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
);
694 btrfs_run_delayed_items(trans
, root
);
699 * helper function to see if a given name and sequence number found
700 * in an inode back reference are already in a directory and correctly
701 * point to this inode
703 static noinline
int inode_in_dir(struct btrfs_root
*root
,
704 struct btrfs_path
*path
,
705 u64 dirid
, u64 objectid
, u64 index
,
706 const char *name
, int name_len
)
708 struct btrfs_dir_item
*di
;
709 struct btrfs_key location
;
712 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
713 index
, name
, name_len
, 0);
714 if (di
&& !IS_ERR(di
)) {
715 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
716 if (location
.objectid
!= objectid
)
720 btrfs_release_path(path
);
722 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
723 if (di
&& !IS_ERR(di
)) {
724 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
725 if (location
.objectid
!= objectid
)
731 btrfs_release_path(path
);
736 * helper function to check a log tree for a named back reference in
737 * an inode. This is used to decide if a back reference that is
738 * found in the subvolume conflicts with what we find in the log.
740 * inode backreferences may have multiple refs in a single item,
741 * during replay we process one reference at a time, and we don't
742 * want to delete valid links to a file from the subvolume if that
743 * link is also in the log.
745 static noinline
int backref_in_log(struct btrfs_root
*log
,
746 struct btrfs_key
*key
,
747 char *name
, int namelen
)
749 struct btrfs_path
*path
;
750 struct btrfs_inode_ref
*ref
;
752 unsigned long ptr_end
;
753 unsigned long name_ptr
;
759 path
= btrfs_alloc_path();
763 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
767 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
768 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
769 ptr_end
= ptr
+ item_size
;
770 while (ptr
< ptr_end
) {
771 ref
= (struct btrfs_inode_ref
*)ptr
;
772 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
773 if (found_name_len
== namelen
) {
774 name_ptr
= (unsigned long)(ref
+ 1);
775 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
782 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
785 btrfs_free_path(path
);
791 * replay one inode back reference item found in the log tree.
792 * eb, slot and key refer to the buffer and key found in the log tree.
793 * root is the destination we are replaying into, and path is for temp
794 * use by this function. (it should be released on return).
796 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
797 struct btrfs_root
*root
,
798 struct btrfs_root
*log
,
799 struct btrfs_path
*path
,
800 struct extent_buffer
*eb
, int slot
,
801 struct btrfs_key
*key
)
803 struct btrfs_inode_ref
*ref
;
804 struct btrfs_dir_item
*di
;
807 unsigned long ref_ptr
;
808 unsigned long ref_end
;
815 * it is possible that we didn't log all the parent directories
816 * for a given inode. If we don't find the dir, just don't
817 * copy the back ref in. The link count fixup code will take
820 dir
= read_one_inode(root
, key
->offset
);
824 inode
= read_one_inode(root
, key
->objectid
);
830 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
831 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
834 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
836 namelen
= btrfs_inode_ref_name_len(eb
, ref
);
837 name
= kmalloc(namelen
, GFP_NOFS
);
840 read_extent_buffer(eb
, name
, (unsigned long)(ref
+ 1), namelen
);
842 /* if we already have a perfect match, we're done */
843 if (inode_in_dir(root
, path
, btrfs_ino(dir
), btrfs_ino(inode
),
844 btrfs_inode_ref_index(eb
, ref
),
850 * look for a conflicting back reference in the metadata.
851 * if we find one we have to unlink that name of the file
852 * before we add our new link. Later on, we overwrite any
853 * existing back reference, and we don't want to create
854 * dangling pointers in the directory.
860 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
864 struct btrfs_inode_ref
*victim_ref
;
866 unsigned long ptr_end
;
867 struct extent_buffer
*leaf
= path
->nodes
[0];
869 /* are we trying to overwrite a back ref for the root directory
870 * if so, just jump out, we're done
872 if (key
->objectid
== key
->offset
)
875 /* check all the names in this back reference to see
876 * if they are in the log. if so, we allow them to stay
877 * otherwise they must be unlinked as a conflict
879 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
880 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
881 while (ptr
< ptr_end
) {
882 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
883 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
885 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
886 BUG_ON(!victim_name
);
888 read_extent_buffer(leaf
, victim_name
,
889 (unsigned long)(victim_ref
+ 1),
892 if (!backref_in_log(log
, key
, victim_name
,
894 btrfs_inc_nlink(inode
);
895 btrfs_release_path(path
);
897 ret
= btrfs_unlink_inode(trans
, root
, dir
,
900 btrfs_run_delayed_items(trans
, root
);
903 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
908 * NOTE: we have searched root tree and checked the
909 * coresponding ref, it does not need to check again.
913 btrfs_release_path(path
);
915 /* look for a conflicting sequence number */
916 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, btrfs_ino(dir
),
917 btrfs_inode_ref_index(eb
, ref
),
919 if (di
&& !IS_ERR(di
)) {
920 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
923 btrfs_release_path(path
);
925 /* look for a conflicing name */
926 di
= btrfs_lookup_dir_item(trans
, root
, path
, btrfs_ino(dir
),
928 if (di
&& !IS_ERR(di
)) {
929 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
932 btrfs_release_path(path
);
935 /* insert our name */
936 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
, 0,
937 btrfs_inode_ref_index(eb
, ref
));
940 btrfs_update_inode(trans
, root
, inode
);
943 ref_ptr
= (unsigned long)(ref
+ 1) + namelen
;
945 if (ref_ptr
< ref_end
)
948 /* finally write the back reference in the inode */
949 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
953 btrfs_release_path(path
);
959 static int insert_orphan_item(struct btrfs_trans_handle
*trans
,
960 struct btrfs_root
*root
, u64 offset
)
963 ret
= btrfs_find_orphan_item(root
, offset
);
965 ret
= btrfs_insert_orphan_item(trans
, root
, offset
);
971 * There are a few corners where the link count of the file can't
972 * be properly maintained during replay. So, instead of adding
973 * lots of complexity to the log code, we just scan the backrefs
974 * for any file that has been through replay.
976 * The scan will update the link count on the inode to reflect the
977 * number of back refs found. If it goes down to zero, the iput
978 * will free the inode.
