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 void 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
);
226 * the walk control struct is used to pass state down the chain when
227 * processing the log tree. The stage field tells us which part
228 * of the log tree processing we are currently doing. The others
229 * are state fields used for that specific part
231 struct walk_control
{
232 /* should we free the extent on disk when done? This is used
233 * at transaction commit time while freeing a log tree
237 /* should we write out the extent buffer? This is used
238 * while flushing the log tree to disk during a sync
242 /* should we wait for the extent buffer io to finish? Also used
243 * while flushing the log tree to disk for a sync
247 /* pin only walk, we record which extents on disk belong to the
252 /* what stage of the replay code we're currently in */
255 /* the root we are currently replaying */
256 struct btrfs_root
*replay_dest
;
258 /* the trans handle for the current replay */
259 struct btrfs_trans_handle
*trans
;
261 /* the function that gets used to process blocks we find in the
262 * tree. Note the extent_buffer might not be up to date when it is
263 * passed in, and it must be checked or read if you need the data
266 int (*process_func
)(struct btrfs_root
*log
, struct extent_buffer
*eb
,
267 struct walk_control
*wc
, u64 gen
);
271 * process_func used to pin down extents, write them or wait on them
273 static int process_one_buffer(struct btrfs_root
*log
,
274 struct extent_buffer
*eb
,
275 struct walk_control
*wc
, u64 gen
)
278 btrfs_pin_extent_for_log_replay(wc
->trans
,
279 log
->fs_info
->extent_root
,
282 if (btrfs_buffer_uptodate(eb
, gen
, 0)) {
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 btrfs_extend_item(trans
, root
, path
,
384 item_size
- found_size
);
388 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0],
391 /* don't overwrite an existing inode if the generation number
392 * was logged as zero. This is done when the tree logging code
393 * is just logging an inode to make sure it exists after recovery.
395 * Also, don't overwrite i_size on directories during replay.
396 * log replay inserts and removes directory items based on the
397 * state of the tree found in the subvolume, and i_size is modified
400 if (key
->type
== BTRFS_INODE_ITEM_KEY
&& ret
== -EEXIST
) {
401 struct btrfs_inode_item
*src_item
;
402 struct btrfs_inode_item
*dst_item
;
404 src_item
= (struct btrfs_inode_item
*)src_ptr
;
405 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
407 if (btrfs_inode_generation(eb
, src_item
) == 0)
410 if (overwrite_root
&&
411 S_ISDIR(btrfs_inode_mode(eb
, src_item
)) &&
412 S_ISDIR(btrfs_inode_mode(path
->nodes
[0], dst_item
))) {
414 saved_i_size
= btrfs_inode_size(path
->nodes
[0],
419 copy_extent_buffer(path
->nodes
[0], eb
, dst_ptr
,
422 if (save_old_i_size
) {
423 struct btrfs_inode_item
*dst_item
;
424 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
425 btrfs_set_inode_size(path
->nodes
[0], dst_item
, saved_i_size
);
428 /* make sure the generation is filled in */
429 if (key
->type
== BTRFS_INODE_ITEM_KEY
) {
430 struct btrfs_inode_item
*dst_item
;
431 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
432 if (btrfs_inode_generation(path
->nodes
[0], dst_item
) == 0) {
433 btrfs_set_inode_generation(path
->nodes
[0], dst_item
,
438 btrfs_mark_buffer_dirty(path
->nodes
[0]);
439 btrfs_release_path(path
);
444 * simple helper to read an inode off the disk from a given root
445 * This can only be called for subvolume roots and not for the log
447 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
450 struct btrfs_key key
;
453 key
.objectid
= objectid
;
454 key
.type
= BTRFS_INODE_ITEM_KEY
;
456 inode
= btrfs_iget(root
->fs_info
->sb
, &key
, root
, NULL
);
459 } else if (is_bad_inode(inode
)) {
466 /* replays a single extent in 'eb' at 'slot' with 'key' into the
467 * subvolume 'root'. path is released on entry and should be released
470 * extents in the log tree have not been allocated out of the extent
471 * tree yet. So, this completes the allocation, taking a reference
472 * as required if the extent already exists or creating a new extent
473 * if it isn't in the extent allocation tree yet.
475 * The extent is inserted into the file, dropping any existing extents
476 * from the file that overlap the new one.
478 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
479 struct btrfs_root
*root
,
480 struct btrfs_path
*path
,
481 struct extent_buffer
*eb
, int slot
,
482 struct btrfs_key
*key
)
485 u64 mask
= root
->sectorsize
- 1;
488 u64 start
= key
->offset
;
490 struct btrfs_file_extent_item
*item
;
491 struct inode
*inode
= NULL
;
495 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
496 found_type
= btrfs_file_extent_type(eb
, item
);
498 if (found_type
== BTRFS_FILE_EXTENT_REG
||
499 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
500 extent_end
= start
+ btrfs_file_extent_num_bytes(eb
, item
);
501 else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
502 size
= btrfs_file_extent_inline_len(eb
, item
);
503 extent_end
= (start
+ size
+ mask
) & ~mask
;
509 inode
= read_one_inode(root
, key
->objectid
);
516 * first check to see if we already have this extent in the
517 * file. This must be done before the btrfs_drop_extents run
518 * so we don't try to drop this extent.
520 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
524 (found_type
== BTRFS_FILE_EXTENT_REG
||
525 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)) {
526 struct btrfs_file_extent_item cmp1
;
527 struct btrfs_file_extent_item cmp2
;
528 struct btrfs_file_extent_item
*existing
;
529 struct extent_buffer
*leaf
;
531 leaf
= path
->nodes
[0];
532 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
533 struct btrfs_file_extent_item
);
535 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
537 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
541 * we already have a pointer to this exact extent,
542 * we don't have to do anything
544 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
545 btrfs_release_path(path
);
549 btrfs_release_path(path
);
551 saved_nbytes
= inode_get_bytes(inode
);
552 /* drop any overlapping extents */
553 ret
= btrfs_drop_extents(trans
, inode
, start
, extent_end
,
557 if (found_type
== BTRFS_FILE_EXTENT_REG
||
558 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
560 unsigned long dest_offset
;
561 struct btrfs_key ins
;
563 ret
= btrfs_insert_empty_item(trans
, root
, path
, key
,
566 dest_offset
= btrfs_item_ptr_offset(path
->nodes
[0],
568 copy_extent_buffer(path
->nodes
[0], eb
, dest_offset
,
569 (unsigned long)item
, sizeof(*item
));
571 ins
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
572 ins
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
573 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
574 offset
= key
->offset
- btrfs_file_extent_offset(eb
, item
);
576 if (ins
.objectid
> 0) {
579 LIST_HEAD(ordered_sums
);
581 * is this extent already allocated in the extent
582 * allocation tree? If so, just add a reference
584 ret
= btrfs_lookup_extent(root
, ins
.objectid
,
587 ret
= btrfs_inc_extent_ref(trans
, root
,
588 ins
.objectid
, ins
.offset
,
589 0, root
->root_key
.objectid
,
590 key
->objectid
, offset
, 0);
594 * insert the extent pointer in the extent
597 ret
= btrfs_alloc_logged_file_extent(trans
,
598 root
, root
->root_key
.objectid
,
599 key
->objectid
, offset
, &ins
);
602 btrfs_release_path(path
);
604 if (btrfs_file_extent_compression(eb
, item
)) {
605 csum_start
= ins
.objectid
;
606 csum_end
= csum_start
+ ins
.offset
;
608 csum_start
= ins
.objectid
+
609 btrfs_file_extent_offset(eb
, item
);
610 csum_end
= csum_start
+
611 btrfs_file_extent_num_bytes(eb
, item
);
614 ret
= btrfs_lookup_csums_range(root
->log_root
,
615 csum_start
, csum_end
- 1,
618 while (!list_empty(&ordered_sums
)) {
619 struct btrfs_ordered_sum
*sums
;
620 sums
= list_entry(ordered_sums
.next
,
621 struct btrfs_ordered_sum
,
623 ret
= btrfs_csum_file_blocks(trans
,
624 root
->fs_info
->csum_root
,
627 list_del(&sums
->list
);
631 btrfs_release_path(path
);
633 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
634 /* inline extents are easy, we just overwrite them */
635 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
639 inode_set_bytes(inode
, saved_nbytes
);
640 btrfs_update_inode(trans
, root
, inode
);
648 * when cleaning up conflicts between the directory names in the
649 * subvolume, directory names in the log and directory names in the
650 * inode back references, we may have to unlink inodes from directories.
