2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
53 #include "tree-checker.h"
54 #include "ref-verify.h"
57 #include <asm/cpufeature.h>
60 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
61 BTRFS_HEADER_FLAG_RELOC |\
62 BTRFS_SUPER_FLAG_ERROR |\
63 BTRFS_SUPER_FLAG_SEEDING |\
64 BTRFS_SUPER_FLAG_METADUMP)
66 static const struct extent_io_ops btree_extent_io_ops
;
67 static void end_workqueue_fn(struct btrfs_work
*work
);
68 static void free_fs_root(struct btrfs_root
*root
);
69 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
);
70 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
71 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
72 struct btrfs_fs_info
*fs_info
);
73 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
74 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
75 struct extent_io_tree
*dirty_pages
,
77 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
78 struct extent_io_tree
*pinned_extents
);
79 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
);
80 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
);
83 * btrfs_end_io_wq structs are used to do processing in task context when an IO
84 * is complete. This is used during reads to verify checksums, and it is used
85 * by writes to insert metadata for new file extents after IO is complete.
87 struct btrfs_end_io_wq
{
91 struct btrfs_fs_info
*info
;
93 enum btrfs_wq_endio_type metadata
;
94 struct btrfs_work work
;
97 static struct kmem_cache
*btrfs_end_io_wq_cache
;
99 int __init
btrfs_end_io_wq_init(void)
101 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq
),
106 if (!btrfs_end_io_wq_cache
)
111 void btrfs_end_io_wq_exit(void)
113 kmem_cache_destroy(btrfs_end_io_wq_cache
);
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
121 struct async_submit_bio
{
123 struct btrfs_fs_info
*fs_info
;
125 extent_submit_bio_hook_t
*submit_bio_start
;
126 extent_submit_bio_hook_t
*submit_bio_done
;
128 unsigned long bio_flags
;
130 * bio_offset is optional, can be used if the pages in the bio
131 * can't tell us where in the file the bio should go
134 struct btrfs_work work
;
139 * Lockdep class keys for extent_buffer->lock's in this root. For a given
140 * eb, the lockdep key is determined by the btrfs_root it belongs to and
141 * the level the eb occupies in the tree.
143 * Different roots are used for different purposes and may nest inside each
144 * other and they require separate keysets. As lockdep keys should be
145 * static, assign keysets according to the purpose of the root as indicated
146 * by btrfs_root->objectid. This ensures that all special purpose roots
147 * have separate keysets.
149 * Lock-nesting across peer nodes is always done with the immediate parent
150 * node locked thus preventing deadlock. As lockdep doesn't know this, use
151 * subclass to avoid triggering lockdep warning in such cases.
153 * The key is set by the readpage_end_io_hook after the buffer has passed
154 * csum validation but before the pages are unlocked. It is also set by
155 * btrfs_init_new_buffer on freshly allocated blocks.
157 * We also add a check to make sure the highest level of the tree is the
158 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
159 * needs update as well.
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
166 static struct btrfs_lockdep_keyset
{
167 u64 id
; /* root objectid */
168 const char *name_stem
; /* lock name stem */
169 char names
[BTRFS_MAX_LEVEL
+ 1][20];
170 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
171 } btrfs_lockdep_keysets
[] = {
172 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
173 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
174 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
175 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
176 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
177 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
178 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
179 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
180 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
181 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
182 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
183 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
184 { .id
= 0, .name_stem
= "tree" },
187 void __init
btrfs_init_lockdep(void)
191 /* initialize lockdep class names */
192 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
193 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
195 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
196 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
197 "btrfs-%s-%02d", ks
->name_stem
, j
);
201 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
204 struct btrfs_lockdep_keyset
*ks
;
206 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
208 /* find the matching keyset, id 0 is the default entry */
209 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
210 if (ks
->id
== objectid
)
213 lockdep_set_class_and_name(&eb
->lock
,
214 &ks
->keys
[level
], ks
->names
[level
]);
220 * extents on the btree inode are pretty simple, there's one extent
221 * that covers the entire device
223 static struct extent_map
*btree_get_extent(struct btrfs_inode
*inode
,
224 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
227 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
228 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
229 struct extent_map
*em
;
232 read_lock(&em_tree
->lock
);
233 em
= lookup_extent_mapping(em_tree
, start
, len
);
235 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
236 read_unlock(&em_tree
->lock
);
239 read_unlock(&em_tree
->lock
);
241 em
= alloc_extent_map();
243 em
= ERR_PTR(-ENOMEM
);
248 em
->block_len
= (u64
)-1;
250 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
252 write_lock(&em_tree
->lock
);
253 ret
= add_extent_mapping(em_tree
, em
, 0);
254 if (ret
== -EEXIST
) {
256 em
= lookup_extent_mapping(em_tree
, start
, len
);
263 write_unlock(&em_tree
->lock
);
269 u32
btrfs_csum_data(const char *data
, u32 seed
, size_t len
)
271 return btrfs_crc32c(seed
, data
, len
);
274 void btrfs_csum_final(u32 crc
, u8
*result
)
276 put_unaligned_le32(~crc
, result
);
280 * compute the csum for a btree block, and either verify it or write it
281 * into the csum field of the block.
283 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
284 struct extent_buffer
*buf
,
287 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
290 unsigned long cur_len
;
291 unsigned long offset
= BTRFS_CSUM_SIZE
;
293 unsigned long map_start
;
294 unsigned long map_len
;
297 unsigned long inline_result
;
299 len
= buf
->len
- offset
;
301 err
= map_private_extent_buffer(buf
, offset
, 32,
302 &kaddr
, &map_start
, &map_len
);
305 cur_len
= min(len
, map_len
- (offset
- map_start
));
306 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
311 if (csum_size
> sizeof(inline_result
)) {
312 result
= kzalloc(csum_size
, GFP_NOFS
);
316 result
= (char *)&inline_result
;
319 btrfs_csum_final(crc
, result
);
322 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
325 memcpy(&found
, result
, csum_size
);
327 read_extent_buffer(buf
, &val
, 0, csum_size
);
328 btrfs_warn_rl(fs_info
,
329 "%s checksum verify failed on %llu wanted %X found %X level %d",
330 fs_info
->sb
->s_id
, buf
->start
,
331 val
, found
, btrfs_header_level(buf
));
332 if (result
!= (char *)&inline_result
)
337 write_extent_buffer(buf
, result
, 0, csum_size
);
339 if (result
!= (char *)&inline_result
)
345 * we can't consider a given block up to date unless the transid of the
346 * block matches the transid in the parent node's pointer. This is how we
347 * detect blocks that either didn't get written at all or got written
348 * in the wrong place.
350 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
351 struct extent_buffer
*eb
, u64 parent_transid
,
354 struct extent_state
*cached_state
= NULL
;
356 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
358 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
365 btrfs_tree_read_lock(eb
);
366 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
369 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
371 if (extent_buffer_uptodate(eb
) &&
372 btrfs_header_generation(eb
) == parent_transid
) {
376 btrfs_err_rl(eb
->fs_info
,
377 "parent transid verify failed on %llu wanted %llu found %llu",
379 parent_transid
, btrfs_header_generation(eb
));
383 * Things reading via commit roots that don't have normal protection,
384 * like send, can have a really old block in cache that may point at a
385 * block that has been freed and re-allocated. So don't clear uptodate
386 * if we find an eb that is under IO (dirty/writeback) because we could
387 * end up reading in the stale data and then writing it back out and
388 * making everybody very sad.
390 if (!extent_buffer_under_io(eb
))
391 clear_extent_buffer_uptodate(eb
);
393 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
394 &cached_state
, GFP_NOFS
);
396 btrfs_tree_read_unlock_blocking(eb
);
401 * Return 0 if the superblock checksum type matches the checksum value of that
402 * algorithm. Pass the raw disk superblock data.
404 static int btrfs_check_super_csum(struct btrfs_fs_info
*fs_info
,
407 struct btrfs_super_block
*disk_sb
=
408 (struct btrfs_super_block
*)raw_disk_sb
;
409 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
412 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
414 const int csum_size
= sizeof(crc
);
415 char result
[csum_size
];
418 * The super_block structure does not span the whole
419 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420 * is filled with zeros and is included in the checksum.
422 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
423 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
424 btrfs_csum_final(crc
, result
);
426 if (memcmp(raw_disk_sb
, result
, csum_size
))
430 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
431 btrfs_err(fs_info
, "unsupported checksum algorithm %u",
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
443 static int btree_read_extent_buffer_pages(struct btrfs_fs_info
*fs_info
,
444 struct extent_buffer
*eb
,
447 struct extent_io_tree
*io_tree
;
452 int failed_mirror
= 0;
454 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
455 io_tree
= &BTRFS_I(fs_info
->btree_inode
)->io_tree
;
457 ret
= read_extent_buffer_pages(io_tree
, eb
, WAIT_COMPLETE
,
458 btree_get_extent
, mirror_num
);
460 if (!verify_parent_transid(io_tree
, eb
,
468 * This buffer's crc is fine, but its contents are corrupted, so
469 * there is no reason to read the other copies, they won't be
472 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
475 num_copies
= btrfs_num_copies(fs_info
,
480 if (!failed_mirror
) {
482 failed_mirror
= eb
->read_mirror
;
486 if (mirror_num
== failed_mirror
)
489 if (mirror_num
> num_copies
)
493 if (failed
&& !ret
&& failed_mirror
)
494 repair_eb_io_failure(fs_info
, eb
, failed_mirror
);
500 * checksum a dirty tree block before IO. This has extra checks to make sure
501 * we only fill in the checksum field in the first page of a multi-page block
504 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
506 u64 start
= page_offset(page
);
508 struct extent_buffer
*eb
;
510 eb
= (struct extent_buffer
*)page
->private;
511 if (page
!= eb
->pages
[0])
514 found_start
= btrfs_header_bytenr(eb
);
516 * Please do not consolidate these warnings into a single if.
517 * It is useful to know what went wrong.