980 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
981 struct btrfs_root
*root
,
984 struct btrfs_path
*path
;
986 struct btrfs_key key
;
989 unsigned long ptr_end
;
991 u64 ino
= btrfs_ino(inode
);
994 key
.type
= BTRFS_INODE_REF_KEY
;
995 key
.offset
= (u64
)-1;
997 path
= btrfs_alloc_path();
1002 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1006 if (path
->slots
[0] == 0)
1010 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1012 if (key
.objectid
!= ino
||
1013 key
.type
!= BTRFS_INODE_REF_KEY
)
1015 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1016 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1018 while (ptr
< ptr_end
) {
1019 struct btrfs_inode_ref
*ref
;
1021 ref
= (struct btrfs_inode_ref
*)ptr
;
1022 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1024 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1028 if (key
.offset
== 0)
1031 btrfs_release_path(path
);
1033 btrfs_release_path(path
);
1034 if (nlink
!= inode
->i_nlink
) {
1035 set_nlink(inode
, nlink
);
1036 btrfs_update_inode(trans
, root
, inode
);
1038 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1040 if (inode
->i_nlink
== 0) {
1041 if (S_ISDIR(inode
->i_mode
)) {
1042 ret
= replay_dir_deletes(trans
, root
, NULL
, path
,
1046 ret
= insert_orphan_item(trans
, root
, ino
);
1049 btrfs_free_path(path
);
1054 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1055 struct btrfs_root
*root
,
1056 struct btrfs_path
*path
)
1059 struct btrfs_key key
;
1060 struct inode
*inode
;
1062 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1063 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1064 key
.offset
= (u64
)-1;
1066 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1071 if (path
->slots
[0] == 0)
1076 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1077 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1078 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1081 ret
= btrfs_del_item(trans
, root
, path
);
1085 btrfs_release_path(path
);
1086 inode
= read_one_inode(root
, key
.offset
);
1090 ret
= fixup_inode_link_count(trans
, root
, inode
);
1096 * fixup on a directory may create new entries,
1097 * make sure we always look for the highset possible
1100 key
.offset
= (u64
)-1;
1104 btrfs_release_path(path
);
1110 * record a given inode in the fixup dir so we can check its link
1111 * count when replay is done. The link count is incremented here
1112 * so the inode won't go away until we check it
1114 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1115 struct btrfs_root
*root
,
1116 struct btrfs_path
*path
,
1119 struct btrfs_key key
;
1121 struct inode
*inode
;
1123 inode
= read_one_inode(root
, objectid
);
1127 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1128 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1129 key
.offset
= objectid
;
1131 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1133 btrfs_release_path(path
);
1135 btrfs_inc_nlink(inode
);
1136 btrfs_update_inode(trans
, root
, inode
);
1137 } else if (ret
== -EEXIST
) {
1148 * when replaying the log for a directory, we only insert names
1149 * for inodes that actually exist. This means an fsync on a directory
1150 * does not implicitly fsync all the new files in it
1152 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1153 struct btrfs_root
*root
,
1154 struct btrfs_path
*path
,
1155 u64 dirid
, u64 index
,
1156 char *name
, int name_len
, u8 type
,
1157 struct btrfs_key
*location
)
1159 struct inode
*inode
;
1163 inode
= read_one_inode(root
, location
->objectid
);
1167 dir
= read_one_inode(root
, dirid
);
1172 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1174 /* FIXME, put inode into FIXUP list */
1182 * take a single entry in a log directory item and replay it into
1185 * if a conflicting item exists in the subdirectory already,
1186 * the inode it points to is unlinked and put into the link count
1189 * If a name from the log points to a file or directory that does
1190 * not exist in the FS, it is skipped. fsyncs on directories
1191 * do not force down inodes inside that directory, just changes to the
1192 * names or unlinks in a directory.
1194 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1195 struct btrfs_root
*root
,
1196 struct btrfs_path
*path
,
1197 struct extent_buffer
*eb
,
1198 struct btrfs_dir_item
*di
,
1199 struct btrfs_key
*key
)
1203 struct btrfs_dir_item
*dst_di
;
1204 struct btrfs_key found_key
;
1205 struct btrfs_key log_key
;
1211 dir
= read_one_inode(root
, key
->objectid
);
1215 name_len
= btrfs_dir_name_len(eb
, di
);
1216 name
= kmalloc(name_len
, GFP_NOFS
);
1220 log_type
= btrfs_dir_type(eb
, di
);
1221 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1224 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1225 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1230 btrfs_release_path(path
);
1232 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1233 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1235 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1236 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1243 if (IS_ERR_OR_NULL(dst_di
)) {
1244 /* we need a sequence number to insert, so we only
1245 * do inserts for the BTRFS_DIR_INDEX_KEY types
1247 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1252 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1253 /* the existing item matches the logged item */
1254 if (found_key
.objectid
== log_key
.objectid
&&
1255 found_key
.type
== log_key
.type
&&
1256 found_key
.offset
== log_key
.offset
&&
1257 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1262 * don't drop the conflicting directory entry if the inode
1263 * for the new entry doesn't exist
1268 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1271 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1274 btrfs_release_path(path
);
1280 btrfs_release_path(path
);
1281 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1282 name
, name_len
, log_type
, &log_key
);
1284 BUG_ON(ret
&& ret
!= -ENOENT
);
1289 * find all the names in a directory item and reconcile them into
1290 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1291 * one name in a directory item, but the same code gets used for
1292 * both directory index types
1294 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1295 struct btrfs_root
*root
,
1296 struct btrfs_path
*path
,
1297 struct extent_buffer
*eb
, int slot
,
1298 struct btrfs_key
*key
)
1301 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1302 struct btrfs_dir_item
*di
;
1305 unsigned long ptr_end
;
1307 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1308 ptr_end
= ptr
+ item_size
;
1309 while (ptr
< ptr_end
) {
1310 di
= (struct btrfs_dir_item
*)ptr
;
1311 if (verify_dir_item(root
, eb
, di
))
1313 name_len
= btrfs_dir_name_len(eb
, di
);
1314 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1316 ptr
= (unsigned long)(di
+ 1);
1323 * directory replay has two parts. There are the standard directory
1324 * items in the log copied from the subvolume, and range items
1325 * created in the log while the subvolume was logged.
1327 * The range items tell us which parts of the key space the log
1328 * is authoritative for. During replay, if a key in the subvolume
1329 * directory is in a logged range item, but not actually in the log
1330 * that means it was deleted from the directory before the fsync
1331 * and should be removed.
1333 static noinline
int find_dir_range(struct btrfs_root
*root
,
1334 struct btrfs_path
*path
,
1335 u64 dirid
, int key_type
,
1336 u64
*start_ret
, u64
*end_ret
)
1338 struct btrfs_key key
;
1340 struct btrfs_dir_log_item
*item
;
1344 if (*start_ret
== (u64
)-1)
1347 key
.objectid
= dirid
;
1348 key
.type
= key_type
;
1349 key
.offset
= *start_ret
;
1351 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1355 if (path
->slots
[0] == 0)
1360 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1362 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1366 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1367 struct btrfs_dir_log_item
);
1368 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1370 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1372 *start_ret
= key
.offset
;
1373 *end_ret
= found_end
;
1378 /* check the next slot in the tree to see if it is a valid item */
1379 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1380 if (path
->slots
[0] >= nritems
) {
1381 ret
= btrfs_next_leaf(root
, path
);
1388 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1390 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1394 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1395 struct btrfs_dir_log_item
);
1396 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1397 *start_ret
= key
.offset
;
1398 *end_ret
= found_end
;
1401 btrfs_release_path(path
);
1406 * this looks for a given directory item in the log. If the directory
1407 * item is not in the log, the item is removed and the inode it points
1410 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1411 struct btrfs_root
*root
,
1412 struct btrfs_root
*log
,
1413 struct btrfs_path
*path
,
1414 struct btrfs_path
*log_path
,
1416 struct btrfs_key
*dir_key
)
1419 struct extent_buffer
*eb
;
1422 struct btrfs_dir_item
*di
;
1423 struct btrfs_dir_item
*log_di
;
1426 unsigned long ptr_end
;
1428 struct inode
*inode
;
1429 struct btrfs_key location
;
1432 eb
= path
->nodes
[0];
1433 slot
= path
->slots
[0];
1434 item_size
= btrfs_item_size_nr(eb
, slot
);
1435 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1436 ptr_end
= ptr
+ item_size
;
1437 while (ptr
< ptr_end
) {
1438 di
= (struct btrfs_dir_item
*)ptr
;
1439 if (verify_dir_item(root
, eb
, di
)) {
1444 name_len
= btrfs_dir_name_len(eb
, di
);
1445 name
= kmalloc(name_len
, GFP_NOFS
);
1450 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1453 if (log
&& dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1454 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1457 } else if (log
&& dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1458 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1464 if (IS_ERR_OR_NULL(log_di
)) {
1465 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1466 btrfs_release_path(path
);
1467 btrfs_release_path(log_path
);
1468 inode
= read_one_inode(root
, location
.objectid
);
1474 ret
= link_to_fixup_dir(trans
, root
,
1475 path
, location
.objectid
);
1477 btrfs_inc_nlink(inode
);
1478 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1482 btrfs_run_delayed_items(trans
, root
);
1487 /* there might still be more names under this key
1488 * check and repeat if required
1490 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1497 btrfs_release_path(log_path
);
1500 ptr
= (unsigned long)(di
+ 1);
1505 btrfs_release_path(path
);
1506 btrfs_release_path(log_path
);
1511 * deletion replay happens before we copy any new directory items
1512 * out of the log or out of backreferences from inodes. It
1513 * scans the log to find ranges of keys that log is authoritative for,
1514 * and then scans the directory to find items in those ranges that are
1515 * not present in the log.
