2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
55 #include <asm/cpufeature.h>
58 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
64 static const struct extent_io_ops btree_extent_io_ops
;
65 static void end_workqueue_fn(struct btrfs_work
*work
);
66 static void free_fs_root(struct btrfs_root
*root
);
67 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
69 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
71 struct btrfs_root
*root
);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
73 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
74 struct extent_io_tree
*dirty_pages
,
76 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
77 struct extent_io_tree
*pinned_extents
);
78 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
79 static void btrfs_error_commit_super(struct btrfs_root
*root
);
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_end_io_wq
{
90 struct btrfs_fs_info
*info
;
92 enum btrfs_wq_endio_type metadata
;
93 struct list_head list
;
94 struct btrfs_work work
;
97 static struct kmem_cache
*btrfs_end_io_wq_cache
;
99 int __init
btrfs_end_io_wq_init(void)
101 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq
),
104 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
106 if (!btrfs_end_io_wq_cache
)
111 void btrfs_end_io_wq_exit(void)
113 kmem_cache_destroy(btrfs_end_io_wq_cache
);
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
121 struct async_submit_bio
{
124 struct list_head list
;
125 extent_submit_bio_hook_t
*submit_bio_start
;
126 extent_submit_bio_hook_t
*submit_bio_done
;
129 unsigned long bio_flags
;
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
135 struct btrfs_work work
;
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
162 #ifdef CONFIG_DEBUG_LOCK_ALLOC
163 # if BTRFS_MAX_LEVEL != 8
167 static struct btrfs_lockdep_keyset
{
168 u64 id
; /* root objectid */
169 const char *name_stem
; /* lock name stem */
170 char names
[BTRFS_MAX_LEVEL
+ 1][20];
171 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
172 } btrfs_lockdep_keysets
[] = {
173 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
174 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
175 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
176 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
177 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
178 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
179 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
180 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
181 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
182 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
183 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
184 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
185 { .id
= 0, .name_stem
= "tree" },
188 void __init
btrfs_init_lockdep(void)
192 /* initialize lockdep class names */
193 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
194 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
196 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
197 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
198 "btrfs-%s-%02d", ks
->name_stem
, j
);
202 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
205 struct btrfs_lockdep_keyset
*ks
;
207 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
211 if (ks
->id
== objectid
)
214 lockdep_set_class_and_name(&eb
->lock
,
215 &ks
->keys
[level
], ks
->names
[level
]);
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
224 static struct extent_map
*btree_get_extent(struct inode
*inode
,
225 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
228 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
229 struct extent_map
*em
;
232 read_lock(&em_tree
->lock
);
233 em
= lookup_extent_mapping(em_tree
, start
, len
);
236 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
237 read_unlock(&em_tree
->lock
);
240 read_unlock(&em_tree
->lock
);
242 em
= alloc_extent_map();
244 em
= ERR_PTR(-ENOMEM
);
249 em
->block_len
= (u64
)-1;
251 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
253 write_lock(&em_tree
->lock
);
254 ret
= add_extent_mapping(em_tree
, em
, 0);
255 if (ret
== -EEXIST
) {
257 em
= lookup_extent_mapping(em_tree
, start
, len
);
264 write_unlock(&em_tree
->lock
);
270 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
272 return btrfs_crc32c(seed
, data
, len
);
275 void btrfs_csum_final(u32 crc
, char *result
)
277 put_unaligned_le32(~crc
, result
);
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
284 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
285 struct extent_buffer
*buf
,
288 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
291 unsigned long cur_len
;
292 unsigned long offset
= BTRFS_CSUM_SIZE
;
294 unsigned long map_start
;
295 unsigned long map_len
;
298 unsigned long inline_result
;
300 len
= buf
->len
- offset
;
302 err
= map_private_extent_buffer(buf
, offset
, 32,
303 &kaddr
, &map_start
, &map_len
);
306 cur_len
= min(len
, map_len
- (offset
- map_start
));
307 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
312 if (csum_size
> sizeof(inline_result
)) {
313 result
= kzalloc(csum_size
, GFP_NOFS
);
317 result
= (char *)&inline_result
;
320 btrfs_csum_final(crc
, result
);
323 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
326 memcpy(&found
, result
, csum_size
);
328 read_extent_buffer(buf
, &val
, 0, csum_size
);
329 btrfs_warn_rl(fs_info
,
330 "%s checksum verify failed on %llu wanted %X found %X "
332 fs_info
->sb
->s_id
, buf
->start
,
333 val
, found
, btrfs_header_level(buf
));
334 if (result
!= (char *)&inline_result
)
339 write_extent_buffer(buf
, result
, 0, csum_size
);
341 if (result
!= (char *)&inline_result
)
347 * we can't consider a given block up to date unless the transid of the
348 * block matches the transid in the parent node's pointer. This is how we
349 * detect blocks that either didn't get written at all or got written
350 * in the wrong place.
352 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
353 struct extent_buffer
*eb
, u64 parent_transid
,
356 struct extent_state
*cached_state
= NULL
;
358 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
360 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
367 btrfs_tree_read_lock(eb
);
368 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
371 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
373 if (extent_buffer_uptodate(eb
) &&
374 btrfs_header_generation(eb
) == parent_transid
) {
378 btrfs_err_rl(eb
->fs_info
,
379 "parent transid verify failed on %llu wanted %llu found %llu",
381 parent_transid
, btrfs_header_generation(eb
));
385 * Things reading via commit roots that don't have normal protection,
386 * like send, can have a really old block in cache that may point at a
387 * block that has been freed and re-allocated. So don't clear uptodate
388 * if we find an eb that is under IO (dirty/writeback) because we could
389 * end up reading in the stale data and then writing it back out and
390 * making everybody very sad.
392 if (!extent_buffer_under_io(eb
))
393 clear_extent_buffer_uptodate(eb
);
395 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
396 &cached_state
, GFP_NOFS
);
398 btrfs_tree_read_unlock_blocking(eb
);
403 * Return 0 if the superblock checksum type matches the checksum value of that
404 * algorithm. Pass the raw disk superblock data.
406 static int btrfs_check_super_csum(char *raw_disk_sb
)
408 struct btrfs_super_block
*disk_sb
=
409 (struct btrfs_super_block
*)raw_disk_sb
;
410 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
413 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
415 const int csum_size
= sizeof(crc
);
416 char result
[csum_size
];
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checksum.
423 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
424 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
425 btrfs_csum_final(crc
, result
);
427 if (memcmp(raw_disk_sb
, result
, csum_size
))
431 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
432 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
444 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
445 struct extent_buffer
*eb
,
446 u64 start
, u64 parent_transid
)
448 struct extent_io_tree
*io_tree
;
453 int failed_mirror
= 0;
455 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
456 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
458 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
460 btree_get_extent
, mirror_num
);
462 if (!verify_parent_transid(io_tree
, eb
,
470 * This buffer's crc is fine, but its contents are corrupted, so
471 * there is no reason to read the other copies, they won't be
474 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
477 num_copies
= btrfs_num_copies(root
->fs_info
,
482 if (!failed_mirror
) {
484 failed_mirror
= eb
->read_mirror
;
488 if (mirror_num
== failed_mirror
)
491 if (mirror_num
> num_copies
)
495 if (failed
&& !ret
&& failed_mirror
)
496 repair_eb_io_failure(root
, eb
, failed_mirror
);
502 * checksum a dirty tree block before IO. This has extra checks to make sure
503 * we only fill in the checksum field in the first page of a multi-page block
506 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
508 u64 start
= page_offset(page
);
510 struct extent_buffer
*eb
;
512 eb
= (struct extent_buffer
*)page
->private;
513 if (page
!= eb
->pages
[0])
516 found_start
= btrfs_header_bytenr(eb
);
518 * Please do not consolidate these warnings into a single if.
519 * It is useful to know what went wrong.
521 if (WARN_ON(found_start
!= start
))
523 if (WARN_ON(!PageUptodate(page
)))
526 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fsid
,
527 btrfs_header_fsid(), BTRFS_FSID_SIZE
) == 0);
529 return csum_tree_block(fs_info
, eb
, 0);
532 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
533 struct extent_buffer
*eb
)
535 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
536 u8 fsid
[BTRFS_UUID_SIZE
];
539 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
541 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
545 fs_devices
= fs_devices
->seed
;
550 #define CORRUPT(reason, eb, root, slot) \
551 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
552 "root=%llu, slot=%d", reason, \
553 btrfs_header_bytenr(eb), root->objectid, slot)
555 static noinline
int check_leaf(struct btrfs_root
*root
,
556 struct extent_buffer
*leaf
)
558 struct btrfs_key key
;
559 struct btrfs_key leaf_key
;
560 u32 nritems
= btrfs_header_nritems(leaf
);
566 /* Check the 0 item */
567 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
568 BTRFS_LEAF_DATA_SIZE(root
)) {
569 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
574 * Check to make sure each items keys are in the correct order and their
575 * offsets make sense. We only have to loop through nritems-1 because
576 * we check the current slot against the next slot, which verifies the
577 * next slot's offset+size makes sense and that the current's slot
580 for (slot
= 0; slot
< nritems
- 1; slot
++) {
581 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
582 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
584 /* Make sure the keys are in the right order */
585 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
586 CORRUPT("bad key order", leaf
, root
, slot
);
591 * Make sure the offset and ends are right, remember that the
592 * item data starts at the end of the leaf and grows towards the
595 if (btrfs_item_offset_nr(leaf
, slot
) !=
596 btrfs_item_end_nr(leaf
, slot
+ 1)) {
597 CORRUPT("slot offset bad", leaf
, root
, slot
);
602 * Check to make sure that we don't point outside of the leaf,
603 * just in case all the items are consistent to each other, but
604 * all point outside of the leaf.