652 * This is a helper function to do the unlink of a specific directory
655 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
656 struct btrfs_root
*root
,
657 struct btrfs_path
*path
,
659 struct btrfs_dir_item
*di
)
664 struct extent_buffer
*leaf
;
665 struct btrfs_key location
;
668 leaf
= path
->nodes
[0];
670 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
671 name_len
= btrfs_dir_name_len(leaf
, di
);
672 name
= kmalloc(name_len
, GFP_NOFS
);
676 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
677 btrfs_release_path(path
);
679 inode
= read_one_inode(root
, location
.objectid
);
685 ret
= link_to_fixup_dir(trans
, root
, path
, location
.objectid
);
688 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
697 * helper function to see if a given name and sequence number found
698 * in an inode back reference are already in a directory and correctly
699 * point to this inode
701 static noinline
int inode_in_dir(struct btrfs_root
*root
,
702 struct btrfs_path
*path
,
703 u64 dirid
, u64 objectid
, u64 index
,
704 const char *name
, int name_len
)
706 struct btrfs_dir_item
*di
;
707 struct btrfs_key location
;
710 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
711 index
, name
, name_len
, 0);
712 if (di
&& !IS_ERR(di
)) {
713 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
714 if (location
.objectid
!= objectid
)
718 btrfs_release_path(path
);
720 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
721 if (di
&& !IS_ERR(di
)) {
722 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
723 if (location
.objectid
!= objectid
)
729 btrfs_release_path(path
);
734 * helper function to check a log tree for a named back reference in
735 * an inode. This is used to decide if a back reference that is
736 * found in the subvolume conflicts with what we find in the log.
738 * inode backreferences may have multiple refs in a single item,
739 * during replay we process one reference at a time, and we don't
740 * want to delete valid links to a file from the subvolume if that
741 * link is also in the log.
743 static noinline
int backref_in_log(struct btrfs_root
*log
,
744 struct btrfs_key
*key
,
745 char *name
, int namelen
)
747 struct btrfs_path
*path
;
748 struct btrfs_inode_ref
*ref
;
750 unsigned long ptr_end
;
751 unsigned long name_ptr
;
757 path
= btrfs_alloc_path();
761 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
765 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
766 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
767 ptr_end
= ptr
+ item_size
;
768 while (ptr
< ptr_end
) {
769 ref
= (struct btrfs_inode_ref
*)ptr
;
770 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
771 if (found_name_len
== namelen
) {
772 name_ptr
= (unsigned long)(ref
+ 1);
773 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
780 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
783 btrfs_free_path(path
);
789 * replay one inode back reference item found in the log tree.
790 * eb, slot and key refer to the buffer and key found in the log tree.
791 * root is the destination we are replaying into, and path is for temp
792 * use by this function. (it should be released on return).
794 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
795 struct btrfs_root
*root
,
796 struct btrfs_root
*log
,
797 struct btrfs_path
*path
,
798 struct extent_buffer
*eb
, int slot
,
799 struct btrfs_key
*key
)
801 struct btrfs_inode_ref
*ref
;
802 struct btrfs_dir_item
*di
;
805 unsigned long ref_ptr
;
806 unsigned long ref_end
;
813 * it is possible that we didn't log all the parent directories
814 * for a given inode. If we don't find the dir, just don't
815 * copy the back ref in. The link count fixup code will take
818 dir
= read_one_inode(root
, key
->offset
);
822 inode
= read_one_inode(root
, key
->objectid
);
828 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
829 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
832 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
834 namelen
= btrfs_inode_ref_name_len(eb
, ref
);
835 name
= kmalloc(namelen
, GFP_NOFS
);
838 read_extent_buffer(eb
, name
, (unsigned long)(ref
+ 1), namelen
);
840 /* if we already have a perfect match, we're done */
841 if (inode_in_dir(root
, path
, btrfs_ino(dir
), btrfs_ino(inode
),
842 btrfs_inode_ref_index(eb
, ref
),
848 * look for a conflicting back reference in the metadata.
849 * if we find one we have to unlink that name of the file
850 * before we add our new link. Later on, we overwrite any
851 * existing back reference, and we don't want to create
852 * dangling pointers in the directory.
858 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
862 struct btrfs_inode_ref
*victim_ref
;
864 unsigned long ptr_end
;
865 struct extent_buffer
*leaf
= path
->nodes
[0];
867 /* are we trying to overwrite a back ref for the root directory
868 * if so, just jump out, we're done
870 if (key
->objectid
== key
->offset
)
873 /* check all the names in this back reference to see
874 * if they are in the log. if so, we allow them to stay
875 * otherwise they must be unlinked as a conflict
877 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
878 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
879 while (ptr
< ptr_end
) {
880 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
881 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
883 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
884 BUG_ON(!victim_name
);
886 read_extent_buffer(leaf
, victim_name
,
887 (unsigned long)(victim_ref
+ 1),
890 if (!backref_in_log(log
, key
, victim_name
,
892 btrfs_inc_nlink(inode
);
893 btrfs_release_path(path
);
895 ret
= btrfs_unlink_inode(trans
, root
, dir
,
900 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
905 * NOTE: we have searched root tree and checked the
906 * coresponding ref, it does not need to check again.
910 btrfs_release_path(path
);
912 /* look for a conflicting sequence number */
913 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, btrfs_ino(dir
),
914 btrfs_inode_ref_index(eb
, ref
),
916 if (di
&& !IS_ERR(di
)) {
917 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
920 btrfs_release_path(path
);
922 /* look for a conflicing name */
923 di
= btrfs_lookup_dir_item(trans
, root
, path
, btrfs_ino(dir
),
925 if (di
&& !IS_ERR(di
)) {
926 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
929 btrfs_release_path(path
);
932 /* insert our name */
933 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
, 0,
934 btrfs_inode_ref_index(eb
, ref
));
937 btrfs_update_inode(trans
, root
, inode
);
940 ref_ptr
= (unsigned long)(ref
+ 1) + namelen
;
942 if (ref_ptr
< ref_end
)
945 /* finally write the back reference in the inode */
946 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
950 btrfs_release_path(path
);
956 static int insert_orphan_item(struct btrfs_trans_handle
*trans
,
957 struct btrfs_root
*root
, u64 offset
)
960 ret
= btrfs_find_orphan_item(root
, offset
);
962 ret
= btrfs_insert_orphan_item(trans
, root
, offset
);
968 * There are a few corners where the link count of the file can't
969 * be properly maintained during replay. So, instead of adding
970 * lots of complexity to the log code, we just scan the backrefs
971 * for any file that has been through replay.
973 * The scan will update the link count on the inode to reflect the
974 * number of back refs found. If it goes down to zero, the iput
975 * will free the inode.
977 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
978 struct btrfs_root
*root
,
981 struct btrfs_path
*path
;
983 struct btrfs_key key
;
986 unsigned long ptr_end
;
988 u64 ino
= btrfs_ino(inode
);
991 key
.type
= BTRFS_INODE_REF_KEY
;
992 key
.offset
= (u64
)-1;
994 path
= btrfs_alloc_path();
999 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1003 if (path
->slots
[0] == 0)
1007 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1009 if (key
.objectid
!= ino
||
1010 key
.type
!= BTRFS_INODE_REF_KEY
)
1012 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1013 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1015 while (ptr
< ptr_end
) {
1016 struct btrfs_inode_ref
*ref
;
1018 ref
= (struct btrfs_inode_ref
*)ptr
;
1019 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1021 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1025 if (key
.offset
== 0)
1028 btrfs_release_path(path
);
1030 btrfs_release_path(path
);
1031 if (nlink
!= inode
->i_nlink
) {
1032 set_nlink(inode
, nlink
);
1033 btrfs_update_inode(trans
, root
, inode
);
1035 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1037 if (inode
->i_nlink
== 0) {
1038 if (S_ISDIR(inode
->i_mode
)) {
1039 ret
= replay_dir_deletes(trans
, root
, NULL
, path
,
1043 ret
= insert_orphan_item(trans
, root
, ino
);
1046 btrfs_free_path(path
);
1051 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1052 struct btrfs_root
*root
,
1053 struct btrfs_path
*path
)
1056 struct btrfs_key key
;
1057 struct inode
*inode
;
1059 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1060 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1061 key
.offset
= (u64
)-1;
1063 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1068 if (path
->slots
[0] == 0)
1073 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1074 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1075 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1078 ret
= btrfs_del_item(trans
, root
, path
);
1082 btrfs_release_path(path
);
1083 inode
= read_one_inode(root
, key
.offset
);
1087 ret
= fixup_inode_link_count(trans
, root
, inode
);
1093 * fixup on a directory may create new entries,
1094 * make sure we always look for the highset possible
1097 key
.offset
= (u64
)-1;
1101 btrfs_release_path(path
);
1107 * record a given inode in the fixup dir so we can check its link
1108 * count when replay is done. The link count is incremented here
1109 * so the inode won't go away until we check it
1111 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1112 struct btrfs_root
*root
,
1113 struct btrfs_path
*path
,
1116 struct btrfs_key key
;
1118 struct inode
*inode
;
1120 inode
= read_one_inode(root
, objectid
);
1124 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1125 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1126 key
.offset
= objectid
;
1128 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1130 btrfs_release_path(path
);
1132 btrfs_inc_nlink(inode
);
1133 btrfs_update_inode(trans
, root
, inode
);
1134 } else if (ret
== -EEXIST
) {
1145 * when replaying the log for a directory, we only insert names
1146 * for inodes that actually exist. This means an fsync on a directory
1147 * does not implicitly fsync all the new files in it
1149 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1150 struct btrfs_root
*root
,
1151 struct btrfs_path
*path
,
1152 u64 dirid
, u64 index
,
1153 char *name
, int name_len
, u8 type
,
1154 struct btrfs_key
*location
)
1156 struct inode
*inode
;
1160 inode
= read_one_inode(root
, location
->objectid
);
1164 dir
= read_one_inode(root
, dirid
);
1169 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1171 /* FIXME, put inode into FIXUP list */
1179 * take a single entry in a log directory item and replay it into
1182 * if a conflicting item exists in the subdirectory already,
1183 * the inode it points to is unlinked and put into the link count
1186 * If a name from the log points to a file or directory that does
1187 * not exist in the FS, it is skipped. fsyncs on directories
1188 * do not force down inodes inside that directory, just changes to the
1189 * names or unlinks in a directory.
