1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
28 #include "print-tree.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
44 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
45 BTRFS_HEADER_FLAG_RELOC |\
46 BTRFS_SUPER_FLAG_ERROR |\
47 BTRFS_SUPER_FLAG_SEEDING |\
48 BTRFS_SUPER_FLAG_METADUMP |\
49 BTRFS_SUPER_FLAG_METADUMP_V2)
51 static const struct extent_io_ops btree_extent_io_ops
;
52 static void end_workqueue_fn(struct btrfs_work
*work
);
53 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
55 struct btrfs_fs_info
*fs_info
);
56 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
57 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
58 struct extent_io_tree
*dirty_pages
,
60 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
61 struct extent_io_tree
*pinned_extents
);
62 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
);
63 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
);
66 * btrfs_end_io_wq structs are used to do processing in task context when an IO
67 * is complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
70 struct btrfs_end_io_wq
{
74 struct btrfs_fs_info
*info
;
76 enum btrfs_wq_endio_type metadata
;
77 struct btrfs_work work
;
80 static struct kmem_cache
*btrfs_end_io_wq_cache
;
82 int __init
btrfs_end_io_wq_init(void)
84 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
85 sizeof(struct btrfs_end_io_wq
),
89 if (!btrfs_end_io_wq_cache
)
94 void __cold
btrfs_end_io_wq_exit(void)
96 kmem_cache_destroy(btrfs_end_io_wq_cache
);
100 * async submit bios are used to offload expensive checksumming
101 * onto the worker threads. They checksum file and metadata bios
102 * just before they are sent down the IO stack.
104 struct async_submit_bio
{
107 extent_submit_bio_start_t
*submit_bio_start
;
110 * bio_offset is optional, can be used if the pages in the bio
111 * can't tell us where in the file the bio should go
114 struct btrfs_work work
;
119 * Lockdep class keys for extent_buffer->lock's in this root. For a given
120 * eb, the lockdep key is determined by the btrfs_root it belongs to and
121 * the level the eb occupies in the tree.
123 * Different roots are used for different purposes and may nest inside each
124 * other and they require separate keysets. As lockdep keys should be
125 * static, assign keysets according to the purpose of the root as indicated
126 * by btrfs_root->root_key.objectid. This ensures that all special purpose
127 * roots have separate keysets.
129 * Lock-nesting across peer nodes is always done with the immediate parent
130 * node locked thus preventing deadlock. As lockdep doesn't know this, use
131 * subclass to avoid triggering lockdep warning in such cases.
133 * The key is set by the readpage_end_io_hook after the buffer has passed
134 * csum validation but before the pages are unlocked. It is also set by
135 * btrfs_init_new_buffer on freshly allocated blocks.
137 * We also add a check to make sure the highest level of the tree is the
138 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
139 * needs update as well.
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 # if BTRFS_MAX_LEVEL != 8
146 static struct btrfs_lockdep_keyset
{
147 u64 id
; /* root objectid */
148 const char *name_stem
; /* lock name stem */
149 char names
[BTRFS_MAX_LEVEL
+ 1][20];
150 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
151 } btrfs_lockdep_keysets
[] = {
152 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
153 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
154 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
155 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
156 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
157 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
158 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
159 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
160 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
161 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
162 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
163 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
164 { .id
= 0, .name_stem
= "tree" },
167 void __init
btrfs_init_lockdep(void)
171 /* initialize lockdep class names */
172 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
173 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
175 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
176 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
177 "btrfs-%s-%02d", ks
->name_stem
, j
);
181 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
184 struct btrfs_lockdep_keyset
*ks
;
186 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
188 /* find the matching keyset, id 0 is the default entry */
189 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
190 if (ks
->id
== objectid
)
193 lockdep_set_class_and_name(&eb
->lock
,
194 &ks
->keys
[level
], ks
->names
[level
]);
200 * extents on the btree inode are pretty simple, there's one extent
201 * that covers the entire device
203 struct extent_map
*btree_get_extent(struct btrfs_inode
*inode
,
204 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
207 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
208 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
209 struct extent_map
*em
;
212 read_lock(&em_tree
->lock
);
213 em
= lookup_extent_mapping(em_tree
, start
, len
);
215 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
216 read_unlock(&em_tree
->lock
);
219 read_unlock(&em_tree
->lock
);
221 em
= alloc_extent_map();
223 em
= ERR_PTR(-ENOMEM
);
228 em
->block_len
= (u64
)-1;
230 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
232 write_lock(&em_tree
->lock
);
233 ret
= add_extent_mapping(em_tree
, em
, 0);
234 if (ret
== -EEXIST
) {
236 em
= lookup_extent_mapping(em_tree
, start
, len
);
243 write_unlock(&em_tree
->lock
);
250 * Compute the csum of a btree block and store the result to provided buffer.
252 * Returns error if the extent buffer cannot be mapped.
254 static int csum_tree_block(struct extent_buffer
*buf
, u8
*result
)
256 struct btrfs_fs_info
*fs_info
= buf
->fs_info
;
257 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
259 unsigned long cur_len
;
260 unsigned long offset
= BTRFS_CSUM_SIZE
;
262 unsigned long map_start
;
263 unsigned long map_len
;
266 shash
->tfm
= fs_info
->csum_shash
;
267 crypto_shash_init(shash
);
269 len
= buf
->len
- offset
;
273 * Note: we don't need to check for the err == 1 case here, as
274 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
275 * and 'min_len = 32' and the currently implemented mapping
276 * algorithm we cannot cross a page boundary.
278 err
= map_private_extent_buffer(buf
, offset
, 32,
279 &kaddr
, &map_start
, &map_len
);
282 cur_len
= min(len
, map_len
- (offset
- map_start
));
283 crypto_shash_update(shash
, kaddr
+ offset
- map_start
, cur_len
);
287 memset(result
, 0, BTRFS_CSUM_SIZE
);
289 crypto_shash_final(shash
, result
);
295 * we can't consider a given block up to date unless the transid of the
296 * block matches the transid in the parent node's pointer. This is how we
297 * detect blocks that either didn't get written at all or got written
298 * in the wrong place.
300 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
301 struct extent_buffer
*eb
, u64 parent_transid
,
304 struct extent_state
*cached_state
= NULL
;
306 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
308 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
315 btrfs_tree_read_lock(eb
);
316 btrfs_set_lock_blocking_read(eb
);
319 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
321 if (extent_buffer_uptodate(eb
) &&
322 btrfs_header_generation(eb
) == parent_transid
) {
326 btrfs_err_rl(eb
->fs_info
,
327 "parent transid verify failed on %llu wanted %llu found %llu",
329 parent_transid
, btrfs_header_generation(eb
));
333 * Things reading via commit roots that don't have normal protection,
334 * like send, can have a really old block in cache that may point at a
335 * block that has been freed and re-allocated. So don't clear uptodate
336 * if we find an eb that is under IO (dirty/writeback) because we could
337 * end up reading in the stale data and then writing it back out and
338 * making everybody very sad.
340 if (!extent_buffer_under_io(eb
))
341 clear_extent_buffer_uptodate(eb
);
343 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
346 btrfs_tree_read_unlock_blocking(eb
);
350 static bool btrfs_supported_super_csum(u16 csum_type
)
353 case BTRFS_CSUM_TYPE_CRC32
:
361 * Return 0 if the superblock checksum type matches the checksum value of that
362 * algorithm. Pass the raw disk superblock data.
364 static int btrfs_check_super_csum(struct btrfs_fs_info
*fs_info
,
367 struct btrfs_super_block
*disk_sb
=
368 (struct btrfs_super_block
*)raw_disk_sb
;
369 char result
[BTRFS_CSUM_SIZE
];
370 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
372 shash
->tfm
= fs_info
->csum_shash
;
373 crypto_shash_init(shash
);
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
378 * filled with zeros and is included in the checksum.
380 crypto_shash_update(shash
, raw_disk_sb
+ BTRFS_CSUM_SIZE
,
381 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
382 crypto_shash_final(shash
, result
);
384 if (memcmp(disk_sb
->csum
, result
, btrfs_super_csum_size(disk_sb
)))
390 int btrfs_verify_level_key(struct extent_buffer
*eb
, int level
,
391 struct btrfs_key
*first_key
, u64 parent_transid
)
393 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
395 struct btrfs_key found_key
;
398 found_level
= btrfs_header_level(eb
);
399 if (found_level
!= level
) {
400 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG
),
401 KERN_ERR
"BTRFS: tree level check failed\n");
403 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
404 eb
->start
, level
, found_level
);
412 * For live tree block (new tree blocks in current transaction),
413 * we need proper lock context to avoid race, which is impossible here.
414 * So we only checks tree blocks which is read from disk, whose
415 * generation <= fs_info->last_trans_committed.
417 if (btrfs_header_generation(eb
) > fs_info
->last_trans_committed
)
420 /* We have @first_key, so this @eb must have at least one item */
421 if (btrfs_header_nritems(eb
) == 0) {
423 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
425 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
430 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
432 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
433 ret
= btrfs_comp_cpu_keys(first_key
, &found_key
);
436 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG
),
437 KERN_ERR
"BTRFS: tree first key check failed\n");
439 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
440 eb
->start
, parent_transid
, first_key
->objectid
,
441 first_key
->type
, first_key
->offset
,
442 found_key
.objectid
, found_key
.type
,
449 * helper to read a given tree block, doing retries as required when
450 * the checksums don't match and we have alternate mirrors to try.
452 * @parent_transid: expected transid, skip check if 0
453 * @level: expected level, mandatory check
454 * @first_key: expected key of first slot, skip check if NULL
456 static int btree_read_extent_buffer_pages(struct extent_buffer
*eb
,
457 u64 parent_transid
, int level
,
458 struct btrfs_key
*first_key
)
460 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
461 struct extent_io_tree
*io_tree
;
466 int failed_mirror
= 0;
468 io_tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
470 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
471 ret
= read_extent_buffer_pages(eb
, WAIT_COMPLETE
, mirror_num
);
473 if (verify_parent_transid(io_tree
, eb
,
476 else if (btrfs_verify_level_key(eb
, level
,
477 first_key
, parent_transid
))
483 num_copies
= btrfs_num_copies(fs_info
,
488 if (!failed_mirror
) {
490 failed_mirror
= eb
->read_mirror
;
494 if (mirror_num
== failed_mirror
)
497 if (mirror_num
> num_copies
)
501 if (failed
&& !ret
&& failed_mirror
)
502 btrfs_repair_eb_io_failure(eb
, failed_mirror
);
508 * checksum a dirty tree block before IO. This has extra checks to make sure
509 * we only fill in the checksum field in the first page of a multi-page block
512 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
514 u64 start
= page_offset(page
);
516 u8 result
[BTRFS_CSUM_SIZE
];
517 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
518 struct extent_buffer
*eb
;
521 eb
= (struct extent_buffer
*)page
->private;
522 if (page
!= eb
->pages
[0])
525 found_start
= btrfs_header_bytenr(eb
);
527 * Please do not consolidate these warnings into a single if.
528 * It is useful to know what went wrong.
530 if (WARN_ON(found_start
!= start
))
532 if (WARN_ON(!PageUptodate(page
)))
535 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fs_devices
->metadata_uuid
,
536 btrfs_header_fsid(), BTRFS_FSID_SIZE
) == 0);
538 if (csum_tree_block(eb
, result
))
541 if (btrfs_header_level(eb
))
542 ret
= btrfs_check_node(eb
);
544 ret
= btrfs_check_leaf_full(eb
);
548 "block=%llu write time tree block corruption detected",
552 write_extent_buffer(eb
, result
, 0, csum_size
);
557 static int check_tree_block_fsid(struct extent_buffer
*eb
)
559 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
560 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
561 u8 fsid
[BTRFS_FSID_SIZE
];
564 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
569 * Checking the incompat flag is only valid for the current
570 * fs. For seed devices it's forbidden to have their uuid
571 * changed so reading ->fsid in this case is fine
573 if (fs_devices
== fs_info
->fs_devices
&&
574 btrfs_fs_incompat(fs_info
, METADATA_UUID
))
575 metadata_uuid
= fs_devices
->metadata_uuid
;
577 metadata_uuid
= fs_devices
->fsid
;
579 if (!memcmp(fsid
, metadata_uuid
, BTRFS_FSID_SIZE
)) {
583 fs_devices
= fs_devices
->seed
;
588 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
589 u64 phy_offset
, struct page
*page
,
590 u64 start
, u64 end
, int mirror
)
594 struct extent_buffer
*eb
;
595 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
596 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
597 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
599 u8 result
[BTRFS_CSUM_SIZE
];
605 eb
= (struct extent_buffer
*)page
->private;
607 /* the pending IO might have been the only thing that kept this buffer
608 * in memory. Make sure we have a ref for all this other checks
610 extent_buffer_get(eb
);
612 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
616 eb
->read_mirror
= mirror
;
617 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
622 found_start
= btrfs_header_bytenr(eb
);
623 if (found_start
!= eb
->start
) {
624 btrfs_err_rl(fs_info
, "bad tree block start, want %llu have %llu",
625 eb
->start
, found_start
);
629 if (check_tree_block_fsid(eb
)) {
630 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
635 found_level
= btrfs_header_level(eb
);
636 if (found_level
>= BTRFS_MAX_LEVEL
) {
637 btrfs_err(fs_info
, "bad tree block level %d on %llu",
638 (int)btrfs_header_level(eb
), eb
->start
);
643 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
646 ret
= csum_tree_block(eb
, result
);
650 if (memcmp_extent_buffer(eb
, result
, 0, csum_size
)) {
654 memcpy(&found
, result
, csum_size
);
656 read_extent_buffer(eb
, &val
, 0, csum_size
);
657 btrfs_warn_rl(fs_info
,
658 "%s checksum verify failed on %llu wanted %x found %x level %d",
659 fs_info
->sb
->s_id
, eb
->start
,
660 val
, found
, btrfs_header_level(eb
));
666 * If this is a leaf block and it is corrupt, set the corrupt bit so
667 * that we don't try and read the other copies of this block, just
670 if (found_level
== 0 && btrfs_check_leaf_full(eb
)) {
671 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
675 if (found_level
> 0 && btrfs_check_node(eb
))
679 set_extent_buffer_uptodate(eb
);
682 "block=%llu read time tree block corruption detected",
686 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
687 btree_readahead_hook(eb
, ret
);
691 * our io error hook is going to dec the io pages
692 * again, we have to make sure it has something
695 atomic_inc(&eb
->io_pages
);
696 clear_extent_buffer_uptodate(eb
);
698 free_extent_buffer(eb
);
703 static void end_workqueue_bio(struct bio
*bio
)
705 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
706 struct btrfs_fs_info
*fs_info
;
707 struct btrfs_workqueue
*wq
;
708 btrfs_work_func_t func
;
710 fs_info
= end_io_wq
->info
;
711 end_io_wq
->status
= bio
->bi_status
;
713 if (bio_op(bio
) == REQ_OP_WRITE
) {
714 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
715 wq
= fs_info
->endio_meta_write_workers
;
716 func
= btrfs_endio_meta_write_helper
;
717 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
718 wq
= fs_info
->endio_freespace_worker
;
719 func
= btrfs_freespace_write_helper
;
720 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
721 wq
= fs_info
->endio_raid56_workers
;
722 func
= btrfs_endio_raid56_helper
;
724 wq
= fs_info
->endio_write_workers
;
725 func
= btrfs_endio_write_helper
;
728 if (unlikely(end_io_wq
->metadata
==
729 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
730 wq
= fs_info
->endio_repair_workers
;
731 func
= btrfs_endio_repair_helper
;
732 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
733 wq
= fs_info
->endio_raid56_workers
;
734 func
= btrfs_endio_raid56_helper
;
735 } else if (end_io_wq
->metadata
) {
736 wq
= fs_info
->endio_meta_workers
;
737 func
= btrfs_endio_meta_helper
;
739 wq
= fs_info
->endio_workers
;
740 func
= btrfs_endio_helper
;
744 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
745 btrfs_queue_work(wq
, &end_io_wq
->work
);
748 blk_status_t
btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
749 enum btrfs_wq_endio_type metadata
)
751 struct btrfs_end_io_wq
*end_io_wq
;
753 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
755 return BLK_STS_RESOURCE
;
757 end_io_wq
->private = bio
->bi_private
;
758 end_io_wq
->end_io
= bio
->bi_end_io
;
759 end_io_wq
->info
= info
;
760 end_io_wq
->status
= 0;
761 end_io_wq
->bio
= bio
;
762 end_io_wq
->metadata
= metadata
;
764 bio
->bi_private
= end_io_wq
;
765 bio
->bi_end_io
= end_workqueue_bio
;
769 static void run_one_async_start(struct btrfs_work
*work
)
771 struct async_submit_bio
*async
;
774 async
= container_of(work
, struct async_submit_bio
, work
);
775 ret
= async
->submit_bio_start(async
->private_data
, async
->bio
,
782 * In order to insert checksums into the metadata in large chunks, we wait
783 * until bio submission time. All the pages in the bio are checksummed and
784 * sums are attached onto the ordered extent record.
