
, u64 objectid
, u64 index
,
650 const char *name
, int name_len
)
652 struct btrfs_dir_item
*di
;
653 struct btrfs_key location
;
656 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
657 index
, name
, name_len
, 0);
658 if (di
&& !IS_ERR(di
)) {
659 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
660 if (location
.objectid
!= objectid
)
664 btrfs_release_path(root
, path
);
666 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
667 if (di
&& !IS_ERR(di
)) {
668 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
669 if (location
.objectid
!= objectid
)
675 btrfs_release_path(root
, path
);
680 * helper function to check a log tree for a named back reference in
681 * an inode. This is used to decide if a back reference that is
682 * found in the subvolume conflicts with what we find in the log.
684 * inode backreferences may have multiple refs in a single item,
685 * during replay we process one reference at a time, and we don't
686 * want to delete valid links to a file from the subvolume if that
687 * link is also in the log.
689 static noinline
int backref_in_log(struct btrfs_root
*log
,
690 struct btrfs_key
*key
,
691 char *name
, int namelen
)
693 struct btrfs_path
*path
;
694 struct btrfs_inode_ref
*ref
;
696 unsigned long ptr_end
;
697 unsigned long name_ptr
;
703 path
= btrfs_alloc_path();
704 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
708 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
709 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
710 ptr_end
= ptr
+ item_size
;
711 while (ptr
< ptr_end
) {
712 ref
= (struct btrfs_inode_ref
*)ptr
;
713 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
714 if (found_name_len
== namelen
) {
715 name_ptr
= (unsigned long)(ref
+ 1);
716 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
723 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
726 btrfs_free_path(path
);
732 * replay one inode back reference item found in the log tree.
733 * eb, slot and key refer to the buffer and key found in the log tree.
734 * root is the destination we are replaying into, and path is for temp
735 * use by this function. (it should be released on return).
737 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
738 struct btrfs_root
*root
,
739 struct btrfs_root
*log
,
740 struct btrfs_path
*path
,
741 struct extent_buffer
*eb
, int slot
,
742 struct btrfs_key
*key
)
746 struct btrfs_key location
;
747 struct btrfs_inode_ref
*ref
;
748 struct btrfs_dir_item
*di
;
752 unsigned long ref_ptr
;
753 unsigned long ref_end
;
755 location
.objectid
= key
->objectid
;
756 location
.type
= BTRFS_INODE_ITEM_KEY
;
760 * it is possible that we didn't log all the parent directories
761 * for a given inode. If we don't find the dir, just don't
762 * copy the back ref in. The link count fixup code will take
765 dir
= read_one_inode(root
, key
->offset
);
769 inode
= read_one_inode(root
, key
->objectid
);
772 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
773 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
776 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
778 namelen
= btrfs_inode_ref_name_len(eb
, ref
);
779 name
= kmalloc(namelen
, GFP_NOFS
);
782 read_extent_buffer(eb
, name
, (unsigned long)(ref
+ 1), namelen
);
784 /* if we already have a perfect match, we're done */
785 if (inode_in_dir(root
, path
, dir
->i_ino
, inode
->i_ino
,
786 btrfs_inode_ref_index(eb
, ref
),
792 * look for a conflicting back reference in the metadata.
793 * if we find one we have to unlink that name of the file
794 * before we add our new link. Later on, we overwrite any
795 * existing back reference, and we don't want to create
796 * dangling pointers in the directory.
799 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
803 struct btrfs_inode_ref
*victim_ref
;
805 unsigned long ptr_end
;
806 struct extent_buffer
*leaf
= path
->nodes
[0];
808 /* are we trying to overwrite a back ref for the root directory
809 * if so, just jump out, we're done
811 if (key
->objectid
== key
->offset
)
814 /* check all the names in this back reference to see
815 * if they are in the log. if so, we allow them to stay
816 * otherwise they must be unlinked as a conflict
818 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
819 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
820 while(ptr
< ptr_end
) {
821 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
822 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
824 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
825 BUG_ON(!victim_name
);
827 read_extent_buffer(leaf
, victim_name
,
828 (unsigned long)(victim_ref
+ 1),
831 if (!backref_in_log(log
, key
, victim_name
,
833 btrfs_inc_nlink(inode
);
834 btrfs_release_path(root
, path
);
835 ret
= btrfs_unlink_inode(trans
, root
, dir
,
839 btrfs_release_path(root
, path
);
843 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
847 btrfs_release_path(root
, path
);
849 /* look for a conflicting sequence number */
850 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
851 btrfs_inode_ref_index(eb
, ref
),
853 if (di
&& !IS_ERR(di
)) {
854 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
857 btrfs_release_path(root
, path
);
860 /* look for a conflicting name */
861 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
863 if (di
&& !IS_ERR(di
)) {
864 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
867 btrfs_release_path(root
, path
);
869 /* insert our name */
870 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
, 0,
871 btrfs_inode_ref_index(eb
, ref
));
874 btrfs_update_inode(trans
, root
, inode
);
877 ref_ptr
= (unsigned long)(ref
+ 1) + namelen
;
879 if (ref_ptr
< ref_end
)
882 /* finally write the back reference in the inode */
883 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
887 btrfs_release_path(root
, path
);
894 * replay one csum item from the log tree into the subvolume 'root'
895 * eb, slot and key all refer to the log tree
896 * path is for temp use by this function and should be released on return
898 * This copies the checksums out of the log tree and inserts them into
899 * the subvolume. Any existing checksums for this range in the file
900 * are overwritten, and new items are added where required.
902 * We keep this simple by reusing the btrfs_ordered_sum code from
903 * the data=ordered mode. This basically means making a copy
904 * of all the checksums in ram, which we have to do anyway for kmap
907 * The copy is then sent down to btrfs_csum_file_blocks, which
908 * does all the hard work of finding existing items in the file
909 * or adding new ones.