1191 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1192 struct btrfs_root
*root
,
1193 struct btrfs_path
*path
,
1194 struct extent_buffer
*eb
,
1195 struct btrfs_dir_item
*di
,
1196 struct btrfs_key
*key
)
1200 struct btrfs_dir_item
*dst_di
;
1201 struct btrfs_key found_key
;
1202 struct btrfs_key log_key
;
1208 dir
= read_one_inode(root
, key
->objectid
);
1212 name_len
= btrfs_dir_name_len(eb
, di
);
1213 name
= kmalloc(name_len
, GFP_NOFS
);
1217 log_type
= btrfs_dir_type(eb
, di
);
1218 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1221 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1222 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1227 btrfs_release_path(path
);
1229 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1230 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1232 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1233 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1240 if (IS_ERR_OR_NULL(dst_di
)) {
1241 /* we need a sequence number to insert, so we only
1242 * do inserts for the BTRFS_DIR_INDEX_KEY types
1244 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1249 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1250 /* the existing item matches the logged item */
1251 if (found_key
.objectid
== log_key
.objectid
&&
1252 found_key
.type
== log_key
.type
&&
1253 found_key
.offset
== log_key
.offset
&&
1254 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1259 * don't drop the conflicting directory entry if the inode
1260 * for the new entry doesn't exist
1265 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1268 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1271 btrfs_release_path(path
);
1277 btrfs_release_path(path
);
1278 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1279 name
, name_len
, log_type
, &log_key
);
1281 BUG_ON(ret
&& ret
!= -ENOENT
);
1286 * find all the names in a directory item and reconcile them into
1287 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1288 * one name in a directory item, but the same code gets used for
1289 * both directory index types
1291 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1292 struct btrfs_root
*root
,
1293 struct btrfs_path
*path
,
1294 struct extent_buffer
*eb
, int slot
,
1295 struct btrfs_key
*key
)
1298 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1299 struct btrfs_dir_item
*di
;
1302 unsigned long ptr_end
;
1304 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1305 ptr_end
= ptr
+ item_size
;
1306 while (ptr
< ptr_end
) {
1307 di
= (struct btrfs_dir_item
*)ptr
;
1308 if (verify_dir_item(root
, eb
, di
))
1310 name_len
= btrfs_dir_name_len(eb
, di
);
1311 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1313 ptr
= (unsigned long)(di
+ 1);
1320 * directory replay has two parts. There are the standard directory
1321 * items in the log copied from the subvolume, and range items
1322 * created in the log while the subvolume was logged.
1324 * The range items tell us which parts of the key space the log
1325 * is authoritative for. During replay, if a key in the subvolume
1326 * directory is in a logged range item, but not actually in the log
1327 * that means it was deleted from the directory before the fsync
1328 * and should be removed.
1330 static noinline
int find_dir_range(struct btrfs_root
*root
,
1331 struct btrfs_path
*path
,
1332 u64 dirid
, int key_type
,
1333 u64
*start_ret
, u64
*end_ret
)
1335 struct btrfs_key key
;
1337 struct btrfs_dir_log_item
*item
;
1341 if (*start_ret
== (u64
)-1)
1344 key
.objectid
= dirid
;
1345 key
.type
= key_type
;
1346 key
.offset
= *start_ret
;
1348 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1352 if (path
->slots
[0] == 0)
1357 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1359 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1363 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1364 struct btrfs_dir_log_item
);
1365 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1367 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1369 *start_ret
= key
.offset
;
1370 *end_ret
= found_end
;
1375 /* check the next slot in the tree to see if it is a valid item */
1376 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1377 if (path
->slots
[0] >= nritems
) {
1378 ret
= btrfs_next_leaf(root
, path
);
1385 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1387 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1391 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1392 struct btrfs_dir_log_item
);
1393 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1394 *start_ret
= key
.offset
;
1395 *end_ret
= found_end
;
1398 btrfs_release_path(path
);
1403 * this looks for a given directory item in the log. If the directory
1404 * item is not in the log, the item is removed and the inode it points
1407 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1408 struct btrfs_root
*root
,
1409 struct btrfs_root
*log
,
1410 struct btrfs_path
*path
,
1411 struct btrfs_path
*log_path
,
1413 struct btrfs_key
*dir_key
)
1416 struct extent_buffer
*eb
;
1419 struct btrfs_dir_item
*di
;
1420 struct btrfs_dir_item
*log_di
;
1423 unsigned long ptr_end
;
1425 struct inode
*inode
;
1426 struct btrfs_key location
;
1429 eb
= path
->nodes
[0];
1430 slot
= path
->slots
[0];
1431 item_size
= btrfs_item_size_nr(eb
, slot
);
1432 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1433 ptr_end
= ptr
+ item_size
;
1434 while (ptr
< ptr_end
) {
1435 di
= (struct btrfs_dir_item
*)ptr
;
1436 if (verify_dir_item(root
, eb
, di
)) {
1441 name_len
= btrfs_dir_name_len(eb
, di
);
1442 name
= kmalloc(name_len
, GFP_NOFS
);
1447 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1450 if (log
&& dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1451 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1454 } else if (log
&& dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1455 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1461 if (IS_ERR_OR_NULL(log_di
)) {
1462 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1463 btrfs_release_path(path
);
1464 btrfs_release_path(log_path
);
1465 inode
= read_one_inode(root
, location
.objectid
);
1471 ret
= link_to_fixup_dir(trans
, root
,
1472 path
, location
.objectid
);
1474 btrfs_inc_nlink(inode
);
1475 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1481 /* there might still be more names under this key
1482 * check and repeat if required
1484 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1491 btrfs_release_path(log_path
);
1494 ptr
= (unsigned long)(di
+ 1);
1499 btrfs_release_path(path
);
1500 btrfs_release_path(log_path
);
1505 * deletion replay happens before we copy any new directory items
1506 * out of the log or out of backreferences from inodes. It
1507 * scans the log to find ranges of keys that log is authoritative for,
1508 * and then scans the directory to find items in those ranges that are
1509 * not present in the log.
1511 * Anything we don't find in the log is unlinked and removed from the
1514 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1515 struct btrfs_root
*root
,
1516 struct btrfs_root
*log
,
1517 struct btrfs_path
*path
,
1518 u64 dirid
, int del_all
)
1522 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1524 struct btrfs_key dir_key
;
1525 struct btrfs_key found_key
;
1526 struct btrfs_path
*log_path
;
1529 dir_key
.objectid
= dirid
;
1530 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1531 log_path
= btrfs_alloc_path();
1535 dir
= read_one_inode(root
, dirid
);
1536 /* it isn't an error if the inode isn't there, that can happen
1537 * because we replay the deletes before we copy in the inode item
1541 btrfs_free_path(log_path
);
1549 range_end
= (u64
)-1;
1551 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1552 &range_start
, &range_end
);
1557 dir_key
.offset
= range_start
;
1560 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1565 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1566 if (path
->slots
[0] >= nritems
) {
1567 ret
= btrfs_next_leaf(root
, path
);
1571 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1573 if (found_key
.objectid
!= dirid
||
1574 found_key
.type
!= dir_key
.type
)
1577 if (found_key
.offset
> range_end
)
1580 ret
= check_item_in_log(trans
, root
, log
, path
,
1584 if (found_key
.offset
== (u64
)-1)
1586 dir_key
.offset
= found_key
.offset
+ 1;
1588 btrfs_release_path(path
);
1589 if (range_end
== (u64
)-1)
1591 range_start
= range_end
+ 1;
1596 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1597 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1598 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1599 btrfs_release_path(path
);
1603 btrfs_release_path(path
);
1604 btrfs_free_path(log_path
);
1610 * the process_func used to replay items from the log tree. This
1611 * gets called in two different stages. The first stage just looks
1612 * for inodes and makes sure they are all copied into the subvolume.