786 * At IO completion time the csums attached on the ordered extent record are
787 * inserted into the tree.
789 static void run_one_async_done(struct btrfs_work
*work
)
791 struct async_submit_bio
*async
;
795 async
= container_of(work
, struct async_submit_bio
, work
);
796 inode
= async
->private_data
;
798 /* If an error occurred we just want to clean up the bio and move on */
800 async
->bio
->bi_status
= async
->status
;
801 bio_endio(async
->bio
);
805 ret
= btrfs_map_bio(btrfs_sb(inode
->i_sb
), async
->bio
,
806 async
->mirror_num
, 1);
808 async
->bio
->bi_status
= ret
;
809 bio_endio(async
->bio
);
813 static void run_one_async_free(struct btrfs_work
*work
)
815 struct async_submit_bio
*async
;
817 async
= container_of(work
, struct async_submit_bio
, work
);
821 blk_status_t
btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
822 int mirror_num
, unsigned long bio_flags
,
823 u64 bio_offset
, void *private_data
,
824 extent_submit_bio_start_t
*submit_bio_start
)
826 struct async_submit_bio
*async
;
828 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
830 return BLK_STS_RESOURCE
;
832 async
->private_data
= private_data
;
834 async
->mirror_num
= mirror_num
;
835 async
->submit_bio_start
= submit_bio_start
;
837 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
838 run_one_async_done
, run_one_async_free
);
840 async
->bio_offset
= bio_offset
;
844 if (op_is_sync(bio
->bi_opf
))
845 btrfs_set_work_high_priority(&async
->work
);
847 btrfs_queue_work(fs_info
->workers
, &async
->work
);
851 static blk_status_t
btree_csum_one_bio(struct bio
*bio
)
853 struct bio_vec
*bvec
;
854 struct btrfs_root
*root
;
856 struct bvec_iter_all iter_all
;
858 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
859 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
860 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
861 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
866 return errno_to_blk_status(ret
);
869 static blk_status_t
btree_submit_bio_start(void *private_data
, struct bio
*bio
,
873 * when we're called for a write, we're already in the async
874 * submission context. Just jump into btrfs_map_bio
876 return btree_csum_one_bio(bio
);
879 static int check_async_write(struct btrfs_fs_info
*fs_info
,
880 struct btrfs_inode
*bi
)
882 if (atomic_read(&bi
->sync_writers
))
884 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST
, &fs_info
->flags
))
889 static blk_status_t
btree_submit_bio_hook(struct inode
*inode
, struct bio
*bio
,
891 unsigned long bio_flags
)
893 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
894 int async
= check_async_write(fs_info
, BTRFS_I(inode
));
897 if (bio_op(bio
) != REQ_OP_WRITE
) {
899 * called for a read, do the setup so that checksum validation
900 * can happen in the async kernel threads
902 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
903 BTRFS_WQ_ENDIO_METADATA
);
906 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
908 ret
= btree_csum_one_bio(bio
);
911 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
914 * kthread helpers are used to submit writes so that
915 * checksumming can happen in parallel across all CPUs
917 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, 0,
918 0, inode
, btree_submit_bio_start
);
926 bio
->bi_status
= ret
;
931 #ifdef CONFIG_MIGRATION
932 static int btree_migratepage(struct address_space
*mapping
,
933 struct page
*newpage
, struct page
*page
,
934 enum migrate_mode mode
)
937 * we can't safely write a btree page from here,
938 * we haven't done the locking hook
943 * Buffers may be managed in a filesystem specific way.
944 * We must have no buffers or drop them.
946 if (page_has_private(page
) &&
947 !try_to_release_page(page
, GFP_KERNEL
))
949 return migrate_page(mapping
, newpage
, page
, mode
);
954 static int btree_writepages(struct address_space
*mapping
,
955 struct writeback_control
*wbc
)
957 struct btrfs_fs_info
*fs_info
;
960 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
962 if (wbc
->for_kupdate
)
965 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
966 /* this is a bit racy, but that's ok */
967 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
968 BTRFS_DIRTY_METADATA_THRESH
,
969 fs_info
->dirty_metadata_batch
);
973 return btree_write_cache_pages(mapping
, wbc
);
976 static int btree_readpage(struct file
*file
, struct page
*page
)
978 struct extent_io_tree
*tree
;
979 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
980 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
983 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
985 if (PageWriteback(page
) || PageDirty(page
))
988 return try_release_extent_buffer(page
);
991 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
994 struct extent_io_tree
*tree
;
995 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
996 extent_invalidatepage(tree
, page
, offset
);
997 btree_releasepage(page
, GFP_NOFS
);
998 if (PagePrivate(page
)) {
999 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1000 "page private not zero on page %llu",
1001 (unsigned long long)page_offset(page
));
1002 ClearPagePrivate(page
);
1003 set_page_private(page
, 0);
1008 static int btree_set_page_dirty(struct page
*page
)
1011 struct extent_buffer
*eb
;
1013 BUG_ON(!PagePrivate(page
));
1014 eb
= (struct extent_buffer
*)page
->private;
1016 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1017 BUG_ON(!atomic_read(&eb
->refs
));
1018 btrfs_assert_tree_locked(eb
);
1020 return __set_page_dirty_nobuffers(page
);
1023 static const struct address_space_operations btree_aops
= {
1024 .readpage
= btree_readpage
,
1025 .writepages
= btree_writepages
,
1026 .releasepage
= btree_releasepage
,
1027 .invalidatepage
= btree_invalidatepage
,
1028 #ifdef CONFIG_MIGRATION
1029 .migratepage
= btree_migratepage
,
1031 .set_page_dirty
= btree_set_page_dirty
,
1034 void readahead_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
1036 struct extent_buffer
*buf
= NULL
;
1039 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1043 ret
= read_extent_buffer_pages(buf
, WAIT_NONE
, 0);
1045 free_extent_buffer_stale(buf
);
1047 free_extent_buffer(buf
);
1050 int reada_tree_block_flagged(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1051 int mirror_num
, struct extent_buffer
**eb
)
1053 struct extent_buffer
*buf
= NULL
;
1056 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1060 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1062 ret
= read_extent_buffer_pages(buf
, WAIT_PAGE_LOCK
, mirror_num
);
1064 free_extent_buffer_stale(buf
);
1068 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1069 free_extent_buffer_stale(buf
);
1071 } else if (extent_buffer_uptodate(buf
)) {
1074 free_extent_buffer(buf
);
1079 struct extent_buffer
*btrfs_find_create_tree_block(
1080 struct btrfs_fs_info
*fs_info
,
1083 if (btrfs_is_testing(fs_info
))
1084 return alloc_test_extent_buffer(fs_info
, bytenr
);
1085 return alloc_extent_buffer(fs_info
, bytenr
);
1089 * Read tree block at logical address @bytenr and do variant basic but critical
1092 * @parent_transid: expected transid of this tree block, skip check if 0
1093 * @level: expected level, mandatory check
1094 * @first_key: expected key in slot 0, skip check if NULL
1096 struct extent_buffer
*read_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1097 u64 parent_transid
, int level
,
1098 struct btrfs_key
*first_key
)
1100 struct extent_buffer
*buf
= NULL
;
1103 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1107 ret
= btree_read_extent_buffer_pages(buf
, parent_transid
,
1110 free_extent_buffer_stale(buf
);
1111 return ERR_PTR(ret
);
1117 void btrfs_clean_tree_block(struct extent_buffer
*buf
)
1119 struct btrfs_fs_info
*fs_info
= buf
->fs_info
;
1120 if (btrfs_header_generation(buf
) ==
1121 fs_info
->running_transaction
->transid
) {
1122 btrfs_assert_tree_locked(buf
);
1124 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1125 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
1127 fs_info
->dirty_metadata_batch
);
1128 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1129 btrfs_set_lock_blocking_write(buf
);
1130 clear_extent_buffer_dirty(buf
);
1135 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1137 struct btrfs_subvolume_writers
*writers
;
1140 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1142 return ERR_PTR(-ENOMEM
);
1144 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_NOFS
);
1147 return ERR_PTR(ret
);
1150 init_waitqueue_head(&writers
->wait
);
1155 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1157 percpu_counter_destroy(&writers
->counter
);
1161 static void __setup_root(struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1164 bool dummy
= test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO
, &fs_info
->fs_state
);
1166 root
->commit_root
= NULL
;
1168 root
->orphan_cleanup_state
= 0;
1170 root
->last_trans
= 0;
1171 root
->highest_objectid
= 0;
1172 root
->nr_delalloc_inodes
= 0;
1173 root
->nr_ordered_extents
= 0;
1174 root
->inode_tree
= RB_ROOT
;
1175 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1176 root
->block_rsv
= NULL
;
1178 INIT_LIST_HEAD(&root
->dirty_list
);
1179 INIT_LIST_HEAD(&root
->root_list
);
1180 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1181 INIT_LIST_HEAD(&root
->delalloc_root
);
1182 INIT_LIST_HEAD(&root
->ordered_extents
);
1183 INIT_LIST_HEAD(&root
->ordered_root
);
1184 INIT_LIST_HEAD(&root
->reloc_dirty_list
);
1185 INIT_LIST_HEAD(&root
->logged_list
[0]);
1186 INIT_LIST_HEAD(&root
->logged_list
[1]);
1187 spin_lock_init(&root
->inode_lock
);
1188 spin_lock_init(&root
->delalloc_lock
);
1189 spin_lock_init(&root
->ordered_extent_lock
);
1190 spin_lock_init(&root
->accounting_lock
);
1191 spin_lock_init(&root
->log_extents_lock
[0]);
1192 spin_lock_init(&root
->log_extents_lock
[1]);
1193 spin_lock_init(&root
->qgroup_meta_rsv_lock
);
1194 mutex_init(&root
->objectid_mutex
);
1195 mutex_init(&root
->log_mutex
);
1196 mutex_init(&root
->ordered_extent_mutex
);
1197 mutex_init(&root
->delalloc_mutex
);
1198 init_waitqueue_head(&root
->log_writer_wait
);
1199 init_waitqueue_head(&root
->log_commit_wait
[0]);
1200 init_waitqueue_head(&root
->log_commit_wait
[1]);
1201 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1202 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1203 atomic_set(&root
->log_commit
[0], 0);
1204 atomic_set(&root
->log_commit
[1], 0);
1205 atomic_set(&root
->log_writers
, 0);
1206 atomic_set(&root
->log_batch
, 0);
1207 refcount_set(&root
->refs
, 1);
1208 atomic_set(&root
->will_be_snapshotted
, 0);
1209 atomic_set(&root
->snapshot_force_cow
, 0);
1210 atomic_set(&root
->nr_swapfiles
, 0);
1211 root
->log_transid
= 0;
1212 root
->log_transid_committed
= -1;
1213 root
->last_log_commit
= 0;
1215 extent_io_tree_init(fs_info
, &root
->dirty_log_pages
,
1216 IO_TREE_ROOT_DIRTY_LOG_PAGES
, NULL
);
1218 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1219 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1220 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1222 root
->defrag_trans_start
= fs_info
->generation
;
1224 root
->defrag_trans_start
= 0;
1225 root
->root_key
.objectid
= objectid
;
1228 spin_lock_init(&root
->root_item_lock
);
1229 btrfs_qgroup_init_swapped_blocks(&root
->swapped_blocks
);
1232 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1235 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1237 root
->fs_info
= fs_info
;
1241 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1242 /* Should only be used by the testing infrastructure */
1243 struct btrfs_root
*btrfs_alloc_dummy_root(struct btrfs_fs_info
*fs_info
)
1245 struct btrfs_root
*root
;
1248 return ERR_PTR(-EINVAL
);
1250 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1252 return ERR_PTR(-ENOMEM
);
1254 /* We don't use the stripesize in selftest, set it as sectorsize */
1255 __setup_root(root
, fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1256 root
->alloc_bytenr
= 0;
1262 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1265 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1266 struct extent_buffer
*leaf
;
1267 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1268 struct btrfs_root
*root
;
1269 struct btrfs_key key
;
1270 unsigned int nofs_flag
;
1272 uuid_le uuid
= NULL_UUID_LE
;
1275 * We're holding a transaction handle, so use a NOFS memory allocation
1276 * context to avoid deadlock if reclaim happens.