519 if (WARN_ON(found_start
!= start
))
521 if (WARN_ON(!PageUptodate(page
)))
524 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fsid
,
525 btrfs_header_fsid(), BTRFS_FSID_SIZE
) == 0);
527 return csum_tree_block(fs_info
, eb
, 0);
530 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
531 struct extent_buffer
*eb
)
533 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
534 u8 fsid
[BTRFS_FSID_SIZE
];
537 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
539 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
543 fs_devices
= fs_devices
->seed
;
548 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
549 u64 phy_offset
, struct page
*page
,
550 u64 start
, u64 end
, int mirror
)
554 struct extent_buffer
*eb
;
555 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
556 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
563 eb
= (struct extent_buffer
*)page
->private;
565 /* the pending IO might have been the only thing that kept this buffer
566 * in memory. Make sure we have a ref for all this other checks
568 extent_buffer_get(eb
);
570 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
574 eb
->read_mirror
= mirror
;
575 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
580 found_start
= btrfs_header_bytenr(eb
);
581 if (found_start
!= eb
->start
) {
582 btrfs_err_rl(fs_info
, "bad tree block start %llu %llu",
583 found_start
, eb
->start
);
587 if (check_tree_block_fsid(fs_info
, eb
)) {
588 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
593 found_level
= btrfs_header_level(eb
);
594 if (found_level
>= BTRFS_MAX_LEVEL
) {
595 btrfs_err(fs_info
, "bad tree block level %d",
596 (int)btrfs_header_level(eb
));
601 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
604 ret
= csum_tree_block(fs_info
, eb
, 1);
609 * If this is a leaf block and it is corrupt, set the corrupt bit so
610 * that we don't try and read the other copies of this block, just
613 if (found_level
== 0 && btrfs_check_leaf_full(root
, eb
)) {
614 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
618 if (found_level
> 0 && btrfs_check_node(root
, eb
))
622 set_extent_buffer_uptodate(eb
);
625 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
626 btree_readahead_hook(eb
, ret
);
630 * our io error hook is going to dec the io pages
631 * again, we have to make sure it has something
634 atomic_inc(&eb
->io_pages
);
635 clear_extent_buffer_uptodate(eb
);
637 free_extent_buffer(eb
);
642 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
644 struct extent_buffer
*eb
;
646 eb
= (struct extent_buffer
*)page
->private;
647 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
648 eb
->read_mirror
= failed_mirror
;
649 atomic_dec(&eb
->io_pages
);
650 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
651 btree_readahead_hook(eb
, -EIO
);
652 return -EIO
; /* we fixed nothing */
655 static void end_workqueue_bio(struct bio
*bio
)
657 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
658 struct btrfs_fs_info
*fs_info
;
659 struct btrfs_workqueue
*wq
;
660 btrfs_work_func_t func
;
662 fs_info
= end_io_wq
->info
;
663 end_io_wq
->status
= bio
->bi_status
;
665 if (bio_op(bio
) == REQ_OP_WRITE
) {
666 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
667 wq
= fs_info
->endio_meta_write_workers
;
668 func
= btrfs_endio_meta_write_helper
;
669 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
670 wq
= fs_info
->endio_freespace_worker
;
671 func
= btrfs_freespace_write_helper
;
672 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
673 wq
= fs_info
->endio_raid56_workers
;
674 func
= btrfs_endio_raid56_helper
;
676 wq
= fs_info
->endio_write_workers
;
677 func
= btrfs_endio_write_helper
;
680 if (unlikely(end_io_wq
->metadata
==
681 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
682 wq
= fs_info
->endio_repair_workers
;
683 func
= btrfs_endio_repair_helper
;
684 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
685 wq
= fs_info
->endio_raid56_workers
;
686 func
= btrfs_endio_raid56_helper
;
687 } else if (end_io_wq
->metadata
) {
688 wq
= fs_info
->endio_meta_workers
;
689 func
= btrfs_endio_meta_helper
;
691 wq
= fs_info
->endio_workers
;
692 func
= btrfs_endio_helper
;
696 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
697 btrfs_queue_work(wq
, &end_io_wq
->work
);
700 blk_status_t
btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
701 enum btrfs_wq_endio_type metadata
)
703 struct btrfs_end_io_wq
*end_io_wq
;
705 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
707 return BLK_STS_RESOURCE
;
709 end_io_wq
->private = bio
->bi_private
;
710 end_io_wq
->end_io
= bio
->bi_end_io
;
711 end_io_wq
->info
= info
;
712 end_io_wq
->status
= 0;
713 end_io_wq
->bio
= bio
;
714 end_io_wq
->metadata
= metadata
;
716 bio
->bi_private
= end_io_wq
;
717 bio
->bi_end_io
= end_workqueue_bio
;
721 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
723 unsigned long limit
= min_t(unsigned long,
724 info
->thread_pool_size
,
725 info
->fs_devices
->open_devices
);
729 static void run_one_async_start(struct btrfs_work
*work
)
731 struct async_submit_bio
*async
;
734 async
= container_of(work
, struct async_submit_bio
, work
);
735 ret
= async
->submit_bio_start(async
->private_data
, async
->bio
,
736 async
->mirror_num
, async
->bio_flags
,
742 static void run_one_async_done(struct btrfs_work
*work
)
744 struct async_submit_bio
*async
;
746 async
= container_of(work
, struct async_submit_bio
, work
);
748 /* If an error occurred we just want to clean up the bio and move on */
750 async
->bio
->bi_status
= async
->status
;
751 bio_endio(async
->bio
);
755 async
->submit_bio_done(async
->private_data
, async
->bio
, async
->mirror_num
,
756 async
->bio_flags
, async
->bio_offset
);
759 static void run_one_async_free(struct btrfs_work
*work
)
761 struct async_submit_bio
*async
;
763 async
= container_of(work
, struct async_submit_bio
, work
);
767 blk_status_t
btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
768 int mirror_num
, unsigned long bio_flags
,
769 u64 bio_offset
, void *private_data
,
770 extent_submit_bio_hook_t
*submit_bio_start
,
771 extent_submit_bio_hook_t
*submit_bio_done
)
773 struct async_submit_bio
*async
;
775 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
777 return BLK_STS_RESOURCE
;
779 async
->private_data
= private_data
;
780 async
->fs_info
= fs_info
;
782 async
->mirror_num
= mirror_num
;
783 async
->submit_bio_start
= submit_bio_start
;
784 async
->submit_bio_done
= submit_bio_done
;
786 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
787 run_one_async_done
, run_one_async_free
);
789 async
->bio_flags
= bio_flags
;
790 async
->bio_offset
= bio_offset
;
794 if (op_is_sync(bio
->bi_opf
))
795 btrfs_set_work_high_priority(&async
->work
);
797 btrfs_queue_work(fs_info
->workers
, &async
->work
);
801 static blk_status_t
btree_csum_one_bio(struct bio
*bio
)
803 struct bio_vec
*bvec
;
804 struct btrfs_root
*root
;
807 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
808 bio_for_each_segment_all(bvec
, bio
, i
) {
809 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
810 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
815 return errno_to_blk_status(ret
);
818 static blk_status_t
__btree_submit_bio_start(void *private_data
, struct bio
*bio
,
819 int mirror_num
, unsigned long bio_flags
,
823 * when we're called for a write, we're already in the async
824 * submission context. Just jump into btrfs_map_bio
826 return btree_csum_one_bio(bio
);
829 static blk_status_t
__btree_submit_bio_done(void *private_data
, struct bio
*bio
,
830 int mirror_num
, unsigned long bio_flags
,
833 struct inode
*inode
= private_data
;
837 * when we're called for a write, we're already in the async
838 * submission context. Just jump into btrfs_map_bio
840 ret
= btrfs_map_bio(btrfs_sb(inode
->i_sb
), bio
, mirror_num
, 1);
842 bio
->bi_status
= ret
;
848 static int check_async_write(struct btrfs_inode
*bi
)
850 if (atomic_read(&bi
->sync_writers
))
853 if (static_cpu_has(X86_FEATURE_XMM4_2
))
859 static blk_status_t
btree_submit_bio_hook(void *private_data
, struct bio
*bio
,
860 int mirror_num
, unsigned long bio_flags
,
863 struct inode
*inode
= private_data
;
864 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
865 int async
= check_async_write(BTRFS_I(inode
));
868 if (bio_op(bio
) != REQ_OP_WRITE
) {
870 * called for a read, do the setup so that checksum validation
871 * can happen in the async kernel threads
873 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
874 BTRFS_WQ_ENDIO_METADATA
);
877 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
879 ret
= btree_csum_one_bio(bio
);
882 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
885 * kthread helpers are used to submit writes so that
886 * checksumming can happen in parallel across all CPUs
888 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, 0,
889 bio_offset
, private_data
,
890 __btree_submit_bio_start
,
891 __btree_submit_bio_done
);
899 bio
->bi_status
= ret
;
904 #ifdef CONFIG_MIGRATION
905 static int btree_migratepage(struct address_space
*mapping
,
906 struct page
*newpage
, struct page
*page
,
907 enum migrate_mode mode
)
910 * we can't safely write a btree page from here,
911 * we haven't done the locking hook
916 * Buffers may be managed in a filesystem specific way.
917 * We must have no buffers or drop them.
919 if (page_has_private(page
) &&
920 !try_to_release_page(page
, GFP_KERNEL
))
922 return migrate_page(mapping
, newpage
, page
, mode
);
927 static int btree_writepages(struct address_space
*mapping
,
928 struct writeback_control
*wbc
)
930 struct btrfs_fs_info
*fs_info
;
933 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
935 if (wbc
->for_kupdate
)
938 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
939 /* this is a bit racy, but that's ok */
940 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
941 BTRFS_DIRTY_METADATA_THRESH
);
945 return btree_write_cache_pages(mapping
, wbc
);
948 static int btree_readpage(struct file
*file
, struct page
*page
)
950 struct extent_io_tree
*tree
;
951 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
952 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
955 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
957 if (PageWriteback(page
) || PageDirty(page
))
960 return try_release_extent_buffer(page
);
963 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
966 struct extent_io_tree
*tree
;
967 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
968 extent_invalidatepage(tree
, page
, offset
);
969 btree_releasepage(page
, GFP_NOFS
);
970 if (PagePrivate(page
)) {
971 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
972 "page private not zero on page %llu",
973 (unsigned long long)page_offset(page
));
974 ClearPagePrivate(page
);
975 set_page_private(page
, 0);
980 static int btree_set_page_dirty(struct page
*page
)
983 struct extent_buffer
*eb
;
985 BUG_ON(!PagePrivate(page
));
986 eb
= (struct extent_buffer
*)page
->private;
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
989 BUG_ON(!atomic_read(&eb
->refs
));
990 btrfs_assert_tree_locked(eb
);
992 return __set_page_dirty_nobuffers(page
);
995 static const struct address_space_operations btree_aops
= {
996 .readpage
= btree_readpage
,
997 .writepages
= btree_writepages
,
998 .releasepage
= btree_releasepage
,
999 .invalidatepage
= btree_invalidatepage
,
1000 #ifdef CONFIG_MIGRATION
1001 .migratepage
= btree_migratepage
,
1003 .set_page_dirty
= btree_set_page_dirty
,
1006 void readahead_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
1008 struct extent_buffer
*buf
= NULL
;
1009 struct inode
*btree_inode
= fs_info
->btree_inode
;
1011 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1014 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1015 buf
, WAIT_NONE
, btree_get_extent
, 0);
1016 free_extent_buffer(buf
);
1019 int reada_tree_block_flagged(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1020 int mirror_num
, struct extent_buffer
**eb
)
1022 struct extent_buffer
*buf
= NULL
;
1023 struct inode
*btree_inode
= fs_info
->btree_inode
;
1024 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1027 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1031 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1033 ret
= read_extent_buffer_pages(io_tree
, buf
, WAIT_PAGE_LOCK
,
1034 btree_get_extent
, mirror_num
);
1036 free_extent_buffer(buf
);
1040 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1041 free_extent_buffer(buf
);
1043 } else if (extent_buffer_uptodate(buf
)) {
1046 free_extent_buffer(buf
);
1051 struct extent_buffer
*btrfs_find_create_tree_block(
1052 struct btrfs_fs_info
*fs_info
,
1055 if (btrfs_is_testing(fs_info
))
1056 return alloc_test_extent_buffer(fs_info
, bytenr
);
1057 return alloc_extent_buffer(fs_info
, bytenr
);
1061 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1063 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1064 buf
->start
+ buf
->len
- 1);
1067 void btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1069 filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1070 buf
->start
, buf
->start
+ buf
->len
- 1);
1073 struct extent_buffer
*read_tree_block(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
1076 struct extent_buffer
*buf
= NULL
;
1079 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
1083 ret
= btree_read_extent_buffer_pages(fs_info
, buf
, parent_transid
);
1085 free_extent_buffer(buf
);
1086 return ERR_PTR(ret
);
1092 void clean_tree_block(struct btrfs_fs_info
*fs_info
,
1093 struct extent_buffer
*buf
)
1095 if (btrfs_header_generation(buf
) ==
1096 fs_info
->running_transaction
->transid
) {
1097 btrfs_assert_tree_locked(buf
);
1099 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1100 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
1102 fs_info
->dirty_metadata_batch
);
1103 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1104 btrfs_set_lock_blocking(buf
);
1105 clear_extent_buffer_dirty(buf
);
1110 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1112 struct btrfs_subvolume_writers
*writers
;
1115 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1117 return ERR_PTR(-ENOMEM
);
1119 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1122 return ERR_PTR(ret
);
1125 init_waitqueue_head(&writers
->wait
);
1130 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1132 percpu_counter_destroy(&writers
->counter
);
1136 static void __setup_root(struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1139 bool dummy
= test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO
, &fs_info
->fs_state
);
1141 root
->commit_root
= NULL
;
1143 root
->orphan_cleanup_state
= 0;
1145 root
->objectid
= objectid
;
1146 root
->last_trans
= 0;
1147 root
->highest_objectid
= 0;
1148 root
->nr_delalloc_inodes
= 0;
1149 root
->nr_ordered_extents
= 0;
1151 root
->inode_tree
= RB_ROOT
;
1152 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1153 root
->block_rsv
= NULL
;
1154 root
->orphan_block_rsv
= NULL
;
1156 INIT_LIST_HEAD(&root
->dirty_list
);
1157 INIT_LIST_HEAD(&root
->root_list
);
1158 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1159 INIT_LIST_HEAD(&root
->delalloc_root
);
1160 INIT_LIST_HEAD(&root
->ordered_extents
);
1161 INIT_LIST_HEAD(&root
->ordered_root
);
1162 INIT_LIST_HEAD(&root
->logged_list
[0]);
1163 INIT_LIST_HEAD(&root
->logged_list
[1]);
1164 spin_lock_init(&root
->orphan_lock
);
1165 spin_lock_init(&root
->inode_lock
);
1166 spin_lock_init(&root
->delalloc_lock
);
1167 spin_lock_init(&root
->ordered_extent_lock
);
1168 spin_lock_init(&root
->accounting_lock
);
1169 spin_lock_init(&root
->log_extents_lock
[0]);
1170 spin_lock_init(&root
->log_extents_lock
[1]);
1171 mutex_init(&root
->objectid_mutex
);
1172 mutex_init(&root
->log_mutex
);
1173 mutex_init(&root
->ordered_extent_mutex
);
1174 mutex_init(&root
->delalloc_mutex
);
1175 init_waitqueue_head(&root
->log_writer_wait
);
1176 init_waitqueue_head(&root
->log_commit_wait
[0]);
1177 init_waitqueue_head(&root
->log_commit_wait
[1]);
1178 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1179 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1180 atomic_set(&root
->log_commit
[0], 0);
1181 atomic_set(&root
->log_commit
[1], 0);
1182 atomic_set(&root
->log_writers
, 0);
1183 atomic_set(&root
->log_batch
, 0);
1184 atomic_set(&root
->orphan_inodes
, 0);
1185 refcount_set(&root
->refs
, 1);
1186 atomic_set(&root
->will_be_snapshotted
, 0);
1187 atomic64_set(&root
->qgroup_meta_rsv
, 0);
1188 root
->log_transid
= 0;
1189 root
->log_transid_committed
= -1;
1190 root
->last_log_commit
= 0;
1192 extent_io_tree_init(&root
->dirty_log_pages
, NULL
);
1194 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1195 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1196 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1198 root
->defrag_trans_start
= fs_info
->generation
;
1200 root
->defrag_trans_start
= 0;
1201 root
->root_key
.objectid
= objectid
;
1204 spin_lock_init(&root
->root_item_lock
);
1207 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1210 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1212 root
->fs_info
= fs_info
;
1216 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1217 /* Should only be used by the testing infrastructure */
1218 struct btrfs_root
*btrfs_alloc_dummy_root(struct btrfs_fs_info
*fs_info
)
1220 struct btrfs_root
*root
;
1223 return ERR_PTR(-EINVAL
);
1225 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1227 return ERR_PTR(-ENOMEM
);
1229 /* We don't use the stripesize in selftest, set it as sectorsize */
1230 __setup_root(root
, fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1231 root
->alloc_bytenr
= 0;
1237 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1238 struct btrfs_fs_info
*fs_info
,
1241 struct extent_buffer
*leaf
;
1242 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1243 struct btrfs_root
*root
;
1244 struct btrfs_key key
;
1248 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1250 return ERR_PTR(-ENOMEM
);
1252 __setup_root(root
, fs_info
, objectid
);
1253 root
->root_key
.objectid
= objectid
;
1254 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1255 root
->root_key
.offset
= 0;
1257 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1259 ret
= PTR_ERR(leaf
);
1264 memzero_extent_buffer(leaf
, 0, sizeof(struct btrfs_header
));
1265 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1266 btrfs_set_header_generation(leaf
, trans
->transid
);
1267 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1268 btrfs_set_header_owner(leaf
, objectid
);
1271 write_extent_buffer_fsid(leaf
, fs_info
->fsid
);
1272 write_extent_buffer_chunk_tree_uuid(leaf
, fs_info
->chunk_tree_uuid
);
1273 btrfs_mark_buffer_dirty(leaf
);
1275 root
->commit_root
= btrfs_root_node(root
);
1276 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1278 root
->root_item
.flags
= 0;
1279 root
->root_item
.byte_limit
= 0;
1280 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1281 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1282 btrfs_set_root_level(&root
->root_item
, 0);
1283 btrfs_set_root_refs(&root
->root_item
, 1);
1284 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1285 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1286 btrfs_set_root_dirid(&root
->root_item
, 0);
1288 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1289 root
->root_item
.drop_level
= 0;
1291 key
.objectid
= objectid
;
1292 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1294 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1298 btrfs_tree_unlock(leaf
);
1304 btrfs_tree_unlock(leaf
);
1305 free_extent_buffer(root
->commit_root
);
1306 free_extent_buffer(leaf
);
1310 return ERR_PTR(ret
);
1313 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1314 struct btrfs_fs_info
*fs_info
)
1316 struct btrfs_root
*root
;
1317 struct extent_buffer
*leaf
;
1319 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1321 return ERR_PTR(-ENOMEM
);
1323 __setup_root(root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1325 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1326 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1327 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1330 * DON'T set REF_COWS for log trees
1332 * log trees do not get reference counted because they go away
1333 * before a real commit is actually done. They do store pointers
1334 * to file data extents, and those reference counts still get
1335 * updated (along with back refs to the log tree).
