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"
28 /* magic values for the inode_only field in btrfs_log_inode:
30 * LOG_INODE_ALL means to log everything
31 * LOG_INODE_EXISTS means to log just enough to recreate the inode
34 #define LOG_INODE_ALL 0
35 #define LOG_INODE_EXISTS 1
38 * stages for the tree walking. The first
39 * stage (0) is to only pin down the blocks we find
40 * the second stage (1) is to make sure that all the inodes
41 * we find in the log are created in the subvolume.
43 * The last stage is to deal with directories and links and extents
44 * and all the other fun semantics
46 #define LOG_WALK_PIN_ONLY 0
47 #define LOG_WALK_REPLAY_INODES 1
48 #define LOG_WALK_REPLAY_ALL 2
50 static int __btrfs_log_inode(struct btrfs_trans_handle
*trans
,
51 struct btrfs_root
*root
, struct inode
*inode
,
53 static int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
54 struct btrfs_root
*root
,
55 struct btrfs_path
*path
, u64 objectid
);
58 * tree logging is a special write ahead log used to make sure that
59 * fsyncs and O_SYNCs can happen without doing full tree commits.
61 * Full tree commits are expensive because they require commonly
62 * modified blocks to be recowed, creating many dirty pages in the
63 * extent tree an 4x-6x higher write load than ext3.
65 * Instead of doing a tree commit on every fsync, we use the
66 * key ranges and transaction ids to find items for a given file or directory
67 * that have changed in this transaction. Those items are copied into
68 * a special tree (one per subvolume root), that tree is written to disk
69 * and then the fsync is considered complete.
71 * After a crash, items are copied out of the log-tree back into the
72 * subvolume tree. Any file data extents found are recorded in the extent
73 * allocation tree, and the log-tree freed.
75 * The log tree is read three times, once to pin down all the extents it is
76 * using in ram and once, once to create all the inodes logged in the tree
77 * and once to do all the other items.
81 * btrfs_add_log_tree adds a new per-subvolume log tree into the
82 * tree of log tree roots. This must be called with a tree log transaction
83 * running (see start_log_trans).
85 static int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
86 struct btrfs_root
*root
)
89 struct btrfs_root_item root_item
;
90 struct btrfs_inode_item
*inode_item
;
91 struct extent_buffer
*leaf
;
92 struct btrfs_root
*new_root
= root
;
94 u64 objectid
= root
->root_key
.objectid
;
96 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
97 BTRFS_TREE_LOG_OBJECTID
,
98 trans
->transid
, 0, 0, 0);
104 btrfs_set_header_nritems(leaf
, 0);
105 btrfs_set_header_level(leaf
, 0);
106 btrfs_set_header_bytenr(leaf
, leaf
->start
);
107 btrfs_set_header_generation(leaf
, trans
->transid
);
108 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
110 write_extent_buffer(leaf
, root
->fs_info
->fsid
,
111 (unsigned long)btrfs_header_fsid(leaf
),
113 btrfs_mark_buffer_dirty(leaf
);
115 inode_item
= &root_item
.inode
;
116 memset(inode_item
, 0, sizeof(*inode_item
));
117 inode_item
->generation
= cpu_to_le64(1);
118 inode_item
->size
= cpu_to_le64(3);
119 inode_item
->nlink
= cpu_to_le32(1);
120 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
121 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
123 btrfs_set_root_bytenr(&root_item
, leaf
->start
);
124 btrfs_set_root_generation(&root_item
, trans
->transid
);
125 btrfs_set_root_level(&root_item
, 0);
126 btrfs_set_root_refs(&root_item
, 0);
127 btrfs_set_root_used(&root_item
, 0);
129 memset(&root_item
.drop_progress
, 0, sizeof(root_item
.drop_progress
));
130 root_item
.drop_level
= 0;
132 btrfs_tree_unlock(leaf
);
133 free_extent_buffer(leaf
);
136 btrfs_set_root_dirid(&root_item
, 0);
138 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
139 key
.offset
= objectid
;
140 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
141 ret
= btrfs_insert_root(trans
, root
->fs_info
->log_root_tree
, &key
,
146 new_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->log_root_tree
,
150 WARN_ON(root
->log_root
);
151 root
->log_root
= new_root
;
154 * log trees do not get reference counted because they go away
155 * before a real commit is actually done. They do store pointers
156 * to file data extents, and those reference counts still get
157 * updated (along with back refs to the log tree).
159 new_root
->ref_cows
= 0;
160 new_root
->last_trans
= trans
->transid
;
166 * start a sub transaction and setup the log tree
167 * this increments the log tree writer count to make the people
168 * syncing the tree wait for us to finish
170 static int start_log_trans(struct btrfs_trans_handle
*trans
,
171 struct btrfs_root
*root
)
174 mutex_lock(&root
->fs_info
->tree_log_mutex
);
175 if (!root
->fs_info
->log_root_tree
) {
176 ret
= btrfs_init_log_root_tree(trans
, root
->fs_info
);
179 if (!root
->log_root
) {
180 ret
= btrfs_add_log_tree(trans
, root
);
183 atomic_inc(&root
->fs_info
->tree_log_writers
);
184 root
->fs_info
->tree_log_batch
++;
185 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
190 * returns 0 if there was a log transaction running and we were able
191 * to join, or returns -ENOENT if there were not transactions
194 static int join_running_log_trans(struct btrfs_root
*root
)
202 mutex_lock(&root
->fs_info
->tree_log_mutex
);
203 if (root
->log_root
) {
205 atomic_inc(&root
->fs_info
->tree_log_writers
);
206 root
->fs_info
->tree_log_batch
++;
208 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
213 * indicate we're done making changes to the log tree
214 * and wake up anyone waiting to do a sync
216 static int end_log_trans(struct btrfs_root
*root
)
218 atomic_dec(&root
->fs_info
->tree_log_writers
);
220 if (waitqueue_active(&root
->fs_info
->tree_log_wait
))
221 wake_up(&root
->fs_info
->tree_log_wait
);
227 * the walk control struct is used to pass state down the chain when
228 * processing the log tree. The stage field tells us which part
229 * of the log tree processing we are currently doing. The others
230 * are state fields used for that specific part
232 struct walk_control
{
233 /* should we free the extent on disk when done? This is used
234 * at transaction commit time while freeing a log tree
238 /* should we write out the extent buffer? This is used
239 * while flushing the log tree to disk during a sync
243 /* should we wait for the extent buffer io to finish? Also used
244 * while flushing the log tree to disk for a sync
248 /* pin only walk, we record which extents on disk belong to the
253 /* what stage of the replay code we're currently in */
256 /* the root we are currently replaying */
257 struct btrfs_root
*replay_dest
;
259 /* the trans handle for the current replay */
260 struct btrfs_trans_handle
*trans
;
262 /* the function that gets used to process blocks we find in the
263 * tree. Note the extent_buffer might not be up to date when it is
264 * passed in, and it must be checked or read if you need the data
267 int (*process_func
)(struct btrfs_root
*log
, struct extent_buffer
*eb
,
268 struct walk_control
*wc
, u64 gen
);
272 * process_func used to pin down extents, write them or wait on them
274 static int process_one_buffer(struct btrfs_root
*log
,
275 struct extent_buffer
*eb
,
276 struct walk_control
*wc
, u64 gen
)
279 mutex_lock(&log
->fs_info
->pinned_mutex
);
280 btrfs_update_pinned_extents(log
->fs_info
->extent_root
,
281 eb
->start
, eb
->len
, 1);
282 mutex_unlock(&log
->fs_info
->pinned_mutex
);
285 if (btrfs_buffer_uptodate(eb
, gen
)) {
287 btrfs_write_tree_block(eb
);
289 btrfs_wait_tree_block_writeback(eb
);
295 * Item overwrite used by replay and tree logging. eb, slot and key all refer
296 * to the src data we are copying out.
298 * root is the tree we are copying into, and path is a scratch
299 * path for use in this function (it should be released on entry and
300 * will be released on exit).
302 * If the key is already in the destination tree the existing item is
303 * overwritten. If the existing item isn't big enough, it is extended.
304 * If it is too large, it is truncated.
306 * If the key isn't in the destination yet, a new item is inserted.
