1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <asm/unaligned.h>
23 #include "transaction.h"
24 #include "btrfs_inode.h"
26 #include "print-tree.h"
29 #include "free-space-cache.h"
30 #include "free-space-tree.h"
31 #include "inode-map.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
43 #include <asm/cpufeature.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
);
102 * async submit bios are used to offload expensive checksumming
103 * onto the worker threads. They checksum file and metadata bios
104 * just before they are sent down the IO stack.
106 struct async_submit_bio
{
109 extent_submit_bio_start_t
*submit_bio_start
;
112 * bio_offset is optional, can be used if the pages in the bio
113 * can't tell us where in the file the bio should go
116 struct btrfs_work work
;
121 * Lockdep class keys for extent_buffer->lock's in this root. For a given
122 * eb, the lockdep key is determined by the btrfs_root it belongs to and
123 * the level the eb occupies in the tree.
125 * Different roots are used for different purposes and may nest inside each
126 * other and they require separate keysets. As lockdep keys should be
127 * static, assign keysets according to the purpose of the root as indicated
128 * by btrfs_root->root_key.objectid. This ensures that all special purpose
129 * roots have separate keysets.
131 * Lock-nesting across peer nodes is always done with the immediate parent
132 * node locked thus preventing deadlock. As lockdep doesn't know this, use
133 * subclass to avoid triggering lockdep warning in such cases.
135 * The key is set by the readpage_end_io_hook after the buffer has passed
136 * csum validation but before the pages are unlocked. It is also set by
137 * btrfs_init_new_buffer on freshly allocated blocks.
139 * We also add a check to make sure the highest level of the tree is the
140 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
141 * needs update as well.
143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
144 # if BTRFS_MAX_LEVEL != 8
148 static struct btrfs_lockdep_keyset
{
149 u64 id
; /* root objectid */
150 const char *name_stem
; /* lock name stem */
151 char names
[BTRFS_MAX_LEVEL
+ 1][20];
152 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
153 } btrfs_lockdep_keysets
[] = {
154 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
155 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
156 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
157 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
158 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
159 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
160 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
161 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
162 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
163 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
164 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
165 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
166 { .id
= 0, .name_stem
= "tree" },
169 void __init
btrfs_init_lockdep(void)
173 /* initialize lockdep class names */
174 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
175 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
177 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
178 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
179 "btrfs-%s-%02d", ks
->name_stem
, j
);
183 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
186 struct btrfs_lockdep_keyset
*ks
;
188 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
190 /* find the matching keyset, id 0 is the default entry */
191 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
192 if (ks
->id
== objectid
)
195 lockdep_set_class_and_name(&eb
->lock
,
196 &ks
->keys
[level
], ks
->names
[level
]);
202 * extents on the btree inode are pretty simple, there's one extent
203 * that covers the entire device
205 struct extent_map
*btree_get_extent(struct btrfs_inode
*inode
,
206 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
209 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
210 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
211 struct extent_map
*em
;
214 read_lock(&em_tree
->lock
);
215 em
= lookup_extent_mapping(em_tree
, start
, len
);
217 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
218 read_unlock(&em_tree
->lock
);
221 read_unlock(&em_tree
->lock
);
223 em
= alloc_extent_map();
225 em
= ERR_PTR(-ENOMEM
);
230 em
->block_len
= (u64
)-1;
232 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
234 write_lock(&em_tree
->lock
);
235 ret
= add_extent_mapping(em_tree
, em
, 0);
236 if (ret
== -EEXIST
) {
238 em
= lookup_extent_mapping(em_tree
, start
, len
);
245 write_unlock(&em_tree
->lock
);
251 u32
btrfs_csum_data(const char *data
, u32 seed
, size_t len
)
253 return crc32c(seed
, data
, len
);
256 void btrfs_csum_final(u32 crc
, u8
*result
)
258 put_unaligned_le32(~crc
, result
);
262 * compute the csum for a btree block, and either verify it or write it
263 * into the csum field of the block.
265 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
266 struct extent_buffer
*buf
,
269 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
270 char result
[BTRFS_CSUM_SIZE
];
272 unsigned long cur_len
;
273 unsigned long offset
= BTRFS_CSUM_SIZE
;
275 unsigned long map_start
;
276 unsigned long map_len
;
280 len
= buf
->len
- offset
;
283 * Note: we don't need to check for the err == 1 case here, as
284 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
285 * and 'min_len = 32' and the currently implemented mapping
286 * algorithm we cannot cross a page boundary.
288 err
= map_private_extent_buffer(buf
, offset
, 32,
289 &kaddr
, &map_start
, &map_len
);
292 cur_len
= min(len
, map_len
- (offset
- map_start
));
293 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
298 memset(result
, 0, BTRFS_CSUM_SIZE
);
300 btrfs_csum_final(crc
, result
);
303 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
306 memcpy(&found
, result
, csum_size
);
308 read_extent_buffer(buf
, &val
, 0, csum_size
);
309 btrfs_warn_rl(fs_info
,
310 "%s checksum verify failed on %llu wanted %X found %X level %d",
311 fs_info
->sb
->s_id
, buf
->start
,
312 val
, found
, btrfs_header_level(buf
));
316 write_extent_buffer(buf
, result
, 0, csum_size
);
323 * we can't consider a given block up to date unless the transid of the
324 * block matches the transid in the parent node's pointer. This is how we
325 * detect blocks that either didn't get written at all or got written
326 * in the wrong place.
328 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
329 struct extent_buffer
*eb
, u64 parent_transid
,
332 struct extent_state
*cached_state
= NULL
;
334 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
336 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
343 btrfs_tree_read_lock(eb
);
344 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
347 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
349 if (extent_buffer_uptodate(eb
) &&
350 btrfs_header_generation(eb
) == parent_transid
) {
354 btrfs_err_rl(eb
->fs_info
,
355 "parent transid verify failed on %llu wanted %llu found %llu",
357 parent_transid
, btrfs_header_generation(eb
));
361 * Things reading via commit roots that don't have normal protection,
362 * like send, can have a really old block in cache that may point at a
363 * block that has been freed and re-allocated. So don't clear uptodate
364 * if we find an eb that is under IO (dirty/writeback) because we could
365 * end up reading in the stale data and then writing it back out and
366 * making everybody very sad.
368 if (!extent_buffer_under_io(eb
))
369 clear_extent_buffer_uptodate(eb
);
371 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
374 btrfs_tree_read_unlock_blocking(eb
);
379 * Return 0 if the superblock checksum type matches the checksum value of that
380 * algorithm. Pass the raw disk superblock data.
382 static int btrfs_check_super_csum(struct btrfs_fs_info
*fs_info
,
385 struct btrfs_super_block
*disk_sb
=
386 (struct btrfs_super_block
*)raw_disk_sb
;
387 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
390 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
392 char result
[sizeof(crc
)];
395 * The super_block structure does not span the whole
396 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
397 * is filled with zeros and is included in the checksum.
399 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
400 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
401 btrfs_csum_final(crc
, result
);
403 if (memcmp(raw_disk_sb
, result
, sizeof(result
)))
407 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
408 btrfs_err(fs_info
, "unsupported checksum algorithm %u",
416 static int verify_level_key(struct btrfs_fs_info
*fs_info
,
417 struct extent_buffer
*eb
, int level
,
418 struct btrfs_key
*first_key
, u64 parent_transid
)
421 struct btrfs_key found_key
;
424 found_level
= btrfs_header_level(eb
);
425 if (found_level
!= level
) {
426 #ifdef CONFIG_BTRFS_DEBUG
429 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
430 eb
->start
, level
, found_level
);
439 * For live tree block (new tree blocks in current transaction),
440 * we need proper lock context to avoid race, which is impossible here.
441 * So we only checks tree blocks which is read from disk, whose
442 * generation <= fs_info->last_trans_committed.
444 if (btrfs_header_generation(eb
) > fs_info
->last_trans_committed
)
447 btrfs_node_key_to_cpu(eb
, &found_key
, 0);
449 btrfs_item_key_to_cpu(eb
, &found_key
, 0);
450 ret
= btrfs_comp_cpu_keys(first_key
, &found_key
);
452 #ifdef CONFIG_BTRFS_DEBUG
456 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
457 eb
->start
, parent_transid
, first_key
->objectid
,
458 first_key
->type
, first_key
->offset
,
459 found_key
.objectid
, found_key
.type
,
467 * helper to read a given tree block, doing retries as required when
468 * the checksums don't match and we have alternate mirrors to try.
470 * @parent_transid: expected transid, skip check if 0
471 * @level: expected level, mandatory check
472 * @first_key: expected key of first slot, skip check if NULL
474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info
*fs_info
,
475 struct extent_buffer
*eb
,
476 u64 parent_transid
, int level
,
477 struct btrfs_key
*first_key
)
479 struct extent_io_tree
*io_tree
;
484 int failed_mirror
= 0;
486 io_tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
488 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
489 ret
= read_extent_buffer_pages(io_tree
, eb
, WAIT_COMPLETE
,
492 if (verify_parent_transid(io_tree
, eb
,
495 else if (verify_level_key(fs_info
, eb
, level
,
496 first_key
, parent_transid
))
502 num_copies
= btrfs_num_copies(fs_info
,
507 if (!failed_mirror
) {
509 failed_mirror
= eb
->read_mirror
;
513 if (mirror_num
== failed_mirror
)
516 if (mirror_num
> num_copies
)
520 if (failed
&& !ret
&& failed_mirror
)
521 repair_eb_io_failure(fs_info
, eb
, failed_mirror
);
527 * checksum a dirty tree block before IO. This has extra checks to make sure
528 * we only fill in the checksum field in the first page of a multi-page block
531 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
533 u64 start
= page_offset(page
);
535 struct extent_buffer
*eb
;
537 eb
= (struct extent_buffer
*)page
->private;
538 if (page
!= eb
->pages
[0])
541 found_start
= btrfs_header_bytenr(eb
);
543 * Please do not consolidate these warnings into a single if.
544 * It is useful to know what went wrong.
546 if (WARN_ON(found_start
!= start
))
548 if (WARN_ON(!PageUptodate(page
)))
551 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fs_devices
->metadata_uuid
,
552 btrfs_header_fsid(), BTRFS_FSID_SIZE
) == 0);
554 return csum_tree_block(fs_info
, eb
, 0);
557 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
558 struct extent_buffer
*eb
)
560 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
561 u8 fsid
[BTRFS_FSID_SIZE
];
564 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
569 * Checking the incompat flag is only valid for the current
570 * fs. For seed devices it's forbidden to have their uuid
571 * changed so reading ->fsid in this case is fine
573 if (fs_devices
== fs_info
->fs_devices
&&
574 btrfs_fs_incompat(fs_info
, METADATA_UUID
))
575 metadata_uuid
= fs_devices
->metadata_uuid
;
577 metadata_uuid
= fs_devices
->fsid
;
579 if (!memcmp(fsid
, metadata_uuid
, BTRFS_FSID_SIZE
)) {
583 fs_devices
= fs_devices
->seed
;
588 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
589 u64 phy_offset
, struct page
*page
,
590 u64 start
, u64 end
, int mirror
)
594 struct extent_buffer
*eb
;
595 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
596 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
603 eb
= (struct extent_buffer
*)page
->private;
605 /* the pending IO might have been the only thing that kept this buffer
606 * in memory. Make sure we have a ref for all this other checks
608 extent_buffer_get(eb
);
610 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
614 eb
->read_mirror
= mirror
;
615 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
620 found_start
= btrfs_header_bytenr(eb
);
621 if (found_start
!= eb
->start
) {
622 btrfs_err_rl(fs_info
, "bad tree block start, want %llu have %llu",
623 eb
->start
, found_start
);
627 if (check_tree_block_fsid(fs_info
, eb
)) {
628 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
633 found_level
= btrfs_header_level(eb
);
634 if (found_level
>= BTRFS_MAX_LEVEL
) {
635 btrfs_err(fs_info
, "bad tree block level %d on %llu",
636 (int)btrfs_header_level(eb
), eb
->start
);
641 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
644 ret
= csum_tree_block(fs_info
, eb
, 1);
649 * If this is a leaf block and it is corrupt, set the corrupt bit so
650 * that we don't try and read the other copies of this block, just
653 if (found_level
== 0 && btrfs_check_leaf_full(fs_info
, eb
)) {
654 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
658 if (found_level
> 0 && btrfs_check_node(fs_info
, eb
))
662 set_extent_buffer_uptodate(eb
);
665 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
666 btree_readahead_hook(eb
, ret
);
670 * our io error hook is going to dec the io pages
671 * again, we have to make sure it has something
674 atomic_inc(&eb
->io_pages
);
675 clear_extent_buffer_uptodate(eb
);
677 free_extent_buffer(eb
);
682 static void end_workqueue_bio(struct bio
*bio
)
684 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
685 struct btrfs_fs_info
*fs_info
;
686 struct btrfs_workqueue
*wq
;
687 btrfs_work_func_t func
;
689 fs_info
= end_io_wq
->info
;
690 end_io_wq
->status
= bio
->bi_status
;
692 if (bio_op(bio
) == REQ_OP_WRITE
) {
693 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
694 wq
= fs_info
->endio_meta_write_workers
;
695 func
= btrfs_endio_meta_write_helper
;
696 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
697 wq
= fs_info
->endio_freespace_worker
;
698 func
= btrfs_freespace_write_helper
;
699 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
700 wq
= fs_info
->endio_raid56_workers
;
701 func
= btrfs_endio_raid56_helper
;
703 wq
= fs_info
->endio_write_workers
;
704 func
= btrfs_endio_write_helper
;
707 if (unlikely(end_io_wq
->metadata
==
708 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
709 wq
= fs_info
->endio_repair_workers
;
710 func
= btrfs_endio_repair_helper
;
711 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
712 wq
= fs_info
->endio_raid56_workers
;
713 func
= btrfs_endio_raid56_helper
;
714 } else if (end_io_wq
->metadata
) {
715 wq
= fs_info
->endio_meta_workers
;
716 func
= btrfs_endio_meta_helper
;
718 wq
= fs_info
->endio_workers
;
719 func
= btrfs_endio_helper
;
723 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
724 btrfs_queue_work(wq
, &end_io_wq
->work
);
727 blk_status_t
btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
728 enum btrfs_wq_endio_type metadata
)
730 struct btrfs_end_io_wq
*end_io_wq
;
732 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
734 return BLK_STS_RESOURCE
;
736 end_io_wq
->private = bio
->bi_private
;
737 end_io_wq
->end_io
= bio
->bi_end_io
;
738 end_io_wq
->info
= info
;
739 end_io_wq
->status
= 0;
740 end_io_wq
->bio
= bio
;
741 end_io_wq
->metadata
= metadata
;
743 bio
->bi_private
= end_io_wq
;
744 bio
->bi_end_io
= end_workqueue_bio
;
748 static void run_one_async_start(struct btrfs_work
*work
)
750 struct async_submit_bio
*async
;
753 async
= container_of(work
, struct async_submit_bio
, work
);
754 ret
= async
->submit_bio_start(async
->private_data
, async
->bio
,
761 * In order to insert checksums into the metadata in large chunks, we wait
762 * until bio submission time. All the pages in the bio are checksummed and
763 * sums are attached onto the ordered extent record.