1338 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1342 return ERR_CAST(leaf
);
1345 memzero_extent_buffer(leaf
, 0, sizeof(struct btrfs_header
));
1346 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1347 btrfs_set_header_generation(leaf
, trans
->transid
);
1348 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1349 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1352 write_extent_buffer_fsid(root
->node
, fs_info
->fsid
);
1353 btrfs_mark_buffer_dirty(root
->node
);
1354 btrfs_tree_unlock(root
->node
);
1358 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1359 struct btrfs_fs_info
*fs_info
)
1361 struct btrfs_root
*log_root
;
1363 log_root
= alloc_log_tree(trans
, fs_info
);
1364 if (IS_ERR(log_root
))
1365 return PTR_ERR(log_root
);
1366 WARN_ON(fs_info
->log_root_tree
);
1367 fs_info
->log_root_tree
= log_root
;
1371 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1372 struct btrfs_root
*root
)
1374 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1375 struct btrfs_root
*log_root
;
1376 struct btrfs_inode_item
*inode_item
;
1378 log_root
= alloc_log_tree(trans
, fs_info
);
1379 if (IS_ERR(log_root
))
1380 return PTR_ERR(log_root
);
1382 log_root
->last_trans
= trans
->transid
;
1383 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1385 inode_item
= &log_root
->root_item
.inode
;
1386 btrfs_set_stack_inode_generation(inode_item
, 1);
1387 btrfs_set_stack_inode_size(inode_item
, 3);
1388 btrfs_set_stack_inode_nlink(inode_item
, 1);
1389 btrfs_set_stack_inode_nbytes(inode_item
,
1391 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1393 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1395 WARN_ON(root
->log_root
);
1396 root
->log_root
= log_root
;
1397 root
->log_transid
= 0;
1398 root
->log_transid_committed
= -1;
1399 root
->last_log_commit
= 0;
1403 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1404 struct btrfs_key
*key
)
1406 struct btrfs_root
*root
;
1407 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1408 struct btrfs_path
*path
;
1412 path
= btrfs_alloc_path();
1414 return ERR_PTR(-ENOMEM
);
1416 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1422 __setup_root(root
, fs_info
, key
->objectid
);
1424 ret
= btrfs_find_root(tree_root
, key
, path
,
1425 &root
->root_item
, &root
->root_key
);
1432 generation
= btrfs_root_generation(&root
->root_item
);
1433 root
->node
= read_tree_block(fs_info
,
1434 btrfs_root_bytenr(&root
->root_item
),
1436 if (IS_ERR(root
->node
)) {
1437 ret
= PTR_ERR(root
->node
);
1439 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1441 free_extent_buffer(root
->node
);
1444 root
->commit_root
= btrfs_root_node(root
);
1446 btrfs_free_path(path
);
1452 root
= ERR_PTR(ret
);
1456 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1457 struct btrfs_key
*location
)
1459 struct btrfs_root
*root
;
1461 root
= btrfs_read_tree_root(tree_root
, location
);
1465 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1466 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1467 btrfs_check_and_init_root_item(&root
->root_item
);
1473 int btrfs_init_fs_root(struct btrfs_root
*root
)
1476 struct btrfs_subvolume_writers
*writers
;
1478 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1479 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1481 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1486 writers
= btrfs_alloc_subvolume_writers();
1487 if (IS_ERR(writers
)) {
1488 ret
= PTR_ERR(writers
);
1491 root
->subv_writers
= writers
;
1493 btrfs_init_free_ino_ctl(root
);
1494 spin_lock_init(&root
->ino_cache_lock
);
1495 init_waitqueue_head(&root
->ino_cache_wait
);
1497 ret
= get_anon_bdev(&root
->anon_dev
);
1501 mutex_lock(&root
->objectid_mutex
);
1502 ret
= btrfs_find_highest_objectid(root
,
1503 &root
->highest_objectid
);
1505 mutex_unlock(&root
->objectid_mutex
);
1509 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1511 mutex_unlock(&root
->objectid_mutex
);
1515 /* the caller is responsible to call free_fs_root */
1519 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1522 struct btrfs_root
*root
;
1524 spin_lock(&fs_info
->fs_roots_radix_lock
);
1525 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1526 (unsigned long)root_id
);
1527 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1531 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1532 struct btrfs_root
*root
)
1536 ret
= radix_tree_preload(GFP_NOFS
);
1540 spin_lock(&fs_info
->fs_roots_radix_lock
);
1541 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1542 (unsigned long)root
->root_key
.objectid
,
1545 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1546 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1547 radix_tree_preload_end();
1552 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1553 struct btrfs_key
*location
,
1556 struct btrfs_root
*root
;
1557 struct btrfs_path
*path
;
1558 struct btrfs_key key
;
1561 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1562 return fs_info
->tree_root
;
1563 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1564 return fs_info
->extent_root
;
1565 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1566 return fs_info
->chunk_root
;
1567 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1568 return fs_info
->dev_root
;
1569 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1570 return fs_info
->csum_root
;
1571 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1572 return fs_info
->quota_root
? fs_info
->quota_root
:
1574 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1575 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1577 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1578 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1581 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1583 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1584 return ERR_PTR(-ENOENT
);
1588 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1592 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1597 ret
= btrfs_init_fs_root(root
);
1601 path
= btrfs_alloc_path();
1606 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1607 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1608 key
.offset
= location
->objectid
;
1610 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1611 btrfs_free_path(path
);
1615 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1617 ret
= btrfs_insert_fs_root(fs_info
, root
);
1619 if (ret
== -EEXIST
) {
1628 return ERR_PTR(ret
);
1631 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1633 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1635 struct btrfs_device
*device
;
1636 struct backing_dev_info
*bdi
;
1639 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1642 bdi
= device
->bdev
->bd_bdi
;
1643 if (bdi_congested(bdi
, bdi_bits
)) {
1653 * called by the kthread helper functions to finally call the bio end_io
1654 * functions. This is where read checksum verification actually happens
1656 static void end_workqueue_fn(struct btrfs_work
*work
)
1659 struct btrfs_end_io_wq
*end_io_wq
;
1661 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1662 bio
= end_io_wq
->bio
;
1664 bio
->bi_status
= end_io_wq
->status
;
1665 bio
->bi_private
= end_io_wq
->private;
1666 bio
->bi_end_io
= end_io_wq
->end_io
;
1667 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1671 static int cleaner_kthread(void *arg
)
1673 struct btrfs_root
*root
= arg
;
1674 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1676 struct btrfs_trans_handle
*trans
;
1681 /* Make the cleaner go to sleep early. */
1682 if (btrfs_need_cleaner_sleep(fs_info
))
1686 * Do not do anything if we might cause open_ctree() to block
1687 * before we have finished mounting the filesystem.
1689 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
1692 if (!mutex_trylock(&fs_info
->cleaner_mutex
))
1696 * Avoid the problem that we change the status of the fs
1697 * during the above check and trylock.
1699 if (btrfs_need_cleaner_sleep(fs_info
)) {
1700 mutex_unlock(&fs_info
->cleaner_mutex
);
1704 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
1705 btrfs_run_delayed_iputs(fs_info
);
1706 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
1708 again
= btrfs_clean_one_deleted_snapshot(root
);
1709 mutex_unlock(&fs_info
->cleaner_mutex
);
1712 * The defragger has dealt with the R/O remount and umount,
1713 * needn't do anything special here.
1715 btrfs_run_defrag_inodes(fs_info
);
1718 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1719 * with relocation (btrfs_relocate_chunk) and relocation
1720 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1721 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1722 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1723 * unused block groups.
1725 btrfs_delete_unused_bgs(fs_info
);
1728 set_current_state(TASK_INTERRUPTIBLE
);
1729 if (!kthread_should_stop())
1731 __set_current_state(TASK_RUNNING
);
1733 } while (!kthread_should_stop());
1736 * Transaction kthread is stopped before us and wakes us up.
1737 * However we might have started a new transaction and COWed some
1738 * tree blocks when deleting unused block groups for example. So
1739 * make sure we commit the transaction we started to have a clean
1740 * shutdown when evicting the btree inode - if it has dirty pages
1741 * when we do the final iput() on it, eviction will trigger a
1742 * writeback for it which will fail with null pointer dereferences
1743 * since work queues and other resources were already released and
1744 * destroyed by the time the iput/eviction/writeback is made.
1746 trans
= btrfs_attach_transaction(root
);
1747 if (IS_ERR(trans
)) {
1748 if (PTR_ERR(trans
) != -ENOENT
)
1750 "cleaner transaction attach returned %ld",
1755 ret
= btrfs_commit_transaction(trans
);
1758 "cleaner open transaction commit returned %d",
1765 static int transaction_kthread(void *arg
)
1767 struct btrfs_root
*root
= arg
;
1768 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1769 struct btrfs_trans_handle
*trans
;
1770 struct btrfs_transaction
*cur
;
1773 unsigned long delay
;
1777 cannot_commit
= false;
1778 delay
= HZ
* fs_info
->commit_interval
;
1779 mutex_lock(&fs_info
->transaction_kthread_mutex
);
1781 spin_lock(&fs_info
->trans_lock
);
1782 cur
= fs_info
->running_transaction
;
1784 spin_unlock(&fs_info
->trans_lock
);
1788 now
= get_seconds();
1789 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1790 (now
< cur
->start_time
||
1791 now
- cur
->start_time
< fs_info
->commit_interval
)) {
1792 spin_unlock(&fs_info
->trans_lock
);
1796 transid
= cur
->transid
;
1797 spin_unlock(&fs_info
->trans_lock
);
1799 /* If the file system is aborted, this will always fail. */
1800 trans
= btrfs_attach_transaction(root
);
1801 if (IS_ERR(trans
)) {
1802 if (PTR_ERR(trans
) != -ENOENT
)
1803 cannot_commit
= true;
1806 if (transid
== trans
->transid
) {
1807 btrfs_commit_transaction(trans
);
1809 btrfs_end_transaction(trans
);
1812 wake_up_process(fs_info
->cleaner_kthread
);
1813 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
1815 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1816 &fs_info
->fs_state
)))
1817 btrfs_cleanup_transaction(fs_info
);
1818 set_current_state(TASK_INTERRUPTIBLE
);
1819 if (!kthread_should_stop() &&
1820 (!btrfs_transaction_blocked(fs_info
) ||
1822 schedule_timeout(delay
);
1823 __set_current_state(TASK_RUNNING
);
1824 } while (!kthread_should_stop());
1829 * this will find the highest generation in the array of
1830 * root backups. The index of the highest array is returned,
1831 * or -1 if we can't find anything.
1833 * We check to make sure the array is valid by comparing the
1834 * generation of the latest root in the array with the generation
1835 * in the super block. If they don't match we pitch it.
1837 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1840 int newest_index
= -1;
1841 struct btrfs_root_backup
*root_backup
;
1844 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1845 root_backup
= info
->super_copy
->super_roots
+ i
;
1846 cur
= btrfs_backup_tree_root_gen(root_backup
);
1847 if (cur
== newest_gen
)
1851 /* check to see if we actually wrapped around */
1852 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1853 root_backup
= info
->super_copy
->super_roots
;
1854 cur
= btrfs_backup_tree_root_gen(root_backup
);
1855 if (cur
== newest_gen
)
1858 return newest_index
;
1863 * find the oldest backup so we know where to store new entries
1864 * in the backup array. This will set the backup_root_index
1865 * field in the fs_info struct
1867 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1870 int newest_index
= -1;
1872 newest_index
= find_newest_super_backup(info
, newest_gen
);
1873 /* if there was garbage in there, just move along */
1874 if (newest_index
== -1) {
1875 info
->backup_root_index
= 0;
1877 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1882 * copy all the root pointers into the super backup array.
