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"
33 /* magic values for the inode_only field in btrfs_log_inode:
35 * LOG_INODE_ALL means to log everything
36 * LOG_INODE_EXISTS means to log just enough to recreate the inode
39 #define LOG_INODE_ALL 0
40 #define LOG_INODE_EXISTS 1
43 * directory trouble cases
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
52 * rename foo/some_dir foo2/some_dir
54 * fsync foo/some_dir/some_file
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
86 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
94 #define LOG_WALK_PIN_ONLY 0
95 #define LOG_WALK_REPLAY_INODES 1
96 #define LOG_WALK_REPLAY_ALL 2
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
;
397 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
398 struct btrfs_inode_item
);
399 nbytes
= btrfs_inode_nbytes(path
->nodes
[0], item
);
400 item
= btrfs_item_ptr(eb
, slot
,
401 struct btrfs_inode_item
);
402 btrfs_set_inode_nbytes(eb
, item
, nbytes
);
404 } else if (inode_item
) {
405 struct btrfs_inode_item
*item
;
408 * New inode, set nbytes to 0 so that the nbytes comes out
409 * properly when we replay the extents.
411 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
412 btrfs_set_inode_nbytes(eb
, item
, 0);
415 btrfs_release_path(path
);
416 /* try to insert the key into the destination tree */
417 ret
= btrfs_insert_empty_item(trans
, root
, path
,
420 /* make sure any existing item is the correct size */
421 if (ret
== -EEXIST
) {
423 found_size
= btrfs_item_size_nr(path
->nodes
[0],
425 if (found_size
> item_size
)
426 btrfs_truncate_item(root
, path
, item_size
, 1);
427 else if (found_size
< item_size
)
428 btrfs_extend_item(root
, path
,
429 item_size
- found_size
);
433 dst_ptr
= btrfs_item_ptr_offset(path
->nodes
[0],
436 /* don't overwrite an existing inode if the generation number
437 * was logged as zero. This is done when the tree logging code
438 * is just logging an inode to make sure it exists after recovery.
440 * Also, don't overwrite i_size on directories during replay.
441 * log replay inserts and removes directory items based on the
442 * state of the tree found in the subvolume, and i_size is modified
445 if (key
->type
== BTRFS_INODE_ITEM_KEY
&& ret
== -EEXIST
) {
446 struct btrfs_inode_item
*src_item
;
447 struct btrfs_inode_item
*dst_item
;
449 src_item
= (struct btrfs_inode_item
*)src_ptr
;
450 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
452 if (btrfs_inode_generation(eb
, src_item
) == 0)
455 if (overwrite_root
&&
456 S_ISDIR(btrfs_inode_mode(eb
, src_item
)) &&
457 S_ISDIR(btrfs_inode_mode(path
->nodes
[0], dst_item
))) {
459 saved_i_size
= btrfs_inode_size(path
->nodes
[0],
464 copy_extent_buffer(path
->nodes
[0], eb
, dst_ptr
,
467 if (save_old_i_size
) {
468 struct btrfs_inode_item
*dst_item
;
469 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
470 btrfs_set_inode_size(path
->nodes
[0], dst_item
, saved_i_size
);
473 /* make sure the generation is filled in */
474 if (key
->type
== BTRFS_INODE_ITEM_KEY
) {
475 struct btrfs_inode_item
*dst_item
;
476 dst_item
= (struct btrfs_inode_item
*)dst_ptr
;
477 if (btrfs_inode_generation(path
->nodes
[0], dst_item
) == 0) {
478 btrfs_set_inode_generation(path
->nodes
[0], dst_item
,
483 btrfs_mark_buffer_dirty(path
->nodes
[0]);
484 btrfs_release_path(path
);
489 * simple helper to read an inode off the disk from a given root
490 * This can only be called for subvolume roots and not for the log
492 static noinline
struct inode
*read_one_inode(struct btrfs_root
*root
,
495 struct btrfs_key key
;
498 key
.objectid
= objectid
;
499 key
.type
= BTRFS_INODE_ITEM_KEY
;
501 inode
= btrfs_iget(root
->fs_info
->sb
, &key
, root
, NULL
);
504 } else if (is_bad_inode(inode
)) {
511 /* replays a single extent in 'eb' at 'slot' with 'key' into the
512 * subvolume 'root'. path is released on entry and should be released
515 * extents in the log tree have not been allocated out of the extent
516 * tree yet. So, this completes the allocation, taking a reference
517 * as required if the extent already exists or creating a new extent
518 * if it isn't in the extent allocation tree yet.
520 * The extent is inserted into the file, dropping any existing extents
521 * from the file that overlap the new one.
523 static noinline
int replay_one_extent(struct btrfs_trans_handle
*trans
,
524 struct btrfs_root
*root
,
525 struct btrfs_path
*path
,
526 struct extent_buffer
*eb
, int slot
,
527 struct btrfs_key
*key
)
531 u64 start
= key
->offset
;
533 struct btrfs_file_extent_item
*item
;
534 struct inode
*inode
= NULL
;
538 item
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
539 found_type
= btrfs_file_extent_type(eb
, item
);
541 if (found_type
== BTRFS_FILE_EXTENT_REG
||
542 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
543 nbytes
= btrfs_file_extent_num_bytes(eb
, item
);
544 extent_end
= start
+ nbytes
;
547 * We don't add to the inodes nbytes if we are prealloc or a
550 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
552 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
553 size
= btrfs_file_extent_inline_len(eb
, item
);
554 nbytes
= btrfs_file_extent_ram_bytes(eb
, item
);
555 extent_end
= ALIGN(start
+ size
, root
->sectorsize
);
561 inode
= read_one_inode(root
, key
->objectid
);
568 * first check to see if we already have this extent in the
569 * file. This must be done before the btrfs_drop_extents run
570 * so we don't try to drop this extent.
572 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
576 (found_type
== BTRFS_FILE_EXTENT_REG
||
577 found_type
== BTRFS_FILE_EXTENT_PREALLOC
)) {
578 struct btrfs_file_extent_item cmp1
;
579 struct btrfs_file_extent_item cmp2
;
580 struct btrfs_file_extent_item
*existing
;
581 struct extent_buffer
*leaf
;
583 leaf
= path
->nodes
[0];
584 existing
= btrfs_item_ptr(leaf
, path
->slots
[0],
585 struct btrfs_file_extent_item
);
587 read_extent_buffer(eb
, &cmp1
, (unsigned long)item
,
589 read_extent_buffer(leaf
, &cmp2
, (unsigned long)existing
,
593 * we already have a pointer to this exact extent,
594 * we don't have to do anything
596 if (memcmp(&cmp1
, &cmp2
, sizeof(cmp1
)) == 0) {
597 btrfs_release_path(path
);
601 btrfs_release_path(path
);
603 /* drop any overlapping extents */
604 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, extent_end
, 1);
608 if (found_type
== BTRFS_FILE_EXTENT_REG
||
609 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
611 unsigned long dest_offset
;
612 struct btrfs_key ins
;
614 ret
= btrfs_insert_empty_item(trans
, root
, path
, key
,
618 dest_offset
= btrfs_item_ptr_offset(path
->nodes
[0],
620 copy_extent_buffer(path
->nodes
[0], eb
, dest_offset
,
621 (unsigned long)item
, sizeof(*item
));
623 ins
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
624 ins
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
625 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
626 offset
= key
->offset
- btrfs_file_extent_offset(eb
, item
);
628 if (ins
.objectid
> 0) {
631 LIST_HEAD(ordered_sums
);
633 * is this extent already allocated in the extent
634 * allocation tree? If so, just add a reference
636 ret
= btrfs_lookup_extent(root
, ins
.objectid
,
639 ret
= btrfs_inc_extent_ref(trans
, root
,
640 ins
.objectid
, ins
.offset
,
641 0, root
->root_key
.objectid
,
642 key
->objectid
, offset
, 0);
647 * insert the extent pointer in the extent
650 ret
= btrfs_alloc_logged_file_extent(trans
,
651 root
, root
->root_key
.objectid
,
652 key
->objectid
, offset
, &ins
);
656 btrfs_release_path(path
);
658 if (btrfs_file_extent_compression(eb
, item
)) {
659 csum_start
= ins
.objectid
;
660 csum_end
= csum_start
+ ins
.offset
;
662 csum_start
= ins
.objectid
+
663 btrfs_file_extent_offset(eb
, item
);
664 csum_end
= csum_start
+
665 btrfs_file_extent_num_bytes(eb
, item
);
668 ret
= btrfs_lookup_csums_range(root
->log_root
,
669 csum_start
, csum_end
- 1,
673 while (!list_empty(&ordered_sums
)) {
674 struct btrfs_ordered_sum
*sums
;
675 sums
= list_entry(ordered_sums
.next
,
676 struct btrfs_ordered_sum
,
679 ret
= btrfs_csum_file_blocks(trans
,
680 root
->fs_info
->csum_root
,
682 list_del(&sums
->list
);
688 btrfs_release_path(path
);
690 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
691 /* inline extents are easy, we just overwrite them */
692 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
697 inode_add_bytes(inode
, nbytes
);
698 ret
= btrfs_update_inode(trans
, root
, inode
);
706 * when cleaning up conflicts between the directory names in the
707 * subvolume, directory names in the log and directory names in the
708 * inode back references, we may have to unlink inodes from directories.
710 * This is a helper function to do the unlink of a specific directory
713 static noinline
int drop_one_dir_item(struct btrfs_trans_handle
*trans
,
714 struct btrfs_root
*root
,
715 struct btrfs_path
*path
,
717 struct btrfs_dir_item
*di
)
722 struct extent_buffer
*leaf
;
723 struct btrfs_key location
;
726 leaf
= path
->nodes
[0];
728 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
729 name_len
= btrfs_dir_name_len(leaf
, di
);
730 name
= kmalloc(name_len
, GFP_NOFS
);
734 read_extent_buffer(leaf
, name
, (unsigned long)(di
+ 1), name_len
);
735 btrfs_release_path(path
);
737 inode
= read_one_inode(root
, location
.objectid
);
743 ret
= link_to_fixup_dir(trans
, root
, path
, location
.objectid
);
747 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
751 ret
= btrfs_run_delayed_items(trans
, root
);
759 * helper function to see if a given name and sequence number found
760 * in an inode back reference are already in a directory and correctly
761 * point to this inode
763 static noinline
int inode_in_dir(struct btrfs_root
*root
,
764 struct btrfs_path
*path
,
765 u64 dirid
, u64 objectid
, u64 index
,
766 const char *name
, int name_len
)
768 struct btrfs_dir_item
*di
;
769 struct btrfs_key location
;
772 di
= btrfs_lookup_dir_index_item(NULL
, root
, path
, dirid
,
773 index
, name
, name_len
, 0);
774 if (di
&& !IS_ERR(di
)) {
775 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
776 if (location
.objectid
!= objectid
)
780 btrfs_release_path(path
);
782 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dirid
, name
, name_len
, 0);
783 if (di
&& !IS_ERR(di
)) {
784 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &location
);
785 if (location
.objectid
!= objectid
)
791 btrfs_release_path(path
);
796 * helper function to check a log tree for a named back reference in
797 * an inode. This is used to decide if a back reference that is
798 * found in the subvolume conflicts with what we find in the log.
800 * inode backreferences may have multiple refs in a single item,
801 * during replay we process one reference at a time, and we don't
802 * want to delete valid links to a file from the subvolume if that
803 * link is also in the log.