1517 * Anything we don't find in the log is unlinked and removed from the
1520 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1521 struct btrfs_root
*root
,
1522 struct btrfs_root
*log
,
1523 struct btrfs_path
*path
,
1524 u64 dirid
, int del_all
)
1528 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1530 struct btrfs_key dir_key
;
1531 struct btrfs_key found_key
;
1532 struct btrfs_path
*log_path
;
1535 dir_key
.objectid
= dirid
;
1536 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1537 log_path
= btrfs_alloc_path();
1541 dir
= read_one_inode(root
, dirid
);
1542 /* it isn't an error if the inode isn't there, that can happen
1543 * because we replay the deletes before we copy in the inode item
1547 btrfs_free_path(log_path
);
1555 range_end
= (u64
)-1;
1557 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1558 &range_start
, &range_end
);
1563 dir_key
.offset
= range_start
;
1566 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1571 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1572 if (path
->slots
[0] >= nritems
) {
1573 ret
= btrfs_next_leaf(root
, path
);
1577 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1579 if (found_key
.objectid
!= dirid
||
1580 found_key
.type
!= dir_key
.type
)
1583 if (found_key
.offset
> range_end
)
1586 ret
= check_item_in_log(trans
, root
, log
, path
,
1590 if (found_key
.offset
== (u64
)-1)
1592 dir_key
.offset
= found_key
.offset
+ 1;
1594 btrfs_release_path(path
);
1595 if (range_end
== (u64
)-1)
1597 range_start
= range_end
+ 1;
1602 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1603 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1604 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1605 btrfs_release_path(path
);
1609 btrfs_release_path(path
);
1610 btrfs_free_path(log_path
);
1616 * the process_func used to replay items from the log tree. This
1617 * gets called in two different stages. The first stage just looks
1618 * for inodes and makes sure they are all copied into the subvolume.
1620 * The second stage copies all the other item types from the log into
1621 * the subvolume. The two stage approach is slower, but gets rid of
1622 * lots of complexity around inodes referencing other inodes that exist
1623 * only in the log (references come from either directory items or inode
1626 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
1627 struct walk_control
*wc
, u64 gen
)
1630 struct btrfs_path
*path
;
1631 struct btrfs_root
*root
= wc
->replay_dest
;
1632 struct btrfs_key key
;
1637 ret
= btrfs_read_buffer(eb
, gen
);
1641 level
= btrfs_header_level(eb
);
1646 path
= btrfs_alloc_path();
1650 nritems
= btrfs_header_nritems(eb
);
1651 for (i
= 0; i
< nritems
; i
++) {
1652 btrfs_item_key_to_cpu(eb
, &key
, i
);
1654 /* inode keys are done during the first stage */
1655 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
1656 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
1657 struct btrfs_inode_item
*inode_item
;
1660 inode_item
= btrfs_item_ptr(eb
, i
,
1661 struct btrfs_inode_item
);
1662 mode
= btrfs_inode_mode(eb
, inode_item
);
1663 if (S_ISDIR(mode
)) {
1664 ret
= replay_dir_deletes(wc
->trans
,
1665 root
, log
, path
, key
.objectid
, 0);
1668 ret
= overwrite_item(wc
->trans
, root
, path
,
1672 /* for regular files, make sure corresponding
1673 * orhpan item exist. extents past the new EOF
1674 * will be truncated later by orphan cleanup.
1676 if (S_ISREG(mode
)) {
1677 ret
= insert_orphan_item(wc
->trans
, root
,
1682 ret
= link_to_fixup_dir(wc
->trans
, root
,
1683 path
, key
.objectid
);
1686 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
1689 /* these keys are simply copied */
1690 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
1691 ret
= overwrite_item(wc
->trans
, root
, path
,
1694 } else if (key
.type
== BTRFS_INODE_REF_KEY
) {
1695 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
1697 BUG_ON(ret
&& ret
!= -ENOENT
);
1698 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
1699 ret
= replay_one_extent(wc
->trans
, root
, path
,
1702 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
||
1703 key
.type
== BTRFS_DIR_INDEX_KEY
) {
1704 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
1709 btrfs_free_path(path
);
1713 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
1714 struct btrfs_root
*root
,
1715 struct btrfs_path
*path
, int *level
,
1716 struct walk_control
*wc
)
1721 struct extent_buffer
*next
;
1722 struct extent_buffer
*cur
;
1723 struct extent_buffer
*parent
;
1727 WARN_ON(*level
< 0);
1728 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1730 while (*level
> 0) {
1731 WARN_ON(*level
< 0);
1732 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1733 cur
= path
->nodes
[*level
];
1735 if (btrfs_header_level(cur
) != *level
)
1738 if (path
->slots
[*level
] >=
1739 btrfs_header_nritems(cur
))
1742 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
1743 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
1744 blocksize
= btrfs_level_size(root
, *level
- 1);
1746 parent
= path
->nodes
[*level
];
1747 root_owner
= btrfs_header_owner(parent
);
1749 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1754 ret
= wc
->process_func(root
, next
, wc
, ptr_gen
);
1758 path
->slots
[*level
]++;
1760 ret
= btrfs_read_buffer(next
, ptr_gen
);
1762 free_extent_buffer(next
);
1766 btrfs_tree_lock(next
);
1767 btrfs_set_lock_blocking(next
);
1768 clean_tree_block(trans
, root
, next
);
1769 btrfs_wait_tree_block_writeback(next
);
1770 btrfs_tree_unlock(next
);
1772 WARN_ON(root_owner
!=
1773 BTRFS_TREE_LOG_OBJECTID
);
1774 ret
= btrfs_free_and_pin_reserved_extent(root
,
1776 BUG_ON(ret
); /* -ENOMEM or logic errors */
1778 free_extent_buffer(next
);
1781 ret
= btrfs_read_buffer(next
, ptr_gen
);
1783 free_extent_buffer(next
);
1787 WARN_ON(*level
<= 0);
1788 if (path
->nodes
[*level
-1])
1789 free_extent_buffer(path
->nodes
[*level
-1]);
1790 path
->nodes
[*level
-1] = next
;
1791 *level
= btrfs_header_level(next
);
1792 path
->slots
[*level
] = 0;
1795 WARN_ON(*level
< 0);
1796 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
1798 path
->slots
[*level
] = btrfs_header_nritems(path
->nodes
[*level
]);
1804 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
1805 struct btrfs_root
*root
,
1806 struct btrfs_path
*path
, int *level
,
1807 struct walk_control
*wc
)
1814 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
1815 slot
= path
->slots
[i
];
1816 if (slot
+ 1 < btrfs_header_nritems(path
->nodes
[i
])) {
1819 WARN_ON(*level
== 0);
1822 struct extent_buffer
*parent
;
1823 if (path
->nodes
[*level
] == root
->node
)
1824 parent
= path
->nodes
[*level