765 * At IO completion time the csums attached on the ordered extent record are
766 * inserted into the tree.
768 static void run_one_async_done(struct btrfs_work
*work
)
770 struct async_submit_bio
*async
;
774 async
= container_of(work
, struct async_submit_bio
, work
);
775 inode
= async
->private_data
;
777 /* If an error occurred we just want to clean up the bio and move on */
779 async
->bio
->bi_status
= async
->status
;
780 bio_endio(async
->bio
);
784 ret
= btrfs_map_bio(btrfs_sb(inode
->i_sb
), async
->bio
,
785 async
->mirror_num
, 1);
787 async
->bio
->bi_status
= ret
;
788 bio_endio(async
->bio
);
792 static void run_one_async_free(struct btrfs_work
*work
)
794 struct async_submit_bio
*async
;
796 async
= container_of(work
, struct async_submit_bio
, work
);
800 blk_status_t
btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
801 int mirror_num
, unsigned long bio_flags
,
802 u64 bio_offset
, void *private_data
,
803 extent_submit_bio_start_t
*submit_bio_start
)
805 struct async_submit_bio
*async
;
807 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
809 return BLK_STS_RESOURCE
;
811 async
->private_data
= private_data
;
813 async
->mirror_num
= mirror_num
;
814 async
->submit_bio_start
= submit_bio_start
;
816 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
817 run_one_async_done
, run_one_async_free
);
819 async
->bio_offset
= bio_offset
;
823 if (op_is_sync(bio
->bi_opf
))
824 btrfs_set_work_high_priority(&async
->work
);
826 btrfs_queue_work(fs_info
->workers
, &async
->work
);
830 static blk_status_t
btree_csum_one_bio(struct bio
*bio
)
832 struct bio_vec
*bvec
;
833 struct btrfs_root
*root
;
835 struct bvec_iter_all iter_all
;
837 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
838 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
839 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
840 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
845 return errno_to_blk_status(ret
);
848 static blk_status_t
btree_submit_bio_start(void *private_data
, struct bio
*bio
,
852 * when we're called for a write, we're already in the async
853 * submission context. Just jump into btrfs_map_bio
855 return btree_csum_one_bio(bio
);
858 static int check_async_write(struct btrfs_inode
*bi
)
860 if (atomic_read(&bi
->sync_writers
))
863 if (static_cpu_has(X86_FEATURE_XMM4_2
))
869 static blk_status_t
btree_submit_bio_hook(void *private_data
, struct bio
*bio
,
870 int mirror_num
, unsigned long bio_flags
,
873 struct inode
*inode
= private_data
;
874 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
875 int async
= check_async_write(BTRFS_I(inode
));
878 if (bio_op(bio
) != REQ_OP_WRITE
) {
880 * called for a read, do the setup so that checksum validation
881 * can happen in the async kernel threads
883 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
884 BTRFS_WQ_ENDIO_METADATA
);
887 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
889 ret
= btree_csum_one_bio(bio
);
892 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
895 * kthread helpers are used to submit writes so that
896 * checksumming can happen in parallel across all CPUs
898 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, 0,
899 bio_offset
, private_data
,
900 btree_submit_bio_start
);
908 bio
->bi_status
= ret
;
913 #ifdef CONFIG_MIGRATION
914 static int btree_migratepage(struct address_space
*mapping
,
915 struct page
*newpage
, struct page
*page
,
916 enum migrate_mode mode
)
919 * we can't safely write a btree page from here,
920 * we haven't done the locking hook
925 * Buffers may be managed in a filesystem specific way.
926 * We must have no buffers or drop them.
928 if (page_has_private(page
) &&
929 !try_to_release_page(page
, GFP_KERNEL
))
931 return migrate_page(mapping
, newpage
, page
, mode
);
936 static int btree_writepages(struct address_space
*mapping
,
937 struct writeback_control
*wbc
)
939 struct btrfs_fs_info
*fs_info
;
942 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
944 if (wbc
->for_kupdate
)
947 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
948 /* this is a bit racy, but that's ok */
949 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
950 BTRFS_DIRTY_METADATA_THRESH
,
951 fs_info
->dirty_metadata_batch
);
955 return btree_write_cache_pages(mapping
, wbc
);
958 static int btree_readpage(struct file
*file
, struct page
*page
)
960 struct extent_io_tree
*tree
;
961 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
962 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
965 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
967 if (PageWriteback(page
) || PageDirty(page
))
970 return try_release_extent_buffer(page
);
973 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
976 struct extent_io_tree
*tree
;
977 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
978 extent_invalidatepage(tree
, page
, offset
);
979 btree_releasepage(page
, GFP_NOFS
);
980 if (PagePrivate(page
)) {
981 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
982 "page private not zero on page %llu",
983 (unsigned long long)page_offset(page
));
984 ClearPagePrivate(page
);
985 set_page_private(page
, 0);
990 static int btree_set_page_dirty(struct page
*page
)
993 struct extent_buffer
*eb
;
995 BUG_ON(!PagePrivate(page
));
996 eb
= (struct extent_buffer
*)page
->private;
998 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
999 BUG_ON(!atomic_read(&eb
->refs
));
1000 btrfs_assert_tree_locked(eb
);
1002 return __set_page_dirty_nobuffers(page
);
1005 static const struct address_space_operations btree_aops
= {
1006 .readpage
= btree_readpage
,
1007 .writepages
= btree_writepages
,
1008 .releasepage
= btree_releasepage
,
1009 .invalidatepage
= btree_invalidatepage
,
1010 #ifdef CONFIG_MIGRATION
1011 .migratepage
= btree_migratepage
,
1013 .set_page_dirty
= btree_set_page_dirty
,
1016 void readahead_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
1018 struct extent_buffer
*buf
= NULL
;
1019 struct inode
*btree_inode
= fs_info
->btree_inode
;
1021 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1024 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1026 free_extent_buffer(buf
);
1029 int reada_tree_block_flagged(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1030 int mirror_num
, struct extent_buffer
**eb
)
1032 struct extent_buffer
*buf
= NULL
;
1033 struct inode
*btree_inode
= fs_info
->btree_inode
;
1034 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1037 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1041 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1043 ret
= read_extent_buffer_pages(io_tree
, buf
, WAIT_PAGE_LOCK
,
1046 free_extent_buffer(buf
);
1050 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1051 free_extent_buffer(buf
);
1053 } else if (extent_buffer_uptodate(buf
)) {
1056 free_extent_buffer(buf
);
1061 struct extent_buffer
*btrfs_find_create_tree_block(
1062 struct btrfs_fs_info
*fs_info
,
1065 if (btrfs_is_testing(fs_info
))
1066 return alloc_test_extent_buffer(fs_info
, bytenr
);
1067 return alloc_extent_buffer(fs_info
, bytenr
);
1071 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1073 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1074 buf
->start
+ buf
->len
- 1);
1077 void btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1079 filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1080 buf
->start
, buf
->start
+ buf
->len
- 1);
1084 * Read tree block at logical address @bytenr and do variant basic but critical
1087 * @parent_transid: expected transid of this tree block, skip check if 0
1088 * @level: expected level, mandatory check
1089 * @first_key: expected key in slot 0, skip check if NULL
1091 struct extent_buffer
*read_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1092 u64 parent_transid
, int level
,
1093 struct btrfs_key
*first_key
)
1095 struct extent_buffer
*buf
= NULL
;
1098 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1102 ret
= btree_read_extent_buffer_pages(fs_info
, buf
, parent_transid
,
1105 free_extent_buffer(buf
);
1106 return ERR_PTR(ret
);
1112 void clean_tree_block(struct btrfs_fs_info
*fs_info
,
1113 struct extent_buffer
*buf
)
1115 if (btrfs_header_generation(buf
) ==
1116 fs_info
->running_transaction
->transid
) {
1117 btrfs_assert_tree_locked(buf
);
1119 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1120 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
1122 fs_info
->dirty_metadata_batch
);
1123 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1124 btrfs_set_lock_blocking(buf
);
1125 clear_extent_buffer_dirty(buf
);
1130 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1132 struct btrfs_subvolume_writers
*writers
;
1135 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1137 return ERR_PTR(-ENOMEM
);
1139 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_NOFS
);
1142 return ERR_PTR(ret
);
1145 init_waitqueue_head(&writers
->wait
);
1150 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1152 percpu_counter_destroy(&writers
->counter
);
1156 static void __setup_root(struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1159 bool dummy
= test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO
, &fs_info
->fs_state
);
1161 root
->commit_root
= NULL
;
1163 root
->orphan_cleanup_state
= 0;
1165 root
->last_trans
= 0;
1166 root
->highest_objectid
= 0;
1167 root
->nr_delalloc_inodes
= 0;
1168 root
->nr_ordered_extents
= 0;
1169 root
->inode_tree
= RB_ROOT
;
1170 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1171 root
->block_rsv
= NULL
;
1173 INIT_LIST_HEAD(&root
->dirty_list
);
1174 INIT_LIST_HEAD(&root
->root_list
);
1175 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1176 INIT_LIST_HEAD(&root
->delalloc_root
);
1177 INIT_LIST_HEAD(&root
->ordered_extents
);
1178 INIT_LIST_HEAD(&root
->ordered_root
);
1179 INIT_LIST_HEAD(&root
->logged_list
[0]);
1180 INIT_LIST_HEAD(&root
->logged_list
[1]);
1181 spin_lock_init(&root
->inode_lock
);
1182 spin_lock_init(&root
->delalloc_lock
);
1183 spin_lock_init(&root
->ordered_extent_lock
);
1184 spin_lock_init(&root
->accounting_lock
);
1185 spin_lock_init(&root
->log_extents_lock
[0]);
1186 spin_lock_init(&root
->log_extents_lock
[1]);
1187 spin_lock_init(&root
->qgroup_meta_rsv_lock
);
1188 mutex_init(&root
->objectid_mutex
);
1189 mutex_init(&root
->log_mutex
);
1190 mutex_init(&root
->ordered_extent_mutex
);
1191 mutex_init(&root
->delalloc_mutex
);
1192 init_waitqueue_head(&root
->log_writer_wait
);
1193 init_waitqueue_head(&root
->log_commit_wait
[0]);
1194 init_waitqueue_head(&root
->log_commit_wait
[1]);
1195 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1196 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1197 atomic_set(&root
->log_commit
[0], 0);
1198 atomic_set(&root
->log_commit
[1], 0);
1199 atomic_set(&root
->log_writers
, 0);
1200 atomic_set(&root
->log_batch
, 0);
1201 refcount_set(&root
->refs
, 1);
1202 atomic_set(&root
->will_be_snapshotted
, 0);
1203 atomic_set(&root
->snapshot_force_cow
, 0);
1204 atomic_set(&root
->nr_swapfiles
, 0);
1205 root
->log_transid
= 0;
1206 root
->log_transid_committed
= -1;
1207 root
->last_log_commit
= 0;
1209 extent_io_tree_init(&root
->dirty_log_pages
, NULL
);
1211 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1212 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1213 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1215 root
->defrag_trans_start
= fs_info
->generation
;
1217 root
->defrag_trans_start
= 0;
1218 root
->root_key
.objectid
= objectid
;
1221 spin_lock_init(&root
->root_item_lock
);
1224 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1227 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1229 root
->fs_info
= fs_info
;
1233 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1234 /* Should only be used by the testing infrastructure */
1235 struct btrfs_root
*btrfs_alloc_dummy_root(struct btrfs_fs_info
*fs_info
)
1237 struct btrfs_root
*root
;
1240 return ERR_PTR(-EINVAL
);
1242 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1244 return ERR_PTR(-ENOMEM
);
1246 /* We don't use the stripesize in selftest, set it as sectorsize */
1247 __setup_root(root
, fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1248 root
->alloc_bytenr
= 0;
1254 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1255 struct btrfs_fs_info
*fs_info
,
1258 struct extent_buffer
*leaf
;
1259 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1260 struct btrfs_root
*root
;
1261 struct btrfs_key key
;
1263 uuid_le uuid
= NULL_UUID_LE
;
1265 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1267 return ERR_PTR(-ENOMEM
);
1269 __setup_root(root
, fs_info
, objectid
);
1270 root
->root_key
.objectid
= objectid
;
1271 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1272 root
->root_key
.offset
= 0;
1274 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1276 ret
= PTR_ERR(leaf
);
1282 btrfs_mark_buffer_dirty(leaf
);
1284 root
->commit_root
= btrfs_root_node(root
);
1285 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1287 root
->root_item
.flags
= 0;
1288 root
->root_item
.byte_limit
= 0;
1289 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1290 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1291 btrfs_set_root_level(&root
->root_item
, 0);
1292 btrfs_set_root_refs(&root
->root_item
, 1);
1293 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1294 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1295 btrfs_set_root_dirid(&root
->root_item
, 0);
1296 if (is_fstree(objectid
))
1298 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1299 root
->root_item
.drop_level
= 0;
1301 key
.objectid
= objectid
;
1302 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1304 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1308 btrfs_tree_unlock(leaf
);
1314 btrfs_tree_unlock(leaf
);
1315 free_extent_buffer(root
->commit_root
);
1316 free_extent_buffer(leaf
);
1320 return ERR_PTR(ret
);
1323 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1324 struct btrfs_fs_info
*fs_info
)
1326 struct btrfs_root
*root
;
1327 struct extent_buffer
*leaf
;
1329 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1331 return ERR_PTR(-ENOMEM
);
1333 __setup_root(root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1335 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1336 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1337 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1340 * DON'T set REF_COWS for log trees
1342 * log trees do not get reference counted because they go away
1343 * before a real commit is actually done. They do store pointers
1344 * to file data extents, and those reference counts still get
1345 * updated (along with back refs to the log tree).
