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
, *seed_devs
;
549 u8 fsid
[BTRFS_FSID_SIZE
];
552 read_extent_buffer(eb
, fsid
, offsetof(struct btrfs_header
, fsid
),
555 * Checking the incompat flag is only valid for the current fs. For
556 * seed devices it's forbidden to have their uuid changed so reading
557 * ->fsid in this case is fine
559 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
))
560 metadata_uuid
= fs_devices
->metadata_uuid
;
562 metadata_uuid
= fs_devices
->fsid
;
564 if (!memcmp(fsid
, metadata_uuid
, BTRFS_FSID_SIZE
))
567 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
)
568 if (!memcmp(fsid
, seed_devs
->fsid
, BTRFS_FSID_SIZE
))
574 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
575 u64 phy_offset
, struct page
*page
,
576 u64 start
, u64 end
, int mirror
)
580 struct extent_buffer
*eb
;
581 struct btrfs_fs_info
*fs_info
;
584 u8 result
[BTRFS_CSUM_SIZE
];
590 eb
= (struct extent_buffer
*)page
->private;
591 fs_info
= eb
->fs_info
;
592 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
594 /* the pending IO might have been the only thing that kept this buffer
595 * in memory. Make sure we have a ref for all this other checks
597 atomic_inc(&eb
->refs
);
599 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
603 eb
->read_mirror
= mirror
;
604 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
609 found_start
= btrfs_header_bytenr(eb
);
610 if (found_start
!= eb
->start
) {
611 btrfs_err_rl(fs_info
, "bad tree block start, want %llu have %llu",
612 eb
->start
, found_start
);
616 if (check_tree_block_fsid(eb
)) {
617 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
622 found_level
= btrfs_header_level(eb
);
623 if (found_level
>= BTRFS_MAX_LEVEL
) {
624 btrfs_err(fs_info
, "bad tree block level %d on %llu",
625 (int)btrfs_header_level(eb
), eb
->start
);
630 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
633 csum_tree_block(eb
, result
);
635 if (memcmp_extent_buffer(eb
, result
, 0, csum_size
)) {
636 u8 val
[BTRFS_CSUM_SIZE
] = { 0 };
638 read_extent_buffer(eb
, &val
, 0, csum_size
);
639 btrfs_warn_rl(fs_info
,
640 "%s checksum verify failed on %llu wanted " CSUM_FMT
" found " CSUM_FMT
" level %d",
641 fs_info
->sb
->s_id
, eb
->start
,
642 CSUM_FMT_VALUE(csum_size
, val
),
643 CSUM_FMT_VALUE(csum_size
, result
),
644 btrfs_header_level(eb
));
650 * If this is a leaf block and it is corrupt, set the corrupt bit so
651 * that we don't try and read the other copies of this block, just
654 if (found_level
== 0 && btrfs_check_leaf_full(eb
)) {
655 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
659 if (found_level
> 0 && btrfs_check_node(eb
))
663 set_extent_buffer_uptodate(eb
);
666 "block=%llu read time tree block corruption detected",
670 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
671 btree_readahead_hook(eb
, ret
);
675 * our io error hook is going to dec the io pages
676 * again, we have to make sure it has something
679 atomic_inc(&eb
->io_pages
);
680 clear_extent_buffer_uptodate(eb
);
682 free_extent_buffer(eb
);
687 static void end_workqueue_bio(struct bio
*bio
)
689 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
690 struct btrfs_fs_info
*fs_info
;
691 struct btrfs_workqueue
*wq
;
693 fs_info
= end_io_wq
->info
;
694 end_io_wq
->status
= bio
->bi_status
;
696 if (bio_op(bio
) == REQ_OP_WRITE
) {
697 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
698 wq
= fs_info
->endio_meta_write_workers
;
699 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
700 wq
= fs_info
->endio_freespace_worker
;
701 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
702 wq
= fs_info
->endio_raid56_workers
;
704 wq
= fs_info
->endio_write_workers
;
706 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
707 wq
= fs_info
->endio_raid56_workers
;
708 else if (end_io_wq
->metadata
)
709 wq
= fs_info
->endio_meta_workers
;
711 wq
= fs_info
->endio_workers
;
714 btrfs_init_work(&end_io_wq
->work
, end_workqueue_fn
, NULL
, NULL
);
715 btrfs_queue_work(wq
, &end_io_wq
->work
);
718 blk_status_t
btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
719 enum btrfs_wq_endio_type metadata
)
721 struct btrfs_end_io_wq
*end_io_wq
;
723 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
725 return BLK_STS_RESOURCE
;
727 end_io_wq
->private = bio
->bi_private
;
728 end_io_wq
->end_io
= bio
->bi_end_io
;
729 end_io_wq
->info
= info
;
730 end_io_wq
->status
= 0;
731 end_io_wq
->bio
= bio
;
732 end_io_wq
->metadata
= metadata
;
734 bio
->bi_private
= end_io_wq
;
735 bio
->bi_end_io
= end_workqueue_bio
;
739 static void run_one_async_start(struct btrfs_work
*work
)
741 struct async_submit_bio
*async
;
744 async
= container_of(work
, struct async_submit_bio
, work
);
745 ret
= async
->submit_bio_start(async
->private_data
, async
->bio
,
752 * In order to insert checksums into the metadata in large chunks, we wait
753 * until bio submission time. All the pages in the bio are checksummed and
754 * sums are attached onto the ordered extent record.
756 * At IO completion time the csums attached on the ordered extent record are
757 * inserted into the tree.
759 static void run_one_async_done(struct btrfs_work
*work
)
761 struct async_submit_bio
*async
;
765 async
= container_of(work
, struct async_submit_bio
, work
);
766 inode
= async
->private_data
;
768 /* If an error occurred we just want to clean up the bio and move on */
770 async
->bio
->bi_status
= async
->status
;
771 bio_endio(async
->bio
);
776 * All of the bios that pass through here are from async helpers.
777 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
778 * This changes nothing when cgroups aren't in use.
780 async
->bio
->bi_opf
|= REQ_CGROUP_PUNT
;
781 ret
= btrfs_map_bio(btrfs_sb(inode
->i_sb
), async
->bio
, async
->mirror_num
);
783 async
->bio
->bi_status
= ret
;
784 bio_endio(async
->bio
);
788 static void run_one_async_free(struct btrfs_work
*work
)
790 struct async_submit_bio
*async
;
792 async
= container_of(work
, struct async_submit_bio
, work
);
796 blk_status_t
btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
797 int mirror_num
, unsigned long bio_flags
,
798 u64 bio_offset
, void *private_data
,
799 extent_submit_bio_start_t
*submit_bio_start
)
801 struct async_submit_bio
*async
;
803 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
805 return BLK_STS_RESOURCE
;
807 async
->private_data
= private_data
;
809 async
->mirror_num
= mirror_num
;
810 async
->submit_bio_start
= submit_bio_start
;
812 btrfs_init_work(&async
->work
, run_one_async_start
, run_one_async_done
,
815 async
->bio_offset
= bio_offset
;
819 if (op_is_sync(bio
->bi_opf
))
820 btrfs_set_work_high_priority(&async
->work
);
822 btrfs_queue_work(fs_info
->workers
, &async
->work
);
826 static blk_status_t
btree_csum_one_bio(struct bio
*bio
)
828 struct bio_vec
*bvec
;
829 struct btrfs_root
*root
;
831 struct bvec_iter_all iter_all
;
833 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
834 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
835 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
836 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
841 return errno_to_blk_status(ret
);
844 static blk_status_t
btree_submit_bio_start(void *private_data
, struct bio
*bio
,
848 * when we're called for a write, we're already in the async
849 * submission context. Just jump into btrfs_map_bio
851 return btree_csum_one_bio(bio
);
854 static int check_async_write(struct btrfs_fs_info
*fs_info
,
855 struct btrfs_inode
*bi
)
857 if (atomic_read(&bi
->sync_writers
))
859 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST
, &fs_info
->flags
))
864 static blk_status_t
btree_submit_bio_hook(struct inode
*inode
, struct bio
*bio
,
866 unsigned long bio_flags
)
868 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
869 int async
= check_async_write(fs_info
, BTRFS_I(inode
));
872 if (bio_op(bio
) != REQ_OP_WRITE
) {
874 * called for a read, do the setup so that checksum validation
875 * can happen in the async kernel threads
877 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
878 BTRFS_WQ_ENDIO_METADATA
);
881 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
);
883 ret
= btree_csum_one_bio(bio
);
886 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
);
889 * kthread helpers are used to submit writes so that
890 * checksumming can happen in parallel across all CPUs
892 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, 0,
893 0, inode
, btree_submit_bio_start
);
901 bio
->bi_status
= ret
;
906 #ifdef CONFIG_MIGRATION
907 static int btree_migratepage(struct address_space
*mapping
,
908 struct page
*newpage
, struct page
*page
,
909 enum migrate_mode mode
)
912 * we can't safely write a btree page from here,
913 * we haven't done the locking hook
918 * Buffers may be managed in a filesystem specific way.
919 * We must have no buffers or drop them.
921 if (page_has_private(page
) &&
922 !try_to_release_page(page
, GFP_KERNEL
))
924 return migrate_page(mapping
, newpage
, page
, mode
);
929 static int btree_writepages(struct address_space
*mapping
,
930 struct writeback_control
*wbc
)
932 struct btrfs_fs_info
*fs_info
;
935 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
937 if (wbc
->for_kupdate
)
940 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
941 /* this is a bit racy, but that's ok */
942 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
943 BTRFS_DIRTY_METADATA_THRESH
,
944 fs_info
->dirty_metadata_batch
);
948 return btree_write_cache_pages(mapping
, wbc
);
951 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
953 if (PageWriteback(page
) || PageDirty(page
))
956 return try_release_extent_buffer(page
);
959 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
962 struct extent_io_tree
*tree
;
963 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
964 extent_invalidatepage(tree
, page
, offset
);
965 btree_releasepage(page
, GFP_NOFS
);
966 if (PagePrivate(page
)) {
967 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
968 "page private not zero on page %llu",
969 (unsigned long long)page_offset(page
));
970 detach_page_private(page
);
974 static int btree_set_page_dirty(struct page
*page
)
977 struct extent_buffer
*eb
;
979 BUG_ON(!PagePrivate(page
));
980 eb
= (struct extent_buffer
*)page
->private;
982 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
983 BUG_ON(!atomic_read(&eb
->refs
));
984 btrfs_assert_tree_locked(eb
);
986 return __set_page_dirty_nobuffers(page
);
989 static const struct address_space_operations btree_aops
= {
990 .writepages
= btree_writepages
,
991 .releasepage
= btree_releasepage
,
992 .invalidatepage
= btree_invalidatepage
,
993 #ifdef CONFIG_MIGRATION
994 .migratepage
= btree_migratepage
,
996 .set_page_dirty
= btree_set_page_dirty
,
999 void readahead_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
1001 struct extent_buffer
*buf
= NULL
;
1004 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1008 ret
= read_extent_buffer_pages(buf
, WAIT_NONE
, 0);
1010 free_extent_buffer_stale(buf
);
1012 free_extent_buffer(buf
);
1015 struct extent_buffer
*btrfs_find_create_tree_block(
1016 struct btrfs_fs_info
*fs_info
,
1019 if (btrfs_is_testing(fs_info
))
1020 return alloc_test_extent_buffer(fs_info
, bytenr
);
1021 return alloc_extent_buffer(fs_info
, bytenr
);
1025 * Read tree block at logical address @bytenr and do variant basic but critical
1028 * @parent_transid: expected transid of this tree block, skip check if 0
1029 * @level: expected level, mandatory check
1030 * @first_key: expected key in slot 0, skip check if NULL
1032 struct extent_buffer
*read_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1033 u64 parent_transid
, int level
,
1034 struct btrfs_key
*first_key
)
1036 struct extent_buffer
*buf
= NULL
;
1039 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1043 ret
= btree_read_extent_buffer_pages(buf
, parent_transid
,
1046 free_extent_buffer_stale(buf
);
1047 return ERR_PTR(ret
);
1053 void btrfs_clean_tree_block(struct extent_buffer
*buf
)
1055 struct btrfs_fs_info
*fs_info
= buf
->fs_info
;
1056 if (btrfs_header_generation(buf
) ==
1057 fs_info
->running_transaction
->transid
) {
1058 btrfs_assert_tree_locked(buf
);
1060 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1061 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
1063 fs_info
->dirty_metadata_batch
);
1064 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1065 btrfs_set_lock_blocking_write(buf
);
1066 clear_extent_buffer_dirty(buf
);
1071 static void __setup_root(struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1074 bool dummy
= test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO
, &fs_info
->fs_state
);
1075 root
->fs_info
= fs_info
;
1077 root
->commit_root
= NULL
;
1079 root
->orphan_cleanup_state
= 0;
1081 root
->last_trans
= 0;
1082 root
->highest_objectid
= 0;
1083 root
->nr_delalloc_inodes
= 0;
1084 root
->nr_ordered_extents
= 0;
1085 root
->inode_tree
= RB_ROOT
;
1086 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1087 root
->block_rsv
= NULL
;
1089 INIT_LIST_HEAD(&root
->dirty_list
);
1090 INIT_LIST_HEAD(&root
->root_list
);
1091 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1092 INIT_LIST_HEAD(&root
->delalloc_root
);
1093 INIT_LIST_HEAD(&root
->ordered_extents
);
1094 INIT_LIST_HEAD(&root
->ordered_root
);
1095 INIT_LIST_HEAD(&root
->reloc_dirty_list
);
1096 INIT_LIST_HEAD(&root
->logged_list
[0]);
1097 INIT_LIST_HEAD(&root
->logged_list
[1]);
1098 spin_lock_init(&root
->inode_lock
);
1099 spin_lock_init(&root
->delalloc_lock
);
1100 spin_lock_init(&root
->ordered_extent_lock
);
1101 spin_lock_init(&root
->accounting_lock
);
1102 spin_lock_init(&root
->log_extents_lock
[0]);
1103 spin_lock_init(&root
->log_extents_lock
[1]);
1104 spin_lock_init(&root
->qgroup_meta_rsv_lock
);
1105 mutex_init(&root
->objectid_mutex
);
1106 mutex_init(&root
->log_mutex
);
1107 mutex_init(&root
->ordered_extent_mutex
);
1108 mutex_init(&root
->delalloc_mutex
);
1109 init_waitqueue_head(&root
->qgroup_flush_wait
);
1110 init_waitqueue_head(&root
->log_writer_wait
);
1111 init_waitqueue_head(&root
->log_commit_wait
[0]);
1112 init_waitqueue_head(&root
->log_commit_wait
[1]);
1113 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1114 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1115 atomic_set(&root
->log_commit
[0], 0);
1116 atomic_set(&root
->log_commit
[1], 0);
1117 atomic_set(&root
->log_writers
, 0);
1118 atomic_set(&root
->log_batch
, 0);
1119 refcount_set(&root
->refs
, 1);
1120 atomic_set(&root
->snapshot_force_cow
, 0);
1121 atomic_set(&root
->nr_swapfiles
, 0);
1122 root
->log_transid
= 0;
1123 root
->log_transid_committed
= -1;
1124 root
->last_log_commit
= 0;
1126 extent_io_tree_init(fs_info
, &root
->dirty_log_pages
,
1127 IO_TREE_ROOT_DIRTY_LOG_PAGES
, NULL
);
1128 extent_io_tree_init(fs_info
, &root
->log_csum_range
,
1129 IO_TREE_LOG_CSUM_RANGE
, NULL
);
1132 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1133 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1134 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1135 root
->root_key
.objectid
= objectid
;
1138 spin_lock_init(&root
->root_item_lock
);
1139 btrfs_qgroup_init_swapped_blocks(&root
->swapped_blocks
);
1140 #ifdef CONFIG_BTRFS_DEBUG
1141 INIT_LIST_HEAD(&root
->leak_list
);
1142 spin_lock(&fs_info
->fs_roots_radix_lock
);
1143 list_add_tail(&root
->leak_list
, &fs_info
->allocated_roots
);
1144 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1148 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1149 u64 objectid
, gfp_t flags
)
1151 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1153 __setup_root(root
, fs_info
, objectid
);
1157 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1158 /* Should only be used by the testing infrastructure */
1159 struct btrfs_root
*btrfs_alloc_dummy_root(struct btrfs_fs_info
*fs_info
)
1161 struct btrfs_root
*root
;
1164 return ERR_PTR(-EINVAL
);
1166 root
= btrfs_alloc_root(fs_info
, BTRFS_ROOT_TREE_OBJECTID
, GFP_KERNEL
);
1168 return ERR_PTR(-ENOMEM
);
1170 /* We don't use the stripesize in selftest, set it as sectorsize */
1171 root
->alloc_bytenr
= 0;
1177 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1180 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1181 struct extent_buffer
*leaf
;
1182 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1183 struct btrfs_root
*root
;
1184 struct btrfs_key key
;
1185 unsigned int nofs_flag
;
1189 * We're holding a transaction handle, so use a NOFS memory allocation
1190 * context to avoid deadlock if reclaim happens.
