2 * Copyright (C) 2007 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.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static const struct extent_io_ops btree_extent_io_ops
;
58 static void end_workqueue_fn(struct btrfs_work
*work
);
59 static void free_fs_root(struct btrfs_root
*root
);
60 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
62 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
64 struct btrfs_root
*root
);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
66 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
67 struct extent_io_tree
*dirty_pages
,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
70 struct extent_io_tree
*pinned_extents
);
71 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
72 static void btrfs_error_commit_super(struct btrfs_root
*root
);
75 * btrfs_end_io_wq structs are used to do processing in task context when an IO
76 * is complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
79 struct btrfs_end_io_wq
{
83 struct btrfs_fs_info
*info
;
85 enum btrfs_wq_endio_type metadata
;
86 struct list_head list
;
87 struct btrfs_work work
;
90 static struct kmem_cache
*btrfs_end_io_wq_cache
;
92 int __init
btrfs_end_io_wq_init(void)
94 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
95 sizeof(struct btrfs_end_io_wq
),
97 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
99 if (!btrfs_end_io_wq_cache
)
104 void btrfs_end_io_wq_exit(void)
106 if (btrfs_end_io_wq_cache
)
107 kmem_cache_destroy(btrfs_end_io_wq_cache
);
111 * async submit bios are used to offload expensive checksumming
112 * onto the worker threads. They checksum file and metadata bios
113 * just before they are sent down the IO stack.
115 struct async_submit_bio
{
118 struct list_head list
;
119 extent_submit_bio_hook_t
*submit_bio_start
;
120 extent_submit_bio_hook_t
*submit_bio_done
;
123 unsigned long bio_flags
;
125 * bio_offset is optional, can be used if the pages in the bio
126 * can't tell us where in the file the bio should go
129 struct btrfs_work work
;
134 * Lockdep class keys for extent_buffer->lock's in this root. For a given
135 * eb, the lockdep key is determined by the btrfs_root it belongs to and
136 * the level the eb occupies in the tree.
138 * Different roots are used for different purposes and may nest inside each
139 * other and they require separate keysets. As lockdep keys should be
140 * static, assign keysets according to the purpose of the root as indicated
141 * by btrfs_root->objectid. This ensures that all special purpose roots
142 * have separate keysets.
144 * Lock-nesting across peer nodes is always done with the immediate parent
145 * node locked thus preventing deadlock. As lockdep doesn't know this, use
146 * subclass to avoid triggering lockdep warning in such cases.
148 * The key is set by the readpage_end_io_hook after the buffer has passed
149 * csum validation but before the pages are unlocked. It is also set by
150 * btrfs_init_new_buffer on freshly allocated blocks.
152 * We also add a check to make sure the highest level of the tree is the
153 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
154 * needs update as well.
156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
157 # if BTRFS_MAX_LEVEL != 8
161 static struct btrfs_lockdep_keyset
{
162 u64 id
; /* root objectid */
163 const char *name_stem
; /* lock name stem */
164 char names
[BTRFS_MAX_LEVEL
+ 1][20];
165 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
166 } btrfs_lockdep_keysets
[] = {
167 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
168 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
169 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
170 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
171 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
172 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
173 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
174 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
175 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
176 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
177 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
178 { .id
= 0, .name_stem
= "tree" },
181 void __init
btrfs_init_lockdep(void)
185 /* initialize lockdep class names */
186 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
187 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
189 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
190 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
191 "btrfs-%s-%02d", ks
->name_stem
, j
);
195 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
198 struct btrfs_lockdep_keyset
*ks
;
200 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
202 /* find the matching keyset, id 0 is the default entry */
203 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
204 if (ks
->id
== objectid
)
207 lockdep_set_class_and_name(&eb
->lock
,
208 &ks
->keys
[level
], ks
->names
[level
]);
214 * extents on the btree inode are pretty simple, there's one extent
215 * that covers the entire device
217 static struct extent_map
*btree_get_extent(struct inode
*inode
,
218 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
221 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
222 struct extent_map
*em
;
225 read_lock(&em_tree
->lock
);
226 em
= lookup_extent_mapping(em_tree
, start
, len
);
229 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
230 read_unlock(&em_tree
->lock
);
233 read_unlock(&em_tree
->lock
);
235 em
= alloc_extent_map();
237 em
= ERR_PTR(-ENOMEM
);
242 em
->block_len
= (u64
)-1;
244 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
246 write_lock(&em_tree
->lock
);
247 ret
= add_extent_mapping(em_tree
, em
, 0);
248 if (ret
== -EEXIST
) {
250 em
= lookup_extent_mapping(em_tree
, start
, len
);
257 write_unlock(&em_tree
->lock
);
263 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
265 return btrfs_crc32c(seed
, data
, len
);
268 void btrfs_csum_final(u32 crc
, char *result
)
270 put_unaligned_le32(~crc
, result
);
274 * compute the csum for a btree block, and either verify it or write it
275 * into the csum field of the block.
277 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
278 struct extent_buffer
*buf
,
281 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
284 unsigned long cur_len
;
285 unsigned long offset
= BTRFS_CSUM_SIZE
;
287 unsigned long map_start
;
288 unsigned long map_len
;
291 unsigned long inline_result
;
293 len
= buf
->len
- offset
;
295 err
= map_private_extent_buffer(buf
, offset
, 32,
296 &kaddr
, &map_start
, &map_len
);
299 cur_len
= min(len
, map_len
- (offset
- map_start
));
300 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
305 if (csum_size
> sizeof(inline_result
)) {
306 result
= kzalloc(csum_size
, GFP_NOFS
);
310 result
= (char *)&inline_result
;
313 btrfs_csum_final(crc
, result
);
316 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
319 memcpy(&found
, result
, csum_size
);
321 read_extent_buffer(buf
, &val
, 0, csum_size
);
322 btrfs_warn_rl(fs_info
,
323 "%s checksum verify failed on %llu wanted %X found %X "
325 fs_info
->sb
->s_id
, buf
->start
,
326 val
, found
, btrfs_header_level(buf
));
327 if (result
!= (char *)&inline_result
)
332 write_extent_buffer(buf
, result
, 0, csum_size
);
334 if (result
!= (char *)&inline_result
)
340 * we can't consider a given block up to date unless the transid of the
341 * block matches the transid in the parent node's pointer. This is how we
342 * detect blocks that either didn't get written at all or got written
343 * in the wrong place.
345 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
346 struct extent_buffer
*eb
, u64 parent_transid
,
349 struct extent_state
*cached_state
= NULL
;
351 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
353 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
360 btrfs_tree_read_lock(eb
);
361 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
364 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
366 if (extent_buffer_uptodate(eb
) &&
367 btrfs_header_generation(eb
) == parent_transid
) {
371 btrfs_err_rl(eb
->fs_info
,
372 "parent transid verify failed on %llu wanted %llu found %llu",
374 parent_transid
, btrfs_header_generation(eb
));
378 * Things reading via commit roots that don't have normal protection,
379 * like send, can have a really old block in cache that may point at a
380 * block that has been free'd and re-allocated. So don't clear uptodate
381 * if we find an eb that is under IO (dirty/writeback) because we could
382 * end up reading in the stale data and then writing it back out and
383 * making everybody very sad.
385 if (!extent_buffer_under_io(eb
))
386 clear_extent_buffer_uptodate(eb
);
388 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
389 &cached_state
, GFP_NOFS
);
391 btrfs_tree_read_unlock_blocking(eb
);
396 * Return 0 if the superblock checksum type matches the checksum value of that
397 * algorithm. Pass the raw disk superblock data.
399 static int btrfs_check_super_csum(char *raw_disk_sb
)
401 struct btrfs_super_block
*disk_sb
=
402 (struct btrfs_super_block
*)raw_disk_sb
;
403 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
406 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
408 const int csum_size
= sizeof(crc
);
409 char result
[csum_size
];
412 * The super_block structure does not span the whole
413 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 * is filled with zeros and is included in the checkum.
416 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
417 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
418 btrfs_csum_final(crc
, result
);
420 if (memcmp(raw_disk_sb
, result
, csum_size
))
424 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
425 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
434 * helper to read a given tree block, doing retries as required when
435 * the checksums don't match and we have alternate mirrors to try.
437 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
438 struct extent_buffer
*eb
,
439 u64 start
, u64 parent_transid
)
441 struct extent_io_tree
*io_tree
;
446 int failed_mirror
= 0;
448 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
449 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
451 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
453 btree_get_extent
, mirror_num
);
455 if (!verify_parent_transid(io_tree
, eb
,
463 * This buffer's crc is fine, but its contents are corrupted, so
464 * there is no reason to read the other copies, they won't be
467 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
470 num_copies
= btrfs_num_copies(root
->fs_info
,
475 if (!failed_mirror
) {
477 failed_mirror
= eb
->read_mirror
;
481 if (mirror_num
== failed_mirror
)
484 if (mirror_num
> num_copies
)
488 if (failed
&& !ret
&& failed_mirror
)
489 repair_eb_io_failure(root
, eb
, failed_mirror
);
495 * checksum a dirty tree block before IO. This has extra checks to make sure
496 * we only fill in the checksum field in the first page of a multi-page block
499 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
501 u64 start
= page_offset(page
);
503 struct extent_buffer
*eb
;
505 eb
= (struct extent_buffer
*)page
->private;
506 if (page
!= eb
->pages
[0])
508 found_start
= btrfs_header_bytenr(eb
);
509 if (WARN_ON(found_start
!= start
|| !PageUptodate(page
)))
511 csum_tree_block(fs_info
, eb
, 0);
515 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
516 struct extent_buffer
*eb
)
518 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
519 u8 fsid
[BTRFS_UUID_SIZE
];
522 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
524 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
528 fs_devices
= fs_devices
->seed
;
533 #define CORRUPT(reason, eb, root, slot) \
534 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
535 "root=%llu, slot=%d", reason, \
536 btrfs_header_bytenr(eb), root->objectid, slot)
538 static noinline
int check_leaf(struct btrfs_root
*root
,
539 struct extent_buffer
*leaf
)
541 struct btrfs_key key
;
542 struct btrfs_key leaf_key
;
543 u32 nritems
= btrfs_header_nritems(leaf
);
549 /* Check the 0 item */
550 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
551 BTRFS_LEAF_DATA_SIZE(root
)) {
552 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
557 * Check to make sure each items keys are in the correct order and their
558 * offsets make sense. We only have to loop through nritems-1 because
559 * we check the current slot against the next slot, which verifies the
560 * next slot's offset+size makes sense and that the current's slot
563 for (slot
= 0; slot
< nritems
- 1; slot
++) {
564 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
565 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
567 /* Make sure the keys are in the right order */
568 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
569 CORRUPT("bad key order", leaf
, root
, slot
);
574 * Make sure the offset and ends are right, remember that the
575 * item data starts at the end of the leaf and grows towards the
578 if (btrfs_item_offset_nr(leaf
, slot
) !=
579 btrfs_item_end_nr(leaf
, slot
+ 1)) {
580 CORRUPT("slot offset bad", leaf
, root
, slot
);
585 * Check to make sure that we don't point outside of the leaf,
586 * just incase all the items are consistent to eachother, but
587 * all point outside of the leaf.