308 static noinline
int overwrite_item(struct btrfs_trans_handle
*trans
,
309 struct btrfs_root
*root
,
310 struct btrfs_path
*path
,
311 struct extent_buffer
*eb
, int slot
,
312 struct btrfs_key
*key
)
316 u64 saved_i_size
= 0;
317 int save_old_i_size
= 0;
318 unsigned long src_ptr
;
319 unsigned long dst_ptr
;
320 int overwrite_root
= 0;
322 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
325 item_size
= btrfs_item_size_nr(eb
, slot
);
326 src_ptr
= btrfs_item_ptr_offset(eb
, slot
);
328 /* look for the key in the destination tree */
329 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
333 u32 dst_size
= btrfs_item_size_nr(path
->nodes
[0],
335 if (dst_size
!= item_size
)
338 if (item_size
== 0) {
339 btrfs_release_path(root
, path
);
342 dst_copy
= kmalloc(item_size
, GFP_NOFS
);
343 src_copy
= kmalloc(item_size
, GFP_NOFS
);
345 read_extent_buffer(eb
, src_copy
, src_ptr
, item_size
);
347 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
348 read_extent_buffer(path
->nodes
[0], dst_copy
, dst_ptr
,
350 ret
= memcmp(dst_copy
, src_copy
, item_size
);
355 * they have the same contents, just return, this saves
356 * us from cowing blocks in the destination tree and doing
357 * extra writes that may not have been done by a previous
361 btrfs_release_path(root
, path
);
367 btrfs_release_path(root
, path
);
368 /* try to insert the key into the destination tree */
369 ret
= btrfs_insert_empty_item(trans
, root
, path
,
372 /* make sure any existing item is the correct size */
373 if (ret
== -EEXIST
) {
375 found_size
= btrfs_item_size_nr(path
->nodes
[0],
377 if (found_size
> item_size
) {
378 btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
379 } else if (found_size
< item_size
) {
380 ret
= btrfs_extend_item(trans
, root
, path
,
381 item_size
- found_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 if (overwrite_root
&&
438 key
->type
== BTRFS_EXTENT_DATA_KEY
) {
440 struct btrfs_file_extent_item
*fi
;
442 fi
= (struct btrfs_file_extent_item
*)dst_ptr
;
443 extent_type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
444 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
445 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
446 struct btrfs_key ins
;
447 ins
.objectid
= btrfs_file_extent_disk_bytenr(
449 ins
.offset
= btrfs_file_extent_disk_num_bytes(
451 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
454 * is this extent already allocated in the extent
455 * allocation tree? If so, just add a reference
457 ret
= btrfs_lookup_extent(root
, ins
.objectid
,
460 ret
= btrfs_inc_extent_ref(trans
, root
,
461 ins
.objectid
, ins
.offset
,
462 path
->nodes
[0]->start
,
463 root
->root_key
.objectid
,
464 trans
->transid
, key
->objectid
);
467 * insert the extent pointer in the extent
470 ret
= btrfs_alloc_logged_extent(trans
, root
,
471 path
->nodes
[0]->start
,
472 root
->root_key
.objectid
,
473 trans
->transid
, key
->objectid
,
480 btrfs_mark_buffer_dirty(path
->nodes
[0]);
481 btrfs_release_path(root
, path
);
486 * simple helper to read an inode off the disk from a given root
487 * This can only be called for subvolume roots and not for the log
489 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
493 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
494 if (inode
->i_state
& I_NEW
) {
495 BTRFS_I(inode
)->root
= root
;
496 BTRFS_I(inode
)->location
.objectid
= objectid
;
497 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
498 BTRFS_I(inode
)->location
.offset
= 0;
499 btrfs_read_locked_inode(inode
);
500 unlock_new_inode(inode
);
503 if (is_bad_inode(inode
)) {
510 /* replays a single extent in 'eb' at 'slot' with 'key' into the
511 * subvolume 'root'. path is released on entry and should be released
514 * extents in the log tree have not been allocated out of the extent
515 * tree yet. So, this completes the allocation, taking a reference
516 * as required if the extent already exists or creating a new extent
517 * if it isn't in the extent allocation tree yet.
519 * The extent is inserted into the file, dropping any existing extents
520 * from the file that overlap the new one.
522 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
523 struct btrfs_root
*root
,
524 struct btrfs_path
*path
,
525 struct extent_buffer
*eb
, int slot
,
526 struct btrfs_key
*key
)
529 u64 mask
= root
->sectorsize
- 1;
532 u64 start
= key
->offset
;
533 struct btrfs_file_extent_item
*item
;
534 struct inode
*inode
= NULL
;
538 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
539 found_type
= btrfs_file_extent_type(eb
, item
);
541 if (found_type
== BTRFS_FILE_EXTENT_REG
||
542 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
543 extent_end
= start
+ btrfs_file_extent_num_bytes(eb
, item
);
544 else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
545 size
= btrfs_file_extent_inline_len(eb
, item
);
546 extent_end
= (start
+ size
+ mask
) & ~mask
;
552 inode
= read_one_inode(root
, key
->objectid
);
559 * first check to see if we already have this extent in the
560 * file. This must be done before the btrfs_drop_extents run
561 * so we don't try to drop this extent.
563 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
567 (found_type
== BTRFS_FILE_EXTENT_REG
||
568 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)) {
569 struct btrfs_file_extent_item cmp1
;
570 struct btrfs_file_extent_item cmp2
;
571 struct btrfs_file_extent_item
*existing
;
572 struct extent_buffer
*leaf
;
574 leaf
= path
->nodes
[0];
575 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
576 struct btrfs_file_extent_item
);
578 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
580 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
584 * we already have a pointer to this exact extent,
585 * we don't have to do anything
587 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
588 btrfs_release_path(root
, path
);
592 btrfs_release_path(root
, path
);
594 /* drop any overlapping extents */
595 ret
= btrfs_drop_extents(trans
, root
, inode
,
596 start
, extent_end
, start
, &alloc_hint
);
599 /* insert the extent */
600 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
603 /* btrfs_drop_extents changes i_bytes & i_blocks, update it here */
604 inode_add_bytes(inode
, extent_end
- start
);
605 btrfs_update_inode(trans
, root
, inode
);
613 * when cleaning up conflicts between the directory names in the
614 * subvolume, directory names in the log and directory names in the
615 * inode back references, we may have to unlink inodes from directories.
617 * This is a helper function to do the unlink of a specific directory
620 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
621 struct btrfs_root
*root
,
622 struct btrfs_path
*path
,
624 struct btrfs_dir_item
*di
)
629 struct extent_buffer
*leaf
;
630 struct btrfs_key location
;
633 leaf
= path
->nodes
[0];
635 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
636 name_len
= btrfs_dir_name_len(leaf
, di
);
637 name
= kmalloc(name_len
, GFP_NOFS
);
638 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
639 btrfs_release_path(root
, path
);
641 inode
= read_one_inode(root
, location
.objectid
);
644 ret
= link_to_fixup_dir(trans
, root
, path
, location
.objectid
);
646 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
655 * helper function to see if a given name and sequence number found
656 * in an inode back reference are already in a directory and correctly
657 * point to this inode
659 static noinline
int inode_in_dir(struct btrfs_root
*root
,
660 struct btrfs_path
*path
,
661 u64 dirid
, u64 objectid
, u64 index
,
662 const char *name
, int name_len
)
664 struct btrfs_dir_item
*di
;
665 struct btrfs_key location
;
668 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
669 index
, name
, name_len
, 0);
670 if (di
&& !IS_ERR(di
)) {
671 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
672 if (location
.objectid
!= objectid
)
676 btrfs_release_path(root
, path
);
678 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
679 if (di
&& !IS_ERR(di
)) {
680 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
681 if (location
.objectid
!= objectid
)
687 btrfs_release_path(root
, path
);
692 * helper function to check a log tree for a named back reference in
693 * an inode. This is used to decide if a back reference that is
694 * found in the subvolume conflicts with what we find in the log.
696 * inode backreferences may have multiple refs in a single item,
697 * during replay we process one reference at a time, and we don't
698 * want to delete valid links to a file from the subvolume if that
699 * link is also in the log.
701 static noinline
int backref_in_log(struct btrfs_root
*log
,
702 struct btrfs_key
*key
,
703 char *name
, int namelen
)
705 struct btrfs_path
*path
;
706 struct btrfs_inode_ref
*ref
;
708 unsigned long ptr_end
;
709 unsigned long name_ptr
;
715 path
= btrfs_alloc_path();
716 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
720 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
721 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
722 ptr_end
= ptr
+ item_size
;
723 while (ptr
< ptr_end
) {
724 ref
= (struct btrfs_inode_ref
*)ptr
;
725 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
726 if (found_name_len
== namelen
) {
727 name_ptr
= (unsigned long)(ref
+ 1);
728 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
735 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
738 btrfs_free_path(path
);
744 * replay one inode back reference item found in the log tree.
745 * eb, slot and key refer to the buffer and key found in the log tree.
746 * root is the destination we are replaying into, and path is for temp
747 * use by this function. (it should be released on return).