1278 nofs_flag
= memalloc_nofs_save();
1279 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1280 memalloc_nofs_restore(nofs_flag
);
1282 return ERR_PTR(-ENOMEM
);
1284 __setup_root(root
, fs_info
, objectid
);
1285 root
->root_key
.objectid
= objectid
;
1286 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1287 root
->root_key
.offset
= 0;
1289 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1291 ret
= PTR_ERR(leaf
);
1297 btrfs_mark_buffer_dirty(leaf
);
1299 root
->commit_root
= btrfs_root_node(root
);
1300 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1302 root
->root_item
.flags
= 0;
1303 root
->root_item
.byte_limit
= 0;
1304 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1305 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1306 btrfs_set_root_level(&root
->root_item
, 0);
1307 btrfs_set_root_refs(&root
->root_item
, 1);
1308 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1309 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1310 btrfs_set_root_dirid(&root
->root_item
, 0);
1311 if (is_fstree(objectid
))
1313 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1314 root
->root_item
.drop_level
= 0;
1316 key
.objectid
= objectid
;
1317 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1319 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1323 btrfs_tree_unlock(leaf
);
1329 btrfs_tree_unlock(leaf
);
1330 free_extent_buffer(root
->commit_root
);
1331 free_extent_buffer(leaf
);
1335 return ERR_PTR(ret
);
1338 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1339 struct btrfs_fs_info
*fs_info
)
1341 struct btrfs_root
*root
;
1342 struct extent_buffer
*leaf
;
1344 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1346 return ERR_PTR(-ENOMEM
);
1348 __setup_root(root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1350 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1351 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1352 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1355 * DON'T set REF_COWS for log trees
1357 * log trees do not get reference counted because they go away
1358 * before a real commit is actually done. They do store pointers
1359 * to file data extents, and those reference counts still get
1360 * updated (along with back refs to the log tree).
1363 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1367 return ERR_CAST(leaf
);
1372 btrfs_mark_buffer_dirty(root
->node
);
1373 btrfs_tree_unlock(root
->node
);
1377 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1378 struct btrfs_fs_info
*fs_info
)
1380 struct btrfs_root
*log_root
;
1382 log_root
= alloc_log_tree(trans
, fs_info
);
1383 if (IS_ERR(log_root
))
1384 return PTR_ERR(log_root
);
1385 WARN_ON(fs_info
->log_root_tree
);
1386 fs_info
->log_root_tree
= log_root
;
1390 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1391 struct btrfs_root
*root
)
1393 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1394 struct btrfs_root
*log_root
;
1395 struct btrfs_inode_item
*inode_item
;
1397 log_root
= alloc_log_tree(trans
, fs_info
);
1398 if (IS_ERR(log_root
))
1399 return PTR_ERR(log_root
);
1401 log_root
->last_trans
= trans
->transid
;
1402 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1404 inode_item
= &log_root
->root_item
.inode
;
1405 btrfs_set_stack_inode_generation(inode_item
, 1);
1406 btrfs_set_stack_inode_size(inode_item
, 3);
1407 btrfs_set_stack_inode_nlink(inode_item
, 1);
1408 btrfs_set_stack_inode_nbytes(inode_item
,
1410 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1412 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1414 WARN_ON(root
->log_root
);
1415 root
->log_root
= log_root
;
1416 root
->log_transid
= 0;
1417 root
->log_transid_committed
= -1;
1418 root
->last_log_commit
= 0;
1422 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1423 struct btrfs_key
*key
)
1425 struct btrfs_root
*root
;
1426 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1427 struct btrfs_path
*path
;
1432 path
= btrfs_alloc_path();
1434 return ERR_PTR(-ENOMEM
);
1436 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1442 __setup_root(root
, fs_info
, key
->objectid
);
1444 ret
= btrfs_find_root(tree_root
, key
, path
,
1445 &root
->root_item
, &root
->root_key
);
1452 generation
= btrfs_root_generation(&root
->root_item
);
1453 level
= btrfs_root_level(&root
->root_item
);
1454 root
->node
= read_tree_block(fs_info
,
1455 btrfs_root_bytenr(&root
->root_item
),
1456 generation
, level
, NULL
);
1457 if (IS_ERR(root
->node
)) {
1458 ret
= PTR_ERR(root
->node
);
1460 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1462 free_extent_buffer(root
->node
);
1465 root
->commit_root
= btrfs_root_node(root
);
1467 btrfs_free_path(path
);
1473 root
= ERR_PTR(ret
);
1477 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1478 struct btrfs_key
*location
)
1480 struct btrfs_root
*root
;
1482 root
= btrfs_read_tree_root(tree_root
, location
);
1486 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1487 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1488 btrfs_check_and_init_root_item(&root
->root_item
);
1494 int btrfs_init_fs_root(struct btrfs_root
*root
)
1497 struct btrfs_subvolume_writers
*writers
;
1499 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1500 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1502 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1507 writers
= btrfs_alloc_subvolume_writers();
1508 if (IS_ERR(writers
)) {
1509 ret
= PTR_ERR(writers
);
1512 root
->subv_writers
= writers
;
1514 btrfs_init_free_ino_ctl(root
);
1515 spin_lock_init(&root
->ino_cache_lock
);
1516 init_waitqueue_head(&root
->ino_cache_wait
);
1518 ret
= get_anon_bdev(&root
->anon_dev
);
1522 mutex_lock(&root
->objectid_mutex
);
1523 ret
= btrfs_find_highest_objectid(root
,
1524 &root
->highest_objectid
);
1526 mutex_unlock(&root
->objectid_mutex
);
1530 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1532 mutex_unlock(&root
->objectid_mutex
);
1536 /* The caller is responsible to call btrfs_free_fs_root */
1540 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1543 struct btrfs_root
*root
;
1545 spin_lock(&fs_info
->fs_roots_radix_lock
);
1546 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1547 (unsigned long)root_id
);
1548 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1552 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1553 struct btrfs_root
*root
)
1557 ret
= radix_tree_preload(GFP_NOFS
);
1561 spin_lock(&fs_info
->fs_roots_radix_lock
);
1562 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1563 (unsigned long)root
->root_key
.objectid
,
1566 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1567 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1568 radix_tree_preload_end();
1573 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1574 struct btrfs_key
*location
,
1577 struct btrfs_root
*root
;
1578 struct btrfs_path
*path
;
1579 struct btrfs_key key
;
1582 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1583 return fs_info
->tree_root
;
1584 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1585 return fs_info
->extent_root
;
1586 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1587 return fs_info
->chunk_root
;
1588 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1589 return fs_info
->dev_root
;
1590 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1591 return fs_info
->csum_root
;
1592 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1593 return fs_info
->quota_root
? fs_info
->quota_root
:
1595 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1596 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1598 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1599 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1602 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1604 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1605 return ERR_PTR(-ENOENT
);
1609 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1613 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1618 ret
= btrfs_init_fs_root(root
);
1622 path
= btrfs_alloc_path();
1627 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1628 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1629 key
.offset
= location
->objectid
;
1631 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1632 btrfs_free_path(path
);
1636 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1638 ret
= btrfs_insert_fs_root(fs_info
, root
);
1640 if (ret
== -EEXIST
) {
1641 btrfs_free_fs_root(root
);
1648 btrfs_free_fs_root(root
);
1649 return ERR_PTR(ret
);
1652 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1654 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1656 struct btrfs_device
*device
;
1657 struct backing_dev_info
*bdi
;
1660 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1663 bdi
= device
->bdev
->bd_bdi
;
1664 if (bdi_congested(bdi
, bdi_bits
)) {
1674 * called by the kthread helper functions to finally call the bio end_io
1675 * functions. This is where read checksum verification actually happens
1677 static void end_workqueue_fn(struct btrfs_work
*work
)
1680 struct btrfs_end_io_wq
*end_io_wq
;
1682 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1683 bio
= end_io_wq
->bio
;
1685 bio
->bi_status
= end_io_wq
->status
;
1686 bio
->bi_private
= end_io_wq
->private;
1687 bio
->bi_end_io
= end_io_wq
->end_io
;
1688 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1692 static int cleaner_kthread(void *arg
)
1694 struct btrfs_root
*root
= arg
;
1695 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1701 set_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1703 /* Make the cleaner go to sleep early. */
1704 if (btrfs_need_cleaner_sleep(fs_info
))
1708 * Do not do anything if we might cause open_ctree() to block
1709 * before we have finished mounting the filesystem.
1711 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1714 if (!mutex_trylock(&fs_info
->cleaner_mutex
))
1718 * Avoid the problem that we change the status of the fs
1719 * during the above check and trylock.
1721 if (btrfs_need_cleaner_sleep(fs_info
)) {
1722 mutex_unlock(&fs_info
->cleaner_mutex
);
1726 btrfs_run_delayed_iputs(fs_info
);
1728 again
= btrfs_clean_one_deleted_snapshot(root
);
1729 mutex_unlock(&fs_info
->cleaner_mutex
);
1732 * The defragger has dealt with the R/O remount and umount,
1733 * needn't do anything special here.
1735 btrfs_run_defrag_inodes(fs_info
);
1738 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1739 * with relocation (btrfs_relocate_chunk) and relocation
1740 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1741 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1742 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1743 * unused block groups.
1745 btrfs_delete_unused_bgs(fs_info
);
1747 clear_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1748 if (kthread_should_park())
1750 if (kthread_should_stop())
1753 set_current_state(TASK_INTERRUPTIBLE
);
1755 __set_current_state(TASK_RUNNING
);
1760 static int transaction_kthread(void *arg
)
1762 struct btrfs_root
*root
= arg
;
1763 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1764 struct btrfs_trans_handle
*trans
;
1765 struct btrfs_transaction
*cur
;
1768 unsigned long delay
;
1772 cannot_commit
= false;
1773 delay
= HZ
* fs_info
->commit_interval
;
1774 mutex_lock(&fs_info
->transaction_kthread_mutex
);
1776 spin_lock(&fs_info
->trans_lock
);
1777 cur
= fs_info
->running_transaction
;
1779 spin_unlock(&fs_info
->trans_lock
);
1783 now
= ktime_get_seconds();
1784 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1785 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT
, &fs_info
->flags
) &&
1786 (now
< cur
->start_time
||
1787 now
- cur
->start_time
< fs_info
->commit_interval
)) {
1788 spin_unlock(&fs_info
->trans_lock
);
1792 transid
= cur
->transid
;
1793 spin_unlock(&fs_info
->trans_lock
);
1795 /* If the file system is aborted, this will always fail. */
1796 trans
= btrfs_attach_transaction(root
);
1797 if (IS_ERR(trans
)) {
1798 if (PTR_ERR(trans
) != -ENOENT
)
1799 cannot_commit
= true;
1802 if (transid
== trans
->transid
) {
1803 btrfs_commit_transaction(trans
);
1805 btrfs_end_transaction(trans
);
1808 wake_up_process(fs_info
->cleaner_kthread
);
1809 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
1811 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1812 &fs_info
->fs_state
)))
1813 btrfs_cleanup_transaction(fs_info
);
1814 if (!kthread_should_stop() &&
1815 (!btrfs_transaction_blocked(fs_info
) ||
1817 schedule_timeout_interruptible(delay
);
1818 } while (!kthread_should_stop());
1823 * this will find the highest generation in the array of
1824 * root backups. The index of the highest array is returned,
1825 * or -1 if we can't find anything.
1827 * We check to make sure the array is valid by comparing the
1828 * generation of the latest root in the array with the generation
1829 * in the super block. If they don't match we pitch it.
1831 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1834 int newest_index
= -1;
1835 struct btrfs_root_backup
*root_backup
;
1838 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1839 root_backup
= info
->super_copy
->super_roots
+ i
;
1840 cur
= btrfs_backup_tree_root_gen(root_backup
);
1841 if (cur
== newest_gen
)
1845 /* check to see if we actually wrapped around */
1846 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1847 root_backup
= info
->super_copy
->super_roots
;
1848 cur
= btrfs_backup_tree_root_gen(root_backup
);
1849 if (cur
== newest_gen
)
1852 return newest_index
;
1857 * find the oldest backup so we know where to store new entries
1858 * in the backup array. This will set the backup_root_index
1859 * field in the fs_info struct
1861 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1864 int newest_index
= -1;
1866 newest_index
= find_newest_super_backup(info
, newest_gen
);
1867 /* if there was garbage in there, just move along */
1868 if (newest_index
== -1) {
1869 info
->backup_root_index
= 0;
1871 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1876 * copy all the root pointers into the super backup array.
1877 * this will bump the backup pointer by one when it is
1880 static void backup_super_roots(struct btrfs_fs_info
*info
)
1883 struct btrfs_root_backup
*root_backup
;
1886 next_backup
= info
->backup_root_index
;
1887 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1888 BTRFS_NUM_BACKUP_ROOTS
;
1891 * just overwrite the last backup if we're at the same generation
1892 * this happens only at umount
1894 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1895 if (btrfs_backup_tree_root_gen(root_backup
) ==
1896 btrfs_header_generation(info
->tree_root
->node
))
1897 next_backup
= last_backup
;
1899 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1902 * make sure all of our padding and empty slots get zero filled
1903 * regardless of which ones we use today
1905 memset(root_backup
, 0, sizeof(*root_backup
));
1907 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1909 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1910 btrfs_set_backup_tree_root_gen(root_backup
,
1911 btrfs_header_generation(info
->tree_root
->node
));
1913 btrfs_set_backup_tree_root_level(root_backup
,
1914 btrfs_header_level(info
->tree_root
->node
));
1916 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1917 btrfs_set_backup_chunk_root_gen(root_backup
,
1918 btrfs_header_generation(info
->chunk_root
->node
));
1919 btrfs_set_backup_chunk_root_level(root_backup
,
1920 btrfs_header_level(info
->chunk_root
->node
));
1922 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1923 btrfs_set_backup_extent_root_gen(root_backup
,
1924 btrfs_header_generation(info
->extent_root
->node
));
1925 btrfs_set_backup_extent_root_level(root_backup
,
1926 btrfs_header_level(info
->extent_root
->node
));
1929 * we might commit during log recovery, which happens before we set
1930 * the fs_root. Make sure it is valid before we fill it in.