606 if (btrfs_item_end_nr(leaf
, slot
) >
607 BTRFS_LEAF_DATA_SIZE(root
)) {
608 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
616 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
617 u64 phy_offset
, struct page
*page
,
618 u64 start
, u64 end
, int mirror
)
622 struct extent_buffer
*eb
;
623 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
624 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
631 eb
= (struct extent_buffer
*)page
->private;
633 /* the pending IO might have been the only thing that kept this buffer
634 * in memory. Make sure we have a ref for all this other checks
636 extent_buffer_get(eb
);
638 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
642 eb
->read_mirror
= mirror
;
643 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
648 found_start
= btrfs_header_bytenr(eb
);
649 if (found_start
!= eb
->start
) {
650 btrfs_err_rl(fs_info
, "bad tree block start %llu %llu",
651 found_start
, eb
->start
);
655 if (check_tree_block_fsid(fs_info
, eb
)) {
656 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
661 found_level
= btrfs_header_level(eb
);
662 if (found_level
>= BTRFS_MAX_LEVEL
) {
663 btrfs_err(fs_info
, "bad tree block level %d",
664 (int)btrfs_header_level(eb
));
669 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
672 ret
= csum_tree_block(fs_info
, eb
, 1);
677 * If this is a leaf block and it is corrupt, set the corrupt bit so
678 * that we don't try and read the other copies of this block, just
681 if (found_level
== 0 && check_leaf(root
, eb
)) {
682 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
687 set_extent_buffer_uptodate(eb
);
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
691 btree_readahead_hook(fs_info
, eb
, eb
->start
, ret
);
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
699 atomic_inc(&eb
->io_pages
);
700 clear_extent_buffer_uptodate(eb
);
702 free_extent_buffer(eb
);
707 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
709 struct extent_buffer
*eb
;
711 eb
= (struct extent_buffer
*)page
->private;
712 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
713 eb
->read_mirror
= failed_mirror
;
714 atomic_dec(&eb
->io_pages
);
715 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
716 btree_readahead_hook(eb
->fs_info
, eb
, eb
->start
, -EIO
);
717 return -EIO
; /* we fixed nothing */
720 static void end_workqueue_bio(struct bio
*bio
)
722 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
723 struct btrfs_fs_info
*fs_info
;
724 struct btrfs_workqueue
*wq
;
725 btrfs_work_func_t func
;
727 fs_info
= end_io_wq
->info
;
728 end_io_wq
->error
= bio
->bi_error
;
730 if (bio
->bi_rw
& REQ_WRITE
) {
731 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
732 wq
= fs_info
->endio_meta_write_workers
;
733 func
= btrfs_endio_meta_write_helper
;
734 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
735 wq
= fs_info
->endio_freespace_worker
;
736 func
= btrfs_freespace_write_helper
;
737 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
738 wq
= fs_info
->endio_raid56_workers
;
739 func
= btrfs_endio_raid56_helper
;
741 wq
= fs_info
->endio_write_workers
;
742 func
= btrfs_endio_write_helper
;
745 if (unlikely(end_io_wq
->metadata
==
746 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
747 wq
= fs_info
->endio_repair_workers
;
748 func
= btrfs_endio_repair_helper
;
749 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
750 wq
= fs_info
->endio_raid56_workers
;
751 func
= btrfs_endio_raid56_helper
;
752 } else if (end_io_wq
->metadata
) {
753 wq
= fs_info
->endio_meta_workers
;
754 func
= btrfs_endio_meta_helper
;
756 wq
= fs_info
->endio_workers
;
757 func
= btrfs_endio_helper
;
761 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
762 btrfs_queue_work(wq
, &end_io_wq
->work
);
765 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
766 enum btrfs_wq_endio_type metadata
)
768 struct btrfs_end_io_wq
*end_io_wq
;
770 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
774 end_io_wq
->private = bio
->bi_private
;
775 end_io_wq
->end_io
= bio
->bi_end_io
;
776 end_io_wq
->info
= info
;
777 end_io_wq
->error
= 0;
778 end_io_wq
->bio
= bio
;
779 end_io_wq
->metadata
= metadata
;
781 bio
->bi_private
= end_io_wq
;
782 bio
->bi_end_io
= end_workqueue_bio
;
786 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
788 unsigned long limit
= min_t(unsigned long,
789 info
->thread_pool_size
,
790 info
->fs_devices
->open_devices
);
794 static void run_one_async_start(struct btrfs_work
*work
)
796 struct async_submit_bio
*async
;
799 async
= container_of(work
, struct async_submit_bio
, work
);
800 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
801 async
->mirror_num
, async
->bio_flags
,
807 static void run_one_async_done(struct btrfs_work
*work
)
809 struct btrfs_fs_info
*fs_info
;
810 struct async_submit_bio
*async
;
813 async
= container_of(work
, struct async_submit_bio
, work
);
814 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
816 limit
= btrfs_async_submit_limit(fs_info
);
817 limit
= limit
* 2 / 3;
820 * atomic_dec_return implies a barrier for waitqueue_active
822 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
823 waitqueue_active(&fs_info
->async_submit_wait
))
824 wake_up(&fs_info
->async_submit_wait
);
826 /* If an error occurred we just want to clean up the bio and move on */
828 async
->bio
->bi_error
= async
->error
;
829 bio_endio(async
->bio
);
833 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
834 async
->mirror_num
, async
->bio_flags
,
838 static void run_one_async_free(struct btrfs_work
*work
)
840 struct async_submit_bio
*async
;
842 async
= container_of(work
, struct async_submit_bio
, work
);
846 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
847 int rw
, struct bio
*bio
, int mirror_num
,
848 unsigned long bio_flags
,
850 extent_submit_bio_hook_t
*submit_bio_start
,
851 extent_submit_bio_hook_t
*submit_bio_done
)
853 struct async_submit_bio
*async
;
855 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
859 async
->inode
= inode
;
862 async
->mirror_num
= mirror_num
;
863 async
->submit_bio_start
= submit_bio_start
;
864 async
->submit_bio_done
= submit_bio_done
;
866 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
867 run_one_async_done
, run_one_async_free
);
869 async
->bio_flags
= bio_flags
;
870 async
->bio_offset
= bio_offset
;
874 atomic_inc(&fs_info
->nr_async_submits
);
877 btrfs_set_work_high_priority(&async
->work
);
879 btrfs_queue_work(fs_info
->workers
, &async
->work
);
881 while (atomic_read(&fs_info
->async_submit_draining
) &&
882 atomic_read(&fs_info
->nr_async_submits
)) {
883 wait_event(fs_info
->async_submit_wait
,
884 (atomic_read(&fs_info
->nr_async_submits
) == 0));
890 static int btree_csum_one_bio(struct bio
*bio
)
892 struct bio_vec
*bvec
;
893 struct btrfs_root
*root
;
896 bio_for_each_segment_all(bvec
, bio
, i
) {
897 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
898 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
906 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
907 struct bio
*bio
, int mirror_num
,
908 unsigned long bio_flags
,
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
915 return btree_csum_one_bio(bio
);
918 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
919 int mirror_num
, unsigned long bio_flags
,
925 * when we're called for a write, we're already in the async
926 * submission context. Just jump into btrfs_map_bio
928 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
936 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
938 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
941 if (static_cpu_has(X86_FEATURE_XMM4_2
))
947 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
948 int mirror_num
, unsigned long bio_flags
,
951 int async
= check_async_write(inode
, bio_flags
);
954 if (!(rw
& REQ_WRITE
)) {
956 * called for a read, do the setup so that checksum validation
957 * can happen in the async kernel threads
959 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
960 bio
, BTRFS_WQ_ENDIO_METADATA
);
963 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
966 ret
= btree_csum_one_bio(bio
);
969 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
973 * kthread helpers are used to submit writes so that
974 * checksumming can happen in parallel across all CPUs
976 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
977 inode
, rw
, bio
, mirror_num
, 0,
979 __btree_submit_bio_start
,
980 __btree_submit_bio_done
);
993 #ifdef CONFIG_MIGRATION
994 static int btree_migratepage(struct address_space
*mapping
,
995 struct page
*newpage
, struct page
*page
,
996 enum migrate_mode mode
)
999 * we can't safely write a btree page from here,
1000 * we haven't done the locking hook
1002 if (PageDirty(page
))
1005 * Buffers may be managed in a filesystem specific way.
1006 * We must have no buffers or drop them.
1008 if (page_has_private(page
) &&
1009 !try_to_release_page(page
, GFP_KERNEL
))
1011 return migrate_page(mapping
, newpage
, page
, mode
);
1016 static int btree_writepages(struct address_space
*mapping
,
1017 struct writeback_control
*wbc
)
1019 struct btrfs_fs_info
*fs_info
;
1022 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1024 if (wbc
->for_kupdate
)
1027 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1028 /* this is a bit racy, but that's ok */
1029 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1030 BTRFS_DIRTY_METADATA_THRESH
);
1034 return btree_write_cache_pages(mapping
, wbc
);
1037 static int btree_readpage(struct file
*file
, struct page
*page
)
1039 struct extent_io_tree
*tree
;
1040 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1041 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1044 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1046 if (PageWriteback(page
) || PageDirty(page
))
1049 return try_release_extent_buffer(page
);
1052 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1053 unsigned int length
)
1055 struct extent_io_tree
*tree
;
1056 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1057 extent_invalidatepage(tree
, page
, offset
);
1058 btree_releasepage(page
, GFP_NOFS
);
1059 if (PagePrivate(page
)) {
1060 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1061 "page private not zero on page %llu",
1062 (unsigned long long)page_offset(page
));
1063 ClearPagePrivate(page
);
1064 set_page_private(page
, 0);
1069 static int btree_set_page_dirty(struct page
*page
)
1072 struct extent_buffer
*eb
;
1074 BUG_ON(!PagePrivate(page
));
1075 eb
= (struct extent_buffer
*)page
->private;
1077 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1078 BUG_ON(!atomic_read(&eb
->refs
));
1079 btrfs_assert_tree_locked(eb
);
1081 return __set_page_dirty_nobuffers(page
);
1084 static const struct address_space_operations btree_aops
= {
1085 .readpage
= btree_readpage
,
1086 .writepages
= btree_writepages
,
1087 .releasepage
= btree_releasepage
,
1088 .invalidatepage
= btree_invalidatepage
,
1089 #ifdef CONFIG_MIGRATION
1090 .migratepage
= btree_migratepage
,
1092 .set_page_dirty
= btree_set_page_dirty
,
1095 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1097 struct extent_buffer
*buf
= NULL
;
1098 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1100 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1103 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1104 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1105 free_extent_buffer(buf
);
1108 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1109 int mirror_num
, struct extent_buffer
**eb
)
1111 struct extent_buffer
*buf
= NULL
;
1112 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1113 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1116 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1120 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1122 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1123 btree_get_extent
, mirror_num
);
1125 free_extent_buffer(buf
);
1129 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1130 free_extent_buffer(buf
);
1132 } else if (extent_buffer_uptodate(buf
)) {
1135 free_extent_buffer(buf
);
1140 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1143 return find_extent_buffer(fs_info
, bytenr
);
1146 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1149 if (btrfs_test_is_dummy_root(root
))
1150 return alloc_test_extent_buffer(root
->fs_info
, bytenr
,
1152 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1156 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1158 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1159 buf
->start
+ buf
->len
- 1);
1162 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1164 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1165 buf
->start
, buf
->start
+ buf
->len
- 1);
1168 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1171 struct extent_buffer
*buf
= NULL
;
1174 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1176 return ERR_PTR(-ENOMEM
);
1178 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1180 free_extent_buffer(buf
);
1181 return ERR_PTR(ret
);
1187 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1188 struct btrfs_fs_info
*fs_info
,
1189 struct extent_buffer
*buf
)
1191 if (btrfs_header_generation(buf
) ==
1192 fs_info
->running_transaction
->transid
) {
1193 btrfs_assert_tree_locked(buf
);
1195 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1196 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1198 fs_info
->dirty_metadata_batch
);
1199 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1200 btrfs_set_lock_blocking(buf
);
1201 clear_extent_buffer_dirty(buf
);
1206 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1208 struct btrfs_subvolume_writers
*writers
;
1211 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1213 return ERR_PTR(-ENOMEM
);
1215 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1218 return ERR_PTR(ret
);
1221 init_waitqueue_head(&writers
->wait
);
1226 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1228 percpu_counter_destroy(&writers
->counter
);
1232 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1233 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1237 root
->commit_root
= NULL
;
1238 root
->sectorsize
= sectorsize
;
1239 root
->nodesize
= nodesize
;
1240 root
->stripesize
= stripesize
;
1242 root
->orphan_cleanup_state
= 0;
1244 root
->objectid
= objectid
;
1245 root
->last_trans
= 0;
1246 root
->highest_objectid
= 0;
1247 root
->nr_delalloc_inodes
= 0;
1248 root
->nr_ordered_extents
= 0;
1250 root
->inode_tree
= RB_ROOT
;
1251 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1252 root
->block_rsv
= NULL
;
1253 root
->orphan_block_rsv
= NULL
;
1255 INIT_LIST_HEAD(&root
->dirty_list
);
1256 INIT_LIST_HEAD(&root
->root_list
);
1257 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1258 INIT_LIST_HEAD(&root
->delalloc_root
);
1259 INIT_LIST_HEAD(&root
->ordered_extents
);
1260 INIT_LIST_HEAD(&root
->ordered_root
);
1261 INIT_LIST_HEAD(&root
->logged_list
[0]);
1262 INIT_LIST_HEAD(&root
->logged_list
[1]);
1263 spin_lock_init(&root
->orphan_lock
);
1264 spin_lock_init(&root
->inode_lock
);
1265 spin_lock_init(&root
->delalloc_lock
);
1266 spin_lock_init(&root
->ordered_extent_lock
);
1267 spin_lock_init(&root
->accounting_lock
);
1268 spin_lock_init(&root
->log_extents_lock
[0]);
1269 spin_lock_init(&root
->log_extents_lock
[1]);
1270 mutex_init(&root
->objectid_mutex
);
1271 mutex_init(&root
->log_mutex
);
1272 mutex_init(&root
->ordered_extent_mutex
);
1273 mutex_init(&root
->delalloc_mutex
);
1274 init_waitqueue_head(&root
->log_writer_wait
);
1275 init_waitqueue_head(&root
->log_commit_wait
[0]);
1276 init_waitqueue_head(&root
->log_commit_wait
[1]);
1277 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1278 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1279 atomic_set(&root
->log_commit
[0], 0);
1280 atomic_set(&root
->log_commit
[1], 0);
1281 atomic_set(&root
->log_writers
, 0);
1282 atomic_set(&root
->log_batch
, 0);
1283 atomic_set(&root
->orphan_inodes
, 0);
1284 atomic_set(&root
->refs
, 1);
1285 atomic_set(&root
->will_be_snapshoted
, 0);
1286 atomic_set(&root
->qgroup_meta_rsv
, 0);
1287 root
->log_transid
= 0;
1288 root
->log_transid_committed
= -1;
1289 root
->last_log_commit
= 0;
1291 extent_io_tree_init(&root
->dirty_log_pages
,
1292 fs_info
->btree_inode
->i_mapping
);
1294 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1295 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1296 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1298 root
->defrag_trans_start
= fs_info
->generation
;
1300 root
->defrag_trans_start
= 0;
1301 root
->root_key
.objectid
= objectid
;
1304 spin_lock_init(&root
->root_item_lock
);
1307 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1310 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1312 root
->fs_info
= fs_info
;
1316 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1317 /* Should only be used by the testing infrastructure */
1318 struct btrfs_root
*btrfs_alloc_dummy_root(u32 sectorsize
, u32 nodesize
)
1320 struct btrfs_root
*root
;
1322 root
= btrfs_alloc_root(NULL
, GFP_KERNEL
);
1324 return ERR_PTR(-ENOMEM
);
1325 /* We don't use the stripesize in selftest, set it as sectorsize */
1326 __setup_root(nodesize
, sectorsize
, sectorsize
, root
, NULL
,
1327 BTRFS_ROOT_TREE_OBJECTID
);
1328 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1329 root
->alloc_bytenr
= 0;
1335 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1336 struct btrfs_fs_info
*fs_info
,
1339 struct extent_buffer
*leaf
;
1340 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1341 struct btrfs_root
*root
;
1342 struct btrfs_key key
;
1346 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1348 return ERR_PTR(-ENOMEM
);
1350 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1351 tree_root
->stripesize
, root
, fs_info
, objectid
);
1352 root
->root_key
.objectid
= objectid
;
1353 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1354 root
->root_key
.offset
= 0;
1356 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1358 ret
= PTR_ERR(leaf
);
1363 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1364 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1365 btrfs_set_header_generation(leaf
, trans
->transid
);
1366 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1367 btrfs_set_header_owner(leaf
, objectid
);
1370 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1372 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1373 btrfs_header_chunk_tree_uuid(leaf
),
1375 btrfs_mark_buffer_dirty(leaf
);
1377 root
->commit_root
= btrfs_root_node(root
);
1378 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1380 root
->root_item
.flags
= 0;
1381 root
->root_item
.byte_limit
= 0;
1382 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1383 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1384 btrfs_set_root_level(&root
->root_item
, 0);
1385 btrfs_set_root_refs(&root
->root_item
, 1);
1386 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1387 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1388 btrfs_set_root_dirid(&root
->root_item
, 0);
1390 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1391 root
->root_item
.drop_level
= 0;
1393 key
.objectid
= objectid
;
1394 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1396 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1400 btrfs_tree_unlock(leaf
);
1406 btrfs_tree_unlock(leaf
);
1407 free_extent_buffer(root
->commit_root
);
1408 free_extent_buffer(leaf
);
1412 return ERR_PTR(ret
);
1415 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1416 struct btrfs_fs_info
*fs_info
)
1418 struct btrfs_root
*root
;
1419 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1420 struct extent_buffer
*leaf
;
1422 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1424 return ERR_PTR(-ENOMEM
);
1426 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1427 tree_root
->stripesize
, root
, fs_info
,
1428 BTRFS_TREE_LOG_OBJECTID
);
1430 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1431 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1432 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1435 * DON'T set REF_COWS for log trees
1437 * log trees do not get reference counted because they go away
1438 * before a real commit is actually done. They do store pointers
1439 * to file data extents, and those reference counts still get
1440 * updated (along with back refs to the log tree).