1348 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1352 return ERR_CAST(leaf
);
1357 btrfs_mark_buffer_dirty(root
->node
);
1358 btrfs_tree_unlock(root
->node
);
1362 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1363 struct btrfs_fs_info
*fs_info
)
1365 struct btrfs_root
*log_root
;
1367 log_root
= alloc_log_tree(trans
, fs_info
);
1368 if (IS_ERR(log_root
))
1369 return PTR_ERR(log_root
);
1370 WARN_ON(fs_info
->log_root_tree
);
1371 fs_info
->log_root_tree
= log_root
;
1375 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1376 struct btrfs_root
*root
)
1378 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1379 struct btrfs_root
*log_root
;
1380 struct btrfs_inode_item
*inode_item
;
1382 log_root
= alloc_log_tree(trans
, fs_info
);
1383 if (IS_ERR(log_root
))
1384 return PTR_ERR(log_root
);
1386 log_root
->last_trans
= trans
->transid
;
1387 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1389 inode_item
= &log_root
->root_item
.inode
;
1390 btrfs_set_stack_inode_generation(inode_item
, 1);
1391 btrfs_set_stack_inode_size(inode_item
, 3);
1392 btrfs_set_stack_inode_nlink(inode_item
, 1);
1393 btrfs_set_stack_inode_nbytes(inode_item
,
1395 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1397 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1399 WARN_ON(root
->log_root
);
1400 root
->log_root
= log_root
;
1401 root
->log_transid
= 0;
1402 root
->log_transid_committed
= -1;
1403 root
->last_log_commit
= 0;
1407 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1408 struct btrfs_key
*key
)
1410 struct btrfs_root
*root
;
1411 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1412 struct btrfs_path
*path
;
1417 path
= btrfs_alloc_path();
1419 return ERR_PTR(-ENOMEM
);
1421 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1427 __setup_root(root
, fs_info
, key
->objectid
);
1429 ret
= btrfs_find_root(tree_root
, key
, path
,
1430 &root
->root_item
, &root
->root_key
);
1437 generation
= btrfs_root_generation(&root
->root_item
);
1438 level
= btrfs_root_level(&root
->root_item
);
1439 root
->node
= read_tree_block(fs_info
,
1440 btrfs_root_bytenr(&root
->root_item
),
1441 generation
, level
, NULL
);
1442 if (IS_ERR(root
->node
)) {
1443 ret
= PTR_ERR(root
->node
);
1445 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1447 free_extent_buffer(root
->node
);
1450 root
->commit_root
= btrfs_root_node(root
);
1452 btrfs_free_path(path
);
1458 root
= ERR_PTR(ret
);
1462 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1463 struct btrfs_key
*location
)
1465 struct btrfs_root
*root
;
1467 root
= btrfs_read_tree_root(tree_root
, location
);
1471 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1472 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1473 btrfs_check_and_init_root_item(&root
->root_item
);
1479 int btrfs_init_fs_root(struct btrfs_root
*root
)
1482 struct btrfs_subvolume_writers
*writers
;
1484 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1485 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1487 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1492 writers
= btrfs_alloc_subvolume_writers();
1493 if (IS_ERR(writers
)) {
1494 ret
= PTR_ERR(writers
);
1497 root
->subv_writers
= writers
;
1499 btrfs_init_free_ino_ctl(root
);
1500 spin_lock_init(&root
->ino_cache_lock
);
1501 init_waitqueue_head(&root
->ino_cache_wait
);
1503 ret
= get_anon_bdev(&root
->anon_dev
);
1507 mutex_lock(&root
->objectid_mutex
);
1508 ret
= btrfs_find_highest_objectid(root
,
1509 &root
->highest_objectid
);
1511 mutex_unlock(&root
->objectid_mutex
);
1515 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1517 mutex_unlock(&root
->objectid_mutex
);
1521 /* The caller is responsible to call btrfs_free_fs_root */
1525 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1528 struct btrfs_root
*root
;
1530 spin_lock(&fs_info
->fs_roots_radix_lock
);
1531 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1532 (unsigned long)root_id
);
1533 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1537 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1538 struct btrfs_root
*root
)
1542 ret
= radix_tree_preload(GFP_NOFS
);
1546 spin_lock(&fs_info
->fs_roots_radix_lock
);
1547 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1548 (unsigned long)root
->root_key
.objectid
,
1551 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1552 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1553 radix_tree_preload_end();
1558 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1559 struct btrfs_key
*location
,
1562 struct btrfs_root
*root
;
1563 struct btrfs_path
*path
;
1564 struct btrfs_key key
;
1567 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1568 return fs_info
->tree_root
;
1569 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1570 return fs_info
->extent_root
;
1571 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1572 return fs_info
->chunk_root
;
1573 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1574 return fs_info
->dev_root
;
1575 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1576 return fs_info
->csum_root
;
1577 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1578 return fs_info
->quota_root
? fs_info
->quota_root
:
1580 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1581 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1583 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1584 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1587 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1589 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1590 return ERR_PTR(-ENOENT
);
1594 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1598 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1603 ret
= btrfs_init_fs_root(root
);
1607 path
= btrfs_alloc_path();
1612 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1613 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1614 key
.offset
= location
->objectid
;
1616 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1617 btrfs_free_path(path
);
1621 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1623 ret
= btrfs_insert_fs_root(fs_info
, root
);
1625 if (ret
== -EEXIST
) {
1626 btrfs_free_fs_root(root
);
1633 btrfs_free_fs_root(root
);
1634 return ERR_PTR(ret
);
1637 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1639 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1641 struct btrfs_device
*device
;
1642 struct backing_dev_info
*bdi
;
1645 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1648 bdi
= device
->bdev
->bd_bdi
;
1649 if (bdi_congested(bdi
, bdi_bits
)) {
1659 * called by the kthread helper functions to finally call the bio end_io
1660 * functions. This is where read checksum verification actually happens
1662 static void end_workqueue_fn(struct btrfs_work
*work
)
1665 struct btrfs_end_io_wq
*end_io_wq
;
1667 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1668 bio
= end_io_wq
->bio
;
1670 bio
->bi_status
= end_io_wq
->status
;
1671 bio
->bi_private
= end_io_wq
->private;
1672 bio
->bi_end_io
= end_io_wq
->end_io
;
1673 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1677 static int cleaner_kthread(void *arg
)
1679 struct btrfs_root
*root
= arg
;
1680 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1686 set_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1688 /* Make the cleaner go to sleep early. */
1689 if (btrfs_need_cleaner_sleep(fs_info
))
1693 * Do not do anything if we might cause open_ctree() to block
1694 * before we have finished mounting the filesystem.
1696 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1699 if (!mutex_trylock(&fs_info
->cleaner_mutex
))
1703 * Avoid the problem that we change the status of the fs
1704 * during the above check and trylock.
1706 if (btrfs_need_cleaner_sleep(fs_info
)) {
1707 mutex_unlock(&fs_info
->cleaner_mutex
);
1711 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
1712 btrfs_run_delayed_iputs(fs_info
);
1713 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
1715 again
= btrfs_clean_one_deleted_snapshot(root
);
1716 mutex_unlock(&fs_info
->cleaner_mutex
);
1719 * The defragger has dealt with the R/O remount and umount,
1720 * needn't do anything special here.
1722 btrfs_run_defrag_inodes(fs_info
);
1725 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1726 * with relocation (btrfs_relocate_chunk) and relocation
1727 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1728 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1729 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1730 * unused block groups.
1732 btrfs_delete_unused_bgs(fs_info
);
1734 clear_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
);
1735 if (kthread_should_park())
1737 if (kthread_should_stop())
1740 set_current_state(TASK_INTERRUPTIBLE
);
1742 __set_current_state(TASK_RUNNING
);
1747 static int transaction_kthread(void *arg
)
1749 struct btrfs_root
*root
= arg
;
1750 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1751 struct btrfs_trans_handle
*trans
;
1752 struct btrfs_transaction
*cur
;
1755 unsigned long delay
;
1759 cannot_commit
= false;
1760 delay
= HZ
* fs_info
->commit_interval
;
1761 mutex_lock(&fs_info
->transaction_kthread_mutex
);
1763 spin_lock(&fs_info
->trans_lock
);
1764 cur
= fs_info
->running_transaction
;
1766 spin_unlock(&fs_info
->trans_lock
);
1770 now
= ktime_get_seconds();
1771 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1772 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT
, &fs_info
->flags
) &&
1773 (now
< cur
->start_time
||
1774 now
- cur
->start_time
< fs_info
->commit_interval
)) {
1775 spin_unlock(&fs_info
->trans_lock
);
1779 transid
= cur
->transid
;
1780 spin_unlock(&fs_info
->trans_lock
);
1782 /* If the file system is aborted, this will always fail. */
1783 trans
= btrfs_attach_transaction(root
);
1784 if (IS_ERR(trans
)) {
1785 if (PTR_ERR(trans
) != -ENOENT
)
1786 cannot_commit
= true;
1789 if (transid
== trans
->transid
) {
1790 btrfs_commit_transaction(trans
);
1792 btrfs_end_transaction(trans
);
1795 wake_up_process(fs_info
->cleaner_kthread
);
1796 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
1798 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1799 &fs_info
->fs_state
)))
1800 btrfs_cleanup_transaction(fs_info
);
1801 if (!kthread_should_stop() &&
1802 (!btrfs_transaction_blocked(fs_info
) ||
1804 schedule_timeout_interruptible(delay
);
1805 } while (!kthread_should_stop());
1810 * this will find the highest generation in the array of
1811 * root backups. The index of the highest array is returned,
1812 * or -1 if we can't find anything.
1814 * We check to make sure the array is valid by comparing the
1815 * generation of the latest root in the array with the generation
1816 * in the super block. If they don't match we pitch it.
1818 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1821 int newest_index
= -1;
1822 struct btrfs_root_backup
*root_backup
;
1825 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1826 root_backup
= info
->super_copy
->super_roots
+ i
;
1827 cur
= btrfs_backup_tree_root_gen(root_backup
);
1828 if (cur
== newest_gen
)
1832 /* check to see if we actually wrapped around */
1833 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1834 root_backup
= info
->super_copy
->super_roots
;
1835 cur
= btrfs_backup_tree_root_gen(root_backup
);
1836 if (cur
== newest_gen
)
1839 return newest_index
;
1844 * find the oldest backup so we know where to store new entries
1845 * in the backup array. This will set the backup_root_index
1846 * field in the fs_info struct
1848 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1851 int newest_index
= -1;
1853 newest_index
= find_newest_super_backup(info
, newest_gen
);
1854 /* if there was garbage in there, just move along */
1855 if (newest_index
== -1) {
1856 info
->backup_root_index
= 0;
1858 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1863 * copy all the root pointers into the super backup array.
1864 * this will bump the backup pointer by one when it is
1867 static void backup_super_roots(struct btrfs_fs_info
*info
)
1870 struct btrfs_root_backup
*root_backup
;
1873 next_backup
= info
->backup_root_index
;
1874 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1875 BTRFS_NUM_BACKUP_ROOTS
;
1878 * just overwrite the last backup if we're at the same generation
1879 * this happens only at umount
1881 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1882 if (btrfs_backup_tree_root_gen(root_backup
) ==
1883 btrfs_header_generation(info
->tree_root
->node
))
1884 next_backup
= last_backup
;
1886 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1889 * make sure all of our padding and empty slots get zero filled
1890 * regardless of which ones we use today
1892 memset(root_backup
, 0, sizeof(*root_backup
));
1894 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1896 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1897 btrfs_set_backup_tree_root_gen(root_backup
,
1898 btrfs_header_generation(info
->tree_root
->node
));
1900 btrfs_set_backup_tree_root_level(root_backup
,
1901 btrfs_header_level(info
->tree_root
->node
));
1903 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1904 btrfs_set_backup_chunk_root_gen(root_backup
,
1905 btrfs_header_generation(info
->chunk_root
->node
));
1906 btrfs_set_backup_chunk_root_level(root_backup
,
1907 btrfs_header_level(info
->chunk_root
->node
));
1909 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1910 btrfs_set_backup_extent_root_gen(root_backup
,
1911 btrfs_header_generation(info
->extent_root
->node
));
1912 btrfs_set_backup_extent_root_level(root_backup
,
1913 btrfs_header_level(info
->extent_root
->node
));
1916 * we might commit during log recovery, which happens before we set
1917 * the fs_root. Make sure it is valid before we fill it in.
