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_ORPHAN_OBJECTID
, .name_stem
= "orphan" },
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 * helper to read a given tree block, doing retries as required when
361 * the checksums don't match and we have alternate mirrors to try.
363 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
364 struct extent_buffer
*eb
,
365 u64 start
, u64 parent_transid
)
367 struct extent_io_tree
*io_tree
;
372 int failed_mirror
= 0;
374 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
375 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
377 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
379 btree_get_extent
, mirror_num
);
381 if (!verify_parent_transid(io_tree
, eb
,
389 * This buffer's crc is fine, but its contents are corrupted, so
390 * there is no reason to read the other copies, they won't be
393 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
396 num_copies
= btrfs_num_copies(root
->fs_info
,
401 if (!failed_mirror
) {
403 failed_mirror
= eb
->read_mirror
;
407 if (mirror_num
== failed_mirror
)
410 if (mirror_num
> num_copies
)
414 if (failed
&& !ret
&& failed_mirror
)
415 repair_eb_io_failure(root
, eb
, failed_mirror
);
421 * checksum a dirty tree block before IO. This has extra checks to make sure
422 * we only fill in the checksum field in the first page of a multi-page block
425 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
427 struct extent_io_tree
*tree
;
428 u64 start
= page_offset(page
);
430 struct extent_buffer
*eb
;
432 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
434 eb
= (struct extent_buffer
*)page
->private;
435 if (page
!= eb
->pages
[0])
437 found_start
= btrfs_header_bytenr(eb
);
438 if (found_start
!= start
) {
442 if (!PageUptodate(page
)) {
446 csum_tree_block(root
, eb
, 0);
450 static int check_tree_block_fsid(struct btrfs_root
*root
,
451 struct extent_buffer
*eb
)
453 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
454 u8 fsid
[BTRFS_UUID_SIZE
];
457 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
460 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
464 fs_devices
= fs_devices
->seed
;
469 #define CORRUPT(reason, eb, root, slot) \
470 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471 "root=%llu, slot=%d\n", reason, \
472 (unsigned long long)btrfs_header_bytenr(eb), \
473 (unsigned long long)root->objectid, slot)
475 static noinline
int check_leaf(struct btrfs_root
*root
,
476 struct extent_buffer
*leaf
)
478 struct btrfs_key key
;
479 struct btrfs_key leaf_key
;
480 u32 nritems
= btrfs_header_nritems(leaf
);
486 /* Check the 0 item */
487 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
488 BTRFS_LEAF_DATA_SIZE(root
)) {
489 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
494 * Check to make sure each items keys are in the correct order and their
495 * offsets make sense. We only have to loop through nritems-1 because
496 * we check the current slot against the next slot, which verifies the
497 * next slot's offset+size makes sense and that the current's slot
500 for (slot
= 0; slot
< nritems
- 1; slot
++) {
501 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
502 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
504 /* Make sure the keys are in the right order */
505 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
506 CORRUPT("bad key order", leaf
, root
, slot
);
511 * Make sure the offset and ends are right, remember that the
512 * item data starts at the end of the leaf and grows towards the
515 if (btrfs_item_offset_nr(leaf
, slot
) !=
516 btrfs_item_end_nr(leaf
, slot
+ 1)) {
517 CORRUPT("slot offset bad", leaf
, root
, slot
);
522 * Check to make sure that we don't point outside of the leaf,
523 * just incase all the items are consistent to eachother, but
524 * all point outside of the leaf.