1192 nofs_flag
= memalloc_nofs_save();
1193 root
= btrfs_alloc_root(fs_info
, objectid
, GFP_KERNEL
);
1194 memalloc_nofs_restore(nofs_flag
);
1196 return ERR_PTR(-ENOMEM
);
1198 root
->root_key
.objectid
= objectid
;
1199 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1200 root
->root_key
.offset
= 0;
1202 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0,
1203 BTRFS_NESTING_NORMAL
);
1205 ret
= PTR_ERR(leaf
);
1211 btrfs_mark_buffer_dirty(leaf
);
1213 root
->commit_root
= btrfs_root_node(root
);
1214 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1216 root
->root_item
.flags
= 0;
1217 root
->root_item
.byte_limit
= 0;
1218 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1219 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1220 btrfs_set_root_level(&root
->root_item
, 0);
1221 btrfs_set_root_refs(&root
->root_item
, 1);
1222 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1223 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1224 btrfs_set_root_dirid(&root
->root_item
, 0);
1225 if (is_fstree(objectid
))
1226 generate_random_guid(root
->root_item
.uuid
);
1228 export_guid(root
->root_item
.uuid
, &guid_null
);
1229 root
->root_item
.drop_level
= 0;
1231 key
.objectid
= objectid
;
1232 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1234 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1238 btrfs_tree_unlock(leaf
);
1244 btrfs_tree_unlock(leaf
);
1245 btrfs_put_root(root
);
1247 return ERR_PTR(ret
);
1250 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1251 struct btrfs_fs_info
*fs_info
)
1253 struct btrfs_root
*root
;
1254 struct extent_buffer
*leaf
;
1256 root
= btrfs_alloc_root(fs_info
, BTRFS_TREE_LOG_OBJECTID
, GFP_NOFS
);
1258 return ERR_PTR(-ENOMEM
);
1260 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1261 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1262 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1265 * DON'T set SHAREABLE bit for log trees.
1267 * Log trees are not exposed to user space thus can't be snapshotted,
1268 * and they go away before a real commit is actually done.
1270 * They do store pointers to file data extents, and those reference
1271 * counts still get updated (along with back refs to the log tree).
1274 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1275 NULL
, 0, 0, 0, BTRFS_NESTING_NORMAL
);
1277 btrfs_put_root(root
);
1278 return ERR_CAST(leaf
);
1283 btrfs_mark_buffer_dirty(root
->node
);
1284 btrfs_tree_unlock(root
->node
);
1288 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1289 struct btrfs_fs_info
*fs_info
)
1291 struct btrfs_root
*log_root
;
1293 log_root
= alloc_log_tree(trans
, fs_info
);
1294 if (IS_ERR(log_root
))
1295 return PTR_ERR(log_root
);
1296 WARN_ON(fs_info
->log_root_tree
);
1297 fs_info
->log_root_tree
= log_root
;
1301 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1302 struct btrfs_root
*root
)
1304 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1305 struct btrfs_root
*log_root
;
1306 struct btrfs_inode_item
*inode_item
;
1308 log_root
= alloc_log_tree(trans
, fs_info
);
1309 if (IS_ERR(log_root
))
1310 return PTR_ERR(log_root
);
1312 log_root
->last_trans
= trans
->transid
;
1313 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1315 inode_item
= &log_root
->root_item
.inode
;
1316 btrfs_set_stack_inode_generation(inode_item
, 1);
1317 btrfs_set_stack_inode_size(inode_item
, 3);
1318 btrfs_set_stack_inode_nlink(inode_item
, 1);
1319 btrfs_set_stack_inode_nbytes(inode_item
,
1321 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1323 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1325 WARN_ON(root
->log_root
);
1326 root
->log_root
= log_root
;
1327 root
->log_transid
= 0;
1328 root
->log_transid_committed
= -1;
1329 root
->last_log_commit
= 0;
1333 struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1334 struct btrfs_key
*key
)
1336 struct btrfs_root
*root
;
1337 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1338 struct btrfs_path
*path
;
1343 path
= btrfs_alloc_path();
1345 return ERR_PTR(-ENOMEM
);
1347 root
= btrfs_alloc_root(fs_info
, key
->objectid
, GFP_NOFS
);
1353 ret
= btrfs_find_root(tree_root
, key
, path
,
1354 &root
->root_item
, &root
->root_key
);
1361 generation
= btrfs_root_generation(&root
->root_item
);
1362 level
= btrfs_root_level(&root
->root_item
);
1363 root
->node
= read_tree_block(fs_info
,
1364 btrfs_root_bytenr(&root
->root_item
),
1365 generation
, level
, NULL
);
1366 if (IS_ERR(root
->node
)) {
1367 ret
= PTR_ERR(root
->node
);
1370 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1374 root
->commit_root
= btrfs_root_node(root
);
1376 btrfs_free_path(path
);
1380 btrfs_put_root(root
);
1382 root
= ERR_PTR(ret
);
1387 * Initialize subvolume root in-memory structure
1389 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1391 static int btrfs_init_fs_root(struct btrfs_root
*root
, dev_t anon_dev
)
1394 unsigned int nofs_flag
;
1396 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1397 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1399 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1405 * We might be called under a transaction (e.g. indirect backref
1406 * resolution) which could deadlock if it triggers memory reclaim
1408 nofs_flag
= memalloc_nofs_save();
1409 ret
= btrfs_drew_lock_init(&root
->snapshot_lock
);
1410 memalloc_nofs_restore(nofs_flag
);
1414 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
&&
1415 root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1416 set_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
);
1417 btrfs_check_and_init_root_item(&root
->root_item
);
1420 btrfs_init_free_ino_ctl(root
);
1421 spin_lock_init(&root
->ino_cache_lock
);
1422 init_waitqueue_head(&root
->ino_cache_wait
);
1425 * Don't assign anonymous block device to roots that are not exposed to
1426 * userspace, the id pool is limited to 1M
1428 if (is_fstree(root
->root_key
.objectid
) &&
1429 btrfs_root_refs(&root
->root_item
) > 0) {
1431 ret
= get_anon_bdev(&root
->anon_dev
);
1435 root
->anon_dev
= anon_dev
;
1439 mutex_lock(&root
->objectid_mutex
);
1440 ret
= btrfs_find_highest_objectid(root
,
1441 &root
->highest_objectid
);
1443 mutex_unlock(&root
->objectid_mutex
);
1447 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1449 mutex_unlock(&root
->objectid_mutex
);
1453 /* The caller is responsible to call btrfs_free_fs_root */
1457 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1460 struct btrfs_root
*root
;
1462 spin_lock(&fs_info
->fs_roots_radix_lock
);
1463 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1464 (unsigned long)root_id
);
1466 root
= btrfs_grab_root(root
);
1467 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1471 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1472 struct btrfs_root
*root
)
1476 ret
= radix_tree_preload(GFP_NOFS
);
1480 spin_lock(&fs_info
->fs_roots_radix_lock
);
1481 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1482 (unsigned long)root
->root_key
.objectid
,
1485 btrfs_grab_root(root
);
1486 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1488 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1489 radix_tree_preload_end();
1494 void btrfs_check_leaked_roots(struct btrfs_fs_info
*fs_info
)
1496 #ifdef CONFIG_BTRFS_DEBUG
1497 struct btrfs_root
*root
;
1499 while (!list_empty(&fs_info
->allocated_roots
)) {
1500 char buf
[BTRFS_ROOT_NAME_BUF_LEN
];
1502 root
= list_first_entry(&fs_info
->allocated_roots
,
1503 struct btrfs_root
, leak_list
);
1504 btrfs_err(fs_info
, "leaked root %s refcount %d",
1505 btrfs_root_name(root
->root_key
.objectid
, buf
),
1506 refcount_read(&root
->refs
));
1507 while (refcount_read(&root
->refs
) > 1)
1508 btrfs_put_root(root
);
1509 btrfs_put_root(root
);
1514 void btrfs_free_fs_info(struct btrfs_fs_info
*fs_info
)
1516 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
1517 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
1518 percpu_counter_destroy(&fs_info
->dio_bytes
);
1519 percpu_counter_destroy(&fs_info
->dev_replace
.bio_counter
);
1520 btrfs_free_csum_hash(fs_info
);
1521 btrfs_free_stripe_hash_table(fs_info
);
1522 btrfs_free_ref_cache(fs_info
);
1523 kfree(fs_info
->balance_ctl
);
1524 kfree(fs_info
->delayed_root
);
1525 btrfs_put_root(fs_info
->extent_root
);
1526 btrfs_put_root(fs_info
->tree_root
);
1527 btrfs_put_root(fs_info
->chunk_root
);
1528 btrfs_put_root(fs_info
->dev_root
);
1529 btrfs_put_root(fs_info
->csum_root
);
1530 btrfs_put_root(fs_info
->quota_root
);
1531 btrfs_put_root(fs_info
->uuid_root
);
1532 btrfs_put_root(fs_info
->free_space_root
);
1533 btrfs_put_root(fs_info
->fs_root
);
1534 btrfs_put_root(fs_info
->data_reloc_root
);
1535 btrfs_check_leaked_roots(fs_info
);
1536 btrfs_extent_buffer_leak_debug_check(fs_info
);
1537 kfree(fs_info
->super_copy
);
1538 kfree(fs_info
->super_for_commit
);
1544 * Get an in-memory reference of a root structure.
1546 * For essential trees like root/extent tree, we grab it from fs_info directly.
1547 * For subvolume trees, we check the cached filesystem roots first. If not
1548 * found, then read it from disk and add it to cached fs roots.
1550 * Caller should release the root by calling btrfs_put_root() after the usage.
1552 * NOTE: Reloc and log trees can't be read by this function as they share the
1553 * same root objectid.
1555 * @objectid: root id
1556 * @anon_dev: preallocated anonymous block device number for new roots,
1557 * pass 0 for new allocation.
1558 * @check_ref: whether to check root item references, If true, return -ENOENT
1561 static struct btrfs_root
*btrfs_get_root_ref(struct btrfs_fs_info
*fs_info
,
1562 u64 objectid
, dev_t anon_dev
,
1565 struct btrfs_root
*root
;
1566 struct btrfs_path
*path
;
1567 struct btrfs_key key
;
1570 if (objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1571 return btrfs_grab_root(fs_info
->tree_root
);
1572 if (objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1573 return btrfs_grab_root(fs_info
->extent_root
);
1574 if (objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1575 return btrfs_grab_root(fs_info
->chunk_root
);
1576 if (objectid
== BTRFS_DEV_TREE_OBJECTID
)
1577 return btrfs_grab_root(fs_info
->dev_root
);
1578 if (objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1579 return btrfs_grab_root(fs_info
->csum_root
);
1580 if (objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1581 return btrfs_grab_root(fs_info
->quota_root
) ?
1582 fs_info
->quota_root
: ERR_PTR(-ENOENT
);
1583 if (objectid
== BTRFS_UUID_TREE_OBJECTID
)
1584 return btrfs_grab_root(fs_info
->uuid_root
) ?
1585 fs_info
->uuid_root
: ERR_PTR(-ENOENT
);
1586 if (objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1587 return btrfs_grab_root(fs_info
->free_space_root
) ?
