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/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42 #include "block-group.h"
44 #include "space-info.h"
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static const struct extent_io_ops btree_extent_io_ops
;
54 static void end_workqueue_fn(struct btrfs_work
*work
);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
57 struct btrfs_fs_info
*fs_info
);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
60 struct extent_io_tree
*dirty_pages
,
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
63 struct extent_io_tree
*pinned_extents
);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
);
65 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
);
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
72 struct btrfs_end_io_wq
{
76 struct btrfs_fs_info
*info
;
78 enum btrfs_wq_endio_type metadata
;
79 struct btrfs_work work
;
82 static struct kmem_cache
*btrfs_end_io_wq_cache
;
84 int __init
btrfs_end_io_wq_init(void)
86 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq
),
91 if (!btrfs_end_io_wq_cache
)
96 void __cold
btrfs_end_io_wq_exit(void)
98 kmem_cache_destroy(btrfs_end_io_wq_cache
);
101 static void btrfs_free_csum_hash(struct btrfs_fs_info
*fs_info
)
103 if (fs_info
->csum_shash
)
104 crypto_free_shash(fs_info
->csum_shash
);
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
112 struct async_submit_bio
{
115 extent_submit_bio_start_t
*submit_bio_start
;
118 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
122 struct btrfs_work work
;
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->root_key.objectid. This ensures that all special purpose
135 * roots have separate keysets.
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
149 #ifdef CONFIG_DEBUG_LOCK_ALLOC
150 # if BTRFS_MAX_LEVEL != 8
154 static struct btrfs_lockdep_keyset
{
155 u64 id
; /* root objectid */
156 const char *name_stem
; /* lock name stem */
157 char names
[BTRFS_MAX_LEVEL
+ 1][20];
158 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
159 } btrfs_lockdep_keysets
[] = {
160 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
161 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
162 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
163 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
164 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
165 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
166 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
167 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
168 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
169 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
170 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
171 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
172 { .id
= 0, .name_stem
= "tree" },
175 void __init
btrfs_init_lockdep(void)
179 /* initialize lockdep class names */
180 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
181 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
183 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
184 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
185 "btrfs-%s-%02d", ks
->name_stem
, j
);
189 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
192 struct btrfs_lockdep_keyset
*ks
;
194 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
198 if (ks
->id
== objectid
)
201 lockdep_set_class_and_name(&eb
->lock
,
202 &ks
->keys
[level
], ks
->names
[level
]);
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
211 struct extent_map
*btree_get_extent(struct btrfs_inode
*inode
,
212 struct page
*page
, size_t pg_offset
,
215 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
216 struct extent_map
*em
;
219 read_lock(&em_tree
->lock
);
220 em
= lookup_extent_mapping(em_tree
, start
, len
);
222 read_unlock(&em_tree
->lock
);
225 read_unlock(&em_tree
->lock
);
227 em
= alloc_extent_map();
229 em
= ERR_PTR(-ENOMEM
);
234 em
->block_len
= (u64
)-1;
237 write_lock(&em_tree
->lock
);
238 ret
= add_extent_mapping(em_tree
, em
, 0);
239 if (ret
== -EEXIST
) {
241 em
= lookup_extent_mapping(em_tree
, start
, len
);
248 write_unlock(&em_tree
->lock
);
255 * Compute the csum of a btree block and store the result to provided buffer.
257 static void csum_tree_block(struct extent_buffer
*buf
, u8
*result
)
259 struct btrfs_fs_info
*fs_info
= buf
->fs_info
;
260 const int num_pages
= fs_info
->nodesize
>> PAGE_SHIFT
;
261 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
265 shash
->tfm
= fs_info
->csum_shash
;
266 crypto_shash_init(shash
);
267 kaddr
= page_address(buf
->pages
[0]);
268 crypto_shash_update(shash
, kaddr
+ BTRFS_CSUM_SIZE
,
269 PAGE_SIZE
- BTRFS_CSUM_SIZE
);
271 for (i
= 1; i
< num_pages
; i
++) {
272 kaddr
= page_address(buf
->pages
[i
]);
273 crypto_shash_update(shash
, kaddr
, PAGE_SIZE
);
275 memset(result
, 0, BTRFS_CSUM_SIZE
);
276 crypto_shash_final(shash
, result
);
280 * we can't consider a given block up to date unless the transid of the
281 * block matches the transid in the parent node's pointer. This is how we
282 * detect blocks that either didn't get written at all or got written
283 * in the wrong place.
285 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
286 struct extent_buffer
*eb
, u64 parent_transid
,
289 struct extent_state
*cached_state
= NULL
;
291 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
293 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
300 btrfs_tree_read_lock(eb
);
301 btrfs_set_lock_blocking_read(eb
);
304 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
306 if (extent_buffer_uptodate(eb
) &&
307 btrfs_header_generation(eb
) == parent_transid
) {
311 btrfs_err_rl(eb
->fs_info
,
312 "parent transid verify failed on %llu wanted %llu found %llu",
314 parent_transid
, btrfs_header_generation(eb
));
318 * Things reading via commit roots that don't have normal protection,
319 * like send, can have a really old block in cache that may point at a
320 * block that has been freed and re-allocated. So don't clear uptodate
321 * if we find an eb that is under IO (dirty/writeback) because we could
322 * end up reading in the stale data and then writing it back out and
323 * making everybody very sad.
325 if (!extent_buffer_under_io(eb
))
326 clear_extent_buffer_uptodate(eb
);
328 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
331 btrfs_tree_read_unlock_blocking(eb
);
335 static bool btrfs_supported_super_csum(u16 csum_type
)
338 case BTRFS_CSUM_TYPE_CRC32
:
339 case BTRFS_CSUM_TYPE_XXHASH
:
340 case BTRFS_CSUM_TYPE_SHA256
:
341 case BTRFS_CSUM_TYPE_BLAKE2
:
349 * Return 0 if the superblock checksum type matches the checksum value of that
350 * algorithm. Pass the raw disk superblock data.
352 static int btrfs_check_super_csum(struct btrfs_fs_info
*fs_info
,
355 struct btrfs_super_block
*disk_sb
=
356 (struct btrfs_super_block
*)raw_disk_sb
;
357 char result
[BTRFS_CSUM_SIZE
];
358 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
360 shash
->tfm
= fs_info
->csum_shash
;
363 * The super_block structure does not span the whole
364 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
365 * filled with zeros and is included in the checksum.
367 crypto_shash_digest(shash
, raw_disk_sb
+ BTRFS_CSUM_SIZE
,
368 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
, result
);
370 if (memcmp(disk_sb
->csum
, result
, btrfs_super_csum_size(disk_sb
)))
376 int btrfs_verify_level_key(struct extent_buffer
*eb
, int level
,
377 struct btrfs_key
*first_key
, u64 parent_transid
)
379 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
381 struct btrfs_key found_key
;
384 found_level
= btrfs_header_level(eb
);
385 if (found_level
!= level
) {
386 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG
),
387 KERN_ERR
"BTRFS: tree level check failed\n");
389 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
390 eb
->start
, level
, found_level
);
398 * For live tree block (new tree blocks in current transaction),
399 * we need proper lock context to avoid race, which is impossible here.
400 * So we only checks tree blocks which is read from disk, whose
401 * generation <= fs_info->last_trans_committed.
403 if (btrfs_header_generation(eb
) > fs_info
->last_trans_committed
)
406 /* We have @first_key, so this @eb must have at least one item */
407 if (btrfs_header_nritems(eb
) == 0) {
409 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
411 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
416 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
418 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
419 ret
= btrfs_comp_cpu_keys(first_key
, &found_key
);
422 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG
),
423 KERN_ERR
"BTRFS: tree first key check failed\n");
425 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
426 eb
->start
, parent_transid
, first_key
->objectid
,
427 first_key
->type
, first_key
->offset
,
428 found_key
.objectid
, found_key
.type
,
435 * helper to read a given tree block, doing retries as required when
436 * the checksums don't match and we have alternate mirrors to try.
438 * @parent_transid: expected transid, skip check if 0
439 * @level: expected level, mandatory check
440 * @first_key: expected key of first slot, skip check if NULL
442 static int btree_read_extent_buffer_pages(struct extent_buffer
*eb
,
443 u64 parent_transid
, int level
,
444 struct btrfs_key
*first_key
)
446 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
447 struct extent_io_tree
*io_tree
;
452 int failed_mirror
= 0;
454 io_tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
456 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
457 ret
= read_extent_buffer_pages(eb
, WAIT_COMPLETE
, mirror_num
);
459 if (verify_parent_transid(io_tree
, eb
,
462 else if (btrfs_verify_level_key(eb
, level
,
463 first_key
, parent_transid
))
469 num_copies
= btrfs_num_copies(fs_info
,
474 if (!failed_mirror
) {
476 failed_mirror
= eb
->read_mirror
;
480 if (mirror_num
== failed_mirror
)
483 if (mirror_num
> num_copies
)
487 if (failed
&& !ret
&& failed_mirror
)
488 btrfs_repair_eb_io_failure(eb
, failed_mirror
);
494 * checksum a dirty tree block before IO. This has extra checks to make sure
495 * we only fill in the checksum field in the first page of a multi-page block
498 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
500 u64 start
= page_offset(page
);
502 u8 result
[BTRFS_CSUM_SIZE
];
503 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
504 struct extent_buffer
*eb
;
507 eb
= (struct extent_buffer
*)page
->private;
508 if (page
!= eb
->pages
[0])
511 found_start
= btrfs_header_bytenr(eb
);
513 * Please do not consolidate these warnings into a single if.
514 * It is useful to know what went wrong.
516 if (WARN_ON(found_start
!= start
))
518 if (WARN_ON(!PageUptodate(page
)))
521 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fs_devices
->metadata_uuid
,
522 offsetof(struct btrfs_header
, fsid
),
523 BTRFS_FSID_SIZE
) == 0);
525 csum_tree_block(eb
, result
);
527 if (btrfs_header_level(eb
))
528 ret
= btrfs_check_node(eb
);
530 ret
= btrfs_check_leaf_full(eb
);
533 btrfs_print_tree(eb
, 0);
535 "block=%llu write time tree block corruption detected",
537 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
540 write_extent_buffer(eb
, result
, 0, csum_size
);
545 static int check_tree_block_fsid(struct extent_buffer
*eb
)
547 struct btrfs_fs_info
*fs_info
= eb
->fs_info
;
548 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
549 u8 fsid
[BTRFS_FSID_SIZE
];
552 read_extent_buffer(eb
, fsid
, offsetof(struct btrfs_header
, fsid
),
558 * Checking the incompat flag is only valid for the current
559 * fs. For seed devices it's forbidden to have their uuid
560 * changed so reading ->fsid in this case is fine
562 if (fs_devices
== fs_info
->fs_devices
&&
563 btrfs_fs_incompat(fs_info
, METADATA_UUID
))
564 metadata_uuid
= fs_devices
->metadata_uuid
;
566 metadata_uuid
= fs_devices
->fsid
;
568 if (!memcmp(fsid
, metadata_uuid
, BTRFS_FSID_SIZE
)) {
572 fs_devices
= fs_devices
->seed
;
577 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
578 u64 phy_offset
, struct page
*page
,
579 u64 start
, u64 end
, int mirror
)
583 struct extent_buffer
*eb
;
584 struct btrfs_fs_info
*fs_info
;
587 u8 result
[BTRFS_CSUM_SIZE
];
593 eb
= (struct extent_buffer
*)page
->private;
594 fs_info
= eb
->fs_info
;
595 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
600 atomic_inc(&eb
->refs
);
602 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
606 eb
->read_mirror
= mirror
;
607 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
612 found_start
= btrfs_header_bytenr(eb
);
613 if (found_start
!= eb
->start
) {
614 btrfs_err_rl(fs_info
, "bad tree block start, want %llu have %llu",
615 eb
->start
, found_start
);
619 if (check_tree_block_fsid(eb
)) {
620 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
625 found_level
= btrfs_header_level(eb
);
626 if (found_level
>= BTRFS_MAX_LEVEL
) {
627 btrfs_err(fs_info
, "bad tree block level %d on %llu",
628 (int)btrfs_header_level(eb
), eb
->start
);
633 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
636 csum_tree_block(eb
, result
);
638 if (memcmp_extent_buffer(eb
, result
, 0, csum_size
)) {
639 u8 val
[BTRFS_CSUM_SIZE
] = { 0 };
641 read_extent_buffer(eb
, &val
, 0, csum_size
);
642 btrfs_warn_rl(fs_info
,
643 "%s checksum verify failed on %llu wanted " CSUM_FMT
" found " CSUM_FMT
" level %d",
644 fs_info
->sb
->s_id
, eb
->start
,
645 CSUM_FMT_VALUE(csum_size
, val
),
646 CSUM_FMT_VALUE(csum_size
, result
),
647 btrfs_header_level(eb
));
653 * If this is a leaf block and it is corrupt, set the corrupt bit so
654 * that we don't try and read the other copies of this block, just
657 if (found_level
== 0 && btrfs_check_leaf_full(eb
)) {
658 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
662 if (found_level
> 0 && btrfs_check_node(eb
))
666 set_extent_buffer_uptodate(eb
);
669 "block=%llu read time tree block corruption detected",
673 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
674 btree_readahead_hook(eb
, ret
);
678 * our io error hook is going to dec the io pages
679 * again, we have to make sure it has something
682 atomic_inc(&eb
->io_pages
);
683 clear_extent_buffer_uptodate(eb
);
685 free_extent_buffer(eb
);
690 static void end_workqueue_bio(struct bio
*bio
)
692 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
693 struct btrfs_fs_info
*fs_info
;
694 struct btrfs_workqueue
*wq
;
696 fs_info
= end_io_wq
->info
;
697 end_io_wq
->status
= bio
->bi_status
;
699 if (bio_op(bio
) == REQ_OP_WRITE
) {
700 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
701 wq
= fs_info
->endio_meta_write_workers
;
702 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
703 wq
= fs_info
->endio_freespace_worker
;
704 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
705 wq
= fs_info
->endio_raid56_workers
;
707 wq
= fs_info
->endio_write_workers
;
709 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
710 wq
= fs_info
->endio_raid56_workers
;
711 else if (end_io_wq
->metadata
)
712 wq
= fs_info
->endio_meta_workers
;
714 wq
= fs_info
->endio_workers
;
717 btrfs_init_work(&end_io_wq
->work
, end_workqueue_fn
, NULL
, NULL
);
718 btrfs_queue_work(wq
, &end_io_wq
->work
);
721 blk_status_t
btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
722 enum btrfs_wq_endio_type metadata
)
724 struct btrfs_end_io_wq
*end_io_wq
;
726 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
728 return BLK_STS_RESOURCE
;
730 end_io_wq
->private = bio
->bi_private
;
731 end_io_wq
->end_io
= bio
->bi_end_io
;
732 end_io_wq
->info
= info
;
733 end_io_wq
->status
= 0;
734 end_io_wq
->bio
= bio
;
735 end_io_wq
->metadata
= metadata
;
737 bio
->bi_private
= end_io_wq
;
738 bio
->bi_end_io
= end_workqueue_bio
;
742 static void run_one_async_start(struct btrfs_work
*work
)
744 struct async_submit_bio
*async
;
747 async
= container_of(work
, struct async_submit_bio
, work
);
748 ret
= async
->submit_bio_start(async
->private_data
, async
->bio
,
755 * In order to insert checksums into the metadata in large chunks, we wait
756 * until bio submission time. All the pages in the bio are checksummed and
757 * sums are attached onto the ordered extent record.