1614 * The second stage copies all the other item types from the log into
1615 * the subvolume. The two stage approach is slower, but gets rid of
1616 * lots of complexity around inodes referencing other inodes that exist
1617 * only in the log (references come from either directory items or inode
1620 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
1621 struct walk_control
*wc
, u64 gen
)
1624 struct btrfs_path
*path
;
1625 struct btrfs_root
*root
= wc
->replay_dest
;
1626 struct btrfs_key key
;
1631 btrfs_read_buffer(eb
, gen
);
1633 level
= btrfs_header_level(eb
);
1638 path
= btrfs_alloc_path();
1642 nritems
= btrfs_header_nritems(eb
);
1643 for (i
= 0; i
< nritems
; i
++) {
1644 btrfs_item_key_to_cpu(eb
, &key
, i
);
1646 /* inode keys are done during the first stage */
1647 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
1648 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
1649 struct btrfs_inode_item
*inode_item
;
1652 inode_item
= btrfs_item_ptr(eb
, i
,
1653 struct btrfs_inode_item
);
1654 mode
= btrfs_inode_mode(eb
, inode_item
);
1655 if (S_ISDIR(mode
)) {
1656 ret
= replay_dir_deletes(wc
->trans
,
1657 root
, log
, path
, key
.objectid
, 0);
1660 ret
= overwrite_item(wc
->trans
, root
, path
,
1664 /* for regular files, make sure corresponding
1665 * orhpan item exist. extents past the new EOF
1666 * will be truncated later by orphan cleanup.
1668 if (S_ISREG(mode
)) {
1669 ret
= insert_orphan_item(wc
->trans
, root
,
1674 ret
= link_to_fixup_dir(wc
->trans
, root
,
1675 path
, key
.objectid
);
1678 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
1681 /* these keys are simply copied */
1682 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
1683 ret
= overwrite_item(wc
->trans
, root
, path
,
1686 } else if (key
.type
== BTRFS_INODE_REF_KEY
) {
1687 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
1689 BUG_ON(ret
&& ret
!= -ENOENT
);
1690 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
1691 ret
= replay_one_extent(wc
->trans
, root
, path
,
1694 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
||
1695 key
.type
== BTRFS_DIR_INDEX_KEY
) {
1696 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
1701 btrfs_free_path(path
);
1705 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
1706 struct btrfs_root
*root
,
1707 struct btrfs_path
*path
, int *level
,
1708 struct walk_control
*wc
)
1713 struct extent_buffer
*next
;
1714 struct extent_buffer
*cur
;
1715 struct extent_buffer
*parent
;
1719 WARN_ON(*level
< 0);
1720 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1722 while (*level
> 0) {
1723 WARN_ON(*level
< 0);
1724 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1725 cur
= path
->nodes
[*level
];
1727 if (btrfs_header_level(cur
) != *level
)
1730 if (path
->slots
[*level
] >=
1731 btrfs_header_nritems(cur
))
1734 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
1735 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
1736 blocksize
= btrfs_level_size(root
, *level
- 1);
1738 parent
= path
->nodes
[*level
];
1739 root_owner
= btrfs_header_owner(parent
);
1741 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1746 ret
= wc
->process_func(root
, next
, wc
, ptr_gen
);
1750 path
->slots
[*level
]++;
1752 btrfs_read_buffer(next
, ptr_gen
);
1754 btrfs_tree_lock(next
);
1755 btrfs_set_lock_blocking(next
);
1756 clean_tree_block(trans
, root
, next
);
1757 btrfs_wait_tree_block_writeback(next
);
1758 btrfs_tree_unlock(next
);
1760 WARN_ON(root_owner
!=
1761 BTRFS_TREE_LOG_OBJECTID
);
1762 ret
= btrfs_free_and_pin_reserved_extent(root
,
1764 BUG_ON(ret
); /* -ENOMEM or logic errors */
1766 free_extent_buffer(next
);
1769 btrfs_read_buffer(next
, ptr_gen
);
1771 WARN_ON(*level
<= 0);
1772 if (path
->nodes
[*level
-1])
1773 free_extent_buffer(path
->nodes
[*level
-1]);
1774 path
->nodes
[*level
-1] = next
;
1775 *level
= btrfs_header_level(next
);
1776 path
->slots
[*level
] = 0;
1779 WARN_ON(*level
< 0);
1780 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1782 path
->slots
[*level
] = btrfs_header_nritems(path
->nodes
[*level
]);
1788 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
1789 struct btrfs_root
*root
,
1790 struct btrfs_path
*path
, int *level
,
1791 struct walk_control
*wc
)
1798 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
1799 slot
= path
->slots
[i
];
1800 if (slot
+ 1 < btrfs_header_nritems(path
->nodes
[i
])) {
1803 WARN_ON(*level
== 0);
1806 struct extent_buffer
*parent
;
1807 if (path
->nodes
[*level
] == root
->node
)
1808 parent
= path
->nodes
[*level
];
1810 parent
= path
->nodes
[*level
+ 1];
1812 root_owner
= btrfs_header_owner(parent
);
1813 ret
= wc
->process_func(root
, path
->nodes
[*level
], wc
,
1814 btrfs_header_generation(path
->nodes
[*level
]));
1819 struct extent_buffer
*next
;
1821 next
= path
->nodes
[*level
];
1823 btrfs_tree_lock(next
);
1824 btrfs_set_lock_blocking(next
);
1825 clean_tree_block(trans
, root
, next
);
1826 btrfs_wait_tree_block_writeback(next
);
1827 btrfs_tree_unlock(next
);
1829 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1830 ret
= btrfs_free_and_pin_reserved_extent(root
,
1831 path
->nodes
[*level
]->start
,
1832 path
->nodes
[*level
]->len
);
1835 free_extent_buffer(path
->nodes
[*level
]);
1836 path
->nodes
[*level
] = NULL
;
1844 * drop the reference count on the tree rooted at 'snap'. This traverses
1845 * the tree freeing any blocks that have a ref count of zero after being
1848 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
1849 struct btrfs_root
*log
, struct walk_control
*wc
)
1854 struct btrfs_path
*path
;
1858 path
= btrfs_alloc_path();
1862 level
= btrfs_header_level(log
->node
);
1864 path
->nodes
[level
] = log
->node
;
1865 extent_buffer_get(log
->node
);
1866 path
->slots
[level
] = 0;
1869 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
1877 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
1886 /* was the root node processed? if not, catch it here */
1887 if (path
->nodes
[orig_level
]) {
1888 ret
= wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
1889 btrfs_header_generation(path
->nodes
[orig_level
]));
1893 struct extent_buffer
*next
;
1895 next
= path
->nodes
[orig_level
];
1897 btrfs_tree_lock(next
);
1898 btrfs_set_lock_blocking(next
);
1899 clean_tree_block(trans
, log
, next
);
1900 btrfs_wait_tree_block_writeback(next
);
1901 btrfs_tree_unlock(next
);
1903 WARN_ON(log
->root_key
.objectid
!=
1904 BTRFS_TREE_LOG_OBJECTID
);
1905 ret
= btrfs_free_and_pin_reserved_extent(log
, next
->start
,
1907 BUG_ON(ret
); /* -ENOMEM or logic errors */
1912 for (i
= 0; i
<= orig_level
; i
++) {
1913 if (path
->nodes
[i
]) {
1914 free_extent_buffer(path
->nodes
[i
]);
1915 path
->nodes
[i
] = NULL
;
1918 btrfs_free_path(path
);
1923 * helper function to update the item for a given subvolumes log root
1924 * in the tree of log roots
1926 static int update_log_root(struct btrfs_trans_handle
*trans
,
1927 struct btrfs_root
*log
)
1931 if (log
->log_transid
== 1) {
1932 /* insert root item on the first sync */
1933 ret
= btrfs_insert_root(trans
, log
->fs_info
->log_root_tree
,
1934 &log
->root_key
, &log
->root_item
);
1936 ret
= btrfs_update_root(trans
, log
->fs_info
->log_root_tree
,
1937 &log
->root_key
, &log
->root_item
);
1942 static int wait_log_commit(struct btrfs_trans_handle
*trans
,
1943 struct btrfs_root
*root
, unsigned long transid
)
1946 int index
= transid
% 2;
1949 * we only allow two pending log transactions at a time,
1950 * so we know that if ours is more than 2 older than the
1951 * current transaction, we're done
1954 prepare_to_wait(&root
->log_commit_wait
[index
],
1955 &wait
, TASK_UNINTERRUPTIBLE
);
1956 mutex_unlock(&root
->log_mutex
);
1958 if (root
->fs_info
->last_trans_log_full_commit
!=
1959 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
1960 atomic_read(&root
->log_commit
[index
]))
1963 finish_wait(&root
->log_commit_wait
[index
], &wait
);
1964 mutex_lock(&root
->log_mutex
);
1965 } while (root
->fs_info
->last_trans_log_full_commit
!=
1966 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
1967 atomic_read(&root
->log_commit
[index
]));
1971 static void wait_for_writer(struct btrfs_trans_handle
*trans
,
1972 struct btrfs_root
*root
)
1975 while (root
->fs_info
->last_trans_log_full_commit
!=
1976 trans
->transid
&& atomic_read(&root
->log_writers
)) {
1977 prepare_to_wait(&root
->log_writer_wait
,
1978 &wait
, TASK_UNINTERRUPTIBLE
);
1979 mutex_unlock(&root
->log_mutex
);
1980 if (root
->fs_info
->last_trans_log_full_commit
!=
1981 trans
->transid
&& atomic_read(&root
->log_writers
))
1983 mutex_lock(&root
->log_mutex
);
1984 finish_wait(&root
->log_writer_wait
, &wait
);
1989 * btrfs_sync_log does sends a given tree log down to the disk and
1990 * updates the super blocks to record it. When this call is done,
1991 * you know that any inodes previously logged are safely on disk only
1994 * Any other return value means you need to call btrfs_commit_transaction.
