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 static const struct extent_io_ops btree_extent_io_ops
;
59 static void end_workqueue_fn(struct btrfs_work
*work
);
60 static void free_fs_root(struct btrfs_root
*root
);
61 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
63 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
65 struct btrfs_root
*root
);
66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
67 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
68 struct extent_io_tree
*dirty_pages
,
70 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
71 struct extent_io_tree
*pinned_extents
);
72 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
73 static void btrfs_error_commit_super(struct btrfs_root
*root
);
76 * btrfs_end_io_wq structs are used to do processing in task context when an IO
77 * is complete. This is used during reads to verify checksums, and it is used
78 * by writes to insert metadata for new file extents after IO is complete.
80 struct btrfs_end_io_wq
{
84 struct btrfs_fs_info
*info
;
86 enum btrfs_wq_endio_type metadata
;
87 struct list_head list
;
88 struct btrfs_work work
;
91 static struct kmem_cache
*btrfs_end_io_wq_cache
;
93 int __init
btrfs_end_io_wq_init(void)
95 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
96 sizeof(struct btrfs_end_io_wq
),
98 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
100 if (!btrfs_end_io_wq_cache
)
105 void btrfs_end_io_wq_exit(void)
107 if (btrfs_end_io_wq_cache
)
108 kmem_cache_destroy(btrfs_end_io_wq_cache
);
112 * async submit bios are used to offload expensive checksumming
113 * onto the worker threads. They checksum file and metadata bios
114 * just before they are sent down the IO stack.
116 struct async_submit_bio
{
119 struct list_head list
;
120 extent_submit_bio_hook_t
*submit_bio_start
;
121 extent_submit_bio_hook_t
*submit_bio_done
;
124 unsigned long bio_flags
;
126 * bio_offset is optional, can be used if the pages in the bio
127 * can't tell us where in the file the bio should go
130 struct btrfs_work work
;
135 * Lockdep class keys for extent_buffer->lock's in this root. For a given
136 * eb, the lockdep key is determined by the btrfs_root it belongs to and
137 * the level the eb occupies in the tree.
139 * Different roots are used for different purposes and may nest inside each
140 * other and they require separate keysets. As lockdep keys should be
141 * static, assign keysets according to the purpose of the root as indicated
142 * by btrfs_root->objectid. This ensures that all special purpose roots
143 * have separate keysets.
145 * Lock-nesting across peer nodes is always done with the immediate parent
146 * node locked thus preventing deadlock. As lockdep doesn't know this, use
147 * subclass to avoid triggering lockdep warning in such cases.
149 * The key is set by the readpage_end_io_hook after the buffer has passed
150 * csum validation but before the pages are unlocked. It is also set by
151 * btrfs_init_new_buffer on freshly allocated blocks.
153 * We also add a check to make sure the highest level of the tree is the
154 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
155 * needs update as well.
157 #ifdef CONFIG_DEBUG_LOCK_ALLOC
158 # if BTRFS_MAX_LEVEL != 8
162 static struct btrfs_lockdep_keyset
{
163 u64 id
; /* root objectid */
164 const char *name_stem
; /* lock name stem */
165 char names
[BTRFS_MAX_LEVEL
+ 1][20];
166 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
167 } btrfs_lockdep_keysets
[] = {
168 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
169 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
170 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
171 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
172 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
173 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
174 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
175 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
176 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
177 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
178 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
179 { .id
= 0, .name_stem
= "tree" },
182 void __init
btrfs_init_lockdep(void)
186 /* initialize lockdep class names */
187 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
188 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
190 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
191 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
192 "btrfs-%s-%02d", ks
->name_stem
, j
);
196 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
199 struct btrfs_lockdep_keyset
*ks
;
201 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
203 /* find the matching keyset, id 0 is the default entry */
204 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
205 if (ks
->id
== objectid
)
208 lockdep_set_class_and_name(&eb
->lock
,
209 &ks
->keys
[level
], ks
->names
[level
]);
215 * extents on the btree inode are pretty simple, there's one extent
216 * that covers the entire device
218 static struct extent_map
*btree_get_extent(struct inode
*inode
,
219 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
222 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
223 struct extent_map
*em
;
226 read_lock(&em_tree
->lock
);
227 em
= lookup_extent_mapping(em_tree
, start
, len
);
230 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
231 read_unlock(&em_tree
->lock
);
234 read_unlock(&em_tree
->lock
);
236 em
= alloc_extent_map();
238 em
= ERR_PTR(-ENOMEM
);
243 em
->block_len
= (u64
)-1;
245 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
247 write_lock(&em_tree
->lock
);
248 ret
= add_extent_mapping(em_tree
, em
, 0);
249 if (ret
== -EEXIST
) {
251 em
= lookup_extent_mapping(em_tree
, start
, len
);
258 write_unlock(&em_tree
->lock
);
264 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
266 return btrfs_crc32c(seed
, data
, len
);
269 void btrfs_csum_final(u32 crc
, char *result
)
271 put_unaligned_le32(~crc
, result
);
275 * compute the csum for a btree block, and either verify it or write it
276 * into the csum field of the block.
278 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
279 struct extent_buffer
*buf
,
282 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
285 unsigned long cur_len
;
286 unsigned long offset
= BTRFS_CSUM_SIZE
;
288 unsigned long map_start
;
289 unsigned long map_len
;
292 unsigned long inline_result
;
294 len
= buf
->len
- offset
;
296 err
= map_private_extent_buffer(buf
, offset
, 32,
297 &kaddr
, &map_start
, &map_len
);
300 cur_len
= min(len
, map_len
- (offset
- map_start
));
301 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
306 if (csum_size
> sizeof(inline_result
)) {
307 result
= kzalloc(csum_size
, GFP_NOFS
);
311 result
= (char *)&inline_result
;
314 btrfs_csum_final(crc
, result
);
317 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
320 memcpy(&found
, result
, csum_size
);
322 read_extent_buffer(buf
, &val
, 0, csum_size
);
323 btrfs_warn_rl(fs_info
,
324 "%s checksum verify failed on %llu wanted %X found %X "
326 fs_info
->sb
->s_id
, buf
->start
,
327 val
, found
, btrfs_header_level(buf
));
328 if (result
!= (char *)&inline_result
)
333 write_extent_buffer(buf
, result
, 0, csum_size
);
335 if (result
!= (char *)&inline_result
)
341 * we can't consider a given block up to date unless the transid of the
342 * block matches the transid in the parent node's pointer. This is how we
343 * detect blocks that either didn't get written at all or got written
344 * in the wrong place.
346 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
347 struct extent_buffer
*eb
, u64 parent_transid
,
350 struct extent_state
*cached_state
= NULL
;
352 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
354 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
361 btrfs_tree_read_lock(eb
);
362 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
365 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
367 if (extent_buffer_uptodate(eb
) &&
368 btrfs_header_generation(eb
) == parent_transid
) {
372 btrfs_err_rl(eb
->fs_info
,
373 "parent transid verify failed on %llu wanted %llu found %llu",
375 parent_transid
, btrfs_header_generation(eb
));
379 * Things reading via commit roots that don't have normal protection,
380 * like send, can have a really old block in cache that may point at a
381 * block that has been free'd and re-allocated. So don't clear uptodate
382 * if we find an eb that is under IO (dirty/writeback) because we could
383 * end up reading in the stale data and then writing it back out and
384 * making everybody very sad.
386 if (!extent_buffer_under_io(eb
))
387 clear_extent_buffer_uptodate(eb
);
389 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
390 &cached_state
, GFP_NOFS
);
392 btrfs_tree_read_unlock_blocking(eb
);
397 * Return 0 if the superblock checksum type matches the checksum value of that
398 * algorithm. Pass the raw disk superblock data.
400 static int btrfs_check_super_csum(char *raw_disk_sb
)
402 struct btrfs_super_block
*disk_sb
=
403 (struct btrfs_super_block
*)raw_disk_sb
;
404 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
407 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
409 const int csum_size
= sizeof(crc
);
410 char result
[csum_size
];
413 * The super_block structure does not span the whole
414 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
415 * is filled with zeros and is included in the checkum.
417 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
418 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
419 btrfs_csum_final(crc
, result
);
421 if (memcmp(raw_disk_sb
, result
, csum_size
))
425 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
426 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
435 * helper to read a given tree block, doing retries as required when
436 * the checksums don't match and we have alternate mirrors to try.
438 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
439 struct extent_buffer
*eb
,
440 u64 start
, u64 parent_transid
)
442 struct extent_io_tree
*io_tree
;
447 int failed_mirror
= 0;
449 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
450 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
452 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
454 btree_get_extent
, mirror_num
);
456 if (!verify_parent_transid(io_tree
, eb
,
464 * This buffer's crc is fine, but its contents are corrupted, so
465 * there is no reason to read the other copies, they won't be
468 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
471 num_copies
= btrfs_num_copies(root
->fs_info
,
476 if (!failed_mirror
) {
478 failed_mirror
= eb
->read_mirror
;
482 if (mirror_num
== failed_mirror
)
485 if (mirror_num
> num_copies
)
489 if (failed
&& !ret
&& failed_mirror
)
490 repair_eb_io_failure(root
, eb
, failed_mirror
);
496 * checksum a dirty tree block before IO. This has extra checks to make sure
497 * we only fill in the checksum field in the first page of a multi-page block
500 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
502 u64 start
= page_offset(page
);
504 struct extent_buffer
*eb
;
506 eb
= (struct extent_buffer
*)page
->private;
507 if (page
!= eb
->pages
[0])
509 found_start
= btrfs_header_bytenr(eb
);
510 if (WARN_ON(found_start
!= start
|| !PageUptodate(page
)))
512 csum_tree_block(fs_info
, eb
, 0);
516 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
517 struct extent_buffer
*eb
)
519 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
520 u8 fsid
[BTRFS_UUID_SIZE
];
523 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
525 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
529 fs_devices
= fs_devices
->seed
;
534 #define CORRUPT(reason, eb, root, slot) \
535 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
536 "root=%llu, slot=%d", reason, \
537 btrfs_header_bytenr(eb), root->objectid, slot)
539 static noinline
int check_leaf(struct btrfs_root
*root
,
540 struct extent_buffer
*leaf
)
542 struct btrfs_key key
;
543 struct btrfs_key leaf_key
;
544 u32 nritems
= btrfs_header_nritems(leaf
);
550 /* Check the 0 item */
551 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
552 BTRFS_LEAF_DATA_SIZE(root
)) {
553 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
558 * Check to make sure each items keys are in the correct order and their
559 * offsets make sense. We only have to loop through nritems-1 because
560 * we check the current slot against the next slot, which verifies the
561 * next slot's offset+size makes sense and that the current's slot
564 for (slot
= 0; slot
< nritems
- 1; slot
++) {
565 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
566 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
568 /* Make sure the keys are in the right order */
569 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
570 CORRUPT("bad key order", leaf
, root
, slot
);
575 * Make sure the offset and ends are right, remember that the
576 * item data starts at the end of the leaf and grows towards the
579 if (btrfs_item_offset_nr(leaf
, slot
) !=
580 btrfs_item_end_nr(leaf
, slot
+ 1)) {
581 CORRUPT("slot offset bad", leaf
, root
, slot
);
586 * Check to make sure that we don't point outside of the leaf,
587 * just incase all the items are consistent to eachother, but
588 * all point outside of the leaf.