759 * At IO completion time the csums attached on the ordered extent record are
760 * inserted into the tree.
762 static void run_one_async_done(struct btrfs_work
*work
)
764 struct async_submit_bio
*async
;
768 async
= container_of(work
, struct async_submit_bio
, work
);
769 inode
= async
->private_data
;
771 /* If an error occurred we just want to clean up the bio and move on */
773 async
->bio
->bi_status
= async
->status
;
774 bio_endio(async
->bio
);
779 * All of the bios that pass through here are from async helpers.
780 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
781 * This changes nothing when cgroups aren't in use.
783 async
->bio
->bi_opf
|= REQ_CGROUP_PUNT
;
784 ret
= btrfs_map_bio(btrfs_sb(inode
->i_sb
), async
->bio
, async
->mirror_num
);
786 async
->bio
->bi_status
= ret
;
787 bio_endio(async
->bio
);
791 static void run_one_async_free(struct btrfs_work
*work
)
793 struct async_submit_bio
*async
;
795 async
= container_of(work
, struct async_submit_bio
, work
);
799 blk_status_t
btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
800 int mirror_num
, unsigned long bio_flags
,
801 u64 bio_offset
, void *private_data
,
802 extent_submit_bio_start_t
*submit_bio_start
)
804 struct async_submit_bio
*async
;
806 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
808 return BLK_STS_RESOURCE
;
810 async
->private_data
= private_data
;
812 async
->mirror_num
= mirror_num
;
813 async
->submit_bio_start
= submit_bio_start
;
815 btrfs_init_work(&async
->work
, run_one_async_start
, run_one_async_done
,
818 async
->bio_offset
= bio_offset
;
822 if (op_is_sync(bio
->bi_opf
))
823 btrfs_set_work_high_priority(&async
->work
);
825 btrfs_queue_work(fs_info
->workers
, &async
->work
);
829 static blk_status_t
btree_csum_one_bio(struct bio
*bio
)
831 struct bio_vec
*bvec
;
832 struct btrfs_root
*root
;
834 struct bvec_iter_all iter_all
;
836 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
837 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
838 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
839 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
844 return errno_to_blk_status(ret
);
847 static blk_status_t
btree_submit_bio_start(void *private_data
, struct bio
*bio
,
851 * when we're called for a write, we're already in the async
852 * submission context. Just jump into btrfs_map_bio
854 return btree_csum_one_bio(bio
);
857 static int check_async_write(struct btrfs_fs_info
*fs_info
,
858 struct btrfs_inode
*bi
)
860 if (atomic_read(&bi
->sync_writers
))
862 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST
, &fs_info
->flags
))
867 static blk_status_t
btree_submit_bio_hook(struct inode
*inode
, struct bio
*bio
,
869 unsigned long bio_flags
)
871 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
872 int async
= check_async_write(fs_info
, BTRFS_I(inode
));
875 if (bio_op(bio
) != REQ_OP_WRITE
) {
877 * called for a read, do the setup so that checksum validation
878 * can happen in the async kernel threads
880 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
881 BTRFS_WQ_ENDIO_METADATA
);
884 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
);
886 ret
= btree_csum_one_bio(bio
);
889 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
);
892 * kthread helpers are used to submit writes so that
893 * checksumming can happen in parallel across all CPUs
895 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, 0,
896 0, inode
, btree_submit_bio_start
);
904 bio
->bi_status
= ret
;
909 #ifdef CONFIG_MIGRATION
910 static int btree_migratepage(struct address_space
*mapping
,
911 struct page
*newpage
, struct page
*page
,
912 enum migrate_mode mode
)
915 * we can't safely write a btree page from here,
916 * we haven't done the locking hook
921 * Buffers may be managed in a filesystem specific way.
922 * We must have no buffers or drop them.
924 if (page_has_private(page
) &&
925 !try_to_release_page(page
, GFP_KERNEL
))
927 return migrate_page(mapping
, newpage
, page
, mode
);
932 static int btree_writepages(struct address_space
*mapping
,
933 struct writeback_control
*wbc
)
935 struct btrfs_fs_info
*fs_info
;
938 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
940 if (wbc
->for_kupdate
)
943 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
944 /* this is a bit racy, but that's ok */
945 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
946 BTRFS_DIRTY_METADATA_THRESH
,
947 fs_info
->dirty_metadata_batch
);
951 return btree_write_cache_pages(mapping
, wbc
);
954 static int btree_readpage(struct file
*file
, struct page
*page
)
956 return extent_read_full_page(page
, btree_get_extent
, 0);
959 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
961 if (PageWriteback(page
) || PageDirty(page
))
964 return try_release_extent_buffer(page
);
967 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
970 struct extent_io_tree
*tree
;
971 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
972 extent_invalidatepage(tree
, page
, offset
);
973 btree_releasepage(page
, GFP_NOFS
);
974 if (PagePrivate(page
)) {
975 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
976 "page private not zero on page %llu",
977 (unsigned long long)page_offset(page
));
978 detach_page_private(page
);
982 static int btree_set_page_dirty(struct page
*page
)
985 struct extent_buffer
*eb
;
987 BUG_ON(!PagePrivate(page
));
988 eb
= (struct extent_buffer
*)page
->private;
990 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
991 BUG_ON(!atomic_read(&eb
->refs
));
992 btrfs_assert_tree_locked(eb
);
994 return __set_page_dirty_nobuffers(page
);
997 static const struct address_space_operations btree_aops
= {
998 .readpage
= btree_readpage
,
999 .writepages
= btree_writepages
,
1000 .releasepage
= btree_releasepage
,
1001 .invalidatepage
= btree_invalidatepage
,
1002 #ifdef CONFIG_MIGRATION
1003 .migratepage
= btree_migratepage
,
1005 .set_page_dirty
= btree_set_page_dirty
,
1008 void readahead_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
1010 struct extent_buffer
*buf
= NULL
;
1013 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1017 ret
= read_extent_buffer_pages(buf
, WAIT_NONE
, 0);
1019 free_extent_buffer_stale(buf
);
1021 free_extent_buffer(buf
);
1024 struct extent_buffer
*btrfs_find_create_tree_block(
1025 struct btrfs_fs_info
*fs_info
,
1028 if (btrfs_is_testing(fs_info
))
1029 return alloc_test_extent_buffer(fs_info
, bytenr
);
1030 return alloc_extent_buffer(fs_info
, bytenr
);
1034 * Read tree block at logical address @bytenr and do variant basic but critical
1037 * @parent_transid: expected transid of this tree block, skip check if 0
1038 * @level: expected level, mandatory check
1039 * @first_key: expected key in slot 0, skip check if NULL
1041 struct extent_buffer
*read_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1042 u64 parent_transid
, int level
,
1043 struct btrfs_key
*first_key
)
1045 struct extent_buffer
*buf
= NULL
;
1048 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1052 ret
= btree_read_extent_buffer_pages(buf
, parent_transid
,
1055 free_extent_buffer_stale(buf
);
1056 return ERR_PTR(ret
);
1062 void btrfs_clean_tree_block(struct extent_buffer
*buf
)
1064 struct btrfs_fs_info
*fs_info
= buf
->fs_info
;
1065 if (btrfs_header_generation(buf
) ==
1066 fs_info
->running_transaction
->transid
) {
1067 btrfs_assert_tree_locked(buf
);
1069 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1070 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
1072 fs_info
->dirty_metadata_batch
);
1073 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1074 btrfs_set_lock_blocking_write(buf
);
1075 clear_extent_buffer_dirty(buf
);
1080 static void __setup_root(struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1083 bool dummy
= test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO
, &fs_info
->fs_state
);
1084 root
->fs_info
= fs_info
;
1086 root
->commit_root
= NULL
;
1088 root
->orphan_cleanup_state
= 0;
1090 root
->last_trans
= 0;
1091 root
->highest_objectid
= 0;
1092 root
->nr_delalloc_inodes
= 0;
1093 root
->nr_ordered_extents
= 0;
1094 root
->inode_tree
= RB_ROOT
;
1095 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1096 root
->block_rsv
= NULL
;
1098 INIT_LIST_HEAD(&root
->dirty_list
);
1099 INIT_LIST_HEAD(&root
->root_list
);
1100 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1101 INIT_LIST_HEAD(&root
->delalloc_root
);
1102 INIT_LIST_HEAD(&root
->ordered_extents
);
1103 INIT_LIST_HEAD(&root
->ordered_root
);
1104 INIT_LIST_HEAD(&root
->reloc_dirty_list
);
1105 INIT_LIST_HEAD(&root
->logged_list
[0]);
1106 INIT_LIST_HEAD(&root
->logged_list
[1]);
1107 spin_lock_init(&root
->inode_lock
);
1108 spin_lock_init(&root
->delalloc_lock
);
1109 spin_lock_init(&root
->ordered_extent_lock
);
1110 spin_lock_init(&root
->accounting_lock
);
1111 spin_lock_init(&root
->log_extents_lock
[0]);
1112 spin_lock_init(&root
->log_extents_lock
[1]);
1113 spin_lock_init(&root
->qgroup_meta_rsv_lock
);
1114 mutex_init(&root
->objectid_mutex
);
1115 mutex_init(&root
->log_mutex
);
1116 mutex_init(&root
->ordered_extent_mutex
);
1117 mutex_init(&root
->delalloc_mutex
);
1118 init_waitqueue_head(&root
->qgroup_flush_wait
);
1119 init_waitqueue_head(&root
->log_writer_wait
);
1120 init_waitqueue_head(&root
->log_commit_wait
[0]);
1121 init_waitqueue_head(&root
->log_commit_wait
[1]);
1122 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1123 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1124 atomic_set(&root
->log_commit
[0], 0);
1125 atomic_set(&root
->log_commit
[1], 0);
1126 atomic_set(&root
->log_writers
, 0);
1127 atomic_set(&root
->log_batch
, 0);
1128 refcount_set(&root
->refs
, 1);
1129 atomic_set(&root
->snapshot_force_cow
, 0);
1130 atomic_set(&root
->nr_swapfiles
, 0);
1131 root
->log_transid
= 0;
1132 root
->log_transid_committed
= -1;
1133 root
->last_log_commit
= 0;
1135 extent_io_tree_init(fs_info
, &root
->dirty_log_pages
,
1136 IO_TREE_ROOT_DIRTY_LOG_PAGES
, NULL
);
1137 extent_io_tree_init(fs_info
, &root
->log_csum_range
,
1138 IO_TREE_LOG_CSUM_RANGE
, NULL
);
1141 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1142 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1143 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1144 root
->root_key
.objectid
= objectid
;
1147 spin_lock_init(&root
->root_item_lock
);
1148 btrfs_qgroup_init_swapped_blocks(&root
->swapped_blocks
);
1149 #ifdef CONFIG_BTRFS_DEBUG
1150 INIT_LIST_HEAD(&root
->leak_list
);
1151 spin_lock(&fs_info
->fs_roots_radix_lock
);
1152 list_add_tail(&root
->leak_list
, &fs_info
->allocated_roots
);
1153 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1157 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1158 u64 objectid
, gfp_t flags
)
1160 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1162 __setup_root(root
, fs_info
, objectid
);
1166 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1167 /* Should only be used by the testing infrastructure */
1168 struct btrfs_root
*btrfs_alloc_dummy_root(struct btrfs_fs_info
*fs_info
)
1170 struct btrfs_root
*root
;
1173 return ERR_PTR(-EINVAL
);
1175 root
= btrfs_alloc_root(fs_info
, BTRFS_ROOT_TREE_OBJECTID
, GFP_KERNEL
);
1177 return ERR_PTR(-ENOMEM
);
1179 /* We don't use the stripesize in selftest, set it as sectorsize */
1180 root
->alloc_bytenr
= 0;
1186 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1189 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1190 struct extent_buffer
*leaf
;
1191 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1192 struct btrfs_root
*root
;
1193 struct btrfs_key key
;
1194 unsigned int nofs_flag
;
1198 * We're holding a transaction handle, so use a NOFS memory allocation
1199 * context to avoid deadlock if reclaim happens.
1201 nofs_flag
= memalloc_nofs_save();
1202 root
= btrfs_alloc_root(fs_info
, objectid
, GFP_KERNEL
);
1203 memalloc_nofs_restore(nofs_flag
);
1205 return ERR_PTR(-ENOMEM
);
1207 root
->root_key
.objectid
= objectid
;
1208 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1209 root
->root_key
.offset
= 0;
1211 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1213 ret
= PTR_ERR(leaf
);
1219 btrfs_mark_buffer_dirty(leaf
);
1221 root
->commit_root
= btrfs_root_node(root
);
1222 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1224 root
->root_item
.flags
= 0;
1225 root
->root_item
.byte_limit
= 0;
1226 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1227 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1228 btrfs_set_root_level(&root
->root_item
, 0);
1229 btrfs_set_root_refs(&root
->root_item
, 1);
1230 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1231 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1232 btrfs_set_root_dirid(&root
->root_item
, 0);
1233 if (is_fstree(objectid
))
1234 generate_random_guid(root
->root_item
.uuid
);
1236 export_guid(root
->root_item
.uuid
, &guid_null
);
1237 root
->root_item
.drop_level
= 0;
1239 key
.objectid
= objectid
;
1240 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1242 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1246 btrfs_tree_unlock(leaf
);
1252 btrfs_tree_unlock(leaf
);
1253 btrfs_put_root(root
);
1255 return ERR_PTR(ret
);
1258 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1259 struct btrfs_fs_info
*fs_info
)
1261 struct btrfs_root
*root
;
1262 struct extent_buffer
*leaf
;
1264 root
= btrfs_alloc_root(fs_info
, BTRFS_TREE_LOG_OBJECTID
, GFP_NOFS
);
1266 return ERR_PTR(-ENOMEM
);
1268 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1269 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1270 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1273 * DON'T set SHAREABLE bit for log trees.
1275 * Log trees are not exposed to user space thus can't be snapshotted,
1276 * and they go away before a real commit is actually done.
1278 * They do store pointers to file data extents, and those reference
1279 * counts still get updated (along with back refs to the log tree).
