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/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
53 #include <asm/cpufeature.h>
56 static struct extent_io_ops btree_extent_io_ops
;
57 static void end_workqueue_fn(struct btrfs_work
*work
);
58 static void free_fs_root(struct btrfs_root
*root
);
59 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
61 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
62 struct btrfs_root
*root
);
63 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
65 struct btrfs_root
*root
);
66 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
68 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
69 struct extent_io_tree
*dirty_pages
,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
72 struct extent_io_tree
*pinned_extents
);
73 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
74 static void btrfs_error_commit_super(struct btrfs_root
*root
);
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
85 struct btrfs_fs_info
*info
;
88 struct list_head list
;
89 struct btrfs_work work
;
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
97 struct async_submit_bio
{
100 struct list_head list
;
101 extent_submit_bio_hook_t
*submit_bio_start
;
102 extent_submit_bio_hook_t
*submit_bio_done
;
105 unsigned long bio_flags
;
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
111 struct btrfs_work work
;
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
143 static struct btrfs_lockdep_keyset
{
144 u64 id
; /* root objectid */
145 const char *name_stem
; /* lock name stem */
146 char names
[BTRFS_MAX_LEVEL
+ 1][20];
147 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
148 } btrfs_lockdep_keysets
[] = {
149 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
150 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
151 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
152 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
153 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
154 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
155 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
156 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
157 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
158 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
159 { .id
= 0, .name_stem
= "tree" },
162 void __init
btrfs_init_lockdep(void)
166 /* initialize lockdep class names */
167 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
168 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
170 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
171 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
172 "btrfs-%s-%02d", ks
->name_stem
, j
);
176 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
179 struct btrfs_lockdep_keyset
*ks
;
181 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
185 if (ks
->id
== objectid
)
188 lockdep_set_class_and_name(&eb
->lock
,
189 &ks
->keys
[level
], ks
->names
[level
]);
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
198 static struct extent_map
*btree_get_extent(struct inode
*inode
,
199 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
202 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
203 struct extent_map
*em
;
206 read_lock(&em_tree
->lock
);
207 em
= lookup_extent_mapping(em_tree
, start
, len
);
210 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
211 read_unlock(&em_tree
->lock
);
214 read_unlock(&em_tree
->lock
);
216 em
= alloc_extent_map();
218 em
= ERR_PTR(-ENOMEM
);
223 em
->block_len
= (u64
)-1;
225 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
227 write_lock(&em_tree
->lock
);
228 ret
= add_extent_mapping(em_tree
, em
, 0);
229 if (ret
== -EEXIST
) {
231 em
= lookup_extent_mapping(em_tree
, start
, len
);
238 write_unlock(&em_tree
->lock
);
244 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
246 return crc32c(seed
, data
, len
);
249 void btrfs_csum_final(u32 crc
, char *result
)
251 put_unaligned_le32(~crc
, result
);
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
258 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
261 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
264 unsigned long cur_len
;
265 unsigned long offset
= BTRFS_CSUM_SIZE
;
267 unsigned long map_start
;
268 unsigned long map_len
;
271 unsigned long inline_result
;
273 len
= buf
->len
- offset
;
275 err
= map_private_extent_buffer(buf
, offset
, 32,
276 &kaddr
, &map_start
, &map_len
);
279 cur_len
= min(len
, map_len
- (offset
- map_start
));
280 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
285 if (csum_size
> sizeof(inline_result
)) {
286 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
290 result
= (char *)&inline_result
;
293 btrfs_csum_final(crc
, result
);
296 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
299 memcpy(&found
, result
, csum_size
);
301 read_extent_buffer(buf
, &val
, 0, csum_size
);
302 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
305 root
->fs_info
->sb
->s_id
,
306 (unsigned long long)buf
->start
, val
, found
,
307 btrfs_header_level(buf
));
308 if (result
!= (char *)&inline_result
)
313 write_extent_buffer(buf
, result
, 0, csum_size
);
315 if (result
!= (char *)&inline_result
)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
327 struct extent_buffer
*eb
, u64 parent_transid
,
330 struct extent_state
*cached_state
= NULL
;
333 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
339 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
341 if (extent_buffer_uptodate(eb
) &&
342 btrfs_header_generation(eb
) == parent_transid
) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 (unsigned long long)eb
->start
,
349 (unsigned long long)parent_transid
,
350 (unsigned long long)btrfs_header_generation(eb
));
352 clear_extent_buffer_uptodate(eb
);
354 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
355 &cached_state
, GFP_NOFS
);
360 * Return 0 if the superblock checksum type matches the checksum value of that
361 * algorithm. Pass the raw disk superblock data.
363 static int btrfs_check_super_csum(char *raw_disk_sb
)
365 struct btrfs_super_block
*disk_sb
=
366 (struct btrfs_super_block
*)raw_disk_sb
;
367 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
370 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
372 const int csum_size
= sizeof(crc
);
373 char result
[csum_size
];
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 * is filled with zeros and is included in the checkum.
380 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
381 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
382 btrfs_csum_final(crc
, result
);
384 if (memcmp(raw_disk_sb
, result
, csum_size
))
387 if (ret
&& btrfs_super_generation(disk_sb
) < 10) {
388 printk(KERN_WARNING
"btrfs: super block crcs don't match, older mkfs detected\n");
393 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
394 printk(KERN_ERR
"btrfs: unsupported checksum algorithm %u\n",
403 * helper to read a given tree block, doing retries as required when
404 * the checksums don't match and we have alternate mirrors to try.
406 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
407 struct extent_buffer
*eb
,
408 u64 start
, u64 parent_transid
)
410 struct extent_io_tree
*io_tree
;
415 int failed_mirror
= 0;
417 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
418 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
420 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
422 btree_get_extent
, mirror_num
);
424 if (!verify_parent_transid(io_tree
, eb
,
432 * This buffer's crc is fine, but its contents are corrupted, so
433 * there is no reason to read the other copies, they won't be
436 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
439 num_copies
= btrfs_num_copies(root
->fs_info
,
444 if (!failed_mirror
) {
446 failed_mirror
= eb
->read_mirror
;
450 if (mirror_num
== failed_mirror
)
453 if (mirror_num
> num_copies
)
457 if (failed
&& !ret
&& failed_mirror
)
458 repair_eb_io_failure(root
, eb
, failed_mirror
);
464 * checksum a dirty tree block before IO. This has extra checks to make sure
465 * we only fill in the checksum field in the first page of a multi-page block
468 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
470 struct extent_io_tree
*tree
;
471 u64 start
= page_offset(page
);
473 struct extent_buffer
*eb
;
475 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
477 eb
= (struct extent_buffer
*)page
->private;
478 if (page
!= eb
->pages
[0])
480 found_start
= btrfs_header_bytenr(eb
);
481 if (found_start
!= start
) {
485 if (!PageUptodate(page
)) {
489 csum_tree_block(root
, eb
, 0);
493 static int check_tree_block_fsid(struct btrfs_root
*root
,
494 struct extent_buffer
*eb
)
496 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
497 u8 fsid
[BTRFS_UUID_SIZE
];
500 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
503 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
507 fs_devices
= fs_devices
->seed
;
512 #define CORRUPT(reason, eb, root, slot) \
513 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
514 "root=%llu, slot=%d\n", reason, \
515 (unsigned long long)btrfs_header_bytenr(eb), \
516 (unsigned long long)root->objectid, slot)
518 static noinline
int check_leaf(struct btrfs_root
*root
,
519 struct extent_buffer
*leaf
)
521 struct btrfs_key key
;
522 struct btrfs_key leaf_key
;
523 u32 nritems
= btrfs_header_nritems(leaf
);
529 /* Check the 0 item */
530 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
531 BTRFS_LEAF_DATA_SIZE(root
)) {
532 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
537 * Check to make sure each items keys are in the correct order and their
538 * offsets make sense. We only have to loop through nritems-1 because
539 * we check the current slot against the next slot, which verifies the
540 * next slot's offset+size makes sense and that the current's slot
543 for (slot
= 0; slot
< nritems
- 1; slot
++) {
544 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
545 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
547 /* Make sure the keys are in the right order */
548 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
549 CORRUPT("bad key order", leaf
, root
, slot
);
554 * Make sure the offset and ends are right, remember that the
555 * item data starts at the end of the leaf and grows towards the
558 if (btrfs_item_offset_nr(leaf
, slot
) !=
559 btrfs_item_end_nr(leaf
, slot
+ 1)) {
560 CORRUPT("slot offset bad", leaf
, root
, slot
);
565 * Check to make sure that we don't point outside of the leaf,
566 * just incase all the items are consistent to eachother, but
567 * all point outside of the leaf.
