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/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "free-space-tree.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
55 #include <asm/cpufeature.h>
58 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
64 static const struct extent_io_ops btree_extent_io_ops
;
65 static void end_workqueue_fn(struct btrfs_work
*work
);
66 static void free_fs_root(struct btrfs_root
*root
);
67 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
69 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
71 struct btrfs_root
*root
);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
73 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
74 struct extent_io_tree
*dirty_pages
,
76 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
77 struct extent_io_tree
*pinned_extents
);
78 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
79 static void btrfs_error_commit_super(struct btrfs_root
*root
);
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_end_io_wq
{
90 struct btrfs_fs_info
*info
;
92 enum btrfs_wq_endio_type metadata
;
93 struct list_head list
;
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
),
104 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
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
{
124 struct list_head list
;
125 extent_submit_bio_hook_t
*submit_bio_start
;
126 extent_submit_bio_hook_t
*submit_bio_done
;
129 unsigned long bio_flags
;
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
135 struct btrfs_work work
;
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
162 #ifdef CONFIG_DEBUG_LOCK_ALLOC
163 # if BTRFS_MAX_LEVEL != 8
167 static struct btrfs_lockdep_keyset
{
168 u64 id
; /* root objectid */
169 const char *name_stem
; /* lock name stem */
170 char names
[BTRFS_MAX_LEVEL
+ 1][20];
171 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
172 } btrfs_lockdep_keysets
[] = {
173 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
174 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
175 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
176 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
177 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
178 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
179 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
180 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
181 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
182 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
183 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
184 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
185 { .id
= 0, .name_stem
= "tree" },
188 void __init
btrfs_init_lockdep(void)
192 /* initialize lockdep class names */
193 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
194 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
196 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
197 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
198 "btrfs-%s-%02d", ks
->name_stem
, j
);
202 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
205 struct btrfs_lockdep_keyset
*ks
;
207 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
211 if (ks
->id
== objectid
)
214 lockdep_set_class_and_name(&eb
->lock
,
215 &ks
->keys
[level
], ks
->names
[level
]);
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
224 static struct extent_map
*btree_get_extent(struct inode
*inode
,
225 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
228 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
229 struct extent_map
*em
;
232 read_lock(&em_tree
->lock
);
233 em
= lookup_extent_mapping(em_tree
, start
, len
);
236 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
237 read_unlock(&em_tree
->lock
);
240 read_unlock(&em_tree
->lock
);
242 em
= alloc_extent_map();
244 em
= ERR_PTR(-ENOMEM
);
249 em
->block_len
= (u64
)-1;
251 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
253 write_lock(&em_tree
->lock
);
254 ret
= add_extent_mapping(em_tree
, em
, 0);
255 if (ret
== -EEXIST
) {
257 em
= lookup_extent_mapping(em_tree
, start
, len
);
264 write_unlock(&em_tree
->lock
);
270 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
272 return btrfs_crc32c(seed
, data
, len
);
275 void btrfs_csum_final(u32 crc
, char *result
)
277 put_unaligned_le32(~crc
, result
);
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
284 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
285 struct extent_buffer
*buf
,
288 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
291 unsigned long cur_len
;
292 unsigned long offset
= BTRFS_CSUM_SIZE
;
294 unsigned long map_start
;
295 unsigned long map_len
;
298 unsigned long inline_result
;
300 len
= buf
->len
- offset
;
302 err
= map_private_extent_buffer(buf
, offset
, 32,
303 &kaddr
, &map_start
, &map_len
);
306 cur_len
= min(len
, map_len
- (offset
- map_start
));
307 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
312 if (csum_size
> sizeof(inline_result
)) {
313 result
= kzalloc(csum_size
, GFP_NOFS
);
317 result
= (char *)&inline_result
;
320 btrfs_csum_final(crc
, result
);
323 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
326 memcpy(&found
, result
, csum_size
);
328 read_extent_buffer(buf
, &val
, 0, csum_size
);
329 btrfs_warn_rl(fs_info
,
330 "%s checksum verify failed on %llu wanted %X found %X "
332 fs_info
->sb
->s_id
, buf
->start
,
333 val
, found
, btrfs_header_level(buf
));
334 if (result
!= (char *)&inline_result
)
339 write_extent_buffer(buf
, result
, 0, csum_size
);
341 if (result
!= (char *)&inline_result
)
347 * we can't consider a given block up to date unless the transid of the
348 * block matches the transid in the parent node's pointer. This is how we
349 * detect blocks that either didn't get written at all or got written
350 * in the wrong place.
352 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
353 struct extent_buffer
*eb
, u64 parent_transid
,
356 struct extent_state
*cached_state
= NULL
;
358 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
360 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
367 btrfs_tree_read_lock(eb
);
368 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
371 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
373 if (extent_buffer_uptodate(eb
) &&
374 btrfs_header_generation(eb
) == parent_transid
) {
378 btrfs_err_rl(eb
->fs_info
,
379 "parent transid verify failed on %llu wanted %llu found %llu",
381 parent_transid
, btrfs_header_generation(eb
));
385 * Things reading via commit roots that don't have normal protection,
386 * like send, can have a really old block in cache that may point at a
387 * block that has been free'd and re-allocated. So don't clear uptodate
388 * if we find an eb that is under IO (dirty/writeback) because we could
389 * end up reading in the stale data and then writing it back out and
390 * making everybody very sad.
392 if (!extent_buffer_under_io(eb
))
393 clear_extent_buffer_uptodate(eb
);
395 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
396 &cached_state
, GFP_NOFS
);
398 btrfs_tree_read_unlock_blocking(eb
);
403 * Return 0 if the superblock checksum type matches the checksum value of that
404 * algorithm. Pass the raw disk superblock data.
406 static int btrfs_check_super_csum(char *raw_disk_sb
)
408 struct btrfs_super_block
*disk_sb
=
409 (struct btrfs_super_block
*)raw_disk_sb
;
410 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
413 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
415 const int csum_size
= sizeof(crc
);
416 char result
[csum_size
];
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checkum.
423 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
424 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
425 btrfs_csum_final(crc
, result
);
427 if (memcmp(raw_disk_sb
, result
, csum_size
))
431 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
432 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
444 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
445 struct extent_buffer
*eb
,
446 u64 start
, u64 parent_transid
)
448 struct extent_io_tree
*io_tree
;
453 int failed_mirror
= 0;
455 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
456 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
458 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
460 btree_get_extent
, mirror_num
);
462 if (!verify_parent_transid(io_tree
, eb
,
470 * This buffer's crc is fine, but its contents are corrupted, so
471 * there is no reason to read the other copies, they won't be
474 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
477 num_copies
= btrfs_num_copies(root
->fs_info
,
482 if (!failed_mirror
) {
484 failed_mirror
= eb
->read_mirror
;
488 if (mirror_num
== failed_mirror
)
491 if (mirror_num
> num_copies
)
495 if (failed
&& !ret
&& failed_mirror
)
496 repair_eb_io_failure(root
, eb
, failed_mirror
);
502 * checksum a dirty tree block before IO. This has extra checks to make sure
503 * we only fill in the checksum field in the first page of a multi-page block
506 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
508 u64 start
= page_offset(page
);
510 struct extent_buffer
*eb
;
512 eb
= (struct extent_buffer
*)page
->private;
513 if (page
!= eb
->pages
[0])
516 found_start
= btrfs_header_bytenr(eb
);
518 * Please do not consolidate these warnings into a single if.
519 * It is useful to know what went wrong.
521 if (WARN_ON(found_start
!= start
))
523 if (WARN_ON(!PageUptodate(page
)))
526 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fsid
,
527 btrfs_header_fsid(), BTRFS_FSID_SIZE
) == 0);
529 return csum_tree_block(fs_info
, eb
, 0);
532 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
533 struct extent_buffer
*eb
)
535 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
536 u8 fsid
[BTRFS_UUID_SIZE
];
539 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
541 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
545 fs_devices
= fs_devices
->seed
;
550 #define CORRUPT(reason, eb, root, slot) \
551 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
552 "root=%llu, slot=%d", reason, \
553 btrfs_header_bytenr(eb), root->objectid, slot)
555 static noinline
int check_leaf(struct btrfs_root
*root
,
556 struct extent_buffer
*leaf
)
558 struct btrfs_key key
;
559 struct btrfs_key leaf_key
;
560 u32 nritems
= btrfs_header_nritems(leaf
);
566 /* Check the 0 item */
567 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
568 BTRFS_LEAF_DATA_SIZE(root
)) {
569 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
574 * Check to make sure each items keys are in the correct order and their
575 * offsets make sense. We only have to loop through nritems-1 because
576 * we check the current slot against the next slot, which verifies the
577 * next slot's offset+size makes sense and that the current's slot
580 for (slot
= 0; slot
< nritems
- 1; slot
++) {
581 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
582 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
584 /* Make sure the keys are in the right order */
585 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
586 CORRUPT("bad key order", leaf
, root
, slot
);
591 * Make sure the offset and ends are right, remember that the
592 * item data starts at the end of the leaf and grows towards the
595 if (btrfs_item_offset_nr(leaf
, slot
) !=
596 btrfs_item_end_nr(leaf
, slot
+ 1)) {
597 CORRUPT("slot offset bad", leaf
, root
, slot
);
602 * Check to make sure that we don't point outside of the leaf,
603 * just incase all the items are consistent to eachother, but
604 * all point outside of the leaf.