1883 * this will bump the backup pointer by one when it is
1886 static void backup_super_roots(struct btrfs_fs_info
*info
)
1889 struct btrfs_root_backup
*root_backup
;
1892 next_backup
= info
->backup_root_index
;
1893 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1894 BTRFS_NUM_BACKUP_ROOTS
;
1897 * just overwrite the last backup if we're at the same generation
1898 * this happens only at umount
1900 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1901 if (btrfs_backup_tree_root_gen(root_backup
) ==
1902 btrfs_header_generation(info
->tree_root
->node
))
1903 next_backup
= last_backup
;
1905 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1908 * make sure all of our padding and empty slots get zero filled
1909 * regardless of which ones we use today
1911 memset(root_backup
, 0, sizeof(*root_backup
));
1913 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1915 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1916 btrfs_set_backup_tree_root_gen(root_backup
,
1917 btrfs_header_generation(info
->tree_root
->node
));
1919 btrfs_set_backup_tree_root_level(root_backup
,
1920 btrfs_header_level(info
->tree_root
->node
));
1922 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1923 btrfs_set_backup_chunk_root_gen(root_backup
,
1924 btrfs_header_generation(info
->chunk_root
->node
));
1925 btrfs_set_backup_chunk_root_level(root_backup
,
1926 btrfs_header_level(info
->chunk_root
->node
));
1928 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1929 btrfs_set_backup_extent_root_gen(root_backup
,
1930 btrfs_header_generation(info
->extent_root
->node
));
1931 btrfs_set_backup_extent_root_level(root_backup
,
1932 btrfs_header_level(info
->extent_root
->node
));
1935 * we might commit during log recovery, which happens before we set
1936 * the fs_root. Make sure it is valid before we fill it in.
1938 if (info
->fs_root
&& info
->fs_root
->node
) {
1939 btrfs_set_backup_fs_root(root_backup
,
1940 info
->fs_root
->node
->start
);
1941 btrfs_set_backup_fs_root_gen(root_backup
,
1942 btrfs_header_generation(info
->fs_root
->node
));
1943 btrfs_set_backup_fs_root_level(root_backup
,
1944 btrfs_header_level(info
->fs_root
->node
));
1947 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1948 btrfs_set_backup_dev_root_gen(root_backup
,
1949 btrfs_header_generation(info
->dev_root
->node
));
1950 btrfs_set_backup_dev_root_level(root_backup
,
1951 btrfs_header_level(info
->dev_root
->node
));
1953 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1954 btrfs_set_backup_csum_root_gen(root_backup
,
1955 btrfs_header_generation(info
->csum_root
->node
));
1956 btrfs_set_backup_csum_root_level(root_backup
,
1957 btrfs_header_level(info
->csum_root
->node
));
1959 btrfs_set_backup_total_bytes(root_backup
,
1960 btrfs_super_total_bytes(info
->super_copy
));
1961 btrfs_set_backup_bytes_used(root_backup
,
1962 btrfs_super_bytes_used(info
->super_copy
));
1963 btrfs_set_backup_num_devices(root_backup
,
1964 btrfs_super_num_devices(info
->super_copy
));
1967 * if we don't copy this out to the super_copy, it won't get remembered
1968 * for the next commit
1970 memcpy(&info
->super_copy
->super_roots
,
1971 &info
->super_for_commit
->super_roots
,
1972 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1976 * this copies info out of the root backup array and back into
1977 * the in-memory super block. It is meant to help iterate through
1978 * the array, so you send it the number of backups you've already
1979 * tried and the last backup index you used.
1981 * this returns -1 when it has tried all the backups
1983 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1984 struct btrfs_super_block
*super
,
1985 int *num_backups_tried
, int *backup_index
)
1987 struct btrfs_root_backup
*root_backup
;
1988 int newest
= *backup_index
;
1990 if (*num_backups_tried
== 0) {
1991 u64 gen
= btrfs_super_generation(super
);
1993 newest
= find_newest_super_backup(info
, gen
);
1997 *backup_index
= newest
;
1998 *num_backups_tried
= 1;
1999 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2000 /* we've tried all the backups, all done */
2003 /* jump to the next oldest backup */
2004 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2005 BTRFS_NUM_BACKUP_ROOTS
;
2006 *backup_index
= newest
;
2007 *num_backups_tried
+= 1;
2009 root_backup
= super
->super_roots
+ newest
;
2011 btrfs_set_super_generation(super
,
2012 btrfs_backup_tree_root_gen(root_backup
));
2013 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2014 btrfs_set_super_root_level(super
,
2015 btrfs_backup_tree_root_level(root_backup
));
2016 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2019 * fixme: the total bytes and num_devices need to match or we should
2022 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2023 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2027 /* helper to cleanup workers */
2028 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2030 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2031 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2032 btrfs_destroy_workqueue(fs_info
->workers
);
2033 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2034 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2035 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2036 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2037 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2038 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2039 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2040 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2041 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2042 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2043 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2044 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2045 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2047 * Now that all other work queues are destroyed, we can safely destroy
2048 * the queues used for metadata I/O, since tasks from those other work
2049 * queues can do metadata I/O operations.
2051 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2052 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2055 static void free_root_extent_buffers(struct btrfs_root
*root
)
2058 free_extent_buffer(root
->node
);
2059 free_extent_buffer(root
->commit_root
);
2061 root
->commit_root
= NULL
;
2065 /* helper to cleanup tree roots */
2066 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2068 free_root_extent_buffers(info
->tree_root
);
2070 free_root_extent_buffers(info
->dev_root
);
2071 free_root_extent_buffers(info
->extent_root
);
2072 free_root_extent_buffers(info
->csum_root
);
2073 free_root_extent_buffers(info
->quota_root
);
2074 free_root_extent_buffers(info
->uuid_root
);
2076 free_root_extent_buffers(info
->chunk_root
);
2077 free_root_extent_buffers(info
->free_space_root
);
2080 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2083 struct btrfs_root
*gang
[8];
2086 while (!list_empty(&fs_info
->dead_roots
)) {
2087 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2088 struct btrfs_root
, root_list
);
2089 list_del(&gang
[0]->root_list
);
2091 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2092 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2094 free_extent_buffer(gang
[0]->node
);
2095 free_extent_buffer(gang
[0]->commit_root
);
2096 btrfs_put_fs_root(gang
[0]);
2101 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2106 for (i
= 0; i
< ret
; i
++)
2107 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2110 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2111 btrfs_free_log_root_tree(NULL
, fs_info
);
2112 btrfs_destroy_pinned_extent(fs_info
, fs_info
->pinned_extents
);
2116 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2118 mutex_init(&fs_info
->scrub_lock
);
2119 atomic_set(&fs_info
->scrubs_running
, 0);
2120 atomic_set(&fs_info
->scrub_pause_req
, 0);
2121 atomic_set(&fs_info
->scrubs_paused
, 0);
2122 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2123 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2124 fs_info
->scrub_workers_refcnt
= 0;
2127 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2129 spin_lock_init(&fs_info
->balance_lock
);
2130 mutex_init(&fs_info
->balance_mutex
);
2131 atomic_set(&fs_info
->balance_running
, 0);
2132 atomic_set(&fs_info
->balance_pause_req
, 0);
2133 atomic_set(&fs_info
->balance_cancel_req
, 0);
2134 fs_info
->balance_ctl
= NULL
;
2135 init_waitqueue_head(&fs_info
->balance_wait_q
);
2138 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
)
2140 struct inode
*inode
= fs_info
->btree_inode
;
2142 inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2143 set_nlink(inode
, 1);
2145 * we set the i_size on the btree inode to the max possible int.
2146 * the real end of the address space is determined by all of
2147 * the devices in the system
2149 inode
->i_size
= OFFSET_MAX
;
2150 inode
->i_mapping
->a_ops
= &btree_aops
;
2152 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
2153 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
, inode
);
2154 BTRFS_I(inode
)->io_tree
.track_uptodate
= 0;
2155 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
);
2157 BTRFS_I(inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2159 BTRFS_I(inode
)->root
= fs_info
->tree_root
;
2160 memset(&BTRFS_I(inode
)->location
, 0, sizeof(struct btrfs_key
));
2161 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
2162 btrfs_insert_inode_hash(inode
);
2165 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2167 fs_info
->dev_replace
.lock_owner
= 0;
2168 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2169 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2170 rwlock_init(&fs_info
->dev_replace
.lock
);
2171 atomic_set(&fs_info
->dev_replace
.read_locks
, 0);
2172 atomic_set(&fs_info
->dev_replace
.blocking_readers
, 0);
2173 init_waitqueue_head(&fs_info
->replace_wait
);
2174 init_waitqueue_head(&fs_info
->dev_replace
.read_lock_wq
);
2177 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2179 spin_lock_init(&fs_info
->qgroup_lock
);
2180 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2181 fs_info
->qgroup_tree
= RB_ROOT
;
2182 fs_info
->qgroup_op_tree
= RB_ROOT
;
2183 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2184 fs_info
->qgroup_seq
= 1;
2185 fs_info
->qgroup_ulist
= NULL
;
2186 fs_info
->qgroup_rescan_running
= false;
2187 mutex_init(&fs_info
->qgroup_rescan_lock
);
2190 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2191 struct btrfs_fs_devices
*fs_devices
)
2193 int max_active
= fs_info
->thread_pool_size
;
2194 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2197 btrfs_alloc_workqueue(fs_info
, "worker",
2198 flags
| WQ_HIGHPRI
, max_active
, 16);
2200 fs_info
->delalloc_workers
=
2201 btrfs_alloc_workqueue(fs_info
, "delalloc",
2202 flags
, max_active
, 2);
2204 fs_info
->flush_workers
=
2205 btrfs_alloc_workqueue(fs_info
, "flush_delalloc",
2206 flags
, max_active
, 0);
2208 fs_info
->caching_workers
=
2209 btrfs_alloc_workqueue(fs_info
, "cache", flags
, max_active
, 0);
2212 * a higher idle thresh on the submit workers makes it much more
2213 * likely that bios will be send down in a sane order to the
2216 fs_info
->submit_workers
=
2217 btrfs_alloc_workqueue(fs_info
, "submit", flags
,
2218 min_t(u64
, fs_devices
->num_devices
,
2221 fs_info
->fixup_workers
=
2222 btrfs_alloc_workqueue(fs_info
, "fixup", flags
, 1, 0);
2225 * endios are largely parallel and should have a very
2228 fs_info
->endio_workers
=
2229 btrfs_alloc_workqueue(fs_info
, "endio", flags
, max_active
, 4);
2230 fs_info
->endio_meta_workers
=
2231 btrfs_alloc_workqueue(fs_info
, "endio-meta", flags
,
2233 fs_info
->endio_meta_write_workers
=
2234 btrfs_alloc_workqueue(fs_info
, "endio-meta-write", flags
,
2236 fs_info
->endio_raid56_workers
=
2237 btrfs_alloc_workqueue(fs_info
, "endio-raid56", flags
,
2239 fs_info
->endio_repair_workers
=
2240 btrfs_alloc_workqueue(fs_info
, "endio-repair", flags
, 1, 0);
2241 fs_info
->rmw_workers
=
2242 btrfs_alloc_workqueue(fs_info
, "rmw", flags
, max_active
, 2);
2243 fs_info
->endio_write_workers
=
2244 btrfs_alloc_workqueue(fs_info
, "endio-write", flags
,
2246 fs_info
->endio_freespace_worker
=
2247 btrfs_alloc_workqueue(fs_info
, "freespace-write", flags
,
2249 fs_info
->delayed_workers
=
2250 btrfs_alloc_workqueue(fs_info
, "delayed-meta", flags
,
2252 fs_info
->readahead_workers
=
2253 btrfs_alloc_workqueue(fs_info
, "readahead", flags
,
2255 fs_info
->qgroup_rescan_workers
=
2256 btrfs_alloc_workqueue(fs_info
, "qgroup-rescan", flags
, 1, 0);
2257 fs_info
->extent_workers
=
2258 btrfs_alloc_workqueue(fs_info
, "extent-refs", flags
,
2259 min_t(u64
, fs_devices
->num_devices
,
2262 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2263 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2264 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2265 fs_info
->endio_meta_write_workers
&&
2266 fs_info
->endio_repair_workers
&&
2267 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2268 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2269 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2270 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2271 fs_info
->extent_workers
&&
2272 fs_info
->qgroup_rescan_workers
)) {
2279 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2280 struct btrfs_fs_devices
*fs_devices
)
2283 struct btrfs_root
*log_tree_root
;
2284 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2285 u64 bytenr
= btrfs_super_log_root(disk_super
);
2287 if (fs_devices
->rw_devices
== 0) {
2288 btrfs_warn(fs_info
, "log replay required on RO media");
2292 log_tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2296 __setup_root(log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2298 log_tree_root
->node
= read_tree_block(fs_info
, bytenr
,
2299 fs_info
->generation
+ 1);
2300 if (IS_ERR(log_tree_root
->node
)) {
2301 btrfs_warn(fs_info
, "failed to read log tree");
2302 ret
= PTR_ERR(log_tree_root
->node
);
2303 kfree(log_tree_root
);
2305 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2306 btrfs_err(fs_info
, "failed to read log tree");
2307 free_extent_buffer(log_tree_root
->node
);
2308 kfree(log_tree_root
);
2311 /* returns with log_tree_root freed on success */
2312 ret
= btrfs_recover_log_trees(log_tree_root
);
2314 btrfs_handle_fs_error(fs_info
, ret
,
2315 "Failed to recover log tree");