569 if (btrfs_item_end_nr(leaf
, slot
) >
570 BTRFS_LEAF_DATA_SIZE(root
)) {
571 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
579 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
580 u64 phy_offset
, struct page
*page
,
581 u64 start
, u64 end
, int mirror
)
583 struct extent_io_tree
*tree
;
586 struct extent_buffer
*eb
;
587 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
594 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
595 eb
= (struct extent_buffer
*)page
->private;
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
600 extent_buffer_get(eb
);
602 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
606 eb
->read_mirror
= mirror
;
607 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
612 found_start
= btrfs_header_bytenr(eb
);
613 if (found_start
!= eb
->start
) {
614 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
616 (unsigned long long)found_start
,
617 (unsigned long long)eb
->start
);
621 if (check_tree_block_fsid(root
, eb
)) {
622 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
623 (unsigned long long)eb
->start
);
627 found_level
= btrfs_header_level(eb
);
628 if (found_level
>= BTRFS_MAX_LEVEL
) {
629 btrfs_info(root
->fs_info
, "bad tree block level %d\n",
630 (int)btrfs_header_level(eb
));
635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
638 ret
= csum_tree_block(root
, eb
, 1);
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
649 if (found_level
== 0 && check_leaf(root
, eb
)) {
650 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
655 set_extent_buffer_uptodate(eb
);
658 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
659 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
663 * our io error hook is going to dec the io pages
664 * again, we have to make sure it has something
667 atomic_inc(&eb
->io_pages
);
668 clear_extent_buffer_uptodate(eb
);
670 free_extent_buffer(eb
);
675 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
677 struct extent_buffer
*eb
;
678 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
680 eb
= (struct extent_buffer
*)page
->private;
681 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
682 eb
->read_mirror
= failed_mirror
;
683 atomic_dec(&eb
->io_pages
);
684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
685 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
686 return -EIO
; /* we fixed nothing */
689 static void end_workqueue_bio(struct bio
*bio
, int err
)
691 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
692 struct btrfs_fs_info
*fs_info
;
694 fs_info
= end_io_wq
->info
;
695 end_io_wq
->error
= err
;
696 end_io_wq
->work
.func
= end_workqueue_fn
;
697 end_io_wq
->work
.flags
= 0;
699 if (bio
->bi_rw
& REQ_WRITE
) {
700 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
701 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
703 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
704 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
706 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
707 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
710 btrfs_queue_worker(&fs_info
->endio_write_workers
,
713 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
714 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
716 else if (end_io_wq
->metadata
)
717 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
720 btrfs_queue_worker(&fs_info
->endio_workers
,
726 * For the metadata arg you want
729 * 1 - if normal metadta
730 * 2 - if writing to the free space cache area
731 * 3 - raid parity work
733 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
736 struct end_io_wq
*end_io_wq
;
737 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
741 end_io_wq
->private = bio
->bi_private
;
742 end_io_wq
->end_io
= bio
->bi_end_io
;
743 end_io_wq
->info
= info
;
744 end_io_wq
->error
= 0;
745 end_io_wq
->bio
= bio
;
746 end_io_wq
->metadata
= metadata
;
748 bio
->bi_private
= end_io_wq
;
749 bio
->bi_end_io
= end_workqueue_bio
;
753 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
755 unsigned long limit
= min_t(unsigned long,
756 info
->workers
.max_workers
,
757 info
->fs_devices
->open_devices
);
761 static void run_one_async_start(struct btrfs_work
*work
)
763 struct async_submit_bio
*async
;
766 async
= container_of(work
, struct async_submit_bio
, work
);
767 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
768 async
->mirror_num
, async
->bio_flags
,
774 static void run_one_async_done(struct btrfs_work
*work
)
776 struct btrfs_fs_info
*fs_info
;
777 struct async_submit_bio
*async
;
780 async
= container_of(work
, struct async_submit_bio
, work
);
781 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
783 limit
= btrfs_async_submit_limit(fs_info
);
784 limit
= limit
* 2 / 3;
786 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
787 waitqueue_active(&fs_info
->async_submit_wait
))
788 wake_up(&fs_info
->async_submit_wait
);
790 /* If an error occured we just want to clean up the bio and move on */
792 bio_endio(async
->bio
, async
->error
);
796 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
797 async
->mirror_num
, async
->bio_flags
,
801 static void run_one_async_free(struct btrfs_work
*work
)
803 struct async_submit_bio
*async
;
805 async
= container_of(work
, struct async_submit_bio
, work
);
809 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
810 int rw
, struct bio
*bio
, int mirror_num
,
811 unsigned long bio_flags
,
813 extent_submit_bio_hook_t
*submit_bio_start
,
814 extent_submit_bio_hook_t
*submit_bio_done
)
816 struct async_submit_bio
*async
;
818 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
822 async
->inode
= inode
;
825 async
->mirror_num
= mirror_num
;
826 async
->submit_bio_start
= submit_bio_start
;
827 async
->submit_bio_done
= submit_bio_done
;
829 async
->work
.func
= run_one_async_start
;
830 async
->work
.ordered_func
= run_one_async_done
;
831 async
->work
.ordered_free
= run_one_async_free
;
833 async
->work
.flags
= 0;
834 async
->bio_flags
= bio_flags
;
835 async
->bio_offset
= bio_offset
;
839 atomic_inc(&fs_info
->nr_async_submits
);
842 btrfs_set_work_high_prio(&async
->work
);
844 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
846 while (atomic_read(&fs_info
->async_submit_draining
) &&
847 atomic_read(&fs_info
->nr_async_submits
)) {
848 wait_event(fs_info
->async_submit_wait
,
849 (atomic_read(&fs_info
->nr_async_submits
) == 0));
855 static int btree_csum_one_bio(struct bio
*bio
)
857 struct bio_vec
*bvec
= bio
->bi_io_vec
;
859 struct btrfs_root
*root
;
862 WARN_ON(bio
->bi_vcnt
<= 0);
863 while (bio_index
< bio
->bi_vcnt
) {
864 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
865 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
874 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
875 struct bio
*bio
, int mirror_num
,
876 unsigned long bio_flags
,
880 * when we're called for a write, we're already in the async
881 * submission context. Just jump into btrfs_map_bio
883 return btree_csum_one_bio(bio
);
886 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
887 int mirror_num
, unsigned long bio_flags
,
893 * when we're called for a write, we're already in the async
894 * submission context. Just jump into btrfs_map_bio
896 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
902 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
904 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
913 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
914 int mirror_num
, unsigned long bio_flags
,
917 int async
= check_async_write(inode
, bio_flags
);
920 if (!(rw
& REQ_WRITE
)) {
922 * called for a read, do the setup so that checksum validation
923 * can happen in the async kernel threads
925 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
929 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
932 ret
= btree_csum_one_bio(bio
);
935 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
942 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
943 inode
, rw
, bio
, mirror_num
, 0,
945 __btree_submit_bio_start
,
946 __btree_submit_bio_done
);
956 #ifdef CONFIG_MIGRATION
957 static int btree_migratepage(struct address_space
*mapping
,
958 struct page
*newpage
, struct page
*page
,
959 enum migrate_mode mode
)
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
971 if (page_has_private(page
) &&
972 !try_to_release_page(page
, GFP_KERNEL
))
974 return migrate_page(mapping
, newpage
, page
, mode
);
979 static int btree_writepages(struct address_space
*mapping
,
980 struct writeback_control
*wbc
)
982 struct extent_io_tree
*tree
;
983 struct btrfs_fs_info
*fs_info
;
986 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
987 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
989 if (wbc
->for_kupdate
)
992 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
993 /* this is a bit racy, but that's ok */
994 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
995 BTRFS_DIRTY_METADATA_THRESH
);
999 return btree_write_cache_pages(mapping
, wbc
);
1002 static int btree_readpage(struct file
*file
, struct page
*page
)
1004 struct extent_io_tree
*tree
;
1005 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1006 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1009 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1011 if (PageWriteback(page
) || PageDirty(page
))
1014 return try_release_extent_buffer(page
);
1017 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1018 unsigned int length
)
1020 struct extent_io_tree
*tree
;
1021 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1022 extent_invalidatepage(tree
, page
, offset
);
1023 btree_releasepage(page
, GFP_NOFS
);
1024 if (PagePrivate(page
)) {
1025 printk(KERN_WARNING
"btrfs warning page private not zero "
1026 "on page %llu\n", (unsigned long long)page_offset(page
));
1027 ClearPagePrivate(page
);
1028 set_page_private(page
, 0);
1029 page_cache_release(page
);
1033 static int btree_set_page_dirty(struct page
*page
)
1036 struct extent_buffer
*eb
;
1038 BUG_ON(!PagePrivate(page
));
1039 eb
= (struct extent_buffer
*)page
->private;
1041 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1042 BUG_ON(!atomic_read(&eb
->refs
));
1043 btrfs_assert_tree_locked(eb
);
1045 return __set_page_dirty_nobuffers(page
);
1048 static const struct address_space_operations btree_aops
= {
1049 .readpage
= btree_readpage
,
1050 .writepages
= btree_writepages
,
1051 .releasepage
= btree_releasepage
,
1052 .invalidatepage
= btree_invalidatepage
,
1053 #ifdef CONFIG_MIGRATION
1054 .migratepage
= btree_migratepage
,
1056 .set_page_dirty
= btree_set_page_dirty
,
1059 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1062 struct extent_buffer
*buf
= NULL
;
1063 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1066 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1069 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1070 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1071 free_extent_buffer(buf
);
1075 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1076 int mirror_num
, struct extent_buffer
**eb
)
1078 struct extent_buffer
*buf
= NULL
;
1079 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1080 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1083 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1087 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1089 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1090 btree_get_extent
, mirror_num
);
1092 free_extent_buffer(buf
);
1096 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1097 free_extent_buffer(buf
);
1099 } else if (extent_buffer_uptodate(buf
)) {
1102 free_extent_buffer(buf
);
1107 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1108 u64 bytenr
, u32 blocksize
)
1110 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1111 struct extent_buffer
*eb
;
1112 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1117 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1118 u64 bytenr
, u32 blocksize
)
1120 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1121 struct extent_buffer
*eb
;
1123 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1129 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1131 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1132 buf
->start
+ buf
->len
- 1);
1135 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1137 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1138 buf
->start
, buf
->start
+ buf
->len
- 1);
1141 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1142 u32 blocksize
, u64 parent_transid
)
1144 struct extent_buffer
*buf
= NULL
;
1147 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1151 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1153 free_extent_buffer(buf
);
1160 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1161 struct extent_buffer
*buf
)
1163 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1165 if (btrfs_header_generation(buf
) ==
1166 fs_info
->running_transaction
->transid
) {
1167 btrfs_assert_tree_locked(buf
);
1169 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1170 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1172 fs_info
->dirty_metadata_batch
);
1173 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1174 btrfs_set_lock_blocking(buf
);
1175 clear_extent_buffer_dirty(buf
);
1180 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1181 u32 stripesize
, struct btrfs_root
*root
,
1182 struct btrfs_fs_info
*fs_info
,
1186 root
->commit_root
= NULL
;
1187 root
->sectorsize
= sectorsize
;
1188 root
->nodesize
= nodesize
;
1189 root
->leafsize
= leafsize
;
1190 root
->stripesize
= stripesize
;
1192 root
->track_dirty
= 0;
1194 root
->orphan_item_inserted
= 0;
1195 root
->orphan_cleanup_state
= 0;
1197 root
->objectid
= objectid
;
1198 root
->last_trans
= 0;
1199 root
->highest_objectid
= 0;
1200 root
->nr_delalloc_inodes
= 0;
1201 root
->nr_ordered_extents
= 0;
1203 root
->inode_tree
= RB_ROOT
;
1204 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1205 root
->block_rsv
= NULL
;
1206 root
->orphan_block_rsv
= NULL
;
1208 INIT_LIST_HEAD(&root
->dirty_list
);
1209 INIT_LIST_HEAD(&root
->root_list
);
1210 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1211 INIT_LIST_HEAD(&root
->delalloc_root
);
1212 INIT_LIST_HEAD(&root
->ordered_extents
);
1213 INIT_LIST_HEAD(&root
->ordered_root
);
1214 INIT_LIST_HEAD(&root
->logged_list
[0]);
1215 INIT_LIST_HEAD(&root
->logged_list
[1]);
1216 spin_lock_init(&root
->orphan_lock
);
1217 spin_lock_init(&root
->inode_lock
);
1218 spin_lock_init(&root
->delalloc_lock
);
1219 spin_lock_init(&root
->ordered_extent_lock
);
1220 spin_lock_init(&root
->accounting_lock
);
1221 spin_lock_init(&root
->log_extents_lock
[0]);
1222 spin_lock_init(&root
->log_extents_lock
[1]);
1223 mutex_init(&root
->objectid_mutex
);
1224 mutex_init(&root
->log_mutex
);
1225 init_waitqueue_head(&root
->log_writer_wait
);
1226 init_waitqueue_head(&root
->log_commit_wait
[0]);
1227 init_waitqueue_head(&root
->log_commit_wait
[1]);
1228 atomic_set(&root
->log_commit
[0], 0);
1229 atomic_set(&root
->log_commit
[1], 0);
1230 atomic_set(&root
->log_writers
, 0);
1231 atomic_set(&root
->log_batch
, 0);
1232 atomic_set(&root
->orphan_inodes
, 0);
1233 atomic_set(&root
->refs
, 1);
1234 root
->log_transid
= 0;
1235 root
->last_log_commit
= 0;
1236 extent_io_tree_init(&root
->dirty_log_pages
,
1237 fs_info
->btree_inode
->i_mapping
);
1239 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1240 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1241 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1242 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1243 root
->defrag_trans_start
= fs_info
->generation
;
1244 init_completion(&root
->kobj_unregister
);
1245 root
->defrag_running
= 0;
1246 root
->root_key
.objectid
= objectid
;
1249 spin_lock_init(&root
->root_item_lock
);
1252 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1254 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1256 root
->fs_info
= fs_info
;
1260 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1261 struct btrfs_fs_info
*fs_info
,
1264 struct extent_buffer
*leaf
;
1265 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1266 struct btrfs_root
*root
;
1267 struct btrfs_key key
;
1272 root
= btrfs_alloc_root(fs_info
);
1274 return ERR_PTR(-ENOMEM
);
1276 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1277 tree_root
->sectorsize
, tree_root
->stripesize
,
1278 root
, fs_info
, objectid
);
1279 root
->root_key
.objectid
= objectid
;
1280 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1281 root
->root_key
.offset
= 0;
1283 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1284 0, objectid
, NULL
, 0, 0, 0);
1286 ret
= PTR_ERR(leaf
);
1291 bytenr
= leaf
->start
;
1292 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1293 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1294 btrfs_set_header_generation(leaf
, trans
->transid
);
1295 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1296 btrfs_set_header_owner(leaf
, objectid
);
1299 write_extent_buffer(leaf
, fs_info
->fsid
,
1300 (unsigned long)btrfs_header_fsid(leaf
),
1302 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1303 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1305 btrfs_mark_buffer_dirty(leaf
);
1307 root
->commit_root
= btrfs_root_node(root
);
1308 root
->track_dirty
= 1;
1311 root
->root_item
.flags
= 0;
1312 root
->root_item
.byte_limit
= 0;
1313 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1314 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1315 btrfs_set_root_level(&root
->root_item
, 0);
1316 btrfs_set_root_refs(&root
->root_item
, 1);
1317 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1318 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1319 btrfs_set_root_dirid(&root
->root_item
, 0);
1321 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1322 root
->root_item
.drop_level
= 0;
1324 key
.objectid
= objectid
;
1325 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1327 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1331 btrfs_tree_unlock(leaf
);
1337 btrfs_tree_unlock(leaf
);
1338 free_extent_buffer(leaf
);
1342 return ERR_PTR(ret
);
1345 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1346 struct btrfs_fs_info
*fs_info
)
1348 struct btrfs_root
*root
;
1349 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1350 struct extent_buffer
*leaf
;
1352 root
= btrfs_alloc_root(fs_info
);
1354 return ERR_PTR(-ENOMEM
);
1356 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1357 tree_root
->sectorsize
, tree_root
->stripesize
,
1358 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1360 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1361 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1362 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1364 * log trees do not get reference counted because they go away
1365 * before a real commit is actually done. They do store pointers
1366 * to file data extents, and those reference counts still get
1367 * updated (along with back refs to the log tree).
