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>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
24 #include "transaction.h"
27 #include "print-tree.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
42 * directory trouble cases
44 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
45 * log, we must force a full commit before doing an fsync of the directory
46 * where the unlink was done.
47 * ---> record transid of last unlink/rename per directory
51 * rename foo/some_dir foo2/some_dir
53 * fsync foo/some_dir/some_file
55 * The fsync above will unlink the original some_dir without recording
56 * it in its new location (foo2). After a crash, some_dir will be gone
57 * unless the fsync of some_file forces a full commit
59 * 2) we must log any new names for any file or dir that is in the fsync
60 * log. ---> check inode while renaming/linking.
62 * 2a) we must log any new names for any file or dir during rename
63 * when the directory they are being removed from was logged.
64 * ---> check inode and old parent dir during rename
66 * 2a is actually the more important variant. With the extra logging
67 * a crash might unlink the old name without recreating the new one
69 * 3) after a crash, we must go through any directories with a link count
70 * of zero and redo the rm -rf
77 * The directory f1 was fully removed from the FS, but fsync was never
78 * called on f1, only its parent dir. After a crash the rm -rf must
79 * be replayed. This must be able to recurse down the entire
80 * directory tree. The inode link count fixup code takes care of the
85 * stages for the tree walking. The first
86 * stage (0) is to only pin down the blocks we find
87 * the second stage (1) is to make sure that all the inodes
88 * we find in the log are created in the subvolume.
90 * The last stage is to deal with directories and links and extents
91 * and all the other fun semantics
93 #define LOG_WALK_PIN_ONLY 0
94 #define LOG_WALK_REPLAY_INODES 1
95 #define LOG_WALK_REPLAY_DIR_INDEX 2
96 #define LOG_WALK_REPLAY_ALL 3
98 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
99 struct btrfs_root
*root
, struct inode
*inode
,
101 static int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
102 struct btrfs_root
*root
,
103 struct btrfs_path
*path
, u64 objectid
);
104 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
105 struct btrfs_root
*root
,
106 struct btrfs_root
*log
,
107 struct btrfs_path
*path
,
108 u64 dirid
, int del_all
);
111 * tree logging is a special write ahead log used to make sure that
112 * fsyncs and O_SYNCs can happen without doing full tree commits.
114 * Full tree commits are expensive because they require commonly
115 * modified blocks to be recowed, creating many dirty pages in the
116 * extent tree an 4x-6x higher write load than ext3.
118 * Instead of doing a tree commit on every fsync, we use the
119 * key ranges and transaction ids to find items for a given file or directory
120 * that have changed in this transaction. Those items are copied into
121 * a special tree (one per subvolume root), that tree is written to disk
122 * and then the fsync is considered complete.
124 * After a crash, items are copied out of the log-tree back into the
125 * subvolume tree. Any file data extents found are recorded in the extent
126 * allocation tree, and the log-tree freed.
128 * The log tree is read three times, once to pin down all the extents it is
129 * using in ram and once, once to create all the inodes logged in the tree
130 * and once to do all the other items.
134 * start a sub transaction and setup the log tree
135 * this increments the log tree writer count to make the people
136 * syncing the tree wait for us to finish
138 static int start_log_trans(struct btrfs_trans_handle
*trans
,
139 struct btrfs_root
*root
)
144 mutex_lock(&root
->log_mutex
);
145 if (root
->log_root
) {
146 if (!root
->log_start_pid
) {
147 root
->log_start_pid
= current
->pid
;
148 root
->log_multiple_pids
= false;
149 } else if (root
->log_start_pid
!= current
->pid
) {
150 root
->log_multiple_pids
= true;
153 atomic_inc(&root
->log_batch
);
154 atomic_inc(&root
->log_writers
);
155 mutex_unlock(&root
->log_mutex
);
158 root
->log_multiple_pids
= false;
159 root
->log_start_pid
= current
->pid
;
160 mutex_lock(&root
->fs_info
->tree_log_mutex
);
161 if (!root
->fs_info
->log_root_tree
) {
162 ret
= btrfs_init_log_root_tree(trans
, root
->fs_info
);
166 if (err
== 0 && !root
->log_root
) {
167 ret
= btrfs_add_log_tree(trans
, root
);
171 mutex_unlock(&root
->fs_info
->tree_log_mutex
);
172 atomic_inc(&root
->log_batch
);
173 atomic_inc(&root
->log_writers
);
174 mutex_unlock(&root
->log_mutex
);
179 * returns 0 if there was a log transaction running and we were able
180 * to join, or returns -ENOENT if there were not transactions
183 static int join_running_log_trans(struct btrfs_root
*root
)
191 mutex_lock(&root
->log_mutex
);
192 if (root
->log_root
) {
194 atomic_inc(&root
->log_writers
);
196 mutex_unlock(&root
->log_mutex
);
201 * This either makes the current running log transaction wait
202 * until you call btrfs_end_log_trans() or it makes any future
203 * log transactions wait until you call btrfs_end_log_trans()
205 int btrfs_pin_log_trans(struct btrfs_root
*root
)
209 mutex_lock(&root
->log_mutex
);
210 atomic_inc(&root
->log_writers
);
211 mutex_unlock(&root
->log_mutex
);
216 * indicate we're done making changes to the log tree
217 * and wake up anyone waiting to do a sync
219 void btrfs_end_log_trans(struct btrfs_root
*root
)
221 if (atomic_dec_and_test(&root
->log_writers
)) {
223 if (waitqueue_active(&root
->log_writer_wait
))
224 wake_up(&root
->log_writer_wait
);
230 * the walk control struct is used to pass state down the chain when
231 * processing the log tree. The stage field tells us which part
232 * of the log tree processing we are currently doing. The others
233 * are state fields used for that specific part
235 struct walk_control
{
236 /* should we free the extent on disk when done? This is used
237 * at transaction commit time while freeing a log tree
241 /* should we write out the extent buffer? This is used
242 * while flushing the log tree to disk during a sync
246 /* should we wait for the extent buffer io to finish? Also used
247 * while flushing the log tree to disk for a sync
251 /* pin only walk, we record which extents on disk belong to the
256 /* what stage of the replay code we're currently in */
259 /* the root we are currently replaying */
260 struct btrfs_root
*replay_dest
;
262 /* the trans handle for the current replay */
263 struct btrfs_trans_handle
*trans
;
265 /* the function that gets used to process blocks we find in the
266 * tree. Note the extent_buffer might not be up to date when it is
267 * passed in, and it must be checked or read if you need the data
270 int (*process_func
)(struct btrfs_root
*log
, struct extent_buffer
*eb
,
271 struct walk_control
*wc
, u64 gen
);
275 * process_func used to pin down extents, write them or wait on them
277 static int process_one_buffer(struct btrfs_root
*log
,
278 struct extent_buffer
*eb
,
279 struct walk_control
*wc
, u64 gen
)
284 * If this fs is mixed then we need to be able to process the leaves to
285 * pin down any logged extents, so we have to read the block.
287 if (btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
)) {
288 ret
= btrfs_read_buffer(eb
, gen
);
294 ret
= btrfs_pin_extent_for_log_replay(log
->fs_info
->extent_root
,
297 if (!ret
&& btrfs_buffer_uptodate(eb
, gen
, 0)) {
298 if (wc
->pin
&& btrfs_header_level(eb
) == 0)
299 ret
= btrfs_exclude_logged_extents(log
, eb
);
301 btrfs_write_tree_block(eb
);
303 btrfs_wait_tree_block_writeback(eb
);
309 * Item overwrite used by replay and tree logging. eb, slot and key all refer
310 * to the src data we are copying out.
312 * root is the tree we are copying into, and path is a scratch
313 * path for use in this function (it should be released on entry and
314 * will be released on exit).
316 * If the key is already in the destination tree the existing item is
317 * overwritten. If the existing item isn't big enough, it is extended.
318 * If it is too large, it is truncated.
320 * If the key isn't in the destination yet, a new item is inserted.
322 static noinline
int overwrite_item(struct btrfs_trans_handle
*trans
,
323 struct btrfs_root
*root
,
324 struct btrfs_path
*path
,
325 struct extent_buffer
*eb
, int slot
,
326 struct btrfs_key
*key
)
330 u64 saved_i_size
= 0;
331 int save_old_i_size
= 0;
332 unsigned long src_ptr
;
333 unsigned long dst_ptr
;
334 int overwrite_root
= 0;
335 bool inode_item
= key
->type
== BTRFS_INODE_ITEM_KEY
;
337 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
340 item_size
= btrfs_item_size_nr(eb
, slot
);
341 src_ptr
= btrfs_item_ptr_offset(eb
, slot
);
343 /* look for the key in the destination tree */
344 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
351 u32 dst_size
= btrfs_item_size_nr(path
->nodes
[0],
353 if (dst_size
!= item_size
)
356 if (item_size
== 0) {
357 btrfs_release_path(path
);
360 dst_copy
= kmalloc(item_size
, GFP_NOFS
);
361 src_copy
= kmalloc(item_size
, GFP_NOFS
);
362 if (!dst_copy
|| !src_copy
) {
363 btrfs_release_path(path
);
369 read_extent_buffer(eb
, src_copy
, src_ptr
, item_size
);
371 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
372 read_extent_buffer(path
->nodes
[0], dst_copy
, dst_ptr
,
374 ret
= memcmp(dst_copy
, src_copy
, item_size
);
379 * they have the same contents, just return, this saves
380 * us from cowing blocks in the destination tree and doing
381 * extra writes that may not have been done by a previous
385 btrfs_release_path(path
);
390 * We need to load the old nbytes into the inode so when we
391 * replay the extents we've logged we get the right nbytes.
394 struct btrfs_inode_item
*item
;
398 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
399 struct btrfs_inode_item
);
400 nbytes
= btrfs_inode_nbytes(path
->nodes
[0], item
);
401 item
= btrfs_item_ptr(eb
, slot
,
402 struct btrfs_inode_item
);
403 btrfs_set_inode_nbytes(eb
, item
, nbytes
);
406 * If this is a directory we need to reset the i_size to
407 * 0 so that we can set it up properly when replaying
408 * the rest of the items in this log.
410 mode
= btrfs_inode_mode(eb
, item
);
412 btrfs_set_inode_size(eb
, item
, 0);
414 } else if (inode_item
) {
415 struct btrfs_inode_item
*item
;
419 * New inode, set nbytes to 0 so that the nbytes comes out
420 * properly when we replay the extents.
422 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
423 btrfs_set_inode_nbytes(eb
, item
, 0);
426 * If this is a directory we need to reset the i_size to 0 so
427 * that we can set it up properly when replaying the rest of
428 * the items in this log.
430 mode
= btrfs_inode_mode(eb
, item
);
432 btrfs_set_inode_size(eb
, item
, 0);
435 btrfs_release_path(path
);
436 /* try to insert the key into the destination tree */
437 ret
= btrfs_insert_empty_item(trans
, root
, path
,
440 /* make sure any existing item is the correct size */
441 if (ret
== -EEXIST
) {
443 found_size
= btrfs_item_size_nr(path
->nodes
[0],
445 if (found_size
> item_size
)
446 btrfs_truncate_item(root
, path
, item_size
, 1);
447 else if (found_size
< item_size
)
448 btrfs_extend_item(root
, path
,
449 item_size
- found_size
);
453 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0],
456 /* don't overwrite an existing inode if the generation number
457 * was logged as zero. This is done when the tree logging code
458 * is just logging an inode to make sure it exists after recovery.
460 * Also, don't overwrite i_size on directories during replay.