2316 free_extent_buffer(log_tree_root
->node
);
2317 kfree(log_tree_root
);
2321 if (sb_rdonly(fs_info
->sb
)) {
2322 ret
= btrfs_commit_super(fs_info
);
2330 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
)
2332 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2333 struct btrfs_root
*root
;
2334 struct btrfs_key location
;
2337 BUG_ON(!fs_info
->tree_root
);
2339 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2340 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2341 location
.offset
= 0;
2343 root
= btrfs_read_tree_root(tree_root
, &location
);
2345 return PTR_ERR(root
);
2346 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2347 fs_info
->extent_root
= root
;
2349 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2350 root
= btrfs_read_tree_root(tree_root
, &location
);
2352 return PTR_ERR(root
);
2353 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2354 fs_info
->dev_root
= root
;
2355 btrfs_init_devices_late(fs_info
);
2357 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2358 root
= btrfs_read_tree_root(tree_root
, &location
);
2360 return PTR_ERR(root
);
2361 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2362 fs_info
->csum_root
= root
;
2364 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2365 root
= btrfs_read_tree_root(tree_root
, &location
);
2366 if (!IS_ERR(root
)) {
2367 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2368 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2369 fs_info
->quota_root
= root
;
2372 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2373 root
= btrfs_read_tree_root(tree_root
, &location
);
2375 ret
= PTR_ERR(root
);
2379 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2380 fs_info
->uuid_root
= root
;
2383 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2384 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2385 root
= btrfs_read_tree_root(tree_root
, &location
);
2387 return PTR_ERR(root
);
2388 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2389 fs_info
->free_space_root
= root
;
2395 int open_ctree(struct super_block
*sb
,
2396 struct btrfs_fs_devices
*fs_devices
,
2404 struct btrfs_key location
;
2405 struct buffer_head
*bh
;
2406 struct btrfs_super_block
*disk_super
;
2407 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2408 struct btrfs_root
*tree_root
;
2409 struct btrfs_root
*chunk_root
;
2412 int num_backups_tried
= 0;
2413 int backup_index
= 0;
2415 int clear_free_space_tree
= 0;
2417 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2418 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2419 if (!tree_root
|| !chunk_root
) {
2424 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2430 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2435 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2436 (1 + ilog2(nr_cpu_ids
));
2438 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2441 goto fail_dirty_metadata_bytes
;
2444 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2447 goto fail_delalloc_bytes
;
2450 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2451 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2452 INIT_LIST_HEAD(&fs_info
->trans_list
);
2453 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2454 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2455 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2456 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2457 spin_lock_init(&fs_info
->delalloc_root_lock
);
2458 spin_lock_init(&fs_info
->trans_lock
);
2459 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2460 spin_lock_init(&fs_info
->delayed_iput_lock
);
2461 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2462 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2463 spin_lock_init(&fs_info
->super_lock
);
2464 spin_lock_init(&fs_info
->qgroup_op_lock
);
2465 spin_lock_init(&fs_info
->buffer_lock
);
2466 spin_lock_init(&fs_info
->unused_bgs_lock
);
2467 rwlock_init(&fs_info
->tree_mod_log_lock
);
2468 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2469 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2470 mutex_init(&fs_info
->reloc_mutex
);
2471 mutex_init(&fs_info
->delalloc_root_mutex
);
2472 mutex_init(&fs_info
->cleaner_delayed_iput_mutex
);
2473 seqlock_init(&fs_info
->profiles_lock
);
2475 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2476 INIT_LIST_HEAD(&fs_info
->space_info
);
2477 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2478 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2479 btrfs_mapping_init(&fs_info
->mapping_tree
);
2480 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2481 BTRFS_BLOCK_RSV_GLOBAL
);
2482 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2483 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2484 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2485 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2486 BTRFS_BLOCK_RSV_DELOPS
);
2487 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2488 atomic_set(&fs_info
->defrag_running
, 0);
2489 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2490 atomic_set(&fs_info
->reada_works_cnt
, 0);
2491 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2493 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2494 fs_info
->metadata_ratio
= 0;
2495 fs_info
->defrag_inodes
= RB_ROOT
;
2496 atomic64_set(&fs_info
->free_chunk_space
, 0);
2497 fs_info
->tree_mod_log
= RB_ROOT
;
2498 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2499 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2500 /* readahead state */
2501 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2502 spin_lock_init(&fs_info
->reada_lock
);
2503 btrfs_init_ref_verify(fs_info
);
2505 fs_info
->thread_pool_size
= min_t(unsigned long,
2506 num_online_cpus() + 2, 8);
2508 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2509 spin_lock_init(&fs_info
->ordered_root_lock
);
2511 fs_info
->btree_inode
= new_inode(sb
);
2512 if (!fs_info
->btree_inode
) {
2514 goto fail_bio_counter
;
2516 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2518 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2520 if (!fs_info
->delayed_root
) {
2524 btrfs_init_delayed_root(fs_info
->delayed_root
);
2526 btrfs_init_scrub(fs_info
);
2527 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2528 fs_info
->check_integrity_print_mask
= 0;
2530 btrfs_init_balance(fs_info
);
2531 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2533 sb
->s_blocksize
= BTRFS_BDEV_BLOCKSIZE
;
2534 sb
->s_blocksize_bits
= blksize_bits(BTRFS_BDEV_BLOCKSIZE
);
2536 btrfs_init_btree_inode(fs_info
);
2538 spin_lock_init(&fs_info
->block_group_cache_lock
);
2539 fs_info
->block_group_cache_tree
= RB_ROOT
;
2540 fs_info
->first_logical_byte
= (u64
)-1;
2542 extent_io_tree_init(&fs_info
->freed_extents
[0], NULL
);
2543 extent_io_tree_init(&fs_info
->freed_extents
[1], NULL
);
2544 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2545 set_bit(BTRFS_FS_BARRIER
, &fs_info
->flags
);
2547 mutex_init(&fs_info
->ordered_operations_mutex
);
2548 mutex_init(&fs_info
->tree_log_mutex
);
2549 mutex_init(&fs_info
->chunk_mutex
);
2550 mutex_init(&fs_info
->transaction_kthread_mutex
);
2551 mutex_init(&fs_info
->cleaner_mutex
);
2552 mutex_init(&fs_info
->volume_mutex
);
2553 mutex_init(&fs_info
->ro_block_group_mutex
);
2554 init_rwsem(&fs_info
->commit_root_sem
);
2555 init_rwsem(&fs_info
->cleanup_work_sem
);
2556 init_rwsem(&fs_info
->subvol_sem
);
2557 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2559 btrfs_init_dev_replace_locks(fs_info
);
2560 btrfs_init_qgroup(fs_info
);
2562 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2563 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2565 init_waitqueue_head(&fs_info
->transaction_throttle
);
2566 init_waitqueue_head(&fs_info
->transaction_wait
);
2567 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2568 init_waitqueue_head(&fs_info
->async_submit_wait
);
2570 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2572 /* Usable values until the real ones are cached from the superblock */
2573 fs_info
->nodesize
= 4096;
2574 fs_info
->sectorsize
= 4096;
2575 fs_info
->stripesize
= 4096;
2577 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2583 __setup_root(tree_root
, fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2585 invalidate_bdev(fs_devices
->latest_bdev
);
2588 * Read super block and check the signature bytes only
2590 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2597 * We want to check superblock checksum, the type is stored inside.
2598 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2600 if (btrfs_check_super_csum(fs_info
, bh
->b_data
)) {
2601 btrfs_err(fs_info
, "superblock checksum mismatch");
2608 * super_copy is zeroed at allocation time and we never touch the
2609 * following bytes up to INFO_SIZE, the checksum is calculated from
2610 * the whole block of INFO_SIZE
2612 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2613 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2614 sizeof(*fs_info
->super_for_commit
));
2617 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2619 ret
= btrfs_check_super_valid(fs_info
);
2621 btrfs_err(fs_info
, "superblock contains fatal errors");
2626 disk_super
= fs_info
->super_copy
;
2627 if (!btrfs_super_root(disk_super
))
2630 /* check FS state, whether FS is broken. */
2631 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2632 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2635 * run through our array of backup supers and setup
2636 * our ring pointer to the oldest one
2638 generation
= btrfs_super_generation(disk_super
);
2639 find_oldest_super_backup(fs_info
, generation
);
2642 * In the long term, we'll store the compression type in the super
2643 * block, and it'll be used for per file compression control.
2645 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2647 ret
= btrfs_parse_options(fs_info
, options
, sb
->s_flags
);
2653 features
= btrfs_super_incompat_flags(disk_super
) &
2654 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2657 "cannot mount because of unsupported optional features (%llx)",
2663 features
= btrfs_super_incompat_flags(disk_super
);
2664 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2665 if (fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2666 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2667 else if (fs_info
->compress_type
== BTRFS_COMPRESS_ZSTD
)
2668 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD
;
2670 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2671 btrfs_info(fs_info
, "has skinny extents");
2674 * flag our filesystem as having big metadata blocks if
2675 * they are bigger than the page size
2677 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
2678 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2680 "flagging fs with big metadata feature");
2681 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2684 nodesize
= btrfs_super_nodesize(disk_super
);
2685 sectorsize
= btrfs_super_sectorsize(disk_super
);
2686 stripesize
= sectorsize
;
2687 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2688 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2690 /* Cache block sizes */
2691 fs_info
->nodesize
= nodesize
;
2692 fs_info
->sectorsize
= sectorsize
;
2693 fs_info
->stripesize
= stripesize
;
2696 * mixed block groups end up with duplicate but slightly offset
2697 * extent buffers for the same range. It leads to corruptions
2699 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2700 (sectorsize
!= nodesize
)) {
2702 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2703 nodesize
, sectorsize
);
2708 * Needn't use the lock because there is no other task which will
2711 btrfs_set_super_incompat_flags(disk_super
, features
);
2713 features
= btrfs_super_compat_ro_flags(disk_super
) &
2714 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2715 if (!sb_rdonly(sb
) && features
) {
2717 "cannot mount read-write because of unsupported optional features (%llx)",
2723 max_active
= fs_info
->thread_pool_size
;
2725 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2728 goto fail_sb_buffer
;
2731 sb
->s_bdi
->congested_fn
= btrfs_congested_fn
;
2732 sb
->s_bdi
->congested_data
= fs_info
;
2733 sb
->s_bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
2734 sb
->s_bdi
->ra_pages
= VM_MAX_READAHEAD
* SZ_1K
/ PAGE_SIZE
;
2735 sb
->s_bdi
->ra_pages
*= btrfs_super_num_devices(disk_super
);
2736 sb
->s_bdi
->ra_pages
= max(sb
->s_bdi
->ra_pages
, SZ_4M
/ PAGE_SIZE
);
2738 sb
->s_blocksize
= sectorsize
;
2739 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2740 memcpy(&sb
->s_uuid
, fs_info
->fsid
, BTRFS_FSID_SIZE
);
2742 mutex_lock(&fs_info
->chunk_mutex
);
2743 ret
= btrfs_read_sys_array(fs_info
);
2744 mutex_unlock(&fs_info
->chunk_mutex
);
2746 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
2747 goto fail_sb_buffer
;
2750 generation
= btrfs_super_chunk_root_generation(disk_super
);
2752 __setup_root(chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2754 chunk_root
->node
= read_tree_block(fs_info
,
2755 btrfs_super_chunk_root(disk_super
),
2757 if (IS_ERR(chunk_root
->node
) ||
2758 !extent_buffer_uptodate(chunk_root
->node
)) {
2759 btrfs_err(fs_info
, "failed to read chunk root");
2760 if (!IS_ERR(chunk_root
->node
))
2761 free_extent_buffer(chunk_root
->node
);
2762 chunk_root
->node
= NULL
;
2763 goto fail_tree_roots
;
2765 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2766 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2768 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2769 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2771 ret
= btrfs_read_chunk_tree(fs_info
);
2773 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
2774 goto fail_tree_roots
;
2778 * keep the device that is marked to be the target device for the
2779 * dev_replace procedure
2781 btrfs_close_extra_devices(fs_devices
, 0);
2783 if (!fs_devices
->latest_bdev
) {
2784 btrfs_err(fs_info
, "failed to read devices");
2785 goto fail_tree_roots
;
2789 generation
= btrfs_super_generation(disk_super
);
2791 tree_root
->node
= read_tree_block(fs_info
,
2792 btrfs_super_root(disk_super
),
2794 if (IS_ERR(tree_root
->node
) ||
2795 !extent_buffer_uptodate(tree_root
->node
)) {
2796 btrfs_warn(fs_info
, "failed to read tree root");
2797 if (!IS_ERR(tree_root
->node
))
2798 free_extent_buffer(tree_root
->node
);
2799 tree_root
->node
= NULL
;