1371 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1372 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1376 return ERR_CAST(leaf
);
1379 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1380 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1381 btrfs_set_header_generation(leaf
, trans
->transid
);
1382 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1383 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1386 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1387 (unsigned long)btrfs_header_fsid(root
->node
),
1389 btrfs_mark_buffer_dirty(root
->node
);
1390 btrfs_tree_unlock(root
->node
);
1394 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1395 struct btrfs_fs_info
*fs_info
)
1397 struct btrfs_root
*log_root
;
1399 log_root
= alloc_log_tree(trans
, fs_info
);
1400 if (IS_ERR(log_root
))
1401 return PTR_ERR(log_root
);
1402 WARN_ON(fs_info
->log_root_tree
);
1403 fs_info
->log_root_tree
= log_root
;
1407 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1408 struct btrfs_root
*root
)
1410 struct btrfs_root
*log_root
;
1411 struct btrfs_inode_item
*inode_item
;
1413 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1414 if (IS_ERR(log_root
))
1415 return PTR_ERR(log_root
);
1417 log_root
->last_trans
= trans
->transid
;
1418 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1420 inode_item
= &log_root
->root_item
.inode
;
1421 btrfs_set_stack_inode_generation(inode_item
, 1);
1422 btrfs_set_stack_inode_size(inode_item
, 3);
1423 btrfs_set_stack_inode_nlink(inode_item
, 1);
1424 btrfs_set_stack_inode_nbytes(inode_item
, root
->leafsize
);
1425 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1427 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1429 WARN_ON(root
->log_root
);
1430 root
->log_root
= log_root
;
1431 root
->log_transid
= 0;
1432 root
->last_log_commit
= 0;
1436 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1437 struct btrfs_key
*key
)
1439 struct btrfs_root
*root
;
1440 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1441 struct btrfs_path
*path
;
1446 path
= btrfs_alloc_path();
1448 return ERR_PTR(-ENOMEM
);
1450 root
= btrfs_alloc_root(fs_info
);
1456 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1457 tree_root
->sectorsize
, tree_root
->stripesize
,
1458 root
, fs_info
, key
->objectid
);
1460 ret
= btrfs_find_root(tree_root
, key
, path
,
1461 &root
->root_item
, &root
->root_key
);
1468 generation
= btrfs_root_generation(&root
->root_item
);
1469 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1470 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1471 blocksize
, generation
);
1475 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1479 root
->commit_root
= btrfs_root_node(root
);
1481 btrfs_free_path(path
);
1485 free_extent_buffer(root
->node
);
1489 root
= ERR_PTR(ret
);
1493 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1494 struct btrfs_key
*location
)
1496 struct btrfs_root
*root
;
1498 root
= btrfs_read_tree_root(tree_root
, location
);
1502 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1504 btrfs_check_and_init_root_item(&root
->root_item
);
1510 int btrfs_init_fs_root(struct btrfs_root
*root
)
1514 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1515 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1517 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1522 btrfs_init_free_ino_ctl(root
);
1523 mutex_init(&root
->fs_commit_mutex
);
1524 spin_lock_init(&root
->cache_lock
);
1525 init_waitqueue_head(&root
->cache_wait
);
1527 ret
= get_anon_bdev(&root
->anon_dev
);
1532 kfree(root
->free_ino_ctl
);
1533 kfree(root
->free_ino_pinned
);
1537 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1540 struct btrfs_root
*root
;
1542 spin_lock(&fs_info
->fs_roots_radix_lock
);
1543 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1544 (unsigned long)root_id
);
1545 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1549 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1550 struct btrfs_root
*root
)
1554 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1558 spin_lock(&fs_info
->fs_roots_radix_lock
);
1559 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1560 (unsigned long)root
->root_key
.objectid
,
1564 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1565 radix_tree_preload_end();
1570 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1571 struct btrfs_key
*location
)
1573 struct btrfs_root
*root
;
1576 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1577 return fs_info
->tree_root
;
1578 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1579 return fs_info
->extent_root
;
1580 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1581 return fs_info
->chunk_root
;
1582 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1583 return fs_info
->dev_root
;
1584 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1585 return fs_info
->csum_root
;
1586 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1587 return fs_info
->quota_root
? fs_info
->quota_root
:
1589 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1590 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1593 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1597 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1601 if (btrfs_root_refs(&root
->root_item
) == 0) {
1606 ret
= btrfs_init_fs_root(root
);
1610 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1614 root
->orphan_item_inserted
= 1;
1616 ret
= btrfs_insert_fs_root(fs_info
, root
);
1618 if (ret
== -EEXIST
) {
1627 return ERR_PTR(ret
);
1630 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1632 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1634 struct btrfs_device
*device
;
1635 struct backing_dev_info
*bdi
;
1638 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1641 bdi
= blk_get_backing_dev_info(device
->bdev
);
1642 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1652 * If this fails, caller must call bdi_destroy() to get rid of the
1655 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1659 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1660 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1664 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1665 bdi
->congested_fn
= btrfs_congested_fn
;
1666 bdi
->congested_data
= info
;
1671 * called by the kthread helper functions to finally call the bio end_io
1672 * functions. This is where read checksum verification actually happens
1674 static void end_workqueue_fn(struct btrfs_work
*work
)
1677 struct end_io_wq
*end_io_wq
;
1678 struct btrfs_fs_info
*fs_info
;
1681 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1682 bio
= end_io_wq
->bio
;
1683 fs_info
= end_io_wq
->info
;
1685 error
= end_io_wq
->error
;
1686 bio
->bi_private
= end_io_wq
->private;
1687 bio
->bi_end_io
= end_io_wq
->end_io
;
1689 bio_endio(bio
, error
);
1692 static int cleaner_kthread(void *arg
)
1694 struct btrfs_root
*root
= arg
;
1700 /* Make the cleaner go to sleep early. */
1701 if (btrfs_need_cleaner_sleep(root
))
1704 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1708 * Avoid the problem that we change the status of the fs
1709 * during the above check and trylock.
1711 if (btrfs_need_cleaner_sleep(root
)) {
1712 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1716 btrfs_run_delayed_iputs(root
);
1717 again
= btrfs_clean_one_deleted_snapshot(root
);
1718 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1721 * The defragger has dealt with the R/O remount and umount,
1722 * needn't do anything special here.
1724 btrfs_run_defrag_inodes(root
->fs_info
);
1726 if (!try_to_freeze() && !again
) {
1727 set_current_state(TASK_INTERRUPTIBLE
);
1728 if (!kthread_should_stop())
1730 __set_current_state(TASK_RUNNING
);
1732 } while (!kthread_should_stop());
1736 static int transaction_kthread(void *arg
)
1738 struct btrfs_root
*root
= arg
;
1739 struct btrfs_trans_handle
*trans
;
1740 struct btrfs_transaction
*cur
;
1743 unsigned long delay
;
1747 cannot_commit
= false;
1748 delay
= HZ
* root
->fs_info
->commit_interval
;
1749 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1751 spin_lock(&root
->fs_info
->trans_lock
);
1752 cur
= root
->fs_info
->running_transaction
;
1754 spin_unlock(&root
->fs_info
->trans_lock
);
1758 now
= get_seconds();
1759 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1760 (now
< cur
->start_time
||
1761 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1762 spin_unlock(&root
->fs_info
->trans_lock
);
1766 transid
= cur
->transid
;
1767 spin_unlock(&root
->fs_info
->trans_lock
);
1769 /* If the file system is aborted, this will always fail. */
1770 trans
= btrfs_attach_transaction(root
);
1771 if (IS_ERR(trans
)) {
1772 if (PTR_ERR(trans
) != -ENOENT
)
1773 cannot_commit
= true;
1776 if (transid
== trans
->transid
) {
1777 btrfs_commit_transaction(trans
, root
);
1779 btrfs_end_transaction(trans
, root
);
1782 wake_up_process(root
->fs_info
->cleaner_kthread
);
1783 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1785 if (!try_to_freeze()) {
1786 set_current_state(TASK_INTERRUPTIBLE
);
1787 if (!kthread_should_stop() &&
1788 (!btrfs_transaction_blocked(root
->fs_info
) ||
1790 schedule_timeout(delay
);
1791 __set_current_state(TASK_RUNNING
);
1793 } while (!kthread_should_stop());
1798 * this will find the highest generation in the array of
1799 * root backups. The index of the highest array is returned,
1800 * or -1 if we can't find anything.
1802 * We check to make sure the array is valid by comparing the
1803 * generation of the latest root in the array with the generation
1804 * in the super block. If they don't match we pitch it.