1995 * Some of the edge cases for fsyncing directories that have had unlinks
1996 * or renames done in the past mean that sometimes the only safe
1997 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1998 * that has happened.
2000 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
2001 struct btrfs_root
*root
)
2007 struct btrfs_root
*log
= root
->log_root
;
2008 struct btrfs_root
*log_root_tree
= root
->fs_info
->log_root_tree
;
2009 unsigned long log_transid
= 0;
2011 mutex_lock(&root
->log_mutex
);
2012 index1
= root
->log_transid
% 2;
2013 if (atomic_read(&root
->log_commit
[index1
])) {
2014 wait_log_commit(trans
, root
, root
->log_transid
);
2015 mutex_unlock(&root
->log_mutex
);
2018 atomic_set(&root
->log_commit
[index1
], 1);
2020 /* wait for previous tree log sync to complete */
2021 if (atomic_read(&root
->log_commit
[(index1
+ 1) % 2]))
2022 wait_log_commit(trans
, root
, root
->log_transid
- 1);
2024 unsigned long batch
= root
->log_batch
;
2025 /* when we're on an ssd, just kick the log commit out */
2026 if (!btrfs_test_opt(root
, SSD
) && root
->log_multiple_pids
) {
2027 mutex_unlock(&root
->log_mutex
);
2028 schedule_timeout_uninterruptible(1);
2029 mutex_lock(&root
->log_mutex
);
2031 wait_for_writer(trans
, root
);
2032 if (batch
== root
->log_batch
)
2036 /* bail out if we need to do a full commit */
2037 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2039 mutex_unlock(&root
->log_mutex
);
2043 log_transid
= root
->log_transid
;
2044 if (log_transid
% 2 == 0)
2045 mark
= EXTENT_DIRTY
;
2049 /* we start IO on all the marked extents here, but we don't actually
2050 * wait for them until later.
2052 ret
= btrfs_write_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2054 btrfs_abort_transaction(trans
, root
, ret
);
2055 mutex_unlock(&root
->log_mutex
);
2059 btrfs_set_root_node(&log
->root_item
, log
->node
);
2061 root
->log_batch
= 0;
2062 root
->log_transid
++;
2063 log
->log_transid
= root
->log_transid
;
2064 root
->log_start_pid
= 0;
2067 * IO has been started, blocks of the log tree have WRITTEN flag set
2068 * in their headers. new modifications of the log will be written to
2069 * new positions. so it's safe to allow log writers to go in.
2071 mutex_unlock(&root
->log_mutex
);
2073 mutex_lock(&log_root_tree
->log_mutex
);
2074 log_root_tree
->log_batch
++;
2075 atomic_inc(&log_root_tree
->log_writers
);
2076 mutex_unlock(&log_root_tree
->log_mutex
);
2078 ret
= update_log_root(trans
, log
);
2080 mutex_lock(&log_root_tree
->log_mutex
);
2081 if (atomic_dec_and_test(&log_root_tree
->log_writers
)) {
2083 if (waitqueue_active(&log_root_tree
->log_writer_wait
))
2084 wake_up(&log_root_tree
->log_writer_wait
);
2088 if (ret
!= -ENOSPC
) {
2089 btrfs_abort_transaction(trans
, root
, ret
);
2090 mutex_unlock(&log_root_tree
->log_mutex
);
2093 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2094 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2095 mutex_unlock(&log_root_tree
->log_mutex
);
2100 index2
= log_root_tree
->log_transid
% 2;
2101 if (atomic_read(&log_root_tree
->log_commit
[index2
])) {
2102 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2103 wait_log_commit(trans
, log_root_tree
,
2104 log_root_tree
->log_transid
);
2105 mutex_unlock(&log_root_tree
->log_mutex
);
2109 atomic_set(&log_root_tree
->log_commit
[index2
], 1);
2111 if (atomic_read(&log_root_tree
->log_commit
[(index2
+ 1) % 2])) {
2112 wait_log_commit(trans
, log_root_tree
,
2113 log_root_tree
->log_transid
- 1);
2116 wait_for_writer(trans
, log_root_tree
);
2119 * now that we've moved on to the tree of log tree roots,
2120 * check the full commit flag again
2122 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2123 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2124 mutex_unlock(&log_root_tree
->log_mutex
);
2126 goto out_wake_log_root
;
2129 ret
= btrfs_write_and_wait_marked_extents(log_root_tree
,
2130 &log_root_tree
->dirty_log_pages
,
2131 EXTENT_DIRTY
| EXTENT_NEW
);
2133 btrfs_abort_transaction(trans
, root
, ret
);
2134 mutex_unlock(&log_root_tree
->log_mutex
);
2135 goto out_wake_log_root
;
2137 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2139 btrfs_set_super_log_root(root
->fs_info
->super_for_commit
,
2140 log_root_tree
->node
->start
);
2141 btrfs_set_super_log_root_level(root
->fs_info
->super_for_commit
,
2142 btrfs_header_level(log_root_tree
->node
));
2144 log_root_tree
->log_batch
= 0;
2145 log_root_tree
->log_transid
++;
2148 mutex_unlock(&log_root_tree
->log_mutex
);
2151 * nobody else is going to jump in and write the the ctree
2152 * super here because the log_commit atomic below is protecting
2153 * us. We must be called with a transaction handle pinning
2154 * the running transaction open, so a full commit can't hop
2155 * in and cause problems either.
2157 btrfs_scrub_pause_super(root
);
2158 write_ctree_super(trans
, root
->fs_info
->tree_root
, 1);
2159 btrfs_scrub_continue_super(root
);
2162 mutex_lock(&root
->log_mutex
);
2163 if (root
->last_log_commit
< log_transid
)
2164 root
->last_log_commit
= log_transid
;
2165 mutex_unlock(&root
->log_mutex
);
2168 atomic_set(&log_root_tree
->log_commit
[index2
], 0);
2170 if (waitqueue_active(&log_root_tree
->log_commit_wait
[index2
]))
2171 wake_up(&log_root_tree
->log_commit_wait
[index2
]);
2173 atomic_set(&root
->log_commit
[index1
], 0);
2175 if (waitqueue_active(&root
->log_commit_wait
[index1
]))
2176 wake_up(&root
->log_commit_wait
[index1
]);
2180 static void free_log_tree(struct btrfs_trans_handle
*trans
,
2181 struct btrfs_root
*log
)
2186 struct walk_control wc
= {
2188 .process_func
= process_one_buffer
2191 ret
= walk_log_tree(trans
, log
, &wc
);
2195 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
2196 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
);
2200 clear_extent_bits(&log
->dirty_log_pages
, start
, end
,
2201 EXTENT_DIRTY
| EXTENT_NEW
, GFP_NOFS
);
2204 free_extent_buffer(log
->node
);
2209 * free all the extents used by the tree log. This should be called
2210 * at commit time of the full transaction
2212 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2214 if (root
->log_root
) {
2215 free_log_tree(trans
, root
->log_root
);
2216 root
->log_root
= NULL
;
2221 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
2222 struct btrfs_fs_info
*fs_info
)
2224 if (fs_info
->log_root_tree
) {
2225 free_log_tree(trans
, fs_info
->log_root_tree
);
2226 fs_info
->log_root_tree
= NULL
;
2232 * If both a file and directory are logged, and unlinks or renames are
2233 * mixed in, we have a few interesting corners:
2235 * create file X in dir Y
2236 * link file X to X.link in dir Y
2238 * unlink file X but leave X.link
2241 * After a crash we would expect only X.link to exist. But file X
2242 * didn't get fsync'd again so the log has back refs for X and X.link.
