2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
22 #include "transaction.h"
25 #include "print-tree.h"
29 /* magic values for the inode_only field in btrfs_log_inode:
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
35 #define LOG_INODE_ALL 0
36 #define LOG_INODE_EXISTS 1
39 * directory trouble cases
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
48 * rename foo/some_dir foo2/some_dir
50 * fsync foo/some_dir/some_file
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
82 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
90 #define LOG_WALK_PIN_ONLY 0
91 #define LOG_WALK_REPLAY_INODES 1
92 #define LOG_WALK_REPLAY_ALL 2
94 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
95 struct btrfs_root
*root
, struct inode
*inode
,
97 static int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
98 struct btrfs_root
*root
,
99 struct btrfs_path
*path
, u64 objectid
);
100 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
101 struct btrfs_root
*root
,
102 struct btrfs_root
*log
,
103 struct btrfs_path
*path
,
104 u64 dirid
, int del_all
);
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction. Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree. Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
134 static int start_log_trans(struct btrfs_trans_handle
*trans
,
135 struct btrfs_root
*root
)
140 mutex_lock(&root
->log_mutex
);
141 if (root
->log_root
) {
142 if (!root
->log_start_pid
) {
143 root
->log_start_pid
= current
->pid
;
144 root
->log_multiple_pids
= false;
145 } else if (root
->log_start_pid
!= current
->pid
) {
146 root
->log_multiple_pids
= true;
150 atomic_inc(&root
->log_writers
);
151 mutex_unlock(&root
->log_mutex
);
154 root
->log_multiple_pids
= false;
155 root
->log_start_pid
= current
->pid
;
156 mutex_lock(&root
->fs_info
->tree_log_mutex
);
157 if (!root
->fs_info
->log_root_tree
) {
158 ret
= btrfs_init_log_root_tree(trans
, root
->fs_info
);
162 if (err
== 0 && !root
->log_root
) {
163 ret
= btrfs_add_log_tree(trans
, root
);
167 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
169 atomic_inc(&root
->log_writers
);
170 mutex_unlock(&root
->log_mutex
);
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
179 static int join_running_log_trans(struct btrfs_root
*root
)
187 mutex_lock(&root
->log_mutex
);
188 if (root
->log_root
) {
190 atomic_inc(&root
->log_writers
);
192 mutex_unlock(&root
->log_mutex
);
197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
201 int btrfs_pin_log_trans(struct btrfs_root
*root
)
205 mutex_lock(&root
->log_mutex
);
206 atomic_inc(&root
->log_writers
);
207 mutex_unlock(&root
->log_mutex
);
212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
215 int btrfs_end_log_trans(struct btrfs_root
*root
)
217 if (atomic_dec_and_test(&root
->log_writers
)) {
219 if (waitqueue_active(&root
->log_writer_wait
))
220 wake_up(&root
->log_writer_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 btrfs_pin_extent(log
->fs_info
->extent_root
,
280 eb
->start
, eb
->len
, 0);
282 if (btrfs_buffer_uptodate(eb
, gen
)) {
284 btrfs_write_tree_block(eb
);
286 btrfs_wait_tree_block_writeback(eb
);
292 * Item overwrite used by replay and tree logging. eb, slot and key all refer
293 * to the src data we are copying out.
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
299 * If the key is already in the destination tree the existing item is
300 * overwritten. If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
303 * If the key isn't in the destination yet, a new item is inserted.
305 static noinline
int overwrite_item(struct btrfs_trans_handle
*trans
,
306 struct btrfs_root
*root
,
307 struct btrfs_path
*path
,
308 struct extent_buffer
*eb
, int slot
,
309 struct btrfs_key
*key
)
313 u64 saved_i_size
= 0;
314 int save_old_i_size
= 0;
315 unsigned long src_ptr
;
316 unsigned long dst_ptr
;
317 int overwrite_root
= 0;
319 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
322 item_size
= btrfs_item_size_nr(eb
, slot
);
323 src_ptr
= btrfs_item_ptr_offset(eb
, slot
);
325 /* look for the key in the destination tree */
326 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
330 u32 dst_size
= btrfs_item_size_nr(path
->nodes
[0],
332 if (dst_size
!= item_size
)
335 if (item_size
== 0) {
336 btrfs_release_path(path
);
339 dst_copy
= kmalloc(item_size
, GFP_NOFS
);
340 src_copy
= kmalloc(item_size
, GFP_NOFS
);
341 if (!dst_copy
|| !src_copy
) {
342 btrfs_release_path(path
);
348 read_extent_buffer(eb
, src_copy
, src_ptr
, item_size
);
350 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
351 read_extent_buffer(path
->nodes
[0], dst_copy
, dst_ptr
,
353 ret
= memcmp(dst_copy
, src_copy
, item_size
);
358 * they have the same contents, just return, this saves
359 * us from cowing blocks in the destination tree and doing
360 * extra writes that may not have been done by a previous
364 btrfs_release_path(path
);
370 btrfs_release_path(path
);
371 /* try to insert the key into the destination tree */
372 ret
= btrfs_insert_empty_item(trans
, root
, path
,
375 /* make sure any existing item is the correct size */
376 if (ret
== -EEXIST
) {
378 found_size
= btrfs_item_size_nr(path
->nodes
[0],
380 if (found_size
> item_size
) {
381 btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
382 } else if (found_size
< item_size
) {
383 ret
= btrfs_extend_item(trans
, root
, path
,
384 item_size
- found_size
);
390 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0],
393 /* don't overwrite an existing inode if the generation number
394 * was logged as zero. This is done when the tree logging code
395 * is just logging an inode to make sure it exists after recovery.
397 * Also, don't overwrite i_size on directories during replay.
398 * log replay inserts and removes directory items based on the
399 * state of the tree found in the subvolume, and i_size is modified
402 if (key
->type
== BTRFS_INODE_ITEM_KEY
&& ret
== -EEXIST
) {
403 struct btrfs_inode_item
*src_item
;
404 struct btrfs_inode_item
*dst_item
;
406 src_item
= (struct btrfs_inode_item
*)src_ptr
;
407 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
409 if (btrfs_inode_generation(eb
, src_item
) == 0)
412 if (overwrite_root
&&
413 S_ISDIR(btrfs_inode_mode(eb
, src_item
)) &&
414 S_ISDIR(btrfs_inode_mode(path
->nodes
[0], dst_item
))) {
416 saved_i_size
= btrfs_inode_size(path
->nodes
[0],
421 copy_extent_buffer(path
->nodes
[0], eb
, dst_ptr
,
424 if (save_old_i_size
) {
425 struct btrfs_inode_item
*dst_item
;
426 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
427 btrfs_set_inode_size(path
->nodes
[0], dst_item
, saved_i_size
);
430 /* make sure the generation is filled in */
431 if (key
->type
== BTRFS_INODE_ITEM_KEY
) {
432 struct btrfs_inode_item
*dst_item
;
433 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
434 if (btrfs_inode_generation(path
->nodes
[0], dst_item
) == 0) {
435 btrfs_set_inode_generation(path
->nodes
[0], dst_item
,
440 btrfs_mark_buffer_dirty(path
->nodes
[0]);
441 btrfs_release_path(path
);
446 * simple helper to read an inode off the disk from a given root
447 * This can only be called for subvolume roots and not for the log
449 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
452 struct btrfs_key key
;
455 key
.objectid
= objectid
;
456 key
.type
= BTRFS_INODE_ITEM_KEY
;
458 inode
= btrfs_iget(root
->fs_info
->sb
, &key
, root
, NULL
);
461 } else if (is_bad_inode(inode
)) {
468 /* replays a single extent in 'eb' at 'slot' with 'key' into the
469 * subvolume 'root'. path is released on entry and should be released
472 * extents in the log tree have not been allocated out of the extent
473 * tree yet. So, this completes the allocation, taking a reference
474 * as required if the extent already exists or creating a new extent
475 * if it isn't in the extent allocation tree yet.
477 * The extent is inserted into the file, dropping any existing extents
478 * from the file that overlap the new one.
480 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
481 struct btrfs_root
*root
,
482 struct btrfs_path
*path
,
483 struct extent_buffer
*eb
, int slot
,
484 struct btrfs_key
*key
)
487 u64 mask
= root
->sectorsize
- 1;
490 u64 start
= key
->offset
;
492 struct btrfs_file_extent_item
*item
;
493 struct inode
*inode
= NULL
;
497 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
498 found_type
= btrfs_file_extent_type(eb
, item
);
500 if (found_type
== BTRFS_FILE_EXTENT_REG
||
501 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
502 extent_end
= start
+ btrfs_file_extent_num_bytes(eb
, item
);
503 else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
504 size
= btrfs_file_extent_inline_len(eb
, item
);
505 extent_end
= (start
+ size
+ mask
) & ~mask
;
511 inode
= read_one_inode(root
, key
->objectid
);
518 * first check to see if we already have this extent in the
519 * file. This must be done before the btrfs_drop_extents run
520 * so we don't try to drop this extent.
