2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static const struct extent_io_ops btree_extent_io_ops
;
58 static void end_workqueue_fn(struct btrfs_work
*work
);
59 static void free_fs_root(struct btrfs_root
*root
);
60 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
62 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
64 struct btrfs_root
*root
);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
66 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
67 struct extent_io_tree
*dirty_pages
,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
70 struct extent_io_tree
*pinned_extents
);
71 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
72 static void btrfs_error_commit_super(struct btrfs_root
*root
);
75 * btrfs_end_io_wq structs are used to do processing in task context when an IO
76 * is complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
79 struct btrfs_end_io_wq
{
83 struct btrfs_fs_info
*info
;
85 enum btrfs_wq_endio_type metadata
;
86 struct list_head list
;
87 struct btrfs_work work
;
90 static struct kmem_cache
*btrfs_end_io_wq_cache
;
92 int __init
btrfs_end_io_wq_init(void)
94 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
95 sizeof(struct btrfs_end_io_wq
),
97 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
99 if (!btrfs_end_io_wq_cache
)
104 void btrfs_end_io_wq_exit(void)
106 if (btrfs_end_io_wq_cache
)
107 kmem_cache_destroy(btrfs_end_io_wq_cache
);
111 * async submit bios are used to offload expensive checksumming
112 * onto the worker threads. They checksum file and metadata bios
113 * just before they are sent down the IO stack.
115 struct async_submit_bio
{
118 struct list_head list
;
119 extent_submit_bio_hook_t
*submit_bio_start
;
120 extent_submit_bio_hook_t
*submit_bio_done
;
123 unsigned long bio_flags
;
125 * bio_offset is optional, can be used if the pages in the bio
126 * can't tell us where in the file the bio should go
129 struct btrfs_work work
;
134 * Lockdep class keys for extent_buffer->lock's in this root. For a given
135 * eb, the lockdep key is determined by the btrfs_root it belongs to and
136 * the level the eb occupies in the tree.
138 * Different roots are used for different purposes and may nest inside each
139 * other and they require separate keysets. As lockdep keys should be
140 * static, assign keysets according to the purpose of the root as indicated
141 * by btrfs_root->objectid. This ensures that all special purpose roots
142 * have separate keysets.
144 * Lock-nesting across peer nodes is always done with the immediate parent
145 * node locked thus preventing deadlock. As lockdep doesn't know this, use
146 * subclass to avoid triggering lockdep warning in such cases.
148 * The key is set by the readpage_end_io_hook after the buffer has passed
149 * csum validation but before the pages are unlocked. It is also set by
150 * btrfs_init_new_buffer on freshly allocated blocks.
152 * We also add a check to make sure the highest level of the tree is the
153 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
154 * needs update as well.
156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
157 # if BTRFS_MAX_LEVEL != 8
161 static struct btrfs_lockdep_keyset
{
162 u64 id
; /* root objectid */
163 const char *name_stem
; /* lock name stem */
164 char names
[BTRFS_MAX_LEVEL
+ 1][20];
165 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
166 } btrfs_lockdep_keysets
[] = {
167 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
168 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
169 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
170 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
171 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
172 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
173 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
174 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
175 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
176 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
177 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
178 { .id
= 0, .name_stem
= "tree" },
181 void __init
btrfs_init_lockdep(void)
185 /* initialize lockdep class names */
186 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
187 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
189 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
190 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
191 "btrfs-%s-%02d", ks
->name_stem
, j
);
195 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
198 struct btrfs_lockdep_keyset
*ks
;
200 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
202 /* find the matching keyset, id 0 is the default entry */
203 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
204 if (ks
->id
== objectid
)
207 lockdep_set_class_and_name(&eb
->lock
,
208 &ks
->keys
[level
], ks
->names
[level
]);
214 * extents on the btree inode are pretty simple, there's one extent
215 * that covers the entire device
217 static struct extent_map
*btree_get_extent(struct inode
*inode
,
218 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
221 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
222 struct extent_map
*em
;
225 read_lock(&em_tree
->lock
);
226 em
= lookup_extent_mapping(em_tree
, start
, len
);
229 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
230 read_unlock(&em_tree
->lock
);
233 read_unlock(&em_tree
->lock
);
235 em
= alloc_extent_map();
237 em
= ERR_PTR(-ENOMEM
);
242 em
->block_len
= (u64
)-1;
244 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
246 write_lock(&em_tree
->lock
);
247 ret
= add_extent_mapping(em_tree
, em
, 0);
248 if (ret
== -EEXIST
) {
250 em
= lookup_extent_mapping(em_tree
, start
, len
);
257 write_unlock(&em_tree
->lock
);
263 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
265 return btrfs_crc32c(seed
, data
, len
);
268 void btrfs_csum_final(u32 crc
, char *result
)
270 put_unaligned_le32(~crc
, result
);
274 * compute the csum for a btree block, and either verify it or write it
275 * into the csum field of the block.
277 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
278 struct extent_buffer
*buf
,
281 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
284 unsigned long cur_len
;
285 unsigned long offset
= BTRFS_CSUM_SIZE
;
287 unsigned long map_start
;
288 unsigned long map_len
;
291 unsigned long inline_result
;
293 len
= buf
->len
- offset
;
295 err
= map_private_extent_buffer(buf
, offset
, 32,
296 &kaddr
, &map_start
, &map_len
);
299 cur_len
= min(len
, map_len
- (offset
- map_start
));
300 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
305 if (csum_size
> sizeof(inline_result
)) {
306 result
= kzalloc(csum_size
, GFP_NOFS
);
310 result
= (char *)&inline_result
;
313 btrfs_csum_final(crc
, result
);
316 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
319 memcpy(&found
, result
, csum_size
);
321 read_extent_buffer(buf
, &val
, 0, csum_size
);
322 btrfs_warn_rl(fs_info
,
323 "%s checksum verify failed on %llu wanted %X found %X "
325 fs_info
->sb
->s_id
, buf
->start
,
326 val
, found
, btrfs_header_level(buf
));
327 if (result
!= (char *)&inline_result
)
332 write_extent_buffer(buf
, result
, 0, csum_size
);
334 if (result
!= (char *)&inline_result
)
340 * we can't consider a given block up to date unless the transid of the
341 * block matches the transid in the parent node's pointer. This is how we
342 * detect blocks that either didn't get written at all or got written
343 * in the wrong place.
345 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
346 struct extent_buffer
*eb
, u64 parent_transid
,
349 struct extent_state
*cached_state
= NULL
;
351 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
353 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
360 btrfs_tree_read_lock(eb
);
361 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
364 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
366 if (extent_buffer_uptodate(eb
) &&
367 btrfs_header_generation(eb
) == parent_transid
) {
371 btrfs_err_rl(eb
->fs_info
,
372 "parent transid verify failed on %llu wanted %llu found %llu",
374 parent_transid
, btrfs_header_generation(eb
));
378 * Things reading via commit roots that don't have normal protection,
379 * like send, can have a really old block in cache that may point at a
380 * block that has been free'd and re-allocated. So don't clear uptodate
381 * if we find an eb that is under IO (dirty/writeback) because we could
382 * end up reading in the stale data and then writing it back out and
383 * making everybody very sad.
385 if (!extent_buffer_under_io(eb
))
386 clear_extent_buffer_uptodate(eb
);
388 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
389 &cached_state
, GFP_NOFS
);
391 btrfs_tree_read_unlock_blocking(eb
);
396 * Return 0 if the superblock checksum type matches the checksum value of that
397 * algorithm. Pass the raw disk superblock data.
399 static int btrfs_check_super_csum(char *raw_disk_sb
)
401 struct btrfs_super_block
*disk_sb
=
402 (struct btrfs_super_block
*)raw_disk_sb
;
403 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
406 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
408 const int csum_size
= sizeof(crc
);
409 char result
[csum_size
];
412 * The super_block structure does not span the whole
413 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 * is filled with zeros and is included in the checkum.
416 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
417 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
418 btrfs_csum_final(crc
, result
);
420 if (memcmp(raw_disk_sb
, result
, csum_size
))
424 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
425 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
434 * helper to read a given tree block, doing retries as required when
435 * the checksums don't match and we have alternate mirrors to try.
437 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
438 struct extent_buffer
*eb
,
439 u64 start
, u64 parent_transid
)
441 struct extent_io_tree
*io_tree
;
446 int failed_mirror
= 0;
448 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
449 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
451 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
453 btree_get_extent
, mirror_num
);
455 if (!verify_parent_transid(io_tree
, eb
,
463 * This buffer's crc is fine, but its contents are corrupted, so
464 * there is no reason to read the other copies, they won't be
467 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
470 num_copies
= btrfs_num_copies(root
->fs_info
,
475 if (!failed_mirror
) {
477 failed_mirror
= eb
->read_mirror
;
481 if (mirror_num
== failed_mirror
)
484 if (mirror_num
> num_copies
)
488 if (failed
&& !ret
&& failed_mirror
)
489 repair_eb_io_failure(root
, eb
, failed_mirror
);
495 * checksum a dirty tree block before IO. This has extra checks to make sure
496 * we only fill in the checksum field in the first page of a multi-page block
499 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
501 u64 start
= page_offset(page
);
503 struct extent_buffer
*eb
;
505 eb
= (struct extent_buffer
*)page
->private;
506 if (page
!= eb
->pages
[0])
508 found_start
= btrfs_header_bytenr(eb
);
509 if (WARN_ON(found_start
!= start
|| !PageUptodate(page
)))
511 csum_tree_block(fs_info
, eb
, 0);
515 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
516 struct extent_buffer
*eb
)
518 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
519 u8 fsid
[BTRFS_UUID_SIZE
];
522 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
524 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
528 fs_devices
= fs_devices
->seed
;
533 #define CORRUPT(reason, eb, root, slot) \
534 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
535 "root=%llu, slot=%d", reason, \
536 btrfs_header_bytenr(eb), root->objectid, slot)
538 static noinline
int check_leaf(struct btrfs_root
*root
,
539 struct extent_buffer
*leaf
)
541 struct btrfs_key key
;
542 struct btrfs_key leaf_key
;
543 u32 nritems
= btrfs_header_nritems(leaf
);
549 /* Check the 0 item */
550 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
551 BTRFS_LEAF_DATA_SIZE(root
)) {
552 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
557 * Check to make sure each items keys are in the correct order and their
558 * offsets make sense. We only have to loop through nritems-1 because
559 * we check the current slot against the next slot, which verifies the
560 * next slot's offset+size makes sense and that the current's slot
563 for (slot
= 0; slot
< nritems
- 1; slot
++) {
564 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
565 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
567 /* Make sure the keys are in the right order */
568 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
569 CORRUPT("bad key order", leaf
, root
, slot
);
574 * Make sure the offset and ends are right, remember that the
575 * item data starts at the end of the leaf and grows towards the
578 if (btrfs_item_offset_nr(leaf
, slot
) !=
579 btrfs_item_end_nr(leaf
, slot
+ 1)) {
580 CORRUPT("slot offset bad", leaf
, root
, slot
);
585 * Check to make sure that we don't point outside of the leaf,
586 * just incase all the items are consistent to eachother, but
587 * all point outside of the leaf.