1588 fs_info
->free_space_root
: ERR_PTR(-ENOENT
);
1590 root
= btrfs_lookup_fs_root(fs_info
, objectid
);
1592 /* Shouldn't get preallocated anon_dev for cached roots */
1594 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1595 btrfs_put_root(root
);
1596 return ERR_PTR(-ENOENT
);
1601 key
.objectid
= objectid
;
1602 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1603 key
.offset
= (u64
)-1;
1604 root
= btrfs_read_tree_root(fs_info
->tree_root
, &key
);
1608 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1613 ret
= btrfs_init_fs_root(root
, anon_dev
);
1617 path
= btrfs_alloc_path();
1622 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1623 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1624 key
.offset
= objectid
;
1626 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1627 btrfs_free_path(path
);
1631 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1633 ret
= btrfs_insert_fs_root(fs_info
, root
);
1635 btrfs_put_root(root
);
1642 btrfs_put_root(root
);
1643 return ERR_PTR(ret
);
1647 * Get in-memory reference of a root structure
1649 * @objectid: tree objectid
1650 * @check_ref: if set, verify that the tree exists and the item has at least
1653 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1654 u64 objectid
, bool check_ref
)
1656 return btrfs_get_root_ref(fs_info
, objectid
, 0, check_ref
);
1660 * Get in-memory reference of a root structure, created as new, optionally pass
1661 * the anonymous block device id
1663 * @objectid: tree objectid
1664 * @anon_dev: if zero, allocate a new anonymous block device or use the
1667 struct btrfs_root
*btrfs_get_new_fs_root(struct btrfs_fs_info
*fs_info
,
1668 u64 objectid
, dev_t anon_dev
)
1670 return btrfs_get_root_ref(fs_info
, objectid
, anon_dev
, true);
1674 * called by the kthread helper functions to finally call the bio end_io
1675 * functions. This is where read checksum verification actually happens
1677 static void end_workqueue_fn(struct btrfs_work
*work
)
1680 struct btrfs_end_io_wq
*end_io_wq
;
1682 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1683 bio
= end_io_wq
->bio
;
1685 bio
->bi_status
= end_io_wq
->status
;
1686 bio
->bi_private
= end_io_wq
->private;
1687 bio
->bi_end_io
= end_io_wq
->end_io
;
1689 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1692 static int cleaner_kthread(void *arg
)
1694 struct btrfs_root
*root
= arg
;
1695 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1701 set_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1703 /* Make the cleaner go to sleep early. */
1704 if (btrfs_need_cleaner_sleep(fs_info
))
1708 * Do not do anything if we might cause open_ctree() to block
1709 * before we have finished mounting the filesystem.
1711 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1714 if (!mutex_trylock(&fs_info
->cleaner_mutex
))
1718 * Avoid the problem that we change the status of the fs
1719 * during the above check and trylock.
1721 if (btrfs_need_cleaner_sleep(fs_info
)) {
1722 mutex_unlock(&fs_info
->cleaner_mutex
);
1726 btrfs_run_delayed_iputs(fs_info
);
1728 again
= btrfs_clean_one_deleted_snapshot(root
);
1729 mutex_unlock(&fs_info
->cleaner_mutex
);
1732 * The defragger has dealt with the R/O remount and umount,
1733 * needn't do anything special here.
1735 btrfs_run_defrag_inodes(fs_info
);
1738 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1739 * with relocation (btrfs_relocate_chunk) and relocation
1740 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1741 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1742 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1743 * unused block groups.
1745 btrfs_delete_unused_bgs(fs_info
);
1747 clear_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1748 if (kthread_should_park())
1750 if (kthread_should_stop())
1753 set_current_state(TASK_INTERRUPTIBLE
);
1755 __set_current_state(TASK_RUNNING
);
1760 static int transaction_kthread(void *arg
)
1762 struct btrfs_root
*root
= arg
;
1763 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1764 struct btrfs_trans_handle
*trans
;
1765 struct btrfs_transaction
*cur
;
1768 unsigned long delay
;
1772 cannot_commit
= false;
1773 delay
= HZ
* fs_info
->commit_interval
;
1774 mutex_lock(&fs_info
->transaction_kthread_mutex
);
1776 spin_lock(&fs_info
->trans_lock
);
1777 cur
= fs_info
->running_transaction
;
1779 spin_unlock(&fs_info
->trans_lock
);
1783 now
= ktime_get_seconds();
1784 if (cur
->state
< TRANS_STATE_COMMIT_START
&&
1785 (now
< cur
->start_time
||
1786 now
- cur
->start_time
< fs_info
->commit_interval
)) {
1787 spin_unlock(&fs_info
->trans_lock
);
1791 transid
= cur
->transid
;
1792 spin_unlock(&fs_info
->trans_lock
);
1794 /* If the file system is aborted, this will always fail. */
1795 trans
= btrfs_attach_transaction(root
);
1796 if (IS_ERR(trans
)) {
1797 if (PTR_ERR(trans
) != -ENOENT
)
1798 cannot_commit
= true;
1801 if (transid
== trans
->transid
) {
1802 btrfs_commit_transaction(trans
);
1804 btrfs_end_transaction(trans
);
1807 wake_up_process(fs_info
->cleaner_kthread
);
1808 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
1810 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1811 &fs_info
->fs_state
)))
1812 btrfs_cleanup_transaction(fs_info
);
1813 if (!kthread_should_stop() &&
1814 (!btrfs_transaction_blocked(fs_info
) ||
1816 schedule_timeout_interruptible(delay
);
1817 } while (!kthread_should_stop());
1822 * This will find the highest generation in the array of root backups. The
1823 * index of the highest array is returned, or -EINVAL if we can't find
1826 * We check to make sure the array is valid by comparing the
1827 * generation of the latest root in the array with the generation
1828 * in the super block. If they don't match we pitch it.
1830 static int find_newest_super_backup(struct btrfs_fs_info
*info
)
1832 const u64 newest_gen
= btrfs_super_generation(info
->super_copy
);
1834 struct btrfs_root_backup
*root_backup
;
1837 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1838 root_backup
= info
->super_copy
->super_roots
+ i
;
1839 cur
= btrfs_backup_tree_root_gen(root_backup
);
1840 if (cur
== newest_gen
)
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1852 static void backup_super_roots(struct btrfs_fs_info
*info
)
1854 const int next_backup
= info
->backup_root_index
;
1855 struct btrfs_root_backup
*root_backup
;
1857 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1860 * make sure all of our padding and empty slots get zero filled
1861 * regardless of which ones we use today
1863 memset(root_backup
, 0, sizeof(*root_backup
));
1865 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1867 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1868 btrfs_set_backup_tree_root_gen(root_backup
,
1869 btrfs_header_generation(info
->tree_root
->node
));
1871 btrfs_set_backup_tree_root_level(root_backup
,
1872 btrfs_header_level(info
->tree_root
->node
));
1874 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1875 btrfs_set_backup_chunk_root_gen(root_backup
,
1876 btrfs_header_generation(info
->chunk_root
->node
));
1877 btrfs_set_backup_chunk_root_level(root_backup
,
1878 btrfs_header_level(info
->chunk_root
->node
));
1880 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1881 btrfs_set_backup_extent_root_gen(root_backup
,
1882 btrfs_header_generation(info
->extent_root
->node
));
1883 btrfs_set_backup_extent_root_level(root_backup
,
1884 btrfs_header_level(info
->extent_root
->node
));
1887 * we might commit during log recovery, which happens before we set
1888 * the fs_root. Make sure it is valid before we fill it in.
1890 if (info
->fs_root
&& info
->fs_root
->node
) {
1891 btrfs_set_backup_fs_root(root_backup
,
1892 info
->fs_root
->node
->start
);
1893 btrfs_set_backup_fs_root_gen(root_backup
,
1894 btrfs_header_generation(info
->fs_root
->node
));
1895 btrfs_set_backup_fs_root_level(root_backup
,
1896 btrfs_header_level(info
->fs_root
->node
));
1899 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1900 btrfs_set_backup_dev_root_gen(root_backup
,
1901 btrfs_header_generation(info
->dev_root
->node
));
1902 btrfs_set_backup_dev_root_level(root_backup
,
1903 btrfs_header_level(info
->dev_root
->node
));
1905 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1906 btrfs_set_backup_csum_root_gen(root_backup
,
1907 btrfs_header_generation(info
->csum_root
->node
));
1908 btrfs_set_backup_csum_root_level(root_backup
,
1909 btrfs_header_level(info
->csum_root
->node
));
1911 btrfs_set_backup_total_bytes(root_backup
,
1912 btrfs_super_total_bytes(info
->super_copy
));
1913 btrfs_set_backup_bytes_used(root_backup
,
1914 btrfs_super_bytes_used(info
->super_copy
));
1915 btrfs_set_backup_num_devices(root_backup
,
1916 btrfs_super_num_devices(info
->super_copy
));
1919 * if we don't copy this out to the super_copy, it won't get remembered
1920 * for the next commit
1922 memcpy(&info
->super_copy
->super_roots
,
1923 &info
->super_for_commit
->super_roots
,
1924 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1928 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1929 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1931 * fs_info - filesystem whose backup roots need to be read
1932 * priority - priority of backup root required
1934 * Returns backup root index on success and -EINVAL otherwise.
1936 static int read_backup_root(struct btrfs_fs_info
*fs_info
, u8 priority
)
1938 int backup_index
= find_newest_super_backup(fs_info
);
1939 struct btrfs_super_block
*super
= fs_info
->super_copy
;
1940 struct btrfs_root_backup
*root_backup
;
1942 if (priority
< BTRFS_NUM_BACKUP_ROOTS
&& backup_index
>= 0) {
1944 return backup_index
;
1946 backup_index
= backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- priority
;
1947 backup_index
%= BTRFS_NUM_BACKUP_ROOTS
;
1952 root_backup
= super
->super_roots
+ backup_index
;
1954 btrfs_set_super_generation(super
,
1955 btrfs_backup_tree_root_gen(root_backup
));
1956 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1957 btrfs_set_super_root_level(super
,
1958 btrfs_backup_tree_root_level(root_backup
));
1959 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1962 * Fixme: the total bytes and num_devices need to match or we should
1965 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1966 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1968 return backup_index
;
1971 /* helper to cleanup workers */
1972 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
1974 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
1975 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
1976 btrfs_destroy_workqueue(fs_info
->workers
);
1977 btrfs_destroy_workqueue(fs_info
->endio_workers
);
1978 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
1979 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
1980 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
1981 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
1982 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
1983 btrfs_destroy_workqueue(fs_info
->caching_workers
);
1984 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
1985 btrfs_destroy_workqueue(fs_info
->flush_workers
);
1986 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
1987 if (fs_info
->discard_ctl
.discard_workers
)
1988 destroy_workqueue(fs_info
->discard_ctl
.discard_workers
);
1990 * Now that all other work queues are destroyed, we can safely destroy
1991 * the queues used for metadata I/O, since tasks from those other work
1992 * queues can do metadata I/O operations.
1994 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
1995 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
1998 static void free_root_extent_buffers(struct btrfs_root
*root
)
2001 free_extent_buffer(root
->node
);
2002 free_extent_buffer(root
->commit_root
);
2004 root
->commit_root
= NULL
;
2008 /* helper to cleanup tree roots */
2009 static void free_root_pointers(struct btrfs_fs_info
*info
, bool free_chunk_root
)
2011 free_root_extent_buffers(info
->tree_root
);
2013 free_root_extent_buffers(info
->dev_root
);
2014 free_root_extent_buffers(info
->extent_root
);
2015 free_root_extent_buffers(info
->csum_root
);
2016 free_root_extent_buffers(info
->quota_root
);
2017 free_root_extent_buffers(info
->uuid_root
);
2018 free_root_extent_buffers(info
->fs_root
);
2019 free_root_extent_buffers(info
->data_reloc_root
);
2020 if (free_chunk_root
)
2021 free_root_extent_buffers(info
->chunk_root
);
2022 free_root_extent_buffers(info
->free_space_root
);
2025 void btrfs_put_root(struct btrfs_root
*root
)
2030 if (refcount_dec_and_test(&root
->refs
)) {
2031 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
2032 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE
, &root
->state
));
2034 free_anon_bdev(root
->anon_dev
);
2035 btrfs_drew_lock_destroy(&root
->snapshot_lock
);
2036 free_root_extent_buffers(root
);
2037 kfree(root
->free_ino_ctl
);
2038 kfree(root
->free_ino_pinned
);
2039 #ifdef CONFIG_BTRFS_DEBUG
2040 spin_lock(&root
->fs_info
->fs_roots_radix_lock
);
2041 list_del_init(&root
->leak_list
);
2042 spin_unlock(&root
->fs_info
->fs_roots_radix_lock
);
2048 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2051 struct btrfs_root
*gang
[8];
2054 while (!list_empty(&fs_info
->dead_roots
)) {
2055 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2056 struct btrfs_root
, root_list
);
2057 list_del(&gang
[0]->root_list
);
2059 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
))
2060 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2061 btrfs_put_root(gang
[0]);
2065 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2070 for (i
= 0; i
< ret
; i
++)
2071 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2075 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2077 mutex_init(&fs_info
->scrub_lock
);
2078 atomic_set(&fs_info
->scrubs_running
, 0);
2079 atomic_set(&fs_info
->scrub_pause_req
, 0);
2080 atomic_set(&fs_info
->scrubs_paused
, 0);
2081 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2082 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2083 refcount_set(&fs_info
->scrub_workers_refcnt
, 0);
2086 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2088 spin_lock_init(&fs_info
->balance_lock
);
2089 mutex_init(&fs_info
->balance_mutex
);
2090 atomic_set(&fs_info
->balance_pause_req
, 0);
2091 atomic_set(&fs_info
->balance_cancel_req
, 0);
2092 fs_info
->balance_ctl
= NULL
;
2093 init_waitqueue_head(&fs_info
->balance_wait_q
);
2096 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
)
2098 struct inode
*inode
= fs_info
->btree_inode
;
2100 inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2101 set_nlink(inode
, 1);
2103 * we set the i_size on the btree inode to the max possible int.