805 static noinline
int backref_in_log(struct btrfs_root
*log
,
806 struct btrfs_key
*key
,
808 char *name
, int namelen
)
810 struct btrfs_path
*path
;
811 struct btrfs_inode_ref
*ref
;
813 unsigned long ptr_end
;
814 unsigned long name_ptr
;
820 path
= btrfs_alloc_path();
824 ret
= btrfs_search_slot(NULL
, log
, key
, path
, 0, 0);
828 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
830 if (key
->type
== BTRFS_INODE_EXTREF_KEY
) {
831 if (btrfs_find_name_in_ext_backref(path
, ref_objectid
,
832 name
, namelen
, NULL
))
838 item_size
= btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]);
839 ptr_end
= ptr
+ item_size
;
840 while (ptr
< ptr_end
) {
841 ref
= (struct btrfs_inode_ref
*)ptr
;
842 found_name_len
= btrfs_inode_ref_name_len(path
->nodes
[0], ref
);
843 if (found_name_len
== namelen
) {
844 name_ptr
= (unsigned long)(ref
+ 1);
845 ret
= memcmp_extent_buffer(path
->nodes
[0], name
,
852 ptr
= (unsigned long)(ref
+ 1) + found_name_len
;
855 btrfs_free_path(path
);
859 static inline int __add_inode_ref(struct btrfs_trans_handle
*trans
,
860 struct btrfs_root
*root
,
861 struct btrfs_path
*path
,
862 struct btrfs_root
*log_root
,
863 struct inode
*dir
, struct inode
*inode
,
864 struct extent_buffer
*eb
,
865 u64 inode_objectid
, u64 parent_objectid
,
866 u64 ref_index
, char *name
, int namelen
,
872 struct extent_buffer
*leaf
;
873 struct btrfs_dir_item
*di
;
874 struct btrfs_key search_key
;
875 struct btrfs_inode_extref
*extref
;
878 /* Search old style refs */
879 search_key
.objectid
= inode_objectid
;
880 search_key
.type
= BTRFS_INODE_REF_KEY
;
881 search_key
.offset
= parent_objectid
;
882 ret
= btrfs_search_slot(NULL
, root
, &search_key
, path
, 0, 0);
884 struct btrfs_inode_ref
*victim_ref
;
886 unsigned long ptr_end
;
888 leaf
= path
->nodes
[0];
890 /* are we trying to overwrite a back ref for the root directory
891 * if so, just jump out, we're done
893 if (search_key
.objectid
== search_key
.offset
)
896 /* check all the names in this back reference to see
897 * if they are in the log. if so, we allow them to stay
898 * otherwise they must be unlinked as a conflict
900 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
901 ptr_end
= ptr
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
902 while (ptr
< ptr_end
) {
903 victim_ref
= (struct btrfs_inode_ref
*)ptr
;
904 victim_name_len
= btrfs_inode_ref_name_len(leaf
,
906 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
910 read_extent_buffer(leaf
, victim_name
,
911 (unsigned long)(victim_ref
+ 1),
914 if (!backref_in_log(log_root
, &search_key
,
918 btrfs_inc_nlink(inode
);
919 btrfs_release_path(path
);
921 ret
= btrfs_unlink_inode(trans
, root
, dir
,
927 ret
= btrfs_run_delayed_items(trans
, root
);
935 ptr
= (unsigned long)(victim_ref
+ 1) + victim_name_len
;
939 * NOTE: we have searched root tree and checked the
940 * coresponding ref, it does not need to check again.
944 btrfs_release_path(path
);
946 /* Same search but for extended refs */
947 extref
= btrfs_lookup_inode_extref(NULL
, root
, path
, name
, namelen
,
948 inode_objectid
, parent_objectid
, 0,
950 if (!IS_ERR_OR_NULL(extref
)) {
954 struct inode
*victim_parent
;
956 leaf
= path
->nodes
[0];
958 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
959 base
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
961 while (cur_offset
< item_size
) {
962 extref
= (struct btrfs_inode_extref
*)base
+ cur_offset
;
964 victim_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
966 if (btrfs_inode_extref_parent(leaf
, extref
) != parent_objectid
)
969 victim_name
= kmalloc(victim_name_len
, GFP_NOFS
);
972 read_extent_buffer(leaf
, victim_name
, (unsigned long)&extref
->name
,
975 search_key
.objectid
= inode_objectid
;
976 search_key
.type
= BTRFS_INODE_EXTREF_KEY
;
977 search_key
.offset
= btrfs_extref_hash(parent_objectid
,
981 if (!backref_in_log(log_root
, &search_key
,
982 parent_objectid
, victim_name
,
985 victim_parent
= read_one_inode(root
,
988 btrfs_inc_nlink(inode
);
989 btrfs_release_path(path
);
991 ret
= btrfs_unlink_inode(trans
, root
,
997 ret
= btrfs_run_delayed_items(
1000 iput(victim_parent
);
1011 cur_offset
+= victim_name_len
+ sizeof(*extref
);
1015 btrfs_release_path(path
);
1017 /* look for a conflicting sequence number */
1018 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, btrfs_ino(dir
),
1019 ref_index
, name
, namelen
, 0);
1020 if (di
&& !IS_ERR(di
)) {
1021 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
1025 btrfs_release_path(path
);
1027 /* look for a conflicing name */
1028 di
= btrfs_lookup_dir_item(trans
, root
, path
, btrfs_ino(dir
),
1030 if (di
&& !IS_ERR(di
)) {
1031 ret
= drop_one_dir_item(trans
, root
, path
, dir
, di
);
1035 btrfs_release_path(path
);
1040 static int extref_get_fields(struct extent_buffer
*eb
, unsigned long ref_ptr
,
1041 u32
*namelen
, char **name
, u64
*index
,
1042 u64
*parent_objectid
)
1044 struct btrfs_inode_extref
*extref
;
1046 extref
= (struct btrfs_inode_extref
*)ref_ptr
;
1048 *namelen
= btrfs_inode_extref_name_len(eb
, extref
);
1049 *name
= kmalloc(*namelen
, GFP_NOFS
);
1053 read_extent_buffer(eb
, *name
, (unsigned long)&extref
->name
,
1056 *index
= btrfs_inode_extref_index(eb
, extref
);
1057 if (parent_objectid
)
1058 *parent_objectid
= btrfs_inode_extref_parent(eb
, extref
);
1063 static int ref_get_fields(struct extent_buffer
*eb
, unsigned long ref_ptr
,
1064 u32
*namelen
, char **name
, u64
*index
)
1066 struct btrfs_inode_ref
*ref
;
1068 ref
= (struct btrfs_inode_ref
*)ref_ptr
;
1070 *namelen
= btrfs_inode_ref_name_len(eb
, ref
);
1071 *name
= kmalloc(*namelen
, GFP_NOFS
);
1075 read_extent_buffer(eb
, *name
, (unsigned long)(ref
+ 1), *namelen
);
1077 *index
= btrfs_inode_ref_index(eb
, ref
);
1083 * replay one inode back reference item found in the log tree.
1084 * eb, slot and key refer to the buffer and key found in the log tree.
1085 * root is the destination we are replaying into, and path is for temp
1086 * use by this function. (it should be released on return).
1088 static noinline
int add_inode_ref(struct btrfs_trans_handle
*trans
,
1089 struct btrfs_root
*root
,
1090 struct btrfs_root
*log
,
1091 struct btrfs_path
*path
,
1092 struct extent_buffer
*eb
, int slot
,
1093 struct btrfs_key
*key
)
1096 struct inode
*inode
;
1097 unsigned long ref_ptr
;
1098 unsigned long ref_end
;
1102 int search_done
= 0;
1103 int log_ref_ver
= 0;
1104 u64 parent_objectid
;
1107 int ref_struct_size
;
1109 ref_ptr
= btrfs_item_ptr_offset(eb
, slot
);
1110 ref_end
= ref_ptr
+ btrfs_item_size_nr(eb
, slot
);
1112 if (key
->type
== BTRFS_INODE_EXTREF_KEY
) {
1113 struct btrfs_inode_extref
*r
;
1115 ref_struct_size
= sizeof(struct btrfs_inode_extref
);
1117 r
= (struct btrfs_inode_extref
*)ref_ptr
;
1118 parent_objectid
= btrfs_inode_extref_parent(eb
, r
);
1120 ref_struct_size
= sizeof(struct btrfs_inode_ref
);
1121 parent_objectid
= key
->offset
;
1123 inode_objectid
= key
->objectid
;
1126 * it is possible that we didn't log all the parent directories
1127 * for a given inode. If we don't find the dir, just don't
1128 * copy the back ref in. The link count fixup code will take
1131 dir
= read_one_inode(root
, parent_objectid
);
1135 inode
= read_one_inode(root
, inode_objectid
);
1141 while (ref_ptr
< ref_end
) {
1143 ret
= extref_get_fields(eb
, ref_ptr
, &namelen
, &name
,
1144 &ref_index
, &parent_objectid
);
1146 * parent object can change from one array
1150 dir
= read_one_inode(root
, parent_objectid
);
1154 ret
= ref_get_fields(eb
, ref_ptr
, &namelen
, &name
,
1160 /* if we already have a perfect match, we're done */
1161 if (!inode_in_dir(root
, path
, btrfs_ino(dir
), btrfs_ino(inode
),
1162 ref_index
, name
, namelen
)) {
1164 * look for a conflicting back reference in the
1165 * metadata. if we find one we have to unlink that name
1166 * of the file before we add our new link. Later on, we
1167 * overwrite any existing back reference, and we don't
1168 * want to create dangling pointers in the directory.
1172 ret
= __add_inode_ref(trans
, root
, path
, log
,
1176 ref_index
, name
, namelen
,
1186 /* insert our name */
1187 ret
= btrfs_add_link(trans
, dir
, inode
, name
, namelen
,
1192 btrfs_update_inode(trans
, root
, inode
);
1195 ref_ptr
= (unsigned long)(ref_ptr
+ ref_struct_size
) + namelen
;
1203 /* finally write the back reference in the inode */
1204 ret
= overwrite_item(trans
, root
, path
, eb
, slot
, key
);
1206 btrfs_release_path(path
);
1212 static int insert_orphan_item(struct btrfs_trans_handle
*trans
,
1213 struct btrfs_root
*root
, u64 offset
)
1216 ret
= btrfs_find_orphan_item(root
, offset
);
1218 ret
= btrfs_insert_orphan_item(trans
, root
, offset
);
1222 static int count_inode_extrefs(struct btrfs_root
*root
,
1223 struct inode
*inode
, struct btrfs_path
*path
)
1227 unsigned int nlink
= 0;
1230 u64 inode_objectid
= btrfs_ino(inode
);
1233 struct btrfs_inode_extref
*extref
;
1234 struct extent_buffer
*leaf
;
1237 ret
= btrfs_find_one_extref(root
, inode_objectid
, offset
, path
,
1242 leaf
= path
->nodes
[0];
1243 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1244 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1246 while (cur_offset
< item_size
) {
1247 extref
= (struct btrfs_inode_extref
*) (ptr
+ cur_offset
);
1248 name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
1252 cur_offset
+= name_len
+ sizeof(*extref
);
1256 btrfs_release_path(path
);
1258 btrfs_release_path(path
);
1265 static int count_inode_refs(struct btrfs_root
*root
,
1266 struct inode
*inode
, struct btrfs_path
*path
)
1269 struct btrfs_key key
;
1270 unsigned int nlink
= 0;
1272 unsigned long ptr_end
;
1274 u64 ino
= btrfs_ino(inode
);
1277 key
.type
= BTRFS_INODE_REF_KEY
;
1278 key
.offset
= (u64
)-1;
1281 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1285 if (path
->slots
[0] == 0)
1289 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1291 if (key
.objectid
!= ino
||
1292 key
.type
!= BTRFS_INODE_REF_KEY
)
1294 ptr
= btrfs_item_ptr_offset(path
->nodes
[0], path
->slots
[0]);
1295 ptr_end
= ptr
+ btrfs_item_size_nr(path
->nodes
[0],
1297 while (ptr
< ptr_end
) {
1298 struct btrfs_inode_ref
*ref
;
1300 ref
= (struct btrfs_inode_ref
*)ptr
;
1301 name_len
= btrfs_inode_ref_name_len(path
->nodes
[0],
1303 ptr
= (unsigned long)(ref
+ 1) + name_len
;
1307 if (key
.offset
== 0)
1310 btrfs_release_path(path
);
1312 btrfs_release_path(path
);
1318 * There are a few corners where the link count of the file can't
1319 * be properly maintained during replay. So, instead of adding
1320 * lots of complexity to the log code, we just scan the backrefs
1321 * for any file that has been through replay.
1323 * The scan will update the link count on the inode to reflect the
1324 * number of back refs found. If it goes down to zero, the iput
1325 * will free the inode.