589 if (btrfs_item_end_nr(leaf
, slot
) >
590 BTRFS_LEAF_DATA_SIZE(root
)) {
591 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
599 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
600 u64 phy_offset
, struct page
*page
,
601 u64 start
, u64 end
, int mirror
)
605 struct extent_buffer
*eb
;
606 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
613 eb
= (struct extent_buffer
*)page
->private;
615 /* the pending IO might have been the only thing that kept this buffer
616 * in memory. Make sure we have a ref for all this other checks
618 extent_buffer_get(eb
);
620 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
624 eb
->read_mirror
= mirror
;
625 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
630 found_start
= btrfs_header_bytenr(eb
);
631 if (found_start
!= eb
->start
) {
632 btrfs_err_rl(eb
->fs_info
, "bad tree block start %llu %llu",
633 found_start
, eb
->start
);
637 if (check_tree_block_fsid(root
->fs_info
, eb
)) {
638 btrfs_err_rl(eb
->fs_info
, "bad fsid on block %llu",
643 found_level
= btrfs_header_level(eb
);
644 if (found_level
>= BTRFS_MAX_LEVEL
) {
645 btrfs_err(root
->fs_info
, "bad tree block level %d",
646 (int)btrfs_header_level(eb
));
651 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
654 ret
= csum_tree_block(root
->fs_info
, eb
, 1);
661 * If this is a leaf block and it is corrupt, set the corrupt bit so
662 * that we don't try and read the other copies of this block, just
665 if (found_level
== 0 && check_leaf(root
, eb
)) {
666 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
671 set_extent_buffer_uptodate(eb
);
674 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
675 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
679 * our io error hook is going to dec the io pages
680 * again, we have to make sure it has something
683 atomic_inc(&eb
->io_pages
);
684 clear_extent_buffer_uptodate(eb
);
686 free_extent_buffer(eb
);
691 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
693 struct extent_buffer
*eb
;
694 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
696 eb
= (struct extent_buffer
*)page
->private;
697 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
698 eb
->read_mirror
= failed_mirror
;
699 atomic_dec(&eb
->io_pages
);
700 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
701 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
702 return -EIO
; /* we fixed nothing */
705 static void end_workqueue_bio(struct bio
*bio
)
707 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
708 struct btrfs_fs_info
*fs_info
;
709 struct btrfs_workqueue
*wq
;
710 btrfs_work_func_t func
;
712 fs_info
= end_io_wq
->info
;
713 end_io_wq
->error
= bio
->bi_error
;
715 if (bio
->bi_rw
& REQ_WRITE
) {
716 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
717 wq
= fs_info
->endio_meta_write_workers
;
718 func
= btrfs_endio_meta_write_helper
;
719 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
720 wq
= fs_info
->endio_freespace_worker
;
721 func
= btrfs_freespace_write_helper
;
722 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
723 wq
= fs_info
->endio_raid56_workers
;
724 func
= btrfs_endio_raid56_helper
;
726 wq
= fs_info
->endio_write_workers
;
727 func
= btrfs_endio_write_helper
;
730 if (unlikely(end_io_wq
->metadata
==
731 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
732 wq
= fs_info
->endio_repair_workers
;
733 func
= btrfs_endio_repair_helper
;
734 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
735 wq
= fs_info
->endio_raid56_workers
;
736 func
= btrfs_endio_raid56_helper
;
737 } else if (end_io_wq
->metadata
) {
738 wq
= fs_info
->endio_meta_workers
;
739 func
= btrfs_endio_meta_helper
;
741 wq
= fs_info
->endio_workers
;
742 func
= btrfs_endio_helper
;
746 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
747 btrfs_queue_work(wq
, &end_io_wq
->work
);
750 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
751 enum btrfs_wq_endio_type metadata
)
753 struct btrfs_end_io_wq
*end_io_wq
;
755 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
759 end_io_wq
->private = bio
->bi_private
;
760 end_io_wq
->end_io
= bio
->bi_end_io
;
761 end_io_wq
->info
= info
;
762 end_io_wq
->error
= 0;
763 end_io_wq
->bio
= bio
;
764 end_io_wq
->metadata
= metadata
;
766 bio
->bi_private
= end_io_wq
;
767 bio
->bi_end_io
= end_workqueue_bio
;
771 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
773 unsigned long limit
= min_t(unsigned long,
774 info
->thread_pool_size
,
775 info
->fs_devices
->open_devices
);
779 static void run_one_async_start(struct btrfs_work
*work
)
781 struct async_submit_bio
*async
;
784 async
= container_of(work
, struct async_submit_bio
, work
);
785 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
786 async
->mirror_num
, async
->bio_flags
,
792 static void run_one_async_done(struct btrfs_work
*work
)
794 struct btrfs_fs_info
*fs_info
;
795 struct async_submit_bio
*async
;
798 async
= container_of(work
, struct async_submit_bio
, work
);
799 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
801 limit
= btrfs_async_submit_limit(fs_info
);
802 limit
= limit
* 2 / 3;
805 * atomic_dec_return implies a barrier for waitqueue_active
807 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
808 waitqueue_active(&fs_info
->async_submit_wait
))
809 wake_up(&fs_info
->async_submit_wait
);
811 /* If an error occured we just want to clean up the bio and move on */
813 async
->bio
->bi_error
= async
->error
;
814 bio_endio(async
->bio
);
818 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
819 async
->mirror_num
, async
->bio_flags
,
823 static void run_one_async_free(struct btrfs_work
*work
)
825 struct async_submit_bio
*async
;
827 async
= container_of(work
, struct async_submit_bio
, work
);
831 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
832 int rw
, struct bio
*bio
, int mirror_num
,
833 unsigned long bio_flags
,
835 extent_submit_bio_hook_t
*submit_bio_start
,
836 extent_submit_bio_hook_t
*submit_bio_done
)
838 struct async_submit_bio
*async
;
840 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
844 async
->inode
= inode
;
847 async
->mirror_num
= mirror_num
;
848 async
->submit_bio_start
= submit_bio_start
;
849 async
->submit_bio_done
= submit_bio_done
;
851 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
852 run_one_async_done
, run_one_async_free
);
854 async
->bio_flags
= bio_flags
;
855 async
->bio_offset
= bio_offset
;
859 atomic_inc(&fs_info
->nr_async_submits
);
862 btrfs_set_work_high_priority(&async
->work
);
864 btrfs_queue_work(fs_info
->workers
, &async
->work
);
866 while (atomic_read(&fs_info
->async_submit_draining
) &&
867 atomic_read(&fs_info
->nr_async_submits
)) {
868 wait_event(fs_info
->async_submit_wait
,
869 (atomic_read(&fs_info
->nr_async_submits
) == 0));
875 static int btree_csum_one_bio(struct bio
*bio
)
877 struct bio_vec
*bvec
;
878 struct btrfs_root
*root
;
881 bio_for_each_segment_all(bvec
, bio
, i
) {
882 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
883 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
891 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
892 struct bio
*bio
, int mirror_num
,
893 unsigned long bio_flags
,
897 * when we're called for a write, we're already in the async
898 * submission context. Just jump into btrfs_map_bio
900 return btree_csum_one_bio(bio
);
903 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
904 int mirror_num
, unsigned long bio_flags
,
910 * when we're called for a write, we're already in the async
911 * submission context. Just jump into btrfs_map_bio
913 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
921 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
923 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
932 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
933 int mirror_num
, unsigned long bio_flags
,
936 int async
= check_async_write(inode
, bio_flags
);
939 if (!(rw
& REQ_WRITE
)) {
941 * called for a read, do the setup so that checksum validation
942 * can happen in the async kernel threads
944 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
945 bio
, BTRFS_WQ_ENDIO_METADATA
);
948 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
951 ret
= btree_csum_one_bio(bio
);
954 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
958 * kthread helpers are used to submit writes so that
959 * checksumming can happen in parallel across all CPUs
961 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
962 inode
, rw
, bio
, mirror_num
, 0,
964 __btree_submit_bio_start
,
965 __btree_submit_bio_done
);
978 #ifdef CONFIG_MIGRATION
979 static int btree_migratepage(struct address_space
*mapping
,
980 struct page
*newpage
, struct page
*page
,
981 enum migrate_mode mode
)
984 * we can't safely write a btree page from here,
985 * we haven't done the locking hook
990 * Buffers may be managed in a filesystem specific way.
991 * We must have no buffers or drop them.
993 if (page_has_private(page
) &&
994 !try_to_release_page(page
, GFP_KERNEL
))
996 return migrate_page(mapping
, newpage
, page
, mode
);
1001 static int btree_writepages(struct address_space
*mapping
,
1002 struct writeback_control
*wbc
)
1004 struct btrfs_fs_info
*fs_info
;
1007 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1009 if (wbc
->for_kupdate
)
1012 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1013 /* this is a bit racy, but that's ok */
1014 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1015 BTRFS_DIRTY_METADATA_THRESH
);
1019 return btree_write_cache_pages(mapping
, wbc
);
1022 static int btree_readpage(struct file
*file
, struct page
*page
)
1024 struct extent_io_tree
*tree
;
1025 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1026 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1029 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1031 if (PageWriteback(page
) || PageDirty(page
))
1034 return try_release_extent_buffer(page
);
1037 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1038 unsigned int length
)
1040 struct extent_io_tree
*tree
;
1041 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1042 extent_invalidatepage(tree
, page
, offset
);
1043 btree_releasepage(page
, GFP_NOFS
);
1044 if (PagePrivate(page
)) {
1045 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1046 "page private not zero on page %llu",
1047 (unsigned long long)page_offset(page
));
1048 ClearPagePrivate(page
);
1049 set_page_private(page
, 0);
1050 page_cache_release(page
);
1054 static int btree_set_page_dirty(struct page
*page
)
1057 struct extent_buffer
*eb
;
1059 BUG_ON(!PagePrivate(page
));
1060 eb
= (struct extent_buffer
*)page
->private;
1062 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1063 BUG_ON(!atomic_read(&eb
->refs
));
1064 btrfs_assert_tree_locked(eb
);
1066 return __set_page_dirty_nobuffers(page
);
1069 static const struct address_space_operations btree_aops
= {
1070 .readpage
= btree_readpage
,
1071 .writepages
= btree_writepages
,
1072 .releasepage
= btree_releasepage
,
1073 .invalidatepage
= btree_invalidatepage
,
1074 #ifdef CONFIG_MIGRATION
1075 .migratepage
= btree_migratepage
,
1077 .set_page_dirty
= btree_set_page_dirty
,
1080 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1082 struct extent_buffer
*buf
= NULL
;
1083 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1085 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1088 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1089 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1090 free_extent_buffer(buf
);
1093 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1094 int mirror_num
, struct extent_buffer
**eb
)
1096 struct extent_buffer
*buf
= NULL
;
1097 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1098 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1101 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1105 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1107 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1108 btree_get_extent
, mirror_num
);
1110 free_extent_buffer(buf
);
1114 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1115 free_extent_buffer(buf
);
1117 } else if (extent_buffer_uptodate(buf
)) {
1120 free_extent_buffer(buf
);
1125 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1128 return find_extent_buffer(fs_info
, bytenr
);
1131 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1134 if (btrfs_test_is_dummy_root(root
))
1135 return alloc_test_extent_buffer(root
->fs_info
, bytenr
);
1136 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1140 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1142 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1143 buf
->start
+ buf
->len
- 1);
1146 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1148 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1149 buf
->start
, buf
->start
+ buf
->len
- 1);
1152 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1155 struct extent_buffer
*buf
= NULL
;
1158 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1160 return ERR_PTR(-ENOMEM
);
1162 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1164 free_extent_buffer(buf
);
1165 return ERR_PTR(ret
);
1171 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1172 struct btrfs_fs_info
*fs_info
,
1173 struct extent_buffer
*buf
)
1175 if (btrfs_header_generation(buf
) ==
1176 fs_info
->running_transaction
->transid
) {
1177 btrfs_assert_tree_locked(buf
);
1179 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1180 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1182 fs_info
->dirty_metadata_batch
);
1183 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1184 btrfs_set_lock_blocking(buf
);
1185 clear_extent_buffer_dirty(buf
);
1190 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1192 struct btrfs_subvolume_writers
*writers
;
1195 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1197 return ERR_PTR(-ENOMEM
);
1199 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1202 return ERR_PTR(ret
);
1205 init_waitqueue_head(&writers
->wait
);
1210 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1212 percpu_counter_destroy(&writers
->counter
);
1216 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1217 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1221 root
->commit_root
= NULL
;
1222 root
->sectorsize
= sectorsize
;
1223 root
->nodesize
= nodesize
;
1224 root
->stripesize
= stripesize
;
1226 root
->orphan_cleanup_state
= 0;
1228 root
->objectid
= objectid
;
1229 root
->last_trans
= 0;
1230 root
->highest_objectid
= 0;
1231 root
->nr_delalloc_inodes
= 0;
1232 root
->nr_ordered_extents
= 0;
1234 root
->inode_tree
= RB_ROOT
;
1235 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1236 root
->block_rsv
= NULL
;
1237 root
->orphan_block_rsv
= NULL
;
1239 INIT_LIST_HEAD(&root
->dirty_list
);
1240 INIT_LIST_HEAD(&root
->root_list