911 static noinline
int replay_one_csum(struct btrfs_trans_handle
*trans
,
912 struct btrfs_root
*root
,
913 struct btrfs_path
*path
,
914 struct extent_buffer
*eb
, int slot
,
915 struct btrfs_key
*key
)
918 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
920 unsigned long file_bytes
;
921 struct btrfs_ordered_sum
*sums
;
922 struct btrfs_sector_sum
*sector_sum
;
926 file_bytes
= (item_size
/ BTRFS_CRC32_SIZE
) * root
->sectorsize
;
927 inode
= read_one_inode(root
, key
->objectid
);
932 sums
= kzalloc(btrfs_ordered_sum_size(root
, file_bytes
), GFP_NOFS
);
938 INIT_LIST_HEAD(&sums
->list
);
939 sums
->len
= file_bytes
;
940 sums
->file_offset
= key
->offset
;
943 * copy all the sums into the ordered sum struct
945 sector_sum
= sums
->sums
;
946 cur_offset
= key
->offset
;
947 ptr
= btrfs_item_ptr_offset(eb
, slot
);
948 while(item_size
> 0) {
949 sector_sum
->offset
= cur_offset
;
950 read_extent_buffer(eb
, §or_sum
->sum
, ptr
, BTRFS_CRC32_SIZE
);
952 item_size
-= BTRFS_CRC32_SIZE
;
953 ptr
+= BTRFS_CRC32_SIZE
;
954 cur_offset
+= root
->sectorsize
;
957 /* let btrfs_csum_file_blocks add them into the file */
958 ret
= btrfs_csum_file_blocks(trans
, root
, inode
, sums
);
966 * There are a few corners where the link count of the file can't
967 * be properly maintained during replay. So, instead of adding
968 * lots of complexity to the log code, we just scan the backrefs
969 * for any file that has been through replay.
971 * The scan will update the link count on the inode to reflect the
972 * number of back refs found. If it goes down to zero, the iput
973 * will free the inode.
975 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
976 struct btrfs_root
*root
,
979 struct btrfs_path
*path
;
981 struct btrfs_key key
;
984 unsigned long ptr_end
;
987 key
.objectid
= inode
->i_ino
;
988 key
.type
= BTRFS_INODE_REF_KEY
;
989 key
.offset
= (u64
)-1;
991 path
= btrfs_alloc_path();
994 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
998 if (path
->slots
[0] == 0)
1002 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1004 if (key
.objectid
!= inode
->i_ino
||
1005 key
.type
!= BTRFS_INODE_REF_KEY
)
1007 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1008 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1010 while(ptr
< ptr_end
) {
1011 struct btrfs_inode_ref
*ref
;
1013 ref
= (struct btrfs_inode_ref
*)ptr
;
1014 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1016 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1020 if (key
.offset
== 0)
1023 btrfs_release_path(root
, path
);
1025 btrfs_free_path(path
);
1026 if (nlink
!= inode
->i_nlink
) {
1027 inode
->i_nlink
= nlink
;
1028 btrfs_update_inode(trans
, root
, inode
);
1034 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1035 struct btrfs_root
*root
,
1036 struct btrfs_path
*path
)
1039 struct btrfs_key key
;
1040 struct inode
*inode
;
1042 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1043 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1044 key
.offset
= (u64
)-1;
1046 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1051 if (path
->slots
[0] == 0)
1056 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1057 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1058 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1061 ret
= btrfs_del_item(trans
, root
, path
);
1064 btrfs_release_path(root
, path
);
1065 inode
= read_one_inode(root
, key
.offset
);
1068 ret
= fixup_inode_link_count(trans
, root
, inode
);
1073 if (key
.offset
== 0)
1077 btrfs_release_path(root
, path
);
1083 * record a given inode in the fixup dir so we can check its link
1084 * count when replay is done. The link count is incremented here
1085 * so the inode won't go away until we check it
1087 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1088 struct btrfs_root
*root
,
1089 struct btrfs_path
*path
,
1092 struct btrfs_key key
;
1094 struct inode
*inode
;
1096 inode
= read_one_inode(root
, objectid
);
1099 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1100 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1101 key
.offset
= objectid
;
1103 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1105 btrfs_release_path(root
, path
);
1107 btrfs_inc_nlink(inode
);
1108 btrfs_update_inode(trans
, root
, inode
);
1109 } else if (ret
== -EEXIST
) {
1120 * when replaying the log for a directory, we only insert names
1121 * for inodes that actually exist. This means an fsync on a directory
1122 * does not implicitly fsync all the new files in it
1124 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1125 struct btrfs_root
*root
,
1126 struct btrfs_path
*path
,
1127 u64 dirid
, u64 index
,
1128 char *name
, int name_len
, u8 type
,
1129 struct btrfs_key
*location
)
1131 struct inode
*inode
;
1135 inode
= read_one_inode(root
, location
->objectid
);
1139 dir
= read_one_inode(root
, dirid
);
1144 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1146 /* FIXME, put inode into FIXUP list */
1154 * take a single entry in a log directory item and replay it into
1157 * if a conflicting item exists in the subdirectory already,
1158 * the inode it points to is unlinked and put into the link count
1161 * If a name from the log points to a file or directory that does
1162 * not exist in the FS, it is skipped. fsyncs on directories
1163 * do not force down inodes inside that directory, just changes to the
1164 * names or unlinks in a directory.
1166 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1167 struct btrfs_root
*root
,
1168 struct btrfs_path
*path
,
1169 struct extent_buffer
*eb
,
1170 struct btrfs_dir_item
*di
,
1171 struct btrfs_key
*key
)
1175 struct btrfs_dir_item
*dst_di
;
1176 struct btrfs_key found_key
;
1177 struct btrfs_key log_key
;
1179 struct inode
*inode
;
1183 dir
= read_one_inode(root
, key
->objectid
);
1186 name_len
= btrfs_dir_name_len(eb
, di
);
1187 name
= kmalloc(name_len
, GFP_NOFS
);
1188 log_type
= btrfs_dir_type(eb
, di
);
1189 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1192 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1193 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1194 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1197 else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1198 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1205 if (!dst_di
|| IS_ERR(dst_di
)) {
1206 /* we need a sequence number to insert, so we only
1207 * do inserts for the BTRFS_DIR_INDEX_KEY types
1209 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1214 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1215 /* the existing item matches the logged item */
1216 if (found_key
.objectid
== log_key
.objectid
&&
1217 found_key
.type
== log_key
.type
&&
1218 found_key
.offset
== log_key
.offset
&&
1219 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1224 * don't drop the conflicting directory entry if the inode
1225 * for the new entry doesn't exist
1227 inode
= read_one_inode(root
, log_key
.objectid
);
1232 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1235 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1238 btrfs_release_path(root
, path
);
1244 btrfs_release_path(root
, path
);
1245 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1246 name
, name_len
, log_type
, &log_key
);
1248 if (ret
&& ret
!= -ENOENT
)
1254 * find all the names in a directory item and reconcile them into
1255 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1256 * one name in a directory item, but the same code gets used for
1257 * both directory index types
1259 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1260 struct btrfs_root
*root
,
1261 struct btrfs_path
*path
,
1262 struct extent_buffer
*eb
, int slot
,
1263 struct btrfs_key
*key
)
1266 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1267 struct btrfs_dir_item
*di
;
1270 unsigned long ptr_end
;
1272 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1273 ptr_end
= ptr
+ item_size
;
1274 while(ptr
< ptr_end
) {
1275 di
= (struct btrfs_dir_item
*)ptr
;
1276 name_len
= btrfs_dir_name_len(eb
, di
);
1277 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1279 ptr
= (unsigned long)(di
+ 1);
1286 * directory replay has two parts. There are the standard directory
1287 * items in the log copied from the subvolume, and range items
1288 * created in the log while the subvolume was logged.