1932 if (info
->fs_root
&& info
->fs_root
->node
) {
1933 btrfs_set_backup_fs_root(root_backup
,
1934 info
->fs_root
->node
->start
);
1935 btrfs_set_backup_fs_root_gen(root_backup
,
1936 btrfs_header_generation(info
->fs_root
->node
));
1937 btrfs_set_backup_fs_root_level(root_backup
,
1938 btrfs_header_level(info
->fs_root
->node
));
1941 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1942 btrfs_set_backup_dev_root_gen(root_backup
,
1943 btrfs_header_generation(info
->dev_root
->node
));
1944 btrfs_set_backup_dev_root_level(root_backup
,
1945 btrfs_header_level(info
->dev_root
->node
));
1947 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1948 btrfs_set_backup_csum_root_gen(root_backup
,
1949 btrfs_header_generation(info
->csum_root
->node
));
1950 btrfs_set_backup_csum_root_level(root_backup
,
1951 btrfs_header_level(info
->csum_root
->node
));
1953 btrfs_set_backup_total_bytes(root_backup
,
1954 btrfs_super_total_bytes(info
->super_copy
));
1955 btrfs_set_backup_bytes_used(root_backup
,
1956 btrfs_super_bytes_used(info
->super_copy
));
1957 btrfs_set_backup_num_devices(root_backup
,
1958 btrfs_super_num_devices(info
->super_copy
));
1961 * if we don't copy this out to the super_copy, it won't get remembered
1962 * for the next commit
1964 memcpy(&info
->super_copy
->super_roots
,
1965 &info
->super_for_commit
->super_roots
,
1966 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1970 * this copies info out of the root backup array and back into
1971 * the in-memory super block. It is meant to help iterate through
1972 * the array, so you send it the number of backups you've already
1973 * tried and the last backup index you used.
1975 * this returns -1 when it has tried all the backups
1977 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1978 struct btrfs_super_block
*super
,
1979 int *num_backups_tried
, int *backup_index
)
1981 struct btrfs_root_backup
*root_backup
;
1982 int newest
= *backup_index
;
1984 if (*num_backups_tried
== 0) {
1985 u64 gen
= btrfs_super_generation(super
);
1987 newest
= find_newest_super_backup(info
, gen
);
1991 *backup_index
= newest
;
1992 *num_backups_tried
= 1;
1993 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1994 /* we've tried all the backups, all done */
1997 /* jump to the next oldest backup */
1998 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1999 BTRFS_NUM_BACKUP_ROOTS
;
2000 *backup_index
= newest
;
2001 *num_backups_tried
+= 1;
2003 root_backup
= super
->super_roots
+ newest
;
2005 btrfs_set_super_generation(super
,
2006 btrfs_backup_tree_root_gen(root_backup
));
2007 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2008 btrfs_set_super_root_level(super
,
2009 btrfs_backup_tree_root_level(root_backup
));
2010 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2013 * fixme: the total bytes and num_devices need to match or we should
2016 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2017 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2021 /* helper to cleanup workers */
2022 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2024 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2025 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2026 btrfs_destroy_workqueue(fs_info
->workers
);
2027 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2028 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2029 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2030 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2031 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2032 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2033 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2034 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2035 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2036 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2037 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2038 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2040 * Now that all other work queues are destroyed, we can safely destroy
2041 * the queues used for metadata I/O, since tasks from those other work
2042 * queues can do metadata I/O operations.
2044 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2045 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2048 static void free_root_extent_buffers(struct btrfs_root
*root
)
2051 free_extent_buffer(root
->node
);
2052 free_extent_buffer(root
->commit_root
);
2054 root
->commit_root
= NULL
;
2058 /* helper to cleanup tree roots */
2059 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2061 free_root_extent_buffers(info
->tree_root
);
2063 free_root_extent_buffers(info
->dev_root
);
2064 free_root_extent_buffers(info
->extent_root
);
2065 free_root_extent_buffers(info
->csum_root
);
2066 free_root_extent_buffers(info
->quota_root
);
2067 free_root_extent_buffers(info
->uuid_root
);
2069 free_root_extent_buffers(info
->chunk_root
);
2070 free_root_extent_buffers(info
->free_space_root
);
2073 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2076 struct btrfs_root
*gang
[8];
2079 while (!list_empty(&fs_info
->dead_roots
)) {
2080 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2081 struct btrfs_root
, root_list
);
2082 list_del(&gang
[0]->root_list
);
2084 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2085 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2087 free_extent_buffer(gang
[0]->node
);
2088 free_extent_buffer(gang
[0]->commit_root
);
2089 btrfs_put_fs_root(gang
[0]);
2094 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2099 for (i
= 0; i
< ret
; i
++)
2100 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2103 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2104 btrfs_free_log_root_tree(NULL
, fs_info
);
2105 btrfs_destroy_pinned_extent(fs_info
, fs_info
->pinned_extents
);
2109 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2111 mutex_init(&fs_info
->scrub_lock
);
2112 atomic_set(&fs_info
->scrubs_running
, 0);
2113 atomic_set(&fs_info
->scrub_pause_req
, 0);
2114 atomic_set(&fs_info
->scrubs_paused
, 0);
2115 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2116 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2117 refcount_set(&fs_info
->scrub_workers_refcnt
, 0);
2120 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2122 spin_lock_init(&fs_info
->balance_lock
);
2123 mutex_init(&fs_info
->balance_mutex
);
2124 atomic_set(&fs_info
->balance_pause_req
, 0);
2125 atomic_set(&fs_info
->balance_cancel_req
, 0);
2126 fs_info
->balance_ctl
= NULL
;
2127 init_waitqueue_head(&fs_info
->balance_wait_q
);
2130 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
)
2132 struct inode
*inode
= fs_info
->btree_inode
;
2134 inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2135 set_nlink(inode
, 1);
2137 * we set the i_size on the btree inode to the max possible int.
2138 * the real end of the address space is determined by all of
2139 * the devices in the system
2141 inode
->i_size
= OFFSET_MAX
;
2142 inode
->i_mapping
->a_ops
= &btree_aops
;
2144 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
2145 extent_io_tree_init(fs_info
, &BTRFS_I(inode
)->io_tree
,
2146 IO_TREE_INODE_IO
, inode
);
2147 BTRFS_I(inode
)->io_tree
.track_uptodate
= false;
2148 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
);
2150 BTRFS_I(inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2152 BTRFS_I(inode
)->root
= fs_info
->tree_root
;
2153 memset(&BTRFS_I(inode
)->location
, 0, sizeof(struct btrfs_key
));
2154 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
2155 btrfs_insert_inode_hash(inode
);
2158 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2160 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2161 init_rwsem(&fs_info
->dev_replace
.rwsem
);
2162 init_waitqueue_head(&fs_info
->dev_replace
.replace_wait
);
2165 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2167 spin_lock_init(&fs_info
->qgroup_lock
);
2168 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2169 fs_info
->qgroup_tree
= RB_ROOT
;
2170 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2171 fs_info
->qgroup_seq
= 1;
2172 fs_info
->qgroup_ulist
= NULL
;
2173 fs_info
->qgroup_rescan_running
= false;
2174 mutex_init(&fs_info
->qgroup_rescan_lock
);
2177 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2178 struct btrfs_fs_devices
*fs_devices
)
2180 u32 max_active
= fs_info
->thread_pool_size
;
2181 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2184 btrfs_alloc_workqueue(fs_info
, "worker",
2185 flags
| WQ_HIGHPRI
, max_active
, 16);
2187 fs_info
->delalloc_workers
=
2188 btrfs_alloc_workqueue(fs_info
, "delalloc",
2189 flags
, max_active
, 2);
2191 fs_info
->flush_workers
=
2192 btrfs_alloc_workqueue(fs_info
, "flush_delalloc",
2193 flags
, max_active
, 0);
2195 fs_info
->caching_workers
=
2196 btrfs_alloc_workqueue(fs_info
, "cache", flags
, max_active
, 0);
2199 * a higher idle thresh on the submit workers makes it much more
2200 * likely that bios will be send down in a sane order to the
2203 fs_info
->submit_workers
=
2204 btrfs_alloc_workqueue(fs_info
, "submit", flags
,
2205 min_t(u64
, fs_devices
->num_devices
,
2208 fs_info
->fixup_workers
=
2209 btrfs_alloc_workqueue(fs_info
, "fixup", flags
, 1, 0);
2212 * endios are largely parallel and should have a very
2215 fs_info
->endio_workers
=
2216 btrfs_alloc_workqueue(fs_info
, "endio", flags
, max_active
, 4);
2217 fs_info
->endio_meta_workers
=
2218 btrfs_alloc_workqueue(fs_info
, "endio-meta", flags
,
2220 fs_info
->endio_meta_write_workers
=
2221 btrfs_alloc_workqueue(fs_info
, "endio-meta-write", flags
,
2223 fs_info
->endio_raid56_workers
=
2224 btrfs_alloc_workqueue(fs_info
, "endio-raid56", flags
,
2226 fs_info
->endio_repair_workers
=
2227 btrfs_alloc_workqueue(fs_info
, "endio-repair", flags
, 1, 0);
2228 fs_info
->rmw_workers
=
2229 btrfs_alloc_workqueue(fs_info
, "rmw", flags
, max_active
, 2);
2230 fs_info
->endio_write_workers
=
2231 btrfs_alloc_workqueue(fs_info
, "endio-write", flags
,
2233 fs_info
->endio_freespace_worker
=
2234 btrfs_alloc_workqueue(fs_info
, "freespace-write", flags
,
2236 fs_info
->delayed_workers
=
2237 btrfs_alloc_workqueue(fs_info
, "delayed-meta", flags
,
2239 fs_info
->readahead_workers
=
2240 btrfs_alloc_workqueue(fs_info
, "readahead", flags
,
2242 fs_info
->qgroup_rescan_workers
=
2243 btrfs_alloc_workqueue(fs_info
, "qgroup-rescan", flags
, 1, 0);
2245 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2246 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2247 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2248 fs_info
->endio_meta_write_workers
&&
2249 fs_info
->endio_repair_workers
&&
2250 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2251 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2252 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2253 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2254 fs_info
->qgroup_rescan_workers
)) {
2261 static int btrfs_init_csum_hash(struct btrfs_fs_info
*fs_info
, u16 csum_type
)
2263 struct crypto_shash
*csum_shash
;
2264 const char *csum_name
= btrfs_super_csum_name(csum_type
);
2266 csum_shash
= crypto_alloc_shash(csum_name
, 0, 0);
2268 if (IS_ERR(csum_shash
)) {
2269 btrfs_err(fs_info
, "error allocating %s hash for checksum",
2271 return PTR_ERR(csum_shash
);
2274 fs_info
->csum_shash
= csum_shash
;
2279 static void btrfs_free_csum_hash(struct btrfs_fs_info
*fs_info
)
2281 crypto_free_shash(fs_info
->csum_shash
);
2284 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2285 struct btrfs_fs_devices
*fs_devices
)
2288 struct btrfs_root
*log_tree_root
;
2289 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2290 u64 bytenr
= btrfs_super_log_root(disk_super
);
2291 int level
= btrfs_super_log_root_level(disk_super
);
2293 if (fs_devices
->rw_devices
== 0) {
2294 btrfs_warn(fs_info
, "log replay required on RO media");
2298 log_tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2302 __setup_root(log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2304 log_tree_root
->node
= read_tree_block(fs_info
, bytenr
,
2305 fs_info
->generation
+ 1,
2307 if (IS_ERR(log_tree_root
->node
)) {
2308 btrfs_warn(fs_info
, "failed to read log tree");
2309 ret
= PTR_ERR(log_tree_root
->node
);
2310 kfree(log_tree_root
);
2312 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2313 btrfs_err(fs_info
, "failed to read log tree");
2314 free_extent_buffer(log_tree_root
->node
);
2315 kfree(log_tree_root
);
2318 /* returns with log_tree_root freed on success */
2319 ret
= btrfs_recover_log_trees(log_tree_root
);
2321 btrfs_handle_fs_error(fs_info
, ret
,
2322 "Failed to recover log tree");
2323 free_extent_buffer(log_tree_root
->node
);
2324 kfree(log_tree_root
);
2328 if (sb_rdonly(fs_info
->sb
)) {
2329 ret
= btrfs_commit_super(fs_info
);
2337 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
)
2339 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2340 struct btrfs_root
*root
;
2341 struct btrfs_key location
;
2344 BUG_ON(!fs_info
->tree_root
);
2346 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2347 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2348 location
.offset
= 0;
2350 root
= btrfs_read_tree_root(tree_root
, &location
);
2352 ret
= PTR_ERR(root
);
2355 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2356 fs_info
->extent_root
= root
;
2358 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2359 root
= btrfs_read_tree_root(tree_root
, &location
);
2361 ret
= PTR_ERR(root
);
2364 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2365 fs_info
->dev_root
= root
;
2366 btrfs_init_devices_late(fs_info
);
2368 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2369 root
= btrfs_read_tree_root(tree_root
, &location
);
2371 ret
= PTR_ERR(root
);
2374 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2375 fs_info
->csum_root
= root
;
2377 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2378 root
= btrfs_read_tree_root(tree_root
, &location
);
2379 if (!IS_ERR(root
)) {
2380 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2381 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2382 fs_info
->quota_root
= root
;
2385 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2386 root
= btrfs_read_tree_root(tree_root
, &location
);
2388 ret
= PTR_ERR(root
);
2392 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2393 fs_info
->uuid_root
= root
;
2396 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2397 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2398 root
= btrfs_read_tree_root(tree_root
, &location
);
2400 ret
= PTR_ERR(root
);
2403 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2404 fs_info
->free_space_root
= root
;
2409 btrfs_warn(fs_info
, "failed to read root (objectid=%llu): %d",
2410 location
.objectid
, ret
);
2415 * Real super block validation
2416 * NOTE: super csum type and incompat features will not be checked here.