2244 * We solve this by removing directory entries and inode backrefs from the
2245 * log when a file that was logged in the current transaction is
2246 * unlinked. Any later fsync will include the updated log entries, and
2247 * we'll be able to reconstruct the proper directory items from backrefs.
2249 * This optimizations allows us to avoid relogging the entire inode
2250 * or the entire directory.
2252 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
2253 struct btrfs_root
*root
,
2254 const char *name
, int name_len
,
2255 struct inode
*dir
, u64 index
)
2257 struct btrfs_root
*log
;
2258 struct btrfs_dir_item
*di
;
2259 struct btrfs_path
*path
;
2263 u64 dir_ino
= btrfs_ino(dir
);
2265 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
2268 ret
= join_running_log_trans(root
);
2272 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2274 log
= root
->log_root
;
2275 path
= btrfs_alloc_path();
2281 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir_ino
,
2282 name
, name_len
, -1);
2288 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2289 bytes_del
+= name_len
;
2292 btrfs_release_path(path
);
2293 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir_ino
,
2294 index
, name
, name_len
, -1);
2300 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2301 bytes_del
+= name_len
;
2305 /* update the directory size in the log to reflect the names
2309 struct btrfs_key key
;
2311 key
.objectid
= dir_ino
;
2313 key
.type
= BTRFS_INODE_ITEM_KEY
;
2314 btrfs_release_path(path
);
2316 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2322 struct btrfs_inode_item
*item
;
2325 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2326 struct btrfs_inode_item
);
2327 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2328 if (i_size
> bytes_del
)
2329 i_size
-= bytes_del
;
2332 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2333 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2336 btrfs_release_path(path
);
2339 btrfs_free_path(path
);
2341 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2342 if (ret
== -ENOSPC
) {
2343 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2346 btrfs_abort_transaction(trans
, root
, ret
);
2348 btrfs_end_log_trans(root
);
2353 /* see comments for btrfs_del_dir_entries_in_log */
2354 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2355 struct btrfs_root
*root
,
2356 const char *name
, int name_len
,
2357 struct inode
*inode
, u64 dirid
)
2359 struct btrfs_root
*log
;
2363 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
2366 ret
= join_running_log_trans(root
);
2369 log
= root
->log_root
;
2370 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2372 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, btrfs_ino(inode
),
2374 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2375 if (ret
== -ENOSPC
) {
2376 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2378 } else if (ret
< 0 && ret
!= -ENOENT
)
2379 btrfs_abort_transaction(trans
, root
, ret
);
2380 btrfs_end_log_trans(root
);
2386 * creates a range item in the log for 'dirid'. first_offset and
2387 * last_offset tell us which parts of the key space the log should
2388 * be considered authoritative for.
2390 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2391 struct btrfs_root
*log
,
2392 struct btrfs_path
*path
,
2393 int key_type
, u64 dirid
,
2394 u64 first_offset
, u64 last_offset
)
2397 struct btrfs_key key
;
2398 struct btrfs_dir_log_item
*item
;
2400 key
.objectid
= dirid
;
2401 key
.offset
= first_offset
;
2402 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2403 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2405 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2406 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2410 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2411 struct btrfs_dir_log_item
);
2412 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2413 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2414 btrfs_release_path(path
);
2419 * log all the items included in the current transaction for a given
2420 * directory. This also creates the range items in the log tree required
2421 * to replay anything deleted before the fsync
2423 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2424 struct btrfs_root
*root
, struct inode
*inode
,
2425 struct btrfs_path
*path
,
2426 struct btrfs_path
*dst_path
, int key_type
,
2427 u64 min_offset
, u64
*last_offset_ret
)
2429 struct btrfs_key min_key
;
2430 struct btrfs_key max_key
;
2431 struct btrfs_root
*log
= root
->log_root
;
2432 struct extent_buffer
*src
;
2437 u64 first_offset
= min_offset
;
2438 u64 last_offset
= (u64
)-1;
2439 u64 ino
= btrfs_ino(inode
);
2441 log
= root
->log_root
;
2442 max_key
.objectid
= ino
;
2443 max_key
.offset
= (u64
)-1;
2444 max_key
.type
= key_type
;
2446 min_key
.objectid
= ino
;
2447 min_key
.type
= key_type
;
2448 min_key
.offset
= min_offset
;
2450 path
->keep_locks
= 1;
2452 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2453 path
, 0, trans
->transid
);
2456 * we didn't find anything from this transaction, see if there
2457 * is anything at all
2459 if (ret
!= 0 || min_key
.objectid
!= ino
|| min_key
.type
!= key_type
) {
2460 min_key
.objectid
= ino
;
2461 min_key
.type
= key_type
;
2462 min_key
.offset
= (u64
)-1;
2463 btrfs_release_path(path
);
2464 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2466 btrfs_release_path(path
);
2469 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2471 /* if ret == 0 there are items for this type,
2472 * create a range to tell us the last key of this type.
2473 * otherwise, there are no items in this directory after
2474 * *min_offset, and we create a range to indicate that.
2477 struct btrfs_key tmp
;
2478 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2480 if (key_type
== tmp
.type
)
2481 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2486 /* go backward to find any previous key */
2487 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2489 struct btrfs_key tmp
;
2490 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2491 if (key_type
== tmp
.type
) {
2492 first_offset
= tmp
.offset
;
2493 ret
= overwrite_item(trans
, log
, dst_path
,
2494 path
->nodes
[0], path
->slots
[0],
2502 btrfs_release_path(path
);
2504 /* find the first key from this transaction again */
2505 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2512 * we have a block from this transaction, log every item in it
2513 * from our directory
2516 struct btrfs_key tmp
;
2517 src
= path
->nodes
[0];
2518 nritems
= btrfs_header_nritems(src
);
2519 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2520 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2522 if (min_key
.objectid
!= ino
|| min_key
.type
!= key_type
)
2524 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2531 path
->slots
[0] = nritems
;
2534 * look ahead to the next item and see if it is also
2535 * from this directory and from this transaction
2537 ret
= btrfs_next_leaf(root
, path
);
2539 last_offset
= (u64
)-1;
2542 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2543 if (tmp
.objectid
!= ino
|| tmp
.type
!= key_type
) {
2544 last_offset
= (u64
)-1;
2547 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2548 ret
= overwrite_item(trans
, log
, dst_path
,
2549 path
->nodes
[0], path
->slots
[0],
2554 last_offset
= tmp
.offset
;
2559 btrfs_release_path(path
);
2560 btrfs_release_path(dst_path
);
2563 *last_offset_ret
= last_offset
;
2565 * insert the log range keys to indicate where the log
2568 ret
= insert_dir_log_key(trans
, log
, path
, key_type
,
2569 ino
, first_offset
, last_offset
);
2577 * logging directories is very similar to logging inodes, We find all the items
2578 * from the current transaction and write them to the log.
2580 * The recovery code scans the directory in the subvolume, and if it finds a
2581 * key in the range logged that is not present in the log tree, then it means
2582 * that dir entry was unlinked during the transaction.
2584 * In order for that scan to work, we must include one key smaller than
2585 * the smallest logged by this transaction and one key larger than the largest
2586 * key logged by this transaction.
2588 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
2589 struct btrfs_root
*root
, struct inode
*inode
,
2590 struct btrfs_path
*path
,
2591 struct btrfs_path
*dst_path
)
2596 int key_type
= BTRFS_DIR_ITEM_KEY
;
2602 ret
= log_dir_items(trans
, root
, inode
, path
,
2603 dst_path
, key_type
, min_key
,
2607 if (max_key
== (u64
)-1)
2609 min_key
= max_key
+ 1;
2612 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
2613 key_type
= BTRFS_DIR_INDEX_KEY
;
2620 * a helper function to drop items from the log before we relog an
2621 * inode. max_key_type indicates the highest item type to remove.