1282 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1285 btrfs_put_root(root
);
1286 return ERR_CAST(leaf
);
1291 btrfs_mark_buffer_dirty(root
->node
);
1292 btrfs_tree_unlock(root
->node
);
1296 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1297 struct btrfs_fs_info
*fs_info
)
1299 struct btrfs_root
*log_root
;
1301 log_root
= alloc_log_tree(trans
, fs_info
);
1302 if (IS_ERR(log_root
))
1303 return PTR_ERR(log_root
);
1304 WARN_ON(fs_info
->log_root_tree
);
1305 fs_info
->log_root_tree
= log_root
;
1309 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1310 struct btrfs_root
*root
)
1312 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1313 struct btrfs_root
*log_root
;
1314 struct btrfs_inode_item
*inode_item
;
1316 log_root
= alloc_log_tree(trans
, fs_info
);
1317 if (IS_ERR(log_root
))
1318 return PTR_ERR(log_root
);
1320 log_root
->last_trans
= trans
->transid
;
1321 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1323 inode_item
= &log_root
->root_item
.inode
;
1324 btrfs_set_stack_inode_generation(inode_item
, 1);
1325 btrfs_set_stack_inode_size(inode_item
, 3);
1326 btrfs_set_stack_inode_nlink(inode_item
, 1);
1327 btrfs_set_stack_inode_nbytes(inode_item
,
1329 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1331 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1333 WARN_ON(root
->log_root
);
1334 root
->log_root
= log_root
;
1335 root
->log_transid
= 0;
1336 root
->log_transid_committed
= -1;
1337 root
->last_log_commit
= 0;
1341 struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1342 struct btrfs_key
*key
)
1344 struct btrfs_root
*root
;
1345 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1346 struct btrfs_path
*path
;
1351 path
= btrfs_alloc_path();
1353 return ERR_PTR(-ENOMEM
);
1355 root
= btrfs_alloc_root(fs_info
, key
->objectid
, GFP_NOFS
);
1361 ret
= btrfs_find_root(tree_root
, key
, path
,
1362 &root
->root_item
, &root
->root_key
);
1369 generation
= btrfs_root_generation(&root
->root_item
);
1370 level
= btrfs_root_level(&root
->root_item
);
1371 root
->node
= read_tree_block(fs_info
,
1372 btrfs_root_bytenr(&root
->root_item
),
1373 generation
, level
, NULL
);
1374 if (IS_ERR(root
->node
)) {
1375 ret
= PTR_ERR(root
->node
);
1378 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1382 root
->commit_root
= btrfs_root_node(root
);
1384 btrfs_free_path(path
);
1388 btrfs_put_root(root
);
1390 root
= ERR_PTR(ret
);
1395 * Initialize subvolume root in-memory structure
1397 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1399 static int btrfs_init_fs_root(struct btrfs_root
*root
, dev_t anon_dev
)
1402 unsigned int nofs_flag
;
1404 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1405 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1407 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1413 * We might be called under a transaction (e.g. indirect backref
1414 * resolution) which could deadlock if it triggers memory reclaim
1416 nofs_flag
= memalloc_nofs_save();
1417 ret
= btrfs_drew_lock_init(&root
->snapshot_lock
);
1418 memalloc_nofs_restore(nofs_flag
);
1422 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
&&
1423 root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1424 set_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
);
1425 btrfs_check_and_init_root_item(&root
->root_item
);
1428 btrfs_init_free_ino_ctl(root
);
1429 spin_lock_init(&root
->ino_cache_lock
);
1430 init_waitqueue_head(&root
->ino_cache_wait
);
1433 * Don't assign anonymous block device to roots that are not exposed to
1434 * userspace, the id pool is limited to 1M
1436 if (is_fstree(root
->root_key
.objectid
) &&
1437 btrfs_root_refs(&root
->root_item
) > 0) {
1439 ret
= get_anon_bdev(&root
->anon_dev
);
1443 root
->anon_dev
= anon_dev
;
1447 mutex_lock(&root
->objectid_mutex
);
1448 ret
= btrfs_find_highest_objectid(root
,
1449 &root
->highest_objectid
);
1451 mutex_unlock(&root
->objectid_mutex
);
1455 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1457 mutex_unlock(&root
->objectid_mutex
);
1461 /* The caller is responsible to call btrfs_free_fs_root */
1465 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1468 struct btrfs_root
*root
;
1470 spin_lock(&fs_info
->fs_roots_radix_lock
);
1471 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1472 (unsigned long)root_id
);
1474 root
= btrfs_grab_root(root
);
1475 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1479 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1480 struct btrfs_root
*root
)
1484 ret
= radix_tree_preload(GFP_NOFS
);
1488 spin_lock(&fs_info
->fs_roots_radix_lock
);
1489 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1490 (unsigned long)root
->root_key
.objectid
,
1493 btrfs_grab_root(root
);
1494 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1496 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1497 radix_tree_preload_end();
1502 void btrfs_check_leaked_roots(struct btrfs_fs_info
*fs_info
)
1504 #ifdef CONFIG_BTRFS_DEBUG
1505 struct btrfs_root
*root
;
1507 while (!list_empty(&fs_info
->allocated_roots
)) {
1508 root
= list_first_entry(&fs_info
->allocated_roots
,
1509 struct btrfs_root
, leak_list
);
1510 btrfs_err(fs_info
, "leaked root %llu-%llu refcount %d",
1511 root
->root_key
.objectid
, root
->root_key
.offset
,
1512 refcount_read(&root
->refs
));
1513 while (refcount_read(&root
->refs
) > 1)
1514 btrfs_put_root(root
);
1515 btrfs_put_root(root
);
1520 void btrfs_free_fs_info(struct btrfs_fs_info
*fs_info
)
1522 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
1523 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
1524 percpu_counter_destroy(&fs_info
->dio_bytes
);
1525 percpu_counter_destroy(&fs_info
->dev_replace
.bio_counter
);
1526 btrfs_free_csum_hash(fs_info
);
1527 btrfs_free_stripe_hash_table(fs_info
);
1528 btrfs_free_ref_cache(fs_info
);
1529 kfree(fs_info
->balance_ctl
);
1530 kfree(fs_info
->delayed_root
);
1531 btrfs_put_root(fs_info
->extent_root
);
1532 btrfs_put_root(fs_info
->tree_root
);
1533 btrfs_put_root(fs_info
->chunk_root
);
1534 btrfs_put_root(fs_info
->dev_root
);
1535 btrfs_put_root(fs_info
->csum_root
);
1536 btrfs_put_root(fs_info
->quota_root
);
1537 btrfs_put_root(fs_info
->uuid_root
);
1538 btrfs_put_root(fs_info
->free_space_root
);
1539 btrfs_put_root(fs_info
->fs_root
);
1540 btrfs_put_root(fs_info
->data_reloc_root
);
1541 btrfs_check_leaked_roots(fs_info
);
1542 btrfs_extent_buffer_leak_debug_check(fs_info
);
1543 kfree(fs_info
->super_copy
);
1544 kfree(fs_info
->super_for_commit
);
1550 * Get an in-memory reference of a root structure.
1552 * For essential trees like root/extent tree, we grab it from fs_info directly.
1553 * For subvolume trees, we check the cached filesystem roots first. If not
1554 * found, then read it from disk and add it to cached fs roots.
1556 * Caller should release the root by calling btrfs_put_root() after the usage.
1558 * NOTE: Reloc and log trees can't be read by this function as they share the
1559 * same root objectid.
1561 * @objectid: root id
1562 * @anon_dev: preallocated anonymous block device number for new roots,
1563 * pass 0 for new allocation.
1564 * @check_ref: whether to check root item references, If true, return -ENOENT
1567 static struct btrfs_root
*btrfs_get_root_ref(struct btrfs_fs_info
*fs_info
,
1568 u64 objectid
, dev_t anon_dev
,
1571 struct btrfs_root
*root
;
1572 struct btrfs_path
*path
;
1573 struct btrfs_key key
;
1576 if (objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1577 return btrfs_grab_root(fs_info
->tree_root
);
1578 if (objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1579 return btrfs_grab_root(fs_info
->extent_root
);
1580 if (objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1581 return btrfs_grab_root(fs_info
->chunk_root
);
1582 if (objectid
== BTRFS_DEV_TREE_OBJECTID
)
1583 return btrfs_grab_root(fs_info
->dev_root
);
1584 if (objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1585 return btrfs_grab_root(fs_info
->csum_root
);
1586 if (objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1587 return btrfs_grab_root(fs_info
->quota_root
) ?
1588 fs_info
->quota_root
: ERR_PTR(-ENOENT
);
1589 if (objectid
== BTRFS_UUID_TREE_OBJECTID
)
1590 return btrfs_grab_root(fs_info
->uuid_root
) ?
1591 fs_info
->uuid_root
: ERR_PTR(-ENOENT
);
1592 if (objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1593 return btrfs_grab_root(fs_info
->free_space_root
) ?
1594 fs_info
->free_space_root
: ERR_PTR(-ENOENT
);
1596 root
= btrfs_lookup_fs_root(fs_info
, objectid
);
1598 /* Shouldn't get preallocated anon_dev for cached roots */
1600 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1601 btrfs_put_root(root
);
1602 return ERR_PTR(-ENOENT
);
1607 key
.objectid
= objectid
;
1608 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1609 key
.offset
= (u64
)-1;
1610 root
= btrfs_read_tree_root(fs_info
->tree_root
, &key
);
1614 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1619 ret
= btrfs_init_fs_root(root
, anon_dev
);
1623 path
= btrfs_alloc_path();
1628 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1629 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1630 key
.offset
= objectid
;
1632 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1633 btrfs_free_path(path
);
1637 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1639 ret
= btrfs_insert_fs_root(fs_info
, root
);
1641 btrfs_put_root(root
);
1648 btrfs_put_root(root
);
1649 return ERR_PTR(ret
);
1653 * Get in-memory reference of a root structure
1655 * @objectid: tree objectid
1656 * @check_ref: if set, verify that the tree exists and the item has at least
1659 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1660 u64 objectid
, bool check_ref
)
1662 return btrfs_get_root_ref(fs_info
, objectid
, 0, check_ref
);
1666 * Get in-memory reference of a root structure, created as new, optionally pass
1667 * the anonymous block device id
1669 * @objectid: tree objectid
1670 * @anon_dev: if zero, allocate a new anonymous block device or use the
1673 struct btrfs_root
*btrfs_get_new_fs_root(struct btrfs_fs_info
*fs_info
,
1674 u64 objectid
, dev_t anon_dev
)
1676 return btrfs_get_root_ref(fs_info
, objectid
, anon_dev
, true);
1679 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1681 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1683 struct btrfs_device
*device
;
1684 struct backing_dev_info
*bdi
;
1687 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1690 bdi
= device
->bdev
->bd_bdi
;
1691 if (bdi_congested(bdi
, bdi_bits
)) {
1701 * called by the kthread helper functions to finally call the bio end_io
1702 * functions. This is where read checksum verification actually happens
1704 static void end_workqueue_fn(struct btrfs_work
*work
)
1707 struct btrfs_end_io_wq
*end_io_wq
;
1709 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1710 bio
= end_io_wq
->bio
;
1712 bio
->bi_status
= end_io_wq
->status
;
1713 bio
->bi_private
= end_io_wq
->private;
1714 bio
->bi_end_io
= end_io_wq
->end_io
;
1716 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1719 static int cleaner_kthread(void *arg
)
1721 struct btrfs_root
*root
= arg
;
1722 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1728 set_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1730 /* Make the cleaner go to sleep early. */
1731 if (btrfs_need_cleaner_sleep(fs_info
))
1735 * Do not do anything if we might cause open_ctree() to block
1736 * before we have finished mounting the filesystem.
1738 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1741 if (!mutex_trylock(&fs_info
->cleaner_mutex
))
1745 * Avoid the problem that we change the status of the fs
1746 * during the above check and trylock.
1748 if (btrfs_need_cleaner_sleep(fs_info
)) {
1749 mutex_unlock(&fs_info
->cleaner_mutex
);
1753 btrfs_run_delayed_iputs(fs_info
);
1755 again
= btrfs_clean_one_deleted_snapshot(root
);
1756 mutex_unlock(&fs_info
->cleaner_mutex
);
1759 * The defragger has dealt with the R/O remount and umount,
1760 * needn't do anything special here.
1762 btrfs_run_defrag_inodes(fs_info
);
1765 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1766 * with relocation (btrfs_relocate_chunk) and relocation
1767 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1768 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1769 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1770 * unused block groups.
1772 btrfs_delete_unused_bgs(fs_info
);
1774 clear_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1775 if (kthread_should_park())
1777 if (kthread_should_stop())
1780 set_current_state(TASK_INTERRUPTIBLE
);
1782 __set_current_state(TASK_RUNNING
);
1787 static int transaction_kthread(void *arg
)
1789 struct btrfs_root
*root
= arg
;
1790 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1791 struct btrfs_trans_handle
*trans
;
1792 struct btrfs_transaction
*cur
;
1795 unsigned long delay
;
1799 cannot_commit
= false;
1800 delay
= HZ
* fs_info
->commit_interval
;
1801 mutex_lock(&fs_info
->transaction_kthread_mutex
);
1803 spin_lock(&fs_info
->trans_lock
);
1804 cur
= fs_info
->running_transaction
;
1806 spin_unlock(&fs_info
->trans_lock
);
1810 now
= ktime_get_seconds();
1811 if (cur
->state
< TRANS_STATE_COMMIT_START
&&
1812 (now
< cur
->start_time
||
1813 now
- cur
->start_time
< fs_info
->commit_interval
)) {
1814 spin_unlock(&fs_info
->trans_lock
);
1818 transid
= cur
->transid
;
1819 spin_unlock(&fs_info
->trans_lock
);
1821 /* If the file system is aborted, this will always fail. */
1822 trans
= btrfs_attach_transaction(root
);
1823 if (IS_ERR(trans
)) {
1824 if (PTR_ERR(trans
) != -ENOENT
)
1825 cannot_commit
= true;
1828 if (transid
== trans
->transid
) {
1829 btrfs_commit_transaction(trans
);
1831 btrfs_end_transaction(trans
);
1834 wake_up_process(fs_info
->cleaner_kthread
);
1835 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
1837 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1838 &fs_info
->fs_state
)))
1839 btrfs_cleanup_transaction(fs_info
);
1840 if (!kthread_should_stop() &&
1841 (!btrfs_transaction_blocked(fs_info
) ||
1843 schedule_timeout_interruptible(delay
);
1844 } while (!kthread_should_stop());
1849 * This will find the highest generation in the array of root backups. The
1850 * index of the highest array is returned, or -EINVAL if we can't find
1853 * We check to make sure the array is valid by comparing the
1854 * generation of the latest root in the array with the generation
1855 * in the super block. If they don't match we pitch it.
1857 static int find_newest_super_backup(struct btrfs_fs_info
*info
)
1859 const u64 newest_gen
= btrfs_super_generation(info
->super_copy
);
1861 struct btrfs_root_backup
*root_backup
;
1864 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1865 root_backup
= info
->super_copy
->super_roots
+ i
;
1866 cur
= btrfs_backup_tree_root_gen(root_backup
);
1867 if (cur
== newest_gen
)
1875 * copy all the root pointers into the super backup array.
1876 * this will bump the backup pointer by one when it is
1879 static void backup_super_roots(struct btrfs_fs_info
*info
)
1881 const int next_backup
= info
->backup_root_index
;
1882 struct btrfs_root_backup
*root_backup
;
1884 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1887 * make sure all of our padding and empty slots get zero filled
1888 * regardless of which ones we use today
1890 memset(root_backup
, 0, sizeof(*root_backup
));
1892 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1894 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1895 btrfs_set_backup_tree_root_gen(root_backup
,
1896 btrfs_header_generation(info
->tree_root
->node
));
1898 btrfs_set_backup_tree_root_level(root_backup
,
1899 btrfs_header_level(info
->tree_root
->node
));
1901 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1902 btrfs_set_backup_chunk_root_gen(root_backup
,
1903 btrfs_header_generation(info
->chunk_root
->node
));
1904 btrfs_set_backup_chunk_root_level(root_backup
,
1905 btrfs_header_level(info
->chunk_root
->node
));
1907 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1908 btrfs_set_backup_extent_root_gen(root_backup
,
1909 btrfs_header_generation(info
->extent_root
->node
));
1910 btrfs_set_backup_extent_root_level(root_backup
,
1911 btrfs_header_level(info
->extent_root
->node
));
1914 * we might commit during log recovery, which happens before we set
1915 * the fs_root. Make sure it is valid before we fill it in.