2418 * @sb: super block to check
2419 * @mirror_num: the super block number to check its bytenr:
2420 * 0 the primary (1st) sb
2421 * 1, 2 2nd and 3rd backup copy
2422 * -1 skip bytenr check
2424 static int validate_super(struct btrfs_fs_info
*fs_info
,
2425 struct btrfs_super_block
*sb
, int mirror_num
)
2427 u64 nodesize
= btrfs_super_nodesize(sb
);
2428 u64 sectorsize
= btrfs_super_sectorsize(sb
);
2431 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
2432 btrfs_err(fs_info
, "no valid FS found");
2435 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
) {
2436 btrfs_err(fs_info
, "unrecognized or unsupported super flag: %llu",
2437 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
2440 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2441 btrfs_err(fs_info
, "tree_root level too big: %d >= %d",
2442 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
2445 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2446 btrfs_err(fs_info
, "chunk_root level too big: %d >= %d",
2447 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
2450 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2451 btrfs_err(fs_info
, "log_root level too big: %d >= %d",
2452 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
2457 * Check sectorsize and nodesize first, other check will need it.
2458 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2460 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
2461 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2462 btrfs_err(fs_info
, "invalid sectorsize %llu", sectorsize
);
2465 /* Only PAGE SIZE is supported yet */
2466 if (sectorsize
!= PAGE_SIZE
) {
2468 "sectorsize %llu not supported yet, only support %lu",
2469 sectorsize
, PAGE_SIZE
);
2472 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
2473 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2474 btrfs_err(fs_info
, "invalid nodesize %llu", nodesize
);
2477 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
2478 btrfs_err(fs_info
, "invalid leafsize %u, should be %llu",
2479 le32_to_cpu(sb
->__unused_leafsize
), nodesize
);
2483 /* Root alignment check */
2484 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
2485 btrfs_warn(fs_info
, "tree_root block unaligned: %llu",
2486 btrfs_super_root(sb
));
2489 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
2490 btrfs_warn(fs_info
, "chunk_root block unaligned: %llu",
2491 btrfs_super_chunk_root(sb
));
2494 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
2495 btrfs_warn(fs_info
, "log_root block unaligned: %llu",
2496 btrfs_super_log_root(sb
));
2500 if (memcmp(fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
,
2501 BTRFS_FSID_SIZE
) != 0) {
2503 "dev_item UUID does not match metadata fsid: %pU != %pU",
2504 fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
);
2509 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2512 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
2513 btrfs_err(fs_info
, "bytes_used is too small %llu",
2514 btrfs_super_bytes_used(sb
));
2517 if (!is_power_of_2(btrfs_super_stripesize(sb
))) {
2518 btrfs_err(fs_info
, "invalid stripesize %u",
2519 btrfs_super_stripesize(sb
));
2522 if (btrfs_super_num_devices(sb
) > (1UL << 31))
2523 btrfs_warn(fs_info
, "suspicious number of devices: %llu",
2524 btrfs_super_num_devices(sb
));
2525 if (btrfs_super_num_devices(sb
) == 0) {
2526 btrfs_err(fs_info
, "number of devices is 0");
2530 if (mirror_num
>= 0 &&
2531 btrfs_super_bytenr(sb
) != btrfs_sb_offset(mirror_num
)) {
2532 btrfs_err(fs_info
, "super offset mismatch %llu != %u",
2533 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
2538 * Obvious sys_chunk_array corruptions, it must hold at least one key
2541 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
2542 btrfs_err(fs_info
, "system chunk array too big %u > %u",
2543 btrfs_super_sys_array_size(sb
),
2544 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
2547 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
2548 + sizeof(struct btrfs_chunk
)) {
2549 btrfs_err(fs_info
, "system chunk array too small %u < %zu",
2550 btrfs_super_sys_array_size(sb
),
2551 sizeof(struct btrfs_disk_key
)
2552 + sizeof(struct btrfs_chunk
));
2557 * The generation is a global counter, we'll trust it more than the others
2558 * but it's still possible that it's the one that's wrong.
2560 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
2562 "suspicious: generation < chunk_root_generation: %llu < %llu",
2563 btrfs_super_generation(sb
),
2564 btrfs_super_chunk_root_generation(sb
));
2565 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
2566 && btrfs_super_cache_generation(sb
) != (u64
)-1)
2568 "suspicious: generation < cache_generation: %llu < %llu",
2569 btrfs_super_generation(sb
),
2570 btrfs_super_cache_generation(sb
));
2576 * Validation of super block at mount time.
2577 * Some checks already done early at mount time, like csum type and incompat
2578 * flags will be skipped.
2580 static int btrfs_validate_mount_super(struct btrfs_fs_info
*fs_info
)
2582 return validate_super(fs_info
, fs_info
->super_copy
, 0);
2586 * Validation of super block at write time.
2587 * Some checks like bytenr check will be skipped as their values will be
2589 * Extra checks like csum type and incompat flags will be done here.
2591 static int btrfs_validate_write_super(struct btrfs_fs_info
*fs_info
,
2592 struct btrfs_super_block
*sb
)
2596 ret
= validate_super(fs_info
, sb
, -1);
2599 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb
))) {
2601 btrfs_err(fs_info
, "invalid csum type, has %u want %u",
2602 btrfs_super_csum_type(sb
), BTRFS_CSUM_TYPE_CRC32
);
2605 if (btrfs_super_incompat_flags(sb
) & ~BTRFS_FEATURE_INCOMPAT_SUPP
) {
2608 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2609 btrfs_super_incompat_flags(sb
),
2610 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP
);
2616 "super block corruption detected before writing it to disk");
2620 int open_ctree(struct super_block
*sb
,
2621 struct btrfs_fs_devices
*fs_devices
,
2630 struct btrfs_key location
;
2631 struct buffer_head
*bh
;
2632 struct btrfs_super_block
*disk_super
;
2633 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2634 struct btrfs_root
*tree_root
;
2635 struct btrfs_root
*chunk_root
;
2638 int num_backups_tried
= 0;
2639 int backup_index
= 0;
2640 int clear_free_space_tree
= 0;
2643 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2644 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2645 if (!tree_root
|| !chunk_root
) {
2650 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2656 ret
= percpu_counter_init(&fs_info
->dio_bytes
, 0, GFP_KERNEL
);
2662 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2665 goto fail_dio_bytes
;
2667 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2668 (1 + ilog2(nr_cpu_ids
));
2670 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2673 goto fail_dirty_metadata_bytes
;
2676 ret
= percpu_counter_init(&fs_info
->dev_replace
.bio_counter
, 0,
2680 goto fail_delalloc_bytes
;
2683 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2684 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2685 INIT_LIST_HEAD(&fs_info
->trans_list
);
2686 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2687 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2688 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2689 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2690 spin_lock_init(&fs_info
->delalloc_root_lock
);
2691 spin_lock_init(&fs_info
->trans_lock
);
2692 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2693 spin_lock_init(&fs_info
->delayed_iput_lock
);
2694 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2695 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2696 spin_lock_init(&fs_info
->super_lock
);
2697 spin_lock_init(&fs_info
->buffer_lock
);
2698 spin_lock_init(&fs_info
->unused_bgs_lock
);
2699 rwlock_init(&fs_info
->tree_mod_log_lock
);
2700 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2701 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2702 mutex_init(&fs_info
->reloc_mutex
);
2703 mutex_init(&fs_info
->delalloc_root_mutex
);
2704 seqlock_init(&fs_info
->profiles_lock
);
2706 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2707 INIT_LIST_HEAD(&fs_info
->space_info
);
2708 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2709 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2710 extent_map_tree_init(&fs_info
->mapping_tree
);
2711 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2712 BTRFS_BLOCK_RSV_GLOBAL
);
2713 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2714 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2715 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2716 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2717 BTRFS_BLOCK_RSV_DELOPS
);
2718 btrfs_init_block_rsv(&fs_info
->delayed_refs_rsv
,
2719 BTRFS_BLOCK_RSV_DELREFS
);
2721 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2722 atomic_set(&fs_info
->defrag_running
, 0);
2723 atomic_set(&fs_info
->reada_works_cnt
, 0);
2724 atomic_set(&fs_info
->nr_delayed_iputs
, 0);
2725 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2727 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2728 fs_info
->metadata_ratio
= 0;
2729 fs_info
->defrag_inodes
= RB_ROOT
;
2730 atomic64_set(&fs_info
->free_chunk_space
, 0);
2731 fs_info
->tree_mod_log
= RB_ROOT
;
2732 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2733 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2734 /* readahead state */
2735 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2736 spin_lock_init(&fs_info
->reada_lock
);
2737 btrfs_init_ref_verify(fs_info
);
2739 fs_info
->thread_pool_size
= min_t(unsigned long,
2740 num_online_cpus() + 2, 8);
2742 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2743 spin_lock_init(&fs_info
->ordered_root_lock
);
2745 fs_info
->btree_inode
= new_inode(sb
);
2746 if (!fs_info
->btree_inode
) {
2748 goto fail_bio_counter
;
2750 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2752 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2754 if (!fs_info
->delayed_root
) {
2758 btrfs_init_delayed_root(fs_info
->delayed_root
);
2760 btrfs_init_scrub(fs_info
);
2761 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2762 fs_info
->check_integrity_print_mask
= 0;
2764 btrfs_init_balance(fs_info
);
2765 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2767 sb
->s_blocksize
= BTRFS_BDEV_BLOCKSIZE
;
2768 sb
->s_blocksize_bits
= blksize_bits(BTRFS_BDEV_BLOCKSIZE
);
2770 btrfs_init_btree_inode(fs_info
);
2772 spin_lock_init(&fs_info
->block_group_cache_lock
);
2773 fs_info
->block_group_cache_tree
= RB_ROOT
;
2774 fs_info
->first_logical_byte
= (u64
)-1;
2776 extent_io_tree_init(fs_info
, &fs_info
->freed_extents
[0],
2777 IO_TREE_FS_INFO_FREED_EXTENTS0
, NULL
);
2778 extent_io_tree_init(fs_info
, &fs_info
->freed_extents
[1],
2779 IO_TREE_FS_INFO_FREED_EXTENTS1
, NULL
);
2780 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2781 set_bit(BTRFS_FS_BARRIER
, &fs_info
->flags
);
2783 mutex_init(&fs_info
->ordered_operations_mutex
);
2784 mutex_init(&fs_info
->tree_log_mutex
);
2785 mutex_init(&fs_info
->chunk_mutex
);
2786 mutex_init(&fs_info
->transaction_kthread_mutex
);
2787 mutex_init(&fs_info
->cleaner_mutex
);
2788 mutex_init(&fs_info
->ro_block_group_mutex
);
2789 init_rwsem(&fs_info
->commit_root_sem
);
2790 init_rwsem(&fs_info
->cleanup_work_sem
);
2791 init_rwsem(&fs_info
->subvol_sem
);
2792 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2794 btrfs_init_dev_replace_locks(fs_info
);
2795 btrfs_init_qgroup(fs_info
);
2797 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2798 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2800 init_waitqueue_head(&fs_info
->transaction_throttle
);
2801 init_waitqueue_head(&fs_info
->transaction_wait
);
2802 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2803 init_waitqueue_head(&fs_info
->async_submit_wait
);
2804 init_waitqueue_head(&fs_info
->delayed_iputs_wait
);
2806 /* Usable values until the real ones are cached from the superblock */
2807 fs_info
->nodesize
= 4096;
2808 fs_info
->sectorsize
= 4096;
2809 fs_info
->stripesize
= 4096;
2811 spin_lock_init(&fs_info
->swapfile_pins_lock
);
2812 fs_info
->swapfile_pins
= RB_ROOT
;
2814 fs_info
->send_in_progress
= 0;
2816 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2822 __setup_root(tree_root
, fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2824 invalidate_bdev(fs_devices
->latest_bdev
);
2827 * Read super block and check the signature bytes only
2829 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2836 * Verify the type first, if that or the the checksum value are
2837 * corrupted, we'll find out
2839 csum_type
= btrfs_super_csum_type((struct btrfs_super_block
*)bh
->b_data
);
2840 if (!btrfs_supported_super_csum(csum_type
)) {
2841 btrfs_err(fs_info
, "unsupported checksum algorithm: %u",
2848 ret
= btrfs_init_csum_hash(fs_info
, csum_type
);
2855 * We want to check superblock checksum, the type is stored inside.
2856 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2858 if (btrfs_check_super_csum(fs_info
, bh
->b_data
)) {
2859 btrfs_err(fs_info
, "superblock checksum mismatch");
2866 * super_copy is zeroed at allocation time and we never touch the
2867 * following bytes up to INFO_SIZE, the checksum is calculated from
2868 * the whole block of INFO_SIZE
2870 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2873 disk_super
= fs_info
->super_copy
;
2875 ASSERT(!memcmp(fs_info
->fs_devices
->fsid
, fs_info
->super_copy
->fsid
,
2878 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
)) {
2879 ASSERT(!memcmp(fs_info
->fs_devices
->metadata_uuid
,
2880 fs_info
->super_copy
->metadata_uuid
,
2884 features
= btrfs_super_flags(disk_super
);
2885 if (features
& BTRFS_SUPER_FLAG_CHANGING_FSID_V2
) {
2886 features
&= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2
;
2887 btrfs_set_super_flags(disk_super
, features
);
2889 "found metadata UUID change in progress flag, clearing");
2892 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2893 sizeof(*fs_info
->super_for_commit
));
2895 ret
= btrfs_validate_mount_super(fs_info
);
2897 btrfs_err(fs_info
, "superblock contains fatal errors");
2902 if (!btrfs_super_root(disk_super
))
2905 /* check FS state, whether FS is broken. */
2906 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2907 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2910 * run through our array of backup supers and setup
2911 * our ring pointer to the oldest one
2913 generation
= btrfs_super_generation(disk_super
);
2914 find_oldest_super_backup(fs_info
, generation
);
2917 * In the long term, we'll store the compression type in the super
2918 * block, and it'll be used for per file compression control.