526 if (btrfs_item_end_nr(leaf
, slot
) >
527 BTRFS_LEAF_DATA_SIZE(root
)) {
528 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
536 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
537 struct extent_state
*state
, int mirror
)
539 struct extent_io_tree
*tree
;
542 struct extent_buffer
*eb
;
543 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
550 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
551 eb
= (struct extent_buffer
*)page
->private;
553 /* the pending IO might have been the only thing that kept this buffer
554 * in memory. Make sure we have a ref for all this other checks
556 extent_buffer_get(eb
);
558 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
562 eb
->read_mirror
= mirror
;
563 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
568 found_start
= btrfs_header_bytenr(eb
);
569 if (found_start
!= eb
->start
) {
570 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
572 (unsigned long long)found_start
,
573 (unsigned long long)eb
->start
);
577 if (check_tree_block_fsid(root
, eb
)) {
578 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
579 (unsigned long long)eb
->start
);
583 found_level
= btrfs_header_level(eb
);
584 if (found_level
>= BTRFS_MAX_LEVEL
) {
585 btrfs_info(root
->fs_info
, "bad tree block level %d\n",
586 (int)btrfs_header_level(eb
));
591 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
594 ret
= csum_tree_block(root
, eb
, 1);
601 * If this is a leaf block and it is corrupt, set the corrupt bit so
602 * that we don't try and read the other copies of this block, just
605 if (found_level
== 0 && check_leaf(root
, eb
)) {
606 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
611 set_extent_buffer_uptodate(eb
);
614 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
615 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
619 * our io error hook is going to dec the io pages
620 * again, we have to make sure it has something
623 atomic_inc(&eb
->io_pages
);
624 clear_extent_buffer_uptodate(eb
);
626 free_extent_buffer(eb
);
631 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
633 struct extent_buffer
*eb
;
634 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
636 eb
= (struct extent_buffer
*)page
->private;
637 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
638 eb
->read_mirror
= failed_mirror
;
639 atomic_dec(&eb
->io_pages
);
640 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
641 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
642 return -EIO
; /* we fixed nothing */
645 static void end_workqueue_bio(struct bio
*bio
, int err
)
647 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
648 struct btrfs_fs_info
*fs_info
;
650 fs_info
= end_io_wq
->info
;
651 end_io_wq
->error
= err
;
652 end_io_wq
->work
.func
= end_workqueue_fn
;
653 end_io_wq
->work
.flags
= 0;
655 if (bio
->bi_rw
& REQ_WRITE
) {
656 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
657 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
659 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
660 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
662 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
663 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
666 btrfs_queue_worker(&fs_info
->endio_write_workers
,
669 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
670 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
672 else if (end_io_wq
->metadata
)
673 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
676 btrfs_queue_worker(&fs_info
->endio_workers
,
682 * For the metadata arg you want
685 * 1 - if normal metadta
686 * 2 - if writing to the free space cache area
687 * 3 - raid parity work
689 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
692 struct end_io_wq
*end_io_wq
;
693 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
697 end_io_wq
->private = bio
->bi_private
;
698 end_io_wq
->end_io
= bio
->bi_end_io
;
699 end_io_wq
->info
= info
;
700 end_io_wq
->error
= 0;
701 end_io_wq
->bio
= bio
;
702 end_io_wq
->metadata
= metadata
;
704 bio
->bi_private
= end_io_wq
;
705 bio
->bi_end_io
= end_workqueue_bio
;
709 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
711 unsigned long limit
= min_t(unsigned long,
712 info
->workers
.max_workers
,
713 info
->fs_devices
->open_devices
);
717 static void run_one_async_start(struct btrfs_work
*work
)
719 struct async_submit_bio
*async
;
722 async
= container_of(work
, struct async_submit_bio
, work
);
723 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
724 async
->mirror_num
, async
->bio_flags
,
730 static void run_one_async_done(struct btrfs_work
*work
)
732 struct btrfs_fs_info
*fs_info
;
733 struct async_submit_bio
*async
;
736 async
= container_of(work
, struct async_submit_bio
, work
);
737 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
739 limit
= btrfs_async_submit_limit(fs_info
);
740 limit
= limit
* 2 / 3;
742 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
743 waitqueue_active(&fs_info
->async_submit_wait
))
744 wake_up(&fs_info
->async_submit_wait
);
746 /* If an error occured we just want to clean up the bio and move on */
748 bio_endio(async
->bio
, async
->error
);
752 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
753 async
->mirror_num
, async
->bio_flags
,
757 static void run_one_async_free(struct btrfs_work
*work
)
759 struct async_submit_bio
*async
;
761 async
= container_of(work
, struct async_submit_bio
, work
);
765 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
766 int rw
, struct bio
*bio
, int mirror_num
,
767 unsigned long bio_flags
,
769 extent_submit_bio_hook_t
*submit_bio_start
,
770 extent_submit_bio_hook_t
*submit_bio_done
)
772 struct async_submit_bio
*async
;
774 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
778 async
->inode