1806 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1809 int newest_index
= -1;
1810 struct btrfs_root_backup
*root_backup
;
1813 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1814 root_backup
= info
->super_copy
->super_roots
+ i
;
1815 cur
= btrfs_backup_tree_root_gen(root_backup
);
1816 if (cur
== newest_gen
)
1820 /* check to see if we actually wrapped around */
1821 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1822 root_backup
= info
->super_copy
->super_roots
;
1823 cur
= btrfs_backup_tree_root_gen(root_backup
);
1824 if (cur
== newest_gen
)
1827 return newest_index
;
1832 * find the oldest backup so we know where to store new entries
1833 * in the backup array. This will set the backup_root_index
1834 * field in the fs_info struct
1836 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1839 int newest_index
= -1;
1841 newest_index
= find_newest_super_backup(info
, newest_gen
);
1842 /* if there was garbage in there, just move along */
1843 if (newest_index
== -1) {
1844 info
->backup_root_index
= 0;
1846 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1851 * copy all the root pointers into the super backup array.
1852 * this will bump the backup pointer by one when it is
1855 static void backup_super_roots(struct btrfs_fs_info
*info
)
1858 struct btrfs_root_backup
*root_backup
;
1861 next_backup
= info
->backup_root_index
;
1862 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1863 BTRFS_NUM_BACKUP_ROOTS
;
1866 * just overwrite the last backup if we're at the same generation
1867 * this happens only at umount
1869 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1870 if (btrfs_backup_tree_root_gen(root_backup
) ==
1871 btrfs_header_generation(info
->tree_root
->node
))
1872 next_backup
= last_backup
;
1874 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1877 * make sure all of our padding and empty slots get zero filled
1878 * regardless of which ones we use today
1880 memset(root_backup
, 0, sizeof(*root_backup
));
1882 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1884 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1885 btrfs_set_backup_tree_root_gen(root_backup
,
1886 btrfs_header_generation(info
->tree_root
->node
));
1888 btrfs_set_backup_tree_root_level(root_backup
,
1889 btrfs_header_level(info
->tree_root
->node
));
1891 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1892 btrfs_set_backup_chunk_root_gen(root_backup
,
1893 btrfs_header_generation(info
->chunk_root
->node
));
1894 btrfs_set_backup_chunk_root_level(root_backup
,
1895 btrfs_header_level(info
->chunk_root
->node
));
1897 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1898 btrfs_set_backup_extent_root_gen(root_backup
,
1899 btrfs_header_generation(info
->extent_root
->node
));
1900 btrfs_set_backup_extent_root_level(root_backup
,
1901 btrfs_header_level(info
->extent_root
->node
));
1904 * we might commit during log recovery, which happens before we set
1905 * the fs_root. Make sure it is valid before we fill it in.
1907 if (info
->fs_root
&& info
->fs_root
->node
) {
1908 btrfs_set_backup_fs_root(root_backup
,
1909 info
->fs_root
->node
->start
);
1910 btrfs_set_backup_fs_root_gen(root_backup
,
1911 btrfs_header_generation(info
->fs_root
->node
));
1912 btrfs_set_backup_fs_root_level(root_backup
,
1913 btrfs_header_level(info
->fs_root
->node
));
1916 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1917 btrfs_set_backup_dev_root_gen(root_backup
,
1918 btrfs_header_generation(info
->dev_root
->node
));
1919 btrfs_set_backup_dev_root_level(root_backup
,
1920 btrfs_header_level(info
->dev_root
->node
));
1922 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1923 btrfs_set_backup_csum_root_gen(root_backup
,
1924 btrfs_header_generation(info
->csum_root
->node
));
1925 btrfs_set_backup_csum_root_level(root_backup
,
1926 btrfs_header_level(info
->csum_root
->node
));
1928 btrfs_set_backup_total_bytes(root_backup
,
1929 btrfs_super_total_bytes(info
->super_copy
));
1930 btrfs_set_backup_bytes_used(root_backup
,
1931 btrfs_super_bytes_used(info
->super_copy
));
1932 btrfs_set_backup_num_devices(root_backup
,
1933 btrfs_super_num_devices(info
->super_copy
));
1936 * if we don't copy this out to the super_copy, it won't get remembered
1937 * for the next commit
1939 memcpy(&info
->super_copy
->super_roots
,
1940 &info
->super_for_commit
->super_roots
,
1941 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1945 * this copies info out of the root backup array and back into
1946 * the in-memory super block. It is meant to help iterate through
1947 * the array, so you send it the number of backups you've already
1948 * tried and the last backup index you used.
1950 * this returns -1 when it has tried all the backups
1952 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1953 struct btrfs_super_block
*super
,
1954 int *num_backups_tried
, int *backup_index
)
1956 struct btrfs_root_backup
*root_backup
;
1957 int newest
= *backup_index
;
1959 if (*num_backups_tried
== 0) {
1960 u64 gen
= btrfs_super_generation(super
);
1962 newest
= find_newest_super_backup(info
, gen
);
1966 *backup_index
= newest
;
1967 *num_backups_tried
= 1;
1968 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1969 /* we've tried all the backups, all done */
1972 /* jump to the next oldest backup */
1973 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1974 BTRFS_NUM_BACKUP_ROOTS
;
1975 *backup_index
= newest
;
1976 *num_backups_tried
+= 1;
1978 root_backup
= super
->super_roots
+ newest
;
1980 btrfs_set_super_generation(super
,
1981 btrfs_backup_tree_root_gen(root_backup
));
1982 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1983 btrfs_set_super_root_level(super
,
1984 btrfs_backup_tree_root_level(root_backup
));
1985 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1988 * fixme: the total bytes and num_devices need to match or we should
1991 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1992 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1996 /* helper to cleanup workers */
1997 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
1999 btrfs_stop_workers(&fs_info
->generic_worker
);
2000 btrfs_stop_workers(&fs_info
->fixup_workers
);
2001 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2002 btrfs_stop_workers(&fs_info
->workers
);
2003 btrfs_stop_workers(&fs_info
->endio_workers
);
2004 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2005 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
2006 btrfs_stop_workers(&fs_info
->rmw_workers
);
2007 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2008 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2009 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2010 btrfs_stop_workers(&fs_info
->submit_workers
);
2011 btrfs_stop_workers(&fs_info
->delayed_workers
);
2012 btrfs_stop_workers(&fs_info
->caching_workers
);
2013 btrfs_stop_workers(&fs_info
->readahead_workers
);
2014 btrfs_stop_workers(&fs_info
->flush_workers
);
2015 btrfs_stop_workers(&fs_info
->qgroup_rescan_workers
);
2018 /* helper to cleanup tree roots */
2019 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2021 free_extent_buffer(info
->tree_root
->node
);
2022 free_extent_buffer(info
->tree_root
->commit_root
);
2023 info
->tree_root
->node
= NULL
;
2024 info
->tree_root
->commit_root
= NULL
;
2026 if (info
->dev_root
) {
2027 free_extent_buffer(info
->dev_root
->node
);
2028 free_extent_buffer(info
->dev_root
->commit_root
);
2029 info
->dev_root
->node
= NULL
;
2030 info
->dev_root
->commit_root
= NULL
;
2032 if (info
->extent_root
) {
2033 free_extent_buffer(info
->extent_root
->node
);
2034 free_extent_buffer(info
->extent_root
->commit_root
);
2035 info
->extent_root
->node
= NULL
;
2036 info
->extent_root
->commit_root
= NULL
;
2038 if (info
->csum_root
) {
2039 free_extent_buffer(info
->csum_root
->node
);
2040 free_extent_buffer(info
->csum_root
->commit_root
);
2041 info
->csum_root
->node
= NULL
;
2042 info
->csum_root
->commit_root
= NULL
;
2044 if (info
->quota_root
) {
2045 free_extent_buffer(info
->quota_root
->node
);
2046 free_extent_buffer(info
->quota_root
->commit_root
);
2047 info
->quota_root
->node
= NULL
;
2048 info
->quota_root
->commit_root
= NULL
;
2050 if (info
->uuid_root
) {
2051 free_extent_buffer(info
->uuid_root
->node
);
2052 free_extent_buffer(info
->uuid_root
->commit_root
);
2053 info
->uuid_root
->node
= NULL
;
2054 info
->uuid_root
->commit_root
= NULL
;
2057 free_extent_buffer(info
->chunk_root
->node
);
2058 free_extent_buffer(info
->chunk_root
->commit_root
);
2059 info
->chunk_root
->node
= NULL
;
2060 info
->chunk_root
->commit_root
= NULL
;
2064 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
2067 struct btrfs_root
*gang
[8];
2070 while (!list_empty(&fs_info
->dead_roots
)) {
2071 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2072 struct btrfs_root
, root_list
);
2073 list_del(&gang
[0]->root_list
);
2075 if (gang
[0]->in_radix
) {
2076 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2078 free_extent_buffer(gang
[0]->node
);
2079 free_extent_buffer(gang
[0]->commit_root
);
2080 btrfs_put_fs_root(gang
[0]);
2085 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2090 for (i
= 0; i
< ret
; i
++)
2091 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2095 int open_ctree(struct super_block
*sb
,
2096 struct btrfs_fs_devices
*fs_devices
,
2106 struct btrfs_key location
;
2107 struct buffer_head
*bh
;
2108 struct btrfs_super_block
*disk_super
;
2109 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2110 struct btrfs_root
*tree_root
;
2111 struct btrfs_root
*extent_root
;
2112 struct btrfs_root
*csum_root
;
2113 struct btrfs_root
*chunk_root
;
2114 struct btrfs_root
*dev_root
;
2115 struct btrfs_root
*quota_root
;
2116 struct btrfs_root
*uuid_root
;
2117 struct btrfs_root
*log_tree_root
;
2120 int num_backups_tried
= 0;
2121 int backup_index
= 0;
2122 bool create_uuid_tree
= false;
2124 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2125 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2126 if (!tree_root
|| !chunk_root
) {
2131 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2137 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2143 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2148 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2149 (1 + ilog2(nr_cpu_ids
));
2151 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2154 goto fail_dirty_metadata_bytes
;
2157 fs_info
->btree_inode
= new_inode(sb
);
2158 if (!fs_info
->btree_inode
) {
2160 goto fail_delalloc_bytes
;
2163 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2165 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2166 INIT_LIST_HEAD(&fs_info
->trans_list
);
2167 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2168 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2169 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2170 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2171 spin_lock_init(&fs_info
->delalloc_root_lock
);
2172 spin_lock_init(&fs_info
->trans_lock
);
2173 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2174 spin_lock_init(&fs_info
->delayed_iput_lock
);
2175 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2176 spin_lock_init(&fs_info
->free_chunk_lock
);
2177 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2178 spin_lock_init(&fs_info
->super_lock
);
2179 rwlock_init(&fs_info
->tree_mod_log_lock
);
2180 mutex_init(&fs_info
->reloc_mutex
);
2181 seqlock_init(&fs_info
->profiles_lock
);
2183 init_completion(&fs_info
->kobj_unregister
);
2184 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2185 INIT_LIST_HEAD(&fs_info
->space_info
);
2186 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2187 btrfs_mapping_init(&fs_info
->mapping_tree
);
2188 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2189 BTRFS_BLOCK_RSV_GLOBAL
);
2190 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2191 BTRFS_BLOCK_RSV_DELALLOC
);
2192 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2193 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2194 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2195 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2196 BTRFS_BLOCK_RSV_DELOPS
);
2197 atomic_set(&fs_info
->nr_async_submits
, 0);
2198 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2199 atomic_set(&fs_info
->async_submit_draining
, 0);
2200 atomic_set(&fs_info
->nr_async_bios
, 0);
2201 atomic_set(&fs_info
->defrag_running
, 0);
2202 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2204 fs_info
->max_inline
= 8192 * 1024;
2205 fs_info
->metadata_ratio
= 0;
2206 fs_info
->defrag_inodes
= RB_ROOT
;
2207 fs_info
->free_chunk_space
= 0;
2208 fs_info
->tree_mod_log
= RB_ROOT
;
2209 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2211 /* readahead state */
2212 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2213 spin_lock_init(&fs_info
->reada_lock
);
2215 fs_info
->thread_pool_size
= min_t(unsigned long,
2216 num_online_cpus() + 2, 8);
2218 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2219 spin_lock_init(&fs_info
->ordered_root_lock
);
2220 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2222 if (!fs_info
->delayed_root
) {
2226 btrfs_init_delayed_root(fs_info
->delayed_root
);
2228 mutex_init(&fs_info
->scrub_lock
);
2229 atomic_set(&fs_info
->scrubs_running
, 0);
2230 atomic_set(&fs_info
->scrub_pause_req
, 0);
2231 atomic_set(&fs_info
->scrubs_paused
, 0);
2232 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2233 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2234 init_rwsem(&fs_info
->scrub_super_lock
);
2235 fs_info
->scrub_workers_refcnt
= 0;
2236 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2237 fs_info
->check_integrity_print_mask
= 0;
2240 spin_lock_init(&fs_info
->balance_lock
);
2241 mutex_init(&fs_info
->balance_mutex
);
2242 atomic_set(&fs_info
->balance_running
, 0);
2243 atomic_set(&fs_info
->balance_pause_req
, 0);
2244 atomic_set(&fs_info
->balance_cancel_req
, 0);
2245 fs_info
->balance_ctl
= NULL
;
2246 init_waitqueue_head(&fs_info
->balance_wait_q
);
2248 sb
->s_blocksize
= 4096;
2249 sb
->s_blocksize_bits
= blksize_bits(4096);
2250 sb
->s_bdi
= &fs_info
->bdi
;
2252 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2253 set_nlink(fs_info
->btree_inode
, 1);
2255 * we set the i_size on the btree inode to the max possible int.