1290 * The range items tell us which parts of the key space the log
1291 * is authoritative for. During replay, if a key in the subvolume
1292 * directory is in a logged range item, but not actually in the log
1293 * that means it was deleted from the directory before the fsync
1294 * and should be removed.
1296 static noinline
int find_dir_range(struct btrfs_root
*root
,
1297 struct btrfs_path
*path
,
1298 u64 dirid
, int key_type
,
1299 u64
*start_ret
, u64
*end_ret
)
1301 struct btrfs_key key
;
1303 struct btrfs_dir_log_item
*item
;
1307 if (*start_ret
== (u64
)-1)
1310 key
.objectid
= dirid
;
1311 key
.type
= key_type
;
1312 key
.offset
= *start_ret
;
1314 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1318 if (path
->slots
[0] == 0)
1323 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1325 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1329 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1330 struct btrfs_dir_log_item
);
1331 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1333 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1335 *start_ret
= key
.offset
;
1336 *end_ret
= found_end
;
1341 /* check the next slot in the tree to see if it is a valid item */
1342 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1343 if (path
->slots
[0] >= nritems
) {
1344 ret
= btrfs_next_leaf(root
, path
);
1351 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1353 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1357 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1358 struct btrfs_dir_log_item
);
1359 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1360 *start_ret
= key
.offset
;
1361 *end_ret
= found_end
;
1364 btrfs_release_path(root
, path
);
1369 * this looks for a given directory item in the log. If the directory
1370 * item is not in the log, the item is removed and the inode it points
1373 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1374 struct btrfs_root
*root
,
1375 struct btrfs_root
*log
,
1376 struct btrfs_path
*path
,
1377 struct btrfs_path
*log_path
,
1379 struct btrfs_key
*dir_key
)
1382 struct extent_buffer
*eb
;
1385 struct btrfs_dir_item
*di
;
1386 struct btrfs_dir_item
*log_di
;
1389 unsigned long ptr_end
;
1391 struct inode
*inode
;
1392 struct btrfs_key location
;
1395 eb
= path
->nodes
[0];
1396 slot
= path
->slots
[0];
1397 item_size
= btrfs_item_size_nr(eb
, slot
);
1398 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1399 ptr_end
= ptr
+ item_size
;
1400 while(ptr
< ptr_end
) {
1401 di
= (struct btrfs_dir_item
*)ptr
;
1402 name_len
= btrfs_dir_name_len(eb
, di
);
1403 name
= kmalloc(name_len
, GFP_NOFS
);
1408 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1411 if (dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1412 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1415 } else if (dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1416 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1422 if (!log_di
|| IS_ERR(log_di
)) {
1423 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1424 btrfs_release_path(root
, path
);
1425 btrfs_release_path(log
, log_path
);
1426 inode
= read_one_inode(root
, location
.objectid
);
1429 ret
= link_to_fixup_dir(trans
, root
,
1430 path
, location
.objectid
);
1432 btrfs_inc_nlink(inode
);
1433 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1439 /* there might still be more names under this key
1440 * check and repeat if required
1442 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1449 btrfs_release_path(log
, log_path
);
1452 ptr
= (unsigned long)(di
+ 1);
1457 btrfs_release_path(root
, path
);
1458 btrfs_release_path(log
, log_path
);
1463 * deletion replay happens before we copy any new directory items
1464 * out of the log or out of backreferences from inodes. It
1465 * scans the log to find ranges of keys that log is authoritative for,
1466 * and then scans the directory to find items in those ranges that are
1467 * not present in the log.
1469 * Anything we don't find in the log is unlinked and removed from the
1472 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1473 struct btrfs_root
*root
,
1474 struct btrfs_root
*log
,
1475 struct btrfs_path
*path
,
1480 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1482 struct btrfs_key dir_key
;
1483 struct btrfs_key found_key
;
1484 struct btrfs_path
*log_path
;
1487 dir_key
.objectid
= dirid
;
1488 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1489 log_path
= btrfs_alloc_path();
1493 dir
= read_one_inode(root
, dirid
);
1494 /* it isn't an error if the inode isn't there, that can happen
1495 * because we replay the deletes before we copy in the inode item
1499 btrfs_free_path(log_path
);
1506 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1507 &range_start
, &range_end
);
1511 dir_key
.offset
= range_start
;
1514 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1519 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1520 if (path
->slots
[0] >= nritems
) {
1521 ret
= btrfs_next_leaf(root
, path
);
1525 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1527 if (found_key
.objectid
!= dirid
||
1528 found_key
.type
!= dir_key
.type
)
1531 if (found_key
.offset
> range_end
)
1534 ret
= check_item_in_log(trans
, root
, log
, path
,
1535 log_path
, dir
, &found_key
);
1537 if (found_key
.offset
== (u64
)-1)
1539 dir_key
.offset
= found_key
.offset
+ 1;
1541 btrfs_release_path(root
, path
);
1542 if (range_end
== (u64
)-1)
1544 range_start
= range_end
+ 1;
1549 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1550 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1551 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1552 btrfs_release_path(root
, path
);
1556 btrfs_release_path(root
, path
);
1557 btrfs_free_path(log_path
);
1563 * the process_func used to replay items from the log tree. This
1564 * gets called in two different stages. The first stage just looks
1565 * for inodes and makes sure they are all copied into the subvolume.