= inode
;
781 async
->mirror_num
= mirror_num
;
782 async
->submit_bio_start
= submit_bio_start
;
783 async
->submit_bio_done
= submit_bio_done
;
785 async
->work
.func
= run_one_async_start
;
786 async
->work
.ordered_func
= run_one_async_done
;
787 async
->work
.ordered_free
= run_one_async_free
;
789 async
->work
.flags
= 0;
790 async
->bio_flags
= bio_flags
;
791 async
->bio_offset
= bio_offset
;
795 atomic_inc(&fs_info
->nr_async_submits
);
798 btrfs_set_work_high_prio(&async
->work
);
800 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
802 while (atomic_read(&fs_info
->async_submit_draining
) &&
803 atomic_read(&fs_info
->nr_async_submits
)) {
804 wait_event(fs_info
->async_submit_wait
,
805 (atomic_read(&fs_info
->nr_async_submits
) == 0));
811 static int btree_csum_one_bio(struct bio
*bio
)
813 struct bio_vec
*bvec
= bio
->bi_io_vec
;
815 struct btrfs_root
*root
;
818 WARN_ON(bio
->bi_vcnt
<= 0);
819 while (bio_index
< bio
->bi_vcnt
) {
820 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
821 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
830 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
831 struct bio
*bio
, int mirror_num
,
832 unsigned long bio_flags
,
836 * when we're called for a write, we're already in the async
837 * submission context. Just jump into btrfs_map_bio
839 return btree_csum_one_bio(bio
);
842 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
843 int mirror_num
, unsigned long bio_flags
,
849 * when we're called for a write, we're already in the async
850 * submission context. Just jump into btrfs_map_bio
852 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
858 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
860 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
869 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
870 int mirror_num
, unsigned long bio_flags
,
873 int async
= check_async_write(inode
, bio_flags
);
876 if (!(rw
& REQ_WRITE
)) {
878 * called for a read, do the setup so that checksum validation
879 * can happen in the async kernel threads
881 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
885 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
888 ret
= btree_csum_one_bio(bio
);
891 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
895 * kthread helpers are used to submit writes so that
896 * checksumming can happen in parallel across all CPUs
898 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
899 inode
, rw
, bio
, mirror_num
, 0,
901 __btree_submit_bio_start
,
902 __btree_submit_bio_done
);
912 #ifdef CONFIG_MIGRATION
913 static int btree_migratepage(struct address_space
*mapping
,
914 struct page
*newpage
, struct page
*page
,
915 enum migrate_mode mode
)
918 * we can't safely write a btree page from here,
919 * we haven't done the locking hook
924 * Buffers may be managed in a filesystem specific way.
925 * We must have no buffers or drop them.
927 if (page_has_private(page
) &&
928 !try_to_release_page(page
, GFP_KERNEL
))
930 return migrate_page(mapping
, newpage
, page
, mode
);
935 static int btree_writepages(struct address_space
*mapping
,
936 struct writeback_control
*wbc
)
938 struct extent_io_tree
*tree
;
939 struct btrfs_fs_info
*fs_info
;
942 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
943 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
945 if (wbc
->for_kupdate
)
948 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
949 /* this is a bit racy, but that's ok */
950 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
951 BTRFS_DIRTY_METADATA_THRESH
);
955 return btree_write_cache_pages(mapping
, wbc
);
958 static int btree_readpage(struct file
*file
, struct page
*page
)
960 struct extent_io_tree
*tree
;
961 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
962 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
965 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
967 if (PageWriteback(page
) || PageDirty(page
))
970 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
971 * slab allocation from alloc_extent_state down the callchain where
972 * it'd hit a BUG_ON as those flags are not allowed.
974 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
976 return try_release_extent_buffer(page
, gfp_flags
);
979 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
981 struct extent_io_tree
*tree
;
982 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
983 extent_invalidatepage(tree
, page
, offset
);
984 btree_releasepage(page
, GFP_NOFS
);
985 if (PagePrivate(page
)) {
986 printk(KERN_WARNING
"btrfs warning page private not zero "
987 "on page %llu\n", (unsigned long long)page_offset(page
));
988 ClearPagePrivate(page
);
989 set_page_private(page
, 0);
990 page_cache_release(page
);
994 static int btree_set_page_dirty(struct page
*page
)
997 struct extent_buffer
*eb
;
999 BUG_ON(!PagePrivate(page
));
1000 eb
= (struct extent_buffer
*)page
->private;
1002 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1003 BUG_ON(!atomic_read(&eb
->refs
));
1004 btrfs_assert_tree_locked(eb
);
1006 return __set_page_dirty_nobuffers(page
);
1009 static const struct address_space_operations btree_aops
= {
1010 .readpage
= btree_readpage
,
1011 .writepages
= btree_writepages
,
1012 .releasepage
= btree_releasepage
,
1013 .invalidatepage
= btree_invalidatepage
,
1014 #ifdef CONFIG_MIGRATION
1015 .migratepage
= btree_migratepage
,
1017 .set_page_dirty
= btree_set_page_dirty
,
1020 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1023 struct extent_buffer
*buf
= NULL
;
1024 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1027 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1030 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1031 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1032 free_extent_buffer(buf