589 if (btrfs_item_end_nr(leaf
, slot
) >
590 BTRFS_LEAF_DATA_SIZE(root
)) {
591 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
599 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
600 u64 phy_offset
, struct page
*page
,
601 u64 start
, u64 end
, int mirror
)
605 struct extent_buffer
*eb
;
606 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
613 eb
= (struct extent_buffer
*)page
->private;
615 /* the pending IO might have been the only thing that kept this buffer
616 * in memory. Make sure we have a ref for all this other checks
618 extent_buffer_get(eb
);
620 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
624 eb
->read_mirror
= mirror
;
625 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
630 found_start
= btrfs_header_bytenr(eb
);
631 if (found_start
!= eb
->start
) {
632 btrfs_err_rl(eb
->fs_info
, "bad tree block start %llu %llu",
633 found_start
, eb
->start
);
637 if (check_tree_block_fsid(root
->fs_info
, eb
)) {
638 btrfs_err_rl(eb
->fs_info
, "bad fsid on block %llu",
643 found_level
= btrfs_header_level(eb
);
644 if (found_level
>= BTRFS_MAX_LEVEL
) {
645 btrfs_err(root
->fs_info
, "bad tree block level %d",
646 (int)btrfs_header_level(eb
));
651 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
654 ret
= csum_tree_block(root
->fs_info
, eb
, 1);
661 * If this is a leaf block and it is corrupt, set the corrupt bit so
662 * that we don't try and read the other copies of this block, just
665 if (found_level
== 0 && check_leaf(root
, eb
)) {
666 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
671 set_extent_buffer_uptodate(eb
);
674 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
675 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
679 * our io error hook is going to dec the io pages
680 * again, we have to make sure it has something
683 atomic_inc(&eb
->io_pages
);
684 clear_extent_buffer_uptodate(eb
);
686 free_extent_buffer(eb
);
691 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
693 struct extent_buffer
*eb
;
694 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
696 eb
= (struct extent_buffer
*)page
->private;
697 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
698 eb
->read_mirror
= failed_mirror
;
699 atomic_dec(&eb
->io_pages
);
700 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
701 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
702 return -EIO
; /* we fixed nothing */
705 static void end_workqueue_bio(struct bio
*bio
)
707 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
708 struct btrfs_fs_info
*fs_info
;
709 struct btrfs_workqueue
*wq
;
710 btrfs_work_func_t func
;
712 fs_info
= end_io_wq
->info
;
713 end_io_wq
->error
= bio
->bi_error
;
715 if (bio
->bi_rw
& REQ_WRITE
) {
716 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
717 wq
= fs_info
->endio_meta_write_workers
;
718 func
= btrfs_endio_meta_write_helper
;
719 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
720 wq
= fs_info
->endio_freespace_worker
;
721 func
= btrfs_freespace_write_helper
;
722 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
723 wq
= fs_info
->endio_raid56_workers
;
724 func
= btrfs_endio_raid56_helper
;
726 wq
= fs_info
->endio_write_workers
;
727 func
= btrfs_endio_write_helper
;
730 if (unlikely(end_io_wq
->metadata
==
731 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
732 wq
= fs_info
->endio_repair_workers
;
733 func
= btrfs_endio_repair_helper
;
734 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
735 wq
= fs_info
->endio_raid56_workers
;
736 func
= btrfs_endio_raid56_helper
;
737 } else if (end_io_wq
->metadata
) {
738 wq
= fs_info
->endio_meta_workers
;
739 func
= btrfs_endio_meta_helper
;
741 wq
= fs_info
->endio_workers
;
742 func
= btrfs_endio_helper
;
746 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
747 btrfs_queue_work(wq
, &end_io_wq
->work
);
750 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
751 enum btrfs_wq_endio_type metadata
)
753 struct btrfs_end_io_wq
*end_io_wq
;
755 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
759 end_io_wq
->private = bio
->bi_private
;
760 end_io_wq
->end_io
= bio
->bi_end_io
;
761 end_io_wq
->info
= info
;
762 end_io_wq
->error
= 0;
763 end_io_wq
->bio
= bio
;
764 end_io_wq
->metadata
= metadata
;
766 bio
->bi_private
= end_io_wq
;
767 bio
->bi_end_io
= end_workqueue_bio
;
771 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
773 unsigned long limit
= min_t(unsigned long,
774 info
->thread_pool_size
,
775 info
->fs_devices
->open_devices
);
779 static void run_one_async_start(struct btrfs_work
*work
)
781 struct async_submit_bio
*async
;
784 async
= container_of(work
, struct async_submit_bio
, work
);
785 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
786 async
->mirror_num
, async
->bio_flags
,
792 static void run_one_async_done(struct btrfs_work
*work
)
794 struct btrfs_fs_info
*fs_info
;
795 struct async_submit_bio
*async
;
798 async
= container_of(work
, struct async_submit_bio
, work
);
799 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
801 limit
= btrfs_async_submit_limit(fs_info
);
802 limit
= limit
* 2 / 3;
805 * atomic_dec_return implies a barrier for waitqueue_active
807 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
808 waitqueue_active(&fs_info
->async_submit_wait
))
809 wake_up(&fs_info
->async_submit_wait
);
811 /* If an error occured we just want to clean up the bio and move on */
813 async
->bio
->bi_error
= async
->error
;
814 bio_endio(async
->bio
);
818 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
819 async
->mirror_num
, async
->bio_flags
,
823 static void run_one_async_free(struct btrfs_work
*work
)
825 struct async_submit_bio
*async
;
827 async
= container_of(work
, struct async_submit_bio
, work
);
831 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
832 int rw
, struct bio
*bio
, int mirror_num
,
833 unsigned long bio_flags
,
835 extent_submit_bio_hook_t
*submit_bio_start
,
836 extent_submit_bio_hook_t
*submit_bio_done
)
838 struct async_submit_bio
*async
;
840 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
844 async
->inode
= inode
;
847 async
->mirror_num
= mirror_num
;
848 async
->submit_bio_start
= submit_bio_start
;
849 async
->submit_bio_done
= submit_bio_done
;
851 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
852 run_one_async_done
, run_one_async_free
);
854 async
->bio_flags
= bio_flags
;
855 async
->bio_offset
= bio_offset
;
859 atomic_inc(&fs_info
->nr_async_submits
);
862 btrfs_set_work_high_priority(&async
->work
);
864 btrfs_queue_work(fs_info
->workers
, &async
->work
);
866 while (atomic_read(&fs_info
->async_submit_draining
) &&
867 atomic_read(&fs_info
->nr_async_submits
)) {
868 wait_event(fs_info
->async_submit_wait
,
869 (atomic_read(&fs_info
->nr_async_submits
) == 0));
875 static int btree_csum_one_bio(struct bio
*bio
)
877 struct bio_vec
*bvec
;
878 struct btrfs_root
*root
;
881 bio_for_each_segment_all(bvec
, bio
, i
) {
882 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
883 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
891 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
892 struct bio
*bio
, int mirror_num
,
893 unsigned long bio_flags
,
897 * when we're called for a write, we're already in the async
898 * submission context. Just jump into btrfs_map_bio
900 return btree_csum_one_bio(bio
);
903 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
904 int mirror_num
, unsigned long bio_flags
,
910 * when we're called for a write, we're already in the async
911 * submission context. Just jump into btrfs_map_bio
913 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
921 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
923 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
926 if (static_cpu_has(X86_FEATURE_XMM4_2
))
932 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
933 int mirror_num
, unsigned long bio_flags
,
936 int async
= check_async_write(inode
, bio_flags
);
939 if (!(rw
& REQ_WRITE
)) {
941 * called for a read, do the setup so that checksum validation
942 * can happen in the async kernel threads
944 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
945 bio
, BTRFS_WQ_ENDIO_METADATA
);
948 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
951 ret
= btree_csum_one_bio(bio
);
954 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
958 * kthread helpers are used to submit writes so that
959 * checksumming can happen in parallel across all CPUs
961 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
962 inode
, rw
, bio
, mirror_num
, 0,
964 __btree_submit_bio_start
,
965 __btree_submit_bio_done
);
978 #ifdef CONFIG_MIGRATION
979 static int btree_migratepage(struct address_space
*mapping
,
980 struct page
*newpage
, struct page
*page
,
981 enum migrate_mode mode
)
984 * we can't safely write a btree page from here,
985 * we haven't done the locking hook
990 * Buffers may be managed in a filesystem specific way.
991 * We must have no buffers or drop them.
993 if (page_has_private(page
) &&
994 !try_to_release_page(page
, GFP_KERNEL
))
996 return migrate_page(mapping
, newpage
, page
, mode
);
1001 static int btree_writepages(struct address_space
*mapping
,
1002 struct writeback_control
*wbc
)
1004 struct btrfs_fs_info
*fs_info
;
1007 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1009 if (wbc
->for_kupdate
)
1012 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1013 /* this is a bit racy, but that's ok */
1014 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1015 BTRFS_DIRTY_METADATA_THRESH
,
1016 fs_info
->dirty_metadata_batch
);
1020 return btree_write_cache_pages(mapping
, wbc
);
1023 static int btree_readpage(struct file
*file
, struct page
*page
)
1025 struct extent_io_tree
*tree
;
1026 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1027 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1030 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1032 if (PageWriteback(page
) || PageDirty(page
))
1035 return try_release_extent_buffer(page
);
1038 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1039 unsigned int length
)
1041 struct extent_io_tree
*tree
;
1042 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1043 extent_invalidatepage(tree
, page
, offset
);
1044 btree_releasepage(page
, GFP_NOFS
);
1045 if (PagePrivate(page
)) {
1046 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1047 "page private not zero on page %llu",
1048 (unsigned long long)page_offset(page
));
1049 ClearPagePrivate(page
);
1050 set_page_private(page
, 0);
1051 page_cache_release(page
);
1055 static int btree_set_page_dirty(struct page
*page
)
1058 struct extent_buffer
*eb
;
1060 BUG_ON(!PagePrivate(page
));
1061 eb
= (struct extent_buffer
*)page
->private;
1063 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1064 BUG_ON(!atomic_read(&eb
->refs
));
1065 btrfs_assert_tree_locked(eb
);
1067 return __set_page_dirty_nobuffers(page
);
1070 static const struct address_space_operations btree_aops
= {
1071 .readpage
= btree_readpage
,
1072 .writepages
= btree_writepages
,
1073 .releasepage
= btree_releasepage
,
1074 .invalidatepage
= btree_invalidatepage
,
1075 #ifdef CONFIG_MIGRATION
1076 .migratepage
= btree_migratepage
,
1078 .set_page_dirty
= btree_set_page_dirty
,
1081 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1083 struct extent_buffer
*buf
= NULL
;
1084 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1086 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1089 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1090 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1091 free_extent_buffer(buf
);
1094 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1095 int mirror_num
, struct extent_buffer
**eb
)
1097 struct extent_buffer
*buf
= NULL
;
1098 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1099 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1102 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1106 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1108 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1109 btree_get_extent
, mirror_num
);
1111 free_extent_buffer(buf
);
1115 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1116 free_extent_buffer(buf
);
1118 } else if (extent_buffer_uptodate(buf
)) {
1121 free_extent_buffer(buf
);
1126 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1129 return find_extent_buffer(fs_info
, bytenr
);
1132 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1135 if (btrfs_test_is_dummy_root(root
))
1136 return alloc_test_extent_buffer(root
->fs_info
, bytenr
);
1137 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1141 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1143 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1144 buf
->start
+ buf
->len
- 1);
1147 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1149 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1150 buf
->start
, buf
->start
+ buf
->len
- 1);
1153 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1156 struct extent_buffer
*buf
= NULL
;
1159 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1161 return ERR_PTR(-ENOMEM
);
1163 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1165 free_extent_buffer(buf
);
1166 return ERR_PTR(ret
);
1172 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1173 struct btrfs_fs_info
*fs_info
,
1174 struct extent_buffer
*buf
)
1176 if (btrfs_header_generation(buf
) ==
1177 fs_info
->running_transaction
->transid
) {
1178 btrfs_assert_tree_locked(buf
);
1180 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1181 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1183 fs_info
->dirty_metadata_batch
);
1184 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1185 btrfs_set_lock_blocking(buf
);
1186 clear_extent_buffer_dirty(buf
);
1191 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1193 struct btrfs_subvolume_writers
*writers
;
1196 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1198 return ERR_PTR(-ENOMEM
);
1200 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_NOFS
);
1203 return ERR_PTR(ret
);
1206 init_waitqueue_head(&writers
->wait
);
1211 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1213 percpu_counter_destroy(&writers
->counter
);
1217 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1218 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1222 root
->commit_root
= NULL
;
1223 root
->sectorsize
= sectorsize
;
1224 root
->nodesize
= nodesize
;
1225 root
->stripesize
= stripesize
;
1227 root
->orphan_cleanup_state
= 0;
1229 root
->objectid
= objectid
;
1230 root
->last_trans
= 0;
1231 root
->highest_objectid
= 0;
1232 root
->nr_delalloc_inodes
= 0;
1233 root
->nr_ordered_extents
= 0;
1235 root
->inode_tree
= RB_ROOT
;
1236 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1237 root
->block_rsv
= NULL
;
1238 root
->orphan_block_rsv