2104 * the real end of the address space is determined by all of
2105 * the devices in the system
2107 inode
->i_size
= OFFSET_MAX
;
2108 inode
->i_mapping
->a_ops
= &btree_aops
;
2110 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
2111 extent_io_tree_init(fs_info
, &BTRFS_I(inode
)->io_tree
,
2112 IO_TREE_INODE_IO
, inode
);
2113 BTRFS_I(inode
)->io_tree
.track_uptodate
= false;
2114 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
);
2116 BTRFS_I(inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2118 BTRFS_I(inode
)->root
= btrfs_grab_root(fs_info
->tree_root
);
2119 memset(&BTRFS_I(inode
)->location
, 0, sizeof(struct btrfs_key
));
2120 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
2121 btrfs_insert_inode_hash(inode
);
2124 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2126 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2127 init_rwsem(&fs_info
->dev_replace
.rwsem
);
2128 init_waitqueue_head(&fs_info
->dev_replace
.replace_wait
);
2131 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2133 spin_lock_init(&fs_info
->qgroup_lock
);
2134 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2135 fs_info
->qgroup_tree
= RB_ROOT
;
2136 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2137 fs_info
->qgroup_seq
= 1;
2138 fs_info
->qgroup_ulist
= NULL
;
2139 fs_info
->qgroup_rescan_running
= false;
2140 mutex_init(&fs_info
->qgroup_rescan_lock
);
2143 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2144 struct btrfs_fs_devices
*fs_devices
)
2146 u32 max_active
= fs_info
->thread_pool_size
;
2147 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2150 btrfs_alloc_workqueue(fs_info
, "worker",
2151 flags
| WQ_HIGHPRI
, max_active
, 16);
2153 fs_info
->delalloc_workers
=
2154 btrfs_alloc_workqueue(fs_info
, "delalloc",
2155 flags
, max_active
, 2);
2157 fs_info
->flush_workers
=
2158 btrfs_alloc_workqueue(fs_info
, "flush_delalloc",
2159 flags
, max_active
, 0);
2161 fs_info
->caching_workers
=
2162 btrfs_alloc_workqueue(fs_info
, "cache", flags
, max_active
, 0);
2164 fs_info
->fixup_workers
=
2165 btrfs_alloc_workqueue(fs_info
, "fixup", flags
, 1, 0);
2168 * endios are largely parallel and should have a very
2171 fs_info
->endio_workers
=
2172 btrfs_alloc_workqueue(fs_info
, "endio", flags
, max_active
, 4);
2173 fs_info
->endio_meta_workers
=
2174 btrfs_alloc_workqueue(fs_info
, "endio-meta", flags
,
2176 fs_info
->endio_meta_write_workers
=
2177 btrfs_alloc_workqueue(fs_info
, "endio-meta-write", flags
,
2179 fs_info
->endio_raid56_workers
=
2180 btrfs_alloc_workqueue(fs_info
, "endio-raid56", flags
,
2182 fs_info
->rmw_workers
=
2183 btrfs_alloc_workqueue(fs_info
, "rmw", flags
, max_active
, 2);
2184 fs_info
->endio_write_workers
=
2185 btrfs_alloc_workqueue(fs_info
, "endio-write", flags
,
2187 fs_info
->endio_freespace_worker
=
2188 btrfs_alloc_workqueue(fs_info
, "freespace-write", flags
,
2190 fs_info
->delayed_workers
=
2191 btrfs_alloc_workqueue(fs_info
, "delayed-meta", flags
,
2193 fs_info
->readahead_workers
=
2194 btrfs_alloc_workqueue(fs_info
, "readahead", flags
,
2196 fs_info
->qgroup_rescan_workers
=
2197 btrfs_alloc_workqueue(fs_info
, "qgroup-rescan", flags
, 1, 0);
2198 fs_info
->discard_ctl
.discard_workers
=
2199 alloc_workqueue("btrfs_discard", WQ_UNBOUND
| WQ_FREEZABLE
, 1);
2201 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2202 fs_info
->flush_workers
&&
2203 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2204 fs_info
->endio_meta_write_workers
&&
2205 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2206 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2207 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2208 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2209 fs_info
->qgroup_rescan_workers
&&
2210 fs_info
->discard_ctl
.discard_workers
)) {
2217 static int btrfs_init_csum_hash(struct btrfs_fs_info
*fs_info
, u16 csum_type
)
2219 struct crypto_shash
*csum_shash
;
2220 const char *csum_driver
= btrfs_super_csum_driver(csum_type
);
2222 csum_shash
= crypto_alloc_shash(csum_driver
, 0, 0);
2224 if (IS_ERR(csum_shash
)) {
2225 btrfs_err(fs_info
, "error allocating %s hash for checksum",
2227 return PTR_ERR(csum_shash
);
2230 fs_info
->csum_shash
= csum_shash
;
2235 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2236 struct btrfs_fs_devices
*fs_devices
)
2239 struct btrfs_root
*log_tree_root
;
2240 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2241 u64 bytenr
= btrfs_super_log_root(disk_super
);
2242 int level
= btrfs_super_log_root_level(disk_super
);
2244 if (fs_devices
->rw_devices
== 0) {
2245 btrfs_warn(fs_info
, "log replay required on RO media");
2249 log_tree_root
= btrfs_alloc_root(fs_info
, BTRFS_TREE_LOG_OBJECTID
,
2254 log_tree_root
->node
= read_tree_block(fs_info
, bytenr
,
2255 fs_info
->generation
+ 1,
2257 if (IS_ERR(log_tree_root
->node
)) {
2258 btrfs_warn(fs_info
, "failed to read log tree");
2259 ret
= PTR_ERR(log_tree_root
->node
);
2260 log_tree_root
->node
= NULL
;
2261 btrfs_put_root(log_tree_root
);
2263 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2264 btrfs_err(fs_info
, "failed to read log tree");
2265 btrfs_put_root(log_tree_root
);
2268 /* returns with log_tree_root freed on success */
2269 ret
= btrfs_recover_log_trees(log_tree_root
);
2271 btrfs_handle_fs_error(fs_info
, ret
,
2272 "Failed to recover log tree");
2273 btrfs_put_root(log_tree_root
);
2277 if (sb_rdonly(fs_info
->sb
)) {
2278 ret
= btrfs_commit_super(fs_info
);
2286 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
)
2288 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2289 struct btrfs_root
*root
;
2290 struct btrfs_key location
;
2293 BUG_ON(!fs_info
->tree_root
);
2295 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2296 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2297 location
.offset
= 0;
2299 root
= btrfs_read_tree_root(tree_root
, &location
);
2301 ret
= PTR_ERR(root
);
2304 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2305 fs_info
->extent_root
= root
;
2307 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2308 root
= btrfs_read_tree_root(tree_root
, &location
);
2310 ret
= PTR_ERR(root
);
2313 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2314 fs_info
->dev_root
= root
;
2315 btrfs_init_devices_late(fs_info
);
2317 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2318 root
= btrfs_read_tree_root(tree_root
, &location
);
2320 ret
= PTR_ERR(root
);
2323 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2324 fs_info
->csum_root
= root
;
2327 * This tree can share blocks with some other fs tree during relocation
2328 * and we need a proper setup by btrfs_get_fs_root
2330 root
= btrfs_get_fs_root(tree_root
->fs_info
,
2331 BTRFS_DATA_RELOC_TREE_OBJECTID
, true);
2333 ret
= PTR_ERR(root
);
2336 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2337 fs_info
->data_reloc_root
= root
;
2339 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2340 root
= btrfs_read_tree_root(tree_root
, &location
);
2341 if (!IS_ERR(root
)) {
2342 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2343 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2344 fs_info
->quota_root
= root
;
2347 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2348 root
= btrfs_read_tree_root(tree_root
, &location
);
2350 ret
= PTR_ERR(root
);
2354 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2355 fs_info
->uuid_root
= root
;
2358 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2359 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2360 root
= btrfs_read_tree_root(tree_root
, &location
);
2362 ret
= PTR_ERR(root
);
2365 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2366 fs_info
->free_space_root
= root
;
2371 btrfs_warn(fs_info
, "failed to read root (objectid=%llu): %d",
2372 location
.objectid
, ret
);
2377 * Real super block validation
2378 * NOTE: super csum type and incompat features will not be checked here.
2380 * @sb: super block to check
2381 * @mirror_num: the super block number to check its bytenr:
2382 * 0 the primary (1st) sb
2383 * 1, 2 2nd and 3rd backup copy
2384 * -1 skip bytenr check
2386 static int validate_super(struct btrfs_fs_info
*fs_info
,
2387 struct btrfs_super_block
*sb
, int mirror_num
)
2389 u64 nodesize
= btrfs_super_nodesize(sb
);
2390 u64 sectorsize
= btrfs_super_sectorsize(sb
);
2393 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
2394 btrfs_err(fs_info
, "no valid FS found");
2397 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
) {
2398 btrfs_err(fs_info
, "unrecognized or unsupported super flag: %llu",
2399 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
2402 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2403 btrfs_err(fs_info
, "tree_root level too big: %d >= %d",
2404 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
2407 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2408 btrfs_err(fs_info
, "chunk_root level too big: %d >= %d",
2409 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
2412 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2413 btrfs_err(fs_info
, "log_root level too big: %d >= %d",
2414 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
2419 * Check sectorsize and nodesize first, other check will need it.
2420 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2422 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
2423 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2424 btrfs_err(fs_info
, "invalid sectorsize %llu", sectorsize
);
2427 /* Only PAGE SIZE is supported yet */
2428 if (sectorsize
!= PAGE_SIZE
) {
2430 "sectorsize %llu not supported yet, only support %lu",
2431 sectorsize
, PAGE_SIZE
);
2434 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
2435 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2436 btrfs_err(fs_info
, "invalid nodesize %llu", nodesize
);
2439 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
2440 btrfs_err(fs_info
, "invalid leafsize %u, should be %llu",
2441 le32_to_cpu(sb
->__unused_leafsize
), nodesize
);
2445 /* Root alignment check */
2446 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
2447 btrfs_warn(fs_info
, "tree_root block unaligned: %llu",
2448 btrfs_super_root(sb
));
2451 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
2452 btrfs_warn(fs_info
, "chunk_root block unaligned: %llu",
2453 btrfs_super_chunk_root(sb
));
2456 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
2457 btrfs_warn(fs_info
, "log_root block unaligned: %llu",
2458 btrfs_super_log_root(sb
));
2462 if (memcmp(fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
,
2463 BTRFS_FSID_SIZE
) != 0) {
2465 "dev_item UUID does not match metadata fsid: %pU != %pU",
2466 fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
);
2471 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2474 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
2475 btrfs_err(fs_info
, "bytes_used is too small %llu",
2476 btrfs_super_bytes_used(sb
));
2479 if (!is_power_of_2(btrfs_super_stripesize(sb
))) {
2480 btrfs_err(fs_info
, "invalid stripesize %u",
2481 btrfs_super_stripesize(sb
));
2484 if (btrfs_super_num_devices(sb
) > (1UL << 31))
2485 btrfs_warn(fs_info
, "suspicious number of devices: %llu",
2486 btrfs_super_num_devices(sb
));
2487 if (btrfs_super_num_devices(sb
) == 0) {
2488 btrfs_err(fs_info
, "number of devices is 0");
2492 if (mirror_num
>= 0 &&
2493 btrfs_super_bytenr(sb
) != btrfs_sb_offset(mirror_num
)) {
2494 btrfs_err(fs_info
, "super offset mismatch %llu != %u",
2495 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
2500 * Obvious sys_chunk_array corruptions, it must hold at least one key
2503 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
2504 btrfs_err(fs_info
, "system chunk array too big %u > %u",
2505 btrfs_super_sys_array_size(sb
),
2506 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
2509 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
2510 + sizeof(struct btrfs_chunk
)) {
2511 btrfs_err(fs_info
, "system chunk array too small %u < %zu",
2512 btrfs_super_sys_array_size(sb
),
2513 sizeof(struct btrfs_disk_key
)
2514 + sizeof(struct btrfs_chunk
));
2519 * The generation is a global counter, we'll trust it more than the others
2520 * but it's still possible that it's the one that's wrong.
2522 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
2524 "suspicious: generation < chunk_root_generation: %llu < %llu",
2525 btrfs_super_generation(sb
),
2526 btrfs_super_chunk_root_generation(sb
));
2527 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
2528 && btrfs_super_cache_generation(sb
) != (u64
)-1)
2530 "suspicious: generation < cache_generation: %llu < %llu",
2531 btrfs_super_generation(sb
),
2532 btrfs_super_cache_generation(sb
));
2538 * Validation of super block at mount time.
2539 * Some checks already done early at mount time, like csum type and incompat
2540 * flags will be skipped.
2542 static int btrfs_validate_mount_super(struct btrfs_fs_info
*fs_info
)
2544 return validate_super(fs_info
, fs_info
->super_copy
, 0);
2548 * Validation of super block at write time.
2549 * Some checks like bytenr check will be skipped as their values will be
2551 * Extra checks like csum type and incompat flags will be done here.