1917 if (info
->fs_root
&& info
->fs_root
->node
) {
1918 btrfs_set_backup_fs_root(root_backup
,
1919 info
->fs_root
->node
->start
);
1920 btrfs_set_backup_fs_root_gen(root_backup
,
1921 btrfs_header_generation(info
->fs_root
->node
));
1922 btrfs_set_backup_fs_root_level(root_backup
,
1923 btrfs_header_level(info
->fs_root
->node
));
1926 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1927 btrfs_set_backup_dev_root_gen(root_backup
,
1928 btrfs_header_generation(info
->dev_root
->node
));
1929 btrfs_set_backup_dev_root_level(root_backup
,
1930 btrfs_header_level(info
->dev_root
->node
));
1932 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1933 btrfs_set_backup_csum_root_gen(root_backup
,
1934 btrfs_header_generation(info
->csum_root
->node
));
1935 btrfs_set_backup_csum_root_level(root_backup
,
1936 btrfs_header_level(info
->csum_root
->node
));
1938 btrfs_set_backup_total_bytes(root_backup
,
1939 btrfs_super_total_bytes(info
->super_copy
));
1940 btrfs_set_backup_bytes_used(root_backup
,
1941 btrfs_super_bytes_used(info
->super_copy
));
1942 btrfs_set_backup_num_devices(root_backup
,
1943 btrfs_super_num_devices(info
->super_copy
));
1946 * if we don't copy this out to the super_copy, it won't get remembered
1947 * for the next commit
1949 memcpy(&info
->super_copy
->super_roots
,
1950 &info
->super_for_commit
->super_roots
,
1951 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1955 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1956 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1958 * fs_info - filesystem whose backup roots need to be read
1959 * priority - priority of backup root required
1961 * Returns backup root index on success and -EINVAL otherwise.
1963 static int read_backup_root(struct btrfs_fs_info
*fs_info
, u8 priority
)
1965 int backup_index
= find_newest_super_backup(fs_info
);
1966 struct btrfs_super_block
*super
= fs_info
->super_copy
;
1967 struct btrfs_root_backup
*root_backup
;
1969 if (priority
< BTRFS_NUM_BACKUP_ROOTS
&& backup_index
>= 0) {
1971 return backup_index
;
1973 backup_index
= backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- priority
;
1974 backup_index
%= BTRFS_NUM_BACKUP_ROOTS
;
1979 root_backup
= super
->super_roots
+ backup_index
;
1981 btrfs_set_super_generation(super
,
1982 btrfs_backup_tree_root_gen(root_backup
));
1983 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1984 btrfs_set_super_root_level(super
,
1985 btrfs_backup_tree_root_level(root_backup
));
1986 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1989 * Fixme: the total bytes and num_devices need to match or we should
1992 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1993 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1995 return backup_index
;
1998 /* helper to cleanup workers */
1999 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2001 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2002 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2003 btrfs_destroy_workqueue(fs_info
->workers
);
2004 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2005 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2006 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2007 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2008 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2009 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2010 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2011 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2012 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2013 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2014 if (fs_info
->discard_ctl
.discard_workers
)
2015 destroy_workqueue(fs_info
->discard_ctl
.discard_workers
);
2017 * Now that all other work queues are destroyed, we can safely destroy
2018 * the queues used for metadata I/O, since tasks from those other work
2019 * queues can do metadata I/O operations.
2021 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2022 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2025 static void free_root_extent_buffers(struct btrfs_root
*root
)
2028 free_extent_buffer(root
->node
);
2029 free_extent_buffer(root
->commit_root
);
2031 root
->commit_root
= NULL
;
2035 /* helper to cleanup tree roots */
2036 static void free_root_pointers(struct btrfs_fs_info
*info
, bool free_chunk_root
)
2038 free_root_extent_buffers(info
->tree_root
);
2040 free_root_extent_buffers(info
->dev_root
);
2041 free_root_extent_buffers(info
->extent_root
);
2042 free_root_extent_buffers(info
->csum_root
);
2043 free_root_extent_buffers(info
->quota_root
);
2044 free_root_extent_buffers(info
->uuid_root
);
2045 free_root_extent_buffers(info
->fs_root
);
2046 free_root_extent_buffers(info
->data_reloc_root
);
2047 if (free_chunk_root
)
2048 free_root_extent_buffers(info
->chunk_root
);
2049 free_root_extent_buffers(info
->free_space_root
);
2052 void btrfs_put_root(struct btrfs_root
*root
)
2057 if (refcount_dec_and_test(&root
->refs
)) {
2058 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
2059 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE
, &root
->state
));
2061 free_anon_bdev(root
->anon_dev
);
2062 btrfs_drew_lock_destroy(&root
->snapshot_lock
);
2063 free_root_extent_buffers(root
);
2064 kfree(root
->free_ino_ctl
);
2065 kfree(root
->free_ino_pinned
);
2066 #ifdef CONFIG_BTRFS_DEBUG
2067 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
2068 list_del_init(&root
->leak_list
);
2069 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
2075 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2078 struct btrfs_root
*gang
[8];
2081 while (!list_empty(&fs_info
->dead_roots
)) {
2082 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2083 struct btrfs_root
, root_list
);
2084 list_del(&gang
[0]->root_list
);
2086 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
))
2087 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2088 btrfs_put_root(gang
[0]);
2092 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2097 for (i
= 0; i
< ret
; i
++)
2098 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2102 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2104 mutex_init(&fs_info
->scrub_lock
);
2105 atomic_set(&fs_info
->scrubs_running
, 0);
2106 atomic_set(&fs_info
->scrub_pause_req
, 0);
2107 atomic_set(&fs_info
->scrubs_paused
, 0);
2108 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2109 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2110 refcount_set(&fs_info
->scrub_workers_refcnt
, 0);
2113 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2115 spin_lock_init(&fs_info
->balance_lock
);
2116 mutex_init(&fs_info
->balance_mutex
);
2117 atomic_set(&fs_info
->balance_pause_req
, 0);
2118 atomic_set(&fs_info
->balance_cancel_req
, 0);
2119 fs_info
->balance_ctl
= NULL
;
2120 init_waitqueue_head(&fs_info
->balance_wait_q
);
2123 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
)
2125 struct inode
*inode
= fs_info
->btree_inode
;
2127 inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2128 set_nlink(inode
, 1);
2130 * we set the i_size on the btree inode to the max possible int.
2131 * the real end of the address space is determined by all of
2132 * the devices in the system
2134 inode
->i_size
= OFFSET_MAX
;
2135 inode
->i_mapping
->a_ops
= &btree_aops
;
2137 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
2138 extent_io_tree_init(fs_info
, &BTRFS_I(inode
)->io_tree
,
2139 IO_TREE_INODE_IO
, inode
);
2140 BTRFS_I(inode
)->io_tree
.track_uptodate
= false;
2141 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
);
2143 BTRFS_I(inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2145 BTRFS_I(inode
)->root
= btrfs_grab_root(fs_info
->tree_root
);
2146 memset(&BTRFS_I(inode
)->location
, 0, sizeof(struct btrfs_key
));
2147 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
2148 btrfs_insert_inode_hash(inode
);
2151 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2153 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2154 init_rwsem(&fs_info
->dev_replace
.rwsem
);
2155 init_waitqueue_head(&fs_info
->dev_replace
.replace_wait
);
2158 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2160 spin_lock_init(&fs_info
->qgroup_lock
);
2161 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2162 fs_info
->qgroup_tree
= RB_ROOT
;
2163 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2164 fs_info
->qgroup_seq
= 1;
2165 fs_info
->qgroup_ulist
= NULL
;
2166 fs_info
->qgroup_rescan_running
= false;
2167 mutex_init(&fs_info
->qgroup_rescan_lock
);
2170 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2171 struct btrfs_fs_devices
*fs_devices
)
2173 u32 max_active
= fs_info
->thread_pool_size
;
2174 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2177 btrfs_alloc_workqueue(fs_info
, "worker",
2178 flags
| WQ_HIGHPRI
, max_active
, 16);
2180 fs_info
->delalloc_workers
=
2181 btrfs_alloc_workqueue(fs_info
, "delalloc",
2182 flags
, max_active
, 2);
2184 fs_info
->flush_workers
=
2185 btrfs_alloc_workqueue(fs_info
, "flush_delalloc",
2186 flags
, max_active
, 0);
2188 fs_info
->caching_workers
=
2189 btrfs_alloc_workqueue(fs_info
, "cache", flags
, max_active
, 0);
2191 fs_info
->fixup_workers
=
2192 btrfs_alloc_workqueue(fs_info
, "fixup", flags
, 1, 0);
2195 * endios are largely parallel and should have a very
2198 fs_info
->endio_workers
=
2199 btrfs_alloc_workqueue(fs_info
, "endio", flags
, max_active
, 4);
2200 fs_info
->endio_meta_workers
=
2201 btrfs_alloc_workqueue(fs_info
, "endio-meta", flags
,
2203 fs_info
->endio_meta_write_workers
=
2204 btrfs_alloc_workqueue(fs_info
, "endio-meta-write", flags
,
2206 fs_info
->endio_raid56_workers
=
2207 btrfs_alloc_workqueue(fs_info
, "endio-raid56", flags
,
2209 fs_info
->rmw_workers
=
2210 btrfs_alloc_workqueue(fs_info
, "rmw", flags
, max_active
, 2);
2211 fs_info
->endio_write_workers
=
2212 btrfs_alloc_workqueue(fs_info
, "endio-write", flags
,
2214 fs_info
->endio_freespace_worker
=
2215 btrfs_alloc_workqueue(fs_info
, "freespace-write", flags
,
2217 fs_info
->delayed_workers
=
2218 btrfs_alloc_workqueue(fs_info
, "delayed-meta", flags
,
2220 fs_info
->readahead_workers
=
2221 btrfs_alloc_workqueue(fs_info
, "readahead", flags
,
2223 fs_info
->qgroup_rescan_workers
=
2224 btrfs_alloc_workqueue(fs_info
, "qgroup-rescan", flags
, 1, 0);
2225 fs_info
->discard_ctl
.discard_workers
=
2226 alloc_workqueue("btrfs_discard", WQ_UNBOUND
| WQ_FREEZABLE
, 1);
2228 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2229 fs_info
->flush_workers
&&
2230 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2231 fs_info
->endio_meta_write_workers
&&
2232 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2233 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2234 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2235 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2236 fs_info
->qgroup_rescan_workers
&&
2237 fs_info
->discard_ctl
.discard_workers
)) {
2244 static int btrfs_init_csum_hash(struct btrfs_fs_info
*fs_info
, u16 csum_type
)
2246 struct crypto_shash
*csum_shash
;
2247 const char *csum_driver
= btrfs_super_csum_driver(csum_type
);
2249 csum_shash
= crypto_alloc_shash(csum_driver
, 0, 0);
2251 if (IS_ERR(csum_shash
)) {
2252 btrfs_err(fs_info
, "error allocating %s hash for checksum",
2254 return PTR_ERR(csum_shash
);
2257 fs_info
->csum_shash
= csum_shash
;
2262 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2263 struct btrfs_fs_devices
*fs_devices
)
2266 struct btrfs_root
*log_tree_root
;
2267 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2268 u64 bytenr
= btrfs_super_log_root(disk_super
);
2269 int level
= btrfs_super_log_root_level(disk_super
);
2271 if (fs_devices
->rw_devices
== 0) {
2272 btrfs_warn(fs_info
, "log replay required on RO media");
2276 log_tree_root
= btrfs_alloc_root(fs_info
, BTRFS_TREE_LOG_OBJECTID
,
2281 log_tree_root
->node
= read_tree_block(fs_info
, bytenr
,
2282 fs_info
->generation
+ 1,
2284 if (IS_ERR(log_tree_root
->node
)) {
2285 btrfs_warn(fs_info
, "failed to read log tree");
2286 ret
= PTR_ERR(log_tree_root
->node
);
2287 log_tree_root
->node
= NULL
;
2288 btrfs_put_root(log_tree_root
);
2290 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2291 btrfs_err(fs_info
, "failed to read log tree");
2292 btrfs_put_root(log_tree_root
);
2295 /* returns with log_tree_root freed on success */
2296 ret
= btrfs_recover_log_trees(log_tree_root
);
2298 btrfs_handle_fs_error(fs_info
, ret
,
2299 "Failed to recover log tree");
2300 btrfs_put_root(log_tree_root
);
2304 if (sb_rdonly(fs_info
->sb
)) {
2305 ret
= btrfs_commit_super(fs_info
);
2313 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
)
2315 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2316 struct btrfs_root
*root
;
2317 struct btrfs_key location
;
2320 BUG_ON(!fs_info
->tree_root
);
2322 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2323 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2324 location
.offset
= 0;
2326 root
= btrfs_read_tree_root(tree_root
, &location
);
2328 ret
= PTR_ERR(root
);
2331 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2332 fs_info
->extent_root
= root
;
2334 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2335 root
= btrfs_read_tree_root(tree_root
, &location
);
2337 ret
= PTR_ERR(root
);
2340 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2341 fs_info
->dev_root
= root
;
2342 btrfs_init_devices_late(fs_info
);
2344 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2345 root
= btrfs_read_tree_root(tree_root
, &location
);
2347 ret
= PTR_ERR(root
);
2350 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2351 fs_info
->csum_root
= root
;
2354 * This tree can share blocks with some other fs tree during relocation
2355 * and we need a proper setup by btrfs_get_fs_root
2357 root
= btrfs_get_fs_root(tree_root
->fs_info
,
2358 BTRFS_DATA_RELOC_TREE_OBJECTID
, true);
2360 ret
= PTR_ERR(root
);
2363 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2364 fs_info
->data_reloc_root
= root
;
2366 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2367 root
= btrfs_read_tree_root(tree_root
, &location
);
2368 if (!IS_ERR(root
)) {
2369 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2370 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2371 fs_info
->quota_root
= root
;
2374 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2375 root
= btrfs_read_tree_root(tree_root
, &location
);
2377 ret
= PTR_ERR(root
);
2381 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2382 fs_info
->uuid_root
= root
;
2385 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2386 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2387 root
= btrfs_read_tree_root(tree_root
, &location
);
2389 ret
= PTR_ERR(root
);
2392 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2393 fs_info
->free_space_root
= root
;
2398 btrfs_warn(fs_info
, "failed to read root (objectid=%llu): %d",
2399 location
.objectid
, ret
);
2404 * Real super block validation
2405 * NOTE: super csum type and incompat features will not be checked here.
2407 * @sb: super block to check
2408 * @mirror_num: the super block number to check its bytenr:
2409 * 0 the primary (1st) sb
2410 * 1, 2 2nd and 3rd backup copy
2411 * -1 skip bytenr check
2413 static int validate_super(struct btrfs_fs_info
*fs_info
,
2414 struct btrfs_super_block
*sb
, int mirror_num
)
2416 u64 nodesize
= btrfs_super_nodesize(sb
);
2417 u64 sectorsize
= btrfs_super_sectorsize(sb
);
2420 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
2421 btrfs_err(fs_info
, "no valid FS found");
2424 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
) {
2425 btrfs_err(fs_info
, "unrecognized or unsupported super flag: %llu",
2426 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
2429 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2430 btrfs_err(fs_info
, "tree_root level too big: %d >= %d",
2431 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
2434 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2435 btrfs_err(fs_info
, "chunk_root level too big: %d >= %d",
2436 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
2439 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2440 btrfs_err(fs_info
, "log_root level too big: %d >= %d",
2441 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
2446 * Check sectorsize and nodesize first, other check will need it.