1919 if (info
->fs_root
&& info
->fs_root
->node
) {
1920 btrfs_set_backup_fs_root(root_backup
,
1921 info
->fs_root
->node
->start
);
1922 btrfs_set_backup_fs_root_gen(root_backup
,
1923 btrfs_header_generation(info
->fs_root
->node
));
1924 btrfs_set_backup_fs_root_level(root_backup
,
1925 btrfs_header_level(info
->fs_root
->node
));
1928 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1929 btrfs_set_backup_dev_root_gen(root_backup
,
1930 btrfs_header_generation(info
->dev_root
->node
));
1931 btrfs_set_backup_dev_root_level(root_backup
,
1932 btrfs_header_level(info
->dev_root
->node
));
1934 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1935 btrfs_set_backup_csum_root_gen(root_backup
,
1936 btrfs_header_generation(info
->csum_root
->node
));
1937 btrfs_set_backup_csum_root_level(root_backup
,
1938 btrfs_header_level(info
->csum_root
->node
));
1940 btrfs_set_backup_total_bytes(root_backup
,
1941 btrfs_super_total_bytes(info
->super_copy
));
1942 btrfs_set_backup_bytes_used(root_backup
,
1943 btrfs_super_bytes_used(info
->super_copy
));
1944 btrfs_set_backup_num_devices(root_backup
,
1945 btrfs_super_num_devices(info
->super_copy
));
1948 * if we don't copy this out to the super_copy, it won't get remembered
1949 * for the next commit
1951 memcpy(&info
->super_copy
->super_roots
,
1952 &info
->super_for_commit
->super_roots
,
1953 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1957 * this copies info out of the root backup array and back into
1958 * the in-memory super block. It is meant to help iterate through
1959 * the array, so you send it the number of backups you've already
1960 * tried and the last backup index you used.
1962 * this returns -1 when it has tried all the backups
1964 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1965 struct btrfs_super_block
*super
,
1966 int *num_backups_tried
, int *backup_index
)
1968 struct btrfs_root_backup
*root_backup
;
1969 int newest
= *backup_index
;
1971 if (*num_backups_tried
== 0) {
1972 u64 gen
= btrfs_super_generation(super
);
1974 newest
= find_newest_super_backup(info
, gen
);
1978 *backup_index
= newest
;
1979 *num_backups_tried
= 1;
1980 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1981 /* we've tried all the backups, all done */
1984 /* jump to the next oldest backup */
1985 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1986 BTRFS_NUM_BACKUP_ROOTS
;
1987 *backup_index
= newest
;
1988 *num_backups_tried
+= 1;
1990 root_backup
= super
->super_roots
+ newest
;
1992 btrfs_set_super_generation(super
,
1993 btrfs_backup_tree_root_gen(root_backup
));
1994 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1995 btrfs_set_super_root_level(super
,
1996 btrfs_backup_tree_root_level(root_backup
));
1997 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2000 * fixme: the total bytes and num_devices need to match or we should
2003 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2004 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2008 /* helper to cleanup workers */
2009 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2011 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2012 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2013 btrfs_destroy_workqueue(fs_info
->workers
);
2014 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2015 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2016 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2017 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2018 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2019 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2020 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2021 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2022 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2023 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2024 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2025 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2026 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2028 * Now that all other work queues are destroyed, we can safely destroy
2029 * the queues used for metadata I/O, since tasks from those other work
2030 * queues can do metadata I/O operations.
2032 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2033 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2036 static void free_root_extent_buffers(struct btrfs_root
*root
)
2039 free_extent_buffer(root
->node
);
2040 free_extent_buffer(root
->commit_root
);
2042 root
->commit_root
= NULL
;
2046 /* helper to cleanup tree roots */
2047 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2049 free_root_extent_buffers(info
->tree_root
);
2051 free_root_extent_buffers(info
->dev_root
);
2052 free_root_extent_buffers(info
->extent_root
);
2053 free_root_extent_buffers(info
->csum_root
);
2054 free_root_extent_buffers(info
->quota_root
);
2055 free_root_extent_buffers(info
->uuid_root
);
2057 free_root_extent_buffers(info
->chunk_root
);
2058 free_root_extent_buffers(info
->free_space_root
);
2061 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2064 struct btrfs_root
*gang
[8];
2067 while (!list_empty(&fs_info
->dead_roots
)) {
2068 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2069 struct btrfs_root
, root_list
);
2070 list_del(&gang
[0]->root_list
);
2072 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2073 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2075 free_extent_buffer(gang
[0]->node
);
2076 free_extent_buffer(gang
[0]->commit_root
);
2077 btrfs_put_fs_root(gang
[0]);
2082 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2087 for (i
= 0; i
< ret
; i
++)
2088 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2091 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2092 btrfs_free_log_root_tree(NULL
, fs_info
);
2093 btrfs_destroy_pinned_extent(fs_info
, fs_info
->pinned_extents
);
2097 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2099 mutex_init(&fs_info
->scrub_lock
);
2100 atomic_set(&fs_info
->scrubs_running
, 0);
2101 atomic_set(&fs_info
->scrub_pause_req
, 0);
2102 atomic_set(&fs_info
->scrubs_paused
, 0);
2103 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2104 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2105 fs_info
->scrub_workers_refcnt
= 0;
2108 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2110 spin_lock_init(&fs_info
->balance_lock
);
2111 mutex_init(&fs_info
->balance_mutex
);
2112 atomic_set(&fs_info
->balance_pause_req
, 0);
2113 atomic_set(&fs_info
->balance_cancel_req
, 0);
2114 fs_info
->balance_ctl
= NULL
;
2115 init_waitqueue_head(&fs_info
->balance_wait_q
);
2118 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
)
2120 struct inode
*inode
= fs_info
->btree_inode
;
2122 inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2123 set_nlink(inode
, 1);
2125 * we set the i_size on the btree inode to the max possible int.
2126 * the real end of the address space is determined by all of
2127 * the devices in the system
2129 inode
->i_size
= OFFSET_MAX
;
2130 inode
->i_mapping
->a_ops
= &btree_aops
;
2132 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
2133 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
, inode
);
2134 BTRFS_I(inode
)->io_tree
.track_uptodate
= 0;
2135 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
);
2137 BTRFS_I(inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2139 BTRFS_I(inode
)->root
= fs_info
->tree_root
;
2140 memset(&BTRFS_I(inode
)->location
, 0, sizeof(struct btrfs_key
));
2141 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
2142 btrfs_insert_inode_hash(inode
);
2145 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2147 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2148 init_rwsem(&fs_info
->dev_replace
.rwsem
);
2149 init_waitqueue_head(&fs_info
->dev_replace
.replace_wait
);
2152 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2154 spin_lock_init(&fs_info
->qgroup_lock
);
2155 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2156 fs_info
->qgroup_tree
= RB_ROOT
;
2157 fs_info
->qgroup_op_tree
= RB_ROOT
;
2158 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2159 fs_info
->qgroup_seq
= 1;
2160 fs_info
->qgroup_ulist
= NULL
;
2161 fs_info
->qgroup_rescan_running
= false;
2162 mutex_init(&fs_info
->qgroup_rescan_lock
);
2165 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2166 struct btrfs_fs_devices
*fs_devices
)
2168 u32 max_active
= fs_info
->thread_pool_size
;
2169 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2172 btrfs_alloc_workqueue(fs_info
, "worker",
2173 flags
| WQ_HIGHPRI
, max_active
, 16);
2175 fs_info
->delalloc_workers
=
2176 btrfs_alloc_workqueue(fs_info
, "delalloc",
2177 flags
, max_active
, 2);
2179 fs_info
->flush_workers
=
2180 btrfs_alloc_workqueue(fs_info
, "flush_delalloc",
2181 flags
, max_active
, 0);
2183 fs_info
->caching_workers
=
2184 btrfs_alloc_workqueue(fs_info
, "cache", flags
, max_active
, 0);
2187 * a higher idle thresh on the submit workers makes it much more
2188 * likely that bios will be send down in a sane order to the
2191 fs_info
->submit_workers
=
2192 btrfs_alloc_workqueue(fs_info
, "submit", flags
,
2193 min_t(u64
, fs_devices
->num_devices
,
2196 fs_info
->fixup_workers
=
2197 btrfs_alloc_workqueue(fs_info
, "fixup", flags
, 1, 0);
2200 * endios are largely parallel and should have a very
2203 fs_info
->endio_workers
=
2204 btrfs_alloc_workqueue(fs_info
, "endio", flags
, max_active
, 4);
2205 fs_info
->endio_meta_workers
=
2206 btrfs_alloc_workqueue(fs_info
, "endio-meta", flags
,
2208 fs_info
->endio_meta_write_workers
=
2209 btrfs_alloc_workqueue(fs_info
, "endio-meta-write", flags
,
2211 fs_info
->endio_raid56_workers
=
2212 btrfs_alloc_workqueue(fs_info
, "endio-raid56", flags
,
2214 fs_info
->endio_repair_workers
=
2215 btrfs_alloc_workqueue(fs_info
, "endio-repair", flags
, 1, 0);
2216 fs_info
->rmw_workers
=
2217 btrfs_alloc_workqueue(fs_info
, "rmw", flags
, max_active
, 2);
2218 fs_info
->endio_write_workers
=
2219 btrfs_alloc_workqueue(fs_info
, "endio-write", flags
,
2221 fs_info
->endio_freespace_worker
=
2222 btrfs_alloc_workqueue(fs_info
, "freespace-write", flags
,
2224 fs_info
->delayed_workers
=
2225 btrfs_alloc_workqueue(fs_info
, "delayed-meta", flags
,
2227 fs_info
->readahead_workers
=
2228 btrfs_alloc_workqueue(fs_info
, "readahead", flags
,
2230 fs_info
->qgroup_rescan_workers
=
2231 btrfs_alloc_workqueue(fs_info
, "qgroup-rescan", flags
, 1, 0);
2232 fs_info
->extent_workers
=
2233 btrfs_alloc_workqueue(fs_info
, "extent-refs", flags
,
2234 min_t(u64
, fs_devices
->num_devices
,
2237 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2238 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2239 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2240 fs_info
->endio_meta_write_workers
&&
2241 fs_info
->endio_repair_workers
&&
2242 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2243 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2244 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2245 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2246 fs_info
->extent_workers
&&
2247 fs_info
->qgroup_rescan_workers
)) {
2254 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2255 struct btrfs_fs_devices
*fs_devices
)
2258 struct btrfs_root
*log_tree_root
;
2259 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2260 u64 bytenr
= btrfs_super_log_root(disk_super
);
2261 int level
= btrfs_super_log_root_level(disk_super
);
2263 if (fs_devices
->rw_devices
== 0) {
2264 btrfs_warn(fs_info
, "log replay required on RO media");
2268 log_tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2272 __setup_root(log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2274 log_tree_root
->node
= read_tree_block(fs_info
, bytenr
,
2275 fs_info
->generation
+ 1,
2277 if (IS_ERR(log_tree_root
->node
)) {
2278 btrfs_warn(fs_info
, "failed to read log tree");
2279 ret
= PTR_ERR(log_tree_root
->node
);
2280 kfree(log_tree_root
);
2282 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2283 btrfs_err(fs_info
, "failed to read log tree");
2284 free_extent_buffer(log_tree_root
->node
);
2285 kfree(log_tree_root
);
2288 /* returns with log_tree_root freed on success */
2289 ret
= btrfs_recover_log_trees(log_tree_root
);
2291 btrfs_handle_fs_error(fs_info
, ret
,
2292 "Failed to recover log tree");
2293 free_extent_buffer(log_tree_root
->node
);
2294 kfree(log_tree_root
);
2298 if (sb_rdonly(fs_info
->sb
)) {
2299 ret
= btrfs_commit_super(fs_info
);
2307 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
)
2309 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2310 struct btrfs_root
*root
;
2311 struct btrfs_key location
;
2314 BUG_ON(!fs_info
->tree_root
);
2316 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2317 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2318 location
.offset
= 0;
2320 root
= btrfs_read_tree_root(tree_root
, &location
);
2322 ret
= PTR_ERR(root
);
2325 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2326 fs_info
->extent_root
= root
;
2328 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2329 root
= btrfs_read_tree_root(tree_root
, &location
);
2331 ret
= PTR_ERR(root
);
2334 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2335 fs_info
->dev_root
= root
;
2336 btrfs_init_devices_late(fs_info
);
2338 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2339 root
= btrfs_read_tree_root(tree_root
, &location
);
2341 ret
= PTR_ERR(root
);
2344 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2345 fs_info
->csum_root
= root
;
2347 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2348 root
= btrfs_read_tree_root(tree_root
, &location
);
2349 if (!IS_ERR(root
)) {
2350 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2351 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2352 fs_info
->quota_root
= root
;
2355 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2356 root
= btrfs_read_tree_root(tree_root
, &location
);
2358 ret
= PTR_ERR(root
);
2362 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2363 fs_info
->uuid_root
= root
;
2366 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2367 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2368 root
= btrfs_read_tree_root(tree_root
, &location
);
2370 ret
= PTR_ERR(root
);
2373 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2374 fs_info
->free_space_root
= root
;
2379 btrfs_warn(fs_info
, "failed to read root (objectid=%llu): %d",
2380 location
.objectid
, ret
);
2385 * Real super block validation
2386 * NOTE: super csum type and incompat features will not be checked here.
2388 * @sb: super block to check
2389 * @mirror_num: the super block number to check its bytenr:
2390 * 0 the primary (1st) sb
2391 * 1, 2 2nd and 3rd backup copy
2392 * -1 skip bytenr check
2394 static int validate_super(struct btrfs_fs_info
*fs_info
,
2395 struct btrfs_super_block
*sb
, int mirror_num
)
2397 u64 nodesize
= btrfs_super_nodesize(sb
);
2398 u64 sectorsize
= btrfs_super_sectorsize(sb
);
2401 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
2402 btrfs_err(fs_info
, "no valid FS found");
2405 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
) {
2406 btrfs_err(fs_info
, "unrecognized or unsupported super flag: %llu",
2407 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
2410 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2411 btrfs_err(fs_info
, "tree_root level too big: %d >= %d",
2412 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
2415 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2416 btrfs_err(fs_info
, "chunk_root level too big: %d >= %d",
2417 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
2420 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
2421 btrfs_err(fs_info
, "log_root level too big: %d >= %d",
2422 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
2427 * Check sectorsize and nodesize first, other check will need it.