2920 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2922 ret
= btrfs_parse_options(fs_info
, options
, sb
->s_flags
);
2928 features
= btrfs_super_incompat_flags(disk_super
) &
2929 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2932 "cannot mount because of unsupported optional features (%llx)",
2938 features
= btrfs_super_incompat_flags(disk_super
);
2939 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2940 if (fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2941 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2942 else if (fs_info
->compress_type
== BTRFS_COMPRESS_ZSTD
)
2943 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD
;
2945 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2946 btrfs_info(fs_info
, "has skinny extents");
2949 * flag our filesystem as having big metadata blocks if
2950 * they are bigger than the page size
2952 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
2953 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2955 "flagging fs with big metadata feature");
2956 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2959 nodesize
= btrfs_super_nodesize(disk_super
);
2960 sectorsize
= btrfs_super_sectorsize(disk_super
);
2961 stripesize
= sectorsize
;
2962 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2963 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2965 /* Cache block sizes */
2966 fs_info
->nodesize
= nodesize
;
2967 fs_info
->sectorsize
= sectorsize
;
2968 fs_info
->stripesize
= stripesize
;
2971 * mixed block groups end up with duplicate but slightly offset
2972 * extent buffers for the same range. It leads to corruptions
2974 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2975 (sectorsize
!= nodesize
)) {
2977 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2978 nodesize
, sectorsize
);
2983 * Needn't use the lock because there is no other task which will
2986 btrfs_set_super_incompat_flags(disk_super
, features
);
2988 features
= btrfs_super_compat_ro_flags(disk_super
) &
2989 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2990 if (!sb_rdonly(sb
) && features
) {
2992 "cannot mount read-write because of unsupported optional features (%llx)",
2998 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
3001 goto fail_sb_buffer
;
3004 sb
->s_bdi
->congested_fn
= btrfs_congested_fn
;
3005 sb
->s_bdi
->congested_data
= fs_info
;
3006 sb
->s_bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
3007 sb
->s_bdi
->ra_pages
= VM_READAHEAD_PAGES
;
3008 sb
->s_bdi
->ra_pages
*= btrfs_super_num_devices(disk_super
);
3009 sb
->s_bdi
->ra_pages
= max(sb
->s_bdi
->ra_pages
, SZ_4M
/ PAGE_SIZE
);
3011 sb
->s_blocksize
= sectorsize
;
3012 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
3013 memcpy(&sb
->s_uuid
, fs_info
->fs_devices
->fsid
, BTRFS_FSID_SIZE
);
3015 mutex_lock(&fs_info
->chunk_mutex
);
3016 ret
= btrfs_read_sys_array(fs_info
);
3017 mutex_unlock(&fs_info
->chunk_mutex
);
3019 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
3020 goto fail_sb_buffer
;
3023 generation
= btrfs_super_chunk_root_generation(disk_super
);
3024 level
= btrfs_super_chunk_root_level(disk_super
);
3026 __setup_root(chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
3028 chunk_root
->node
= read_tree_block(fs_info
,
3029 btrfs_super_chunk_root(disk_super
),
3030 generation
, level
, NULL
);
3031 if (IS_ERR(chunk_root
->node
) ||
3032 !extent_buffer_uptodate(chunk_root
->node
)) {
3033 btrfs_err(fs_info
, "failed to read chunk root");
3034 if (!IS_ERR(chunk_root
->node
))
3035 free_extent_buffer(chunk_root
->node
);
3036 chunk_root
->node
= NULL
;
3037 goto fail_tree_roots
;
3039 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
3040 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
3042 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
3043 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
3045 ret
= btrfs_read_chunk_tree(fs_info
);
3047 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
3048 goto fail_tree_roots
;
3052 * Keep the devid that is marked to be the target device for the
3053 * device replace procedure
3055 btrfs_free_extra_devids(fs_devices
, 0);
3057 if (!fs_devices
->latest_bdev
) {
3058 btrfs_err(fs_info
, "failed to read devices");
3059 goto fail_tree_roots
;
3063 generation
= btrfs_super_generation(disk_super
);
3064 level
= btrfs_super_root_level(disk_super
);
3066 tree_root
->node
= read_tree_block(fs_info
,
3067 btrfs_super_root(disk_super
),
3068 generation
, level
, NULL
);
3069 if (IS_ERR(tree_root
->node
) ||
3070 !extent_buffer_uptodate(tree_root
->node
)) {
3071 btrfs_warn(fs_info
, "failed to read tree root");
3072 if (!IS_ERR(tree_root
->node
))
3073 free_extent_buffer(tree_root
->node
);
3074 tree_root
->node
= NULL
;
3075 goto recovery_tree_root
;
3078 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
3079 tree_root
->commit_root
= btrfs_root_node(tree_root
);
3080 btrfs_set_root_refs(&tree_root
->root_item
, 1);
3082 mutex_lock(&tree_root
->objectid_mutex
);
3083 ret
= btrfs_find_highest_objectid(tree_root
,
3084 &tree_root
->highest_objectid
);
3086 mutex_unlock(&tree_root
->objectid_mutex
);
3087 goto recovery_tree_root
;
3090 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
3092 mutex_unlock(&tree_root
->objectid_mutex
);
3094 ret
= btrfs_read_roots(fs_info
);
3096 goto recovery_tree_root
;
3098 fs_info
->generation
= generation
;
3099 fs_info
->last_trans_committed
= generation
;
3101 ret
= btrfs_verify_dev_extents(fs_info
);
3104 "failed to verify dev extents against chunks: %d",
3106 goto fail_block_groups
;
3108 ret
= btrfs_recover_balance(fs_info
);
3110 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
3111 goto fail_block_groups
;
3114 ret
= btrfs_init_dev_stats(fs_info
);
3116 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
3117 goto fail_block_groups
;
3120 ret
= btrfs_init_dev_replace(fs_info
);
3122 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
3123 goto fail_block_groups
;
3126 btrfs_free_extra_devids(fs_devices
, 1);
3128 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
3130 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
3132 goto fail_block_groups
;
3135 ret
= btrfs_sysfs_add_device(fs_devices
);
3137 btrfs_err(fs_info
, "failed to init sysfs device interface: %d",
3139 goto fail_fsdev_sysfs
;
3142 ret
= btrfs_sysfs_add_mounted(fs_info
);
3144 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
3145 goto fail_fsdev_sysfs
;
3148 ret
= btrfs_init_space_info(fs_info
);
3150 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
3154 ret
= btrfs_read_block_groups(fs_info
);
3156 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
3160 if (!sb_rdonly(sb
) && !btrfs_check_rw_degradable(fs_info
, NULL
)) {
3162 "writable mount is not allowed due to too many missing devices");
3166 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3168 if (IS_ERR(fs_info
->cleaner_kthread
))
3171 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3173 "btrfs-transaction");
3174 if (IS_ERR(fs_info
->transaction_kthread
))
3177 if (!btrfs_test_opt(fs_info
, NOSSD
) &&
3178 !fs_info
->fs_devices
->rotating
) {
3179 btrfs_set_and_info(fs_info
, SSD
, "enabling ssd optimizations");
3183 * Mount does not set all options immediately, we can do it now and do
3184 * not have to wait for transaction commit
3186 btrfs_apply_pending_changes(fs_info
);
3188 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3189 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
)) {
3190 ret
= btrfsic_mount(fs_info
, fs_devices
,
3191 btrfs_test_opt(fs_info
,
3192 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3194 fs_info
->check_integrity_print_mask
);
3197 "failed to initialize integrity check module: %d",
3201 ret
= btrfs_read_qgroup_config(fs_info
);
3203 goto fail_trans_kthread
;
3205 if (btrfs_build_ref_tree(fs_info
))
3206 btrfs_err(fs_info
, "couldn't build ref tree");
3208 /* do not make disk changes in broken FS or nologreplay is given */
3209 if (btrfs_super_log_root(disk_super
) != 0 &&
3210 !btrfs_test_opt(fs_info
, NOLOGREPLAY
)) {
3211 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3218 ret
= btrfs_find_orphan_roots(fs_info
);
3222 if (!sb_rdonly(sb
)) {
3223 ret
= btrfs_cleanup_fs_roots(fs_info
);
3227 mutex_lock(&fs_info
->cleaner_mutex
);
3228 ret
= btrfs_recover_relocation(tree_root
);
3229 mutex_unlock(&fs_info
->cleaner_mutex
);
3231 btrfs_warn(fs_info
, "failed to recover relocation: %d",
3238 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3239 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3240 location
.offset
= 0;
3242 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3243 if (IS_ERR(fs_info
->fs_root
)) {
3244 err
= PTR_ERR(fs_info
->fs_root
);
3245 btrfs_warn(fs_info
, "failed to read fs tree: %d", err
);
3252 if (btrfs_test_opt(fs_info
, CLEAR_CACHE
) &&
3253 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3254 clear_free_space_tree
= 1;
3255 } else if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
3256 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE_VALID
)) {
3257 btrfs_warn(fs_info
, "free space tree is invalid");
3258 clear_free_space_tree
= 1;
3261 if (clear_free_space_tree
) {
3262 btrfs_info(fs_info
, "clearing free space tree");
3263 ret
= btrfs_clear_free_space_tree(fs_info
);
3266 "failed to clear free space tree: %d", ret
);
3267 close_ctree(fs_info
);
3272 if (btrfs_test_opt(fs_info
, FREE_SPACE_TREE
) &&
3273 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3274 btrfs_info(fs_info
, "creating free space tree");
3275 ret
= btrfs_create_free_space_tree(fs_info
);
3278 "failed to create free space tree: %d", ret
);
3279 close_ctree(fs_info
);
3284 down_read(&fs_info
->cleanup_work_sem
);
3285 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3286 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3287 up_read(&fs_info
->cleanup_work_sem
);
3288 close_ctree(fs_info
);
3291 up_read(&fs_info
->cleanup_work_sem
);
3293 ret
= btrfs_resume_balance_async(fs_info
);
3295 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3296 close_ctree(fs_info
);
3300 ret
= btrfs_resume_dev_replace_async(fs_info
);
3302 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3303 close_ctree(fs_info
);
3307 btrfs_qgroup_rescan_resume(fs_info
);
3309 if (!fs_info
->uuid_root
) {
3310 btrfs_info(fs_info
, "creating UUID tree");
3311 ret
= btrfs_create_uuid_tree(fs_info
);
3314 "failed to create the UUID tree: %d", ret
);
3315 close_ctree(fs_info
);
3318 } else if (btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) ||
3319 fs_info
->generation
!=
3320 btrfs_super_uuid_tree_generation(disk_super
)) {
3321 btrfs_info(fs_info
, "checking UUID tree");
3322 ret
= btrfs_check_uuid_tree(fs_info
);
3325 "failed to check the UUID tree: %d", ret
);
3326 close_ctree(fs_info
);
3330 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
3332 set_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3335 * backuproot only affect mount behavior, and if open_ctree succeeded,
3336 * no need to keep the flag
3338 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3343 btrfs_free_qgroup_config(fs_info
);
3345 kthread_stop(fs_info
->transaction_kthread
);
3346 btrfs_cleanup_transaction(fs_info
);
3347 btrfs_free_fs_roots(fs_info
);
3349 kthread_stop(fs_info
->cleaner_kthread
);
3352 * make sure we're done with the btree inode before we stop our
3355 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3358 btrfs_sysfs_remove_mounted(fs_info
);
3361 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3364 btrfs_put_block_group_cache(fs_info
);
3367 free_root_pointers(fs_info
, 1);
3368 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3371 btrfs_stop_all_workers(fs_info
);
3372 btrfs_free_block_groups(fs_info
);
3374 btrfs_free_csum_hash(fs_info
);
3377 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3379 iput(fs_info
->btree_inode
);
3381 percpu_counter_destroy(&fs_info
->dev_replace
.bio_counter
);
3382 fail_delalloc_bytes
:
3383 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3384 fail_dirty_metadata_bytes
:
3385 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3387 percpu_counter_destroy(&fs_info
->dio_bytes
);
3389 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3391 btrfs_free_stripe_hash_table(fs_info
);
3392 btrfs_close_devices(fs_info
->fs_devices
);
3396 if (!btrfs_test_opt(fs_info
, USEBACKUPROOT
))
3397 goto fail_tree_roots
;
3399 free_root_pointers(fs_info
, 0);
3401 /* don't use the log in recovery mode, it won't be valid */
3402 btrfs_set_super_log_root(disk_super
, 0);
3404 /* we can't trust the free space cache either */
3405 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3407 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3408 &num_backups_tried
, &backup_index
);
3410 goto fail_block_groups
;
3411 goto retry_root_backup
;
3413 ALLOW_ERROR_INJECTION(open_ctree
, ERRNO
);
3415 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3418 set_buffer_uptodate(bh
);
3420 struct btrfs_device
*device
= (struct btrfs_device
*)
3423 btrfs_warn_rl_in_rcu(device
->fs_info
,
3424 "lost page write due to IO error on %s",
3425 rcu_str_deref(device
->name
));
3426 /* note, we don't set_buffer_write_io_error because we have
3427 * our own ways of dealing with the IO errors
3429 clear_buffer_uptodate(bh
);
3430 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3436 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3437 struct buffer_head
**bh_ret
)
3439 struct buffer_head
*bh
;
3440 struct btrfs_super_block
*super
;
3443 bytenr
= btrfs_sb_offset(copy_num
);
3444 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3447 bh
= __bread(bdev
, bytenr
/ BTRFS_BDEV_BLOCKSIZE
, BTRFS_SUPER_INFO_SIZE
);
3449 * If we fail to read from the underlying devices, as of now
3450 * the best option we have is to mark it EIO.
3455 super
= (struct btrfs_super_block
*)bh
->b_data
;
3456 if (btrfs_super_bytenr(super
) != bytenr
||
3457 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3467 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3469 struct buffer_head
*bh
;
3470 struct buffer_head
*latest
= NULL
;
3471 struct btrfs_super_block
*super
;
3476 /* we would like to check all the supers, but that would make
3477 * a btrfs mount succeed after a mkfs from a different FS.
3478 * So, we need to add a special mount option to scan for
3479 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3481 for (i
= 0; i
< 1; i
++) {
3482 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3486 super
= (struct btrfs_super_block
*)bh
->b_data
;
3488 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3491 transid
= btrfs_super_generation(super
);
3498 return ERR_PTR(ret
);
3504 * Write superblock @sb to the @device. Do not wait for completion, all the
3505 * buffer heads we write are pinned.
3507 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3508 * the expected device size at commit time. Note that max_mirrors must be
3509 * same for write and wait phases.
3511 * Return number of errors when buffer head is not found or submission fails.