590 if (btrfs_item_end_nr(leaf
, slot
) >
591 BTRFS_LEAF_DATA_SIZE(root
)) {
592 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
600 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
601 u64 phy_offset
, struct page
*page
,
602 u64 start
, u64 end
, int mirror
)
606 struct extent_buffer
*eb
;
607 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
614 eb
= (struct extent_buffer
*)page
->private;
616 /* the pending IO might have been the only thing that kept this buffer
617 * in memory. Make sure we have a ref for all this other checks
619 extent_buffer_get(eb
);
621 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
625 eb
->read_mirror
= mirror
;
626 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
631 found_start
= btrfs_header_bytenr(eb
);
632 if (found_start
!= eb
->start
) {
633 btrfs_err_rl(eb
->fs_info
, "bad tree block start %llu %llu",
634 found_start
, eb
->start
);
638 if (check_tree_block_fsid(root
->fs_info
, eb
)) {
639 btrfs_err_rl(eb
->fs_info
, "bad fsid on block %llu",
644 found_level
= btrfs_header_level(eb
);
645 if (found_level
>= BTRFS_MAX_LEVEL
) {
646 btrfs_err(root
->fs_info
, "bad tree block level %d",
647 (int)btrfs_header_level(eb
));
652 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
655 ret
= csum_tree_block(root
->fs_info
, eb
, 1);
662 * If this is a leaf block and it is corrupt, set the corrupt bit so
663 * that we don't try and read the other copies of this block, just
666 if (found_level
== 0 && check_leaf(root
, eb
)) {
667 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
672 set_extent_buffer_uptodate(eb
);
675 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
676 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
680 * our io error hook is going to dec the io pages
681 * again, we have to make sure it has something
684 atomic_inc(&eb
->io_pages
);
685 clear_extent_buffer_uptodate(eb
);
687 free_extent_buffer(eb
);
692 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
694 struct extent_buffer
*eb
;
695 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
697 eb
= (struct extent_buffer
*)page
->private;
698 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
699 eb
->read_mirror
= failed_mirror
;
700 atomic_dec(&eb
->io_pages
);
701 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
702 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
703 return -EIO
; /* we fixed nothing */
706 static void end_workqueue_bio(struct bio
*bio
)
708 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
709 struct btrfs_fs_info
*fs_info
;
710 struct btrfs_workqueue
*wq
;
711 btrfs_work_func_t func
;
713 fs_info
= end_io_wq
->info
;
714 end_io_wq
->error
= bio
->bi_error
;
716 if (bio
->bi_rw
& REQ_WRITE
) {
717 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
718 wq
= fs_info
->endio_meta_write_workers
;
719 func
= btrfs_endio_meta_write_helper
;
720 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
721 wq
= fs_info
->endio_freespace_worker
;
722 func
= btrfs_freespace_write_helper
;
723 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
724 wq
= fs_info
->endio_raid56_workers
;
725 func
= btrfs_endio_raid56_helper
;
727 wq
= fs_info
->endio_write_workers
;
728 func
= btrfs_endio_write_helper
;
731 if (unlikely(end_io_wq
->metadata
==
732 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
733 wq
= fs_info
->endio_repair_workers
;
734 func
= btrfs_endio_repair_helper
;
735 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
736 wq
= fs_info
->endio_raid56_workers
;
737 func
= btrfs_endio_raid56_helper
;
738 } else if (end_io_wq
->metadata
) {
739 wq
= fs_info
->endio_meta_workers
;
740 func
= btrfs_endio_meta_helper
;
742 wq
= fs_info
->endio_workers
;
743 func
= btrfs_endio_helper
;
747 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
748 btrfs_queue_work(wq
, &end_io_wq
->work
);
751 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
752 enum btrfs_wq_endio_type metadata
)
754 struct btrfs_end_io_wq
*end_io_wq
;
756 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
760 end_io_wq
->private = bio
->bi_private
;
761 end_io_wq
->end_io
= bio
->bi_end_io
;
762 end_io_wq
->info
= info
;
763 end_io_wq
->error
= 0;
764 end_io_wq
->bio
= bio
;
765 end_io_wq
->metadata
= metadata
;
767 bio
->bi_private
= end_io_wq
;
768 bio
->bi_end_io
= end_workqueue_bio
;
772 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
774 unsigned long limit
= min_t(unsigned long,
775 info
->thread_pool_size
,
776 info
->fs_devices
->open_devices
);
780 static void run_one_async_start(struct btrfs_work
*work
)
782 struct async_submit_bio
*async
;
785 async
= container_of(work
, struct async_submit_bio
, work
);
786 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
787 async
->mirror_num
, async
->bio_flags
,
793 static void run_one_async_done(struct btrfs_work
*work
)
795 struct btrfs_fs_info
*fs_info
;
796 struct async_submit_bio
*async
;
799 async
= container_of(work
, struct async_submit_bio
, work
);
800 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
802 limit
= btrfs_async_submit_limit(fs_info
);
803 limit
= limit
* 2 / 3;
806 * atomic_dec_return implies a barrier for waitqueue_active
808 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
809 waitqueue_active(&fs_info
->async_submit_wait
))
810 wake_up(&fs_info
->async_submit_wait
);
812 /* If an error occured we just want to clean up the bio and move on */
814 async
->bio
->bi_error
= async
->error
;
815 bio_endio(async
->bio
);
819 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
820 async
->mirror_num
, async
->bio_flags
,
824 static void run_one_async_free(struct btrfs_work
*work
)
826 struct async_submit_bio
*async
;
828 async
= container_of(work
, struct async_submit_bio
, work
);
832 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
833 int rw
, struct bio
*bio
, int mirror_num
,
834 unsigned long bio_flags
,
836 extent_submit_bio_hook_t
*submit_bio_start
,
837 extent_submit_bio_hook_t
*submit_bio_done
)
839 struct async_submit_bio
*async
;
841 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
845 async
->inode
= inode
;
848 async
->mirror_num
= mirror_num
;
849 async
->submit_bio_start
= submit_bio_start
;
850 async
->submit_bio_done
= submit_bio_done
;
852 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
853 run_one_async_done
, run_one_async_free
);
855 async
->bio_flags
= bio_flags
;
856 async
->bio_offset
= bio_offset
;
860 atomic_inc(&fs_info
->nr_async_submits
);
863 btrfs_set_work_high_priority(&async
->work
);
865 btrfs_queue_work(fs_info
->workers
, &async
->work
);
867 while (atomic_read(&fs_info
->async_submit_draining
) &&
868 atomic_read(&fs_info
->nr_async_submits
)) {
869 wait_event(fs_info
->async_submit_wait
,
870 (atomic_read(&fs_info
->nr_async_submits
) == 0));
876 static int btree_csum_one_bio(struct bio
*bio
)
878 struct bio_vec
*bvec
;
879 struct btrfs_root
*root
;
882 bio_for_each_segment_all(bvec
, bio
, i
) {
883 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
884 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
892 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
893 struct bio
*bio
, int mirror_num
,
894 unsigned long bio_flags
,
898 * when we're called for a write, we're already in the async
899 * submission context. Just jump into btrfs_map_bio
901 return btree_csum_one_bio(bio
);
904 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
905 int mirror_num
, unsigned long bio_flags
,
911 * when we're called for a write, we're already in the async
912 * submission context. Just jump into btrfs_map_bio
914 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
922 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
924 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
933 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
934 int mirror_num
, unsigned long bio_flags
,
937 int async
= check_async_write(inode
, bio_flags
);
940 if (!(rw
& REQ_WRITE
)) {
942 * called for a read, do the setup so that checksum validation
943 * can happen in the async kernel threads
945 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
946 bio
, BTRFS_WQ_ENDIO_METADATA
);
949 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
952 ret
= btree_csum_one_bio(bio
);
955 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
959 * kthread helpers are used to submit writes so that
960 * checksumming can happen in parallel across all CPUs
962 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
963 inode
, rw
, bio
, mirror_num
, 0,
965 __btree_submit_bio_start
,
966 __btree_submit_bio_done
);
979 #ifdef CONFIG_MIGRATION
980 static int btree_migratepage(struct address_space
*mapping
,
981 struct page
*newpage
, struct page
*page
,
982 enum migrate_mode mode
)
985 * we can't safely write a btree page from here,
986 * we haven't done the locking hook
991 * Buffers may be managed in a filesystem specific way.
992 * We must have no buffers or drop them.
994 if (page_has_private(page
) &&
995 !try_to_release_page(page
, GFP_KERNEL
))
997 return migrate_page(mapping
, newpage
, page
, mode
);
1002 static int btree_writepages(struct address_space
*mapping
,
1003 struct writeback_control
*wbc
)
1005 struct btrfs_fs_info
*fs_info
;
1008 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1010 if (wbc
->for_kupdate
)
1013 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1014 /* this is a bit racy, but that's ok */
1015 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1016 BTRFS_DIRTY_METADATA_THRESH
);
1020 return btree_write_cache_pages(mapping
, wbc
);
1023 static int btree_readpage(struct file
*file
, struct page
*page
)
1025 struct extent_io_tree
*tree
;
1026 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1027 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1030 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1032 if (PageWriteback(page
) || PageDirty(page
))
1035 return try_release_extent_buffer(page
);
1038 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1039 unsigned int length
)
1041 struct extent_io_tree
*tree
;
1042 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1043 extent_invalidatepage(tree
, page
, offset
);
1044 btree_releasepage(page
, GFP_NOFS
);
1045 if (PagePrivate(page
)) {
1046 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1047 "page private not zero on page %llu",
1048 (unsigned long long)page_offset(page
));
1049 ClearPagePrivate(page
);
1050 set_page_private(page
, 0);
1051 page_cache_release(page
);
1055 static int btree_set_page_dirty(struct page
*page
)
1058 struct extent_buffer
*eb
;
1060 BUG_ON(!PagePrivate(page
));
1061 eb
= (struct extent_buffer
*)page
->private;
1063 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1064 BUG_ON(!atomic_read(&eb
->refs
));
1065 btrfs_assert_tree_locked(eb
);
1067 return __set_page_dirty_nobuffers(page
);
1070 static const struct address_space_operations btree_aops
= {
1071 .readpage
= btree_readpage
,
1072 .writepages
= btree_writepages
,
1073 .releasepage
= btree_releasepage
,
1074 .invalidatepage
= btree_invalidatepage
,
1075 #ifdef CONFIG_MIGRATION
1076 .migratepage
= btree_migratepage
,
1078 .set_page_dirty
= btree_set_page_dirty
,
1081 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1083 struct extent_buffer
*buf
= NULL
;
1084 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1086 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1089 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1090 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1091 free_extent_buffer(buf
);
1094 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1095 int mirror_num
, struct extent_buffer
**eb
)
1097 struct extent_buffer
*buf
= NULL
;
1098 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1099 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1102 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1106 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1108 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1109 btree_get_extent
, mirror_num
);
1111 free_extent_buffer(buf
);
1115 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1116 free_extent_buffer(buf
);
1118 } else if (extent_buffer_uptodate(buf
)) {
1121 free_extent_buffer(buf
);
1126 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1129 return find_extent_buffer(fs_info
, bytenr
);
1132 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1135 if (btrfs_test_is_dummy_root(root
))
1136 return alloc_test_extent_buffer(root
->fs_info
, bytenr
);
1137 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1141 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1143 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1144 buf
->start
+ buf
->len
- 1);
1147 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1149 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1150 buf
->start
, buf
->start
+ buf
->len
- 1);
1153 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1156 struct extent_buffer
*buf
= NULL
;
1159 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1161 return ERR_PTR(-ENOMEM
);
1163 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1165 free_extent_buffer(buf
);
1166 return ERR_PTR(ret
);
1172 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1173 struct btrfs_fs_info
*fs_info
,
1174 struct extent_buffer
*buf
)
1176 if (btrfs_header_generation(buf
) ==
1177 fs_info
->running_transaction
->transid
) {
1178 btrfs_assert_tree_locked(buf
);
1180 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1181 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1183 fs_info
->dirty_metadata_batch
);
1184 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1185 btrfs_set_lock_blocking(buf
);
1186 clear_extent_buffer_dirty(buf
);
1191 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1193 struct btrfs_subvolume_writers
*writers
;
1196 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1198 return ERR_PTR(-ENOMEM
);
1200 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1203 return ERR_PTR(ret
);
1206 init_waitqueue_head(&writers
->wait
);
1211 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1213 percpu_counter_destroy(&writers
->counter
);
1217 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1218 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1222 root
->commit_root
= NULL
;
1223 root
->sectorsize
= sectorsize
;
1224 root
->nodesize
= nodesize
;
1225 root
->stripesize
= stripesize
;
1227 root
->orphan_cleanup_state
= 0;
1229 root
->objectid
= objectid
;
1230 root
->last_trans
= 0;
1231 root
->highest_objectid
= 0;
1232 root
->nr_delalloc_inodes
= 0;
1233 root
->nr_ordered_extents
= 0;
1235 root
->inode_tree
= RB_ROOT
;