522 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
526 (found_type
== BTRFS_FILE_EXTENT_REG
||
527 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)) {
528 struct btrfs_file_extent_item cmp1
;
529 struct btrfs_file_extent_item cmp2
;
530 struct btrfs_file_extent_item
*existing
;
531 struct extent_buffer
*leaf
;
533 leaf
= path
->nodes
[0];
534 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
535 struct btrfs_file_extent_item
);
537 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
539 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
543 * we already have a pointer to this exact extent,
544 * we don't have to do anything
546 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
547 btrfs_release_path(path
);
551 btrfs_release_path(path
);
553 saved_nbytes
= inode_get_bytes(inode
);
554 /* drop any overlapping extents */
555 ret
= btrfs_drop_extents(trans
, inode
, start
, extent_end
,
559 if (found_type
== BTRFS_FILE_EXTENT_REG
||
560 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
562 unsigned long dest_offset
;
563 struct btrfs_key ins
;
565 ret
= btrfs_insert_empty_item(trans
, root
, path
, key
,
568 dest_offset
= btrfs_item_ptr_offset(path
->nodes
[0],
570 copy_extent_buffer(path
->nodes
[0], eb
, dest_offset
,
571 (unsigned long)item
, sizeof(*item
));
573 ins
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
574 ins
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
575 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
576 offset
= key
->offset
- btrfs_file_extent_offset(eb
, item
);
578 if (ins
.objectid
> 0) {
581 LIST_HEAD(ordered_sums
);
583 * is this extent already allocated in the extent
584 * allocation tree? If so, just add a reference
586 ret
= btrfs_lookup_extent(root
, ins
.objectid
,
589 ret
= btrfs_inc_extent_ref(trans
, root
,
590 ins
.objectid
, ins
.offset
,
591 0, root
->root_key
.objectid
,
592 key
->objectid
, offset
);
595 * insert the extent pointer in the extent
598 ret
= btrfs_alloc_logged_file_extent(trans
,
599 root
, root
->root_key
.objectid
,
600 key
->objectid
, offset
, &ins
);
603 btrfs_release_path(path
);
605 if (btrfs_file_extent_compression(eb
, item
)) {
606 csum_start
= ins
.objectid
;
607 csum_end
= csum_start
+ ins
.offset
;
609 csum_start
= ins
.objectid
+
610 btrfs_file_extent_offset(eb
, item
);
611 csum_end
= csum_start
+
612 btrfs_file_extent_num_bytes(eb
, item
);
615 ret
= btrfs_lookup_csums_range(root
->log_root
,
616 csum_start
, csum_end
- 1,
619 while (!list_empty(&ordered_sums
)) {
620 struct btrfs_ordered_sum
*sums
;
621 sums
= list_entry(ordered_sums
.next
,
622 struct btrfs_ordered_sum
,
624 ret
= btrfs_csum_file_blocks(trans
,
625 root
->fs_info
->csum_root
,
628 list_del(&sums
->list
);
632 btrfs_release_path(path
);
634 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
635 /* inline extents are easy, we just overwrite them */
636 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
640 inode_set_bytes(inode
, saved_nbytes
);
641 btrfs_update_inode(trans
, root
, inode
);
649 * when cleaning up conflicts between the directory names in the
650 * subvolume, directory names in the log and directory names in the
651 * inode back references, we may have to unlink inodes from directories.
653 * This is a helper function to do the unlink of a specific directory
656 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
657 struct btrfs_root
*root
,
658 struct btrfs_path
*path
,
660 struct btrfs_dir_item
*di
)
665 struct extent_buffer
*leaf
;
666 struct btrfs_key location
;
669 leaf
= path
->nodes
[0];
671 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
672 name_len
= btrfs_dir_name_len(leaf
, di
);
673 name
= kmalloc(name_len
, GFP_NOFS
);
677 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
678 btrfs_release_path(path
);
680 inode
= read_one_inode(root
, location
.objectid
);
683 ret
= link_to_fixup_dir(trans
, root
, path
, location
.objectid
);
686 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
695 * helper function to see if a given name and sequence number found
696 * in an inode back reference are already in a directory and correctly
697 * point to this inode
699 static noinline
int inode_in_dir(struct btrfs_root
*root
,
700 struct btrfs_path
*path
,
701 u64 dirid
, u64 objectid
, u64 index
,
702 const char *name
, int name_len
)
704 struct btrfs_dir_item
*di
;
705 struct btrfs_key location
;
708 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
709 index
, name
, name_len
, 0);
710 if (di
&& !IS_ERR(di
)) {
711 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
712 if (location
.objectid
!= objectid
)
716 btrfs_release_path(path
);
718 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
719 if (di
&& !IS_ERR(di
)) {
720 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
721 if (location
.objectid
!= objectid
)
727 btrfs_release_path(path
);
732 * helper function to check a log tree for a named back reference in
733 * an inode. This is used to decide if a back reference that is
734 * found in the subvolume conflicts with what we find in the log.
736 * inode backreferences may have multiple refs in a single item,
737 * during replay we process one reference at a time, and we don't
738 * want to delete valid links to a file from the subvolume if that
739 * link is also in the log.
741 static noinline
int backref_in_log(struct btrfs_root
*log
,
742 struct btrfs_key
*key
,
743 char *name
, int namelen
)
745 struct btrfs_path
*path
;
746 struct btrfs_inode_ref
*ref
;
748 unsigned long ptr_end
;
749 unsigned long name_ptr
;
755 path
= btrfs_alloc_path();
759 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
763 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
764 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
765 ptr_end
= ptr
+ item_size
;
766 while (ptr
< ptr_end
) {
767 ref
= (struct btrfs_inode_ref
*)ptr
;
768 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
769 if (found_name_len
== namelen
) {
770 name_ptr
= (unsigned long)(ref
+ 1);
771 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
778 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
781 btrfs_free_path(path
);
787 * replay one inode back reference item found in the log tree.
788 * eb, slot and key refer to the buffer and key found in the log tree.
789 * root is the destination we are replaying into, and path is for temp
790 * use by this function. (it should be released on return).
792 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
793 struct btrfs_root
*root
,
794 struct btrfs_root
*log
,
795 struct btrfs_path
*path
,
796 struct extent_buffer
*eb
, int slot
,
797 struct btrfs_key
*key
)
801 struct btrfs_inode_ref
*ref
;
805 unsigned long ref_ptr
;
806 unsigned long ref_end
;
810 * it is possible that we didn't log all the parent directories
811 * for a given inode. If we don't find the dir, just don't
812 * copy the back ref in. The link count fixup code will take
815 dir
= read_one_inode(root
, key
->offset
);
819 inode
= read_one_inode(root
, key
->objectid
);
822 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
823 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
826 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
828 namelen
= btrfs_inode_ref_name_len(eb
, ref
);
829 name
= kmalloc(namelen
, GFP_NOFS
);
832 read_extent_buffer(eb
, name
, (unsigned long)(ref
+ 1), namelen
);
834 /* if we already have a perfect match, we're done */
835 if (inode_in_dir(root
, path
, btrfs_ino(dir
), btrfs_ino(inode
),
836 btrfs_inode_ref_index(eb
, ref
),
842 * look for a conflicting back reference in the metadata.
843 * if we find one we have to unlink that name of the file
844 * before we add our new link. Later on, we overwrite any
845 * existing back reference, and we don't want to create
846 * dangling pointers in the directory.
852 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
856 struct btrfs_inode_ref
*victim_ref
;
858 unsigned long ptr_end
;
859 struct extent_buffer
*leaf
= path
->nodes
[0];
861 /* are we trying to overwrite a back ref for the root directory
862 * if so, just jump out, we're done
864 if (key
->objectid
== key
->offset
)
867 /* check all the names in this back reference to see
868 * if they are in the log. if so, we allow them to stay
869 * otherwise they must be unlinked as a conflict
871 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
872 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
873 while (ptr
< ptr_end
) {
874 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
875 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
877 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
878 BUG_ON(!victim_name
);
880 read_extent_buffer(leaf
, victim_name
,
881 (unsigned long)(victim_ref
+ 1),
884 if (!backref_in_log(log
, key
, victim_name
,
886 btrfs_inc_nlink(inode
);
887 btrfs_release_path(path
);
889 ret
= btrfs_unlink_inode(trans
, root
, dir
,
894 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
899 * NOTE: we have searched root tree and checked the
900 * coresponding ref, it does not need to check again.