606 if (btrfs_item_end_nr(leaf
, slot
) >
607 BTRFS_LEAF_DATA_SIZE(root
)) {
608 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
616 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
617 u64 phy_offset
, struct page
*page
,
618 u64 start
, u64 end
, int mirror
)
622 struct extent_buffer
*eb
;
623 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
624 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
631 eb
= (struct extent_buffer
*)page
->private;
633 /* the pending IO might have been the only thing that kept this buffer
634 * in memory. Make sure we have a ref for all this other checks
636 extent_buffer_get(eb
);
638 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
642 eb
->read_mirror
= mirror
;
643 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
648 found_start
= btrfs_header_bytenr(eb
);
649 if (found_start
!= eb
->start
) {
650 btrfs_err_rl(fs_info
, "bad tree block start %llu %llu",
651 found_start
, eb
->start
);
655 if (check_tree_block_fsid(fs_info
, eb
)) {
656 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
661 found_level
= btrfs_header_level(eb
);
662 if (found_level
>= BTRFS_MAX_LEVEL
) {
663 btrfs_err(fs_info
, "bad tree block level %d",
664 (int)btrfs_header_level(eb
));
669 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
672 ret
= csum_tree_block(fs_info
, eb
, 1);
677 * If this is a leaf block and it is corrupt, set the corrupt bit so
678 * that we don't try and read the other copies of this block, just
681 if (found_level
== 0 && check_leaf(root
, eb
)) {
682 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
687 set_extent_buffer_uptodate(eb
);
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
691 btree_readahead_hook(fs_info
, eb
, eb
->start
, ret
);
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
699 atomic_inc(&eb
->io_pages
);
700 clear_extent_buffer_uptodate(eb
);
702 free_extent_buffer(eb
);
707 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
709 struct extent_buffer
*eb
;
711 eb
= (struct extent_buffer
*)page
->private;
712 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
713 eb
->read_mirror
= failed_mirror
;
714 atomic_dec(&eb
->io_pages
);
715 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
716 btree_readahead_hook(eb
->fs_info
, eb
, eb
->start
, -EIO
);
717 return -EIO
; /* we fixed nothing */
720 static void end_workqueue_bio(struct bio
*bio
)
722 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
723 struct btrfs_fs_info
*fs_info
;
724 struct btrfs_workqueue
*wq
;
725 btrfs_work_func_t func
;
727 fs_info
= end_io_wq
->info
;
728 end_io_wq
->error
= bio
->bi_error
;
730 if (bio
->bi_rw
& REQ_WRITE
) {
731 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
732 wq
= fs_info
->endio_meta_write_workers
;
733 func
= btrfs_endio_meta_write_helper
;
734 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
735 wq
= fs_info
->endio_freespace_worker
;
736 func
= btrfs_freespace_write_helper
;
737 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
738 wq
= fs_info
->endio_raid56_workers
;
739 func
= btrfs_endio_raid56_helper
;
741 wq
= fs_info
->endio_write_workers
;
742 func
= btrfs_endio_write_helper
;
745 if (unlikely(end_io_wq
->metadata
==
746 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
747 wq
= fs_info
->endio_repair_workers
;
748 func
= btrfs_endio_repair_helper
;
749 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
750 wq
= fs_info
->endio_raid56_workers
;
751 func
= btrfs_endio_raid56_helper
;
752 } else if (end_io_wq
->metadata
) {
753 wq
= fs_info
->endio_meta_workers
;
754 func
= btrfs_endio_meta_helper
;
756 wq
= fs_info
->endio_workers
;
757 func
= btrfs_endio_helper
;
761 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
762 btrfs_queue_work(wq
, &end_io_wq
->work
);
765 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
766 enum btrfs_wq_endio_type metadata
)
768 struct btrfs_end_io_wq
*end_io_wq
;
770 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
774 end_io_wq
->private = bio
->bi_private
;
775 end_io_wq
->end_io
= bio
->bi_end_io
;
776 end_io_wq
->info
= info
;
777 end_io_wq
->error
= 0;
778 end_io_wq
->bio
= bio
;
779 end_io_wq
->metadata
= metadata
;
781 bio
->bi_private
= end_io_wq
;
782 bio
->bi_end_io
= end_workqueue_bio
;
786 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
788 unsigned long limit
= min_t(unsigned long,
789 info
->thread_pool_size
,
790 info
->fs_devices
->open_devices
);
794 static void run_one_async_start(struct btrfs_work
*work
)
796 struct async_submit_bio
*async
;
799 async
= container_of(work
, struct async_submit_bio
, work
);
800 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
801 async
->mirror_num
, async
->bio_flags
,
807 static void run_one_async_done(struct btrfs_work
*work
)
809 struct btrfs_fs_info
*fs_info
;
810 struct async_submit_bio
*async
;
813 async
= container_of(work
, struct async_submit_bio
, work
);
814 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
816 limit
= btrfs_async_submit_limit(fs_info
);
817 limit
= limit
* 2 / 3;
820 * atomic_dec_return implies a barrier for waitqueue_active
822 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
823 waitqueue_active(&fs_info
->async_submit_wait
))
824 wake_up(&fs_info
->async_submit_wait
);
826 /* If an error occured we just want to clean up the bio and move on */
828 async
->bio
->bi_error
= async
->error
;
829 bio_endio(async
->bio
);
833 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
834 async
->mirror_num
, async
->bio_flags
,
838 static void run_one_async_free(struct btrfs_work
*work
)
840 struct async_submit_bio
*async
;
842 async
= container_of(work
, struct async_submit_bio
, work
);
846 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
847 int rw
, struct bio
*bio
, int mirror_num
,
848 unsigned long bio_flags
,
850 extent_submit_bio_hook_t
*submit_bio_start
,
851 extent_submit_bio_hook_t
*submit_bio_done
)
853 struct async_submit_bio
*async
;
855 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
859 async
->inode
= inode
;
862 async
->mirror_num
= mirror_num
;
863 async
->submit_bio_start
= submit_bio_start
;
864 async
->submit_bio_done
= submit_bio_done
;
866 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
867 run_one_async_done
, run_one_async_free
);
869 async
->bio_flags
= bio_flags
;
870 async
->bio_offset
= bio_offset
;
874 atomic_inc(&fs_info
->nr_async_submits
);
877 btrfs_set_work_high_priority(&async
->work
);
879 btrfs_queue_work(fs_info
->workers
, &async
->work
);
881 while (atomic_read(&fs_info
->async_submit_draining
) &&
882 atomic_read(&fs_info
->nr_async_submits
)) {
883 wait_event(fs_info
->async_submit_wait
,
884 (atomic_read(&fs_info
->nr_async_submits
) == 0));
890 static int btree_csum_one_bio(struct bio
*bio
)
892 struct bio_vec
*bvec
;
893 struct btrfs_root
*root
;
896 bio_for_each_segment_all(bvec
, bio
, i
) {
897 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
898 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
906 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
907 struct bio
*bio
, int mirror_num
,
908 unsigned long bio_flags
,
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
915 return btree_csum_one_bio(bio
);
918 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
919 int mirror_num
, unsigned long bio_flags
,
925 * when we're called for a write, we're already in the async
926 * submission context. Just jump into btrfs_map_bio
928 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
936 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
938 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
941 if (static_cpu_has_safe(X86_FEATURE_XMM4_2
))
947 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
948 int mirror_num
, unsigned long bio_flags
,
951 int async
= check_async_write(inode
, bio_flags
);
954 if (!(rw
& REQ_WRITE
)) {
956 * called for a read, do the setup so that checksum validation
957 * can happen in the async kernel threads
959 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
960 bio
, BTRFS_WQ_ENDIO_METADATA
);
963 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
966 ret
= btree_csum_one_bio(bio
);
969 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
973 * kthread helpers are used to submit writes so that
974 * checksumming can happen in parallel across all CPUs
976 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
977 inode
, rw
, bio
, mirror_num
, 0,
979 __btree_submit_bio_start
,
980 __btree_submit_bio_done
);
993 #ifdef CONFIG_MIGRATION
994 static int btree_migratepage(struct address_space
*mapping
,
995 struct page
*newpage
, struct page
*page
,
996 enum migrate_mode mode
)
999 * we can't safely write a btree page from here,
1000 * we haven't done the locking hook
1002 if (PageDirty(page
))
1005 * Buffers may be managed in a filesystem specific way.
1006 * We must have no buffers or drop them.
1008 if (page_has_private(page
) &&
1009 !try_to_release_page(page
, GFP_KERNEL
))
1011 return migrate_page(mapping
, newpage
, page
, mode
);
1016 static int btree_writepages(struct address_space
*mapping
,
1017 struct writeback_control
*wbc
)
1019 struct btrfs_fs_info
*fs_info
;
1022 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1024 if (wbc
->for_kupdate
)
1027 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1028 /* this is a bit racy, but that's ok */
1029 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1030 BTRFS_DIRTY_METADATA_THRESH
);
1034 return btree_write_cache_pages(mapping
, wbc
);
1037 static int btree_readpage(struct file
*file
, struct page
*page
)
1039 struct extent_io_tree
*tree
;
1040 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1041 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1044 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1046 if (PageWriteback(page
) || PageDirty(page
))
1049 return try_release_extent_buffer(page
);
1052 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1053 unsigned int length
)
1055 struct extent_io_tree
*tree
;
1056 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1057 extent_invalidatepage(tree
, page
, offset
);
1058 btree_releasepage(page
, GFP_NOFS
);
1059 if (PagePrivate(page
)) {
1060 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1061 "page private not zero on page %llu",
1062 (unsigned long long)page_offset(page
));
1063 ClearPagePrivate(page
);
1064 set_page_private(page
, 0);
1065 page_cache_release(page
);
1069 static int btree_set_page_dirty(struct page
*page
)
1072 struct extent_buffer
*eb
;
1074 BUG_ON(!PagePrivate(page
));
1075 eb
= (struct extent_buffer
*)page
->private;
1077 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1078 BUG_ON(!atomic_read(&eb
->refs
));
1079 btrfs_assert_tree_locked(eb
);
1081 return __set_page_dirty_nobuffers(page
);
1084 static const struct address_space_operations btree_aops
= {
1085 .readpage
= btree_readpage
,
1086 .writepages
= btree_writepages
,
1087 .releasepage
= btree_releasepage
,
1088 .invalidatepage
= btree_invalidatepage
,
1089 #ifdef CONFIG_MIGRATION
1090 .migratepage
= btree_migratepage
,
1092 .set_page_dirty
= btree_set_page_dirty
,
1095 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1097 struct extent_buffer
*buf
= NULL
;
1098 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1100 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1103 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1104 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1105 free_extent_buffer(buf
);
1108 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1109 int mirror_num
, struct extent_buffer
**eb
)
1111 struct extent_buffer
*buf
= NULL
;
1112 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1113 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1116 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1120 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1122 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1123 btree_get_extent
, mirror_num
);
1125 free_extent_buffer(buf
);
1129 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1130 free_extent_buffer(buf
);
1132 } else if (extent_buffer_uptodate(buf
)) {
1135 free_extent_buffer(buf
);
1140 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1143 return find_extent_buffer(fs_info
, bytenr
);
1146 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1149 if (btrfs_test_is_dummy_root(root
))
1150 return alloc_test_extent_buffer(root
->fs_info
, bytenr
);
1151 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1155 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1157 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1158 buf
->start
+ buf
->len
- 1);
1161 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1163 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1164 buf
->start
, buf
->start
+ buf
->len
- 1);
1167 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1170 struct extent_buffer
*buf
= NULL
;
1173 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1175 return ERR_PTR(-ENOMEM
);
1177 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1179 free_extent_buffer(buf
);
1180 return ERR_PTR(ret
);
1186 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1187 struct btrfs_fs_info
*fs_info
,
1188 struct extent_buffer
*buf
)
1190 if (btrfs_header_generation(buf
) ==
1191 fs_info
->running_transaction
->transid
) {
1192 btrfs_assert_tree_locked(buf
);
1194 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1195 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1197 fs_info
->dirty_metadata_batch
);
1198 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1199 btrfs_set_lock_blocking(buf
);
1200 clear_extent_buffer_dirty(buf
);
1205 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1207 struct btrfs_subvolume_writers
*writers
;
1210 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1212 return ERR_PTR(-ENOMEM
);
1214 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1217 return ERR_PTR(ret
);
1220 init_waitqueue_head(&writers
->wait
);
1225 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1227 percpu_counter_destroy(&writers
->counter
);
1231 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1232 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1236 root
->commit_root
= NULL
;
1237 root
->sectorsize
= sectorsize
;
1238 root
->nodesize
= nodesize
;
1239 root
->stripesize
= stripesize
;
1241 root
->orphan_cleanup_state
= 0;
1243 root
->objectid
= objectid
;
1244 root
->last_trans
= 0;
1245 root
->highest_objectid
= 0;
1246 root
->nr_delalloc_inodes
= 0;
1247 root
->nr_ordered_extents
= 0;
1249 root
->inode_tree
= RB_ROOT
;
1250 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1251 root
->block_rsv
= NULL
;
1252 root
->orphan_block_rsv
= NULL
;
1254 INIT_LIST_HEAD(&root
->dirty_list
);
1255 INIT_LIST_HEAD(&root
->root_list
);
1256 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1257 INIT_LIST_HEAD(&root
->delalloc_root
);
1258 INIT_LIST_HEAD(&root
->ordered_extents
);
1259 INIT_LIST_HEAD(&root
->ordered_root
);
1260 INIT_LIST_HEAD(&root
->logged_list
[0]);
1261 INIT_LIST_HEAD(&root
->logged_list
[1]);
1262 spin_lock_init(&root
->orphan_lock
);
1263 spin_lock_init(&root
->inode_lock
);
1264 spin_lock_init(&root
->delalloc_lock
);
1265 spin_lock_init(&root
->ordered_extent_lock
);
1266 spin_lock_init(&root
->accounting_lock
);
1267 spin_lock_init(&root
->log_extents_lock
[0]);
1268 spin_lock_init(&root
->log_extents_lock
[1]);
1269 mutex_init(&root
->objectid_mutex
);
1270 mutex_init(&root
->log_mutex
);
1271 mutex_init(&root
->ordered_extent_mutex
);
1272 mutex_init(&root
->delalloc_mutex
);
1273 init_waitqueue_head(&root
->log_writer_wait
);
1274 init_waitqueue_head(&root
->log_commit_wait
[0]);
1275 init_waitqueue_head(&root
->log_commit_wait
[1]);
1276 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1277 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1278 atomic_set(&root
->log_commit
[0], 0);
1279 atomic_set(&root
->log_commit
[1], 0);
1280 atomic_set(&root
->log_writers
, 0);
1281 atomic_set(&root
->log_batch
, 0);
1282 atomic_set(&root
->orphan_inodes
, 0);
1283 atomic_set(&root
->refs
, 1);
1284 atomic_set(&root
->will_be_snapshoted
, 0);
1285 atomic_set(&root
->qgroup_meta_rsv
, 0);
1286 root
->log_transid
= 0;
1287 root
->log_transid_committed
= -1;
1288 root
->last_log_commit
= 0;
1290 extent_io_tree_init(&root
->dirty_log_pages
,
1291 fs_info
->btree_inode
->i_mapping
);
1293 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1294 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1295 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1297 root
->defrag_trans_start
= fs_info
->generation
;
1299 root
->defrag_trans_start
= 0;
1300 root
->root_key
.objectid
= objectid
;
1303 spin_lock_init(&root
->root_item_lock
);
1306 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1309 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1311 root
->fs_info
= fs_info
;
1315 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1316 /* Should only be used by the testing infrastructure */
1317 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1319 struct btrfs_root
*root
;
1321 root
= btrfs_alloc_root(NULL
, GFP_KERNEL
);
1323 return ERR_PTR(-ENOMEM
);
1324 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1325 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1326 root
->alloc_bytenr
= 0;
1332 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1333 struct btrfs_fs_info
*fs_info
,
1336 struct extent_buffer
*leaf
;
1337 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1338 struct btrfs_root
*root
;
1339 struct btrfs_key key
;
1343 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1345 return ERR_PTR(-ENOMEM
);
1347 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1348 tree_root
->stripesize
, root
, fs_info
, objectid
);
1349 root
->root_key
.objectid
= objectid
;
1350 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1351 root
->root_key
.offset
= 0;
1353 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1355 ret
= PTR_ERR(leaf
);
1360 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1361 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1362 btrfs_set_header_generation(leaf
, trans
->transid
);
1363 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1364 btrfs_set_header_owner(leaf
, objectid
);
1367 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1369 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1370 btrfs_header_chunk_tree_uuid(leaf
),
1372 btrfs_mark_buffer_dirty(leaf
);
1374 root
->commit_root
= btrfs_root_node(root
);
1375 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1377 root
->root_item
.flags
= 0;
1378 root
->root_item
.byte_limit
= 0;
1379 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1380 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1381 btrfs_set_root_level(&root
->root_item
, 0);
1382 btrfs_set_root_refs(&root
->root_item
, 1);
1383 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1384 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1385 btrfs_set_root_dirid(&root
->root_item
, 0);
1387 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1388 root
->root_item
.drop_level
= 0;
1390 key
.objectid
= objectid
;
1391 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1393 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1397 btrfs_tree_unlock(leaf
);
1403 btrfs_tree_unlock(leaf
);
1404 free_extent_buffer(root
->commit_root
);
1405 free_extent_buffer(leaf
);
1409 return ERR_PTR(ret
);
1412 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1413 struct btrfs_fs_info
*fs_info
)
1415 struct btrfs_root
*root
;
1416 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1417 struct extent_buffer
*leaf
;
1419 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1421 return ERR_PTR(-ENOMEM
);
1423 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1424 tree_root
->stripesize
, root
, fs_info
,
1425 BTRFS_TREE_LOG_OBJECTID
);
1427 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1428 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1429 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1432 * DON'T set REF_COWS for log trees
1434 * log trees do not get reference counted because they go away
1435 * before a real commit is actually done. They do store pointers
1436 * to file data extents, and those reference counts still get
1437 * updated (along with back refs to the log tree).