2428 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2430 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
2431 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2432 btrfs_err(fs_info
, "invalid sectorsize %llu", sectorsize
);
2435 /* Only PAGE SIZE is supported yet */
2436 if (sectorsize
!= PAGE_SIZE
) {
2438 "sectorsize %llu not supported yet, only support %lu",
2439 sectorsize
, PAGE_SIZE
);
2442 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
2443 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
2444 btrfs_err(fs_info
, "invalid nodesize %llu", nodesize
);
2447 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
2448 btrfs_err(fs_info
, "invalid leafsize %u, should be %llu",
2449 le32_to_cpu(sb
->__unused_leafsize
), nodesize
);
2453 /* Root alignment check */
2454 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
2455 btrfs_warn(fs_info
, "tree_root block unaligned: %llu",
2456 btrfs_super_root(sb
));
2459 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
2460 btrfs_warn(fs_info
, "chunk_root block unaligned: %llu",
2461 btrfs_super_chunk_root(sb
));
2464 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
2465 btrfs_warn(fs_info
, "log_root block unaligned: %llu",
2466 btrfs_super_log_root(sb
));
2470 if (memcmp(fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
,
2471 BTRFS_FSID_SIZE
) != 0) {
2473 "dev_item UUID does not match metadata fsid: %pU != %pU",
2474 fs_info
->fs_devices
->metadata_uuid
, sb
->dev_item
.fsid
);
2479 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2482 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
2483 btrfs_err(fs_info
, "bytes_used is too small %llu",
2484 btrfs_super_bytes_used(sb
));
2487 if (!is_power_of_2(btrfs_super_stripesize(sb
))) {
2488 btrfs_err(fs_info
, "invalid stripesize %u",
2489 btrfs_super_stripesize(sb
));
2492 if (btrfs_super_num_devices(sb
) > (1UL << 31))
2493 btrfs_warn(fs_info
, "suspicious number of devices: %llu",
2494 btrfs_super_num_devices(sb
));
2495 if (btrfs_super_num_devices(sb
) == 0) {
2496 btrfs_err(fs_info
, "number of devices is 0");
2500 if (mirror_num
>= 0 &&
2501 btrfs_super_bytenr(sb
) != btrfs_sb_offset(mirror_num
)) {
2502 btrfs_err(fs_info
, "super offset mismatch %llu != %u",
2503 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
2508 * Obvious sys_chunk_array corruptions, it must hold at least one key
2511 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
2512 btrfs_err(fs_info
, "system chunk array too big %u > %u",
2513 btrfs_super_sys_array_size(sb
),
2514 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
2517 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
2518 + sizeof(struct btrfs_chunk
)) {
2519 btrfs_err(fs_info
, "system chunk array too small %u < %zu",
2520 btrfs_super_sys_array_size(sb
),
2521 sizeof(struct btrfs_disk_key
)
2522 + sizeof(struct btrfs_chunk
));
2527 * The generation is a global counter, we'll trust it more than the others
2528 * but it's still possible that it's the one that's wrong.
2530 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
2532 "suspicious: generation < chunk_root_generation: %llu < %llu",
2533 btrfs_super_generation(sb
),
2534 btrfs_super_chunk_root_generation(sb
));
2535 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
2536 && btrfs_super_cache_generation(sb
) != (u64
)-1)
2538 "suspicious: generation < cache_generation: %llu < %llu",
2539 btrfs_super_generation(sb
),
2540 btrfs_super_cache_generation(sb
));
2546 * Validation of super block at mount time.
2547 * Some checks already done early at mount time, like csum type and incompat
2548 * flags will be skipped.
2550 static int btrfs_validate_mount_super(struct btrfs_fs_info
*fs_info
)
2552 return validate_super(fs_info
, fs_info
->super_copy
, 0);
2556 * Validation of super block at write time.
2557 * Some checks like bytenr check will be skipped as their values will be
2559 * Extra checks like csum type and incompat flags will be done here.
2561 static int btrfs_validate_write_super(struct btrfs_fs_info
*fs_info
,
2562 struct btrfs_super_block
*sb
)
2566 ret
= validate_super(fs_info
, sb
, -1);
2569 if (btrfs_super_csum_type(sb
) != BTRFS_CSUM_TYPE_CRC32
) {
2571 btrfs_err(fs_info
, "invalid csum type, has %u want %u",
2572 btrfs_super_csum_type(sb
), BTRFS_CSUM_TYPE_CRC32
);
2575 if (btrfs_super_incompat_flags(sb
) & ~BTRFS_FEATURE_INCOMPAT_SUPP
) {
2578 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2579 btrfs_super_incompat_flags(sb
),
2580 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP
);
2586 "super block corruption detected before writing it to disk");
2590 int open_ctree(struct super_block
*sb
,
2591 struct btrfs_fs_devices
*fs_devices
,
2599 struct btrfs_key location
;
2600 struct buffer_head
*bh
;
2601 struct btrfs_super_block
*disk_super
;
2602 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2603 struct btrfs_root
*tree_root
;
2604 struct btrfs_root
*chunk_root
;
2607 int num_backups_tried
= 0;
2608 int backup_index
= 0;
2609 int clear_free_space_tree
= 0;
2612 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2613 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2614 if (!tree_root
|| !chunk_root
) {
2619 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2625 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2630 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2631 (1 + ilog2(nr_cpu_ids
));
2633 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2636 goto fail_dirty_metadata_bytes
;
2639 ret
= percpu_counter_init(&fs_info
->dev_replace
.bio_counter
, 0,
2643 goto fail_delalloc_bytes
;
2646 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2647 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2648 INIT_LIST_HEAD(&fs_info
->trans_list
);
2649 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2650 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2651 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2652 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2653 INIT_LIST_HEAD(&fs_info
->pending_raid_kobjs
);
2654 spin_lock_init(&fs_info
->pending_raid_kobjs_lock
);
2655 spin_lock_init(&fs_info
->delalloc_root_lock
);
2656 spin_lock_init(&fs_info
->trans_lock
);
2657 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2658 spin_lock_init(&fs_info
->delayed_iput_lock
);
2659 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2660 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2661 spin_lock_init(&fs_info
->super_lock
);
2662 spin_lock_init(&fs_info
->qgroup_op_lock
);
2663 spin_lock_init(&fs_info
->buffer_lock
);
2664 spin_lock_init(&fs_info
->unused_bgs_lock
);
2665 rwlock_init(&fs_info
->tree_mod_log_lock
);
2666 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2667 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2668 mutex_init(&fs_info
->reloc_mutex
);
2669 mutex_init(&fs_info
->delalloc_root_mutex
);
2670 mutex_init(&fs_info
->cleaner_delayed_iput_mutex
);
2671 seqlock_init(&fs_info
->profiles_lock
);
2673 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2674 INIT_LIST_HEAD(&fs_info
->space_info
);
2675 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2676 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2677 btrfs_mapping_init(&fs_info
->mapping_tree
);
2678 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2679 BTRFS_BLOCK_RSV_GLOBAL
);
2680 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2681 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2682 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2683 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2684 BTRFS_BLOCK_RSV_DELOPS
);
2685 btrfs_init_block_rsv(&fs_info
->delayed_refs_rsv
,
2686 BTRFS_BLOCK_RSV_DELREFS
);
2688 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2689 atomic_set(&fs_info
->defrag_running
, 0);
2690 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2691 atomic_set(&fs_info
->reada_works_cnt
, 0);
2692 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2694 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2695 fs_info
->metadata_ratio
= 0;
2696 fs_info
->defrag_inodes
= RB_ROOT
;
2697 atomic64_set(&fs_info
->free_chunk_space
, 0);
2698 fs_info
->tree_mod_log
= RB_ROOT
;
2699 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2700 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2701 /* readahead state */
2702 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2703 spin_lock_init(&fs_info
->reada_lock
);
2704 btrfs_init_ref_verify(fs_info
);
2706 fs_info
->thread_pool_size
= min_t(unsigned long,
2707 num_online_cpus() + 2, 8);
2709 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2710 spin_lock_init(&fs_info
->ordered_root_lock
);
2712 fs_info
->btree_inode
= new_inode(sb
);
2713 if (!fs_info
->btree_inode
) {
2715 goto fail_bio_counter
;
2717 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2719 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2721 if (!fs_info
->delayed_root
) {
2725 btrfs_init_delayed_root(fs_info
->delayed_root
);
2727 btrfs_init_scrub(fs_info
);
2728 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2729 fs_info
->check_integrity_print_mask
= 0;
2731 btrfs_init_balance(fs_info
);
2732 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2734 sb
->s_blocksize
= BTRFS_BDEV_BLOCKSIZE
;
2735 sb
->s_blocksize_bits
= blksize_bits(BTRFS_BDEV_BLOCKSIZE
);
2737 btrfs_init_btree_inode(fs_info
);
2739 spin_lock_init(&fs_info
->block_group_cache_lock
);
2740 fs_info
->block_group_cache_tree
= RB_ROOT
;
2741 fs_info
->first_logical_byte
= (u64
)-1;
2743 extent_io_tree_init(&fs_info
->freed_extents
[0], NULL
);
2744 extent_io_tree_init(&fs_info
->freed_extents
[1], NULL
);
2745 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2746 set_bit(BTRFS_FS_BARRIER
, &fs_info
->flags
);
2748 mutex_init(&fs_info
->ordered_operations_mutex
);
2749 mutex_init(&fs_info
->tree_log_mutex
);
2750 mutex_init(&fs_info
->chunk_mutex
);
2751 mutex_init(&fs_info
->transaction_kthread_mutex
);
2752 mutex_init(&fs_info
->cleaner_mutex
);
2753 mutex_init(&fs_info
->ro_block_group_mutex
);
2754 init_rwsem(&fs_info
->commit_root_sem
);
2755 init_rwsem(&fs_info
->cleanup_work_sem
);
2756 init_rwsem(&fs_info
->subvol_sem
);
2757 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2759 btrfs_init_dev_replace_locks(fs_info
);
2760 btrfs_init_qgroup(fs_info
);
2762 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2763 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2765 init_waitqueue_head(&fs_info
->transaction_throttle
);
2766 init_waitqueue_head(&fs_info
->transaction_wait
);
2767 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2768 init_waitqueue_head(&fs_info
->async_submit_wait
);
2770 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2772 /* Usable values until the real ones are cached from the superblock */
2773 fs_info
->nodesize
= 4096;
2774 fs_info
->sectorsize
= 4096;
2775 fs_info
->stripesize
= 4096;
2777 spin_lock_init(&fs_info
->swapfile_pins_lock
);
2778 fs_info
->swapfile_pins
= RB_ROOT
;
2780 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2786 __setup_root(tree_root
, fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2788 invalidate_bdev(fs_devices
->latest_bdev
);
2791 * Read super block and check the signature bytes only
2793 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2800 * We want to check superblock checksum, the type is stored inside.
2801 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2803 if (btrfs_check_super_csum(fs_info
, bh
->b_data
)) {
2804 btrfs_err(fs_info
, "superblock checksum mismatch");
2811 * super_copy is zeroed at allocation time and we never touch the
2812 * following bytes up to INFO_SIZE, the checksum is calculated from
2813 * the whole block of INFO_SIZE
2815 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2818 disk_super
= fs_info
->super_copy
;
2820 ASSERT(!memcmp(fs_info
->fs_devices
->fsid
, fs_info
->super_copy
->fsid
,
2823 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
)) {
2824 ASSERT(!memcmp(fs_info
->fs_devices
->metadata_uuid
,
2825 fs_info
->super_copy
->metadata_uuid
,
2829 features
= btrfs_super_flags(disk_super
);
2830 if (features
& BTRFS_SUPER_FLAG_CHANGING_FSID_V2
) {
2831 features
&= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2
;
2832 btrfs_set_super_flags(disk_super
, features
);
2834 "found metadata UUID change in progress flag, clearing");
2837 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2838 sizeof(*fs_info
->super_for_commit
));
2840 ret
= btrfs_validate_mount_super(fs_info
);
2842 btrfs_err(fs_info
, "superblock contains fatal errors");
2847 if (!btrfs_super_root(disk_super
))
2850 /* check FS state, whether FS is broken. */
2851 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2852 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2855 * run through our array of backup supers and setup
2856 * our ring pointer to the oldest one
2858 generation
= btrfs_super_generation(disk_super
);
2859 find_oldest_super_backup(fs_info
, generation
);
2862 * In the long term, we'll store the compression type in the super
2863 * block, and it'll be used for per file compression control.