2800 goto recovery_tree_root
;
2803 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2804 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2805 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2807 mutex_lock(&tree_root
->objectid_mutex
);
2808 ret
= btrfs_find_highest_objectid(tree_root
,
2809 &tree_root
->highest_objectid
);
2811 mutex_unlock(&tree_root
->objectid_mutex
);
2812 goto recovery_tree_root
;
2815 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2817 mutex_unlock(&tree_root
->objectid_mutex
);
2819 ret
= btrfs_read_roots(fs_info
);
2821 goto recovery_tree_root
;
2823 fs_info
->generation
= generation
;
2824 fs_info
->last_trans_committed
= generation
;
2826 ret
= btrfs_recover_balance(fs_info
);
2828 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
2829 goto fail_block_groups
;
2832 ret
= btrfs_init_dev_stats(fs_info
);
2834 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
2835 goto fail_block_groups
;
2838 ret
= btrfs_init_dev_replace(fs_info
);
2840 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
2841 goto fail_block_groups
;
2844 btrfs_close_extra_devices(fs_devices
, 1);
2846 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2848 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
2850 goto fail_block_groups
;
2853 ret
= btrfs_sysfs_add_device(fs_devices
);
2855 btrfs_err(fs_info
, "failed to init sysfs device interface: %d",
2857 goto fail_fsdev_sysfs
;
2860 ret
= btrfs_sysfs_add_mounted(fs_info
);
2862 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
2863 goto fail_fsdev_sysfs
;
2866 ret
= btrfs_init_space_info(fs_info
);
2868 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
2872 ret
= btrfs_read_block_groups(fs_info
);
2874 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
2878 if (!sb_rdonly(sb
) && !btrfs_check_rw_degradable(fs_info
)) {
2880 "writeable mount is not allowed due to too many missing devices");
2884 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2886 if (IS_ERR(fs_info
->cleaner_kthread
))
2889 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2891 "btrfs-transaction");
2892 if (IS_ERR(fs_info
->transaction_kthread
))
2895 if (!btrfs_test_opt(fs_info
, NOSSD
) &&
2896 !fs_info
->fs_devices
->rotating
) {
2897 btrfs_set_and_info(fs_info
, SSD
, "enabling ssd optimizations");
2901 * Mount does not set all options immediately, we can do it now and do
2902 * not have to wait for transaction commit
2904 btrfs_apply_pending_changes(fs_info
);
2906 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2907 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
)) {
2908 ret
= btrfsic_mount(fs_info
, fs_devices
,
2909 btrfs_test_opt(fs_info
,
2910 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2912 fs_info
->check_integrity_print_mask
);
2915 "failed to initialize integrity check module: %d",
2919 ret
= btrfs_read_qgroup_config(fs_info
);
2921 goto fail_trans_kthread
;
2923 if (btrfs_build_ref_tree(fs_info
))
2924 btrfs_err(fs_info
, "couldn't build ref tree");
2926 /* do not make disk changes in broken FS or nologreplay is given */
2927 if (btrfs_super_log_root(disk_super
) != 0 &&
2928 !btrfs_test_opt(fs_info
, NOLOGREPLAY
)) {
2929 ret
= btrfs_replay_log(fs_info
, fs_devices
);
2936 ret
= btrfs_find_orphan_roots(fs_info
);
2940 if (!sb_rdonly(sb
)) {
2941 ret
= btrfs_cleanup_fs_roots(fs_info
);
2945 mutex_lock(&fs_info
->cleaner_mutex
);
2946 ret
= btrfs_recover_relocation(tree_root
);
2947 mutex_unlock(&fs_info
->cleaner_mutex
);
2949 btrfs_warn(fs_info
, "failed to recover relocation: %d",
2956 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2957 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2958 location
.offset
= 0;
2960 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2961 if (IS_ERR(fs_info
->fs_root
)) {
2962 err
= PTR_ERR(fs_info
->fs_root
);
2969 if (btrfs_test_opt(fs_info
, CLEAR_CACHE
) &&
2970 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2971 clear_free_space_tree
= 1;
2972 } else if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
2973 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE_VALID
)) {
2974 btrfs_warn(fs_info
, "free space tree is invalid");
2975 clear_free_space_tree
= 1;
2978 if (clear_free_space_tree
) {
2979 btrfs_info(fs_info
, "clearing free space tree");
2980 ret
= btrfs_clear_free_space_tree(fs_info
);
2983 "failed to clear free space tree: %d", ret
);
2984 close_ctree(fs_info
);
2989 if (btrfs_test_opt(fs_info
, FREE_SPACE_TREE
) &&
2990 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2991 btrfs_info(fs_info
, "creating free space tree");
2992 ret
= btrfs_create_free_space_tree(fs_info
);
2995 "failed to create free space tree: %d", ret
);
2996 close_ctree(fs_info
);
3001 down_read(&fs_info
->cleanup_work_sem
);
3002 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3003 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3004 up_read(&fs_info
->cleanup_work_sem
);
3005 close_ctree(fs_info
);
3008 up_read(&fs_info
->cleanup_work_sem
);
3010 ret
= btrfs_resume_balance_async(fs_info
);
3012 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3013 close_ctree(fs_info
);
3017 ret
= btrfs_resume_dev_replace_async(fs_info
);
3019 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3020 close_ctree(fs_info
);
3024 btrfs_qgroup_rescan_resume(fs_info
);
3026 if (!fs_info
->uuid_root
) {
3027 btrfs_info(fs_info
, "creating UUID tree");
3028 ret
= btrfs_create_uuid_tree(fs_info
);
3031 "failed to create the UUID tree: %d", ret
);
3032 close_ctree(fs_info
);
3035 } else if (btrfs_test_opt(fs_info
, RESCAN_UUID_TREE
) ||
3036 fs_info
->generation
!=
3037 btrfs_super_uuid_tree_generation(disk_super
)) {
3038 btrfs_info(fs_info
, "checking UUID tree");
3039 ret
= btrfs_check_uuid_tree(fs_info
);
3042 "failed to check the UUID tree: %d", ret
);
3043 close_ctree(fs_info
);
3047 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
3049 set_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3052 * backuproot only affect mount behavior, and if open_ctree succeeded,
3053 * no need to keep the flag
3055 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3060 btrfs_free_qgroup_config(fs_info
);
3062 kthread_stop(fs_info
->transaction_kthread
);
3063 btrfs_cleanup_transaction(fs_info
);
3064 btrfs_free_fs_roots(fs_info
);
3066 kthread_stop(fs_info
->cleaner_kthread
);
3069 * make sure we're done with the btree inode before we stop our
3072 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3075 btrfs_sysfs_remove_mounted(fs_info
);
3078 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3081 btrfs_put_block_group_cache(fs_info
);
3084 free_root_pointers(fs_info
, 1);
3085 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3088 btrfs_stop_all_workers(fs_info
);
3089 btrfs_free_block_groups(fs_info
);
3092 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3094 iput(fs_info
->btree_inode
);
3096 percpu_counter_destroy(&fs_info
->bio_counter
);
3097 fail_delalloc_bytes
:
3098 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3099 fail_dirty_metadata_bytes
:
3100 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3102 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3104 btrfs_free_stripe_hash_table(fs_info
);
3105 btrfs_close_devices(fs_info
->fs_devices
);
3109 if (!btrfs_test_opt(fs_info
, USEBACKUPROOT
))
3110 goto fail_tree_roots
;
3112 free_root_pointers(fs_info
, 0);
3114 /* don't use the log in recovery mode, it won't be valid */
3115 btrfs_set_super_log_root(disk_super
, 0);
3117 /* we can't trust the free space cache either */
3118 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3120 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3121 &num_backups_tried
, &backup_index
);
3123 goto fail_block_groups
;
3124 goto retry_root_backup
;
3127 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3130 set_buffer_uptodate(bh
);
3132 struct btrfs_device
*device
= (struct btrfs_device
*)
3135 btrfs_warn_rl_in_rcu(device
->fs_info
,
3136 "lost page write due to IO error on %s",
3137 rcu_str_deref(device
->name
));
3138 /* note, we don't set_buffer_write_io_error because we have
3139 * our own ways of dealing with the IO errors
3141 clear_buffer_uptodate(bh
);
3142 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3148 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3149 struct buffer_head
**bh_ret
)
3151 struct buffer_head
*bh
;
3152 struct btrfs_super_block
*super
;
3155 bytenr
= btrfs_sb_offset(copy_num
);
3156 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3159 bh
= __bread(bdev
, bytenr
/ BTRFS_BDEV_BLOCKSIZE
, BTRFS_SUPER_INFO_SIZE
);
3161 * If we fail to read from the underlying devices, as of now
3162 * the best option we have is to mark it EIO.
3167 super
= (struct btrfs_super_block
*)bh
->b_data
;
3168 if (btrfs_super_bytenr(super
) != bytenr
||
3169 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3179 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3181 struct buffer_head
*bh
;
3182 struct buffer_head
*latest
= NULL
;
3183 struct btrfs_super_block
*super
;
3188 /* we would like to check all the supers, but that would make
3189 * a btrfs mount succeed after a mkfs from a different FS.
3190 * So, we need to add a special mount option to scan for
3191 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3193 for (i
= 0; i
< 1; i
++) {
3194 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3198 super
= (struct btrfs_super_block
*)bh
->b_data
;
3200 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3203 transid
= btrfs_super_generation(super
);
3210 return ERR_PTR(ret
);
3216 * Write superblock @sb to the @device. Do not wait for completion, all the
3217 * buffer heads we write are pinned.
3219 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3220 * the expected device size at commit time. Note that max_mirrors must be
3221 * same for write and wait phases.
3223 * Return number of errors when buffer head is not found or submission fails.
3225 static int write_dev_supers(struct btrfs_device
*device
,
3226 struct btrfs_super_block
*sb
, int max_mirrors
)
3228 struct buffer_head
*bh
;
3236 if (max_mirrors
== 0)
3237 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3239 for (i
= 0; i
< max_mirrors
; i
++) {
3240 bytenr
= btrfs_sb_offset(i
);
3241 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3242 device
->commit_total_bytes
)
3245 btrfs_set_super_bytenr(sb
, bytenr
);
3248 crc
= btrfs_csum_data((const char *)sb
+ BTRFS_CSUM_SIZE
, crc
,
3249 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
3250 btrfs_csum_final(crc
, sb
->csum
);
3252 /* One reference for us, and we leave it for the caller */
3253 bh
= __getblk(device
->bdev
, bytenr
/ BTRFS_BDEV_BLOCKSIZE
,
3254 BTRFS_SUPER_INFO_SIZE
);
3256 btrfs_err(device
->fs_info
,
3257 "couldn't get super buffer head for bytenr %llu",
3263 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3265 /* one reference for submit_bh */
3268 set_buffer_uptodate(bh
);
3270 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3271 bh
->b_private
= device
;
3274 * we fua the first super. The others we allow
3277 op_flags
= REQ_SYNC
| REQ_META
| REQ_PRIO
;
3278 if (i
== 0 && !btrfs_test_opt(device
->fs_info
, NOBARRIER
))
3279 op_flags
|= REQ_FUA
;
3280 ret
= btrfsic_submit_bh(REQ_OP_WRITE
, op_flags
, bh
);
3284 return errors
< i
? 0 : -1;
3288 * Wait for write completion of superblocks done by write_dev_supers,
3289 * @max_mirrors same for write and wait phases.
3291 * Return number of errors when buffer head is not found or not marked up to
3294 static int wait_dev_supers(struct btrfs_device
*device
, int max_mirrors
)
3296 struct buffer_head
*bh
;
3301 if (max_mirrors
== 0)
3302 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3304 for (i
= 0; i
< max_mirrors
; i
++) {
3305 bytenr
= btrfs_sb_offset(i
);
3306 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3307 device
->commit_total_bytes
)
3310 bh
= __find_get_block(device
->bdev
,
3311 bytenr
/ BTRFS_BDEV_BLOCKSIZE
,
3312 BTRFS_SUPER_INFO_SIZE
);
3318 if (!buffer_uptodate(bh
))
3321 /* drop our reference */
3324 /* drop the reference from the writing run */
3328 return errors
< i
? 0 : -1;
3332 * endio for the write_dev_flush, this will wake anyone waiting
3333 * for the barrier when it is done
3335 static void btrfs_end_empty_barrier(struct bio
*bio
)
3337 complete(bio
->bi_private
);
3341 * Submit a flush request to the device if it supports it. Error handling is
3342 * done in the waiting counterpart.
3344 static void write_dev_flush(struct btrfs_device
*device
)
3346 struct request_queue
*q
= bdev_get_queue(device
->bdev
);
3347 struct bio
*bio
= device
->flush_bio
;
3349 if (!test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
))
3353 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3354 bio_set_dev(bio
, device
->bdev
);
3355 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
;
3356 init_completion(&device
->flush_wait
);
3357 bio
->bi_private
= &device
->flush_wait
;
3359 btrfsic_submit_bio(bio
);
3360 device
->flush_bio_sent
= 1;
3364 * If the flush bio has been submitted by write_dev_flush, wait for it.
3366 static blk_status_t
wait_dev_flush(struct btrfs_device
*device
)
3368 struct bio
*bio
= device
->flush_bio
;
3370 if (!device
->flush_bio_sent
)
3373 device
->flush_bio_sent
= 0;
3374 wait_for_completion_io(&device
->flush_wait
);
3376 return bio
->bi_status
;
3379 static int check_barrier_error(struct btrfs_fs_info
*fs_info
)
3381 if (!btrfs_check_rw_degradable(fs_info
))
3387 * send an empty flush down to each device in parallel,
3388 * then wait for them
3390 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3392 struct list_head
*head
;
3393 struct btrfs_device
*dev
;
3394 int errors_wait
= 0;
3397 /* send down all the barriers */
3398 head
= &info
->fs_devices
->devices
;
3399 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3404 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3407 write_dev_flush(dev
);
3408 dev
->last_flush_error
= BLK_STS_OK
;
3411 /* wait for all the barriers */
3412 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3419 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3422 ret
= wait_dev_flush(dev
);
3424 dev
->last_flush_error
= ret
;
3425 btrfs_dev_stat_inc_and_print(dev
,
3426 BTRFS_DEV_STAT_FLUSH_ERRS
);
3433 * At some point we need the status of all disks
3434 * to arrive at the volume status. So error checking
3435 * is being pushed to a separate loop.