2622 * This cannot be run for file data extents because it does not
2623 * free the extents they point to.
2625 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
2626 struct btrfs_root
*log
,
2627 struct btrfs_path
*path
,
2628 u64 objectid
, int max_key_type
)
2631 struct btrfs_key key
;
2632 struct btrfs_key found_key
;
2634 key
.objectid
= objectid
;
2635 key
.type
= max_key_type
;
2636 key
.offset
= (u64
)-1;
2639 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
2644 if (path
->slots
[0] == 0)
2648 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2651 if (found_key
.objectid
!= objectid
)
2654 ret
= btrfs_del_item(trans
, log
, path
);
2657 btrfs_release_path(path
);
2659 btrfs_release_path(path
);
2663 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
2664 struct btrfs_root
*log
,
2665 struct btrfs_path
*dst_path
,
2666 struct extent_buffer
*src
,
2667 int start_slot
, int nr
, int inode_only
)
2669 unsigned long src_offset
;
2670 unsigned long dst_offset
;
2671 struct btrfs_file_extent_item
*extent
;
2672 struct btrfs_inode_item
*inode_item
;
2674 struct btrfs_key
*ins_keys
;
2678 struct list_head ordered_sums
;
2680 INIT_LIST_HEAD(&ordered_sums
);
2682 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
2683 nr
* sizeof(u32
), GFP_NOFS
);
2687 ins_sizes
= (u32
*)ins_data
;
2688 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
2690 for (i
= 0; i
< nr
; i
++) {
2691 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
2692 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
2694 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
2695 ins_keys
, ins_sizes
, nr
);
2701 for (i
= 0; i
< nr
; i
++, dst_path
->slots
[0]++) {
2702 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
2703 dst_path
->slots
[0]);
2705 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
2707 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
2708 src_offset
, ins_sizes
[i
]);
2710 if (inode_only
== LOG_INODE_EXISTS
&&
2711 ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
2712 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
2714 struct btrfs_inode_item
);
2715 btrfs_set_inode_size(dst_path
->nodes
[0], inode_item
, 0);
2717 /* set the generation to zero so the recover code
2718 * can tell the difference between an logging
2719 * just to say 'this inode exists' and a logging
2720 * to say 'update this inode with these values'
2722 btrfs_set_inode_generation(dst_path
->nodes
[0],
2725 /* take a reference on file data extents so that truncates
2726 * or deletes of this inode don't have to relog the inode
2729 if (btrfs_key_type(ins_keys
+ i
) == BTRFS_EXTENT_DATA_KEY
) {
2731 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
2732 struct btrfs_file_extent_item
);
2734 if (btrfs_file_extent_generation(src
, extent
) < trans
->transid
)
2737 found_type
= btrfs_file_extent_type(src
, extent
);
2738 if (found_type
== BTRFS_FILE_EXTENT_REG
||
2739 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
2741 ds
= btrfs_file_extent_disk_bytenr(src
,
2743 /* ds == 0 is a hole */
2747 dl
= btrfs_file_extent_disk_num_bytes(src
,
2749 cs
= btrfs_file_extent_offset(src
, extent
);
2750 cl
= btrfs_file_extent_num_bytes(src
,
2752 if (btrfs_file_extent_compression(src
,
2758 ret
= btrfs_lookup_csums_range(
2759 log
->fs_info
->csum_root
,
2760 ds
+ cs
, ds
+ cs
+ cl
- 1,
2767 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
2768 btrfs_release_path(dst_path
);
2772 * we have to do this after the loop above to avoid changing the
2773 * log tree while trying to change the log tree.
2776 while (!list_empty(&ordered_sums
)) {
2777 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
2778 struct btrfs_ordered_sum
,
2781 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
2782 list_del(&sums
->list
);
2788 /* log a single inode in the tree log.
2789 * At least one parent directory for this inode must exist in the tree
2790 * or be logged already.
2792 * Any items from this inode changed by the current transaction are copied
2793 * to the log tree. An extra reference is taken on any extents in this
2794 * file, allowing us to avoid a whole pile of corner cases around logging
2795 * blocks that have been removed from the tree.
2797 * See LOG_INODE_ALL and related defines for a description of what inode_only
2800 * This handles both files and directories.
2802 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2803 struct btrfs_root
*root
, struct inode
*inode
,
2806 struct btrfs_path
*path
;
2807 struct btrfs_path
*dst_path
;
2808 struct btrfs_key min_key
;
2809 struct btrfs_key max_key
;
2810 struct btrfs_root
*log
= root
->log_root
;
2811 struct extent_buffer
*src
= NULL
;
2815 int ins_start_slot
= 0;
2817 u64 ino
= btrfs_ino(inode
);
2819 log
= root
->log_root
;
2821 path
= btrfs_alloc_path();
2824 dst_path
= btrfs_alloc_path();
2826 btrfs_free_path(path
);
2830 min_key
.objectid
= ino
;
2831 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
2834 max_key
.objectid
= ino
;
2836 /* today the code can only do partial logging of directories */
2837 if (!S_ISDIR(inode
->i_mode
))
2838 inode_only
= LOG_INODE_ALL
;
2840 if (inode_only
== LOG_INODE_EXISTS
|| S_ISDIR(inode
->i_mode
))
2841 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
2843 max_key
.type
= (u8
)-1;
2844 max_key
.offset
= (u64
)-1;
2846 ret
= btrfs_commit_inode_delayed_items(trans
, inode
);
2848 btrfs_free_path(path
);
2849 btrfs_free_path(dst_path
);
2853 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2856 * a brute force approach to making sure we get the most uptodate
2857 * copies of everything.
2859 if (S_ISDIR(inode
->i_mode
)) {
2860 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
2862 if (inode_only
== LOG_INODE_EXISTS
)
2863 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
2864 ret
= drop_objectid_items(trans
, log
, path
, ino
, max_key_type
);
2866 ret
= btrfs_truncate_inode_items(trans
, log
, inode
, 0, 0);
2872 path
->keep_locks
= 1;
2876 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2877 path
, 0, trans
->transid
);
2881 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2882 if (min_key
.objectid
!= ino
)
2884 if (min_key
.type
> max_key
.type
)
2887 src
= path
->nodes
[0];
2888 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
2891 } else if (!ins_nr
) {
2892 ins_start_slot
= path
->slots
[0];
2897 ret
= copy_items(trans
, log
, dst_path
, src
, ins_start_slot
,
2898 ins_nr
, inode_only
);
2904 ins_start_slot
= path
->slots
[0];
2907 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2909 if (path
->slots
[0] < nritems
) {
2910 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
2915 ret
= copy_items(trans
, log
, dst_path
, src
,
2917 ins_nr
, inode_only
);
2924 btrfs_release_path(path
);
2926 if (min_key
.offset
< (u64
)-1)
2928 else if (min_key
.type
< (u8
)-1)
2930 else if (min_key
.objectid
< (u64
)-1)
2936 ret
= copy_items(trans
, log
, dst_path
, src
,
2938 ins_nr
, inode_only
);
2946 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
2947 btrfs_release_path(path
);
2948 btrfs_release_path(dst_path
);
2949 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
2955 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
2957 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2959 btrfs_free_path(path
);
2960 btrfs_free_path(dst_path
);
2965 * follow the dentry parent pointers up the chain and see if any
2966 * of the directories in it require a full commit before they can
2967 * be logged. Returns zero if nothing special needs to be done or 1 if
2968 * a full commit is required.
2970 static noinline
int check_parent_dirs_for_sync(struct btrfs_trans_handle
*trans
,
2971 struct inode
*inode
,
2972 struct dentry
*parent
,
2973 struct super_block
*sb
,
2977 struct btrfs_root
*root
;
2978 struct dentry
*old_parent
= NULL
;
2981 * for regular files, if its inode is already on disk, we don't
2982 * have to worry about the parents at all. This is because
2983 * we can use the last_unlink_trans field to record renames
2984 * and other fun in this file.