904 btrfs_release_path(path
);
907 /* insert our name */
908 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
, 0,
909 btrfs_inode_ref_index(eb
, ref
));
912 btrfs_update_inode(trans
, root
, inode
);
915 ref_ptr
= (unsigned long)(ref
+ 1) + namelen
;
917 if (ref_ptr
< ref_end
)
920 /* finally write the back reference in the inode */
921 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
925 btrfs_release_path(path
);
931 static int insert_orphan_item(struct btrfs_trans_handle
*trans
,
932 struct btrfs_root
*root
, u64 offset
)
935 ret
= btrfs_find_orphan_item(root
, offset
);
937 ret
= btrfs_insert_orphan_item(trans
, root
, offset
);
943 * There are a few corners where the link count of the file can't
944 * be properly maintained during replay. So, instead of adding
945 * lots of complexity to the log code, we just scan the backrefs
946 * for any file that has been through replay.
948 * The scan will update the link count on the inode to reflect the
949 * number of back refs found. If it goes down to zero, the iput
950 * will free the inode.
952 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
953 struct btrfs_root
*root
,
956 struct btrfs_path
*path
;
958 struct btrfs_key key
;
961 unsigned long ptr_end
;
963 u64 ino
= btrfs_ino(inode
);
966 key
.type
= BTRFS_INODE_REF_KEY
;
967 key
.offset
= (u64
)-1;
969 path
= btrfs_alloc_path();
974 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
978 if (path
->slots
[0] == 0)
982 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
984 if (key
.objectid
!= ino
||
985 key
.type
!= BTRFS_INODE_REF_KEY
)
987 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
988 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
990 while (ptr
< ptr_end
) {
991 struct btrfs_inode_ref
*ref
;
993 ref
= (struct btrfs_inode_ref
*)ptr
;
994 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
996 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1000 if (key
.offset
== 0)
1003 btrfs_release_path(path
);
1005 btrfs_release_path(path
);
1006 if (nlink
!= inode
->i_nlink
) {
1007 inode
->i_nlink
= nlink
;
1008 btrfs_update_inode(trans
, root
, inode
);
1010 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1012 if (inode
->i_nlink
== 0) {
1013 if (S_ISDIR(inode
->i_mode
)) {
1014 ret
= replay_dir_deletes(trans
, root
, NULL
, path
,
1018 ret
= insert_orphan_item(trans
, root
, ino
);
1021 btrfs_free_path(path
);
1026 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1027 struct btrfs_root
*root
,
1028 struct btrfs_path
*path
)
1031 struct btrfs_key key
;
1032 struct inode
*inode
;
1034 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1035 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1036 key
.offset
= (u64
)-1;
1038 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1043 if (path
->slots
[0] == 0)
1048 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1049 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1050 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1053 ret
= btrfs_del_item(trans
, root
, path
);
1056 btrfs_release_path(path
);
1057 inode
= read_one_inode(root
, key
.offset
);
1060 ret
= fixup_inode_link_count(trans
, root
, inode
);
1066 * fixup on a directory may create new entries,
1067 * make sure we always look for the highset possible
1070 key
.offset
= (u64
)-1;
1072 btrfs_release_path(path
);
1078 * record a given inode in the fixup dir so we can check its link
1079 * count when replay is done. The link count is incremented here
1080 * so the inode won't go away until we check it
1082 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1083 struct btrfs_root
*root
,
1084 struct btrfs_path
*path
,
1087 struct btrfs_key key
;
1089 struct inode
*inode
;
1091 inode
= read_one_inode(root
, objectid
);
1094 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1095 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1096 key
.offset
= objectid
;
1098 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1100 btrfs_release_path(path
);
1102 btrfs_inc_nlink(inode
);
1103 btrfs_update_inode(trans
, root
, inode
);
1104 } else if (ret
== -EEXIST
) {
1115 * when replaying the log for a directory, we only insert names
1116 * for inodes that actually exist. This means an fsync on a directory
1117 * does not implicitly fsync all the new files in it
1119 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1120 struct btrfs_root
*root
,
1121 struct btrfs_path
*path
,
1122 u64 dirid
, u64 index
,
1123 char *name
, int name_len
, u8 type
,
1124 struct btrfs_key
*location
)
1126 struct inode
*inode
;
1130 inode
= read_one_inode(root
, location
->objectid
);
1134 dir
= read_one_inode(root
, dirid
);
1139 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1141 /* FIXME, put inode into FIXUP list */
1149 * take a single entry in a log directory item and replay it into
1152 * if a conflicting item exists in the subdirectory already,
1153 * the inode it points to is unlinked and put into the link count
1156 * If a name from the log points to a file or directory that does
1157 * not exist in the FS, it is skipped. fsyncs on directories
1158 * do not force down inodes inside that directory, just changes to the
1159 * names or unlinks in a directory.
1161 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1162 struct btrfs_root
*root
,
1163 struct btrfs_path
*path
,
1164 struct extent_buffer
*eb
,
1165 struct btrfs_dir_item
*di
,
1166 struct btrfs_key
*key
)
1170 struct btrfs_dir_item
*dst_di
;
1171 struct btrfs_key found_key
;
1172 struct btrfs_key log_key
;
1178 dir
= read_one_inode(root
, key
->objectid
);
1181 name_len
= btrfs_dir_name_len(eb
, di
);
1182 name
= kmalloc(name_len
, GFP_NOFS
);
1186 log_type
= btrfs_dir_type(eb
, di
);
1187 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1190 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1191 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1196 btrfs_release_path(path
);
1198 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1199 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1201 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1202 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1209 if (IS_ERR_OR_NULL(dst_di
)) {
1210 /* we need a sequence number to insert, so we only
1211 * do inserts for the BTRFS_DIR_INDEX_KEY types
1213 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1218 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1219 /* the existing item matches the logged item */
1220 if (found_key
.objectid
== log_key
.objectid
&&
1221 found_key
.type
== log_key
.type
&&
1222 found_key
.offset
== log_key
.offset
&&
1223 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1228 * don't drop the conflicting directory entry if the inode
1229 * for the new entry doesn't exist
1234 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1237 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1240 btrfs_release_path(path
);
1246 btrfs_release_path(path
);
1247 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1248 name
, name_len
, log_type
, &log_key
);
1250 BUG_ON(ret
&& ret
!= -ENOENT
);
1255 * find all the names in a directory item and reconcile them into
1256 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1257 * one name in a directory item, but the same code gets used for
1258 * both directory index types
1260 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1261 struct btrfs_root
*root
,
1262 struct btrfs_path
*path
,
1263 struct extent_buffer
*eb
, int slot
,
1264 struct btrfs_key
*key
)
1267 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1268 struct btrfs_dir_item
*di
;
1271 unsigned long ptr_end
;
1273 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1274 ptr_end
= ptr
+ item_size
;
1275 while (ptr
< ptr_end
) {
1276 di
= (struct btrfs_dir_item
*)ptr
;
1277 if (verify_dir_item(root
, eb
, di
))
1279 name_len
= btrfs_dir_name_len(eb
, di
);
1280 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1282 ptr
= (unsigned long)(di
+ 1);
1289 * directory replay has two parts. There are the standard directory
1290 * items in the log copied from the subvolume, and range items
1291 * created in the log while the subvolume was logged.
1293 * The range items tell us which parts of the key space the log
1294 * is authoritative for. During replay, if a key in the subvolume
1295 * directory is in a logged range item, but not actually in the log
1296 * that means it was deleted from the directory before the fsync
1297 * and should be removed.
1299 static noinline
int find_dir_range(struct btrfs_root
*root
,
1300 struct btrfs_path
*path
,
1301 u64 dirid
, int key_type
,
1302 u64
*start_ret
, u64
*end_ret
)
1304 struct btrfs_key key
;
1306 struct btrfs_dir_log_item
*item
;
1310 if (*start_ret
== (u64
)-1)
1313 key
.objectid
= dirid
;
1314 key
.type
= key_type
;
1315 key
.offset
= *start_ret
;
1317 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1321 if (path
->slots
[0] == 0)
1326 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1328 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1332 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1333 struct btrfs_dir_log_item
);
1334 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1336 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1338 *start_ret
= key
.offset
;
1339 *end_ret
= found_end
;
1344 /* check the next slot in the tree to see if it is a valid item */
1345 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1346 if (path
->slots
[0] >= nritems
) {
1347 ret
= btrfs_next_leaf(root
, path
);
1354 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1356 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1360 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1361 struct btrfs_dir_log_item
);
1362 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1363 *start_ret
= key
.offset
;
1364 *end_ret
= found_end
;
1367 btrfs_release_path(path
);
1372 * this looks for a given directory item in the log. If the directory
1373 * item is not in the log, the item is removed and the inode it points
1376 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1377 struct btrfs_root
*root
,
1378 struct btrfs_root
*log
,
1379 struct btrfs_path
*path
,
1380 struct btrfs_path
*log_path
,
1382 struct btrfs_key
*dir_key
)
1385 struct extent_buffer
*eb
;
1388 struct btrfs_dir_item
*di
;
1389 struct btrfs_dir_item
*log_di
;
1392 unsigned long ptr_end
;
1394 struct inode
*inode
;
1395 struct btrfs_key location
;
1398 eb
= path
->nodes
[0];
1399 slot
= path
->slots
[0];
1400 item_size
= btrfs_item_size_nr(eb
, slot
);
1401 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1402 ptr_end
= ptr
+ item_size
;
1403 while (ptr
< ptr_end
) {
1404 di
= (struct btrfs_dir_item
*)ptr
;
1405 if (verify_dir_item(root
, eb
, di
)) {
1410 name_len
= btrfs_dir_name_len(eb
, di
);
1411 name
= kmalloc(name_len
, GFP_NOFS
);
1416 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1419 if (log
&& dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1420 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1423 } else if (log
&& dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1424 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1430 if (IS_ERR_OR_NULL(log_di
)) {
1431 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1432 btrfs_release_path(path
);
1433 btrfs_release_path(log_path
);
1434 inode
= read_one_inode(root
, location
.objectid
);
1437 ret
= link_to_fixup_dir(trans
, root
,
1438 path
, location
.objectid
);
1440 btrfs_inc_nlink(inode
);
1441 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1447 /* there might still be more names under this key
1448 * check and repeat if required
1450 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1457 btrfs_release_path(log_path
);
1460 ptr
= (unsigned long)(di
+ 1);
1465 btrfs_release_path(path
);
1466 btrfs_release_path(log_path
);
1471 * deletion replay happens before we copy any new directory items
1472 * out of the log or out of backreferences from inodes. It
1473 * scans the log to find ranges of keys that log is authoritative for,
1474 * and then scans the directory to find items in those ranges that are
1475 * not present in the log.