1236 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1237 root
->block_rsv
= NULL
;
1238 root
->orphan_block_rsv
= NULL
;
1240 INIT_LIST_HEAD(&root
->dirty_list
);
1241 INIT_LIST_HEAD(&root
->root_list
);
1242 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1243 INIT_LIST_HEAD(&root
->delalloc_root
);
1244 INIT_LIST_HEAD(&root
->ordered_extents
);
1245 INIT_LIST_HEAD(&root
->ordered_root
);
1246 INIT_LIST_HEAD(&root
->logged_list
[0]);
1247 INIT_LIST_HEAD(&root
->logged_list
[1]);
1248 spin_lock_init(&root
->orphan_lock
);
1249 spin_lock_init(&root
->inode_lock
);
1250 spin_lock_init(&root
->delalloc_lock
);
1251 spin_lock_init(&root
->ordered_extent_lock
);
1252 spin_lock_init(&root
->accounting_lock
);
1253 spin_lock_init(&root
->log_extents_lock
[0]);
1254 spin_lock_init(&root
->log_extents_lock
[1]);
1255 mutex_init(&root
->objectid_mutex
);
1256 mutex_init(&root
->log_mutex
);
1257 mutex_init(&root
->ordered_extent_mutex
);
1258 mutex_init(&root
->delalloc_mutex
);
1259 init_waitqueue_head(&root
->log_writer_wait
);
1260 init_waitqueue_head(&root
->log_commit_wait
[0]);
1261 init_waitqueue_head(&root
->log_commit_wait
[1]);
1262 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1263 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1264 atomic_set(&root
->log_commit
[0], 0);
1265 atomic_set(&root
->log_commit
[1], 0);
1266 atomic_set(&root
->log_writers
, 0);
1267 atomic_set(&root
->log_batch
, 0);
1268 atomic_set(&root
->orphan_inodes
, 0);
1269 atomic_set(&root
->refs
, 1);
1270 atomic_set(&root
->will_be_snapshoted
, 0);
1271 atomic_set(&root
->qgroup_meta_rsv
, 0);
1272 root
->log_transid
= 0;
1273 root
->log_transid_committed
= -1;
1274 root
->last_log_commit
= 0;
1276 extent_io_tree_init(&root
->dirty_log_pages
,
1277 fs_info
->btree_inode
->i_mapping
);
1279 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1280 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1281 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1283 root
->defrag_trans_start
= fs_info
->generation
;
1285 root
->defrag_trans_start
= 0;
1286 root
->root_key
.objectid
= objectid
;
1289 spin_lock_init(&root
->root_item_lock
);
1292 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1294 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1296 root
->fs_info
= fs_info
;
1300 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1301 /* Should only be used by the testing infrastructure */
1302 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1304 struct btrfs_root
*root
;
1306 root
= btrfs_alloc_root(NULL
);
1308 return ERR_PTR(-ENOMEM
);
1309 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1310 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1311 root
->alloc_bytenr
= 0;
1317 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1318 struct btrfs_fs_info
*fs_info
,
1321 struct extent_buffer
*leaf
;
1322 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1323 struct btrfs_root
*root
;
1324 struct btrfs_key key
;
1328 root
= btrfs_alloc_root(fs_info
);
1330 return ERR_PTR(-ENOMEM
);
1332 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1333 tree_root
->stripesize
, root
, fs_info
, objectid
);
1334 root
->root_key
.objectid
= objectid
;
1335 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1336 root
->root_key
.offset
= 0;
1338 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1340 ret
= PTR_ERR(leaf
);
1345 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1346 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1347 btrfs_set_header_generation(leaf
, trans
->transid
);
1348 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1349 btrfs_set_header_owner(leaf
, objectid
);
1352 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1354 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1355 btrfs_header_chunk_tree_uuid(leaf
),
1357 btrfs_mark_buffer_dirty(leaf
);
1359 root
->commit_root
= btrfs_root_node(root
);
1360 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1362 root
->root_item
.flags
= 0;
1363 root
->root_item
.byte_limit
= 0;
1364 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1365 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1366 btrfs_set_root_level(&root
->root_item
, 0);
1367 btrfs_set_root_refs(&root
->root_item
, 1);
1368 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1369 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1370 btrfs_set_root_dirid(&root
->root_item
, 0);
1372 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1373 root
->root_item
.drop_level
= 0;
1375 key
.objectid
= objectid
;
1376 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1378 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1382 btrfs_tree_unlock(leaf
);
1388 btrfs_tree_unlock(leaf
);
1389 free_extent_buffer(root
->commit_root
);
1390 free_extent_buffer(leaf
);
1394 return ERR_PTR(ret
);
1397 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1398 struct btrfs_fs_info
*fs_info
)
1400 struct btrfs_root
*root
;
1401 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1402 struct extent_buffer
*leaf
;
1404 root
= btrfs_alloc_root(fs_info
);
1406 return ERR_PTR(-ENOMEM
);
1408 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1409 tree_root
->stripesize
, root
, fs_info
,
1410 BTRFS_TREE_LOG_OBJECTID
);
1412 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1413 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1414 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1417 * DON'T set REF_COWS for log trees
1419 * log trees do not get reference counted because they go away
1420 * before a real commit is actually done. They do store pointers
1421 * to file data extents, and those reference counts still get
1422 * updated (along with back refs to the log tree).
1425 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1429 return ERR_CAST(leaf
);
1432 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1433 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1434 btrfs_set_header_generation(leaf
, trans
->transid
);
1435 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1436 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1439 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1440 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1441 btrfs_mark_buffer_dirty(root
->node
);
1442 btrfs_tree_unlock(root
->node
);
1446 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1447 struct btrfs_fs_info
*fs_info
)
1449 struct btrfs_root
*log_root
;
1451 log_root
= alloc_log_tree(trans
, fs_info
);
1452 if (IS_ERR(log_root
))
1453 return PTR_ERR(log_root
);
1454 WARN_ON(fs_info
->log_root_tree
);
1455 fs_info
->log_root_tree
= log_root
;
1459 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1460 struct btrfs_root
*root
)
1462 struct btrfs_root
*log_root
;
1463 struct btrfs_inode_item
*inode_item
;
1465 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1466 if (IS_ERR(log_root
))
1467 return PTR_ERR(log_root
);
1469 log_root
->last_trans
= trans
->transid
;
1470 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1472 inode_item
= &log_root
->root_item
.inode
;
1473 btrfs_set_stack_inode_generation(inode_item
, 1);
1474 btrfs_set_stack_inode_size(inode_item
, 3);
1475 btrfs_set_stack_inode_nlink(inode_item
, 1);
1476 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1477 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1479 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1481 WARN_ON(root
->log_root
);
1482 root
->log_root
= log_root
;
1483 root
->log_transid
= 0;
1484 root
->log_transid_committed
= -1;
1485 root
->last_log_commit
= 0;
1489 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1490 struct btrfs_key
*key
)
1492 struct btrfs_root
*root
;
1493 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1494 struct btrfs_path
*path
;
1498 path
= btrfs_alloc_path();
1500 return ERR_PTR(-ENOMEM
);
1502 root
= btrfs_alloc_root(fs_info
);
1508 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1509 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1511 ret
= btrfs_find_root(tree_root
, key
, path
,
1512 &root
->root_item
, &root
->root_key
);
1519 generation
= btrfs_root_generation(&root
->root_item
);
1520 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1522 if (IS_ERR(root
->node
)) {
1523 ret
= PTR_ERR(root
->node
);
1525 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1527 free_extent_buffer(root
->node
);
1530 root
->commit_root
= btrfs_root_node(root
);
1532 btrfs_free_path(path
);
1538 root
= ERR_PTR(ret
);
1542 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1543 struct btrfs_key
*location
)
1545 struct btrfs_root
*root
;
1547 root
= btrfs_read_tree_root(tree_root
, location
);
1551 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1552 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1553 btrfs_check_and_init_root_item(&root
->root_item
);
1559 int btrfs_init_fs_root(struct btrfs_root
*root
)
1562 struct btrfs_subvolume_writers
*writers
;
1564 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1565 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1567 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1572 writers
= btrfs_alloc_subvolume_writers();
1573 if (IS_ERR(writers
)) {
1574 ret
= PTR_ERR(writers
);
1577 root
->subv_writers
= writers
;
1579 btrfs_init_free_ino_ctl(root
);
1580 spin_lock_init(&root
->ino_cache_lock
);
1581 init_waitqueue_head(&root
->ino_cache_wait
);
1583 ret
= get_anon_bdev(&root
->anon_dev
);
1587 mutex_lock(&root
->objectid_mutex
);
1588 ret
= btrfs_find_highest_objectid(root
,
1589 &root
->highest_objectid
);
1591 mutex_unlock(&root
->objectid_mutex
);
1595 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1597 mutex_unlock(&root
->objectid_mutex
);
1602 free_anon_bdev(root
->anon_dev
);
1604 btrfs_free_subvolume_writers(root
->subv_writers
);
1606 kfree(root
->free_ino_ctl
);
1607 kfree(root
->free_ino_pinned
);
1611 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1614 struct btrfs_root
*root
;
1616 spin_lock(&fs_info
->fs_roots_radix_lock
);
1617 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1618 (unsigned long)root_id
);
1619 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1623 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1624 struct btrfs_root
*root
)
1628 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1632 spin_lock(&fs_info
->fs_roots_radix_lock
);
1633 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1634 (unsigned long)root
->root_key
.objectid
,
1637 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1638 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1639 radix_tree_preload_end();
1644 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1645 struct btrfs_key
*location
,
1648 struct btrfs_root
*root
;
1649 struct btrfs_path
*path
;
1650 struct btrfs_key key
;
1653 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1654 return fs_info
->tree_root
;
1655 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1656 return fs_info
->extent_root
;
1657 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1658 return fs_info
->chunk_root
;
1659 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1660 return fs_info
->dev_root
;
1661 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1662 return fs_info
->csum_root
;
1663 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1664 return fs_info
->quota_root
? fs_info
->quota_root
:
1666 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1667 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1669 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1670 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1673 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1675 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1676 return ERR_PTR(-ENOENT
);
1680 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1684 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1689 ret
= btrfs_init_fs_root(root
);
1693 path
= btrfs_alloc_path();
1698 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1699 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1700 key
.offset
= location
->objectid
;
1702 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1703 btrfs_free_path(path
);
1707 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1709 ret
= btrfs_insert_fs_root(fs_info
, root
);
1711 if (ret
== -EEXIST
) {
1720 return ERR_PTR(ret
);
1723 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1725 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1727 struct btrfs_device
*device
;
1728 struct backing_dev_info
*bdi
;
1731 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1734 bdi
= blk_get_backing_dev_info(device
->bdev
);
1735 if (bdi_congested(bdi
, bdi_bits
)) {
1744 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1748 err
= bdi_setup_and_register(bdi
, "btrfs");
1752 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
;
1753 bdi
->congested_fn
= btrfs_congested_fn
;
1754 bdi
->congested_data
= info
;
1755 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1760 * called by the kthread helper functions to finally call the bio end_io
1761 * functions. This is where read checksum verification actually happens
1763 static void end_workqueue_fn(struct btrfs_work
*work
)
1766 struct btrfs_end_io_wq
*end_io_wq
;
1768 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1769 bio
= end_io_wq
->bio
;
1771 bio
->bi_error
= end_io_wq
->error
;
1772 bio
->bi_private
= end_io_wq
->private;
1773 bio
->bi_end_io
= end_io_wq
->end_io
;
1774 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1778 static int cleaner_kthread(void *arg
)
1780 struct btrfs_root
*root
= arg
;
1782 struct btrfs_trans_handle
*trans
;
1788 /* Make the cleaner go to sleep early. */
1789 if (btrfs_need_cleaner_sleep(root
))
1792 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1796 * Avoid the problem that we change the status of the fs
1797 * during the above check and trylock.
1799 if (btrfs_need_cleaner_sleep(root
)) {
1800 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1804 btrfs_run_delayed_iputs(root
);
1805 again
= btrfs_clean_one_deleted_snapshot(root
);
1806 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1809 * The defragger has dealt with the R/O remount and umount,
1810 * needn't do anything special here.
1812 btrfs_run_defrag_inodes(root
->fs_info
);
1815 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1816 * with relocation (btrfs_relocate_chunk) and relocation
1817 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1818 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1819 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1820 * unused block groups.