1440 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1444 return ERR_CAST(leaf
);
1447 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1448 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1449 btrfs_set_header_generation(leaf
, trans
->transid
);
1450 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1451 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1454 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1455 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1456 btrfs_mark_buffer_dirty(root
->node
);
1457 btrfs_tree_unlock(root
->node
);
1461 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1462 struct btrfs_fs_info
*fs_info
)
1464 struct btrfs_root
*log_root
;
1466 log_root
= alloc_log_tree(trans
, fs_info
);
1467 if (IS_ERR(log_root
))
1468 return PTR_ERR(log_root
);
1469 WARN_ON(fs_info
->log_root_tree
);
1470 fs_info
->log_root_tree
= log_root
;
1474 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1475 struct btrfs_root
*root
)
1477 struct btrfs_root
*log_root
;
1478 struct btrfs_inode_item
*inode_item
;
1480 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1481 if (IS_ERR(log_root
))
1482 return PTR_ERR(log_root
);
1484 log_root
->last_trans
= trans
->transid
;
1485 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1487 inode_item
= &log_root
->root_item
.inode
;
1488 btrfs_set_stack_inode_generation(inode_item
, 1);
1489 btrfs_set_stack_inode_size(inode_item
, 3);
1490 btrfs_set_stack_inode_nlink(inode_item
, 1);
1491 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1492 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1494 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1496 WARN_ON(root
->log_root
);
1497 root
->log_root
= log_root
;
1498 root
->log_transid
= 0;
1499 root
->log_transid_committed
= -1;
1500 root
->last_log_commit
= 0;
1504 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1505 struct btrfs_key
*key
)
1507 struct btrfs_root
*root
;
1508 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1509 struct btrfs_path
*path
;
1513 path
= btrfs_alloc_path();
1515 return ERR_PTR(-ENOMEM
);
1517 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1523 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1524 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1526 ret
= btrfs_find_root(tree_root
, key
, path
,
1527 &root
->root_item
, &root
->root_key
);
1534 generation
= btrfs_root_generation(&root
->root_item
);
1535 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1537 if (IS_ERR(root
->node
)) {
1538 ret
= PTR_ERR(root
->node
);
1540 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1542 free_extent_buffer(root
->node
);
1545 root
->commit_root
= btrfs_root_node(root
);
1547 btrfs_free_path(path
);
1553 root
= ERR_PTR(ret
);
1557 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1558 struct btrfs_key
*location
)
1560 struct btrfs_root
*root
;
1562 root
= btrfs_read_tree_root(tree_root
, location
);
1566 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1567 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1568 btrfs_check_and_init_root_item(&root
->root_item
);
1574 int btrfs_init_fs_root(struct btrfs_root
*root
)
1577 struct btrfs_subvolume_writers
*writers
;
1579 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1580 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1582 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1587 writers
= btrfs_alloc_subvolume_writers();
1588 if (IS_ERR(writers
)) {
1589 ret
= PTR_ERR(writers
);
1592 root
->subv_writers
= writers
;
1594 btrfs_init_free_ino_ctl(root
);
1595 spin_lock_init(&root
->ino_cache_lock
);
1596 init_waitqueue_head(&root
->ino_cache_wait
);
1598 ret
= get_anon_bdev(&root
->anon_dev
);
1602 mutex_lock(&root
->objectid_mutex
);
1603 ret
= btrfs_find_highest_objectid(root
,
1604 &root
->highest_objectid
);
1606 mutex_unlock(&root
->objectid_mutex
);
1610 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1612 mutex_unlock(&root
->objectid_mutex
);
1617 free_anon_bdev(root
->anon_dev
);
1619 btrfs_free_subvolume_writers(root
->subv_writers
);
1621 kfree(root
->free_ino_ctl
);
1622 kfree(root
->free_ino_pinned
);
1626 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1629 struct btrfs_root
*root
;
1631 spin_lock(&fs_info
->fs_roots_radix_lock
);
1632 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1633 (unsigned long)root_id
);
1634 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1638 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1639 struct btrfs_root
*root
)
1643 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1647 spin_lock(&fs_info
->fs_roots_radix_lock
);
1648 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1649 (unsigned long)root
->root_key
.objectid
,
1652 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1653 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1654 radix_tree_preload_end();
1659 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1660 struct btrfs_key
*location
,
1663 struct btrfs_root
*root
;
1664 struct btrfs_path
*path
;
1665 struct btrfs_key key
;
1668 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1669 return fs_info
->tree_root
;
1670 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1671 return fs_info
->extent_root
;
1672 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1673 return fs_info
->chunk_root
;
1674 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1675 return fs_info
->dev_root
;
1676 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1677 return fs_info
->csum_root
;
1678 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1679 return fs_info
->quota_root
? fs_info
->quota_root
:
1681 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1682 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1684 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1685 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1688 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1690 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1691 return ERR_PTR(-ENOENT
);
1695 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1699 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1704 ret
= btrfs_init_fs_root(root
);
1708 path
= btrfs_alloc_path();
1713 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1714 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1715 key
.offset
= location
->objectid
;
1717 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1718 btrfs_free_path(path
);
1722 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1724 ret
= btrfs_insert_fs_root(fs_info
, root
);
1726 if (ret
== -EEXIST
) {
1735 return ERR_PTR(ret
);
1738 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1740 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1742 struct btrfs_device
*device
;
1743 struct backing_dev_info
*bdi
;
1746 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1749 bdi
= blk_get_backing_dev_info(device
->bdev
);
1750 if (bdi_congested(bdi
, bdi_bits
)) {
1759 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1763 err
= bdi_setup_and_register(bdi
, "btrfs");
1767 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
;
1768 bdi
->congested_fn
= btrfs_congested_fn
;
1769 bdi
->congested_data
= info
;
1770 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1775 * called by the kthread helper functions to finally call the bio end_io
1776 * functions. This is where read checksum verification actually happens
1778 static void end_workqueue_fn(struct btrfs_work
*work
)
1781 struct btrfs_end_io_wq
*end_io_wq
;
1783 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1784 bio
= end_io_wq
->bio
;
1786 bio
->bi_error
= end_io_wq
->error
;
1787 bio
->bi_private
= end_io_wq
->private;
1788 bio
->bi_end_io
= end_io_wq
->end_io
;
1789 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1793 static int cleaner_kthread(void *arg
)
1795 struct btrfs_root
*root
= arg
;
1797 struct btrfs_trans_handle
*trans
;
1802 /* Make the cleaner go to sleep early. */
1803 if (btrfs_need_cleaner_sleep(root
))
1806 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1810 * Avoid the problem that we change the status of the fs
1811 * during the above check and trylock.
1813 if (btrfs_need_cleaner_sleep(root
)) {
1814 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1818 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1819 btrfs_run_delayed_iputs(root
);
1820 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1822 again
= btrfs_clean_one_deleted_snapshot(root
);
1823 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1826 * The defragger has dealt with the R/O remount and umount,
1827 * needn't do anything special here.
1829 btrfs_run_defrag_inodes(root
->fs_info
);
1832 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1833 * with relocation (btrfs_relocate_chunk) and relocation
1834 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1835 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1836 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1837 * unused block groups.
1839 btrfs_delete_unused_bgs(root
->fs_info
);
1842 set_current_state(TASK_INTERRUPTIBLE
);
1843 if (!kthread_should_stop())
1845 __set_current_state(TASK_RUNNING
);
1847 } while (!kthread_should_stop());
1850 * Transaction kthread is stopped before us and wakes us up.
1851 * However we might have started a new transaction and COWed some
1852 * tree blocks when deleting unused block groups for example. So
1853 * make sure we commit the transaction we started to have a clean
1854 * shutdown when evicting the btree inode - if it has dirty pages
1855 * when we do the final iput() on it, eviction will trigger a
1856 * writeback for it which will fail with null pointer dereferences
1857 * since work queues and other resources were already released and
1858 * destroyed by the time the iput/eviction/writeback is made.