];
1826 parent
= path
->nodes
[*level
+ 1];
1828 root_owner
= btrfs_header_owner(parent
);
1829 ret
= wc
->process_func(root
, path
->nodes
[*level
], wc
,
1830 btrfs_header_generation(path
->nodes
[*level
]));
1835 struct extent_buffer
*next
;
1837 next
= path
->nodes
[*level
];
1839 btrfs_tree_lock(next
);
1840 btrfs_set_lock_blocking(next
);
1841 clean_tree_block(trans
, root
, next
);
1842 btrfs_wait_tree_block_writeback(next
);
1843 btrfs_tree_unlock(next
);
1845 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
1846 ret
= btrfs_free_and_pin_reserved_extent(root
,
1847 path
->nodes
[*level
]->start
,
1848 path
->nodes
[*level
]->len
);
1851 free_extent_buffer(path
->nodes
[*level
]);
1852 path
->nodes
[*level
] = NULL
;
1860 * drop the reference count on the tree rooted at 'snap'. This traverses
1861 * the tree freeing any blocks that have a ref count of zero after being
1864 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
1865 struct btrfs_root
*log
, struct walk_control
*wc
)
1870 struct btrfs_path
*path
;
1874 path
= btrfs_alloc_path();
1878 level
= btrfs_header_level(log
->node
);
1880 path
->nodes
[level
] = log
->node
;
1881 extent_buffer_get(log
->node
);
1882 path
->slots
[level
] = 0;
1885 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
1893 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
1902 /* was the root node processed? if not, catch it here */
1903 if (path
->nodes
[orig_level
]) {
1904 ret
= wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
1905 btrfs_header_generation(path
->nodes
[orig_level
]));
1909 struct extent_buffer
*next
;
1911 next
= path
->nodes
[orig_level
];
1913 btrfs_tree_lock(next
);
1914 btrfs_set_lock_blocking(next
);
1915 clean_tree_block(trans
, log
, next
);
1916 btrfs_wait_tree_block_writeback(next
);
1917 btrfs_tree_unlock(next
);
1919 WARN_ON(log
->root_key
.objectid
!=
1920 BTRFS_TREE_LOG_OBJECTID
);
1921 ret
= btrfs_free_and_pin_reserved_extent(log
, next
->start
,
1923 BUG_ON(ret
); /* -ENOMEM or logic errors */
1928 for (i
= 0; i
<= orig_level
; i
++) {
1929 if (path
->nodes
[i
]) {
1930 free_extent_buffer(path
->nodes
[i
]);
1931 path
->nodes
[i
] = NULL
;
1934 btrfs_free_path(path
);
1939 * helper function to update the item for a given subvolumes log root
1940 * in the tree of log roots
1942 static int update_log_root(struct btrfs_trans_handle
*trans
,
1943 struct btrfs_root
*log
)
1947 if (log
->log_transid
== 1) {
1948 /* insert root item on the first sync */
1949 ret
= btrfs_insert_root(trans
, log
->fs_info
->log_root_tree
,
1950 &log
->root_key
, &log
->root_item
);
1952 ret
= btrfs_update_root(trans
, log
->fs_info
->log_root_tree
,
1953 &log
->root_key
, &log
->root_item
);
1958 static int wait_log_commit(struct btrfs_trans_handle
*trans
,
1959 struct btrfs_root
*root
, unsigned long transid
)
1962 int index
= transid
% 2;
1965 * we only allow two pending log transactions at a time,
1966 * so we know that if ours is more than 2 older than the
1967 * current transaction, we're done
1970 prepare_to_wait(&root
->log_commit_wait
[index
],
1971 &wait
, TASK_UNINTERRUPTIBLE
);
1972 mutex_unlock(&root
->log_mutex
);
1974 if (root
->fs_info
->last_trans_log_full_commit
!=
1975 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
1976 atomic_read(&root
->log_commit
[index
]))
1979 finish_wait(&root
->log_commit_wait
[index
], &wait
);
1980 mutex_lock(&root
->log_mutex
);
1981 } while (root
->fs_info
->last_trans_log_full_commit
!=
1982 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
1983 atomic_read(&root
->log_commit
[index
]));
1987 static void wait_for_writer(struct btrfs_trans_handle
*trans
,
1988 struct btrfs_root
*root
)
1991 while (root
->fs_info
->last_trans_log_full_commit
!=
1992 trans
->transid
&& atomic_read(&root
->log_writers
)) {
1993 prepare_to_wait(&root
->log_writer_wait
,
1994 &wait
, TASK_UNINTERRUPTIBLE
);
1995 mutex_unlock(&root
->log_mutex
);
1996 if (root
->fs_info
->last_trans_log_full_commit
!=
1997 trans
->transid
&& atomic_read(&root
->log_writers
))
1999 mutex_lock(&root
->log_mutex
);
2000 finish_wait(&root
->log_writer_wait
, &wait
);
2005 * btrfs_sync_log does sends a given tree log down to the disk and
2006 * updates the super blocks to record it. When this call is done,
2007 * you know that any inodes previously logged are safely on disk only
2010 * Any other return value means you need to call btrfs_commit_transaction.
2011 * Some of the edge cases for fsyncing directories that have had unlinks
2012 * or renames done in the past mean that sometimes the only safe
2013 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2014 * that has happened.
2016 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
2017 struct btrfs_root
*root
)
2023 struct btrfs_root
*log
= root
->log_root
;
2024 struct btrfs_root
*log_root_tree
= root
->fs_info
->log_root_tree
;
2025 unsigned long log_transid
= 0;
2027 mutex_lock(&root
->log_mutex
);
2028 index1
= root
->log_transid
% 2;
2029 if (atomic_read(&root
->log_commit
[index1
])) {
2030 wait_log_commit(trans
, root
, root
->log_transid
);
2031 mutex_unlock(&root
->log_mutex
);
2034 atomic_set(&root
->log_commit
[index1
], 1);
2036 /* wait for previous tree log sync to complete */
2037 if (atomic_read(&root
->log_commit
[(index1
+ 1) % 2]))
2038 wait_log_commit(trans
, root
, root
->log_transid
- 1);
2040 unsigned long batch
= root
->log_batch
;
2041 /* when we're on an ssd, just kick the log commit out */
2042 if (!btrfs_test_opt(root
, SSD
) && root
->log_multiple_pids
) {
2043 mutex_unlock(&root
->log_mutex
);
2044 schedule_timeout_uninterruptible(1);
2045 mutex_lock(&root
->log_mutex
);
2047 wait_for_writer(trans
, root
);
2048 if (batch
== root
->log_batch
)
2052 /* bail out if we need to do a full commit */
2053 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2055 mutex_unlock(&root
->log_mutex
);
2059 log_transid
= root
->log_transid
;
2060 if (log_transid
% 2 == 0)
2061 mark
= EXTENT_DIRTY
;
2065 /* we start IO on all the marked extents here, but we don't actually
2066 * wait for them until later.
2068 ret
= btrfs_write_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2070 btrfs_abort_transaction(trans
, root
, ret
);
2071 mutex_unlock(&root
->log_mutex
);
2075 btrfs_set_root_node(&log
->root_item
, log
->node
);
2077 root
->log_batch
= 0;
2078 root
->log_transid
++;
2079 log
->log_transid
= root
->log_transid
;
2080 root
->log_start_pid
= 0;
2083 * IO has been started, blocks of the log tree have WRITTEN flag set
2084 * in their headers. new modifications of the log will be written to
2085 * new positions. so it's safe to allow log writers to go in.