1822 btrfs_delete_unused_bgs(root
->fs_info
);
1824 if (!try_to_freeze() && !again
) {
1825 set_current_state(TASK_INTERRUPTIBLE
);
1826 if (!kthread_should_stop())
1828 __set_current_state(TASK_RUNNING
);
1830 } while (!kthread_should_stop());
1833 * Transaction kthread is stopped before us and wakes us up.
1834 * However we might have started a new transaction and COWed some
1835 * tree blocks when deleting unused block groups for example. So
1836 * make sure we commit the transaction we started to have a clean
1837 * shutdown when evicting the btree inode - if it has dirty pages
1838 * when we do the final iput() on it, eviction will trigger a
1839 * writeback for it which will fail with null pointer dereferences
1840 * since work queues and other resources were already released and
1841 * destroyed by the time the iput/eviction/writeback is made.
1843 trans
= btrfs_attach_transaction(root
);
1844 if (IS_ERR(trans
)) {
1845 if (PTR_ERR(trans
) != -ENOENT
)
1846 btrfs_err(root
->fs_info
,
1847 "cleaner transaction attach returned %ld",
1852 ret
= btrfs_commit_transaction(trans
, root
);
1854 btrfs_err(root
->fs_info
,
1855 "cleaner open transaction commit returned %d",
1862 static int transaction_kthread(void *arg
)
1864 struct btrfs_root
*root
= arg
;
1865 struct btrfs_trans_handle
*trans
;
1866 struct btrfs_transaction
*cur
;
1869 unsigned long delay
;
1873 cannot_commit
= false;
1874 delay
= HZ
* root
->fs_info
->commit_interval
;
1875 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1877 spin_lock(&root
->fs_info
->trans_lock
);
1878 cur
= root
->fs_info
->running_transaction
;
1880 spin_unlock(&root
->fs_info
->trans_lock
);
1884 now
= get_seconds();
1885 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1886 (now
< cur
->start_time
||
1887 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1888 spin_unlock(&root
->fs_info
->trans_lock
);
1892 transid
= cur
->transid
;
1893 spin_unlock(&root
->fs_info
->trans_lock
);
1895 /* If the file system is aborted, this will always fail. */
1896 trans
= btrfs_attach_transaction(root
);
1897 if (IS_ERR(trans
)) {
1898 if (PTR_ERR(trans
) != -ENOENT
)
1899 cannot_commit
= true;
1902 if (transid
== trans
->transid
) {
1903 btrfs_commit_transaction(trans
, root
);
1905 btrfs_end_transaction(trans
, root
);
1908 wake_up_process(root
->fs_info
->cleaner_kthread
);
1909 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1911 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1912 &root
->fs_info
->fs_state
)))
1913 btrfs_cleanup_transaction(root
);
1914 if (!try_to_freeze()) {
1915 set_current_state(TASK_INTERRUPTIBLE
);
1916 if (!kthread_should_stop() &&
1917 (!btrfs_transaction_blocked(root
->fs_info
) ||
1919 schedule_timeout(delay
);
1920 __set_current_state(TASK_RUNNING
);
1922 } while (!kthread_should_stop());
1927 * this will find the highest generation in the array of
1928 * root backups. The index of the highest array is returned,
1929 * or -1 if we can't find anything.
1931 * We check to make sure the array is valid by comparing the
1932 * generation of the latest root in the array with the generation
1933 * in the super block. If they don't match we pitch it.
1935 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1938 int newest_index
= -1;
1939 struct btrfs_root_backup
*root_backup
;
1942 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1943 root_backup
= info
->super_copy
->super_roots
+ i
;
1944 cur
= btrfs_backup_tree_root_gen(root_backup
);
1945 if (cur
== newest_gen
)
1949 /* check to see if we actually wrapped around */
1950 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1951 root_backup
= info
->super_copy
->super_roots
;
1952 cur
= btrfs_backup_tree_root_gen(root_backup
);
1953 if (cur
== newest_gen
)
1956 return newest_index
;
1961 * find the oldest backup so we know where to store new entries
1962 * in the backup array. This will set the backup_root_index
1963 * field in the fs_info struct
1965 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1968 int newest_index
= -1;
1970 newest_index
= find_newest_super_backup(info
, newest_gen
);
1971 /* if there was garbage in there, just move along */
1972 if (newest_index
== -1) {
1973 info
->backup_root_index
= 0;
1975 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1980 * copy all the root pointers into the super backup array.
1981 * this will bump the backup pointer by one when it is
1984 static void backup_super_roots(struct btrfs_fs_info
*info
)
1987 struct btrfs_root_backup
*root_backup
;
1990 next_backup
= info
->backup_root_index
;
1991 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1992 BTRFS_NUM_BACKUP_ROOTS
;
1995 * just overwrite the last backup if we're at the same generation
1996 * this happens only at umount
1998 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1999 if (btrfs_backup_tree_root_gen(root_backup
) ==
2000 btrfs_header_generation(info
->tree_root
->node
))
2001 next_backup
= last_backup
;
2003 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
2006 * make sure all of our padding and empty slots get zero filled
2007 * regardless of which ones we use today
2009 memset(root_backup
, 0, sizeof(*root_backup
));
2011 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
2013 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
2014 btrfs_set_backup_tree_root_gen(root_backup
,
2015 btrfs_header_generation(info
->tree_root
->node
));
2017 btrfs_set_backup_tree_root_level(root_backup
,
2018 btrfs_header_level(info
->tree_root
->node
));
2020 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2021 btrfs_set_backup_chunk_root_gen(root_backup
,
2022 btrfs_header_generation(info
->chunk_root
->node
));
2023 btrfs_set_backup_chunk_root_level(root_backup
,
2024 btrfs_header_level(info
->chunk_root
->node
));
2026 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2027 btrfs_set_backup_extent_root_gen(root_backup
,
2028 btrfs_header_generation(info
->extent_root
->node
));
2029 btrfs_set_backup_extent_root_level(root_backup
,
2030 btrfs_header_level(info
->extent_root
->node
));
2033 * we might commit during log recovery, which happens before we set
2034 * the fs_root. Make sure it is valid before we fill it in.
2036 if (info
->fs_root
&& info
->fs_root
->node
) {
2037 btrfs_set_backup_fs_root(root_backup
,
2038 info
->fs_root
->node
->start
);
2039 btrfs_set_backup_fs_root_gen(root_backup
,
2040 btrfs_header_generation(info
->fs_root
->node
));
2041 btrfs_set_backup_fs_root_level(root_backup
,
2042 btrfs_header_level(info
->fs_root
->node
));
2045 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2046 btrfs_set_backup_dev_root_gen(root_backup
,
2047 btrfs_header_generation(info
->dev_root
->node
));
2048 btrfs_set_backup_dev_root_level(root_backup
,
2049 btrfs_header_level(info
->dev_root
->node
));
2051 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2052 btrfs_set_backup_csum_root_gen(root_backup
,
2053 btrfs_header_generation(info
->csum_root
->node
));
2054 btrfs_set_backup_csum_root_level(root_backup
,
2055 btrfs_header_level(info
->csum_root
->node
));
2057 btrfs_set_backup_total_bytes(root_backup
,
2058 btrfs_super_total_bytes(info
->super_copy
));
2059 btrfs_set_backup_bytes_used(root_backup
,
2060 btrfs_super_bytes_used(info
->super_copy
));
2061 btrfs_set_backup_num_devices(root_backup
,
2062 btrfs_super_num_devices(info
->super_copy
));
2065 * if we don't copy this out to the super_copy, it won't get remembered
2066 * for the next commit
2068 memcpy(&info
->super_copy
->super_roots
,
2069 &info
->super_for_commit
->super_roots
,
2070 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2074 * this copies info out of the root backup array and back into
2075 * the in-memory super block. It is meant to help iterate through
2076 * the array, so you send it the number of backups you've already
2077 * tried and the last backup index you used.
2079 * this returns -1 when it has tried all the backups
2081 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2082 struct btrfs_super_block
*super
,
2083 int *num_backups_tried
, int *backup_index
)
2085 struct btrfs_root_backup
*root_backup
;
2086 int newest
= *backup_index
;
2088 if (*num_backups_tried
== 0) {
2089 u64 gen
= btrfs_super_generation(super
);
2091 newest
= find_newest_super_backup(info
, gen
);
2095 *backup_index
= newest
;
2096 *num_backups_tried
= 1;
2097 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2098 /* we've tried all the backups, all done */
2101 /* jump to the next oldest backup */
2102 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2103 BTRFS_NUM_BACKUP_ROOTS
;
2104 *backup_index
= newest
;
2105 *num_backups_tried
+= 1;
2107 root_backup
= super
->super_roots
+ newest
;
2109 btrfs_set_super_generation(super
,
2110 btrfs_backup_tree_root_gen(root_backup
));
2111 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2112 btrfs_set_super_root_level(super
,
2113 btrfs_backup_tree_root_level(root_backup
));
2114 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2117 * fixme: the total bytes and num_devices need to match or we should
2120 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2121 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2125 /* helper to cleanup workers */
2126 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2128 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2129 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2130 btrfs_destroy_workqueue(fs_info
->workers
);
2131 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2132 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2133 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2134 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2135 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2136 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2137 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2138 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2139 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2140 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2141 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2142 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2143 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2144 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2145 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2148 static void free_root_extent_buffers(struct btrfs_root
*root
)
2151 free_extent_buffer(root
->node
);
2152 free_extent_buffer(root
->commit_root
);
2154 root
->commit_root
= NULL
;
2158 /* helper to cleanup tree roots */
2159 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2161 free_root_extent_buffers(info
->tree_root
);
2163 free_root_extent_buffers(info
->dev_root
);
2164 free_root_extent_buffers(info
->extent_root
);
2165 free_root_extent_buffers(info
->csum_root
);
2166 free_root_extent_buffers(info
->quota_root
);
2167 free_root_extent_buffers(info
->uuid_root
);
2169 free_root_extent_buffers(info
->chunk_root
);
2170 free_root_extent_buffers(info
->free_space_root
);
2173 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2176 struct btrfs_root
*gang
[8];
2179 while (!list_empty(&fs_info
->dead_roots
)) {
2180 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2181 struct btrfs_root
, root_list
);
2182 list_del(&gang
[0]->root_list
);
2184 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2185 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2187 free_extent_buffer(gang
[0]->node
);
2188 free_extent_buffer(gang
[0]->commit_root
);
2189 btrfs_put_fs_root(gang
[0]);
2194 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2199 for (i
= 0; i
< ret
; i
++)
2200 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2203 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2204 btrfs_free_log_root_tree(NULL
, fs_info
);
2205 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2206 fs_info
->pinned_extents
);
2210 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2212 mutex_init(&fs_info
->scrub_lock
);
2213 atomic_set(&fs_info
->scrubs_running
, 0);
2214 atomic_set(&fs_info
->scrub_pause_req
, 0);
2215 atomic_set(&fs_info
->scrubs_paused
, 0);
2216 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2217 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2218 fs_info
->scrub_workers_refcnt
= 0;
2221 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2223 spin_lock_init(&fs_info
->balance_lock
);
2224 mutex_init(&fs_info
->balance_mutex
);
2225 atomic_set(&fs_info
->balance_running
, 0);
2226 atomic_set(&fs_info
->balance_pause_req
, 0);
2227 atomic_set(&fs_info
->balance_cancel_req
, 0);
2228 fs_info
->balance_ctl
= NULL
;
2229 init_waitqueue_head(&fs_info
->balance_wait_q
);
2232 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2233 struct btrfs_root
*tree_root
)
2235 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2236 set_nlink(fs_info
->btree_inode
, 1);
2238 * we set the i_size on the btree inode to the max possible int.