2865 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2867 ret
= btrfs_parse_options(fs_info
, options
, sb
->s_flags
);
2873 features
= btrfs_super_incompat_flags(disk_super
) &
2874 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2877 "cannot mount because of unsupported optional features (%llx)",
2883 features
= btrfs_super_incompat_flags(disk_super
);
2884 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2885 if (fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2886 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2887 else if (fs_info
->compress_type
== BTRFS_COMPRESS_ZSTD
)
2888 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD
;
2890 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2891 btrfs_info(fs_info
, "has skinny extents");
2894 * flag our filesystem as having big metadata blocks if
2895 * they are bigger than the page size
2897 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
2898 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2900 "flagging fs with big metadata feature");
2901 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2904 nodesize
= btrfs_super_nodesize(disk_super
);
2905 sectorsize
= btrfs_super_sectorsize(disk_super
);
2906 stripesize
= sectorsize
;
2907 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2908 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2910 /* Cache block sizes */
2911 fs_info
->nodesize
= nodesize
;
2912 fs_info
->sectorsize
= sectorsize
;
2913 fs_info
->stripesize
= stripesize
;
2916 * mixed block groups end up with duplicate but slightly offset
2917 * extent buffers for the same range. It leads to corruptions
2919 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2920 (sectorsize
!= nodesize
)) {
2922 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2923 nodesize
, sectorsize
);
2928 * Needn't use the lock because there is no other task which will
2931 btrfs_set_super_incompat_flags(disk_super
, features
);
2933 features
= btrfs_super_compat_ro_flags(disk_super
) &
2934 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2935 if (!sb_rdonly(sb
) && features
) {
2937 "cannot mount read-write because of unsupported optional features (%llx)",
2943 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2946 goto fail_sb_buffer
;
2949 sb
->s_bdi
->congested_fn
= btrfs_congested_fn
;
2950 sb
->s_bdi
->congested_data
= fs_info
;
2951 sb
->s_bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
2952 sb
->s_bdi
->ra_pages
= VM_MAX_READAHEAD
* SZ_1K
/ PAGE_SIZE
;
2953 sb
->s_bdi
->ra_pages
*= btrfs_super_num_devices(disk_super
);
2954 sb
->s_bdi
->ra_pages
= max(sb
->s_bdi
->ra_pages
, SZ_4M
/ PAGE_SIZE
);
2956 sb
->s_blocksize
= sectorsize
;
2957 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2958 memcpy(&sb
->s_uuid
, fs_info
->fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2960 mutex_lock(&fs_info
->chunk_mutex
);
2961 ret
= btrfs_read_sys_array(fs_info
);
2962 mutex_unlock(&fs_info
->chunk_mutex
);
2964 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
2965 goto fail_sb_buffer
;
2968 generation
= btrfs_super_chunk_root_generation(disk_super
);
2969 level
= btrfs_super_chunk_root_level(disk_super
);
2971 __setup_root(chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2973 chunk_root
->node
= read_tree_block(fs_info
,
2974 btrfs_super_chunk_root(disk_super
),
2975 generation
, level
, NULL
);
2976 if (IS_ERR(chunk_root
->node
) ||
2977 !extent_buffer_uptodate(chunk_root
->node
)) {
2978 btrfs_err(fs_info
, "failed to read chunk root");
2979 if (!IS_ERR(chunk_root
->node
))
2980 free_extent_buffer(chunk_root
->node
);
2981 chunk_root
->node
= NULL
;
2982 goto fail_tree_roots
;
2984 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2985 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2987 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2988 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2990 ret
= btrfs_read_chunk_tree(fs_info
);
2992 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
2993 goto fail_tree_roots
;
2997 * Keep the devid that is marked to be the target device for the
2998 * device replace procedure
3000 btrfs_free_extra_devids(fs_devices
, 0);
3002 if (!fs_devices
->latest_bdev
) {
3003 btrfs_err(fs_info
, "failed to read devices");
3004 goto fail_tree_roots
;
3008 generation
= btrfs_super_generation(disk_super
);
3009 level
= btrfs_super_root_level(disk_super
);
3011 tree_root
->node
= read_tree_block(fs_info
,
3012 btrfs_super_root(disk_super
),
3013 generation
, level
, NULL
);
3014 if (IS_ERR(tree_root
->node
) ||
3015 !extent_buffer_uptodate(tree_root
->node
)) {
3016 btrfs_warn(fs_info
, "failed to read tree root");
3017 if (!IS_ERR(tree_root
->node
))
3018 free_extent_buffer(tree_root
->node
);
3019 tree_root
->node
= NULL
;
3020 goto recovery_tree_root
;
3023 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
3024 tree_root
->commit_root
= btrfs_root_node(tree_root
);
3025 btrfs_set_root_refs(&tree_root
->root_item
, 1);
3027 mutex_lock(&tree_root
->objectid_mutex
);
3028 ret
= btrfs_find_highest_objectid(tree_root
,
3029 &tree_root
->highest_objectid
);
3031 mutex_unlock(&tree_root
->objectid_mutex
);
3032 goto recovery_tree_root
;
3035 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
3037 mutex_unlock(&tree_root
->objectid_mutex
);
3039 ret
= btrfs_read_roots(fs_info
);
3041 goto recovery_tree_root
;
3043 fs_info
->generation
= generation
;
3044 fs_info
->last_trans_committed
= generation
;
3046 ret
= btrfs_verify_dev_extents(fs_info
);
3049 "failed to verify dev extents against chunks: %d",
3051 goto fail_block_groups
;
3053 ret
= btrfs_recover_balance(fs_info
);
3055 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
3056 goto fail_block_groups
;
3059 ret
= btrfs_init_dev_stats(fs_info
);
3061 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
3062 goto fail_block_groups
;
3065 ret
= btrfs_init_dev_replace(fs_info
);
3067 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
3068 goto fail_block_groups
;
3071 btrfs_free_extra_devids(fs_devices
, 1);
3073 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
3075 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
3077 goto fail_block_groups
;
3080 ret
= btrfs_sysfs_add_device(fs_devices
);
3082 btrfs_err(fs_info
, "failed to init sysfs device interface: %d",
3084 goto fail_fsdev_sysfs
;
3087 ret
= btrfs_sysfs_add_mounted(fs_info
);
3089 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
3090 goto fail_fsdev_sysfs
;
3093 ret
= btrfs_init_space_info(fs_info
);
3095 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
3099 ret
= btrfs_read_block_groups(fs_info
);
3101 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
3105 if (!sb_rdonly(sb
) && !btrfs_check_rw_degradable(fs_info
, NULL
)) {
3107 "writable mount is not allowed due to too many missing devices");
3111 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3113 if (IS_ERR(fs_info
->cleaner_kthread
))
3116 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3118 "btrfs-transaction");
3119 if (IS_ERR(fs_info
->transaction_kthread
))
3122 if (!btrfs_test_opt(fs_info
, NOSSD
) &&
3123 !fs_info
->fs_devices
->rotating
) {
3124 btrfs_set_and_info(fs_info
, SSD
, "enabling ssd optimizations");
3128 * Mount does not set all options immediately, we can do it now and do
3129 * not have to wait for transaction commit
3131 btrfs_apply_pending_changes(fs_info
);
3133 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3134 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
)) {
3135 ret
= btrfsic_mount(fs_info
, fs_devices
,
3136 btrfs_test_opt(fs_info
,
3137 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3139 fs_info
->check_integrity_print_mask
);
3142 "failed to initialize integrity check module: %d",
3146 ret
= btrfs_read_qgroup_config(fs_info
);
3148 goto fail_trans_kthread
;
3150 if (btrfs_build_ref_tree(fs_info
))
3151 btrfs_err(fs_info
, "couldn't build ref tree");
3153 /* do not make disk changes in broken FS or nologreplay is given */
3154 if (btrfs_super_log_root(disk_super
) != 0 &&
3155 !btrfs_test_opt(fs_info
, NOLOGREPLAY
)) {
3156 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3163 ret
= btrfs_find_orphan_roots(fs_info
);
3167 if (!sb_rdonly(sb
)) {
3168 ret
= btrfs_cleanup_fs_roots(fs_info
);
3172 mutex_lock(&fs_info
->cleaner_mutex
);
3173 ret
= btrfs_recover_relocation(tree_root
);
3174 mutex_unlock(&fs_info
->cleaner_mutex
);
3176 btrfs_warn(fs_info
, "failed to recover relocation: %d",
3183 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3184 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3185 location
.offset
= 0;
3187 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3188 if (IS_ERR(fs_info
->fs_root
)) {
3189 err
= PTR_ERR(fs_info
->fs_root
);
3190 btrfs_warn(fs_info
, "failed to read fs tree: %d", err
);
3197 if (btrfs_test_opt(fs_info
, CLEAR_CACHE
) &&
3198 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3199 clear_free_space_tree
= 1;
3200 } else if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
3201 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE_VALID
)) {
3202 btrfs_warn(fs_info
, "free space tree is invalid");
3203 clear_free_space_tree
= 1;
3206 if (clear_free_space_tree
) {
3207 btrfs_info(fs_info
, "clearing free space tree");
3208 ret
= btrfs_clear_free_space_tree(fs_info
);
3211 "failed to clear free space tree: %d", ret
);
3212 close_ctree(fs_info
);
3217 if (btrfs_test_opt(fs_info
, FREE_SPACE_TREE
) &&
3218 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3219 btrfs_info(fs_info
, "creating free space tree");
3220 ret
= btrfs_create_free_space_tree(fs_info
);
3223 "failed to create free space tree: %d", ret
);
3224 close_ctree(fs_info
);
3229 down_read(&fs_info
->cleanup_work_sem
);
3230 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3231 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3232 up_read(&fs_info
->cleanup_work_sem
);
3233 close_ctree(fs_info
);
3236 up_read(&fs_info
->cleanup_work_sem
);
3238 ret
= btrfs_resume_balance_async(fs_info
);
3240 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3241 close_ctree(fs_info
);
3245 ret
= btrfs_resume_dev_replace_async(fs_info
);
3247 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3248 close_ctree(fs_info
);
3252 btrfs_qgroup_rescan_resume(fs_info
);
3254 if (!fs_info
->uuid_root
) {
3255 btrfs_info(fs_info
, "creating UUID tree");
3256 ret
= btrfs_create_uuid_tree(fs_info
);
3259 "failed to create the UUID tree: %d", ret
);
3260 close_ctree(fs_info
);
3263 } else if (btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) ||
3264 fs_info
->generation
!=
3265 btrfs_super_uuid_tree_generation(disk_super
)) {
3266 btrfs_info(fs_info
, "checking UUID tree");
3267 ret
= btrfs_check_uuid_tree(fs_info
);
3270 "failed to check the UUID tree: %d", ret
);
3271 close_ctree(fs_info
);
3275 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
3277 set_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3280 * backuproot only affect mount behavior, and if open_ctree succeeded,
3281 * no need to keep the flag
3283 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3288 btrfs_free_qgroup_config(fs_info
);
3290 kthread_stop(fs_info
->transaction_kthread
);
3291 btrfs_cleanup_transaction(fs_info
);
3292 btrfs_free_fs_roots(fs_info
);
3294 kthread_stop(fs_info
->cleaner_kthread
);
3297 * make sure we're done with the btree inode before we stop our
3300 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3303 btrfs_sysfs_remove_mounted(fs_info
);
3306 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3309 btrfs_put_block_group_cache(fs_info
);
3312 free_root_pointers(fs_info
, 1);
3313 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3316 btrfs_stop_all_workers(fs_info
);
3317 btrfs_free_block_groups(fs_info
);
3320 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3322 iput(fs_info
->btree_inode
);
3324 percpu_counter_destroy(&fs_info
->dev_replace
.bio_counter
);
3325 fail_delalloc_bytes
:
3326 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3327 fail_dirty_metadata_bytes
:
3328 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3330 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3332 btrfs_free_stripe_hash_table(fs_info
);
3333 btrfs_close_devices(fs_info
->fs_devices
);
3337 if (!btrfs_test_opt(fs_info
, USEBACKUPROOT
))
3338 goto fail_tree_roots
;
3340 free_root_pointers(fs_info
, 0);
3342 /* don't use the log in recovery mode, it won't be valid */
3343 btrfs_set_super_log_root(disk_super
, 0);
3345 /* we can't trust the free space cache either */
3346 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3348 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3349 &num_backups_tried
, &backup_index
);
3351 goto fail_block_groups
;
3352 goto retry_root_backup
;
3354 ALLOW_ERROR_INJECTION(open_ctree
, ERRNO
);
3356 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3359 set_buffer_uptodate(bh
);
3361 struct btrfs_device
*device
= (struct btrfs_device
*)
3364 btrfs_warn_rl_in_rcu(device
->fs_info
,
3365 "lost page write due to IO error on %s",
3366 rcu_str_deref(device
->name
));
3367 /* note, we don't set_buffer_write_io_error because we have
3368 * our own ways of dealing with the IO errors
3370 clear_buffer_uptodate(bh
);
3371 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3377 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3378 struct buffer_head
**bh_ret
)
3380 struct buffer_head
*bh
;
3381 struct btrfs_super_block
*super
;
3384 bytenr
= btrfs_sb_offset(copy_num
);
3385 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3388 bh
= __bread(bdev
, bytenr
/ BTRFS_BDEV_BLOCKSIZE
, BTRFS_SUPER_INFO_SIZE
);
3390 * If we fail to read from the underlying devices, as of now
3391 * the best option we have is to mark it EIO.
3396 super
= (struct btrfs_super_block
*)bh
->b_data
;
3397 if (btrfs_super_bytenr(super
) != bytenr
||
3398 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3408 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3410 struct buffer_head
*bh
;
3411 struct buffer_head
*latest
= NULL
;
3412 struct btrfs_super_block
*super
;
3417 /* we would like to check all the supers, but that would make
3418 * a btrfs mount succeed after a mkfs from a different FS.
3419 * So, we need to add a special mount option to scan for
3420 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3422 for (i
= 0; i
< 1; i
++) {
3423 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3427 super
= (struct btrfs_super_block
*)bh
->b_data
;
3429 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3432 transid
= btrfs_super_generation(super
);
3439 return ERR_PTR(ret
);
3445 * Write superblock @sb to the @device. Do not wait for completion, all the
3446 * buffer heads we write are pinned.
3448 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3449 * the expected device size at commit time. Note that max_mirrors must be
3450 * same for write and wait phases.
3452 * Return number of errors when buffer head is not found or submission fails.
3454 static int write_dev_supers(struct btrfs_device
*device
,
3455 struct btrfs_super_block
*sb
, int max_mirrors
)
3457 struct buffer_head
*bh
;
3465 if (max_mirrors
== 0)
3466 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3468 for (i
= 0; i
< max_mirrors
; i
++) {
3469 bytenr
= btrfs_sb_offset(i
);
3470 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3471 device
->commit_total_bytes
)
3474 btrfs_set_super_bytenr(sb
, bytenr
);
3477 crc
= btrfs_csum_data((const char *)sb
+ BTRFS_CSUM_SIZE
, crc
,
3478 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
3479 btrfs_csum_final(crc
, sb
->csum
);
3481 /* One reference for us, and we leave it for the caller */
3482 bh
= __getblk(device
->bdev
, bytenr
/ BTRFS_BDEV_BLOCKSIZE
,
3483 BTRFS_SUPER_INFO_SIZE
);
3485 btrfs_err(device
->fs_info
,
3486 "couldn't get super buffer head for bytenr %llu",
3492 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3494 /* one reference for submit_bh */
3497 set_buffer_uptodate(bh
);
3499 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3500 bh
->b_private
= device
;
3503 * we fua the first super. The others we allow
3506 op_flags
= REQ_SYNC
| REQ_META
| REQ_PRIO
;
3507 if (i
== 0 && !btrfs_test_opt(device
->fs_info
, NOBARRIER
))
3508 op_flags
|= REQ_FUA
;
3509 ret
= btrfsic_submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3513 return errors
< i
? 0 : -1;
3517 * Wait for write completion of superblocks done by write_dev_supers,
3518 * @max_mirrors same for write and wait phases.