461 * log replay inserts and removes directory items based on the
462 * state of the tree found in the subvolume, and i_size is modified
465 if (key
->type
== BTRFS_INODE_ITEM_KEY
&& ret
== -EEXIST
) {
466 struct btrfs_inode_item
*src_item
;
467 struct btrfs_inode_item
*dst_item
;
469 src_item
= (struct btrfs_inode_item
*)src_ptr
;
470 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
472 if (btrfs_inode_generation(eb
, src_item
) == 0)
475 if (overwrite_root
&&
476 S_ISDIR(btrfs_inode_mode(eb
, src_item
)) &&
477 S_ISDIR(btrfs_inode_mode(path
->nodes
[0], dst_item
))) {
479 saved_i_size
= btrfs_inode_size(path
->nodes
[0],
484 copy_extent_buffer(path
->nodes
[0], eb
, dst_ptr
,
487 if (save_old_i_size
) {
488 struct btrfs_inode_item
*dst_item
;
489 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
490 btrfs_set_inode_size(path
->nodes
[0], dst_item
, saved_i_size
);
493 /* make sure the generation is filled in */
494 if (key
->type
== BTRFS_INODE_ITEM_KEY
) {
495 struct btrfs_inode_item
*dst_item
;
496 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
497 if (btrfs_inode_generation(path
->nodes
[0], dst_item
) == 0) {
498 btrfs_set_inode_generation(path
->nodes
[0], dst_item
,
503 btrfs_mark_buffer_dirty(path
->nodes
[0]);
504 btrfs_release_path(path
);
509 * simple helper to read an inode off the disk from a given root
510 * This can only be called for subvolume roots and not for the log
512 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
515 struct btrfs_key key
;
518 key
.objectid
= objectid
;
519 key
.type
= BTRFS_INODE_ITEM_KEY
;
521 inode
= btrfs_iget(root
->fs_info
->sb
, &key
, root
, NULL
);
524 } else if (is_bad_inode(inode
)) {
531 /* replays a single extent in 'eb' at 'slot' with 'key' into the
532 * subvolume 'root'. path is released on entry and should be released
535 * extents in the log tree have not been allocated out of the extent
536 * tree yet. So, this completes the allocation, taking a reference
537 * as required if the extent already exists or creating a new extent
538 * if it isn't in the extent allocation tree yet.
540 * The extent is inserted into the file, dropping any existing extents
541 * from the file that overlap the new one.
543 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
544 struct btrfs_root
*root
,
545 struct btrfs_path
*path
,
546 struct extent_buffer
*eb
, int slot
,
547 struct btrfs_key
*key
)
551 u64 start
= key
->offset
;
553 struct btrfs_file_extent_item
*item
;
554 struct inode
*inode
= NULL
;
558 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
559 found_type
= btrfs_file_extent_type(eb
, item
);
561 if (found_type
== BTRFS_FILE_EXTENT_REG
||
562 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
563 nbytes
= btrfs_file_extent_num_bytes(eb
, item
);
564 extent_end
= start
+ nbytes
;
567 * We don't add to the inodes nbytes if we are prealloc or a
570 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
572 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
573 size
= btrfs_file_extent_inline_len(eb
, item
);
574 nbytes
= btrfs_file_extent_ram_bytes(eb
, item
);
575 extent_end
= ALIGN(start
+ size
, root
->sectorsize
);
581 inode
= read_one_inode(root
, key
->objectid
);
588 * first check to see if we already have this extent in the
589 * file. This must be done before the btrfs_drop_extents run
590 * so we don't try to drop this extent.
592 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
596 (found_type
== BTRFS_FILE_EXTENT_REG
||
597 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)) {
598 struct btrfs_file_extent_item cmp1
;
599 struct btrfs_file_extent_item cmp2
;
600 struct btrfs_file_extent_item
*existing
;
601 struct extent_buffer
*leaf
;
603 leaf
= path
->nodes
[0];
604 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
605 struct btrfs_file_extent_item
);
607 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
609 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
613 * we already have a pointer to this exact extent,
614 * we don't have to do anything
616 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
617 btrfs_release_path(path
);
621 btrfs_release_path(path
);
623 /* drop any overlapping extents */
624 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, extent_end
, 1);
628 if (found_type
== BTRFS_FILE_EXTENT_REG
||
629 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
631 unsigned long dest_offset
;
632 struct btrfs_key ins
;
634 ret
= btrfs_insert_empty_item(trans
, root
, path
, key
,
638 dest_offset
= btrfs_item_ptr_offset(path
->nodes
[0],
640 copy_extent_buffer(path
->nodes
[0], eb
, dest_offset
,
641 (unsigned long)item
, sizeof(*item
));
643 ins
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
644 ins
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
645 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
646 offset
= key
->offset
- btrfs_file_extent_offset(eb
, item
);
648 if (ins
.objectid
> 0) {
651 LIST_HEAD(ordered_sums
);
653 * is this extent already allocated in the extent
654 * allocation tree? If so, just add a reference
656 ret
= btrfs_lookup_extent(root
, ins
.objectid
,
659 ret
= btrfs_inc_extent_ref(trans
, root
,
660 ins
.objectid
, ins
.offset
,
661 0, root
->root_key
.objectid
,
662 key
->objectid
, offset
, 0);
667 * insert the extent pointer in the extent
670 ret
= btrfs_alloc_logged_file_extent(trans
,
671 root
, root
->root_key
.objectid
,
672 key
->objectid
, offset
, &ins
);
676 btrfs_release_path(path
);
678 if (btrfs_file_extent_compression(eb
, item
)) {
679 csum_start
= ins
.objectid
;
680 csum_end
= csum_start
+ ins
.offset
;
682 csum_start
= ins
.objectid
+
683 btrfs_file_extent_offset(eb
, item
);
684 csum_end
= csum_start
+
685 btrfs_file_extent_num_bytes(eb
, item
);
688 ret
= btrfs_lookup_csums_range(root
->log_root
,
689 csum_start
, csum_end
- 1,
693 while (!list_empty(&ordered_sums
)) {
694 struct btrfs_ordered_sum
*sums
;
695 sums
= list_entry(ordered_sums
.next
,
696 struct btrfs_ordered_sum
,
699 ret
= btrfs_csum_file_blocks(trans
,
700 root
->fs_info
->csum_root
,
702 list_del(&sums
->list
);
708 btrfs_release_path(path
);
710 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
711 /* inline extents are easy, we just overwrite them */
712 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
717 inode_add_bytes(inode
, nbytes
);
718 ret
= btrfs_update_inode(trans
, root
, inode
);
726 * when cleaning up conflicts between the directory names in the
727 * subvolume, directory names in the log and directory names in the
728 * inode back references, we may have to unlink inodes from directories.
730 * This is a helper function to do the unlink of a specific directory
733 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
734 struct btrfs_root
*root
,
735 struct btrfs_path
*path
,
737 struct btrfs_dir_item
*di
)
742 struct extent_buffer
*leaf
;
743 struct btrfs_key location
;
746 leaf
= path
->nodes
[0];
748 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
749 name_len
= btrfs_dir_name_len(leaf
, di
);
750 name
= kmalloc(name_len
, GFP_NOFS
);
754 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
755 btrfs_release_path(path
);
757 inode
= read_one_inode(root
, location
.objectid
);
763 ret
= link_to_fixup_dir(trans
, root
, path
, location
.objectid
);
767 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
771 ret
= btrfs_run_delayed_items(trans
, root
);
779 * helper function to see if a given name and sequence number found
780 * in an inode back reference are already in a directory and correctly
781 * point to this inode
783 static noinline
int inode_in_dir(struct btrfs_root
*root
,
784 struct btrfs_path
*path
,
785 u64 dirid
, u64 objectid
, u64 index
,
786 const char *name
, int name_len
)
788 struct btrfs_dir_item
*di
;
789 struct btrfs_key location
;
792 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
793 index
, name
, name_len
, 0);
794 if (di
&& !IS_ERR(di
)) {
795 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
796 if (location
.objectid
!= objectid
)
800 btrfs_release_path(path
);
802 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
803 if (di
&& !IS_ERR(di
)) {
804 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
805 if (location
.objectid
!= objectid
)
811 btrfs_release_path(path
);
816 * helper function to check a log tree for a named back reference in
817 * an inode. This is used to decide if a back reference that is
818 * found in the subvolume conflicts with what we find in the log.
820 * inode backreferences may have multiple refs in a single item,
821 * during replay we process one reference at a time, and we don't
822 * want to delete valid links to a file from the subvolume if that
823 * link is also in the log.
825 static noinline
int backref_in_log(struct btrfs_root
*log
,
826 struct btrfs_key
*key
,
828 char *name
, int namelen
)
830 struct btrfs_path
*path
;
831 struct btrfs_inode_ref
*ref
;
833 unsigned long ptr_end
;
834 unsigned long name_ptr
;
840 path
= btrfs_alloc_path();
844 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
848 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
850 if (key
->type
== BTRFS_INODE_EXTREF_KEY
) {
851 if (btrfs_find_name_in_ext_backref(path
, ref_objectid
,
852 name
, namelen
, NULL
))
858 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
859 ptr_end
= ptr
+ item_size
;
860 while (ptr
< ptr_end
) {
861 ref
= (struct btrfs_inode_ref
*)ptr
;
862 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
863 if (found_name_len
== namelen
) {
864 name_ptr
= (unsigned long)(ref
+ 1);
865 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
872 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
875 btrfs_free_path(path
);
879 static inline int __add_inode_ref(struct btrfs_trans_handle
*trans
,
880 struct btrfs_root
*root
,
881 struct btrfs_path
*path
,
882 struct btrfs_root
*log_root
,
883 struct inode
*dir
, struct inode
*inode
,
884 struct extent_buffer
*eb
,
885 u64 inode_objectid
, u64 parent_objectid
,
886 u64 ref_index
, char *name
, int namelen
,
892 struct extent_buffer
*leaf
;
893 struct btrfs_dir_item
*di
;
894 struct btrfs_key search_key
;
895 struct btrfs_inode_extref
*extref
;
898 /* Search old style refs */
899 search_key
.objectid
= inode_objectid
;
900 search_key
.type
= BTRFS_INODE_REF_KEY
;
901 search_key
.offset
= parent_objectid
;
902 ret
= btrfs_search_slot(NULL
, root
, &search_key
, path
, 0, 0);
904 struct btrfs_inode_ref
*victim_ref
;
906 unsigned long ptr_end
;
908 leaf
= path
->nodes
[0];
910 /* are we trying to overwrite a back ref for the root directory
911 * if so, just jump out, we're done
913 if (search_key
.objectid
== search_key
.offset
)
916 /* check all the names in this back reference to see
917 * if they are in the log. if so, we allow them to stay
918 * otherwise they must be unlinked as a conflict
920 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
921 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
922 while (ptr
< ptr_end
) {
923 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
924 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
926 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
930 read_extent_buffer(leaf
, victim_name
,
931 (unsigned long)(victim_ref
+ 1),
934 if (!backref_in_log(log_root
, &search_key
,
939 btrfs_release_path(path
);
941 ret
= btrfs_unlink_inode(trans
, root
, dir
,
947 ret
= btrfs_run_delayed_items(trans
, root
);
955 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
959 * NOTE: we have searched root tree and checked the
960 * coresponding ref, it does not need to check again.
964 btrfs_release_path(path
);
966 /* Same search but for extended refs */
967 extref
= btrfs_lookup_inode_extref(NULL
, root
, path
, name
, namelen
,
968 inode_objectid
, parent_objectid
, 0,
970 if (!IS_ERR_OR_NULL(extref
)) {
974 struct inode
*victim_parent
;
976 leaf
= path
->nodes
[0];
978 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
979 base
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
981 while (cur_offset
< item_size
) {
982 extref
= (struct btrfs_inode_extref
*)base
+ cur_offset
;
984 victim_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
986 if (btrfs_inode_extref_parent(leaf
, extref
) != parent_objectid
)
989 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
992 read_extent_buffer(leaf
, victim_name
, (unsigned long)&extref
->name
,
995 search_key
.objectid
= inode_objectid
;
996 search_key
.type
= BTRFS_INODE_EXTREF_KEY
;
997 search_key
.offset
= btrfs_extref_hash(parent_objectid
,
1001 if (!backref_in_log(log_root
, &search_key
,
1002 parent_objectid
, victim_name
,
1005 victim_parent
= read_one_inode(root
,
1007 if (victim_parent
) {
1009 btrfs_release_path(path
);
1011 ret
= btrfs_unlink_inode(trans
, root
,
1017 ret
= btrfs_run_delayed_items(
1020 iput(victim_parent
);
1031 cur_offset
+= victim_name_len
+ sizeof(*extref
);
1035 btrfs_release_path(path
);
1037 /* look for a conflicting sequence number */
1038 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, btrfs_ino(dir
),
1039 ref_index
, name
, namelen
, 0);
1040 if (di
&& !IS_ERR(di
)) {
1041 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
1045 btrfs_release_path(path
);
1047 /* look for a conflicing name */
1048 di
= btrfs_lookup_dir_item(trans
, root
, path
, btrfs_ino(dir
),
1050 if (di
&& !IS_ERR(di
)) {
1051 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
1055 btrfs_release_path(path
);
1060 static int extref_get_fields(struct extent_buffer
*eb
, unsigned long ref_ptr
,
1061 u32
*namelen
, char **name
, u64
*index
,
1062 u64
*parent_objectid
)
1064 struct btrfs_inode_extref
*extref
;
1066 extref
= (struct btrfs_inode_extref
*)ref_ptr
;
1068 *namelen
= btrfs_inode_extref_name_len(eb
, extref
);
1069 *name
= kmalloc(*namelen
, GFP_NOFS
);
1073 read_extent_buffer(eb
, *name
, (unsigned long)&extref
->name
,
1076 *index
= btrfs_inode_extref_index(eb
, extref
);
1077 if (parent_objectid
)
1078 *parent_objectid
= btrfs_inode_extref_parent(eb
, extref
);
1083 static int ref_get_fields(struct extent_buffer
*eb
, unsigned long ref_ptr
,
1084 u32
*namelen
, char **name
, u64
*index
)
1086 struct btrfs_inode_ref
*ref
;
1088 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
1090 *namelen
= btrfs_inode_ref_name_len(eb
, ref
);
1091 *name
= kmalloc(*namelen
, GFP_NOFS
);
1095 read_extent_buffer(eb
, *name
, (unsigned long)(ref
+ 1), *namelen
);
1097 *index
= btrfs_inode_ref_index(eb
, ref
);
1103 * replay one inode back reference item found in the log tree.