2239 * the real end of the address space is determined by all of
2240 * the devices in the system
2242 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2243 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2245 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2246 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2247 fs_info
->btree_inode
->i_mapping
);
2248 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2249 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2251 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2253 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2254 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2255 sizeof(struct btrfs_key
));
2256 set_bit(BTRFS_INODE_DUMMY
,
2257 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2258 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2261 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2263 fs_info
->dev_replace
.lock_owner
= 0;
2264 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2265 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2266 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2267 mutex_init(&fs_info
->dev_replace
.lock
);
2268 init_waitqueue_head(&fs_info
->replace_wait
);
2271 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2273 spin_lock_init(&fs_info
->qgroup_lock
);
2274 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2275 fs_info
->qgroup_tree
= RB_ROOT
;
2276 fs_info
->qgroup_op_tree
= RB_ROOT
;
2277 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2278 fs_info
->qgroup_seq
= 1;
2279 fs_info
->quota_enabled
= 0;
2280 fs_info
->pending_quota_state
= 0;
2281 fs_info
->qgroup_ulist
= NULL
;
2282 mutex_init(&fs_info
->qgroup_rescan_lock
);
2285 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2286 struct btrfs_fs_devices
*fs_devices
)
2288 int max_active
= fs_info
->thread_pool_size
;
2289 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2292 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2295 fs_info
->delalloc_workers
=
2296 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2298 fs_info
->flush_workers
=
2299 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2301 fs_info
->caching_workers
=
2302 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2305 * a higher idle thresh on the submit workers makes it much more
2306 * likely that bios will be send down in a sane order to the
2309 fs_info
->submit_workers
=
2310 btrfs_alloc_workqueue("submit", flags
,
2311 min_t(u64
, fs_devices
->num_devices
,
2314 fs_info
->fixup_workers
=
2315 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2318 * endios are largely parallel and should have a very
2321 fs_info
->endio_workers
=
2322 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2323 fs_info
->endio_meta_workers
=
2324 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2325 fs_info
->endio_meta_write_workers
=
2326 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2327 fs_info
->endio_raid56_workers
=
2328 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2329 fs_info
->endio_repair_workers
=
2330 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2331 fs_info
->rmw_workers
=
2332 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2333 fs_info
->endio_write_workers
=
2334 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2335 fs_info
->endio_freespace_worker
=
2336 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2337 fs_info
->delayed_workers
=
2338 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2339 fs_info
->readahead_workers
=
2340 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2341 fs_info
->qgroup_rescan_workers
=
2342 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2343 fs_info
->extent_workers
=
2344 btrfs_alloc_workqueue("extent-refs", flags
,
2345 min_t(u64
, fs_devices
->num_devices
,
2348 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2349 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2350 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2351 fs_info
->endio_meta_write_workers
&&
2352 fs_info
->endio_repair_workers
&&
2353 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2354 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2355 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2356 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2357 fs_info
->extent_workers
&&
2358 fs_info
->qgroup_rescan_workers
)) {
2365 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2366 struct btrfs_fs_devices
*fs_devices
)
2369 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2370 struct btrfs_root
*log_tree_root
;
2371 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2372 u64 bytenr
= btrfs_super_log_root(disk_super
);
2374 if (fs_devices
->rw_devices
== 0) {
2375 btrfs_warn(fs_info
, "log replay required on RO media");
2379 log_tree_root
= btrfs_alloc_root(fs_info
);
2383 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2384 tree_root
->stripesize
, log_tree_root
, fs_info
,
2385 BTRFS_TREE_LOG_OBJECTID
);
2387 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2388 fs_info
->generation
+ 1);
2389 if (IS_ERR(log_tree_root
->node
)) {
2390 btrfs_warn(fs_info
, "failed to read log tree");
2391 ret
= PTR_ERR(log_tree_root
->node
);
2392 kfree(log_tree_root
);
2394 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2395 btrfs_err(fs_info
, "failed to read log tree");
2396 free_extent_buffer(log_tree_root
->node
);
2397 kfree(log_tree_root
);
2400 /* returns with log_tree_root freed on success */
2401 ret
= btrfs_recover_log_trees(log_tree_root
);
2403 btrfs_std_error(tree_root
->fs_info
, ret
,
2404 "Failed to recover log tree");
2405 free_extent_buffer(log_tree_root
->node
);
2406 kfree(log_tree_root
);
2410 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2411 ret
= btrfs_commit_super(tree_root
);
2419 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2420 struct btrfs_root
*tree_root
)
2422 struct btrfs_root
*root
;
2423 struct btrfs_key location
;
2426 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2427 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2428 location
.offset
= 0;
2430 root
= btrfs_read_tree_root(tree_root
, &location
);
2432 return PTR_ERR(root
);
2433 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2434 fs_info
->extent_root
= root
;
2436 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2437 root
= btrfs_read_tree_root(tree_root
, &location
);
2439 return PTR_ERR(root
);
2440 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2441 fs_info
->dev_root
= root
;
2442 btrfs_init_devices_late(fs_info
);
2444 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2445 root
= btrfs_read_tree_root(tree_root
, &location
);
2447 return PTR_ERR(root
);
2448 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2449 fs_info
->csum_root
= root
;
2451 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2452 root
= btrfs_read_tree_root(tree_root
, &location
);
2453 if (!IS_ERR(root
)) {
2454 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2455 fs_info
->quota_enabled
= 1;
2456 fs_info
->pending_quota_state
= 1;
2457 fs_info
->quota_root
= root
;
2460 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2461 root
= btrfs_read_tree_root(tree_root
, &location
);
2463 ret
= PTR_ERR(root
);
2467 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2468 fs_info
->uuid_root
= root
;
2471 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2472 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2473 root
= btrfs_read_tree_root(tree_root
, &location
);
2475 return PTR_ERR(root
);
2476 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2477 fs_info
->free_space_root
= root
;
2483 int open_ctree(struct super_block
*sb
,
2484 struct btrfs_fs_devices
*fs_devices
,
2492 struct btrfs_key location
;
2493 struct buffer_head
*bh
;
2494 struct btrfs_super_block
*disk_super
;
2495 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2496 struct btrfs_root
*tree_root
;
2497 struct btrfs_root
*chunk_root
;
2500 int num_backups_tried
= 0;
2501 int backup_index
= 0;
2504 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2505 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2506 if (!tree_root
|| !chunk_root
) {
2511 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2517 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2523 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2528 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2529 (1 + ilog2(nr_cpu_ids
));
2531 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2534 goto fail_dirty_metadata_bytes
;
2537 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2540 goto fail_delalloc_bytes
;
2543 fs_info
->btree_inode
= new_inode(sb
);
2544 if (!fs_info
->btree_inode
) {
2546 goto fail_bio_counter
;
2549 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2551 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2552 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2553 INIT_LIST_HEAD(&fs_info
->trans_list
);
2554 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2555 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2556 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2557 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2558 spin_lock_init(&fs_info
->delalloc_root_lock
);
2559 spin_lock_init(&fs_info
->trans_lock
);
2560 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2561 spin_lock_init(&fs_info
->delayed_iput_lock
);
2562 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2563 spin_lock_init(&fs_info
->free_chunk_lock
);
2564 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2565 spin_lock_init(&fs_info
->super_lock
);
2566 spin_lock_init(&fs_info
->qgroup_op_lock
);
2567 spin_lock_init(&fs_info
->buffer_lock
);
2568 spin_lock_init(&fs_info
->unused_bgs_lock
);
2569 rwlock_init(&fs_info
->tree_mod_log_lock
);
2570 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2571 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2572 mutex_init(&fs_info
->reloc_mutex
);
2573 mutex_init(&fs_info
->delalloc_root_mutex
);
2574 seqlock_init(&fs_info
->profiles_lock
);
2575 init_rwsem(&fs_info
->delayed_iput_sem
);
2577 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2578 INIT_LIST_HEAD(&fs_info
->space_info
);
2579 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2580 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2581 btrfs_mapping_init(&fs_info
->mapping_tree
);
2582 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2583 BTRFS_BLOCK_RSV_GLOBAL
);
2584 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2585 BTRFS_BLOCK_RSV_DELALLOC
);
2586 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2587 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2588 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2589 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2590 BTRFS_BLOCK_RSV_DELOPS
);
2591 atomic_set(&fs_info
->nr_async_submits
, 0);
2592 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2593 atomic_set(&fs_info
->async_submit_draining
, 0);
2594 atomic_set(&fs_info
->nr_async_bios
, 0);
2595 atomic_set(&fs_info
->defrag_running
, 0);
2596 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2597 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2599 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2600 fs_info
->metadata_ratio
= 0;
2601 fs_info
->defrag_inodes
= RB_ROOT
;
2602 fs_info
->free_chunk_space
= 0;
2603 fs_info
->tree_mod_log
= RB_ROOT
;
2604 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2605 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2606 /* readahead state */
2607 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2608 spin_lock_init(&fs_info
->reada_lock
);
2610 fs_info
->thread_pool_size
= min_t(unsigned long,
2611 num_online_cpus() + 2, 8);
2613 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2614 spin_lock_init(&fs_info
->ordered_root_lock
);
2615 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2617 if (!fs_info
->delayed_root
) {
2621 btrfs_init_delayed_root(fs_info
->delayed_root
);
2623 btrfs_init_scrub(fs_info
);
2624 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2625 fs_info
->check_integrity_print_mask
= 0;
2627 btrfs_init_balance(fs_info
);
2628 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2630 sb
->s_blocksize
= 4096;
2631 sb
->s_blocksize_bits
= blksize_bits(4096);
2632 sb
->s_bdi
= &fs_info
->bdi
;
2634 btrfs_init_btree_inode(fs_info
, tree_root
);
2636 spin_lock_init(&fs_info
->block_group_cache_lock
);
2637 fs_info
->block_group_cache_tree
= RB_ROOT
;
2638 fs_info
->first_logical_byte
= (u64
)-1;
2640 extent_io_tree_init(&fs_info
->freed_extents
[0],
2641 fs_info
->btree_inode
->i_mapping
);
2642 extent_io_tree_init(&fs_info
->freed_extents
[1],
2643 fs_info
->btree_inode
->i_mapping
);
2644 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2645 fs_info
->do_barriers
= 1;
2648 mutex_init(&fs_info
->ordered_operations_mutex
);
2649 mutex_init(&fs_info
->tree_log_mutex
);
2650 mutex_init(&fs_info
->chunk_mutex
);
2651 mutex_init(&fs_info
->transaction_kthread_mutex
);
2652 mutex_init(&fs_info
->cleaner_mutex
);
2653 mutex_init(&fs_info
->volume_mutex
);
2654 mutex_init(&fs_info
->ro_block_group_mutex
);
2655 init_rwsem(&fs_info
->commit_root_sem
);
2656 init_rwsem(&fs_info
->cleanup_work_sem
);
2657 init_rwsem(&fs_info
->subvol_sem
);
2658 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2660 btrfs_init_dev_replace_locks(fs_info
);
2661 btrfs_init_qgroup(fs_info
);
2663 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2664 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2666 init_waitqueue_head(&fs_info
->transaction_throttle
);
2667 init_waitqueue_head(&fs_info
->transaction_wait
);
2668 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2669 init_waitqueue_head(&fs_info
->async_submit_wait
);
2671 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2673 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2679 __setup_root(4096, 4096, 4096, tree_root
,
2680 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2682 invalidate_bdev(fs_devices
->latest_bdev
);
2685 * Read super block and check the signature bytes only
2687 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2694 * We want to check superblock checksum, the type is stored inside.
2695 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2697 if (btrfs_check_super_csum(bh
->b_data
)) {
2698 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2705 * super_copy is zeroed at allocation time and we never touch the
2706 * following bytes up to INFO_SIZE, the checksum is calculated from
2707 * the whole block of INFO_SIZE
2709 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2710 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2711 sizeof(*fs_info
->super_for_commit
));
2714 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2716 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2718 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2723 disk_super
= fs_info
->super_copy
;
2724 if (!btrfs_super_root(disk_super
))
2727 /* check FS state, whether FS is broken. */
2728 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2729 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2732 * run through our array of backup supers and setup
2733 * our ring pointer to the oldest one
2735 generation
= btrfs_super_generation(disk_super
);
2736 find_oldest_super_backup(fs_info
, generation
);
2739 * In the long term, we'll store the compression type in the super
2740 * block, and it'll be used for per file compression control.
2742 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2744 ret
= btrfs_parse_options(tree_root
, options
);
2750 features
= btrfs_super_incompat_flags(disk_super
) &
2751 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2753 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2754 "unsupported optional features (%Lx).\n",
2761 * Leafsize and nodesize were always equal, this is only a sanity check.