1327 static noinline
int fixup_inode_link_count(struct btrfs_trans_handle
*trans
,
1328 struct btrfs_root
*root
,
1329 struct inode
*inode
)
1331 struct btrfs_path
*path
;
1334 u64 ino
= btrfs_ino(inode
);
1336 path
= btrfs_alloc_path();
1340 ret
= count_inode_refs(root
, inode
, path
);
1346 ret
= count_inode_extrefs(root
, inode
, path
);
1357 if (nlink
!= inode
->i_nlink
) {
1358 set_nlink(inode
, nlink
);
1359 btrfs_update_inode(trans
, root
, inode
);
1361 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1363 if (inode
->i_nlink
== 0) {
1364 if (S_ISDIR(inode
->i_mode
)) {
1365 ret
= replay_dir_deletes(trans
, root
, NULL
, path
,
1370 ret
= insert_orphan_item(trans
, root
, ino
);
1374 btrfs_free_path(path
);
1378 static noinline
int fixup_inode_link_counts(struct btrfs_trans_handle
*trans
,
1379 struct btrfs_root
*root
,
1380 struct btrfs_path
*path
)
1383 struct btrfs_key key
;
1384 struct inode
*inode
;
1386 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1387 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1388 key
.offset
= (u64
)-1;
1390 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1395 if (path
->slots
[0] == 0)
1400 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1401 if (key
.objectid
!= BTRFS_TREE_LOG_FIXUP_OBJECTID
||
1402 key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
1405 ret
= btrfs_del_item(trans
, root
, path
);
1409 btrfs_release_path(path
);
1410 inode
= read_one_inode(root
, key
.offset
);
1414 ret
= fixup_inode_link_count(trans
, root
, inode
);
1420 * fixup on a directory may create new entries,
1421 * make sure we always look for the highset possible
1424 key
.offset
= (u64
)-1;
1428 btrfs_release_path(path
);
1434 * record a given inode in the fixup dir so we can check its link
1435 * count when replay is done. The link count is incremented here
1436 * so the inode won't go away until we check it
1438 static noinline
int link_to_fixup_dir(struct btrfs_trans_handle
*trans
,
1439 struct btrfs_root
*root
,
1440 struct btrfs_path
*path
,
1443 struct btrfs_key key
;
1445 struct inode
*inode
;
1447 inode
= read_one_inode(root
, objectid
);
1451 key
.objectid
= BTRFS_TREE_LOG_FIXUP_OBJECTID
;
1452 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1453 key
.offset
= objectid
;
1455 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1457 btrfs_release_path(path
);
1459 if (!inode
->i_nlink
)
1460 set_nlink(inode
, 1);
1462 btrfs_inc_nlink(inode
);
1463 ret
= btrfs_update_inode(trans
, root
, inode
);
1464 } else if (ret
== -EEXIST
) {
1467 BUG(); /* Logic Error */
1475 * when replaying the log for a directory, we only insert names
1476 * for inodes that actually exist. This means an fsync on a directory
1477 * does not implicitly fsync all the new files in it
1479 static noinline
int insert_one_name(struct btrfs_trans_handle
*trans
,
1480 struct btrfs_root
*root
,
1481 struct btrfs_path
*path
,
1482 u64 dirid
, u64 index
,
1483 char *name
, int name_len
, u8 type
,
1484 struct btrfs_key
*location
)
1486 struct inode
*inode
;
1490 inode
= read_one_inode(root
, location
->objectid
);
1494 dir
= read_one_inode(root
, dirid
);
1499 ret
= btrfs_add_link(trans
, dir
, inode
, name
, name_len
, 1, index
);
1501 /* FIXME, put inode into FIXUP list */
1509 * take a single entry in a log directory item and replay it into
1512 * if a conflicting item exists in the subdirectory already,
1513 * the inode it points to is unlinked and put into the link count
1516 * If a name from the log points to a file or directory that does
1517 * not exist in the FS, it is skipped. fsyncs on directories
1518 * do not force down inodes inside that directory, just changes to the
1519 * names or unlinks in a directory.
1521 static noinline
int replay_one_name(struct btrfs_trans_handle
*trans
,
1522 struct btrfs_root
*root
,
1523 struct btrfs_path
*path
,
1524 struct extent_buffer
*eb
,
1525 struct btrfs_dir_item
*di
,
1526 struct btrfs_key
*key
)
1530 struct btrfs_dir_item
*dst_di
;
1531 struct btrfs_key found_key
;
1532 struct btrfs_key log_key
;
1538 dir
= read_one_inode(root
, key
->objectid
);
1542 name_len
= btrfs_dir_name_len(eb
, di
);
1543 name
= kmalloc(name_len
, GFP_NOFS
);
1549 log_type
= btrfs_dir_type(eb
, di
);
1550 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1553 btrfs_dir_item_key_to_cpu(eb
, di
, &log_key
);
1554 exists
= btrfs_lookup_inode(trans
, root
, path
, &log_key
, 0);
1559 btrfs_release_path(path
);
1561 if (key
->type
== BTRFS_DIR_ITEM_KEY
) {
1562 dst_di
= btrfs_lookup_dir_item(trans
, root
, path
, key
->objectid
,
1564 } else if (key
->type
== BTRFS_DIR_INDEX_KEY
) {
1565 dst_di
= btrfs_lookup_dir_index_item(trans
, root
, path
,
1574 if (IS_ERR_OR_NULL(dst_di
)) {
1575 /* we need a sequence number to insert, so we only
1576 * do inserts for the BTRFS_DIR_INDEX_KEY types
1578 if (key
->type
!= BTRFS_DIR_INDEX_KEY
)
1583 btrfs_dir_item_key_to_cpu(path
->nodes
[0], dst_di
, &found_key
);
1584 /* the existing item matches the logged item */
1585 if (found_key
.objectid
== log_key
.objectid
&&
1586 found_key
.type
== log_key
.type
&&
1587 found_key
.offset
== log_key
.offset
&&
1588 btrfs_dir_type(path
->nodes
[0], dst_di
) == log_type
) {
1593 * don't drop the conflicting directory entry if the inode
1594 * for the new entry doesn't exist
1599 ret
= drop_one_dir_item(trans
, root
, path
, dir
, dst_di
);
1603 if (key
->type
== BTRFS_DIR_INDEX_KEY
)
1606 btrfs_release_path(path
);
1612 btrfs_release_path(path
);
1613 ret
= insert_one_name(trans
, root
, path
, key
->objectid
, key
->offset
,
1614 name
, name_len
, log_type
, &log_key
);
1615 if (ret
&& ret
!= -ENOENT
)
1622 * find all the names in a directory item and reconcile them into
1623 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1624 * one name in a directory item, but the same code gets used for
1625 * both directory index types
1627 static noinline
int replay_one_dir_item(struct btrfs_trans_handle
*trans
,
1628 struct btrfs_root
*root
,
1629 struct btrfs_path
*path
,
1630 struct extent_buffer
*eb
, int slot
,
1631 struct btrfs_key
*key
)
1634 u32 item_size
= btrfs_item_size_nr(eb
, slot
);
1635 struct btrfs_dir_item
*di
;
1638 unsigned long ptr_end
;
1640 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1641 ptr_end
= ptr
+ item_size
;
1642 while (ptr
< ptr_end
) {
1643 di
= (struct btrfs_dir_item
*)ptr
;
1644 if (verify_dir_item(root
, eb
, di
))
1646 name_len
= btrfs_dir_name_len(eb
, di
);
1647 ret
= replay_one_name(trans
, root
, path
, eb
, di
, key
);
1650 ptr
= (unsigned long)(di
+ 1);
1657 * directory replay has two parts. There are the standard directory
1658 * items in the log copied from the subvolume, and range items
1659 * created in the log while the subvolume was logged.
1661 * The range items tell us which parts of the key space the log
1662 * is authoritative for. During replay, if a key in the subvolume
1663 * directory is in a logged range item, but not actually in the log
1664 * that means it was deleted from the directory before the fsync
1665 * and should be removed.
1667 static noinline
int find_dir_range(struct btrfs_root
*root
,
1668 struct btrfs_path
*path
,
1669 u64 dirid
, int key_type
,
1670 u64
*start_ret
, u64
*end_ret
)
1672 struct btrfs_key key
;
1674 struct btrfs_dir_log_item
*item
;
1678 if (*start_ret
== (u64
)-1)
1681 key
.objectid
= dirid
;
1682 key
.type
= key_type
;
1683 key
.offset
= *start_ret
;
1685 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1689 if (path
->slots
[0] == 0)
1694 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1696 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1700 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1701 struct btrfs_dir_log_item
);
1702 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1704 if (*start_ret
>= key
.offset
&& *start_ret
<= found_end
) {
1706 *start_ret
= key
.offset
;
1707 *end_ret
= found_end
;
1712 /* check the next slot in the tree to see if it is a valid item */
1713 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1714 if (path
->slots
[0] >= nritems
) {
1715 ret
= btrfs_next_leaf(root
, path
);
1722 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1724 if (key
.type
!= key_type
|| key
.objectid
!= dirid
) {
1728 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1729 struct btrfs_dir_log_item
);
1730 found_end
= btrfs_dir_log_end(path
->nodes
[0], item
);
1731 *start_ret
= key
.offset
;
1732 *end_ret
= found_end
;
1735 btrfs_release_path(path
);
1740 * this looks for a given directory item in the log. If the directory
1741 * item is not in the log, the item is removed and the inode it points
1744 static noinline
int check_item_in_log(struct btrfs_trans_handle
*trans
,
1745 struct btrfs_root
*root
,
1746 struct btrfs_root
*log
,
1747 struct btrfs_path
*path
,
1748 struct btrfs_path
*log_path
,
1750 struct btrfs_key
*dir_key
)
1753 struct extent_buffer
*eb
;
1756 struct btrfs_dir_item
*di
;
1757 struct btrfs_dir_item
*log_di
;
1760 unsigned long ptr_end
;
1762 struct inode
*inode
;
1763 struct btrfs_key location
;
1766 eb
= path
->nodes
[0];
1767 slot
= path
->slots
[0];
1768 item_size
= btrfs_item_size_nr(eb
, slot
);
1769 ptr
= btrfs_item_ptr_offset(eb
, slot
);
1770 ptr_end
= ptr
+ item_size
;
1771 while (ptr
< ptr_end
) {
1772 di
= (struct btrfs_dir_item
*)ptr
;
1773 if (verify_dir_item(root
, eb
, di
)) {
1778 name_len
= btrfs_dir_name_len(eb
, di
);
1779 name
= kmalloc(name_len
, GFP_NOFS
);
1784 read_extent_buffer(eb
, name
, (unsigned long)(di
+ 1),
1787 if (log
&& dir_key
->type
== BTRFS_DIR_ITEM_KEY
) {
1788 log_di
= btrfs_lookup_dir_item(trans
, log
, log_path
,
1791 } else if (log
&& dir_key
->type
== BTRFS_DIR_INDEX_KEY
) {
1792 log_di
= btrfs_lookup_dir_index_item(trans
, log
,
1798 if (IS_ERR_OR_NULL(log_di
)) {
1799 btrfs_dir_item_key_to_cpu(eb
, di
, &location
);
1800 btrfs_release_path(path
);
1801 btrfs_release_path(log_path
);
1802 inode
= read_one_inode(root
, location
.objectid
);
1808 ret
= link_to_fixup_dir(trans
, root
,
1809 path
, location
.objectid
);
1816 btrfs_inc_nlink(inode
);
1817 ret
= btrfs_unlink_inode(trans
, root
, dir
, inode
,
1820 ret
= btrfs_run_delayed_items(trans
, root
);
1826 /* there might still be more names under this key
1827 * check and repeat if required
1829 ret
= btrfs_search_slot(NULL
, root
, dir_key
, path
,
1836 btrfs_release_path(log_path
);
1839 ptr
= (unsigned long)(di
+ 1);
1844 btrfs_release_path(path
);
1845 btrfs_release_path(log_path
);
1850 * deletion replay happens before we copy any new directory items
1851 * out of the log or out of backreferences from inodes. It
1852 * scans the log to find ranges of keys that log is authoritative for,
1853 * and then scans the directory to find items in those ranges that are
1854 * not present in the log.
1856 * Anything we don't find in the log is unlinked and removed from the
1859 static noinline
int replay_dir_deletes(struct btrfs_trans_handle
*trans
,
1860 struct btrfs_root
*root
,
1861 struct btrfs_root
*log
,
1862 struct btrfs_path
*path
,
1863 u64 dirid
, int del_all
)
1867 int key_type
= BTRFS_DIR_LOG_ITEM_KEY
;
1869 struct btrfs_key dir_key
;
1870 struct btrfs_key found_key
;
1871 struct btrfs_path
*log_path
;
1874 dir_key
.objectid
= dirid
;
1875 dir_key
.type
= BTRFS_DIR_ITEM_KEY
;
1876 log_path
= btrfs_alloc_path();
1880 dir
= read_one_inode(root
, dirid
);
1881 /* it isn't an error if the inode isn't there, that can happen
1882 * because we replay the deletes before we copy in the inode item
1886 btrfs_free_path(log_path
);
1894 range_end
= (u64
)-1;
1896 ret
= find_dir_range(log
, path
, dirid
, key_type
,
1897 &range_start
, &range_end
);
1902 dir_key
.offset
= range_start
;
1905 ret
= btrfs_search_slot(NULL
, root
, &dir_key
, path
,
1910 nritems
= btrfs_header_nritems(path
->nodes
[0]);
1911 if (path
->slots
[0] >= nritems
) {
1912 ret
= btrfs_next_leaf(root
, path
);
1916 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1918 if (found_key
.objectid
!= dirid
||
1919 found_key
.type
!= dir_key
.type
)
1922 if (found_key
.offset
> range_end
)
1925 ret
= check_item_in_log(trans
, root
, log
, path
,
1930 if (found_key
.offset
== (u64
)-1)
1932 dir_key
.offset
= found_key
.offset
+ 1;
1934 btrfs_release_path(path
);
1935 if (range_end
== (u64
)-1)
1937 range_start
= range_end
+ 1;
1942 if (key_type
== BTRFS_DIR_LOG_ITEM_KEY
) {
1943 key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
1944 dir_key
.type
= BTRFS_DIR_INDEX_KEY
;
1945 btrfs_release_path(path
);
1949 btrfs_release_path(path
);
1950 btrfs_free_path(log_path
);
1956 * the process_func used to replay items from the log tree. This
1957 * gets called in two different stages. The first stage just looks
1958 * for inodes and makes sure they are all copied into the subvolume.