1477 * Anything we don't find in the log is unlinked and removed from the
1480 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1481 struct btrfs_root
*root
,
1482 struct btrfs_root
*log
,
1483 struct btrfs_path
*path
,
1484 u64 dirid
, int del_all
)
1488 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1490 struct btrfs_key dir_key
;
1491 struct btrfs_key found_key
;
1492 struct btrfs_path
*log_path
;
1495 dir_key
.objectid
= dirid
;
1496 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1497 log_path
= btrfs_alloc_path();
1501 dir
= read_one_inode(root
, dirid
);
1502 /* it isn't an error if the inode isn't there, that can happen
1503 * because we replay the deletes before we copy in the inode item
1507 btrfs_free_path(log_path
);
1515 range_end
= (u64
)-1;
1517 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1518 &range_start
, &range_end
);
1523 dir_key
.offset
= range_start
;
1526 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1531 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1532 if (path
->slots
[0] >= nritems
) {
1533 ret
= btrfs_next_leaf(root
, path
);
1537 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1539 if (found_key
.objectid
!= dirid
||
1540 found_key
.type
!= dir_key
.type
)
1543 if (found_key
.offset
> range_end
)
1546 ret
= check_item_in_log(trans
, root
, log
, path
,
1550 if (found_key
.offset
== (u64
)-1)
1552 dir_key
.offset
= found_key
.offset
+ 1;
1554 btrfs_release_path(path
);
1555 if (range_end
== (u64
)-1)
1557 range_start
= range_end
+ 1;
1562 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1563 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1564 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1565 btrfs_release_path(path
);
1569 btrfs_release_path(path
);
1570 btrfs_free_path(log_path
);
1576 * the process_func used to replay items from the log tree. This
1577 * gets called in two different stages. The first stage just looks
1578 * for inodes and makes sure they are all copied into the subvolume.
1580 * The second stage copies all the other item types from the log into
1581 * the subvolume. The two stage approach is slower, but gets rid of
1582 * lots of complexity around inodes referencing other inodes that exist
1583 * only in the log (references come from either directory items or inode
1586 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
1587 struct walk_control
*wc
, u64 gen
)
1590 struct btrfs_path
*path
;
1591 struct btrfs_root
*root
= wc
->replay_dest
;
1592 struct btrfs_key key
;
1597 btrfs_read_buffer(eb
, gen
);
1599 level
= btrfs_header_level(eb
);
1604 path
= btrfs_alloc_path();
1607 nritems
= btrfs_header_nritems(eb
);
1608 for (i
= 0; i
< nritems
; i
++) {
1609 btrfs_item_key_to_cpu(eb
, &key
, i
);
1611 /* inode keys are done during the first stage */
1612 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
1613 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
1614 struct btrfs_inode_item
*inode_item
;
1617 inode_item
= btrfs_item_ptr(eb
, i
,
1618 struct btrfs_inode_item
);
1619 mode
= btrfs_inode_mode(eb
, inode_item
);
1620 if (S_ISDIR(mode
)) {
1621 ret
= replay_dir_deletes(wc
->trans
,
1622 root
, log
, path
, key
.objectid
, 0);
1625 ret
= overwrite_item(wc
->trans
, root
, path
,
1629 /* for regular files, make sure corresponding
1630 * orhpan item exist. extents past the new EOF
1631 * will be truncated later by orphan cleanup.
1633 if (S_ISREG(mode
)) {
1634 ret
= insert_orphan_item(wc
->trans
, root
,
1639 ret
= link_to_fixup_dir(wc
->trans
, root
,
1640 path
, key
.objectid
);
1643 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
1646 /* these keys are simply copied */
1647 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
1648 ret
= overwrite_item(wc
->trans
, root
, path
,
1651 } else if (key
.type
== BTRFS_INODE_REF_KEY
) {
1652 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
1654 BUG_ON(ret
&& ret
!= -ENOENT
);
1655 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
1656 ret
= replay_one_extent(wc
->trans
, root
, path
,
1659 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
||
1660 key
.type
== BTRFS_DIR_INDEX_KEY
) {
1661 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
1666 btrfs_free_path(path
);
1670 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
1671 struct btrfs_root
*root
,
1672 struct btrfs_path
*path
, int *level
,
1673 struct walk_control
*wc
)
1678 struct extent_buffer
*next
;
1679 struct extent_buffer
*cur
;
1680 struct extent_buffer
*parent
;
1684 WARN_ON(*level
< 0);
1685 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1687 while (*level
> 0) {
1688 WARN_ON(*level
< 0);
1689 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1690 cur
= path
->nodes
[*level
];
1692 if (btrfs_header_level(cur
) != *level
)
1695 if (path
->slots
[*level
] >=
1696 btrfs_header_nritems(cur
))
1699 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
1700 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
1701 blocksize
= btrfs_level_size(root
, *level
- 1);
1703 parent
= path
->nodes
[*level
];
1704 root_owner
= btrfs_header_owner(parent
);
1706 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1711 wc
->process_func(root
, next
, wc
, ptr_gen
);
1713 path
->slots
[*level
]++;
1715 btrfs_read_buffer(next
, ptr_gen
);
1717 btrfs_tree_lock(next
);
1718 clean_tree_block(trans
, root
, next
);
1719 btrfs_set_lock_blocking(next
);
1720 btrfs_wait_tree_block_writeback(next
);
1721 btrfs_tree_unlock(next
);
1723 WARN_ON(root_owner
!=
1724 BTRFS_TREE_LOG_OBJECTID
);
1725 ret
= btrfs_free_reserved_extent(root
,
1729 free_extent_buffer(next
);
1732 btrfs_read_buffer(next
, ptr_gen
);
1734 WARN_ON(*level
<= 0);
1735 if (path
->nodes
[*level
-1])
1736 free_extent_buffer(path
->nodes
[*level
-1]);
1737 path
->nodes
[*level
-1] = next
;
1738 *level
= btrfs_header_level(next
);
1739 path
->slots
[*level
] = 0;
1742 WARN_ON(*level
< 0);
1743 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1745 path
->slots
[*level
] = btrfs_header_nritems(path
->nodes
[*level
]);
1751 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
1752 struct btrfs_root
*root
,
1753 struct btrfs_path
*path
, int *level
,
1754 struct walk_control
*wc
)
1761 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
1762 slot
= path
->slots
[i
];
1763 if (slot
+ 1 < btrfs_header_nritems(path
->nodes
[i
])) {
1766 WARN_ON(*level
== 0);
1769 struct extent_buffer
*parent
;
1770 if (path
->nodes
[*level
] == root
->node
)
1771 parent
= path
->nodes
[*level
];
1773 parent
= path
->nodes
[*level
+ 1];
1775 root_owner
= btrfs_header_owner(parent
);
1776 wc
->process_func(root
, path
->nodes
[*level
], wc
,
1777 btrfs_header_generation(path
->nodes
[*level
]));
1779 struct extent_buffer
*next
;
1781 next
= path
->nodes
[*level
];
1783 btrfs_tree_lock(next
);
1784 clean_tree_block(trans
, root
, next
);
1785 btrfs_set_lock_blocking(next
);
1786 btrfs_wait_tree_block_writeback(next
);
1787 btrfs_tree_unlock(next
);
1789 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1790 ret
= btrfs_free_reserved_extent(root
,
1791 path
->nodes
[*level
]->start
,
1792 path
->nodes
[*level
]->len
);
1795 free_extent_buffer(path
->nodes
[*level
]);
1796 path
->nodes
[*level
] = NULL
;
1804 * drop the reference count on the tree rooted at 'snap'. This traverses
1805 * the tree freeing any blocks that have a ref count of zero after being
1808 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
1809 struct btrfs_root
*log
, struct walk_control
*wc
)
1814 struct btrfs_path
*path
;
1818 path
= btrfs_alloc_path();