);
1241 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1242 INIT_LIST_HEAD(&root
->delalloc_root
);
1243 INIT_LIST_HEAD(&root
->ordered_extents
);
1244 INIT_LIST_HEAD(&root
->ordered_root
);
1245 INIT_LIST_HEAD(&root
->logged_list
[0]);
1246 INIT_LIST_HEAD(&root
->logged_list
[1]);
1247 spin_lock_init(&root
->orphan_lock
);
1248 spin_lock_init(&root
->inode_lock
);
1249 spin_lock_init(&root
->delalloc_lock
);
1250 spin_lock_init(&root
->ordered_extent_lock
);
1251 spin_lock_init(&root
->accounting_lock
);
1252 spin_lock_init(&root
->log_extents_lock
[0]);
1253 spin_lock_init(&root
->log_extents_lock
[1]);
1254 mutex_init(&root
->objectid_mutex
);
1255 mutex_init(&root
->log_mutex
);
1256 mutex_init(&root
->ordered_extent_mutex
);
1257 mutex_init(&root
->delalloc_mutex
);
1258 init_waitqueue_head(&root
->log_writer_wait
);
1259 init_waitqueue_head(&root
->log_commit_wait
[0]);
1260 init_waitqueue_head(&root
->log_commit_wait
[1]);
1261 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1262 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1263 atomic_set(&root
->log_commit
[0], 0);
1264 atomic_set(&root
->log_commit
[1], 0);
1265 atomic_set(&root
->log_writers
, 0);
1266 atomic_set(&root
->log_batch
, 0);
1267 atomic_set(&root
->orphan_inodes
, 0);
1268 atomic_set(&root
->refs
, 1);
1269 atomic_set(&root
->will_be_snapshoted
, 0);
1270 root
->log_transid
= 0;
1271 root
->log_transid_committed
= -1;
1272 root
->last_log_commit
= 0;
1274 extent_io_tree_init(&root
->dirty_log_pages
,
1275 fs_info
->btree_inode
->i_mapping
);
1277 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1278 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1279 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1281 root
->defrag_trans_start
= fs_info
->generation
;
1283 root
->defrag_trans_start
= 0;
1284 root
->root_key
.objectid
= objectid
;
1287 spin_lock_init(&root
->root_item_lock
);
1290 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1292 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1294 root
->fs_info
= fs_info
;
1298 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1299 /* Should only be used by the testing infrastructure */
1300 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1302 struct btrfs_root
*root
;
1304 root
= btrfs_alloc_root(NULL
);
1306 return ERR_PTR(-ENOMEM
);
1307 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1308 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1309 root
->alloc_bytenr
= 0;
1315 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1316 struct btrfs_fs_info
*fs_info
,
1319 struct extent_buffer
*leaf
;
1320 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1321 struct btrfs_root
*root
;
1322 struct btrfs_key key
;
1326 root
= btrfs_alloc_root(fs_info
);
1328 return ERR_PTR(-ENOMEM
);
1330 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1331 tree_root
->stripesize
, root
, fs_info
, objectid
);
1332 root
->root_key
.objectid
= objectid
;
1333 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1334 root
->root_key
.offset
= 0;
1336 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1338 ret
= PTR_ERR(leaf
);
1343 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1344 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1345 btrfs_set_header_generation(leaf
, trans
->transid
);
1346 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1347 btrfs_set_header_owner(leaf
, objectid
);
1350 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1352 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1353 btrfs_header_chunk_tree_uuid(leaf
),
1355 btrfs_mark_buffer_dirty(leaf
);
1357 root
->commit_root
= btrfs_root_node(root
);
1358 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1360 root
->root_item
.flags
= 0;
1361 root
->root_item
.byte_limit
= 0;
1362 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1363 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1364 btrfs_set_root_level(&root
->root_item
, 0);
1365 btrfs_set_root_refs(&root
->root_item
, 1);
1366 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1367 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1368 btrfs_set_root_dirid(&root
->root_item
, 0);
1370 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1371 root
->root_item
.drop_level
= 0;
1373 key
.objectid
= objectid
;
1374 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1376 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1380 btrfs_tree_unlock(leaf
);
1386 btrfs_tree_unlock(leaf
);
1387 free_extent_buffer(root
->commit_root
);
1388 free_extent_buffer(leaf
);
1392 return ERR_PTR(ret
);
1395 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1396 struct btrfs_fs_info
*fs_info
)
1398 struct btrfs_root
*root
;
1399 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1400 struct extent_buffer
*leaf
;
1402 root
= btrfs_alloc_root(fs_info
);
1404 return ERR_PTR(-ENOMEM
);
1406 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1407 tree_root
->stripesize
, root
, fs_info
,
1408 BTRFS_TREE_LOG_OBJECTID
);
1410 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1411 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1412 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1415 * DON'T set REF_COWS for log trees
1417 * log trees do not get reference counted because they go away
1418 * before a real commit is actually done. They do store pointers
1419 * to file data extents, and those reference counts still get
1420 * updated (along with back refs to the log tree).
1423 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1427 return ERR_CAST(leaf
);
1430 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1431 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1432 btrfs_set_header_generation(leaf
, trans
->transid
);
1433 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1434 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1437 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1438 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1439 btrfs_mark_buffer_dirty(root
->node
);
1440 btrfs_tree_unlock(root
->node
);
1444 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1445 struct btrfs_fs_info
*fs_info
)
1447 struct btrfs_root
*log_root
;
1449 log_root
= alloc_log_tree(trans
, fs_info
);
1450 if (IS_ERR(log_root
))
1451 return PTR_ERR(log_root
);
1452 WARN_ON(fs_info
->log_root_tree
);
1453 fs_info
->log_root_tree
= log_root
;
1457 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1458 struct btrfs_root
*root
)
1460 struct btrfs_root
*log_root
;
1461 struct btrfs_inode_item
*inode_item
;
1463 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1464 if (IS_ERR(log_root
))
1465 return PTR_ERR(log_root
);
1467 log_root
->last_trans
= trans
->transid
;
1468 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1470 inode_item
= &log_root
->root_item
.inode
;
1471 btrfs_set_stack_inode_generation(inode_item
, 1);
1472 btrfs_set_stack_inode_size(inode_item
, 3);
1473 btrfs_set_stack_inode_nlink(inode_item
, 1);
1474 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1475 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1477 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1479 WARN_ON(root
->log_root
);
1480 root
->log_root
= log_root
;
1481 root
->log_transid
= 0;
1482 root
->log_transid_committed
= -1;
1483 root
->last_log_commit
= 0;
1487 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1488 struct btrfs_key
*key
)
1490 struct btrfs_root
*root
;
1491 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1492 struct btrfs_path
*path
;
1496 path
= btrfs_alloc_path();
1498 return ERR_PTR(-ENOMEM
);
1500 root
= btrfs_alloc_root(fs_info
);
1506 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1507 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1509 ret
= btrfs_find_root(tree_root
, key
, path
,
1510 &root
->root_item
, &root
->root_key
);
1517 generation
= btrfs_root_generation(&root
->root_item
);
1518 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1520 if (IS_ERR(root
->node
)) {
1521 ret
= PTR_ERR(root
->node
);
1523 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1525 free_extent_buffer(root
->node
);
1528 root
->commit_root
= btrfs_root_node(root
);
1530 btrfs_free_path(path
);
1536 root
= ERR_PTR(ret
);
1540 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1541 struct btrfs_key
*location
)
1543 struct btrfs_root
*root
;
1545 root
= btrfs_read_tree_root(tree_root
, location
);
1549 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1550 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1551 btrfs_check_and_init_root_item(&root
->root_item
);
1557 int btrfs_init_fs_root(struct btrfs_root
*root
)
1560 struct btrfs_subvolume_writers
*writers
;
1562 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1563 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1565 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1570 writers
= btrfs_alloc_subvolume_writers();
1571 if (IS_ERR(writers
)) {
1572 ret
= PTR_ERR(writers
);
1575 root
->subv_writers
= writers
;
1577 btrfs_init_free_ino_ctl(root
);
1578 spin_lock_init(&root
->ino_cache_lock
);
1579 init_waitqueue_head(&root
->ino_cache_wait
);
1581 ret
= get_anon_bdev(&root
->anon_dev
);
1587 btrfs_free_subvolume_writers(root
->subv_writers
);
1589 kfree(root
->free_ino_ctl
);
1590 kfree(root
->free_ino_pinned
);
1594 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1597 struct btrfs_root
*root
;
1599 spin_lock(&fs_info
->fs_roots_radix_lock
);
1600 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1601 (unsigned long)root_id
);
1602 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1606 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1607 struct btrfs_root
*root
)
1611 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1615 spin_lock(&fs_info
->fs_roots_radix_lock
);
1616 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1617 (unsigned long)root
->root_key
.objectid
,
1620 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1621 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1622 radix_tree_preload_end();
1627 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1628 struct btrfs_key
*location
,
1631 struct btrfs_root
*root
;
1632 struct btrfs_path
*path
;
1633 struct btrfs_key key
;
1636 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1637 return fs_info
->tree_root
;
1638 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1639 return fs_info
->extent_root
;
1640 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1641 return fs_info
->chunk_root
;
1642 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1643 return fs_info
->dev_root
;
1644 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1645 return fs_info
->csum_root
;
1646 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1647 return fs_info
->quota_root
? fs_info
->quota_root
:
1649 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1650 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1653 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1655 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1656 return ERR_PTR(-ENOENT
);
1660 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1664 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1669 ret
= btrfs_init_fs_root(root
);
1673 path
= btrfs_alloc_path();
1678 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1679 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1680 key
.offset
= location
->objectid
;
1682 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1683 btrfs_free_path(path
);
1687 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1689 ret
= btrfs_insert_fs_root(fs_info
, root
);
1691 if (ret
== -EEXIST
) {
1700 return ERR_PTR(ret
);
1703 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1705 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1707 struct btrfs_device
*device
;
1708 struct backing_dev_info
*bdi
;
1711 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1714 bdi
= blk_get_backing_dev_info(device
->bdev
);
1715 if (bdi_congested(bdi
, bdi_bits
)) {
1724 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1728 err
= bdi_setup_and_register(bdi
, "btrfs");
1732 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
;
1733 bdi
->congested_fn
= btrfs_congested_fn
;
1734 bdi
->congested_data
= info
;
1735 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1740 * called by the kthread helper functions to finally call the bio end_io
1741 * functions. This is where read checksum verification actually happens
1743 static void end_workqueue_fn(struct btrfs_work
*work
)
1746 struct btrfs_end_io_wq
*end_io_wq
;
1748 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1749 bio
= end_io_wq
->bio
;
1751 bio
->bi_error
= end_io_wq
->error
;
1752 bio
->bi_private
= end_io_wq
->private;
1753 bio
->bi_end_io
= end_io_wq
->end_io
;
1754 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1758 static int cleaner_kthread(void *arg
)
1760 struct btrfs_root
*root
= arg
;
1762 struct btrfs_trans_handle
*trans
;
1767 /* Make the cleaner go to sleep early. */
1768 if (btrfs_need_cleaner_sleep(root
))
1771 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1775 * Avoid the problem that we change the status of the fs
1776 * during the above check and trylock.
1778 if (btrfs_need_cleaner_sleep(root
)) {
1779 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1783 btrfs_run_delayed_iputs(root
);
1784 again
= btrfs_clean_one_deleted_snapshot(root
);
1785 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1788 * The defragger has dealt with the R/O remount and umount,
1789 * needn't do anything special here.
1791 btrfs_run_defrag_inodes(root
->fs_info
);
1794 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1795 * with relocation (btrfs_relocate_chunk) and relocation
1796 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1797 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1798 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1799 * unused block groups.
1801 btrfs_delete_unused_bgs(root
->fs_info
);
1803 if (!try_to_freeze() && !again
) {
1804 set_current_state(TASK_INTERRUPTIBLE
);
1805 if (!kthread_should_stop())
1807 __set_current_state(TASK_RUNNING
);
1809 } while (!kthread_should_stop());
1812 * Transaction kthread is stopped before us and wakes us up.
1813 * However we might have started a new transaction and COWed some
1814 * tree blocks when deleting unused block groups for example. So
1815 * make sure we commit the transaction we started to have a clean
1816 * shutdown when evicting the btree inode - if it has dirty pages
1817 * when we do the final iput() on it, eviction will trigger a
1818 * writeback for it which will fail with null pointer dereferences
1819 * since work queues and other resources were already released and
1820 * destroyed by the time the iput/eviction/writeback is made.