3437 return check_barrier_error(info
);
3442 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3445 int min_tolerated
= INT_MAX
;
3447 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3448 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3449 min_tolerated
= min(min_tolerated
,
3450 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3451 tolerated_failures
);
3453 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3454 if (raid_type
== BTRFS_RAID_SINGLE
)
3456 if (!(flags
& btrfs_raid_group
[raid_type
]))
3458 min_tolerated
= min(min_tolerated
,
3459 btrfs_raid_array
[raid_type
].
3460 tolerated_failures
);
3463 if (min_tolerated
== INT_MAX
) {
3464 pr_warn("BTRFS: unknown raid flag: %llu", flags
);
3468 return min_tolerated
;
3471 int write_all_supers(struct btrfs_fs_info
*fs_info
, int max_mirrors
)
3473 struct list_head
*head
;
3474 struct btrfs_device
*dev
;
3475 struct btrfs_super_block
*sb
;
3476 struct btrfs_dev_item
*dev_item
;
3480 int total_errors
= 0;
3483 do_barriers
= !btrfs_test_opt(fs_info
, NOBARRIER
);
3486 * max_mirrors == 0 indicates we're from commit_transaction,
3487 * not from fsync where the tree roots in fs_info have not
3488 * been consistent on disk.
3490 if (max_mirrors
== 0)
3491 backup_super_roots(fs_info
);
3493 sb
= fs_info
->super_for_commit
;
3494 dev_item
= &sb
->dev_item
;
3496 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3497 head
= &fs_info
->fs_devices
->devices
;
3498 max_errors
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
3501 ret
= barrier_all_devices(fs_info
);
3504 &fs_info
->fs_devices
->device_list_mutex
);
3505 btrfs_handle_fs_error(fs_info
, ret
,
3506 "errors while submitting device barriers.");
3511 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3516 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3519 btrfs_set_stack_device_generation(dev_item
, 0);
3520 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3521 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3522 btrfs_set_stack_device_total_bytes(dev_item
,
3523 dev
->commit_total_bytes
);
3524 btrfs_set_stack_device_bytes_used(dev_item
,
3525 dev
->commit_bytes_used
);
3526 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3527 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3528 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3529 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3530 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_FSID_SIZE
);
3532 flags
= btrfs_super_flags(sb
);
3533 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3535 ret
= write_dev_supers(dev
, sb
, max_mirrors
);
3539 if (total_errors
> max_errors
) {
3540 btrfs_err(fs_info
, "%d errors while writing supers",
3542 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3544 /* FUA is masked off if unsupported and can't be the reason */
3545 btrfs_handle_fs_error(fs_info
, -EIO
,
3546 "%d errors while writing supers",
3552 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3555 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3558 ret
= wait_dev_supers(dev
, max_mirrors
);
3562 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3563 if (total_errors
> max_errors
) {
3564 btrfs_handle_fs_error(fs_info
, -EIO
,
3565 "%d errors while writing supers",
3572 /* Drop a fs root from the radix tree and free it. */
3573 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3574 struct btrfs_root
*root
)
3576 spin_lock(&fs_info
->fs_roots_radix_lock
);
3577 radix_tree_delete(&fs_info
->fs_roots_radix
,
3578 (unsigned long)root
->root_key
.objectid
);
3579 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3581 if (btrfs_root_refs(&root
->root_item
) == 0)
3582 synchronize_srcu(&fs_info
->subvol_srcu
);
3584 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3585 btrfs_free_log(NULL
, root
);
3586 if (root
->reloc_root
) {
3587 free_extent_buffer(root
->reloc_root
->node
);
3588 free_extent_buffer(root
->reloc_root
->commit_root
);
3589 btrfs_put_fs_root(root
->reloc_root
);
3590 root
->reloc_root
= NULL
;
3594 if (root
->free_ino_pinned
)
3595 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3596 if (root
->free_ino_ctl
)
3597 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3601 static void free_fs_root(struct btrfs_root
*root
)
3603 iput(root
->ino_cache_inode
);
3604 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3605 btrfs_free_block_rsv(root
->fs_info
, root
->orphan_block_rsv
);
3606 root
->orphan_block_rsv
= NULL
;
3608 free_anon_bdev(root
->anon_dev
);
3609 if (root
->subv_writers
)
3610 btrfs_free_subvolume_writers(root
->subv_writers
);
3611 free_extent_buffer(root
->node
);
3612 free_extent_buffer(root
->commit_root
);
3613 kfree(root
->free_ino_ctl
);
3614 kfree(root
->free_ino_pinned
);
3616 btrfs_put_fs_root(root
);
3619 void btrfs_free_fs_root(struct btrfs_root
*root
)
3624 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3626 u64 root_objectid
= 0;
3627 struct btrfs_root
*gang
[8];
3630 unsigned int ret
= 0;
3634 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3635 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3636 (void **)gang
, root_objectid
,
3639 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3642 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3644 for (i
= 0; i
< ret
; i
++) {
3645 /* Avoid to grab roots in dead_roots */
3646 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3650 /* grab all the search result for later use */
3651 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3653 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3655 for (i
= 0; i
< ret
; i
++) {
3658 root_objectid
= gang
[i
]->root_key
.objectid
;
3659 err
= btrfs_orphan_cleanup(gang
[i
]);
3662 btrfs_put_fs_root(gang
[i
]);
3667 /* release the uncleaned roots due to error */
3668 for (; i
< ret
; i
++) {
3670 btrfs_put_fs_root(gang
[i
]);
3675 int btrfs_commit_super(struct btrfs_fs_info
*fs_info
)
3677 struct btrfs_root
*root
= fs_info
->tree_root
;
3678 struct btrfs_trans_handle
*trans
;
3680 mutex_lock(&fs_info
->cleaner_mutex
);
3681 btrfs_run_delayed_iputs(fs_info
);
3682 mutex_unlock(&fs_info
->cleaner_mutex
);
3683 wake_up_process(fs_info
->cleaner_kthread
);
3685 /* wait until ongoing cleanup work done */
3686 down_write(&fs_info
->cleanup_work_sem
);
3687 up_write(&fs_info
->cleanup_work_sem
);
3689 trans
= btrfs_join_transaction(root
);
3691 return PTR_ERR(trans
);
3692 return btrfs_commit_transaction(trans
);
3695 void close_ctree(struct btrfs_fs_info
*fs_info
)
3697 struct btrfs_root
*root
= fs_info
->tree_root
;
3700 set_bit(BTRFS_FS_CLOSING_START
, &fs_info
->flags
);
3702 /* wait for the qgroup rescan worker to stop */
3703 btrfs_qgroup_wait_for_completion(fs_info
, false);
3705 /* wait for the uuid_scan task to finish */
3706 down(&fs_info
->uuid_tree_rescan_sem
);
3707 /* avoid complains from lockdep et al., set sem back to initial state */
3708 up(&fs_info
->uuid_tree_rescan_sem
);
3710 /* pause restriper - we want to resume on mount */
3711 btrfs_pause_balance(fs_info
);
3713 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3715 btrfs_scrub_cancel(fs_info
);
3717 /* wait for any defraggers to finish */
3718 wait_event(fs_info
->transaction_wait
,
3719 (atomic_read(&fs_info
->defrag_running
) == 0));
3721 /* clear out the rbtree of defraggable inodes */
3722 btrfs_cleanup_defrag_inodes(fs_info
);
3724 cancel_work_sync(&fs_info
->async_reclaim_work
);
3726 if (!sb_rdonly(fs_info
->sb
)) {
3728 * If the cleaner thread is stopped and there are
3729 * block groups queued for removal, the deletion will be
3730 * skipped when we quit the cleaner thread.
3732 btrfs_delete_unused_bgs(fs_info
);
3734 ret
= btrfs_commit_super(fs_info
);
3736 btrfs_err(fs_info
, "commit super ret %d", ret
);
3739 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3740 btrfs_error_commit_super(fs_info
);
3742 kthread_stop(fs_info
->transaction_kthread
);
3743 kthread_stop(fs_info
->cleaner_kthread
);
3745 set_bit(BTRFS_FS_CLOSING_DONE
, &fs_info
->flags
);
3747 btrfs_free_qgroup_config(fs_info
);
3749 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3750 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3751 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3754 btrfs_sysfs_remove_mounted(fs_info
);
3755 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3757 btrfs_free_fs_roots(fs_info
);
3759 btrfs_put_block_group_cache(fs_info
);
3762 * we must make sure there is not any read request to
3763 * submit after we stopping all workers.
3765 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3766 btrfs_stop_all_workers(fs_info
);
3768 btrfs_free_block_groups(fs_info
);
3770 clear_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3771 free_root_pointers(fs_info
, 1);
3773 iput(fs_info
->btree_inode
);
3775 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3776 if (btrfs_test_opt(fs_info
, CHECK_INTEGRITY
))
3777 btrfsic_unmount(fs_info
->fs_devices
);
3780 btrfs_close_devices(fs_info
->fs_devices
);
3781 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3783 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3784 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3785 percpu_counter_destroy(&fs_info
->bio_counter
);
3786 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3788 btrfs_free_stripe_hash_table(fs_info
);
3789 btrfs_free_ref_cache(fs_info
);
3791 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3792 root
->orphan_block_rsv
= NULL
;
3794 while (!list_empty(&fs_info
->pinned_chunks
)) {
3795 struct extent_map
*em
;
3797 em
= list_first_entry(&fs_info
->pinned_chunks
,
3798 struct extent_map
, list
);
3799 list_del_init(&em
->list
);
3800 free_extent_map(em
);
3804 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3808 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3810 ret
= extent_buffer_uptodate(buf
);
3814 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3815 parent_transid
, atomic
);
3821 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3823 struct btrfs_fs_info
*fs_info
;
3824 struct btrfs_root
*root
;
3825 u64 transid
= btrfs_header_generation(buf
);
3828 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3830 * This is a fast path so only do this check if we have sanity tests
3831 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3832 * outside of the sanity tests.
3834 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3837 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3838 fs_info
= root
->fs_info
;
3839 btrfs_assert_tree_locked(buf
);
3840 if (transid
!= fs_info
->generation
)
3841 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3842 buf
->start
, transid
, fs_info
->generation
);
3843 was_dirty
= set_extent_buffer_dirty(buf
);
3845 percpu_counter_add_batch(&fs_info
->dirty_metadata_bytes
,
3847 fs_info
->dirty_metadata_batch
);
3848 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3850 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3851 * but item data not updated.
3852 * So here we should only check item pointers, not item data.
3854 if (btrfs_header_level(buf
) == 0 &&
3855 btrfs_check_leaf_relaxed(root
, buf
)) {
3856 btrfs_print_leaf(buf
);
3862 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
,
3866 * looks as though older kernels can get into trouble with
3867 * this code, they end up stuck in balance_dirty_pages forever
3871 if (current
->flags
& PF_MEMALLOC
)
3875 btrfs_balance_delayed_items(fs_info
);
3877 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
3878 BTRFS_DIRTY_METADATA_THRESH
);
3880 balance_dirty_pages_ratelimited(fs_info
->btree_inode
->i_mapping
);
3884 void btrfs_btree_balance_dirty(struct btrfs_fs_info
*fs_info
)
3886 __btrfs_btree_balance_dirty(fs_info
, 1);
3889 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info
*fs_info
)
3891 __btrfs_btree_balance_dirty(fs_info
, 0);
3894 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3896 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3897 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3899 return btree_read_extent_buffer_pages(fs_info
, buf
, parent_transid
);
3902 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
)
3904 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
3905 u64 nodesize
= btrfs_super_nodesize(sb
);
3906 u64 sectorsize
= btrfs_super_sectorsize(sb
);
3909 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
3910 btrfs_err(fs_info
, "no valid FS found");
3913 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
)
3914 btrfs_warn(fs_info
, "unrecognized super flag: %llu",
3915 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
3916 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3917 btrfs_err(fs_info
, "tree_root level too big: %d >= %d",
3918 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
3921 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3922 btrfs_err(fs_info
, "chunk_root level too big: %d >= %d",
3923 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
3926 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3927 btrfs_err(fs_info
, "log_root level too big: %d >= %d",
3928 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
3933 * Check sectorsize and nodesize first, other check will need it.
3934 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3936 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
3937 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
3938 btrfs_err(fs_info
, "invalid sectorsize %llu", sectorsize
);
3941 /* Only PAGE SIZE is supported yet */
3942 if (sectorsize
!= PAGE_SIZE
) {
3944 "sectorsize %llu not supported yet, only support %lu",
3945 sectorsize
, PAGE_SIZE
);
3948 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
3949 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
3950 btrfs_err(fs_info
, "invalid nodesize %llu", nodesize
);
3953 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
3954 btrfs_err(fs_info
, "invalid leafsize %u, should be %llu",
3955 le32_to_cpu(sb
->__unused_leafsize
), nodesize
);
3959 /* Root alignment check */
3960 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
3961 btrfs_warn(fs_info
, "tree_root block unaligned: %llu",
3962 btrfs_super_root(sb
));
3965 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
3966 btrfs_warn(fs_info
, "chunk_root block unaligned: %llu",
3967 btrfs_super_chunk_root(sb
));
3970 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
3971 btrfs_warn(fs_info
, "log_root block unaligned: %llu",
3972 btrfs_super_log_root(sb
));
3976 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_FSID_SIZE
) != 0) {
3978 "dev_item UUID does not match fsid: %pU != %pU",
3979 fs_info
->fsid
, sb
->dev_item
.fsid
);
3984 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3987 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
3988 btrfs_err(fs_info
, "bytes_used is too small %llu",
3989 btrfs_super_bytes_used(sb
));
3992 if (!is_power_of_2(btrfs_super_stripesize(sb
))) {
3993 btrfs_err(fs_info
, "invalid stripesize %u",
3994 btrfs_super_stripesize(sb
));
3997 if (btrfs_super_num_devices(sb
) > (1UL << 31))
3998 btrfs_warn(fs_info
, "suspicious number of devices: %llu",
3999 btrfs_super_num_devices(sb
));
4000 if (btrfs_super_num_devices(sb
) == 0) {
4001 btrfs_err(fs_info
, "number of devices is 0");
4005 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4006 btrfs_err(fs_info
, "super offset mismatch %llu != %u",
4007 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4012 * Obvious sys_chunk_array corruptions, it must hold at least one key
4015 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4016 btrfs_err(fs_info
, "system chunk array too big %u > %u",
4017 btrfs_super_sys_array_size(sb
),
4018 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4021 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4022 + sizeof(struct btrfs_chunk
)) {
4023 btrfs_err(fs_info
, "system chunk array too small %u < %zu",
4024 btrfs_super_sys_array_size(sb
),
4025 sizeof(struct btrfs_disk_key
)
4026 + sizeof(struct btrfs_chunk
));
4031 * The generation is a global counter, we'll trust it more than the others
4032 * but it's still possible that it's the one that's wrong.