1822 level
= btrfs_header_level(log
->node
);
1824 path
->nodes
[level
] = log
->node
;
1825 extent_buffer_get(log
->node
);
1826 path
->slots
[level
] = 0;
1829 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
1835 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
1842 /* was the root node processed? if not, catch it here */
1843 if (path
->nodes
[orig_level
]) {
1844 wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
1845 btrfs_header_generation(path
->nodes
[orig_level
]));
1847 struct extent_buffer
*next
;
1849 next
= path
->nodes
[orig_level
];
1851 btrfs_tree_lock(next
);
1852 clean_tree_block(trans
, log
, next
);
1853 btrfs_set_lock_blocking(next
);
1854 btrfs_wait_tree_block_writeback(next
);
1855 btrfs_tree_unlock(next
);
1857 WARN_ON(log
->root_key
.objectid
!=
1858 BTRFS_TREE_LOG_OBJECTID
);
1859 ret
= btrfs_free_reserved_extent(log
, next
->start
,
1865 for (i
= 0; i
<= orig_level
; i
++) {
1866 if (path
->nodes
[i
]) {
1867 free_extent_buffer(path
->nodes
[i
]);
1868 path
->nodes
[i
] = NULL
;
1871 btrfs_free_path(path
);
1876 * helper function to update the item for a given subvolumes log root
1877 * in the tree of log roots
1879 static int update_log_root(struct btrfs_trans_handle
*trans
,
1880 struct btrfs_root
*log
)
1884 if (log
->log_transid
== 1) {
1885 /* insert root item on the first sync */
1886 ret
= btrfs_insert_root(trans
, log
->fs_info
->log_root_tree
,
1887 &log
->root_key
, &log
->root_item
);
1889 ret
= btrfs_update_root(trans
, log
->fs_info
->log_root_tree
,
1890 &log
->root_key
, &log
->root_item
);
1895 static int wait_log_commit(struct btrfs_trans_handle
*trans
,
1896 struct btrfs_root
*root
, unsigned long transid
)
1899 int index
= transid
% 2;
1902 * we only allow two pending log transactions at a time,
1903 * so we know that if ours is more than 2 older than the
1904 * current transaction, we're done
1907 prepare_to_wait(&root
->log_commit_wait
[index
],
1908 &wait
, TASK_UNINTERRUPTIBLE
);
1909 mutex_unlock(&root
->log_mutex
);
1911 if (root
->fs_info
->last_trans_log_full_commit
!=
1912 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
1913 atomic_read(&root
->log_commit
[index
]))
1916 finish_wait(&root
->log_commit_wait
[index
], &wait
);
1917 mutex_lock(&root
->log_mutex
);
1918 } while (root
->log_transid
< transid
+ 2 &&
1919 atomic_read(&root
->log_commit
[index
]));
1923 static int wait_for_writer(struct btrfs_trans_handle
*trans
,
1924 struct btrfs_root
*root
)
1927 while (atomic_read(&root
->log_writers
)) {
1928 prepare_to_wait(&root
->log_writer_wait
,
1929 &wait
, TASK_UNINTERRUPTIBLE
);
1930 mutex_unlock(&root
->log_mutex
);
1931 if (root
->fs_info
->last_trans_log_full_commit
!=
1932 trans
->transid
&& atomic_read(&root
->log_writers
))
1934 mutex_lock(&root
->log_mutex
);
1935 finish_wait(&root
->log_writer_wait
, &wait
);
1941 * btrfs_sync_log does sends a given tree log down to the disk and
1942 * updates the super blocks to record it. When this call is done,
1943 * you know that any inodes previously logged are safely on disk only
1946 * Any other return value means you need to call btrfs_commit_transaction.
1947 * Some of the edge cases for fsyncing directories that have had unlinks
1948 * or renames done in the past mean that sometimes the only safe
1949 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1950 * that has happened.
1952 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
1953 struct btrfs_root
*root
)
1959 struct btrfs_root
*log
= root
->log_root
;
1960 struct btrfs_root
*log_root_tree
= root
->fs_info
->log_root_tree
;
1961 unsigned long log_transid
= 0;
1963 mutex_lock(&root
->log_mutex
);
1964 index1
= root
->log_transid
% 2;
1965 if (atomic_read(&root
->log_commit
[index1
])) {
1966 wait_log_commit(trans
, root
, root
->log_transid
);
1967 mutex_unlock(&root
->log_mutex
);
1970 atomic_set(&root
->log_commit
[index1
], 1);
1972 /* wait for previous tree log sync to complete */
1973 if (atomic_read(&root
->log_commit
[(index1
+ 1) % 2]))
1974 wait_log_commit(trans
, root
, root
->log_transid
- 1);
1977 unsigned long batch
= root
->log_batch
;
1978 if (root
->log_multiple_pids
) {
1979 mutex_unlock(&root
->log_mutex
);
1980 schedule_timeout_uninterruptible(1);
1981 mutex_lock(&root
->log_mutex
);
1983 wait_for_writer(trans
, root
);
1984 if (batch
== root
->log_batch
)
1988 /* bail out if we need to do a full commit */
1989 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
1991 mutex_unlock(&root
->log_mutex
);
1995 log_transid
= root
->log_transid
;
1996 if (log_transid
% 2 == 0)
1997 mark
= EXTENT_DIRTY
;
2001 /* we start IO on all the marked extents here, but we don't actually
2002 * wait for them until later.
2004 ret
= btrfs_write_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2007 btrfs_set_root_node(&log
->root_item
, log
->node
);
2009 root
->log_batch
= 0;
2010 root
->log_transid
++;
2011 log
->log_transid
= root
->log_transid
;
2012 root
->log_start_pid
= 0;
2015 * IO has been started, blocks of the log tree have WRITTEN flag set
2016 * in their headers. new modifications of the log will be written to
2017 * new positions. so it's safe to allow log writers to go in.
2019 mutex_unlock(&root
->log_mutex
);
2021 mutex_lock(&log_root_tree
->log_mutex
);
2022 log_root_tree
->log_batch
++;
2023 atomic_inc(&log_root_tree
->log_writers
);
2024 mutex_unlock(&log_root_tree
->log_mutex
);
2026 ret
= update_log_root(trans
, log
);
2028 mutex_lock(&log_root_tree
->log_mutex
);
2029 if (atomic_dec_and_test(&log_root_tree
->log_writers
)) {
2031 if (waitqueue_active(&log_root_tree
->log_writer_wait
))
2032 wake_up(&log_root_tree
->log_writer_wait
);
2036 BUG_ON(ret
!= -ENOSPC
);
2037 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2038 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2039 mutex_unlock(&log_root_tree
->log_mutex
);
2044 index2
= log_root_tree
->log_transid
% 2;
2045 if (atomic_read(&log_root_tree
->log_commit
[index2
])) {
2046 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2047 wait_log_commit(trans
, log_root_tree
,
2048 log_root_tree
->log_transid
);
2049 mutex_unlock(&log_root_tree
->log_mutex
);
2053 atomic_set(&log_root_tree
->log_commit
[index2
], 1);
2055 if (atomic_read(&log_root_tree
->log_commit
[(index2
+ 1) % 2])) {
2056 wait_log_commit(trans
, log_root_tree
,
2057 log_root_tree
->log_transid
- 1);
2060 wait_for_writer(trans
, log_root_tree
);
2063 * now that we've moved on to the tree of log tree roots,
2064 * check the full commit flag again
2066 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2067 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2068 mutex_unlock(&log_root_tree
->log_mutex
);
2070 goto out_wake_log_root
;
2073 ret
= btrfs_write_and_wait_marked_extents(log_root_tree
,
2074 &log_root_tree
->dirty_log_pages
,
2075 EXTENT_DIRTY
| EXTENT_NEW
);
2077 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2079 btrfs_set_super_log_root(&root
->fs_info
->super_for_commit
,
2080 log_root_tree
->node
->start
);
2081 btrfs_set_super_log_root_level(&root
->fs_info
->super_for_commit
,
2082 btrfs_header_level(log_root_tree
->node
));
2084 log_root_tree
->log_batch
= 0;
2085 log_root_tree
->log_transid
++;
2088 mutex_unlock(&log_root_tree
->log_mutex
);
2091 * nobody else is going to jump in and write the the ctree
2092 * super here because the log_commit atomic below is protecting
2093 * us. We must be called with a transaction handle pinning
2094 * the running transaction open, so a full commit can't hop
2095 * in and cause problems either.