2256 * the real end of the address space is determined by all of
2257 * the devices in the system
2259 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2260 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2261 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2263 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2264 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2265 fs_info
->btree_inode
->i_mapping
);
2266 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2267 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2269 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2271 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2272 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2273 sizeof(struct btrfs_key
));
2274 set_bit(BTRFS_INODE_DUMMY
,
2275 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2276 insert_inode_hash(fs_info
->btree_inode
);
2278 spin_lock_init(&fs_info
->block_group_cache_lock
);
2279 fs_info
->block_group_cache_tree
= RB_ROOT
;
2280 fs_info
->first_logical_byte
= (u64
)-1;
2282 extent_io_tree_init(&fs_info
->freed_extents
[0],
2283 fs_info
->btree_inode
->i_mapping
);
2284 extent_io_tree_init(&fs_info
->freed_extents
[1],
2285 fs_info
->btree_inode
->i_mapping
);
2286 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2287 fs_info
->do_barriers
= 1;
2290 mutex_init(&fs_info
->ordered_operations_mutex
);
2291 mutex_init(&fs_info
->ordered_extent_flush_mutex
);
2292 mutex_init(&fs_info
->tree_log_mutex
);
2293 mutex_init(&fs_info
->chunk_mutex
);
2294 mutex_init(&fs_info
->transaction_kthread_mutex
);
2295 mutex_init(&fs_info
->cleaner_mutex
);
2296 mutex_init(&fs_info
->volume_mutex
);
2297 init_rwsem(&fs_info
->extent_commit_sem
);
2298 init_rwsem(&fs_info
->cleanup_work_sem
);
2299 init_rwsem(&fs_info
->subvol_sem
);
2300 fs_info
->dev_replace
.lock_owner
= 0;
2301 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2302 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2303 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2304 mutex_init(&fs_info
->dev_replace
.lock
);
2306 spin_lock_init(&fs_info
->qgroup_lock
);
2307 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2308 fs_info
->qgroup_tree
= RB_ROOT
;
2309 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2310 fs_info
->qgroup_seq
= 1;
2311 fs_info
->quota_enabled
= 0;
2312 fs_info
->pending_quota_state
= 0;
2313 fs_info
->qgroup_ulist
= NULL
;
2314 mutex_init(&fs_info
->qgroup_rescan_lock
);
2316 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2317 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2319 init_waitqueue_head(&fs_info
->transaction_throttle
);
2320 init_waitqueue_head(&fs_info
->transaction_wait
);
2321 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2322 init_waitqueue_head(&fs_info
->async_submit_wait
);
2324 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2330 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2331 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2333 invalidate_bdev(fs_devices
->latest_bdev
);
2336 * Read super block and check the signature bytes only
2338 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2345 * We want to check superblock checksum, the type is stored inside.
2346 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2348 if (btrfs_check_super_csum(bh
->b_data
)) {
2349 printk(KERN_ERR
"btrfs: superblock checksum mismatch\n");
2355 * super_copy is zeroed at allocation time and we never touch the
2356 * following bytes up to INFO_SIZE, the checksum is calculated from
2357 * the whole block of INFO_SIZE
2359 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2360 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2361 sizeof(*fs_info
->super_for_commit
));
2364 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2366 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2368 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2373 disk_super
= fs_info
->super_copy
;
2374 if (!btrfs_super_root(disk_super
))
2377 /* check FS state, whether FS is broken. */
2378 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2379 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2382 * run through our array of backup supers and setup
2383 * our ring pointer to the oldest one
2385 generation
= btrfs_super_generation(disk_super
);
2386 find_oldest_super_backup(fs_info
, generation
);
2389 * In the long term, we'll store the compression type in the super
2390 * block, and it'll be used for per file compression control.
2392 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2394 ret
= btrfs_parse_options(tree_root
, options
);
2400 features
= btrfs_super_incompat_flags(disk_super
) &
2401 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2403 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2404 "unsupported optional features (%Lx).\n",
2405 (unsigned long long)features
);
2410 if (btrfs_super_leafsize(disk_super
) !=
2411 btrfs_super_nodesize(disk_super
)) {
2412 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2413 "blocksizes don't match. node %d leaf %d\n",
2414 btrfs_super_nodesize(disk_super
),
2415 btrfs_super_leafsize(disk_super
));
2419 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2420 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2421 "blocksize (%d) was too large\n",
2422 btrfs_super_leafsize(disk_super
));
2427 features
= btrfs_super_incompat_flags(disk_super
);
2428 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2429 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2430 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2432 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2433 printk(KERN_ERR
"btrfs: has skinny extents\n");
2436 * flag our filesystem as having big metadata blocks if
2437 * they are bigger than the page size
2439 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2440 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2441 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2442 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2445 nodesize
= btrfs_super_nodesize(disk_super
);
2446 leafsize
= btrfs_super_leafsize(disk_super
);
2447 sectorsize
= btrfs_super_sectorsize(disk_super
);
2448 stripesize
= btrfs_super_stripesize(disk_super
);
2449 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2450 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2453 * mixed block groups end up with duplicate but slightly offset
2454 * extent buffers for the same range. It leads to corruptions
2456 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2457 (sectorsize
!= leafsize
)) {
2458 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2459 "are not allowed for mixed block groups on %s\n",
2465 * Needn't use the lock because there is no other task which will
2468 btrfs_set_super_incompat_flags(disk_super
, features
);
2470 features
= btrfs_super_compat_ro_flags(disk_super
) &
2471 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2472 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2473 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2474 "unsupported option features (%Lx).\n",
2475 (unsigned long long)features
);
2480 btrfs_init_workers(&fs_info
->generic_worker
,
2481 "genwork", 1, NULL
);
2483 btrfs_init_workers(&fs_info
->workers
, "worker",
2484 fs_info
->thread_pool_size
,
2485 &fs_info
->generic_worker
);
2487 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2488 fs_info
->thread_pool_size
,
2489 &fs_info
->generic_worker
);
2491 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2492 fs_info
->thread_pool_size
,
2493 &fs_info
->generic_worker
);
2495 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2496 min_t(u64
, fs_devices
->num_devices
,
2497 fs_info
->thread_pool_size
),
2498 &fs_info
->generic_worker
);
2500 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2501 2, &fs_info
->generic_worker
);
2503 /* a higher idle thresh on the submit workers makes it much more
2504 * likely that bios will be send down in a sane order to the
2507 fs_info
->submit_workers
.idle_thresh
= 64;
2509 fs_info
->workers
.idle_thresh
= 16;
2510 fs_info
->workers
.ordered
= 1;
2512 fs_info
->delalloc_workers
.idle_thresh
= 2;
2513 fs_info
->delalloc_workers
.ordered
= 1;
2515 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2516 &fs_info
->generic_worker
);
2517 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2518 fs_info
->thread_pool_size
,
2519 &fs_info
->generic_worker
);
2520 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2521 fs_info
->thread_pool_size
,
2522 &fs_info
->generic_worker
);
2523 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2524 "endio-meta-write", fs_info
->thread_pool_size
,
2525 &fs_info
->generic_worker
);
2526 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2527 "endio-raid56", fs_info
->thread_pool_size
,
2528 &fs_info
->generic_worker
);
2529 btrfs_init_workers(&fs_info
->rmw_workers
,
2530 "rmw", fs_info
->thread_pool_size
,
2531 &fs_info
->generic_worker
);
2532 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2533 fs_info
->thread_pool_size
,
2534 &fs_info
->generic_worker
);
2535 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2536 1, &fs_info
->generic_worker
);
2537 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2538 fs_info
->thread_pool_size
,
2539 &fs_info
->generic_worker
);
2540 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2541 fs_info
->thread_pool_size
,
2542 &fs_info
->generic_worker
);
2543 btrfs_init_workers(&fs_info
->qgroup_rescan_workers
, "qgroup-rescan", 1,
2544 &fs_info
->generic_worker
);
2547 * endios are largely parallel and should have a very
2550 fs_info
->endio_workers
.idle_thresh
= 4;
2551 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2552 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2553 fs_info
->rmw_workers
.idle_thresh
= 2;
2555 fs_info
->endio_write_workers
.idle_thresh
= 2;
2556 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2557 fs_info
->readahead_workers
.idle_thresh
= 2;
2560 * btrfs_start_workers can really only fail because of ENOMEM so just
2561 * return -ENOMEM if any of these fail.