1567 * The second stage copies all the other item types from the log into
1568 * the subvolume. The two stage approach is slower, but gets rid of
1569 * lots of complexity around inodes referencing other inodes that exist
1570 * only in the log (references come from either directory items or inode
1573 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
1574 struct walk_control
*wc
, u64 gen
)
1577 struct btrfs_path
*path
;
1578 struct btrfs_root
*root
= wc
->replay_dest
;
1579 struct btrfs_key key
;
1585 btrfs_read_buffer(eb
, gen
);
1587 level
= btrfs_header_level(eb
);
1592 path
= btrfs_alloc_path();
1595 nritems
= btrfs_header_nritems(eb
);
1596 for (i
= 0; i
< nritems
; i
++) {
1597 btrfs_item_key_to_cpu(eb
, &key
, i
);
1598 item_size
= btrfs_item_size_nr(eb
, i
);
1600 /* inode keys are done during the first stage */
1601 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
1602 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
1603 struct inode
*inode
;
1604 struct btrfs_inode_item
*inode_item
;
1607 inode_item
= btrfs_item_ptr(eb
, i
,
1608 struct btrfs_inode_item
);
1609 mode
= btrfs_inode_mode(eb
, inode_item
);
1610 if (S_ISDIR(mode
)) {
1611 ret
= replay_dir_deletes(wc
->trans
,
1612 root
, log
, path
, key
.objectid
);
1615 ret
= overwrite_item(wc
->trans
, root
, path
,
1619 /* for regular files, truncate away
1620 * extents past the new EOF
1622 if (S_ISREG(mode
)) {
1623 inode
= read_one_inode(root
,
1627 ret
= btrfs_truncate_inode_items(wc
->trans
,
1628 root
, inode
, inode
->i_size
,
1629 BTRFS_EXTENT_DATA_KEY
);
1633 ret
= link_to_fixup_dir(wc
->trans
, root
,
1634 path
, key
.objectid
);
1637 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
1640 /* these keys are simply copied */
1641 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
1642 ret
= overwrite_item(wc
->trans
, root
, path
,
1645 } else if (key
.type
== BTRFS_INODE_REF_KEY
) {
1646 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
1648 BUG_ON(ret
&& ret
!= -ENOENT
);
1649 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
1650 ret
= replay_one_extent(wc
->trans
, root
, path
,
1653 } else if (key
.type
== BTRFS_CSUM_ITEM_KEY
) {
1654 ret
= replay_one_csum(wc
->trans
, root
, path
,
1657 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
||
1658 key
.type
== BTRFS_DIR_INDEX_KEY
) {
1659 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
1664 btrfs_free_path(path
);
1668 static int noinline
walk_down_log_tree(struct btrfs_trans_handle
*trans
,
1669 struct btrfs_root
*root
,
1670 struct btrfs_path
*path
, int *level
,
1671 struct walk_control
*wc
)
1677 struct extent_buffer
*next
;
1678 struct extent_buffer
*cur
;
1679 struct extent_buffer
*parent
;
1683 WARN_ON(*level
< 0);
1684 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1687 WARN_ON(*level
< 0);
1688 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1689 cur
= path
->nodes
[*level
];
1691 if (btrfs_header_level(cur
) != *level
)
1694 if (path
->slots
[*level
] >=
1695 btrfs_header_nritems(cur
))
1698 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
1699 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
1700 blocksize
= btrfs_level_size(root
, *level
- 1);
1702 parent
= path
->nodes
[*level
];
1703 root_owner
= btrfs_header_owner(parent
);
1704 root_gen
= btrfs_header_generation(parent
);
1706 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1708 wc
->process_func(root
, next
, wc
, ptr_gen
);
1711 path
->slots
[*level
]++;
1713 btrfs_read_buffer(next
, ptr_gen
);
1715 btrfs_tree_lock(next
);
1716 clean_tree_block(trans
, root
, next
);
1717 btrfs_wait_tree_block_writeback(next
);
1718 btrfs_tree_unlock(next
);
1720 ret
= btrfs_drop_leaf_ref(trans
, root
, next
);
1723 WARN_ON(root_owner
!=
1724 BTRFS_TREE_LOG_OBJECTID
);
1725 ret
= btrfs_free_extent(trans
, root
, bytenr
,
1726 blocksize
, root_owner
,
1730 free_extent_buffer(next
);
1733 btrfs_read_buffer(next
, ptr_gen
);
1735 WARN_ON(*level
<= 0);
1736 if (path
->nodes
[*level
-1])
1737 free_extent_buffer(path
->nodes
[*level
-1]);
1738 path
->nodes
[*level
-1] = next
;
1739 *level
= btrfs_header_level(next
);
1740 path
->slots
[*level
] = 0;
1743 WARN_ON(*level
< 0);
1744 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1746 if (path
->nodes
[*level
] == root
->node
) {
1747 parent
= path
->nodes
[*level
];
1749 parent
= path
->nodes
[*level
+ 1];
1751 bytenr
= path
->nodes
[*level
]->start
;
1753 blocksize
= btrfs_level_size(root
, *level
);
1754 root_owner
= btrfs_header_owner(parent
);
1755 root_gen
= btrfs_header_generation(parent
);
1757 wc
->process_func(root
, path
->nodes
[*level
], wc
,
1758 btrfs_header_generation(path
->nodes
[*level
]));
1761 next
= path
->nodes
[*level
];
1762 btrfs_tree_lock(next
);
1763 clean_tree_block(trans
, root
, next
);
1764 btrfs_wait_tree_block_writeback(next
);
1765 btrfs_tree_unlock(next
);
1768 ret
= btrfs_drop_leaf_ref(trans
, root
, next
);
1771 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1772 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
1773 root_owner
, root_gen
, 0, 0, 1);
1776 free_extent_buffer(path
->nodes
[*level
]);
1777 path
->nodes
[*level
] = NULL
;
1784 static int noinline
walk_up_log_tree(struct btrfs_trans_handle
*trans
,
1785 struct btrfs_root
*root
,
1786 struct btrfs_path
*path
, int *level
,
1787 struct walk_control
*wc
)
1795 for(i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
1796 slot
= path
->slots
[i
];
1797 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
1798 struct extent_buffer
*node
;
1799 node
= path
->nodes
[i
];
1802 WARN_ON(*level
== 0);
1805 if (path
->nodes
[*level
] == root
->node
) {
1806 root_owner
= root
->root_key
.objectid
;
1808 btrfs_header_generation(path
->nodes
[*level
]);
1810 struct extent_buffer
*node
;
1811 node
= path
->nodes
[*level
+ 1];
1812 root_owner
= btrfs_header_owner(node
);
1813 root_gen
= btrfs_header_generation(node
);
1815 wc
->process_func(root
, path
->nodes
[*level
], wc
,
1816 btrfs_header_generation(path
->nodes
[*level
]));
1818 struct extent_buffer
*next
;
1820 next
= path
->nodes
[*level
];
1822 btrfs_tree_lock(next
);
1823 clean_tree_block(trans
, root
, next
);