1860 trans
= btrfs_attach_transaction(root
);
1861 if (IS_ERR(trans
)) {
1862 if (PTR_ERR(trans
) != -ENOENT
)
1863 btrfs_err(root
->fs_info
,
1864 "cleaner transaction attach returned %ld",
1869 ret
= btrfs_commit_transaction(trans
, root
);
1871 btrfs_err(root
->fs_info
,
1872 "cleaner open transaction commit returned %d",
1879 static int transaction_kthread(void *arg
)
1881 struct btrfs_root
*root
= arg
;
1882 struct btrfs_trans_handle
*trans
;
1883 struct btrfs_transaction
*cur
;
1886 unsigned long delay
;
1890 cannot_commit
= false;
1891 delay
= HZ
* root
->fs_info
->commit_interval
;
1892 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1894 spin_lock(&root
->fs_info
->trans_lock
);
1895 cur
= root
->fs_info
->running_transaction
;
1897 spin_unlock(&root
->fs_info
->trans_lock
);
1901 now
= get_seconds();
1902 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1903 (now
< cur
->start_time
||
1904 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1905 spin_unlock(&root
->fs_info
->trans_lock
);
1909 transid
= cur
->transid
;
1910 spin_unlock(&root
->fs_info
->trans_lock
);
1912 /* If the file system is aborted, this will always fail. */
1913 trans
= btrfs_attach_transaction(root
);
1914 if (IS_ERR(trans
)) {
1915 if (PTR_ERR(trans
) != -ENOENT
)
1916 cannot_commit
= true;
1919 if (transid
== trans
->transid
) {
1920 btrfs_commit_transaction(trans
, root
);
1922 btrfs_end_transaction(trans
, root
);
1925 wake_up_process(root
->fs_info
->cleaner_kthread
);
1926 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1928 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1929 &root
->fs_info
->fs_state
)))
1930 btrfs_cleanup_transaction(root
);
1931 if (!try_to_freeze()) {
1932 set_current_state(TASK_INTERRUPTIBLE
);
1933 if (!kthread_should_stop() &&
1934 (!btrfs_transaction_blocked(root
->fs_info
) ||
1936 schedule_timeout(delay
);
1937 __set_current_state(TASK_RUNNING
);
1939 } while (!kthread_should_stop());
1944 * this will find the highest generation in the array of
1945 * root backups. The index of the highest array is returned,
1946 * or -1 if we can't find anything.
1948 * We check to make sure the array is valid by comparing the
1949 * generation of the latest root in the array with the generation
1950 * in the super block. If they don't match we pitch it.
1952 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1955 int newest_index
= -1;
1956 struct btrfs_root_backup
*root_backup
;
1959 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1960 root_backup
= info
->super_copy
->super_roots
+ i
;
1961 cur
= btrfs_backup_tree_root_gen(root_backup
);
1962 if (cur
== newest_gen
)
1966 /* check to see if we actually wrapped around */
1967 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1968 root_backup
= info
->super_copy
->super_roots
;
1969 cur
= btrfs_backup_tree_root_gen(root_backup
);
1970 if (cur
== newest_gen
)
1973 return newest_index
;
1978 * find the oldest backup so we know where to store new entries
1979 * in the backup array. This will set the backup_root_index
1980 * field in the fs_info struct
1982 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1985 int newest_index
= -1;
1987 newest_index
= find_newest_super_backup(info
, newest_gen
);
1988 /* if there was garbage in there, just move along */
1989 if (newest_index
== -1) {
1990 info
->backup_root_index
= 0;
1992 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1997 * copy all the root pointers into the super backup array.
1998 * this will bump the backup pointer by one when it is
2001 static void backup_super_roots(struct btrfs_fs_info
*info
)
2004 struct btrfs_root_backup
*root_backup
;
2007 next_backup
= info
->backup_root_index
;
2008 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2009 BTRFS_NUM_BACKUP_ROOTS
;
2012 * just overwrite the last backup if we're at the same generation
2013 * this happens only at umount
2015 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
2016 if (btrfs_backup_tree_root_gen(root_backup
) ==
2017 btrfs_header_generation(info
->tree_root
->node
))
2018 next_backup
= last_backup
;
2020 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
2023 * make sure all of our padding and empty slots get zero filled
2024 * regardless of which ones we use today
2026 memset(root_backup
, 0, sizeof(*root_backup
));
2028 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
2030 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
2031 btrfs_set_backup_tree_root_gen(root_backup
,
2032 btrfs_header_generation(info
->tree_root
->node
));
2034 btrfs_set_backup_tree_root_level(root_backup
,
2035 btrfs_header_level(info
->tree_root
->node
));
2037 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2038 btrfs_set_backup_chunk_root_gen(root_backup
,
2039 btrfs_header_generation(info
->chunk_root
->node
));
2040 btrfs_set_backup_chunk_root_level(root_backup
,
2041 btrfs_header_level(info
->chunk_root
->node
));
2043 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2044 btrfs_set_backup_extent_root_gen(root_backup
,
2045 btrfs_header_generation(info
->extent_root
->node
));
2046 btrfs_set_backup_extent_root_level(root_backup
,
2047 btrfs_header_level(info
->extent_root
->node
));
2050 * we might commit during log recovery, which happens before we set
2051 * the fs_root. Make sure it is valid before we fill it in.
2053 if (info
->fs_root
&& info
->fs_root
->node
) {
2054 btrfs_set_backup_fs_root(root_backup
,
2055 info
->fs_root
->node
->start
);
2056 btrfs_set_backup_fs_root_gen(root_backup
,
2057 btrfs_header_generation(info
->fs_root
->node
));
2058 btrfs_set_backup_fs_root_level(root_backup
,
2059 btrfs_header_level(info
->fs_root
->node
));
2062 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2063 btrfs_set_backup_dev_root_gen(root_backup
,
2064 btrfs_header_generation(info
->dev_root
->node
));
2065 btrfs_set_backup_dev_root_level(root_backup
,
2066 btrfs_header_level(info
->dev_root
->node
));
2068 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2069 btrfs_set_backup_csum_root_gen(root_backup
,
2070 btrfs_header_generation(info
->csum_root
->node
));
2071 btrfs_set_backup_csum_root_level(root_backup
,
2072 btrfs_header_level(info
->csum_root
->node
));
2074 btrfs_set_backup_total_bytes(root_backup
,
2075 btrfs_super_total_bytes(info
->super_copy
));
2076 btrfs_set_backup_bytes_used(root_backup
,
2077 btrfs_super_bytes_used(info
->super_copy
));
2078 btrfs_set_backup_num_devices(root_backup
,
2079 btrfs_super_num_devices(info
->super_copy
));
2082 * if we don't copy this out to the super_copy, it won't get remembered
2083 * for the next commit
2085 memcpy(&info
->super_copy
->super_roots
,
2086 &info
->super_for_commit
->super_roots
,
2087 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2091 * this copies info out of the root backup array and back into
2092 * the in-memory super block. It is meant to help iterate through
2093 * the array, so you send it the number of backups you've already
2094 * tried and the last backup index you used.
2096 * this returns -1 when it has tried all the backups
2098 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2099 struct btrfs_super_block
*super
,
2100 int *num_backups_tried
, int *backup_index
)
2102 struct btrfs_root_backup
*root_backup
;
2103 int newest
= *backup_index
;
2105 if (*num_backups_tried
== 0) {
2106 u64 gen
= btrfs_super_generation(super
);
2108 newest
= find_newest_super_backup(info
, gen
);
2112 *backup_index
= newest
;
2113 *num_backups_tried
= 1;
2114 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2115 /* we've tried all the backups, all done */
2118 /* jump to the next oldest backup */
2119 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2120 BTRFS_NUM_BACKUP_ROOTS
;
2121 *backup_index
= newest
;
2122 *num_backups_tried
+= 1;
2124 root_backup
= super
->super_roots
+ newest
;
2126 btrfs_set_super_generation(super
,
2127 btrfs_backup_tree_root_gen(root_backup
));
2128 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2129 btrfs_set_super_root_level(super
,
2130 btrfs_backup_tree_root_level(root_backup
));
2131 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2134 * fixme: the total bytes and num_devices need to match or we should
2137 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2138 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2142 /* helper to cleanup workers */
2143 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2145 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2146 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2147 btrfs_destroy_workqueue(fs_info
->workers
);
2148 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2149 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2150 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2151 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2152 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2153 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2154 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2155 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2156 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2157 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2158 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2159 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2160 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2161 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2162 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2165 static void free_root_extent_buffers(struct btrfs_root
*root
)
2168 free_extent_buffer(root
->node
);
2169 free_extent_buffer(root
->commit_root
);
2171 root
->commit_root
= NULL
;
2175 /* helper to cleanup tree roots */
2176 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2178 free_root_extent_buffers(info
->tree_root
);
2180 free_root_extent_buffers(info
->dev_root
);
2181 free_root_extent_buffers(info
->extent_root
);
2182 free_root_extent_buffers(info
->csum_root
);
2183 free_root_extent_buffers(info
->quota_root
);
2184 free_root_extent_buffers(info
->uuid_root
);
2186 free_root_extent_buffers(info
->chunk_root
);
2187 free_root_extent_buffers(info
->free_space_root
);
2190 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2193 struct btrfs_root
*gang
[8];
2196 while (!list_empty(&fs_info
->dead_roots
)) {
2197 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2198 struct btrfs_root
, root_list
);
2199 list_del(&gang
[0]->root_list
);
2201 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2202 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2204 free_extent_buffer(gang
[0]->node
);
2205 free_extent_buffer(gang
[0]->commit_root
);
2206 btrfs_put_fs_root(gang
[0]);
2211 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2216 for (i
= 0; i
< ret
; i
++)
2217 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2220 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2221 btrfs_free_log_root_tree(NULL
, fs_info
);
2222 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2223 fs_info
->pinned_extents
);
2227 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2229 mutex_init(&fs_info
->scrub_lock
);
2230 atomic_set(&fs_info
->scrubs_running
, 0);
2231 atomic_set(&fs_info
->scrub_pause_req
, 0);
2232 atomic_set(&fs_info
->scrubs_paused
, 0);
2233 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2234 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2235 fs_info
->scrub_workers_refcnt
= 0;
2238 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2240 spin_lock_init(&fs_info
->balance_lock
);
2241 mutex_init(&fs_info
->balance_mutex
);
2242 atomic_set(&fs_info
->balance_running
, 0);
2243 atomic_set(&fs_info
->balance_pause_req
, 0);
2244 atomic_set(&fs_info
->balance_cancel_req
, 0);
2245 fs_info
->balance_ctl
= NULL
;
2246 init_waitqueue_head(&fs_info
->balance_wait_q
);
2249 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2250 struct btrfs_root
*tree_root
)
2252 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2253 set_nlink(fs_info
->btree_inode
, 1);
2255 * we set the i_size on the btree inode to the max possible int.