1104 * eb, slot and key refer to the buffer and key found in the log tree.
1105 * root is the destination we are replaying into, and path is for temp
1106 * use by this function. (it should be released on return).
1108 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
1109 struct btrfs_root
*root
,
1110 struct btrfs_root
*log
,
1111 struct btrfs_path
*path
,
1112 struct extent_buffer
*eb
, int slot
,
1113 struct btrfs_key
*key
)
1115 struct inode
*dir
= NULL
;
1116 struct inode
*inode
= NULL
;
1117 unsigned long ref_ptr
;
1118 unsigned long ref_end
;
1122 int search_done
= 0;
1123 int log_ref_ver
= 0;
1124 u64 parent_objectid
;
1127 int ref_struct_size
;
1129 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
1130 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
1132 if (key
->type
== BTRFS_INODE_EXTREF_KEY
) {
1133 struct btrfs_inode_extref
*r
;
1135 ref_struct_size
= sizeof(struct btrfs_inode_extref
);
1137 r
= (struct btrfs_inode_extref
*)ref_ptr
;
1138 parent_objectid
= btrfs_inode_extref_parent(eb
, r
);
1140 ref_struct_size
= sizeof(struct btrfs_inode_ref
);
1141 parent_objectid
= key
->offset
;
1143 inode_objectid
= key
->objectid
;
1146 * it is possible that we didn't log all the parent directories
1147 * for a given inode. If we don't find the dir, just don't
1148 * copy the back ref in. The link count fixup code will take
1151 dir
= read_one_inode(root
, parent_objectid
);
1157 inode
= read_one_inode(root
, inode_objectid
);
1163 while (ref_ptr
< ref_end
) {
1165 ret
= extref_get_fields(eb
, ref_ptr
, &namelen
, &name
,
1166 &ref_index
, &parent_objectid
);
1168 * parent object can change from one array
1172 dir
= read_one_inode(root
, parent_objectid
);
1178 ret
= ref_get_fields(eb
, ref_ptr
, &namelen
, &name
,
1184 /* if we already have a perfect match, we're done */
1185 if (!inode_in_dir(root
, path
, btrfs_ino(dir
), btrfs_ino(inode
),
1186 ref_index
, name
, namelen
)) {
1188 * look for a conflicting back reference in the
1189 * metadata. if we find one we have to unlink that name
1190 * of the file before we add our new link. Later on, we
1191 * overwrite any existing back reference, and we don't
1192 * want to create dangling pointers in the directory.
1196 ret
= __add_inode_ref(trans
, root
, path
, log
,
1200 ref_index
, name
, namelen
,
1209 /* insert our name */
1210 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
,
1215 btrfs_update_inode(trans
, root
, inode
);
1218 ref_ptr
= (unsigned long)(ref_ptr
+ ref_struct_size
) + namelen
;
1227 /* finally write the back reference in the inode */
1228 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
1230 btrfs_release_path(path
);
1237 static int insert_orphan_item(struct btrfs_trans_handle
*trans
,
1238 struct btrfs_root
*root
, u64 offset
)
1241 ret
= btrfs_find_orphan_item(root
, offset
);
1243 ret
= btrfs_insert_orphan_item(trans
, root
, offset
);
1247 static int count_inode_extrefs(struct btrfs_root
*root
,
1248 struct inode
*inode
, struct btrfs_path
*path
)
1252 unsigned int nlink
= 0;
1255 u64 inode_objectid
= btrfs_ino(inode
);
1258 struct btrfs_inode_extref
*extref
;
1259 struct extent_buffer
*leaf
;
1262 ret
= btrfs_find_one_extref(root
, inode_objectid
, offset
, path
,
1267 leaf
= path
->nodes
[0];
1268 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1269 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1271 while (cur_offset
< item_size
) {
1272 extref
= (struct btrfs_inode_extref
*) (ptr
+ cur_offset
);
1273 name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
1277 cur_offset
+= name_len
+ sizeof(*extref
);
1281 btrfs_release_path(path
);
1283 btrfs_release_path(path
);
1290 static int count_inode_refs(struct btrfs_root
*root
,
1291 struct inode
*inode
, struct btrfs_path
*path
)
1294 struct btrfs_key key
;
1295 unsigned int nlink
= 0;
1297 unsigned long ptr_end
;
1299 u64 ino
= btrfs_ino(inode
);
1302 key
.type
= BTRFS_INODE_REF_KEY
;
1303 key
.offset
= (u64
)-1;
1306 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1310 if (path
->slots
[0] == 0)
1315 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1317 if (key
.objectid
!= ino
||
1318 key
.type
!= BTRFS_INODE_REF_KEY
)
1320 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1321 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1323 while (ptr
< ptr_end
) {
1324 struct btrfs_inode_ref
*ref
;
1326 ref
= (struct btrfs_inode_ref
*)ptr
;
1327 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1329 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1333 if (key
.offset
== 0)
1335 if (path
->slots
[0] > 0) {
1340 btrfs_release_path(path
);
1342 btrfs_release_path(path
);
1348 * There are a few corners where the link count of the file can't
1349 * be properly maintained during replay. So, instead of adding
1350 * lots of complexity to the log code, we just scan the backrefs
1351 * for any file that has been through replay.
1353 * The scan will update the link count on the inode to reflect the
1354 * number of back refs found. If it goes down to zero, the iput
1355 * will free the inode.
1357 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
1358 struct btrfs_root
*root
,
1359 struct inode
*inode
)
1361 struct btrfs_path
*path
;
1364 u64 ino
= btrfs_ino(inode
);
1366 path
= btrfs_alloc_path();
1370 ret
= count_inode_refs(root
, inode
, path
);
1376 ret
= count_inode_extrefs(root
, inode
, path
);
1387 if (nlink
!= inode
->i_nlink
) {
1388 set_nlink(inode
, nlink
);
1389 btrfs_update_inode(trans
, root
, inode
);
1391 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1393 if (inode
->i_nlink
== 0) {
1394 if (S_ISDIR(inode
->i_mode
)) {
1395 ret
= replay_dir_deletes(trans
, root
, NULL
, path
,
1400 ret
= insert_orphan_item(trans
, root
, ino
);
1404 btrfs_free_path(path
);
1408 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1409 struct btrfs_root
*root
,
1410 struct btrfs_path
*path
)
1413 struct btrfs_key key
;
1414 struct inode
*inode
;
1416 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1417 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1418 key
.offset
= (u64
)-1;
1420 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1425 if (path
->slots
[0] == 0)
1430 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1431 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1432 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1435 ret
= btrfs_del_item(trans
, root
, path
);
1439 btrfs_release_path(path
);
1440 inode
= read_one_inode(root
, key
.offset
);
1444 ret
= fixup_inode_link_count(trans
, root
, inode
);
1450 * fixup on a directory may create new entries,
1451 * make sure we always look for the highset possible
1454 key
.offset
= (u64
)-1;
1458 btrfs_release_path(path
);
1464 * record a given inode in the fixup dir so we can check its link
1465 * count when replay is done. The link count is incremented here
1466 * so the inode won't go away until we check it
1468 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1469 struct btrfs_root
*root
,
1470 struct btrfs_path
*path
,
1473 struct btrfs_key key
;
1475 struct inode
*inode
;
1477 inode
= read_one_inode(root
, objectid
);
1481 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1482 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1483 key
.offset
= objectid
;
1485 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1487 btrfs_release_path(path
);
1489 if (!inode
->i_nlink
)
1490 set_nlink(inode
, 1);
1493 ret
= btrfs_update_inode(trans
, root
, inode
);
1494 } else if (ret
== -EEXIST
) {
1497 BUG(); /* Logic Error */
1505 * when replaying the log for a directory, we only insert names
1506 * for inodes that actually exist. This means an fsync on a directory
1507 * does not implicitly fsync all the new files in it
1509 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1510 struct btrfs_root
*root
,
1511 struct btrfs_path
*path
,
1512 u64 dirid
, u64 index
,
1513 char *name
, int name_len
, u8 type
,
1514 struct btrfs_key
*location
)
1516 struct inode
*inode
;
1520 inode
= read_one_inode(root
, location
->objectid
);
1524 dir
= read_one_inode(root
, dirid
);
1530 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1532 /* FIXME, put inode into FIXUP list */
1540 * take a single entry in a log directory item and replay it into
1543 * if a conflicting item exists in the subdirectory already,
1544 * the inode it points to is unlinked and put into the link count
1547 * If a name from the log points to a file or directory that does
1548 * not exist in the FS, it is skipped. fsyncs on directories
1549 * do not force down inodes inside that directory, just changes to the
1550 * names or unlinks in a directory.
1552 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1553 struct btrfs_root
*root
,
1554 struct btrfs_path
*path
,
1555 struct extent_buffer
*eb
,
1556 struct btrfs_dir_item
*di
,
1557 struct btrfs_key
*key
)
1561 struct btrfs_dir_item
*dst_di
;
1562 struct btrfs_key found_key
;
1563 struct btrfs_key log_key
;
1568 bool update_size
= (key
->type
== BTRFS_DIR_INDEX_KEY
);
1570 dir
= read_one_inode(root
, key
->objectid
);
1574 name_len
= btrfs_dir_name_len(eb
, di
);
1575 name
= kmalloc(name_len
, GFP_NOFS
);
1581 log_type
= btrfs_dir_type(eb
, di
);
1582 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1585 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1586 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1591 btrfs_release_path(path
);
1593 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1594 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1596 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1597 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1606 if (IS_ERR_OR_NULL(dst_di
)) {
1607 /* we need a sequence number to insert, so we only
1608 * do inserts for the BTRFS_DIR_INDEX_KEY types
1610 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1615 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1616 /* the existing item matches the logged item */
1617 if (found_key
.objectid
== log_key
.objectid
&&
1618 found_key
.type
== log_key
.type
&&
1619 found_key
.offset
== log_key
.offset
&&
1620 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1625 * don't drop the conflicting directory entry if the inode
1626 * for the new entry doesn't exist
1631 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1635 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1638 btrfs_release_path(path
);
1639 if (!ret
&& update_size
) {
1640 btrfs_i_size_write(dir
, dir
->i_size
+ name_len
* 2);
1641 ret
= btrfs_update_inode(trans
, root
, dir
);
1648 btrfs_release_path(path
);
1649 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1650 name
, name_len
, log_type
, &log_key
);
1651 if (ret
&& ret
!= -ENOENT
)
1653 update_size
= false;
1659 * find all the names in a directory item and reconcile them into
1660 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1661 * one name in a directory item, but the same code gets used for
1662 * both directory index types
1664 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1665 struct btrfs_root
*root
,
1666 struct btrfs_path
*path
,
1667 struct extent_buffer
*eb
, int slot
,
1668 struct btrfs_key
*key
)
1671 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1672 struct btrfs_dir_item
*di
;
1675 unsigned long ptr_end
;
1677 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1678 ptr_end
= ptr
+ item_size
;
1679 while (ptr
< ptr_end
) {
1680 di
= (struct btrfs_dir_item
*)ptr
;
1681 if (verify_dir_item(root
, eb
, di
))
1683 name_len
= btrfs_dir_name_len(eb
, di
);
1684 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1687 ptr
= (unsigned long)(di
+ 1);
1694 * directory replay has two parts. There are the standard directory
1695 * items in the log copied from the subvolume, and range items
1696 * created in the log while the subvolume was logged.