2447 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2449 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
2450 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2451 btrfs_err(fs_info
, "invalid sectorsize %llu", sectorsize
);
2454 /* Only PAGE SIZE is supported yet */
2455 if (sectorsize
!= PAGE_SIZE
) {
2457 "sectorsize %llu not supported yet, only support %lu",
2458 sectorsize
, PAGE_SIZE
);
2461 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
2462 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2463 btrfs_err(fs_info
, "invalid nodesize %llu", nodesize
);
2466 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
2467 btrfs_err(fs_info
, "invalid leafsize %u, should be %llu",
2468 le32_to_cpu(sb
->__unused_leafsize
), nodesize
);
2472 /* Root alignment check */
2473 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
2474 btrfs_warn(fs_info
, "tree_root block unaligned: %llu",
2475 btrfs_super_root(sb
));
2478 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
2479 btrfs_warn(fs_info
, "chunk_root block unaligned: %llu",
2480 btrfs_super_chunk_root(sb
));
2483 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
2484 btrfs_warn(fs_info
, "log_root block unaligned: %llu",
2485 btrfs_super_log_root(sb
));
2489 if (memcmp(fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
,
2490 BTRFS_FSID_SIZE
) != 0) {
2492 "dev_item UUID does not match metadata fsid: %pU != %pU",
2493 fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
);
2498 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2501 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
2502 btrfs_err(fs_info
, "bytes_used is too small %llu",
2503 btrfs_super_bytes_used(sb
));
2506 if (!is_power_of_2(btrfs_super_stripesize(sb
))) {
2507 btrfs_err(fs_info
, "invalid stripesize %u",
2508 btrfs_super_stripesize(sb
));
2511 if (btrfs_super_num_devices(sb
) > (1UL << 31))
2512 btrfs_warn(fs_info
, "suspicious number of devices: %llu",
2513 btrfs_super_num_devices(sb
));
2514 if (btrfs_super_num_devices(sb
) == 0) {
2515 btrfs_err(fs_info
, "number of devices is 0");
2519 if (mirror_num
>= 0 &&
2520 btrfs_super_bytenr(sb
) != btrfs_sb_offset(mirror_num
)) {
2521 btrfs_err(fs_info
, "super offset mismatch %llu != %u",
2522 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
2527 * Obvious sys_chunk_array corruptions, it must hold at least one key
2530 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
2531 btrfs_err(fs_info
, "system chunk array too big %u > %u",
2532 btrfs_super_sys_array_size(sb
),
2533 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
2536 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
2537 + sizeof(struct btrfs_chunk
)) {
2538 btrfs_err(fs_info
, "system chunk array too small %u < %zu",
2539 btrfs_super_sys_array_size(sb
),
2540 sizeof(struct btrfs_disk_key
)
2541 + sizeof(struct btrfs_chunk
));
2546 * The generation is a global counter, we'll trust it more than the others
2547 * but it's still possible that it's the one that's wrong.
2549 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
2551 "suspicious: generation < chunk_root_generation: %llu < %llu",
2552 btrfs_super_generation(sb
),
2553 btrfs_super_chunk_root_generation(sb
));
2554 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
2555 && btrfs_super_cache_generation(sb
) != (u64
)-1)
2557 "suspicious: generation < cache_generation: %llu < %llu",
2558 btrfs_super_generation(sb
),
2559 btrfs_super_cache_generation(sb
));
2565 * Validation of super block at mount time.
2566 * Some checks already done early at mount time, like csum type and incompat
2567 * flags will be skipped.
2569 static int btrfs_validate_mount_super(struct btrfs_fs_info
*fs_info
)
2571 return validate_super(fs_info
, fs_info
->super_copy
, 0);
2575 * Validation of super block at write time.
2576 * Some checks like bytenr check will be skipped as their values will be
2578 * Extra checks like csum type and incompat flags will be done here.
2580 static int btrfs_validate_write_super(struct btrfs_fs_info
*fs_info
,
2581 struct btrfs_super_block
*sb
)
2585 ret
= validate_super(fs_info
, sb
, -1);
2588 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb
))) {
2590 btrfs_err(fs_info
, "invalid csum type, has %u want %u",
2591 btrfs_super_csum_type(sb
), BTRFS_CSUM_TYPE_CRC32
);
2594 if (btrfs_super_incompat_flags(sb
) & ~BTRFS_FEATURE_INCOMPAT_SUPP
) {
2597 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2598 btrfs_super_incompat_flags(sb
),
2599 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP
);
2605 "super block corruption detected before writing it to disk");
2609 static int __cold
init_tree_roots(struct btrfs_fs_info
*fs_info
)
2611 int backup_index
= find_newest_super_backup(fs_info
);
2612 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
2613 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2614 bool handle_error
= false;
2618 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
2623 if (!IS_ERR(tree_root
->node
))
2624 free_extent_buffer(tree_root
->node
);
2625 tree_root
->node
= NULL
;
2627 if (!btrfs_test_opt(fs_info
, USEBACKUPROOT
))
2630 free_root_pointers(fs_info
, 0);
2633 * Don't use the log in recovery mode, it won't be
2636 btrfs_set_super_log_root(sb
, 0);
2638 /* We can't trust the free space cache either */
2639 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2641 ret
= read_backup_root(fs_info
, i
);
2646 generation
= btrfs_super_generation(sb
);
2647 level
= btrfs_super_root_level(sb
);
2648 tree_root
->node
= read_tree_block(fs_info
, btrfs_super_root(sb
),
2649 generation
, level
, NULL
);
2650 if (IS_ERR(tree_root
->node
) ||
2651 !extent_buffer_uptodate(tree_root
->node
)) {
2652 handle_error
= true;
2654 if (IS_ERR(tree_root
->node
)) {
2655 ret
= PTR_ERR(tree_root
->node
);
2656 tree_root
->node
= NULL
;
2657 } else if (!extent_buffer_uptodate(tree_root
->node
)) {
2661 btrfs_warn(fs_info
, "failed to read tree root");
2665 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2666 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2667 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2670 * No need to hold btrfs_root::objectid_mutex since the fs
2671 * hasn't been fully initialised and we are the only user
2673 ret
= btrfs_find_highest_objectid(tree_root
,
2674 &tree_root
->highest_objectid
);
2676 handle_error
= true;
2680 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2682 ret
= btrfs_read_roots(fs_info
);
2684 handle_error
= true;
2688 /* All successful */
2689 fs_info
->generation
= generation
;
2690 fs_info
->last_trans_committed
= generation
;
2692 /* Always begin writing backup roots after the one being used */
2693 if (backup_index
< 0) {
2694 fs_info
->backup_root_index
= 0;
2696 fs_info
->backup_root_index
= backup_index
+ 1;
2697 fs_info
->backup_root_index
%= BTRFS_NUM_BACKUP_ROOTS
;
2705 void btrfs_init_fs_info(struct btrfs_fs_info
*fs_info
)
2707 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2708 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2709 INIT_LIST_HEAD(&fs_info
->trans_list
);
2710 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2711 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2712 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2713 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2714 spin_lock_init(&fs_info
->delalloc_root_lock
);
2715 spin_lock_init(&fs_info
->trans_lock
);
2716 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2717 spin_lock_init(&fs_info
->delayed_iput_lock
);
2718 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2719 spin_lock_init(&fs_info
->super_lock
);
2720 spin_lock_init(&fs_info
->buffer_lock
);
2721 spin_lock_init(&fs_info
->unused_bgs_lock
);
2722 rwlock_init(&fs_info
->tree_mod_log_lock
);
2723 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2724 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2725 mutex_init(&fs_info
->reloc_mutex
);
2726 mutex_init(&fs_info
->delalloc_root_mutex
);
2727 seqlock_init(&fs_info
->profiles_lock
);
2729 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2730 INIT_LIST_HEAD(&fs_info
->space_info
);
2731 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2732 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2733 #ifdef CONFIG_BTRFS_DEBUG
2734 INIT_LIST_HEAD(&fs_info
->allocated_roots
);
2735 INIT_LIST_HEAD(&fs_info
->allocated_ebs
);
2736 spin_lock_init(&fs_info
->eb_leak_lock
);
2738 extent_map_tree_init(&fs_info
->mapping_tree
);
2739 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2740 BTRFS_BLOCK_RSV_GLOBAL
);
2741 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2742 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2743 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2744 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2745 BTRFS_BLOCK_RSV_DELOPS
);
2746 btrfs_init_block_rsv(&fs_info
->delayed_refs_rsv
,
2747 BTRFS_BLOCK_RSV_DELREFS
);
2749 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2750 atomic_set(&fs_info
->defrag_running
, 0);
2751 atomic_set(&fs_info
->reada_works_cnt
, 0);
2752 atomic_set(&fs_info
->nr_delayed_iputs
, 0);
2753 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2754 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2755 fs_info
->metadata_ratio
= 0;
2756 fs_info
->defrag_inodes
= RB_ROOT
;
2757 atomic64_set(&fs_info
->free_chunk_space
, 0);
2758 fs_info
->tree_mod_log
= RB_ROOT
;
2759 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2760 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2761 /* readahead state */
2762 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2763 spin_lock_init(&fs_info
->reada_lock
);
2764 btrfs_init_ref_verify(fs_info
);
2766 fs_info
->thread_pool_size
= min_t(unsigned long,
2767 num_online_cpus() + 2, 8);
2769 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2770 spin_lock_init(&fs_info
->ordered_root_lock
);
2772 btrfs_init_scrub(fs_info
);
2773 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2774 fs_info
->check_integrity_print_mask
= 0;
2776 btrfs_init_balance(fs_info
);
2777 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2779 spin_lock_init(&fs_info
->block_group_cache_lock
);
2780 fs_info
->block_group_cache_tree
= RB_ROOT
;
2781 fs_info
->first_logical_byte
= (u64
)-1;
2783 extent_io_tree_init(fs_info
, &fs_info
->excluded_extents
,
2784 IO_TREE_FS_EXCLUDED_EXTENTS
, NULL
);
2785 set_bit(BTRFS_FS_BARRIER
, &fs_info
->flags
);
2787 mutex_init(&fs_info
->ordered_operations_mutex
);
2788 mutex_init(&fs_info
->tree_log_mutex
);
2789 mutex_init(&fs_info
->chunk_mutex
);
2790 mutex_init(&fs_info
->transaction_kthread_mutex
);
2791 mutex_init(&fs_info
->cleaner_mutex
);
2792 mutex_init(&fs_info
->ro_block_group_mutex
);
2793 init_rwsem(&fs_info
->commit_root_sem
);
2794 init_rwsem(&fs_info
->cleanup_work_sem
);
2795 init_rwsem(&fs_info
->subvol_sem
);
2796 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2798 btrfs_init_dev_replace_locks(fs_info
);
2799 btrfs_init_qgroup(fs_info
);
2800 btrfs_discard_init(fs_info
);
2802 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2803 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2805 init_waitqueue_head(&fs_info
->transaction_throttle
);
2806 init_waitqueue_head(&fs_info
->transaction_wait
);
2807 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2808 init_waitqueue_head(&fs_info
->async_submit_wait
);
2809 init_waitqueue_head(&fs_info
->delayed_iputs_wait
);
2811 /* Usable values until the real ones are cached from the superblock */
2812 fs_info
->nodesize
= 4096;
2813 fs_info
->sectorsize
= 4096;
2814 fs_info
->stripesize
= 4096;
2816 spin_lock_init(&fs_info
->swapfile_pins_lock
);
2817 fs_info
->swapfile_pins
= RB_ROOT
;
2819 fs_info
->send_in_progress
= 0;
2822 static int init_mount_fs_info(struct btrfs_fs_info
*fs_info
, struct super_block
*sb
)
2827 sb
->s_blocksize
= BTRFS_BDEV_BLOCKSIZE
;
2828 sb
->s_blocksize_bits
= blksize_bits(BTRFS_BDEV_BLOCKSIZE
);
2830 ret
= percpu_counter_init(&fs_info
->dio_bytes
, 0, GFP_KERNEL
);
2834 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2838 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2839 (1 + ilog2(nr_cpu_ids
));
2841 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2845 ret
= percpu_counter_init(&fs_info
->dev_replace
.bio_counter
, 0,
2850 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2852 if (!fs_info
->delayed_root
)
2854 btrfs_init_delayed_root(fs_info
->delayed_root
);
2856 return btrfs_alloc_stripe_hash_table(fs_info
);
2859 static int btrfs_uuid_rescan_kthread(void *data
)
2861 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
2865 * 1st step is to iterate through the existing UUID tree and
2866 * to delete all entries that contain outdated data.
2867 * 2nd step is to add all missing entries to the UUID tree.
2869 ret
= btrfs_uuid_tree_iterate(fs_info
);
2872 btrfs_warn(fs_info
, "iterating uuid_tree failed %d",
2874 up(&fs_info
->uuid_tree_rescan_sem
);
2877 return btrfs_uuid_scan_kthread(data
);
2880 static int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
2882 struct task_struct
*task
;
2884 down(&fs_info
->uuid_tree_rescan_sem
);
2885 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
2887 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2888 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
2889 up(&fs_info
->uuid_tree_rescan_sem
);
2890 return PTR_ERR(task
);
2896 int __cold
open_ctree(struct super_block
*sb
, struct btrfs_fs_devices
*fs_devices
,
2905 struct btrfs_super_block
*disk_super
;
2906 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2907 struct btrfs_root
*tree_root
;
2908 struct btrfs_root
*chunk_root
;
2911 int clear_free_space_tree
= 0;
2914 ret
= init_mount_fs_info(fs_info
, sb
);
2920 /* These need to be init'ed before we start creating inodes and such. */
2921 tree_root
= btrfs_alloc_root(fs_info
, BTRFS_ROOT_TREE_OBJECTID
,
2923 fs_info
->tree_root
= tree_root
;
2924 chunk_root
= btrfs_alloc_root(fs_info
, BTRFS_CHUNK_TREE_OBJECTID
,
2926 fs_info
->chunk_root
= chunk_root
;
2927 if (!tree_root
|| !chunk_root
) {
2932 fs_info
->btree_inode
= new_inode(sb
);
2933 if (!fs_info
->btree_inode
) {
2937 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2938 btrfs_init_btree_inode(fs_info
);
2940 invalidate_bdev(fs_devices
->latest_bdev
);
2943 * Read super block and check the signature bytes only
2945 disk_super
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2946 if (IS_ERR(disk_super
)) {
2947 err
= PTR_ERR(disk_super
);
2952 * Verify the type first, if that or the the checksum value are
2953 * corrupted, we'll find out
2955 csum_type
= btrfs_super_csum_type(disk_super
);
2956 if (!btrfs_supported_super_csum(csum_type
)) {
2957 btrfs_err(fs_info
, "unsupported checksum algorithm: %u",
2960 btrfs_release_disk_super(disk_super
);
2964 ret
= btrfs_init_csum_hash(fs_info
, csum_type
);
2967 btrfs_release_disk_super(disk_super
);
2972 * We want to check superblock checksum, the type is stored inside.