749 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
750 struct btrfs_root
*root
,
751 struct btrfs_root
*log
,
752 struct btrfs_path
*path
,
753 struct extent_buffer
*eb
, int slot
,
754 struct btrfs_key
*key
)
758 struct btrfs_key location
;
759 struct btrfs_inode_ref
*ref
;
760 struct btrfs_dir_item
*di
;
764 unsigned long ref_ptr
;
765 unsigned long ref_end
;
767 location
.objectid
= key
->objectid
;
768 location
.type
= BTRFS_INODE_ITEM_KEY
;
772 * it is possible that we didn't log all the parent directories
773 * for a given inode. If we don't find the dir, just don't
774 * copy the back ref in. The link count fixup code will take
777 dir
= read_one_inode(root
, key
->offset
);
781 inode
= read_one_inode(root
, key
->objectid
);
784 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
785 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
788 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
790 namelen
= btrfs_inode_ref_name_len(eb
, ref
);
791 name
= kmalloc(namelen
, GFP_NOFS
);
794 read_extent_buffer(eb
, name
, (unsigned long)(ref
+ 1), namelen
);
796 /* if we already have a perfect match, we're done */
797 if (inode_in_dir(root
, path
, dir
->i_ino
, inode
->i_ino
,
798 btrfs_inode_ref_index(eb
, ref
),
804 * look for a conflicting back reference in the metadata.
805 * if we find one we have to unlink that name of the file
806 * before we add our new link. Later on, we overwrite any
807 * existing back reference, and we don't want to create
808 * dangling pointers in the directory.
811 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
815 struct btrfs_inode_ref
*victim_ref
;
817 unsigned long ptr_end
;
818 struct extent_buffer
*leaf
= path
->nodes
[0];
820 /* are we trying to overwrite a back ref for the root directory
821 * if so, just jump out, we're done
823 if (key
->objectid
== key
->offset
)
826 /* check all the names in this back reference to see
827 * if they are in the log. if so, we allow them to stay
828 * otherwise they must be unlinked as a conflict
830 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
831 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
832 while (ptr
< ptr_end
) {
833 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
834 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
836 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
837 BUG_ON(!victim_name
);
839 read_extent_buffer(leaf
, victim_name
,
840 (unsigned long)(victim_ref
+ 1),
843 if (!backref_in_log(log
, key
, victim_name
,
845 btrfs_inc_nlink(inode
);
846 btrfs_release_path(root
, path
);
847 ret
= btrfs_unlink_inode(trans
, root
, dir
,
851 btrfs_release_path(root
, path
);
855 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
859 btrfs_release_path(root
, path
);
861 /* look for a conflicting sequence number */
862 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
863 btrfs_inode_ref_index(eb
, ref
),
865 if (di
&& !IS_ERR(di
)) {
866 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
869 btrfs_release_path(root
, path
);
872 /* look for a conflicting name */
873 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
875 if (di
&& !IS_ERR(di
)) {
876 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
879 btrfs_release_path(root
, path
);
881 /* insert our name */
882 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
, 0,
883 btrfs_inode_ref_index(eb
, ref
));
886 btrfs_update_inode(trans
, root
, inode
);
889 ref_ptr
= (unsigned long)(ref
+ 1) + namelen
;
891 if (ref_ptr
< ref_end
)
894 /* finally write the back reference in the inode */
895 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
899 btrfs_release_path(root
, path
);
906 * replay one csum item from the log tree into the subvolume 'root'
907 * eb, slot and key all refer to the log tree
908 * path is for temp use by this function and should be released on return
910 * This copies the checksums out of the log tree and inserts them into
911 * the subvolume. Any existing checksums for this range in the file
912 * are overwritten, and new items are added where required.
914 * We keep this simple by reusing the btrfs_ordered_sum code from
915 * the data=ordered mode. This basically means making a copy
916 * of all the checksums in ram, which we have to do anyway for kmap
919 * The copy is then sent down to btrfs_csum_file_blocks, which
920 * does all the hard work of finding existing items in the file
921 * or adding new ones.
923 static noinline
int replay_one_csum(struct btrfs_trans_handle
*trans
,
924 struct btrfs_root
*root
,
925 struct btrfs_path
*path
,
926 struct extent_buffer
*eb
, int slot
,
927 struct btrfs_key
*key
)
930 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
933 btrfs_super_csum_size(&root
->fs_info
->super_copy
);
934 unsigned long file_bytes
;
935 struct btrfs_ordered_sum
*sums
;
936 struct btrfs_sector_sum
*sector_sum
;
939 file_bytes
= (item_size
/ csum_size
) * root
->sectorsize
;
940 sums
= kzalloc(btrfs_ordered_sum_size(root
, file_bytes
), GFP_NOFS
);
944 INIT_LIST_HEAD(&sums
->list
);
945 sums
->len
= file_bytes
;
946 sums
->bytenr
= key
->offset
;
949 * copy all the sums into the ordered sum struct
951 sector_sum
= sums
->sums
;
952 cur_offset
= key
->offset
;
953 ptr
= btrfs_item_ptr_offset(eb
, slot
);
954 while (item_size
> 0) {
955 sector_sum
->bytenr
= cur_offset
;
956 read_extent_buffer(eb
, §or_sum
->sum
, ptr
, csum_size
);
958 item_size
-= csum_size
;
960 cur_offset
+= root
->sectorsize
;
963 /* let btrfs_csum_file_blocks add them into the file */
964 ret
= btrfs_csum_file_blocks(trans
, root
->fs_info
->csum_root
, sums
);
970 * There are a few corners where the link count of the file can't
971 * be properly maintained during replay. So, instead of adding
972 * lots of complexity to the log code, we just scan the backrefs
973 * for any file that has been through replay.
975 * The scan will update the link count on the inode to reflect the
976 * number of back refs found. If it goes down to zero, the iput
977 * will free the inode.
979 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
980 struct btrfs_root
*root
,
983 struct btrfs_path
*path
;
985 struct btrfs_key key
;
988 unsigned long ptr_end
;
991 key
.objectid
= inode
->i_ino
;
992 key
.type
= BTRFS_INODE_REF_KEY
;
993 key
.offset
= (u64
)-1;
995 path
= btrfs_alloc_path();
998 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1002 if (path
->slots
[0] == 0)
1006 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1008 if (key
.objectid
!= inode
->i_ino
||
1009 key
.type
!= BTRFS_INODE_REF_KEY
)
1011 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1012 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1014 while (ptr
< ptr_end
) {
1015 struct btrfs_inode_ref
*ref
;
1017 ref
= (struct btrfs_inode_ref
*)ptr
;
1018 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1020 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1024 if (key
.offset
== 0)
1027 btrfs_release_path(root
, path
);
1029 btrfs_free_path(path
);
1030 if (nlink
!= inode
->i_nlink
) {
1031 inode
->i_nlink
= nlink
;
1032 btrfs_update_inode(trans
, root
, inode
);
1034 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1039 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1040 struct btrfs_root
*root
,
1041 struct btrfs_path
*path
)
1044 struct btrfs_key key
;
1045 struct inode
*inode
;
1047 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1048 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1049 key
.offset
= (u64
)-1;
1051 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1056 if (path
->slots
[0] == 0)
1061 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1062 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1063 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1066 ret
= btrfs_del_item(trans
, root
, path
);
1069 btrfs_release_path(root
, path
);
1070 inode
= read_one_inode(root
, key
.offset
);
1073 ret
= fixup_inode_link_count(trans
, root
, inode
);
1078 if (key
.offset
== 0)
1082 btrfs_release_path(root
, path
);
1088 * record a given inode in the fixup dir so we can check its link
1089 * count when replay is done. The link count is incremented here
1090 * so the inode won't go away until we check it
1092 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1093 struct btrfs_root
*root
,
1094 struct btrfs_path
*path
,
1097 struct btrfs_key key
;
1099 struct inode
*inode
;
1101 inode
= read_one_inode(root
, objectid
);
1104 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1105 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1106 key
.offset
= objectid
;
1108 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1110 btrfs_release_path(root
, path
);
1112 btrfs_inc_nlink(inode
);
1113 btrfs_update_inode(trans
, root
, inode
);
1114 } else if (ret
== -EEXIST
) {
1125 * when replaying the log for a directory, we only insert names
1126 * for inodes that actually exist. This means an fsync on a directory
1127 * does not implicitly fsync all the new files in it
1129 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1130 struct btrfs_root
*root
,
1131 struct btrfs_path
*path
,
1132 u64 dirid
, u64 index
,
1133 char *name
, int name_len
, u8 type
,
1134 struct btrfs_key
*location
)
1136 struct inode
*inode
;
1140 inode
= read_one_inode(root
, location
->objectid
);
1144 dir
= read_one_inode(root
, dirid
);
1149 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1151 /* FIXME, put inode into FIXUP list */
1159 * take a single entry in a log directory item and replay it into
1162 * if a conflicting item exists in the subdirectory already,
1163 * the inode it points to is unlinked and put into the link count
1166 * If a name from the log points to a file or directory that does
1167 * not exist in the FS, it is skipped. fsyncs on directories
1168 * do not force down inodes inside that directory, just changes to the
1169 * names or unlinks in a directory.