1698 * The range items tell us which parts of the key space the log
1699 * is authoritative for. During replay, if a key in the subvolume
1700 * directory is in a logged range item, but not actually in the log
1701 * that means it was deleted from the directory before the fsync
1702 * and should be removed.
1704 static noinline
int find_dir_range(struct btrfs_root
*root
,
1705 struct btrfs_path
*path
,
1706 u64 dirid
, int key_type
,
1707 u64
*start_ret
, u64
*end_ret
)
1709 struct btrfs_key key
;
1711 struct btrfs_dir_log_item
*item
;
1715 if (*start_ret
== (u64
)-1)
1718 key
.objectid
= dirid
;
1719 key
.type
= key_type
;
1720 key
.offset
= *start_ret
;
1722 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1726 if (path
->slots
[0] == 0)
1731 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1733 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1737 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1738 struct btrfs_dir_log_item
);
1739 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1741 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1743 *start_ret
= key
.offset
;
1744 *end_ret
= found_end
;
1749 /* check the next slot in the tree to see if it is a valid item */
1750 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1751 if (path
->slots
[0] >= nritems
) {
1752 ret
= btrfs_next_leaf(root
, path
);
1759 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1761 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1765 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1766 struct btrfs_dir_log_item
);
1767 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1768 *start_ret
= key
.offset
;
1769 *end_ret
= found_end
;
1772 btrfs_release_path(path
);
1777 * this looks for a given directory item in the log. If the directory
1778 * item is not in the log, the item is removed and the inode it points
1781 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1782 struct btrfs_root
*root
,
1783 struct btrfs_root
*log
,
1784 struct btrfs_path
*path
,
1785 struct btrfs_path
*log_path
,
1787 struct btrfs_key
*dir_key
)
1790 struct extent_buffer
*eb
;
1793 struct btrfs_dir_item
*di
;
1794 struct btrfs_dir_item
*log_di
;
1797 unsigned long ptr_end
;
1799 struct inode
*inode
;
1800 struct btrfs_key location
;
1803 eb
= path
->nodes
[0];
1804 slot
= path
->slots
[0];
1805 item_size
= btrfs_item_size_nr(eb
, slot
);
1806 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1807 ptr_end
= ptr
+ item_size
;
1808 while (ptr
< ptr_end
) {
1809 di
= (struct btrfs_dir_item
*)ptr
;
1810 if (verify_dir_item(root
, eb
, di
)) {
1815 name_len
= btrfs_dir_name_len(eb
, di
);
1816 name
= kmalloc(name_len
, GFP_NOFS
);
1821 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1824 if (log
&& dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1825 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1828 } else if (log
&& dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1829 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1835 if (IS_ERR_OR_NULL(log_di
)) {
1836 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1837 btrfs_release_path(path
);
1838 btrfs_release_path(log_path
);
1839 inode
= read_one_inode(root
, location
.objectid
);
1845 ret
= link_to_fixup_dir(trans
, root
,
1846 path
, location
.objectid
);
1854 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1857 ret
= btrfs_run_delayed_items(trans
, root
);
1863 /* there might still be more names under this key
1864 * check and repeat if required
1866 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1873 btrfs_release_path(log_path
);
1876 ptr
= (unsigned long)(di
+ 1);
1881 btrfs_release_path(path
);
1882 btrfs_release_path(log_path
);
1887 * deletion replay happens before we copy any new directory items
1888 * out of the log or out of backreferences from inodes. It
1889 * scans the log to find ranges of keys that log is authoritative for,
1890 * and then scans the directory to find items in those ranges that are
1891 * not present in the log.
1893 * Anything we don't find in the log is unlinked and removed from the
1896 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1897 struct btrfs_root
*root
,
1898 struct btrfs_root
*log
,
1899 struct btrfs_path
*path
,
1900 u64 dirid
, int del_all
)
1904 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1906 struct btrfs_key dir_key
;
1907 struct btrfs_key found_key
;
1908 struct btrfs_path
*log_path
;
1911 dir_key
.objectid
= dirid
;
1912 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1913 log_path
= btrfs_alloc_path();
1917 dir
= read_one_inode(root
, dirid
);
1918 /* it isn't an error if the inode isn't there, that can happen
1919 * because we replay the deletes before we copy in the inode item
1923 btrfs_free_path(log_path
);
1931 range_end
= (u64
)-1;
1933 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1934 &range_start
, &range_end
);
1939 dir_key
.offset
= range_start
;
1942 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1947 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1948 if (path
->slots
[0] >= nritems
) {
1949 ret
= btrfs_next_leaf(root
, path
);
1953 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1955 if (found_key
.objectid
!= dirid
||
1956 found_key
.type
!= dir_key
.type
)
1959 if (found_key
.offset
> range_end
)
1962 ret
= check_item_in_log(trans
, root
, log
, path
,
1967 if (found_key
.offset
== (u64
)-1)
1969 dir_key
.offset
= found_key
.offset
+ 1;
1971 btrfs_release_path(path
);
1972 if (range_end
== (u64
)-1)
1974 range_start
= range_end
+ 1;
1979 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1980 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1981 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1982 btrfs_release_path(path
);
1986 btrfs_release_path(path
);
1987 btrfs_free_path(log_path
);
1993 * the process_func used to replay items from the log tree. This
1994 * gets called in two different stages. The first stage just looks
1995 * for inodes and makes sure they are all copied into the subvolume.
1997 * The second stage copies all the other item types from the log into
1998 * the subvolume. The two stage approach is slower, but gets rid of
1999 * lots of complexity around inodes referencing other inodes that exist
2000 * only in the log (references come from either directory items or inode
2003 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
2004 struct walk_control
*wc
, u64 gen
)
2007 struct btrfs_path
*path
;
2008 struct btrfs_root
*root
= wc
->replay_dest
;
2009 struct btrfs_key key
;
2014 ret
= btrfs_read_buffer(eb
, gen
);
2018 level
= btrfs_header_level(eb
);
2023 path
= btrfs_alloc_path();
2027 nritems
= btrfs_header_nritems(eb
);
2028 for (i
= 0; i
< nritems
; i
++) {
2029 btrfs_item_key_to_cpu(eb
, &key
, i
);
2031 /* inode keys are done during the first stage */
2032 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
2033 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
2034 struct btrfs_inode_item
*inode_item
;
2037 inode_item
= btrfs_item_ptr(eb
, i
,
2038 struct btrfs_inode_item
);
2039 mode
= btrfs_inode_mode(eb
, inode_item
);
2040 if (S_ISDIR(mode
)) {
2041 ret
= replay_dir_deletes(wc
->trans
,
2042 root
, log
, path
, key
.objectid
, 0);
2046 ret
= overwrite_item(wc
->trans
, root
, path
,
2051 /* for regular files, make sure corresponding
2052 * orhpan item exist. extents past the new EOF
2053 * will be truncated later by orphan cleanup.
2055 if (S_ISREG(mode
)) {
2056 ret
= insert_orphan_item(wc
->trans
, root
,
2062 ret
= link_to_fixup_dir(wc
->trans
, root
,
2063 path
, key
.objectid
);
2068 if (key
.type
== BTRFS_DIR_INDEX_KEY
&&
2069 wc
->stage
== LOG_WALK_REPLAY_DIR_INDEX
) {
2070 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
2076 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
2079 /* these keys are simply copied */
2080 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
2081 ret
= overwrite_item(wc
->trans
, root
, path
,
2085 } else if (key
.type
== BTRFS_INODE_REF_KEY
||
2086 key
.type
== BTRFS_INODE_EXTREF_KEY
) {
2087 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
2089 if (ret
&& ret
!= -ENOENT
)
2092 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
2093 ret
= replay_one_extent(wc
->trans
, root
, path
,
2097 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
) {
2098 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
2104 btrfs_free_path(path
);
2108 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
2109 struct btrfs_root
*root
,
2110 struct btrfs_path
*path
, int *level
,
2111 struct walk_control
*wc
)
2116 struct extent_buffer
*next
;
2117 struct extent_buffer
*cur
;
2118 struct extent_buffer
*parent
;
2122 WARN_ON(*level
< 0);
2123 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2125 while (*level
> 0) {
2126 WARN_ON(*level
< 0);
2127 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2128 cur
= path
->nodes
[*level
];
2130 if (btrfs_header_level(cur
) != *level
)
2133 if (path
->slots
[*level
] >=
2134 btrfs_header_nritems(cur
))
2137 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2138 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2139 blocksize
= btrfs_level_size(root
, *level
- 1);
2141 parent
= path
->nodes
[*level
];
2142 root_owner
= btrfs_header_owner(parent
);
2144 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
2149 ret
= wc
->process_func(root
, next
, wc
, ptr_gen
);
2151 free_extent_buffer(next
);
2155 path
->slots
[*level
]++;
2157 ret
= btrfs_read_buffer(next
, ptr_gen
);
2159 free_extent_buffer(next
);
2164 btrfs_tree_lock(next
);
2165 btrfs_set_lock_blocking(next
);
2166 clean_tree_block(trans
, root
, next
);
2167 btrfs_wait_tree_block_writeback(next
);
2168 btrfs_tree_unlock(next
);
2171 WARN_ON(root_owner
!=
2172 BTRFS_TREE_LOG_OBJECTID
);
2173 ret
= btrfs_free_and_pin_reserved_extent(root
,
2176 free_extent_buffer(next
);
2180 free_extent_buffer(next
);
2183 ret
= btrfs_read_buffer(next
, ptr_gen
);
2185 free_extent_buffer(next
);
2189 WARN_ON(*level
<= 0);
2190 if (path
->nodes
[*level
-1])
2191 free_extent_buffer(path
->nodes
[*level
-1]);
2192 path
->nodes
[*level
-1] = next
;
2193 *level
= btrfs_header_level(next
);
2194 path
->slots
[*level
] = 0;
2197 WARN_ON(*level
< 0);
2198 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2200 path
->slots
[*level
] = btrfs_header_nritems(path
->nodes
[*level
]);
2206 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
2207 struct btrfs_root
*root
,
2208 struct btrfs_path
*path
, int *level
,
2209 struct walk_control
*wc
)
2216 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2217 slot
= path
->slots
[i
];
2218 if (slot
+ 1 < btrfs_header_nritems(path
->nodes
[i
])) {
2221 WARN_ON(*level
== 0);
2224 struct extent_buffer
*parent
;
2225 if (path
->nodes
[*level
] == root
->node
)
2226 parent
= path
->nodes
[*level
];
2228 parent
= path
->nodes
[*level
+ 1];
2230 root_owner
= btrfs_header_owner(parent
);
2231 ret
= wc
->process_func(root
, path
->nodes
[*level
], wc
,
2232 btrfs_header_generation(path
->nodes
[*level
]));
2237 struct extent_buffer
*next
;
2239 next
= path
->nodes
[*level
];
2242 btrfs_tree_lock(next
);
2243 btrfs_set_lock_blocking(next
);
2244 clean_tree_block(trans
, root
, next
);
2245 btrfs_wait_tree_block_writeback(next
);
2246 btrfs_tree_unlock(next
);
2249 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
2250 ret
= btrfs_free_and_pin_reserved_extent(root
,
2251 path
->nodes
[*level
]->start
,
2252 path
->nodes
[*level
]->len
);
2256 free_extent_buffer(path
->nodes
[*level
]);
2257 path
->nodes
[*level
] = NULL
;
2265 * drop the reference count on the tree rooted at 'snap'. This traverses
2266 * the tree freeing any blocks that have a ref count of zero after being
2269 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
2270 struct btrfs_root
*log
, struct walk_control
*wc
)
2275 struct btrfs_path
*path
;
2278 path
= btrfs_alloc_path();
2282 level
= btrfs_header_level(log
->node
);
2284 path
->nodes
[level
] = log
->node
;
2285 extent_buffer_get(log
->node
);
2286 path
->slots
[level
] = 0;
2289 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
2297 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
2306 /* was the root node processed? if not, catch it here */
2307 if (path
->nodes
[orig_level
]) {
2308 ret
= wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
2309 btrfs_header_generation(path
->nodes
[orig_level
]));
2313 struct extent_buffer
*next
;
2315 next
= path
->nodes
[orig_level
];
2318 btrfs_tree_lock(next
);
2319 btrfs_set_lock_blocking(next
);
2320 clean_tree_block(trans
, log
, next
);
2321 btrfs_wait_tree_block_writeback(next
);
2322 btrfs_tree_unlock(next
);
2325 WARN_ON(log
->root_key
.objectid
!=
2326 BTRFS_TREE_LOG_OBJECTID
);
2327 ret
= btrfs_free_and_pin_reserved_extent(log
, next
->start
,
2335 btrfs_free_path(path
);
2340 * helper function to update the item for a given subvolumes log root
2341 * in the tree of log roots
2343 static int update_log_root(struct btrfs_trans_handle
*trans
,
2344 struct btrfs_root
*log
)
2348 if (log
->log_transid
== 1) {
2349 /* insert root item on the first sync */
2350 ret
= btrfs_insert_root(trans
, log
->fs_info
->log_root_tree
,
2351 &log
->root_key
, &log
->root_item
);
2353 ret
= btrfs_update_root(trans
, log
->fs_info
->log_root_tree
,
2354 &log
->root_key
, &log
->root_item
);
2359 static int wait_log_commit(struct btrfs_trans_handle
*trans
,
2360 struct btrfs_root
*root
, unsigned long transid
)
2363 int index
= transid
% 2;
2366 * we only allow two pending log transactions at a time,
2367 * so we know that if ours is more than 2 older than the
2368 * current transaction, we're done
2371 prepare_to_wait(&root
->log_commit_wait
[index
],
2372 &wait
, TASK_UNINTERRUPTIBLE
);
2373 mutex_unlock(&root
->log_mutex
);
2375 if (root
->fs_info
->last_trans_log_full_commit
!=
2376 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
2377 atomic_read(&root
->log_commit
[index
]))
2380 finish_wait(&root
->log_commit_wait
[index
], &wait
);
2381 mutex_lock(&root
->log_mutex
);
2382 } while (root
->fs_info
->last_trans_log_full_commit
!=
2383 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
2384 atomic_read(&root
->log_commit
[index
]));
2388 static void wait_for_writer(struct btrfs_trans_handle
*trans
,
2389 struct btrfs_root
*root
)
2392 while (root
->fs_info
->last_trans_log_full_commit
!=
2393 trans
->transid
&& atomic_read(&root
->log_writers
)) {
2394 prepare_to_wait(&root
->log_writer_wait
,
2395 &wait
, TASK_UNINTERRUPTIBLE
);
2396 mutex_unlock(&root
->log_mutex
);
2397 if (root
->fs_info
->last_trans_log_full_commit
!=
2398 trans
->transid
&& atomic_read(&root
->log_writers
))
2400 mutex_lock(&root
->log_mutex
);
2401 finish_wait(&root
->log_writer_wait
, &wait
);
2406 * btrfs_sync_log does sends a given tree log down to the disk and
2407 * updates the super blocks to record it. When this call is done,
2408 * you know that any inodes previously logged are safely on disk only
2411 * Any other return value means you need to call btrfs_commit_transaction.