= NULL
;
1240 INIT_LIST_HEAD(&root
->dirty_list
);
1241 INIT_LIST_HEAD(&root
->root_list
);
1242 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1243 INIT_LIST_HEAD(&root
->delalloc_root
);
1244 INIT_LIST_HEAD(&root
->ordered_extents
);
1245 INIT_LIST_HEAD(&root
->ordered_root
);
1246 INIT_LIST_HEAD(&root
->logged_list
[0]);
1247 INIT_LIST_HEAD(&root
->logged_list
[1]);
1248 spin_lock_init(&root
->orphan_lock
);
1249 spin_lock_init(&root
->inode_lock
);
1250 spin_lock_init(&root
->delalloc_lock
);
1251 spin_lock_init(&root
->ordered_extent_lock
);
1252 spin_lock_init(&root
->accounting_lock
);
1253 spin_lock_init(&root
->log_extents_lock
[0]);
1254 spin_lock_init(&root
->log_extents_lock
[1]);
1255 mutex_init(&root
->objectid_mutex
);
1256 mutex_init(&root
->log_mutex
);
1257 mutex_init(&root
->ordered_extent_mutex
);
1258 mutex_init(&root
->delalloc_mutex
);
1259 init_waitqueue_head(&root
->log_writer_wait
);
1260 init_waitqueue_head(&root
->log_commit_wait
[0]);
1261 init_waitqueue_head(&root
->log_commit_wait
[1]);
1262 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1263 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1264 atomic_set(&root
->log_commit
[0], 0);
1265 atomic_set(&root
->log_commit
[1], 0);
1266 atomic_set(&root
->log_writers
, 0);
1267 atomic_set(&root
->log_batch
, 0);
1268 atomic_set(&root
->orphan_inodes
, 0);
1269 atomic_set(&root
->refs
, 1);
1270 atomic_set(&root
->will_be_snapshoted
, 0);
1271 atomic_set(&root
->qgroup_meta_rsv
, 0);
1272 root
->log_transid
= 0;
1273 root
->log_transid_committed
= -1;
1274 root
->last_log_commit
= 0;
1276 extent_io_tree_init(&root
->dirty_log_pages
,
1277 fs_info
->btree_inode
->i_mapping
);
1279 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1280 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1281 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1283 root
->defrag_trans_start
= fs_info
->generation
;
1285 root
->defrag_trans_start
= 0;
1286 root
->root_key
.objectid
= objectid
;
1289 spin_lock_init(&root
->root_item_lock
);
1292 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1294 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1296 root
->fs_info
= fs_info
;
1300 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1301 /* Should only be used by the testing infrastructure */
1302 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1304 struct btrfs_root
*root
;
1306 root
= btrfs_alloc_root(NULL
);
1308 return ERR_PTR(-ENOMEM
);
1309 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1310 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1311 root
->alloc_bytenr
= 0;
1317 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1318 struct btrfs_fs_info
*fs_info
,
1321 struct extent_buffer
*leaf
;
1322 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1323 struct btrfs_root
*root
;
1324 struct btrfs_key key
;
1328 root
= btrfs_alloc_root(fs_info
);
1330 return ERR_PTR(-ENOMEM
);
1332 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1333 tree_root
->stripesize
, root
, fs_info
, objectid
);
1334 root
->root_key
.objectid
= objectid
;
1335 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1336 root
->root_key
.offset
= 0;
1338 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1340 ret
= PTR_ERR(leaf
);
1345 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1346 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1347 btrfs_set_header_generation(leaf
, trans
->transid
);
1348 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1349 btrfs_set_header_owner(leaf
, objectid
);
1352 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1354 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1355 btrfs_header_chunk_tree_uuid(leaf
),
1357 btrfs_mark_buffer_dirty(leaf
);
1359 root
->commit_root
= btrfs_root_node(root
);
1360 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1362 root
->root_item
.flags
= 0;
1363 root
->root_item
.byte_limit
= 0;
1364 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1365 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1366 btrfs_set_root_level(&root
->root_item
, 0);
1367 btrfs_set_root_refs(&root
->root_item
, 1);
1368 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1369 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1370 btrfs_set_root_dirid(&root
->root_item
, 0);
1372 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1373 root
->root_item
.drop_level
= 0;
1375 key
.objectid
= objectid
;
1376 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1378 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1382 btrfs_tree_unlock(leaf
);
1388 btrfs_tree_unlock(leaf
);
1389 free_extent_buffer(root
->commit_root
);
1390 free_extent_buffer(leaf
);
1394 return ERR_PTR(ret
);
1397 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1398 struct btrfs_fs_info
*fs_info
)
1400 struct btrfs_root
*root
;
1401 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1402 struct extent_buffer
*leaf
;
1404 root
= btrfs_alloc_root(fs_info
);
1406 return ERR_PTR(-ENOMEM
);
1408 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1409 tree_root
->stripesize
, root
, fs_info
,
1410 BTRFS_TREE_LOG_OBJECTID
);
1412 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1413 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1414 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1417 * DON'T set REF_COWS for log trees
1419 * log trees do not get reference counted because they go away
1420 * before a real commit is actually done. They do store pointers
1421 * to file data extents, and those reference counts still get
1422 * updated (along with back refs to the log tree).
1425 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1429 return ERR_CAST(leaf
);
1432 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1433 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1434 btrfs_set_header_generation(leaf
, trans
->transid
);
1435 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1436 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1439 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1440 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1441 btrfs_mark_buffer_dirty(root
->node
);
1442 btrfs_tree_unlock(root
->node
);
1446 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1447 struct btrfs_fs_info
*fs_info
)
1449 struct btrfs_root
*log_root
;
1451 log_root
= alloc_log_tree(trans
, fs_info
);
1452 if (IS_ERR(log_root
))
1453 return PTR_ERR(log_root
);
1454 WARN_ON(fs_info
->log_root_tree
);
1455 fs_info
->log_root_tree
= log_root
;
1459 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1460 struct btrfs_root
*root
)
1462 struct btrfs_root
*log_root
;
1463 struct btrfs_inode_item
*inode_item
;
1465 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1466 if (IS_ERR(log_root
))
1467 return PTR_ERR(log_root
);
1469 log_root
->last_trans
= trans
->transid
;
1470 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1472 inode_item
= &log_root
->root_item
.inode
;
1473 btrfs_set_stack_inode_generation(inode_item
, 1);
1474 btrfs_set_stack_inode_size(inode_item
, 3);
1475 btrfs_set_stack_inode_nlink(inode_item
, 1);
1476 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1477 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1479 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1481 WARN_ON(root
->log_root
);
1482 root
->log_root
= log_root
;
1483 root
->log_transid
= 0;
1484 root
->log_transid_committed
= -1;
1485 root
->last_log_commit
= 0;
1489 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1490 struct btrfs_key
*key
)
1492 struct btrfs_root
*root
;
1493 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1494 struct btrfs_path
*path
;
1498 path
= btrfs_alloc_path();
1500 return ERR_PTR(-ENOMEM
);
1502 root
= btrfs_alloc_root(fs_info
);
1508 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1509 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1511 ret
= btrfs_find_root(tree_root
, key
, path
,
1512 &root
->root_item
, &root
->root_key
);
1519 generation
= btrfs_root_generation(&root
->root_item
);
1520 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1522 if (IS_ERR(root
->node
)) {
1523 ret
= PTR_ERR(root
->node
);
1525 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1527 free_extent_buffer(root
->node
);
1530 root
->commit_root
= btrfs_root_node(root
);
1532 btrfs_free_path(path
);
1538 root
= ERR_PTR(ret
);
1542 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1543 struct btrfs_key
*location
)
1545 struct btrfs_root
*root
;
1547 root
= btrfs_read_tree_root(tree_root
, location
);
1551 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1552 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1553 btrfs_check_and_init_root_item(&root
->root_item
);
1559 int btrfs_init_fs_root(struct btrfs_root
*root
)
1562 struct btrfs_subvolume_writers
*writers
;
1564 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1565 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1567 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1572 writers
= btrfs_alloc_subvolume_writers();
1573 if (IS_ERR(writers
)) {
1574 ret
= PTR_ERR(writers
);
1577 root
->subv_writers
= writers
;
1579 btrfs_init_free_ino_ctl(root
);
1580 spin_lock_init(&root
->ino_cache_lock
);
1581 init_waitqueue_head(&root
->ino_cache_wait
);
1583 ret
= get_anon_bdev(&root
->anon_dev
);
1587 mutex_lock(&root
->objectid_mutex
);
1588 ret
= btrfs_find_highest_objectid(root
,
1589 &root
->highest_objectid
);
1591 mutex_unlock(&root
->objectid_mutex
);
1595 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1597 mutex_unlock(&root
->objectid_mutex
);
1602 free_anon_bdev(root
->anon_dev
);
1604 btrfs_free_subvolume_writers(root
->subv_writers
);
1606 kfree(root
->free_ino_ctl
);
1607 kfree(root
->free_ino_pinned
);
1611 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1614 struct btrfs_root
*root
;
1616 spin_lock(&fs_info
->fs_roots_radix_lock
);
1617 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1618 (unsigned long)root_id
);
1619 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1623 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1624 struct btrfs_root
*root
)
1628 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1632 spin_lock(&fs_info
->fs_roots_radix_lock
);
1633 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1634 (unsigned long)root
->root_key
.objectid
,
1637 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1638 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1639 radix_tree_preload_end();
1644 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1645 struct btrfs_key
*location
,
1648 struct btrfs_root
*root
;
1649 struct btrfs_path
*path
;
1650 struct btrfs_key key
;
1653 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1654 return fs_info
->tree_root
;
1655 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1656 return fs_info
->extent_root
;
1657 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1658 return fs_info
->chunk_root
;
1659 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1660 return fs_info
->dev_root
;
1661 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1662 return fs_info
->csum_root
;
1663 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1664 return fs_info
->quota_root
? fs_info
->quota_root
:
1666 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1667 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1670 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1672 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1673 return ERR_PTR(-ENOENT
);
1677 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1681 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1686 ret
= btrfs_init_fs_root(root
);
1690 path
= btrfs_alloc_path();
1695 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1696 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1697 key
.offset
= location
->objectid
;
1699 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1700 btrfs_free_path(path
);
1704 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1706 ret
= btrfs_insert_fs_root(fs_info
, root
);
1708 if (ret
== -EEXIST
) {
1717 return ERR_PTR(ret
);
1720 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1722 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1724 struct btrfs_device
*device
;
1725 struct backing_dev_info
*bdi
;
1728 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1731 bdi
= blk_get_backing_dev_info(device
->bdev
);
1732 if (bdi_congested(bdi
, bdi_bits
)) {
1741 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1745 err
= bdi_setup_and_register(bdi
, "btrfs");
1749 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
;
1750 bdi
->congested_fn
= btrfs_congested_fn
;
1751 bdi
->congested_data
= info
;
1752 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1757 * called by the kthread helper functions to finally call the bio end_io
1758 * functions. This is where read checksum verification actually happens
1760 static void end_workqueue_fn(struct btrfs_work
*work
)
1763 struct btrfs_end_io_wq
*end_io_wq
;
1765 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1766 bio
= end_io_wq
->bio
;
1768 bio
->bi_error
= end_io_wq
->error
;
1769 bio
->bi_private
= end_io_wq
->private;
1770 bio
->bi_end_io
= end_io_wq
->end_io
;
1771 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1775 static int cleaner_kthread(void *arg
)
1777 struct btrfs_root
*root
= arg
;
1779 struct btrfs_trans_handle
*trans
;
1784 /* Make the cleaner go to sleep early. */
1785 if (btrfs_need_cleaner_sleep(root
))
1788 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1792 * Avoid the problem that we change the status of the fs
1793 * during the above check and trylock.
1795 if (btrfs_need_cleaner_sleep(root
)) {
1796 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1800 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1801 btrfs_run_delayed_iputs(root
);
1802 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1804 again
= btrfs_clean_one_deleted_snapshot(root
);
1805 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1808 * The defragger has dealt with the R/O remount and umount,
1809 * needn't do anything special here.
1811 btrfs_run_defrag_inodes(root
->fs_info
);
1814 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1815 * with relocation (btrfs_relocate_chunk) and relocation
1816 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1817 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1818 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1819 * unused block groups.
1821 btrfs_delete_unused_bgs(root
->fs_info
);
1823 if (!try_to_freeze() && !again
) {
1824 set_current_state(TASK_INTERRUPTIBLE
);
1825 if (!kthread_should_stop())
1827 __set_current_state(TASK_RUNNING
);
1829 } while (!kthread_should_stop());
1832 * Transaction kthread is stopped before us and wakes us up.