2256 * the real end of the address space is determined by all of
2257 * the devices in the system
2259 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2260 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2262 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2263 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2264 fs_info
->btree_inode
->i_mapping
);
2265 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2266 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2268 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2270 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2271 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2272 sizeof(struct btrfs_key
));
2273 set_bit(BTRFS_INODE_DUMMY
,
2274 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2275 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2278 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2280 fs_info
->dev_replace
.lock_owner
= 0;
2281 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2282 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2283 rwlock_init(&fs_info
->dev_replace
.lock
);
2284 atomic_set(&fs_info
->dev_replace
.read_locks
, 0);
2285 atomic_set(&fs_info
->dev_replace
.blocking_readers
, 0);
2286 init_waitqueue_head(&fs_info
->replace_wait
);
2287 init_waitqueue_head(&fs_info
->dev_replace
.read_lock_wq
);
2290 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2292 spin_lock_init(&fs_info
->qgroup_lock
);
2293 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2294 fs_info
->qgroup_tree
= RB_ROOT
;
2295 fs_info
->qgroup_op_tree
= RB_ROOT
;
2296 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2297 fs_info
->qgroup_seq
= 1;
2298 fs_info
->quota_enabled
= 0;
2299 fs_info
->pending_quota_state
= 0;
2300 fs_info
->qgroup_ulist
= NULL
;
2301 mutex_init(&fs_info
->qgroup_rescan_lock
);
2304 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2305 struct btrfs_fs_devices
*fs_devices
)
2307 int max_active
= fs_info
->thread_pool_size
;
2308 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2311 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2314 fs_info
->delalloc_workers
=
2315 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2317 fs_info
->flush_workers
=
2318 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2320 fs_info
->caching_workers
=
2321 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2324 * a higher idle thresh on the submit workers makes it much more
2325 * likely that bios will be send down in a sane order to the
2328 fs_info
->submit_workers
=
2329 btrfs_alloc_workqueue("submit", flags
,
2330 min_t(u64
, fs_devices
->num_devices
,
2333 fs_info
->fixup_workers
=
2334 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2337 * endios are largely parallel and should have a very
2340 fs_info
->endio_workers
=
2341 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2342 fs_info
->endio_meta_workers
=
2343 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2344 fs_info
->endio_meta_write_workers
=
2345 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2346 fs_info
->endio_raid56_workers
=
2347 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2348 fs_info
->endio_repair_workers
=
2349 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2350 fs_info
->rmw_workers
=
2351 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2352 fs_info
->endio_write_workers
=
2353 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2354 fs_info
->endio_freespace_worker
=
2355 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2356 fs_info
->delayed_workers
=
2357 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2358 fs_info
->readahead_workers
=
2359 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2360 fs_info
->qgroup_rescan_workers
=
2361 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2362 fs_info
->extent_workers
=
2363 btrfs_alloc_workqueue("extent-refs", flags
,
2364 min_t(u64
, fs_devices
->num_devices
,
2367 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2368 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2369 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2370 fs_info
->endio_meta_write_workers
&&
2371 fs_info
->endio_repair_workers
&&
2372 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2373 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2374 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2375 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2376 fs_info
->extent_workers
&&
2377 fs_info
->qgroup_rescan_workers
)) {
2384 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2385 struct btrfs_fs_devices
*fs_devices
)
2388 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2389 struct btrfs_root
*log_tree_root
;
2390 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2391 u64 bytenr
= btrfs_super_log_root(disk_super
);
2393 if (fs_devices
->rw_devices
== 0) {
2394 btrfs_warn(fs_info
, "log replay required on RO media");
2398 log_tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2402 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2403 tree_root
->stripesize
, log_tree_root
, fs_info
,
2404 BTRFS_TREE_LOG_OBJECTID
);
2406 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2407 fs_info
->generation
+ 1);
2408 if (IS_ERR(log_tree_root
->node
)) {
2409 btrfs_warn(fs_info
, "failed to read log tree");
2410 ret
= PTR_ERR(log_tree_root
->node
);
2411 kfree(log_tree_root
);
2413 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2414 btrfs_err(fs_info
, "failed to read log tree");
2415 free_extent_buffer(log_tree_root
->node
);
2416 kfree(log_tree_root
);
2419 /* returns with log_tree_root freed on success */
2420 ret
= btrfs_recover_log_trees(log_tree_root
);
2422 btrfs_std_error(tree_root
->fs_info
, ret
,
2423 "Failed to recover log tree");
2424 free_extent_buffer(log_tree_root
->node
);
2425 kfree(log_tree_root
);
2429 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2430 ret
= btrfs_commit_super(tree_root
);
2438 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2439 struct btrfs_root
*tree_root
)
2441 struct btrfs_root
*root
;
2442 struct btrfs_key location
;
2445 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2446 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2447 location
.offset
= 0;
2449 root
= btrfs_read_tree_root(tree_root
, &location
);
2451 return PTR_ERR(root
);
2452 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2453 fs_info
->extent_root
= root
;
2455 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2456 root
= btrfs_read_tree_root(tree_root
, &location
);
2458 return PTR_ERR(root
);
2459 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2460 fs_info
->dev_root
= root
;
2461 btrfs_init_devices_late(fs_info
);
2463 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2464 root
= btrfs_read_tree_root(tree_root
, &location
);
2466 return PTR_ERR(root
);
2467 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2468 fs_info
->csum_root
= root
;
2470 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2471 root
= btrfs_read_tree_root(tree_root
, &location
);
2472 if (!IS_ERR(root
)) {
2473 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2474 fs_info
->quota_enabled
= 1;
2475 fs_info
->pending_quota_state
= 1;
2476 fs_info
->quota_root
= root
;
2479 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2480 root
= btrfs_read_tree_root(tree_root
, &location
);
2482 ret
= PTR_ERR(root
);
2486 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2487 fs_info
->uuid_root
= root
;
2490 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2491 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2492 root
= btrfs_read_tree_root(tree_root
, &location
);
2494 return PTR_ERR(root
);
2495 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2496 fs_info
->free_space_root
= root
;
2502 int open_ctree(struct super_block
*sb
,
2503 struct btrfs_fs_devices
*fs_devices
,
2511 struct btrfs_key location
;
2512 struct buffer_head
*bh
;
2513 struct btrfs_super_block
*disk_super
;
2514 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2515 struct btrfs_root
*tree_root
;
2516 struct btrfs_root
*chunk_root
;
2519 int num_backups_tried
= 0;
2520 int backup_index
= 0;
2523 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2524 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2525 if (!tree_root
|| !chunk_root
) {
2530 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2536 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2542 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2547 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2548 (1 + ilog2(nr_cpu_ids
));
2550 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2553 goto fail_dirty_metadata_bytes
;
2556 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2559 goto fail_delalloc_bytes
;
2562 fs_info
->btree_inode
= new_inode(sb
);
2563 if (!fs_info
->btree_inode
) {
2565 goto fail_bio_counter
;
2568 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2570 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2571 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2572 INIT_LIST_HEAD(&fs_info
->trans_list
);
2573 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2574 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2575 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2576 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2577 spin_lock_init(&fs_info
->delalloc_root_lock
);
2578 spin_lock_init(&fs_info
->trans_lock
);
2579 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2580 spin_lock_init(&fs_info
->delayed_iput_lock
);
2581 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2582 spin_lock_init(&fs_info
->free_chunk_lock
);
2583 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2584 spin_lock_init(&fs_info
->super_lock
);
2585 spin_lock_init(&fs_info
->qgroup_op_lock
);
2586 spin_lock_init(&fs_info
->buffer_lock
);
2587 spin_lock_init(&fs_info
->unused_bgs_lock
);
2588 rwlock_init(&fs_info
->tree_mod_log_lock
);
2589 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2590 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2591 mutex_init(&fs_info
->reloc_mutex
);
2592 mutex_init(&fs_info
->delalloc_root_mutex
);
2593 mutex_init(&fs_info
->cleaner_delayed_iput_mutex
);
2594 seqlock_init(&fs_info
->profiles_lock
);
2596 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2597 INIT_LIST_HEAD(&fs_info
->space_info
);
2598 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2599 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2600 btrfs_mapping_init(&fs_info
->mapping_tree
);
2601 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2602 BTRFS_BLOCK_RSV_GLOBAL
);
2603 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2604 BTRFS_BLOCK_RSV_DELALLOC
);
2605 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2606 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2607 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2608 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2609 BTRFS_BLOCK_RSV_DELOPS
);
2610 atomic_set(&fs_info
->nr_async_submits
, 0);
2611 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2612 atomic_set(&fs_info
->async_submit_draining
, 0);
2613 atomic_set(&fs_info
->nr_async_bios
, 0);
2614 atomic_set(&fs_info
->defrag_running
, 0);
2615 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2616 atomic_set(&fs_info
->reada_works_cnt
, 0);
2617 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2619 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2620 fs_info
->metadata_ratio
= 0;
2621 fs_info
->defrag_inodes
= RB_ROOT
;
2622 fs_info
->free_chunk_space
= 0;
2623 fs_info
->tree_mod_log
= RB_ROOT
;
2624 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2625 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2626 /* readahead state */
2627 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2628 spin_lock_init(&fs_info
->reada_lock
);
2630 fs_info
->thread_pool_size
= min_t(unsigned long,
2631 num_online_cpus() + 2, 8);
2633 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2634 spin_lock_init(&fs_info
->ordered_root_lock
);
2635 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2637 if (!fs_info
->delayed_root
) {
2641 btrfs_init_delayed_root(fs_info
->delayed_root
);
2643 btrfs_init_scrub(fs_info
);
2644 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2645 fs_info
->check_integrity_print_mask
= 0;
2647 btrfs_init_balance(fs_info
);
2648 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2650 sb
->s_blocksize
= 4096;
2651 sb
->s_blocksize_bits
= blksize_bits(4096);
2652 sb
->s_bdi
= &fs_info
->bdi
;
2654 btrfs_init_btree_inode(fs_info
, tree_root
);
2656 spin_lock_init(&fs_info
->block_group_cache_lock
);
2657 fs_info
->block_group_cache_tree
= RB_ROOT
;
2658 fs_info
->first_logical_byte
= (u64
)-1;
2660 extent_io_tree_init(&fs_info
->freed_extents
[0],
2661 fs_info
->btree_inode
->i_mapping
);
2662 extent_io_tree_init(&fs_info
->freed_extents
[1],
2663 fs_info
->btree_inode
->i_mapping
);
2664 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2665 fs_info
->do_barriers
= 1;
2668 mutex_init(&fs_info
->ordered_operations_mutex
);
2669 mutex_init(&fs_info
->tree_log_mutex
);
2670 mutex_init(&fs_info
->chunk_mutex
);
2671 mutex_init(&fs_info
->transaction_kthread_mutex
);
2672 mutex_init(&fs_info
->cleaner_mutex
);
2673 mutex_init(&fs_info
->volume_mutex
);
2674 mutex_init(&fs_info
->ro_block_group_mutex
);
2675 init_rwsem(&fs_info
->commit_root_sem
);
2676 init_rwsem(&fs_info
->cleanup_work_sem
);
2677 init_rwsem(&fs_info
->subvol_sem
);
2678 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2680 btrfs_init_dev_replace_locks(fs_info
);
2681 btrfs_init_qgroup(fs_info
);
2683 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2684 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2686 init_waitqueue_head(&fs_info
->transaction_throttle
);
2687 init_waitqueue_head(&fs_info
->transaction_wait
);
2688 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2689 init_waitqueue_head(&fs_info
->async_submit_wait
);
2691 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2693 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2699 __setup_root(4096, 4096, 4096, tree_root
,
2700 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2702 invalidate_bdev(fs_devices
->latest_bdev
);
2705 * Read super block and check the signature bytes only
2707 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2714 * We want to check superblock checksum, the type is stored inside.
2715 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2717 if (btrfs_check_super_csum(bh
->b_data
)) {
2718 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2725 * super_copy is zeroed at allocation time and we never touch the
2726 * following bytes up to INFO_SIZE, the checksum is calculated from
2727 * the whole block of INFO_SIZE
2729 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2730 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2731 sizeof(*fs_info
->super_for_commit
));
2734 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2736 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2738 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2743 disk_super
= fs_info
->super_copy
;
2744 if (!btrfs_super_root(disk_super
))
2747 /* check FS state, whether FS is broken. */
2748 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2749 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2752 * run through our array of backup supers and setup
2753 * our ring pointer to the oldest one
2755 generation
= btrfs_super_generation(disk_super
);
2756 find_oldest_super_backup(fs_info
, generation
);
2759 * In the long term, we'll store the compression type in the super
2760 * block, and it'll be used for per file compression control.