1960 * The second stage copies all the other item types from the log into
1961 * the subvolume. The two stage approach is slower, but gets rid of
1962 * lots of complexity around inodes referencing other inodes that exist
1963 * only in the log (references come from either directory items or inode
1966 static int replay_one_buffer(struct btrfs_root
*log
, struct extent_buffer
*eb
,
1967 struct walk_control
*wc
, u64 gen
)
1970 struct btrfs_path
*path
;
1971 struct btrfs_root
*root
= wc
->replay_dest
;
1972 struct btrfs_key key
;
1977 ret
= btrfs_read_buffer(eb
, gen
);
1981 level
= btrfs_header_level(eb
);
1986 path
= btrfs_alloc_path();
1990 nritems
= btrfs_header_nritems(eb
);
1991 for (i
= 0; i
< nritems
; i
++) {
1992 btrfs_item_key_to_cpu(eb
, &key
, i
);
1994 /* inode keys are done during the first stage */
1995 if (key
.type
== BTRFS_INODE_ITEM_KEY
&&
1996 wc
->stage
== LOG_WALK_REPLAY_INODES
) {
1997 struct btrfs_inode_item
*inode_item
;
2000 inode_item
= btrfs_item_ptr(eb
, i
,
2001 struct btrfs_inode_item
);
2002 mode
= btrfs_inode_mode(eb
, inode_item
);
2003 if (S_ISDIR(mode
)) {
2004 ret
= replay_dir_deletes(wc
->trans
,
2005 root
, log
, path
, key
.objectid
, 0);
2009 ret
= overwrite_item(wc
->trans
, root
, path
,
2014 /* for regular files, make sure corresponding
2015 * orhpan item exist. extents past the new EOF
2016 * will be truncated later by orphan cleanup.
2018 if (S_ISREG(mode
)) {
2019 ret
= insert_orphan_item(wc
->trans
, root
,
2025 ret
= link_to_fixup_dir(wc
->trans
, root
,
2026 path
, key
.objectid
);
2030 if (wc
->stage
< LOG_WALK_REPLAY_ALL
)
2033 /* these keys are simply copied */
2034 if (key
.type
== BTRFS_XATTR_ITEM_KEY
) {
2035 ret
= overwrite_item(wc
->trans
, root
, path
,
2039 } else if (key
.type
== BTRFS_INODE_REF_KEY
||
2040 key
.type
== BTRFS_INODE_EXTREF_KEY
) {
2041 ret
= add_inode_ref(wc
->trans
, root
, log
, path
,
2043 if (ret
&& ret
!= -ENOENT
)
2046 } else if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
2047 ret
= replay_one_extent(wc
->trans
, root
, path
,
2051 } else if (key
.type
== BTRFS_DIR_ITEM_KEY
||
2052 key
.type
== BTRFS_DIR_INDEX_KEY
) {
2053 ret
= replay_one_dir_item(wc
->trans
, root
, path
,
2059 btrfs_free_path(path
);
2063 static noinline
int walk_down_log_tree(struct btrfs_trans_handle
*trans
,
2064 struct btrfs_root
*root
,
2065 struct btrfs_path
*path
, int *level
,
2066 struct walk_control
*wc
)
2071 struct extent_buffer
*next
;
2072 struct extent_buffer
*cur
;
2073 struct extent_buffer
*parent
;
2077 WARN_ON(*level
< 0);
2078 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2080 while (*level
> 0) {
2081 WARN_ON(*level
< 0);
2082 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2083 cur
= path
->nodes
[*level
];
2085 if (btrfs_header_level(cur
) != *level
)
2088 if (path
->slots
[*level
] >=
2089 btrfs_header_nritems(cur
))
2092 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
2093 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
2094 blocksize
= btrfs_level_size(root
, *level
- 1);
2096 parent
= path
->nodes
[*level
];
2097 root_owner
= btrfs_header_owner(parent
);
2099 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
2104 ret
= wc
->process_func(root
, next
, wc
, ptr_gen
);
2106 free_extent_buffer(next
);
2110 path
->slots
[*level
]++;
2112 ret
= btrfs_read_buffer(next
, ptr_gen
);
2114 free_extent_buffer(next
);
2118 btrfs_tree_lock(next
);
2119 btrfs_set_lock_blocking(next
);
2120 clean_tree_block(trans
, root
, next
);
2121 btrfs_wait_tree_block_writeback(next
);
2122 btrfs_tree_unlock(next
);
2124 WARN_ON(root_owner
!=
2125 BTRFS_TREE_LOG_OBJECTID
);
2126 ret
= btrfs_free_and_pin_reserved_extent(root
,
2129 free_extent_buffer(next
);
2133 free_extent_buffer(next
);
2136 ret
= btrfs_read_buffer(next
, ptr_gen
);
2138 free_extent_buffer(next
);
2142 WARN_ON(*level
<= 0);
2143 if (path
->nodes
[*level
-1])
2144 free_extent_buffer(path
->nodes
[*level
-1]);
2145 path
->nodes
[*level
-1] = next
;
2146 *level
= btrfs_header_level(next
);
2147 path
->slots
[*level
] = 0;
2150 WARN_ON(*level
< 0);
2151 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
2153 path
->slots
[*level
] = btrfs_header_nritems(path
->nodes
[*level
]);
2159 static noinline
int walk_up_log_tree(struct btrfs_trans_handle
*trans
,
2160 struct btrfs_root
*root
,
2161 struct btrfs_path
*path
, int *level
,
2162 struct walk_control
*wc
)
2169 for (i
= *level
; i
< BTRFS_MAX_LEVEL
- 1 && path
->nodes
[i
]; i
++) {
2170 slot
= path
->slots
[i
];
2171 if (slot
+ 1 < btrfs_header_nritems(path
->nodes
[i
])) {
2174 WARN_ON(*level
== 0);
2177 struct extent_buffer
*parent
;
2178 if (path
->nodes
[*level
] == root
->node
)
2179 parent
= path
->nodes
[*level
];
2181 parent
= path
->nodes
[*level
+ 1];
2183 root_owner
= btrfs_header_owner(parent
);
2184 ret
= wc
->process_func(root
, path
->nodes
[*level
], wc
,
2185 btrfs_header_generation(path
->nodes
[*level
]));
2190 struct extent_buffer
*next
;
2192 next
= path
->nodes
[*level
];
2194 btrfs_tree_lock(next
);
2195 btrfs_set_lock_blocking(next
);
2196 clean_tree_block(trans
, root
, next
);
2197 btrfs_wait_tree_block_writeback(next
);
2198 btrfs_tree_unlock(next
);
2200 WARN_ON(root_owner
!= BTRFS_TREE_LOG_OBJECTID
);
2201 ret
= btrfs_free_and_pin_reserved_extent(root
,
2202 path
->nodes
[*level
]->start
,
2203 path
->nodes
[*level
]->len
);
2207 free_extent_buffer(path
->nodes
[*level
]);
2208 path
->nodes
[*level
] = NULL
;
2216 * drop the reference count on the tree rooted at 'snap'. This traverses
2217 * the tree freeing any blocks that have a ref count of zero after being
2220 static int walk_log_tree(struct btrfs_trans_handle
*trans
,
2221 struct btrfs_root
*log
, struct walk_control
*wc
)
2226 struct btrfs_path
*path
;
2229 path
= btrfs_alloc_path();
2233 level
= btrfs_header_level(log
->node
);
2235 path
->nodes
[level
] = log
->node
;
2236 extent_buffer_get(log
->node
);
2237 path
->slots
[level
] = 0;
2240 wret
= walk_down_log_tree(trans
, log
, path
, &level
, wc
);
2248 wret
= walk_up_log_tree(trans
, log
, path
, &level
, wc
);
2257 /* was the root node processed? if not, catch it here */
2258 if (path
->nodes
[orig_level
]) {
2259 ret
= wc
->process_func(log
, path
->nodes
[orig_level
], wc
,
2260 btrfs_header_generation(path
->nodes
[orig_level
]));
2264 struct extent_buffer
*next
;
2266 next
= path
->nodes
[orig_level
];
2268 btrfs_tree_lock(next
);
2269 btrfs_set_lock_blocking(next
);
2270 clean_tree_block(trans
, log
, next
);
2271 btrfs_wait_tree_block_writeback(next
);
2272 btrfs_tree_unlock(next
);
2274 WARN_ON(log
->root_key
.objectid
!=
2275 BTRFS_TREE_LOG_OBJECTID
);
2276 ret
= btrfs_free_and_pin_reserved_extent(log
, next
->start
,
2284 btrfs_free_path(path
);
2289 * helper function to update the item for a given subvolumes log root
2290 * in the tree of log roots
2292 static int update_log_root(struct btrfs_trans_handle
*trans
,
2293 struct btrfs_root
*log
)
2297 if (log
->log_transid
== 1) {
2298 /* insert root item on the first sync */
2299 ret
= btrfs_insert_root(trans
, log
->fs_info
->log_root_tree
,
2300 &log
->root_key
, &log
->root_item
);
2302 ret
= btrfs_update_root(trans
, log
->fs_info
->log_root_tree
,
2303 &log
->root_key
, &log
->root_item
);
2308 static int wait_log_commit(struct btrfs_trans_handle
*trans
,
2309 struct btrfs_root
*root
, unsigned long transid
)
2312 int index
= transid
% 2;
2315 * we only allow two pending log transactions at a time,
2316 * so we know that if ours is more than 2 older than the
2317 * current transaction, we're done
2320 prepare_to_wait(&root
->log_commit_wait
[index
],
2321 &wait
, TASK_UNINTERRUPTIBLE
);
2322 mutex_unlock(&root
->log_mutex
);
2324 if (root
->fs_info
->last_trans_log_full_commit
!=
2325 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
2326 atomic_read(&root
->log_commit
[index
]))
2329 finish_wait(&root
->log_commit_wait
[index
], &wait
);
2330 mutex_lock(&root
->log_mutex
);
2331 } while (root
->fs_info
->last_trans_log_full_commit
!=
2332 trans
->transid
&& root
->log_transid
< transid
+ 2 &&
2333 atomic_read(&root
->log_commit
[index
]));
2337 static void wait_for_writer(struct btrfs_trans_handle
*trans
,
2338 struct btrfs_root
*root
)
2341 while (root
->fs_info
->last_trans_log_full_commit
!=
2342 trans
->transid
&& atomic_read(&root
->log_writers
)) {
2343 prepare_to_wait(&root
->log_writer_wait
,
2344 &wait
, TASK_UNINTERRUPTIBLE
);
2345 mutex_unlock(&root
->log_mutex
);
2346 if (root
->fs_info
->last_trans_log_full_commit
!=
2347 trans
->transid
&& atomic_read(&root
->log_writers
))
2349 mutex_lock(&root
->log_mutex
);
2350 finish_wait(&root
->log_writer_wait
, &wait
);
2355 * btrfs_sync_log does sends a given tree log down to the disk and
2356 * updates the super blocks to record it. When this call is done,
2357 * you know that any inodes previously logged are safely on disk only
2360 * Any other return value means you need to call btrfs_commit_transaction.
2361 * Some of the edge cases for fsyncing directories that have had unlinks
2362 * or renames done in the past mean that sometimes the only safe
2363 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2364 * that has happened.