1443 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1447 return ERR_CAST(leaf
);
1450 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1451 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1452 btrfs_set_header_generation(leaf
, trans
->transid
);
1453 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1454 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1457 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1458 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1459 btrfs_mark_buffer_dirty(root
->node
);
1460 btrfs_tree_unlock(root
->node
);
1464 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1465 struct btrfs_fs_info
*fs_info
)
1467 struct btrfs_root
*log_root
;
1469 log_root
= alloc_log_tree(trans
, fs_info
);
1470 if (IS_ERR(log_root
))
1471 return PTR_ERR(log_root
);
1472 WARN_ON(fs_info
->log_root_tree
);
1473 fs_info
->log_root_tree
= log_root
;
1477 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1478 struct btrfs_root
*root
)
1480 struct btrfs_root
*log_root
;
1481 struct btrfs_inode_item
*inode_item
;
1483 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1484 if (IS_ERR(log_root
))
1485 return PTR_ERR(log_root
);
1487 log_root
->last_trans
= trans
->transid
;
1488 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1490 inode_item
= &log_root
->root_item
.inode
;
1491 btrfs_set_stack_inode_generation(inode_item
, 1);
1492 btrfs_set_stack_inode_size(inode_item
, 3);
1493 btrfs_set_stack_inode_nlink(inode_item
, 1);
1494 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1495 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1497 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1499 WARN_ON(root
->log_root
);
1500 root
->log_root
= log_root
;
1501 root
->log_transid
= 0;
1502 root
->log_transid_committed
= -1;
1503 root
->last_log_commit
= 0;
1507 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1508 struct btrfs_key
*key
)
1510 struct btrfs_root
*root
;
1511 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1512 struct btrfs_path
*path
;
1516 path
= btrfs_alloc_path();
1518 return ERR_PTR(-ENOMEM
);
1520 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1526 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1527 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1529 ret
= btrfs_find_root(tree_root
, key
, path
,
1530 &root
->root_item
, &root
->root_key
);
1537 generation
= btrfs_root_generation(&root
->root_item
);
1538 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1540 if (IS_ERR(root
->node
)) {
1541 ret
= PTR_ERR(root
->node
);
1543 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1545 free_extent_buffer(root
->node
);
1548 root
->commit_root
= btrfs_root_node(root
);
1550 btrfs_free_path(path
);
1556 root
= ERR_PTR(ret
);
1560 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1561 struct btrfs_key
*location
)
1563 struct btrfs_root
*root
;
1565 root
= btrfs_read_tree_root(tree_root
, location
);
1569 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1570 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1571 btrfs_check_and_init_root_item(&root
->root_item
);
1577 int btrfs_init_fs_root(struct btrfs_root
*root
)
1580 struct btrfs_subvolume_writers
*writers
;
1582 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1583 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1585 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1590 writers
= btrfs_alloc_subvolume_writers();
1591 if (IS_ERR(writers
)) {
1592 ret
= PTR_ERR(writers
);
1595 root
->subv_writers
= writers
;
1597 btrfs_init_free_ino_ctl(root
);
1598 spin_lock_init(&root
->ino_cache_lock
);
1599 init_waitqueue_head(&root
->ino_cache_wait
);
1601 ret
= get_anon_bdev(&root
->anon_dev
);
1605 mutex_lock(&root
->objectid_mutex
);
1606 ret
= btrfs_find_highest_objectid(root
,
1607 &root
->highest_objectid
);
1609 mutex_unlock(&root
->objectid_mutex
);
1613 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1615 mutex_unlock(&root
->objectid_mutex
);
1620 free_anon_bdev(root
->anon_dev
);
1622 btrfs_free_subvolume_writers(root
->subv_writers
);
1624 kfree(root
->free_ino_ctl
);
1625 kfree(root
->free_ino_pinned
);
1629 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1632 struct btrfs_root
*root
;
1634 spin_lock(&fs_info
->fs_roots_radix_lock
);
1635 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1636 (unsigned long)root_id
);
1637 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1641 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1642 struct btrfs_root
*root
)
1646 ret
= radix_tree_preload(GFP_NOFS
);
1650 spin_lock(&fs_info
->fs_roots_radix_lock
);
1651 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1652 (unsigned long)root
->root_key
.objectid
,
1655 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1656 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1657 radix_tree_preload_end();
1662 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1663 struct btrfs_key
*location
,
1666 struct btrfs_root
*root
;
1667 struct btrfs_path
*path
;
1668 struct btrfs_key key
;
1671 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1672 return fs_info
->tree_root
;
1673 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1674 return fs_info
->extent_root
;
1675 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1676 return fs_info
->chunk_root
;
1677 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1678 return fs_info
->dev_root
;
1679 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1680 return fs_info
->csum_root
;
1681 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1682 return fs_info
->quota_root
? fs_info
->quota_root
:
1684 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1685 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1687 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1688 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1691 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1693 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1694 return ERR_PTR(-ENOENT
);
1698 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1702 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1707 ret
= btrfs_init_fs_root(root
);
1711 path
= btrfs_alloc_path();
1716 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1717 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1718 key
.offset
= location
->objectid
;
1720 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1721 btrfs_free_path(path
);
1725 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1727 ret
= btrfs_insert_fs_root(fs_info
, root
);
1729 if (ret
== -EEXIST
) {
1738 return ERR_PTR(ret
);
1741 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1743 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1745 struct btrfs_device
*device
;
1746 struct backing_dev_info
*bdi
;
1749 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1752 bdi
= blk_get_backing_dev_info(device
->bdev
);
1753 if (bdi_congested(bdi
, bdi_bits
)) {
1762 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1766 err
= bdi_setup_and_register(bdi
, "btrfs");
1770 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_SIZE
;
1771 bdi
->congested_fn
= btrfs_congested_fn
;
1772 bdi
->congested_data
= info
;
1773 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1778 * called by the kthread helper functions to finally call the bio end_io
1779 * functions. This is where read checksum verification actually happens
1781 static void end_workqueue_fn(struct btrfs_work
*work
)
1784 struct btrfs_end_io_wq
*end_io_wq
;
1786 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1787 bio
= end_io_wq
->bio
;
1789 bio
->bi_error
= end_io_wq
->error
;
1790 bio
->bi_private
= end_io_wq
->private;
1791 bio
->bi_end_io
= end_io_wq
->end_io
;
1792 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1796 static int cleaner_kthread(void *arg
)
1798 struct btrfs_root
*root
= arg
;
1800 struct btrfs_trans_handle
*trans
;
1805 /* Make the cleaner go to sleep early. */
1806 if (btrfs_need_cleaner_sleep(root
))
1809 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1813 * Avoid the problem that we change the status of the fs
1814 * during the above check and trylock.
1816 if (btrfs_need_cleaner_sleep(root
)) {
1817 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1821 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1822 btrfs_run_delayed_iputs(root
);
1823 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1825 again
= btrfs_clean_one_deleted_snapshot(root
);
1826 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1829 * The defragger has dealt with the R/O remount and umount,
1830 * needn't do anything special here.
1832 btrfs_run_defrag_inodes(root
->fs_info
);
1835 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1836 * with relocation (btrfs_relocate_chunk) and relocation
1837 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1838 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1839 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1840 * unused block groups.
1842 btrfs_delete_unused_bgs(root
->fs_info
);
1845 set_current_state(TASK_INTERRUPTIBLE
);
1846 if (!kthread_should_stop())
1848 __set_current_state(TASK_RUNNING
);
1850 } while (!kthread_should_stop());
1853 * Transaction kthread is stopped before us and wakes us up.
1854 * However we might have started a new transaction and COWed some
1855 * tree blocks when deleting unused block groups for example. So
1856 * make sure we commit the transaction we started to have a clean
1857 * shutdown when evicting the btree inode - if it has dirty pages
1858 * when we do the final iput() on it, eviction will trigger a
1859 * writeback for it which will fail with null pointer dereferences
1860 * since work queues and other resources were already released and
1861 * destroyed by the time the iput/eviction/writeback is made.