1824 btrfs_wait_tree_block_writeback(next
);
1825 btrfs_tree_unlock(next
);
1828 ret
= btrfs_drop_leaf_ref(trans
, root
,
1833 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1834 ret
= btrfs_free_extent(trans
, root
,
1835 path
->nodes
[*level
]->start
,
1836 path
->nodes
[*level
]->len
,
1837 root_owner
, root_gen
, 0, 0, 1);
1840 free_extent_buffer(path
->nodes
[*level
]);
1841 path
->nodes
[*level
] = NULL
;
1849 * drop the reference count on the tree rooted at 'snap'. This traverses
1850 * the tree freeing any blocks that have a ref count of zero after being
1853 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
1854 struct btrfs_root
*log
, struct walk_control
*wc
)
1859 struct btrfs_path
*path
;
1863 path
= btrfs_alloc_path();
1866 level
= btrfs_header_level(log
->node
);
1868 path
->nodes
[level
] = log
->node
;
1869 extent_buffer_get(log
->node
);
1870 path
->slots
[level
] = 0;
1873 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
1879 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
1886 /* was the root node processed? if not, catch it here */
1887 if (path
->nodes
[orig_level
]) {
1888 wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
1889 btrfs_header_generation(path
->nodes
[orig_level
]));
1891 struct extent_buffer
*next
;
1893 next
= path
->nodes
[orig_level
];
1895 btrfs_tree_lock(next
);
1896 clean_tree_block(trans
, log
, next
);
1897 btrfs_wait_tree_block_writeback(next
);
1898 btrfs_tree_unlock(next
);
1900 if (orig_level
== 0) {
1901 ret
= btrfs_drop_leaf_ref(trans
, log
,
1905 WARN_ON(log
->root_key
.objectid
!=
1906 BTRFS_TREE_LOG_OBJECTID
);
1907 ret
= btrfs_free_extent(trans
, log
,
1908 next
->start
, next
->len
,
1909 log
->root_key
.objectid
,
1910 btrfs_header_generation(next
),
1916 for (i
= 0; i
<= orig_level
; i
++) {
1917 if (path
->nodes
[i
]) {
1918 free_extent_buffer(path
->nodes
[i
]);
1919 path
->nodes
[i
] = NULL
;
1922 btrfs_free_path(path
);
1924 free_extent_buffer(log
->node
);
1928 int wait_log_commit(struct btrfs_root
*log
)
1931 u64 transid
= log
->fs_info
->tree_log_transid
;
1934 prepare_to_wait(&log
->fs_info
->tree_log_wait
, &wait
,
1935 TASK_UNINTERRUPTIBLE
);
1936 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
1937 if (atomic_read(&log
->fs_info
->tree_log_commit
))
1939 finish_wait(&log
->fs_info
->tree_log_wait
, &wait
);
1940 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1941 } while(transid
== log
->fs_info
->tree_log_transid
&&
1942 atomic_read(&log
->fs_info
->tree_log_commit
));
1947 * btrfs_sync_log does sends a given tree log down to the disk and
1948 * updates the super blocks to record it. When this call is done,
1949 * you know that any inodes previously logged are safely on disk
1951 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
1952 struct btrfs_root
*root
)
1955 unsigned long batch
;
1956 struct btrfs_root
*log
= root
->log_root
;
1957 struct walk_control wc
= {
1959 .process_func
= process_one_buffer
1962 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1963 if (atomic_read(&log
->fs_info
->tree_log_commit
)) {
1964 wait_log_commit(log
);
1967 atomic_set(&log
->fs_info
->tree_log_commit
, 1);
1970 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
1971 schedule_timeout_uninterruptible(1);
1972 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1973 batch
= log
->fs_info
->tree_log_batch
;
1975 while(atomic_read(&log
->fs_info
->tree_log_writers
)) {
1977 prepare_to_wait(&log
->fs_info
->tree_log_wait
, &wait
,
1978 TASK_UNINTERRUPTIBLE
);
1979 batch
= log
->fs_info
->tree_log_batch
;
1980 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
1981 if (atomic_read(&log
->fs_info
->tree_log_writers
))
1983 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1984 finish_wait(&log
->fs_info
->tree_log_wait
, &wait
);
1986 if (batch
== log
->fs_info
->tree_log_batch
)
1989 ret
= walk_log_tree(trans
, log
, &wc
);
1992 ret
= walk_log_tree(trans
, log
->fs_info
->log_root_tree
, &wc
);
1997 ret
= walk_log_tree(trans
, log
, &wc
);
2000 ret
= walk_log_tree(trans
, log
->fs_info
->log_root_tree
, &wc
);
2003 btrfs_set_super_log_root(&root
->fs_info
->super_for_commit
,
2004 log
->fs_info
->log_root_tree
->node
->start
);
2005 btrfs_set_super_log_root_level(&root
->fs_info
->super_for_commit
,
2006 btrfs_header_level(log
->fs_info
->log_root_tree
->node
));
2008 write_ctree_super(trans
, log
->fs_info
->tree_root
);
2009 log
->fs_info
->tree_log_transid
++;
2010 log
->fs_info
->tree_log_batch
= 0;
2011 atomic_set(&log
->fs_info
->tree_log_commit
, 0);
2013 if (waitqueue_active(&log
->fs_info
->tree_log_wait
))
2014 wake_up(&log
->fs_info
->tree_log_wait
);
2016 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
2022 * free all the extents used by the tree log. This should be called
2023 * at commit time of the full transaction
2025 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2028 struct btrfs_root
*log
;
2030 struct walk_control wc
= {
2032 .process_func
= process_one_buffer
2035 if (!root
->log_root
)
2038 log
= root
->log_root
;
2039 ret
= walk_log_tree(trans
, log
, &wc
);
2042 log
= root
->log_root
;
2043 ret
= btrfs_del_root(trans
, root
->fs_info
->log_root_tree
,
2046 root
->log_root
= NULL
;
2047 kfree(root
->log_root
);
2052 * helper function to update the item for a given subvolumes log root
2053 * in the tree of log roots
2055 static int update_log_root(struct btrfs_trans_handle
*trans
,
2056 struct btrfs_root
*log
)
2058 u64 bytenr
= btrfs_root_bytenr(&log
->root_item
);
2061 if (log
->node
->start
== bytenr
)
2064 btrfs_set_root_bytenr(&log
->root_item
, log
->node
->start
);
2065 btrfs_set_root_level(&log
->root_item
, btrfs_header_level(log
->node
));
2066 ret
= btrfs_update_root(trans
, log
->fs_info
->log_root_tree
,
2067 &log
->root_key
, &log
->root_item
);
2073 * If both a file and directory are logged, and unlinks or renames are
2074 * mixed in, we have a few interesting corners:
2076 * create file X in dir Y
2077 * link file X to X.link in dir Y
2079 * unlink file X but leave X.link
2082 * After a crash we would expect only X.link to exist. But file X
2083 * didn't get fsync'd again so the log has back refs for X and X.link.