2087 mutex_unlock(&root
->log_mutex
);
2089 mutex_lock(&log_root_tree
->log_mutex
);
2090 log_root_tree
->log_batch
++;
2091 atomic_inc(&log_root_tree
->log_writers
);
2092 mutex_unlock(&log_root_tree
->log_mutex
);
2094 ret
= update_log_root(trans
, log
);
2096 mutex_lock(&log_root_tree
->log_mutex
);
2097 if (atomic_dec_and_test(&log_root_tree
->log_writers
)) {
2099 if (waitqueue_active(&log_root_tree
->log_writer_wait
))
2100 wake_up(&log_root_tree
->log_writer_wait
);
2104 if (ret
!= -ENOSPC
) {
2105 btrfs_abort_transaction(trans
, root
, ret
);
2106 mutex_unlock(&log_root_tree
->log_mutex
);
2109 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2110 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2111 mutex_unlock(&log_root_tree
->log_mutex
);
2116 index2
= log_root_tree
->log_transid
% 2;
2117 if (atomic_read(&log_root_tree
->log_commit
[index2
])) {
2118 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2119 wait_log_commit(trans
, log_root_tree
,
2120 log_root_tree
->log_transid
);
2121 mutex_unlock(&log_root_tree
->log_mutex
);
2125 atomic_set(&log_root_tree
->log_commit
[index2
], 1);
2127 if (atomic_read(&log_root_tree
->log_commit
[(index2
+ 1) % 2])) {
2128 wait_log_commit(trans
, log_root_tree
,
2129 log_root_tree
->log_transid
- 1);
2132 wait_for_writer(trans
, log_root_tree
);
2135 * now that we've moved on to the tree of log tree roots,
2136 * check the full commit flag again
2138 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2139 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2140 mutex_unlock(&log_root_tree
->log_mutex
);
2142 goto out_wake_log_root
;
2145 ret
= btrfs_write_and_wait_marked_extents(log_root_tree
,
2146 &log_root_tree
->dirty_log_pages
,
2147 EXTENT_DIRTY
| EXTENT_NEW
);
2149 btrfs_abort_transaction(trans
, root
, ret
);
2150 mutex_unlock(&log_root_tree
->log_mutex
);
2151 goto out_wake_log_root
;
2153 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2155 btrfs_set_super_log_root(root
->fs_info
->super_for_commit
,
2156 log_root_tree
->node
->start
);
2157 btrfs_set_super_log_root_level(root
->fs_info
->super_for_commit
,
2158 btrfs_header_level(log_root_tree
->node
));
2160 log_root_tree
->log_batch
= 0;
2161 log_root_tree
->log_transid
++;
2164 mutex_unlock(&log_root_tree
->log_mutex
);
2167 * nobody else is going to jump in and write the the ctree
2168 * super here because the log_commit atomic below is protecting
2169 * us. We must be called with a transaction handle pinning
2170 * the running transaction open, so a full commit can't hop
2171 * in and cause problems either.
2173 btrfs_scrub_pause_super(root
);
2174 write_ctree_super(trans
, root
->fs_info
->tree_root
, 1);
2175 btrfs_scrub_continue_super(root
);
2178 mutex_lock(&root
->log_mutex
);
2179 if (root
->last_log_commit
< log_transid
)
2180 root
->last_log_commit
= log_transid
;
2181 mutex_unlock(&root
->log_mutex
);
2184 atomic_set(&log_root_tree
->log_commit
[index2
], 0);
2186 if (waitqueue_active(&log_root_tree
->log_commit_wait
[index2
]))
2187 wake_up(&log_root_tree
->log_commit_wait
[index2
]);
2189 atomic_set(&root
->log_commit
[index1
], 0);
2191 if (waitqueue_active(&root
->log_commit_wait
[index1
]))
2192 wake_up(&root
->log_commit_wait
[index1
]);
2196 static void free_log_tree(struct btrfs_trans_handle
*trans
,
2197 struct btrfs_root
*log
)
2202 struct walk_control wc
= {
2204 .process_func
= process_one_buffer
2207 ret
= walk_log_tree(trans
, log
, &wc
);
2211 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
2212 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
);
2216 clear_extent_bits(&log
->dirty_log_pages
, start
, end
,
2217 EXTENT_DIRTY
| EXTENT_NEW
, GFP_NOFS
);
2220 free_extent_buffer(log
->node
);
2225 * free all the extents used by the tree log. This should be called
2226 * at commit time of the full transaction
2228 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2230 if (root
->log_root
) {
2231 free_log_tree(trans
, root
->log_root
);
2232 root
->log_root
= NULL
;
2237 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
2238 struct btrfs_fs_info
*fs_info
)
2240 if (fs_info
->log_root_tree
) {
2241 free_log_tree(trans
, fs_info
->log_root_tree
);
2242 fs_info
->log_root_tree
= NULL
;
2248 * If both a file and directory are logged, and unlinks or renames are
2249 * mixed in, we have a few interesting corners:
2251 * create file X in dir Y
2252 * link file X to X.link in dir Y
2254 * unlink file X but leave X.link
2257 * After a crash we would expect only X.link to exist. But file X
2258 * didn't get fsync'd again so the log has back refs for X and X.link.
2260 * We solve this by removing directory entries and inode backrefs from the
2261 * log when a file that was logged in the current transaction is
2262 * unlinked. Any later fsync will include the updated log entries, and
2263 * we'll be able to reconstruct the proper directory items from backrefs.
2265 * This optimizations allows us to avoid relogging the entire inode
2266 * or the entire directory.
2268 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
2269 struct btrfs_root
*root
,
2270 const char *name
, int name_len
,
2271 struct inode
*dir
, u64 index
)
2273 struct btrfs_root
*log
;
2274 struct btrfs_dir_item
*di
;
2275 struct btrfs_path
*path
;
2279 u64 dir_ino
= btrfs_ino(dir
);
2281 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
2284 ret
= join_running_log_trans(root
);
2288 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2290 log
= root
->log_root
;
2291 path
= btrfs_alloc_path();
2297 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir_ino
,
2298 name
, name_len
, -1);
2304 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2305 bytes_del
+= name_len
;
2308 btrfs_release_path(path
);
2309 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir_ino
,
2310 index
, name
, name_len
, -1);
2316 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2317 bytes_del
+= name_len
;
2321 /* update the directory size in the log to reflect the names
2325 struct btrfs_key key
;
2327 key
.objectid
= dir_ino
;
2329 key
.type
= BTRFS_INODE_ITEM_KEY
;
2330 btrfs_release_path(path
);
2332 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2338 struct btrfs_inode_item
*item
;
2341 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2342 struct btrfs_inode_item
);
2343 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2344 if (i_size
> bytes_del
)
2345 i_size
-= bytes_del
;
2348 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2349 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2352 btrfs_release_path(path
);
2355 btrfs_free_path(path
);
2357 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2358 if (ret
== -ENOSPC
) {
2359 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2362 btrfs_abort_transaction(trans
, root
, ret
);
2364 btrfs_end_log_trans(root
);
2369 /* see comments for btrfs_del_dir_entries_in_log */
2370 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2371 struct btrfs_root
*root
,
2372 const char *name
, int name_len
,
2373 struct inode
*inode
, u64 dirid
)
2375 struct btrfs_root
*log
;
2379 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
2382 ret
= join_running_log_trans(root
);
2385 log
= root
->log_root
;
2386 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2388 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, btrfs_ino(inode
),
2390 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2391 if (ret
== -ENOSPC
) {
2392 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2394 } else if (ret
< 0 && ret
!= -ENOENT
)
2395 btrfs_abort_transaction(trans
, root
, ret
);
2396 btrfs_end_log_trans(root
);
2402 * creates a range item in the log for 'dirid'. first_offset and
2403 * last_offset tell us which parts of the key space the log should
2404 * be considered authoritative for.