1863 trans
= btrfs_attach_transaction(root
);
1864 if (IS_ERR(trans
)) {
1865 if (PTR_ERR(trans
) != -ENOENT
)
1866 btrfs_err(root
->fs_info
,
1867 "cleaner transaction attach returned %ld",
1872 ret
= btrfs_commit_transaction(trans
, root
);
1874 btrfs_err(root
->fs_info
,
1875 "cleaner open transaction commit returned %d",
1882 static int transaction_kthread(void *arg
)
1884 struct btrfs_root
*root
= arg
;
1885 struct btrfs_trans_handle
*trans
;
1886 struct btrfs_transaction
*cur
;
1889 unsigned long delay
;
1893 cannot_commit
= false;
1894 delay
= HZ
* root
->fs_info
->commit_interval
;
1895 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1897 spin_lock(&root
->fs_info
->trans_lock
);
1898 cur
= root
->fs_info
->running_transaction
;
1900 spin_unlock(&root
->fs_info
->trans_lock
);
1904 now
= get_seconds();
1905 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1906 (now
< cur
->start_time
||
1907 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1908 spin_unlock(&root
->fs_info
->trans_lock
);
1912 transid
= cur
->transid
;
1913 spin_unlock(&root
->fs_info
->trans_lock
);
1915 /* If the file system is aborted, this will always fail. */
1916 trans
= btrfs_attach_transaction(root
);
1917 if (IS_ERR(trans
)) {
1918 if (PTR_ERR(trans
) != -ENOENT
)
1919 cannot_commit
= true;
1922 if (transid
== trans
->transid
) {
1923 btrfs_commit_transaction(trans
, root
);
1925 btrfs_end_transaction(trans
, root
);
1928 wake_up_process(root
->fs_info
->cleaner_kthread
);
1929 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1931 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1932 &root
->fs_info
->fs_state
)))
1933 btrfs_cleanup_transaction(root
);
1934 set_current_state(TASK_INTERRUPTIBLE
);
1935 if (!kthread_should_stop() &&
1936 (!btrfs_transaction_blocked(root
->fs_info
) ||
1938 schedule_timeout(delay
);
1939 __set_current_state(TASK_RUNNING
);
1940 } while (!kthread_should_stop());
1945 * this will find the highest generation in the array of
1946 * root backups. The index of the highest array is returned,
1947 * or -1 if we can't find anything.
1949 * We check to make sure the array is valid by comparing the
1950 * generation of the latest root in the array with the generation
1951 * in the super block. If they don't match we pitch it.
1953 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1956 int newest_index
= -1;
1957 struct btrfs_root_backup
*root_backup
;
1960 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1961 root_backup
= info
->super_copy
->super_roots
+ i
;
1962 cur
= btrfs_backup_tree_root_gen(root_backup
);
1963 if (cur
== newest_gen
)
1967 /* check to see if we actually wrapped around */
1968 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1969 root_backup
= info
->super_copy
->super_roots
;
1970 cur
= btrfs_backup_tree_root_gen(root_backup
);
1971 if (cur
== newest_gen
)
1974 return newest_index
;
1979 * find the oldest backup so we know where to store new entries
1980 * in the backup array. This will set the backup_root_index
1981 * field in the fs_info struct
1983 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1986 int newest_index
= -1;
1988 newest_index
= find_newest_super_backup(info
, newest_gen
);
1989 /* if there was garbage in there, just move along */
1990 if (newest_index
== -1) {
1991 info
->backup_root_index
= 0;
1993 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1998 * copy all the root pointers into the super backup array.
1999 * this will bump the backup pointer by one when it is
2002 static void backup_super_roots(struct btrfs_fs_info
*info
)
2005 struct btrfs_root_backup
*root_backup
;
2008 next_backup
= info
->backup_root_index
;
2009 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2010 BTRFS_NUM_BACKUP_ROOTS
;
2013 * just overwrite the last backup if we're at the same generation
2014 * this happens only at umount
2016 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
2017 if (btrfs_backup_tree_root_gen(root_backup
) ==
2018 btrfs_header_generation(info
->tree_root
->node
))
2019 next_backup
= last_backup
;
2021 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
2024 * make sure all of our padding and empty slots get zero filled
2025 * regardless of which ones we use today
2027 memset(root_backup
, 0, sizeof(*root_backup
));
2029 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
2031 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
2032 btrfs_set_backup_tree_root_gen(root_backup
,
2033 btrfs_header_generation(info
->tree_root
->node
));
2035 btrfs_set_backup_tree_root_level(root_backup
,
2036 btrfs_header_level(info
->tree_root
->node
));
2038 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2039 btrfs_set_backup_chunk_root_gen(root_backup
,
2040 btrfs_header_generation(info
->chunk_root
->node
));
2041 btrfs_set_backup_chunk_root_level(root_backup
,
2042 btrfs_header_level(info
->chunk_root
->node
));
2044 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2045 btrfs_set_backup_extent_root_gen(root_backup
,
2046 btrfs_header_generation(info
->extent_root
->node
));
2047 btrfs_set_backup_extent_root_level(root_backup
,
2048 btrfs_header_level(info
->extent_root
->node
));
2051 * we might commit during log recovery, which happens before we set
2052 * the fs_root. Make sure it is valid before we fill it in.
2054 if (info
->fs_root
&& info
->fs_root
->node
) {
2055 btrfs_set_backup_fs_root(root_backup
,
2056 info
->fs_root
->node
->start
);
2057 btrfs_set_backup_fs_root_gen(root_backup
,
2058 btrfs_header_generation(info
->fs_root
->node
));
2059 btrfs_set_backup_fs_root_level(root_backup
,
2060 btrfs_header_level(info
->fs_root
->node
));
2063 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2064 btrfs_set_backup_dev_root_gen(root_backup
,
2065 btrfs_header_generation(info
->dev_root
->node
));
2066 btrfs_set_backup_dev_root_level(root_backup
,
2067 btrfs_header_level(info
->dev_root
->node
));
2069 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2070 btrfs_set_backup_csum_root_gen(root_backup
,
2071 btrfs_header_generation(info
->csum_root
->node
));
2072 btrfs_set_backup_csum_root_level(root_backup
,
2073 btrfs_header_level(info
->csum_root
->node
));
2075 btrfs_set_backup_total_bytes(root_backup
,
2076 btrfs_super_total_bytes(info
->super_copy
));
2077 btrfs_set_backup_bytes_used(root_backup
,
2078 btrfs_super_bytes_used(info
->super_copy
));
2079 btrfs_set_backup_num_devices(root_backup
,
2080 btrfs_super_num_devices(info
->super_copy
));
2083 * if we don't copy this out to the super_copy, it won't get remembered
2084 * for the next commit
2086 memcpy(&info
->super_copy
->super_roots
,
2087 &info
->super_for_commit
->super_roots
,
2088 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2092 * this copies info out of the root backup array and back into
2093 * the in-memory super block. It is meant to help iterate through
2094 * the array, so you send it the number of backups you've already
2095 * tried and the last backup index you used.
2097 * this returns -1 when it has tried all the backups
2099 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2100 struct btrfs_super_block
*super
,
2101 int *num_backups_tried
, int *backup_index
)
2103 struct btrfs_root_backup
*root_backup
;
2104 int newest
= *backup_index
;
2106 if (*num_backups_tried
== 0) {
2107 u64 gen
= btrfs_super_generation(super
);
2109 newest
= find_newest_super_backup(info
, gen
);
2113 *backup_index
= newest
;
2114 *num_backups_tried
= 1;
2115 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2116 /* we've tried all the backups, all done */
2119 /* jump to the next oldest backup */
2120 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2121 BTRFS_NUM_BACKUP_ROOTS
;
2122 *backup_index
= newest
;
2123 *num_backups_tried
+= 1;
2125 root_backup
= super
->super_roots
+ newest
;
2127 btrfs_set_super_generation(super
,
2128 btrfs_backup_tree_root_gen(root_backup
));
2129 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2130 btrfs_set_super_root_level(super
,
2131 btrfs_backup_tree_root_level(root_backup
));
2132 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2135 * fixme: the total bytes and num_devices need to match or we should
2138 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2139 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2143 /* helper to cleanup workers */
2144 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2146 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2147 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2148 btrfs_destroy_workqueue(fs_info
->workers
);
2149 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2150 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2151 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2152 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2153 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2154 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2155 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2156 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2157 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2158 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2159 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2160 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2161 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2162 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2163 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2166 static void free_root_extent_buffers(struct btrfs_root
*root
)
2169 free_extent_buffer(root
->node
);
2170 free_extent_buffer(root
->commit_root
);
2172 root
->commit_root
= NULL
;
2176 /* helper to cleanup tree roots */
2177 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2179 free_root_extent_buffers(info
->tree_root
);
2181 free_root_extent_buffers(info
->dev_root
);
2182 free_root_extent_buffers(info
->extent_root
);
2183 free_root_extent_buffers(info
->csum_root
);
2184 free_root_extent_buffers(info
->quota_root
);
2185 free_root_extent_buffers(info
->uuid_root
);
2187 free_root_extent_buffers(info
->chunk_root
);
2188 free_root_extent_buffers(info
->free_space_root
);
2191 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2194 struct btrfs_root
*gang
[8];
2197 while (!list_empty(&fs_info
->dead_roots
)) {
2198 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2199 struct btrfs_root
, root_list
);
2200 list_del(&gang
[0]->root_list
);
2202 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2203 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2205 free_extent_buffer(gang
[0]->node
);
2206 free_extent_buffer(gang
[0]->commit_root
);
2207 btrfs_put_fs_root(gang
[0]);
2212 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2217 for (i
= 0; i
< ret
; i
++)
2218 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2221 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2222 btrfs_free_log_root_tree(NULL
, fs_info
);
2223 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2224 fs_info
->pinned_extents
);
2228 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2230 mutex_init(&fs_info
->scrub_lock
);
2231 atomic_set(&fs_info
->scrubs_running
, 0);
2232 atomic_set(&fs_info
->scrub_pause_req
, 0);
2233 atomic_set(&fs_info
->scrubs_paused
, 0);
2234 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2235 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2236 fs_info
->scrub_workers_refcnt
= 0;
2239 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2241 spin_lock_init(&fs_info
->balance_lock
);
2242 mutex_init(&fs_info
->balance_mutex
);
2243 atomic_set(&fs_info
->balance_running
, 0);
2244 atomic_set(&fs_info
->balance_pause_req
, 0);
2245 atomic_set(&fs_info
->balance_cancel_req
, 0);
2246 fs_info
->balance_ctl
= NULL
;
2247 init_waitqueue_head(&fs_info
->balance_wait_q
);
2250 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2251 struct btrfs_root
*tree_root
)
2253 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2254 set_nlink(fs_info
->btree_inode
, 1);
2256 * we set the i_size on the btree inode to the max possible int.