2085 * We solve this by removing directory entries and inode backrefs from the
2086 * log when a file that was logged in the current transaction is
2087 * unlinked. Any later fsync will include the updated log entries, and
2088 * we'll be able to reconstruct the proper directory items from backrefs.
2090 * This optimizations allows us to avoid relogging the entire inode
2091 * or the entire directory.
2093 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
2094 struct btrfs_root
*root
,
2095 const char *name
, int name_len
,
2096 struct inode
*dir
, u64 index
)
2098 struct btrfs_root
*log
;
2099 struct btrfs_dir_item
*di
;
2100 struct btrfs_path
*path
;
2104 ret
= join_running_log_trans(root
);
2108 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2110 log
= root
->log_root
;
2111 path
= btrfs_alloc_path();
2112 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir
->i_ino
,
2113 name
, name_len
, -1);
2114 if (di
&& !IS_ERR(di
)) {
2115 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2116 bytes_del
+= name_len
;
2119 btrfs_release_path(log
, path
);
2120 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir
->i_ino
,
2121 index
, name
, name_len
, -1);
2122 if (di
&& !IS_ERR(di
)) {
2123 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2124 bytes_del
+= name_len
;
2128 /* update the directory size in the log to reflect the names
2132 struct btrfs_key key
;
2134 key
.objectid
= dir
->i_ino
;
2136 key
.type
= BTRFS_INODE_ITEM_KEY
;
2137 btrfs_release_path(log
, path
);
2139 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2141 struct btrfs_inode_item
*item
;
2144 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2145 struct btrfs_inode_item
);
2146 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2147 if (i_size
> bytes_del
)
2148 i_size
-= bytes_del
;
2151 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2152 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2155 btrfs_release_path(log
, path
);
2158 btrfs_free_path(path
);
2159 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2160 end_log_trans(root
);
2165 /* see comments for btrfs_del_dir_entries_in_log */
2166 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2167 struct btrfs_root
*root
,
2168 const char *name
, int name_len
,
2169 struct inode
*inode
, u64 dirid
)
2171 struct btrfs_root
*log
;
2175 ret
= join_running_log_trans(root
);
2178 log
= root
->log_root
;
2179 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2181 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, inode
->i_ino
,
2183 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2184 end_log_trans(root
);
2186 if (ret
== 0 || ret
== -ENOENT
)
2192 * creates a range item in the log for 'dirid'. first_offset and
2193 * last_offset tell us which parts of the key space the log should
2194 * be considered authoritative for.
2196 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2197 struct btrfs_root
*log
,
2198 struct btrfs_path
*path
,
2199 int key_type
, u64 dirid
,
2200 u64 first_offset
, u64 last_offset
)
2203 struct btrfs_key key
;
2204 struct btrfs_dir_log_item
*item
;
2206 key
.objectid
= dirid
;
2207 key
.offset
= first_offset
;
2208 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2209 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2211 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2212 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2215 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2216 struct btrfs_dir_log_item
);
2217 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2218 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2219 btrfs_release_path(log
, path
);
2224 * log all the items included in the current transaction for a given
2225 * directory. This also creates the range items in the log tree required
2226 * to replay anything deleted before the fsync
2228 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2229 struct btrfs_root
*root
, struct inode
*inode
,
2230 struct btrfs_path
*path
,
2231 struct btrfs_path
*dst_path
, int key_type
,
2232 u64 min_offset
, u64
*last_offset_ret
)
2234 struct btrfs_key min_key
;
2235 struct btrfs_key max_key
;
2236 struct btrfs_root
*log
= root
->log_root
;
2237 struct extent_buffer
*src
;
2241 u64 first_offset
= min_offset
;
2242 u64 last_offset
= (u64
)-1;
2244 log
= root
->log_root
;
2245 max_key
.objectid
= inode
->i_ino
;
2246 max_key
.offset
= (u64
)-1;
2247 max_key
.type
= key_type
;
2249 min_key
.objectid
= inode
->i_ino
;
2250 min_key
.type
= key_type
;
2251 min_key
.offset
= min_offset
;
2253 path
->keep_locks
= 1;
2255 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2256 path
, 0, trans
->transid
);
2259 * we didn't find anything from this transaction, see if there
2260 * is anything at all
2262 if (ret
!= 0 || min_key
.objectid
!= inode
->i_ino
||
2263 min_key
.type
!= key_type
) {
2264 min_key
.objectid
= inode
->i_ino
;
2265 min_key
.type
= key_type
;
2266 min_key
.offset
= (u64
)-1;
2267 btrfs_release_path(root
, path
);
2268 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2270 btrfs_release_path(root
, path
);
2273 ret
= btrfs_previous_item(root
, path
, inode
->i_ino
, key_type
);
2275 /* if ret == 0 there are items for this type,
2276 * create a range to tell us the last key of this type.
2277 * otherwise, there are no items in this directory after
2278 * *min_offset, and we create a range to indicate that.