2762 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2764 ret
= btrfs_parse_options(tree_root
, options
, sb
->s_flags
);
2770 features
= btrfs_super_incompat_flags(disk_super
) &
2771 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2773 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2774 "unsupported optional features (%Lx).\n",
2780 features
= btrfs_super_incompat_flags(disk_super
);
2781 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2782 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2783 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2785 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2786 printk(KERN_INFO
"BTRFS: has skinny extents\n");
2789 * flag our filesystem as having big metadata blocks if
2790 * they are bigger than the page size
2792 if (btrfs_super_nodesize(disk_super
) > PAGE_CACHE_SIZE
) {
2793 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2794 printk(KERN_INFO
"BTRFS: flagging fs with big metadata feature\n");
2795 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2798 nodesize
= btrfs_super_nodesize(disk_super
);
2799 sectorsize
= btrfs_super_sectorsize(disk_super
);
2800 stripesize
= btrfs_super_stripesize(disk_super
);
2801 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2802 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2805 * mixed block groups end up with duplicate but slightly offset
2806 * extent buffers for the same range. It leads to corruptions
2808 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2809 (sectorsize
!= nodesize
)) {
2810 printk(KERN_ERR
"BTRFS: unequal leaf/node/sector sizes "
2811 "are not allowed for mixed block groups on %s\n",
2817 * Needn't use the lock because there is no other task which will
2820 btrfs_set_super_incompat_flags(disk_super
, features
);
2822 features
= btrfs_super_compat_ro_flags(disk_super
) &
2823 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2824 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2825 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2826 "unsupported option features (%Lx).\n",
2832 max_active
= fs_info
->thread_pool_size
;
2834 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2837 goto fail_sb_buffer
;
2840 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2841 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2842 SZ_4M
/ PAGE_CACHE_SIZE
);
2844 tree_root
->nodesize
= nodesize
;
2845 tree_root
->sectorsize
= sectorsize
;
2846 tree_root
->stripesize
= stripesize
;
2848 sb
->s_blocksize
= sectorsize
;
2849 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2851 mutex_lock(&fs_info
->chunk_mutex
);
2852 ret
= btrfs_read_sys_array(tree_root
);
2853 mutex_unlock(&fs_info
->chunk_mutex
);
2855 printk(KERN_ERR
"BTRFS: failed to read the system "
2856 "array on %s\n", sb
->s_id
);
2857 goto fail_sb_buffer
;
2860 generation
= btrfs_super_chunk_root_generation(disk_super
);
2862 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2863 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2865 chunk_root
->node
= read_tree_block(chunk_root
,
2866 btrfs_super_chunk_root(disk_super
),
2868 if (IS_ERR(chunk_root
->node
) ||
2869 !extent_buffer_uptodate(chunk_root
->node
)) {
2870 printk(KERN_ERR
"BTRFS: failed to read chunk root on %s\n",
2872 if (!IS_ERR(chunk_root
->node
))
2873 free_extent_buffer(chunk_root
->node
);
2874 chunk_root
->node
= NULL
;
2875 goto fail_tree_roots
;
2877 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2878 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2880 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2881 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2883 ret
= btrfs_read_chunk_tree(chunk_root
);
2885 printk(KERN_ERR
"BTRFS: failed to read chunk tree on %s\n",
2887 goto fail_tree_roots
;
2891 * keep the device that is marked to be the target device for the
2892 * dev_replace procedure
2894 btrfs_close_extra_devices(fs_devices
, 0);
2896 if (!fs_devices
->latest_bdev
) {
2897 printk(KERN_ERR
"BTRFS: failed to read devices on %s\n",
2899 goto fail_tree_roots
;
2903 generation
= btrfs_super_generation(disk_super
);
2905 tree_root
->node
= read_tree_block(tree_root
,
2906 btrfs_super_root(disk_super
),
2908 if (IS_ERR(tree_root
->node
) ||
2909 !extent_buffer_uptodate(tree_root
->node
)) {
2910 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2912 if (!IS_ERR(tree_root
->node
))
2913 free_extent_buffer(tree_root
->node
);
2914 tree_root
->node
= NULL
;
2915 goto recovery_tree_root
;
2918 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2919 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2920 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2922 mutex_lock(&tree_root
->objectid_mutex
);
2923 ret
= btrfs_find_highest_objectid(tree_root
,
2924 &tree_root
->highest_objectid
);
2926 mutex_unlock(&tree_root
->objectid_mutex
);
2927 goto recovery_tree_root
;
2930 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2932 mutex_unlock(&tree_root
->objectid_mutex
);
2934 ret
= btrfs_read_roots(fs_info
, tree_root
);
2936 goto recovery_tree_root
;
2938 fs_info
->generation
= generation
;
2939 fs_info
->last_trans_committed
= generation
;
2941 ret
= btrfs_recover_balance(fs_info
);
2943 printk(KERN_ERR
"BTRFS: failed to recover balance\n");
2944 goto fail_block_groups
;
2947 ret
= btrfs_init_dev_stats(fs_info
);
2949 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2951 goto fail_block_groups
;
2954 ret
= btrfs_init_dev_replace(fs_info
);
2956 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2957 goto fail_block_groups
;
2960 btrfs_close_extra_devices(fs_devices
, 1);
2962 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2964 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret
);
2965 goto fail_block_groups
;
2968 ret
= btrfs_sysfs_add_device(fs_devices
);
2970 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret
);
2971 goto fail_fsdev_sysfs
;
2974 ret
= btrfs_sysfs_add_mounted(fs_info
);
2976 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2977 goto fail_fsdev_sysfs
;
2980 ret
= btrfs_init_space_info(fs_info
);
2982 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2986 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2988 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
2991 fs_info
->num_tolerated_disk_barrier_failures
=
2992 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2993 if (fs_info
->fs_devices
->missing_devices
>
2994 fs_info
->num_tolerated_disk_barrier_failures
&&
2995 !(sb
->s_flags
& MS_RDONLY
)) {
2996 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2997 fs_info
->fs_devices
->missing_devices
,
2998 fs_info
->num_tolerated_disk_barrier_failures
);
3002 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3004 if (IS_ERR(fs_info
->cleaner_kthread
))
3007 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3009 "btrfs-transaction");
3010 if (IS_ERR(fs_info
->transaction_kthread
))
3013 if (!btrfs_test_opt(tree_root
, SSD
) &&
3014 !btrfs_test_opt(tree_root
, NOSSD
) &&
3015 !fs_info
->fs_devices
->rotating
) {
3016 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
3018 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
3022 * Mount does not set all options immediatelly, we can do it now and do
3023 * not have to wait for transaction commit
3025 btrfs_apply_pending_changes(fs_info
);
3027 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3028 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
3029 ret
= btrfsic_mount(tree_root
, fs_devices
,
3030 btrfs_test_opt(tree_root
,
3031 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3033 fs_info
->check_integrity_print_mask
);
3035 printk(KERN_WARNING
"BTRFS: failed to initialize"
3036 " integrity check module %s\n", sb
->s_id
);
3039 ret
= btrfs_read_qgroup_config(fs_info
);
3041 goto fail_trans_kthread
;
3043 /* do not make disk changes in broken FS or nologreplay is given */
3044 if (btrfs_super_log_root(disk_super
) != 0 &&
3045 !btrfs_test_opt(tree_root
, NOLOGREPLAY
)) {
3046 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3053 ret
= btrfs_find_orphan_roots(tree_root
);
3057 if (!(sb
->s_flags
& MS_RDONLY
)) {
3058 ret
= btrfs_cleanup_fs_roots(fs_info
);
3062 mutex_lock(&fs_info
->cleaner_mutex
);
3063 ret
= btrfs_recover_relocation(tree_root
);
3064 mutex_unlock(&fs_info
->cleaner_mutex
);
3067 "BTRFS: failed to recover relocation\n");
3073 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3074 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3075 location
.offset
= 0;
3077 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3078 if (IS_ERR(fs_info
->fs_root
)) {
3079 err
= PTR_ERR(fs_info
->fs_root
);
3083 if (sb
->s_flags
& MS_RDONLY
)
3086 if (btrfs_test_opt(tree_root
, FREE_SPACE_TREE
) &&
3087 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3088 pr_info("BTRFS: creating free space tree\n");
3089 ret
= btrfs_create_free_space_tree(fs_info
);
3091 pr_warn("BTRFS: failed to create free space tree %d\n",
3093 close_ctree(tree_root
);
3098 down_read(&fs_info
->cleanup_work_sem
);
3099 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3100 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3101 up_read(&fs_info
->cleanup_work_sem
);
3102 close_ctree(tree_root
);
3105 up_read(&fs_info
->cleanup_work_sem
);
3107 ret
= btrfs_resume_balance_async(fs_info
);
3109 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
3110 close_ctree(tree_root
);
3114 ret
= btrfs_resume_dev_replace_async(fs_info
);
3116 pr_warn("BTRFS: failed to resume dev_replace\n");
3117 close_ctree(tree_root
);
3121 btrfs_qgroup_rescan_resume(fs_info
);
3123 if (btrfs_test_opt(tree_root
, CLEAR_CACHE
) &&
3124 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3125 pr_info("BTRFS: clearing free space tree\n");
3126 ret
= btrfs_clear_free_space_tree(fs_info
);
3128 pr_warn("BTRFS: failed to clear free space tree %d\n",
3130 close_ctree(tree_root
);
3135 if (!fs_info
->uuid_root
) {
3136 pr_info("BTRFS: creating UUID tree\n");
3137 ret
= btrfs_create_uuid_tree(fs_info
);
3139 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3141 close_ctree(tree_root
);
3144 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3145 fs_info
->generation
!=
3146 btrfs_super_uuid_tree_generation(disk_super
)) {
3147 pr_info("BTRFS: checking UUID tree\n");
3148 ret
= btrfs_check_uuid_tree(fs_info
);
3150 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3152 close_ctree(tree_root
);
3156 fs_info
->update_uuid_tree_gen
= 1;
3162 * backuproot only affect mount behavior, and if open_ctree succeeded,
3163 * no need to keep the flag
3165 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3170 btrfs_free_qgroup_config(fs_info
);
3172 kthread_stop(fs_info
->transaction_kthread
);
3173 btrfs_cleanup_transaction(fs_info
->tree_root
);
3174 btrfs_free_fs_roots(fs_info
);
3176 kthread_stop(fs_info
->cleaner_kthread
);
3179 * make sure we're done with the btree inode before we stop our
3182 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3185 btrfs_sysfs_remove_mounted(fs_info
);
3188 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3191 btrfs_put_block_group_cache(fs_info
);
3192 btrfs_free_block_groups(fs_info
);
3195 free_root_pointers(fs_info
, 1);
3196 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3199 btrfs_stop_all_workers(fs_info
);
3202 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3204 iput(fs_info
->btree_inode
);
3206 percpu_counter_destroy(&fs_info
->bio_counter
);
3207 fail_delalloc_bytes
:
3208 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3209 fail_dirty_metadata_bytes
:
3210 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3212 bdi_destroy(&fs_info
->bdi
);
3214 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3216 btrfs_free_stripe_hash_table(fs_info
);
3217 btrfs_close_devices(fs_info
->fs_devices
);
3221 if (!btrfs_test_opt(tree_root
, USEBACKUPROOT
))
3222 goto fail_tree_roots
;
3224 free_root_pointers(fs_info
, 0);
3226 /* don't use the log in recovery mode, it won't be valid */
3227 btrfs_set_super_log_root(disk_super
, 0);
3229 /* we can't trust the free space cache either */
3230 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3232 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3233 &num_backups_tried
, &backup_index
);
3235 goto fail_block_groups
;
3236 goto retry_root_backup
;
3239 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3242 set_buffer_uptodate(bh
);
3244 struct btrfs_device
*device
= (struct btrfs_device
*)
3247 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3248 "lost page write due to IO error on %s",
3249 rcu_str_deref(device
->name
));
3250 /* note, we dont' set_buffer_write_io_error because we have
3251 * our own ways of dealing with the IO errors
3253 clear_buffer_uptodate(bh
);
3254 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3260 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3261 struct buffer_head
**bh_ret
)
3263 struct buffer_head
*bh
;
3264 struct btrfs_super_block
*super
;
3267 bytenr
= btrfs_sb_offset(copy_num
);
3268 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3271 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3273 * If we fail to read from the underlying devices, as of now
3274 * the best option we have is to mark it EIO.
3279 super
= (struct btrfs_super_block
*)bh
->b_data
;
3280 if (btrfs_super_bytenr(super
) != bytenr
||
3281 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3291 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3293 struct buffer_head
*bh
;
3294 struct buffer_head
*latest
= NULL
;
3295 struct btrfs_super_block
*super
;
3300 /* we would like to check all the supers, but that would make
3301 * a btrfs mount succeed after a mkfs from a different FS.
3302 * So, we need to add a special mount option to scan for
3303 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3305 for (i
= 0; i
< 1; i
++) {
3306 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3310 super
= (struct btrfs_super_block
*)bh
->b_data
;
3312 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3315 transid
= btrfs_super_generation(super
);
3322 return ERR_PTR(ret
);
3328 * this should be called twice, once with wait == 0 and
3329 * once with wait == 1. When wait == 0 is done, all the buffer heads
3330 * we write are pinned.
3332 * They are released when wait == 1 is done.
3333 * max_mirrors must be the same for both runs, and it indicates how
3334 * many supers on this one device should be written.
3336 * max_mirrors == 0 means to write them all.