3513 static int write_dev_supers(struct btrfs_device
*device
,
3514 struct btrfs_super_block
*sb
, int max_mirrors
)
3516 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
3517 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
3518 struct buffer_head
*bh
;
3525 if (max_mirrors
== 0)
3526 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3528 shash
->tfm
= fs_info
->csum_shash
;
3530 for (i
= 0; i
< max_mirrors
; i
++) {
3531 bytenr
= btrfs_sb_offset(i
);
3532 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3533 device
->commit_total_bytes
)
3536 btrfs_set_super_bytenr(sb
, bytenr
);
3538 crypto_shash_init(shash
);
3539 crypto_shash_update(shash
, (const char *)sb
+ BTRFS_CSUM_SIZE
,
3540 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
3541 crypto_shash_final(shash
, sb
->csum
);
3543 /* One reference for us, and we leave it for the caller */
3544 bh
= __getblk(device
->bdev
, bytenr
/ BTRFS_BDEV_BLOCKSIZE
,
3545 BTRFS_SUPER_INFO_SIZE
);
3547 btrfs_err(device
->fs_info
,
3548 "couldn't get super buffer head for bytenr %llu",
3554 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3556 /* one reference for submit_bh */
3559 set_buffer_uptodate(bh
);
3561 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3562 bh
->b_private
= device
;
3565 * we fua the first super. The others we allow
3568 op_flags
= REQ_SYNC
| REQ_META
| REQ_PRIO
;
3569 if (i
== 0 && !btrfs_test_opt(device
->fs_info
, NOBARRIER
))
3570 op_flags
|= REQ_FUA
;
3571 ret
= btrfsic_submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3575 return errors
< i
? 0 : -1;
3579 * Wait for write completion of superblocks done by write_dev_supers,
3580 * @max_mirrors same for write and wait phases.
3582 * Return number of errors when buffer head is not found or not marked up to
3585 static int wait_dev_supers(struct btrfs_device
*device
, int max_mirrors
)
3587 struct buffer_head
*bh
;
3590 bool primary_failed
= false;
3593 if (max_mirrors
== 0)
3594 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3596 for (i
= 0; i
< max_mirrors
; i
++) {
3597 bytenr
= btrfs_sb_offset(i
);
3598 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3599 device
->commit_total_bytes
)
3602 bh
= __find_get_block(device
->bdev
,
3603 bytenr
/ BTRFS_BDEV_BLOCKSIZE
,
3604 BTRFS_SUPER_INFO_SIZE
);
3608 primary_failed
= true;
3612 if (!buffer_uptodate(bh
)) {
3615 primary_failed
= true;
3618 /* drop our reference */
3621 /* drop the reference from the writing run */
3625 /* log error, force error return */
3626 if (primary_failed
) {
3627 btrfs_err(device
->fs_info
, "error writing primary super block to device %llu",
3632 return errors
< i
? 0 : -1;
3636 * endio for the write_dev_flush, this will wake anyone waiting
3637 * for the barrier when it is done
3639 static void btrfs_end_empty_barrier(struct bio
*bio
)
3641 complete(bio
->bi_private
);
3645 * Submit a flush request to the device if it supports it. Error handling is
3646 * done in the waiting counterpart.
3648 static void write_dev_flush(struct btrfs_device
*device
)
3650 struct request_queue
*q
= bdev_get_queue(device
->bdev
);
3651 struct bio
*bio
= device
->flush_bio
;
3653 if (!test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
))
3657 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3658 bio_set_dev(bio
, device
->bdev
);
3659 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
;
3660 init_completion(&device
->flush_wait
);
3661 bio
->bi_private
= &device
->flush_wait
;
3663 btrfsic_submit_bio(bio
);
3664 set_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3668 * If the flush bio has been submitted by write_dev_flush, wait for it.
3670 static blk_status_t
wait_dev_flush(struct btrfs_device
*device
)
3672 struct bio
*bio
= device
->flush_bio
;
3674 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
))
3677 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3678 wait_for_completion_io(&device
->flush_wait
);
3680 return bio
->bi_status
;
3683 static int check_barrier_error(struct btrfs_fs_info
*fs_info
)
3685 if (!btrfs_check_rw_degradable(fs_info
, NULL
))
3691 * send an empty flush down to each device in parallel,
3692 * then wait for them
3694 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3696 struct list_head
*head
;
3697 struct btrfs_device
*dev
;
3698 int errors_wait
= 0;
3701 lockdep_assert_held(&info
->fs_devices
->device_list_mutex
);
3702 /* send down all the barriers */
3703 head
= &info
->fs_devices
->devices
;
3704 list_for_each_entry(dev
, head
, dev_list
) {
3705 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3709 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3710 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3713 write_dev_flush(dev
);
3714 dev
->last_flush_error
= BLK_STS_OK
;
3717 /* wait for all the barriers */
3718 list_for_each_entry(dev
, head
, dev_list
) {
3719 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3725 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3726 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3729 ret
= wait_dev_flush(dev
);
3731 dev
->last_flush_error
= ret
;
3732 btrfs_dev_stat_inc_and_print(dev
,
3733 BTRFS_DEV_STAT_FLUSH_ERRS
);
3740 * At some point we need the status of all disks
3741 * to arrive at the volume status. So error checking
3742 * is being pushed to a separate loop.
3744 return check_barrier_error(info
);
3749 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3752 int min_tolerated
= INT_MAX
;
3754 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3755 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3756 min_tolerated
= min_t(int, min_tolerated
,
3757 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3758 tolerated_failures
);
3760 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3761 if (raid_type
== BTRFS_RAID_SINGLE
)
3763 if (!(flags
& btrfs_raid_array
[raid_type
].bg_flag
))
3765 min_tolerated
= min_t(int, min_tolerated
,
3766 btrfs_raid_array
[raid_type
].
3767 tolerated_failures
);
3770 if (min_tolerated
== INT_MAX
) {
3771 pr_warn("BTRFS: unknown raid flag: %llu", flags
);
3775 return min_tolerated
;
3778 int write_all_supers(struct btrfs_fs_info
*fs_info
, int max_mirrors
)
3780 struct list_head
*head
;
3781 struct btrfs_device
*dev
;
3782 struct btrfs_super_block
*sb
;
3783 struct btrfs_dev_item
*dev_item
;
3787 int total_errors
= 0;
3790 do_barriers
= !btrfs_test_opt(fs_info
, NOBARRIER
);
3793 * max_mirrors == 0 indicates we're from commit_transaction,
3794 * not from fsync where the tree roots in fs_info have not
3795 * been consistent on disk.
3797 if (max_mirrors
== 0)
3798 backup_super_roots(fs_info
);
3800 sb
= fs_info
->super_for_commit
;
3801 dev_item
= &sb
->dev_item
;
3803 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3804 head
= &fs_info
->fs_devices
->devices
;
3805 max_errors
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
3808 ret
= barrier_all_devices(fs_info
);
3811 &fs_info
->fs_devices
->device_list_mutex
);
3812 btrfs_handle_fs_error(fs_info
, ret
,
3813 "errors while submitting device barriers.");
3818 list_for_each_entry(dev
, head
, dev_list
) {
3823 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3824 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3827 btrfs_set_stack_device_generation(dev_item
, 0);
3828 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3829 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3830 btrfs_set_stack_device_total_bytes(dev_item
,
3831 dev
->commit_total_bytes
);
3832 btrfs_set_stack_device_bytes_used(dev_item
,
3833 dev
->commit_bytes_used
);
3834 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3835 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3836 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3837 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3838 memcpy(dev_item
->fsid
, dev
->fs_devices
->metadata_uuid
,
3841 flags
= btrfs_super_flags(sb
);
3842 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3844 ret
= btrfs_validate_write_super(fs_info
, sb
);
3846 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3847 btrfs_handle_fs_error(fs_info
, -EUCLEAN
,
3848 "unexpected superblock corruption detected");
3852 ret
= write_dev_supers(dev
, sb
, max_mirrors
);
3856 if (total_errors
> max_errors
) {
3857 btrfs_err(fs_info
, "%d errors while writing supers",
3859 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3861 /* FUA is masked off if unsupported and can't be the reason */
3862 btrfs_handle_fs_error(fs_info
, -EIO
,
3863 "%d errors while writing supers",
3869 list_for_each_entry(dev
, head
, dev_list
) {
3872 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3873 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3876 ret
= wait_dev_supers(dev
, max_mirrors
);
3880 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3881 if (total_errors
> max_errors
) {
3882 btrfs_handle_fs_error(fs_info
, -EIO
,
3883 "%d errors while writing supers",
3890 /* Drop a fs root from the radix tree and free it. */
3891 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3892 struct btrfs_root
*root
)
3894 spin_lock(&fs_info
->fs_roots_radix_lock
);
3895 radix_tree_delete(&fs_info
->fs_roots_radix
,
3896 (unsigned long)root
->root_key
.objectid
);
3897 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3899 if (btrfs_root_refs(&root
->root_item
) == 0)
3900 synchronize_srcu(&fs_info
->subvol_srcu
);
3902 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3903 btrfs_free_log(NULL
, root
);
3904 if (root
->reloc_root
) {
3905 free_extent_buffer(root
->reloc_root
->node
);
3906 free_extent_buffer(root
->reloc_root
->commit_root
);
3907 btrfs_put_fs_root(root
->reloc_root
);
3908 root
->reloc_root
= NULL
;
3912 if (root
->free_ino_pinned
)
3913 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3914 if (root
->free_ino_ctl
)
3915 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3916 btrfs_free_fs_root(root
);
3919 void btrfs_free_fs_root(struct btrfs_root
*root
)
3921 iput(root
->ino_cache_inode
);
3922 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3924 free_anon_bdev(root
->anon_dev
);
3925 if (root
->subv_writers
)
3926 btrfs_free_subvolume_writers(root
->subv_writers
);
3927 free_extent_buffer(root
->node
);
3928 free_extent_buffer(root
->commit_root
);
3929 kfree(root
->free_ino_ctl
);
3930 kfree(root
->free_ino_pinned
);
3931 btrfs_put_fs_root(root
);
3934 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3936 u64 root_objectid
= 0;
3937 struct btrfs_root
*gang
[8];
3940 unsigned int ret
= 0;
3944 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3945 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3946 (void **)gang
, root_objectid
,
3949 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3952 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3954 for (i
= 0; i
< ret
; i
++) {
3955 /* Avoid to grab roots in dead_roots */
3956 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3960 /* grab all the search result for later use */
3961 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3963 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3965 for (i
= 0; i
< ret
; i
++) {
3968 root_objectid
= gang
[i
]->root_key
.objectid
;
3969 err
= btrfs_orphan_cleanup(gang
[i
]);
3972 btrfs_put_fs_root(gang
[i
]);
3977 /* release the uncleaned roots due to error */
3978 for (; i
< ret
; i
++) {
3980 btrfs_put_fs_root(gang
[i
]);
3985 int btrfs_commit_super(struct btrfs_fs_info
*fs_info
)
3987 struct btrfs_root
*root
= fs_info
->tree_root
;
3988 struct btrfs_trans_handle
*trans
;
3990 mutex_lock(&fs_info
->cleaner_mutex
);
3991 btrfs_run_delayed_iputs(fs_info
);
3992 mutex_unlock(&fs_info
->cleaner_mutex
);
3993 wake_up_process(fs_info
->cleaner_kthread
);
3995 /* wait until ongoing cleanup work done */
3996 down_write(&fs_info
->cleanup_work_sem
);
3997 up_write(&fs_info
->cleanup_work_sem
);
3999 trans
= btrfs_join_transaction(root
);
4001 return PTR_ERR(trans
);
4002 return btrfs_commit_transaction(trans
);
4005 void close_ctree(struct btrfs_fs_info
*fs_info
)
4009 set_bit(BTRFS_FS_CLOSING_START
, &fs_info
->flags
);
4011 * We don't want the cleaner to start new transactions, add more delayed
4012 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4013 * because that frees the task_struct, and the transaction kthread might
4014 * still try to wake up the cleaner.
4016 kthread_park(fs_info
->cleaner_kthread
);
4018 /* wait for the qgroup rescan worker to stop */
4019 btrfs_qgroup_wait_for_completion(fs_info
, false);
4021 /* wait for the uuid_scan task to finish */
4022 down(&fs_info
->uuid_tree_rescan_sem
);
4023 /* avoid complains from lockdep et al., set sem back to initial state */
4024 up(&fs_info
->uuid_tree_rescan_sem
);
4026 /* pause restriper - we want to resume on mount */
4027 btrfs_pause_balance(fs_info
);
4029 btrfs_dev_replace_suspend_for_unmount(fs_info
);
4031 btrfs_scrub_cancel(fs_info
);
4033 /* wait for any defraggers to finish */
4034 wait_event(fs_info
->transaction_wait
,
4035 (atomic_read(&fs_info
->defrag_running
) == 0));
4037 /* clear out the rbtree of defraggable inodes */
4038 btrfs_cleanup_defrag_inodes(fs_info
);
4040 cancel_work_sync(&fs_info
->async_reclaim_work
);
4042 if (!sb_rdonly(fs_info
->sb
)) {
4044 * The cleaner kthread is stopped, so do one final pass over
4045 * unused block groups.
4047 btrfs_delete_unused_bgs(fs_info
);
4049 ret
= btrfs_commit_super(fs_info
);
4051 btrfs_err(fs_info
, "commit super ret %d", ret
);
4054 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
) ||
4055 test_bit(BTRFS_FS_STATE_TRANS_ABORTED
, &fs_info
->fs_state
))
4056 btrfs_error_commit_super(fs_info
);
4058 kthread_stop(fs_info
->transaction_kthread
);
4059 kthread_stop(fs_info
->cleaner_kthread
);
4061 ASSERT(list_empty(&fs_info
->delayed_iputs
));
4062 set_bit(BTRFS_FS_CLOSING_DONE
, &fs_info
->flags
);
4064 btrfs_free_qgroup_config(fs_info
);
4065 ASSERT(list_empty(&fs_info
->delalloc_roots
));
4067 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
4068 btrfs_info(fs_info
, "at unmount delalloc count %lld",
4069 percpu_counter_sum(&fs_info
->delalloc_bytes
));
4072 if (percpu_counter_sum(&fs_info
->dio_bytes
))
4073 btrfs_info(fs_info
, "at unmount dio bytes count %lld",
4074 percpu_counter_sum(&fs_info
->dio_bytes
));
4076 btrfs_sysfs_remove_mounted(fs_info
);
4077 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
4079 btrfs_free_fs_roots(fs_info
);
4081 btrfs_put_block_group_cache(fs_info
);
4084 * we must make sure there is not any read request to
4085 * submit after we stopping all workers.