2257 * the real end of the address space is determined by all of
2258 * the devices in the system
2260 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2261 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2263 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2264 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2265 fs_info
->btree_inode
->i_mapping
);
2266 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2267 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2269 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2271 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2272 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2273 sizeof(struct btrfs_key
));
2274 set_bit(BTRFS_INODE_DUMMY
,
2275 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2276 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2279 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2281 fs_info
->dev_replace
.lock_owner
= 0;
2282 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2283 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2284 rwlock_init(&fs_info
->dev_replace
.lock
);
2285 atomic_set(&fs_info
->dev_replace
.read_locks
, 0);
2286 atomic_set(&fs_info
->dev_replace
.blocking_readers
, 0);
2287 init_waitqueue_head(&fs_info
->replace_wait
);
2288 init_waitqueue_head(&fs_info
->dev_replace
.read_lock_wq
);
2291 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2293 spin_lock_init(&fs_info
->qgroup_lock
);
2294 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2295 fs_info
->qgroup_tree
= RB_ROOT
;
2296 fs_info
->qgroup_op_tree
= RB_ROOT
;
2297 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2298 fs_info
->qgroup_seq
= 1;
2299 fs_info
->quota_enabled
= 0;
2300 fs_info
->pending_quota_state
= 0;
2301 fs_info
->qgroup_ulist
= NULL
;
2302 mutex_init(&fs_info
->qgroup_rescan_lock
);
2305 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2306 struct btrfs_fs_devices
*fs_devices
)
2308 int max_active
= fs_info
->thread_pool_size
;
2309 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2312 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2315 fs_info
->delalloc_workers
=
2316 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2318 fs_info
->flush_workers
=
2319 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2321 fs_info
->caching_workers
=
2322 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2325 * a higher idle thresh on the submit workers makes it much more
2326 * likely that bios will be send down in a sane order to the
2329 fs_info
->submit_workers
=
2330 btrfs_alloc_workqueue("submit", flags
,
2331 min_t(u64
, fs_devices
->num_devices
,
2334 fs_info
->fixup_workers
=
2335 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2338 * endios are largely parallel and should have a very
2341 fs_info
->endio_workers
=
2342 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2343 fs_info
->endio_meta_workers
=
2344 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2345 fs_info
->endio_meta_write_workers
=
2346 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2347 fs_info
->endio_raid56_workers
=
2348 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2349 fs_info
->endio_repair_workers
=
2350 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2351 fs_info
->rmw_workers
=
2352 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2353 fs_info
->endio_write_workers
=
2354 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2355 fs_info
->endio_freespace_worker
=
2356 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2357 fs_info
->delayed_workers
=
2358 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2359 fs_info
->readahead_workers
=
2360 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2361 fs_info
->qgroup_rescan_workers
=
2362 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2363 fs_info
->extent_workers
=
2364 btrfs_alloc_workqueue("extent-refs", flags
,
2365 min_t(u64
, fs_devices
->num_devices
,
2368 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2369 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2370 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2371 fs_info
->endio_meta_write_workers
&&
2372 fs_info
->endio_repair_workers
&&
2373 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2374 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2375 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2376 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2377 fs_info
->extent_workers
&&
2378 fs_info
->qgroup_rescan_workers
)) {
2385 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2386 struct btrfs_fs_devices
*fs_devices
)
2389 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2390 struct btrfs_root
*log_tree_root
;
2391 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2392 u64 bytenr
= btrfs_super_log_root(disk_super
);
2394 if (fs_devices
->rw_devices
== 0) {
2395 btrfs_warn(fs_info
, "log replay required on RO media");
2399 log_tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2403 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2404 tree_root
->stripesize
, log_tree_root
, fs_info
,
2405 BTRFS_TREE_LOG_OBJECTID
);
2407 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2408 fs_info
->generation
+ 1);
2409 if (IS_ERR(log_tree_root
->node
)) {
2410 btrfs_warn(fs_info
, "failed to read log tree");
2411 ret
= PTR_ERR(log_tree_root
->node
);
2412 kfree(log_tree_root
);
2414 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2415 btrfs_err(fs_info
, "failed to read log tree");
2416 free_extent_buffer(log_tree_root
->node
);
2417 kfree(log_tree_root
);
2420 /* returns with log_tree_root freed on success */
2421 ret
= btrfs_recover_log_trees(log_tree_root
);
2423 btrfs_handle_fs_error(tree_root
->fs_info
, ret
,
2424 "Failed to recover log tree");
2425 free_extent_buffer(log_tree_root
->node
);
2426 kfree(log_tree_root
);
2430 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2431 ret
= btrfs_commit_super(tree_root
);
2439 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2440 struct btrfs_root
*tree_root
)
2442 struct btrfs_root
*root
;
2443 struct btrfs_key location
;
2446 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2447 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2448 location
.offset
= 0;
2450 root
= btrfs_read_tree_root(tree_root
, &location
);
2452 return PTR_ERR(root
);
2453 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2454 fs_info
->extent_root
= root
;
2456 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2457 root
= btrfs_read_tree_root(tree_root
, &location
);
2459 return PTR_ERR(root
);
2460 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2461 fs_info
->dev_root
= root
;
2462 btrfs_init_devices_late(fs_info
);
2464 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2465 root
= btrfs_read_tree_root(tree_root
, &location
);
2467 return PTR_ERR(root
);
2468 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2469 fs_info
->csum_root
= root
;
2471 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2472 root
= btrfs_read_tree_root(tree_root
, &location
);
2473 if (!IS_ERR(root
)) {
2474 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2475 fs_info
->quota_enabled
= 1;
2476 fs_info
->pending_quota_state
= 1;
2477 fs_info
->quota_root
= root
;
2480 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2481 root
= btrfs_read_tree_root(tree_root
, &location
);
2483 ret
= PTR_ERR(root
);
2487 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2488 fs_info
->uuid_root
= root
;
2491 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2492 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2493 root
= btrfs_read_tree_root(tree_root
, &location
);
2495 return PTR_ERR(root
);
2496 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2497 fs_info
->free_space_root
= root
;
2503 int open_ctree(struct super_block
*sb
,
2504 struct btrfs_fs_devices
*fs_devices
,
2512 struct btrfs_key location
;
2513 struct buffer_head
*bh
;
2514 struct btrfs_super_block
*disk_super
;
2515 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2516 struct btrfs_root
*tree_root
;
2517 struct btrfs_root
*chunk_root
;
2520 int num_backups_tried
= 0;
2521 int backup_index
= 0;
2523 bool cleaner_mutex_locked
= false;
2525 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2526 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2527 if (!tree_root
|| !chunk_root
) {
2532 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2538 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2544 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2549 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2550 (1 + ilog2(nr_cpu_ids
));
2552 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2555 goto fail_dirty_metadata_bytes
;
2558 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2561 goto fail_delalloc_bytes
;
2564 fs_info
->btree_inode
= new_inode(sb
);
2565 if (!fs_info
->btree_inode
) {
2567 goto fail_bio_counter
;
2570 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2572 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2573 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2574 INIT_LIST_HEAD(&fs_info
->trans_list
);
2575 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2576 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2577 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2578 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2579 spin_lock_init(&fs_info
->delalloc_root_lock
);
2580 spin_lock_init(&fs_info
->trans_lock
);
2581 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2582 spin_lock_init(&fs_info
->delayed_iput_lock
);
2583 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2584 spin_lock_init(&fs_info
->free_chunk_lock
);
2585 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2586 spin_lock_init(&fs_info
->super_lock
);
2587 spin_lock_init(&fs_info
->qgroup_op_lock
);
2588 spin_lock_init(&fs_info
->buffer_lock
);
2589 spin_lock_init(&fs_info
->unused_bgs_lock
);
2590 rwlock_init(&fs_info
->tree_mod_log_lock
);
2591 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2592 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2593 mutex_init(&fs_info
->reloc_mutex
);
2594 mutex_init(&fs_info
->delalloc_root_mutex
);
2595 mutex_init(&fs_info
->cleaner_delayed_iput_mutex
);
2596 seqlock_init(&fs_info
->profiles_lock
);
2598 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2599 INIT_LIST_HEAD(&fs_info
->space_info
);
2600 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2601 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2602 btrfs_mapping_init(&fs_info
->mapping_tree
);
2603 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2604 BTRFS_BLOCK_RSV_GLOBAL
);
2605 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2606 BTRFS_BLOCK_RSV_DELALLOC
);
2607 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2608 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2609 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2610 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2611 BTRFS_BLOCK_RSV_DELOPS
);
2612 atomic_set(&fs_info
->nr_async_submits
, 0);
2613 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2614 atomic_set(&fs_info
->async_submit_draining
, 0);
2615 atomic_set(&fs_info
->nr_async_bios
, 0);
2616 atomic_set(&fs_info
->defrag_running
, 0);
2617 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2618 atomic_set(&fs_info
->reada_works_cnt
, 0);
2619 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2621 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2622 fs_info
->metadata_ratio
= 0;
2623 fs_info
->defrag_inodes
= RB_ROOT
;
2624 fs_info
->free_chunk_space
= 0;
2625 fs_info
->tree_mod_log
= RB_ROOT
;
2626 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2627 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2628 /* readahead state */
2629 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2630 spin_lock_init(&fs_info
->reada_lock
);
2632 fs_info
->thread_pool_size
= min_t(unsigned long,
2633 num_online_cpus() + 2, 8);
2635 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2636 spin_lock_init(&fs_info
->ordered_root_lock
);
2637 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2639 if (!fs_info
->delayed_root
) {
2643 btrfs_init_delayed_root(fs_info
->delayed_root
);
2645 btrfs_init_scrub(fs_info
);
2646 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2647 fs_info
->check_integrity_print_mask
= 0;
2649 btrfs_init_balance(fs_info
);
2650 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2652 sb
->s_blocksize
= 4096;
2653 sb
->s_blocksize_bits
= blksize_bits(4096);
2654 sb
->s_bdi
= &fs_info
->bdi
;
2656 btrfs_init_btree_inode(fs_info
, tree_root
);
2658 spin_lock_init(&fs_info
->block_group_cache_lock
);
2659 fs_info
->block_group_cache_tree
= RB_ROOT
;
2660 fs_info
->first_logical_byte
= (u64
)-1;
2662 extent_io_tree_init(&fs_info
->freed_extents
[0],
2663 fs_info
->btree_inode
->i_mapping
);
2664 extent_io_tree_init(&fs_info
->freed_extents
[1],
2665 fs_info
->btree_inode
->i_mapping
);
2666 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2667 fs_info
->do_barriers
= 1;
2670 mutex_init(&fs_info
->ordered_operations_mutex
);
2671 mutex_init(&fs_info
->tree_log_mutex
);
2672 mutex_init(&fs_info
->chunk_mutex
);
2673 mutex_init(&fs_info
->transaction_kthread_mutex
);
2674 mutex_init(&fs_info
->cleaner_mutex
);
2675 mutex_init(&fs_info
->volume_mutex
);
2676 mutex_init(&fs_info
->ro_block_group_mutex
);
2677 init_rwsem(&fs_info
->commit_root_sem
);
2678 init_rwsem(&fs_info
->cleanup_work_sem
);
2679 init_rwsem(&fs_info
->subvol_sem
);
2680 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2682 btrfs_init_dev_replace_locks(fs_info
);
2683 btrfs_init_qgroup(fs_info
);
2685 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2686 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2688 init_waitqueue_head(&fs_info
->transaction_throttle
);
2689 init_waitqueue_head(&fs_info
->transaction_wait
);
2690 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2691 init_waitqueue_head(&fs_info
->async_submit_wait
);
2693 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2695 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2701 __setup_root(4096, 4096, 4096, tree_root
,
2702 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2704 invalidate_bdev(fs_devices
->latest_bdev
);
2707 * Read super block and check the signature bytes only
2709 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2716 * We want to check superblock checksum, the type is stored inside.
2717 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2719 if (btrfs_check_super_csum(bh
->b_data
)) {
2720 btrfs_err(fs_info
, "superblock checksum mismatch");
2727 * super_copy is zeroed at allocation time and we never touch the
2728 * following bytes up to INFO_SIZE, the checksum is calculated from
2729 * the whole block of INFO_SIZE
2731 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2732 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2733 sizeof(*fs_info
->super_for_commit
));
2736 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2738 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2740 btrfs_err(fs_info
, "superblock contains fatal errors");
2745 disk_super
= fs_info
->super_copy
;
2746 if (!btrfs_super_root(disk_super
))
2749 /* check FS state, whether FS is broken. */
2750 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2751 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2754 * run through our array of backup supers and setup
2755 * our ring pointer to the oldest one
2757 generation
= btrfs_super_generation(disk_super
);
2758 find_oldest_super_backup(fs_info
, generation
);
2761 * In the long term, we'll store the compression type in the super
2762 * block, and it'll be used for per file compression control.