2366 int btrfs_sync_log(struct btrfs_trans_handle
*trans
,
2367 struct btrfs_root
*root
)
2373 struct btrfs_root
*log
= root
->log_root
;
2374 struct btrfs_root
*log_root_tree
= root
->fs_info
->log_root_tree
;
2375 unsigned long log_transid
= 0;
2376 struct blk_plug plug
;
2378 mutex_lock(&root
->log_mutex
);
2379 log_transid
= root
->log_transid
;
2380 index1
= root
->log_transid
% 2;
2381 if (atomic_read(&root
->log_commit
[index1
])) {
2382 wait_log_commit(trans
, root
, root
->log_transid
);
2383 mutex_unlock(&root
->log_mutex
);
2386 atomic_set(&root
->log_commit
[index1
], 1);
2388 /* wait for previous tree log sync to complete */
2389 if (atomic_read(&root
->log_commit
[(index1
+ 1) % 2]))
2390 wait_log_commit(trans
, root
, root
->log_transid
- 1);
2392 int batch
= atomic_read(&root
->log_batch
);
2393 /* when we're on an ssd, just kick the log commit out */
2394 if (!btrfs_test_opt(root
, SSD
) && root
->log_multiple_pids
) {
2395 mutex_unlock(&root
->log_mutex
);
2396 schedule_timeout_uninterruptible(1);
2397 mutex_lock(&root
->log_mutex
);
2399 wait_for_writer(trans
, root
);
2400 if (batch
== atomic_read(&root
->log_batch
))
2404 /* bail out if we need to do a full commit */
2405 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2407 btrfs_free_logged_extents(log
, log_transid
);
2408 mutex_unlock(&root
->log_mutex
);
2412 if (log_transid
% 2 == 0)
2413 mark
= EXTENT_DIRTY
;
2417 /* we start IO on all the marked extents here, but we don't actually
2418 * wait for them until later.
2420 blk_start_plug(&plug
);
2421 ret
= btrfs_write_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2423 blk_finish_plug(&plug
);
2424 btrfs_abort_transaction(trans
, root
, ret
);
2425 btrfs_free_logged_extents(log
, log_transid
);
2426 mutex_unlock(&root
->log_mutex
);
2430 btrfs_set_root_node(&log
->root_item
, log
->node
);
2432 root
->log_transid
++;
2433 log
->log_transid
= root
->log_transid
;
2434 root
->log_start_pid
= 0;
2437 * IO has been started, blocks of the log tree have WRITTEN flag set
2438 * in their headers. new modifications of the log will be written to
2439 * new positions. so it's safe to allow log writers to go in.
2441 mutex_unlock(&root
->log_mutex
);
2443 mutex_lock(&log_root_tree
->log_mutex
);
2444 atomic_inc(&log_root_tree
->log_batch
);
2445 atomic_inc(&log_root_tree
->log_writers
);
2446 mutex_unlock(&log_root_tree
->log_mutex
);
2448 ret
= update_log_root(trans
, log
);
2450 mutex_lock(&log_root_tree
->log_mutex
);
2451 if (atomic_dec_and_test(&log_root_tree
->log_writers
)) {
2453 if (waitqueue_active(&log_root_tree
->log_writer_wait
))
2454 wake_up(&log_root_tree
->log_writer_wait
);
2458 blk_finish_plug(&plug
);
2459 if (ret
!= -ENOSPC
) {
2460 btrfs_abort_transaction(trans
, root
, ret
);
2461 mutex_unlock(&log_root_tree
->log_mutex
);
2464 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2465 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2466 btrfs_free_logged_extents(log
, log_transid
);
2467 mutex_unlock(&log_root_tree
->log_mutex
);
2472 index2
= log_root_tree
->log_transid
% 2;
2473 if (atomic_read(&log_root_tree
->log_commit
[index2
])) {
2474 blk_finish_plug(&plug
);
2475 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2476 wait_log_commit(trans
, log_root_tree
,
2477 log_root_tree
->log_transid
);
2478 btrfs_free_logged_extents(log
, log_transid
);
2479 mutex_unlock(&log_root_tree
->log_mutex
);
2483 atomic_set(&log_root_tree
->log_commit
[index2
], 1);
2485 if (atomic_read(&log_root_tree
->log_commit
[(index2
+ 1) % 2])) {
2486 wait_log_commit(trans
, log_root_tree
,
2487 log_root_tree
->log_transid
- 1);
2490 wait_for_writer(trans
, log_root_tree
);
2493 * now that we've moved on to the tree of log tree roots,
2494 * check the full commit flag again
2496 if (root
->fs_info
->last_trans_log_full_commit
== trans
->transid
) {
2497 blk_finish_plug(&plug
);
2498 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2499 btrfs_free_logged_extents(log
, log_transid
);
2500 mutex_unlock(&log_root_tree
->log_mutex
);
2502 goto out_wake_log_root
;
2505 ret
= btrfs_write_marked_extents(log_root_tree
,
2506 &log_root_tree
->dirty_log_pages
,
2507 EXTENT_DIRTY
| EXTENT_NEW
);
2508 blk_finish_plug(&plug
);
2510 btrfs_abort_transaction(trans
, root
, ret
);
2511 btrfs_free_logged_extents(log
, log_transid
);
2512 mutex_unlock(&log_root_tree
->log_mutex
);
2513 goto out_wake_log_root
;
2515 btrfs_wait_marked_extents(log
, &log
->dirty_log_pages
, mark
);
2516 btrfs_wait_marked_extents(log_root_tree
,
2517 &log_root_tree
->dirty_log_pages
,
2518 EXTENT_NEW
| EXTENT_DIRTY
);
2519 btrfs_wait_logged_extents(log
, log_transid
);
2521 btrfs_set_super_log_root(root
->fs_info
->super_for_commit
,
2522 log_root_tree
->node
->start
);
2523 btrfs_set_super_log_root_level(root
->fs_info
->super_for_commit
,
2524 btrfs_header_level(log_root_tree
->node
));
2526 log_root_tree
->log_transid
++;
2529 mutex_unlock(&log_root_tree
->log_mutex
);
2532 * nobody else is going to jump in and write the the ctree
2533 * super here because the log_commit atomic below is protecting
2534 * us. We must be called with a transaction handle pinning
2535 * the running transaction open, so a full commit can't hop
2536 * in and cause problems either.
2538 btrfs_scrub_pause_super(root
);
2539 ret
= write_ctree_super(trans
, root
->fs_info
->tree_root
, 1);
2540 btrfs_scrub_continue_super(root
);
2542 btrfs_abort_transaction(trans
, root
, ret
);
2543 goto out_wake_log_root
;
2546 mutex_lock(&root
->log_mutex
);
2547 if (root
->last_log_commit
< log_transid
)
2548 root
->last_log_commit
= log_transid
;
2549 mutex_unlock(&root
->log_mutex
);
2552 atomic_set(&log_root_tree
->log_commit
[index2
], 0);
2554 if (waitqueue_active(&log_root_tree
->log_commit_wait
[index2
]))
2555 wake_up(&log_root_tree
->log_commit_wait
[index2
]);
2557 atomic_set(&root
->log_commit
[index1
], 0);
2559 if (waitqueue_active(&root
->log_commit_wait
[index1
]))
2560 wake_up(&root
->log_commit_wait
[index1
]);
2564 static void free_log_tree(struct btrfs_trans_handle
*trans
,
2565 struct btrfs_root
*log
)
2570 struct walk_control wc
= {
2572 .process_func
= process_one_buffer
2576 ret
= walk_log_tree(trans
, log
, &wc
);
2578 /* I don't think this can happen but just in case */
2580 btrfs_abort_transaction(trans
, log
, ret
);
2584 ret
= find_first_extent_bit(&log
->dirty_log_pages
,
2585 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
,
2590 clear_extent_bits(&log
->dirty_log_pages
, start
, end
,
2591 EXTENT_DIRTY
| EXTENT_NEW
, GFP_NOFS
);
2595 * We may have short-circuited the log tree with the full commit logic
2596 * and left ordered extents on our list, so clear these out to keep us
2597 * from leaking inodes and memory.
2599 btrfs_free_logged_extents(log
, 0);
2600 btrfs_free_logged_extents(log
, 1);
2602 free_extent_buffer(log
->node
);
2607 * free all the extents used by the tree log. This should be called
2608 * at commit time of the full transaction
2610 int btrfs_free_log(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
)
2612 if (root
->log_root
) {
2613 free_log_tree(trans
, root
->log_root
);
2614 root
->log_root
= NULL
;
2619 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
2620 struct btrfs_fs_info
*fs_info
)
2622 if (fs_info
->log_root_tree
) {
2623 free_log_tree(trans
, fs_info
->log_root_tree
);
2624 fs_info
->log_root_tree
= NULL
;
2630 * If both a file and directory are logged, and unlinks or renames are
2631 * mixed in, we have a few interesting corners:
2633 * create file X in dir Y
2634 * link file X to X.link in dir Y
2636 * unlink file X but leave X.link
2639 * After a crash we would expect only X.link to exist. But file X
2640 * didn't get fsync'd again so the log has back refs for X and X.link.
2642 * We solve this by removing directory entries and inode backrefs from the
2643 * log when a file that was logged in the current transaction is
2644 * unlinked. Any later fsync will include the updated log entries, and
2645 * we'll be able to reconstruct the proper directory items from backrefs.
2647 * This optimizations allows us to avoid relogging the entire inode
2648 * or the entire directory.
2650 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle
*trans
,
2651 struct btrfs_root
*root
,
2652 const char *name
, int name_len
,
2653 struct inode
*dir
, u64 index
)
2655 struct btrfs_root
*log
;
2656 struct btrfs_dir_item
*di
;
2657 struct btrfs_path
*path
;
2661 u64 dir_ino
= btrfs_ino(dir
);
2663 if (BTRFS_I(dir
)->logged_trans
< trans
->transid
)
2666 ret
= join_running_log_trans(root
);
2670 mutex_lock(&BTRFS_I(dir
)->log_mutex
);
2672 log
= root
->log_root
;
2673 path
= btrfs_alloc_path();
2679 di
= btrfs_lookup_dir_item(trans
, log
, path
, dir_ino
,
2680 name
, name_len
, -1);
2686 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2687 bytes_del
+= name_len
;
2693 btrfs_release_path(path
);
2694 di
= btrfs_lookup_dir_index_item(trans
, log
, path
, dir_ino
,
2695 index
, name
, name_len
, -1);
2701 ret
= btrfs_delete_one_dir_name(trans
, log
, path
, di
);
2702 bytes_del
+= name_len
;
2709 /* update the directory size in the log to reflect the names
2713 struct btrfs_key key
;
2715 key
.objectid
= dir_ino
;
2717 key
.type
= BTRFS_INODE_ITEM_KEY
;
2718 btrfs_release_path(path
);
2720 ret
= btrfs_search_slot(trans
, log
, &key
, path
, 0, 1);
2726 struct btrfs_inode_item
*item
;
2729 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2730 struct btrfs_inode_item
);
2731 i_size
= btrfs_inode_size(path
->nodes
[0], item
);
2732 if (i_size
> bytes_del
)
2733 i_size
-= bytes_del
;
2736 btrfs_set_inode_size(path
->nodes
[0], item
, i_size
);
2737 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2740 btrfs_release_path(path
);
2743 btrfs_free_path(path
);
2745 mutex_unlock(&BTRFS_I(dir
)->log_mutex
);
2746 if (ret
== -ENOSPC
) {
2747 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2750 btrfs_abort_transaction(trans
, root
, ret
);
2752 btrfs_end_log_trans(root
);
2757 /* see comments for btrfs_del_dir_entries_in_log */
2758 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle
*trans
,
2759 struct btrfs_root
*root
,
2760 const char *name
, int name_len
,
2761 struct inode
*inode
, u64 dirid
)
2763 struct btrfs_root
*log
;
2767 if (BTRFS_I(inode
)->logged_trans
< trans
->transid
)
2770 ret
= join_running_log_trans(root
);
2773 log
= root
->log_root
;
2774 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
2776 ret
= btrfs_del_inode_ref(trans
, log
, name
, name_len
, btrfs_ino(inode
),
2778 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
2779 if (ret
== -ENOSPC
) {
2780 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
2782 } else if (ret
< 0 && ret
!= -ENOENT
)
2783 btrfs_abort_transaction(trans
, root
, ret
);
2784 btrfs_end_log_trans(root
);
2790 * creates a range item in the log for 'dirid'. first_offset and
2791 * last_offset tell us which parts of the key space the log should
2792 * be considered authoritative for.