2281 struct btrfs_key tmp
;
2282 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2284 if (key_type
== tmp
.type
) {
2285 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2291 /* go backward to find any previous key */
2292 ret
= btrfs_previous_item(root
, path
, inode
->i_ino
, key_type
);
2294 struct btrfs_key tmp
;
2295 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2296 if (key_type
== tmp
.type
) {
2297 first_offset
= tmp
.offset
;
2298 ret
= overwrite_item(trans
, log
, dst_path
,
2299 path
->nodes
[0], path
->slots
[0],
2303 btrfs_release_path(root
, path
);
2305 /* find the first key from this transaction again */
2306 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2313 * we have a block from this transaction, log every item in it
2314 * from our directory
2317 struct btrfs_key tmp
;
2318 src
= path
->nodes
[0];
2319 nritems
= btrfs_header_nritems(src
);
2320 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2321 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2323 if (min_key
.objectid
!= inode
->i_ino
||
2324 min_key
.type
!= key_type
)
2326 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2330 path
->slots
[0] = nritems
;
2333 * look ahead to the next item and see if it is also
2334 * from this directory and from this transaction
2336 ret
= btrfs_next_leaf(root
, path
);
2338 last_offset
= (u64
)-1;
2341 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2342 if (tmp
.objectid
!= inode
->i_ino
|| tmp
.type
!= key_type
) {
2343 last_offset
= (u64
)-1;
2346 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2347 ret
= overwrite_item(trans
, log
, dst_path
,
2348 path
->nodes
[0], path
->slots
[0],
2352 last_offset
= tmp
.offset
;
2357 *last_offset_ret
= last_offset
;
2358 btrfs_release_path(root
, path
);
2359 btrfs_release_path(log
, dst_path
);
2361 /* insert the log range keys to indicate where the log is valid */
2362 ret
= insert_dir_log_key(trans
, log
, path
, key_type
, inode
->i_ino
,
2363 first_offset
, last_offset
);
2369 * logging directories is very similar to logging inodes, We find all the items
2370 * from the current transaction and write them to the log.
2372 * The recovery code scans the directory in the subvolume, and if it finds a
2373 * key in the range logged that is not present in the log tree, then it means
2374 * that dir entry was unlinked during the transaction.
2376 * In order for that scan to work, we must include one key smaller than
2377 * the smallest logged by this transaction and one key larger than the largest
2378 * key logged by this transaction.
2380 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
2381 struct btrfs_root
*root
, struct inode
*inode
,
2382 struct btrfs_path
*path
,
2383 struct btrfs_path
*dst_path
)
2388 int key_type
= BTRFS_DIR_ITEM_KEY
;
2394 ret
= log_dir_items(trans
, root
, inode
, path
,
2395 dst_path
, key_type
, min_key
,
2398 if (max_key
== (u64
)-1)
2400 min_key
= max_key
+ 1;
2403 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
2404 key_type
= BTRFS_DIR_INDEX_KEY
;
2411 * a helper function to drop items from the log before we relog an
2412 * inode. max_key_type indicates the highest item type to remove.
2413 * This cannot be run for file data extents because it does not
2414 * free the extents they point to.
2416 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
2417 struct btrfs_root
*log
,
2418 struct btrfs_path
*path
,
2419 u64 objectid
, int max_key_type
)
2422 struct btrfs_key key
;
2423 struct btrfs_key found_key
;
2425 key
.objectid
= objectid
;
2426 key
.type
= max_key_type
;
2427 key
.offset
= (u64
)-1;
2430 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
2435 if (path
->slots
[0] == 0)
2439 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2442 if (found_key
.objectid
!= objectid
)
2445 ret
= btrfs_del_item(trans
, log
, path
);
2447 btrfs_release_path(log
, path
);
2449 btrfs_release_path(log
, path
);
2453 /* log a single inode in the tree log.
2454 * At least one parent directory for this inode must exist in the tree
2455 * or be logged already.
2457 * Any items from this inode changed by the current transaction are copied
2458 * to the log tree. An extra reference is taken on any extents in this
2459 * file, allowing us to avoid a whole pile of corner cases around logging
2460 * blocks that have been removed from the tree.
2462 * See LOG_INODE_ALL and related defines for a description of what inode_only
2465 * This handles both files and directories.
2467 static int __btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2468 struct btrfs_root
*root
, struct inode
*inode
,
2471 struct btrfs_path
*path
;
2472 struct btrfs_path
*dst_path
;
2473 struct btrfs_key min_key
;
2474 struct btrfs_key max_key
;
2475 struct btrfs_root
*log
= root
->log_root
;
2476 unsigned long src_offset
;
2477 unsigned long dst_offset
;
2478 struct extent_buffer
*src
;
2479 struct btrfs_file_extent_item
*extent
;
2480 struct btrfs_inode_item
*inode_item
;
2484 log
= root
->log_root
;
2486 path
= btrfs_alloc_path();
2487 dst_path
= btrfs_alloc_path();
2489 min_key
.objectid
= inode
->i_ino
;
2490 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
2493 max_key
.objectid
= inode
->i_ino
;
2494 if (inode_only
== LOG_INODE_EXISTS
|| S_ISDIR(inode
->i_mode
))
2495 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
2497 max_key
.type
= (u8
)-1;
2498 max_key
.offset
= (u64
)-1;
2501 * if this inode has already been logged and we're in inode_only
2502 * mode, we don't want to delete the things that have already
2503 * been written to the log.
2505 * But, if the inode has been through an inode_only log,
2506 * the logged_trans field is not set. This allows us to catch
2507 * any new names for this inode in the backrefs by logging it
2510 if (inode_only
== LOG_INODE_EXISTS
&&
2511 BTRFS_I(inode
)->logged_trans
== trans
->transid
) {
2512 btrfs_free_path(path
);
2513 btrfs_free_path(dst_path
);
2516 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2519 * a brute force approach to making sure we get the most uptodate
2520 * copies of everything.