2553 static int btrfs_validate_write_super(struct btrfs_fs_info
*fs_info
,
2554 struct btrfs_super_block
*sb
)
2558 ret
= validate_super(fs_info
, sb
, -1);
2561 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb
))) {
2563 btrfs_err(fs_info
, "invalid csum type, has %u want %u",
2564 btrfs_super_csum_type(sb
), BTRFS_CSUM_TYPE_CRC32
);
2567 if (btrfs_super_incompat_flags(sb
) & ~BTRFS_FEATURE_INCOMPAT_SUPP
) {
2570 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2571 btrfs_super_incompat_flags(sb
),
2572 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP
);
2578 "super block corruption detected before writing it to disk");
2582 static int __cold
init_tree_roots(struct btrfs_fs_info
*fs_info
)
2584 int backup_index
= find_newest_super_backup(fs_info
);
2585 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
2586 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2587 bool handle_error
= false;
2591 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
2596 if (!IS_ERR(tree_root
->node
))
2597 free_extent_buffer(tree_root
->node
);
2598 tree_root
->node
= NULL
;
2600 if (!btrfs_test_opt(fs_info
, USEBACKUPROOT
))
2603 free_root_pointers(fs_info
, 0);
2606 * Don't use the log in recovery mode, it won't be
2609 btrfs_set_super_log_root(sb
, 0);
2611 /* We can't trust the free space cache either */
2612 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2614 ret
= read_backup_root(fs_info
, i
);
2619 generation
= btrfs_super_generation(sb
);
2620 level
= btrfs_super_root_level(sb
);
2621 tree_root
->node
= read_tree_block(fs_info
, btrfs_super_root(sb
),
2622 generation
, level
, NULL
);
2623 if (IS_ERR(tree_root
->node
)) {
2624 handle_error
= true;
2625 ret
= PTR_ERR(tree_root
->node
);
2626 tree_root
->node
= NULL
;
2627 btrfs_warn(fs_info
, "couldn't read tree root");
2630 } else if (!extent_buffer_uptodate(tree_root
->node
)) {
2631 handle_error
= true;
2633 btrfs_warn(fs_info
, "error while reading tree root");
2637 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2638 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2639 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2642 * No need to hold btrfs_root::objectid_mutex since the fs
2643 * hasn't been fully initialised and we are the only user
2645 ret
= btrfs_find_highest_objectid(tree_root
,
2646 &tree_root
->highest_objectid
);
2648 handle_error
= true;
2652 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2654 ret
= btrfs_read_roots(fs_info
);
2656 handle_error
= true;
2660 /* All successful */
2661 fs_info
->generation
= generation
;
2662 fs_info
->last_trans_committed
= generation
;
2664 /* Always begin writing backup roots after the one being used */
2665 if (backup_index
< 0) {
2666 fs_info
->backup_root_index
= 0;
2668 fs_info
->backup_root_index
= backup_index
+ 1;
2669 fs_info
->backup_root_index
%= BTRFS_NUM_BACKUP_ROOTS
;
2677 void btrfs_init_fs_info(struct btrfs_fs_info
*fs_info
)
2679 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2680 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2681 INIT_LIST_HEAD(&fs_info
->trans_list
);
2682 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2683 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2684 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2685 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2686 spin_lock_init(&fs_info
->delalloc_root_lock
);
2687 spin_lock_init(&fs_info
->trans_lock
);
2688 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2689 spin_lock_init(&fs_info
->delayed_iput_lock
);
2690 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2691 spin_lock_init(&fs_info
->super_lock
);
2692 spin_lock_init(&fs_info
->buffer_lock
);
2693 spin_lock_init(&fs_info
->unused_bgs_lock
);
2694 rwlock_init(&fs_info
->tree_mod_log_lock
);
2695 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2696 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2697 mutex_init(&fs_info
->reloc_mutex
);
2698 mutex_init(&fs_info
->delalloc_root_mutex
);
2699 seqlock_init(&fs_info
->profiles_lock
);
2701 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2702 INIT_LIST_HEAD(&fs_info
->space_info
);
2703 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2704 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2705 #ifdef CONFIG_BTRFS_DEBUG
2706 INIT_LIST_HEAD(&fs_info
->allocated_roots
);
2707 INIT_LIST_HEAD(&fs_info
->allocated_ebs
);
2708 spin_lock_init(&fs_info
->eb_leak_lock
);
2710 extent_map_tree_init(&fs_info
->mapping_tree
);
2711 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2712 BTRFS_BLOCK_RSV_GLOBAL
);
2713 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2714 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2715 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2716 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2717 BTRFS_BLOCK_RSV_DELOPS
);
2718 btrfs_init_block_rsv(&fs_info
->delayed_refs_rsv
,
2719 BTRFS_BLOCK_RSV_DELREFS
);
2721 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2722 atomic_set(&fs_info
->defrag_running
, 0);
2723 atomic_set(&fs_info
->reada_works_cnt
, 0);
2724 atomic_set(&fs_info
->nr_delayed_iputs
, 0);
2725 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2726 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2727 fs_info
->metadata_ratio
= 0;
2728 fs_info
->defrag_inodes
= RB_ROOT
;
2729 atomic64_set(&fs_info
->free_chunk_space
, 0);
2730 fs_info
->tree_mod_log
= RB_ROOT
;
2731 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2732 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2733 /* readahead state */
2734 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2735 spin_lock_init(&fs_info
->reada_lock
);
2736 btrfs_init_ref_verify(fs_info
);
2738 fs_info
->thread_pool_size
= min_t(unsigned long,
2739 num_online_cpus() + 2, 8);
2741 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2742 spin_lock_init(&fs_info
->ordered_root_lock
);
2744 btrfs_init_scrub(fs_info
);
2745 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2746 fs_info
->check_integrity_print_mask
= 0;
2748 btrfs_init_balance(fs_info
);
2749 btrfs_init_async_reclaim_work(fs_info
);
2751 spin_lock_init(&fs_info
->block_group_cache_lock
);
2752 fs_info
->block_group_cache_tree
= RB_ROOT
;
2753 fs_info
->first_logical_byte
= (u64
)-1;
2755 extent_io_tree_init(fs_info
, &fs_info
->excluded_extents
,
2756 IO_TREE_FS_EXCLUDED_EXTENTS
, NULL
);
2757 set_bit(BTRFS_FS_BARRIER
, &fs_info
->flags
);
2759 mutex_init(&fs_info
->ordered_operations_mutex
);
2760 mutex_init(&fs_info
->tree_log_mutex
);
2761 mutex_init(&fs_info
->chunk_mutex
);
2762 mutex_init(&fs_info
->transaction_kthread_mutex
);
2763 mutex_init(&fs_info
->cleaner_mutex
);
2764 mutex_init(&fs_info
->ro_block_group_mutex
);
2765 init_rwsem(&fs_info
->commit_root_sem
);
2766 init_rwsem(&fs_info
->cleanup_work_sem
);
2767 init_rwsem(&fs_info
->subvol_sem
);
2768 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2770 btrfs_init_dev_replace_locks(fs_info
);
2771 btrfs_init_qgroup(fs_info
);
2772 btrfs_discard_init(fs_info
);
2774 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2775 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2777 init_waitqueue_head(&fs_info
->transaction_throttle
);
2778 init_waitqueue_head(&fs_info
->transaction_wait
);
2779 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2780 init_waitqueue_head(&fs_info
->async_submit_wait
);
2781 init_waitqueue_head(&fs_info
->delayed_iputs_wait
);
2783 /* Usable values until the real ones are cached from the superblock */
2784 fs_info
->nodesize
= 4096;
2785 fs_info
->sectorsize
= 4096;
2786 fs_info
->stripesize
= 4096;
2788 spin_lock_init(&fs_info
->swapfile_pins_lock
);
2789 fs_info
->swapfile_pins
= RB_ROOT
;
2791 fs_info
->send_in_progress
= 0;
2794 static int init_mount_fs_info(struct btrfs_fs_info
*fs_info
, struct super_block
*sb
)
2799 sb
->s_blocksize
= BTRFS_BDEV_BLOCKSIZE
;
2800 sb
->s_blocksize_bits
= blksize_bits(BTRFS_BDEV_BLOCKSIZE
);
2802 ret
= percpu_counter_init(&fs_info
->dio_bytes
, 0, GFP_KERNEL
);
2806 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2810 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2811 (1 + ilog2(nr_cpu_ids
));
2813 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2817 ret
= percpu_counter_init(&fs_info
->dev_replace
.bio_counter
, 0,
2822 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2824 if (!fs_info
->delayed_root
)
2826 btrfs_init_delayed_root(fs_info
->delayed_root
);
2828 return btrfs_alloc_stripe_hash_table(fs_info
);
2831 static int btrfs_uuid_rescan_kthread(void *data
)
2833 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
2837 * 1st step is to iterate through the existing UUID tree and
2838 * to delete all entries that contain outdated data.
2839 * 2nd step is to add all missing entries to the UUID tree.
2841 ret
= btrfs_uuid_tree_iterate(fs_info
);
2844 btrfs_warn(fs_info
, "iterating uuid_tree failed %d",
2846 up(&fs_info
->uuid_tree_rescan_sem
);
2849 return btrfs_uuid_scan_kthread(data
);
2852 static int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
2854 struct task_struct
*task
;
2856 down(&fs_info
->uuid_tree_rescan_sem
);
2857 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
2859 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2860 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
2861 up(&fs_info
->uuid_tree_rescan_sem
);
2862 return PTR_ERR(task
);
2868 int __cold
open_ctree(struct super_block
*sb
, struct btrfs_fs_devices
*fs_devices
,
2877 struct btrfs_super_block
*disk_super
;
2878 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2879 struct btrfs_root
*tree_root
;
2880 struct btrfs_root
*chunk_root
;
2883 int clear_free_space_tree
= 0;
2886 ret
= init_mount_fs_info(fs_info
, sb
);
2892 /* These need to be init'ed before we start creating inodes and such. */
2893 tree_root
= btrfs_alloc_root(fs_info
, BTRFS_ROOT_TREE_OBJECTID
,
2895 fs_info
->tree_root
= tree_root
;
2896 chunk_root
= btrfs_alloc_root(fs_info
, BTRFS_CHUNK_TREE_OBJECTID
,
2898 fs_info
->chunk_root
= chunk_root
;
2899 if (!tree_root
|| !chunk_root
) {
2904 fs_info
->btree_inode
= new_inode(sb
);
2905 if (!fs_info
->btree_inode
) {
2909 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2910 btrfs_init_btree_inode(fs_info
);
2912 invalidate_bdev(fs_devices
->latest_bdev
);
2915 * Read super block and check the signature bytes only
2917 disk_super
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2918 if (IS_ERR(disk_super
)) {
2919 err
= PTR_ERR(disk_super
);
2924 * Verify the type first, if that or the checksum value are
2925 * corrupted, we'll find out
2927 csum_type
= btrfs_super_csum_type(disk_super
);
2928 if (!btrfs_supported_super_csum(csum_type
)) {
2929 btrfs_err(fs_info
, "unsupported checksum algorithm: %u",
2932 btrfs_release_disk_super(disk_super
);
2936 ret
= btrfs_init_csum_hash(fs_info
, csum_type
);
2939 btrfs_release_disk_super(disk_super
);
2944 * We want to check superblock checksum, the type is stored inside.
2945 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2947 if (btrfs_check_super_csum(fs_info
, (u8
*)disk_super
)) {
2948 btrfs_err(fs_info
, "superblock checksum mismatch");
2950 btrfs_release_disk_super(disk_super
);
2955 * super_copy is zeroed at allocation time and we never touch the
2956 * following bytes up to INFO_SIZE, the checksum is calculated from
2957 * the whole block of INFO_SIZE
2959 memcpy(fs_info
->super_copy
, disk_super
, sizeof(*fs_info
->super_copy
));
2960 btrfs_release_disk_super(disk_super
);
2962 disk_super
= fs_info
->super_copy
;
2964 ASSERT(!memcmp(fs_info
->fs_devices
->fsid
, fs_info
->super_copy
->fsid
,
2967 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
)) {
2968 ASSERT(!memcmp(fs_info
->fs_devices
->metadata_uuid
,
2969 fs_info
->super_copy
->metadata_uuid
,
2973 features
= btrfs_super_flags(disk_super
);
2974 if (features
& BTRFS_SUPER_FLAG_CHANGING_FSID_V2
) {
2975 features
&= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2
;
2976 btrfs_set_super_flags(disk_super
, features
);
2978 "found metadata UUID change in progress flag, clearing");
2981 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2982 sizeof(*fs_info
->super_for_commit
));
2984 ret
= btrfs_validate_mount_super(fs_info
);
2986 btrfs_err(fs_info
, "superblock contains fatal errors");
2991 if (!btrfs_super_root(disk_super
))
2994 /* check FS state, whether FS is broken. */
2995 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2996 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2999 * In the long term, we'll store the compression type in the super
3000 * block, and it'll be used for per file compression control.
3002 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
3004 ret
= btrfs_parse_options(fs_info
, options
, sb
->s_flags
);
3010 features
= btrfs_super_incompat_flags(disk_super
) &
3011 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
3014 "cannot mount because of unsupported optional features (%llx)",
3020 features
= btrfs_super_incompat_flags(disk_super
);
3021 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
3022 if (fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
3023 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
3024 else if (fs_info
->compress_type
== BTRFS_COMPRESS_ZSTD
)
3025 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD
;
3027 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
3028 btrfs_info(fs_info
, "has skinny extents");
3031 * flag our filesystem as having big metadata blocks if
3032 * they are bigger than the page size
3034 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
3035 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
3037 "flagging fs with big metadata feature");
3038 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
3041 nodesize
= btrfs_super_nodesize(disk_super
);
3042 sectorsize
= btrfs_super_sectorsize(disk_super
);
3043 stripesize
= sectorsize
;
3044 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
3045 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
3047 /* Cache block sizes */
3048 fs_info
->nodesize
= nodesize
;
3049 fs_info
->sectorsize
= sectorsize
;
3050 fs_info
->stripesize
= stripesize
;
3053 * mixed block groups end up with duplicate but slightly offset
3054 * extent buffers for the same range. It leads to corruptions
3056 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
3057 (sectorsize
!= nodesize
)) {
3059 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3060 nodesize
, sectorsize
);
3065 * Needn't use the lock because there is no other task which will
3068 btrfs_set_super_incompat_flags(disk_super
, features
);
3070 features
= btrfs_super_compat_ro_flags(disk_super
) &
3071 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
3072 if (!sb_rdonly(sb
) && features
) {
3074 "cannot mount read-write because of unsupported optional features (%llx)",
3080 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
3083 goto fail_sb_buffer
;
3086 sb
->s_bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
3087 sb
->s_bdi
->ra_pages
= VM_READAHEAD_PAGES
;
3088 sb
->s_bdi
->ra_pages
*= btrfs_super_num_devices(disk_super
);
3089 sb
->s_bdi
->ra_pages
= max(sb
->s_bdi
->ra_pages
, SZ_4M
/ PAGE_SIZE
);
3091 sb
->s_blocksize
= sectorsize
;
3092 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
3093 memcpy(&sb
->s_uuid
, fs_info
->fs_devices
->fsid
, BTRFS_FSID_SIZE
);
3095 mutex_lock(&fs_info
->chunk_mutex
);
3096 ret
= btrfs_read_sys_array(fs_info
);
3097 mutex_unlock(&fs_info
->chunk_mutex
);
3099 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
3100 goto fail_sb_buffer
;
3103 generation
= btrfs_super_chunk_root_generation(disk_super
);
3104 level
= btrfs_super_chunk_root_level(disk_super
);
3106 chunk_root
->node
= read_tree_block(fs_info
,
3107 btrfs_super_chunk_root(disk_super
),
3108 generation
, level
, NULL
);
3109 if (IS_ERR(chunk_root
->node
) ||
3110 !extent_buffer_uptodate(chunk_root
->node
)) {
3111 btrfs_err(fs_info
, "failed to read chunk root");
3112 if (!IS_ERR(chunk_root
->node
))
3113 free_extent_buffer(chunk_root
->node
);
3114 chunk_root
->node
= NULL
;
3115 goto fail_tree_roots
;
3117 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
3118 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
3120 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
3121 offsetof(struct btrfs_header
, chunk_tree_uuid
),
3124 ret
= btrfs_read_chunk_tree(fs_info
);
3126 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
3127 goto fail_tree_roots
;
3131 * Keep the devid that is marked to be the target device for the
3132 * device replace procedure
3134 btrfs_free_extra_devids(fs_devices
, 0);
3136 if (!fs_devices
->latest_bdev
) {
3137 btrfs_err(fs_info
, "failed to read devices");
3138 goto fail_tree_roots
;
3141 ret
= init_tree_roots(fs_info
);
3143 goto fail_tree_roots
;
3146 * If we have a uuid root and we're not being told to rescan we need to
3147 * check the generation here so we can set the
3148 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3149 * transaction during a balance or the log replay without updating the
3150 * uuid generation, and then if we crash we would rescan the uuid tree,
3151 * even though it was perfectly fine.