3520 * Return number of errors when buffer head is not found or not marked up to
3523 static int wait_dev_supers(struct btrfs_device
*device
, int max_mirrors
)
3525 struct buffer_head
*bh
;
3528 bool primary_failed
= false;
3531 if (max_mirrors
== 0)
3532 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3534 for (i
= 0; i
< max_mirrors
; i
++) {
3535 bytenr
= btrfs_sb_offset(i
);
3536 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3537 device
->commit_total_bytes
)
3540 bh
= __find_get_block(device
->bdev
,
3541 bytenr
/ BTRFS_BDEV_BLOCKSIZE
,
3542 BTRFS_SUPER_INFO_SIZE
);
3546 primary_failed
= true;
3550 if (!buffer_uptodate(bh
)) {
3553 primary_failed
= true;
3556 /* drop our reference */
3559 /* drop the reference from the writing run */
3563 /* log error, force error return */
3564 if (primary_failed
) {
3565 btrfs_err(device
->fs_info
, "error writing primary super block to device %llu",
3570 return errors
< i
? 0 : -1;
3574 * endio for the write_dev_flush, this will wake anyone waiting
3575 * for the barrier when it is done
3577 static void btrfs_end_empty_barrier(struct bio
*bio
)
3579 complete(bio
->bi_private
);
3583 * Submit a flush request to the device if it supports it. Error handling is
3584 * done in the waiting counterpart.
3586 static void write_dev_flush(struct btrfs_device
*device
)
3588 struct request_queue
*q
= bdev_get_queue(device
->bdev
);
3589 struct bio
*bio
= device
->flush_bio
;
3591 if (!test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
))
3595 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3596 bio_set_dev(bio
, device
->bdev
);
3597 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
;
3598 init_completion(&device
->flush_wait
);
3599 bio
->bi_private
= &device
->flush_wait
;
3601 btrfsic_submit_bio(bio
);
3602 set_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3606 * If the flush bio has been submitted by write_dev_flush, wait for it.
3608 static blk_status_t
wait_dev_flush(struct btrfs_device
*device
)
3610 struct bio
*bio
= device
->flush_bio
;
3612 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
))
3615 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
);
3616 wait_for_completion_io(&device
->flush_wait
);
3618 return bio
->bi_status
;
3621 static int check_barrier_error(struct btrfs_fs_info
*fs_info
)
3623 if (!btrfs_check_rw_degradable(fs_info
, NULL
))
3629 * send an empty flush down to each device in parallel,
3630 * then wait for them
3632 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3634 struct list_head
*head
;
3635 struct btrfs_device
*dev
;
3636 int errors_wait
= 0;
3639 lockdep_assert_held(&info
->fs_devices
->device_list_mutex
);
3640 /* send down all the barriers */
3641 head
= &info
->fs_devices
->devices
;
3642 list_for_each_entry(dev
, head
, dev_list
) {
3643 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3647 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3648 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3651 write_dev_flush(dev
);
3652 dev
->last_flush_error
= BLK_STS_OK
;
3655 /* wait for all the barriers */
3656 list_for_each_entry(dev
, head
, dev_list
) {
3657 if (test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
))
3663 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3664 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3667 ret
= wait_dev_flush(dev
);
3669 dev
->last_flush_error
= ret
;
3670 btrfs_dev_stat_inc_and_print(dev
,
3671 BTRFS_DEV_STAT_FLUSH_ERRS
);
3678 * At some point we need the status of all disks
3679 * to arrive at the volume status. So error checking
3680 * is being pushed to a separate loop.
3682 return check_barrier_error(info
);
3687 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3690 int min_tolerated
= INT_MAX
;
3692 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3693 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3694 min_tolerated
= min(min_tolerated
,
3695 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3696 tolerated_failures
);
3698 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3699 if (raid_type
== BTRFS_RAID_SINGLE
)
3701 if (!(flags
& btrfs_raid_array
[raid_type
].bg_flag
))
3703 min_tolerated
= min(min_tolerated
,
3704 btrfs_raid_array
[raid_type
].
3705 tolerated_failures
);
3708 if (min_tolerated
== INT_MAX
) {
3709 pr_warn("BTRFS: unknown raid flag: %llu", flags
);
3713 return min_tolerated
;
3716 int write_all_supers(struct btrfs_fs_info
*fs_info
, int max_mirrors
)
3718 struct list_head
*head
;
3719 struct btrfs_device
*dev
;
3720 struct btrfs_super_block
*sb
;
3721 struct btrfs_dev_item
*dev_item
;
3725 int total_errors
= 0;
3728 do_barriers
= !btrfs_test_opt(fs_info
, NOBARRIER
);
3731 * max_mirrors == 0 indicates we're from commit_transaction,
3732 * not from fsync where the tree roots in fs_info have not
3733 * been consistent on disk.
3735 if (max_mirrors
== 0)
3736 backup_super_roots(fs_info
);
3738 sb
= fs_info
->super_for_commit
;
3739 dev_item
= &sb
->dev_item
;
3741 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3742 head
= &fs_info
->fs_devices
->devices
;
3743 max_errors
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
3746 ret
= barrier_all_devices(fs_info
);
3749 &fs_info
->fs_devices
->device_list_mutex
);
3750 btrfs_handle_fs_error(fs_info
, ret
,
3751 "errors while submitting device barriers.");
3756 list_for_each_entry(dev
, head
, dev_list
) {
3761 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3762 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3765 btrfs_set_stack_device_generation(dev_item
, 0);
3766 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3767 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3768 btrfs_set_stack_device_total_bytes(dev_item
,
3769 dev
->commit_total_bytes
);
3770 btrfs_set_stack_device_bytes_used(dev_item
,
3771 dev
->commit_bytes_used
);
3772 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3773 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3774 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3775 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3776 memcpy(dev_item
->fsid
, dev
->fs_devices
->metadata_uuid
,
3779 flags
= btrfs_super_flags(sb
);
3780 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3782 ret
= btrfs_validate_write_super(fs_info
, sb
);
3784 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3785 btrfs_handle_fs_error(fs_info
, -EUCLEAN
,
3786 "unexpected superblock corruption detected");
3790 ret
= write_dev_supers(dev
, sb
, max_mirrors
);
3794 if (total_errors
> max_errors
) {
3795 btrfs_err(fs_info
, "%d errors while writing supers",
3797 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3799 /* FUA is masked off if unsupported and can't be the reason */
3800 btrfs_handle_fs_error(fs_info
, -EIO
,
3801 "%d errors while writing supers",
3807 list_for_each_entry(dev
, head
, dev_list
) {
3810 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &dev
->dev_state
) ||
3811 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))
3814 ret
= wait_dev_supers(dev
, max_mirrors
);
3818 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3819 if (total_errors
> max_errors
) {
3820 btrfs_handle_fs_error(fs_info
, -EIO
,
3821 "%d errors while writing supers",
3828 /* Drop a fs root from the radix tree and free it. */
3829 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3830 struct btrfs_root
*root
)
3832 spin_lock(&fs_info
->fs_roots_radix_lock
);
3833 radix_tree_delete(&fs_info
->fs_roots_radix
,
3834 (unsigned long)root
->root_key
.objectid
);
3835 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3837 if (btrfs_root_refs(&root
->root_item
) == 0)
3838 synchronize_srcu(&fs_info
->subvol_srcu
);
3840 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3841 btrfs_free_log(NULL
, root
);
3842 if (root
->reloc_root
) {
3843 free_extent_buffer(root
->reloc_root
->node
);
3844 free_extent_buffer(root
->reloc_root
->commit_root
);
3845 btrfs_put_fs_root(root
->reloc_root
);
3846 root
->reloc_root
= NULL
;
3850 if (root
->free_ino_pinned
)
3851 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3852 if (root
->free_ino_ctl
)
3853 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3854 btrfs_free_fs_root(root
);
3857 void btrfs_free_fs_root(struct btrfs_root
*root
)
3859 iput(root
->ino_cache_inode
);
3860 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3862 free_anon_bdev(root
->anon_dev
);
3863 if (root
->subv_writers
)
3864 btrfs_free_subvolume_writers(root
->subv_writers
);
3865 free_extent_buffer(root
->node
);
3866 free_extent_buffer(root
->commit_root
);
3867 kfree(root
->free_ino_ctl
);
3868 kfree(root
->free_ino_pinned
);
3869 btrfs_put_fs_root(root
);
3872 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3874 u64 root_objectid
= 0;
3875 struct btrfs_root
*gang
[8];
3878 unsigned int ret
= 0;
3882 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3883 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3884 (void **)gang
, root_objectid
,
3887 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3890 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3892 for (i
= 0; i
< ret
; i
++) {
3893 /* Avoid to grab roots in dead_roots */
3894 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3898 /* grab all the search result for later use */
3899 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3901 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3903 for (i
= 0; i
< ret
; i
++) {
3906 root_objectid
= gang
[i
]->root_key
.objectid
;
3907 err
= btrfs_orphan_cleanup(gang
[i
]);
3910 btrfs_put_fs_root(gang
[i
]);
3915 /* release the uncleaned roots due to error */
3916 for (; i
< ret
; i
++) {
3918 btrfs_put_fs_root(gang
[i
]);
3923 int btrfs_commit_super(struct btrfs_fs_info
*fs_info
)
3925 struct btrfs_root
*root
= fs_info
->tree_root
;
3926 struct btrfs_trans_handle
*trans
;
3928 mutex_lock(&fs_info
->cleaner_mutex
);
3929 btrfs_run_delayed_iputs(fs_info
);
3930 mutex_unlock(&fs_info
->cleaner_mutex
);
3931 wake_up_process(fs_info
->cleaner_kthread
);
3933 /* wait until ongoing cleanup work done */
3934 down_write(&fs_info
->cleanup_work_sem
);
3935 up_write(&fs_info
->cleanup_work_sem
);
3937 trans
= btrfs_join_transaction(root
);
3939 return PTR_ERR(trans
);
3940 return btrfs_commit_transaction(trans
);
3943 void close_ctree(struct btrfs_fs_info
*fs_info
)
3947 set_bit(BTRFS_FS_CLOSING_START
, &fs_info
->flags
);
3949 * We don't want the cleaner to start new transactions, add more delayed
3950 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3951 * because that frees the task_struct, and the transaction kthread might
3952 * still try to wake up the cleaner.
3954 kthread_park(fs_info
->cleaner_kthread
);
3956 /* wait for the qgroup rescan worker to stop */
3957 btrfs_qgroup_wait_for_completion(fs_info
, false);
3959 /* wait for the uuid_scan task to finish */
3960 down(&fs_info
->uuid_tree_rescan_sem
);
3961 /* avoid complains from lockdep et al., set sem back to initial state */
3962 up(&fs_info
->uuid_tree_rescan_sem
);
3964 /* pause restriper - we want to resume on mount */
3965 btrfs_pause_balance(fs_info
);
3967 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3969 btrfs_scrub_cancel(fs_info
);
3971 /* wait for any defraggers to finish */
3972 wait_event(fs_info
->transaction_wait
,
3973 (atomic_read(&fs_info
->defrag_running
) == 0));
3975 /* clear out the rbtree of defraggable inodes */
3976 btrfs_cleanup_defrag_inodes(fs_info
);
3978 cancel_work_sync(&fs_info
->async_reclaim_work
);
3980 if (!sb_rdonly(fs_info
->sb
)) {
3982 * The cleaner kthread is stopped, so do one final pass over
3983 * unused block groups.
3985 btrfs_delete_unused_bgs(fs_info
);
3987 ret
= btrfs_commit_super(fs_info
);
3989 btrfs_err(fs_info
, "commit super ret %d", ret
);
3992 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
) ||
3993 test_bit(BTRFS_FS_STATE_TRANS_ABORTED
, &fs_info
->fs_state
))
3994 btrfs_error_commit_super(fs_info
);
3996 kthread_stop(fs_info
->transaction_kthread
);
3997 kthread_stop(fs_info
->cleaner_kthread
);
3999 ASSERT(list_empty(&fs_info
->delayed_iputs
));
4000 set_bit(BTRFS_FS_CLOSING_DONE
, &fs_info
->flags
);
4002 btrfs_free_qgroup_config(fs_info
);
4003 ASSERT(list_empty(&fs_info
->delalloc_roots
));
4005 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
4006 btrfs_info(fs_info
, "at unmount delalloc count %lld",
4007 percpu_counter_sum(&fs_info
->delalloc_bytes
));
4010 btrfs_sysfs_remove_mounted(fs_info
);
4011 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
4013 btrfs_free_fs_roots(fs_info
);
4015 btrfs_put_block_group_cache(fs_info
);
4018 * we must make sure there is not any read request to
4019 * submit after we stopping all workers.
4021 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
4022 btrfs_stop_all_workers(fs_info
);
4024 btrfs_free_block_groups(fs_info
);
4026 clear_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
4027 free_root_pointers(fs_info
, 1);
4029 iput(fs_info
->btree_inode
);
4031 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4032 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
))
4033 btrfsic_unmount(fs_info
->fs_devices
);
4036 btrfs_close_devices(fs_info
->fs_devices
);
4037 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
4039 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
4040 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
4041 percpu_counter_destroy(&fs_info
->dev_replace
.bio_counter
);
4042 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
4044 btrfs_free_stripe_hash_table(fs_info
);
4045 btrfs_free_ref_cache(fs_info
);
4047 while (!list_empty(&fs_info
->pinned_chunks
)) {
4048 struct extent_map
*em
;
4050 em
= list_first_entry(&fs_info
->pinned_chunks
,
4051 struct extent_map
, list
);
4052 list_del_init(&em
->list
);
4053 free_extent_map(em
);
4057 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
4061 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
4063 ret
= extent_buffer_uptodate(buf
);
4067 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
4068 parent_transid
, atomic
);
4074 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
4076 struct btrfs_fs_info
*fs_info
;
4077 struct btrfs_root
*root
;
4078 u64 transid
= btrfs_header_generation(buf
);
4081 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4083 * This is a fast path so only do this check if we have sanity tests
4084 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4085 * outside of the sanity tests.