1822 trans
= btrfs_attach_transaction(root
);
1823 if (IS_ERR(trans
)) {
1824 if (PTR_ERR(trans
) != -ENOENT
)
1825 btrfs_err(root
->fs_info
,
1826 "cleaner transaction attach returned %ld",
1831 ret
= btrfs_commit_transaction(trans
, root
);
1833 btrfs_err(root
->fs_info
,
1834 "cleaner open transaction commit returned %d",
1841 static int transaction_kthread(void *arg
)
1843 struct btrfs_root
*root
= arg
;
1844 struct btrfs_trans_handle
*trans
;
1845 struct btrfs_transaction
*cur
;
1848 unsigned long delay
;
1852 cannot_commit
= false;
1853 delay
= HZ
* root
->fs_info
->commit_interval
;
1854 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1856 spin_lock(&root
->fs_info
->trans_lock
);
1857 cur
= root
->fs_info
->running_transaction
;
1859 spin_unlock(&root
->fs_info
->trans_lock
);
1863 now
= get_seconds();
1864 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1865 (now
< cur
->start_time
||
1866 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1867 spin_unlock(&root
->fs_info
->trans_lock
);
1871 transid
= cur
->transid
;
1872 spin_unlock(&root
->fs_info
->trans_lock
);
1874 /* If the file system is aborted, this will always fail. */
1875 trans
= btrfs_attach_transaction(root
);
1876 if (IS_ERR(trans
)) {
1877 if (PTR_ERR(trans
) != -ENOENT
)
1878 cannot_commit
= true;
1881 if (transid
== trans
->transid
) {
1882 btrfs_commit_transaction(trans
, root
);
1884 btrfs_end_transaction(trans
, root
);
1887 wake_up_process(root
->fs_info
->cleaner_kthread
);
1888 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1890 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1891 &root
->fs_info
->fs_state
)))
1892 btrfs_cleanup_transaction(root
);
1893 if (!try_to_freeze()) {
1894 set_current_state(TASK_INTERRUPTIBLE
);
1895 if (!kthread_should_stop() &&
1896 (!btrfs_transaction_blocked(root
->fs_info
) ||
1898 schedule_timeout(delay
);
1899 __set_current_state(TASK_RUNNING
);
1901 } while (!kthread_should_stop());
1906 * this will find the highest generation in the array of
1907 * root backups. The index of the highest array is returned,
1908 * or -1 if we can't find anything.
1910 * We check to make sure the array is valid by comparing the
1911 * generation of the latest root in the array with the generation
1912 * in the super block. If they don't match we pitch it.
1914 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1917 int newest_index
= -1;
1918 struct btrfs_root_backup
*root_backup
;
1921 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1922 root_backup
= info
->super_copy
->super_roots
+ i
;
1923 cur
= btrfs_backup_tree_root_gen(root_backup
);
1924 if (cur
== newest_gen
)
1928 /* check to see if we actually wrapped around */
1929 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1930 root_backup
= info
->super_copy
->super_roots
;
1931 cur
= btrfs_backup_tree_root_gen(root_backup
);
1932 if (cur
== newest_gen
)
1935 return newest_index
;
1940 * find the oldest backup so we know where to store new entries
1941 * in the backup array. This will set the backup_root_index
1942 * field in the fs_info struct
1944 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1947 int newest_index
= -1;
1949 newest_index
= find_newest_super_backup(info
, newest_gen
);
1950 /* if there was garbage in there, just move along */
1951 if (newest_index
== -1) {
1952 info
->backup_root_index
= 0;
1954 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1959 * copy all the root pointers into the super backup array.
1960 * this will bump the backup pointer by one when it is
1963 static void backup_super_roots(struct btrfs_fs_info
*info
)
1966 struct btrfs_root_backup
*root_backup
;
1969 next_backup
= info
->backup_root_index
;
1970 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1971 BTRFS_NUM_BACKUP_ROOTS
;
1974 * just overwrite the last backup if we're at the same generation
1975 * this happens only at umount
1977 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1978 if (btrfs_backup_tree_root_gen(root_backup
) ==
1979 btrfs_header_generation(info
->tree_root
->node
))
1980 next_backup
= last_backup
;
1982 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1985 * make sure all of our padding and empty slots get zero filled
1986 * regardless of which ones we use today
1988 memset(root_backup
, 0, sizeof(*root_backup
));
1990 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1992 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1993 btrfs_set_backup_tree_root_gen(root_backup
,
1994 btrfs_header_generation(info
->tree_root
->node
));
1996 btrfs_set_backup_tree_root_level(root_backup
,
1997 btrfs_header_level(info
->tree_root
->node
));
1999 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2000 btrfs_set_backup_chunk_root_gen(root_backup
,
2001 btrfs_header_generation(info
->chunk_root
->node
));
2002 btrfs_set_backup_chunk_root_level(root_backup
,
2003 btrfs_header_level(info
->chunk_root
->node
));
2005 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2006 btrfs_set_backup_extent_root_gen(root_backup
,
2007 btrfs_header_generation(info
->extent_root
->node
));
2008 btrfs_set_backup_extent_root_level(root_backup
,
2009 btrfs_header_level(info
->extent_root
->node
));
2012 * we might commit during log recovery, which happens before we set
2013 * the fs_root. Make sure it is valid before we fill it in.
2015 if (info
->fs_root
&& info
->fs_root
->node
) {
2016 btrfs_set_backup_fs_root(root_backup
,
2017 info
->fs_root
->node
->start
);
2018 btrfs_set_backup_fs_root_gen(root_backup
,
2019 btrfs_header_generation(info
->fs_root
->node
));
2020 btrfs_set_backup_fs_root_level(root_backup
,
2021 btrfs_header_level(info
->fs_root
->node
));
2024 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2025 btrfs_set_backup_dev_root_gen(root_backup
,
2026 btrfs_header_generation(info
->dev_root
->node
));
2027 btrfs_set_backup_dev_root_level(root_backup
,
2028 btrfs_header_level(info
->dev_root
->node
));
2030 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2031 btrfs_set_backup_csum_root_gen(root_backup
,
2032 btrfs_header_generation(info
->csum_root
->node
));
2033 btrfs_set_backup_csum_root_level(root_backup
,
2034 btrfs_header_level(info
->csum_root
->node
));
2036 btrfs_set_backup_total_bytes(root_backup
,
2037 btrfs_super_total_bytes(info
->super_copy
));
2038 btrfs_set_backup_bytes_used(root_backup
,
2039 btrfs_super_bytes_used(info
->super_copy
));
2040 btrfs_set_backup_num_devices(root_backup
,
2041 btrfs_super_num_devices(info
->super_copy
));
2044 * if we don't copy this out to the super_copy, it won't get remembered
2045 * for the next commit
2047 memcpy(&info
->super_copy
->super_roots
,
2048 &info
->super_for_commit
->super_roots
,
2049 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2053 * this copies info out of the root backup array and back into
2054 * the in-memory super block. It is meant to help iterate through
2055 * the array, so you send it the number of backups you've already
2056 * tried and the last backup index you used.
2058 * this returns -1 when it has tried all the backups
2060 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2061 struct btrfs_super_block
*super
,
2062 int *num_backups_tried
, int *backup_index
)
2064 struct btrfs_root_backup
*root_backup
;
2065 int newest
= *backup_index
;
2067 if (*num_backups_tried
== 0) {
2068 u64 gen
= btrfs_super_generation(super
);
2070 newest
= find_newest_super_backup(info
, gen
);
2074 *backup_index
= newest
;
2075 *num_backups_tried
= 1;
2076 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2077 /* we've tried all the backups, all done */
2080 /* jump to the next oldest backup */
2081 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2082 BTRFS_NUM_BACKUP_ROOTS
;
2083 *backup_index
= newest
;
2084 *num_backups_tried
+= 1;
2086 root_backup
= super
->super_roots
+ newest
;
2088 btrfs_set_super_generation(super
,
2089 btrfs_backup_tree_root_gen(root_backup
));
2090 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2091 btrfs_set_super_root_level(super
,
2092 btrfs_backup_tree_root_level(root_backup
));
2093 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2096 * fixme: the total bytes and num_devices need to match or we should
2099 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2100 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2104 /* helper to cleanup workers */
2105 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2107 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2108 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2109 btrfs_destroy_workqueue(fs_info
->workers
);
2110 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2111 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2112 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2113 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2114 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2115 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2116 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2117 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2118 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2119 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2120 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2121 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2122 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2123 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2124 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2127 static void free_root_extent_buffers(struct btrfs_root
*root
)
2130 free_extent_buffer(root
->node
);
2131 free_extent_buffer(root
->commit_root
);
2133 root
->commit_root
= NULL
;
2137 /* helper to cleanup tree roots */
2138 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2140 free_root_extent_buffers(info
->tree_root
);
2142 free_root_extent_buffers(info
->dev_root
);
2143 free_root_extent_buffers(info
->extent_root
);
2144 free_root_extent_buffers(info
->csum_root
);
2145 free_root_extent_buffers(info
->quota_root
);
2146 free_root_extent_buffers(info
->uuid_root
);
2148 free_root_extent_buffers(info
->chunk_root
);
2151 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2154 struct btrfs_root
*gang
[8];
2157 while (!list_empty(&fs_info
->dead_roots
)) {
2158 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2159 struct btrfs_root
, root_list
);
2160 list_del(&gang
[0]->root_list
);
2162 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2163 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2165 free_extent_buffer(gang
[0]->node
);
2166 free_extent_buffer(gang
[0]->commit_root
);
2167 btrfs_put_fs_root(gang
[0]);
2172 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2177 for (i
= 0; i
< ret
; i
++)
2178 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2181 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2182 btrfs_free_log_root_tree(NULL
, fs_info
);
2183 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2184 fs_info
->pinned_extents
);
2188 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2190 mutex_init(&fs_info
->scrub_lock
);
2191 atomic_set(&fs_info
->scrubs_running
, 0);
2192 atomic_set(&fs_info
->scrub_pause_req
, 0);
2193 atomic_set(&fs_info
->scrubs_paused
, 0);
2194 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2195 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2196 fs_info
->scrub_workers_refcnt
= 0;
2199 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2201 spin_lock_init(&fs_info
->balance_lock
);
2202 mutex_init(&fs_info
->balance_mutex
);
2203 atomic_set(&fs_info
->balance_running
, 0);
2204 atomic_set(&fs_info
->balance_pause_req
, 0);
2205 atomic_set(&fs_info
->balance_cancel_req
, 0);
2206 fs_info
->balance_ctl
= NULL
;
2207 init_waitqueue_head(&fs_info
->balance_wait_q
);
2210 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2211 struct btrfs_root
*tree_root
)
2213 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2214 set_nlink(fs_info
->btree_inode
, 1);
2216 * we set the i_size on the btree inode to the max possible int.