3153 if (fs_info
->uuid_root
&& !btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) &&
3154 fs_info
->generation
== btrfs_super_uuid_tree_generation(disk_super
))
3155 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
3157 ret
= btrfs_verify_dev_extents(fs_info
);
3160 "failed to verify dev extents against chunks: %d",
3162 goto fail_block_groups
;
3164 ret
= btrfs_recover_balance(fs_info
);
3166 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
3167 goto fail_block_groups
;
3170 ret
= btrfs_init_dev_stats(fs_info
);
3172 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
3173 goto fail_block_groups
;
3176 ret
= btrfs_init_dev_replace(fs_info
);
3178 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
3179 goto fail_block_groups
;
3182 btrfs_free_extra_devids(fs_devices
, 1);
3184 ret
= btrfs_sysfs_add_fsid(fs_devices
);
3186 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
3188 goto fail_block_groups
;
3191 ret
= btrfs_sysfs_add_mounted(fs_info
);
3193 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
3194 goto fail_fsdev_sysfs
;
3197 ret
= btrfs_init_space_info(fs_info
);
3199 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
3203 ret
= btrfs_read_block_groups(fs_info
);
3205 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
3209 if (!sb_rdonly(sb
) && !btrfs_check_rw_degradable(fs_info
, NULL
)) {
3211 "writable mount is not allowed due to too many missing devices");
3215 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3217 if (IS_ERR(fs_info
->cleaner_kthread
))
3220 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3222 "btrfs-transaction");
3223 if (IS_ERR(fs_info
->transaction_kthread
))
3226 if (!btrfs_test_opt(fs_info
, NOSSD
) &&
3227 !fs_info
->fs_devices
->rotating
) {
3228 btrfs_set_and_info(fs_info
, SSD
, "enabling ssd optimizations");
3232 * Mount does not set all options immediately, we can do it now and do
3233 * not have to wait for transaction commit
3235 btrfs_apply_pending_changes(fs_info
);
3237 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3238 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
)) {
3239 ret
= btrfsic_mount(fs_info
, fs_devices
,
3240 btrfs_test_opt(fs_info
,
3241 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3243 fs_info
->check_integrity_print_mask
);
3246 "failed to initialize integrity check module: %d",
3250 ret
= btrfs_read_qgroup_config(fs_info
);
3252 goto fail_trans_kthread
;
3254 if (btrfs_build_ref_tree(fs_info
))
3255 btrfs_err(fs_info
, "couldn't build ref tree");
3257 /* do not make disk changes in broken FS or nologreplay is given */
3258 if (btrfs_super_log_root(disk_super
) != 0 &&
3259 !btrfs_test_opt(fs_info
, NOLOGREPLAY
)) {
3260 btrfs_info(fs_info
, "start tree-log replay");
3261 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3268 ret
= btrfs_find_orphan_roots(fs_info
);
3272 if (!sb_rdonly(sb
)) {
3273 ret
= btrfs_cleanup_fs_roots(fs_info
);
3277 mutex_lock(&fs_info
->cleaner_mutex
);
3278 ret
= btrfs_recover_relocation(tree_root
);
3279 mutex_unlock(&fs_info
->cleaner_mutex
);
3281 btrfs_warn(fs_info
, "failed to recover relocation: %d",
3288 fs_info
->fs_root
= btrfs_get_fs_root(fs_info
, BTRFS_FS_TREE_OBJECTID
, true);
3289 if (IS_ERR(fs_info
->fs_root
)) {
3290 err
= PTR_ERR(fs_info
->fs_root
);
3291 btrfs_warn(fs_info
, "failed to read fs tree: %d", err
);
3292 fs_info
->fs_root
= NULL
;
3299 if (btrfs_test_opt(fs_info
, CLEAR_CACHE
) &&
3300 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3301 clear_free_space_tree
= 1;
3302 } else if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
3303 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE_VALID
)) {
3304 btrfs_warn(fs_info
, "free space tree is invalid");
3305 clear_free_space_tree
= 1;
3308 if (clear_free_space_tree
) {
3309 btrfs_info(fs_info
, "clearing free space tree");
3310 ret
= btrfs_clear_free_space_tree(fs_info
);
3313 "failed to clear free space tree: %d", ret
);
3314 close_ctree(fs_info
);
3319 if (btrfs_test_opt(fs_info
, FREE_SPACE_TREE
) &&
3320 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3321 btrfs_info(fs_info
, "creating free space tree");
3322 ret
= btrfs_create_free_space_tree(fs_info
);
3325 "failed to create free space tree: %d", ret
);
3326 close_ctree(fs_info
);
3331 down_read(&fs_info
->cleanup_work_sem
);
3332 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3333 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3334 up_read(&fs_info
->cleanup_work_sem
);
3335 close_ctree(fs_info
);
3338 up_read(&fs_info
->cleanup_work_sem
);
3340 ret
= btrfs_resume_balance_async(fs_info
);
3342 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3343 close_ctree(fs_info
);
3347 ret
= btrfs_resume_dev_replace_async(fs_info
);
3349 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3350 close_ctree(fs_info
);
3354 btrfs_qgroup_rescan_resume(fs_info
);
3355 btrfs_discard_resume(fs_info
);
3357 if (!fs_info
->uuid_root
) {
3358 btrfs_info(fs_info
, "creating UUID tree");
3359 ret
= btrfs_create_uuid_tree(fs_info
);
3362 "failed to create the UUID tree: %d", ret
);
3363 close_ctree(fs_info
);
3366 } else if (btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) ||
3367 fs_info
->generation
!=
3368 btrfs_super_uuid_tree_generation(disk_super
)) {
3369 btrfs_info(fs_info
, "checking UUID tree");
3370 ret
= btrfs_check_uuid_tree(fs_info
);
3373 "failed to check the UUID tree: %d", ret
);
3374 close_ctree(fs_info
);
3378 set_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3381 * backuproot only affect mount behavior, and if open_ctree succeeded,
3382 * no need to keep the flag
3384 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3389 btrfs_free_qgroup_config(fs_info
);
3391 kthread_stop(fs_info
->transaction_kthread
);
3392 btrfs_cleanup_transaction(fs_info
);
3393 btrfs_free_fs_roots(fs_info
);
3395 kthread_stop(fs_info
->cleaner_kthread
);
3398 * make sure we're done with the btree inode before we stop our
3401 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3404 btrfs_sysfs_remove_mounted(fs_info
);
3407 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3410 btrfs_put_block_group_cache(fs_info
);
3413 if (fs_info
->data_reloc_root
)
3414 btrfs_drop_and_free_fs_root(fs_info
, fs_info
->data_reloc_root
);
3415 free_root_pointers(fs_info
, true);
3416 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3419 btrfs_stop_all_workers(fs_info
);
3420 btrfs_free_block_groups(fs_info
);
3422 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3424 iput(fs_info
->btree_inode
);
3426 btrfs_close_devices(fs_info
->fs_devices
);
3429 ALLOW_ERROR_INJECTION(open_ctree
, ERRNO
);
3431 static void btrfs_end_super_write(struct bio
*bio
)
3433 struct btrfs_device
*device
= bio
->bi_private
;
3434 struct bio_vec
*bvec
;
3435 struct bvec_iter_all iter_all
;
3438 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
3439 page
= bvec
->bv_page
;
3441 if (bio
->bi_status
) {
3442 btrfs_warn_rl_in_rcu(device
->fs_info
,
3443 "lost page write due to IO error on %s (%d)",
3444 rcu_str_deref(device
->name
),
3445 blk_status_to_errno(bio
->bi_status
));
3446 ClearPageUptodate(page
);
3448 btrfs_dev_stat_inc_and_print(device
,
3449 BTRFS_DEV_STAT_WRITE_ERRS
);
3451 SetPageUptodate(page
);
3461 struct btrfs_super_block
*btrfs_read_dev_one_super(struct block_device
*bdev
,
3464 struct btrfs_super_block
*super
;
3467 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
3469 bytenr
= btrfs_sb_offset(copy_num
);
3470 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3471 return ERR_PTR(-EINVAL
);
3473 page
= read_cache_page_gfp(mapping
, bytenr
>> PAGE_SHIFT
, GFP_NOFS
);
3475 return ERR_CAST(page
);
3477 super
= page_address(page
);
3478 if (btrfs_super_bytenr(super
) != bytenr
||
3479 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3480 btrfs_release_disk_super(super
);
3481 return ERR_PTR(-EINVAL
);
3488 struct btrfs_super_block
*btrfs_read_dev_super(struct block_device
*bdev
)
3490 struct btrfs_super_block
*super
, *latest
= NULL
;
3494 /* we would like to check all the supers, but that would make
3495 * a btrfs mount succeed after a mkfs from a different FS.
3496 * So, we need to add a special mount option to scan for
3497 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3499 for (i
= 0; i
< 1; i
++) {
3500 super
= btrfs_read_dev_one_super(bdev
, i
);
3504 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3506 btrfs_release_disk_super(super
);
3509 transid
= btrfs_super_generation(super
);
3517 * Write superblock @sb to the @device. Do not wait for completion, all the
3518 * pages we use for writing are locked.
3520 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3521 * the expected device size at commit time. Note that max_mirrors must be
3522 * same for write and wait phases.
3524 * Return number of errors when page is not found or submission fails.
3526 static int write_dev_supers(struct btrfs_device
*device
,
3527 struct btrfs_super_block
*sb
, int max_mirrors
)
3529 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
3530 struct address_space
*mapping
= device
->bdev
->bd_inode
->i_mapping
;
3531 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
3536 if (max_mirrors
== 0)
3537 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3539 shash
->tfm
= fs_info
->csum_shash
;
3541 for (i
= 0; i
< max_mirrors
; i
++) {
3544 struct btrfs_super_block
*disk_super
;
3546 bytenr
= btrfs_sb_offset(i
);
3547 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3548 device
->commit_total_bytes
)
3551 btrfs_set_super_bytenr(sb
, bytenr
);
3553 crypto_shash_digest(shash
, (const char *)sb
+ BTRFS_CSUM_SIZE
,
3554 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
,
3557 page
= find_or_create_page(mapping
, bytenr
>> PAGE_SHIFT
,
3560 btrfs_err(device
->fs_info
,
3561 "couldn't get super block page for bytenr %llu",
3567 /* Bump the refcount for wait_dev_supers() */
3570 disk_super
= page_address(page
);
3571 memcpy(disk_super
, sb
, BTRFS_SUPER_INFO_SIZE
);
3574 * Directly use bios here instead of relying on the page cache
3575 * to do I/O, so we don't lose the ability to do integrity
3578 bio
= bio_alloc(GFP_NOFS
, 1);
3579 bio_set_dev(bio
, device
->bdev
);
3580 bio
->bi_iter
.bi_sector
= bytenr
>> SECTOR_SHIFT
;
3581 bio
->bi_private
= device
;
3582 bio
->bi_end_io
= btrfs_end_super_write
;
3583 __bio_add_page(bio
, page
, BTRFS_SUPER_INFO_SIZE
,
3584 offset_in_page(bytenr
));
3587 * We FUA only the first super block. The others we allow to
3588 * go down lazy and there's a short window where the on-disk
3589 * copies might still contain the older version.
3591 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_META
| REQ_PRIO
;
3592 if (i
== 0 && !btrfs_test_opt(device
->fs_info
, NOBARRIER
))
3593 bio
->bi_opf
|= REQ_FUA
;
3595 btrfsic_submit_bio(bio
);
3597 return errors
< i
? 0 : -1;
3601 * Wait for write completion of superblocks done by write_dev_supers,
3602 * @max_mirrors same for write and wait phases.
3604 * Return number of errors when page is not found or not marked up to
3607 static int wait_dev_supers(struct btrfs_device
*device
, int max_mirrors
)
3611 bool primary_failed
= false;
3614 if (max_mirrors
== 0)
3615 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3617 for (i
= 0; i
< max_mirrors
; i
++) {
3620 bytenr
= btrfs_sb_offset(i
);
3621 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3622 device
->commit_total_bytes
)
3625 page
= find_get_page(device
->bdev
->bd_inode
->i_mapping
,
3626 bytenr
>> PAGE_SHIFT
);
3630 primary_failed
= true;
3633 /* Page is submitted locked and unlocked once the IO completes */
3634 wait_on_page_locked(page
);
3635 if (PageError(page
)) {
3638 primary_failed
= true;
3641 /* Drop our reference */
3644 /* Drop the reference from the writing run */
3648 /* log error, force error return */
3649 if (primary_failed
) {
3650 btrfs_err(device
->fs_info
, "error writing primary super block to device %llu",
3655 return errors
< i
? 0 : -1;
3659 * endio for the write_dev_flush, this will wake anyone waiting
3660 * for the barrier when it is done
3662 static void btrfs_end_empty_barrier(struct bio
*bio
)
3664 complete(bio
->bi_private
);
3668 * Submit a flush request to the device if it supports it. Error handling is
3669 * done in the waiting counterpart.