2406 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2407 struct btrfs_root
*log
,
2408 struct btrfs_path
*path
,
2409 int key_type
, u64 dirid
,
2410 u64 first_offset
, u64 last_offset
)
2413 struct btrfs_key key
;
2414 struct btrfs_dir_log_item
*item
;
2416 key
.objectid
= dirid
;
2417 key
.offset
= first_offset
;
2418 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2419 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2421 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2422 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2426 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2427 struct btrfs_dir_log_item
);
2428 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2429 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2430 btrfs_release_path(path
);
2435 * log all the items included in the current transaction for a given
2436 * directory. This also creates the range items in the log tree required
2437 * to replay anything deleted before the fsync
2439 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2440 struct btrfs_root
*root
, struct inode
*inode
,
2441 struct btrfs_path
*path
,
2442 struct btrfs_path
*dst_path
, int key_type
,
2443 u64 min_offset
, u64
*last_offset_ret
)
2445 struct btrfs_key min_key
;
2446 struct btrfs_key max_key
;
2447 struct btrfs_root
*log
= root
->log_root
;
2448 struct extent_buffer
*src
;
2453 u64 first_offset
= min_offset
;
2454 u64 last_offset
= (u64
)-1;
2455 u64 ino
= btrfs_ino(inode
);
2457 log
= root
->log_root
;
2458 max_key
.objectid
= ino
;
2459 max_key
.offset
= (u64
)-1;
2460 max_key
.type
= key_type
;
2462 min_key
.objectid
= ino
;
2463 min_key
.type
= key_type
;
2464 min_key
.offset
= min_offset
;
2466 path
->keep_locks
= 1;
2468 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2469 path
, 0, trans
->transid
);
2472 * we didn't find anything from this transaction, see if there
2473 * is anything at all
2475 if (ret
!= 0 || min_key
.objectid
!= ino
|| min_key
.type
!= key_type
) {
2476 min_key
.objectid
= ino
;
2477 min_key
.type
= key_type
;
2478 min_key
.offset
= (u64
)-1;
2479 btrfs_release_path(path
);
2480 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2482 btrfs_release_path(path
);
2485 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2487 /* if ret == 0 there are items for this type,
2488 * create a range to tell us the last key of this type.
2489 * otherwise, there are no items in this directory after
2490 * *min_offset, and we create a range to indicate that.
2493 struct btrfs_key tmp
;
2494 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2496 if (key_type
== tmp
.type
)
2497 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2502 /* go backward to find any previous key */
2503 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2505 struct btrfs_key tmp
;
2506 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2507 if (key_type
== tmp
.type
) {
2508 first_offset
= tmp
.offset
;
2509 ret
= overwrite_item(trans
, log
, dst_path
,
2510 path
->nodes
[0], path
->slots
[0],
2518 btrfs_release_path(path
);
2520 /* find the first key from this transaction again */
2521 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2528 * we have a block from this transaction, log every item in it
2529 * from our directory
2532 struct btrfs_key tmp
;
2533 src
= path
->nodes
[0];
2534 nritems
= btrfs_header_nritems(src
);
2535 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2536 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2538 if (min_key
.objectid
!= ino
|| min_key
.type
!= key_type
)
2540 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2547 path
->slots
[0] = nritems
;
2550 * look ahead to the next item and see if it is also
2551 * from this directory and from this transaction
2553 ret
= btrfs_next_leaf(root
, path
);
2555 last_offset
= (u64
)-1;
2558 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2559 if (tmp
.objectid
!= ino
|| tmp
.type
!= key_type
) {
2560 last_offset
= (u64
)-1;
2563 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2564 ret
= overwrite_item(trans
, log
, dst_path
,
2565 path
->nodes
[0], path
->slots
[0],
2570 last_offset
= tmp
.offset
;
2575 btrfs_release_path(path
);
2576 btrfs_release_path(dst_path
);
2579 *last_offset_ret
= last_offset
;
2581 * insert the log range keys to indicate where the log
2584 ret
= insert_dir_log_key(trans
, log
, path
, key_type
,
2585 ino
, first_offset
, last_offset
);
2593 * logging directories is very similar to logging inodes, We find all the items
2594 * from the current transaction and write them to the log.
2596 * The recovery code scans the directory in the subvolume, and if it finds a
2597 * key in the range logged that is not present in the log tree, then it means
2598 * that dir entry was unlinked during the transaction.
2600 * In order for that scan to work, we must include one key smaller than
2601 * the smallest logged by this transaction and one key larger than the largest
2602 * key logged by this transaction.
2604 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
2605 struct btrfs_root
*root
, struct inode
*inode
,
2606 struct btrfs_path
*path
,
2607 struct btrfs_path
*dst_path
)
2612 int key_type
= BTRFS_DIR_ITEM_KEY
;
2618 ret
= log_dir_items(trans
, root
, inode
, path
,
2619 dst_path
, key_type
, min_key
,
2623 if (max_key
== (u64
)-1)
2625 min_key
= max_key
+ 1;
2628 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
2629 key_type
= BTRFS_DIR_INDEX_KEY
;
2636 * a helper function to drop items from the log before we relog an
2637 * inode. max_key_type indicates the highest item type to remove.
2638 * This cannot be run for file data extents because it does not
2639 * free the extents they point to.
2641 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
2642 struct btrfs_root
*log
,
2643 struct btrfs_path
*path
,
2644 u64 objectid
, int max_key_type
)
2647 struct btrfs_key key
;
2648 struct btrfs_key found_key
;
2650 key
.objectid
= objectid
;
2651 key
.type
= max_key_type
;
2652 key
.offset
= (u64
)-1;
2655 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
2660 if (path
->slots
[0] == 0)
2664 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2667 if (found_key
.objectid
!= objectid
)
2670 ret
= btrfs_del_item(trans
, log
, path
);
2673 btrfs_release_path(path
);
2675 btrfs_release_path(path
);
2681 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
2682 struct btrfs_root
*log
,
2683 struct btrfs_path
*dst_path
,
2684 struct extent_buffer
*src
,
2685 int start_slot
, int nr
, int inode_only
)
2687 unsigned long src_offset
;
2688 unsigned long dst_offset
;
2689 struct btrfs_file_extent_item
*extent
;
2690 struct btrfs_inode_item
*inode_item
;
2692 struct btrfs_key
*ins_keys
;
2696 struct list_head ordered_sums
;
2698 INIT_LIST_HEAD(&ordered_sums
);
2700 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
2701 nr
* sizeof(u32
), GFP_NOFS
);
2705 ins_sizes
= (u32
*)ins_data
;
2706 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
2708 for (i
= 0; i
< nr
; i
++) {
2709 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
2710 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
2712 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
2713 ins_keys
, ins_sizes
, nr
);
2719 for (i
= 0; i
< nr
; i
++, dst_path
->slots
[0]++) {
2720 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
2721 dst_path
->slots
[0]);
2723 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
2725 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
2726 src_offset
, ins_sizes
[i
]);
2728 if (inode_only
== LOG_INODE_EXISTS
&&
2729 ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
2730 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
2732 struct btrfs_inode_item
);
2733 btrfs_set_inode_size(dst_path
->nodes
[0], inode_item
, 0);
2735 /* set the generation to zero so the recover code
2736 * can tell the difference between an logging
2737 * just to say 'this inode exists' and a logging
2738 * to say 'update this inode with these values'
2740 btrfs_set_inode_generation(dst_path
->nodes
[0],
2743 /* take a reference on file data extents so that truncates
2744 * or deletes of this inode don't have to relog the inode
2747 if (btrfs_key_type(ins_keys
+ i
) == BTRFS_EXTENT_DATA_KEY
) {
2749 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
2750 struct btrfs_file_extent_item
);
2752 if (btrfs_file_extent_generation(src
, extent
) < trans
->transid
)
2755 found_type
= btrfs_file_extent_type(src
, extent
);
2756 if (found_type
== BTRFS_FILE_EXTENT_REG
||
2757 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
2759 ds
= btrfs_file_extent_disk_bytenr(src
,
2761 /* ds == 0 is a hole */
2765 dl
= btrfs_file_extent_disk_num_bytes(src
,
2767 cs
= btrfs_file_extent_offset(src
, extent
);
2768 cl
= btrfs_file_extent_num_bytes(src
,
2770 if (btrfs_file_extent_compression(src
,
2776 ret
= btrfs_lookup_csums_range(
2777 log
->fs_info
->csum_root
,
2778 ds
+ cs
, ds
+ cs
+ cl
- 1,
2785 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
2786 btrfs_release_path(dst_path
);
2790 * we have to do this after the loop above to avoid changing the
2791 * log tree while trying to change the log tree.