4087 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED
, &buf
->bflags
)))
4090 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4091 fs_info
= root
->fs_info
;
4092 btrfs_assert_tree_locked(buf
);
4093 if (transid
!= fs_info
->generation
)
4094 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4095 buf
->start
, transid
, fs_info
->generation
);
4096 was_dirty
= set_extent_buffer_dirty(buf
);
4098 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
4100 fs_info
->dirty_metadata_batch
);
4101 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4103 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4104 * but item data not updated.
4105 * So here we should only check item pointers, not item data.
4107 if (btrfs_header_level(buf
) == 0 &&
4108 btrfs_check_leaf_relaxed(fs_info
, buf
)) {
4109 btrfs_print_leaf(buf
);
4115 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
,
4119 * looks as though older kernels can get into trouble with
4120 * this code, they end up stuck in balance_dirty_pages forever
4124 if (current
->flags
& PF_MEMALLOC
)
4128 btrfs_balance_delayed_items(fs_info
);
4130 ret
= __percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
4131 BTRFS_DIRTY_METADATA_THRESH
,
4132 fs_info
->dirty_metadata_batch
);
4134 balance_dirty_pages_ratelimited(fs_info
->btree_inode
->i_mapping
);
4138 void btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
)
4140 __btrfs_btree_balance_dirty(fs_info
, 1);
4143 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info
*fs_info
)
4145 __btrfs_btree_balance_dirty(fs_info
, 0);
4148 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
, int level
,
4149 struct btrfs_key
*first_key
)
4151 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4152 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4154 return btree_read_extent_buffer_pages(fs_info
, buf
, parent_transid
,
4158 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
)
4160 /* cleanup FS via transaction */
4161 btrfs_cleanup_transaction(fs_info
);
4163 mutex_lock(&fs_info
->cleaner_mutex
);
4164 btrfs_run_delayed_iputs(fs_info
);
4165 mutex_unlock(&fs_info
->cleaner_mutex
);
4167 down_write(&fs_info
->cleanup_work_sem
);
4168 up_write(&fs_info
->cleanup_work_sem
);
4171 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4173 struct btrfs_ordered_extent
*ordered
;
4175 spin_lock(&root
->ordered_extent_lock
);
4177 * This will just short circuit the ordered completion stuff which will
4178 * make sure the ordered extent gets properly cleaned up.
4180 list_for_each_entry(ordered
, &root
->ordered_extents
,
4182 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4183 spin_unlock(&root
->ordered_extent_lock
);
4186 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4188 struct btrfs_root
*root
;
4189 struct list_head splice
;
4191 INIT_LIST_HEAD(&splice
);
4193 spin_lock(&fs_info
->ordered_root_lock
);
4194 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4195 while (!list_empty(&splice
)) {
4196 root
= list_first_entry(&splice
, struct btrfs_root
,
4198 list_move_tail(&root
->ordered_root
,
4199 &fs_info
->ordered_roots
);
4201 spin_unlock(&fs_info
->ordered_root_lock
);
4202 btrfs_destroy_ordered_extents(root
);
4205 spin_lock(&fs_info
->ordered_root_lock
);
4207 spin_unlock(&fs_info
->ordered_root_lock
);
4210 * We need this here because if we've been flipped read-only we won't
4211 * get sync() from the umount, so we need to make sure any ordered
4212 * extents that haven't had their dirty pages IO start writeout yet
4213 * actually get run and error out properly.
4215 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0, (u64
)-1);
4218 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4219 struct btrfs_fs_info
*fs_info
)
4221 struct rb_node
*node
;
4222 struct btrfs_delayed_ref_root
*delayed_refs
;
4223 struct btrfs_delayed_ref_node
*ref
;
4226 delayed_refs
= &trans
->delayed_refs
;
4228 spin_lock(&delayed_refs
->lock
);
4229 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4230 spin_unlock(&delayed_refs
->lock
);
4231 btrfs_info(fs_info
, "delayed_refs has NO entry");
4235 while ((node
= rb_first_cached(&delayed_refs
->href_root
)) != NULL
) {
4236 struct btrfs_delayed_ref_head
*head
;
4238 bool pin_bytes
= false;
4240 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4242 if (!mutex_trylock(&head
->mutex
)) {
4243 refcount_inc(&head
->refs
);
4244 spin_unlock(&delayed_refs
->lock
);
4246 mutex_lock(&head
->mutex
);
4247 mutex_unlock(&head
->mutex
);
4248 btrfs_put_delayed_ref_head(head
);
4249 spin_lock(&delayed_refs
->lock
);
4252 spin_lock(&head
->lock
);
4253 while ((n
= rb_first_cached(&head
->ref_tree
)) != NULL
) {
4254 ref
= rb_entry(n
, struct btrfs_delayed_ref_node
,
4257 rb_erase_cached(&ref
->ref_node
, &head
->ref_tree
);
4258 RB_CLEAR_NODE(&ref
->ref_node
);
4259 if (!list_empty(&ref
->add_list
))
4260 list_del(&ref
->add_list
);
4261 atomic_dec(&delayed_refs
->num_entries
);
4262 btrfs_put_delayed_ref(ref
);
4264 if (head
->must_insert_reserved
)
4266 btrfs_free_delayed_extent_op(head
->extent_op
);
4267 delayed_refs
->num_heads
--;
4268 if (head
->processing
== 0)
4269 delayed_refs
->num_heads_ready
--;
4270 atomic_dec(&delayed_refs
->num_entries
);
4271 rb_erase_cached(&head
->href_node
, &delayed_refs
->href_root
);
4272 RB_CLEAR_NODE(&head
->href_node
);
4273 spin_unlock(&head
->lock
);
4274 spin_unlock(&delayed_refs
->lock
);
4275 mutex_unlock(&head
->mutex
);
4278 btrfs_pin_extent(fs_info
, head
->bytenr
,
4279 head
->num_bytes
, 1);
4280 btrfs_cleanup_ref_head_accounting(fs_info
, delayed_refs
, head
);
4281 btrfs_put_delayed_ref_head(head
);
4283 spin_lock(&delayed_refs
->lock
);
4286 spin_unlock(&delayed_refs
->lock
);
4291 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4293 struct btrfs_inode
*btrfs_inode
;
4294 struct list_head splice
;
4296 INIT_LIST_HEAD(&splice
);
4298 spin_lock(&root
->delalloc_lock
);
4299 list_splice_init(&root
->delalloc_inodes
, &splice
);
4301 while (!list_empty(&splice
)) {
4302 struct inode
*inode
= NULL
;
4303 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4305 __btrfs_del_delalloc_inode(root
, btrfs_inode
);
4306 spin_unlock(&root
->delalloc_lock
);
4309 * Make sure we get a live inode and that it'll not disappear
4312 inode
= igrab(&btrfs_inode
->vfs_inode
);
4314 invalidate_inode_pages2(inode
->i_mapping
);
4317 spin_lock(&root
->delalloc_lock
);
4319 spin_unlock(&root
->delalloc_lock
);
4322 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4324 struct btrfs_root
*root
;
4325 struct list_head splice
;
4327 INIT_LIST_HEAD(&splice
);
4329 spin_lock(&fs_info
->delalloc_root_lock
);
4330 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4331 while (!list_empty(&splice
)) {
4332 root
= list_first_entry(&splice
, struct btrfs_root
,
4334 root
= btrfs_grab_fs_root(root
);
4336 spin_unlock(&fs_info
->delalloc_root_lock
);
4338 btrfs_destroy_delalloc_inodes(root
);
4339 btrfs_put_fs_root(root
);
4341 spin_lock(&fs_info
->delalloc_root_lock
);
4343 spin_unlock(&fs_info
->delalloc_root_lock
);
4346 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
4347 struct extent_io_tree
*dirty_pages
,
4351 struct extent_buffer
*eb
;
4356 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4361 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4362 while (start
<= end
) {
4363 eb
= find_extent_buffer(fs_info
, start
);
4364 start
+= fs_info
->nodesize
;
4367 wait_on_extent_buffer_writeback(eb
);
4369 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4371 clear_extent_buffer_dirty(eb
);
4372 free_extent_buffer_stale(eb
);
4379 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
4380 struct extent_io_tree
*pinned_extents
)
4382 struct extent_io_tree
*unpin
;
4388 unpin
= pinned_extents
;
4391 struct extent_state
*cached_state
= NULL
;
4394 * The btrfs_finish_extent_commit() may get the same range as
4395 * ours between find_first_extent_bit and clear_extent_dirty.
4396 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4397 * the same extent range.
4399 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
4400 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4401 EXTENT_DIRTY
, &cached_state
);
4403 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4407 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
4408 free_extent_state(cached_state
);
4409 btrfs_error_unpin_extent_range(fs_info
, start
, end
);
4410 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
4415 if (unpin
== &fs_info
->freed_extents
[0])
4416 unpin
= &fs_info
->freed_extents
[1];
4418 unpin
= &fs_info
->freed_extents
[0];
4426 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache
*cache
)
4428 struct inode
*inode
;
4430 inode
= cache
->io_ctl
.inode
;
4432 invalidate_inode_pages2(inode
->i_mapping
);
4433 BTRFS_I(inode
)->generation
= 0;
4434 cache
->io_ctl
.inode
= NULL
;
4437 btrfs_put_block_group(cache
);
4440 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction
*cur_trans
,
4441 struct btrfs_fs_info
*fs_info
)
4443 struct btrfs_block_group_cache
*cache
;
4445 spin_lock(&cur_trans
->dirty_bgs_lock
);
4446 while (!list_empty(&cur_trans
->dirty_bgs
)) {
4447 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
4448 struct btrfs_block_group_cache
,
4451 if (!list_empty(&cache
->io_list
)) {
4452 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4453 list_del_init(&cache
->io_list
);
4454 btrfs_cleanup_bg_io(cache
);
4455 spin_lock(&cur_trans
->dirty_bgs_lock
);
4458 list_del_init(&cache
->dirty_list
);
4459 spin_lock(&cache
->lock
);
4460 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4461 spin_unlock(&cache
->lock
);
4463 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4464 btrfs_put_block_group(cache
);
4465 btrfs_delayed_refs_rsv_release(fs_info
, 1);
4466 spin_lock(&cur_trans
->dirty_bgs_lock
);
4468 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4471 * Refer to the definition of io_bgs member for details why it's safe
4472 * to use it without any locking
4474 while (!list_empty(&cur_trans
->io_bgs
)) {
4475 cache
= list_first_entry(&cur_trans
->io_bgs
,
4476 struct btrfs_block_group_cache
,
4479 list_del_init(&cache
->io_list
);
4480 spin_lock(&cache
->lock
);
4481 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4482 spin_unlock(&cache
->lock
);
4483 btrfs_cleanup_bg_io(cache
);
4487 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4488 struct btrfs_fs_info
*fs_info
)
4490 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
4491 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
4492 ASSERT(list_empty(&cur_trans
->io_bgs
));
4494 btrfs_destroy_delayed_refs(cur_trans
, fs_info
);
4496 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4497 wake_up(&fs_info
->transaction_blocked_wait
);
4499 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4500 wake_up(&fs_info
->transaction_wait
);
4502 btrfs_destroy_delayed_inodes(fs_info
);
4503 btrfs_assert_delayed_root_empty(fs_info
);
4505 btrfs_destroy_marked_extents(fs_info
, &cur_trans
->dirty_pages
,
4507 btrfs_destroy_pinned_extent(fs_info
,
4508 fs_info
->pinned_extents
);
4510 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4511 wake_up(&cur_trans
->commit_wait
);
4514 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
)
4516 struct btrfs_transaction
*t
;
4518 mutex_lock(&fs_info
->transaction_kthread_mutex
);
4520 spin_lock(&fs_info
->trans_lock
);
4521 while (!list_empty(&fs_info
->trans_list
)) {
4522 t
= list_first_entry(&fs_info
->trans_list
,
4523 struct btrfs_transaction
, list
);
4524 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4525 refcount_inc(&t
->use_count
);
4526 spin_unlock(&fs_info
->trans_lock
);
4527 btrfs_wait_for_commit(fs_info
, t
->transid
);
4528 btrfs_put_transaction(t
);
4529 spin_lock(&fs_info
->trans_lock
);
4532 if (t
== fs_info
->running_transaction
) {
4533 t
->state
= TRANS_STATE_COMMIT_DOING
;
4534 spin_unlock(&fs_info
->trans_lock
);
4536 * We wait for 0 num_writers since we don't hold a trans
4537 * handle open currently for this transaction.
4539 wait_event(t
->writer_wait
,
4540 atomic_read(&t
->num_writers
) == 0);
4542 spin_unlock(&fs_info
->trans_lock
);
4544 btrfs_cleanup_one_transaction(t
, fs_info
);
4546 spin_lock(&fs_info
->trans_lock
);
4547 if (t
== fs_info
->running_transaction
)
4548 fs_info
->running_transaction
= NULL
;
4549 list_del_init(&t
->list
);
4550 spin_unlock(&fs_info
->trans_lock
);
4552 btrfs_put_transaction(t
);
4553 trace_btrfs_transaction_commit(fs_info
->tree_root
);
4554 spin_lock(&fs_info
->trans_lock
);
4556 spin_unlock(&fs_info
->trans_lock
);
4557 btrfs_destroy_all_ordered_extents(fs_info
);
4558 btrfs_destroy_delayed_inodes(fs_info
);
4559 btrfs_assert_delayed_root_empty(fs_info
);
4560 btrfs_destroy_pinned_extent(fs_info
, fs_info
->pinned_extents
);
4561 btrfs_destroy_all_delalloc_inodes(fs_info
);
4562 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
4567 static const struct extent_io_ops btree_extent_io_ops
= {
4568 /* mandatory callbacks */
4569 .submit_bio_hook
= btree_submit_bio_hook
,
4570 .readpage_end_io_hook
= btree_readpage_end_io_hook
,