);
1036 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1037 int mirror_num
, struct extent_buffer
**eb
)
1039 struct extent_buffer
*buf
= NULL
;
1040 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1041 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1044 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1048 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1050 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1051 btree_get_extent
, mirror_num
);
1053 free_extent_buffer(buf
);
1057 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1058 free_extent_buffer(buf
);
1060 } else if (extent_buffer_uptodate(buf
)) {
1063 free_extent_buffer(buf
);
1068 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1069 u64 bytenr
, u32 blocksize
)
1071 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1072 struct extent_buffer
*eb
;
1073 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1078 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1079 u64 bytenr
, u32 blocksize
)
1081 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1082 struct extent_buffer
*eb
;
1084 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1090 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1092 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1093 buf
->start
+ buf
->len
- 1);
1096 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1098 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1099 buf
->start
, buf
->start
+ buf
->len
- 1);
1102 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1103 u32 blocksize
, u64 parent_transid
)
1105 struct extent_buffer
*buf
= NULL
;
1108 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1112 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1117 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1118 struct extent_buffer
*buf
)
1120 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1122 if (btrfs_header_generation(buf
) ==
1123 fs_info
->running_transaction
->transid
) {
1124 btrfs_assert_tree_locked(buf
);
1126 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1127 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1129 fs_info
->dirty_metadata_batch
);
1130 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1131 btrfs_set_lock_blocking(buf
);
1132 clear_extent_buffer_dirty(buf
);
1137 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1138 u32 stripesize
, struct btrfs_root
*root
,
1139 struct btrfs_fs_info
*fs_info
,
1143 root
->commit_root
= NULL
;
1144 root
->sectorsize
= sectorsize
;
1145 root
->nodesize
= nodesize
;
1146 root
->leafsize
= leafsize
;
1147 root
->stripesize
= stripesize
;
1149 root
->track_dirty
= 0;
1151 root
->orphan_item_inserted
= 0;
1152 root
->orphan_cleanup_state
= 0;
1154 root
->objectid
= objectid
;
1155 root
->last_trans
= 0;
1156 root
->highest_objectid
= 0;
1158 root
->inode_tree
= RB_ROOT
;
1159 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1160 root
->block_rsv
= NULL
;
1161 root
->orphan_block_rsv
= NULL
;
1163 INIT_LIST_HEAD(&root
->dirty_list
);
1164 INIT_LIST_HEAD(&root
->root_list
);
1165 INIT_LIST_HEAD(&root
->logged_list
[0]);
1166 INIT_LIST_HEAD(&root
->logged_list
[1]);
1167 spin_lock_init(&root
->orphan_lock
);
1168 spin_lock_init(&root
->inode_lock
);
1169 spin_lock_init(&root
->accounting_lock
);
1170 spin_lock_init(&root
->log_extents_lock
[0]);
1171 spin_lock_init(&root
->log_extents_lock
[1]);
1172 mutex_init(&root
->objectid_mutex
);
1173 mutex_init(&root
->log_mutex
);
1174 init_waitqueue_head(&root
->log_writer_wait
);
1175 init_waitqueue_head(&root
->log_commit_wait
[0]);
1176 init_waitqueue_head(&root
->log_commit_wait
[1]);
1177 atomic_set(&root
->log_commit
[0], 0);
1178 atomic_set(&root
->log_commit
[1], 0);
1179 atomic_set(&root
->log_writers
, 0);
1180 atomic_set(&root
->log_batch
, 0);
1181 atomic_set(&root
->orphan_inodes
, 0);
1182 root
->log_transid
= 0;
1183 root
->last_log_commit
= 0;
1184 extent_io_tree_init(&root
->dirty_log_pages
,
1185 fs_info
->btree_inode
->i_mapping
);
1187 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1188 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1189 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1190 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1191 root
->defrag_trans_start
= fs_info
->generation
;
1192 init_completion(&root
->kobj_unregister
);
1193 root
->defrag_running
= 0;
1194 root
->root_key
.objectid
= objectid
;
1197 spin_lock_init(&root
->root_item_lock
);
1200 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1201 struct btrfs_fs_info
*fs_info
,
1203 struct btrfs_root
*root
)
1209 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1210 tree_root
->sectorsize
, tree_root
->stripesize
,
1211 root
, fs_info
, objectid
);
1212 ret
= btrfs_find_last_root(tree_root
, objectid
,
1213 &root
->root_item
, &root
->root_key
);
1219 generation
= btrfs_root_generation(&root
->root_item
);
1220 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1221 root
->commit_root
= NULL
;
1222 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1223 blocksize
, generation
);
1224 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1225 free_extent_buffer(root
->node
);
1229 root
->commit_root
= btrfs_root_node(root
);
1233 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1235 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1237 root
->fs_info
= fs_info
;
1241 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1242 struct btrfs_fs_info
*fs_info
,
1245 struct extent_buffer
*leaf
;
1246 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1247 struct btrfs_root
*root
;
1248 struct btrfs_key key
;
1253 root
= btrfs_alloc_root(fs_info
);
1255 return ERR_PTR(-ENOMEM
);
1257 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1258 tree_root
->sectorsize