2412 * Some of the edge cases for fsyncing directories that have had unlinks
2413 * or renames done in the past mean that sometimes the only safe
2414 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2415 * that has happened.
2417 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
2418 struct btrfs_root
*root
)
2424 struct btrfs_root
*log
= root
->log_root
;
2425 struct btrfs_root
*log_root_tree
= root
->fs_info
->log_root_tree
;
2426 unsigned long log_transid
= 0;
2427 struct blk_plug plug
;
2429 mutex_lock(&root
->log_mutex
);
2430 log_transid
= root
->log_transid
;
2431 index1
= root
->log_transid
% 2;
2432 if (atomic_read(&root
->log_commit
[index1
])) {
2433 wait_log_commit(trans
, root
, root
->log_transid
);
2434 mutex_unlock(&root
->log_mutex
);
2437 atomic_set(&root
->log_commit
[index1
], 1);
2439 /* wait for previous tree log sync to complete */
2440 if (atomic_read(&root
->log_commit
[(index1
+ 1) % 2]))
2441 wait_log_commit(trans
, root
, root
->log_transid
- 1);
2443 int batch
= atomic_read(&root
->log_batch
);
2444 /* when we're on an ssd, just kick the log commit out */
2445 if (!btrfs_test_opt(root
, SSD
) && root
->log_multiple_pids
) {
2446 mutex_unlock(&root
->log_mutex
);
2447 schedule_timeout_uninterruptible(1);
2448 mutex_lock(&root
->log_mutex
);
2450 wait_for_writer(trans
, root
);
2451 if (batch
== atomic_read(&root
->log_batch
))
2455 /* bail out if we need to do a full commit */
2456 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2458 btrfs_free_logged_extents(log
, log_transid
);
2459 mutex_unlock(&root
->log_mutex
);
2463 if (log_transid
% 2 == 0)
2464 mark
= EXTENT_DIRTY
;
2468 /* we start IO on all the marked extents here, but we don't actually
2469 * wait for them until later.
2471 blk_start_plug(&plug
);
2472 ret
= btrfs_write_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2474 blk_finish_plug(&plug
);
2475 btrfs_abort_transaction(trans
, root
, ret
);
2476 btrfs_free_logged_extents(log
, log_transid
);
2477 mutex_unlock(&root
->log_mutex
);
2481 btrfs_set_root_node(&log
->root_item
, log
->node
);
2483 root
->log_transid
++;
2484 log
->log_transid
= root
->log_transid
;
2485 root
->log_start_pid
= 0;
2488 * IO has been started, blocks of the log tree have WRITTEN flag set
2489 * in their headers. new modifications of the log will be written to
2490 * new positions. so it's safe to allow log writers to go in.
2492 mutex_unlock(&root
->log_mutex
);
2494 mutex_lock(&log_root_tree
->log_mutex
);
2495 atomic_inc(&log_root_tree
->log_batch
);
2496 atomic_inc(&log_root_tree
->log_writers
);
2497 mutex_unlock(&log_root_tree
->log_mutex
);
2499 ret
= update_log_root(trans
, log
);
2501 mutex_lock(&log_root_tree
->log_mutex
);
2502 if (atomic_dec_and_test(&log_root_tree
->log_writers
)) {
2504 if (waitqueue_active(&log_root_tree
->log_writer_wait
))
2505 wake_up(&log_root_tree
->log_writer_wait
);
2509 blk_finish_plug(&plug
);
2510 if (ret
!= -ENOSPC
) {
2511 btrfs_abort_transaction(trans
, root
, ret
);
2512 mutex_unlock(&log_root_tree
->log_mutex
);
2515 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2516 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2517 btrfs_free_logged_extents(log
, log_transid
);
2518 mutex_unlock(&log_root_tree
->log_mutex
);
2523 index2
= log_root_tree
->log_transid
% 2;
2524 if (atomic_read(&log_root_tree
->log_commit
[index2
])) {
2525 blk_finish_plug(&plug
);
2526 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2527 wait_log_commit(trans
, log_root_tree
,
2528 log_root_tree
->log_transid
);
2529 btrfs_free_logged_extents(log
, log_transid
);
2530 mutex_unlock(&log_root_tree
->log_mutex
);
2534 atomic_set(&log_root_tree
->log_commit
[index2
], 1);
2536 if (atomic_read(&log_root_tree
->log_commit
[(index2
+ 1) % 2])) {
2537 wait_log_commit(trans
, log_root_tree
,
2538 log_root_tree
->log_transid
- 1);
2541 wait_for_writer(trans
, log_root_tree
);
2544 * now that we've moved on to the tree of log tree roots,
2545 * check the full commit flag again
2547 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2548 blk_finish_plug(&plug
);
2549 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2550 btrfs_free_logged_extents(log
, log_transid
);
2551 mutex_unlock(&log_root_tree
->log_mutex
);
2553 goto out_wake_log_root
;
2556 ret
= btrfs_write_marked_extents(log_root_tree
,
2557 &log_root_tree
->dirty_log_pages
,
2558 EXTENT_DIRTY
| EXTENT_NEW
);
2559 blk_finish_plug(&plug
);
2561 btrfs_abort_transaction(trans
, root
, ret
);
2562 btrfs_free_logged_extents(log
, log_transid
);
2563 mutex_unlock(&log_root_tree
->log_mutex
);
2564 goto out_wake_log_root
;
2566 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2567 btrfs_wait_marked_extents(log_root_tree
,
2568 &log_root_tree
->dirty_log_pages
,
2569 EXTENT_NEW
| EXTENT_DIRTY
);
2570 btrfs_wait_logged_extents(log
, log_transid
);
2572 btrfs_set_super_log_root(root
->fs_info
->super_for_commit
,
2573 log_root_tree
->node
->start
);
2574 btrfs_set_super_log_root_level(root
->fs_info
->super_for_commit
,
2575 btrfs_header_level(log_root_tree
->node
));
2577 log_root_tree
->log_transid
++;
2580 mutex_unlock(&log_root_tree
->log_mutex
);
2583 * nobody else is going to jump in and write the the ctree
2584 * super here because the log_commit atomic below is protecting
2585 * us. We must be called with a transaction handle pinning
2586 * the running transaction open, so a full commit can't hop
2587 * in and cause problems either.
2589 btrfs_scrub_pause_super(root
);
2590 ret
= write_ctree_super(trans
, root
->fs_info
->tree_root
, 1);
2591 btrfs_scrub_continue_super(root
);
2593 btrfs_abort_transaction(trans
, root
, ret
);
2594 goto out_wake_log_root
;
2597 mutex_lock(&root
->log_mutex
);
2598 if (root
->last_log_commit
< log_transid
)
2599 root
->last_log_commit
= log_transid
;
2600 mutex_unlock(&root
->log_mutex
);
2603 atomic_set(&log_root_tree
->log_commit
[index2
], 0);
2605 if (waitqueue_active(&log_root_tree
->log_commit_wait
[index2
]))
2606 wake_up(&log_root_tree
->log_commit_wait
[index2
]);
2608 atomic_set(&root
->log_commit
[index1
], 0);
2610 if (waitqueue_active(&root
->log_commit_wait
[index1
]))
2611 wake_up(&root
->log_commit_wait
[index1
]);
2615 static void free_log_tree(struct btrfs_trans_handle
*trans
,
2616 struct btrfs_root
*log
)
2621 struct walk_control wc
= {
2623 .process_func
= process_one_buffer
2626 ret
= walk_log_tree(trans
, log
, &wc
);
2627 /* I don't think this can happen but just in case */
2629 btrfs_abort_transaction(trans
, log
, ret
);
2632 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
2633 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
,
2638 clear_extent_bits(&log
->dirty_log_pages
, start
, end
,
2639 EXTENT_DIRTY
| EXTENT_NEW
, GFP_NOFS
);
2643 * We may have short-circuited the log tree with the full commit logic
2644 * and left ordered extents on our list, so clear these out to keep us
2645 * from leaking inodes and memory.
2647 btrfs_free_logged_extents(log
, 0);
2648 btrfs_free_logged_extents(log
, 1);
2650 free_extent_buffer(log
->node
);
2655 * free all the extents used by the tree log. This should be called
2656 * at commit time of the full transaction
2658 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2660 if (root
->log_root
) {
2661 free_log_tree(trans
, root
->log_root
);
2662 root
->log_root
= NULL
;
2667 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
2668 struct btrfs_fs_info
*fs_info
)
2670 if (fs_info
->log_root_tree
) {
2671 free_log_tree(trans
, fs_info
->log_root_tree
);
2672 fs_info
->log_root_tree
= NULL
;
2678 * If both a file and directory are logged, and unlinks or renames are
2679 * mixed in, we have a few interesting corners:
2681 * create file X in dir Y
2682 * link file X to X.link in dir Y
2684 * unlink file X but leave X.link
2687 * After a crash we would expect only X.link to exist. But file X
2688 * didn't get fsync'd again so the log has back refs for X and X.link.