1171 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1172 struct btrfs_root
*root
,
1173 struct btrfs_path
*path
,
1174 struct extent_buffer
*eb
,
1175 struct btrfs_dir_item
*di
,
1176 struct btrfs_key
*key
)
1180 struct btrfs_dir_item
*dst_di
;
1181 struct btrfs_key found_key
;
1182 struct btrfs_key log_key
;
1188 dir
= read_one_inode(root
, key
->objectid
);
1191 name_len
= btrfs_dir_name_len(eb
, di
);
1192 name
= kmalloc(name_len
, GFP_NOFS
);
1193 log_type
= btrfs_dir_type(eb
, di
);
1194 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1197 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1198 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1203 btrfs_release_path(root
, path
);
1205 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1206 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1208 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1209 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1216 if (!dst_di
|| IS_ERR(dst_di
)) {
1217 /* we need a sequence number to insert, so we only
1218 * do inserts for the BTRFS_DIR_INDEX_KEY types
1220 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1225 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1226 /* the existing item matches the logged item */
1227 if (found_key
.objectid
== log_key
.objectid
&&
1228 found_key
.type
== log_key
.type
&&
1229 found_key
.offset
== log_key
.offset
&&
1230 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1235 * don't drop the conflicting directory entry if the inode
1236 * for the new entry doesn't exist
1241 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1244 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1247 btrfs_release_path(root
, path
);
1253 btrfs_release_path(root
, path
);
1254 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1255 name
, name_len
, log_type
, &log_key
);
1257 if (ret
&& ret
!= -ENOENT
)
1263 * find all the names in a directory item and reconcile them into
1264 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1265 * one name in a directory item, but the same code gets used for
1266 * both directory index types
1268 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1269 struct btrfs_root
*root
,
1270 struct btrfs_path
*path
,
1271 struct extent_buffer
*eb
, int slot
,
1272 struct btrfs_key
*key
)
1275 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1276 struct btrfs_dir_item
*di
;
1279 unsigned long ptr_end
;
1281 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1282 ptr_end
= ptr
+ item_size
;
1283 while (ptr
< ptr_end
) {
1284 di
= (struct btrfs_dir_item
*)ptr
;
1285 name_len
= btrfs_dir_name_len(eb
, di
);
1286 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1288 ptr
= (unsigned long)(di
+ 1);
1295 * directory replay has two parts. There are the standard directory
1296 * items in the log copied from the subvolume, and range items
1297 * created in the log while the subvolume was logged.
1299 * The range items tell us which parts of the key space the log
1300 * is authoritative for. During replay, if a key in the subvolume
1301 * directory is in a logged range item, but not actually in the log
1302 * that means it was deleted from the directory before the fsync
1303 * and should be removed.
1305 static noinline
int find_dir_range(struct btrfs_root
*root
,
1306 struct btrfs_path
*path
,
1307 u64 dirid
, int key_type
,
1308 u64
*start_ret
, u64
*end_ret
)
1310 struct btrfs_key key
;
1312 struct btrfs_dir_log_item
*item
;
1316 if (*start_ret
== (u64
)-1)
1319 key
.objectid
= dirid
;
1320 key
.type
= key_type
;
1321 key
.offset
= *start_ret
;
1323 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1327 if (path
->slots
[0] == 0)
1332 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1334 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1338 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1339 struct btrfs_dir_log_item
);
1340 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1342 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1344 *start_ret
= key
.offset
;
1345 *end_ret
= found_end
;
1350 /* check the next slot in the tree to see if it is a valid item */
1351 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1352 if (path
->slots
[0] >= nritems
) {
1353 ret
= btrfs_next_leaf(root
, path
);
1360 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1362 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1366 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1367 struct btrfs_dir_log_item
);
1368 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1369 *start_ret
= key
.offset
;
1370 *end_ret
= found_end
;
1373 btrfs_release_path(root
, path
);
1378 * this looks for a given directory item in the log. If the directory
1379 * item is not in the log, the item is removed and the inode it points
1382 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1383 struct btrfs_root
*root
,
1384 struct btrfs_root
*log
,
1385 struct btrfs_path
*path
,
1386 struct btrfs_path
*log_path
,
1388 struct btrfs_key
*dir_key
)
1391 struct extent_buffer
*eb
;
1394 struct btrfs_dir_item
*di
;
1395 struct btrfs_dir_item
*log_di
;
1398 unsigned long ptr_end
;
1400 struct inode
*inode
;
1401 struct btrfs_key location
;
1404 eb
= path
->nodes
[0];
1405 slot
= path
->slots
[0];
1406 item_size
= btrfs_item_size_nr(eb
, slot
);
1407 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1408 ptr_end
= ptr
+ item_size
;
1409 while (ptr
< ptr_end
) {
1410 di
= (struct btrfs_dir_item
*)ptr
;
1411 name_len
= btrfs_dir_name_len(eb
, di
);
1412 name
= kmalloc(name_len
, GFP_NOFS
);
1417 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1420 if (dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1421 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1424 } else if (dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1425 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1431 if (!log_di
|| IS_ERR(log_di
)) {
1432 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1433 btrfs_release_path(root
, path
);
1434 btrfs_release_path(log
, log_path
);
1435 inode
= read_one_inode(root
, location
.objectid
);
1438 ret
= link_to_fixup_dir(trans
, root
,
1439 path
, location
.objectid
);
1441 btrfs_inc_nlink(inode
);
1442 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1448 /* there might still be more names under this key
1449 * check and repeat if required
1451 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1458 btrfs_release_path(log
, log_path
);
1461 ptr
= (unsigned long)(di
+ 1);
1466 btrfs_release_path(root
, path
);
1467 btrfs_release_path(log
, log_path
);
1472 * deletion replay happens before we copy any new directory items
1473 * out of the log or out of backreferences from inodes. It
1474 * scans the log to find ranges of keys that log is authoritative for,
1475 * and then scans the directory to find items in those ranges that are
1476 * not present in the log.
1478 * Anything we don't find in the log is unlinked and removed from the
1481 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1482 struct btrfs_root
*root
,
1483 struct btrfs_root
*log
,
1484 struct btrfs_path
*path
,
1489 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1491 struct btrfs_key dir_key
;
1492 struct btrfs_key found_key
;
1493 struct btrfs_path
*log_path
;
1496 dir_key
.objectid
= dirid
;
1497 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1498 log_path
= btrfs_alloc_path();
1502 dir
= read_one_inode(root
, dirid
);
1503 /* it isn't an error if the inode isn't there, that can happen
1504 * because we replay the deletes before we copy in the inode item
1508 btrfs_free_path(log_path
);
1515 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1516 &range_start
, &range_end
);
1520 dir_key
.offset
= range_start
;
1523 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1528 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1529 if (path
->slots
[0] >= nritems
) {
1530 ret
= btrfs_next_leaf(root
, path
);
1534 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1536 if (found_key
.objectid
!= dirid
||
1537 found_key
.type
!= dir_key
.type
)
1540 if (found_key
.offset
> range_end
)
1543 ret
= check_item_in_log(trans
, root
, log
, path
,
1544 log_path
, dir
, &found_key
);
1546 if (found_key
.offset
== (u64
)-1)
1548 dir_key
.offset
= found_key
.offset
+ 1;
1550 btrfs_release_path(root
, path
);
1551 if (range_end
== (u64
)-1)
1553 range_start
= range_end
+ 1;
1558 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1559 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1560 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1561 btrfs_release_path(root
, path
);
1565 btrfs_release_path(root
, path
);
1566 btrfs_free_path(log_path
);
1572 * the process_func used to replay items from the log tree. This
1573 * gets called in two different stages. The first stage just looks
1574 * for inodes and makes sure they are all copied into the subvolume.