2764 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2766 ret
= btrfs_parse_options(tree_root
, options
, sb
->s_flags
);
2772 features
= btrfs_super_incompat_flags(disk_super
) &
2773 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2776 "cannot mount because of unsupported optional features (%llx)",
2782 features
= btrfs_super_incompat_flags(disk_super
);
2783 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2784 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2785 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2787 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2788 btrfs_info(fs_info
, "has skinny extents");
2791 * flag our filesystem as having big metadata blocks if
2792 * they are bigger than the page size
2794 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
2795 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2797 "flagging fs with big metadata feature");
2798 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2801 nodesize
= btrfs_super_nodesize(disk_super
);
2802 sectorsize
= btrfs_super_sectorsize(disk_super
);
2803 stripesize
= btrfs_super_stripesize(disk_super
);
2804 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2805 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2808 * mixed block groups end up with duplicate but slightly offset
2809 * extent buffers for the same range. It leads to corruptions
2811 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2812 (sectorsize
!= nodesize
)) {
2814 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2815 nodesize
, sectorsize
);
2820 * Needn't use the lock because there is no other task which will
2823 btrfs_set_super_incompat_flags(disk_super
, features
);
2825 features
= btrfs_super_compat_ro_flags(disk_super
) &
2826 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2827 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2829 "cannot mount read-write because of unsupported optional features (%llx)",
2835 max_active
= fs_info
->thread_pool_size
;
2837 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2840 goto fail_sb_buffer
;
2843 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2844 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2847 tree_root
->nodesize
= nodesize
;
2848 tree_root
->sectorsize
= sectorsize
;
2849 tree_root
->stripesize
= stripesize
;
2851 sb
->s_blocksize
= sectorsize
;
2852 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2854 mutex_lock(&fs_info
->chunk_mutex
);
2855 ret
= btrfs_read_sys_array(tree_root
);
2856 mutex_unlock(&fs_info
->chunk_mutex
);
2858 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
2859 goto fail_sb_buffer
;
2862 generation
= btrfs_super_chunk_root_generation(disk_super
);
2864 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2865 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2867 chunk_root
->node
= read_tree_block(chunk_root
,
2868 btrfs_super_chunk_root(disk_super
),
2870 if (IS_ERR(chunk_root
->node
) ||
2871 !extent_buffer_uptodate(chunk_root
->node
)) {
2872 btrfs_err(fs_info
, "failed to read chunk root");
2873 if (!IS_ERR(chunk_root
->node
))
2874 free_extent_buffer(chunk_root
->node
);
2875 chunk_root
->node
= NULL
;
2876 goto fail_tree_roots
;
2878 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2879 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2881 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2882 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2884 ret
= btrfs_read_chunk_tree(chunk_root
);
2886 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
2887 goto fail_tree_roots
;
2891 * keep the device that is marked to be the target device for the
2892 * dev_replace procedure
2894 btrfs_close_extra_devices(fs_devices
, 0);
2896 if (!fs_devices
->latest_bdev
) {
2897 btrfs_err(fs_info
, "failed to read devices");
2898 goto fail_tree_roots
;
2902 generation
= btrfs_super_generation(disk_super
);
2904 tree_root
->node
= read_tree_block(tree_root
,
2905 btrfs_super_root(disk_super
),
2907 if (IS_ERR(tree_root
->node
) ||
2908 !extent_buffer_uptodate(tree_root
->node
)) {
2909 btrfs_warn(fs_info
, "failed to read tree root");
2910 if (!IS_ERR(tree_root
->node
))
2911 free_extent_buffer(tree_root
->node
);
2912 tree_root
->node
= NULL
;
2913 goto recovery_tree_root
;
2916 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2917 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2918 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2920 mutex_lock(&tree_root
->objectid_mutex
);
2921 ret
= btrfs_find_highest_objectid(tree_root
,
2922 &tree_root
->highest_objectid
);
2924 mutex_unlock(&tree_root
->objectid_mutex
);
2925 goto recovery_tree_root
;
2928 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2930 mutex_unlock(&tree_root
->objectid_mutex
);
2932 ret
= btrfs_read_roots(fs_info
, tree_root
);
2934 goto recovery_tree_root
;
2936 fs_info
->generation
= generation
;
2937 fs_info
->last_trans_committed
= generation
;
2939 ret
= btrfs_recover_balance(fs_info
);
2941 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
2942 goto fail_block_groups
;
2945 ret
= btrfs_init_dev_stats(fs_info
);
2947 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
2948 goto fail_block_groups
;
2951 ret
= btrfs_init_dev_replace(fs_info
);
2953 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
2954 goto fail_block_groups
;
2957 btrfs_close_extra_devices(fs_devices
, 1);
2959 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2961 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
2963 goto fail_block_groups
;
2966 ret
= btrfs_sysfs_add_device(fs_devices
);
2968 btrfs_err(fs_info
, "failed to init sysfs device interface: %d",
2970 goto fail_fsdev_sysfs
;
2973 ret
= btrfs_sysfs_add_mounted(fs_info
);
2975 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
2976 goto fail_fsdev_sysfs
;
2979 ret
= btrfs_init_space_info(fs_info
);
2981 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
2985 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2987 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
2990 fs_info
->num_tolerated_disk_barrier_failures
=
2991 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2992 if (fs_info
->fs_devices
->missing_devices
>
2993 fs_info
->num_tolerated_disk_barrier_failures
&&
2994 !(sb
->s_flags
& MS_RDONLY
)) {
2996 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
2997 fs_info
->fs_devices
->missing_devices
,
2998 fs_info
->num_tolerated_disk_barrier_failures
);
3003 * Hold the cleaner_mutex thread here so that we don't block
3004 * for a long time on btrfs_recover_relocation. cleaner_kthread
3005 * will wait for us to finish mounting the filesystem.
3007 mutex_lock(&fs_info
->cleaner_mutex
);
3008 cleaner_mutex_locked
= true;
3009 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3011 if (IS_ERR(fs_info
->cleaner_kthread
))
3014 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3016 "btrfs-transaction");
3017 if (IS_ERR(fs_info
->transaction_kthread
))
3020 if (!btrfs_test_opt(tree_root
, SSD
) &&
3021 !btrfs_test_opt(tree_root
, NOSSD
) &&
3022 !fs_info
->fs_devices
->rotating
) {
3023 btrfs_info(fs_info
, "detected SSD devices, enabling SSD mode");
3024 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
3028 * Mount does not set all options immediately, we can do it now and do
3029 * not have to wait for transaction commit
3031 btrfs_apply_pending_changes(fs_info
);
3033 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3034 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
3035 ret
= btrfsic_mount(tree_root
, fs_devices
,
3036 btrfs_test_opt(tree_root
,
3037 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3039 fs_info
->check_integrity_print_mask
);
3042 "failed to initialize integrity check module: %d",
3046 ret
= btrfs_read_qgroup_config(fs_info
);
3048 goto fail_trans_kthread
;
3050 /* do not make disk changes in broken FS or nologreplay is given */
3051 if (btrfs_super_log_root(disk_super
) != 0 &&
3052 !btrfs_test_opt(tree_root
, NOLOGREPLAY
)) {
3053 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3060 ret
= btrfs_find_orphan_roots(tree_root
);
3064 if (!(sb
->s_flags
& MS_RDONLY
)) {
3065 ret
= btrfs_cleanup_fs_roots(fs_info
);
3068 /* We locked cleaner_mutex before creating cleaner_kthread. */
3069 ret
= btrfs_recover_relocation(tree_root
);
3071 btrfs_warn(fs_info
, "failed to recover relocation: %d",
3077 mutex_unlock(&fs_info
->cleaner_mutex
);
3078 cleaner_mutex_locked
= false;
3080 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3081 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3082 location
.offset
= 0;
3084 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3085 if (IS_ERR(fs_info
->fs_root
)) {
3086 err
= PTR_ERR(fs_info
->fs_root
);
3090 if (sb
->s_flags
& MS_RDONLY
)
3093 if (btrfs_test_opt(tree_root
, FREE_SPACE_TREE
) &&
3094 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3095 btrfs_info(fs_info
, "creating free space tree");
3096 ret
= btrfs_create_free_space_tree(fs_info
);
3099 "failed to create free space tree: %d", ret
);
3100 close_ctree(tree_root
);
3105 down_read(&fs_info
->cleanup_work_sem
);
3106 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3107 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3108 up_read(&fs_info
->cleanup_work_sem
);
3109 close_ctree(tree_root
);
3112 up_read(&fs_info
->cleanup_work_sem
);
3114 ret
= btrfs_resume_balance_async(fs_info
);
3116 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3117 close_ctree(tree_root
);
3121 ret
= btrfs_resume_dev_replace_async(fs_info
);
3123 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3124 close_ctree(tree_root
);
3128 btrfs_qgroup_rescan_resume(fs_info
);
3130 if (btrfs_test_opt(tree_root
, CLEAR_CACHE
) &&
3131 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3132 btrfs_info(fs_info
, "clearing free space tree");
3133 ret
= btrfs_clear_free_space_tree(fs_info
);
3136 "failed to clear free space tree: %d", ret
);
3137 close_ctree(tree_root
);
3142 if (!fs_info
->uuid_root
) {
3143 btrfs_info(fs_info
, "creating UUID tree");
3144 ret
= btrfs_create_uuid_tree(fs_info
);
3147 "failed to create the UUID tree: %d", ret
);
3148 close_ctree(tree_root
);
3151 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3152 fs_info
->generation
!=
3153 btrfs_super_uuid_tree_generation(disk_super
)) {
3154 btrfs_info(fs_info
, "checking UUID tree");
3155 ret
= btrfs_check_uuid_tree(fs_info
);
3158 "failed to check the UUID tree: %d", ret
);
3159 close_ctree(tree_root
);
3163 fs_info
->update_uuid_tree_gen
= 1;
3169 * backuproot only affect mount behavior, and if open_ctree succeeded,
3170 * no need to keep the flag
3172 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3177 btrfs_free_qgroup_config(fs_info
);
3179 kthread_stop(fs_info
->transaction_kthread
);
3180 btrfs_cleanup_transaction(fs_info
->tree_root
);
3181 btrfs_free_fs_roots(fs_info
);
3183 kthread_stop(fs_info
->cleaner_kthread
);
3186 * make sure we're done with the btree inode before we stop our
3189 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3192 if (cleaner_mutex_locked
) {
3193 mutex_unlock(&fs_info
->cleaner_mutex
);
3194 cleaner_mutex_locked
= false;
3196 btrfs_sysfs_remove_mounted(fs_info
);
3199 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3202 btrfs_put_block_group_cache(fs_info
);
3203 btrfs_free_block_groups(fs_info
);
3206 free_root_pointers(fs_info
, 1);
3207 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3210 btrfs_stop_all_workers(fs_info
);
3213 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3215 iput(fs_info
->btree_inode
);
3217 percpu_counter_destroy(&fs_info
->bio_counter
);
3218 fail_delalloc_bytes
:
3219 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3220 fail_dirty_metadata_bytes
:
3221 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3223 bdi_destroy(&fs_info
->bdi
);
3225 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3227 btrfs_free_stripe_hash_table(fs_info
);
3228 btrfs_close_devices(fs_info
->fs_devices
);
3232 if (!btrfs_test_opt(tree_root
, USEBACKUPROOT
))
3233 goto fail_tree_roots
;
3235 free_root_pointers(fs_info
, 0);
3237 /* don't use the log in recovery mode, it won't be valid */
3238 btrfs_set_super_log_root(disk_super
, 0);
3240 /* we can't trust the free space cache either */
3241 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3243 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3244 &num_backups_tried
, &backup_index
);
3246 goto fail_block_groups
;
3247 goto retry_root_backup
;
3250 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3253 set_buffer_uptodate(bh
);
3255 struct btrfs_device
*device
= (struct btrfs_device
*)
3258 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3259 "lost page write due to IO error on %s",
3260 rcu_str_deref(device
->name
));
3261 /* note, we don't set_buffer_write_io_error because we have
3262 * our own ways of dealing with the IO errors
3264 clear_buffer_uptodate(bh
);
3265 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3271 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3272 struct buffer_head
**bh_ret
)
3274 struct buffer_head
*bh
;
3275 struct btrfs_super_block
*super
;
3278 bytenr
= btrfs_sb_offset(copy_num
);
3279 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3282 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3284 * If we fail to read from the underlying devices, as of now
3285 * the best option we have is to mark it EIO.
3290 super
= (struct btrfs_super_block
*)bh
->b_data
;
3291 if (btrfs_super_bytenr(super
) != bytenr
||
3292 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3302 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3304 struct buffer_head
*bh
;
3305 struct buffer_head
*latest
= NULL
;
3306 struct btrfs_super_block
*super
;
3311 /* we would like to check all the supers, but that would make
3312 * a btrfs mount succeed after a mkfs from a different FS.
3313 * So, we need to add a special mount option to scan for
3314 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3316 for (i
= 0; i
< 1; i
++) {
3317 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3321 super
= (struct btrfs_super_block
*)bh
->b_data
;
3323 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3326 transid
= btrfs_super_generation(super
);
3333 return ERR_PTR(ret
);
3339 * this should be called twice, once with wait == 0 and
3340 * once with wait == 1. When wait == 0 is done, all the buffer heads
3341 * we write are pinned.
3343 * They are released when wait == 1 is done.
3344 * max_mirrors must be the same for both runs, and it indicates how
3345 * many supers on this one device should be written.
3347 * max_mirrors == 0 means to write them all.