2097 btrfs_scrub_pause_super(root
);
2098 write_ctree_super(trans
, root
->fs_info
->tree_root
, 1);
2099 btrfs_scrub_continue_super(root
);
2102 mutex_lock(&root
->log_mutex
);
2103 if (root
->last_log_commit
< log_transid
)
2104 root
->last_log_commit
= log_transid
;
2105 mutex_unlock(&root
->log_mutex
);
2108 atomic_set(&log_root_tree
->log_commit
[index2
], 0);
2110 if (waitqueue_active(&log_root_tree
->log_commit_wait
[index2
]))
2111 wake_up(&log_root_tree
->log_commit_wait
[index2
]);
2113 atomic_set(&root
->log_commit
[index1
], 0);
2115 if (waitqueue_active(&root
->log_commit_wait
[index1
]))
2116 wake_up(&root
->log_commit_wait
[index1
]);
2120 static void free_log_tree(struct btrfs_trans_handle
*trans
,
2121 struct btrfs_root
*log
)
2126 struct walk_control wc
= {
2128 .process_func
= process_one_buffer
2131 ret
= walk_log_tree(trans
, log
, &wc
);
2135 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
2136 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
);
2140 clear_extent_bits(&log
->dirty_log_pages
, start
, end
,
2141 EXTENT_DIRTY
| EXTENT_NEW
, GFP_NOFS
);
2144 free_extent_buffer(log
->node
);
2149 * free all the extents used by the tree log. This should be called
2150 * at commit time of the full transaction
2152 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2154 if (root
->log_root
) {
2155 free_log_tree(trans
, root
->log_root
);
2156 root
->log_root
= NULL
;
2161 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
2162 struct btrfs_fs_info
*fs_info
)
2164 if (fs_info
->log_root_tree
) {
2165 free_log_tree(trans
, fs_info
->log_root_tree
);
2166 fs_info
->log_root_tree
= NULL
;
2172 * If both a file and directory are logged, and unlinks or renames are
2173 * mixed in, we have a few interesting corners:
2175 * create file X in dir Y
2176 * link file X to X.link in dir Y
2178 * unlink file X but leave X.link
2181 * After a crash we would expect only X.link to exist. But file X
2182 * didn't get fsync'd again so the log has back refs for X and X.link.
2184 * We solve this by removing directory entries and inode backrefs from the
2185 * log when a file that was logged in the current transaction is
2186 * unlinked. Any later fsync will include the updated log entries, and
2187 * we'll be able to reconstruct the proper directory items from backrefs.
2189 * This optimizations allows us to avoid relogging the entire inode
2190 * or the entire directory.
2192 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
2193 struct btrfs_root
*root
,
2194 const char *name
, int name_len
,
2195 struct inode
*dir
, u64 index
)
2197 struct btrfs_root
*log
;
2198 struct btrfs_dir_item
*di
;
2199 struct btrfs_path
*path
;
2203 u64 dir_ino
= btrfs_ino(dir
);
2205 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
2208 ret
= join_running_log_trans(root
);
2212 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2214 log
= root
->log_root
;
2215 path
= btrfs_alloc_path();
2221 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir_ino
,
2222 name
, name_len
, -1);
2228 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2229 bytes_del
+= name_len
;
2232 btrfs_release_path(path
);
2233 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir_ino
,
2234 index
, name
, name_len
, -1);
2240 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2241 bytes_del
+= name_len
;
2245 /* update the directory size in the log to reflect the names
2249 struct btrfs_key key
;
2251 key
.objectid
= dir_ino
;
2253 key
.type
= BTRFS_INODE_ITEM_KEY
;
2254 btrfs_release_path(path
);
2256 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2262 struct btrfs_inode_item
*item
;
2265 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2266 struct btrfs_inode_item
);
2267 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2268 if (i_size
> bytes_del
)
2269 i_size
-= bytes_del
;
2272 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2273 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2276 btrfs_release_path(path
);
2279 btrfs_free_path(path
);
2281 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2282 if (ret
== -ENOSPC
) {
2283 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2286 btrfs_end_log_trans(root
);
2291 /* see comments for btrfs_del_dir_entries_in_log */
2292 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2293 struct btrfs_root
*root
,
2294 const char *name
, int name_len
,
2295 struct inode
*inode
, u64 dirid
)
2297 struct btrfs_root
*log
;
2301 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
2304 ret
= join_running_log_trans(root
);
2307 log
= root
->log_root
;
2308 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2310 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, btrfs_ino(inode
),
2312 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2313 if (ret
== -ENOSPC
) {
2314 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2317 btrfs_end_log_trans(root
);
2323 * creates a range item in the log for 'dirid'. first_offset and
2324 * last_offset tell us which parts of the key space the log should
2325 * be considered authoritative for.
2327 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2328 struct btrfs_root
*log
,
2329 struct btrfs_path
*path
,
2330 int key_type
, u64 dirid
,
2331 u64 first_offset
, u64 last_offset
)
2334 struct btrfs_key key
;
2335 struct btrfs_dir_log_item
*item
;
2337 key
.objectid
= dirid
;
2338 key
.offset
= first_offset
;
2339 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2340 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2342 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2343 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2347 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2348 struct btrfs_dir_log_item
);
2349 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2350 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2351 btrfs_release_path(path
);
2356 * log all the items included in the current transaction for a given
2357 * directory. This also creates the range items in the log tree required
2358 * to replay anything deleted before the fsync
2360 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2361 struct btrfs_root
*root
, struct inode
*inode
,
2362 struct btrfs_path
*path
,
2363 struct btrfs_path
*dst_path
, int key_type
,
2364 u64 min_offset
, u64
*last_offset_ret
)
2366 struct btrfs_key min_key
;
2367 struct btrfs_key max_key
;
2368 struct btrfs_root
*log
= root
->log_root
;
2369 struct extent_buffer
*src
;
2374 u64 first_offset
= min_offset
;
2375 u64 last_offset
= (u64
)-1;
2376 u64 ino
= btrfs_ino(inode
);
2378 log
= root
->log_root
;
2379 max_key
.objectid
= ino
;
2380 max_key
.offset
= (u64
)-1;
2381 max_key
.type
= key_type
;
2383 min_key
.objectid
= ino
;
2384 min_key
.type
= key_type
;
2385 min_key
.offset
= min_offset
;
2387 path
->keep_locks
= 1;
2389 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2390 path
, 0, trans
->transid
);
2393 * we didn't find anything from this transaction, see if there
2394 * is anything at all
2396 if (ret
!= 0 || min_key
.objectid
!= ino
|| min_key
.type
!= key_type
) {
2397 min_key
.objectid
= ino
;
2398 min_key
.type
= key_type
;
2399 min_key
.offset
= (u64
)-1;
2400 btrfs_release_path(path
);
2401 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2403 btrfs_release_path(path
);
2406 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2408 /* if ret == 0 there are items for this type,
2409 * create a range to tell us the last key of this type.
2410 * otherwise, there are no items in this directory after
2411 * *min_offset, and we create a range to indicate that.
2414 struct btrfs_key tmp
;
2415 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2417 if (key_type
== tmp
.type
)
2418 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2423 /* go backward to find any previous key */
2424 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2426 struct btrfs_key tmp
;
2427 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2428 if (key_type
== tmp
.type
) {
2429 first_offset
= tmp
.offset
;
2430 ret
= overwrite_item(trans
, log
, dst_path
,
2431 path
->nodes
[0], path
->slots
[0],
2439 btrfs_release_path(path
);
2441 /* find the first key from this transaction again */
2442 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2449 * we have a block from this transaction, log every item in it
2450 * from our directory
2453 struct btrfs_key tmp
;
2454 src
= path
->nodes
[0];
2455 nritems
= btrfs_header_nritems(src
);
2456 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2457 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2459 if (min_key
.objectid
!= ino
|| min_key
.type
!= key_type
)
2461 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2468 path
->slots
[0] = nritems
;
2471 * look ahead to the next item and see if it is also
2472 * from this directory and from this transaction
2474 ret
= btrfs_next_leaf(root
, path
);
2476 last_offset
= (u64
)-1;
2479 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2480 if (tmp
.objectid
!= ino
|| tmp
.type
!= key_type
) {
2481 last_offset
= (u64
)-1;
2484 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2485 ret
= overwrite_item(trans
, log
, dst_path
,
2486 path
->nodes
[0], path
->slots
[0],
2491 last_offset
= tmp
.offset
;
2496 btrfs_release_path(path
);
2497 btrfs_release_path(dst_path
);
2500 *last_offset_ret
= last_offset
;
2502 * insert the log range keys to indicate where the log
2505 ret
= insert_dir_log_key(trans
, log
, path
, key_type
,
2506 ino
, first_offset
, last_offset
);
2514 * logging directories is very similar to logging inodes, We find all the items
2515 * from the current transaction and write them to the log.