2217 * the real end of the address space is determined by all of
2218 * the devices in the system
2220 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2221 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2223 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2224 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2225 fs_info
->btree_inode
->i_mapping
);
2226 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2227 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2229 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2231 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2232 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2233 sizeof(struct btrfs_key
));
2234 set_bit(BTRFS_INODE_DUMMY
,
2235 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2236 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2239 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2241 fs_info
->dev_replace
.lock_owner
= 0;
2242 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2243 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2244 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2245 mutex_init(&fs_info
->dev_replace
.lock
);
2246 init_waitqueue_head(&fs_info
->replace_wait
);
2249 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2251 spin_lock_init(&fs_info
->qgroup_lock
);
2252 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2253 fs_info
->qgroup_tree
= RB_ROOT
;
2254 fs_info
->qgroup_op_tree
= RB_ROOT
;
2255 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2256 fs_info
->qgroup_seq
= 1;
2257 fs_info
->quota_enabled
= 0;
2258 fs_info
->pending_quota_state
= 0;
2259 fs_info
->qgroup_ulist
= NULL
;
2260 mutex_init(&fs_info
->qgroup_rescan_lock
);
2263 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2264 struct btrfs_fs_devices
*fs_devices
)
2266 int max_active
= fs_info
->thread_pool_size
;
2267 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2270 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2273 fs_info
->delalloc_workers
=
2274 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2276 fs_info
->flush_workers
=
2277 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2279 fs_info
->caching_workers
=
2280 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2283 * a higher idle thresh on the submit workers makes it much more
2284 * likely that bios will be send down in a sane order to the
2287 fs_info
->submit_workers
=
2288 btrfs_alloc_workqueue("submit", flags
,
2289 min_t(u64
, fs_devices
->num_devices
,
2292 fs_info
->fixup_workers
=
2293 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2296 * endios are largely parallel and should have a very
2299 fs_info
->endio_workers
=
2300 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2301 fs_info
->endio_meta_workers
=
2302 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2303 fs_info
->endio_meta_write_workers
=
2304 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2305 fs_info
->endio_raid56_workers
=
2306 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2307 fs_info
->endio_repair_workers
=
2308 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2309 fs_info
->rmw_workers
=
2310 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2311 fs_info
->endio_write_workers
=
2312 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2313 fs_info
->endio_freespace_worker
=
2314 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2315 fs_info
->delayed_workers
=
2316 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2317 fs_info
->readahead_workers
=
2318 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2319 fs_info
->qgroup_rescan_workers
=
2320 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2321 fs_info
->extent_workers
=
2322 btrfs_alloc_workqueue("extent-refs", flags
,
2323 min_t(u64
, fs_devices
->num_devices
,
2326 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2327 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2328 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2329 fs_info
->endio_meta_write_workers
&&
2330 fs_info
->endio_repair_workers
&&
2331 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2332 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2333 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2334 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2335 fs_info
->extent_workers
&&
2336 fs_info
->qgroup_rescan_workers
)) {
2343 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2344 struct btrfs_fs_devices
*fs_devices
)
2347 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2348 struct btrfs_root
*log_tree_root
;
2349 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2350 u64 bytenr
= btrfs_super_log_root(disk_super
);
2352 if (fs_devices
->rw_devices
== 0) {
2353 btrfs_warn(fs_info
, "log replay required on RO media");
2357 log_tree_root
= btrfs_alloc_root(fs_info
);
2361 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2362 tree_root
->stripesize
, log_tree_root
, fs_info
,
2363 BTRFS_TREE_LOG_OBJECTID
);
2365 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2366 fs_info
->generation
+ 1);
2367 if (IS_ERR(log_tree_root
->node
)) {
2368 btrfs_warn(fs_info
, "failed to read log tree");
2369 ret
= PTR_ERR(log_tree_root
->node
);
2370 kfree(log_tree_root
);
2372 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2373 btrfs_err(fs_info
, "failed to read log tree");
2374 free_extent_buffer(log_tree_root
->node
);
2375 kfree(log_tree_root
);
2378 /* returns with log_tree_root freed on success */
2379 ret
= btrfs_recover_log_trees(log_tree_root
);
2381 btrfs_std_error(tree_root
->fs_info
, ret
,
2382 "Failed to recover log tree");
2383 free_extent_buffer(log_tree_root
->node
);
2384 kfree(log_tree_root
);
2388 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2389 ret
= btrfs_commit_super(tree_root
);
2397 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2398 struct btrfs_root
*tree_root
)
2400 struct btrfs_root
*root
;
2401 struct btrfs_key location
;
2404 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2405 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2406 location
.offset
= 0;
2408 root
= btrfs_read_tree_root(tree_root
, &location
);
2410 return PTR_ERR(root
);
2411 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2412 fs_info
->extent_root
= root
;
2414 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2415 root
= btrfs_read_tree_root(tree_root
, &location
);
2417 return PTR_ERR(root
);
2418 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2419 fs_info
->dev_root
= root
;
2420 btrfs_init_devices_late(fs_info
);
2422 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2423 root
= btrfs_read_tree_root(tree_root
, &location
);
2425 return PTR_ERR(root
);
2426 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2427 fs_info
->csum_root
= root
;
2429 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2430 root
= btrfs_read_tree_root(tree_root
, &location
);
2431 if (!IS_ERR(root
)) {
2432 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2433 fs_info
->quota_enabled
= 1;
2434 fs_info
->pending_quota_state
= 1;
2435 fs_info
->quota_root
= root
;
2438 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2439 root
= btrfs_read_tree_root(tree_root
, &location
);
2441 ret
= PTR_ERR(root
);
2445 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2446 fs_info
->uuid_root
= root
;
2452 int open_ctree(struct super_block
*sb
,
2453 struct btrfs_fs_devices
*fs_devices
,
2461 struct btrfs_key location
;
2462 struct buffer_head
*bh
;
2463 struct btrfs_super_block
*disk_super
;
2464 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2465 struct btrfs_root
*tree_root
;
2466 struct btrfs_root
*chunk_root
;
2469 int num_backups_tried
= 0;
2470 int backup_index
= 0;
2473 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2474 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2475 if (!tree_root
|| !chunk_root
) {
2480 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2486 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2492 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2497 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2498 (1 + ilog2(nr_cpu_ids
));
2500 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2503 goto fail_dirty_metadata_bytes
;
2506 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2509 goto fail_delalloc_bytes
;
2512 fs_info
->btree_inode
= new_inode(sb
);
2513 if (!fs_info
->btree_inode
) {
2515 goto fail_bio_counter
;
2518 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2520 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2521 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2522 INIT_LIST_HEAD(&fs_info
->trans_list
);
2523 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2524 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2525 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2526 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2527 spin_lock_init(&fs_info
->delalloc_root_lock
);
2528 spin_lock_init(&fs_info
->trans_lock
);
2529 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2530 spin_lock_init(&fs_info
->delayed_iput_lock
);
2531 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2532 spin_lock_init(&fs_info
->free_chunk_lock
);
2533 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2534 spin_lock_init(&fs_info
->super_lock
);
2535 spin_lock_init(&fs_info
->qgroup_op_lock
);
2536 spin_lock_init(&fs_info
->buffer_lock
);
2537 spin_lock_init(&fs_info
->unused_bgs_lock
);
2538 rwlock_init(&fs_info
->tree_mod_log_lock
);
2539 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2540 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2541 mutex_init(&fs_info
->reloc_mutex
);
2542 mutex_init(&fs_info
->delalloc_root_mutex
);
2543 seqlock_init(&fs_info
->profiles_lock
);
2544 init_rwsem(&fs_info
->delayed_iput_sem
);
2546 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2547 INIT_LIST_HEAD(&fs_info
->space_info
);
2548 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2549 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2550 btrfs_mapping_init(&fs_info
->mapping_tree
);
2551 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2552 BTRFS_BLOCK_RSV_GLOBAL
);
2553 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2554 BTRFS_BLOCK_RSV_DELALLOC
);
2555 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2556 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2557 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2558 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2559 BTRFS_BLOCK_RSV_DELOPS
);
2560 atomic_set(&fs_info
->nr_async_submits
, 0);
2561 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2562 atomic_set(&fs_info
->async_submit_draining
, 0);
2563 atomic_set(&fs_info
->nr_async_bios
, 0);
2564 atomic_set(&fs_info
->defrag_running
, 0);
2565 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2566 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2568 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2569 fs_info
->metadata_ratio
= 0;
2570 fs_info
->defrag_inodes
= RB_ROOT
;
2571 fs_info
->free_chunk_space
= 0;
2572 fs_info
->tree_mod_log
= RB_ROOT
;
2573 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2574 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2575 /* readahead state */
2576 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2577 spin_lock_init(&fs_info
->reada_lock
);
2579 fs_info
->thread_pool_size
= min_t(unsigned long,
2580 num_online_cpus() + 2, 8);
2582 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2583 spin_lock_init(&fs_info
->ordered_root_lock
);
2584 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2586 if (!fs_info
->delayed_root
) {
2590 btrfs_init_delayed_root(fs_info
->delayed_root
);
2592 btrfs_init_scrub(fs_info
);
2593 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2594 fs_info
->check_integrity_print_mask
= 0;
2596 btrfs_init_balance(fs_info
);
2597 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2599 sb
->s_blocksize
= 4096;
2600 sb
->s_blocksize_bits
= blksize_bits(4096);
2601 sb
->s_bdi
= &fs_info
->bdi
;
2603 btrfs_init_btree_inode(fs_info
, tree_root
);
2605 spin_lock_init(&fs_info
->block_group_cache_lock
);
2606 fs_info
->block_group_cache_tree
= RB_ROOT
;
2607 fs_info
->first_logical_byte
= (u64
)-1;
2609 extent_io_tree_init(&fs_info
->freed_extents
[0],
2610 fs_info
->btree_inode
->i_mapping
);
2611 extent_io_tree_init(&fs_info
->freed_extents
[1],
2612 fs_info
->btree_inode
->i_mapping
);
2613 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2614 fs_info
->do_barriers
= 1;
2617 mutex_init(&fs_info
->ordered_operations_mutex
);
2618 mutex_init(&fs_info
->tree_log_mutex
);
2619 mutex_init(&fs_info
->chunk_mutex
);
2620 mutex_init(&fs_info
->transaction_kthread_mutex
);
2621 mutex_init(&fs_info
->cleaner_mutex
);
2622 mutex_init(&fs_info
->volume_mutex
);
2623 mutex_init(&fs_info
->ro_block_group_mutex
);
2624 init_rwsem(&fs_info
->commit_root_sem
);
2625 init_rwsem(&fs_info
->cleanup_work_sem
);
2626 init_rwsem(&fs_info
->subvol_sem
);
2627 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2629 btrfs_init_dev_replace_locks(fs_info
);
2630 btrfs_init_qgroup(fs_info
);
2632 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2633 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2635 init_waitqueue_head(&fs_info
->transaction_throttle
);
2636 init_waitqueue_head(&fs_info
->transaction_wait
);
2637 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2638 init_waitqueue_head(&fs_info
->async_submit_wait
);
2640 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2642 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2648 __setup_root(4096, 4096, 4096, tree_root
,
2649 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2651 invalidate_bdev(fs_devices
->latest_bdev
);
2654 * Read super block and check the signature bytes only
2656 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2663 * We want to check superblock checksum, the type is stored inside.
2664 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2666 if (btrfs_check_super_csum(bh
->b_data
)) {
2667 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2673 * super_copy is zeroed at allocation time and we never touch the
2674 * following bytes up to INFO_SIZE, the checksum is calculated from
2675 * the whole block of INFO_SIZE
2677 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2678 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2679 sizeof(*fs_info
->super_for_commit
));
2682 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2684 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2686 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2691 disk_super
= fs_info
->super_copy
;
2692 if (!btrfs_super_root(disk_super
))
2695 /* check FS state, whether FS is broken. */
2696 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2697 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2700 * run through our array of backup supers and setup
2701 * our ring pointer to the oldest one
2703 generation
= btrfs_super_generation(disk_super
);
2704 find_oldest_super_backup(fs_info
, generation
);
2707 * In the long term, we'll store the compression type in the super
2708 * block, and it'll be used for per file compression control.
2710 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2712 ret
= btrfs_parse_options(tree_root
, options
);
2718 features
= btrfs_super_incompat_flags(disk_super
) &
2719 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2721 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2722 "unsupported optional features (%Lx).\n",
2729 * Leafsize and nodesize were always equal, this is only a sanity check.