2794 static noinline
int insert_dir_log_key(struct btrfs_trans_handle
*trans
,
2795 struct btrfs_root
*log
,
2796 struct btrfs_path
*path
,
2797 int key_type
, u64 dirid
,
2798 u64 first_offset
, u64 last_offset
)
2801 struct btrfs_key key
;
2802 struct btrfs_dir_log_item
*item
;
2804 key
.objectid
= dirid
;
2805 key
.offset
= first_offset
;
2806 if (key_type
== BTRFS_DIR_ITEM_KEY
)
2807 key
.type
= BTRFS_DIR_LOG_ITEM_KEY
;
2809 key
.type
= BTRFS_DIR_LOG_INDEX_KEY
;
2810 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*item
));
2814 item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2815 struct btrfs_dir_log_item
);
2816 btrfs_set_dir_log_end(path
->nodes
[0], item
, last_offset
);
2817 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2818 btrfs_release_path(path
);
2823 * log all the items included in the current transaction for a given
2824 * directory. This also creates the range items in the log tree required
2825 * to replay anything deleted before the fsync
2827 static noinline
int log_dir_items(struct btrfs_trans_handle
*trans
,
2828 struct btrfs_root
*root
, struct inode
*inode
,
2829 struct btrfs_path
*path
,
2830 struct btrfs_path
*dst_path
, int key_type
,
2831 u64 min_offset
, u64
*last_offset_ret
)
2833 struct btrfs_key min_key
;
2834 struct btrfs_key max_key
;
2835 struct btrfs_root
*log
= root
->log_root
;
2836 struct extent_buffer
*src
;
2841 u64 first_offset
= min_offset
;
2842 u64 last_offset
= (u64
)-1;
2843 u64 ino
= btrfs_ino(inode
);
2845 log
= root
->log_root
;
2846 max_key
.objectid
= ino
;
2847 max_key
.offset
= (u64
)-1;
2848 max_key
.type
= key_type
;
2850 min_key
.objectid
= ino
;
2851 min_key
.type
= key_type
;
2852 min_key
.offset
= min_offset
;
2854 path
->keep_locks
= 1;
2856 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
2857 path
, trans
->transid
);
2860 * we didn't find anything from this transaction, see if there
2861 * is anything at all
2863 if (ret
!= 0 || min_key
.objectid
!= ino
|| min_key
.type
!= key_type
) {
2864 min_key
.objectid
= ino
;
2865 min_key
.type
= key_type
;
2866 min_key
.offset
= (u64
)-1;
2867 btrfs_release_path(path
);
2868 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2870 btrfs_release_path(path
);
2873 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2875 /* if ret == 0 there are items for this type,
2876 * create a range to tell us the last key of this type.
2877 * otherwise, there are no items in this directory after
2878 * *min_offset, and we create a range to indicate that.
2881 struct btrfs_key tmp
;
2882 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
,
2884 if (key_type
== tmp
.type
)
2885 first_offset
= max(min_offset
, tmp
.offset
) + 1;
2890 /* go backward to find any previous key */
2891 ret
= btrfs_previous_item(root
, path
, ino
, key_type
);
2893 struct btrfs_key tmp
;
2894 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2895 if (key_type
== tmp
.type
) {
2896 first_offset
= tmp
.offset
;
2897 ret
= overwrite_item(trans
, log
, dst_path
,
2898 path
->nodes
[0], path
->slots
[0],
2906 btrfs_release_path(path
);
2908 /* find the first key from this transaction again */
2909 ret
= btrfs_search_slot(NULL
, root
, &min_key
, path
, 0, 0);
2916 * we have a block from this transaction, log every item in it
2917 * from our directory
2920 struct btrfs_key tmp
;
2921 src
= path
->nodes
[0];
2922 nritems
= btrfs_header_nritems(src
);
2923 for (i
= path
->slots
[0]; i
< nritems
; i
++) {
2924 btrfs_item_key_to_cpu(src
, &min_key
, i
);
2926 if (min_key
.objectid
!= ino
|| min_key
.type
!= key_type
)
2928 ret
= overwrite_item(trans
, log
, dst_path
, src
, i
,
2935 path
->slots
[0] = nritems
;
2938 * look ahead to the next item and see if it is also
2939 * from this directory and from this transaction
2941 ret
= btrfs_next_leaf(root
, path
);
2943 last_offset
= (u64
)-1;
2946 btrfs_item_key_to_cpu(path
->nodes
[0], &tmp
, path
->slots
[0]);
2947 if (tmp
.objectid
!= ino
|| tmp
.type
!= key_type
) {
2948 last_offset
= (u64
)-1;
2951 if (btrfs_header_generation(path
->nodes
[0]) != trans
->transid
) {
2952 ret
= overwrite_item(trans
, log
, dst_path
,
2953 path
->nodes
[0], path
->slots
[0],
2958 last_offset
= tmp
.offset
;
2963 btrfs_release_path(path
);
2964 btrfs_release_path(dst_path
);
2967 *last_offset_ret
= last_offset
;
2969 * insert the log range keys to indicate where the log
2972 ret
= insert_dir_log_key(trans
, log
, path
, key_type
,
2973 ino
, first_offset
, last_offset
);
2981 * logging directories is very similar to logging inodes, We find all the items
2982 * from the current transaction and write them to the log.
2984 * The recovery code scans the directory in the subvolume, and if it finds a
2985 * key in the range logged that is not present in the log tree, then it means
2986 * that dir entry was unlinked during the transaction.
2988 * In order for that scan to work, we must include one key smaller than
2989 * the smallest logged by this transaction and one key larger than the largest
2990 * key logged by this transaction.
2992 static noinline
int log_directory_changes(struct btrfs_trans_handle
*trans
,
2993 struct btrfs_root
*root
, struct inode
*inode
,
2994 struct btrfs_path
*path
,
2995 struct btrfs_path
*dst_path
)
3000 int key_type
= BTRFS_DIR_ITEM_KEY
;
3006 ret
= log_dir_items(trans
, root
, inode
, path
,
3007 dst_path
, key_type
, min_key
,
3011 if (max_key
== (u64
)-1)
3013 min_key
= max_key
+ 1;
3016 if (key_type
== BTRFS_DIR_ITEM_KEY
) {
3017 key_type
= BTRFS_DIR_INDEX_KEY
;
3024 * a helper function to drop items from the log before we relog an
3025 * inode. max_key_type indicates the highest item type to remove.
3026 * This cannot be run for file data extents because it does not
3027 * free the extents they point to.
3029 static int drop_objectid_items(struct btrfs_trans_handle
*trans
,
3030 struct btrfs_root
*log
,
3031 struct btrfs_path
*path
,
3032 u64 objectid
, int max_key_type
)
3035 struct btrfs_key key
;
3036 struct btrfs_key found_key
;
3039 key
.objectid
= objectid
;
3040 key
.type
= max_key_type
;
3041 key
.offset
= (u64
)-1;
3044 ret
= btrfs_search_slot(trans
, log
, &key
, path
, -1, 1);
3045 BUG_ON(ret
== 0); /* Logic error */
3049 if (path
->slots
[0] == 0)
3053 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
3056 if (found_key
.objectid
!= objectid
)
3059 found_key
.offset
= 0;
3061 ret
= btrfs_bin_search(path
->nodes
[0], &found_key
, 0,
3064 ret
= btrfs_del_items(trans
, log
, path
, start_slot
,
3065 path
->slots
[0] - start_slot
+ 1);
3067 * If start slot isn't 0 then we don't need to re-search, we've
3068 * found the last guy with the objectid in this tree.
3070 if (ret
|| start_slot
!= 0)
3072 btrfs_release_path(path
);
3074 btrfs_release_path(path
);
3080 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3081 struct extent_buffer
*leaf
,
3082 struct btrfs_inode_item
*item
,
3083 struct inode
*inode
, int log_inode_only
)
3085 struct btrfs_map_token token
;
3087 btrfs_init_map_token(&token
);
3089 if (log_inode_only
) {
3090 /* set the generation to zero so the recover code
3091 * can tell the difference between an logging
3092 * just to say 'this inode exists' and a logging
3093 * to say 'update this inode with these values'
3095 btrfs_set_token_inode_generation(leaf
, item
, 0, &token
);
3096 btrfs_set_token_inode_size(leaf
, item
, 0, &token
);
3098 btrfs_set_token_inode_generation(leaf
, item
,
3099 BTRFS_I(inode
)->generation
,
3101 btrfs_set_token_inode_size(leaf
, item
, inode
->i_size
, &token
);
3104 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3105 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3106 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3107 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3109 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_atime(item
),
3110 inode
->i_atime
.tv_sec
, &token
);
3111 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_atime(item
),
3112 inode
->i_atime
.tv_nsec
, &token
);
3114 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_mtime(item
),
3115 inode
->i_mtime
.tv_sec
, &token
);
3116 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
3117 inode
->i_mtime
.tv_nsec
, &token
);
3119 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_ctime(item
),
3120 inode
->i_ctime
.tv_sec
, &token
);
3121 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
3122 inode
->i_ctime
.tv_nsec
, &token
);
3124 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3127 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3128 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3129 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3130 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3131 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3134 static int log_inode_item(struct btrfs_trans_handle
*trans
,
3135 struct btrfs_root
*log
, struct btrfs_path
*path
,
3136 struct inode
*inode
)
3138 struct btrfs_inode_item
*inode_item
;
3139 struct btrfs_key key
;
3142 memcpy(&key
, &BTRFS_I(inode
)->location
, sizeof(key
));
3143 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
,
3144 sizeof(*inode_item
));
3145 if (ret
&& ret
!= -EEXIST
)
3147 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3148 struct btrfs_inode_item
);
3149 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
, 0);
3150 btrfs_release_path(path
);
3154 static noinline
int copy_items(struct btrfs_trans_handle
*trans
,
3155 struct inode
*inode
,
3156 struct btrfs_path
*dst_path
,
3157 struct extent_buffer
*src
,
3158 int start_slot
, int nr
, int inode_only
)
3160 unsigned long src_offset
;
3161 unsigned long dst_offset
;
3162 struct btrfs_root
*log
= BTRFS_I(inode
)->root
->log_root
;
3163 struct btrfs_file_extent_item
*extent
;
3164 struct btrfs_inode_item
*inode_item
;
3166 struct btrfs_key
*ins_keys
;
3170 struct list_head ordered_sums
;
3171 int skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
3173 INIT_LIST_HEAD(&ordered_sums
);
3175 ins_data
= kmalloc(nr
* sizeof(struct btrfs_key
) +
3176 nr
* sizeof(u32
), GFP_NOFS
);
3180 ins_sizes
= (u32
*)ins_data
;
3181 ins_keys
= (struct btrfs_key
*)(ins_data
+ nr
* sizeof(u32
));
3183 for (i
= 0; i
< nr
; i
++) {
3184 ins_sizes
[i
] = btrfs_item_size_nr(src
, i
+ start_slot
);
3185 btrfs_item_key_to_cpu(src
, ins_keys
+ i
, i
+ start_slot
);
3187 ret
= btrfs_insert_empty_items(trans
, log
, dst_path
,
3188 ins_keys
, ins_sizes
, nr
);
3194 for (i
= 0; i
< nr
; i
++, dst_path
->slots
[0]++) {
3195 dst_offset
= btrfs_item_ptr_offset(dst_path
->nodes
[0],
3196 dst_path
->slots
[0]);
3198 src_offset
= btrfs_item_ptr_offset(src
, start_slot
+ i
);
3200 if (ins_keys
[i
].type
== BTRFS_INODE_ITEM_KEY
) {
3201 inode_item
= btrfs_item_ptr(dst_path
->nodes
[0],
3203 struct btrfs_inode_item
);
3204 fill_inode_item(trans
, dst_path
->nodes
[0], inode_item
,
3205 inode
, inode_only
== LOG_INODE_EXISTS
);
3207 copy_extent_buffer(dst_path
->nodes
[0], src
, dst_offset
,
3208 src_offset
, ins_sizes
[i
]);
3211 /* take a reference on file data extents so that truncates
3212 * or deletes of this inode don't have to relog the inode
3215 if (btrfs_key_type(ins_keys
+ i
) == BTRFS_EXTENT_DATA_KEY
&&
3218 extent
= btrfs_item_ptr(src
, start_slot
+ i
,
3219 struct btrfs_file_extent_item
);
3221 if (btrfs_file_extent_generation(src
, extent
) < trans
->transid
)
3224 found_type
= btrfs_file_extent_type(src
, extent
);
3225 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
3227 ds
= btrfs_file_extent_disk_bytenr(src
,
3229 /* ds == 0 is a hole */
3233 dl
= btrfs_file_extent_disk_num_bytes(src
,
3235 cs
= btrfs_file_extent_offset(src
, extent
);
3236 cl
= btrfs_file_extent_num_bytes(src
,
3238 if (btrfs_file_extent_compression(src
,
3244 ret
= btrfs_lookup_csums_range(
3245 log
->fs_info
->csum_root
,
3246 ds
+ cs
, ds
+ cs
+ cl
- 1,
3249 btrfs_release_path(dst_path
);
3257 btrfs_mark_buffer_dirty(dst_path
->nodes
[0]);
3258 btrfs_release_path(dst_path
);
3262 * we have to do this after the loop above to avoid changing the
3263 * log tree while trying to change the log tree.