3349 static int write_dev_supers(struct btrfs_device
*device
,
3350 struct btrfs_super_block
*sb
,
3351 int do_barriers
, int wait
, int max_mirrors
)
3353 struct buffer_head
*bh
;
3360 if (max_mirrors
== 0)
3361 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3363 for (i
= 0; i
< max_mirrors
; i
++) {
3364 bytenr
= btrfs_sb_offset(i
);
3365 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3366 device
->commit_total_bytes
)
3370 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3371 BTRFS_SUPER_INFO_SIZE
);
3377 if (!buffer_uptodate(bh
))
3380 /* drop our reference */
3383 /* drop the reference from the wait == 0 run */
3387 btrfs_set_super_bytenr(sb
, bytenr
);
3390 crc
= btrfs_csum_data((char *)sb
+
3391 BTRFS_CSUM_SIZE
, crc
,
3392 BTRFS_SUPER_INFO_SIZE
-
3394 btrfs_csum_final(crc
, sb
->csum
);
3397 * one reference for us, and we leave it for the
3400 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3401 BTRFS_SUPER_INFO_SIZE
);
3403 btrfs_err(device
->dev_root
->fs_info
,
3404 "couldn't get super buffer head for bytenr %llu",
3410 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3412 /* one reference for submit_bh */
3415 set_buffer_uptodate(bh
);
3417 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3418 bh
->b_private
= device
;
3422 * we fua the first super. The others we allow
3426 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3428 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3432 return errors
< i
? 0 : -1;
3436 * endio for the write_dev_flush, this will wake anyone waiting
3437 * for the barrier when it is done
3439 static void btrfs_end_empty_barrier(struct bio
*bio
)
3441 if (bio
->bi_private
)
3442 complete(bio
->bi_private
);
3447 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3448 * sent down. With wait == 1, it waits for the previous flush.
3450 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3453 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3458 if (device
->nobarriers
)
3462 bio
= device
->flush_bio
;
3466 wait_for_completion(&device
->flush_wait
);
3468 if (bio
->bi_error
) {
3469 ret
= bio
->bi_error
;
3470 btrfs_dev_stat_inc_and_print(device
,
3471 BTRFS_DEV_STAT_FLUSH_ERRS
);
3474 /* drop the reference from the wait == 0 run */
3476 device
->flush_bio
= NULL
;
3482 * one reference for us, and we leave it for the
3485 device
->flush_bio
= NULL
;
3486 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3490 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3491 bio
->bi_bdev
= device
->bdev
;
3492 init_completion(&device
->flush_wait
);
3493 bio
->bi_private
= &device
->flush_wait
;
3494 device
->flush_bio
= bio
;
3497 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3503 * send an empty flush down to each device in parallel,
3504 * then wait for them
3506 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3508 struct list_head
*head
;
3509 struct btrfs_device
*dev
;
3510 int errors_send
= 0;
3511 int errors_wait
= 0;
3514 /* send down all the barriers */
3515 head
= &info
->fs_devices
->devices
;
3516 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3523 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3526 ret
= write_dev_flush(dev
, 0);
3531 /* wait for all the barriers */
3532 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3539 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3542 ret
= write_dev_flush(dev
, 1);
3546 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3547 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3552 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3555 int min_tolerated
= INT_MAX
;
3557 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3558 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3559 min_tolerated
= min(min_tolerated
,
3560 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3561 tolerated_failures
);
3563 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3564 if (raid_type
== BTRFS_RAID_SINGLE
)
3566 if (!(flags
& btrfs_raid_group
[raid_type
]))
3568 min_tolerated
= min(min_tolerated
,
3569 btrfs_raid_array
[raid_type
].
3570 tolerated_failures
);
3573 if (min_tolerated
== INT_MAX
) {
3574 pr_warn("BTRFS: unknown raid flag: %llu\n", flags
);
3578 return min_tolerated
;
3581 int btrfs_calc_num_tolerated_disk_barrier_failures(
3582 struct btrfs_fs_info
*fs_info
)
3584 struct btrfs_ioctl_space_info space
;
3585 struct btrfs_space_info
*sinfo
;
3586 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3587 BTRFS_BLOCK_GROUP_SYSTEM
,
3588 BTRFS_BLOCK_GROUP_METADATA
,
3589 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3592 int num_tolerated_disk_barrier_failures
=
3593 (int)fs_info
->fs_devices
->num_devices
;
3595 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3596 struct btrfs_space_info
*tmp
;
3600 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3601 if (tmp
->flags
== types
[i
]) {
3611 down_read(&sinfo
->groups_sem
);
3612 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3615 if (list_empty(&sinfo
->block_groups
[c
]))
3618 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3620 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3622 flags
= space
.flags
;
3624 num_tolerated_disk_barrier_failures
= min(
3625 num_tolerated_disk_barrier_failures
,
3626 btrfs_get_num_tolerated_disk_barrier_failures(
3629 up_read(&sinfo
->groups_sem
);
3632 return num_tolerated_disk_barrier_failures
;
3635 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3637 struct list_head
*head
;
3638 struct btrfs_device
*dev
;
3639 struct btrfs_super_block
*sb
;
3640 struct btrfs_dev_item
*dev_item
;
3644 int total_errors
= 0;
3647 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3648 backup_super_roots(root
->fs_info
);
3650 sb
= root
->fs_info
->super_for_commit
;
3651 dev_item
= &sb
->dev_item
;
3653 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3654 head
= &root
->fs_info
->fs_devices
->devices
;
3655 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3658 ret
= barrier_all_devices(root
->fs_info
);
3661 &root
->fs_info
->fs_devices
->device_list_mutex
);
3662 btrfs_handle_fs_error(root
->fs_info
, ret
,
3663 "errors while submitting device barriers.");
3668 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3673 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3676 btrfs_set_stack_device_generation(dev_item
, 0);
3677 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3678 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3679 btrfs_set_stack_device_total_bytes(dev_item
,
3680 dev
->commit_total_bytes
);
3681 btrfs_set_stack_device_bytes_used(dev_item
,
3682 dev
->commit_bytes_used
);
3683 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3684 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3685 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3686 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3687 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3689 flags
= btrfs_super_flags(sb
);
3690 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3692 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3696 if (total_errors
> max_errors
) {
3697 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3699 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3701 /* FUA is masked off if unsupported and can't be the reason */
3702 btrfs_handle_fs_error(root
->fs_info
, -EIO
,
3703 "%d errors while writing supers", total_errors
);
3708 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3711 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3714 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3718 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3719 if (total_errors
> max_errors
) {
3720 btrfs_handle_fs_error(root
->fs_info
, -EIO
,
3721 "%d errors while writing supers", total_errors
);
3727 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3728 struct btrfs_root
*root
, int max_mirrors
)
3730 return write_all_supers(root
, max_mirrors
);
3733 /* Drop a fs root from the radix tree and free it. */
3734 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3735 struct btrfs_root
*root
)
3737 spin_lock(&fs_info
->fs_roots_radix_lock
);
3738 radix_tree_delete(&fs_info
->fs_roots_radix
,
3739 (unsigned long)root
->root_key
.objectid
);
3740 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3742 if (btrfs_root_refs(&root
->root_item
) == 0)
3743 synchronize_srcu(&fs_info
->subvol_srcu
);
3745 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3746 btrfs_free_log(NULL
, root
);
3748 if (root
->free_ino_pinned
)
3749 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3750 if (root
->free_ino_ctl
)
3751 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3755 static void free_fs_root(struct btrfs_root
*root
)
3757 iput(root
->ino_cache_inode
);
3758 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3759 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3760 root
->orphan_block_rsv
= NULL
;
3762 free_anon_bdev(root
->anon_dev
);
3763 if (root
->subv_writers
)
3764 btrfs_free_subvolume_writers(root
->subv_writers
);
3765 free_extent_buffer(root
->node
);
3766 free_extent_buffer(root
->commit_root
);
3767 kfree(root
->free_ino_ctl
);
3768 kfree(root
->free_ino_pinned
);
3770 btrfs_put_fs_root(root
);
3773 void btrfs_free_fs_root(struct btrfs_root
*root
)
3778 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3780 u64 root_objectid
= 0;
3781 struct btrfs_root
*gang
[8];
3784 unsigned int ret
= 0;
3788 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3789 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3790 (void **)gang
, root_objectid
,
3793 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3796 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3798 for (i
= 0; i
< ret
; i
++) {
3799 /* Avoid to grab roots in dead_roots */
3800 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3804 /* grab all the search result for later use */
3805 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3807 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3809 for (i
= 0; i
< ret
; i
++) {
3812 root_objectid
= gang
[i
]->root_key
.objectid
;
3813 err
= btrfs_orphan_cleanup(gang
[i
]);
3816 btrfs_put_fs_root(gang
[i
]);
3821 /* release the uncleaned roots due to error */
3822 for (; i
< ret
; i
++) {
3824 btrfs_put_fs_root(gang
[i
]);
3829 int btrfs_commit_super(struct btrfs_root
*root
)
3831 struct btrfs_trans_handle
*trans
;
3833 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3834 btrfs_run_delayed_iputs(root
);
3835 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3836 wake_up_process(root
->fs_info
->cleaner_kthread
);
3838 /* wait until ongoing cleanup work done */
3839 down_write(&root
->fs_info
->cleanup_work_sem
);
3840 up_write(&root
->fs_info
->cleanup_work_sem
);
3842 trans
= btrfs_join_transaction(root
);
3844 return PTR_ERR(trans
);
3845 return btrfs_commit_transaction(trans
, root
);
3848 void close_ctree(struct btrfs_root
*root
)
3850 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3853 fs_info
->closing
= 1;
3856 /* wait for the qgroup rescan worker to stop */
3857 btrfs_qgroup_wait_for_completion(fs_info
);
3859 /* wait for the uuid_scan task to finish */
3860 down(&fs_info
->uuid_tree_rescan_sem
);
3861 /* avoid complains from lockdep et al., set sem back to initial state */
3862 up(&fs_info
->uuid_tree_rescan_sem
);
3864 /* pause restriper - we want to resume on mount */
3865 btrfs_pause_balance(fs_info
);
3867 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3869 btrfs_scrub_cancel(fs_info
);
3871 /* wait for any defraggers to finish */
3872 wait_event(fs_info
->transaction_wait
,
3873 (atomic_read(&fs_info
->defrag_running
) == 0));
3875 /* clear out the rbtree of defraggable inodes */
3876 btrfs_cleanup_defrag_inodes(fs_info
);
3878 cancel_work_sync(&fs_info
->async_reclaim_work
);
3880 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3882 * If the cleaner thread is stopped and there are
3883 * block groups queued for removal, the deletion will be
3884 * skipped when we quit the cleaner thread.
3886 btrfs_delete_unused_bgs(root
->fs_info
);
3888 ret
= btrfs_commit_super(root
);
3890 btrfs_err(fs_info
, "commit super ret %d", ret
);
3893 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3894 btrfs_error_commit_super(root
);
3896 kthread_stop(fs_info
->transaction_kthread
);
3897 kthread_stop(fs_info
->cleaner_kthread
);
3899 fs_info
->closing
= 2;
3902 btrfs_free_qgroup_config(fs_info
);
3904 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3905 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3906 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3909 btrfs_sysfs_remove_mounted(fs_info
);
3910 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3912 btrfs_free_fs_roots(fs_info
);
3914 btrfs_put_block_group_cache(fs_info
);
3916 btrfs_free_block_groups(fs_info
);
3919 * we must make sure there is not any read request to
3920 * submit after we stopping all workers.
3922 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3923 btrfs_stop_all_workers(fs_info
);
3926 free_root_pointers(fs_info
, 1);
3928 iput(fs_info
->btree_inode
);
3930 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3931 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3932 btrfsic_unmount(root
, fs_info
->fs_devices
);
3935 btrfs_close_devices(fs_info
->fs_devices
);
3936 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3938 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3939 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3940 percpu_counter_destroy(&fs_info
->bio_counter
);
3941 bdi_destroy(&fs_info
->bdi
);
3942 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3944 btrfs_free_stripe_hash_table(fs_info
);
3946 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3947 root
->orphan_block_rsv
= NULL
;
3950 while (!list_empty(&fs_info
->pinned_chunks
)) {
3951 struct extent_map
*em
;
3953 em
= list_first_entry(&fs_info
->pinned_chunks
,
3954 struct extent_map
, list
);
3955 list_del_init(&em
->list
);
3956 free_extent_map(em
);
3958 unlock_chunks(root
);
3961 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3965 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3967 ret
= extent_buffer_uptodate(buf
);
3971 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3972 parent_transid
, atomic
);
3978 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3980 struct btrfs_root
*root
;
3981 u64 transid
= btrfs_header_generation(buf
);
3984 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3986 * This is a fast path so only do this check if we have sanity tests
3987 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3988 * outside of the sanity tests.