2763 if (le32_to_cpu(disk_super
->__unused_leafsize
) !=
2764 btrfs_super_nodesize(disk_super
)) {
2765 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2766 "blocksizes don't match. node %d leaf %d\n",
2767 btrfs_super_nodesize(disk_super
),
2768 le32_to_cpu(disk_super
->__unused_leafsize
));
2772 if (btrfs_super_nodesize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2773 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2774 "blocksize (%d) was too large\n",
2775 btrfs_super_nodesize(disk_super
));
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 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2852 printk(KERN_ERR
"BTRFS: valid FS not found on %s\n", sb
->s_id
);
2853 goto fail_sb_buffer
;
2856 if (sectorsize
!= PAGE_SIZE
) {
2857 printk(KERN_ERR
"BTRFS: incompatible sector size (%lu) "
2858 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2859 goto fail_sb_buffer
;
2862 mutex_lock(&fs_info
->chunk_mutex
);
2863 ret
= btrfs_read_sys_array(tree_root
);
2864 mutex_unlock(&fs_info
->chunk_mutex
);
2866 printk(KERN_ERR
"BTRFS: failed to read the system "
2867 "array on %s\n", sb
->s_id
);
2868 goto fail_sb_buffer
;
2871 generation
= btrfs_super_chunk_root_generation(disk_super
);
2873 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2874 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2876 chunk_root
->node
= read_tree_block(chunk_root
,
2877 btrfs_super_chunk_root(disk_super
),
2879 if (IS_ERR(chunk_root
->node
) ||
2880 !extent_buffer_uptodate(chunk_root
->node
)) {
2881 printk(KERN_ERR
"BTRFS: failed to read chunk root on %s\n",
2883 if (!IS_ERR(chunk_root
->node
))
2884 free_extent_buffer(chunk_root
->node
);
2885 chunk_root
->node
= NULL
;
2886 goto fail_tree_roots
;
2888 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2889 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2891 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2892 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2894 ret
= btrfs_read_chunk_tree(chunk_root
);
2896 printk(KERN_ERR
"BTRFS: failed to read chunk tree on %s\n",
2898 goto fail_tree_roots
;
2902 * keep the device that is marked to be the target device for the
2903 * dev_replace procedure
2905 btrfs_close_extra_devices(fs_devices
, 0);
2907 if (!fs_devices
->latest_bdev
) {
2908 printk(KERN_ERR
"BTRFS: failed to read devices on %s\n",
2910 goto fail_tree_roots
;
2914 generation
= btrfs_super_generation(disk_super
);
2916 tree_root
->node
= read_tree_block(tree_root
,
2917 btrfs_super_root(disk_super
),
2919 if (IS_ERR(tree_root
->node
) ||
2920 !extent_buffer_uptodate(tree_root
->node
)) {
2921 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2923 if (!IS_ERR(tree_root
->node
))
2924 free_extent_buffer(tree_root
->node
);
2925 tree_root
->node
= NULL
;
2926 goto recovery_tree_root
;
2929 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2930 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2931 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2933 mutex_lock(&tree_root
->objectid_mutex
);
2934 ret
= btrfs_find_highest_objectid(tree_root
,
2935 &tree_root
->highest_objectid
);
2937 mutex_unlock(&tree_root
->objectid_mutex
);
2938 goto recovery_tree_root
;
2941 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2943 mutex_unlock(&tree_root
->objectid_mutex
);
2945 ret
= btrfs_read_roots(fs_info
, tree_root
);
2947 goto recovery_tree_root
;
2949 fs_info
->generation
= generation
;
2950 fs_info
->last_trans_committed
= generation
;
2952 ret
= btrfs_recover_balance(fs_info
);
2954 printk(KERN_ERR
"BTRFS: failed to recover balance\n");
2955 goto fail_block_groups
;
2958 ret
= btrfs_init_dev_stats(fs_info
);
2960 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2962 goto fail_block_groups
;
2965 ret
= btrfs_init_dev_replace(fs_info
);
2967 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2968 goto fail_block_groups
;
2971 btrfs_close_extra_devices(fs_devices
, 1);
2973 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2975 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret
);
2976 goto fail_block_groups
;
2979 ret
= btrfs_sysfs_add_device(fs_devices
);
2981 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret
);
2982 goto fail_fsdev_sysfs
;
2985 ret
= btrfs_sysfs_add_mounted(fs_info
);
2987 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2988 goto fail_fsdev_sysfs
;
2991 ret
= btrfs_init_space_info(fs_info
);
2993 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2997 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2999 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
3002 fs_info
->num_tolerated_disk_barrier_failures
=
3003 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3004 if (fs_info
->fs_devices
->missing_devices
>
3005 fs_info
->num_tolerated_disk_barrier_failures
&&
3006 !(sb
->s_flags
& MS_RDONLY
)) {
3007 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
3008 fs_info
->fs_devices
->missing_devices
,
3009 fs_info
->num_tolerated_disk_barrier_failures
);
3013 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3015 if (IS_ERR(fs_info
->cleaner_kthread
))
3018 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3020 "btrfs-transaction");
3021 if (IS_ERR(fs_info
->transaction_kthread
))
3024 if (!btrfs_test_opt(tree_root
, SSD
) &&
3025 !btrfs_test_opt(tree_root
, NOSSD
) &&
3026 !fs_info
->fs_devices
->rotating
) {
3027 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
3029 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
3033 * Mount does not set all options immediatelly, we can do it now and do
3034 * not have to wait for transaction commit
3036 btrfs_apply_pending_changes(fs_info
);
3038 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3039 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
3040 ret
= btrfsic_mount(tree_root
, fs_devices
,
3041 btrfs_test_opt(tree_root
,
3042 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3044 fs_info
->check_integrity_print_mask
);
3046 printk(KERN_WARNING
"BTRFS: failed to initialize"
3047 " integrity check module %s\n", sb
->s_id
);
3050 ret
= btrfs_read_qgroup_config(fs_info
);
3052 goto fail_trans_kthread
;
3054 /* do not make disk changes in broken FS */
3055 if (btrfs_super_log_root(disk_super
) != 0) {
3056 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3063 ret
= btrfs_find_orphan_roots(tree_root
);
3067 if (!(sb
->s_flags
& MS_RDONLY
)) {
3068 ret
= btrfs_cleanup_fs_roots(fs_info
);
3072 mutex_lock(&fs_info
->cleaner_mutex
);
3073 ret
= btrfs_recover_relocation(tree_root
);
3074 mutex_unlock(&fs_info
->cleaner_mutex
);
3077 "BTRFS: failed to recover relocation\n");
3083 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3084 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3085 location
.offset
= 0;
3087 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3088 if (IS_ERR(fs_info
->fs_root
)) {
3089 err
= PTR_ERR(fs_info
->fs_root
);
3093 if (sb
->s_flags
& MS_RDONLY
)
3096 if (btrfs_test_opt(tree_root
, FREE_SPACE_TREE
) &&
3097 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3098 pr_info("BTRFS: creating free space tree\n");
3099 ret
= btrfs_create_free_space_tree(fs_info
);
3101 pr_warn("BTRFS: failed to create free space tree %d\n",
3103 close_ctree(tree_root
);
3108 down_read(&fs_info
->cleanup_work_sem
);
3109 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3110 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3111 up_read(&fs_info
->cleanup_work_sem
);
3112 close_ctree(tree_root
);
3115 up_read(&fs_info
->cleanup_work_sem
);
3117 ret
= btrfs_resume_balance_async(fs_info
);
3119 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
3120 close_ctree(tree_root
);
3124 ret
= btrfs_resume_dev_replace_async(fs_info
);
3126 pr_warn("BTRFS: failed to resume dev_replace\n");
3127 close_ctree(tree_root
);
3131 btrfs_qgroup_rescan_resume(fs_info
);
3133 if (btrfs_test_opt(tree_root
, CLEAR_CACHE
) &&
3134 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3135 pr_info("BTRFS: clearing free space tree\n");
3136 ret
= btrfs_clear_free_space_tree(fs_info
);
3138 pr_warn("BTRFS: failed to clear free space tree %d\n",
3140 close_ctree(tree_root
);
3145 if (!fs_info
->uuid_root
) {
3146 pr_info("BTRFS: creating UUID tree\n");
3147 ret
= btrfs_create_uuid_tree(fs_info
);
3149 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3151 close_ctree(tree_root
);
3154 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3155 fs_info
->generation
!=
3156 btrfs_super_uuid_tree_generation(disk_super
)) {
3157 pr_info("BTRFS: checking UUID tree\n");
3158 ret
= btrfs_check_uuid_tree(fs_info
);
3160 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3162 close_ctree(tree_root
);
3166 fs_info
->update_uuid_tree_gen
= 1;
3174 btrfs_free_qgroup_config(fs_info
);
3176 kthread_stop(fs_info
->transaction_kthread
);
3177 btrfs_cleanup_transaction(fs_info
->tree_root
);
3178 btrfs_free_fs_roots(fs_info
);
3180 kthread_stop(fs_info
->cleaner_kthread
);
3183 * make sure we're done with the btree inode before we stop our
3186 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3189 btrfs_sysfs_remove_mounted(fs_info
);
3192 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3195 btrfs_put_block_group_cache(fs_info
);
3196 btrfs_free_block_groups(fs_info
);
3199 free_root_pointers(fs_info
, 1);
3200 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3203 btrfs_stop_all_workers(fs_info
);
3206 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3208 iput(fs_info
->btree_inode
);
3210 percpu_counter_destroy(&fs_info
->bio_counter
);
3211 fail_delalloc_bytes
:
3212 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3213 fail_dirty_metadata_bytes
:
3214 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3216 bdi_destroy(&fs_info
->bdi
);
3218 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3220 btrfs_free_stripe_hash_table(fs_info
);
3221 btrfs_close_devices(fs_info
->fs_devices
);
3225 if (!btrfs_test_opt(tree_root
, RECOVERY
))
3226 goto fail_tree_roots
;
3228 free_root_pointers(fs_info
, 0);
3230 /* don't use the log in recovery mode, it won't be valid */
3231 btrfs_set_super_log_root(disk_super
, 0);
3233 /* we can't trust the free space cache either */
3234 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3236 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3237 &num_backups_tried
, &backup_index
);
3239 goto fail_block_groups
;
3240 goto retry_root_backup
;
3243 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3246 set_buffer_uptodate(bh
);
3248 struct btrfs_device
*device
= (struct btrfs_device
*)
3251 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3252 "lost page write due to IO error on %s",
3253 rcu_str_deref(device
->name
));
3254 /* note, we dont' set_buffer_write_io_error because we have
3255 * our own ways of dealing with the IO errors
3257 clear_buffer_uptodate(bh
);
3258 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3264 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3265 struct buffer_head
**bh_ret
)
3267 struct buffer_head
*bh
;
3268 struct btrfs_super_block
*super
;
3271 bytenr
= btrfs_sb_offset(copy_num
);
3272 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3275 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3277 * If we fail to read from the underlying devices, as of now
3278 * the best option we have is to mark it EIO.
3283 super
= (struct btrfs_super_block
*)bh
->b_data
;
3284 if (btrfs_super_bytenr(super
) != bytenr
||
3285 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3295 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3297 struct buffer_head
*bh
;
3298 struct buffer_head
*latest
= NULL
;
3299 struct btrfs_super_block
*super
;
3304 /* we would like to check all the supers, but that would make
3305 * a btrfs mount succeed after a mkfs from a different FS.
3306 * So, we need to add a special mount option to scan for
3307 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3309 for (i
= 0; i
< 1; i
++) {
3310 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3314 super
= (struct btrfs_super_block
*)bh
->b_data
;
3316 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3319 transid
= btrfs_super_generation(super
);
3326 return ERR_PTR(ret
);
3332 * this should be called twice, once with wait == 0 and
3333 * once with wait == 1. When wait == 0 is done, all the buffer heads
3334 * we write are pinned.
3336 * They are released when wait == 1 is done.
3337 * max_mirrors must be the same for both runs, and it indicates how
3338 * many supers on this one device should be written.
3340 * max_mirrors == 0 means to write them all.