2563 ret
= btrfs_start_workers(&fs_info
->workers
);
2564 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2565 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2566 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2567 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2568 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2569 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2570 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2571 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2572 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2573 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2574 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2575 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2576 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2577 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2578 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2579 ret
|= btrfs_start_workers(&fs_info
->qgroup_rescan_workers
);
2582 goto fail_sb_buffer
;
2585 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2586 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2587 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2589 tree_root
->nodesize
= nodesize
;
2590 tree_root
->leafsize
= leafsize
;
2591 tree_root
->sectorsize
= sectorsize
;
2592 tree_root
->stripesize
= stripesize
;
2594 sb
->s_blocksize
= sectorsize
;
2595 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2597 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2598 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2599 goto fail_sb_buffer
;
2602 if (sectorsize
!= PAGE_SIZE
) {
2603 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2604 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2605 goto fail_sb_buffer
;
2608 mutex_lock(&fs_info
->chunk_mutex
);
2609 ret
= btrfs_read_sys_array(tree_root
);
2610 mutex_unlock(&fs_info
->chunk_mutex
);
2612 printk(KERN_WARNING
"btrfs: failed to read the system "
2613 "array on %s\n", sb
->s_id
);
2614 goto fail_sb_buffer
;
2617 blocksize
= btrfs_level_size(tree_root
,
2618 btrfs_super_chunk_root_level(disk_super
));
2619 generation
= btrfs_super_chunk_root_generation(disk_super
);
2621 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2622 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2624 chunk_root
->node
= read_tree_block(chunk_root
,
2625 btrfs_super_chunk_root(disk_super
),
2626 blocksize
, generation
);
2627 if (!chunk_root
->node
||
2628 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2629 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2631 goto fail_tree_roots
;
2633 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2634 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2636 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2637 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2640 ret
= btrfs_read_chunk_tree(chunk_root
);
2642 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2644 goto fail_tree_roots
;
2648 * keep the device that is marked to be the target device for the
2649 * dev_replace procedure
2651 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2653 if (!fs_devices
->latest_bdev
) {
2654 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2656 goto fail_tree_roots
;
2660 blocksize
= btrfs_level_size(tree_root
,
2661 btrfs_super_root_level(disk_super
));
2662 generation
= btrfs_super_generation(disk_super
);
2664 tree_root
->node
= read_tree_block(tree_root
,
2665 btrfs_super_root(disk_super
),
2666 blocksize
, generation
);
2667 if (!tree_root
->node
||
2668 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2669 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2672 goto recovery_tree_root
;
2675 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2676 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2678 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2679 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2680 location
.offset
= 0;
2682 extent_root
= btrfs_read_tree_root(tree_root
, &location
);
2683 if (IS_ERR(extent_root
)) {
2684 ret
= PTR_ERR(extent_root
);
2685 goto recovery_tree_root
;
2687 extent_root
->track_dirty
= 1;
2688 fs_info
->extent_root
= extent_root
;
2690 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2691 dev_root
= btrfs_read_tree_root(tree_root
, &location
);
2692 if (IS_ERR(dev_root
)) {
2693 ret
= PTR_ERR(dev_root
);
2694 goto recovery_tree_root
;
2696 dev_root
->track_dirty
= 1;
2697 fs_info
->dev_root
= dev_root
;
2698 btrfs_init_devices_late(fs_info
);
2700 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2701 csum_root
= btrfs_read_tree_root(tree_root
, &location
);
2702 if (IS_ERR(csum_root
)) {
2703 ret
= PTR_ERR(csum_root
);
2704 goto recovery_tree_root
;
2706 csum_root
->track_dirty
= 1;
2707 fs_info
->csum_root
= csum_root
;
2709 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2710 quota_root
= btrfs_read_tree_root(tree_root
, &location
);
2711 if (!IS_ERR(quota_root
)) {
2712 quota_root
->track_dirty
= 1;
2713 fs_info
->quota_enabled
= 1;
2714 fs_info
->pending_quota_state
= 1;
2715 fs_info
->quota_root
= quota_root
;
2718 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2719 uuid_root
= btrfs_read_tree_root(tree_root
, &location
);
2720 if (IS_ERR(uuid_root
)) {
2721 ret
= PTR_ERR(uuid_root
);
2723 goto recovery_tree_root
;
2724 create_uuid_tree
= true;
2726 uuid_root
->track_dirty
= 1;
2727 fs_info
->uuid_root
= uuid_root
;
2730 fs_info
->generation
= generation
;
2731 fs_info
->last_trans_committed
= generation
;
2733 ret
= btrfs_recover_balance(fs_info
);
2735 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2736 goto fail_block_groups
;
2739 ret
= btrfs_init_dev_stats(fs_info
);
2741 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2743 goto fail_block_groups
;
2746 ret
= btrfs_init_dev_replace(fs_info
);
2748 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2749 goto fail_block_groups
;
2752 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2754 ret
= btrfs_init_space_info(fs_info
);
2756 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2757 goto fail_block_groups
;
2760 ret
= btrfs_read_block_groups(extent_root
);
2762 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2763 goto fail_block_groups
;
2765 fs_info
->num_tolerated_disk_barrier_failures
=
2766 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2767 if (fs_info
->fs_devices
->missing_devices
>
2768 fs_info
->num_tolerated_disk_barrier_failures
&&
2769 !(sb
->s_flags
& MS_RDONLY
)) {
2771 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2772 goto fail_block_groups
;
2775 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2777 if (IS_ERR(fs_info
->cleaner_kthread
))
2778 goto fail_block_groups
;
2780 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2782 "btrfs-transaction");
2783 if (IS_ERR(fs_info
->transaction_kthread
))
2786 if (!btrfs_test_opt(tree_root
, SSD
) &&
2787 !btrfs_test_opt(tree_root
, NOSSD
) &&
2788 !fs_info
->fs_devices
->rotating
) {
2789 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2791 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2794 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2795 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2796 ret
= btrfsic_mount(tree_root
, fs_devices
,
2797 btrfs_test_opt(tree_root
,
2798 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2800 fs_info
->check_integrity_print_mask
);
2802 printk(KERN_WARNING
"btrfs: failed to initialize"
2803 " integrity check module %s\n", sb
->s_id
);
2806 ret
= btrfs_read_qgroup_config(fs_info
);
2808 goto fail_trans_kthread
;
2810 /* do not make disk changes in broken FS */
2811 if (btrfs_super_log_root(disk_super
) != 0) {
2812 u64 bytenr
= btrfs_super_log_root(disk_super
);
2814 if (fs_devices
->rw_devices
== 0) {
2815 printk(KERN_WARNING
"Btrfs log replay required "
2821 btrfs_level_size(tree_root
,
2822 btrfs_super_log_root_level(disk_super
));
2824 log_tree_root
= btrfs_alloc_root(fs_info
);
2825 if (!log_tree_root
) {
2830 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2831 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2833 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2836 if (!log_tree_root
->node
||
2837 !extent_buffer_uptodate(log_tree_root
->node
)) {
2838 printk(KERN_ERR
"btrfs: failed to read log tree\n");
2839 free_extent_buffer(log_tree_root
->node
);
2840 kfree(log_tree_root
);
2841 goto fail_trans_kthread
;
2843 /* returns with log_tree_root freed on success */
2844 ret
= btrfs_recover_log_trees(log_tree_root
);
2846 btrfs_error(tree_root
->fs_info
, ret
,
2847 "Failed to recover log tree");
2848 free_extent_buffer(log_tree_root
->node
);
2849 kfree(log_tree_root
);
2850 goto fail_trans_kthread
;
2853 if (sb
->s_flags
& MS_RDONLY
) {
2854 ret
= btrfs_commit_super(tree_root
);
2856 goto fail_trans_kthread
;
2860 ret
= btrfs_find_orphan_roots(tree_root
);
2862 goto fail_trans_kthread
;
2864 if (!(sb
->s_flags
& MS_RDONLY
)) {
2865 ret
= btrfs_cleanup_fs_roots(fs_info
);
2867 goto fail_trans_kthread
;
2869 ret
= btrfs_recover_relocation(tree_root
);
2872 "btrfs: failed to recover relocation\n");
2878 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2879 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2880 location
.offset
= 0;
2882 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2883 if (IS_ERR(fs_info
->fs_root
)) {
2884 err
= PTR_ERR(fs_info
->fs_root
);
2888 if (sb
->s_flags
& MS_RDONLY
)
2891 down_read(&fs_info
->cleanup_work_sem
);
2892 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2893 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2894 up_read(&fs_info
->cleanup_work_sem
);
2895 close_ctree(tree_root
);
2898 up_read(&fs_info
->cleanup_work_sem
);
2900 ret
= btrfs_resume_balance_async(fs_info
);
2902 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2903 close_ctree(tree_root
);
2907 ret
= btrfs_resume_dev_replace_async(fs_info
);
2909 pr_warn("btrfs: failed to resume dev_replace\n");
2910 close_ctree(tree_root
);
2914 btrfs_qgroup_rescan_resume(fs_info
);
2916 if (create_uuid_tree
) {
2917 pr_info("btrfs: creating UUID tree\n");
2918 ret
= btrfs_create_uuid_tree(fs_info
);
2920 pr_warn("btrfs: failed to create the UUID tree %d\n",
2922 close_ctree(tree_root
);
2930 btrfs_free_qgroup_config(fs_info
);
2932 kthread_stop(fs_info
->transaction_kthread
);
2933 btrfs_cleanup_transaction(fs_info
->tree_root
);
2934 del_fs_roots(fs_info
);
2936 kthread_stop(fs_info
->cleaner_kthread
);
2939 * make sure we're done with the btree inode before we stop our
2942 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2945 btrfs_put_block_group_cache(fs_info
);
2946 btrfs_free_block_groups(fs_info
);
2949 free_root_pointers(fs_info
, 1);
2950 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2953 btrfs_stop_all_workers(fs_info
);
2956 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2958 iput(fs_info
->btree_inode
);
2959 fail_delalloc_bytes
:
2960 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2961 fail_dirty_metadata_bytes
:
2962 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2964 bdi_destroy(&fs_info
->bdi
);
2966 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2968 btrfs_free_stripe_hash_table(fs_info
);
2969 btrfs_close_devices(fs_info
->fs_devices
);
2973 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2974 goto fail_tree_roots
;
2976 free_root_pointers(fs_info
, 0);
2978 /* don't use the log in recovery mode, it won't be valid */
2979 btrfs_set_super_log_root(disk_super
, 0);
2981 /* we can't trust the free space cache either */
2982 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2984 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2985 &num_backups_tried
, &backup_index
);
2987 goto fail_block_groups
;
2988 goto retry_root_backup
;
2991 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2994 set_buffer_uptodate(bh
);
2996 struct btrfs_device
*device
= (struct btrfs_device
*)
2999 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
3000 "I/O error on %s\n",
3001 rcu_str_deref(device
->name
));
3002 /* note, we dont' set_buffer_write_io_error because we have
3003 * our own ways of dealing with the IO errors
3005 clear_buffer_uptodate(bh
);
3006 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3012 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3014 struct buffer_head
*bh
;
3015 struct buffer_head
*latest
= NULL
;
3016 struct btrfs_super_block
*super
;
3021 /* we would like to check all the supers, but that would make
3022 * a btrfs mount succeed after a mkfs from a different FS.
3023 * So, we need to add a special mount option to scan for
3024 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3026 for (i
= 0; i
< 1; i
++) {
3027 bytenr
= btrfs_sb_offset(i
);
3028 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3029 i_size_read(bdev
->bd_inode
))
3031 bh
= __bread(bdev
, bytenr
/ 4096,
3032 BTRFS_SUPER_INFO_SIZE
);
3036 super
= (struct btrfs_super_block
*)bh
->b_data
;
3037 if (btrfs_super_bytenr(super
) != bytenr
||
3038 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3043 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3046 transid
= btrfs_super_generation(super
);
3055 * this should be called twice, once with wait == 0 and
3056 * once with wait == 1. When wait == 0 is done, all the buffer heads
3057 * we write are pinned.