2731 if (le32_to_cpu(disk_super
->__unused_leafsize
) !=
2732 btrfs_super_nodesize(disk_super
)) {
2733 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2734 "blocksizes don't match. node %d leaf %d\n",
2735 btrfs_super_nodesize(disk_super
),
2736 le32_to_cpu(disk_super
->__unused_leafsize
));
2740 if (btrfs_super_nodesize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2741 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2742 "blocksize (%d) was too large\n",
2743 btrfs_super_nodesize(disk_super
));
2748 features
= btrfs_super_incompat_flags(disk_super
);
2749 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2750 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2751 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2753 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2754 printk(KERN_INFO
"BTRFS: has skinny extents\n");
2757 * flag our filesystem as having big metadata blocks if
2758 * they are bigger than the page size
2760 if (btrfs_super_nodesize(disk_super
) > PAGE_CACHE_SIZE
) {
2761 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2762 printk(KERN_INFO
"BTRFS: flagging fs with big metadata feature\n");
2763 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2766 nodesize
= btrfs_super_nodesize(disk_super
);
2767 sectorsize
= btrfs_super_sectorsize(disk_super
);
2768 stripesize
= btrfs_super_stripesize(disk_super
);
2769 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2770 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2773 * mixed block groups end up with duplicate but slightly offset
2774 * extent buffers for the same range. It leads to corruptions
2776 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2777 (sectorsize
!= nodesize
)) {
2778 printk(KERN_ERR
"BTRFS: unequal leaf/node/sector sizes "
2779 "are not allowed for mixed block groups on %s\n",
2785 * Needn't use the lock because there is no other task which will
2788 btrfs_set_super_incompat_flags(disk_super
, features
);
2790 features
= btrfs_super_compat_ro_flags(disk_super
) &
2791 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2792 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2793 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2794 "unsupported option features (%Lx).\n",
2800 max_active
= fs_info
->thread_pool_size
;
2802 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2805 goto fail_sb_buffer
;
2808 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2809 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2810 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2812 tree_root
->nodesize
= nodesize
;
2813 tree_root
->sectorsize
= sectorsize
;
2814 tree_root
->stripesize
= stripesize
;
2816 sb
->s_blocksize
= sectorsize
;
2817 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2819 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2820 printk(KERN_ERR
"BTRFS: valid FS not found on %s\n", sb
->s_id
);
2821 goto fail_sb_buffer
;
2824 if (sectorsize
!= PAGE_SIZE
) {
2825 printk(KERN_ERR
"BTRFS: incompatible sector size (%lu) "
2826 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2827 goto fail_sb_buffer
;
2830 mutex_lock(&fs_info
->chunk_mutex
);
2831 ret
= btrfs_read_sys_array(tree_root
);
2832 mutex_unlock(&fs_info
->chunk_mutex
);
2834 printk(KERN_ERR
"BTRFS: failed to read the system "
2835 "array on %s\n", sb
->s_id
);
2836 goto fail_sb_buffer
;
2839 generation
= btrfs_super_chunk_root_generation(disk_super
);
2841 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2842 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2844 chunk_root
->node
= read_tree_block(chunk_root
,
2845 btrfs_super_chunk_root(disk_super
),
2847 if (IS_ERR(chunk_root
->node
) ||
2848 !extent_buffer_uptodate(chunk_root
->node
)) {
2849 printk(KERN_ERR
"BTRFS: failed to read chunk root on %s\n",
2851 if (!IS_ERR(chunk_root
->node
))
2852 free_extent_buffer(chunk_root
->node
);
2853 chunk_root
->node
= NULL
;
2854 goto fail_tree_roots
;
2856 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2857 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2859 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2860 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2862 ret
= btrfs_read_chunk_tree(chunk_root
);
2864 printk(KERN_ERR
"BTRFS: failed to read chunk tree on %s\n",
2866 goto fail_tree_roots
;
2870 * keep the device that is marked to be the target device for the
2871 * dev_replace procedure
2873 btrfs_close_extra_devices(fs_devices
, 0);
2875 if (!fs_devices
->latest_bdev
) {
2876 printk(KERN_ERR
"BTRFS: failed to read devices on %s\n",
2878 goto fail_tree_roots
;
2882 generation
= btrfs_super_generation(disk_super
);
2884 tree_root
->node
= read_tree_block(tree_root
,
2885 btrfs_super_root(disk_super
),
2887 if (IS_ERR(tree_root
->node
) ||
2888 !extent_buffer_uptodate(tree_root
->node
)) {
2889 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2891 if (!IS_ERR(tree_root
->node
))
2892 free_extent_buffer(tree_root
->node
);
2893 tree_root
->node
= NULL
;
2894 goto recovery_tree_root
;
2897 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2898 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2899 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2901 ret
= btrfs_read_roots(fs_info
, tree_root
);
2903 goto recovery_tree_root
;
2905 fs_info
->generation
= generation
;
2906 fs_info
->last_trans_committed
= generation
;
2908 ret
= btrfs_recover_balance(fs_info
);
2910 printk(KERN_ERR
"BTRFS: failed to recover balance\n");
2911 goto fail_block_groups
;
2914 ret
= btrfs_init_dev_stats(fs_info
);
2916 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2918 goto fail_block_groups
;
2921 ret
= btrfs_init_dev_replace(fs_info
);
2923 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2924 goto fail_block_groups
;
2927 btrfs_close_extra_devices(fs_devices
, 1);
2929 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2931 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret
);
2932 goto fail_block_groups
;
2935 ret
= btrfs_sysfs_add_device(fs_devices
);
2937 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret
);
2938 goto fail_fsdev_sysfs
;
2941 ret
= btrfs_sysfs_add_mounted(fs_info
);
2943 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2944 goto fail_fsdev_sysfs
;
2947 ret
= btrfs_init_space_info(fs_info
);
2949 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2953 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2955 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
2958 fs_info
->num_tolerated_disk_barrier_failures
=
2959 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2960 if (fs_info
->fs_devices
->missing_devices
>
2961 fs_info
->num_tolerated_disk_barrier_failures
&&
2962 !(sb
->s_flags
& MS_RDONLY
)) {
2963 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2964 fs_info
->fs_devices
->missing_devices
,
2965 fs_info
->num_tolerated_disk_barrier_failures
);
2969 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2971 if (IS_ERR(fs_info
->cleaner_kthread
))
2974 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2976 "btrfs-transaction");
2977 if (IS_ERR(fs_info
->transaction_kthread
))
2980 if (!btrfs_test_opt(tree_root
, SSD
) &&
2981 !btrfs_test_opt(tree_root
, NOSSD
) &&
2982 !fs_info
->fs_devices
->rotating
) {
2983 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
2985 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2989 * Mount does not set all options immediatelly, we can do it now and do
2990 * not have to wait for transaction commit
2992 btrfs_apply_pending_changes(fs_info
);
2994 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2995 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2996 ret
= btrfsic_mount(tree_root
, fs_devices
,
2997 btrfs_test_opt(tree_root
,
2998 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3000 fs_info
->check_integrity_print_mask
);
3002 printk(KERN_WARNING
"BTRFS: failed to initialize"
3003 " integrity check module %s\n", sb
->s_id
);
3006 ret
= btrfs_read_qgroup_config(fs_info
);
3008 goto fail_trans_kthread
;
3010 /* do not make disk changes in broken FS */
3011 if (btrfs_super_log_root(disk_super
) != 0) {
3012 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3019 ret
= btrfs_find_orphan_roots(tree_root
);
3023 if (!(sb
->s_flags
& MS_RDONLY
)) {
3024 ret
= btrfs_cleanup_fs_roots(fs_info
);
3028 mutex_lock(&fs_info
->cleaner_mutex
);
3029 ret
= btrfs_recover_relocation(tree_root
);
3030 mutex_unlock(&fs_info
->cleaner_mutex
);
3033 "BTRFS: failed to recover relocation\n");
3039 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3040 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3041 location
.offset
= 0;
3043 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3044 if (IS_ERR(fs_info
->fs_root
)) {
3045 err
= PTR_ERR(fs_info
->fs_root
);
3049 if (sb
->s_flags
& MS_RDONLY
)
3052 down_read(&fs_info
->cleanup_work_sem
);
3053 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3054 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3055 up_read(&fs_info
->cleanup_work_sem
);
3056 close_ctree(tree_root
);
3059 up_read(&fs_info
->cleanup_work_sem
);
3061 ret
= btrfs_resume_balance_async(fs_info
);
3063 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
3064 close_ctree(tree_root
);
3068 ret
= btrfs_resume_dev_replace_async(fs_info
);
3070 pr_warn("BTRFS: failed to resume dev_replace\n");
3071 close_ctree(tree_root
);
3075 btrfs_qgroup_rescan_resume(fs_info
);
3077 if (!fs_info
->uuid_root
) {
3078 pr_info("BTRFS: creating UUID tree\n");
3079 ret
= btrfs_create_uuid_tree(fs_info
);
3081 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3083 close_ctree(tree_root
);
3086 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3087 fs_info
->generation
!=
3088 btrfs_super_uuid_tree_generation(disk_super
)) {
3089 pr_info("BTRFS: checking UUID tree\n");
3090 ret
= btrfs_check_uuid_tree(fs_info
);
3092 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3094 close_ctree(tree_root
);
3098 fs_info
->update_uuid_tree_gen
= 1;
3106 btrfs_free_qgroup_config(fs_info
);
3108 kthread_stop(fs_info
->transaction_kthread
);
3109 btrfs_cleanup_transaction(fs_info
->tree_root
);
3110 btrfs_free_fs_roots(fs_info
);
3112 kthread_stop(fs_info
->cleaner_kthread
);
3115 * make sure we're done with the btree inode before we stop our
3118 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3121 btrfs_sysfs_remove_mounted(fs_info
);
3124 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3127 btrfs_put_block_group_cache(fs_info
);
3128 btrfs_free_block_groups(fs_info
);
3131 free_root_pointers(fs_info
, 1);
3132 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3135 btrfs_stop_all_workers(fs_info
);
3138 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3140 iput(fs_info
->btree_inode
);
3142 percpu_counter_destroy(&fs_info
->bio_counter
);
3143 fail_delalloc_bytes
:
3144 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3145 fail_dirty_metadata_bytes
:
3146 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3148 bdi_destroy(&fs_info
->bdi
);
3150 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3152 btrfs_free_stripe_hash_table(fs_info
);
3153 btrfs_close_devices(fs_info
->fs_devices
);
3157 if (!btrfs_test_opt(tree_root
, RECOVERY
))
3158 goto fail_tree_roots
;
3160 free_root_pointers(fs_info
, 0);
3162 /* don't use the log in recovery mode, it won't be valid */
3163 btrfs_set_super_log_root(disk_super
, 0);
3165 /* we can't trust the free space cache either */
3166 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3168 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3169 &num_backups_tried
, &backup_index
);
3171 goto fail_block_groups
;
3172 goto retry_root_backup
;
3175 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3178 set_buffer_uptodate(bh
);
3180 struct btrfs_device
*device
= (struct btrfs_device
*)
3183 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3184 "lost page write due to IO error on %s",
3185 rcu_str_deref(device
->name
));
3186 /* note, we dont' set_buffer_write_io_error because we have
3187 * our own ways of dealing with the IO errors
3189 clear_buffer_uptodate(bh
);
3190 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3196 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3197 struct buffer_head
**bh_ret
)
3199 struct buffer_head
*bh
;
3200 struct btrfs_super_block
*super
;
3203 bytenr
= btrfs_sb_offset(copy_num
);
3204 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3207 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3209 * If we fail to read from the underlying devices, as of now
3210 * the best option we have is to mark it EIO.
3215 super
= (struct btrfs_super_block
*)bh
->b_data
;
3216 if (btrfs_super_bytenr(super
) != bytenr
||
3217 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3227 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3229 struct buffer_head
*bh
;
3230 struct buffer_head
*latest
= NULL
;
3231 struct btrfs_super_block
*super
;
3236 /* we would like to check all the supers, but that would make
3237 * a btrfs mount succeed after a mkfs from a different FS.
3238 * So, we need to add a special mount option to scan for
3239 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3241 for (i
= 0; i
< 1; i
++) {
3242 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3246 super
= (struct btrfs_super_block
*)bh
->b_data
;
3248 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3251 transid
= btrfs_super_generation(super
);
3258 return ERR_PTR(ret
);
3264 * this should be called twice, once with wait == 0 and
3265 * once with wait == 1. When wait == 0 is done, all the buffer heads
3266 * we write are pinned.
3268 * They are released when wait == 1 is done.
3269 * max_mirrors must be the same for both runs, and it indicates how
3270 * many supers on this one device should be written.
3272 * max_mirrors == 0 means to write them all.