, tree_root
->stripesize
,
1259 root
, fs_info
, objectid
);
1260 root
->root_key
.objectid
= objectid
;
1261 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1262 root
->root_key
.offset
= 0;
1264 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1265 0, objectid
, NULL
, 0, 0, 0);
1267 ret
= PTR_ERR(leaf
);
1272 bytenr
= leaf
->start
;
1273 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1274 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1275 btrfs_set_header_generation(leaf
, trans
->transid
);
1276 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1277 btrfs_set_header_owner(leaf
, objectid
);
1280 write_extent_buffer(leaf
, fs_info
->fsid
,
1281 (unsigned long)btrfs_header_fsid(leaf
),
1283 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1284 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1286 btrfs_mark_buffer_dirty(leaf
);
1288 root
->commit_root
= btrfs_root_node(root
);
1289 root
->track_dirty
= 1;
1292 root
->root_item
.flags
= 0;
1293 root
->root_item
.byte_limit
= 0;
1294 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1295 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1296 btrfs_set_root_level(&root
->root_item
, 0);
1297 btrfs_set_root_refs(&root
->root_item
, 1);
1298 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1299 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1300 btrfs_set_root_dirid(&root
->root_item
, 0);
1302 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1303 root
->root_item
.drop_level
= 0;
1305 key
.objectid
= objectid
;
1306 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1308 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1312 btrfs_tree_unlock(leaf
);
1318 btrfs_tree_unlock(leaf
);
1319 free_extent_buffer(leaf
);
1323 return ERR_PTR(ret
);
1326 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1327 struct btrfs_fs_info
*fs_info
)
1329 struct btrfs_root
*root
;
1330 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1331 struct extent_buffer
*leaf
;
1333 root
= btrfs_alloc_root(fs_info
);
1335 return ERR_PTR(-ENOMEM
);
1337 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1338 tree_root
->sectorsize
, tree_root
->stripesize
,
1339 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1341 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1342 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1343 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1345 * log trees do not get reference counted because they go away
1346 * before a real commit is actually done. They do store pointers
1347 * to file data extents, and those reference counts still get
1348 * updated (along with back refs to the log tree).
1352 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1353 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1357 return ERR_CAST(leaf
);
1360 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1361 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1362 btrfs_set_header_generation(leaf
, trans
->transid
);
1363 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1364 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1367 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1368 (unsigned long)btrfs_header_fsid(root
->node
),
1370 btrfs_mark_buffer_dirty(root
->node
);
1371 btrfs_tree_unlock(root
->node
);
1375 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1376 struct btrfs_fs_info
*fs_info
)
1378 struct btrfs_root
*log_root
;
1380 log_root
= alloc_log_tree(trans
, fs_info
);
1381 if (IS_ERR(log_root
))
1382 return PTR_ERR(log_root
);
1383 WARN_ON(fs_info
->log_root_tree
);
1384 fs_info
->log_root_tree
= log_root
;
1388 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1389 struct btrfs_root
*root
)
1391 struct btrfs_root
*log_root
;
1392 struct btrfs_inode_item
*inode_item
;
1394 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1395 if (IS_ERR(log_root
))
1396 return PTR_ERR(log_root
);
1398 log_root
->last_trans
= trans
->transid
;
1399 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1401 inode_item
= &log_root
->root_item
.inode
;
1402 inode_item
->generation
= cpu_to_le64(1);
1403 inode_item
->size
= cpu_to_le64(3);
1404 inode_item
->nlink
= cpu_to_le32(1);
1405 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1406 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1408 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1410 WARN_ON(root
->log_root
);
1411 root
->log_root
= log_root
;
1412 root
->log_transid
= 0;
1413 root
->last_log_commit
= 0;
1417 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1418 struct btrfs_key
*location
)
1420 struct btrfs_root
*root
;
1421 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1422 struct btrfs_path
*path
;
1423 struct extent_buffer
*l
;
1429 root
= btrfs_alloc_root(fs_info
);
1431 return ERR_PTR(-ENOMEM
);
1432 if (location
->offset
== (u64
)-1) {
1433 ret
= find_and_setup_root(tree_root
, fs_info
,
1434 location
->objectid
, root
);
1437 return ERR_PTR(ret
);
1442 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1443 tree_root
->sectorsize
, tree_root
->stripesize
,
1444 root
, fs_info
, location
->objectid
);
1446 path
= btrfs_alloc_path();
1449 return ERR_PTR(-ENOMEM
);
1451 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1454 slot
= path
->slots
[0];
1455 btrfs_read_root_item(l
, slot
, &root
->root_item
);
1456 memcpy(&root
->root_key
, location
, sizeof(*location
));
1458 btrfs_free_path(path
);
1463 return ERR_PTR(ret
);
1466 generation
= btrfs_root_generation(&root
->root_item
);
1467 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1468 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1469 blocksize
, generation
);
1470 if (!root
->node
|| !extent_buffer_uptodate(root
->node
)) {
1471 ret
= (!root
->node
) ? -ENOMEM
: -EIO
;