3059 * They are released when wait == 1 is done.
3060 * max_mirrors must be the same for both runs, and it indicates how
3061 * many supers on this one device should be written.
3063 * max_mirrors == 0 means to write them all.
3065 static int write_dev_supers(struct btrfs_device
*device
,
3066 struct btrfs_super_block
*sb
,
3067 int do_barriers
, int wait
, int max_mirrors
)
3069 struct buffer_head
*bh
;
3076 if (max_mirrors
== 0)
3077 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3079 for (i
= 0; i
< max_mirrors
; i
++) {
3080 bytenr
= btrfs_sb_offset(i
);
3081 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
3085 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3086 BTRFS_SUPER_INFO_SIZE
);
3092 if (!buffer_uptodate(bh
))
3095 /* drop our reference */
3098 /* drop the reference from the wait == 0 run */
3102 btrfs_set_super_bytenr(sb
, bytenr
);
3105 crc
= btrfs_csum_data((char *)sb
+
3106 BTRFS_CSUM_SIZE
, crc
,
3107 BTRFS_SUPER_INFO_SIZE
-
3109 btrfs_csum_final(crc
, sb
->csum
);
3112 * one reference for us, and we leave it for the
3115 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3116 BTRFS_SUPER_INFO_SIZE
);
3118 printk(KERN_ERR
"btrfs: couldn't get super "
3119 "buffer head for bytenr %Lu\n", bytenr
);
3124 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3126 /* one reference for submit_bh */
3129 set_buffer_uptodate(bh
);
3131 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3132 bh
->b_private
= device
;
3136 * we fua the first super. The others we allow
3139 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3143 return errors
< i
? 0 : -1;
3147 * endio for the write_dev_flush, this will wake anyone waiting
3148 * for the barrier when it is done
3150 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3153 if (err
== -EOPNOTSUPP
)
3154 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3155 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3157 if (bio
->bi_private
)
3158 complete(bio
->bi_private
);
3163 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3164 * sent down. With wait == 1, it waits for the previous flush.
3166 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3169 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3174 if (device
->nobarriers
)
3178 bio
= device
->flush_bio
;
3182 wait_for_completion(&device
->flush_wait
);
3184 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3185 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3186 rcu_str_deref(device
->name
));
3187 device
->nobarriers
= 1;
3188 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3190 btrfs_dev_stat_inc_and_print(device
,
3191 BTRFS_DEV_STAT_FLUSH_ERRS
);
3194 /* drop the reference from the wait == 0 run */
3196 device
->flush_bio
= NULL
;
3202 * one reference for us, and we leave it for the
3205 device
->flush_bio
= NULL
;
3206 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3210 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3211 bio
->bi_bdev
= device
->bdev
;
3212 init_completion(&device
->flush_wait
);
3213 bio
->bi_private
= &device
->flush_wait
;
3214 device
->flush_bio
= bio
;
3217 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3223 * send an empty flush down to each device in parallel,
3224 * then wait for them
3226 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3228 struct list_head
*head
;
3229 struct btrfs_device
*dev
;
3230 int errors_send
= 0;
3231 int errors_wait
= 0;
3234 /* send down all the barriers */
3235 head
= &info
->fs_devices
->devices
;
3236 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3241 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3244 ret
= write_dev_flush(dev
, 0);
3249 /* wait for all the barriers */
3250 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3255 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3258 ret
= write_dev_flush(dev
, 1);
3262 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3263 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3268 int btrfs_calc_num_tolerated_disk_barrier_failures(
3269 struct btrfs_fs_info
*fs_info
)
3271 struct btrfs_ioctl_space_info space
;
3272 struct btrfs_space_info
*sinfo
;
3273 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3274 BTRFS_BLOCK_GROUP_SYSTEM
,
3275 BTRFS_BLOCK_GROUP_METADATA
,
3276 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3280 int num_tolerated_disk_barrier_failures
=
3281 (int)fs_info
->fs_devices
->num_devices
;
3283 for (i
= 0; i
< num_types
; i
++) {
3284 struct btrfs_space_info
*tmp
;
3288 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3289 if (tmp
->flags
== types
[i
]) {
3299 down_read(&sinfo
->groups_sem
);
3300 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3301 if (!list_empty(&sinfo
->block_groups
[c
])) {
3304 btrfs_get_block_group_info(
3305 &sinfo
->block_groups
[c
], &space
);
3306 if (space
.total_bytes
== 0 ||
3307 space
.used_bytes
== 0)
3309 flags
= space
.flags
;
3312 * 0: if dup, single or RAID0 is configured for
3313 * any of metadata, system or data, else
3314 * 1: if RAID5 is configured, or if RAID1 or
3315 * RAID10 is configured and only two mirrors
3317 * 2: if RAID6 is configured, else
3318 * num_mirrors - 1: if RAID1 or RAID10 is
3319 * configured and more than
3320 * 2 mirrors are used.
3322 if (num_tolerated_disk_barrier_failures
> 0 &&
3323 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3324 BTRFS_BLOCK_GROUP_RAID0
)) ||
3325 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3327 num_tolerated_disk_barrier_failures
= 0;
3328 else if (num_tolerated_disk_barrier_failures
> 1) {
3329 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3330 BTRFS_BLOCK_GROUP_RAID5
|
3331 BTRFS_BLOCK_GROUP_RAID10
)) {
3332 num_tolerated_disk_barrier_failures
= 1;
3334 BTRFS_BLOCK_GROUP_RAID6
) {
3335 num_tolerated_disk_barrier_failures
= 2;
3340 up_read(&sinfo
->groups_sem
);
3343 return num_tolerated_disk_barrier_failures
;
3346 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3348 struct list_head
*head
;
3349 struct btrfs_device
*dev
;
3350 struct btrfs_super_block
*sb
;
3351 struct btrfs_dev_item
*dev_item
;
3355 int total_errors
= 0;
3358 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3359 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3360 backup_super_roots(root
->fs_info
);
3362 sb
= root
->fs_info
->super_for_commit
;
3363 dev_item
= &sb
->dev_item
;
3365 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3366 head
= &root
->fs_info
->fs_devices
->devices
;
3369 ret
= barrier_all_devices(root
->fs_info
);
3372 &root
->fs_info
->fs_devices
->device_list_mutex
);
3373 btrfs_error(root
->fs_info
, ret
,
3374 "errors while submitting device barriers.");
3379 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3384 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3387 btrfs_set_stack_device_generation(dev_item
, 0);
3388 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3389 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3390 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3391 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3392 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3393 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3394 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3395 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3396 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3398 flags
= btrfs_super_flags(sb
);
3399 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3401 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3405 if (total_errors
> max_errors
) {
3406 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3409 /* This shouldn't happen. FUA is masked off if unsupported */
3414 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3417 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3420 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3424 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3425 if (total_errors
> max_errors
) {
3426 btrfs_error(root
->fs_info
, -EIO
,
3427 "%d errors while writing supers", total_errors
);
3433 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3434 struct btrfs_root
*root
, int max_mirrors
)
3438 ret
= write_all_supers(root
, max_mirrors
);
3442 /* Drop a fs root from the radix tree and free it. */
3443 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3444 struct btrfs_root
*root
)
3446 spin_lock(&fs_info
->fs_roots_radix_lock
);
3447 radix_tree_delete(&fs_info
->fs_roots_radix
,
3448 (unsigned long)root
->root_key
.objectid
);
3449 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3451 if (btrfs_root_refs(&root
->root_item
) == 0)
3452 synchronize_srcu(&fs_info
->subvol_srcu
);
3454 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3455 btrfs_free_log(NULL
, root
);
3456 btrfs_free_log_root_tree(NULL
, fs_info
);
3459 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3460 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3464 static void free_fs_root(struct btrfs_root
*root
)
3466 iput(root
->cache_inode
);
3467 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3469 free_anon_bdev(root
->anon_dev
);
3470 free_extent_buffer(root
->node
);
3471 free_extent_buffer(root
->commit_root
);
3472 kfree(root
->free_ino_ctl
);
3473 kfree(root
->free_ino_pinned
);
3475 btrfs_put_fs_root(root
);
3478 void btrfs_free_fs_root(struct btrfs_root
*root
)
3483 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3485 u64 root_objectid
= 0;
3486 struct btrfs_root
*gang
[8];
3491 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3492 (void **)gang
, root_objectid
,
3497 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3498 for (i
= 0; i
< ret
; i
++) {
3501 root_objectid
= gang
[i
]->root_key
.objectid
;
3502 err
= btrfs_orphan_cleanup(gang
[i
]);
3511 int btrfs_commit_super(struct btrfs_root
*root
)
3513 struct btrfs_trans_handle
*trans
;
3516 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3517 btrfs_run_delayed_iputs(root
);
3518 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3519 wake_up_process(root
->fs_info
->cleaner_kthread
);
3521 /* wait until ongoing cleanup work done */
3522 down_write(&root
->fs_info
->cleanup_work_sem
);
3523 up_write(&root
->fs_info
->cleanup_work_sem
);
3525 trans
= btrfs_join_transaction(root
);
3527 return PTR_ERR(trans
);
3528 ret
= btrfs_commit_transaction(trans
, root
);
3531 /* run commit again to drop the original snapshot */
3532 trans
= btrfs_join_transaction(root
);
3534 return PTR_ERR(trans
);
3535 ret
= btrfs_commit_transaction(trans
, root
);
3538 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3540 btrfs_error(root
->fs_info
, ret
,
3541 "Failed to sync btree inode to disk.");
3545 ret
= write_ctree_super(NULL
, root
, 0);
3549 int close_ctree(struct btrfs_root
*root
)
3551 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3554 fs_info
->closing
= 1;
3557 /* pause restriper - we want to resume on mount */
3558 btrfs_pause_balance(fs_info
);
3560 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3562 btrfs_scrub_cancel(fs_info
);
3564 /* wait for any defraggers to finish */
3565 wait_event(fs_info
->transaction_wait
,
3566 (atomic_read(&fs_info
->defrag_running
) == 0));
3568 /* clear out the rbtree of defraggable inodes */
3569 btrfs_cleanup_defrag_inodes(fs_info
);
3571 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3572 ret
= btrfs_commit_super(root
);
3574 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3577 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3578 btrfs_error_commit_super(root
);
3580 btrfs_put_block_group_cache(fs_info
);
3582 kthread_stop(fs_info
->transaction_kthread
);
3583 kthread_stop(fs_info
->cleaner_kthread
);
3585 fs_info
->closing
= 2;
3588 btrfs_free_qgroup_config(root
->fs_info
);
3590 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3591 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3592 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3595 btrfs_free_block_groups(fs_info
);
3597 btrfs_stop_all_workers(fs_info
);
3599 del_fs_roots(fs_info
);
3601 free_root_pointers(fs_info
, 1);
3603 iput(fs_info
->btree_inode
);
3605 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3606 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3607 btrfsic_unmount(root
, fs_info
->fs_devices
);
3610 btrfs_close_devices(fs_info
->fs_devices
);
3611 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3613 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3614 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3615 bdi_destroy(&fs_info
->bdi
);
3616 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3618 btrfs_free_stripe_hash_table(fs_info
);
3623 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3627 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3629 ret
= extent_buffer_uptodate(buf
);
3633 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3634 parent_transid
, atomic
);
3640 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3642 return set_extent_buffer_uptodate(buf
);
3645 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3647 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3648 u64 transid
= btrfs_header_generation(buf
);
3651 btrfs_assert_tree_locked(buf
);
3652 if (transid
!= root
->fs_info
->generation
)
3653 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3654 "found %llu running %llu\n",
3655 (unsigned long long)buf
->start
,
3656 (unsigned long long)transid
,
3657 (unsigned long long)root
->fs_info
->generation
);
3658 was_dirty
= set_extent_buffer_dirty(buf
);
3660 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3662 root
->fs_info
->dirty_metadata_batch
);
3665 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3669 * looks as though older kernels can get into trouble with
3670 * this code, they end up stuck in balance_dirty_pages forever
3674 if (current
->flags
& PF_MEMALLOC
)
3678 btrfs_balance_delayed_items(root
);
3680 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3681 BTRFS_DIRTY_METADATA_THRESH
);
3683 balance_dirty_pages_ratelimited(
3684 root
->fs_info
->btree_inode
->i_mapping
);
3689 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3691 __btrfs_btree_balance_dirty(root
, 1);
3694 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3696 __btrfs_btree_balance_dirty(root
, 0);
3699 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3701 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3702 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3705 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3709 * Placeholder for checks
3714 static void btrfs_error_commit_super(struct btrfs_root
*root
)
3716 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3717 btrfs_run_delayed_iputs(root
);
3718 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3720 down_write(&root
->fs_info
->cleanup_work_sem
);
3721 up_write(&root
->fs_info
->cleanup_work_sem
);
3723 /* cleanup FS via transaction */
3724 btrfs_cleanup_transaction(root
);
3727 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
3728 struct btrfs_root
*root
)
3730 struct btrfs_inode
*btrfs_inode
;
3731 struct list_head splice
;
3733 INIT_LIST_HEAD(&splice
);
3735 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3736 spin_lock(&root
->fs_info
->ordered_root_lock
);
3738 list_splice_init(&t
->ordered_operations
, &splice
);
3739 while (!list_empty(&splice
)) {
3740 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3741 ordered_operations
);
3743 list_del_init(&btrfs_inode
->ordered_operations
);
3744 spin_unlock(&root
->fs_info
->ordered_root_lock
);
3746 btrfs_invalidate_inodes(btrfs_inode
->root
);
3748 spin_lock(&root
->fs_info
->ordered_root_lock
);
3751 spin_unlock(&root
->fs_info
->ordered_root_lock
);
3752 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3755 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3757 struct btrfs_ordered_extent
*ordered
;
3759 spin_lock(&root
->ordered_extent_lock
);
3761 * This will just short circuit the ordered completion stuff which will
3762 * make sure the ordered extent gets properly cleaned up.