2986 if (S_ISREG(inode
->i_mode
) &&
2987 BTRFS_I(inode
)->generation
<= last_committed
&&
2988 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
)
2991 if (!S_ISDIR(inode
->i_mode
)) {
2992 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
2994 inode
= parent
->d_inode
;
2998 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
3001 if (BTRFS_I(inode
)->last_unlink_trans
> last_committed
) {
3002 root
= BTRFS_I(inode
)->root
;
3005 * make sure any commits to the log are forced
3006 * to be full commits
3008 root
->fs_info
->last_trans_log_full_commit
=
3014 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3017 if (IS_ROOT(parent
))
3020 parent
= dget_parent(parent
);
3022 old_parent
= parent
;
3023 inode
= parent
->d_inode
;
3031 static int inode_in_log(struct btrfs_trans_handle
*trans
,
3032 struct inode
*inode
)
3034 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3037 mutex_lock(&root
->log_mutex
);
3038 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
&&
3039 BTRFS_I(inode
)->last_sub_trans
<= root
->last_log_commit
)
3041 mutex_unlock(&root
->log_mutex
);
3047 * helper function around btrfs_log_inode to make sure newly created
3048 * parent directories also end up in the log. A minimal inode and backref
3049 * only logging is done of any parent directories that are older than
3050 * the last committed transaction
3052 int btrfs_log_inode_parent(struct btrfs_trans_handle
*trans
,
3053 struct btrfs_root
*root
, struct inode
*inode
,
3054 struct dentry
*parent
, int exists_only
)
3056 int inode_only
= exists_only
? LOG_INODE_EXISTS
: LOG_INODE_ALL
;
3057 struct super_block
*sb
;
3058 struct dentry
*old_parent
= NULL
;
3060 u64 last_committed
= root
->fs_info
->last_trans_committed
;
3064 if (btrfs_test_opt(root
, NOTREELOG
)) {
3069 if (root
->fs_info
->last_trans_log_full_commit
>
3070 root
->fs_info
->last_trans_committed
) {
3075 if (root
!= BTRFS_I(inode
)->root
||
3076 btrfs_root_refs(&root
->root_item
) == 0) {
3081 ret
= check_parent_dirs_for_sync(trans
, inode
, parent
,
3082 sb
, last_committed
);
3086 if (inode_in_log(trans
, inode
)) {
3087 ret
= BTRFS_NO_LOG_SYNC
;
3091 ret
= start_log_trans(trans
, root
);
3095 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3100 * for regular files, if its inode is already on disk, we don't
3101 * have to worry about the parents at all. This is because
3102 * we can use the last_unlink_trans field to record renames
3103 * and other fun in this file.
3105 if (S_ISREG(inode
->i_mode
) &&
3106 BTRFS_I(inode
)->generation
<= last_committed
&&
3107 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
) {
3112 inode_only
= LOG_INODE_EXISTS
;
3114 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3117 inode
= parent
->d_inode
;
3118 if (root
!= BTRFS_I(inode
)->root
)
3121 if (BTRFS_I(inode
)->generation
>
3122 root
->fs_info
->last_trans_committed
) {
3123 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3127 if (IS_ROOT(parent
))
3130 parent
= dget_parent(parent
);
3132 old_parent
= parent
;
3138 BUG_ON(ret
!= -ENOSPC
);
3139 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
3142 btrfs_end_log_trans(root
);
3148 * it is not safe to log dentry if the chunk root has added new
3149 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3150 * If this returns 1, you must commit the transaction to safely get your
3153 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
3154 struct btrfs_root
*root
, struct dentry
*dentry
)
3156 struct dentry
*parent
= dget_parent(dentry
);
3159 ret
= btrfs_log_inode_parent(trans
, root
, dentry
->d_inode
, parent
, 0);
3166 * should be called during mount to recover any replay any log trees
3169 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
3172 struct btrfs_path
*path
;
3173 struct btrfs_trans_handle
*trans
;
3174 struct btrfs_key key
;
3175 struct btrfs_key found_key
;
3176 struct btrfs_key tmp_key
;
3177 struct btrfs_root
*log
;
3178 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
3179 struct walk_control wc
= {
3180 .process_func
= process_one_buffer
,
3184 path
= btrfs_alloc_path();
3188 fs_info
->log_root_recovering
= 1;
3190 trans
= btrfs_start_transaction(fs_info
->tree_root
, 0);
3191 if (IS_ERR(trans
)) {
3192 ret
= PTR_ERR(trans
);
3199 ret
= walk_log_tree(trans
, log_root_tree
, &wc
);
3201 btrfs_error(fs_info
, ret
, "Failed to pin buffers while "
3202 "recovering log root tree.");
3207 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
3208 key
.offset
= (u64
)-1;
3209 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
3212 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
3215 btrfs_error(fs_info
, ret
,
3216 "Couldn't find tree log root.");
3220 if (path
->slots
[0] == 0)
3224 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
3226 btrfs_release_path(path
);
3227 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
3230 log
= btrfs_read_fs_root_no_radix(log_root_tree
,
3234 btrfs_error(fs_info
, ret
,
3235 "Couldn't read tree log root.");
3239 tmp_key
.objectid
= found_key
.offset
;
3240 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3241 tmp_key
.offset
= (u64
)-1;
3243 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
3244 if (IS_ERR(wc
.replay_dest
)) {
3245 ret
= PTR_ERR(wc
.replay_dest
);
3246 btrfs_error(fs_info
, ret
, "Couldn't read target root "
3247 "for tree log recovery.");
3251 wc
.replay_dest
->log_root
= log
;
3252 btrfs_record_root_in_trans(trans
, wc
.replay_dest
);
3253 ret
= walk_log_tree(trans
, log
, &wc
);
3256 if (wc
.stage
== LOG_WALK_REPLAY_ALL
) {
3257 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
3262 key
.offset
= found_key
.offset
- 1;
3263 wc
.replay_dest
->log_root
= NULL
;
3264 free_extent_buffer(log
->node
);
3265 free_extent_buffer(log
->commit_root
);
3268 if (found_key
.offset
== 0)
3271 btrfs_release_path(path
);
3273 /* step one is to pin it all, step two is to replay just inodes */
3276 wc
.process_func
= replay_one_buffer
;
3277 wc
.stage
= LOG_WALK_REPLAY_INODES
;
3280 /* step three is to replay everything */
3281 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
3286 btrfs_free_path(path
);
3288 free_extent_buffer(log_root_tree
->node
);
3289 log_root_tree
->log_root
= NULL
;
3290 fs_info
->log_root_recovering
= 0;
3292 /* step 4: commit the transaction, which also unpins the blocks */
3293 btrfs_commit_transaction(trans
, fs_info
->tree_root
);
3295 kfree(log_root_tree
);
3299 btrfs_free_path(path
);
3304 * there are some corner cases where we want to force a full
3305 * commit instead of allowing a directory to be logged.
3307 * They revolve around files there were unlinked from the directory, and
3308 * this function updates the parent directory so that a full commit is
3309 * properly done if it is fsync'd later after the unlinks are done.
3311 void btrfs_record_unlink_dir(struct btrfs_trans_handle
*trans
,
3312 struct inode
*dir
, struct inode
*inode
,
3316 * when we're logging a file, if it hasn't been renamed
3317 * or unlinked, and its inode is fully committed on disk,
3318 * we don't have to worry about walking up the directory chain
3319 * to log its parents.
3321 * So, we use the last_unlink_trans field to put this transid
3322 * into the file. When the file is logged we check it and
3323 * don't log the parents if the file is fully on disk.
3325 if (S_ISREG(inode
->i_mode
))
3326 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
3329 * if this directory was already logged any new
3330 * names for this file/dir will get recorded
3333 if (BTRFS_I(dir
)->logged_trans
== trans
->transid
)
3337 * if the inode we're about to unlink was logged,
3338 * the log will be properly updated for any new names
3340 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
)
3344 * when renaming files across directories, if the directory
3345 * there we're unlinking from gets fsync'd later on, there's
3346 * no way to find the destination directory later and fsync it
3347 * properly. So, we have to be conservative and force commits
3348 * so the new name gets discovered.
3353 /* we can safely do the unlink without any special recording */
3357 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
3361 * Call this after adding a new name for a file and it will properly
3362 * update the log to reflect the new name.
3364 * It will return zero if all goes well, and it will return 1 if a
3365 * full transaction commit is required.
3367 int btrfs_log_new_name(struct btrfs_trans_handle
*trans
,
3368 struct inode
*inode
, struct inode
*old_dir
,
3369 struct dentry
*parent
)
3371 struct btrfs_root
* root
= BTRFS_I(inode
)->root
;
3374 * this will force the logging code to walk the dentry chain
3377 if (S_ISREG(inode
->i_mode
))
3378 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
3381 * if this inode hasn't been logged and directory we're renaming it
3382 * from hasn't been logged, we don't need to log it
3384 if (BTRFS_I(inode
)->logged_trans
<=
3385 root
->fs_info
->last_trans_committed
&&
3386 (!old_dir
|| BTRFS_I(old_dir
)->logged_trans
<=
3387 root
->fs_info
->last_trans_committed
))
3390 return btrfs_log_inode_parent(trans
, root
, inode
, parent
, 1);