1473 free_extent_buffer(root
->node
);
1475 return ERR_PTR(ret
);
1478 root
->commit_root
= btrfs_root_node(root
);
1479 BUG_ON(!root
->node
); /* -ENOMEM */
1481 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1483 btrfs_check_and_init_root_item(&root
->root_item
);
1489 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1490 struct btrfs_key
*location
)
1492 struct btrfs_root
*root
;
1495 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1496 return fs_info
->tree_root
;
1497 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1498 return fs_info
->extent_root
;
1499 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1500 return fs_info
->chunk_root
;
1501 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1502 return fs_info
->dev_root
;
1503 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1504 return fs_info
->csum_root
;
1505 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1506 return fs_info
->quota_root
? fs_info
->quota_root
:
1509 spin_lock(&fs_info
->fs_roots_radix_lock
);
1510 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1511 (unsigned long)location
->objectid
);
1512 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1516 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1520 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1521 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1523 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1528 btrfs_init_free_ino_ctl(root
);
1529 mutex_init(&root
->fs_commit_mutex
);
1530 spin_lock_init(&root
->cache_lock
);
1531 init_waitqueue_head(&root
->cache_wait
);
1533 ret
= get_anon_bdev(&root
->anon_dev
);
1537 if (btrfs_root_refs(&root
->root_item
) == 0) {
1542 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1546 root
->orphan_item_inserted
= 1;
1548 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1552 spin_lock(&fs_info
->fs_roots_radix_lock
);
1553 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1554 (unsigned long)root
->root_key
.objectid
,
1559 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1560 radix_tree_preload_end();
1562 if (ret
== -EEXIST
) {
1569 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1570 root
->root_key
.objectid
);
1575 return ERR_PTR(ret
);
1578 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1580 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1582 struct btrfs_device
*device
;
1583 struct backing_dev_info
*bdi
;
1586 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1589 bdi
= blk_get_backing_dev_info(device
->bdev
);
1590 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1600 * If this fails, caller must call bdi_destroy() to get rid of the
1603 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1607 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1608 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1612 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1613 bdi
->congested_fn
= btrfs_congested_fn
;
1614 bdi
->congested_data
= info
;
1619 * called by the kthread helper functions to finally call the bio end_io
1620 * functions. This is where read checksum verification actually happens
1622 static void end_workqueue_fn(struct btrfs_work
*work
)
1625 struct end_io_wq
*end_io_wq
;
1626 struct btrfs_fs_info
*fs_info
;
1629 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1630 bio
= end_io_wq
->bio
;
1631 fs_info
= end_io_wq
->info
;
1633 error
= end_io_wq
->error
;
1634 bio
->bi_private
= end_io_wq
->private;
1635 bio
->bi_end_io
= end_io_wq
->end_io
;
1637 bio_endio(bio
, error
);
1640 static int cleaner_kthread(void *arg
)
1642 struct btrfs_root
*root
= arg
;
1647 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1648 down_read_trylock(&root
->fs_info
->sb
->s_umount
)) {
1649 if (mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1650 btrfs_run_delayed_iputs(root
);
1651 again
= btrfs_clean_one_deleted_snapshot(root
);
1652 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1654 btrfs_run_defrag_inodes(root
->fs_info
);
1655 up_read(&root
->fs_info
->sb
->s_umount
);
1658 if (!try_to_freeze() && !again
) {
1659 set_current_state(TASK_INTERRUPTIBLE
);
1660 if (!kthread_should_stop())
1662 __set_current_state(TASK_RUNNING
);
1664 } while (!kthread_should_stop());
1668 static int transaction_kthread(void *arg
)
1670 struct btrfs_root
*root
= arg
;
1671 struct btrfs_trans_handle
*trans
;
1672 struct btrfs_transaction
*cur
;
1675 unsigned long delay
;
1679 cannot_commit
= false;
1681 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1683 spin_lock(&root
->fs_info
->trans_lock
);
1684 cur
= root
->fs_info
->running_transaction
;
1686 spin_unlock(&root
->fs_info
->trans_lock
);
1690 now
= get_seconds();
1691 if (!cur
->blocked
&&
1692 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1693 spin_unlock(&root
->fs_info
->trans_lock
);
1697 transid
= cur
->transid
;
1698 spin_unlock(&root
->fs_info
->trans_lock
);
1700 /* If the file system is aborted, this will always fail. */
1701 trans
= btrfs_attach_transaction(root
);
1702 if (IS_ERR(trans
)) {
1703 if (PTR_ERR(trans
) != -ENOENT
)
1704 cannot_commit
= true;
1707 if (transid
== trans
->transid
) {
1708 btrfs_commit_transaction(trans
, root
);
1710 btrfs_end_transaction(trans
, root
);
1713 wake_up_process(root
->fs_info
->cleaner_kthread
);
1714 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1716 if (!try_to_freeze()) {
1717 set_current_state(TASK_INTERRUPTIBLE
);
1718 if (!kthread_should_stop() &&
1719 (!btrfs_transaction_blocked(root
->fs_info
) ||
1721 schedule_timeout(delay
);
1722 __set_current_state(TASK_RUNNING
);
1724 } while (!kthread_should_stop());
1729 * this will find the highest generation in the array of
1730 * root backups. The index of the highest array is returned,
1731 * or -1 if we can't find anything.