3764 list_for_each_entry(ordered
, &root
->ordered_extents
,
3766 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3767 spin_unlock(&root
->ordered_extent_lock
);
3770 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
3772 struct btrfs_root
*root
;
3773 struct list_head splice
;
3775 INIT_LIST_HEAD(&splice
);
3777 spin_lock(&fs_info
->ordered_root_lock
);
3778 list_splice_init(&fs_info
->ordered_roots
, &splice
);
3779 while (!list_empty(&splice
)) {
3780 root
= list_first_entry(&splice
, struct btrfs_root
,
3782 list_del_init(&root
->ordered_root
);
3784 btrfs_destroy_ordered_extents(root
);
3786 cond_resched_lock(&fs_info
->ordered_root_lock
);
3788 spin_unlock(&fs_info
->ordered_root_lock
);
3791 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3792 struct btrfs_root
*root
)
3794 struct rb_node
*node
;
3795 struct btrfs_delayed_ref_root
*delayed_refs
;
3796 struct btrfs_delayed_ref_node
*ref
;
3799 delayed_refs
= &trans
->delayed_refs
;
3801 spin_lock(&delayed_refs
->lock
);
3802 if (delayed_refs
->num_entries
== 0) {
3803 spin_unlock(&delayed_refs
->lock
);
3804 printk(KERN_INFO
"delayed_refs has NO entry\n");
3808 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3809 struct btrfs_delayed_ref_head
*head
= NULL
;
3810 bool pin_bytes
= false;
3812 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3813 atomic_set(&ref
->refs
, 1);
3814 if (btrfs_delayed_ref_is_head(ref
)) {
3816 head
= btrfs_delayed_node_to_head(ref
);
3817 if (!mutex_trylock(&head
->mutex
)) {
3818 atomic_inc(&ref
->refs
);
3819 spin_unlock(&delayed_refs
->lock
);
3821 /* Need to wait for the delayed ref to run */
3822 mutex_lock(&head
->mutex
);
3823 mutex_unlock(&head
->mutex
);
3824 btrfs_put_delayed_ref(ref
);
3826 spin_lock(&delayed_refs
->lock
);
3830 if (head
->must_insert_reserved
)
3832 btrfs_free_delayed_extent_op(head
->extent_op
);
3833 delayed_refs
->num_heads
--;
3834 if (list_empty(&head
->cluster
))
3835 delayed_refs
->num_heads_ready
--;
3836 list_del_init(&head
->cluster
);
3840 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3841 delayed_refs
->num_entries
--;
3842 spin_unlock(&delayed_refs
->lock
);
3845 btrfs_pin_extent(root
, ref
->bytenr
,
3847 mutex_unlock(&head
->mutex
);
3849 btrfs_put_delayed_ref(ref
);
3852 spin_lock(&delayed_refs
->lock
);
3855 spin_unlock(&delayed_refs
->lock
);
3860 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
)
3862 struct btrfs_pending_snapshot
*snapshot
;
3863 struct list_head splice
;
3865 INIT_LIST_HEAD(&splice
);
3867 list_splice_init(&t
->pending_snapshots
, &splice
);
3869 while (!list_empty(&splice
)) {
3870 snapshot
= list_entry(splice
.next
,
3871 struct btrfs_pending_snapshot
,
3873 snapshot
->error
= -ECANCELED
;
3874 list_del_init(&snapshot
->list
);
3878 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3880 struct btrfs_inode
*btrfs_inode
;
3881 struct list_head splice
;
3883 INIT_LIST_HEAD(&splice
);
3885 spin_lock(&root
->delalloc_lock
);
3886 list_splice_init(&root
->delalloc_inodes
, &splice
);
3888 while (!list_empty(&splice
)) {
3889 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
3892 list_del_init(&btrfs_inode
->delalloc_inodes
);
3893 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3894 &btrfs_inode
->runtime_flags
);
3895 spin_unlock(&root
->delalloc_lock
);
3897 btrfs_invalidate_inodes(btrfs_inode
->root
);
3899 spin_lock(&root
->delalloc_lock
);
3902 spin_unlock(&root
->delalloc_lock
);
3905 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
3907 struct btrfs_root
*root
;
3908 struct list_head splice
;
3910 INIT_LIST_HEAD(&splice
);
3912 spin_lock(&fs_info
->delalloc_root_lock
);
3913 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
3914 while (!list_empty(&splice
)) {
3915 root
= list_first_entry(&splice
, struct btrfs_root
,
3917 list_del_init(&root
->delalloc_root
);
3918 root
= btrfs_grab_fs_root(root
);
3920 spin_unlock(&fs_info
->delalloc_root_lock
);
3922 btrfs_destroy_delalloc_inodes(root
);
3923 btrfs_put_fs_root(root
);
3925 spin_lock(&fs_info
->delalloc_root_lock
);
3927 spin_unlock(&fs_info
->delalloc_root_lock
);
3930 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3931 struct extent_io_tree
*dirty_pages
,
3935 struct extent_buffer
*eb
;
3940 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3945 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3946 while (start
<= end
) {
3947 eb
= btrfs_find_tree_block(root
, start
,
3949 start
+= root
->leafsize
;
3952 wait_on_extent_buffer_writeback(eb
);
3954 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3956 clear_extent_buffer_dirty(eb
);
3957 free_extent_buffer_stale(eb
);
3964 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3965 struct extent_io_tree
*pinned_extents
)
3967 struct extent_io_tree
*unpin
;
3973 unpin
= pinned_extents
;
3976 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3977 EXTENT_DIRTY
, NULL
);
3982 if (btrfs_test_opt(root
, DISCARD
))
3983 ret
= btrfs_error_discard_extent(root
, start
,
3987 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3988 btrfs_error_unpin_extent_range(root
, start
, end
);
3993 if (unpin
== &root
->fs_info
->freed_extents
[0])
3994 unpin
= &root
->fs_info
->freed_extents
[1];
3996 unpin
= &root
->fs_info
->freed_extents
[0];
4004 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4005 struct btrfs_root
*root
)
4007 btrfs_destroy_delayed_refs(cur_trans
, root
);
4008 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
4009 cur_trans
->dirty_pages
.dirty_bytes
);
4011 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4012 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4014 btrfs_evict_pending_snapshots(cur_trans
);
4016 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4017 wake_up(&root
->fs_info
->transaction_wait
);
4019 btrfs_destroy_delayed_inodes(root
);
4020 btrfs_assert_delayed_root_empty(root
);
4022 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4024 btrfs_destroy_pinned_extent(root
,
4025 root
->fs_info
->pinned_extents
);
4027 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4028 wake_up(&cur_trans
->commit_wait
);
4031 memset(cur_trans, 0, sizeof(*cur_trans));
4032 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4036 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4038 struct btrfs_transaction
*t
;
4041 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4043 spin_lock(&root
->fs_info
->trans_lock
);
4044 list_splice_init(&root
->fs_info
->trans_list
, &list
);
4045 root
->fs_info
->running_transaction
= NULL
;
4046 spin_unlock(&root
->fs_info
->trans_lock
);
4048 while (!list_empty(&list
)) {
4049 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
4051 btrfs_destroy_ordered_operations(t
, root
);
4053 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4055 btrfs_destroy_delayed_refs(t
, root
);
4058 * FIXME: cleanup wait for commit
4059 * We needn't acquire the lock here, because we are during
4060 * the umount, there is no other task which will change it.
4062 t
->state
= TRANS_STATE_COMMIT_START
;
4064 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
4065 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4067 btrfs_evict_pending_snapshots(t
);
4069 t
->state
= TRANS_STATE_UNBLOCKED
;
4071 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
4072 wake_up(&root
->fs_info
->transaction_wait
);
4074 btrfs_destroy_delayed_inodes(root
);
4075 btrfs_assert_delayed_root_empty(root
);
4077 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4079 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
4082 btrfs_destroy_pinned_extent(root
,
4083 root
->fs_info
->pinned_extents
);
4085 t
->state
= TRANS_STATE_COMPLETED
;
4087 if (waitqueue_active(&t
->commit_wait
))
4088 wake_up(&t
->commit_wait
);
4090 atomic_set(&t
->use_count
, 0);
4091 list_del_init(&t
->list
);
4092 memset(t
, 0, sizeof(*t
));
4093 kmem_cache_free(btrfs_transaction_cachep
, t
);
4096 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4101 static struct extent_io_ops btree_extent_io_ops
= {
4102 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4103 .readpage_io_failed_hook
= btree_io_failed_hook
,
4104 .submit_bio_hook
= btree_submit_bio_hook
,
4105 /* note we're sharing with inode.c for the merge bio hook */
4106 .merge_bio_hook
= btrfs_merge_bio_hook
,