2794 while (!list_empty(&ordered_sums
)) {
2795 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
2796 struct btrfs_ordered_sum
,
2799 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
2800 list_del(&sums
->list
);
2806 /* log a single inode in the tree log.
2807 * At least one parent directory for this inode must exist in the tree
2808 * or be logged already.
2810 * Any items from this inode changed by the current transaction are copied
2811 * to the log tree. An extra reference is taken on any extents in this
2812 * file, allowing us to avoid a whole pile of corner cases around logging
2813 * blocks that have been removed from the tree.
2815 * See LOG_INODE_ALL and related defines for a description of what inode_only
2818 * This handles both files and directories.
2820 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
2821 struct btrfs_root
*root
, struct inode
*inode
,
2824 struct btrfs_path
*path
;
2825 struct btrfs_path
*dst_path
;
2826 struct btrfs_key min_key
;
2827 struct btrfs_key max_key
;
2828 struct btrfs_root
*log
= root
->log_root
;
2829 struct extent_buffer
*src
= NULL
;
2833 int ins_start_slot
= 0;
2835 u64 ino
= btrfs_ino(inode
);
2837 log
= root
->log_root
;
2839 path
= btrfs_alloc_path();
2842 dst_path
= btrfs_alloc_path();
2844 btrfs_free_path(path
);
2848 min_key
.objectid
= ino
;
2849 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
2852 max_key
.objectid
= ino
;
2854 /* today the code can only do partial logging of directories */
2855 if (!S_ISDIR(inode
->i_mode
))
2856 inode_only
= LOG_INODE_ALL
;
2858 if (inode_only
== LOG_INODE_EXISTS
|| S_ISDIR(inode
->i_mode
))
2859 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
2861 max_key
.type
= (u8
)-1;
2862 max_key
.offset
= (u64
)-1;
2864 ret
= btrfs_commit_inode_delayed_items(trans
, inode
);
2866 btrfs_free_path(path
);
2867 btrfs_free_path(dst_path
);
2871 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2874 * a brute force approach to making sure we get the most uptodate
2875 * copies of everything.
2877 if (S_ISDIR(inode
->i_mode
)) {
2878 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
2880 if (inode_only
== LOG_INODE_EXISTS
)
2881 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
2882 ret
= drop_objectid_items(trans
, log
, path
, ino
, max_key_type
);
2884 ret
= btrfs_truncate_inode_items(trans
, log
, inode
, 0, 0);
2890 path
->keep_locks
= 1;
2894 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2895 path
, 0, trans
->transid
);
2899 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2900 if (min_key
.objectid
!= ino
)
2902 if (min_key
.type
> max_key
.type
)
2905 src
= path
->nodes
[0];
2906 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
2909 } else if (!ins_nr
) {
2910 ins_start_slot
= path
->slots
[0];
2915 ret
= copy_items(trans
, log
, dst_path
, src
, ins_start_slot
,
2916 ins_nr
, inode_only
);
2922 ins_start_slot
= path
->slots
[0];
2925 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2927 if (path
->slots
[0] < nritems
) {
2928 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
2933 ret
= copy_items(trans
, log
, dst_path
, src
,
2935 ins_nr
, inode_only
);
2942 btrfs_release_path(path
);
2944 if (min_key
.offset
< (u64
)-1)
2946 else if (min_key
.type
< (u8
)-1)
2948 else if (min_key
.objectid
< (u64
)-1)
2954 ret
= copy_items(trans
, log
, dst_path
, src
,
2956 ins_nr
, inode_only
);
2964 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
2965 btrfs_release_path(path
);
2966 btrfs_release_path(dst_path
);
2967 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
2973 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
2975 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2977 btrfs_free_path(path
);
2978 btrfs_free_path(dst_path
);
2983 * follow the dentry parent pointers up the chain and see if any
2984 * of the directories in it require a full commit before they can
2985 * be logged. Returns zero if nothing special needs to be done or 1 if
2986 * a full commit is required.
2988 static noinline
int check_parent_dirs_for_sync(struct btrfs_trans_handle
*trans
,
2989 struct inode
*inode
,
2990 struct dentry
*parent
,
2991 struct super_block
*sb
,
2995 struct btrfs_root
*root
;
2996 struct dentry
*old_parent
= NULL
;
2999 * for regular files, if its inode is already on disk, we don't
3000 * have to worry about the parents at all. This is because
3001 * we can use the last_unlink_trans field to record renames
3002 * and other fun in this file.
3004 if (S_ISREG(inode
->i_mode
) &&
3005 BTRFS_I(inode
)->generation
<= last_committed
&&
3006 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
)
3009 if (!S_ISDIR(inode
->i_mode
)) {
3010 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3012 inode
= parent
->d_inode
;
3016 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
3019 if (BTRFS_I(inode
)->last_unlink_trans
> last_committed
) {
3020 root
= BTRFS_I(inode
)->root
;
3023 * make sure any commits to the log are forced
3024 * to be full commits
3026 root
->fs_info
->last_trans_log_full_commit
=
3032 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3035 if (IS_ROOT(parent
))
3038 parent
= dget_parent(parent
);
3040 old_parent
= parent
;
3041 inode
= parent
->d_inode
;
3050 * helper function around btrfs_log_inode to make sure newly created
3051 * parent directories also end up in the log. A minimal inode and backref
3052 * only logging is done of any parent directories that are older than
3053 * the last committed transaction
3055 int btrfs_log_inode_parent(struct btrfs_trans_handle
*trans
,
3056 struct btrfs_root
*root
, struct inode
*inode
,
3057 struct dentry
*parent
, int exists_only
)
3059 int inode_only
= exists_only
? LOG_INODE_EXISTS
: LOG_INODE_ALL
;
3060 struct super_block
*sb
;
3061 struct dentry
*old_parent
= NULL
;
3063 u64 last_committed
= root
->fs_info
->last_trans_committed
;
3067 if (btrfs_test_opt(root
, NOTREELOG
)) {
3072 if (root
->fs_info
->last_trans_log_full_commit
>
3073 root
->fs_info
->last_trans_committed
) {
3078 if (root
!= BTRFS_I(inode
)->root
||
3079 btrfs_root_refs(&root
->root_item
) == 0) {
3084 ret
= check_parent_dirs_for_sync(trans
, inode
, parent
,
3085 sb
, last_committed
);
3089 if (btrfs_inode_in_log(inode
, trans
->transid
)) {
3090 ret
= BTRFS_NO_LOG_SYNC
;
3094 ret
= start_log_trans(trans
, root
);
3098 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3103 * for regular files, if its inode is already on disk, we don't
3104 * have to worry about the parents at all. This is because
3105 * we can use the last_unlink_trans field to record renames
3106 * and other fun in this file.