1833 * However we might have started a new transaction and COWed some
1834 * tree blocks when deleting unused block groups for example. So
1835 * make sure we commit the transaction we started to have a clean
1836 * shutdown when evicting the btree inode - if it has dirty pages
1837 * when we do the final iput() on it, eviction will trigger a
1838 * writeback for it which will fail with null pointer dereferences
1839 * since work queues and other resources were already released and
1840 * destroyed by the time the iput/eviction/writeback is made.
1842 trans
= btrfs_attach_transaction(root
);
1843 if (IS_ERR(trans
)) {
1844 if (PTR_ERR(trans
) != -ENOENT
)
1845 btrfs_err(root
->fs_info
,
1846 "cleaner transaction attach returned %ld",
1851 ret
= btrfs_commit_transaction(trans
, root
);
1853 btrfs_err(root
->fs_info
,
1854 "cleaner open transaction commit returned %d",
1861 static int transaction_kthread(void *arg
)
1863 struct btrfs_root
*root
= arg
;
1864 struct btrfs_trans_handle
*trans
;
1865 struct btrfs_transaction
*cur
;
1868 unsigned long delay
;
1872 cannot_commit
= false;
1873 delay
= HZ
* root
->fs_info
->commit_interval
;
1874 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1876 spin_lock(&root
->fs_info
->trans_lock
);
1877 cur
= root
->fs_info
->running_transaction
;
1879 spin_unlock(&root
->fs_info
->trans_lock
);
1883 now
= get_seconds();
1884 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1885 (now
< cur
->start_time
||
1886 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1887 spin_unlock(&root
->fs_info
->trans_lock
);
1891 transid
= cur
->transid
;
1892 spin_unlock(&root
->fs_info
->trans_lock
);
1894 /* If the file system is aborted, this will always fail. */
1895 trans
= btrfs_attach_transaction(root
);
1896 if (IS_ERR(trans
)) {
1897 if (PTR_ERR(trans
) != -ENOENT
)
1898 cannot_commit
= true;
1901 if (transid
== trans
->transid
) {
1902 btrfs_commit_transaction(trans
, root
);
1904 btrfs_end_transaction(trans
, root
);
1907 wake_up_process(root
->fs_info
->cleaner_kthread
);
1908 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1910 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1911 &root
->fs_info
->fs_state
)))
1912 btrfs_cleanup_transaction(root
);
1913 if (!try_to_freeze()) {
1914 set_current_state(TASK_INTERRUPTIBLE
);
1915 if (!kthread_should_stop() &&
1916 (!btrfs_transaction_blocked(root
->fs_info
) ||
1918 schedule_timeout(delay
);
1919 __set_current_state(TASK_RUNNING
);
1921 } while (!kthread_should_stop());
1926 * this will find the highest generation in the array of
1927 * root backups. The index of the highest array is returned,
1928 * or -1 if we can't find anything.
1930 * We check to make sure the array is valid by comparing the
1931 * generation of the latest root in the array with the generation
1932 * in the super block. If they don't match we pitch it.
1934 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1937 int newest_index
= -1;
1938 struct btrfs_root_backup
*root_backup
;
1941 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1942 root_backup
= info
->super_copy
->super_roots
+ i
;
1943 cur
= btrfs_backup_tree_root_gen(root_backup
);
1944 if (cur
== newest_gen
)
1948 /* check to see if we actually wrapped around */
1949 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1950 root_backup
= info
->super_copy
->super_roots
;
1951 cur
= btrfs_backup_tree_root_gen(root_backup
);
1952 if (cur
== newest_gen
)
1955 return newest_index
;
1960 * find the oldest backup so we know where to store new entries
1961 * in the backup array. This will set the backup_root_index
1962 * field in the fs_info struct
1964 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1967 int newest_index
= -1;
1969 newest_index
= find_newest_super_backup(info
, newest_gen
);
1970 /* if there was garbage in there, just move along */
1971 if (newest_index
== -1) {
1972 info
->backup_root_index
= 0;
1974 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1979 * copy all the root pointers into the super backup array.
1980 * this will bump the backup pointer by one when it is
1983 static void backup_super_roots(struct btrfs_fs_info
*info
)
1986 struct btrfs_root_backup
*root_backup
;
1989 next_backup
= info
->backup_root_index
;
1990 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1991 BTRFS_NUM_BACKUP_ROOTS
;
1994 * just overwrite the last backup if we're at the same generation
1995 * this happens only at umount
1997 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1998 if (btrfs_backup_tree_root_gen(root_backup
) ==
1999 btrfs_header_generation(info
->tree_root
->node
))
2000 next_backup
= last_backup
;
2002 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
2005 * make sure all of our padding and empty slots get zero filled
2006 * regardless of which ones we use today
2008 memset(root_backup
, 0, sizeof(*root_backup
));
2010 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
2012 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
2013 btrfs_set_backup_tree_root_gen(root_backup
,
2014 btrfs_header_generation(info
->tree_root
->node
));
2016 btrfs_set_backup_tree_root_level(root_backup
,
2017 btrfs_header_level(info
->tree_root
->node
));
2019 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2020 btrfs_set_backup_chunk_root_gen(root_backup
,
2021 btrfs_header_generation(info
->chunk_root
->node
));
2022 btrfs_set_backup_chunk_root_level(root_backup
,
2023 btrfs_header_level(info
->chunk_root
->node
));
2025 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2026 btrfs_set_backup_extent_root_gen(root_backup
,
2027 btrfs_header_generation(info
->extent_root
->node
));
2028 btrfs_set_backup_extent_root_level(root_backup
,
2029 btrfs_header_level(info
->extent_root
->node
));
2032 * we might commit during log recovery, which happens before we set
2033 * the fs_root. Make sure it is valid before we fill it in.
2035 if (info
->fs_root
&& info
->fs_root
->node
) {
2036 btrfs_set_backup_fs_root(root_backup
,
2037 info
->fs_root
->node
->start
);
2038 btrfs_set_backup_fs_root_gen(root_backup
,
2039 btrfs_header_generation(info
->fs_root
->node
));
2040 btrfs_set_backup_fs_root_level(root_backup
,
2041 btrfs_header_level(info
->fs_root
->node
));
2044 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2045 btrfs_set_backup_dev_root_gen(root_backup
,
2046 btrfs_header_generation(info
->dev_root
->node
));
2047 btrfs_set_backup_dev_root_level(root_backup
,
2048 btrfs_header_level(info
->dev_root
->node
));
2050 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2051 btrfs_set_backup_csum_root_gen(root_backup
,
2052 btrfs_header_generation(info
->csum_root
->node
));
2053 btrfs_set_backup_csum_root_level(root_backup
,
2054 btrfs_header_level(info
->csum_root
->node
));
2056 btrfs_set_backup_total_bytes(root_backup
,
2057 btrfs_super_total_bytes(info
->super_copy
));
2058 btrfs_set_backup_bytes_used(root_backup
,
2059 btrfs_super_bytes_used(info
->super_copy
));
2060 btrfs_set_backup_num_devices(root_backup
,
2061 btrfs_super_num_devices(info
->super_copy
));
2064 * if we don't copy this out to the super_copy, it won't get remembered
2065 * for the next commit
2067 memcpy(&info
->super_copy
->super_roots
,
2068 &info
->super_for_commit
->super_roots
,
2069 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2073 * this copies info out of the root backup array and back into
2074 * the in-memory super block. It is meant to help iterate through
2075 * the array, so you send it the number of backups you've already
2076 * tried and the last backup index you used.
2078 * this returns -1 when it has tried all the backups
2080 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2081 struct btrfs_super_block
*super
,
2082 int *num_backups_tried
, int *backup_index
)
2084 struct btrfs_root_backup
*root_backup
;
2085 int newest
= *backup_index
;
2087 if (*num_backups_tried
== 0) {
2088 u64 gen
= btrfs_super_generation(super
);
2090 newest
= find_newest_super_backup(info
, gen
);
2094 *backup_index
= newest
;
2095 *num_backups_tried
= 1;
2096 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2097 /* we've tried all the backups, all done */
2100 /* jump to the next oldest backup */
2101 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2102 BTRFS_NUM_BACKUP_ROOTS
;
2103 *backup_index
= newest
;
2104 *num_backups_tried
+= 1;
2106 root_backup
= super
->super_roots
+ newest
;
2108 btrfs_set_super_generation(super
,
2109 btrfs_backup_tree_root_gen(root_backup
));
2110 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2111 btrfs_set_super_root_level(super
,
2112 btrfs_backup_tree_root_level(root_backup
));
2113 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2116 * fixme: the total bytes and num_devices need to match or we should
2119 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2120 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2124 /* helper to cleanup workers */
2125 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2127 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2128 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2129 btrfs_destroy_workqueue(fs_info
->workers
);
2130 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2131 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2132 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2133 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2134 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2135 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2136 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2137 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2138 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2139 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2140 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2141 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2142 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2143 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2144 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2147 static void free_root_extent_buffers(struct btrfs_root
*root
)
2150 free_extent_buffer(root
->node
);
2151 free_extent_buffer(root
->commit_root
);
2153 root
->commit_root
= NULL
;
2157 /* helper to cleanup tree roots */
2158 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2160 free_root_extent_buffers(info
->tree_root
);
2162 free_root_extent_buffers(info
->dev_root
);
2163 free_root_extent_buffers(info
->extent_root
);
2164 free_root_extent_buffers(info
->csum_root
);
2165 free_root_extent_buffers(info
->quota_root
);
2166 free_root_extent_buffers(info
->uuid_root
);
2168 free_root_extent_buffers(info
->chunk_root
);
2171 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2174 struct btrfs_root
*gang
[8];
2177 while (!list_empty(&fs_info
->dead_roots
)) {
2178 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2179 struct btrfs_root
, root_list
);
2180 list_del(&gang
[0]->root_list
);
2182 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2183 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2185 free_extent_buffer(gang
[0]->node
);
2186 free_extent_buffer(gang
[0]->commit_root
);
2187 btrfs_put_fs_root(gang
[0]);
2192 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2197 for (i
= 0; i
< ret
; i
++)
2198 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2201 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2202 btrfs_free_log_root_tree(NULL
, fs_info
);
2203 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2204 fs_info
->pinned_extents
);
2208 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2210 mutex_init(&fs_info
->scrub_lock
);
2211 atomic_set(&fs_info
->scrubs_running
, 0);
2212 atomic_set(&fs_info
->scrub_pause_req
, 0);
2213 atomic_set(&fs_info
->scrubs_paused
, 0);
2214 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2215 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2216 fs_info
->scrub_workers_refcnt
= 0;
2219 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2221 spin_lock_init(&fs_info
->balance_lock
);
2222 mutex_init(&fs_info
->balance_mutex
);
2223 atomic_set(&fs_info
->balance_running
, 0);
2224 atomic_set(&fs_info
->balance_pause_req
, 0);
2225 atomic_set(&fs_info
->balance_cancel_req
, 0);
2226 fs_info
->balance_ctl
= NULL
;
2227 init_waitqueue_head(&fs_info
->balance_wait_q
);
2230 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2231 struct btrfs_root
*tree_root
)
2233 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2234 set_nlink(fs_info
->btree_inode
, 1);
2236 * we set the i_size on the btree inode to the max possible int.