3338 static int write_dev_supers(struct btrfs_device
*device
,
3339 struct btrfs_super_block
*sb
,
3340 int do_barriers
, int wait
, int max_mirrors
)
3342 struct buffer_head
*bh
;
3349 if (max_mirrors
== 0)
3350 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3352 for (i
= 0; i
< max_mirrors
; i
++) {
3353 bytenr
= btrfs_sb_offset(i
);
3354 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3355 device
->commit_total_bytes
)
3359 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3360 BTRFS_SUPER_INFO_SIZE
);
3366 if (!buffer_uptodate(bh
))
3369 /* drop our reference */
3372 /* drop the reference from the wait == 0 run */
3376 btrfs_set_super_bytenr(sb
, bytenr
);
3379 crc
= btrfs_csum_data((char *)sb
+
3380 BTRFS_CSUM_SIZE
, crc
,
3381 BTRFS_SUPER_INFO_SIZE
-
3383 btrfs_csum_final(crc
, sb
->csum
);
3386 * one reference for us, and we leave it for the
3389 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3390 BTRFS_SUPER_INFO_SIZE
);
3392 btrfs_err(device
->dev_root
->fs_info
,
3393 "couldn't get super buffer head for bytenr %llu",
3399 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3401 /* one reference for submit_bh */
3404 set_buffer_uptodate(bh
);
3406 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3407 bh
->b_private
= device
;
3411 * we fua the first super. The others we allow
3415 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3417 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3421 return errors
< i
? 0 : -1;
3425 * endio for the write_dev_flush, this will wake anyone waiting
3426 * for the barrier when it is done
3428 static void btrfs_end_empty_barrier(struct bio
*bio
)
3430 if (bio
->bi_private
)
3431 complete(bio
->bi_private
);
3436 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3437 * sent down. With wait == 1, it waits for the previous flush.
3439 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3442 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3447 if (device
->nobarriers
)
3451 bio
= device
->flush_bio
;
3455 wait_for_completion(&device
->flush_wait
);
3457 if (bio
->bi_error
) {
3458 ret
= bio
->bi_error
;
3459 btrfs_dev_stat_inc_and_print(device
,
3460 BTRFS_DEV_STAT_FLUSH_ERRS
);
3463 /* drop the reference from the wait == 0 run */
3465 device
->flush_bio
= NULL
;
3471 * one reference for us, and we leave it for the
3474 device
->flush_bio
= NULL
;
3475 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3479 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3480 bio
->bi_bdev
= device
->bdev
;
3481 init_completion(&device
->flush_wait
);
3482 bio
->bi_private
= &device
->flush_wait
;
3483 device
->flush_bio
= bio
;
3486 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3492 * send an empty flush down to each device in parallel,
3493 * then wait for them
3495 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3497 struct list_head
*head
;
3498 struct btrfs_device
*dev
;
3499 int errors_send
= 0;
3500 int errors_wait
= 0;
3503 /* send down all the barriers */
3504 head
= &info
->fs_devices
->devices
;
3505 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3512 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3515 ret
= write_dev_flush(dev
, 0);
3520 /* wait for all the barriers */
3521 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3528 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3531 ret
= write_dev_flush(dev
, 1);
3535 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3536 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3541 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3544 int min_tolerated
= INT_MAX
;
3546 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3547 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3548 min_tolerated
= min(min_tolerated
,
3549 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3550 tolerated_failures
);
3552 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3553 if (raid_type
== BTRFS_RAID_SINGLE
)
3555 if (!(flags
& btrfs_raid_group
[raid_type
]))
3557 min_tolerated
= min(min_tolerated
,
3558 btrfs_raid_array
[raid_type
].
3559 tolerated_failures
);
3562 if (min_tolerated
== INT_MAX
) {
3563 pr_warn("BTRFS: unknown raid flag: %llu\n", flags
);
3567 return min_tolerated
;
3570 int btrfs_calc_num_tolerated_disk_barrier_failures(
3571 struct btrfs_fs_info
*fs_info
)
3573 struct btrfs_ioctl_space_info space
;
3574 struct btrfs_space_info
*sinfo
;
3575 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3576 BTRFS_BLOCK_GROUP_SYSTEM
,
3577 BTRFS_BLOCK_GROUP_METADATA
,
3578 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3581 int num_tolerated_disk_barrier_failures
=
3582 (int)fs_info
->fs_devices
->num_devices
;
3584 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3585 struct btrfs_space_info
*tmp
;
3589 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3590 if (tmp
->flags
== types
[i
]) {
3600 down_read(&sinfo
->groups_sem
);
3601 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3604 if (list_empty(&sinfo
->block_groups
[c
]))
3607 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3609 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3611 flags
= space
.flags
;
3613 num_tolerated_disk_barrier_failures
= min(
3614 num_tolerated_disk_barrier_failures
,
3615 btrfs_get_num_tolerated_disk_barrier_failures(
3618 up_read(&sinfo
->groups_sem
);
3621 return num_tolerated_disk_barrier_failures
;
3624 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3626 struct list_head
*head
;
3627 struct btrfs_device
*dev
;
3628 struct btrfs_super_block
*sb
;
3629 struct btrfs_dev_item
*dev_item
;
3633 int total_errors
= 0;
3636 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3637 backup_super_roots(root
->fs_info
);
3639 sb
= root
->fs_info
->super_for_commit
;
3640 dev_item
= &sb
->dev_item
;
3642 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3643 head
= &root
->fs_info
->fs_devices
->devices
;
3644 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3647 ret
= barrier_all_devices(root
->fs_info
);
3650 &root
->fs_info
->fs_devices
->device_list_mutex
);
3651 btrfs_std_error(root
->fs_info
, ret
,
3652 "errors while submitting device barriers.");
3657 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3662 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3665 btrfs_set_stack_device_generation(dev_item
, 0);
3666 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3667 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3668 btrfs_set_stack_device_total_bytes(dev_item
,
3669 dev
->commit_total_bytes
);
3670 btrfs_set_stack_device_bytes_used(dev_item
,
3671 dev
->commit_bytes_used
);
3672 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3673 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3674 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3675 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3676 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3678 flags
= btrfs_super_flags(sb
);
3679 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3681 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3685 if (total_errors
> max_errors
) {
3686 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3688 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3690 /* FUA is masked off if unsupported and can't be the reason */
3691 btrfs_std_error(root
->fs_info
, -EIO
,
3692 "%d errors while writing supers", total_errors
);
3697 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3700 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3703 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3707 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3708 if (total_errors
> max_errors
) {
3709 btrfs_std_error(root
->fs_info
, -EIO
,
3710 "%d errors while writing supers", total_errors
);
3716 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3717 struct btrfs_root
*root
, int max_mirrors
)
3719 return write_all_supers(root
, max_mirrors
);
3722 /* Drop a fs root from the radix tree and free it. */
3723 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3724 struct btrfs_root
*root
)
3726 spin_lock(&fs_info
->fs_roots_radix_lock
);
3727 radix_tree_delete(&fs_info
->fs_roots_radix
,
3728 (unsigned long)root
->root_key
.objectid
);
3729 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3731 if (btrfs_root_refs(&root
->root_item
) == 0)
3732 synchronize_srcu(&fs_info
->subvol_srcu
);
3734 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3735 btrfs_free_log(NULL
, root
);
3737 if (root
->free_ino_pinned
)
3738 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3739 if (root
->free_ino_ctl
)
3740 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3744 static void free_fs_root(struct btrfs_root
*root
)
3746 iput(root
->ino_cache_inode
);
3747 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3748 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3749 root
->orphan_block_rsv
= NULL
;
3751 free_anon_bdev(root
->anon_dev
);
3752 if (root
->subv_writers
)
3753 btrfs_free_subvolume_writers(root
->subv_writers
);
3754 free_extent_buffer(root
->node
);
3755 free_extent_buffer(root
->commit_root
);
3756 kfree(root
->free_ino_ctl
);
3757 kfree(root
->free_ino_pinned
);
3759 btrfs_put_fs_root(root
);
3762 void btrfs_free_fs_root(struct btrfs_root
*root
)
3767 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3769 u64 root_objectid
= 0;
3770 struct btrfs_root
*gang
[8];
3773 unsigned int ret
= 0;
3777 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3778 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3779 (void **)gang
, root_objectid
,
3782 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3785 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3787 for (i
= 0; i
< ret
; i
++) {
3788 /* Avoid to grab roots in dead_roots */
3789 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3793 /* grab all the search result for later use */
3794 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3796 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3798 for (i
= 0; i
< ret
; i
++) {
3801 root_objectid
= gang
[i
]->root_key
.objectid
;
3802 err
= btrfs_orphan_cleanup(gang
[i
]);
3805 btrfs_put_fs_root(gang
[i
]);
3810 /* release the uncleaned roots due to error */
3811 for (; i
< ret
; i
++) {
3813 btrfs_put_fs_root(gang
[i
]);
3818 int btrfs_commit_super(struct btrfs_root
*root
)
3820 struct btrfs_trans_handle
*trans
;
3822 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3823 btrfs_run_delayed_iputs(root
);
3824 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3825 wake_up_process(root
->fs_info
->cleaner_kthread
);
3827 /* wait until ongoing cleanup work done */
3828 down_write(&root
->fs_info
->cleanup_work_sem
);
3829 up_write(&root
->fs_info
->cleanup_work_sem
);
3831 trans
= btrfs_join_transaction(root
);
3833 return PTR_ERR(trans
);
3834 return btrfs_commit_transaction(trans
, root
);
3837 void close_ctree(struct btrfs_root
*root
)
3839 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3842 fs_info
->closing
= 1;
3845 /* wait for the qgroup rescan worker to stop */
3846 btrfs_qgroup_wait_for_completion(fs_info
);
3848 /* wait for the uuid_scan task to finish */
3849 down(&fs_info
->uuid_tree_rescan_sem
);
3850 /* avoid complains from lockdep et al., set sem back to initial state */
3851 up(&fs_info
->uuid_tree_rescan_sem
);
3853 /* pause restriper - we want to resume on mount */
3854 btrfs_pause_balance(fs_info
);
3856 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3858 btrfs_scrub_cancel(fs_info
);
3860 /* wait for any defraggers to finish */
3861 wait_event(fs_info
->transaction_wait
,
3862 (atomic_read(&fs_info
->defrag_running
) == 0));
3864 /* clear out the rbtree of defraggable inodes */
3865 btrfs_cleanup_defrag_inodes(fs_info
);
3867 cancel_work_sync(&fs_info
->async_reclaim_work
);
3869 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3871 * If the cleaner thread is stopped and there are
3872 * block groups queued for removal, the deletion will be
3873 * skipped when we quit the cleaner thread.
3875 btrfs_delete_unused_bgs(root
->fs_info
);
3877 ret
= btrfs_commit_super(root
);
3879 btrfs_err(fs_info
, "commit super ret %d", ret
);
3882 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3883 btrfs_error_commit_super(root
);
3885 kthread_stop(fs_info
->transaction_kthread
);
3886 kthread_stop(fs_info
->cleaner_kthread
);
3888 fs_info
->closing
= 2;
3891 btrfs_free_qgroup_config(fs_info
);
3893 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3894 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3895 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3898 btrfs_sysfs_remove_mounted(fs_info
);
3899 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3901 btrfs_free_fs_roots(fs_info
);
3903 btrfs_put_block_group_cache(fs_info
);
3905 btrfs_free_block_groups(fs_info
);
3908 * we must make sure there is not any read request to
3909 * submit after we stopping all workers.
3911 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3912 btrfs_stop_all_workers(fs_info
);
3915 free_root_pointers(fs_info
, 1);
3917 iput(fs_info
->btree_inode
);
3919 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3920 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3921 btrfsic_unmount(root
, fs_info
->fs_devices
);
3924 btrfs_close_devices(fs_info
->fs_devices
);
3925 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3927 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3928 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3929 percpu_counter_destroy(&fs_info
->bio_counter
);
3930 bdi_destroy(&fs_info
->bdi
);
3931 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3933 btrfs_free_stripe_hash_table(fs_info
);
3935 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3936 root
->orphan_block_rsv
= NULL
;
3939 while (!list_empty(&fs_info
->pinned_chunks
)) {
3940 struct extent_map
*em
;
3942 em
= list_first_entry(&fs_info
->pinned_chunks
,
3943 struct extent_map
, list
);
3944 list_del_init(&em
->list
);
3945 free_extent_map(em
);
3947 unlock_chunks(root
);
3950 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3954 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3956 ret
= extent_buffer_uptodate(buf
);
3960 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3961 parent_transid
, atomic
);
3967 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3969 struct btrfs_root
*root
;
3970 u64 transid
= btrfs_header_generation(buf
);
3973 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3975 * This is a fast path so only do this check if we have sanity tests
3976 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3977 * outside of the sanity tests.