2690 * We solve this by removing directory entries and inode backrefs from the
2691 * log when a file that was logged in the current transaction is
2692 * unlinked. Any later fsync will include the updated log entries, and
2693 * we'll be able to reconstruct the proper directory items from backrefs.
2695 * This optimizations allows us to avoid relogging the entire inode
2696 * or the entire directory.
2698 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
2699 struct btrfs_root
*root
,
2700 const char *name
, int name_len
,
2701 struct inode
*dir
, u64 index
)
2703 struct btrfs_root
*log
;
2704 struct btrfs_dir_item
*di
;
2705 struct btrfs_path
*path
;
2709 u64 dir_ino
= btrfs_ino(dir
);
2711 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
2714 ret
= join_running_log_trans(root
);
2718 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2720 log
= root
->log_root
;
2721 path
= btrfs_alloc_path();
2727 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir_ino
,
2728 name
, name_len
, -1);
2734 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2735 bytes_del
+= name_len
;
2741 btrfs_release_path(path
);
2742 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir_ino
,
2743 index
, name
, name_len
, -1);
2749 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2750 bytes_del
+= name_len
;
2757 /* update the directory size in the log to reflect the names
2761 struct btrfs_key key
;
2763 key
.objectid
= dir_ino
;
2765 key
.type
= BTRFS_INODE_ITEM_KEY
;
2766 btrfs_release_path(path
);
2768 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2774 struct btrfs_inode_item
*item
;
2777 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2778 struct btrfs_inode_item
);
2779 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2780 if (i_size
> bytes_del
)
2781 i_size
-= bytes_del
;
2784 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2785 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2788 btrfs_release_path(path
);
2791 btrfs_free_path(path
);
2793 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2794 if (ret
== -ENOSPC
) {
2795 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2798 btrfs_abort_transaction(trans
, root
, ret
);
2800 btrfs_end_log_trans(root
);
2805 /* see comments for btrfs_del_dir_entries_in_log */
2806 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2807 struct btrfs_root
*root
,
2808 const char *name
, int name_len
,
2809 struct inode
*inode
, u64 dirid
)
2811 struct btrfs_root
*log
;
2815 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
2818 ret
= join_running_log_trans(root
);
2821 log
= root
->log_root
;
2822 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2824 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, btrfs_ino(inode
),
2826 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2827 if (ret
== -ENOSPC
) {
2828 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2830 } else if (ret
< 0 && ret
!= -ENOENT
)
2831 btrfs_abort_transaction(trans
, root
, ret
);
2832 btrfs_end_log_trans(root
);
2838 * creates a range item in the log for 'dirid'. first_offset and
2839 * last_offset tell us which parts of the key space the log should
2840 * be considered authoritative for.
2842 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2843 struct btrfs_root
*log
,
2844 struct btrfs_path
*path
,
2845 int key_type
, u64 dirid
,
2846 u64 first_offset
, u64 last_offset
)
2849 struct btrfs_key key
;
2850 struct btrfs_dir_log_item
*item
;
2852 key
.objectid
= dirid
;
2853 key
.offset
= first_offset
;
2854 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2855 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2857 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2858 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2862 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2863 struct btrfs_dir_log_item
);
2864 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2865 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2866 btrfs_release_path(path
);
2871 * log all the items included in the current transaction for a given
2872 * directory. This also creates the range items in the log tree required
2873 * to replay anything deleted before the fsync
2875 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2876 struct btrfs_root
*root
, struct inode
*inode
,
2877 struct btrfs_path
*path
,
2878 struct btrfs_path
*dst_path
, int key_type
,
2879 u64 min_offset
, u64
*last_offset_ret
)
2881 struct btrfs_key min_key
;
2882 struct btrfs_root
*log
= root
->log_root
;
2883 struct extent_buffer
*src
;
2888 u64 first_offset
= min_offset
;
2889 u64 last_offset
= (u64
)-1;
2890 u64 ino
= btrfs_ino(inode
);
2892 log
= root
->log_root
;
2894 min_key
.objectid
= ino
;
2895 min_key
.type
= key_type
;
2896 min_key
.offset
= min_offset
;
2898 path
->keep_locks
= 1;
2900 ret
= btrfs_search_forward(root
, &min_key
, path
, trans
->transid
);
2903 * we didn't find anything from this transaction, see if there
2904 * is anything at all
2906 if (ret
!= 0 || min_key
.objectid
!= ino
|| min_key
.type
!= key_type
) {
2907 min_key
.objectid
= ino
;
2908 min_key
.type
= key_type
;
2909 min_key
.offset
= (u64
)-1;
2910 btrfs_release_path(path
);
2911 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2913 btrfs_release_path(path
);
2916 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2918 /* if ret == 0 there are items for this type,
2919 * create a range to tell us the last key of this type.
2920 * otherwise, there are no items in this directory after
2921 * *min_offset, and we create a range to indicate that.
2924 struct btrfs_key tmp
;
2925 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2927 if (key_type
== tmp
.type
)
2928 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2933 /* go backward to find any previous key */
2934 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2936 struct btrfs_key tmp
;
2937 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2938 if (key_type
== tmp
.type
) {
2939 first_offset
= tmp
.offset
;
2940 ret
= overwrite_item(trans
, log
, dst_path
,
2941 path
->nodes
[0], path
->slots
[0],
2949 btrfs_release_path(path
);
2951 /* find the first key from this transaction again */
2952 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2959 * we have a block from this transaction, log every item in it
2960 * from our directory
2963 struct btrfs_key tmp
;
2964 src
= path
->nodes
[0];
2965 nritems
= btrfs_header_nritems(src
);
2966 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2967 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2969 if (min_key
.objectid
!= ino
|| min_key
.type
!= key_type
)
2971 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2978 path
->slots
[0] = nritems
;
2981 * look ahead to the next item and see if it is also
2982 * from this directory and from this transaction
2984 ret
= btrfs_next_leaf(root
, path
);
2986 last_offset
= (u64
)-1;
2989 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2990 if (tmp
.objectid
!= ino
|| tmp
.type
!= key_type
) {
2991 last_offset
= (u64
)-1;
2994 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2995 ret
= overwrite_item(trans
, log
, dst_path
,
2996 path
->nodes
[0], path
->slots
[0],
3001 last_offset
= tmp
.offset
;
3006 btrfs_release_path(path
);
3007 btrfs_release_path(dst_path
);
3010 *last_offset_ret
= last_offset
;
3012 * insert the log range keys to indicate where the log
3015 ret
= insert_dir_log_key(trans
, log
, path
, key_type
,
3016 ino
, first_offset
, last_offset
);
3024 * logging directories is very similar to logging inodes, We find all the items
3025 * from the current transaction and write them to the log.
3027 * The recovery code scans the directory in the subvolume, and if it finds a
3028 * key in the range logged that is not present in the log tree, then it means
3029 * that dir entry was unlinked during the transaction.
3031 * In order for that scan to work, we must include one key smaller than
3032 * the smallest logged by this transaction and one key larger than the largest
3033 * key logged by this transaction.
3035 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
3036 struct btrfs_root
*root
, struct inode
*inode
,
3037 struct btrfs_path
*path
,
3038 struct btrfs_path
*dst_path
)
3043 int key_type
= BTRFS_DIR_ITEM_KEY
;
3049 ret
= log_dir_items(trans
, root
, inode
, path
,
3050 dst_path
, key_type
, min_key
,
3054 if (max_key
== (u64
)-1)
3056 min_key
= max_key
+ 1;
3059 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
3060 key_type
= BTRFS_DIR_INDEX_KEY
;
3067 * a helper function to drop items from the log before we relog an
3068 * inode. max_key_type indicates the highest item type to remove.
3069 * This cannot be run for file data extents because it does not
3070 * free the extents they point to.
3072 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
3073 struct btrfs_root
*log
,
3074 struct btrfs_path
*path
,
3075 u64 objectid
, int max_key_type
)
3078 struct btrfs_key key
;
3079 struct btrfs_key found_key
;
3082 key
.objectid
= objectid
;
3083 key
.type
= max_key_type
;
3084 key
.offset
= (u64
)-1;
3087 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
3088 BUG_ON(ret
== 0); /* Logic error */
3092 if (path
->slots
[0] == 0)
3096 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
3099 if (found_key
.objectid
!= objectid
)
3102 found_key
.offset
= 0;
3104 ret
= btrfs_bin_search(path
->nodes
[0], &found_key
, 0,
3107 ret
= btrfs_del_items(trans
, log
, path
, start_slot
,
3108 path
->slots
[0] - start_slot
+ 1);
3110 * If start slot isn't 0 then we don't need to re-search, we've
3111 * found the last guy with the objectid in this tree.
3113 if (ret
|| start_slot
!= 0)
3115 btrfs_release_path(path
);
3117 btrfs_release_path(path
);
3123 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3124 struct extent_buffer
*leaf
,
3125 struct btrfs_inode_item
*item
,
3126 struct inode
*inode
, int log_inode_only
)
3128 struct btrfs_map_token token
;
3130 btrfs_init_map_token(&token
);
3132 if (log_inode_only
) {
3133 /* set the generation to zero so the recover code
3134 * can tell the difference between an logging
3135 * just to say 'this inode exists' and a logging
3136 * to say 'update this inode with these values'
3138 btrfs_set_token_inode_generation(leaf
, item
, 0, &token
);
3139 btrfs_set_token_inode_size(leaf
, item
, 0, &token
);
3141 btrfs_set_token_inode_generation(leaf
, item
,
3142 BTRFS_I(inode
)->generation
,
3144 btrfs_set_token_inode_size(leaf
, item
, inode
->i_size
, &token
);
3147 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3148 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3149 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3150 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3152 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_atime(item
),
3153 inode
->i_atime
.tv_sec
, &token
);
3154 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_atime(item
),
3155 inode
->i_atime
.tv_nsec
, &token
);
3157 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_mtime(item
),
3158 inode
->i_mtime
.tv_sec
, &token
);
3159 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
3160 inode
->i_mtime
.tv_nsec
, &token
);
3162 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_ctime(item
),
3163 inode
->i_ctime
.tv_sec
, &token
);
3164 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
3165 inode
->i_ctime
.tv_nsec
, &token
);
3167 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3170 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3171 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3172 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3173 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3174 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3177 static int log_inode_item(struct btrfs_trans_handle
*trans
,
3178 struct btrfs_root
*log
, struct btrfs_path
*path
,
3179 struct inode
*inode
)
3181 struct btrfs_inode_item
*inode_item
;
3184 ret
= btrfs_insert_empty_item(trans
, log
, path
,
3185 &BTRFS_I(inode
)->location
,
3186 sizeof(*inode_item
));
3187 if (ret
&& ret
!= -EEXIST
)
3189 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3190 struct btrfs_inode_item
);
3191 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
, 0);
3192 btrfs_release_path(path
);
3196 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
3197 struct inode
*inode
,
3198 struct btrfs_path
*dst_path
,
3199 struct extent_buffer
*src
,
3200 int start_slot
, int nr
, int inode_only
)
3202 unsigned long src_offset
;
3203 unsigned long dst_offset
;
3204 struct btrfs_root
*log
= BTRFS_I(inode
)->root
->log_root
;
3205 struct btrfs_file_extent_item
*extent
;
3206 struct btrfs_inode_item
*inode_item
;
3208 struct btrfs_key
*ins_keys
;
3212 struct list_head ordered_sums
;
3213 int skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
3215 INIT_LIST_HEAD(&ordered_sums
);
3217 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
3218 nr
* sizeof(u32
), GFP_NOFS
);
3222 ins_sizes
= (u32
*)ins_data
;
3223 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
3225 for (i
= 0; i
< nr
; i
++) {
3226 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
3227 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
3229 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
3230 ins_keys
, ins_sizes
, nr
);
3236 for (i
= 0; i
< nr
; i
++, dst_path
->slots
[0]++) {
3237 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
3238 dst_path
->slots
[0]);
3240 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
3242 if (ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
3243 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
3245 struct btrfs_inode_item
);
3246 fill_inode_item(trans
, dst_path
->nodes
[0], inode_item
,
3247 inode
, inode_only
== LOG_INODE_EXISTS
);
3249 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
3250 src_offset
, ins_sizes
[i
]);
3253 /* take a reference on file data extents so that truncates
3254 * or deletes of this inode don't have to relog the inode
3257 if (btrfs_key_type(ins_keys
+ i
) == BTRFS_EXTENT_DATA_KEY
&&
3260 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
3261 struct btrfs_file_extent_item
);
3263 if (btrfs_file_extent_generation(src
, extent
) < trans
->transid
)
3266 found_type
= btrfs_file_extent_type(src
, extent
);
3267 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
3269 ds
= btrfs_file_extent_disk_bytenr(src
,
3271 /* ds == 0 is a hole */
3275 dl
= btrfs_file_extent_disk_num_bytes(src
,
3277 cs
= btrfs_file_extent_offset(src
, extent
);
3278 cl
= btrfs_file_extent_num_bytes(src
,
3280 if (btrfs_file_extent_compression(src
,
3286 ret
= btrfs_lookup_csums_range(
3287 log
->fs_info
->csum_root
,
3288 ds
+ cs
, ds
+ cs
+ cl
- 1,
3291 btrfs_release_path(dst_path
);
3299 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
3300 btrfs_release_path(dst_path
);
3304 * we have to do this after the loop above to avoid changing the
3305 * log tree while trying to change the log tree.