3266 while (!list_empty(&ordered_sums
)) {
3267 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
3268 struct btrfs_ordered_sum
,
3271 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
3272 list_del(&sums
->list
);
3278 static int extent_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3280 struct extent_map
*em1
, *em2
;
3282 em1
= list_entry(a
, struct extent_map
, list
);
3283 em2
= list_entry(b
, struct extent_map
, list
);
3285 if (em1
->start
< em2
->start
)
3287 else if (em1
->start
> em2
->start
)
3292 static int log_one_extent(struct btrfs_trans_handle
*trans
,
3293 struct inode
*inode
, struct btrfs_root
*root
,
3294 struct extent_map
*em
, struct btrfs_path
*path
)
3296 struct btrfs_root
*log
= root
->log_root
;
3297 struct btrfs_file_extent_item
*fi
;
3298 struct extent_buffer
*leaf
;
3299 struct btrfs_ordered_extent
*ordered
;
3300 struct list_head ordered_sums
;
3301 struct btrfs_map_token token
;
3302 struct btrfs_key key
;
3303 u64 mod_start
= em
->mod_start
;
3304 u64 mod_len
= em
->mod_len
;
3307 u64 extent_offset
= em
->start
- em
->orig_start
;
3310 int index
= log
->log_transid
% 2;
3311 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
3313 ret
= __btrfs_drop_extents(trans
, log
, inode
, path
, em
->start
,
3314 em
->start
+ em
->len
, NULL
, 0);
3318 INIT_LIST_HEAD(&ordered_sums
);
3319 btrfs_init_map_token(&token
);
3320 key
.objectid
= btrfs_ino(inode
);
3321 key
.type
= BTRFS_EXTENT_DATA_KEY
;
3322 key
.offset
= em
->start
;
3324 ret
= btrfs_insert_empty_item(trans
, log
, path
, &key
, sizeof(*fi
));
3327 leaf
= path
->nodes
[0];
3328 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3329 struct btrfs_file_extent_item
);
3331 btrfs_set_token_file_extent_generation(leaf
, fi
, em
->generation
,
3333 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3335 btrfs_set_token_file_extent_type(leaf
, fi
,
3336 BTRFS_FILE_EXTENT_PREALLOC
,
3339 btrfs_set_token_file_extent_type(leaf
, fi
,
3340 BTRFS_FILE_EXTENT_REG
,
3342 if (em
->block_start
== 0)
3346 block_len
= max(em
->block_len
, em
->orig_block_len
);
3347 if (em
->compress_type
!= BTRFS_COMPRESS_NONE
) {
3348 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
,
3351 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, block_len
,
3353 } else if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
3354 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
,
3356 extent_offset
, &token
);
3357 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, block_len
,
3360 btrfs_set_token_file_extent_disk_bytenr(leaf
, fi
, 0, &token
);
3361 btrfs_set_token_file_extent_disk_num_bytes(leaf
, fi
, 0,
3365 btrfs_set_token_file_extent_offset(leaf
, fi
,
3366 em
->start
- em
->orig_start
,
3368 btrfs_set_token_file_extent_num_bytes(leaf
, fi
, em
->len
, &token
);
3369 btrfs_set_token_file_extent_ram_bytes(leaf
, fi
, em
->ram_bytes
, &token
);
3370 btrfs_set_token_file_extent_compression(leaf
, fi
, em
->compress_type
,
3372 btrfs_set_token_file_extent_encryption(leaf
, fi
, 0, &token
);
3373 btrfs_set_token_file_extent_other_encoding(leaf
, fi
, 0, &token
);
3374 btrfs_mark_buffer_dirty(leaf
);
3376 btrfs_release_path(path
);
3384 if (em
->compress_type
) {
3386 csum_len
= block_len
;
3390 * First check and see if our csums are on our outstanding ordered
3394 spin_lock_irq(&log
->log_extents_lock
[index
]);
3395 list_for_each_entry(ordered
, &log
->logged_list
[index
], log_list
) {
3396 struct btrfs_ordered_sum
*sum
;
3401 if (ordered
->inode
!= inode
)
3404 if (ordered
->file_offset
+ ordered
->len
<= mod_start
||
3405 mod_start
+ mod_len
<= ordered
->file_offset
)
3409 * We are going to copy all the csums on this ordered extent, so
3410 * go ahead and adjust mod_start and mod_len in case this
3411 * ordered extent has already been logged.
3413 if (ordered
->file_offset
> mod_start
) {
3414 if (ordered
->file_offset
+ ordered
->len
>=
3415 mod_start
+ mod_len
)
3416 mod_len
= ordered
->file_offset
- mod_start
;
3418 * If we have this case
3420 * |--------- logged extent ---------|
3421 * |----- ordered extent ----|
3423 * Just don't mess with mod_start and mod_len, we'll
3424 * just end up logging more csums than we need and it
3428 if (ordered
->file_offset
+ ordered
->len
<
3429 mod_start
+ mod_len
) {
3430 mod_len
= (mod_start
+ mod_len
) -
3431 (ordered
->file_offset
+ ordered
->len
);
3432 mod_start
= ordered
->file_offset
+
3440 * To keep us from looping for the above case of an ordered
3441 * extent that falls inside of the logged extent.
3443 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM
,
3446 atomic_inc(&ordered
->refs
);
3447 spin_unlock_irq(&log
->log_extents_lock
[index
]);
3449 * we've dropped the lock, we must either break or
3450 * start over after this.
3453 wait_event(ordered
->wait
, ordered
->csum_bytes_left
== 0);
3455 list_for_each_entry(sum
, &ordered
->list
, list
) {
3456 ret
= btrfs_csum_file_blocks(trans
, log
, sum
);
3458 btrfs_put_ordered_extent(ordered
);
3462 btrfs_put_ordered_extent(ordered
);
3466 spin_unlock_irq(&log
->log_extents_lock
[index
]);
3469 if (!mod_len
|| ret
)
3472 csum_offset
= mod_start
- em
->start
;
3475 /* block start is already adjusted for the file extent offset. */
3476 ret
= btrfs_lookup_csums_range(log
->fs_info
->csum_root
,
3477 em
->block_start
+ csum_offset
,
3478 em
->block_start
+ csum_offset
+
3479 csum_len
- 1, &ordered_sums
, 0);
3483 while (!list_empty(&ordered_sums
)) {
3484 struct btrfs_ordered_sum
*sums
= list_entry(ordered_sums
.next
,
3485 struct btrfs_ordered_sum
,
3488 ret
= btrfs_csum_file_blocks(trans
, log
, sums
);
3489 list_del(&sums
->list
);
3496 static int btrfs_log_changed_extents(struct btrfs_trans_handle
*trans
,
3497 struct btrfs_root
*root
,
3498 struct inode
*inode
,
3499 struct btrfs_path
*path
)
3501 struct extent_map
*em
, *n
;
3502 struct list_head extents
;
3503 struct extent_map_tree
*tree
= &BTRFS_I(inode
)->extent_tree
;
3508 INIT_LIST_HEAD(&extents
);
3510 write_lock(&tree
->lock
);
3511 test_gen
= root
->fs_info
->last_trans_committed
;
3513 list_for_each_entry_safe(em
, n
, &tree
->modified_extents
, list
) {
3514 list_del_init(&em
->list
);
3517 * Just an arbitrary number, this can be really CPU intensive
3518 * once we start getting a lot of extents, and really once we
3519 * have a bunch of extents we just want to commit since it will
3522 if (++num
> 32768) {
3523 list_del_init(&tree
->modified_extents
);
3528 if (em
->generation
<= test_gen
)
3530 /* Need a ref to keep it from getting evicted from cache */
3531 atomic_inc(&em
->refs
);
3532 set_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
3533 list_add_tail(&em
->list
, &extents
);
3537 list_sort(NULL
, &extents
, extent_cmp
);
3540 while (!list_empty(&extents
)) {
3541 em
= list_entry(extents
.next
, struct extent_map
, list
);
3543 list_del_init(&em
->list
);
3546 * If we had an error we just need to delete everybody from our
3550 clear_em_logging(tree
, em
);
3551 free_extent_map(em
);
3555 write_unlock(&tree
->lock
);
3557 ret
= log_one_extent(trans
, inode
, root
, em
, path
);
3558 write_lock(&tree
->lock
);
3559 clear_em_logging(tree
, em
);
3560 free_extent_map(em
);
3562 WARN_ON(!list_empty(&extents
));
3563 write_unlock(&tree
->lock
);
3565 btrfs_release_path(path
);
3569 /* log a single inode in the tree log.
3570 * At least one parent directory for this inode must exist in the tree
3571 * or be logged already.
3573 * Any items from this inode changed by the current transaction are copied
3574 * to the log tree. An extra reference is taken on any extents in this
3575 * file, allowing us to avoid a whole pile of corner cases around logging
3576 * blocks that have been removed from the tree.
3578 * See LOG_INODE_ALL and related defines for a description of what inode_only
3581 * This handles both files and directories.
3583 static int btrfs_log_inode(struct btrfs_trans_handle
*trans
,
3584 struct btrfs_root
*root
, struct inode
*inode
,
3587 struct btrfs_path
*path
;
3588 struct btrfs_path
*dst_path
;
3589 struct btrfs_key min_key
;
3590 struct btrfs_key max_key
;
3591 struct btrfs_root
*log
= root
->log_root
;
3592 struct extent_buffer
*src
= NULL
;
3596 int ins_start_slot
= 0;
3598 bool fast_search
= false;
3599 u64 ino
= btrfs_ino(inode
);
3601 path
= btrfs_alloc_path();
3604 dst_path
= btrfs_alloc_path();
3606 btrfs_free_path(path
);
3610 min_key
.objectid
= ino
;
3611 min_key
.type
= BTRFS_INODE_ITEM_KEY
;
3614 max_key
.objectid
= ino
;
3617 /* today the code can only do partial logging of directories */
3618 if (S_ISDIR(inode
->i_mode
) ||
3619 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3620 &BTRFS_I(inode
)->runtime_flags
) &&
3621 inode_only
== LOG_INODE_EXISTS
))
3622 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
3624 max_key
.type
= (u8
)-1;
3625 max_key
.offset
= (u64
)-1;
3627 /* Only run delayed items if we are a dir or a new file */
3628 if (S_ISDIR(inode
->i_mode
) ||
3629 BTRFS_I(inode
)->generation
> root
->fs_info
->last_trans_committed
) {
3630 ret
= btrfs_commit_inode_delayed_items(trans
, inode
);
3632 btrfs_free_path(path
);
3633 btrfs_free_path(dst_path
);
3638 mutex_lock(&BTRFS_I(inode
)->log_mutex
);
3640 btrfs_get_logged_extents(log
, inode
);
3643 * a brute force approach to making sure we get the most uptodate
3644 * copies of everything.