2237 * the real end of the address space is determined by all of
2238 * the devices in the system
2240 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2241 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2243 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2244 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2245 fs_info
->btree_inode
->i_mapping
);
2246 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2247 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2249 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2251 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2252 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2253 sizeof(struct btrfs_key
));
2254 set_bit(BTRFS_INODE_DUMMY
,
2255 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2256 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2259 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2261 fs_info
->dev_replace
.lock_owner
= 0;
2262 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2263 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2264 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2265 mutex_init(&fs_info
->dev_replace
.lock
);
2266 init_waitqueue_head(&fs_info
->replace_wait
);
2269 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2271 spin_lock_init(&fs_info
->qgroup_lock
);
2272 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2273 fs_info
->qgroup_tree
= RB_ROOT
;
2274 fs_info
->qgroup_op_tree
= RB_ROOT
;
2275 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2276 fs_info
->qgroup_seq
= 1;
2277 fs_info
->quota_enabled
= 0;
2278 fs_info
->pending_quota_state
= 0;
2279 fs_info
->qgroup_ulist
= NULL
;
2280 fs_info
->qgroup_rescan_running
= false;
2281 mutex_init(&fs_info
->qgroup_rescan_lock
);
2284 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2285 struct btrfs_fs_devices
*fs_devices
)
2287 int max_active
= fs_info
->thread_pool_size
;
2288 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2291 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2294 fs_info
->delalloc_workers
=
2295 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2297 fs_info
->flush_workers
=
2298 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2300 fs_info
->caching_workers
=
2301 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2304 * a higher idle thresh on the submit workers makes it much more
2305 * likely that bios will be send down in a sane order to the
2308 fs_info
->submit_workers
=
2309 btrfs_alloc_workqueue("submit", flags
,
2310 min_t(u64
, fs_devices
->num_devices
,
2313 fs_info
->fixup_workers
=
2314 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2317 * endios are largely parallel and should have a very
2320 fs_info
->endio_workers
=
2321 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2322 fs_info
->endio_meta_workers
=
2323 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2324 fs_info
->endio_meta_write_workers
=
2325 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2326 fs_info
->endio_raid56_workers
=
2327 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2328 fs_info
->endio_repair_workers
=
2329 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2330 fs_info
->rmw_workers
=
2331 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2332 fs_info
->endio_write_workers
=
2333 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2334 fs_info
->endio_freespace_worker
=
2335 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2336 fs_info
->delayed_workers
=
2337 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2338 fs_info
->readahead_workers
=
2339 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2340 fs_info
->qgroup_rescan_workers
=
2341 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2342 fs_info
->extent_workers
=
2343 btrfs_alloc_workqueue("extent-refs", flags
,
2344 min_t(u64
, fs_devices
->num_devices
,
2347 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2348 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2349 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2350 fs_info
->endio_meta_write_workers
&&
2351 fs_info
->endio_repair_workers
&&
2352 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2353 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2354 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2355 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2356 fs_info
->extent_workers
&&
2357 fs_info
->qgroup_rescan_workers
)) {
2364 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2365 struct btrfs_fs_devices
*fs_devices
)
2368 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2369 struct btrfs_root
*log_tree_root
;
2370 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2371 u64 bytenr
= btrfs_super_log_root(disk_super
);
2373 if (fs_devices
->rw_devices
== 0) {
2374 btrfs_warn(fs_info
, "log replay required on RO media");
2378 log_tree_root
= btrfs_alloc_root(fs_info
);
2382 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2383 tree_root
->stripesize
, log_tree_root
, fs_info
,
2384 BTRFS_TREE_LOG_OBJECTID
);
2386 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2387 fs_info
->generation
+ 1);
2388 if (IS_ERR(log_tree_root
->node
)) {
2389 btrfs_warn(fs_info
, "failed to read log tree");
2390 ret
= PTR_ERR(log_tree_root
->node
);
2391 kfree(log_tree_root
);
2393 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2394 btrfs_err(fs_info
, "failed to read log tree");
2395 free_extent_buffer(log_tree_root
->node
);
2396 kfree(log_tree_root
);
2399 /* returns with log_tree_root freed on success */
2400 ret
= btrfs_recover_log_trees(log_tree_root
);
2402 btrfs_std_error(tree_root
->fs_info
, ret
,
2403 "Failed to recover log tree");
2404 free_extent_buffer(log_tree_root
->node
);
2405 kfree(log_tree_root
);
2409 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2410 ret
= btrfs_commit_super(tree_root
);
2418 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2419 struct btrfs_root
*tree_root
)
2421 struct btrfs_root
*root
;
2422 struct btrfs_key location
;
2425 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2426 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2427 location
.offset
= 0;
2429 root
= btrfs_read_tree_root(tree_root
, &location
);
2431 return PTR_ERR(root
);
2432 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2433 fs_info
->extent_root
= root
;
2435 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2436 root
= btrfs_read_tree_root(tree_root
, &location
);
2438 return PTR_ERR(root
);
2439 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2440 fs_info
->dev_root
= root
;
2441 btrfs_init_devices_late(fs_info
);
2443 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2444 root
= btrfs_read_tree_root(tree_root
, &location
);
2446 return PTR_ERR(root
);
2447 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2448 fs_info
->csum_root
= root
;
2450 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2451 root
= btrfs_read_tree_root(tree_root
, &location
);
2452 if (!IS_ERR(root
)) {
2453 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2454 fs_info
->quota_enabled
= 1;
2455 fs_info
->pending_quota_state
= 1;
2456 fs_info
->quota_root
= root
;
2459 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2460 root
= btrfs_read_tree_root(tree_root
, &location
);
2462 ret
= PTR_ERR(root
);
2466 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2467 fs_info
->uuid_root
= root
;
2473 int open_ctree(struct super_block
*sb
,
2474 struct btrfs_fs_devices
*fs_devices
,
2482 struct btrfs_key location
;
2483 struct buffer_head
*bh
;
2484 struct btrfs_super_block
*disk_super
;
2485 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2486 struct btrfs_root
*tree_root
;
2487 struct btrfs_root
*chunk_root
;
2490 int num_backups_tried
= 0;
2491 int backup_index
= 0;
2494 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2495 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2496 if (!tree_root
|| !chunk_root
) {
2501 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2507 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2513 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2518 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2519 (1 + ilog2(nr_cpu_ids
));
2521 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2524 goto fail_dirty_metadata_bytes
;
2527 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2530 goto fail_delalloc_bytes
;
2533 fs_info
->btree_inode
= new_inode(sb
);
2534 if (!fs_info
->btree_inode
) {
2536 goto fail_bio_counter
;
2539 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2541 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2542 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2543 INIT_LIST_HEAD(&fs_info
->trans_list
);
2544 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2545 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2546 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2547 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2548 spin_lock_init(&fs_info
->delalloc_root_lock
);
2549 spin_lock_init(&fs_info
->trans_lock
);
2550 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2551 spin_lock_init(&fs_info
->delayed_iput_lock
);
2552 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2553 spin_lock_init(&fs_info
->free_chunk_lock
);
2554 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2555 spin_lock_init(&fs_info
->super_lock
);
2556 spin_lock_init(&fs_info
->qgroup_op_lock
);
2557 spin_lock_init(&fs_info
->buffer_lock
);
2558 spin_lock_init(&fs_info
->unused_bgs_lock
);
2559 rwlock_init(&fs_info
->tree_mod_log_lock
);
2560 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2561 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2562 mutex_init(&fs_info
->reloc_mutex
);
2563 mutex_init(&fs_info
->delalloc_root_mutex
);
2564 mutex_init(&fs_info
->cleaner_delayed_iput_mutex
);
2565 seqlock_init(&fs_info
->profiles_lock
);
2567 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2568 INIT_LIST_HEAD(&fs_info
->space_info
);
2569 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2570 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2571 btrfs_mapping_init(&fs_info
->mapping_tree
);
2572 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2573 BTRFS_BLOCK_RSV_GLOBAL
);
2574 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2575 BTRFS_BLOCK_RSV_DELALLOC
);
2576 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2577 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2578 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2579 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2580 BTRFS_BLOCK_RSV_DELOPS
);
2581 atomic_set(&fs_info
->nr_async_submits
, 0);
2582 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2583 atomic_set(&fs_info
->async_submit_draining
, 0);
2584 atomic_set(&fs_info
->nr_async_bios
, 0);
2585 atomic_set(&fs_info
->defrag_running
, 0);
2586 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2587 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2589 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2590 fs_info
->metadata_ratio
= 0;
2591 fs_info
->defrag_inodes
= RB_ROOT
;
2592 fs_info
->free_chunk_space
= 0;
2593 fs_info
->tree_mod_log
= RB_ROOT
;
2594 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2595 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2596 /* readahead state */
2597 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2598 spin_lock_init(&fs_info
->reada_lock
);
2600 fs_info
->thread_pool_size
= min_t(unsigned long,
2601 num_online_cpus() + 2, 8);
2603 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2604 spin_lock_init(&fs_info
->ordered_root_lock
);
2605 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2607 if (!fs_info
->delayed_root
) {
2611 btrfs_init_delayed_root(fs_info
->delayed_root
);
2613 btrfs_init_scrub(fs_info
);
2614 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2615 fs_info
->check_integrity_print_mask
= 0;
2617 btrfs_init_balance(fs_info
);
2618 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2620 sb
->s_blocksize
= 4096;
2621 sb
->s_blocksize_bits
= blksize_bits(4096);
2622 sb
->s_bdi
= &fs_info
->bdi
;
2624 btrfs_init_btree_inode(fs_info
, tree_root
);
2626 spin_lock_init(&fs_info
->block_group_cache_lock
);
2627 fs_info
->block_group_cache_tree
= RB_ROOT
;
2628 fs_info
->first_logical_byte
= (u64
)-1;
2630 extent_io_tree_init(&fs_info
->freed_extents
[0],
2631 fs_info
->btree_inode
->i_mapping
);
2632 extent_io_tree_init(&fs_info
->freed_extents
[1],
2633 fs_info
->btree_inode
->i_mapping
);
2634 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2635 fs_info
->do_barriers
= 1;
2638 mutex_init(&fs_info
->ordered_operations_mutex
);
2639 mutex_init(&fs_info
->tree_log_mutex
);
2640 mutex_init(&fs_info
->chunk_mutex
);
2641 mutex_init(&fs_info
->transaction_kthread_mutex
);
2642 mutex_init(&fs_info
->cleaner_mutex
);
2643 mutex_init(&fs_info
->volume_mutex
);
2644 mutex_init(&fs_info
->ro_block_group_mutex
);
2645 init_rwsem(&fs_info
->commit_root_sem
);
2646 init_rwsem(&fs_info
->cleanup_work_sem
);
2647 init_rwsem(&fs_info
->subvol_sem
);
2648 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2650 btrfs_init_dev_replace_locks(fs_info
);
2651 btrfs_init_qgroup(fs_info
);
2653 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2654 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2656 init_waitqueue_head(&fs_info
->transaction_throttle
);
2657 init_waitqueue_head(&fs_info
->transaction_wait
);
2658 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2659 init_waitqueue_head(&fs_info
->async_submit_wait
);
2661 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2663 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2669 __setup_root(4096, 4096, 4096, tree_root
,
2670 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2672 invalidate_bdev(fs_devices
->latest_bdev
);
2675 * Read super block and check the signature bytes only
2677 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2684 * We want to check superblock checksum, the type is stored inside.
2685 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2687 if (btrfs_check_super_csum(bh
->b_data
)) {
2688 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2695 * super_copy is zeroed at allocation time and we never touch the
2696 * following bytes up to INFO_SIZE, the checksum is calculated from
2697 * the whole block of INFO_SIZE
2699 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2700 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2701 sizeof(*fs_info
->super_for_commit
));
2704 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2706 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2708 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2713 disk_super
= fs_info
->super_copy
;
2714 if (!btrfs_super_root(disk_super
))
2717 /* check FS state, whether FS is broken. */
2718 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2719 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2722 * run through our array of backup supers and setup
2723 * our ring pointer to the oldest one
2725 generation
= btrfs_super_generation(disk_super
);
2726 find_oldest_super_backup(fs_info
, generation
);
2729 * In the long term, we'll store the compression type in the super
2730 * block, and it'll be used for per file compression control.
2732 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2734 ret
= btrfs_parse_options(tree_root
, options
);
2740 features
= btrfs_super_incompat_flags(disk_super
) &
2741 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2743 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2744 "unsupported optional features (%Lx).\n",
2751 * Leafsize and nodesize were always equal, this is only a sanity check.