3342 static int write_dev_supers(struct btrfs_device
*device
,
3343 struct btrfs_super_block
*sb
,
3344 int do_barriers
, int wait
, int max_mirrors
)
3346 struct buffer_head
*bh
;
3353 if (max_mirrors
== 0)
3354 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3356 for (i
= 0; i
< max_mirrors
; i
++) {
3357 bytenr
= btrfs_sb_offset(i
);
3358 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3359 device
->commit_total_bytes
)
3363 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3364 BTRFS_SUPER_INFO_SIZE
);
3370 if (!buffer_uptodate(bh
))
3373 /* drop our reference */
3376 /* drop the reference from the wait == 0 run */
3380 btrfs_set_super_bytenr(sb
, bytenr
);
3383 crc
= btrfs_csum_data((char *)sb
+
3384 BTRFS_CSUM_SIZE
, crc
,
3385 BTRFS_SUPER_INFO_SIZE
-
3387 btrfs_csum_final(crc
, sb
->csum
);
3390 * one reference for us, and we leave it for the
3393 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3394 BTRFS_SUPER_INFO_SIZE
);
3396 btrfs_err(device
->dev_root
->fs_info
,
3397 "couldn't get super buffer head for bytenr %llu",
3403 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3405 /* one reference for submit_bh */
3408 set_buffer_uptodate(bh
);
3410 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3411 bh
->b_private
= device
;
3415 * we fua the first super. The others we allow
3419 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3421 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3425 return errors
< i
? 0 : -1;
3429 * endio for the write_dev_flush, this will wake anyone waiting
3430 * for the barrier when it is done
3432 static void btrfs_end_empty_barrier(struct bio
*bio
)
3434 if (bio
->bi_private
)
3435 complete(bio
->bi_private
);
3440 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3441 * sent down. With wait == 1, it waits for the previous flush.
3443 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3446 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3451 if (device
->nobarriers
)
3455 bio
= device
->flush_bio
;
3459 wait_for_completion(&device
->flush_wait
);
3461 if (bio
->bi_error
) {
3462 ret
= bio
->bi_error
;
3463 btrfs_dev_stat_inc_and_print(device
,
3464 BTRFS_DEV_STAT_FLUSH_ERRS
);
3467 /* drop the reference from the wait == 0 run */
3469 device
->flush_bio
= NULL
;
3475 * one reference for us, and we leave it for the
3478 device
->flush_bio
= NULL
;
3479 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3483 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3484 bio
->bi_bdev
= device
->bdev
;
3485 init_completion(&device
->flush_wait
);
3486 bio
->bi_private
= &device
->flush_wait
;
3487 device
->flush_bio
= bio
;
3490 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3496 * send an empty flush down to each device in parallel,
3497 * then wait for them
3499 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3501 struct list_head
*head
;
3502 struct btrfs_device
*dev
;
3503 int errors_send
= 0;
3504 int errors_wait
= 0;
3507 /* send down all the barriers */
3508 head
= &info
->fs_devices
->devices
;
3509 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3516 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3519 ret
= write_dev_flush(dev
, 0);
3524 /* wait for all the barriers */
3525 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3532 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3535 ret
= write_dev_flush(dev
, 1);
3539 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3540 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3545 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3548 int min_tolerated
= INT_MAX
;
3550 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3551 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3552 min_tolerated
= min(min_tolerated
,
3553 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3554 tolerated_failures
);
3556 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3557 if (raid_type
== BTRFS_RAID_SINGLE
)
3559 if (!(flags
& btrfs_raid_group
[raid_type
]))
3561 min_tolerated
= min(min_tolerated
,
3562 btrfs_raid_array
[raid_type
].
3563 tolerated_failures
);
3566 if (min_tolerated
== INT_MAX
) {
3567 pr_warn("BTRFS: unknown raid flag: %llu\n", flags
);
3571 return min_tolerated
;
3574 int btrfs_calc_num_tolerated_disk_barrier_failures(
3575 struct btrfs_fs_info
*fs_info
)
3577 struct btrfs_ioctl_space_info space
;
3578 struct btrfs_space_info
*sinfo
;
3579 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3580 BTRFS_BLOCK_GROUP_SYSTEM
,
3581 BTRFS_BLOCK_GROUP_METADATA
,
3582 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3585 int num_tolerated_disk_barrier_failures
=
3586 (int)fs_info
->fs_devices
->num_devices
;
3588 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3589 struct btrfs_space_info
*tmp
;
3593 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3594 if (tmp
->flags
== types
[i
]) {
3604 down_read(&sinfo
->groups_sem
);
3605 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3608 if (list_empty(&sinfo
->block_groups
[c
]))
3611 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3613 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3615 flags
= space
.flags
;
3617 num_tolerated_disk_barrier_failures
= min(
3618 num_tolerated_disk_barrier_failures
,
3619 btrfs_get_num_tolerated_disk_barrier_failures(
3622 up_read(&sinfo
->groups_sem
);
3625 return num_tolerated_disk_barrier_failures
;
3628 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3630 struct list_head
*head
;
3631 struct btrfs_device
*dev
;
3632 struct btrfs_super_block
*sb
;
3633 struct btrfs_dev_item
*dev_item
;
3637 int total_errors
= 0;
3640 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3641 backup_super_roots(root
->fs_info
);
3643 sb
= root
->fs_info
->super_for_commit
;
3644 dev_item
= &sb
->dev_item
;
3646 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3647 head
= &root
->fs_info
->fs_devices
->devices
;
3648 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3651 ret
= barrier_all_devices(root
->fs_info
);
3654 &root
->fs_info
->fs_devices
->device_list_mutex
);
3655 btrfs_std_error(root
->fs_info
, ret
,
3656 "errors while submitting device barriers.");
3661 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3666 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3669 btrfs_set_stack_device_generation(dev_item
, 0);
3670 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3671 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3672 btrfs_set_stack_device_total_bytes(dev_item
,
3673 dev
->commit_total_bytes
);
3674 btrfs_set_stack_device_bytes_used(dev_item
,
3675 dev
->commit_bytes_used
);
3676 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3677 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3678 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3679 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3680 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3682 flags
= btrfs_super_flags(sb
);
3683 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3685 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3689 if (total_errors
> max_errors
) {
3690 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3692 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3694 /* FUA is masked off if unsupported and can't be the reason */
3695 btrfs_std_error(root
->fs_info
, -EIO
,
3696 "%d errors while writing supers", total_errors
);
3701 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3704 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3707 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3711 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3712 if (total_errors
> max_errors
) {
3713 btrfs_std_error(root
->fs_info
, -EIO
,
3714 "%d errors while writing supers", total_errors
);
3720 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3721 struct btrfs_root
*root
, int max_mirrors
)
3723 return write_all_supers(root
, max_mirrors
);
3726 /* Drop a fs root from the radix tree and free it. */
3727 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3728 struct btrfs_root
*root
)
3730 spin_lock(&fs_info
->fs_roots_radix_lock
);
3731 radix_tree_delete(&fs_info
->fs_roots_radix
,
3732 (unsigned long)root
->root_key
.objectid
);
3733 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3735 if (btrfs_root_refs(&root
->root_item
) == 0)
3736 synchronize_srcu(&fs_info
->subvol_srcu
);
3738 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3739 btrfs_free_log(NULL
, root
);
3741 if (root
->free_ino_pinned
)
3742 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3743 if (root
->free_ino_ctl
)
3744 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3748 static void free_fs_root(struct btrfs_root
*root
)
3750 iput(root
->ino_cache_inode
);
3751 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3752 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3753 root
->orphan_block_rsv
= NULL
;
3755 free_anon_bdev(root
->anon_dev
);
3756 if (root
->subv_writers
)
3757 btrfs_free_subvolume_writers(root
->subv_writers
);
3758 free_extent_buffer(root
->node
);
3759 free_extent_buffer(root
->commit_root
);
3760 kfree(root
->free_ino_ctl
);
3761 kfree(root
->free_ino_pinned
);
3763 btrfs_put_fs_root(root
);
3766 void btrfs_free_fs_root(struct btrfs_root
*root
)
3771 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3773 u64 root_objectid
= 0;
3774 struct btrfs_root
*gang
[8];
3777 unsigned int ret
= 0;
3781 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3782 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3783 (void **)gang
, root_objectid
,
3786 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3789 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3791 for (i
= 0; i
< ret
; i
++) {
3792 /* Avoid to grab roots in dead_roots */
3793 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3797 /* grab all the search result for later use */
3798 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3800 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3802 for (i
= 0; i
< ret
; i
++) {
3805 root_objectid
= gang
[i
]->root_key
.objectid
;
3806 err
= btrfs_orphan_cleanup(gang
[i
]);
3809 btrfs_put_fs_root(gang
[i
]);
3814 /* release the uncleaned roots due to error */
3815 for (; i
< ret
; i
++) {
3817 btrfs_put_fs_root(gang
[i
]);
3822 int btrfs_commit_super(struct btrfs_root
*root
)
3824 struct btrfs_trans_handle
*trans
;
3826 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3827 btrfs_run_delayed_iputs(root
);
3828 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3829 wake_up_process(root
->fs_info
->cleaner_kthread
);
3831 /* wait until ongoing cleanup work done */
3832 down_write(&root
->fs_info
->cleanup_work_sem
);
3833 up_write(&root
->fs_info
->cleanup_work_sem
);
3835 trans
= btrfs_join_transaction(root
);
3837 return PTR_ERR(trans
);
3838 return btrfs_commit_transaction(trans
, root
);
3841 void close_ctree(struct btrfs_root
*root
)
3843 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3846 fs_info
->closing
= 1;
3849 /* wait for the qgroup rescan worker to stop */
3850 btrfs_qgroup_wait_for_completion(fs_info
);
3852 /* wait for the uuid_scan task to finish */
3853 down(&fs_info
->uuid_tree_rescan_sem
);
3854 /* avoid complains from lockdep et al., set sem back to initial state */
3855 up(&fs_info
->uuid_tree_rescan_sem
);
3857 /* pause restriper - we want to resume on mount */
3858 btrfs_pause_balance(fs_info
);
3860 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3862 btrfs_scrub_cancel(fs_info
);
3864 /* wait for any defraggers to finish */
3865 wait_event(fs_info
->transaction_wait
,
3866 (atomic_read(&fs_info
->defrag_running
) == 0));
3868 /* clear out the rbtree of defraggable inodes */
3869 btrfs_cleanup_defrag_inodes(fs_info
);
3871 cancel_work_sync(&fs_info
->async_reclaim_work
);
3873 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3875 * If the cleaner thread is stopped and there are
3876 * block groups queued for removal, the deletion will be
3877 * skipped when we quit the cleaner thread.
3879 btrfs_delete_unused_bgs(root
->fs_info
);
3881 ret
= btrfs_commit_super(root
);
3883 btrfs_err(fs_info
, "commit super ret %d", ret
);
3886 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3887 btrfs_error_commit_super(root
);
3889 kthread_stop(fs_info
->transaction_kthread
);
3890 kthread_stop(fs_info
->cleaner_kthread
);
3892 fs_info
->closing
= 2;
3895 btrfs_free_qgroup_config(fs_info
);
3897 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3898 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3899 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3902 btrfs_sysfs_remove_mounted(fs_info
);
3903 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3905 btrfs_free_fs_roots(fs_info
);
3907 btrfs_put_block_group_cache(fs_info
);
3909 btrfs_free_block_groups(fs_info
);
3912 * we must make sure there is not any read request to
3913 * submit after we stopping all workers.
3915 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3916 btrfs_stop_all_workers(fs_info
);
3919 free_root_pointers(fs_info
, 1);
3921 iput(fs_info
->btree_inode
);
3923 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3924 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3925 btrfsic_unmount(root
, fs_info
->fs_devices
);
3928 btrfs_close_devices(fs_info
->fs_devices
);
3929 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3931 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3932 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3933 percpu_counter_destroy(&fs_info
->bio_counter
);
3934 bdi_destroy(&fs_info
->bdi
);
3935 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3937 btrfs_free_stripe_hash_table(fs_info
);
3939 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3940 root
->orphan_block_rsv
= NULL
;
3943 while (!list_empty(&fs_info
->pinned_chunks
)) {
3944 struct extent_map
*em
;
3946 em
= list_first_entry(&fs_info
->pinned_chunks
,
3947 struct extent_map
, list
);
3948 list_del_init(&em
->list
);
3949 free_extent_map(em
);
3951 unlock_chunks(root
);
3954 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3958 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3960 ret
= extent_buffer_uptodate(buf
);
3964 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3965 parent_transid
, atomic
);
3971 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3973 struct btrfs_root
*root
;
3974 u64 transid
= btrfs_header_generation(buf
);
3977 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3979 * This is a fast path so only do this check if we have sanity tests
3980 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3981 * outside of the sanity tests.