2973 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2975 if (btrfs_check_super_csum(fs_info
, (u8
*)disk_super
)) {
2976 btrfs_err(fs_info
, "superblock checksum mismatch");
2978 btrfs_release_disk_super(disk_super
);
2983 * super_copy is zeroed at allocation time and we never touch the
2984 * following bytes up to INFO_SIZE, the checksum is calculated from
2985 * the whole block of INFO_SIZE
2987 memcpy(fs_info
->super_copy
, disk_super
, sizeof(*fs_info
->super_copy
));
2988 btrfs_release_disk_super(disk_super
);
2990 disk_super
= fs_info
->super_copy
;
2992 ASSERT(!memcmp(fs_info
->fs_devices
->fsid
, fs_info
->super_copy
->fsid
,
2995 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
)) {
2996 ASSERT(!memcmp(fs_info
->fs_devices
->metadata_uuid
,
2997 fs_info
->super_copy
->metadata_uuid
,
3001 features
= btrfs_super_flags(disk_super
);
3002 if (features
& BTRFS_SUPER_FLAG_CHANGING_FSID_V2
) {
3003 features
&= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2
;
3004 btrfs_set_super_flags(disk_super
, features
);
3006 "found metadata UUID change in progress flag, clearing");
3009 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
3010 sizeof(*fs_info
->super_for_commit
));
3012 ret
= btrfs_validate_mount_super(fs_info
);
3014 btrfs_err(fs_info
, "superblock contains fatal errors");
3019 if (!btrfs_super_root(disk_super
))
3022 /* check FS state, whether FS is broken. */
3023 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
3024 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
3027 * In the long term, we'll store the compression type in the super
3028 * block, and it'll be used for per file compression control.
3030 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
3032 ret
= btrfs_parse_options(fs_info
, options
, sb
->s_flags
);
3038 features
= btrfs_super_incompat_flags(disk_super
) &
3039 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
3042 "cannot mount because of unsupported optional features (%llx)",
3048 features
= btrfs_super_incompat_flags(disk_super
);
3049 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
3050 if (fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
3051 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
3052 else if (fs_info
->compress_type
== BTRFS_COMPRESS_ZSTD
)
3053 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD
;
3055 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
3056 btrfs_info(fs_info
, "has skinny extents");
3059 * flag our filesystem as having big metadata blocks if
3060 * they are bigger than the page size
3062 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
3063 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
3065 "flagging fs with big metadata feature");
3066 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
3069 nodesize
= btrfs_super_nodesize(disk_super
);
3070 sectorsize
= btrfs_super_sectorsize(disk_super
);
3071 stripesize
= sectorsize
;
3072 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
3073 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
3075 /* Cache block sizes */
3076 fs_info
->nodesize
= nodesize
;
3077 fs_info
->sectorsize
= sectorsize
;
3078 fs_info
->stripesize
= stripesize
;
3081 * mixed block groups end up with duplicate but slightly offset
3082 * extent buffers for the same range. It leads to corruptions
3084 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
3085 (sectorsize
!= nodesize
)) {
3087 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3088 nodesize
, sectorsize
);
3093 * Needn't use the lock because there is no other task which will
3096 btrfs_set_super_incompat_flags(disk_super
, features
);
3098 features
= btrfs_super_compat_ro_flags(disk_super
) &
3099 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
3100 if (!sb_rdonly(sb
) && features
) {
3102 "cannot mount read-write because of unsupported optional features (%llx)",
3108 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
3111 goto fail_sb_buffer
;
3114 sb
->s_bdi
->congested_fn
= btrfs_congested_fn
;
3115 sb
->s_bdi
->congested_data
= fs_info
;
3116 sb
->s_bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
3117 sb
->s_bdi
->ra_pages
= VM_READAHEAD_PAGES
;
3118 sb
->s_bdi
->ra_pages
*= btrfs_super_num_devices(disk_super
);
3119 sb
->s_bdi
->ra_pages
= max(sb
->s_bdi
->ra_pages
, SZ_4M
/ PAGE_SIZE
);
3121 sb
->s_blocksize
= sectorsize
;
3122 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
3123 memcpy(&sb
->s_uuid
, fs_info
->fs_devices
->fsid
, BTRFS_FSID_SIZE
);
3125 mutex_lock(&fs_info
->chunk_mutex
);
3126 ret
= btrfs_read_sys_array(fs_info
);
3127 mutex_unlock(&fs_info
->chunk_mutex
);
3129 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
3130 goto fail_sb_buffer
;
3133 generation
= btrfs_super_chunk_root_generation(disk_super
);
3134 level
= btrfs_super_chunk_root_level(disk_super
);
3136 chunk_root
->node
= read_tree_block(fs_info
,
3137 btrfs_super_chunk_root(disk_super
),
3138 generation
, level
, NULL
);
3139 if (IS_ERR(chunk_root
->node
) ||
3140 !extent_buffer_uptodate(chunk_root
->node
)) {
3141 btrfs_err(fs_info
, "failed to read chunk root");
3142 if (!IS_ERR(chunk_root
->node
))
3143 free_extent_buffer(chunk_root
->node
);
3144 chunk_root
->node
= NULL
;
3145 goto fail_tree_roots
;
3147 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
3148 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
3150 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
3151 offsetof(struct btrfs_header
, chunk_tree_uuid
),
3154 ret
= btrfs_read_chunk_tree(fs_info
);
3156 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
3157 goto fail_tree_roots
;
3161 * Keep the devid that is marked to be the target device for the
3162 * device replace procedure
3164 btrfs_free_extra_devids(fs_devices
, 0);
3166 if (!fs_devices
->latest_bdev
) {
3167 btrfs_err(fs_info
, "failed to read devices");
3168 goto fail_tree_roots
;
3171 ret
= init_tree_roots(fs_info
);
3173 goto fail_tree_roots
;
3176 * If we have a uuid root and we're not being told to rescan we need to
3177 * check the generation here so we can set the
3178 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3179 * transaction during a balance or the log replay without updating the
3180 * uuid generation, and then if we crash we would rescan the uuid tree,
3181 * even though it was perfectly fine.
3183 if (fs_info
->uuid_root
&& !btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) &&
3184 fs_info
->generation
== btrfs_super_uuid_tree_generation(disk_super
))
3185 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
3187 ret
= btrfs_verify_dev_extents(fs_info
);
3190 "failed to verify dev extents against chunks: %d",
3192 goto fail_block_groups
;
3194 ret
= btrfs_recover_balance(fs_info
);
3196 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
3197 goto fail_block_groups
;
3200 ret
= btrfs_init_dev_stats(fs_info
);
3202 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
3203 goto fail_block_groups
;
3206 ret
= btrfs_init_dev_replace(fs_info
);
3208 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
3209 goto fail_block_groups
;
3212 btrfs_free_extra_devids(fs_devices
, 1);
3214 ret
= btrfs_sysfs_add_fsid(fs_devices
);
3216 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
3218 goto fail_block_groups
;
3221 ret
= btrfs_sysfs_add_mounted(fs_info
);
3223 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
3224 goto fail_fsdev_sysfs
;
3227 ret
= btrfs_init_space_info(fs_info
);
3229 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
3233 ret
= btrfs_read_block_groups(fs_info
);
3235 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
3239 if (!sb_rdonly(sb
) && !btrfs_check_rw_degradable(fs_info
, NULL
)) {
3241 "writable mount is not allowed due to too many missing devices");
3245 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3247 if (IS_ERR(fs_info
->cleaner_kthread
))
3250 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3252 "btrfs-transaction");
3253 if (IS_ERR(fs_info
->transaction_kthread
))
3256 if (!btrfs_test_opt(fs_info
, NOSSD
) &&
3257 !fs_info
->fs_devices
->rotating
) {
3258 btrfs_set_and_info(fs_info
, SSD
, "enabling ssd optimizations");
3262 * Mount does not set all options immediately, we can do it now and do
3263 * not have to wait for transaction commit
3265 btrfs_apply_pending_changes(fs_info
);
3267 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3268 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
)) {
3269 ret
= btrfsic_mount(fs_info
, fs_devices
,
3270 btrfs_test_opt(fs_info
,
3271 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3273 fs_info
->check_integrity_print_mask
);
3276 "failed to initialize integrity check module: %d",
3280 ret
= btrfs_read_qgroup_config(fs_info
);
3282 goto fail_trans_kthread
;
3284 if (btrfs_build_ref_tree(fs_info
))
3285 btrfs_err(fs_info
, "couldn't build ref tree");
3287 /* do not make disk changes in broken FS or nologreplay is given */
3288 if (btrfs_super_log_root(disk_super
) != 0 &&
3289 !btrfs_test_opt(fs_info
, NOLOGREPLAY
)) {
3290 btrfs_info(fs_info
, "start tree-log replay");
3291 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3298 ret
= btrfs_find_orphan_roots(fs_info
);
3302 if (!sb_rdonly(sb
)) {
3303 ret
= btrfs_cleanup_fs_roots(fs_info
);
3307 mutex_lock(&fs_info
->cleaner_mutex
);
3308 ret
= btrfs_recover_relocation(tree_root
);
3309 mutex_unlock(&fs_info
->cleaner_mutex
);
3311 btrfs_warn(fs_info
, "failed to recover relocation: %d",
3318 fs_info
->fs_root
= btrfs_get_fs_root(fs_info
, BTRFS_FS_TREE_OBJECTID
, true);
3319 if (IS_ERR(fs_info
->fs_root
)) {
3320 err
= PTR_ERR(fs_info
->fs_root
);
3321 btrfs_warn(fs_info
, "failed to read fs tree: %d", err
);
3322 fs_info
->fs_root
= NULL
;
3329 if (btrfs_test_opt(fs_info
, CLEAR_CACHE
) &&
3330 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3331 clear_free_space_tree
= 1;
3332 } else if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
3333 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE_VALID
)) {
3334 btrfs_warn(fs_info
, "free space tree is invalid");
3335 clear_free_space_tree
= 1;
3338 if (clear_free_space_tree
) {
3339 btrfs_info(fs_info
, "clearing free space tree");
3340 ret
= btrfs_clear_free_space_tree(fs_info
);
3343 "failed to clear free space tree: %d", ret
);
3344 close_ctree(fs_info
);
3349 if (btrfs_test_opt(fs_info
, FREE_SPACE_TREE
) &&
3350 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3351 btrfs_info(fs_info
, "creating free space tree");
3352 ret
= btrfs_create_free_space_tree(fs_info
);
3355 "failed to create free space tree: %d", ret
);
3356 close_ctree(fs_info
);
3361 down_read(&fs_info
->cleanup_work_sem
);
3362 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3363 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3364 up_read(&fs_info
->cleanup_work_sem
);
3365 close_ctree(fs_info
);
3368 up_read(&fs_info
->cleanup_work_sem
);
3370 ret
= btrfs_resume_balance_async(fs_info
);
3372 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3373 close_ctree(fs_info
);
3377 ret
= btrfs_resume_dev_replace_async(fs_info
);
3379 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3380 close_ctree(fs_info
);
3384 btrfs_qgroup_rescan_resume(fs_info
);
3385 btrfs_discard_resume(fs_info
);
3387 if (!fs_info
->uuid_root
) {
3388 btrfs_info(fs_info
, "creating UUID tree");
3389 ret
= btrfs_create_uuid_tree(fs_info
);
3392 "failed to create the UUID tree: %d", ret
);
3393 close_ctree(fs_info
);
3396 } else if (btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) ||
3397 fs_info
->generation
!=
3398 btrfs_super_uuid_tree_generation(disk_super
)) {
3399 btrfs_info(fs_info
, "checking UUID tree");
3400 ret
= btrfs_check_uuid_tree(fs_info
);
3403 "failed to check the UUID tree: %d", ret
);
3404 close_ctree(fs_info
);
3408 set_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3411 * backuproot only affect mount behavior, and if open_ctree succeeded,
3412 * no need to keep the flag
3414 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3419 btrfs_free_qgroup_config(fs_info
);
3421 kthread_stop(fs_info
->transaction_kthread
);
3422 btrfs_cleanup_transaction(fs_info
);
3423 btrfs_free_fs_roots(fs_info
);
3425 kthread_stop(fs_info
->cleaner_kthread
);
3428 * make sure we're done with the btree inode before we stop our
3431 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3434 btrfs_sysfs_remove_mounted(fs_info
);
3437 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3440 btrfs_put_block_group_cache(fs_info
);
3443 if (fs_info
->data_reloc_root
)
3444 btrfs_drop_and_free_fs_root(fs_info
, fs_info
->data_reloc_root
);
3445 free_root_pointers(fs_info
, true);
3446 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3449 btrfs_stop_all_workers(fs_info
);
3450 btrfs_free_block_groups(fs_info
);
3452 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3454 iput(fs_info
->btree_inode
);
3456 btrfs_close_devices(fs_info
->fs_devices
);
3459 ALLOW_ERROR_INJECTION(open_ctree
, ERRNO
);
3461 static void btrfs_end_super_write(struct bio
*bio
)
3463 struct btrfs_device
*device
= bio
->bi_private
;
3464 struct bio_vec
*bvec
;
3465 struct bvec_iter_all iter_all
;
3468 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
3469 page
= bvec
->bv_page
;
3471 if (bio
->bi_status
) {
3472 btrfs_warn_rl_in_rcu(device
->fs_info
,
3473 "lost page write due to IO error on %s (%d)",
3474 rcu_str_deref(device
->name
),
3475 blk_status_to_errno(bio
->bi_status
));
3476 ClearPageUptodate(page
);
3478 btrfs_dev_stat_inc_and_print(device
,
3479 BTRFS_DEV_STAT_WRITE_ERRS
);
3481 SetPageUptodate(page
);
3491 struct btrfs_super_block
*btrfs_read_dev_one_super(struct block_device
*bdev
,
3494 struct btrfs_super_block
*super
;
3497 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
3499 bytenr
= btrfs_sb_offset(copy_num
);
3500 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3501 return ERR_PTR(-EINVAL
);
3503 page
= read_cache_page_gfp(mapping
, bytenr
>> PAGE_SHIFT
, GFP_NOFS
);
3505 return ERR_CAST(page
);
3507 super
= page_address(page
);
3508 if (btrfs_super_bytenr(super
) != bytenr
||
3509 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3510 btrfs_release_disk_super(super
);
3511 return ERR_PTR(-EINVAL
);
3518 struct btrfs_super_block
*btrfs_read_dev_super(struct block_device
*bdev
)
3520 struct btrfs_super_block
*super
, *latest
= NULL
;
3524 /* we would like to check all the supers, but that would make
3525 * a btrfs mount succeed after a mkfs from a different FS.
3526 * So, we need to add a special mount option to scan for
3527 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3529 for (i
= 0; i
< 1; i
++) {
3530 super
= btrfs_read_dev_one_super(bdev
, i
);
3534 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3536 btrfs_release_disk_super(super
);
3539 transid
= btrfs_super_generation(super
);
3547 * Write superblock @sb to the @device. Do not wait for completion, all the
3548 * pages we use for writing are locked.
3550 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3551 * the expected device size at commit time. Note that max_mirrors must be
3552 * same for write and wait phases.
3554 * Return number of errors when page is not found or submission fails.