2753 if (le32_to_cpu(disk_super
->__unused_leafsize
) !=
2754 btrfs_super_nodesize(disk_super
)) {
2755 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2756 "blocksizes don't match. node %d leaf %d\n",
2757 btrfs_super_nodesize(disk_super
),
2758 le32_to_cpu(disk_super
->__unused_leafsize
));
2762 if (btrfs_super_nodesize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2763 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2764 "blocksize (%d) was too large\n",
2765 btrfs_super_nodesize(disk_super
));
2770 features
= btrfs_super_incompat_flags(disk_super
);
2771 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2772 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2773 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2775 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2776 printk(KERN_INFO
"BTRFS: has skinny extents\n");
2779 * flag our filesystem as having big metadata blocks if
2780 * they are bigger than the page size
2782 if (btrfs_super_nodesize(disk_super
) > PAGE_CACHE_SIZE
) {
2783 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2784 printk(KERN_INFO
"BTRFS: flagging fs with big metadata feature\n");
2785 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2788 nodesize
= btrfs_super_nodesize(disk_super
);
2789 sectorsize
= btrfs_super_sectorsize(disk_super
);
2790 stripesize
= btrfs_super_stripesize(disk_super
);
2791 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2792 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2795 * mixed block groups end up with duplicate but slightly offset
2796 * extent buffers for the same range. It leads to corruptions
2798 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2799 (sectorsize
!= nodesize
)) {
2800 printk(KERN_ERR
"BTRFS: unequal leaf/node/sector sizes "
2801 "are not allowed for mixed block groups on %s\n",
2807 * Needn't use the lock because there is no other task which will
2810 btrfs_set_super_incompat_flags(disk_super
, features
);
2812 features
= btrfs_super_compat_ro_flags(disk_super
) &
2813 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2814 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2815 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2816 "unsupported option features (%Lx).\n",
2822 max_active
= fs_info
->thread_pool_size
;
2824 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2827 goto fail_sb_buffer
;
2830 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2831 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2832 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2834 tree_root
->nodesize
= nodesize
;
2835 tree_root
->sectorsize
= sectorsize
;
2836 tree_root
->stripesize
= stripesize
;
2838 sb
->s_blocksize
= sectorsize
;
2839 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2841 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2842 printk(KERN_ERR
"BTRFS: valid FS not found on %s\n", sb
->s_id
);
2843 goto fail_sb_buffer
;
2846 if (sectorsize
!= PAGE_SIZE
) {
2847 printk(KERN_ERR
"BTRFS: incompatible sector size (%lu) "
2848 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2849 goto fail_sb_buffer
;
2852 mutex_lock(&fs_info
->chunk_mutex
);
2853 ret
= btrfs_read_sys_array(tree_root
);
2854 mutex_unlock(&fs_info
->chunk_mutex
);
2856 printk(KERN_ERR
"BTRFS: failed to read the system "
2857 "array on %s\n", sb
->s_id
);
2858 goto fail_sb_buffer
;
2861 generation
= btrfs_super_chunk_root_generation(disk_super
);
2863 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2864 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2866 chunk_root
->node
= read_tree_block(chunk_root
,
2867 btrfs_super_chunk_root(disk_super
),
2869 if (IS_ERR(chunk_root
->node
) ||
2870 !extent_buffer_uptodate(chunk_root
->node
)) {
2871 printk(KERN_ERR
"BTRFS: failed to read chunk root on %s\n",
2873 if (!IS_ERR(chunk_root
->node
))
2874 free_extent_buffer(chunk_root
->node
);
2875 chunk_root
->node
= NULL
;
2876 goto fail_tree_roots
;
2878 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2879 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2881 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2882 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2884 ret
= btrfs_read_chunk_tree(chunk_root
);
2886 printk(KERN_ERR
"BTRFS: failed to read chunk tree on %s\n",
2888 goto fail_tree_roots
;
2892 * keep the device that is marked to be the target device for the
2893 * dev_replace procedure
2895 btrfs_close_extra_devices(fs_devices
, 0);
2897 if (!fs_devices
->latest_bdev
) {
2898 printk(KERN_ERR
"BTRFS: failed to read devices on %s\n",
2900 goto fail_tree_roots
;
2904 generation
= btrfs_super_generation(disk_super
);
2906 tree_root
->node
= read_tree_block(tree_root
,
2907 btrfs_super_root(disk_super
),
2909 if (IS_ERR(tree_root
->node
) ||
2910 !extent_buffer_uptodate(tree_root
->node
)) {
2911 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2913 if (!IS_ERR(tree_root
->node
))
2914 free_extent_buffer(tree_root
->node
);
2915 tree_root
->node
= NULL
;
2916 goto recovery_tree_root
;
2919 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2920 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2921 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2923 mutex_lock(&tree_root
->objectid_mutex
);
2924 ret
= btrfs_find_highest_objectid(tree_root
,
2925 &tree_root
->highest_objectid
);
2927 mutex_unlock(&tree_root
->objectid_mutex
);
2928 goto recovery_tree_root
;
2931 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2933 mutex_unlock(&tree_root
->objectid_mutex
);
2935 ret
= btrfs_read_roots(fs_info
, tree_root
);
2937 goto recovery_tree_root
;
2939 fs_info
->generation
= generation
;
2940 fs_info
->last_trans_committed
= generation
;
2942 ret
= btrfs_recover_balance(fs_info
);
2944 printk(KERN_ERR
"BTRFS: failed to recover balance\n");
2945 goto fail_block_groups
;
2948 ret
= btrfs_init_dev_stats(fs_info
);
2950 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2952 goto fail_block_groups
;
2955 ret
= btrfs_init_dev_replace(fs_info
);
2957 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2958 goto fail_block_groups
;
2961 btrfs_close_extra_devices(fs_devices
, 1);
2963 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2965 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret
);
2966 goto fail_block_groups
;
2969 ret
= btrfs_sysfs_add_device(fs_devices
);
2971 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret
);
2972 goto fail_fsdev_sysfs
;
2975 ret
= btrfs_sysfs_add_mounted(fs_info
);
2977 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2978 goto fail_fsdev_sysfs
;
2981 ret
= btrfs_init_space_info(fs_info
);
2983 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2987 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2989 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
2992 fs_info
->num_tolerated_disk_barrier_failures
=
2993 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2994 if (fs_info
->fs_devices
->missing_devices
>
2995 fs_info
->num_tolerated_disk_barrier_failures
&&
2996 !(sb
->s_flags
& MS_RDONLY
)) {
2997 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2998 fs_info
->fs_devices
->missing_devices
,
2999 fs_info
->num_tolerated_disk_barrier_failures
);
3003 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3005 if (IS_ERR(fs_info
->cleaner_kthread
))
3008 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3010 "btrfs-transaction");
3011 if (IS_ERR(fs_info
->transaction_kthread
))
3014 if (!btrfs_test_opt(tree_root
, SSD
) &&
3015 !btrfs_test_opt(tree_root
, NOSSD
) &&
3016 !fs_info
->fs_devices
->rotating
) {
3017 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
3019 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
3023 * Mount does not set all options immediatelly, we can do it now and do
3024 * not have to wait for transaction commit
3026 btrfs_apply_pending_changes(fs_info
);
3028 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3029 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
3030 ret
= btrfsic_mount(tree_root
, fs_devices
,
3031 btrfs_test_opt(tree_root
,
3032 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3034 fs_info
->check_integrity_print_mask
);
3036 printk(KERN_WARNING
"BTRFS: failed to initialize"
3037 " integrity check module %s\n", sb
->s_id
);
3040 ret
= btrfs_read_qgroup_config(fs_info
);
3042 goto fail_trans_kthread
;
3044 /* do not make disk changes in broken FS */
3045 if (btrfs_super_log_root(disk_super
) != 0) {
3046 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3053 ret
= btrfs_find_orphan_roots(tree_root
);
3057 if (!(sb
->s_flags
& MS_RDONLY
)) {
3058 ret
= btrfs_cleanup_fs_roots(fs_info
);
3062 mutex_lock(&fs_info
->cleaner_mutex
);
3063 ret
= btrfs_recover_relocation(tree_root
);
3064 mutex_unlock(&fs_info
->cleaner_mutex
);
3067 "BTRFS: failed to recover relocation\n");
3073 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3074 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3075 location
.offset
= 0;
3077 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3078 if (IS_ERR(fs_info
->fs_root
)) {
3079 err
= PTR_ERR(fs_info
->fs_root
);
3083 if (sb
->s_flags
& MS_RDONLY
)
3086 down_read(&fs_info
->cleanup_work_sem
);
3087 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3088 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3089 up_read(&fs_info
->cleanup_work_sem
);
3090 close_ctree(tree_root
);
3093 up_read(&fs_info
->cleanup_work_sem
);
3095 ret
= btrfs_resume_balance_async(fs_info
);
3097 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
3098 close_ctree(tree_root
);
3102 ret
= btrfs_resume_dev_replace_async(fs_info
);
3104 pr_warn("BTRFS: failed to resume dev_replace\n");
3105 close_ctree(tree_root
);
3109 btrfs_qgroup_rescan_resume(fs_info
);
3111 if (!fs_info
->uuid_root
) {
3112 pr_info("BTRFS: creating UUID tree\n");
3113 ret
= btrfs_create_uuid_tree(fs_info
);
3115 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3117 close_ctree(tree_root
);
3120 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3121 fs_info
->generation
!=
3122 btrfs_super_uuid_tree_generation(disk_super
)) {
3123 pr_info("BTRFS: checking UUID tree\n");
3124 ret
= btrfs_check_uuid_tree(fs_info
);
3126 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3128 close_ctree(tree_root
);
3132 fs_info
->update_uuid_tree_gen
= 1;
3140 btrfs_free_qgroup_config(fs_info
);
3142 kthread_stop(fs_info
->transaction_kthread
);
3143 btrfs_cleanup_transaction(fs_info
->tree_root
);
3144 btrfs_free_fs_roots(fs_info
);
3146 kthread_stop(fs_info
->cleaner_kthread
);
3149 * make sure we're done with the btree inode before we stop our
3152 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3155 btrfs_sysfs_remove_mounted(fs_info
);
3158 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3161 btrfs_put_block_group_cache(fs_info
);
3162 btrfs_free_block_groups(fs_info
);
3165 free_root_pointers(fs_info
, 1);
3166 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3169 btrfs_stop_all_workers(fs_info
);
3172 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3174 iput(fs_info
->btree_inode
);
3176 percpu_counter_destroy(&fs_info
->bio_counter
);
3177 fail_delalloc_bytes
:
3178 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3179 fail_dirty_metadata_bytes
:
3180 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3182 bdi_destroy(&fs_info
->bdi
);
3184 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3186 btrfs_free_stripe_hash_table(fs_info
);
3187 btrfs_close_devices(fs_info
->fs_devices
);
3191 if (!btrfs_test_opt(tree_root
, RECOVERY
))
3192 goto fail_tree_roots
;
3194 free_root_pointers(fs_info
, 0);
3196 /* don't use the log in recovery mode, it won't be valid */
3197 btrfs_set_super_log_root(disk_super
, 0);
3199 /* we can't trust the free space cache either */
3200 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3202 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3203 &num_backups_tried
, &backup_index
);
3205 goto fail_block_groups
;
3206 goto retry_root_backup
;
3209 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3212 set_buffer_uptodate(bh
);
3214 struct btrfs_device
*device
= (struct btrfs_device
*)
3217 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3218 "lost page write due to IO error on %s",
3219 rcu_str_deref(device
->name
));
3220 /* note, we dont' set_buffer_write_io_error because we have
3221 * our own ways of dealing with the IO errors
3223 clear_buffer_uptodate(bh
);
3224 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3230 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3231 struct buffer_head
**bh_ret
)
3233 struct buffer_head
*bh
;
3234 struct btrfs_super_block
*super
;
3237 bytenr
= btrfs_sb_offset(copy_num
);
3238 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3241 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3243 * If we fail to read from the underlying devices, as of now
3244 * the best option we have is to mark it EIO.
3249 super
= (struct btrfs_super_block
*)bh
->b_data
;
3250 if (btrfs_super_bytenr(super
) != bytenr
||
3251 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3261 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3263 struct buffer_head
*bh
;
3264 struct buffer_head
*latest
= NULL
;
3265 struct btrfs_super_block
*super
;
3270 /* we would like to check all the supers, but that would make
3271 * a btrfs mount succeed after a mkfs from a different FS.
3272 * So, we need to add a special mount option to scan for
3273 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3275 for (i
= 0; i
< 1; i
++) {
3276 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3280 super
= (struct btrfs_super_block
*)bh
->b_data
;
3282 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3285 transid
= btrfs_super_generation(super
);
3292 return ERR_PTR(ret
);
3298 * this should be called twice, once with wait == 0 and
3299 * once with wait == 1. When wait == 0 is done, all the buffer heads
3300 * we write are pinned.
3302 * They are released when wait == 1 is done.
3303 * max_mirrors must be the same for both runs, and it indicates how
3304 * many supers on this one device should be written.
3306 * max_mirrors == 0 means to write them all.