1576 * The second stage copies all the other item types from the log into
1577 * the subvolume. The two stage approach is slower, but gets rid of
1578 * lots of complexity around inodes referencing other inodes that exist
1579 * only in the log (references come from either directory items or inode
1582 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
1583 struct walk_control
*wc
, u64 gen
)
1586 struct btrfs_path
*path
;
1587 struct btrfs_root
*root
= wc
->replay_dest
;
1588 struct btrfs_key key
;
1594 btrfs_read_buffer(eb
, gen
);
1596 level
= btrfs_header_level(eb
);
1601 path
= btrfs_alloc_path();
1604 nritems
= btrfs_header_nritems(eb
);
1605 for (i
= 0; i
< nritems
; i
++) {
1606 btrfs_item_key_to_cpu(eb
, &key
, i
);
1607 item_size
= btrfs_item_size_nr(eb
, i
);
1609 /* inode keys are done during the first stage */
1610 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
1611 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
1612 struct inode
*inode
;
1613 struct btrfs_inode_item
*inode_item
;
1616 inode_item
= btrfs_item_ptr(eb
, i
,
1617 struct btrfs_inode_item
);
1618 mode
= btrfs_inode_mode(eb
, inode_item
);
1619 if (S_ISDIR(mode
)) {
1620 ret
= replay_dir_deletes(wc
->trans
,
1621 root
, log
, path
, key
.objectid
);
1624 ret
= overwrite_item(wc
->trans
, root
, path
,
1628 /* for regular files, truncate away
1629 * extents past the new EOF
1631 if (S_ISREG(mode
)) {
1632 inode
= read_one_inode(root
,
1636 ret
= btrfs_truncate_inode_items(wc
->trans
,
1637 root
, inode
, inode
->i_size
,
1638 BTRFS_EXTENT_DATA_KEY
);
1642 ret
= link_to_fixup_dir(wc
->trans
, root
,
1643 path
, key
.objectid
);
1646 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
1649 /* these keys are simply copied */
1650 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
1651 ret
= overwrite_item(wc
->trans
, root
, path
,
1654 } else if (key
.type
== BTRFS_INODE_REF_KEY
) {
1655 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
1657 BUG_ON(ret
&& ret
!= -ENOENT
);
1658 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
1659 ret
= replay_one_extent(wc
->trans
, root
, path
,
1662 } else if (key
.type
== BTRFS_EXTENT_CSUM_KEY
) {
1663 ret
= replay_one_csum(wc
->trans
, root
, path
,
1666 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
||
1667 key
.type
== BTRFS_DIR_INDEX_KEY
) {
1668 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
1673 btrfs_free_path(path
);
1677 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
1678 struct btrfs_root
*root
,
1679 struct btrfs_path
*path
, int *level
,
1680 struct walk_control
*wc
)
1686 struct extent_buffer
*next
;
1687 struct extent_buffer
*cur
;
1688 struct extent_buffer
*parent
;
1692 WARN_ON(*level
< 0);
1693 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1695 while (*level
> 0) {
1696 WARN_ON(*level
< 0);
1697 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1698 cur
= path
->nodes
[*level
];
1700 if (btrfs_header_level(cur
) != *level
)
1703 if (path
->slots
[*level
] >=
1704 btrfs_header_nritems(cur
))
1707 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
1708 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
1709 blocksize
= btrfs_level_size(root
, *level
- 1);
1711 parent
= path
->nodes
[*level
];
1712 root_owner
= btrfs_header_owner(parent
);
1713 root_gen
= btrfs_header_generation(parent
);
1715 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1717 wc
->process_func(root
, next
, wc
, ptr_gen
);
1720 path
->slots
[*level
]++;
1722 btrfs_read_buffer(next
, ptr_gen
);
1724 btrfs_tree_lock(next
);
1725 clean_tree_block(trans
, root
, next
);
1726 btrfs_wait_tree_block_writeback(next
);
1727 btrfs_tree_unlock(next
);
1729 ret
= btrfs_drop_leaf_ref(trans
, root
, next
);
1732 WARN_ON(root_owner
!=
1733 BTRFS_TREE_LOG_OBJECTID
);
1734 ret
= btrfs_free_reserved_extent(root
,
1738 free_extent_buffer(next
);
1741 btrfs_read_buffer(next
, ptr_gen
);
1743 WARN_ON(*level
<= 0);
1744 if (path
->nodes
[*level
-1])
1745 free_extent_buffer(path
->nodes
[*level
-1]);
1746 path
->nodes
[*level
-1] = next
;
1747 *level
= btrfs_header_level(next
);
1748 path
->slots
[*level
] = 0;
1751 WARN_ON(*level
< 0);
1752 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1754 if (path
->nodes
[*level
] == root
->node
)
1755 parent
= path
->nodes
[*level
];
1757 parent
= path
->nodes
[*level
+ 1];
1759 bytenr
= path
->nodes
[*level
]->start
;
1761 blocksize
= btrfs_level_size(root
, *level
);
1762 root_owner
= btrfs_header_owner(parent
);
1763 root_gen
= btrfs_header_generation(parent
);
1765 wc
->process_func(root
, path
->nodes
[*level
], wc
,
1766 btrfs_header_generation(path
->nodes
[*level
]));
1769 next
= path
->nodes
[*level
];
1770 btrfs_tree_lock(next
);
1771 clean_tree_block(trans
, root
, next
);
1772 btrfs_wait_tree_block_writeback(next
);
1773 btrfs_tree_unlock(next
);
1776 ret
= btrfs_drop_leaf_ref(trans
, root
, next
);
1779 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1780 ret
= btrfs_free_reserved_extent(root
, bytenr
, blocksize
);
1783 free_extent_buffer(path
->nodes
[*level
]);
1784 path
->nodes
[*level
] = NULL
;
1791 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
1792 struct btrfs_root
*root
,
1793 struct btrfs_path
*path
, int *level
,
1794 struct walk_control
*wc
)
1802 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
1803 slot
= path
->slots
[i
];
1804 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
1805 struct extent_buffer
*node
;
1806 node
= path
->nodes
[i
];
1809 WARN_ON(*level
== 0);
1812 struct extent_buffer
*parent
;
1813 if (path
->nodes
[*level
] == root
->node
)
1814 parent
= path
->nodes
[*level
];
1816 parent
= path
->nodes
[*level
+ 1];
1818 root_owner
= btrfs_header_owner(parent
);
1819 root_gen
= btrfs_header_generation(parent
);
1820 wc
->process_func(root
, path
->nodes
[*level
], wc
,
1821 btrfs_header_generation(path
->nodes
[*level
]));
1823 struct extent_buffer
*next
;
1825 next
= path
->nodes
[*level
];
1827 btrfs_tree_lock(next
);
1828 clean_tree_block(trans
, root
, next
);
1829 btrfs_wait_tree_block_writeback(next
);
1830 btrfs_tree_unlock(next
);
1833 ret
= btrfs_drop_leaf_ref(trans
, root
,
1838 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1839 ret
= btrfs_free_reserved_extent(root
,
1840 path
->nodes
[*level
]->start
,
1841 path
->nodes
[*level
]->len
);
1844 free_extent_buffer(path
->nodes
[*level
]);
1845 path
->nodes
[*level
] = NULL
;
1853 * drop the reference count on the tree rooted at 'snap'. This traverses
1854 * the tree freeing any blocks that have a ref count of zero after being
1857 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
1858 struct btrfs_root
*log
, struct walk_control
*wc
)
1863 struct btrfs_path
*path
;
1867 path
= btrfs_alloc_path();
1870 level
= btrfs_header_level(log
->node
);
1872 path
->nodes
[level
] = log
->node
;
1873 extent_buffer_get(log
->node
);
1874 path
->slots
[level
] = 0;
1877 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
1883 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
1890 /* was the root node processed? if not, catch it here */
1891 if (path
->nodes
[orig_level
]) {
1892 wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
1893 btrfs_header_generation(path
->nodes
[orig_level
]));
1895 struct extent_buffer
*next
;
1897 next
= path
->nodes
[orig_level
];
1899 btrfs_tree_lock(next
);
1900 clean_tree_block(trans
, log
, next
);
1901 btrfs_wait_tree_block_writeback(next
);
1902 btrfs_tree_unlock(next
);
1904 if (orig_level
== 0) {
1905 ret
= btrfs_drop_leaf_ref(trans
, log
,
1909 WARN_ON(log
->root_key
.objectid
!=
1910 BTRFS_TREE_LOG_OBJECTID
);
1911 ret
= btrfs_free_reserved_extent(log
, next
->start
,
1917 for (i
= 0; i
<= orig_level
; i
++) {
1918 if (path
->nodes
[i
]) {
1919 free_extent_buffer(path
->nodes
[i
]);
1920 path
->nodes
[i
] = NULL
;
1923 btrfs_free_path(path
);
1925 free_extent_buffer(log
->node
);
1929 static int wait_log_commit(struct btrfs_root
*log
)
1932 u64 transid
= log
->fs_info
->tree_log_transid
;
1935 prepare_to_wait(&log
->fs_info
->tree_log_wait
, &wait
,
1936 TASK_UNINTERRUPTIBLE
);
1937 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
1938 if (atomic_read(&log
->fs_info
->tree_log_commit
))
1940 finish_wait(&log
->fs_info
->tree_log_wait