3274 static int write_dev_supers(struct btrfs_device
*device
,
3275 struct btrfs_super_block
*sb
,
3276 int do_barriers
, int wait
, int max_mirrors
)
3278 struct buffer_head
*bh
;
3285 if (max_mirrors
== 0)
3286 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3288 for (i
= 0; i
< max_mirrors
; i
++) {
3289 bytenr
= btrfs_sb_offset(i
);
3290 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3291 device
->commit_total_bytes
)
3295 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3296 BTRFS_SUPER_INFO_SIZE
);
3302 if (!buffer_uptodate(bh
))
3305 /* drop our reference */
3308 /* drop the reference from the wait == 0 run */
3312 btrfs_set_super_bytenr(sb
, bytenr
);
3315 crc
= btrfs_csum_data((char *)sb
+
3316 BTRFS_CSUM_SIZE
, crc
,
3317 BTRFS_SUPER_INFO_SIZE
-
3319 btrfs_csum_final(crc
, sb
->csum
);
3322 * one reference for us, and we leave it for the
3325 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3326 BTRFS_SUPER_INFO_SIZE
);
3328 btrfs_err(device
->dev_root
->fs_info
,
3329 "couldn't get super buffer head for bytenr %llu",
3335 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3337 /* one reference for submit_bh */
3340 set_buffer_uptodate(bh
);
3342 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3343 bh
->b_private
= device
;
3347 * we fua the first super. The others we allow
3351 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3353 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3357 return errors
< i
? 0 : -1;
3361 * endio for the write_dev_flush, this will wake anyone waiting
3362 * for the barrier when it is done
3364 static void btrfs_end_empty_barrier(struct bio
*bio
)
3366 if (bio
->bi_private
)
3367 complete(bio
->bi_private
);
3372 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3373 * sent down. With wait == 1, it waits for the previous flush.
3375 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3378 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3383 if (device
->nobarriers
)
3387 bio
= device
->flush_bio
;
3391 wait_for_completion(&device
->flush_wait
);
3393 if (bio
->bi_error
) {
3394 ret
= bio
->bi_error
;
3395 btrfs_dev_stat_inc_and_print(device
,
3396 BTRFS_DEV_STAT_FLUSH_ERRS
);
3399 /* drop the reference from the wait == 0 run */
3401 device
->flush_bio
= NULL
;
3407 * one reference for us, and we leave it for the
3410 device
->flush_bio
= NULL
;
3411 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3415 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3416 bio
->bi_bdev
= device
->bdev
;
3417 init_completion(&device
->flush_wait
);
3418 bio
->bi_private
= &device
->flush_wait
;
3419 device
->flush_bio
= bio
;
3422 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3428 * send an empty flush down to each device in parallel,
3429 * then wait for them
3431 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3433 struct list_head
*head
;
3434 struct btrfs_device
*dev
;
3435 int errors_send
= 0;
3436 int errors_wait
= 0;
3439 /* send down all the barriers */
3440 head
= &info
->fs_devices
->devices
;
3441 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3448 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3451 ret
= write_dev_flush(dev
, 0);
3456 /* wait for all the barriers */
3457 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3464 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3467 ret
= write_dev_flush(dev
, 1);
3471 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3472 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3477 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3479 if ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3480 BTRFS_BLOCK_GROUP_RAID0
|
3481 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)) ||
3482 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0))
3485 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3486 BTRFS_BLOCK_GROUP_RAID5
|
3487 BTRFS_BLOCK_GROUP_RAID10
))
3490 if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
3493 pr_warn("BTRFS: unknown raid type: %llu\n", flags
);
3497 int btrfs_calc_num_tolerated_disk_barrier_failures(
3498 struct btrfs_fs_info
*fs_info
)
3500 struct btrfs_ioctl_space_info space
;
3501 struct btrfs_space_info
*sinfo
;
3502 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3503 BTRFS_BLOCK_GROUP_SYSTEM
,
3504 BTRFS_BLOCK_GROUP_METADATA
,
3505 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3508 int num_tolerated_disk_barrier_failures
=
3509 (int)fs_info
->fs_devices
->num_devices
;
3511 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3512 struct btrfs_space_info
*tmp
;
3516 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3517 if (tmp
->flags
== types
[i
]) {
3527 down_read(&sinfo
->groups_sem
);
3528 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3531 if (list_empty(&sinfo
->block_groups
[c
]))
3534 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3536 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3538 flags
= space
.flags
;
3540 num_tolerated_disk_barrier_failures
= min(
3541 num_tolerated_disk_barrier_failures
,
3542 btrfs_get_num_tolerated_disk_barrier_failures(
3545 up_read(&sinfo
->groups_sem
);
3548 return num_tolerated_disk_barrier_failures
;
3551 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3553 struct list_head
*head
;
3554 struct btrfs_device
*dev
;
3555 struct btrfs_super_block
*sb
;
3556 struct btrfs_dev_item
*dev_item
;
3560 int total_errors
= 0;
3563 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3564 backup_super_roots(root
->fs_info
);
3566 sb
= root
->fs_info
->super_for_commit
;
3567 dev_item
= &sb
->dev_item
;
3569 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3570 head
= &root
->fs_info
->fs_devices
->devices
;
3571 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3574 ret
= barrier_all_devices(root
->fs_info
);
3577 &root
->fs_info
->fs_devices
->device_list_mutex
);
3578 btrfs_std_error(root
->fs_info
, ret
,
3579 "errors while submitting device barriers.");
3584 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3589 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3592 btrfs_set_stack_device_generation(dev_item
, 0);
3593 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3594 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3595 btrfs_set_stack_device_total_bytes(dev_item
,
3596 dev
->commit_total_bytes
);
3597 btrfs_set_stack_device_bytes_used(dev_item
,
3598 dev
->commit_bytes_used
);
3599 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3600 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3601 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3602 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3603 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3605 flags
= btrfs_super_flags(sb
);
3606 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3608 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3612 if (total_errors
> max_errors
) {
3613 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3615 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3617 /* FUA is masked off if unsupported and can't be the reason */
3618 btrfs_std_error(root
->fs_info
, -EIO
,
3619 "%d errors while writing supers", total_errors
);
3624 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3627 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3630 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3634 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3635 if (total_errors
> max_errors
) {
3636 btrfs_std_error(root
->fs_info
, -EIO
,
3637 "%d errors while writing supers", total_errors
);
3643 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3644 struct btrfs_root
*root
, int max_mirrors
)
3646 return write_all_supers(root
, max_mirrors
);
3649 /* Drop a fs root from the radix tree and free it. */
3650 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3651 struct btrfs_root
*root
)
3653 spin_lock(&fs_info
->fs_roots_radix_lock
);
3654 radix_tree_delete(&fs_info
->fs_roots_radix
,
3655 (unsigned long)root
->root_key
.objectid
);
3656 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3658 if (btrfs_root_refs(&root
->root_item
) == 0)
3659 synchronize_srcu(&fs_info
->subvol_srcu
);
3661 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3662 btrfs_free_log(NULL
, root
);
3664 if (root
->free_ino_pinned
)
3665 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3666 if (root
->free_ino_ctl
)
3667 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3671 static void free_fs_root(struct btrfs_root
*root
)
3673 iput(root
->ino_cache_inode
);
3674 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3675 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3676 root
->orphan_block_rsv
= NULL
;
3678 free_anon_bdev(root
->anon_dev
);
3679 if (root
->subv_writers
)
3680 btrfs_free_subvolume_writers(root
->subv_writers
);
3681 free_extent_buffer(root
->node
);
3682 free_extent_buffer(root
->commit_root
);
3683 kfree(root
->free_ino_ctl
);
3684 kfree(root
->free_ino_pinned
);
3686 btrfs_put_fs_root(root
);
3689 void btrfs_free_fs_root(struct btrfs_root
*root
)
3694 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3696 u64 root_objectid
= 0;
3697 struct btrfs_root
*gang
[8];
3700 unsigned int ret
= 0;
3704 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3705 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3706 (void **)gang
, root_objectid
,
3709 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3712 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3714 for (i
= 0; i
< ret
; i
++) {
3715 /* Avoid to grab roots in dead_roots */
3716 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3720 /* grab all the search result for later use */
3721 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3723 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3725 for (i
= 0; i
< ret
; i
++) {
3728 root_objectid
= gang
[i
]->root_key
.objectid
;
3729 err
= btrfs_orphan_cleanup(gang
[i
]);
3732 btrfs_put_fs_root(gang
[i
]);
3737 /* release the uncleaned roots due to error */
3738 for (; i
< ret
; i
++) {
3740 btrfs_put_fs_root(gang
[i
]);
3745 int btrfs_commit_super(struct btrfs_root
*root
)
3747 struct btrfs_trans_handle
*trans
;
3749 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3750 btrfs_run_delayed_iputs(root
);
3751 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3752 wake_up_process(root
->fs_info
->cleaner_kthread
);
3754 /* wait until ongoing cleanup work done */
3755 down_write(&root
->fs_info
->cleanup_work_sem
);
3756 up_write(&root
->fs_info
->cleanup_work_sem
);
3758 trans
= btrfs_join_transaction(root
);
3760 return PTR_ERR(trans
);
3761 return btrfs_commit_transaction(trans
, root
);
3764 void close_ctree(struct btrfs_root
*root
)
3766 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3769 fs_info
->closing
= 1;
3772 /* wait for the uuid_scan task to finish */
3773 down(&fs_info
->uuid_tree_rescan_sem
);
3774 /* avoid complains from lockdep et al., set sem back to initial state */
3775 up(&fs_info
->uuid_tree_rescan_sem
);
3777 /* pause restriper - we want to resume on mount */
3778 btrfs_pause_balance(fs_info
);
3780 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3782 btrfs_scrub_cancel(fs_info
);
3784 /* wait for any defraggers to finish */
3785 wait_event(fs_info
->transaction_wait
,
3786 (atomic_read(&fs_info
->defrag_running
) == 0));
3788 /* clear out the rbtree of defraggable inodes */
3789 btrfs_cleanup_defrag_inodes(fs_info
);
3791 cancel_work_sync(&fs_info
->async_reclaim_work
);
3793 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3795 * If the cleaner thread is stopped and there are
3796 * block groups queued for removal, the deletion will be
3797 * skipped when we quit the cleaner thread.
3799 btrfs_delete_unused_bgs(root
->fs_info
);
3801 ret
= btrfs_commit_super(root
);
3803 btrfs_err(fs_info
, "commit super ret %d", ret
);
3806 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3807 btrfs_error_commit_super(root
);
3809 kthread_stop(fs_info
->transaction_kthread
);
3810 kthread_stop(fs_info
->cleaner_kthread
);
3812 fs_info
->closing
= 2;
3815 btrfs_free_qgroup_config(fs_info
);
3817 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3818 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3819 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3822 btrfs_sysfs_remove_mounted(fs_info
);
3823 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3825 btrfs_free_fs_roots(fs_info
);
3827 btrfs_put_block_group_cache(fs_info
);
3829 btrfs_free_block_groups(fs_info
);
3832 * we must make sure there is not any read request to
3833 * submit after we stopping all workers.
3835 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3836 btrfs_stop_all_workers(fs_info
);
3839 free_root_pointers(fs_info
, 1);
3841 iput(fs_info
->btree_inode
);
3843 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3844 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3845 btrfsic_unmount(root
, fs_info
->fs_devices
);
3848 btrfs_close_devices(fs_info
->fs_devices
);
3849 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3851 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3852 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3853 percpu_counter_destroy(&fs_info
->bio_counter
);
3854 bdi_destroy(&fs_info
->bdi
);
3855 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3857 btrfs_free_stripe_hash_table(fs_info
);
3859 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3860 root
->orphan_block_rsv
= NULL
;
3863 while (!list_empty(&fs_info
->pinned_chunks
)) {
3864 struct extent_map
*em
;
3866 em
= list_first_entry(&fs_info
->pinned_chunks
,
3867 struct extent_map
, list
);
3868 list_del_init(&em
->list
);
3869 free_extent_map(em
);
3871 unlock_chunks(root
);
3874 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3878 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3880 ret
= extent_buffer_uptodate(buf
);
3884 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3885 parent_transid
, atomic
);
3891 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3893 return set_extent_buffer_uptodate(buf
);
3896 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3898 struct btrfs_root
*root
;
3899 u64 transid
= btrfs_header_generation(buf
);
3902 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3904 * This is a fast path so only do this check if we have sanity tests
3905 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3906 * outside of the sanity tests.