3308 static int write_dev_supers(struct btrfs_device
*device
,
3309 struct btrfs_super_block
*sb
,
3310 int do_barriers
, int wait
, int max_mirrors
)
3312 struct buffer_head
*bh
;
3319 if (max_mirrors
== 0)
3320 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3322 for (i
= 0; i
< max_mirrors
; i
++) {
3323 bytenr
= btrfs_sb_offset(i
);
3324 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3325 device
->commit_total_bytes
)
3329 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3330 BTRFS_SUPER_INFO_SIZE
);
3336 if (!buffer_uptodate(bh
))
3339 /* drop our reference */
3342 /* drop the reference from the wait == 0 run */
3346 btrfs_set_super_bytenr(sb
, bytenr
);
3349 crc
= btrfs_csum_data((char *)sb
+
3350 BTRFS_CSUM_SIZE
, crc
,
3351 BTRFS_SUPER_INFO_SIZE
-
3353 btrfs_csum_final(crc
, sb
->csum
);
3356 * one reference for us, and we leave it for the
3359 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3360 BTRFS_SUPER_INFO_SIZE
);
3362 btrfs_err(device
->dev_root
->fs_info
,
3363 "couldn't get super buffer head for bytenr %llu",
3369 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3371 /* one reference for submit_bh */
3374 set_buffer_uptodate(bh
);
3376 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3377 bh
->b_private
= device
;
3381 * we fua the first super. The others we allow
3385 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3387 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3391 return errors
< i
? 0 : -1;
3395 * endio for the write_dev_flush, this will wake anyone waiting
3396 * for the barrier when it is done
3398 static void btrfs_end_empty_barrier(struct bio
*bio
)
3400 if (bio
->bi_private
)
3401 complete(bio
->bi_private
);
3406 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3407 * sent down. With wait == 1, it waits for the previous flush.
3409 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3412 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3417 if (device
->nobarriers
)
3421 bio
= device
->flush_bio
;
3425 wait_for_completion(&device
->flush_wait
);
3427 if (bio
->bi_error
) {
3428 ret
= bio
->bi_error
;
3429 btrfs_dev_stat_inc_and_print(device
,
3430 BTRFS_DEV_STAT_FLUSH_ERRS
);
3433 /* drop the reference from the wait == 0 run */
3435 device
->flush_bio
= NULL
;
3441 * one reference for us, and we leave it for the
3444 device
->flush_bio
= NULL
;
3445 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3449 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3450 bio
->bi_bdev
= device
->bdev
;
3451 init_completion(&device
->flush_wait
);
3452 bio
->bi_private
= &device
->flush_wait
;
3453 device
->flush_bio
= bio
;
3456 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3462 * send an empty flush down to each device in parallel,
3463 * then wait for them
3465 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3467 struct list_head
*head
;
3468 struct btrfs_device
*dev
;
3469 int errors_send
= 0;
3470 int errors_wait
= 0;
3473 /* send down all the barriers */
3474 head
= &info
->fs_devices
->devices
;
3475 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3482 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3485 ret
= write_dev_flush(dev
, 0);
3490 /* wait for all the barriers */
3491 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3498 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3501 ret
= write_dev_flush(dev
, 1);
3505 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3506 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3511 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3514 int min_tolerated
= INT_MAX
;
3516 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3517 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3518 min_tolerated
= min(min_tolerated
,
3519 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3520 tolerated_failures
);
3522 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3523 if (raid_type
== BTRFS_RAID_SINGLE
)
3525 if (!(flags
& btrfs_raid_group
[raid_type
]))
3527 min_tolerated
= min(min_tolerated
,
3528 btrfs_raid_array
[raid_type
].
3529 tolerated_failures
);
3532 if (min_tolerated
== INT_MAX
) {
3533 pr_warn("BTRFS: unknown raid flag: %llu\n", flags
);
3537 return min_tolerated
;
3540 int btrfs_calc_num_tolerated_disk_barrier_failures(
3541 struct btrfs_fs_info
*fs_info
)
3543 struct btrfs_ioctl_space_info space
;
3544 struct btrfs_space_info
*sinfo
;
3545 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3546 BTRFS_BLOCK_GROUP_SYSTEM
,
3547 BTRFS_BLOCK_GROUP_METADATA
,
3548 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3551 int num_tolerated_disk_barrier_failures
=
3552 (int)fs_info
->fs_devices
->num_devices
;
3554 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3555 struct btrfs_space_info
*tmp
;
3559 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3560 if (tmp
->flags
== types
[i
]) {
3570 down_read(&sinfo
->groups_sem
);
3571 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3574 if (list_empty(&sinfo
->block_groups
[c
]))
3577 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3579 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3581 flags
= space
.flags
;
3583 num_tolerated_disk_barrier_failures
= min(
3584 num_tolerated_disk_barrier_failures
,
3585 btrfs_get_num_tolerated_disk_barrier_failures(
3588 up_read(&sinfo
->groups_sem
);
3591 return num_tolerated_disk_barrier_failures
;
3594 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3596 struct list_head
*head
;
3597 struct btrfs_device
*dev
;
3598 struct btrfs_super_block
*sb
;
3599 struct btrfs_dev_item
*dev_item
;
3603 int total_errors
= 0;
3606 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3607 backup_super_roots(root
->fs_info
);
3609 sb
= root
->fs_info
->super_for_commit
;
3610 dev_item
= &sb
->dev_item
;
3612 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3613 head
= &root
->fs_info
->fs_devices
->devices
;
3614 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3617 ret
= barrier_all_devices(root
->fs_info
);
3620 &root
->fs_info
->fs_devices
->device_list_mutex
);
3621 btrfs_std_error(root
->fs_info
, ret
,
3622 "errors while submitting device barriers.");
3627 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3632 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3635 btrfs_set_stack_device_generation(dev_item
, 0);
3636 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3637 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3638 btrfs_set_stack_device_total_bytes(dev_item
,
3639 dev
->commit_total_bytes
);
3640 btrfs_set_stack_device_bytes_used(dev_item
,
3641 dev
->commit_bytes_used
);
3642 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3643 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3644 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3645 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3646 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3648 flags
= btrfs_super_flags(sb
);
3649 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3651 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3655 if (total_errors
> max_errors
) {
3656 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3658 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3660 /* FUA is masked off if unsupported and can't be the reason */
3661 btrfs_std_error(root
->fs_info
, -EIO
,
3662 "%d errors while writing supers", total_errors
);
3667 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3670 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3673 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3677 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3678 if (total_errors
> max_errors
) {
3679 btrfs_std_error(root
->fs_info
, -EIO
,
3680 "%d errors while writing supers", total_errors
);
3686 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3687 struct btrfs_root
*root
, int max_mirrors
)
3689 return write_all_supers(root
, max_mirrors
);
3692 /* Drop a fs root from the radix tree and free it. */
3693 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3694 struct btrfs_root
*root
)
3696 spin_lock(&fs_info
->fs_roots_radix_lock
);
3697 radix_tree_delete(&fs_info
->fs_roots_radix
,
3698 (unsigned long)root
->root_key
.objectid
);
3699 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3701 if (btrfs_root_refs(&root
->root_item
) == 0)
3702 synchronize_srcu(&fs_info
->subvol_srcu
);
3704 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3705 btrfs_free_log(NULL
, root
);
3707 if (root
->free_ino_pinned
)
3708 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3709 if (root
->free_ino_ctl
)
3710 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3714 static void free_fs_root(struct btrfs_root
*root
)
3716 iput(root
->ino_cache_inode
);
3717 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3718 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3719 root
->orphan_block_rsv
= NULL
;
3721 free_anon_bdev(root
->anon_dev
);
3722 if (root
->subv_writers
)
3723 btrfs_free_subvolume_writers(root
->subv_writers
);
3724 free_extent_buffer(root
->node
);
3725 free_extent_buffer(root
->commit_root
);
3726 kfree(root
->free_ino_ctl
);
3727 kfree(root
->free_ino_pinned
);
3729 btrfs_put_fs_root(root
);
3732 void btrfs_free_fs_root(struct btrfs_root
*root
)
3737 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3739 u64 root_objectid
= 0;
3740 struct btrfs_root
*gang
[8];
3743 unsigned int ret
= 0;
3747 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3748 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3749 (void **)gang
, root_objectid
,
3752 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3755 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3757 for (i
= 0; i
< ret
; i
++) {
3758 /* Avoid to grab roots in dead_roots */
3759 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3763 /* grab all the search result for later use */
3764 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3766 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3768 for (i
= 0; i
< ret
; i
++) {
3771 root_objectid
= gang
[i
]->root_key
.objectid
;
3772 err
= btrfs_orphan_cleanup(gang
[i
]);
3775 btrfs_put_fs_root(gang
[i
]);
3780 /* release the uncleaned roots due to error */
3781 for (; i
< ret
; i
++) {
3783 btrfs_put_fs_root(gang
[i
]);
3788 int btrfs_commit_super(struct btrfs_root
*root
)
3790 struct btrfs_trans_handle
*trans
;
3792 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3793 btrfs_run_delayed_iputs(root
);
3794 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3795 wake_up_process(root
->fs_info
->cleaner_kthread
);
3797 /* wait until ongoing cleanup work done */
3798 down_write(&root
->fs_info
->cleanup_work_sem
);
3799 up_write(&root
->fs_info
->cleanup_work_sem
);
3801 trans
= btrfs_join_transaction(root
);
3803 return PTR_ERR(trans
);
3804 return btrfs_commit_transaction(trans
, root
);
3807 void close_ctree(struct btrfs_root
*root
)
3809 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3812 fs_info
->closing
= 1;
3815 /* wait for the qgroup rescan worker to stop */
3816 btrfs_qgroup_wait_for_completion(fs_info
, false);
3818 /* wait for the uuid_scan task to finish */
3819 down(&fs_info
->uuid_tree_rescan_sem
);
3820 /* avoid complains from lockdep et al., set sem back to initial state */
3821 up(&fs_info
->uuid_tree_rescan_sem
);
3823 /* pause restriper - we want to resume on mount */
3824 btrfs_pause_balance(fs_info
);
3826 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3828 btrfs_scrub_cancel(fs_info
);
3830 /* wait for any defraggers to finish */
3831 wait_event(fs_info
->transaction_wait
,
3832 (atomic_read(&fs_info
->defrag_running
) == 0));
3834 /* clear out the rbtree of defraggable inodes */
3835 btrfs_cleanup_defrag_inodes(fs_info
);
3837 cancel_work_sync(&fs_info
->async_reclaim_work
);
3839 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3841 * If the cleaner thread is stopped and there are
3842 * block groups queued for removal, the deletion will be
3843 * skipped when we quit the cleaner thread.
3845 btrfs_delete_unused_bgs(root
->fs_info
);
3847 ret
= btrfs_commit_super(root
);
3849 btrfs_err(fs_info
, "commit super ret %d", ret
);
3852 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3853 btrfs_error_commit_super(root
);
3855 kthread_stop(fs_info
->transaction_kthread
);
3856 kthread_stop(fs_info
->cleaner_kthread
);
3858 fs_info
->closing
= 2;
3861 btrfs_free_qgroup_config(fs_info
);
3863 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3864 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3865 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3868 btrfs_sysfs_remove_mounted(fs_info
);
3869 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3871 btrfs_free_fs_roots(fs_info
);
3873 btrfs_put_block_group_cache(fs_info
);
3875 btrfs_free_block_groups(fs_info
);
3878 * we must make sure there is not any read request to
3879 * submit after we stopping all workers.
3881 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3882 btrfs_stop_all_workers(fs_info
);
3885 free_root_pointers(fs_info
, 1);
3887 iput(fs_info
->btree_inode
);
3889 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3890 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3891 btrfsic_unmount(root
, fs_info
->fs_devices
);
3894 btrfs_close_devices(fs_info
->fs_devices
);
3895 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3897 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3898 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3899 percpu_counter_destroy(&fs_info
->bio_counter
);
3900 bdi_destroy(&fs_info
->bdi
);
3901 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3903 btrfs_free_stripe_hash_table(fs_info
);
3905 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3906 root
->orphan_block_rsv
= NULL
;
3909 while (!list_empty(&fs_info
->pinned_chunks
)) {
3910 struct extent_map
*em
;
3912 em
= list_first_entry(&fs_info
->pinned_chunks
,
3913 struct extent_map
, list
);
3914 list_del_init(&em
->list
);
3915 free_extent_map(em
);
3917 unlock_chunks(root
);
3920 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3924 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3926 ret
= extent_buffer_uptodate(buf
);
3930 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3931 parent_transid
, atomic
);
3937 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3939 return set_extent_buffer_uptodate(buf
);
3942 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3944 struct btrfs_root
*root
;
3945 u64 transid
= btrfs_header_generation(buf
);
3948 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3950 * This is a fast path so only do this check if we have sanity tests
3951 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3952 * outside of the sanity tests.