2517 * The recovery code scans the directory in the subvolume, and if it finds a
2518 * key in the range logged that is not present in the log tree, then it means
2519 * that dir entry was unlinked during the transaction.
2521 * In order for that scan to work, we must include one key smaller than
2522 * the smallest logged by this transaction and one key larger than the largest
2523 * key logged by this transaction.
2525 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
2526 struct btrfs_root
*root
, struct inode
*inode
,
2527 struct btrfs_path
*path
,
2528 struct btrfs_path
*dst_path
)
2533 int key_type
= BTRFS_DIR_ITEM_KEY
;
2539 ret
= log_dir_items(trans
, root
, inode
, path
,
2540 dst_path
, key_type
, min_key
,
2544 if (max_key
== (u64
)-1)
2546 min_key
= max_key
+ 1;
2549 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
2550 key_type
= BTRFS_DIR_INDEX_KEY
;
2557 * a helper function to drop items from the log before we relog an
2558 * inode. max_key_type indicates the highest item type to remove.
2559 * This cannot be run for file data extents because it does not
2560 * free the extents they point to.
2562 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
2563 struct btrfs_root
*log
,
2564 struct btrfs_path
*path
,
2565 u64 objectid
, int max_key_type
)
2568 struct btrfs_key key
;
2569 struct btrfs_key found_key
;
2571 key
.objectid
= objectid
;
2572 key
.type
= max_key_type
;
2573 key
.offset
= (u64
)-1;
2576 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
2581 if (path
->slots
[0] == 0)
2585 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2588 if (found_key
.objectid
!= objectid
)
2591 ret
= btrfs_del_item(trans
, log
, path
);
2593 btrfs_release_path(path
);
2595 btrfs_release_path(path
);
2599 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
2600 struct btrfs_root
*log
,
2601 struct btrfs_path
*dst_path
,
2602 struct extent_buffer
*src
,
2603 int start_slot
, int nr
, int inode_only
)
2605 unsigned long src_offset
;
2606 unsigned long dst_offset
;
2607 struct btrfs_file_extent_item
*extent
;
2608 struct btrfs_inode_item
*inode_item
;
2610 struct btrfs_key
*ins_keys
;
2614 struct list_head ordered_sums
;
2616 INIT_LIST_HEAD(&ordered_sums
);
2618 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
2619 nr
* sizeof(u32
), GFP_NOFS
);
2623 ins_sizes
= (u32
*)ins_data
;
2624 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
2626 for (i
= 0; i
< nr
; i
++) {
2627 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
2628 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
2630 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
2631 ins_keys
, ins_sizes
, nr
);
2637 for (i
= 0; i
< nr
; i
++, dst_path
->slots
[0]++) {
2638 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
2639 dst_path
->slots
[0]);
2641 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
2643 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
2644 src_offset
, ins_sizes
[i
]);
2646 if (inode_only
== LOG_INODE_EXISTS
&&
2647 ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
2648 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
2650 struct btrfs_inode_item
);
2651 btrfs_set_inode_size(dst_path
->nodes
[0], inode_item
, 0);
2653 /* set the generation to zero so the recover code
2654 * can tell the difference between an logging
2655 * just to say 'this inode exists' and a logging
2656 * to say 'update this inode with these values'
2658 btrfs_set_inode_generation(dst_path
->nodes
[0],
2661 /* take a reference on file data extents so that truncates
2662 * or deletes of this inode don't have to relog the inode
2665 if (btrfs_key_type(ins_keys
+ i
) == BTRFS_EXTENT_DATA_KEY
) {
2667 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
2668 struct btrfs_file_extent_item
);
2670 found_type
= btrfs_file_extent_type(src
, extent
);
2671 if (found_type
== BTRFS_FILE_EXTENT_REG
||
2672 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
2674 ds
= btrfs_file_extent_disk_bytenr(src
,
2676 /* ds == 0 is a hole */
2680 dl
= btrfs_file_extent_disk_num_bytes(src
,
2682 cs
= btrfs_file_extent_offset(src
, extent
);
2683 cl
= btrfs_file_extent_num_bytes(src
,
2685 if (btrfs_file_extent_compression(src
,
2691 ret
= btrfs_lookup_csums_range(
2692 log
->fs_info
->csum_root
,
2693 ds
+ cs
, ds
+ cs
+ cl
- 1,
2700 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
2701 btrfs_release_path(dst_path
);
2705 * we have to do this after the loop above to avoid changing the
2706 * log tree while trying to change the log tree.
2709 while (!list_empty(&ordered_sums
)) {
2710 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
2711 struct btrfs_ordered_sum
,
2714 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
2715 list_del(&sums
->list
);
2721 /* log a single inode in the tree log.
2722 * At least one parent directory for this inode must exist in the tree
2723 * or be logged already.
2725 * Any items from this inode changed by the current transaction are copied
2726 * to the log tree. An extra reference is taken on any extents in this
2727 * file, allowing us to avoid a whole pile of corner cases around logging
2728 * blocks that have been removed from the tree.
2730 * See LOG_INODE_ALL and related defines for a description of what inode_only
2733 * This handles both files and directories.
2735 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2736 struct btrfs_root
*root
, struct inode
*inode
,
2739 struct btrfs_path
*path
;
2740 struct btrfs_path
*dst_path
;
2741 struct btrfs_key min_key
;
2742 struct btrfs_key max_key
;
2743 struct btrfs_root
*log
= root
->log_root
;
2744 struct extent_buffer
*src
= NULL
;
2748 int ins_start_slot
= 0;
2750 u64 ino
= btrfs_ino(inode
);
2752 log
= root
->log_root
;
2754 path
= btrfs_alloc_path();
2757 dst_path
= btrfs_alloc_path();
2759 btrfs_free_path(path
);
2763 min_key
.objectid
= ino
;
2764 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
2767 max_key
.objectid
= ino
;
2769 /* today the code can only do partial logging of directories */
2770 if (!S_ISDIR(inode
->i_mode
))
2771 inode_only
= LOG_INODE_ALL
;
2773 if (inode_only
== LOG_INODE_EXISTS
|| S_ISDIR(inode
->i_mode
))
2774 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
2776 max_key
.type
= (u8
)-1;
2777 max_key
.offset
= (u64
)-1;
2779 ret
= btrfs_commit_inode_delayed_items(trans
, inode
);
2781 btrfs_free_path(path
);
2782 btrfs_free_path(dst_path
);
2786 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2789 * a brute force approach to making sure we get the most uptodate
2790 * copies of everything.
2792 if (S_ISDIR(inode
->i_mode
)) {
2793 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
2795 if (inode_only
== LOG_INODE_EXISTS
)
2796 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
2797 ret
= drop_objectid_items(trans
, log
, path
, ino
, max_key_type
);
2799 ret
= btrfs_truncate_inode_items(trans
, log
, inode
, 0, 0);
2805 path
->keep_locks
= 1;
2809 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2810 path
, 0, trans
->transid
);
2814 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2815 if (min_key
.objectid
!= ino
)
2817 if (min_key
.type
> max_key
.type
)
2820 src
= path
->nodes
[0];
2821 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
2824 } else if (!ins_nr
) {
2825 ins_start_slot
= path
->slots
[0];
2830 ret
= copy_items(trans
, log
, dst_path
, src
, ins_start_slot
,
2831 ins_nr
, inode_only
);
2837 ins_start_slot
= path
->slots
[0];
2840 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2842 if (path
->slots
[0] < nritems
) {
2843 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
2848 ret
= copy_items(trans
, log
, dst_path
, src
,
2850 ins_nr
, inode_only
);
2857 btrfs_release_path(path
);
2859 if (min_key
.offset
< (u64
)-1)
2861 else if (min_key
.type
< (u8
)-1)
2863 else if (min_key
.objectid
< (u64
)-1)
2869 ret
= copy_items(trans
, log
, dst_path
, src
,
2871 ins_nr
, inode_only
);
2879 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
2880 btrfs_release_path(path
);
2881 btrfs_release_path(dst_path
);
2882 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
2888 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
2890 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2892 btrfs_free_path(path
);
2893 btrfs_free_path(dst_path
);
2898 * follow the dentry parent pointers up the chain and see if any
2899 * of the directories in it require a full commit before they can
2900 * be logged. Returns zero if nothing special needs to be done or 1 if
2901 * a full commit is required.
2903 static noinline
int check_parent_dirs_for_sync(struct btrfs_trans_handle
*trans
,
2904 struct inode
*inode
,
2905 struct dentry
*parent
,
2906 struct super_block
*sb
,
2910 struct btrfs_root
*root
;
2911 struct dentry
*old_parent
= NULL
;
2914 * for regular files, if its inode is already on disk, we don't
2915 * have to worry about the parents at all. This is because
2916 * we can use the last_unlink_trans field to record renames
2917 * and other fun in this file.