3308 while (!list_empty(&ordered_sums
)) {
3309 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
3310 struct btrfs_ordered_sum
,
3313 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
3314 list_del(&sums
->list
);
3320 static int extent_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3322 struct extent_map
*em1
, *em2
;
3324 em1
= list_entry(a
, struct extent_map
, list
);
3325 em2
= list_entry(b
, struct extent_map
, list
);
3327 if (em1
->start
< em2
->start
)
3329 else if (em1
->start
> em2
->start
)
3334 static int log_one_extent(struct btrfs_trans_handle
*trans
,
3335 struct inode
*inode
, struct btrfs_root
*root
,
3336 struct extent_map
*em
, struct btrfs_path
*path
)
3338 struct btrfs_root
*log
= root
->log_root
;
3339 struct btrfs_file_extent_item
*fi
;
3340 struct extent_buffer
*leaf
;
3341 struct btrfs_ordered_extent
*ordered
;
3342 struct list_head ordered_sums
;
3343 struct btrfs_map_token token
;
3344 struct btrfs_key key
;
3345 u64 mod_start
= em
->mod_start
;
3346 u64 mod_len
= em
->mod_len
;
3349 u64 extent_offset
= em
->start
- em
->orig_start
;
3352 int index
= log
->log_transid
% 2;
3353 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
3355 ret
= __btrfs_drop_extents(trans
, log
, inode
, path
, em
->start
,
3356 em
->start
+ em
->len
, NULL
, 0);
3360 INIT_LIST_HEAD(&ordered_sums
);
3361 btrfs_init_map_token(&token
);
3362 key
.objectid
= btrfs_ino(inode
);
3363 key
.type
= BTRFS_EXTENT_DATA_KEY
;
3364 key
.offset
= em
->start
;
3366 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*fi
));
3369 leaf
= path
->nodes
[0];
3370 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3371 struct btrfs_file_extent_item
);
3373 btrfs_set_token_file_extent_generation(leaf
, fi
, em
->generation
,
3375 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3377 btrfs_set_token_file_extent_type(leaf
, fi
,
3378 BTRFS_FILE_EXTENT_PREALLOC
,
3381 btrfs_set_token_file_extent_type(leaf
, fi
,
3382 BTRFS_FILE_EXTENT_REG
,
3384 if (em
->block_start
== EXTENT_MAP_HOLE
)
3388 block_len
= max(em
->block_len
, em
->orig_block_len
);
3389 if (em
->compress_type
!= BTRFS_COMPRESS_NONE
) {
3390 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
,
3393 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, block_len
,
3395 } else if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
3396 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
,
3398 extent_offset
, &token
);
3399 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, block_len
,
3402 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
, 0, &token
);
3403 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, 0,
3407 btrfs_set_token_file_extent_offset(leaf
, fi
,
3408 em
->start
- em
->orig_start
,
3410 btrfs_set_token_file_extent_num_bytes(leaf
, fi
, em
->len
, &token
);
3411 btrfs_set_token_file_extent_ram_bytes(leaf
, fi
, em
->ram_bytes
, &token
);
3412 btrfs_set_token_file_extent_compression(leaf
, fi
, em
->compress_type
,
3414 btrfs_set_token_file_extent_encryption(leaf
, fi
, 0, &token
);
3415 btrfs_set_token_file_extent_other_encoding(leaf
, fi
, 0, &token
);
3416 btrfs_mark_buffer_dirty(leaf
);
3418 btrfs_release_path(path
);
3426 if (em
->compress_type
) {
3428 csum_len
= block_len
;
3432 * First check and see if our csums are on our outstanding ordered
3436 spin_lock_irq(&log
->log_extents_lock
[index
]);
3437 list_for_each_entry(ordered
, &log
->logged_list
[index
], log_list
) {
3438 struct btrfs_ordered_sum
*sum
;
3443 if (ordered
->inode
!= inode
)
3446 if (ordered
->file_offset
+ ordered
->len
<= mod_start
||
3447 mod_start
+ mod_len
<= ordered
->file_offset
)
3451 * We are going to copy all the csums on this ordered extent, so
3452 * go ahead and adjust mod_start and mod_len in case this
3453 * ordered extent has already been logged.
3455 if (ordered
->file_offset
> mod_start
) {
3456 if (ordered
->file_offset
+ ordered
->len
>=
3457 mod_start
+ mod_len
)
3458 mod_len
= ordered
->file_offset
- mod_start
;
3460 * If we have this case
3462 * |--------- logged extent ---------|
3463 * |----- ordered extent ----|
3465 * Just don't mess with mod_start and mod_len, we'll
3466 * just end up logging more csums than we need and it
3470 if (ordered
->file_offset
+ ordered
->len
<
3471 mod_start
+ mod_len
) {
3472 mod_len
= (mod_start
+ mod_len
) -
3473 (ordered
->file_offset
+ ordered
->len
);
3474 mod_start
= ordered
->file_offset
+
3482 * To keep us from looping for the above case of an ordered
3483 * extent that falls inside of the logged extent.
3485 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM
,
3488 atomic_inc(&ordered
->refs
);
3489 spin_unlock_irq(&log
->log_extents_lock
[index
]);
3491 * we've dropped the lock, we must either break or
3492 * start over after this.
3495 wait_event(ordered
->wait
, ordered
->csum_bytes_left
== 0);
3497 list_for_each_entry(sum
, &ordered
->list
, list
) {
3498 ret
= btrfs_csum_file_blocks(trans
, log
, sum
);
3500 btrfs_put_ordered_extent(ordered
);
3504 btrfs_put_ordered_extent(ordered
);
3508 spin_unlock_irq(&log
->log_extents_lock
[index
]);
3511 if (!mod_len
|| ret
)
3514 csum_offset
= mod_start
- em
->start
;
3517 /* block start is already adjusted for the file extent offset. */
3518 ret
= btrfs_lookup_csums_range(log
->fs_info
->csum_root
,
3519 em
->block_start
+ csum_offset
,
3520 em
->block_start
+ csum_offset
+
3521 csum_len
- 1, &ordered_sums
, 0);
3525 while (!list_empty(&ordered_sums
)) {
3526 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
3527 struct btrfs_ordered_sum
,
3530 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
3531 list_del(&sums
->list
);
3538 static int btrfs_log_changed_extents(struct btrfs_trans_handle
*trans
,
3539 struct btrfs_root
*root
,
3540 struct inode
*inode
,
3541 struct btrfs_path
*path
)
3543 struct extent_map
*em
, *n
;
3544 struct list_head extents
;
3545 struct extent_map_tree
*tree
= &BTRFS_I(inode
)->extent_tree
;
3550 INIT_LIST_HEAD(&extents
);
3552 write_lock(&tree
->lock
);
3553 test_gen
= root
->fs_info
->last_trans_committed
;
3555 list_for_each_entry_safe(em
, n
, &tree
->modified_extents
, list
) {
3556 list_del_init(&em
->list
);
3559 * Just an arbitrary number, this can be really CPU intensive
3560 * once we start getting a lot of extents, and really once we
3561 * have a bunch of extents we just want to commit since it will
3564 if (++num
> 32768) {
3565 list_del_init(&tree
->modified_extents
);
3570 if (em
->generation
<= test_gen
)
3572 /* Need a ref to keep it from getting evicted from cache */
3573 atomic_inc(&em
->refs
);
3574 set_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
3575 list_add_tail(&em
->list
, &extents
);
3579 list_sort(NULL
, &extents
, extent_cmp
);
3582 while (!list_empty(&extents
)) {
3583 em
= list_entry(extents
.next
, struct extent_map
, list
);
3585 list_del_init(&em
->list
);
3588 * If we had an error we just need to delete everybody from our
3592 clear_em_logging(tree
, em
);
3593 free_extent_map(em
);
3597 write_unlock(&tree
->lock
);
3599 ret
= log_one_extent(trans
, inode
, root
, em
, path
);
3600 write_lock(&tree
->lock
);
3601 clear_em_logging(tree
, em
);
3602 free_extent_map(em
);
3604 WARN_ON(!list_empty(&extents
));
3605 write_unlock(&tree
->lock
);
3607 btrfs_release_path(path
);
3611 /* log a single inode in the tree log.
3612 * At least one parent directory for this inode must exist in the tree
3613 * or be logged already.
3615 * Any items from this inode changed by the current transaction are copied
3616 * to the log tree. An extra reference is taken on any extents in this
3617 * file, allowing us to avoid a whole pile of corner cases around logging
3618 * blocks that have been removed from the tree.
3620 * See LOG_INODE_ALL and related defines for a description of what inode_only
3623 * This handles both files and directories.
3625 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
3626 struct btrfs_root
*root
, struct inode
*inode
,
3629 struct btrfs_path
*path
;
3630 struct btrfs_path
*dst_path
;
3631 struct btrfs_key min_key
;
3632 struct btrfs_key max_key
;
3633 struct btrfs_root
*log
= root
->log_root
;
3634 struct extent_buffer
*src
= NULL
;
3638 int ins_start_slot
= 0;
3640 bool fast_search
= false;
3641 u64 ino
= btrfs_ino(inode
);
3643 path
= btrfs_alloc_path();
3646 dst_path
= btrfs_alloc_path();
3648 btrfs_free_path(path
);
3652 min_key
.objectid
= ino
;
3653 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
3656 max_key
.objectid
= ino
;
3659 /* today the code can only do partial logging of directories */
3660 if (S_ISDIR(inode
->i_mode
) ||
3661 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3662 &BTRFS_I(inode
)->runtime_flags
) &&
3663 inode_only
== LOG_INODE_EXISTS
))
3664 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
3666 max_key
.type
= (u8
)-1;
3667 max_key
.offset
= (u64
)-1;
3669 /* Only run delayed items if we are a dir or a new file */
3670 if (S_ISDIR(inode
->i_mode
) ||
3671 BTRFS_I(inode
)->generation
> root
->fs_info
->last_trans_committed
) {
3672 ret
= btrfs_commit_inode_delayed_items(trans
, inode
);
3674 btrfs_free_path(path
);
3675 btrfs_free_path(dst_path
);
3680 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
3682 btrfs_get_logged_extents(log
, inode
);
3685 * a brute force approach to making sure we get the most uptodate
3686 * copies of everything.