3671 static void write_dev_flush(struct btrfs_device
*device
)
3673 struct request_queue
*q
= bdev_get_queue(device
->bdev
);
3674 struct bio
*bio
= device
->flush_bio
;
3676 if (!test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
))
3680 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3681 bio_set_dev(bio
, device
->bdev
);
3682 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
;
3683 init_completion(&device
->flush_wait
);
3684 bio
->bi_private
= &device
->flush_wait
;
3686 btrfsic_submit_bio(bio
);
3687 set_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3691 * If the flush bio has been submitted by write_dev_flush, wait for it.
3693 static blk_status_t
wait_dev_flush(struct btrfs_device
*device
)
3695 struct bio
*bio
= device
->flush_bio
;
3697 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
))
3700 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3701 wait_for_completion_io(&device
->flush_wait
);
3703 return bio
->bi_status
;
3706 static int check_barrier_error(struct btrfs_fs_info
*fs_info
)
3708 if (!btrfs_check_rw_degradable(fs_info
, NULL
))
3714 * send an empty flush down to each device in parallel,
3715 * then wait for them
3717 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3719 struct list_head
*head
;
3720 struct btrfs_device
*dev
;
3721 int errors_wait
= 0;
3724 lockdep_assert_held(&info
->fs_devices
->device_list_mutex
);
3725 /* send down all the barriers */
3726 head
= &info
->fs_devices
->devices
;
3727 list_for_each_entry(dev
, head
, dev_list
) {
3728 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3732 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3733 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3736 write_dev_flush(dev
);
3737 dev
->last_flush_error
= BLK_STS_OK
;
3740 /* wait for all the barriers */
3741 list_for_each_entry(dev
, head
, dev_list
) {
3742 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3748 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3749 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3752 ret
= wait_dev_flush(dev
);
3754 dev
->last_flush_error
= ret
;
3755 btrfs_dev_stat_inc_and_print(dev
,
3756 BTRFS_DEV_STAT_FLUSH_ERRS
);
3763 * At some point we need the status of all disks
3764 * to arrive at the volume status. So error checking
3765 * is being pushed to a separate loop.
3767 return check_barrier_error(info
);
3772 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3775 int min_tolerated
= INT_MAX
;
3777 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3778 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3779 min_tolerated
= min_t(int, min_tolerated
,
3780 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3781 tolerated_failures
);
3783 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3784 if (raid_type
== BTRFS_RAID_SINGLE
)
3786 if (!(flags
& btrfs_raid_array
[raid_type
].bg_flag
))
3788 min_tolerated
= min_t(int, min_tolerated
,
3789 btrfs_raid_array
[raid_type
].
3790 tolerated_failures
);
3793 if (min_tolerated
== INT_MAX
) {
3794 pr_warn("BTRFS: unknown raid flag: %llu", flags
);
3798 return min_tolerated
;
3801 int write_all_supers(struct btrfs_fs_info
*fs_info
, int max_mirrors
)
3803 struct list_head
*head
;
3804 struct btrfs_device
*dev
;
3805 struct btrfs_super_block
*sb
;
3806 struct btrfs_dev_item
*dev_item
;
3810 int total_errors
= 0;
3813 do_barriers
= !btrfs_test_opt(fs_info
, NOBARRIER
);
3816 * max_mirrors == 0 indicates we're from commit_transaction,
3817 * not from fsync where the tree roots in fs_info have not
3818 * been consistent on disk.
3820 if (max_mirrors
== 0)
3821 backup_super_roots(fs_info
);
3823 sb
= fs_info
->super_for_commit
;
3824 dev_item
= &sb
->dev_item
;
3826 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3827 head
= &fs_info
->fs_devices
->devices
;
3828 max_errors
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
3831 ret
= barrier_all_devices(fs_info
);
3834 &fs_info
->fs_devices
->device_list_mutex
);
3835 btrfs_handle_fs_error(fs_info
, ret
,
3836 "errors while submitting device barriers.");
3841 list_for_each_entry(dev
, head
, dev_list
) {
3846 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3847 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3850 btrfs_set_stack_device_generation(dev_item
, 0);
3851 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3852 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3853 btrfs_set_stack_device_total_bytes(dev_item
,
3854 dev
->commit_total_bytes
);
3855 btrfs_set_stack_device_bytes_used(dev_item
,
3856 dev
->commit_bytes_used
);
3857 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3858 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3859 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3860 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3861 memcpy(dev_item
->fsid
, dev
->fs_devices
->metadata_uuid
,
3864 flags
= btrfs_super_flags(sb
);
3865 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3867 ret
= btrfs_validate_write_super(fs_info
, sb
);
3869 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3870 btrfs_handle_fs_error(fs_info
, -EUCLEAN
,
3871 "unexpected superblock corruption detected");
3875 ret
= write_dev_supers(dev
, sb
, max_mirrors
);
3879 if (total_errors
> max_errors
) {
3880 btrfs_err(fs_info
, "%d errors while writing supers",
3882 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3884 /* FUA is masked off if unsupported and can't be the reason */
3885 btrfs_handle_fs_error(fs_info
, -EIO
,
3886 "%d errors while writing supers",
3892 list_for_each_entry(dev
, head
, dev_list
) {
3895 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3896 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3899 ret
= wait_dev_supers(dev
, max_mirrors
);
3903 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3904 if (total_errors
> max_errors
) {
3905 btrfs_handle_fs_error(fs_info
, -EIO
,
3906 "%d errors while writing supers",
3913 /* Drop a fs root from the radix tree and free it. */
3914 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3915 struct btrfs_root
*root
)
3917 bool drop_ref
= false;
3919 spin_lock(&fs_info
->fs_roots_radix_lock
);
3920 radix_tree_delete(&fs_info
->fs_roots_radix
,
3921 (unsigned long)root
->root_key
.objectid
);
3922 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
))
3924 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3926 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3927 ASSERT(root
->log_root
== NULL
);
3928 if (root
->reloc_root
) {
3929 btrfs_put_root(root
->reloc_root
);
3930 root
->reloc_root
= NULL
;
3934 if (root
->free_ino_pinned
)
3935 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3936 if (root
->free_ino_ctl
)
3937 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3938 if (root
->ino_cache_inode
) {
3939 iput(root
->ino_cache_inode
);
3940 root
->ino_cache_inode
= NULL
;
3943 btrfs_put_root(root
);
3946 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3948 u64 root_objectid
= 0;
3949 struct btrfs_root
*gang
[8];
3952 unsigned int ret
= 0;
3955 spin_lock(&fs_info
->fs_roots_radix_lock
);
3956 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3957 (void **)gang
, root_objectid
,
3960 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3963 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3965 for (i
= 0; i
< ret
; i
++) {
3966 /* Avoid to grab roots in dead_roots */
3967 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3971 /* grab all the search result for later use */
3972 gang
[i
] = btrfs_grab_root(gang
[i
]);
3974 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3976 for (i
= 0; i
< ret
; i
++) {
3979 root_objectid
= gang
[i
]->root_key
.objectid
;
3980 err
= btrfs_orphan_cleanup(gang
[i
]);
3983 btrfs_put_root(gang
[i
]);
3988 /* release the uncleaned roots due to error */
3989 for (; i
< ret
; i
++) {
3991 btrfs_put_root(gang
[i
]);
3996 int btrfs_commit_super(struct btrfs_fs_info
*fs_info
)
3998 struct btrfs_root
*root
= fs_info
->tree_root
;
3999 struct btrfs_trans_handle
*trans
;
4001 mutex_lock(&fs_info
->cleaner_mutex
);
4002 btrfs_run_delayed_iputs(fs_info
);
4003 mutex_unlock(&fs_info
->cleaner_mutex
);
4004 wake_up_process(fs_info
->cleaner_kthread
);
4006 /* wait until ongoing cleanup work done */
4007 down_write(&fs_info
->cleanup_work_sem
);
4008 up_write(&fs_info
->cleanup_work_sem
);
4010 trans
= btrfs_join_transaction(root
);
4012 return PTR_ERR(trans
);
4013 return btrfs_commit_transaction(trans
);
4016 void __cold
close_ctree(struct btrfs_fs_info
*fs_info
)
4020 set_bit(BTRFS_FS_CLOSING_START
, &fs_info
->flags
);
4022 * We don't want the cleaner to start new transactions, add more delayed
4023 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4024 * because that frees the task_struct, and the transaction kthread might
4025 * still try to wake up the cleaner.
4027 kthread_park(fs_info
->cleaner_kthread
);
4029 /* wait for the qgroup rescan worker to stop */
4030 btrfs_qgroup_wait_for_completion(fs_info
, false);
4032 /* wait for the uuid_scan task to finish */
4033 down(&fs_info
->uuid_tree_rescan_sem
);
4034 /* avoid complains from lockdep et al., set sem back to initial state */
4035 up(&fs_info
->uuid_tree_rescan_sem
);
4037 /* pause restriper - we want to resume on mount */
4038 btrfs_pause_balance(fs_info
);
4040 btrfs_dev_replace_suspend_for_unmount(fs_info
);
4042 btrfs_scrub_cancel(fs_info
);
4044 /* wait for any defraggers to finish */
4045 wait_event(fs_info
->transaction_wait
,
4046 (atomic_read(&fs_info
->defrag_running
) == 0));
4048 /* clear out the rbtree of defraggable inodes */
4049 btrfs_cleanup_defrag_inodes(fs_info
);
4051 cancel_work_sync(&fs_info
->async_reclaim_work
);
4052 cancel_work_sync(&fs_info
->async_data_reclaim_work
);
4054 /* Cancel or finish ongoing discard work */
4055 btrfs_discard_cleanup(fs_info
);
4057 if (!sb_rdonly(fs_info
->sb
)) {
4059 * The cleaner kthread is stopped, so do one final pass over
4060 * unused block groups.
4062 btrfs_delete_unused_bgs(fs_info
);
4065 * There might be existing delayed inode workers still running
4066 * and holding an empty delayed inode item. We must wait for
4067 * them to complete first because they can create a transaction.
4068 * This happens when someone calls btrfs_balance_delayed_items()
4069 * and then a transaction commit runs the same delayed nodes
4070 * before any delayed worker has done something with the nodes.
4071 * We must wait for any worker here and not at transaction
4072 * commit time since that could cause a deadlock.
4073 * This is a very rare case.
4075 btrfs_flush_workqueue(fs_info
->delayed_workers
);
4077 ret
= btrfs_commit_super(fs_info
);
4079 btrfs_err(fs_info
, "commit super ret %d", ret
);
4082 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
) ||
4083 test_bit(BTRFS_FS_STATE_TRANS_ABORTED
, &fs_info
->fs_state
))
4084 btrfs_error_commit_super(fs_info
);
4086 kthread_stop(fs_info
->transaction_kthread
);
4087 kthread_stop(fs_info
->cleaner_kthread
);
4089 ASSERT(list_empty(&fs_info
->delayed_iputs
));
4090 set_bit(BTRFS_FS_CLOSING_DONE
, &fs_info
->flags
);
4092 if (btrfs_check_quota_leak(fs_info
)) {
4093 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
4094 btrfs_err(fs_info
, "qgroup reserved space leaked");
4097 btrfs_free_qgroup_config(fs_info
);
4098 ASSERT(list_empty(&fs_info
->delalloc_roots
));
4100 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
4101 btrfs_info(fs_info
, "at unmount delalloc count %lld",
4102 percpu_counter_sum(&fs_info
->delalloc_bytes
));
4105 if (percpu_counter_sum(&fs_info
->dio_bytes
))
4106 btrfs_info(fs_info
, "at unmount dio bytes count %lld",
4107 percpu_counter_sum(&fs_info
->dio_bytes
));
4109 btrfs_sysfs_remove_mounted(fs_info
);
4110 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
4112 btrfs_put_block_group_cache(fs_info
);
4115 * we must make sure there is not any read request to
4116 * submit after we stopping all workers.
4118 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
4119 btrfs_stop_all_workers(fs_info
);
4121 clear_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
4122 free_root_pointers(fs_info
, true);
4123 btrfs_free_fs_roots(fs_info
);
4126 * We must free the block groups after dropping the fs_roots as we could
4127 * have had an IO error and have left over tree log blocks that aren't
4128 * cleaned up until the fs roots are freed. This makes the block group
4129 * accounting appear to be wrong because there's pending reserved bytes,
4130 * so make sure we do the block group cleanup afterwards.
4132 btrfs_free_block_groups(fs_info
);
4134 iput(fs_info
->btree_inode
);
4136 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4137 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
))
4138 btrfsic_unmount(fs_info
->fs_devices
);
4141 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
4142 btrfs_close_devices(fs_info
->fs_devices
);
4145 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
4149 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
4151 ret
= extent_buffer_uptodate(buf
);
4155 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
4156 parent_transid
, atomic
);
4162 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
4164 struct btrfs_fs_info
*fs_info
;
4165 struct btrfs_root
*root
;
4166 u64 transid
= btrfs_header_generation(buf
);
4169 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4171 * This is a fast path so only do this check if we have sanity tests
4172 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4173 * outside of the sanity tests.
4175 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED
, &buf
->bflags
)))
4178 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4179 fs_info
= root
->fs_info
;
4180 btrfs_assert_tree_locked(buf
);
4181 if (transid
!= fs_info
->generation
)
4182 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4183 buf
->start
, transid
, fs_info
->generation
);
4184 was_dirty
= set_extent_buffer_dirty(buf
);
4186 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
4188 fs_info
->dirty_metadata_batch
);
4189 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4191 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4192 * but item data not updated.
4193 * So here we should only check item pointers, not item data.