2919 if (S_ISREG(inode
->i_mode
) &&
2920 BTRFS_I(inode
)->generation
<= last_committed
&&
2921 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
)
2924 if (!S_ISDIR(inode
->i_mode
)) {
2925 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
2927 inode
= parent
->d_inode
;
2931 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
2934 if (BTRFS_I(inode
)->last_unlink_trans
> last_committed
) {
2935 root
= BTRFS_I(inode
)->root
;
2938 * make sure any commits to the log are forced
2939 * to be full commits
2941 root
->fs_info
->last_trans_log_full_commit
=
2947 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
2950 if (IS_ROOT(parent
))
2953 parent
= dget_parent(parent
);
2955 old_parent
= parent
;
2956 inode
= parent
->d_inode
;
2964 static int inode_in_log(struct btrfs_trans_handle
*trans
,
2965 struct inode
*inode
)
2967 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2970 mutex_lock(&root
->log_mutex
);
2971 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
&&
2972 BTRFS_I(inode
)->last_sub_trans
<= root
->last_log_commit
)
2974 mutex_unlock(&root
->log_mutex
);
2980 * helper function around btrfs_log_inode to make sure newly created
2981 * parent directories also end up in the log. A minimal inode and backref
2982 * only logging is done of any parent directories that are older than
2983 * the last committed transaction
2985 int btrfs_log_inode_parent(struct btrfs_trans_handle
*trans
,
2986 struct btrfs_root
*root
, struct inode
*inode
,
2987 struct dentry
*parent
, int exists_only
)
2989 int inode_only
= exists_only
? LOG_INODE_EXISTS
: LOG_INODE_ALL
;
2990 struct super_block
*sb
;
2991 struct dentry
*old_parent
= NULL
;
2993 u64 last_committed
= root
->fs_info
->last_trans_committed
;
2997 if (btrfs_test_opt(root
, NOTREELOG
)) {
3002 if (root
->fs_info
->last_trans_log_full_commit
>
3003 root
->fs_info
->last_trans_committed
) {
3008 if (root
!= BTRFS_I(inode
)->root
||
3009 btrfs_root_refs(&root
->root_item
) == 0) {
3014 ret
= check_parent_dirs_for_sync(trans
, inode
, parent
,
3015 sb
, last_committed
);
3019 if (inode_in_log(trans
, inode
)) {
3020 ret
= BTRFS_NO_LOG_SYNC
;
3024 ret
= start_log_trans(trans
, root
);
3028 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3033 * for regular files, if its inode is already on disk, we don't
3034 * have to worry about the parents at all. This is because
3035 * we can use the last_unlink_trans field to record renames
3036 * and other fun in this file.
3038 if (S_ISREG(inode
->i_mode
) &&
3039 BTRFS_I(inode
)->generation
<= last_committed
&&
3040 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
) {
3045 inode_only
= LOG_INODE_EXISTS
;
3047 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3050 inode
= parent
->d_inode
;
3051 if (root
!= BTRFS_I(inode
)->root
)
3054 if (BTRFS_I(inode
)->generation
>
3055 root
->fs_info
->last_trans_committed
) {
3056 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3060 if (IS_ROOT(parent
))
3063 parent
= dget_parent(parent
);
3065 old_parent
= parent
;
3071 BUG_ON(ret
!= -ENOSPC
);
3072 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
3075 btrfs_end_log_trans(root
);
3081 * it is not safe to log dentry if the chunk root has added new
3082 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3083 * If this returns 1, you must commit the transaction to safely get your
3086 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
3087 struct btrfs_root
*root
, struct dentry
*dentry
)
3089 struct dentry
*parent
= dget_parent(dentry
);
3092 ret
= btrfs_log_inode_parent(trans
, root
, dentry
->d_inode
, parent
, 0);
3099 * should be called during mount to recover any replay any log trees
3102 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
3105 struct btrfs_path
*path
;
3106 struct btrfs_trans_handle
*trans
;
3107 struct btrfs_key key
;
3108 struct btrfs_key found_key
;
3109 struct btrfs_key tmp_key
;
3110 struct btrfs_root
*log
;
3111 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
3112 struct walk_control wc
= {
3113 .process_func
= process_one_buffer
,
3117 path
= btrfs_alloc_path();
3121 fs_info
->log_root_recovering
= 1;
3123 trans
= btrfs_start_transaction(fs_info
->tree_root
, 0);
3124 BUG_ON(IS_ERR(trans
));
3129 ret
= walk_log_tree(trans
, log_root_tree
, &wc
);
3133 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
3134 key
.offset
= (u64
)-1;
3135 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
3138 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
3142 if (path
->slots
[0] == 0)
3146 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
3148 btrfs_release_path(path
);
3149 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
3152 log
= btrfs_read_fs_root_no_radix(log_root_tree
,
3154 BUG_ON(IS_ERR(log
));
3156 tmp_key
.objectid
= found_key
.offset
;
3157 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3158 tmp_key
.offset
= (u64
)-1;
3160 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
3161 BUG_ON(!wc
.replay_dest
);
3163 wc
.replay_dest
->log_root
= log
;
3164 btrfs_record_root_in_trans(trans
, wc
.replay_dest
);
3165 ret
= walk_log_tree(trans
, log
, &wc
);
3168 if (wc
.stage
== LOG_WALK_REPLAY_ALL
) {
3169 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
3174 key
.offset
= found_key
.offset
- 1;
3175 wc
.replay_dest
->log_root
= NULL
;
3176 free_extent_buffer(log
->node
);
3177 free_extent_buffer(log
->commit_root
);
3180 if (found_key
.offset
== 0)
3183 btrfs_release_path(path
);
3185 /* step one is to pin it all, step two is to replay just inodes */
3188 wc
.process_func
= replay_one_buffer
;
3189 wc
.stage
= LOG_WALK_REPLAY_INODES
;
3192 /* step three is to replay everything */
3193 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
3198 btrfs_free_path(path
);
3200 free_extent_buffer(log_root_tree
->node
);
3201 log_root_tree
->log_root
= NULL
;
3202 fs_info
->log_root_recovering
= 0;
3204 /* step 4: commit the transaction, which also unpins the blocks */
3205 btrfs_commit_transaction(trans
, fs_info
->tree_root
);
3207 kfree(log_root_tree
);
3212 * there are some corner cases where we want to force a full
3213 * commit instead of allowing a directory to be logged.
3215 * They revolve around files there were unlinked from the directory, and
3216 * this function updates the parent directory so that a full commit is
3217 * properly done if it is fsync'd later after the unlinks are done.
3219 void btrfs_record_unlink_dir(struct btrfs_trans_handle
*trans
,
3220 struct inode
*dir
, struct inode
*inode
,
3224 * when we're logging a file, if it hasn't been renamed
3225 * or unlinked, and its inode is fully committed on disk,
3226 * we don't have to worry about walking up the directory chain
3227 * to log its parents.
3229 * So, we use the last_unlink_trans field to put this transid
3230 * into the file. When the file is logged we check it and
3231 * don't log the parents if the file is fully on disk.
3233 if (S_ISREG(inode
->i_mode
))
3234 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
3237 * if this directory was already logged any new
3238 * names for this file/dir will get recorded
3241 if (BTRFS_I(dir
)->logged_trans
== trans
->transid
)
3245 * if the inode we're about to unlink was logged,
3246 * the log will be properly updated for any new names
3248 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
)
3252 * when renaming files across directories, if the directory
3253 * there we're unlinking from gets fsync'd later on, there's
3254 * no way to find the destination directory later and fsync it
3255 * properly. So, we have to be conservative and force commits
3256 * so the new name gets discovered.
3261 /* we can safely do the unlink without any special recording */
3265 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
3269 * Call this after adding a new name for a file and it will properly
3270 * update the log to reflect the new name.
3272 * It will return zero if all goes well, and it will return 1 if a
3273 * full transaction commit is required.
3275 int btrfs_log_new_name(struct btrfs_trans_handle
*trans
,
3276 struct inode
*inode
, struct inode
*old_dir
,
3277 struct dentry
*parent
)
3279 struct btrfs_root
* root
= BTRFS_I(inode
)->root
;
3282 * this will force the logging code to walk the dentry chain
3285 if (S_ISREG(inode
->i_mode
))
3286 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
3289 * if this inode hasn't been logged and directory we're renaming it
3290 * from hasn't been logged, we don't need to log it
3292 if (BTRFS_I(inode
)->logged_trans
<=
3293 root
->fs_info
->last_trans_committed
&&
3294 (!old_dir
|| BTRFS_I(old_dir
)->logged_trans
<=
3295 root
->fs_info
->last_trans_committed
))
3298 return btrfs_log_inode_parent(trans
, root
, inode
, parent
, 1);