3688 if (S_ISDIR(inode
->i_mode
)) {
3689 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
3691 if (inode_only
== LOG_INODE_EXISTS
)
3692 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
3693 ret
= drop_objectid_items(trans
, log
, path
, ino
, max_key_type
);
3695 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3696 &BTRFS_I(inode
)->runtime_flags
)) {
3697 clear_bit(BTRFS_INODE_COPY_EVERYTHING
,
3698 &BTRFS_I(inode
)->runtime_flags
);
3699 ret
= btrfs_truncate_inode_items(trans
, log
,
3701 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING
,
3702 &BTRFS_I(inode
)->runtime_flags
)) {
3703 if (inode_only
== LOG_INODE_ALL
)
3705 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
3706 ret
= drop_objectid_items(trans
, log
, path
, ino
,
3709 if (inode_only
== LOG_INODE_ALL
)
3711 ret
= log_inode_item(trans
, log
, dst_path
, inode
);
3724 path
->keep_locks
= 1;
3728 ret
= btrfs_search_forward(root
, &min_key
,
3729 path
, trans
->transid
);
3733 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3734 if (min_key
.objectid
!= ino
)
3736 if (min_key
.type
> max_key
.type
)
3739 src
= path
->nodes
[0];
3740 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
3743 } else if (!ins_nr
) {
3744 ins_start_slot
= path
->slots
[0];
3749 ret
= copy_items(trans
, inode
, dst_path
, src
, ins_start_slot
,
3750 ins_nr
, inode_only
);
3756 ins_start_slot
= path
->slots
[0];
3759 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3761 if (path
->slots
[0] < nritems
) {
3762 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
3767 ret
= copy_items(trans
, inode
, dst_path
, src
,
3769 ins_nr
, inode_only
);
3776 btrfs_release_path(path
);
3778 if (min_key
.offset
< (u64
)-1) {
3780 } else if (min_key
.type
< max_key
.type
) {
3788 ret
= copy_items(trans
, inode
, dst_path
, src
, ins_start_slot
,
3789 ins_nr
, inode_only
);
3798 btrfs_release_path(path
);
3799 btrfs_release_path(dst_path
);
3801 ret
= btrfs_log_changed_extents(trans
, root
, inode
, dst_path
);
3807 struct extent_map_tree
*tree
= &BTRFS_I(inode
)->extent_tree
;
3808 struct extent_map
*em
, *n
;
3810 write_lock(&tree
->lock
);
3811 list_for_each_entry_safe(em
, n
, &tree
->modified_extents
, list
)
3812 list_del_init(&em
->list
);
3813 write_unlock(&tree
->lock
);
3816 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
3817 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
3823 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
3824 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->last_sub_trans
;
3827 btrfs_free_logged_extents(log
, log
->log_transid
);
3828 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
3830 btrfs_free_path(path
);
3831 btrfs_free_path(dst_path
);
3836 * follow the dentry parent pointers up the chain and see if any
3837 * of the directories in it require a full commit before they can
3838 * be logged. Returns zero if nothing special needs to be done or 1 if
3839 * a full commit is required.
3841 static noinline
int check_parent_dirs_for_sync(struct btrfs_trans_handle
*trans
,
3842 struct inode
*inode
,
3843 struct dentry
*parent
,
3844 struct super_block
*sb
,
3848 struct btrfs_root
*root
;
3849 struct dentry
*old_parent
= NULL
;
3850 struct inode
*orig_inode
= inode
;
3853 * for regular files, if its inode is already on disk, we don't
3854 * have to worry about the parents at all. This is because
3855 * we can use the last_unlink_trans field to record renames
3856 * and other fun in this file.
3858 if (S_ISREG(inode
->i_mode
) &&
3859 BTRFS_I(inode
)->generation
<= last_committed
&&
3860 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
)
3863 if (!S_ISDIR(inode
->i_mode
)) {
3864 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3866 inode
= parent
->d_inode
;
3871 * If we are logging a directory then we start with our inode,
3872 * not our parents inode, so we need to skipp setting the
3873 * logged_trans so that further down in the log code we don't
3874 * think this inode has already been logged.
3876 if (inode
!= orig_inode
)
3877 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
3880 if (BTRFS_I(inode
)->last_unlink_trans
> last_committed
) {
3881 root
= BTRFS_I(inode
)->root
;
3884 * make sure any commits to the log are forced
3885 * to be full commits
3887 root
->fs_info
->last_trans_log_full_commit
=
3893 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3896 if (IS_ROOT(parent
))
3899 parent
= dget_parent(parent
);
3901 old_parent
= parent
;
3902 inode
= parent
->d_inode
;
3911 * helper function around btrfs_log_inode to make sure newly created
3912 * parent directories also end up in the log. A minimal inode and backref
3913 * only logging is done of any parent directories that are older than
3914 * the last committed transaction
3916 static int btrfs_log_inode_parent(struct btrfs_trans_handle
*trans
,
3917 struct btrfs_root
*root
, struct inode
*inode
,
3918 struct dentry
*parent
, int exists_only
)
3920 int inode_only
= exists_only
? LOG_INODE_EXISTS
: LOG_INODE_ALL
;
3921 struct super_block
*sb
;
3922 struct dentry
*old_parent
= NULL
;
3924 u64 last_committed
= root
->fs_info
->last_trans_committed
;
3928 if (btrfs_test_opt(root
, NOTREELOG
)) {
3933 if (root
->fs_info
->last_trans_log_full_commit
>
3934 root
->fs_info
->last_trans_committed
) {
3939 if (root
!= BTRFS_I(inode
)->root
||
3940 btrfs_root_refs(&root
->root_item
) == 0) {
3945 ret
= check_parent_dirs_for_sync(trans
, inode
, parent
,
3946 sb
, last_committed
);
3950 if (btrfs_inode_in_log(inode
, trans
->transid
)) {
3951 ret
= BTRFS_NO_LOG_SYNC
;
3955 ret
= start_log_trans(trans
, root
);
3959 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3964 * for regular files, if its inode is already on disk, we don't
3965 * have to worry about the parents at all. This is because
3966 * we can use the last_unlink_trans field to record renames
3967 * and other fun in this file.
3969 if (S_ISREG(inode
->i_mode
) &&
3970 BTRFS_I(inode
)->generation
<= last_committed
&&
3971 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
) {
3976 inode_only
= LOG_INODE_EXISTS
;
3978 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3981 inode
= parent
->d_inode
;
3982 if (root
!= BTRFS_I(inode
)->root
)
3985 if (BTRFS_I(inode
)->generation
>
3986 root
->fs_info
->last_trans_committed
) {
3987 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3991 if (IS_ROOT(parent
))
3994 parent
= dget_parent(parent
);
3996 old_parent
= parent
;
4002 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
4005 btrfs_end_log_trans(root
);
4011 * it is not safe to log dentry if the chunk root has added new
4012 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
4013 * If this returns 1, you must commit the transaction to safely get your
4016 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
4017 struct btrfs_root
*root
, struct dentry
*dentry
)
4019 struct dentry
*parent
= dget_parent(dentry
);
4022 ret
= btrfs_log_inode_parent(trans
, root
, dentry
->d_inode
, parent
, 0);
4029 * should be called during mount to recover any replay any log trees
4032 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
4035 struct btrfs_path
*path
;
4036 struct btrfs_trans_handle
*trans
;
4037 struct btrfs_key key
;
4038 struct btrfs_key found_key
;
4039 struct btrfs_key tmp_key
;
4040 struct btrfs_root
*log
;
4041 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
4042 struct walk_control wc
= {
4043 .process_func
= process_one_buffer
,
4047 path
= btrfs_alloc_path();
4051 fs_info
->log_root_recovering
= 1;
4053 trans
= btrfs_start_transaction(fs_info
->tree_root
, 0);
4054 if (IS_ERR(trans
)) {
4055 ret
= PTR_ERR(trans
);
4062 ret
= walk_log_tree(trans
, log_root_tree
, &wc
);
4064 btrfs_error(fs_info
, ret
, "Failed to pin buffers while "
4065 "recovering log root tree.");
4070 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
4071 key
.offset
= (u64
)-1;
4072 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
4075 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
4078 btrfs_error(fs_info
, ret
,
4079 "Couldn't find tree log root.");
4083 if (path
->slots
[0] == 0)
4087 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
4089 btrfs_release_path(path
);
4090 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4093 log
= btrfs_read_fs_root(log_root_tree
, &found_key
);
4096 btrfs_error(fs_info
, ret
,
4097 "Couldn't read tree log root.");
4101 tmp_key
.objectid
= found_key
.offset
;
4102 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4103 tmp_key
.offset
= (u64
)-1;
4105 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
4106 if (IS_ERR(wc
.replay_dest
)) {
4107 ret
= PTR_ERR(wc
.replay_dest
);
4108 free_extent_buffer(log
->node
);
4109 free_extent_buffer(log
->commit_root
);
4111 btrfs_error(fs_info
, ret
, "Couldn't read target root "
4112 "for tree log recovery.");
4116 wc
.replay_dest
->log_root
= log
;
4117 btrfs_record_root_in_trans(trans
, wc
.replay_dest
);
4118 ret
= walk_log_tree(trans
, log
, &wc
);
4120 if (!ret
&& wc
.stage
== LOG_WALK_REPLAY_ALL
) {
4121 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
4125 key
.offset
= found_key
.offset
- 1;
4126 wc
.replay_dest
->log_root
= NULL
;
4127 free_extent_buffer(log
->node
);
4128 free_extent_buffer(log
->commit_root
);
4134 if (found_key
.offset
== 0)
4137 btrfs_release_path(path
);
4139 /* step one is to pin it all, step two is to replay just inodes */
4142 wc
.process_func
= replay_one_buffer
;
4143 wc
.stage
= LOG_WALK_REPLAY_INODES
;
4146 /* step three is to replay everything */
4147 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
4152 btrfs_free_path(path
);
4154 /* step 4: commit the transaction, which also unpins the blocks */
4155 ret
= btrfs_commit_transaction(trans
, fs_info
->tree_root
);
4159 free_extent_buffer(log_root_tree
->node
);
4160 log_root_tree
->log_root
= NULL
;
4161 fs_info
->log_root_recovering
= 0;
4162 kfree(log_root_tree
);
4167 btrfs_end_transaction(wc
.trans
, fs_info
->tree_root
);
4168 btrfs_free_path(path
);
4173 * there are some corner cases where we want to force a full
4174 * commit instead of allowing a directory to be logged.
4176 * They revolve around files there were unlinked from the directory, and
4177 * this function updates the parent directory so that a full commit is
4178 * properly done if it is fsync'd later after the unlinks are done.
4180 void btrfs_record_unlink_dir(struct btrfs_trans_handle
*trans
,
4181 struct inode
*dir
, struct inode
*inode
,
4185 * when we're logging a file, if it hasn't been renamed
4186 * or unlinked, and its inode is fully committed on disk,
4187 * we don't have to worry about walking up the directory chain
4188 * to log its parents.
4190 * So, we use the last_unlink_trans field to put this transid
4191 * into the file. When the file is logged we check it and
4192 * don't log the parents if the file is fully on disk.
4194 if (S_ISREG(inode
->i_mode
))
4195 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
4198 * if this directory was already logged any new
4199 * names for this file/dir will get recorded
4202 if (BTRFS_I(dir
)->logged_trans
== trans
->transid
)
4206 * if the inode we're about to unlink was logged,
4207 * the log will be properly updated for any new names
4209 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
)
4213 * when renaming files across directories, if the directory
4214 * there we're unlinking from gets fsync'd later on, there's
4215 * no way to find the destination directory later and fsync it
4216 * properly. So, we have to be conservative and force commits
4217 * so the new name gets discovered.
4222 /* we can safely do the unlink without any special recording */
4226 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
4230 * Call this after adding a new name for a file and it will properly
4231 * update the log to reflect the new name.
4233 * It will return zero if all goes well, and it will return 1 if a
4234 * full transaction commit is required.
4236 int btrfs_log_new_name(struct btrfs_trans_handle
*trans
,
4237 struct inode
*inode
, struct inode
*old_dir
,
4238 struct dentry
*parent
)
4240 struct btrfs_root
* root
= BTRFS_I(inode
)->root
;
4243 * this will force the logging code to walk the dentry chain
4246 if (S_ISREG(inode
->i_mode
))
4247 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
4250 * if this inode hasn't been logged and directory we're renaming it
4251 * from hasn't been logged, we don't need to log it
4253 if (BTRFS_I(inode
)->logged_trans
<=
4254 root
->fs_info
->last_trans_committed
&&
4255 (!old_dir
|| BTRFS_I(old_dir
)->logged_trans
<=
4256 root
->fs_info
->last_trans_committed
))
4259 return btrfs_log_inode_parent(trans
, root
, inode
, parent
, 1);