3108 if (S_ISREG(inode
->i_mode
) &&
3109 BTRFS_I(inode
)->generation
<= last_committed
&&
3110 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
) {
3115 inode_only
= LOG_INODE_EXISTS
;
3117 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3120 inode
= parent
->d_inode
;
3121 if (root
!= BTRFS_I(inode
)->root
)
3124 if (BTRFS_I(inode
)->generation
>
3125 root
->fs_info
->last_trans_committed
) {
3126 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3130 if (IS_ROOT(parent
))
3133 parent
= dget_parent(parent
);
3135 old_parent
= parent
;
3141 BUG_ON(ret
!= -ENOSPC
);
3142 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
3145 btrfs_end_log_trans(root
);
3151 * it is not safe to log dentry if the chunk root has added new
3152 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3153 * If this returns 1, you must commit the transaction to safely get your
3156 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
3157 struct btrfs_root
*root
, struct dentry
*dentry
)
3159 struct dentry
*parent
= dget_parent(dentry
);
3162 ret
= btrfs_log_inode_parent(trans
, root
, dentry
->d_inode
, parent
, 0);
3169 * should be called during mount to recover any replay any log trees
3172 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
3175 struct btrfs_path
*path
;
3176 struct btrfs_trans_handle
*trans
;
3177 struct btrfs_key key
;
3178 struct btrfs_key found_key
;
3179 struct btrfs_key tmp_key
;
3180 struct btrfs_root
*log
;
3181 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
3182 struct walk_control wc
= {
3183 .process_func
= process_one_buffer
,
3187 path
= btrfs_alloc_path();
3191 fs_info
->log_root_recovering
= 1;
3193 trans
= btrfs_start_transaction(fs_info
->tree_root
, 0);
3194 if (IS_ERR(trans
)) {
3195 ret
= PTR_ERR(trans
);
3202 ret
= walk_log_tree(trans
, log_root_tree
, &wc
);
3204 btrfs_error(fs_info
, ret
, "Failed to pin buffers while "
3205 "recovering log root tree.");
3210 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
3211 key
.offset
= (u64
)-1;
3212 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
3215 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
3218 btrfs_error(fs_info
, ret
,
3219 "Couldn't find tree log root.");
3223 if (path
->slots
[0] == 0)
3227 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
3229 btrfs_release_path(path
);
3230 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
3233 log
= btrfs_read_fs_root_no_radix(log_root_tree
,
3237 btrfs_error(fs_info
, ret
,
3238 "Couldn't read tree log root.");
3242 tmp_key
.objectid
= found_key
.offset
;
3243 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3244 tmp_key
.offset
= (u64
)-1;
3246 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
3247 if (IS_ERR(wc
.replay_dest
)) {
3248 ret
= PTR_ERR(wc
.replay_dest
);
3249 btrfs_error(fs_info
, ret
, "Couldn't read target root "
3250 "for tree log recovery.");
3254 wc
.replay_dest
->log_root
= log
;
3255 btrfs_record_root_in_trans(trans
, wc
.replay_dest
);
3256 ret
= walk_log_tree(trans
, log
, &wc
);
3259 if (wc
.stage
== LOG_WALK_REPLAY_ALL
) {
3260 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
3265 key
.offset
= found_key
.offset
- 1;
3266 wc
.replay_dest
->log_root
= NULL
;
3267 free_extent_buffer(log
->node
);
3268 free_extent_buffer(log
->commit_root
);
3271 if (found_key
.offset
== 0)
3274 btrfs_release_path(path
);
3276 /* step one is to pin it all, step two is to replay just inodes */
3279 wc
.process_func
= replay_one_buffer
;
3280 wc
.stage
= LOG_WALK_REPLAY_INODES
;
3283 /* step three is to replay everything */
3284 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
3289 btrfs_free_path(path
);
3291 free_extent_buffer(log_root_tree
->node
);
3292 log_root_tree
->log_root
= NULL
;
3293 fs_info
->log_root_recovering
= 0;
3295 /* step 4: commit the transaction, which also unpins the blocks */
3296 btrfs_commit_transaction(trans
, fs_info
->tree_root
);
3298 kfree(log_root_tree
);
3302 btrfs_free_path(path
);
3307 * there are some corner cases where we want to force a full
3308 * commit instead of allowing a directory to be logged.
3310 * They revolve around files there were unlinked from the directory, and
3311 * this function updates the parent directory so that a full commit is
3312 * properly done if it is fsync'd later after the unlinks are done.
3314 void btrfs_record_unlink_dir(struct btrfs_trans_handle
*trans
,
3315 struct inode
*dir
, struct inode
*inode
,
3319 * when we're logging a file, if it hasn't been renamed
3320 * or unlinked, and its inode is fully committed on disk,
3321 * we don't have to worry about walking up the directory chain
3322 * to log its parents.
3324 * So, we use the last_unlink_trans field to put this transid
3325 * into the file. When the file is logged we check it and
3326 * don't log the parents if the file is fully on disk.
3328 if (S_ISREG(inode
->i_mode
))
3329 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
3332 * if this directory was already logged any new
3333 * names for this file/dir will get recorded
3336 if (BTRFS_I(dir
)->logged_trans
== trans
->transid
)
3340 * if the inode we're about to unlink was logged,
3341 * the log will be properly updated for any new names
3343 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
)
3347 * when renaming files across directories, if the directory
3348 * there we're unlinking from gets fsync'd later on, there's
3349 * no way to find the destination directory later and fsync it
3350 * properly. So, we have to be conservative and force commits
3351 * so the new name gets discovered.
3356 /* we can safely do the unlink without any special recording */
3360 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
3364 * Call this after adding a new name for a file and it will properly
3365 * update the log to reflect the new name.
3367 * It will return zero if all goes well, and it will return 1 if a
3368 * full transaction commit is required.
3370 int btrfs_log_new_name(struct btrfs_trans_handle
*trans
,
3371 struct inode
*inode
, struct inode
*old_dir
,
3372 struct dentry
*parent
)
3374 struct btrfs_root
* root
= BTRFS_I(inode
)->root
;
3377 * this will force the logging code to walk the dentry chain
3380 if (S_ISREG(inode
->i_mode
))
3381 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
3384 * if this inode hasn't been logged and directory we're renaming it
3385 * from hasn't been logged, we don't need to log it
3387 if (BTRFS_I(inode
)->logged_trans
<=
3388 root
->fs_info
->last_trans_committed
&&
3389 (!old_dir
|| BTRFS_I(old_dir
)->logged_trans
<=
3390 root
->fs_info
->last_trans_committed
))
3393 return btrfs_log_inode_parent(trans
, root
, inode
, parent
, 1);