, &wait
);
1941 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1942 } while (transid
== log
->fs_info
->tree_log_transid
&&
1943 atomic_read(&log
->fs_info
->tree_log_commit
));
1948 * btrfs_sync_log does sends a given tree log down to the disk and
1949 * updates the super blocks to record it. When this call is done,
1950 * you know that any inodes previously logged are safely on disk
1952 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
1953 struct btrfs_root
*root
)
1956 unsigned long batch
;
1957 struct btrfs_root
*log
= root
->log_root
;
1959 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1960 if (atomic_read(&log
->fs_info
->tree_log_commit
)) {
1961 wait_log_commit(log
);
1964 atomic_set(&log
->fs_info
->tree_log_commit
, 1);
1967 batch
= log
->fs_info
->tree_log_batch
;
1968 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
1969 schedule_timeout_uninterruptible(1);
1970 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1972 while (atomic_read(&log
->fs_info
->tree_log_writers
)) {
1974 prepare_to_wait(&log
->fs_info
->tree_log_wait
, &wait
,
1975 TASK_UNINTERRUPTIBLE
);
1976 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
1977 if (atomic_read(&log
->fs_info
->tree_log_writers
))
1979 mutex_lock(&log
->fs_info
->tree_log_mutex
);
1980 finish_wait(&log
->fs_info
->tree_log_wait
, &wait
);
1982 if (batch
== log
->fs_info
->tree_log_batch
)
1986 ret
= btrfs_write_and_wait_marked_extents(log
, &log
->dirty_log_pages
);
1988 ret
= btrfs_write_and_wait_marked_extents(root
->fs_info
->log_root_tree
,
1989 &root
->fs_info
->log_root_tree
->dirty_log_pages
);
1992 btrfs_set_super_log_root(&root
->fs_info
->super_for_commit
,
1993 log
->fs_info
->log_root_tree
->node
->start
);
1994 btrfs_set_super_log_root_level(&root
->fs_info
->super_for_commit
,
1995 btrfs_header_level(log
->fs_info
->log_root_tree
->node
));
1997 write_ctree_super(trans
, log
->fs_info
->tree_root
, 2);
1998 log
->fs_info
->tree_log_transid
++;
1999 log
->fs_info
->tree_log_batch
= 0;
2000 atomic_set(&log
->fs_info
->tree_log_commit
, 0);
2002 if (waitqueue_active(&log
->fs_info
->tree_log_wait
))
2003 wake_up(&log
->fs_info
->tree_log_wait
);
2005 mutex_unlock(&log
->fs_info
->tree_log_mutex
);
2009 /* * free all the extents used by the tree log. This should be called
2010 * at commit time of the full transaction
2012 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2015 struct btrfs_root
*log
;
2019 struct walk_control wc
= {
2021 .process_func
= process_one_buffer
2024 if (!root
->log_root
)
2027 log
= root
->log_root
;
2028 ret
= walk_log_tree(trans
, log
, &wc
);
2032 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
2033 0, &start
, &end
, EXTENT_DIRTY
);
2037 clear_extent_dirty(&log
->dirty_log_pages
,
2038 start
, end
, GFP_NOFS
);
2041 log
= root
->log_root
;
2042 ret
= btrfs_del_root(trans
, root
->fs_info
->log_root_tree
,
2045 root
->log_root
= NULL
;
2046 kfree(root
->log_root
);
2051 * helper function to update the item for a given subvolumes log root
2052 * in the tree of log roots
2054 static int update_log_root(struct btrfs_trans_handle
*trans
,
2055 struct btrfs_root
*log
)
2057 u64 bytenr
= btrfs_root_bytenr(&log
->root_item
);
2060 if (log
->node
->start
== bytenr
)
2063 btrfs_set_root_bytenr(&log
->root_item
, log
->node
->start
);
2064 btrfs_set_root_generation(&log
->root_item
, trans
->transid
);
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 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
2107 ret
= join_running_log_trans(root
);
2111 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2113 log
= root
->log_root
;
2114 path
= btrfs_alloc_path();
2115 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir
->i_ino
,
2116 name
, name_len
, -1);
2117 if (di
&& !IS_ERR(di
)) {
2118 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2119 bytes_del
+= name_len
;
2122 btrfs_release_path(log
, path
);
2123 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir
->i_ino
,
2124 index
, name
, name_len
, -1);
2125 if (di
&& !IS_ERR(di
)) {
2126 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2127 bytes_del
+= name_len
;
2131 /* update the directory size in the log to reflect the names
2135 struct btrfs_key key
;
2137 key
.objectid
= dir
->i_ino
;
2139 key
.type
= BTRFS_INODE_ITEM_KEY
;
2140 btrfs_release_path(log
, path
);
2142 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2144 struct btrfs_inode_item
*item
;
2147 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2148 struct btrfs_inode_item
);
2149 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2150 if (i_size
> bytes_del
)
2151 i_size
-= bytes_del
;
2154 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2155 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2158 btrfs_release_path(log
, path
);
2161 btrfs_free_path(path
);
2162 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2163 end_log_trans(root
);
2168 /* see comments for btrfs_del_dir_entries_in_log */
2169 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2170 struct btrfs_root
*root
,
2171 const char *name
, int name_len
,
2172 struct inode
*inode
, u64 dirid
)
2174 struct btrfs_root
*log
;
2178 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
2181 ret
= join_running_log_trans(root
);
2184 log
= root
->log_root
;
2185 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2187 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, inode
->i_ino
,
2189 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2190 end_log_trans(root
);
2196 * creates a range item in the log for 'dirid'. first_offset and
2197 * last_offset tell us which parts of the key space the log should
2198 * be considered authoritative for.
2200 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2201 struct btrfs_root
*log
,
2202 struct btrfs_path
*path
,
2203 int key_type
, u64 dirid
,
2204 u64 first_offset
, u64 last_offset
)
2207 struct btrfs_key key
;
2208 struct btrfs_dir_log_item
*item
;
2210 key
.objectid
= dirid
;
2211 key
.offset
= first_offset
;
2212 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2213 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2215 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2216 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2219 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2220 struct btrfs_dir_log_item
);
2221 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2222 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2223 btrfs_release_path(log
, path
);
2228 * log all the items included in the current transaction for a given
2229 * directory. This also creates the range items in the log tree required
2230 * to replay anything deleted before the fsync
2232 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2233 struct btrfs_root
*root
, struct inode
*inode
,
2234 struct btrfs_path
*path
,
2235 struct btrfs_path
*dst_path
, int key_type
,
2236 u64 min_offset
, u64
*last_offset_ret
)
2238 struct btrfs_key min_key
;
2239 struct btrfs_key max_key
;
2240 struct btrfs_root
*log
= root
->log_root
;
2241 struct extent_buffer
*src
;
2245 u64 first_offset
= min_offset
;
2246 u64 last_offset
= (u64
)-1;
2248 log
= root
->log_root
;
2249 max_key
.objectid
= inode
->i_ino
;
2250 max_key
.offset
= (u64
)-1;
2251 max_key
.type
= key_type
;
2253 min_key
.objectid
= inode
->i_ino
;
2254 min_key
.type
= key_type
;
2255 min_key
.offset
= min_offset
;
2257 path
->keep_locks
= 1;
2259 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2260 path
, 0, trans
->transid
);
2263 * we didn't find anything from this transaction, see if there
2264 * is anything at all
2266 if (ret
!= 0 || min_key
.objectid
!= inode
->i_ino
||
2267 min_key
.type
!= key_type
) {
2268 min_key
.objectid
= inode
->i_ino
;
2269 min_key
.type
= key_type
;
2270 min_key
.offset
= (u64
)-1;
2271 btrfs_release_path(root
, path
);
2272 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2274 btrfs_release_path(root
, path
);
2277 ret
= btrfs_previous_item(root
, path
, inode
->i_ino
, key_type
);
2279 /* if ret == 0 there are items for this type,
2280 * create a range to tell us the last key of this type.
2281 * otherwise, there are no items in this directory after
2282 * *min_offset, and we create a range to indicate that.