3556 static int write_dev_supers(struct btrfs_device
*device
,
3557 struct btrfs_super_block
*sb
, int max_mirrors
)
3559 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
3560 struct address_space
*mapping
= device
->bdev
->bd_inode
->i_mapping
;
3561 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
3566 if (max_mirrors
== 0)
3567 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3569 shash
->tfm
= fs_info
->csum_shash
;
3571 for (i
= 0; i
< max_mirrors
; i
++) {
3574 struct btrfs_super_block
*disk_super
;
3576 bytenr
= btrfs_sb_offset(i
);
3577 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3578 device
->commit_total_bytes
)
3581 btrfs_set_super_bytenr(sb
, bytenr
);
3583 crypto_shash_digest(shash
, (const char *)sb
+ BTRFS_CSUM_SIZE
,
3584 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
,
3587 page
= find_or_create_page(mapping
, bytenr
>> PAGE_SHIFT
,
3590 btrfs_err(device
->fs_info
,
3591 "couldn't get super block page for bytenr %llu",
3597 /* Bump the refcount for wait_dev_supers() */
3600 disk_super
= page_address(page
);
3601 memcpy(disk_super
, sb
, BTRFS_SUPER_INFO_SIZE
);
3604 * Directly use bios here instead of relying on the page cache
3605 * to do I/O, so we don't lose the ability to do integrity
3608 bio
= bio_alloc(GFP_NOFS
, 1);
3609 bio_set_dev(bio
, device
->bdev
);
3610 bio
->bi_iter
.bi_sector
= bytenr
>> SECTOR_SHIFT
;
3611 bio
->bi_private
= device
;
3612 bio
->bi_end_io
= btrfs_end_super_write
;
3613 __bio_add_page(bio
, page
, BTRFS_SUPER_INFO_SIZE
,
3614 offset_in_page(bytenr
));
3617 * We FUA only the first super block. The others we allow to
3618 * go down lazy and there's a short window where the on-disk
3619 * copies might still contain the older version.
3621 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_META
| REQ_PRIO
;
3622 if (i
== 0 && !btrfs_test_opt(device
->fs_info
, NOBARRIER
))
3623 bio
->bi_opf
|= REQ_FUA
;
3625 btrfsic_submit_bio(bio
);
3627 return errors
< i
? 0 : -1;
3631 * Wait for write completion of superblocks done by write_dev_supers,
3632 * @max_mirrors same for write and wait phases.
3634 * Return number of errors when page is not found or not marked up to
3637 static int wait_dev_supers(struct btrfs_device
*device
, int max_mirrors
)
3641 bool primary_failed
= false;
3644 if (max_mirrors
== 0)
3645 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3647 for (i
= 0; i
< max_mirrors
; i
++) {
3650 bytenr
= btrfs_sb_offset(i
);
3651 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3652 device
->commit_total_bytes
)
3655 page
= find_get_page(device
->bdev
->bd_inode
->i_mapping
,
3656 bytenr
>> PAGE_SHIFT
);
3660 primary_failed
= true;
3663 /* Page is submitted locked and unlocked once the IO completes */
3664 wait_on_page_locked(page
);
3665 if (PageError(page
)) {
3668 primary_failed
= true;
3671 /* Drop our reference */
3674 /* Drop the reference from the writing run */
3678 /* log error, force error return */
3679 if (primary_failed
) {
3680 btrfs_err(device
->fs_info
, "error writing primary super block to device %llu",
3685 return errors
< i
? 0 : -1;
3689 * endio for the write_dev_flush, this will wake anyone waiting
3690 * for the barrier when it is done
3692 static void btrfs_end_empty_barrier(struct bio
*bio
)
3694 complete(bio
->bi_private
);
3698 * Submit a flush request to the device if it supports it. Error handling is
3699 * done in the waiting counterpart.
3701 static void write_dev_flush(struct btrfs_device
*device
)
3703 struct request_queue
*q
= bdev_get_queue(device
->bdev
);
3704 struct bio
*bio
= device
->flush_bio
;
3706 if (!test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
))
3710 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3711 bio_set_dev(bio
, device
->bdev
);
3712 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
;
3713 init_completion(&device
->flush_wait
);
3714 bio
->bi_private
= &device
->flush_wait
;
3716 btrfsic_submit_bio(bio
);
3717 set_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3721 * If the flush bio has been submitted by write_dev_flush, wait for it.
3723 static blk_status_t
wait_dev_flush(struct btrfs_device
*device
)
3725 struct bio
*bio
= device
->flush_bio
;
3727 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
))
3730 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3731 wait_for_completion_io(&device
->flush_wait
);
3733 return bio
->bi_status
;
3736 static int check_barrier_error(struct btrfs_fs_info
*fs_info
)
3738 if (!btrfs_check_rw_degradable(fs_info
, NULL
))
3744 * send an empty flush down to each device in parallel,
3745 * then wait for them
3747 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3749 struct list_head
*head
;
3750 struct btrfs_device
*dev
;
3751 int errors_wait
= 0;
3754 lockdep_assert_held(&info
->fs_devices
->device_list_mutex
);
3755 /* send down all the barriers */
3756 head
= &info
->fs_devices
->devices
;
3757 list_for_each_entry(dev
, head
, dev_list
) {
3758 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3762 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3763 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3766 write_dev_flush(dev
);
3767 dev
->last_flush_error
= BLK_STS_OK
;
3770 /* wait for all the barriers */
3771 list_for_each_entry(dev
, head
, dev_list
) {
3772 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3778 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3779 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3782 ret
= wait_dev_flush(dev
);
3784 dev
->last_flush_error
= ret
;
3785 btrfs_dev_stat_inc_and_print(dev
,
3786 BTRFS_DEV_STAT_FLUSH_ERRS
);
3793 * At some point we need the status of all disks
3794 * to arrive at the volume status. So error checking
3795 * is being pushed to a separate loop.
3797 return check_barrier_error(info
);
3802 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3805 int min_tolerated
= INT_MAX
;
3807 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3808 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3809 min_tolerated
= min_t(int, min_tolerated
,
3810 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3811 tolerated_failures
);
3813 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3814 if (raid_type
== BTRFS_RAID_SINGLE
)
3816 if (!(flags
& btrfs_raid_array
[raid_type
].bg_flag
))
3818 min_tolerated
= min_t(int, min_tolerated
,
3819 btrfs_raid_array
[raid_type
].
3820 tolerated_failures
);
3823 if (min_tolerated
== INT_MAX
) {
3824 pr_warn("BTRFS: unknown raid flag: %llu", flags
);
3828 return min_tolerated
;
3831 int write_all_supers(struct btrfs_fs_info
*fs_info
, int max_mirrors
)
3833 struct list_head
*head
;
3834 struct btrfs_device
*dev
;
3835 struct btrfs_super_block
*sb
;
3836 struct btrfs_dev_item
*dev_item
;
3840 int total_errors
= 0;
3843 do_barriers
= !btrfs_test_opt(fs_info
, NOBARRIER
);
3846 * max_mirrors == 0 indicates we're from commit_transaction,
3847 * not from fsync where the tree roots in fs_info have not
3848 * been consistent on disk.
3850 if (max_mirrors
== 0)
3851 backup_super_roots(fs_info
);
3853 sb
= fs_info
->super_for_commit
;
3854 dev_item
= &sb
->dev_item
;
3856 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3857 head
= &fs_info
->fs_devices
->devices
;
3858 max_errors
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
3861 ret
= barrier_all_devices(fs_info
);
3864 &fs_info
->fs_devices
->device_list_mutex
);
3865 btrfs_handle_fs_error(fs_info
, ret
,
3866 "errors while submitting device barriers.");
3871 list_for_each_entry(dev
, head
, dev_list
) {
3876 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3877 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3880 btrfs_set_stack_device_generation(dev_item
, 0);
3881 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3882 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3883 btrfs_set_stack_device_total_bytes(dev_item
,
3884 dev
->commit_total_bytes
);
3885 btrfs_set_stack_device_bytes_used(dev_item
,
3886 dev
->commit_bytes_used
);
3887 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3888 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3889 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3890 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3891 memcpy(dev_item
->fsid
, dev
->fs_devices
->metadata_uuid
,
3894 flags
= btrfs_super_flags(sb
);
3895 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3897 ret
= btrfs_validate_write_super(fs_info
, sb
);
3899 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3900 btrfs_handle_fs_error(fs_info
, -EUCLEAN
,
3901 "unexpected superblock corruption detected");
3905 ret
= write_dev_supers(dev
, sb
, max_mirrors
);
3909 if (total_errors
> max_errors
) {
3910 btrfs_err(fs_info
, "%d errors while writing supers",
3912 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3914 /* FUA is masked off if unsupported and can't be the reason */
3915 btrfs_handle_fs_error(fs_info
, -EIO
,
3916 "%d errors while writing supers",
3922 list_for_each_entry(dev
, head
, dev_list
) {
3925 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3926 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3929 ret
= wait_dev_supers(dev
, max_mirrors
);
3933 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3934 if (total_errors
> max_errors
) {
3935 btrfs_handle_fs_error(fs_info
, -EIO
,
3936 "%d errors while writing supers",
3943 /* Drop a fs root from the radix tree and free it. */
3944 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3945 struct btrfs_root
*root
)
3947 bool drop_ref
= false;
3949 spin_lock(&fs_info
->fs_roots_radix_lock
);
3950 radix_tree_delete(&fs_info
->fs_roots_radix
,
3951 (unsigned long)root
->root_key
.objectid
);
3952 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
))
3954 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3956 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3957 ASSERT(root
->log_root
== NULL
);
3958 if (root
->reloc_root
) {
3959 btrfs_put_root(root
->reloc_root
);
3960 root
->reloc_root
= NULL
;
3964 if (root
->free_ino_pinned
)
3965 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3966 if (root
->free_ino_ctl
)
3967 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3968 if (root
->ino_cache_inode
) {
3969 iput(root
->ino_cache_inode
);
3970 root
->ino_cache_inode
= NULL
;
3973 btrfs_put_root(root
);
3976 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3978 u64 root_objectid
= 0;
3979 struct btrfs_root
*gang
[8];
3982 unsigned int ret
= 0;
3985 spin_lock(&fs_info
->fs_roots_radix_lock
);
3986 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3987 (void **)gang
, root_objectid
,
3990 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3993 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3995 for (i
= 0; i
< ret
; i
++) {
3996 /* Avoid to grab roots in dead_roots */
3997 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
4001 /* grab all the search result for later use */
4002 gang
[i
] = btrfs_grab_root(gang
[i
]);
4004 spin_unlock(&fs_info
->fs_roots_radix_lock
);
4006 for (i
= 0; i
< ret
; i
++) {
4009 root_objectid
= gang
[i
]->root_key
.objectid
;
4010 err
= btrfs_orphan_cleanup(gang
[i
]);
4013 btrfs_put_root(gang
[i
]);
4018 /* release the uncleaned roots due to error */
4019 for (; i
< ret
; i
++) {
4021 btrfs_put_root(gang
[i
]);
4026 int btrfs_commit_super(struct btrfs_fs_info
*fs_info
)
4028 struct btrfs_root
*root
= fs_info
->tree_root
;
4029 struct btrfs_trans_handle
*trans
;
4031 mutex_lock(&fs_info
->cleaner_mutex
);
4032 btrfs_run_delayed_iputs(fs_info
);
4033 mutex_unlock(&fs_info
->cleaner_mutex
);
4034 wake_up_process(fs_info
->cleaner_kthread
);
4036 /* wait until ongoing cleanup work done */
4037 down_write(&fs_info
->cleanup_work_sem
);
4038 up_write(&fs_info
->cleanup_work_sem
);
4040 trans
= btrfs_join_transaction(root
);
4042 return PTR_ERR(trans
);
4043 return btrfs_commit_transaction(trans
);
4046 void __cold
close_ctree(struct btrfs_fs_info
*fs_info
)
4050 set_bit(BTRFS_FS_CLOSING_START
, &fs_info
->flags
);
4052 * We don't want the cleaner to start new transactions, add more delayed
4053 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4054 * because that frees the task_struct, and the transaction kthread might
4055 * still try to wake up the cleaner.
4057 kthread_park(fs_info
->cleaner_kthread
);
4059 /* wait for the qgroup rescan worker to stop */
4060 btrfs_qgroup_wait_for_completion(fs_info
, false);
4062 /* wait for the uuid_scan task to finish */
4063 down(&fs_info
->uuid_tree_rescan_sem
);
4064 /* avoid complains from lockdep et al., set sem back to initial state */
4065 up(&fs_info
->uuid_tree_rescan_sem
);
4067 /* pause restriper - we want to resume on mount */
4068 btrfs_pause_balance(fs_info
);
4070 btrfs_dev_replace_suspend_for_unmount(fs_info
);
4072 btrfs_scrub_cancel(fs_info
);
4074 /* wait for any defraggers to finish */
4075 wait_event(fs_info
->transaction_wait
,
4076 (atomic_read(&fs_info
->defrag_running
) == 0));
4078 /* clear out the rbtree of defraggable inodes */
4079 btrfs_cleanup_defrag_inodes(fs_info
);
4081 cancel_work_sync(&fs_info
->async_reclaim_work
);
4083 /* Cancel or finish ongoing discard work */
4084 btrfs_discard_cleanup(fs_info
);
4086 if (!sb_rdonly(fs_info
->sb
)) {
4088 * The cleaner kthread is stopped, so do one final pass over
4089 * unused block groups.
4091 btrfs_delete_unused_bgs(fs_info
);
4094 * There might be existing delayed inode workers still running
4095 * and holding an empty delayed inode item. We must wait for
4096 * them to complete first because they can create a transaction.
4097 * This happens when someone calls btrfs_balance_delayed_items()
4098 * and then a transaction commit runs the same delayed nodes
4099 * before any delayed worker has done something with the nodes.
4100 * We must wait for any worker here and not at transaction
4101 * commit time since that could cause a deadlock.
4102 * This is a very rare case.
4104 btrfs_flush_workqueue(fs_info
->delayed_workers
);
4106 ret
= btrfs_commit_super(fs_info
);
4108 btrfs_err(fs_info
, "commit super ret %d", ret
);
4111 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
) ||
4112 test_bit(BTRFS_FS_STATE_TRANS_ABORTED
, &fs_info
->fs_state
))
4113 btrfs_error_commit_super(fs_info
);
4115 kthread_stop(fs_info
->transaction_kthread
);
4116 kthread_stop(fs_info
->cleaner_kthread
);
4118 ASSERT(list_empty(&fs_info
->delayed_iputs
));
4119 set_bit(BTRFS_FS_CLOSING_DONE
, &fs_info
->flags
);
4121 if (btrfs_check_quota_leak(fs_info
)) {
4122 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
4123 btrfs_err(fs_info
, "qgroup reserved space leaked");
4126 btrfs_free_qgroup_config(fs_info
);
4127 ASSERT(list_empty(&fs_info
->delalloc_roots
));
4129 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
4130 btrfs_info(fs_info
, "at unmount delalloc count %lld",
4131 percpu_counter_sum(&fs_info
->delalloc_bytes
));
4134 if (percpu_counter_sum(&fs_info
->dio_bytes
))
4135 btrfs_info(fs_info
, "at unmount dio bytes count %lld",
4136 percpu_counter_sum(&fs_info
->dio_bytes
));
4138 btrfs_sysfs_remove_mounted(fs_info
);
4139 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
4141 btrfs_put_block_group_cache(fs_info
);
4144 * we must make sure there is not any read request to
4145 * submit after we stopping all workers.
4147 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
4148 btrfs_stop_all_workers(fs_info
);
4150 clear_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
4151 free_root_pointers(fs_info
, true);
4152 btrfs_free_fs_roots(fs_info
);
4155 * We must free the block groups after dropping the fs_roots as we could
4156 * have had an IO error and have left over tree log blocks that aren't
4157 * cleaned up until the fs roots are freed. This makes the block group
4158 * accounting appear to be wrong because there's pending reserved bytes,
4159 * so make sure we do the block group cleanup afterwards.
4161 btrfs_free_block_groups(fs_info
);
4163 iput(fs_info
->btree_inode
);
4165 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4166 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
))
4167 btrfsic_unmount(fs_info
->fs_devices
);
4170 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
4171 btrfs_close_devices(fs_info
->fs_devices
);
4174 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
4178 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
4180 ret
= extent_buffer_uptodate(buf
);
4184 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
4185 parent_transid
, atomic
);
4191 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
4193 struct btrfs_fs_info
*fs_info
;
4194 struct btrfs_root
*root
;
4195 u64 transid
= btrfs_header_generation(buf
);
4198 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4200 * This is a fast path so only do this check if we have sanity tests
4201 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4202 * outside of the sanity tests.