3646 if (S_ISDIR(inode
->i_mode
)) {
3647 int max_key_type
= BTRFS_DIR_LOG_INDEX_KEY
;
3649 if (inode_only
== LOG_INODE_EXISTS
)
3650 max_key_type
= BTRFS_XATTR_ITEM_KEY
;
3651 ret
= drop_objectid_items(trans
, log
, path
, ino
, max_key_type
);
3653 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3654 &BTRFS_I(inode
)->runtime_flags
)) {
3655 clear_bit(BTRFS_INODE_COPY_EVERYTHING
,
3656 &BTRFS_I(inode
)->runtime_flags
);
3657 ret
= btrfs_truncate_inode_items(trans
, log
,
3659 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING
,
3660 &BTRFS_I(inode
)->runtime_flags
)) {
3661 if (inode_only
== LOG_INODE_ALL
)
3663 max_key
.type
= BTRFS_XATTR_ITEM_KEY
;
3664 ret
= drop_objectid_items(trans
, log
, path
, ino
,
3667 if (inode_only
== LOG_INODE_ALL
)
3669 ret
= log_inode_item(trans
, log
, dst_path
, inode
);
3682 path
->keep_locks
= 1;
3686 ret
= btrfs_search_forward(root
, &min_key
, &max_key
,
3687 path
, trans
->transid
);
3691 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3692 if (min_key
.objectid
!= ino
)
3694 if (min_key
.type
> max_key
.type
)
3697 src
= path
->nodes
[0];
3698 if (ins_nr
&& ins_start_slot
+ ins_nr
== path
->slots
[0]) {
3701 } else if (!ins_nr
) {
3702 ins_start_slot
= path
->slots
[0];
3707 ret
= copy_items(trans
, inode
, dst_path
, src
, ins_start_slot
,
3708 ins_nr
, inode_only
);
3714 ins_start_slot
= path
->slots
[0];
3717 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3719 if (path
->slots
[0] < nritems
) {
3720 btrfs_item_key_to_cpu(path
->nodes
[0], &min_key
,
3725 ret
= copy_items(trans
, inode
, dst_path
, src
,
3727 ins_nr
, inode_only
);
3734 btrfs_release_path(path
);
3736 if (min_key
.offset
< (u64
)-1)
3738 else if (min_key
.type
< (u8
)-1)
3740 else if (min_key
.objectid
< (u64
)-1)
3746 ret
= copy_items(trans
, inode
, dst_path
, src
, ins_start_slot
,
3747 ins_nr
, inode_only
);
3756 btrfs_release_path(path
);
3757 btrfs_release_path(dst_path
);
3759 ret
= btrfs_log_changed_extents(trans
, root
, inode
, dst_path
);
3765 struct extent_map_tree
*tree
= &BTRFS_I(inode
)->extent_tree
;
3766 struct extent_map
*em
, *n
;
3768 write_lock(&tree
->lock
);
3769 list_for_each_entry_safe(em
, n
, &tree
->modified_extents
, list
)
3770 list_del_init(&em
->list
);
3771 write_unlock(&tree
->lock
);
3774 if (inode_only
== LOG_INODE_ALL
&& S_ISDIR(inode
->i_mode
)) {
3775 ret
= log_directory_changes(trans
, root
, inode
, path
, dst_path
);
3781 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
3782 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->last_sub_trans
;
3785 btrfs_free_logged_extents(log
, log
->log_transid
);
3786 mutex_unlock(&BTRFS_I(inode
)->log_mutex
);
3788 btrfs_free_path(path
);
3789 btrfs_free_path(dst_path
);
3794 * follow the dentry parent pointers up the chain and see if any
3795 * of the directories in it require a full commit before they can
3796 * be logged. Returns zero if nothing special needs to be done or 1 if
3797 * a full commit is required.
3799 static noinline
int check_parent_dirs_for_sync(struct btrfs_trans_handle
*trans
,
3800 struct inode
*inode
,
3801 struct dentry
*parent
,
3802 struct super_block
*sb
,
3806 struct btrfs_root
*root
;
3807 struct dentry
*old_parent
= NULL
;
3810 * for regular files, if its inode is already on disk, we don't
3811 * have to worry about the parents at all. This is because
3812 * we can use the last_unlink_trans field to record renames
3813 * and other fun in this file.
3815 if (S_ISREG(inode
->i_mode
) &&
3816 BTRFS_I(inode
)->generation
<= last_committed
&&
3817 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
)
3820 if (!S_ISDIR(inode
->i_mode
)) {
3821 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3823 inode
= parent
->d_inode
;
3827 BTRFS_I(inode
)->logged_trans
= trans
->transid
;
3830 if (BTRFS_I(inode
)->last_unlink_trans
> last_committed
) {
3831 root
= BTRFS_I(inode
)->root
;
3834 * make sure any commits to the log are forced
3835 * to be full commits
3837 root
->fs_info
->last_trans_log_full_commit
=
3843 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3846 if (IS_ROOT(parent
))
3849 parent
= dget_parent(parent
);
3851 old_parent
= parent
;
3852 inode
= parent
->d_inode
;
3861 * helper function around btrfs_log_inode to make sure newly created
3862 * parent directories also end up in the log. A minimal inode and backref
3863 * only logging is done of any parent directories that are older than
3864 * the last committed transaction
3866 static int btrfs_log_inode_parent(struct btrfs_trans_handle
*trans
,
3867 struct btrfs_root
*root
, struct inode
*inode
,
3868 struct dentry
*parent
, int exists_only
)
3870 int inode_only
= exists_only
? LOG_INODE_EXISTS
: LOG_INODE_ALL
;
3871 struct super_block
*sb
;
3872 struct dentry
*old_parent
= NULL
;
3874 u64 last_committed
= root
->fs_info
->last_trans_committed
;
3878 if (btrfs_test_opt(root
, NOTREELOG
)) {
3883 if (root
->fs_info
->last_trans_log_full_commit
>
3884 root
->fs_info
->last_trans_committed
) {
3889 if (root
!= BTRFS_I(inode
)->root
||
3890 btrfs_root_refs(&root
->root_item
) == 0) {
3895 ret
= check_parent_dirs_for_sync(trans
, inode
, parent
,
3896 sb
, last_committed
);
3900 if (btrfs_inode_in_log(inode
, trans
->transid
)) {
3901 ret
= BTRFS_NO_LOG_SYNC
;
3905 ret
= start_log_trans(trans
, root
);
3909 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3914 * for regular files, if its inode is already on disk, we don't
3915 * have to worry about the parents at all. This is because
3916 * we can use the last_unlink_trans field to record renames
3917 * and other fun in this file.
3919 if (S_ISREG(inode
->i_mode
) &&
3920 BTRFS_I(inode
)->generation
<= last_committed
&&
3921 BTRFS_I(inode
)->last_unlink_trans
<= last_committed
) {
3926 inode_only
= LOG_INODE_EXISTS
;
3928 if (!parent
|| !parent
->d_inode
|| sb
!= parent
->d_inode
->i_sb
)
3931 inode
= parent
->d_inode
;
3932 if (root
!= BTRFS_I(inode
)->root
)
3935 if (BTRFS_I(inode
)->generation
>
3936 root
->fs_info
->last_trans_committed
) {
3937 ret
= btrfs_log_inode(trans
, root
, inode
, inode_only
);
3941 if (IS_ROOT(parent
))
3944 parent
= dget_parent(parent
);
3946 old_parent
= parent
;
3952 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
3955 btrfs_end_log_trans(root
);
3961 * it is not safe to log dentry if the chunk root has added new
3962 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3963 * If this returns 1, you must commit the transaction to safely get your
3966 int btrfs_log_dentry_safe(struct btrfs_trans_handle
*trans
,
3967 struct btrfs_root
*root
, struct dentry
*dentry
)
3969 struct dentry
*parent
= dget_parent(dentry
);
3972 ret
= btrfs_log_inode_parent(trans
, root
, dentry
->d_inode
, parent
, 0);
3979 * should be called during mount to recover any replay any log trees
3982 int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree
)
3985 struct btrfs_path
*path
;
3986 struct btrfs_trans_handle
*trans
;
3987 struct btrfs_key key
;
3988 struct btrfs_key found_key
;
3989 struct btrfs_key tmp_key
;
3990 struct btrfs_root
*log
;
3991 struct btrfs_fs_info
*fs_info
= log_root_tree
->fs_info
;
3992 struct walk_control wc
= {
3993 .process_func
= process_one_buffer
,
3997 path
= btrfs_alloc_path();
4001 fs_info
->log_root_recovering
= 1;
4003 trans
= btrfs_start_transaction(fs_info
->tree_root
, 0);
4004 if (IS_ERR(trans
)) {
4005 ret
= PTR_ERR(trans
);
4012 ret
= walk_log_tree(trans
, log_root_tree
, &wc
);
4014 btrfs_error(fs_info
, ret
, "Failed to pin buffers while "
4015 "recovering log root tree.");
4020 key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
4021 key
.offset
= (u64
)-1;
4022 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
4025 ret
= btrfs_search_slot(NULL
, log_root_tree
, &key
, path
, 0, 0);
4028 btrfs_error(fs_info
, ret
,
4029 "Couldn't find tree log root.");
4033 if (path
->slots
[0] == 0)
4037 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
4039 btrfs_release_path(path
);
4040 if (found_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4043 log
= btrfs_read_fs_root(log_root_tree
, &found_key
);
4046 btrfs_error(fs_info
, ret
,
4047 "Couldn't read tree log root.");
4051 tmp_key
.objectid
= found_key
.offset
;
4052 tmp_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4053 tmp_key
.offset
= (u64
)-1;
4055 wc
.replay_dest
= btrfs_read_fs_root_no_name(fs_info
, &tmp_key
);
4056 if (IS_ERR(wc
.replay_dest
)) {
4057 ret
= PTR_ERR(wc
.replay_dest
);
4058 free_extent_buffer(log
->node
);
4059 free_extent_buffer(log
->commit_root
);
4061 btrfs_error(fs_info
, ret
, "Couldn't read target root "
4062 "for tree log recovery.");
4066 wc
.replay_dest
->log_root
= log
;
4067 btrfs_record_root_in_trans(trans
, wc
.replay_dest
);
4068 ret
= walk_log_tree(trans
, log
, &wc
);
4070 if (!ret
&& wc
.stage
== LOG_WALK_REPLAY_ALL
) {
4071 ret
= fixup_inode_link_counts(trans
, wc
.replay_dest
,
4075 key
.offset
= found_key
.offset
- 1;
4076 wc
.replay_dest
->log_root
= NULL
;
4077 free_extent_buffer(log
->node
);
4078 free_extent_buffer(log
->commit_root
);
4084 if (found_key
.offset
== 0)
4087 btrfs_release_path(path
);
4089 /* step one is to pin it all, step two is to replay just inodes */
4092 wc
.process_func
= replay_one_buffer
;
4093 wc
.stage
= LOG_WALK_REPLAY_INODES
;
4096 /* step three is to replay everything */
4097 if (wc
.stage
< LOG_WALK_REPLAY_ALL
) {
4102 btrfs_free_path(path
);
4104 /* step 4: commit the transaction, which also unpins the blocks */
4105 ret
= btrfs_commit_transaction(trans
, fs_info
->tree_root
);
4109 free_extent_buffer(log_root_tree
->node
);
4110 log_root_tree
->log_root
= NULL
;
4111 fs_info
->log_root_recovering
= 0;
4112 kfree(log_root_tree
);
4117 btrfs_end_transaction(wc
.trans
, fs_info
->tree_root
);
4118 btrfs_free_path(path
);
4123 * there are some corner cases where we want to force a full
4124 * commit instead of allowing a directory to be logged.
4126 * They revolve around files there were unlinked from the directory, and
4127 * this function updates the parent directory so that a full commit is
4128 * properly done if it is fsync'd later after the unlinks are done.
4130 void btrfs_record_unlink_dir(struct btrfs_trans_handle
*trans
,
4131 struct inode
*dir
, struct inode
*inode
,
4135 * when we're logging a file, if it hasn't been renamed
4136 * or unlinked, and its inode is fully committed on disk,
4137 * we don't have to worry about walking up the directory chain
4138 * to log its parents.
4140 * So, we use the last_unlink_trans field to put this transid
4141 * into the file. When the file is logged we check it and
4142 * don't log the parents if the file is fully on disk.
4144 if (S_ISREG(inode
->i_mode
))
4145 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
4148 * if this directory was already logged any new
4149 * names for this file/dir will get recorded
4152 if (BTRFS_I(dir
)->logged_trans
== trans
->transid
)
4156 * if the inode we're about to unlink was logged,
4157 * the log will be properly updated for any new names
4159 if (BTRFS_I(inode
)->logged_trans
== trans
->transid
)
4163 * when renaming files across directories, if the directory
4164 * there we're unlinking from gets fsync'd later on, there's
4165 * no way to find the destination directory later and fsync it
4166 * properly. So, we have to be conservative and force commits
4167 * so the new name gets discovered.
4172 /* we can safely do the unlink without any special recording */
4176 BTRFS_I(dir
)->last_unlink_trans
= trans
->transid
;
4180 * Call this after adding a new name for a file and it will properly
4181 * update the log to reflect the new name.
4183 * It will return zero if all goes well, and it will return 1 if a
4184 * full transaction commit is required.
4186 int btrfs_log_new_name(struct btrfs_trans_handle
*trans
,
4187 struct inode
*inode
, struct inode
*old_dir
,
4188 struct dentry
*parent
)
4190 struct btrfs_root
* root
= BTRFS_I(inode
)->root
;
4193 * this will force the logging code to walk the dentry chain
4196 if (S_ISREG(inode
->i_mode
))
4197 BTRFS_I(inode
)->last_unlink_trans
= trans
->transid
;
4200 * if this inode hasn't been logged and directory we're renaming it
4201 * from hasn't been logged, we don't need to log it
4203 if (BTRFS_I(inode
)->logged_trans
<=
4204 root
->fs_info
->last_trans_committed
&&
4205 (!old_dir
|| BTRFS_I(old_dir
)->logged_trans
<=
4206 root
->fs_info
->last_trans_committed
))
4209 return btrfs_log_inode_parent(trans
, root
, inode
, parent
, 1);