3990 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3993 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3994 btrfs_assert_tree_locked(buf
);
3995 if (transid
!= root
->fs_info
->generation
)
3996 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3997 "found %llu running %llu\n",
3998 buf
->start
, transid
, root
->fs_info
->generation
);
3999 was_dirty
= set_extent_buffer_dirty(buf
);
4001 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
4003 root
->fs_info
->dirty_metadata_batch
);
4004 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4005 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
4006 btrfs_print_leaf(root
, buf
);
4012 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
4016 * looks as though older kernels can get into trouble with
4017 * this code, they end up stuck in balance_dirty_pages forever
4021 if (current
->flags
& PF_MEMALLOC
)
4025 btrfs_balance_delayed_items(root
);
4027 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
4028 BTRFS_DIRTY_METADATA_THRESH
);
4030 balance_dirty_pages_ratelimited(
4031 root
->fs_info
->btree_inode
->i_mapping
);
4035 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
4037 __btrfs_btree_balance_dirty(root
, 1);
4040 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
4042 __btrfs_btree_balance_dirty(root
, 0);
4045 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
4047 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4048 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
4051 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
4054 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
4055 u64 nodesize
= btrfs_super_nodesize(sb
);
4056 u64 sectorsize
= btrfs_super_sectorsize(sb
);
4059 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
4060 printk(KERN_ERR
"BTRFS: no valid FS found\n");
4063 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
)
4064 printk(KERN_WARNING
"BTRFS: unrecognized super flag: %llu\n",
4065 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
4066 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4067 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
4068 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
4071 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4072 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
4073 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
4076 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4077 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
4078 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
4083 * Check sectorsize and nodesize first, other check will need it.
4084 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4086 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
4087 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
4088 printk(KERN_ERR
"BTRFS: invalid sectorsize %llu\n", sectorsize
);
4091 /* Only PAGE SIZE is supported yet */
4092 if (sectorsize
!= PAGE_SIZE
) {
4093 printk(KERN_ERR
"BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4094 sectorsize
, PAGE_SIZE
);
4097 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
4098 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
4099 printk(KERN_ERR
"BTRFS: invalid nodesize %llu\n", nodesize
);
4102 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
4103 printk(KERN_ERR
"BTRFS: invalid leafsize %u, should be %llu\n",
4104 le32_to_cpu(sb
->__unused_leafsize
),
4109 /* Root alignment check */
4110 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
4111 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
4112 btrfs_super_root(sb
));
4115 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
4116 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
4117 btrfs_super_chunk_root(sb
));
4120 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
4121 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
4122 btrfs_super_log_root(sb
));
4126 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4127 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4128 fs_info
->fsid
, sb
->dev_item
.fsid
);
4133 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4136 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
4137 btrfs_err(fs_info
, "bytes_used is too small %llu",
4138 btrfs_super_bytes_used(sb
));
4141 if (!is_power_of_2(btrfs_super_stripesize(sb
)) ||
4142 btrfs_super_stripesize(sb
) != sectorsize
) {
4143 btrfs_err(fs_info
, "invalid stripesize %u",
4144 btrfs_super_stripesize(sb
));
4147 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4148 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
4149 btrfs_super_num_devices(sb
));
4150 if (btrfs_super_num_devices(sb
) == 0) {
4151 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
4155 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4156 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
4157 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4162 * Obvious sys_chunk_array corruptions, it must hold at least one key
4165 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4166 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
4167 btrfs_super_sys_array_size(sb
),
4168 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4171 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4172 + sizeof(struct btrfs_chunk
)) {
4173 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
4174 btrfs_super_sys_array_size(sb
),
4175 sizeof(struct btrfs_disk_key
)
4176 + sizeof(struct btrfs_chunk
));
4181 * The generation is a global counter, we'll trust it more than the others
4182 * but it's still possible that it's the one that's wrong.
4184 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4186 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4187 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
4188 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4189 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4191 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4192 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
4197 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4199 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4200 btrfs_run_delayed_iputs(root
);
4201 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4203 down_write(&root
->fs_info
->cleanup_work_sem
);
4204 up_write(&root
->fs_info
->cleanup_work_sem
);
4206 /* cleanup FS via transaction */
4207 btrfs_cleanup_transaction(root
);
4210 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4212 struct btrfs_ordered_extent
*ordered
;
4214 spin_lock(&root
->ordered_extent_lock
);
4216 * This will just short circuit the ordered completion stuff which will
4217 * make sure the ordered extent gets properly cleaned up.
4219 list_for_each_entry(ordered
, &root
->ordered_extents
,
4221 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4222 spin_unlock(&root
->ordered_extent_lock
);
4225 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4227 struct btrfs_root
*root
;
4228 struct list_head splice
;
4230 INIT_LIST_HEAD(&splice
);
4232 spin_lock(&fs_info
->ordered_root_lock
);
4233 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4234 while (!list_empty(&splice
)) {
4235 root
= list_first_entry(&splice
, struct btrfs_root
,
4237 list_move_tail(&root
->ordered_root
,
4238 &fs_info
->ordered_roots
);
4240 spin_unlock(&fs_info
->ordered_root_lock
);
4241 btrfs_destroy_ordered_extents(root
);
4244 spin_lock(&fs_info
->ordered_root_lock
);
4246 spin_unlock(&fs_info
->ordered_root_lock
);
4249 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4250 struct btrfs_root
*root
)
4252 struct rb_node
*node
;
4253 struct btrfs_delayed_ref_root
*delayed_refs
;
4254 struct btrfs_delayed_ref_node
*ref
;
4257 delayed_refs
= &trans
->delayed_refs
;
4259 spin_lock(&delayed_refs
->lock
);
4260 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4261 spin_unlock(&delayed_refs
->lock
);
4262 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4266 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4267 struct btrfs_delayed_ref_head
*head
;
4268 struct btrfs_delayed_ref_node
*tmp
;
4269 bool pin_bytes
= false;
4271 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4273 if (!mutex_trylock(&head
->mutex
)) {
4274 atomic_inc(&head
->node
.refs
);
4275 spin_unlock(&delayed_refs
->lock
);
4277 mutex_lock(&head
->mutex
);
4278 mutex_unlock(&head
->mutex
);
4279 btrfs_put_delayed_ref(&head
->node
);
4280 spin_lock(&delayed_refs
->lock
);
4283 spin_lock(&head
->lock
);
4284 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4287 list_del(&ref
->list
);
4288 atomic_dec(&delayed_refs
->num_entries
);
4289 btrfs_put_delayed_ref(ref
);
4291 if (head
->must_insert_reserved
)
4293 btrfs_free_delayed_extent_op(head
->extent_op
);
4294 delayed_refs
->num_heads
--;
4295 if (head
->processing
== 0)
4296 delayed_refs
->num_heads_ready
--;
4297 atomic_dec(&delayed_refs
->num_entries
);
4298 head
->node
.in_tree
= 0;
4299 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4300 spin_unlock(&head
->lock
);
4301 spin_unlock(&delayed_refs
->lock
);
4302 mutex_unlock(&head
->mutex
);
4305 btrfs_pin_extent(root
, head
->node
.bytenr
,
4306 head
->node
.num_bytes
, 1);
4307 btrfs_put_delayed_ref(&head
->node
);
4309 spin_lock(&delayed_refs
->lock
);
4312 spin_unlock(&delayed_refs
->lock
);
4317 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4319 struct btrfs_inode
*btrfs_inode
;
4320 struct list_head splice
;
4322 INIT_LIST_HEAD(&splice
);
4324 spin_lock(&root
->delalloc_lock
);
4325 list_splice_init(&root
->delalloc_inodes
, &splice
);
4327 while (!list_empty(&splice
)) {
4328 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4331 list_del_init(&btrfs_inode
->delalloc_inodes
);
4332 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4333 &btrfs_inode
->runtime_flags
);
4334 spin_unlock(&root
->delalloc_lock
);
4336 btrfs_invalidate_inodes(btrfs_inode
->root
);
4338 spin_lock(&root
->delalloc_lock
);
4341 spin_unlock(&root
->delalloc_lock
);
4344 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4346 struct btrfs_root
*root
;
4347 struct list_head splice
;
4349 INIT_LIST_HEAD(&splice
);
4351 spin_lock(&fs_info
->delalloc_root_lock
);
4352 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4353 while (!list_empty(&splice
)) {
4354 root
= list_first_entry(&splice
, struct btrfs_root
,
4356 list_del_init(&root
->delalloc_root
);
4357 root
= btrfs_grab_fs_root(root
);
4359 spin_unlock(&fs_info
->delalloc_root_lock
);
4361 btrfs_destroy_delalloc_inodes(root
);
4362 btrfs_put_fs_root(root
);
4364 spin_lock(&fs_info
->delalloc_root_lock
);
4366 spin_unlock(&fs_info
->delalloc_root_lock
);
4369 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4370 struct extent_io_tree
*dirty_pages
,
4374 struct extent_buffer
*eb
;
4379 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4384 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4385 while (start
<= end
) {
4386 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4387 start
+= root
->nodesize
;
4390 wait_on_extent_buffer_writeback(eb
);
4392 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4394 clear_extent_buffer_dirty(eb
);
4395 free_extent_buffer_stale(eb
);
4402 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4403 struct extent_io_tree
*pinned_extents
)
4405 struct extent_io_tree
*unpin
;
4411 unpin
= pinned_extents
;
4414 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4415 EXTENT_DIRTY
, NULL
);
4419 clear_extent_dirty(unpin
, start
, end
);
4420 btrfs_error_unpin_extent_range(root
, start
, end
);
4425 if (unpin
== &root
->fs_info
->freed_extents
[0])
4426 unpin
= &root
->fs_info
->freed_extents
[1];
4428 unpin
= &root
->fs_info
->freed_extents
[0];
4436 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4437 struct btrfs_root
*root
)
4439 btrfs_destroy_delayed_refs(cur_trans
, root
);
4441 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4442 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4444 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4445 wake_up(&root
->fs_info
->transaction_wait
);
4447 btrfs_destroy_delayed_inodes(root
);
4448 btrfs_assert_delayed_root_empty(root
);
4450 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4452 btrfs_destroy_pinned_extent(root
,
4453 root
->fs_info
->pinned_extents
);
4455 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4456 wake_up(&cur_trans
->commit_wait
);
4459 memset(cur_trans, 0, sizeof(*cur_trans));
4460 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4464 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4466 struct btrfs_transaction
*t
;
4468 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4470 spin_lock(&root
->fs_info
->trans_lock
);
4471 while (!list_empty(&root
->fs_info
->trans_list
)) {
4472 t
= list_first_entry(&root
->fs_info
->trans_list
,
4473 struct btrfs_transaction
, list
);
4474 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4475 atomic_inc(&t
->use_count
);
4476 spin_unlock(&root
->fs_info
->trans_lock
);
4477 btrfs_wait_for_commit(root
, t
->transid
);
4478 btrfs_put_transaction(t
);
4479 spin_lock(&root
->fs_info
->trans_lock
);
4482 if (t
== root
->fs_info
->running_transaction
) {
4483 t
->state
= TRANS_STATE_COMMIT_DOING
;
4484 spin_unlock(&root
->fs_info
->trans_lock
);
4486 * We wait for 0 num_writers since we don't hold a trans
4487 * handle open currently for this transaction.
4489 wait_event(t
->writer_wait
,
4490 atomic_read(&t
->num_writers
) == 0);
4492 spin_unlock(&root
->fs_info
->trans_lock
);
4494 btrfs_cleanup_one_transaction(t
, root
);
4496 spin_lock(&root
->fs_info
->trans_lock
);
4497 if (t
== root
->fs_info
->running_transaction
)
4498 root
->fs_info
->running_transaction
= NULL
;
4499 list_del_init(&t
->list
);
4500 spin_unlock(&root
->fs_info
->trans_lock
);
4502 btrfs_put_transaction(t
);
4503 trace_btrfs_transaction_commit(root
);
4504 spin_lock(&root
->fs_info
->trans_lock
);
4506 spin_unlock(&root
->fs_info
->trans_lock
);
4507 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4508 btrfs_destroy_delayed_inodes(root
);
4509 btrfs_assert_delayed_root_empty(root
);
4510 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4511 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4512 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4517 static const struct extent_io_ops btree_extent_io_ops
= {
4518 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4519 .readpage_io_failed_hook
= btree_io_failed_hook
,
4520 .submit_bio_hook
= btree_submit_bio_hook
,
4521 /* note we're sharing with inode.c for the merge bio hook */
4522 .merge_bio_hook
= btrfs_merge_bio_hook
,