4034 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4036 "suspicious: generation < chunk_root_generation: %llu < %llu",
4037 btrfs_super_generation(sb
),
4038 btrfs_super_chunk_root_generation(sb
));
4039 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4040 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4042 "suspicious: generation < cache_generation: %llu < %llu",
4043 btrfs_super_generation(sb
),
4044 btrfs_super_cache_generation(sb
));
4049 static void btrfs_error_commit_super(struct btrfs_fs_info
*fs_info
)
4051 mutex_lock(&fs_info
->cleaner_mutex
);
4052 btrfs_run_delayed_iputs(fs_info
);
4053 mutex_unlock(&fs_info
->cleaner_mutex
);
4055 down_write(&fs_info
->cleanup_work_sem
);
4056 up_write(&fs_info
->cleanup_work_sem
);
4058 /* cleanup FS via transaction */
4059 btrfs_cleanup_transaction(fs_info
);
4062 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4064 struct btrfs_ordered_extent
*ordered
;
4066 spin_lock(&root
->ordered_extent_lock
);
4068 * This will just short circuit the ordered completion stuff which will
4069 * make sure the ordered extent gets properly cleaned up.
4071 list_for_each_entry(ordered
, &root
->ordered_extents
,
4073 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4074 spin_unlock(&root
->ordered_extent_lock
);
4077 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4079 struct btrfs_root
*root
;
4080 struct list_head splice
;
4082 INIT_LIST_HEAD(&splice
);
4084 spin_lock(&fs_info
->ordered_root_lock
);
4085 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4086 while (!list_empty(&splice
)) {
4087 root
= list_first_entry(&splice
, struct btrfs_root
,
4089 list_move_tail(&root
->ordered_root
,
4090 &fs_info
->ordered_roots
);
4092 spin_unlock(&fs_info
->ordered_root_lock
);
4093 btrfs_destroy_ordered_extents(root
);
4096 spin_lock(&fs_info
->ordered_root_lock
);
4098 spin_unlock(&fs_info
->ordered_root_lock
);
4101 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4102 struct btrfs_fs_info
*fs_info
)
4104 struct rb_node
*node
;
4105 struct btrfs_delayed_ref_root
*delayed_refs
;
4106 struct btrfs_delayed_ref_node
*ref
;
4109 delayed_refs
= &trans
->delayed_refs
;
4111 spin_lock(&delayed_refs
->lock
);
4112 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4113 spin_unlock(&delayed_refs
->lock
);
4114 btrfs_info(fs_info
, "delayed_refs has NO entry");
4118 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4119 struct btrfs_delayed_ref_head
*head
;
4121 bool pin_bytes
= false;
4123 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4125 if (!mutex_trylock(&head
->mutex
)) {
4126 refcount_inc(&head
->refs
);
4127 spin_unlock(&delayed_refs
->lock
);
4129 mutex_lock(&head
->mutex
);
4130 mutex_unlock(&head
->mutex
);
4131 btrfs_put_delayed_ref_head(head
);
4132 spin_lock(&delayed_refs
->lock
);
4135 spin_lock(&head
->lock
);
4136 while ((n
= rb_first(&head
->ref_tree
)) != NULL
) {
4137 ref
= rb_entry(n
, struct btrfs_delayed_ref_node
,
4140 rb_erase(&ref
->ref_node
, &head
->ref_tree
);
4141 RB_CLEAR_NODE(&ref
->ref_node
);
4142 if (!list_empty(&ref
->add_list
))
4143 list_del(&ref
->add_list
);
4144 atomic_dec(&delayed_refs
->num_entries
);
4145 btrfs_put_delayed_ref(ref
);
4147 if (head
->must_insert_reserved
)
4149 btrfs_free_delayed_extent_op(head
->extent_op
);
4150 delayed_refs
->num_heads
--;
4151 if (head
->processing
== 0)
4152 delayed_refs
->num_heads_ready
--;
4153 atomic_dec(&delayed_refs
->num_entries
);
4154 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4155 RB_CLEAR_NODE(&head
->href_node
);
4156 spin_unlock(&head
->lock
);
4157 spin_unlock(&delayed_refs
->lock
);
4158 mutex_unlock(&head
->mutex
);
4161 btrfs_pin_extent(fs_info
, head
->bytenr
,
4162 head
->num_bytes
, 1);
4163 btrfs_put_delayed_ref_head(head
);
4165 spin_lock(&delayed_refs
->lock
);
4168 spin_unlock(&delayed_refs
->lock
);
4173 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4175 struct btrfs_inode
*btrfs_inode
;
4176 struct list_head splice
;
4178 INIT_LIST_HEAD(&splice
);
4180 spin_lock(&root
->delalloc_lock
);
4181 list_splice_init(&root
->delalloc_inodes
, &splice
);
4183 while (!list_empty(&splice
)) {
4184 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4187 list_del_init(&btrfs_inode
->delalloc_inodes
);
4188 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4189 &btrfs_inode
->runtime_flags
);
4190 spin_unlock(&root
->delalloc_lock
);
4192 btrfs_invalidate_inodes(btrfs_inode
->root
);
4194 spin_lock(&root
->delalloc_lock
);
4197 spin_unlock(&root
->delalloc_lock
);
4200 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4202 struct btrfs_root
*root
;
4203 struct list_head splice
;
4205 INIT_LIST_HEAD(&splice
);
4207 spin_lock(&fs_info
->delalloc_root_lock
);
4208 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4209 while (!list_empty(&splice
)) {
4210 root
= list_first_entry(&splice
, struct btrfs_root
,
4212 list_del_init(&root
->delalloc_root
);
4213 root
= btrfs_grab_fs_root(root
);
4215 spin_unlock(&fs_info
->delalloc_root_lock
);
4217 btrfs_destroy_delalloc_inodes(root
);
4218 btrfs_put_fs_root(root
);
4220 spin_lock(&fs_info
->delalloc_root_lock
);
4222 spin_unlock(&fs_info
->delalloc_root_lock
);
4225 static int btrfs_destroy_marked_extents(struct btrfs_fs_info
*fs_info
,
4226 struct extent_io_tree
*dirty_pages
,
4230 struct extent_buffer
*eb
;
4235 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4240 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4241 while (start
<= end
) {
4242 eb
= find_extent_buffer(fs_info
, start
);
4243 start
+= fs_info
->nodesize
;
4246 wait_on_extent_buffer_writeback(eb
);
4248 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4250 clear_extent_buffer_dirty(eb
);
4251 free_extent_buffer_stale(eb
);
4258 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info
*fs_info
,
4259 struct extent_io_tree
*pinned_extents
)
4261 struct extent_io_tree
*unpin
;
4267 unpin
= pinned_extents
;
4270 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4271 EXTENT_DIRTY
, NULL
);
4275 clear_extent_dirty(unpin
, start
, end
);
4276 btrfs_error_unpin_extent_range(fs_info
, start
, end
);
4281 if (unpin
== &fs_info
->freed_extents
[0])
4282 unpin
= &fs_info
->freed_extents
[1];
4284 unpin
= &fs_info
->freed_extents
[0];
4292 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache
*cache
)
4294 struct inode
*inode
;
4296 inode
= cache
->io_ctl
.inode
;
4298 invalidate_inode_pages2(inode
->i_mapping
);
4299 BTRFS_I(inode
)->generation
= 0;
4300 cache
->io_ctl
.inode
= NULL
;
4303 btrfs_put_block_group(cache
);
4306 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction
*cur_trans
,
4307 struct btrfs_fs_info
*fs_info
)
4309 struct btrfs_block_group_cache
*cache
;
4311 spin_lock(&cur_trans
->dirty_bgs_lock
);
4312 while (!list_empty(&cur_trans
->dirty_bgs
)) {
4313 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
4314 struct btrfs_block_group_cache
,
4317 btrfs_err(fs_info
, "orphan block group dirty_bgs list");
4318 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4322 if (!list_empty(&cache
->io_list
)) {
4323 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4324 list_del_init(&cache
->io_list
);
4325 btrfs_cleanup_bg_io(cache
);
4326 spin_lock(&cur_trans
->dirty_bgs_lock
);
4329 list_del_init(&cache
->dirty_list
);
4330 spin_lock(&cache
->lock
);
4331 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4332 spin_unlock(&cache
->lock
);
4334 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4335 btrfs_put_block_group(cache
);
4336 spin_lock(&cur_trans
->dirty_bgs_lock
);
4338 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4340 while (!list_empty(&cur_trans
->io_bgs
)) {
4341 cache
= list_first_entry(&cur_trans
->io_bgs
,
4342 struct btrfs_block_group_cache
,
4345 btrfs_err(fs_info
, "orphan block group on io_bgs list");
4349 list_del_init(&cache
->io_list
);
4350 spin_lock(&cache
->lock
);
4351 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4352 spin_unlock(&cache
->lock
);
4353 btrfs_cleanup_bg_io(cache
);
4357 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4358 struct btrfs_fs_info
*fs_info
)
4360 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
4361 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
4362 ASSERT(list_empty(&cur_trans
->io_bgs
));
4364 btrfs_destroy_delayed_refs(cur_trans
, fs_info
);
4366 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4367 wake_up(&fs_info
->transaction_blocked_wait
);
4369 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4370 wake_up(&fs_info
->transaction_wait
);
4372 btrfs_destroy_delayed_inodes(fs_info
);
4373 btrfs_assert_delayed_root_empty(fs_info
);
4375 btrfs_destroy_marked_extents(fs_info
, &cur_trans
->dirty_pages
,
4377 btrfs_destroy_pinned_extent(fs_info
,
4378 fs_info
->pinned_extents
);
4380 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4381 wake_up(&cur_trans
->commit_wait
);
4384 static int btrfs_cleanup_transaction(struct btrfs_fs_info
*fs_info
)
4386 struct btrfs_transaction
*t
;
4388 mutex_lock(&fs_info
->transaction_kthread_mutex
);
4390 spin_lock(&fs_info
->trans_lock
);
4391 while (!list_empty(&fs_info
->trans_list
)) {
4392 t
= list_first_entry(&fs_info
->trans_list
,
4393 struct btrfs_transaction
, list
);
4394 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4395 refcount_inc(&t
->use_count
);
4396 spin_unlock(&fs_info
->trans_lock
);
4397 btrfs_wait_for_commit(fs_info
, t
->transid
);
4398 btrfs_put_transaction(t
);
4399 spin_lock(&fs_info
->trans_lock
);
4402 if (t
== fs_info
->running_transaction
) {
4403 t
->state
= TRANS_STATE_COMMIT_DOING
;
4404 spin_unlock(&fs_info
->trans_lock
);
4406 * We wait for 0 num_writers since we don't hold a trans
4407 * handle open currently for this transaction.
4409 wait_event(t
->writer_wait
,
4410 atomic_read(&t
->num_writers
) == 0);
4412 spin_unlock(&fs_info
->trans_lock
);
4414 btrfs_cleanup_one_transaction(t
, fs_info
);
4416 spin_lock(&fs_info
->trans_lock
);
4417 if (t
== fs_info
->running_transaction
)
4418 fs_info
->running_transaction
= NULL
;
4419 list_del_init(&t
->list
);
4420 spin_unlock(&fs_info
->trans_lock
);
4422 btrfs_put_transaction(t
);
4423 trace_btrfs_transaction_commit(fs_info
->tree_root
);
4424 spin_lock(&fs_info
->trans_lock
);
4426 spin_unlock(&fs_info
->trans_lock
);
4427 btrfs_destroy_all_ordered_extents(fs_info
);
4428 btrfs_destroy_delayed_inodes(fs_info
);
4429 btrfs_assert_delayed_root_empty(fs_info
);
4430 btrfs_destroy_pinned_extent(fs_info
, fs_info
->pinned_extents
);
4431 btrfs_destroy_all_delalloc_inodes(fs_info
);
4432 mutex_unlock(&fs_info
->transaction_kthread_mutex
);
4437 static struct btrfs_fs_info
*btree_fs_info(void *private_data
)
4439 struct inode
*inode
= private_data
;
4440 return btrfs_sb(inode
->i_sb
);
4443 static const struct extent_io_ops btree_extent_io_ops
= {
4444 /* mandatory callbacks */
4445 .submit_bio_hook
= btree_submit_bio_hook
,
4446 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4447 /* note we're sharing with inode.c for the merge bio hook */
4448 .merge_bio_hook
= btrfs_merge_bio_hook
,
4449 .readpage_io_failed_hook
= btree_io_failed_hook
,
4450 .set_range_writeback
= btrfs_set_range_writeback
,
4451 .tree_fs_info
= btree_fs_info
,
4453 /* optional callbacks */