3979 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3982 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3983 btrfs_assert_tree_locked(buf
);
3984 if (transid
!= root
->fs_info
->generation
)
3985 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3986 "found %llu running %llu\n",
3987 buf
->start
, transid
, root
->fs_info
->generation
);
3988 was_dirty
= set_extent_buffer_dirty(buf
);
3990 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3992 root
->fs_info
->dirty_metadata_batch
);
3993 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3994 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3995 btrfs_print_leaf(root
, buf
);
4001 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
4005 * looks as though older kernels can get into trouble with
4006 * this code, they end up stuck in balance_dirty_pages forever
4010 if (current
->flags
& PF_MEMALLOC
)
4014 btrfs_balance_delayed_items(root
);
4016 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
4017 BTRFS_DIRTY_METADATA_THRESH
);
4019 balance_dirty_pages_ratelimited(
4020 root
->fs_info
->btree_inode
->i_mapping
);
4024 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
4026 __btrfs_btree_balance_dirty(root
, 1);
4029 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
4031 __btrfs_btree_balance_dirty(root
, 0);
4034 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
4036 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4037 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
4040 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
4043 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
4044 u64 nodesize
= btrfs_super_nodesize(sb
);
4045 u64 sectorsize
= btrfs_super_sectorsize(sb
);
4048 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
4049 printk(KERN_ERR
"BTRFS: no valid FS found\n");
4052 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
)
4053 printk(KERN_WARNING
"BTRFS: unrecognized super flag: %llu\n",
4054 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
4055 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4056 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
4057 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
4060 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4061 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
4062 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
4065 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4066 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
4067 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
4072 * Check sectorsize and nodesize first, other check will need it.
4073 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4075 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
4076 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
4077 printk(KERN_ERR
"BTRFS: invalid sectorsize %llu\n", sectorsize
);
4080 /* Only PAGE SIZE is supported yet */
4081 if (sectorsize
!= PAGE_CACHE_SIZE
) {
4082 printk(KERN_ERR
"BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4083 sectorsize
, PAGE_CACHE_SIZE
);
4086 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
4087 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
4088 printk(KERN_ERR
"BTRFS: invalid nodesize %llu\n", nodesize
);
4091 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
4092 printk(KERN_ERR
"BTRFS: invalid leafsize %u, should be %llu\n",
4093 le32_to_cpu(sb
->__unused_leafsize
),
4098 /* Root alignment check */
4099 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
4100 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
4101 btrfs_super_root(sb
));
4104 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
4105 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
4106 btrfs_super_chunk_root(sb
));
4109 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
4110 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
4111 btrfs_super_log_root(sb
));
4115 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4116 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4117 fs_info
->fsid
, sb
->dev_item
.fsid
);
4122 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4125 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4126 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
4127 btrfs_super_num_devices(sb
));
4128 if (btrfs_super_num_devices(sb
) == 0) {
4129 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
4133 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4134 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
4135 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4140 * Obvious sys_chunk_array corruptions, it must hold at least one key
4143 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4144 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
4145 btrfs_super_sys_array_size(sb
),
4146 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4149 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4150 + sizeof(struct btrfs_chunk
)) {
4151 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
4152 btrfs_super_sys_array_size(sb
),
4153 sizeof(struct btrfs_disk_key
)
4154 + sizeof(struct btrfs_chunk
));
4159 * The generation is a global counter, we'll trust it more than the others
4160 * but it's still possible that it's the one that's wrong.
4162 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4164 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4165 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
4166 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4167 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4169 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4170 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
4175 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4177 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4178 btrfs_run_delayed_iputs(root
);
4179 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4181 down_write(&root
->fs_info
->cleanup_work_sem
);
4182 up_write(&root
->fs_info
->cleanup_work_sem
);
4184 /* cleanup FS via transaction */
4185 btrfs_cleanup_transaction(root
);
4188 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4190 struct btrfs_ordered_extent
*ordered
;
4192 spin_lock(&root
->ordered_extent_lock
);
4194 * This will just short circuit the ordered completion stuff which will
4195 * make sure the ordered extent gets properly cleaned up.
4197 list_for_each_entry(ordered
, &root
->ordered_extents
,
4199 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4200 spin_unlock(&root
->ordered_extent_lock
);
4203 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4205 struct btrfs_root
*root
;
4206 struct list_head splice
;
4208 INIT_LIST_HEAD(&splice
);
4210 spin_lock(&fs_info
->ordered_root_lock
);
4211 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4212 while (!list_empty(&splice
)) {
4213 root
= list_first_entry(&splice
, struct btrfs_root
,
4215 list_move_tail(&root
->ordered_root
,
4216 &fs_info
->ordered_roots
);
4218 spin_unlock(&fs_info
->ordered_root_lock
);
4219 btrfs_destroy_ordered_extents(root
);
4222 spin_lock(&fs_info
->ordered_root_lock
);
4224 spin_unlock(&fs_info
->ordered_root_lock
);
4227 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4228 struct btrfs_root
*root
)
4230 struct rb_node
*node
;
4231 struct btrfs_delayed_ref_root
*delayed_refs
;
4232 struct btrfs_delayed_ref_node
*ref
;
4235 delayed_refs
= &trans
->delayed_refs
;
4237 spin_lock(&delayed_refs
->lock
);
4238 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4239 spin_unlock(&delayed_refs
->lock
);
4240 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4244 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4245 struct btrfs_delayed_ref_head
*head
;
4246 struct btrfs_delayed_ref_node
*tmp
;
4247 bool pin_bytes
= false;
4249 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4251 if (!mutex_trylock(&head
->mutex
)) {
4252 atomic_inc(&head
->node
.refs
);
4253 spin_unlock(&delayed_refs
->lock
);
4255 mutex_lock(&head
->mutex
);
4256 mutex_unlock(&head
->mutex
);
4257 btrfs_put_delayed_ref(&head
->node
);
4258 spin_lock(&delayed_refs
->lock
);
4261 spin_lock(&head
->lock
);
4262 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4265 list_del(&ref
->list
);
4266 atomic_dec(&delayed_refs
->num_entries
);
4267 btrfs_put_delayed_ref(ref
);
4269 if (head
->must_insert_reserved
)
4271 btrfs_free_delayed_extent_op(head
->extent_op
);
4272 delayed_refs
->num_heads
--;
4273 if (head
->processing
== 0)
4274 delayed_refs
->num_heads_ready
--;
4275 atomic_dec(&delayed_refs
->num_entries
);
4276 head
->node
.in_tree
= 0;
4277 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4278 spin_unlock(&head
->lock
);
4279 spin_unlock(&delayed_refs
->lock
);
4280 mutex_unlock(&head
->mutex
);
4283 btrfs_pin_extent(root
, head
->node
.bytenr
,
4284 head
->node
.num_bytes
, 1);
4285 btrfs_put_delayed_ref(&head
->node
);
4287 spin_lock(&delayed_refs
->lock
);
4290 spin_unlock(&delayed_refs
->lock
);
4295 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4297 struct btrfs_inode
*btrfs_inode
;
4298 struct list_head splice
;
4300 INIT_LIST_HEAD(&splice
);
4302 spin_lock(&root
->delalloc_lock
);
4303 list_splice_init(&root
->delalloc_inodes
, &splice
);
4305 while (!list_empty(&splice
)) {
4306 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4309 list_del_init(&btrfs_inode
->delalloc_inodes
);
4310 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4311 &btrfs_inode
->runtime_flags
);
4312 spin_unlock(&root
->delalloc_lock
);
4314 btrfs_invalidate_inodes(btrfs_inode
->root
);
4316 spin_lock(&root
->delalloc_lock
);
4319 spin_unlock(&root
->delalloc_lock
);
4322 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4324 struct btrfs_root
*root
;
4325 struct list_head splice
;
4327 INIT_LIST_HEAD(&splice
);
4329 spin_lock(&fs_info
->delalloc_root_lock
);
4330 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4331 while (!list_empty(&splice
)) {
4332 root
= list_first_entry(&splice
, struct btrfs_root
,
4334 list_del_init(&root
->delalloc_root
);
4335 root
= btrfs_grab_fs_root(root
);
4337 spin_unlock(&fs_info
->delalloc_root_lock
);
4339 btrfs_destroy_delalloc_inodes(root
);
4340 btrfs_put_fs_root(root
);
4342 spin_lock(&fs_info
->delalloc_root_lock
);
4344 spin_unlock(&fs_info
->delalloc_root_lock
);
4347 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4348 struct extent_io_tree
*dirty_pages
,
4352 struct extent_buffer
*eb
;
4357 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4362 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4363 while (start
<= end
) {
4364 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4365 start
+= root
->nodesize
;
4368 wait_on_extent_buffer_writeback(eb
);
4370 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4372 clear_extent_buffer_dirty(eb
);
4373 free_extent_buffer_stale(eb
);
4380 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4381 struct extent_io_tree
*pinned_extents
)
4383 struct extent_io_tree
*unpin
;
4389 unpin
= pinned_extents
;
4392 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4393 EXTENT_DIRTY
, NULL
);
4397 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4398 btrfs_error_unpin_extent_range(root
, start
, end
);
4403 if (unpin
== &root
->fs_info
->freed_extents
[0])
4404 unpin
= &root
->fs_info
->freed_extents
[1];
4406 unpin
= &root
->fs_info
->freed_extents
[0];
4414 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4415 struct btrfs_root
*root
)
4417 btrfs_destroy_delayed_refs(cur_trans
, root
);
4419 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4420 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4422 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4423 wake_up(&root
->fs_info
->transaction_wait
);
4425 btrfs_destroy_delayed_inodes(root
);
4426 btrfs_assert_delayed_root_empty(root
);
4428 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4430 btrfs_destroy_pinned_extent(root
,
4431 root
->fs_info
->pinned_extents
);
4433 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4434 wake_up(&cur_trans
->commit_wait
);
4437 memset(cur_trans, 0, sizeof(*cur_trans));
4438 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4442 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4444 struct btrfs_transaction
*t
;
4446 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4448 spin_lock(&root
->fs_info
->trans_lock
);
4449 while (!list_empty(&root
->fs_info
->trans_list
)) {
4450 t
= list_first_entry(&root
->fs_info
->trans_list
,
4451 struct btrfs_transaction
, list
);
4452 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4453 atomic_inc(&t
->use_count
);
4454 spin_unlock(&root
->fs_info
->trans_lock
);
4455 btrfs_wait_for_commit(root
, t
->transid
);
4456 btrfs_put_transaction(t
);
4457 spin_lock(&root
->fs_info
->trans_lock
);
4460 if (t
== root
->fs_info
->running_transaction
) {
4461 t
->state
= TRANS_STATE_COMMIT_DOING
;
4462 spin_unlock(&root
->fs_info
->trans_lock
);
4464 * We wait for 0 num_writers since we don't hold a trans
4465 * handle open currently for this transaction.
4467 wait_event(t
->writer_wait
,
4468 atomic_read(&t
->num_writers
) == 0);
4470 spin_unlock(&root
->fs_info
->trans_lock
);
4472 btrfs_cleanup_one_transaction(t
, root
);
4474 spin_lock(&root
->fs_info
->trans_lock
);
4475 if (t
== root
->fs_info
->running_transaction
)
4476 root
->fs_info
->running_transaction
= NULL
;
4477 list_del_init(&t
->list
);
4478 spin_unlock(&root
->fs_info
->trans_lock
);
4480 btrfs_put_transaction(t
);
4481 trace_btrfs_transaction_commit(root
);
4482 spin_lock(&root
->fs_info
->trans_lock
);
4484 spin_unlock(&root
->fs_info
->trans_lock
);
4485 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4486 btrfs_destroy_delayed_inodes(root
);
4487 btrfs_assert_delayed_root_empty(root
);
4488 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4489 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4490 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4495 static const struct extent_io_ops btree_extent_io_ops
= {
4496 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4497 .readpage_io_failed_hook
= btree_io_failed_hook
,
4498 .submit_bio_hook
= btree_submit_bio_hook
,
4499 /* note we're sharing with inode.c for the merge bio hook */
4500 .merge_bio_hook
= btrfs_merge_bio_hook
,