1733 * We check to make sure the array is valid by comparing the
1734 * generation of the latest root in the array with the generation
1735 * in the super block. If they don't match we pitch it.
1737 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1740 int newest_index
= -1;
1741 struct btrfs_root_backup
*root_backup
;
1744 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1745 root_backup
= info
->super_copy
->super_roots
+ i
;
1746 cur
= btrfs_backup_tree_root_gen(root_backup
);
1747 if (cur
== newest_gen
)
1751 /* check to see if we actually wrapped around */
1752 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1753 root_backup
= info
->super_copy
->super_roots
;
1754 cur
= btrfs_backup_tree_root_gen(root_backup
);
1755 if (cur
== newest_gen
)
1758 return newest_index
;
1763 * find the oldest backup so we know where to store new entries
1764 * in the backup array. This will set the backup_root_index
1765 * field in the fs_info struct
1767 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1770 int newest_index
= -1;
1772 newest_index
= find_newest_super_backup(info
, newest_gen
);
1773 /* if there was garbage in there, just move along */
1774 if (newest_index
== -1) {
1775 info
->backup_root_index
= 0;
1777 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1782 * copy all the root pointers into the super backup array.
1783 * this will bump the backup pointer by one when it is
1786 static void backup_super_roots(struct btrfs_fs_info
*info
)
1789 struct btrfs_root_backup
*root_backup
;
1792 next_backup
= info
->backup_root_index
;
1793 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1794 BTRFS_NUM_BACKUP_ROOTS
;
1797 * just overwrite the last backup if we're at the same generation
1798 * this happens only at umount
1800 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1801 if (btrfs_backup_tree_root_gen(root_backup
) ==
1802 btrfs_header_generation(info
->tree_root
->node
))
1803 next_backup
= last_backup
;
1805 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1808 * make sure all of our padding and empty slots get zero filled
1809 * regardless of which ones we use today
1811 memset(root_backup
, 0, sizeof(*root_backup
));
1813 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1815 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1816 btrfs_set_backup_tree_root_gen(root_backup
,
1817 btrfs_header_generation(info
->tree_root
->node
));
1819 btrfs_set_backup_tree_root_level(root_backup
,
1820 btrfs_header_level(info
->tree_root
->node
));
1822 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1823 btrfs_set_backup_chunk_root_gen(root_backup
,
1824 btrfs_header_generation(info
->chunk_root
->node
));
1825 btrfs_set_backup_chunk_root_level(root_backup
,
1826 btrfs_header_level(info
->chunk_root
->node
));
1828 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1829 btrfs_set_backup_extent_root_gen(root_backup
,
1830 btrfs_header_generation(info
->extent_root
->node
));
1831 btrfs_set_backup_extent_root_level(root_backup
,
1832 btrfs_header_level(info
->extent_root
->node
));
1835 * we might commit during log recovery, which happens before we set
1836 * the fs_root. Make sure it is valid before we fill it in.