3983 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3986 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3987 btrfs_assert_tree_locked(buf
);
3988 if (transid
!= root
->fs_info
->generation
)
3989 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3990 "found %llu running %llu\n",
3991 buf
->start
, transid
, root
->fs_info
->generation
);
3992 was_dirty
= set_extent_buffer_dirty(buf
);
3994 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3996 root
->fs_info
->dirty_metadata_batch
);
3997 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3998 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3999 btrfs_print_leaf(root
, buf
);
4005 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
4009 * looks as though older kernels can get into trouble with
4010 * this code, they end up stuck in balance_dirty_pages forever
4014 if (current
->flags
& PF_MEMALLOC
)
4018 btrfs_balance_delayed_items(root
);
4020 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
4021 BTRFS_DIRTY_METADATA_THRESH
);
4023 balance_dirty_pages_ratelimited(
4024 root
->fs_info
->btree_inode
->i_mapping
);
4028 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
4030 __btrfs_btree_balance_dirty(root
, 1);
4033 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
4035 __btrfs_btree_balance_dirty(root
, 0);
4038 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
4040 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4041 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
4044 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
4047 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
4050 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4051 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
4052 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
4055 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4056 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
4057 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
4060 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4061 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
4062 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
4067 * The common minimum, we don't know if we can trust the nodesize/sectorsize
4068 * items yet, they'll be verified later. Issue just a warning.
4070 if (!IS_ALIGNED(btrfs_super_root(sb
), 4096))
4071 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
4072 btrfs_super_root(sb
));
4073 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), 4096))
4074 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
4075 btrfs_super_chunk_root(sb
));
4076 if (!IS_ALIGNED(btrfs_super_log_root(sb
), 4096))
4077 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
4078 btrfs_super_log_root(sb
));
4081 * Check the lower bound, the alignment and other constraints are
4084 if (btrfs_super_nodesize(sb
) < 4096) {
4085 printk(KERN_ERR
"BTRFS: nodesize too small: %u < 4096\n",
4086 btrfs_super_nodesize(sb
));
4089 if (btrfs_super_sectorsize(sb
) < 4096) {
4090 printk(KERN_ERR
"BTRFS: sectorsize too small: %u < 4096\n",
4091 btrfs_super_sectorsize(sb
));
4095 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4096 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4097 fs_info
->fsid
, sb
->dev_item
.fsid
);
4102 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4105 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4106 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
4107 btrfs_super_num_devices(sb
));
4108 if (btrfs_super_num_devices(sb
) == 0) {
4109 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
4113 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4114 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
4115 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4120 * Obvious sys_chunk_array corruptions, it must hold at least one key
4123 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4124 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
4125 btrfs_super_sys_array_size(sb
),
4126 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4129 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4130 + sizeof(struct btrfs_chunk
)) {
4131 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
4132 btrfs_super_sys_array_size(sb
),
4133 sizeof(struct btrfs_disk_key
)
4134 + sizeof(struct btrfs_chunk
));
4139 * The generation is a global counter, we'll trust it more than the others
4140 * but it's still possible that it's the one that's wrong.
4142 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4144 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4145 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
4146 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4147 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4149 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4150 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
4155 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4157 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4158 btrfs_run_delayed_iputs(root
);
4159 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4161 down_write(&root
->fs_info
->cleanup_work_sem
);
4162 up_write(&root
->fs_info
->cleanup_work_sem
);
4164 /* cleanup FS via transaction */
4165 btrfs_cleanup_transaction(root
);
4168 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4170 struct btrfs_ordered_extent
*ordered
;
4172 spin_lock(&root
->ordered_extent_lock
);
4174 * This will just short circuit the ordered completion stuff which will
4175 * make sure the ordered extent gets properly cleaned up.
4177 list_for_each_entry(ordered
, &root
->ordered_extents
,
4179 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4180 spin_unlock(&root
->ordered_extent_lock
);
4183 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4185 struct btrfs_root
*root
;
4186 struct list_head splice
;
4188 INIT_LIST_HEAD(&splice
);
4190 spin_lock(&fs_info
->ordered_root_lock
);
4191 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4192 while (!list_empty(&splice
)) {
4193 root
= list_first_entry(&splice
, struct btrfs_root
,
4195 list_move_tail(&root
->ordered_root
,
4196 &fs_info
->ordered_roots
);
4198 spin_unlock(&fs_info
->ordered_root_lock
);
4199 btrfs_destroy_ordered_extents(root
);
4202 spin_lock(&fs_info
->ordered_root_lock
);
4204 spin_unlock(&fs_info
->ordered_root_lock
);
4207 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4208 struct btrfs_root
*root
)
4210 struct rb_node
*node
;
4211 struct btrfs_delayed_ref_root
*delayed_refs
;
4212 struct btrfs_delayed_ref_node
*ref
;
4215 delayed_refs
= &trans
->delayed_refs
;
4217 spin_lock(&delayed_refs
->lock
);
4218 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4219 spin_unlock(&delayed_refs
->lock
);
4220 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4224 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4225 struct btrfs_delayed_ref_head
*head
;
4226 struct btrfs_delayed_ref_node
*tmp
;
4227 bool pin_bytes
= false;
4229 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4231 if (!mutex_trylock(&head
->mutex
)) {
4232 atomic_inc(&head
->node
.refs
);
4233 spin_unlock(&delayed_refs
->lock
);
4235 mutex_lock(&head
->mutex
);
4236 mutex_unlock(&head
->mutex
);
4237 btrfs_put_delayed_ref(&head
->node
);
4238 spin_lock(&delayed_refs
->lock
);
4241 spin_lock(&head
->lock
);
4242 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4245 list_del(&ref
->list
);
4246 atomic_dec(&delayed_refs
->num_entries
);
4247 btrfs_put_delayed_ref(ref
);
4249 if (head
->must_insert_reserved
)
4251 btrfs_free_delayed_extent_op(head
->extent_op
);
4252 delayed_refs
->num_heads
--;
4253 if (head
->processing
== 0)
4254 delayed_refs
->num_heads_ready
--;
4255 atomic_dec(&delayed_refs
->num_entries
);
4256 head
->node
.in_tree
= 0;
4257 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4258 spin_unlock(&head
->lock
);
4259 spin_unlock(&delayed_refs
->lock
);
4260 mutex_unlock(&head
->mutex
);
4263 btrfs_pin_extent(root
, head
->node
.bytenr
,
4264 head
->node
.num_bytes
, 1);
4265 btrfs_put_delayed_ref(&head
->node
);
4267 spin_lock(&delayed_refs
->lock
);
4270 spin_unlock(&delayed_refs
->lock
);
4275 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4277 struct btrfs_inode
*btrfs_inode
;
4278 struct list_head splice
;
4280 INIT_LIST_HEAD(&splice
);
4282 spin_lock(&root
->delalloc_lock
);
4283 list_splice_init(&root
->delalloc_inodes
, &splice
);
4285 while (!list_empty(&splice
)) {
4286 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4289 list_del_init(&btrfs_inode
->delalloc_inodes
);
4290 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4291 &btrfs_inode
->runtime_flags
);
4292 spin_unlock(&root
->delalloc_lock
);
4294 btrfs_invalidate_inodes(btrfs_inode
->root
);
4296 spin_lock(&root
->delalloc_lock
);
4299 spin_unlock(&root
->delalloc_lock
);
4302 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4304 struct btrfs_root
*root
;
4305 struct list_head splice
;
4307 INIT_LIST_HEAD(&splice
);
4309 spin_lock(&fs_info
->delalloc_root_lock
);
4310 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4311 while (!list_empty(&splice
)) {
4312 root
= list_first_entry(&splice
, struct btrfs_root
,
4314 list_del_init(&root
->delalloc_root
);
4315 root
= btrfs_grab_fs_root(root
);
4317 spin_unlock(&fs_info
->delalloc_root_lock
);
4319 btrfs_destroy_delalloc_inodes(root
);
4320 btrfs_put_fs_root(root
);
4322 spin_lock(&fs_info
->delalloc_root_lock
);
4324 spin_unlock(&fs_info
->delalloc_root_lock
);
4327 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4328 struct extent_io_tree
*dirty_pages
,
4332 struct extent_buffer
*eb
;
4337 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4342 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4343 while (start
<= end
) {
4344 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4345 start
+= root
->nodesize
;
4348 wait_on_extent_buffer_writeback(eb
);
4350 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4352 clear_extent_buffer_dirty(eb
);
4353 free_extent_buffer_stale(eb
);
4360 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4361 struct extent_io_tree
*pinned_extents
)
4363 struct extent_io_tree
*unpin
;
4369 unpin
= pinned_extents
;
4372 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4373 EXTENT_DIRTY
, NULL
);
4377 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4378 btrfs_error_unpin_extent_range(root
, start
, end
);
4383 if (unpin
== &root
->fs_info
->freed_extents
[0])
4384 unpin
= &root
->fs_info
->freed_extents
[1];
4386 unpin
= &root
->fs_info
->freed_extents
[0];
4394 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4395 struct btrfs_root
*root
)
4397 btrfs_destroy_delayed_refs(cur_trans
, root
);
4399 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4400 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4402 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4403 wake_up(&root
->fs_info
->transaction_wait
);
4405 btrfs_destroy_delayed_inodes(root
);
4406 btrfs_assert_delayed_root_empty(root
);
4408 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4410 btrfs_destroy_pinned_extent(root
,
4411 root
->fs_info
->pinned_extents
);
4413 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4414 wake_up(&cur_trans
->commit_wait
);
4417 memset(cur_trans, 0, sizeof(*cur_trans));
4418 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4422 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4424 struct btrfs_transaction
*t
;
4426 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4428 spin_lock(&root
->fs_info
->trans_lock
);
4429 while (!list_empty(&root
->fs_info
->trans_list
)) {
4430 t
= list_first_entry(&root
->fs_info
->trans_list
,
4431 struct btrfs_transaction
, list
);
4432 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4433 atomic_inc(&t
->use_count
);
4434 spin_unlock(&root
->fs_info
->trans_lock
);
4435 btrfs_wait_for_commit(root
, t
->transid
);
4436 btrfs_put_transaction(t
);
4437 spin_lock(&root
->fs_info
->trans_lock
);
4440 if (t
== root
->fs_info
->running_transaction
) {
4441 t
->state
= TRANS_STATE_COMMIT_DOING
;
4442 spin_unlock(&root
->fs_info
->trans_lock
);
4444 * We wait for 0 num_writers since we don't hold a trans
4445 * handle open currently for this transaction.
4447 wait_event(t
->writer_wait
,
4448 atomic_read(&t
->num_writers
) == 0);
4450 spin_unlock(&root
->fs_info
->trans_lock
);
4452 btrfs_cleanup_one_transaction(t
, root
);
4454 spin_lock(&root
->fs_info
->trans_lock
);
4455 if (t
== root
->fs_info
->running_transaction
)
4456 root
->fs_info
->running_transaction
= NULL
;
4457 list_del_init(&t
->list
);
4458 spin_unlock(&root
->fs_info
->trans_lock
);
4460 btrfs_put_transaction(t
);
4461 trace_btrfs_transaction_commit(root
);
4462 spin_lock(&root
->fs_info
->trans_lock
);
4464 spin_unlock(&root
->fs_info
->trans_lock
);
4465 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4466 btrfs_destroy_delayed_inodes(root
);
4467 btrfs_assert_delayed_root_empty(root
);
4468 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4469 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4470 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4475 static const struct extent_io_ops btree_extent_io_ops
= {
4476 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4477 .readpage_io_failed_hook
= btree_io_failed_hook
,
4478 .submit_bio_hook
= btree_submit_bio_hook
,
4479 /* note we're sharing with inode.c for the merge bio hook */
4480 .merge_bio_hook
= btrfs_merge_bio_hook
,