4204 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED
, &buf
->bflags
)))
4207 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4208 fs_info
= root
->fs_info
;
4209 btrfs_assert_tree_locked(buf
);
4210 if (transid
!= fs_info
->generation
)
4211 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4212 buf
->start
, transid
, fs_info
->generation
);
4213 was_dirty
= set_extent_buffer_dirty(buf
);
4215 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
4217 fs_info
->dirty_metadata_batch
);
4218 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4220 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4221 * but item data not updated.
4222 * So here we should only check item pointers, not item data.
4224 if (btrfs_header_level(buf
) == 0 &&
4225 btrfs_check_leaf_relaxed(buf
)) {
4226 btrfs_print_leaf(buf
);
4232 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
,
4236 * looks as though older kernels can get into trouble with
4237 * this code, they end up stuck in balance_dirty_pages forever
4241 if (current
->flags
& PF_MEMALLOC
)
4245 btrfs_balance_delayed_items(fs_info
);
4247 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
4248 BTRFS_DIRTY_METADATA_THRESH
,
4249 fs_info
->dirty_metadata_batch
);
4251 balance_dirty_pages_ratelimited(fs_info
->btree_inode
->i_mapping
);
4255 void btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
)
4257 __btrfs_btree_balance_dirty(fs_info
, 1);
4260 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info
*fs_info
)
4262 __btrfs_btree_balance_dirty(fs_info
, 0);
4265 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
, int level
,
4266 struct btrfs_key
*first_key
)
4268 return btree_read_extent_buffer_pages(buf
, parent_transid
,
4272 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
)
4274 /* cleanup FS via transaction */
4275 btrfs_cleanup_transaction(fs_info
);
4277 mutex_lock(&fs_info
->cleaner_mutex
);
4278 btrfs_run_delayed_iputs(fs_info
);
4279 mutex_unlock(&fs_info
->cleaner_mutex
);
4281 down_write(&fs_info
->cleanup_work_sem
);
4282 up_write(&fs_info
->cleanup_work_sem
);
4285 static void btrfs_drop_all_logs(struct btrfs_fs_info
*fs_info
)
4287 struct btrfs_root
*gang
[8];
4288 u64 root_objectid
= 0;
4291 spin_lock(&fs_info
->fs_roots_radix_lock
);
4292 while ((ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
4293 (void **)gang
, root_objectid
,
4294 ARRAY_SIZE(gang
))) != 0) {
4297 for (i
= 0; i
< ret
; i
++)
4298 gang
[i
] = btrfs_grab_root(gang
[i
]);
4299 spin_unlock(&fs_info
->fs_roots_radix_lock
);
4301 for (i
= 0; i
< ret
; i
++) {
4304 root_objectid
= gang
[i
]->root_key
.objectid
;
4305 btrfs_free_log(NULL
, gang
[i
]);
4306 btrfs_put_root(gang
[i
]);
4309 spin_lock(&fs_info
->fs_roots_radix_lock
);
4311 spin_unlock(&fs_info
->fs_roots_radix_lock
);
4312 btrfs_free_log_root_tree(NULL
, fs_info
);
4315 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4317 struct btrfs_ordered_extent
*ordered
;
4319 spin_lock(&root
->ordered_extent_lock
);
4321 * This will just short circuit the ordered completion stuff which will
4322 * make sure the ordered extent gets properly cleaned up.
4324 list_for_each_entry(ordered
, &root
->ordered_extents
,
4326 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4327 spin_unlock(&root
->ordered_extent_lock
);
4330 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4332 struct btrfs_root
*root
;
4333 struct list_head splice
;
4335 INIT_LIST_HEAD(&splice
);
4337 spin_lock(&fs_info
->ordered_root_lock
);
4338 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4339 while (!list_empty(&splice
)) {
4340 root
= list_first_entry(&splice
, struct btrfs_root
,
4342 list_move_tail(&root
->ordered_root
,
4343 &fs_info
->ordered_roots
);
4345 spin_unlock(&fs_info
->ordered_root_lock
);
4346 btrfs_destroy_ordered_extents(root
);
4349 spin_lock(&fs_info
->ordered_root_lock
);
4351 spin_unlock(&fs_info
->ordered_root_lock
);
4354 * We need this here because if we've been flipped read-only we won't
4355 * get sync() from the umount, so we need to make sure any ordered
4356 * extents that haven't had their dirty pages IO start writeout yet
4357 * actually get run and error out properly.
4359 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0, (u64
)-1);
4362 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4363 struct btrfs_fs_info
*fs_info
)
4365 struct rb_node
*node
;
4366 struct btrfs_delayed_ref_root
*delayed_refs
;
4367 struct btrfs_delayed_ref_node
*ref
;
4370 delayed_refs
= &trans
->delayed_refs
;
4372 spin_lock(&delayed_refs
->lock
);
4373 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4374 spin_unlock(&delayed_refs
->lock
);
4375 btrfs_debug(fs_info
, "delayed_refs has NO entry");
4379 while ((node
= rb_first_cached(&delayed_refs
->href_root
)) != NULL
) {
4380 struct btrfs_delayed_ref_head
*head
;
4382 bool pin_bytes
= false;
4384 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4386 if (btrfs_delayed_ref_lock(delayed_refs
, head
))
4389 spin_lock(&head
->lock
);
4390 while ((n
= rb_first_cached(&head
->ref_tree
)) != NULL
) {
4391 ref
= rb_entry(n
, struct btrfs_delayed_ref_node
,
4394 rb_erase_cached(&ref
->ref_node
, &head
->ref_tree
);
4395 RB_CLEAR_NODE(&ref
->ref_node
);
4396 if (!list_empty(&ref
->add_list
))
4397 list_del(&ref
->add_list
);
4398 atomic_dec(&delayed_refs
->num_entries
);
4399 btrfs_put_delayed_ref(ref
);
4401 if (head
->must_insert_reserved
)
4403 btrfs_free_delayed_extent_op(head
->extent_op
);
4404 btrfs_delete_ref_head(delayed_refs
, head
);
4405 spin_unlock(&head
->lock
);
4406 spin_unlock(&delayed_refs
->lock
);
4407 mutex_unlock(&head
->mutex
);
4410 struct btrfs_block_group
*cache
;
4412 cache
= btrfs_lookup_block_group(fs_info
, head
->bytenr
);
4415 spin_lock(&cache
->space_info
->lock
);
4416 spin_lock(&cache
->lock
);
4417 cache
->pinned
+= head
->num_bytes
;
4418 btrfs_space_info_update_bytes_pinned(fs_info
,
4419 cache
->space_info
, head
->num_bytes
);
4420 cache
->reserved
-= head
->num_bytes
;
4421 cache
->space_info
->bytes_reserved
-= head
->num_bytes
;
4422 spin_unlock(&cache
->lock
);
4423 spin_unlock(&cache
->space_info
->lock
);
4424 percpu_counter_add_batch(
4425 &cache
->space_info
->total_bytes_pinned
,
4426 head
->num_bytes
, BTRFS_TOTAL_BYTES_PINNED_BATCH
);
4428 btrfs_put_block_group(cache
);
4430 btrfs_error_unpin_extent_range(fs_info
, head
->bytenr
,
4431 head
->bytenr
+ head
->num_bytes
- 1);
4433 btrfs_cleanup_ref_head_accounting(fs_info
, delayed_refs
, head
);
4434 btrfs_put_delayed_ref_head(head
);
4436 spin_lock(&delayed_refs
->lock
);
4438 btrfs_qgroup_destroy_extent_records(trans
);
4440 spin_unlock(&delayed_refs
->lock
);
4445 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4447 struct btrfs_inode
*btrfs_inode
;
4448 struct list_head splice
;
4450 INIT_LIST_HEAD(&splice
);
4452 spin_lock(&root
->delalloc_lock
);
4453 list_splice_init(&root
->delalloc_inodes
, &splice
);
4455 while (!list_empty(&splice
)) {
4456 struct inode
*inode
= NULL
;
4457 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4459 __btrfs_del_delalloc_inode(root
, btrfs_inode
);
4460 spin_unlock(&root
->delalloc_lock
);
4463 * Make sure we get a live inode and that it'll not disappear
4466 inode
= igrab(&btrfs_inode
->vfs_inode
);
4468 invalidate_inode_pages2(inode
->i_mapping
);
4471 spin_lock(&root
->delalloc_lock
);
4473 spin_unlock(&root
->delalloc_lock
);
4476 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4478 struct btrfs_root
*root
;
4479 struct list_head splice
;
4481 INIT_LIST_HEAD(&splice
);
4483 spin_lock(&fs_info
->delalloc_root_lock
);
4484 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4485 while (!list_empty(&splice
)) {
4486 root
= list_first_entry(&splice
, struct btrfs_root
,
4488 root
= btrfs_grab_root(root
);
4490 spin_unlock(&fs_info
->delalloc_root_lock
);
4492 btrfs_destroy_delalloc_inodes(root
);
4493 btrfs_put_root(root
);
4495 spin_lock(&fs_info
->delalloc_root_lock
);
4497 spin_unlock(&fs_info
->delalloc_root_lock
);
4500 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
4501 struct extent_io_tree
*dirty_pages
,
4505 struct extent_buffer
*eb
;
4510 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4515 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4516 while (start
<= end
) {
4517 eb
= find_extent_buffer(fs_info
, start
);
4518 start
+= fs_info
->nodesize
;
4521 wait_on_extent_buffer_writeback(eb
);
4523 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4525 clear_extent_buffer_dirty(eb
);
4526 free_extent_buffer_stale(eb
);
4533 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
4534 struct extent_io_tree
*unpin
)
4541 struct extent_state
*cached_state
= NULL
;
4544 * The btrfs_finish_extent_commit() may get the same range as
4545 * ours between find_first_extent_bit and clear_extent_dirty.
4546 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4547 * the same extent range.
4549 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
4550 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4551 EXTENT_DIRTY
, &cached_state
);
4553 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4557 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
4558 free_extent_state(cached_state
);
4559 btrfs_error_unpin_extent_range(fs_info
, start
, end
);
4560 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4567 static void btrfs_cleanup_bg_io(struct btrfs_block_group
*cache
)
4569 struct inode
*inode
;
4571 inode
= cache
->io_ctl
.inode
;
4573 invalidate_inode_pages2(inode
->i_mapping
);
4574 BTRFS_I(inode
)->generation
= 0;
4575 cache
->io_ctl
.inode
= NULL
;
4578 ASSERT(cache
->io_ctl
.pages
== NULL
);
4579 btrfs_put_block_group(cache
);
4582 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction
*cur_trans
,
4583 struct btrfs_fs_info
*fs_info
)
4585 struct btrfs_block_group
*cache
;
4587 spin_lock(&cur_trans
->dirty_bgs_lock
);
4588 while (!list_empty(&cur_trans
->dirty_bgs
)) {
4589 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
4590 struct btrfs_block_group
,
4593 if (!list_empty(&cache
->io_list
)) {
4594 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4595 list_del_init(&cache
->io_list
);
4596 btrfs_cleanup_bg_io(cache
);
4597 spin_lock(&cur_trans
->dirty_bgs_lock
);
4600 list_del_init(&cache
->dirty_list
);
4601 spin_lock(&cache
->lock
);
4602 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4603 spin_unlock(&cache
->lock
);
4605 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4606 btrfs_put_block_group(cache
);
4607 btrfs_delayed_refs_rsv_release(fs_info
, 1);
4608 spin_lock(&cur_trans
->dirty_bgs_lock
);
4610 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4613 * Refer to the definition of io_bgs member for details why it's safe
4614 * to use it without any locking
4616 while (!list_empty(&cur_trans
->io_bgs
)) {
4617 cache
= list_first_entry(&cur_trans
->io_bgs
,
4618 struct btrfs_block_group
,
4621 list_del_init(&cache
->io_list
);
4622 spin_lock(&cache
->lock
);
4623 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4624 spin_unlock(&cache
->lock
);
4625 btrfs_cleanup_bg_io(cache
);
4629 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4630 struct btrfs_fs_info
*fs_info
)
4632 struct btrfs_device
*dev
, *tmp
;
4634 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
4635 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
4636 ASSERT(list_empty(&cur_trans
->io_bgs
));
4638 list_for_each_entry_safe(dev
, tmp
, &cur_trans
->dev_update_list
,
4640 list_del_init(&dev
->post_commit_list
);
4643 btrfs_destroy_delayed_refs(cur_trans
, fs_info
);
4645 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4646 wake_up(&fs_info
->transaction_blocked_wait
);
4648 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4649 wake_up(&fs_info
->transaction_wait
);
4651 btrfs_destroy_delayed_inodes(fs_info
);
4653 btrfs_destroy_marked_extents(fs_info
, &cur_trans
->dirty_pages
,
4655 btrfs_destroy_pinned_extent(fs_info
, &cur_trans
->pinned_extents
);
4657 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4658 wake_up(&cur_trans
->commit_wait
);
4661 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
)
4663 struct btrfs_transaction
*t
;
4665 mutex_lock(&fs_info
->transaction_kthread_mutex
);
4667 spin_lock(&fs_info
->trans_lock
);
4668 while (!list_empty(&fs_info
->trans_list
)) {
4669 t
= list_first_entry(&fs_info
->trans_list
,
4670 struct btrfs_transaction
, list
);
4671 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4672 refcount_inc(&t
->use_count
);
4673 spin_unlock(&fs_info
->trans_lock
);
4674 btrfs_wait_for_commit(fs_info
, t
->transid
);
4675 btrfs_put_transaction(t
);
4676 spin_lock(&fs_info
->trans_lock
);
4679 if (t
== fs_info
->running_transaction
) {
4680 t
->state
= TRANS_STATE_COMMIT_DOING
;
4681 spin_unlock(&fs_info
->trans_lock
);
4683 * We wait for 0 num_writers since we don't hold a trans
4684 * handle open currently for this transaction.
4686 wait_event(t
->writer_wait
,
4687 atomic_read(&t
->num_writers
) == 0);
4689 spin_unlock(&fs_info
->trans_lock
);
4691 btrfs_cleanup_one_transaction(t
, fs_info
);
4693 spin_lock(&fs_info
->trans_lock
);
4694 if (t
== fs_info
->running_transaction
)
4695 fs_info
->running_transaction
= NULL
;
4696 list_del_init(&t
->list
);
4697 spin_unlock(&fs_info
->trans_lock
);
4699 btrfs_put_transaction(t
);
4700 trace_btrfs_transaction_commit(fs_info
->tree_root
);
4701 spin_lock(&fs_info
->trans_lock
);
4703 spin_unlock(&fs_info
->trans_lock
);
4704 btrfs_destroy_all_ordered_extents(fs_info
);
4705 btrfs_destroy_delayed_inodes(fs_info
);
4706 btrfs_assert_delayed_root_empty(fs_info
);
4707 btrfs_destroy_all_delalloc_inodes(fs_info
);
4708 btrfs_drop_all_logs(fs_info
);
4709 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
4714 static const struct extent_io_ops btree_extent_io_ops
= {
4715 /* mandatory callbacks */
4716 .submit_bio_hook
= btree_submit_bio_hook
,
4717 .readpage_end_io_hook
= btree_readpage_end_io_hook
,