4195 if (btrfs_header_level(buf
) == 0 &&
4196 btrfs_check_leaf_relaxed(buf
)) {
4197 btrfs_print_leaf(buf
);
4203 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
,
4207 * looks as though older kernels can get into trouble with
4208 * this code, they end up stuck in balance_dirty_pages forever
4212 if (current
->flags
& PF_MEMALLOC
)
4216 btrfs_balance_delayed_items(fs_info
);
4218 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
4219 BTRFS_DIRTY_METADATA_THRESH
,
4220 fs_info
->dirty_metadata_batch
);
4222 balance_dirty_pages_ratelimited(fs_info
->btree_inode
->i_mapping
);
4226 void btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
)
4228 __btrfs_btree_balance_dirty(fs_info
, 1);
4231 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info
*fs_info
)
4233 __btrfs_btree_balance_dirty(fs_info
, 0);
4236 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
, int level
,
4237 struct btrfs_key
*first_key
)
4239 return btree_read_extent_buffer_pages(buf
, parent_transid
,
4243 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
)
4245 /* cleanup FS via transaction */
4246 btrfs_cleanup_transaction(fs_info
);
4248 mutex_lock(&fs_info
->cleaner_mutex
);
4249 btrfs_run_delayed_iputs(fs_info
);
4250 mutex_unlock(&fs_info
->cleaner_mutex
);
4252 down_write(&fs_info
->cleanup_work_sem
);
4253 up_write(&fs_info
->cleanup_work_sem
);
4256 static void btrfs_drop_all_logs(struct btrfs_fs_info
*fs_info
)
4258 struct btrfs_root
*gang
[8];
4259 u64 root_objectid
= 0;
4262 spin_lock(&fs_info
->fs_roots_radix_lock
);
4263 while ((ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
4264 (void **)gang
, root_objectid
,
4265 ARRAY_SIZE(gang
))) != 0) {
4268 for (i
= 0; i
< ret
; i
++)
4269 gang
[i
] = btrfs_grab_root(gang
[i
]);
4270 spin_unlock(&fs_info
->fs_roots_radix_lock
);
4272 for (i
= 0; i
< ret
; i
++) {
4275 root_objectid
= gang
[i
]->root_key
.objectid
;
4276 btrfs_free_log(NULL
, gang
[i
]);
4277 btrfs_put_root(gang
[i
]);
4280 spin_lock(&fs_info
->fs_roots_radix_lock
);
4282 spin_unlock(&fs_info
->fs_roots_radix_lock
);
4283 btrfs_free_log_root_tree(NULL
, fs_info
);
4286 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4288 struct btrfs_ordered_extent
*ordered
;
4290 spin_lock(&root
->ordered_extent_lock
);
4292 * This will just short circuit the ordered completion stuff which will
4293 * make sure the ordered extent gets properly cleaned up.
4295 list_for_each_entry(ordered
, &root
->ordered_extents
,
4297 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4298 spin_unlock(&root
->ordered_extent_lock
);
4301 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4303 struct btrfs_root
*root
;
4304 struct list_head splice
;
4306 INIT_LIST_HEAD(&splice
);
4308 spin_lock(&fs_info
->ordered_root_lock
);
4309 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4310 while (!list_empty(&splice
)) {
4311 root
= list_first_entry(&splice
, struct btrfs_root
,
4313 list_move_tail(&root
->ordered_root
,
4314 &fs_info
->ordered_roots
);
4316 spin_unlock(&fs_info
->ordered_root_lock
);
4317 btrfs_destroy_ordered_extents(root
);
4320 spin_lock(&fs_info
->ordered_root_lock
);
4322 spin_unlock(&fs_info
->ordered_root_lock
);
4325 * We need this here because if we've been flipped read-only we won't
4326 * get sync() from the umount, so we need to make sure any ordered
4327 * extents that haven't had their dirty pages IO start writeout yet
4328 * actually get run and error out properly.
4330 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0, (u64
)-1);
4333 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4334 struct btrfs_fs_info
*fs_info
)
4336 struct rb_node
*node
;
4337 struct btrfs_delayed_ref_root
*delayed_refs
;
4338 struct btrfs_delayed_ref_node
*ref
;
4341 delayed_refs
= &trans
->delayed_refs
;
4343 spin_lock(&delayed_refs
->lock
);
4344 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4345 spin_unlock(&delayed_refs
->lock
);
4346 btrfs_debug(fs_info
, "delayed_refs has NO entry");
4350 while ((node
= rb_first_cached(&delayed_refs
->href_root
)) != NULL
) {
4351 struct btrfs_delayed_ref_head
*head
;
4353 bool pin_bytes
= false;
4355 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4357 if (btrfs_delayed_ref_lock(delayed_refs
, head
))
4360 spin_lock(&head
->lock
);
4361 while ((n
= rb_first_cached(&head
->ref_tree
)) != NULL
) {
4362 ref
= rb_entry(n
, struct btrfs_delayed_ref_node
,
4365 rb_erase_cached(&ref
->ref_node
, &head
->ref_tree
);
4366 RB_CLEAR_NODE(&ref
->ref_node
);
4367 if (!list_empty(&ref
->add_list
))
4368 list_del(&ref
->add_list
);
4369 atomic_dec(&delayed_refs
->num_entries
);
4370 btrfs_put_delayed_ref(ref
);
4372 if (head
->must_insert_reserved
)
4374 btrfs_free_delayed_extent_op(head
->extent_op
);
4375 btrfs_delete_ref_head(delayed_refs
, head
);
4376 spin_unlock(&head
->lock
);
4377 spin_unlock(&delayed_refs
->lock
);
4378 mutex_unlock(&head
->mutex
);
4381 struct btrfs_block_group
*cache
;
4383 cache
= btrfs_lookup_block_group(fs_info
, head
->bytenr
);
4386 spin_lock(&cache
->space_info
->lock
);
4387 spin_lock(&cache
->lock
);
4388 cache
->pinned
+= head
->num_bytes
;
4389 btrfs_space_info_update_bytes_pinned(fs_info
,
4390 cache
->space_info
, head
->num_bytes
);
4391 cache
->reserved
-= head
->num_bytes
;
4392 cache
->space_info
->bytes_reserved
-= head
->num_bytes
;
4393 spin_unlock(&cache
->lock
);
4394 spin_unlock(&cache
->space_info
->lock
);
4395 percpu_counter_add_batch(
4396 &cache
->space_info
->total_bytes_pinned
,
4397 head
->num_bytes
, BTRFS_TOTAL_BYTES_PINNED_BATCH
);
4399 btrfs_put_block_group(cache
);
4401 btrfs_error_unpin_extent_range(fs_info
, head
->bytenr
,
4402 head
->bytenr
+ head
->num_bytes
- 1);
4404 btrfs_cleanup_ref_head_accounting(fs_info
, delayed_refs
, head
);
4405 btrfs_put_delayed_ref_head(head
);
4407 spin_lock(&delayed_refs
->lock
);
4409 btrfs_qgroup_destroy_extent_records(trans
);
4411 spin_unlock(&delayed_refs
->lock
);
4416 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4418 struct btrfs_inode
*btrfs_inode
;
4419 struct list_head splice
;
4421 INIT_LIST_HEAD(&splice
);
4423 spin_lock(&root
->delalloc_lock
);
4424 list_splice_init(&root
->delalloc_inodes
, &splice
);
4426 while (!list_empty(&splice
)) {
4427 struct inode
*inode
= NULL
;
4428 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4430 __btrfs_del_delalloc_inode(root
, btrfs_inode
);
4431 spin_unlock(&root
->delalloc_lock
);
4434 * Make sure we get a live inode and that it'll not disappear
4437 inode
= igrab(&btrfs_inode
->vfs_inode
);
4439 invalidate_inode_pages2(inode
->i_mapping
);
4442 spin_lock(&root
->delalloc_lock
);
4444 spin_unlock(&root
->delalloc_lock
);
4447 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4449 struct btrfs_root
*root
;
4450 struct list_head splice
;
4452 INIT_LIST_HEAD(&splice
);
4454 spin_lock(&fs_info
->delalloc_root_lock
);
4455 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4456 while (!list_empty(&splice
)) {
4457 root
= list_first_entry(&splice
, struct btrfs_root
,
4459 root
= btrfs_grab_root(root
);
4461 spin_unlock(&fs_info
->delalloc_root_lock
);
4463 btrfs_destroy_delalloc_inodes(root
);
4464 btrfs_put_root(root
);
4466 spin_lock(&fs_info
->delalloc_root_lock
);
4468 spin_unlock(&fs_info
->delalloc_root_lock
);
4471 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
4472 struct extent_io_tree
*dirty_pages
,
4476 struct extent_buffer
*eb
;
4481 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4486 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4487 while (start
<= end
) {
4488 eb
= find_extent_buffer(fs_info
, start
);
4489 start
+= fs_info
->nodesize
;
4492 wait_on_extent_buffer_writeback(eb
);
4494 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4496 clear_extent_buffer_dirty(eb
);
4497 free_extent_buffer_stale(eb
);
4504 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
4505 struct extent_io_tree
*unpin
)
4512 struct extent_state
*cached_state
= NULL
;
4515 * The btrfs_finish_extent_commit() may get the same range as
4516 * ours between find_first_extent_bit and clear_extent_dirty.
4517 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4518 * the same extent range.
4520 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
4521 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4522 EXTENT_DIRTY
, &cached_state
);
4524 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4528 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
4529 free_extent_state(cached_state
);
4530 btrfs_error_unpin_extent_range(fs_info
, start
, end
);
4531 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4538 static void btrfs_cleanup_bg_io(struct btrfs_block_group
*cache
)
4540 struct inode
*inode
;
4542 inode
= cache
->io_ctl
.inode
;
4544 invalidate_inode_pages2(inode
->i_mapping
);
4545 BTRFS_I(inode
)->generation
= 0;
4546 cache
->io_ctl
.inode
= NULL
;
4549 ASSERT(cache
->io_ctl
.pages
== NULL
);
4550 btrfs_put_block_group(cache
);
4553 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction
*cur_trans
,
4554 struct btrfs_fs_info
*fs_info
)
4556 struct btrfs_block_group
*cache
;
4558 spin_lock(&cur_trans
->dirty_bgs_lock
);
4559 while (!list_empty(&cur_trans
->dirty_bgs
)) {
4560 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
4561 struct btrfs_block_group
,
4564 if (!list_empty(&cache
->io_list
)) {
4565 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4566 list_del_init(&cache
->io_list
);
4567 btrfs_cleanup_bg_io(cache
);
4568 spin_lock(&cur_trans
->dirty_bgs_lock
);
4571 list_del_init(&cache
->dirty_list
);
4572 spin_lock(&cache
->lock
);
4573 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4574 spin_unlock(&cache
->lock
);
4576 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4577 btrfs_put_block_group(cache
);
4578 btrfs_delayed_refs_rsv_release(fs_info
, 1);
4579 spin_lock(&cur_trans
->dirty_bgs_lock
);
4581 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4584 * Refer to the definition of io_bgs member for details why it's safe
4585 * to use it without any locking
4587 while (!list_empty(&cur_trans
->io_bgs
)) {
4588 cache
= list_first_entry(&cur_trans
->io_bgs
,
4589 struct btrfs_block_group
,
4592 list_del_init(&cache
->io_list
);
4593 spin_lock(&cache
->lock
);
4594 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4595 spin_unlock(&cache
->lock
);
4596 btrfs_cleanup_bg_io(cache
);
4600 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4601 struct btrfs_fs_info
*fs_info
)
4603 struct btrfs_device
*dev
, *tmp
;
4605 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
4606 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
4607 ASSERT(list_empty(&cur_trans
->io_bgs
));
4609 list_for_each_entry_safe(dev
, tmp
, &cur_trans
->dev_update_list
,
4611 list_del_init(&dev
->post_commit_list
);
4614 btrfs_destroy_delayed_refs(cur_trans
, fs_info
);
4616 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4617 wake_up(&fs_info
->transaction_blocked_wait
);
4619 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4620 wake_up(&fs_info
->transaction_wait
);
4622 btrfs_destroy_delayed_inodes(fs_info
);
4624 btrfs_destroy_marked_extents(fs_info
, &cur_trans
->dirty_pages
,
4626 btrfs_destroy_pinned_extent(fs_info
, &cur_trans
->pinned_extents
);
4628 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4629 wake_up(&cur_trans
->commit_wait
);
4632 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
)
4634 struct btrfs_transaction
*t
;
4636 mutex_lock(&fs_info
->transaction_kthread_mutex
);
4638 spin_lock(&fs_info
->trans_lock
);
4639 while (!list_empty(&fs_info
->trans_list
)) {
4640 t
= list_first_entry(&fs_info
->trans_list
,
4641 struct btrfs_transaction
, list
);
4642 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4643 refcount_inc(&t
->use_count
);
4644 spin_unlock(&fs_info
->trans_lock
);
4645 btrfs_wait_for_commit(fs_info
, t
->transid
);
4646 btrfs_put_transaction(t
);
4647 spin_lock(&fs_info
->trans_lock
);
4650 if (t
== fs_info
->running_transaction
) {
4651 t
->state
= TRANS_STATE_COMMIT_DOING
;
4652 spin_unlock(&fs_info
->trans_lock
);
4654 * We wait for 0 num_writers since we don't hold a trans
4655 * handle open currently for this transaction.
4657 wait_event(t
->writer_wait
,
4658 atomic_read(&t
->num_writers
) == 0);
4660 spin_unlock(&fs_info
->trans_lock
);
4662 btrfs_cleanup_one_transaction(t
, fs_info
);
4664 spin_lock(&fs_info
->trans_lock
);
4665 if (t
== fs_info
->running_transaction
)
4666 fs_info
->running_transaction
= NULL
;
4667 list_del_init(&t
->list
);
4668 spin_unlock(&fs_info
->trans_lock
);
4670 btrfs_put_transaction(t
);
4671 trace_btrfs_transaction_commit(fs_info
->tree_root
);
4672 spin_lock(&fs_info
->trans_lock
);
4674 spin_unlock(&fs_info
->trans_lock
);
4675 btrfs_destroy_all_ordered_extents(fs_info
);
4676 btrfs_destroy_delayed_inodes(fs_info
);
4677 btrfs_assert_delayed_root_empty(fs_info
);
4678 btrfs_destroy_all_delalloc_inodes(fs_info
);
4679 btrfs_drop_all_logs(fs_info
);
4680 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
4685 static const struct extent_io_ops btree_extent_io_ops
= {
4686 /* mandatory callbacks */
4687 .submit_bio_hook
= btree_submit_bio_hook
,
4688 .readpage_end_io_hook
= btree_readpage_end_io_hook
,