4087 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
4088 btrfs_stop_all_workers(fs_info
);
4090 btrfs_free_block_groups(fs_info
);
4092 clear_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
4093 free_root_pointers(fs_info
, 1);
4095 iput(fs_info
->btree_inode
);
4097 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4098 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
))
4099 btrfsic_unmount(fs_info
->fs_devices
);
4102 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
4103 btrfs_close_devices(fs_info
->fs_devices
);
4105 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
4106 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
4107 percpu_counter_destroy(&fs_info
->dio_bytes
);
4108 percpu_counter_destroy(&fs_info
->dev_replace
.bio_counter
);
4109 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
4111 btrfs_free_csum_hash(fs_info
);
4112 btrfs_free_stripe_hash_table(fs_info
);
4113 btrfs_free_ref_cache(fs_info
);
4116 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
4120 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
4122 ret
= extent_buffer_uptodate(buf
);
4126 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
4127 parent_transid
, atomic
);
4133 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
4135 struct btrfs_fs_info
*fs_info
;
4136 struct btrfs_root
*root
;
4137 u64 transid
= btrfs_header_generation(buf
);
4140 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4142 * This is a fast path so only do this check if we have sanity tests
4143 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4144 * outside of the sanity tests.
4146 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED
, &buf
->bflags
)))
4149 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4150 fs_info
= root
->fs_info
;
4151 btrfs_assert_tree_locked(buf
);
4152 if (transid
!= fs_info
->generation
)
4153 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4154 buf
->start
, transid
, fs_info
->generation
);
4155 was_dirty
= set_extent_buffer_dirty(buf
);
4157 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
4159 fs_info
->dirty_metadata_batch
);
4160 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4162 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4163 * but item data not updated.
4164 * So here we should only check item pointers, not item data.
4166 if (btrfs_header_level(buf
) == 0 &&
4167 btrfs_check_leaf_relaxed(buf
)) {
4168 btrfs_print_leaf(buf
);
4174 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
,
4178 * looks as though older kernels can get into trouble with
4179 * this code, they end up stuck in balance_dirty_pages forever
4183 if (current
->flags
& PF_MEMALLOC
)
4187 btrfs_balance_delayed_items(fs_info
);
4189 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
4190 BTRFS_DIRTY_METADATA_THRESH
,
4191 fs_info
->dirty_metadata_batch
);
4193 balance_dirty_pages_ratelimited(fs_info
->btree_inode
->i_mapping
);
4197 void btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
)
4199 __btrfs_btree_balance_dirty(fs_info
, 1);
4202 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info
*fs_info
)
4204 __btrfs_btree_balance_dirty(fs_info
, 0);
4207 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
, int level
,
4208 struct btrfs_key
*first_key
)
4210 return btree_read_extent_buffer_pages(buf
, parent_transid
,
4214 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
)
4216 /* cleanup FS via transaction */
4217 btrfs_cleanup_transaction(fs_info
);
4219 mutex_lock(&fs_info
->cleaner_mutex
);
4220 btrfs_run_delayed_iputs(fs_info
);
4221 mutex_unlock(&fs_info
->cleaner_mutex
);
4223 down_write(&fs_info
->cleanup_work_sem
);
4224 up_write(&fs_info
->cleanup_work_sem
);
4227 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4229 struct btrfs_ordered_extent
*ordered
;
4231 spin_lock(&root
->ordered_extent_lock
);
4233 * This will just short circuit the ordered completion stuff which will
4234 * make sure the ordered extent gets properly cleaned up.
4236 list_for_each_entry(ordered
, &root
->ordered_extents
,
4238 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4239 spin_unlock(&root
->ordered_extent_lock
);
4242 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4244 struct btrfs_root
*root
;
4245 struct list_head splice
;
4247 INIT_LIST_HEAD(&splice
);
4249 spin_lock(&fs_info
->ordered_root_lock
);
4250 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4251 while (!list_empty(&splice
)) {
4252 root
= list_first_entry(&splice
, struct btrfs_root
,
4254 list_move_tail(&root
->ordered_root
,
4255 &fs_info
->ordered_roots
);
4257 spin_unlock(&fs_info
->ordered_root_lock
);
4258 btrfs_destroy_ordered_extents(root
);
4261 spin_lock(&fs_info
->ordered_root_lock
);
4263 spin_unlock(&fs_info
->ordered_root_lock
);
4266 * We need this here because if we've been flipped read-only we won't
4267 * get sync() from the umount, so we need to make sure any ordered
4268 * extents that haven't had their dirty pages IO start writeout yet
4269 * actually get run and error out properly.
4271 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0, (u64
)-1);
4274 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4275 struct btrfs_fs_info
*fs_info
)
4277 struct rb_node
*node
;
4278 struct btrfs_delayed_ref_root
*delayed_refs
;
4279 struct btrfs_delayed_ref_node
*ref
;
4282 delayed_refs
= &trans
->delayed_refs
;
4284 spin_lock(&delayed_refs
->lock
);
4285 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4286 spin_unlock(&delayed_refs
->lock
);
4287 btrfs_info(fs_info
, "delayed_refs has NO entry");
4291 while ((node
= rb_first_cached(&delayed_refs
->href_root
)) != NULL
) {
4292 struct btrfs_delayed_ref_head
*head
;
4294 bool pin_bytes
= false;
4296 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4298 if (btrfs_delayed_ref_lock(delayed_refs
, head
))
4301 spin_lock(&head
->lock
);
4302 while ((n
= rb_first_cached(&head
->ref_tree
)) != NULL
) {
4303 ref
= rb_entry(n
, struct btrfs_delayed_ref_node
,
4306 rb_erase_cached(&ref
->ref_node
, &head
->ref_tree
);
4307 RB_CLEAR_NODE(&ref
->ref_node
);
4308 if (!list_empty(&ref
->add_list
))
4309 list_del(&ref
->add_list
);
4310 atomic_dec(&delayed_refs
->num_entries
);
4311 btrfs_put_delayed_ref(ref
);
4313 if (head
->must_insert_reserved
)
4315 btrfs_free_delayed_extent_op(head
->extent_op
);
4316 btrfs_delete_ref_head(delayed_refs
, head
);
4317 spin_unlock(&head
->lock
);
4318 spin_unlock(&delayed_refs
->lock
);
4319 mutex_unlock(&head
->mutex
);
4322 btrfs_pin_extent(fs_info
, head
->bytenr
,
4323 head
->num_bytes
, 1);
4324 btrfs_cleanup_ref_head_accounting(fs_info
, delayed_refs
, head
);
4325 btrfs_put_delayed_ref_head(head
);
4327 spin_lock(&delayed_refs
->lock
);
4330 spin_unlock(&delayed_refs
->lock
);
4335 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4337 struct btrfs_inode
*btrfs_inode
;
4338 struct list_head splice
;
4340 INIT_LIST_HEAD(&splice
);
4342 spin_lock(&root
->delalloc_lock
);
4343 list_splice_init(&root
->delalloc_inodes
, &splice
);
4345 while (!list_empty(&splice
)) {
4346 struct inode
*inode
= NULL
;
4347 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4349 __btrfs_del_delalloc_inode(root
, btrfs_inode
);
4350 spin_unlock(&root
->delalloc_lock
);
4353 * Make sure we get a live inode and that it'll not disappear
4356 inode
= igrab(&btrfs_inode
->vfs_inode
);
4358 invalidate_inode_pages2(inode
->i_mapping
);
4361 spin_lock(&root
->delalloc_lock
);
4363 spin_unlock(&root
->delalloc_lock
);
4366 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4368 struct btrfs_root
*root
;
4369 struct list_head splice
;
4371 INIT_LIST_HEAD(&splice
);
4373 spin_lock(&fs_info
->delalloc_root_lock
);
4374 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4375 while (!list_empty(&splice
)) {
4376 root
= list_first_entry(&splice
, struct btrfs_root
,
4378 root
= btrfs_grab_fs_root(root
);
4380 spin_unlock(&fs_info
->delalloc_root_lock
);
4382 btrfs_destroy_delalloc_inodes(root
);
4383 btrfs_put_fs_root(root
);
4385 spin_lock(&fs_info
->delalloc_root_lock
);
4387 spin_unlock(&fs_info
->delalloc_root_lock
);
4390 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
4391 struct extent_io_tree
*dirty_pages
,
4395 struct extent_buffer
*eb
;
4400 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4405 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4406 while (start
<= end
) {
4407 eb
= find_extent_buffer(fs_info
, start
);
4408 start
+= fs_info
->nodesize
;
4411 wait_on_extent_buffer_writeback(eb
);
4413 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4415 clear_extent_buffer_dirty(eb
);
4416 free_extent_buffer_stale(eb
);
4423 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
4424 struct extent_io_tree
*pinned_extents
)
4426 struct extent_io_tree
*unpin
;
4432 unpin
= pinned_extents
;
4435 struct extent_state
*cached_state
= NULL
;
4438 * The btrfs_finish_extent_commit() may get the same range as
4439 * ours between find_first_extent_bit and clear_extent_dirty.
4440 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4441 * the same extent range.
4443 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
4444 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4445 EXTENT_DIRTY
, &cached_state
);
4447 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4451 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
4452 free_extent_state(cached_state
);
4453 btrfs_error_unpin_extent_range(fs_info
, start
, end
);
4454 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4459 if (unpin
== &fs_info
->freed_extents
[0])
4460 unpin
= &fs_info
->freed_extents
[1];
4462 unpin
= &fs_info
->freed_extents
[0];
4470 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache
*cache
)
4472 struct inode
*inode
;
4474 inode
= cache
->io_ctl
.inode
;
4476 invalidate_inode_pages2(inode
->i_mapping
);
4477 BTRFS_I(inode
)->generation
= 0;
4478 cache
->io_ctl
.inode
= NULL
;
4481 btrfs_put_block_group(cache
);
4484 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction
*cur_trans
,
4485 struct btrfs_fs_info
*fs_info
)
4487 struct btrfs_block_group_cache
*cache
;
4489 spin_lock(&cur_trans
->dirty_bgs_lock
);
4490 while (!list_empty(&cur_trans
->dirty_bgs
)) {
4491 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
4492 struct btrfs_block_group_cache
,
4495 if (!list_empty(&cache
->io_list
)) {
4496 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4497 list_del_init(&cache
->io_list
);
4498 btrfs_cleanup_bg_io(cache
);
4499 spin_lock(&cur_trans
->dirty_bgs_lock
);
4502 list_del_init(&cache
->dirty_list
);
4503 spin_lock(&cache
->lock
);
4504 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4505 spin_unlock(&cache
->lock
);
4507 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4508 btrfs_put_block_group(cache
);
4509 btrfs_delayed_refs_rsv_release(fs_info
, 1);
4510 spin_lock(&cur_trans
->dirty_bgs_lock
);
4512 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4515 * Refer to the definition of io_bgs member for details why it's safe
4516 * to use it without any locking
4518 while (!list_empty(&cur_trans
->io_bgs
)) {
4519 cache
= list_first_entry(&cur_trans
->io_bgs
,
4520 struct btrfs_block_group_cache
,
4523 list_del_init(&cache
->io_list
);
4524 spin_lock(&cache
->lock
);
4525 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4526 spin_unlock(&cache
->lock
);
4527 btrfs_cleanup_bg_io(cache
);
4531 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4532 struct btrfs_fs_info
*fs_info
)
4534 struct btrfs_device
*dev
, *tmp
;
4536 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
4537 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
4538 ASSERT(list_empty(&cur_trans
->io_bgs
));
4540 list_for_each_entry_safe(dev
, tmp
, &cur_trans
->dev_update_list
,
4542 list_del_init(&dev
->post_commit_list
);
4545 btrfs_destroy_delayed_refs(cur_trans
, fs_info
);
4547 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4548 wake_up(&fs_info
->transaction_blocked_wait
);
4550 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4551 wake_up(&fs_info
->transaction_wait
);
4553 btrfs_destroy_delayed_inodes(fs_info
);
4554 btrfs_assert_delayed_root_empty(fs_info
);
4556 btrfs_destroy_marked_extents(fs_info
, &cur_trans
->dirty_pages
,
4558 btrfs_destroy_pinned_extent(fs_info
,
4559 fs_info
->pinned_extents
);
4561 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4562 wake_up(&cur_trans
->commit_wait
);
4565 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
)
4567 struct btrfs_transaction
*t
;
4569 mutex_lock(&fs_info
->transaction_kthread_mutex
);
4571 spin_lock(&fs_info
->trans_lock
);
4572 while (!list_empty(&fs_info
->trans_list
)) {
4573 t
= list_first_entry(&fs_info
->trans_list
,
4574 struct btrfs_transaction
, list
);
4575 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4576 refcount_inc(&t
->use_count
);
4577 spin_unlock(&fs_info
->trans_lock
);
4578 btrfs_wait_for_commit(fs_info
, t
->transid
);
4579 btrfs_put_transaction(t
);
4580 spin_lock(&fs_info
->trans_lock
);
4583 if (t
== fs_info
->running_transaction
) {
4584 t
->state
= TRANS_STATE_COMMIT_DOING
;
4585 spin_unlock(&fs_info
->trans_lock
);
4587 * We wait for 0 num_writers since we don't hold a trans
4588 * handle open currently for this transaction.
4590 wait_event(t
->writer_wait
,
4591 atomic_read(&t
->num_writers
) == 0);
4593 spin_unlock(&fs_info
->trans_lock
);
4595 btrfs_cleanup_one_transaction(t
, fs_info
);
4597 spin_lock(&fs_info
->trans_lock
);
4598 if (t
== fs_info
->running_transaction
)
4599 fs_info
->running_transaction
= NULL
;
4600 list_del_init(&t
->list
);
4601 spin_unlock(&fs_info
->trans_lock
);
4603 btrfs_put_transaction(t
);
4604 trace_btrfs_transaction_commit(fs_info
->tree_root
);
4605 spin_lock(&fs_info
->trans_lock
);
4607 spin_unlock(&fs_info
->trans_lock
);
4608 btrfs_destroy_all_ordered_extents(fs_info
);
4609 btrfs_destroy_delayed_inodes(fs_info
);
4610 btrfs_assert_delayed_root_empty(fs_info
);
4611 btrfs_destroy_pinned_extent(fs_info
, fs_info
->pinned_extents
);
4612 btrfs_destroy_all_delalloc_inodes(fs_info
);
4613 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
4618 static const struct extent_io_ops btree_extent_io_ops
= {
4619 /* mandatory callbacks */
4620 .submit_bio_hook
= btree_submit_bio_hook
,
4621 .readpage_end_io_hook
= btree_readpage_end_io_hook
,