2285 struct btrfs_key tmp
;
2286 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2288 if (key_type
== tmp
.type
)
2289 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2294 /* go backward to find any previous key */
2295 ret
= btrfs_previous_item(root
, path
, inode
->i_ino
, key_type
);
2297 struct btrfs_key tmp
;
2298 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2299 if (key_type
== tmp
.type
) {
2300 first_offset
= tmp
.offset
;
2301 ret
= overwrite_item(trans
, log
, dst_path
,
2302 path
->nodes
[0], path
->slots
[0],
2306 btrfs_release_path(root
, path
);
2308 /* find the first key from this transaction again */
2309 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2316 * we have a block from this transaction, log every item in it
2317 * from our directory
2320 struct btrfs_key tmp
;
2321 src
= path
->nodes
[0];
2322 nritems
= btrfs_header_nritems(src
);
2323 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2324 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2326 if (min_key
.objectid
!= inode
->i_ino
||
2327 min_key
.type
!= key_type
)
2329 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2333 path
->slots
[0] = nritems
;
2336 * look ahead to the next item and see if it is also
2337 * from this directory and from this transaction
2339 ret
= btrfs_next_leaf(root
, path
);
2341 last_offset
= (u64
)-1;
2344 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2345 if (tmp
.objectid
!= inode
->i_ino
|| tmp
.type
!= key_type
) {
2346 last_offset
= (u64
)-1;
2349 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2350 ret
= overwrite_item(trans
, log
, dst_path
,
2351 path
->nodes
[0], path
->slots
[0],
2355 last_offset
= tmp
.offset
;
2360 *last_offset_ret
= last_offset
;
2361 btrfs_release_path(root
, path
);
2362 btrfs_release_path(log
, dst_path
);
2364 /* insert the log range keys to indicate where the log is valid */
2365 ret
= insert_dir_log_key(trans
, log
, path
, key_type
, inode
->i_ino
,
2366 first_offset
, last_offset
);
2372 * logging directories is very similar to logging inodes, We find all the items
2373 * from the current transaction and write them to the log.
2375 * The recovery code scans the directory in the subvolume, and if it finds a
2376 * key in the range logged that is not present in the log tree, then it means
2377 * that dir entry was unlinked during the transaction.
2379 * In order for that scan to work, we must include one key smaller than
2380 * the smallest logged by this transaction and one key larger than the largest
2381 * key logged by this transaction.
2383 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
2384 struct btrfs_root
*root
, struct inode
*inode
,
2385 struct btrfs_path
*path
,
2386 struct btrfs_path
*dst_path
)
2391 int key_type
= BTRFS_DIR_ITEM_KEY
;
2397 ret
= log_dir_items(trans
, root
, inode
, path
,
2398 dst_path
, key_type
, min_key
,
2401 if (max_key
== (u64
)-1)
2403 min_key
= max_key
+ 1;
2406 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
2407 key_type
= BTRFS_DIR_INDEX_KEY
;
2414 * a helper function to drop items from the log before we relog an
2415 * inode. max_key_type indicates the highest item type to remove.
2416 * This cannot be run for file data extents because it does not
2417 * free the extents they point to.
2419 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
2420 struct btrfs_root
*log
,
2421 struct btrfs_path
*path
,
2422 u64 objectid
, int max_key_type
)
2425 struct btrfs_key key
;
2426 struct btrfs_key found_key
;
2428 key
.objectid
= objectid
;
2429 key
.type
= max_key_type
;
2430 key
.offset
= (u64
)-1;
2433 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
2438 if (path
->slots
[0] == 0)
2442 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2445 if (found_key
.objectid
!= objectid
)
2448 ret
= btrfs_del_item(trans
, log
, path
);
2450 btrfs_release_path(log
, path
);
2452 btrfs_release_path(log
, path
);
2456 static noinline
int copy_extent_csums(struct btrfs_trans_handle
*trans
,
2457 struct list_head
*list
,
2458 struct btrfs_root
*root
,
2459 u64 disk_bytenr
, u64 len
)
2461 struct btrfs_ordered_sum
*sums
;
2462 struct btrfs_sector_sum
*sector_sum
;
2464 struct btrfs_path
*path
;
2465 struct btrfs_csum_item
*item
= NULL
;
2466 u64 end
= disk_bytenr
+ len
;
2467 u64 item_start_offset
= 0;
2468 u64 item_last_offset
= 0;
2471 u16 csum_size
= btrfs_super_csum_size(&root
->fs_info
->super_copy
);
2473 sums
= kzalloc(btrfs_ordered_sum_size(root
, len
), GFP_NOFS
);
2475 sector_sum
= sums
->sums
;
2476 sums
->bytenr
= disk_bytenr
;
2478 list_add_tail(&sums
->list
, list
);
2480 path
= btrfs_alloc_path();
2481 while (disk_bytenr
< end
) {
2482 if (!item
|| disk_bytenr
< item_start_offset
||
2483 disk_bytenr
>= item_last_offset
) {
2484 struct btrfs_key found_key
;
2488 btrfs_release_path(root
, path
);
2489 item
= btrfs_lookup_csum(NULL
, root
, path
,
2492 ret
= PTR_ERR(item
);
2493 if (ret
== -ENOENT
|| ret
== -EFBIG
)
2496 printk(KERN_INFO
"log no csum found for "
2498 (unsigned long long)disk_bytenr
);
2500 btrfs_release_path(root
, path
);
2503 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2506 item_start_offset
= found_key
.offset
;
2507 item_size
= btrfs_item_size_nr(path
->nodes
[0],
2509 item_last_offset
= item_start_offset
+
2510 (item_size
/ csum_size
) *
2512 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2513 struct btrfs_csum_item
);
2516 * this byte range must be able to fit inside
2517 * a single leaf so it will also fit inside a u32
2519 diff
= disk_bytenr
- item_start_offset
;
2520 diff
= diff
/ root
->sectorsize
;
2521 diff
= diff
* csum_size
;
2523 read_extent_buffer(path
->nodes
[0], &sum
,
2524 ((unsigned long)item
) + diff
,
2527 sector_sum
->bytenr
= disk_bytenr
;
2528 sector_sum
->sum
= sum
;
2529 disk_bytenr
+= root
->sectorsize
;
2532 btrfs_free_path(path
);
2536 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
2537 struct btrfs_root
*log
,
2538 struct btrfs_path
*dst_path
,
2539 struct extent_buffer
*src
,
2540 int start_slot
, int nr
, int inode_only
)
2542 unsigned long src_offset
;
2543 unsigned long dst_offset
;
2544 struct btrfs_file_extent_item
*extent
;
2545 struct btrfs_inode_item
*inode_item
;
2547 struct btrfs_key
*ins_keys
;
2551 struct list_head ordered_sums
;
2553 INIT_LIST_HEAD(&ordered_sums
);
2555 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
2556 nr
* sizeof(u32
), GFP_NOFS
);
2557 ins_sizes
= (u32
*)ins_data
;
2558 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
2560 for (i
= 0; i
< nr
; i
++) {
2561 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
2562 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
2564 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
2565 ins_keys
, ins_sizes
, nr
);
2568 for (i
= 0; i
< nr
; i
++) {
2569 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
2570 dst_path
->slots
[0]);
2572 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
2574 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
2575 src_offset
, ins_sizes
[i
]);
2577 if (inode_only
== LOG_INODE_EXISTS
&&
2578 ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
2579 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
2581 struct btrfs_inode_item
);
2582 btrfs_set_inode_size(dst_path
->nodes
[0], inode_item
, 0);
2584 /* set the generation to zero so the recover code
2585 * can tell the difference between an logging
2586 * just to say 'this inode exists' and a logging
2587 * to say 'update this inode with these values'
2589 btrfs_set_inode_generation(dst_path
->nodes
[0],
2592 /* take a reference on file data extents so that truncates
2593 * or deletes of this inode don't have to relog the inode
2596 if (btrfs_key_type(ins_keys
+ i
) == BTRFS_EXTENT_DATA_KEY
) {
2598 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
2599 struct btrfs_file_extent_item
);
2601 found_type
= btrfs_file_extent_type(src
, extent
);
2602 if (found_type
== BTRFS_FILE_EXTENT_REG
||
2603 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
2604 u64 ds
= btrfs_file_extent_disk_bytenr(src
,
2606 u64 dl
= btrfs_file_extent_disk_num_bytes(src
,
2608 u64 cs
= btrfs_file_extent_offset(src
, extent
);
2609 u64 cl
= btrfs_file_extent_num_bytes(src
,
2611 if (btrfs_file_extent_compression(src
,
2616 /* ds == 0 is a hole */
2618 ret
= btrfs_inc_extent_ref(trans
, log
,
2620 dst_path
->nodes
[0]->start
,
2621 BTRFS_TREE_LOG_OBJECTID
,
2623 ins_keys
[i
].objectid
);
2625 ret
= copy_extent_csums(trans
,
2627 log
->fs_info
->csum_root
,
2633 dst_path
->slots
[0]++;
2636 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
2637 btrfs_release_path(log
, dst_path
);
2641 * we have to do this after the loop above to avoid changing the
2642 * log tree while trying to change the log tree.