2522 if (S_ISDIR(inode
->i_mode
)) {
2523 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
2525 if (inode_only
== LOG_INODE_EXISTS
)
2526 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
2527 ret
= drop_objectid_items(trans
, log
, path
,
2528 inode
->i_ino
, max_key_type
);
2530 ret
= btrfs_truncate_inode_items(trans
, log
, inode
, 0, 0);
2533 path
->keep_locks
= 1;
2536 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2537 path
, 0, trans
->transid
);
2541 if (min_key
.objectid
!= inode
->i_ino
)
2543 if (min_key
.type
> max_key
.type
)
2546 src
= path
->nodes
[0];
2547 size
= btrfs_item_size_nr(src
, path
->slots
[0]);
2548 ret
= btrfs_insert_empty_item(trans
, log
, dst_path
, &min_key
,
2553 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
2554 dst_path
->slots
[0]);
2556 src_offset
= btrfs_item_ptr_offset(src
, path
->slots
[0]);
2558 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
2561 if (inode_only
== LOG_INODE_EXISTS
&&
2562 min_key
.type
== BTRFS_INODE_ITEM_KEY
) {
2563 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
2565 struct btrfs_inode_item
);
2566 btrfs_set_inode_size(dst_path
->nodes
[0], inode_item
, 0);
2568 /* set the generation to zero so the recover code
2569 * can tell the difference between an logging
2570 * just to say 'this inode exists' and a logging
2571 * to say 'update this inode with these values'
2573 btrfs_set_inode_generation(dst_path
->nodes
[0],
2576 /* take a reference on file data extents so that truncates
2577 * or deletes of this inode don't have to relog the inode
2580 if (btrfs_key_type(&min_key
) == BTRFS_EXTENT_DATA_KEY
) {
2582 extent
= btrfs_item_ptr(src
, path
->slots
[0],
2583 struct btrfs_file_extent_item
);
2585 found_type
= btrfs_file_extent_type(src
, extent
);
2586 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
2587 u64 ds
= btrfs_file_extent_disk_bytenr(src
,
2589 u64 dl
= btrfs_file_extent_disk_num_bytes(src
,
2591 /* ds == 0 is a hole */
2593 ret
= btrfs_inc_extent_ref(trans
, log
,
2595 log
->root_key
.objectid
,
2604 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
2605 btrfs_release_path(root
, path
);
2606 btrfs_release_path(log
, dst_path
);
2608 if (min_key
.offset
< (u64
)-1)
2610 else if (min_key
.type
< (u8
)-1)
2612 else if (min_key
.objectid
< (u64
)-1)
2617 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
2618 btrfs_release_path(root
, path
);
2619 btrfs_release_path(log
, dst_path
);
2620 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
2623 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2625 btrfs_free_path(path
);
2626 btrfs_free_path(dst_path
);
2628 mutex_lock(&root
->fs_info
->tree_log_mutex
);
2629 ret
= update_log_root(trans
, log
);
2631 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2636 int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2637 struct btrfs_root
*root
, struct inode
*inode
,
2642 start_log_trans(trans
, root
);
2643 ret
= __btrfs_log_inode(trans
, root
, inode
, inode_only
);
2644 end_log_trans(root
);
2649 * helper function around btrfs_log_inode to make sure newly created
2650 * parent directories also end up in the log. A minimal inode and backref
2651 * only logging is done of any parent directories that are older than
2652 * the last committed transaction
2654 int btrfs_log_dentry(struct btrfs_trans_handle
*trans
,
2655 struct btrfs_root
*root
, struct dentry
*dentry
)
2657 int inode_only
= LOG_INODE_ALL
;
2658 struct super_block
*sb
;
2661 start_log_trans(trans
, root
);
2662 sb
= dentry
->d_inode
->i_sb
;
2664 ret
= __btrfs_log_inode(trans
, root
, dentry
->d_inode
,
2667 inode_only
= LOG_INODE_EXISTS
;
2669 dentry
= dentry
->d_parent
;
2670 if (!dentry
|| !dentry
->d_inode
|| sb
!= dentry
->d_inode
->i_sb
)
2673 if (BTRFS_I(dentry
->d_inode
)->generation
<=
2674 root
->fs_info
->last_trans_committed
)
2677 end_log_trans(root
);
2682 * it is not safe to log dentry if the chunk root has added new
2683 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2684 * If this returns 1, you must commit the transaction to safely get your
2687 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
2688 struct btrfs_root
*root
, struct dentry
*dentry
)
2691 gen
= root
->fs_info
->last_trans_new_blockgroup
;
2692 if (gen
> root
->fs_info
->last_trans_committed
)
2695 return btrfs_log_dentry(trans
, root
, dentry
);
2699 * should be called during mount to recover any replay any log trees
2702 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
2705 struct btrfs_path
*path
;
2706 struct btrfs_trans_handle
*trans
;
2707 struct btrfs_key key
;
2708 struct btrfs_key found_key
;
2709 struct btrfs_key tmp_key
;
2710 struct btrfs_root
*log
;
2711 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
2712 struct walk_control wc
= {
2713 .process_func
= process_one_buffer
,
2717 fs_info
->log_root_recovering
= 1;
2718 path
= btrfs_alloc_path();
2721 trans
= btrfs_start_transaction(fs_info
->tree_root
, 1);
2726 walk_log_tree(trans
, log_root_tree
, &wc
);
2729 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
2730 key
.offset
= (u64
)-1;
2731 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
2734 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
2738 if (path
->slots
[0] == 0)
2742 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2744 btrfs_release_path(log_root_tree
, path
);
2745 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
2748 log
= btrfs_read_fs_root_no_radix(log_root_tree
,
2753 tmp_key
.objectid
= found_key
.offset
;
2754 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
2755 tmp_key
.offset
= (u64
)-1;
2757 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
2759 BUG_ON(!wc
.replay_dest
);
2761 btrfs_record_root_in_trans(wc
.replay_dest
);
2762 ret
= walk_log_tree(trans
, log
, &wc
);
2765 if (wc
.stage
== LOG_WALK_REPLAY_ALL
) {
2766 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
2771 key
.offset
= found_key
.offset
- 1;
2772 free_extent_buffer(log
->node
);
2775 if (found_key
.offset
== 0)
2778 btrfs_release_path(log_root_tree
, path
);
2780 /* step one is to pin it all, step two is to replay just inodes */
2783 wc
.process_func
= replay_one_buffer
;
2784 wc
.stage
= LOG_WALK_REPLAY_INODES
;
2787 /* step three is to replay everything */
2788 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
2793 btrfs_free_path(path
);
2795 free_extent_buffer(log_root_tree
->node
);
2796 log_root_tree
->log_root
= NULL
;
2797 fs_info
->log_root_recovering
= 0;
2799 /* step 4: commit the transaction, which also unpins the blocks */
2800 btrfs_commit_transaction(trans
, fs_info
->tree_root
);
2802 kfree(log_root_tree
);