3908 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3911 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3912 btrfs_assert_tree_locked(buf
);
3913 if (transid
!= root
->fs_info
->generation
)
3914 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3915 "found %llu running %llu\n",
3916 buf
->start
, transid
, root
->fs_info
->generation
);
3917 was_dirty
= set_extent_buffer_dirty(buf
);
3919 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3921 root
->fs_info
->dirty_metadata_batch
);
3922 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3923 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3924 btrfs_print_leaf(root
, buf
);
3930 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3934 * looks as though older kernels can get into trouble with
3935 * this code, they end up stuck in balance_dirty_pages forever
3939 if (current
->flags
& PF_MEMALLOC
)
3943 btrfs_balance_delayed_items(root
);
3945 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3946 BTRFS_DIRTY_METADATA_THRESH
);
3948 balance_dirty_pages_ratelimited(
3949 root
->fs_info
->btree_inode
->i_mapping
);
3954 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3956 __btrfs_btree_balance_dirty(root
, 1);
3959 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3961 __btrfs_btree_balance_dirty(root
, 0);
3964 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3966 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3967 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3970 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3973 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
3976 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3977 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
3978 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
3981 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3982 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
3983 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
3986 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3987 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
3988 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
3993 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3994 * items yet, they'll be verified later. Issue just a warning.
3996 if (!IS_ALIGNED(btrfs_super_root(sb
), 4096))
3997 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
3998 btrfs_super_root(sb
));
3999 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), 4096))
4000 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
4001 btrfs_super_chunk_root(sb
));
4002 if (!IS_ALIGNED(btrfs_super_log_root(sb
), 4096))
4003 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
4004 btrfs_super_log_root(sb
));
4007 * Check the lower bound, the alignment and other constraints are
4010 if (btrfs_super_nodesize(sb
) < 4096) {
4011 printk(KERN_ERR
"BTRFS: nodesize too small: %u < 4096\n",
4012 btrfs_super_nodesize(sb
));
4015 if (btrfs_super_sectorsize(sb
) < 4096) {
4016 printk(KERN_ERR
"BTRFS: sectorsize too small: %u < 4096\n",
4017 btrfs_super_sectorsize(sb
));
4021 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4022 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4023 fs_info
->fsid
, sb
->dev_item
.fsid
);
4028 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4031 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4032 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
4033 btrfs_super_num_devices(sb
));
4034 if (btrfs_super_num_devices(sb
) == 0) {
4035 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
4039 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4040 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
4041 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4046 * Obvious sys_chunk_array corruptions, it must hold at least one key
4049 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4050 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
4051 btrfs_super_sys_array_size(sb
),
4052 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4055 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4056 + sizeof(struct btrfs_chunk
)) {
4057 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
4058 btrfs_super_sys_array_size(sb
),
4059 sizeof(struct btrfs_disk_key
)
4060 + sizeof(struct btrfs_chunk
));
4065 * The generation is a global counter, we'll trust it more than the others
4066 * but it's still possible that it's the one that's wrong.
4068 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4070 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4071 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
4072 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4073 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4075 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4076 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
4081 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4083 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4084 btrfs_run_delayed_iputs(root
);
4085 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4087 down_write(&root
->fs_info
->cleanup_work_sem
);
4088 up_write(&root
->fs_info
->cleanup_work_sem
);
4090 /* cleanup FS via transaction */
4091 btrfs_cleanup_transaction(root
);
4094 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4096 struct btrfs_ordered_extent
*ordered
;
4098 spin_lock(&root
->ordered_extent_lock
);
4100 * This will just short circuit the ordered completion stuff which will
4101 * make sure the ordered extent gets properly cleaned up.
4103 list_for_each_entry(ordered
, &root
->ordered_extents
,
4105 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4106 spin_unlock(&root
->ordered_extent_lock
);
4109 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4111 struct btrfs_root
*root
;
4112 struct list_head splice
;
4114 INIT_LIST_HEAD(&splice
);
4116 spin_lock(&fs_info
->ordered_root_lock
);
4117 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4118 while (!list_empty(&splice
)) {
4119 root
= list_first_entry(&splice
, struct btrfs_root
,
4121 list_move_tail(&root
->ordered_root
,
4122 &fs_info
->ordered_roots
);
4124 spin_unlock(&fs_info
->ordered_root_lock
);
4125 btrfs_destroy_ordered_extents(root
);
4128 spin_lock(&fs_info
->ordered_root_lock
);
4130 spin_unlock(&fs_info
->ordered_root_lock
);
4133 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4134 struct btrfs_root
*root
)
4136 struct rb_node
*node
;
4137 struct btrfs_delayed_ref_root
*delayed_refs
;
4138 struct btrfs_delayed_ref_node
*ref
;
4141 delayed_refs
= &trans
->delayed_refs
;
4143 spin_lock(&delayed_refs
->lock
);
4144 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4145 spin_unlock(&delayed_refs
->lock
);
4146 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4150 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4151 struct btrfs_delayed_ref_head
*head
;
4152 struct btrfs_delayed_ref_node
*tmp
;
4153 bool pin_bytes
= false;
4155 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4157 if (!mutex_trylock(&head
->mutex
)) {
4158 atomic_inc(&head
->node
.refs
);
4159 spin_unlock(&delayed_refs
->lock
);
4161 mutex_lock(&head
->mutex
);
4162 mutex_unlock(&head
->mutex
);
4163 btrfs_put_delayed_ref(&head
->node
);
4164 spin_lock(&delayed_refs
->lock
);
4167 spin_lock(&head
->lock
);
4168 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4171 list_del(&ref
->list
);
4172 atomic_dec(&delayed_refs
->num_entries
);
4173 btrfs_put_delayed_ref(ref
);
4175 if (head
->must_insert_reserved
)
4177 btrfs_free_delayed_extent_op(head
->extent_op
);
4178 delayed_refs
->num_heads
--;
4179 if (head
->processing
== 0)
4180 delayed_refs
->num_heads_ready
--;
4181 atomic_dec(&delayed_refs
->num_entries
);
4182 head
->node
.in_tree
= 0;
4183 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4184 spin_unlock(&head
->lock
);
4185 spin_unlock(&delayed_refs
->lock
);
4186 mutex_unlock(&head
->mutex
);
4189 btrfs_pin_extent(root
, head
->node
.bytenr
,
4190 head
->node
.num_bytes
, 1);
4191 btrfs_put_delayed_ref(&head
->node
);
4193 spin_lock(&delayed_refs
->lock
);
4196 spin_unlock(&delayed_refs
->lock
);
4201 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4203 struct btrfs_inode
*btrfs_inode
;
4204 struct list_head splice
;
4206 INIT_LIST_HEAD(&splice
);
4208 spin_lock(&root
->delalloc_lock
);
4209 list_splice_init(&root
->delalloc_inodes
, &splice
);
4211 while (!list_empty(&splice
)) {
4212 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4215 list_del_init(&btrfs_inode
->delalloc_inodes
);
4216 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4217 &btrfs_inode
->runtime_flags
);
4218 spin_unlock(&root
->delalloc_lock
);
4220 btrfs_invalidate_inodes(btrfs_inode
->root
);
4222 spin_lock(&root
->delalloc_lock
);
4225 spin_unlock(&root
->delalloc_lock
);
4228 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4230 struct btrfs_root
*root
;
4231 struct list_head splice
;
4233 INIT_LIST_HEAD(&splice
);
4235 spin_lock(&fs_info
->delalloc_root_lock
);
4236 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4237 while (!list_empty(&splice
)) {
4238 root
= list_first_entry(&splice
, struct btrfs_root
,
4240 list_del_init(&root
->delalloc_root
);
4241 root
= btrfs_grab_fs_root(root
);
4243 spin_unlock(&fs_info
->delalloc_root_lock
);
4245 btrfs_destroy_delalloc_inodes(root
);
4246 btrfs_put_fs_root(root
);
4248 spin_lock(&fs_info
->delalloc_root_lock
);
4250 spin_unlock(&fs_info
->delalloc_root_lock
);
4253 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4254 struct extent_io_tree
*dirty_pages
,
4258 struct extent_buffer
*eb
;
4263 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4268 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4269 while (start
<= end
) {
4270 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4271 start
+= root
->nodesize
;
4274 wait_on_extent_buffer_writeback(eb
);
4276 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4278 clear_extent_buffer_dirty(eb
);
4279 free_extent_buffer_stale(eb
);
4286 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4287 struct extent_io_tree
*pinned_extents
)
4289 struct extent_io_tree
*unpin
;
4295 unpin
= pinned_extents
;
4298 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4299 EXTENT_DIRTY
, NULL
);
4303 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4304 btrfs_error_unpin_extent_range(root
, start
, end
);
4309 if (unpin
== &root
->fs_info
->freed_extents
[0])
4310 unpin
= &root
->fs_info
->freed_extents
[1];
4312 unpin
= &root
->fs_info
->freed_extents
[0];
4320 static void btrfs_free_pending_ordered(struct btrfs_transaction
*cur_trans
,
4321 struct btrfs_fs_info
*fs_info
)
4323 struct btrfs_ordered_extent
*ordered
;
4325 spin_lock(&fs_info
->trans_lock
);
4326 while (!list_empty(&cur_trans
->pending_ordered
)) {
4327 ordered
= list_first_entry(&cur_trans
->pending_ordered
,
4328 struct btrfs_ordered_extent
,
4330 list_del_init(&ordered
->trans_list
);
4331 spin_unlock(&fs_info
->trans_lock
);
4333 btrfs_put_ordered_extent(ordered
);
4334 spin_lock(&fs_info
->trans_lock
);
4336 spin_unlock(&fs_info
->trans_lock
);
4339 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4340 struct btrfs_root
*root
)
4342 btrfs_destroy_delayed_refs(cur_trans
, root
);
4344 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4345 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4347 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4348 wake_up(&root
->fs_info
->transaction_wait
);
4350 btrfs_free_pending_ordered(cur_trans
, root
->fs_info
);
4351 btrfs_destroy_delayed_inodes(root
);
4352 btrfs_assert_delayed_root_empty(root
);
4354 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4356 btrfs_destroy_pinned_extent(root
,
4357 root
->fs_info
->pinned_extents
);
4359 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4360 wake_up(&cur_trans
->commit_wait
);
4363 memset(cur_trans, 0, sizeof(*cur_trans));
4364 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4368 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4370 struct btrfs_transaction
*t
;
4372 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4374 spin_lock(&root
->fs_info
->trans_lock
);
4375 while (!list_empty(&root
->fs_info
->trans_list
)) {
4376 t
= list_first_entry(&root
->fs_info
->trans_list
,
4377 struct btrfs_transaction
, list
);
4378 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4379 atomic_inc(&t
->use_count
);
4380 spin_unlock(&root
->fs_info
->trans_lock
);
4381 btrfs_wait_for_commit(root
, t
->transid
);
4382 btrfs_put_transaction(t
);
4383 spin_lock(&root
->fs_info
->trans_lock
);
4386 if (t
== root
->fs_info
->running_transaction
) {
4387 t
->state
= TRANS_STATE_COMMIT_DOING
;
4388 spin_unlock(&root
->fs_info
->trans_lock
);
4390 * We wait for 0 num_writers since we don't hold a trans
4391 * handle open currently for this transaction.
4393 wait_event(t
->writer_wait
,
4394 atomic_read(&t
->num_writers
) == 0);
4396 spin_unlock(&root
->fs_info
->trans_lock
);
4398 btrfs_cleanup_one_transaction(t
, root
);
4400 spin_lock(&root
->fs_info
->trans_lock
);
4401 if (t
== root
->fs_info
->running_transaction
)
4402 root
->fs_info
->running_transaction
= NULL
;
4403 list_del_init(&t
->list
);
4404 spin_unlock(&root
->fs_info
->trans_lock
);
4406 btrfs_put_transaction(t
);
4407 trace_btrfs_transaction_commit(root
);
4408 spin_lock(&root
->fs_info
->trans_lock
);
4410 spin_unlock(&root
->fs_info
->trans_lock
);
4411 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4412 btrfs_destroy_delayed_inodes(root
);
4413 btrfs_assert_delayed_root_empty(root
);
4414 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4415 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4416 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4421 static const struct extent_io_ops btree_extent_io_ops
= {
4422 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4423 .readpage_io_failed_hook
= btree_io_failed_hook
,
4424 .submit_bio_hook
= btree_submit_bio_hook
,
4425 /* note we're sharing with inode.c for the merge bio hook */
4426 .merge_bio_hook
= btrfs_merge_bio_hook
,