2644 while (!list_empty(&ordered_sums
)) {
2645 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
2646 struct btrfs_ordered_sum
,
2648 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
2650 list_del(&sums
->list
);
2656 /* log a single inode in the tree log.
2657 * At least one parent directory for this inode must exist in the tree
2658 * or be logged already.
2660 * Any items from this inode changed by the current transaction are copied
2661 * to the log tree. An extra reference is taken on any extents in this
2662 * file, allowing us to avoid a whole pile of corner cases around logging
2663 * blocks that have been removed from the tree.
2665 * See LOG_INODE_ALL and related defines for a description of what inode_only
2668 * This handles both files and directories.
2670 static int __btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2671 struct btrfs_root
*root
, struct inode
*inode
,
2674 struct btrfs_path
*path
;
2675 struct btrfs_path
*dst_path
;
2676 struct btrfs_key min_key
;
2677 struct btrfs_key max_key
;
2678 struct btrfs_root
*log
= root
->log_root
;
2679 struct extent_buffer
*src
= NULL
;
2683 int ins_start_slot
= 0;
2686 log
= root
->log_root
;
2688 path
= btrfs_alloc_path();
2689 dst_path
= btrfs_alloc_path();
2691 min_key
.objectid
= inode
->i_ino
;
2692 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
2695 max_key
.objectid
= inode
->i_ino
;
2696 if (inode_only
== LOG_INODE_EXISTS
|| S_ISDIR(inode
->i_mode
))
2697 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
2699 max_key
.type
= (u8
)-1;
2700 max_key
.offset
= (u64
)-1;
2703 * if this inode has already been logged and we're in inode_only
2704 * mode, we don't want to delete the things that have already
2705 * been written to the log.
2707 * But, if the inode has been through an inode_only log,
2708 * the logged_trans field is not set. This allows us to catch
2709 * any new names for this inode in the backrefs by logging it
2712 if (inode_only
== LOG_INODE_EXISTS
&&
2713 BTRFS_I(inode
)->logged_trans
== trans
->transid
) {
2714 btrfs_free_path(path
);
2715 btrfs_free_path(dst_path
);
2718 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2721 * a brute force approach to making sure we get the most uptodate
2722 * copies of everything.
2724 if (S_ISDIR(inode
->i_mode
)) {
2725 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
2727 if (inode_only
== LOG_INODE_EXISTS
)
2728 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
2729 ret
= drop_objectid_items(trans
, log
, path
,
2730 inode
->i_ino
, max_key_type
);
2732 ret
= btrfs_truncate_inode_items(trans
, log
, inode
, 0, 0);
2735 path
->keep_locks
= 1;
2739 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2740 path
, 0, trans
->transid
);
2744 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2745 if (min_key
.objectid
!= inode
->i_ino
)
2747 if (min_key
.type
> max_key
.type
)
2750 src
= path
->nodes
[0];
2751 size
= btrfs_item_size_nr(src
, path
->slots
[0]);
2752 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
2755 } else if (!ins_nr
) {
2756 ins_start_slot
= path
->slots
[0];
2761 ret
= copy_items(trans
, log
, dst_path
, src
, ins_start_slot
,
2762 ins_nr
, inode_only
);
2765 ins_start_slot
= path
->slots
[0];
2768 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2770 if (path
->slots
[0] < nritems
) {
2771 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
2776 ret
= copy_items(trans
, log
, dst_path
, src
,
2778 ins_nr
, inode_only
);
2782 btrfs_release_path(root
, path
);
2784 if (min_key
.offset
< (u64
)-1)
2786 else if (min_key
.type
< (u8
)-1)
2788 else if (min_key
.objectid
< (u64
)-1)
2794 ret
= copy_items(trans
, log
, dst_path
, src
,
2796 ins_nr
, inode_only
);
2801 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
2802 btrfs_release_path(root
, path
);
2803 btrfs_release_path(log
, dst_path
);
2804 BTRFS_I(inode
)->log_dirty_trans
= 0;
2805 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
2808 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
2809 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2811 btrfs_free_path(path
);
2812 btrfs_free_path(dst_path
);
2814 mutex_lock(&root
->fs_info
->tree_log_mutex
);
2815 ret
= update_log_root(trans
, log
);
2817 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
2822 int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2823 struct btrfs_root
*root
, struct inode
*inode
,
2828 start_log_trans(trans
, root
);
2829 ret
= __btrfs_log_inode(trans
, root
, inode
, inode_only
);
2830 end_log_trans(root
);
2835 * helper function around btrfs_log_inode to make sure newly created
2836 * parent directories also end up in the log. A minimal inode and backref
2837 * only logging is done of any parent directories that are older than
2838 * the last committed transaction
2840 int btrfs_log_dentry(struct btrfs_trans_handle
*trans
,
2841 struct btrfs_root
*root
, struct dentry
*dentry
)
2843 int inode_only
= LOG_INODE_ALL
;
2844 struct super_block
*sb
;
2847 start_log_trans(trans
, root
);
2848 sb
= dentry
->d_inode
->i_sb
;
2850 ret
= __btrfs_log_inode(trans
, root
, dentry
->d_inode
,
2853 inode_only
= LOG_INODE_EXISTS
;
2855 dentry
= dentry
->d_parent
;
2856 if (!dentry
|| !dentry
->d_inode
|| sb
!= dentry
->d_inode
->i_sb
)
2859 if (BTRFS_I(dentry
->d_inode
)->generation
<=
2860 root
->fs_info
->last_trans_committed
)
2863 end_log_trans(root
);
2868 * it is not safe to log dentry if the chunk root has added new
2869 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2870 * If this returns 1, you must commit the transaction to safely get your
2873 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
2874 struct btrfs_root
*root
, struct dentry
*dentry
)
2877 gen
= root
->fs_info
->last_trans_new_blockgroup
;
2878 if (gen
> root
->fs_info
->last_trans_committed
)
2881 return btrfs_log_dentry(trans
, root
, dentry
);
2885 * should be called during mount to recover any replay any log trees
2888 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
2891 struct btrfs_path
*path
;
2892 struct btrfs_trans_handle
*trans
;
2893 struct btrfs_key key
;
2894 struct btrfs_key found_key
;
2895 struct btrfs_key tmp_key
;
2896 struct btrfs_root
*log
;
2897 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
2899 struct walk_control wc
= {
2900 .process_func
= process_one_buffer
,
2904 fs_info
->log_root_recovering
= 1;
2905 path
= btrfs_alloc_path();
2908 trans
= btrfs_start_transaction(fs_info
->tree_root
, 1);
2913 walk_log_tree(trans
, log_root_tree
, &wc
);
2916 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
2917 key
.offset
= (u64
)-1;
2918 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
2921 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
2925 if (path
->slots
[0] == 0)
2929 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2931 btrfs_release_path(log_root_tree
, path
);
2932 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
2935 log
= btrfs_read_fs_root_no_radix(log_root_tree
,
2940 tmp_key
.objectid
= found_key
.offset
;
2941 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
2942 tmp_key
.offset
= (u64
)-1;
2944 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
2946 BUG_ON(!wc
.replay_dest
);
2948 btrfs_record_root_in_trans(wc
.replay_dest
);
2949 ret
= walk_log_tree(trans
, log
, &wc
);
2952 if (wc
.stage
== LOG_WALK_REPLAY_ALL
) {
2953 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
2957 ret
= btrfs_find_highest_inode(wc
.replay_dest
, &highest_inode
);
2959 wc
.replay_dest
->highest_inode
= highest_inode
;
2960 wc
.replay_dest
->last_inode_alloc
= highest_inode
;
2963 key
.offset
= found_key
.offset
- 1;
2964 free_extent_buffer(log
->node
);
2967 if (found_key
.offset
== 0)
2970 btrfs_release_path(log_root_tree
, path
);
2972 /* step one is to pin it all, step two is to replay just inodes */
2975 wc
.process_func
= replay_one_buffer
;
2976 wc
.stage
= LOG_WALK_REPLAY_INODES
;
2979 /* step three is to replay everything */
2980 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
2985 btrfs_free_path(path
);
2987 free_extent_buffer(log_root_tree
->node
);
2988 log_root_tree
->log_root
= NULL
;
2989 fs_info
->log_root_recovering
= 0;
2991 /* step 4: commit the transaction, which also unpins the blocks */
2992 btrfs_commit_transaction(trans
, fs_info
->tree_root
);
2994 kfree(log_root_tree
);