3954 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3957 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3958 btrfs_assert_tree_locked(buf
);
3959 if (transid
!= root
->fs_info
->generation
)
3960 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3961 "found %llu running %llu\n",
3962 buf
->start
, transid
, root
->fs_info
->generation
);
3963 was_dirty
= set_extent_buffer_dirty(buf
);
3965 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3967 root
->fs_info
->dirty_metadata_batch
);
3968 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3969 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3970 btrfs_print_leaf(root
, buf
);
3976 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3980 * looks as though older kernels can get into trouble with
3981 * this code, they end up stuck in balance_dirty_pages forever
3985 if (current
->flags
& PF_MEMALLOC
)
3989 btrfs_balance_delayed_items(root
);
3991 ret
= __percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3992 BTRFS_DIRTY_METADATA_THRESH
,
3993 root
->fs_info
->dirty_metadata_batch
);
3995 balance_dirty_pages_ratelimited(
3996 root
->fs_info
->btree_inode
->i_mapping
);
4001 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
4003 __btrfs_btree_balance_dirty(root
, 1);
4006 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
4008 __btrfs_btree_balance_dirty(root
, 0);
4011 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
4013 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4014 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
4017 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
4020 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
4023 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4024 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
4025 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
4028 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4029 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
4030 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
4033 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4034 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
4035 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
4040 * The common minimum, we don't know if we can trust the nodesize/sectorsize
4041 * items yet, they'll be verified later. Issue just a warning.
4043 if (!IS_ALIGNED(btrfs_super_root(sb
), 4096))
4044 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
4045 btrfs_super_root(sb
));
4046 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), 4096))
4047 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
4048 btrfs_super_chunk_root(sb
));
4049 if (!IS_ALIGNED(btrfs_super_log_root(sb
), 4096))
4050 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
4051 btrfs_super_log_root(sb
));
4054 * Check the lower bound, the alignment and other constraints are
4057 if (btrfs_super_nodesize(sb
) < 4096) {
4058 printk(KERN_ERR
"BTRFS: nodesize too small: %u < 4096\n",
4059 btrfs_super_nodesize(sb
));
4062 if (btrfs_super_sectorsize(sb
) < 4096) {
4063 printk(KERN_ERR
"BTRFS: sectorsize too small: %u < 4096\n",
4064 btrfs_super_sectorsize(sb
));
4068 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4069 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4070 fs_info
->fsid
, sb
->dev_item
.fsid
);
4075 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4078 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4079 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
4080 btrfs_super_num_devices(sb
));
4081 if (btrfs_super_num_devices(sb
) == 0) {
4082 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
4086 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4087 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
4088 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4093 * Obvious sys_chunk_array corruptions, it must hold at least one key
4096 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4097 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
4098 btrfs_super_sys_array_size(sb
),
4099 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4102 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4103 + sizeof(struct btrfs_chunk
)) {
4104 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
4105 btrfs_super_sys_array_size(sb
),
4106 sizeof(struct btrfs_disk_key
)
4107 + sizeof(struct btrfs_chunk
));
4112 * The generation is a global counter, we'll trust it more than the others
4113 * but it's still possible that it's the one that's wrong.
4115 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4117 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4118 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
4119 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4120 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4122 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4123 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
4128 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4130 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4131 btrfs_run_delayed_iputs(root
);
4132 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4134 down_write(&root
->fs_info
->cleanup_work_sem
);
4135 up_write(&root
->fs_info
->cleanup_work_sem
);
4137 /* cleanup FS via transaction */
4138 btrfs_cleanup_transaction(root
);
4141 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4143 struct btrfs_ordered_extent
*ordered
;
4145 spin_lock(&root
->ordered_extent_lock
);
4147 * This will just short circuit the ordered completion stuff which will
4148 * make sure the ordered extent gets properly cleaned up.
4150 list_for_each_entry(ordered
, &root
->ordered_extents
,
4152 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4153 spin_unlock(&root
->ordered_extent_lock
);
4156 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4158 struct btrfs_root
*root
;
4159 struct list_head splice
;
4161 INIT_LIST_HEAD(&splice
);
4163 spin_lock(&fs_info
->ordered_root_lock
);
4164 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4165 while (!list_empty(&splice
)) {
4166 root
= list_first_entry(&splice
, struct btrfs_root
,
4168 list_move_tail(&root
->ordered_root
,
4169 &fs_info
->ordered_roots
);
4171 spin_unlock(&fs_info
->ordered_root_lock
);
4172 btrfs_destroy_ordered_extents(root
);
4175 spin_lock(&fs_info
->ordered_root_lock
);
4177 spin_unlock(&fs_info
->ordered_root_lock
);
4180 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4181 struct btrfs_root
*root
)
4183 struct rb_node
*node
;
4184 struct btrfs_delayed_ref_root
*delayed_refs
;
4185 struct btrfs_delayed_ref_node
*ref
;
4188 delayed_refs
= &trans
->delayed_refs
;
4190 spin_lock(&delayed_refs
->lock
);
4191 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4192 spin_unlock(&delayed_refs
->lock
);
4193 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4197 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4198 struct btrfs_delayed_ref_head
*head
;
4199 struct btrfs_delayed_ref_node
*tmp
;
4200 bool pin_bytes
= false;
4202 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4204 if (!mutex_trylock(&head
->mutex
)) {
4205 atomic_inc(&head
->node
.refs
);
4206 spin_unlock(&delayed_refs
->lock
);
4208 mutex_lock(&head
->mutex
);
4209 mutex_unlock(&head
->mutex
);
4210 btrfs_put_delayed_ref(&head
->node
);
4211 spin_lock(&delayed_refs
->lock
);
4214 spin_lock(&head
->lock
);
4215 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4218 list_del(&ref
->list
);
4219 atomic_dec(&delayed_refs
->num_entries
);
4220 btrfs_put_delayed_ref(ref
);
4222 if (head
->must_insert_reserved
)
4224 btrfs_free_delayed_extent_op(head
->extent_op
);
4225 delayed_refs
->num_heads
--;
4226 if (head
->processing
== 0)
4227 delayed_refs
->num_heads_ready
--;
4228 atomic_dec(&delayed_refs
->num_entries
);
4229 head
->node
.in_tree
= 0;
4230 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4231 spin_unlock(&head
->lock
);
4232 spin_unlock(&delayed_refs
->lock
);
4233 mutex_unlock(&head
->mutex
);
4236 btrfs_pin_extent(root
, head
->node
.bytenr
,
4237 head
->node
.num_bytes
, 1);
4238 btrfs_put_delayed_ref(&head
->node
);
4240 spin_lock(&delayed_refs
->lock
);
4243 spin_unlock(&delayed_refs
->lock
);
4248 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4250 struct btrfs_inode
*btrfs_inode
;
4251 struct list_head splice
;
4253 INIT_LIST_HEAD(&splice
);
4255 spin_lock(&root
->delalloc_lock
);
4256 list_splice_init(&root
->delalloc_inodes
, &splice
);
4258 while (!list_empty(&splice
)) {
4259 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4262 list_del_init(&btrfs_inode
->delalloc_inodes
);
4263 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4264 &btrfs_inode
->runtime_flags
);
4265 spin_unlock(&root
->delalloc_lock
);
4267 btrfs_invalidate_inodes(btrfs_inode
->root
);
4269 spin_lock(&root
->delalloc_lock
);
4272 spin_unlock(&root
->delalloc_lock
);
4275 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4277 struct btrfs_root
*root
;
4278 struct list_head splice
;
4280 INIT_LIST_HEAD(&splice
);
4282 spin_lock(&fs_info
->delalloc_root_lock
);
4283 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4284 while (!list_empty(&splice
)) {
4285 root
= list_first_entry(&splice
, struct btrfs_root
,
4287 list_del_init(&root
->delalloc_root
);
4288 root
= btrfs_grab_fs_root(root
);
4290 spin_unlock(&fs_info
->delalloc_root_lock
);
4292 btrfs_destroy_delalloc_inodes(root
);
4293 btrfs_put_fs_root(root
);
4295 spin_lock(&fs_info
->delalloc_root_lock
);
4297 spin_unlock(&fs_info
->delalloc_root_lock
);
4300 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4301 struct extent_io_tree
*dirty_pages
,
4305 struct extent_buffer
*eb
;
4310 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4315 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4316 while (start
<= end
) {
4317 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4318 start
+= root
->nodesize
;
4321 wait_on_extent_buffer_writeback(eb
);
4323 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4325 clear_extent_buffer_dirty(eb
);
4326 free_extent_buffer_stale(eb
);
4333 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4334 struct extent_io_tree
*pinned_extents
)
4336 struct extent_io_tree
*unpin
;
4342 unpin
= pinned_extents
;
4345 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4346 EXTENT_DIRTY
, NULL
);
4350 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4351 btrfs_error_unpin_extent_range(root
, start
, end
);
4356 if (unpin
== &root
->fs_info
->freed_extents
[0])
4357 unpin
= &root
->fs_info
->freed_extents
[1];
4359 unpin
= &root
->fs_info
->freed_extents
[0];
4367 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4368 struct btrfs_root
*root
)
4370 btrfs_destroy_delayed_refs(cur_trans
, root
);
4372 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4373 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4375 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4376 wake_up(&root
->fs_info
->transaction_wait
);
4378 btrfs_destroy_delayed_inodes(root
);
4379 btrfs_assert_delayed_root_empty(root
);
4381 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4383 btrfs_destroy_pinned_extent(root
,
4384 root
->fs_info
->pinned_extents
);
4386 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4387 wake_up(&cur_trans
->commit_wait
);
4390 memset(cur_trans, 0, sizeof(*cur_trans));
4391 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4395 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4397 struct btrfs_transaction
*t
;
4399 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4401 spin_lock(&root
->fs_info
->trans_lock
);
4402 while (!list_empty(&root
->fs_info
->trans_list
)) {
4403 t
= list_first_entry(&root
->fs_info
->trans_list
,
4404 struct btrfs_transaction
, list
);
4405 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4406 atomic_inc(&t
->use_count
);
4407 spin_unlock(&root
->fs_info
->trans_lock
);
4408 btrfs_wait_for_commit(root
, t
->transid
);
4409 btrfs_put_transaction(t
);
4410 spin_lock(&root
->fs_info
->trans_lock
);
4413 if (t
== root
->fs_info
->running_transaction
) {
4414 t
->state
= TRANS_STATE_COMMIT_DOING
;
4415 spin_unlock(&root
->fs_info
->trans_lock
);
4417 * We wait for 0 num_writers since we don't hold a trans
4418 * handle open currently for this transaction.
4420 wait_event(t
->writer_wait
,
4421 atomic_read(&t
->num_writers
) == 0);
4423 spin_unlock(&root
->fs_info
->trans_lock
);
4425 btrfs_cleanup_one_transaction(t
, root
);
4427 spin_lock(&root
->fs_info
->trans_lock
);
4428 if (t
== root
->fs_info
->running_transaction
)
4429 root
->fs_info
->running_transaction
= NULL
;
4430 list_del_init(&t
->list
);
4431 spin_unlock(&root
->fs_info
->trans_lock
);
4433 btrfs_put_transaction(t
);
4434 trace_btrfs_transaction_commit(root
);
4435 spin_lock(&root
->fs_info
->trans_lock
);
4437 spin_unlock(&root
->fs_info
->trans_lock
);
4438 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4439 btrfs_destroy_delayed_inodes(root
);
4440 btrfs_assert_delayed_root_empty(root
);
4441 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4442 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4443 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4448 static const struct extent_io_ops btree_extent_io_ops
= {
4449 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4450 .readpage_io_failed_hook
= btree_io_failed_hook
,
4451 .submit_bio_hook
= btree_submit_bio_hook
,
4452 /* note we're sharing with inode.c for the merge bio hook */
4453 .merge_bio_hook
= btrfs_merge_bio_hook
,