1838 if (info
->fs_root
&& info
->fs_root
->node
) {
1839 btrfs_set_backup_fs_root(root_backup
,
1840 info
->fs_root
->node
->start
);
1841 btrfs_set_backup_fs_root_gen(root_backup
,
1842 btrfs_header_generation(info
->fs_root
->node
));
1843 btrfs_set_backup_fs_root_level(root_backup
,
1844 btrfs_header_level(info
->fs_root
->node
));
1847 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1848 btrfs_set_backup_dev_root_gen(root_backup
,
1849 btrfs_header_generation(info
->dev_root
->node
));
1850 btrfs_set_backup_dev_root_level(root_backup
,
1851 btrfs_header_level(info
->dev_root
->node
));
1853 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1854 btrfs_set_backup_csum_root_gen(root_backup
,
1855 btrfs_header_generation(info
->csum_root
->node
));
1856 btrfs_set_backup_csum_root_level(root_backup
,
1857 btrfs_header_level(info
->csum_root
->node
));
1859 btrfs_set_backup_total_bytes(root_backup
,
1860 btrfs_super_total_bytes(info
->super_copy
));
1861 btrfs_set_backup_bytes_used(root_backup
,
1862 btrfs_super_bytes_used(info
->super_copy
));
1863 btrfs_set_backup_num_devices(root_backup
,
1864 btrfs_super_num_devices(info
->super_copy
));
1867 * if we don't copy this out to the super_copy, it won't get remembered
1868 * for the next commit
1870 memcpy(&info
->super_copy
->super_roots
,
1871 &info
->super_for_commit
->super_roots
,
1872 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1876 * this copies info out of the root backup array and back into
1877 * the in-memory super block. It is meant to help iterate through
1878 * the array, so you send it the number of backups you've already
1879 * tried and the last backup index you used.
1881 * this returns -1 when it has tried all the backups
1883 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1884 struct btrfs_super_block
*super
,
1885 int *num_backups_tried
, int *backup_index
)
1887 struct btrfs_root_backup
*root_backup
;
1888 int newest
= *backup_index
;
1890 if (*num_backups_tried
== 0) {
1891 u64 gen
= btrfs_super_generation(super
);
1893 newest
= find_newest_super_backup(info
, gen
);
1897 *backup_index
= newest
;
1898 *num_backups_tried
= 1;
1899 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1900 /* we've tried all the backups, all done */
1903 /* jump to the next oldest backup */
1904 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1905 BTRFS_NUM_BACKUP_ROOTS
;
1906 *backup_index
= newest
;
1907 *num_backups_tried
+= 1;
1909 root_backup
= super
->super_roots
+ newest
;
1911 btrfs_set_super_generation(super
,
1912 btrfs_backup_tree_root_gen(root_backup
));
1913 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1914 btrfs_set_super_root_level(super
,
1915 btrfs_backup_tree_root_level(root_backup
));
1916 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1919 * fixme: the total bytes and num_devices need to match or we should
1922 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1923 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1927 /* helper to cleanup workers */
1928 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
1930 btrfs_stop_workers(&fs_info
->generic_worker
);
1931 btrfs_stop_workers(&fs_info
->fixup_workers
);
1932 btrfs_stop_workers(&fs_info
->delalloc_workers
);
1933 btrfs_stop_workers(&fs_info
->workers
);
1934 btrfs_stop_workers(&fs_info
->endio_workers
);
1935 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
1936 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
1937 btrfs_stop_workers(&fs_info
->rmw_workers
);
1938 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
1939 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1940 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
1941 btrfs_stop_workers(&fs_info
->submit_workers
);
1942 btrfs_stop_workers(&fs_info
->delayed_workers
);
1943 btrfs_stop_workers(&fs_info
->caching_workers
);
1944 btrfs_stop_workers(&fs_info
->readahead_workers
);
1945 btrfs_stop_workers(&fs_info
->flush_workers
);
1946 btrfs_stop_workers(&fs_info
->qgroup_rescan_workers
);
1949 /* helper to cleanup tree roots */
1950 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1952 free_extent_buffer(info
->tree_root
->node
);
1953 free_extent_buffer(info
->tree_root
->commit_root
);
1954 free_extent_buffer(info
->dev_root
->node
);
1955 free_extent_buffer(info
->dev_root
->commit_root
);
1956 free_extent_buffer(info
->extent_root
->node
);
1957 free_extent_buffer(info
->extent_root
->commit_root
);
1958 free_extent_buffer(info
->csum_root
->node
);
1959 free_extent_buffer(info
->csum_root
->commit_root
);
1960 if (info
->quota_root
) {
1961 free_extent_buffer(info
->quota_root
->node
);
1962 free_extent_buffer(info
->quota_root
->commit_root
);
1965 info
->tree_root
->node
= NULL
;
1966 info
->tree_root
->commit_root
= NULL
;
1967 info
->dev_root
->node
= NULL
;
1968 info
->dev_root
->commit_root
= NULL
;
1969 info
->extent_root
->node
= NULL
;
1970 info
->extent_root
->commit_root
= NULL
;
1971 info
->csum_root
->node
= NULL
;
1972 info
->csum_root
->commit_root
= NULL
;
1973 if (info
->quota_root
) {
1974 info
->quota_root
->node
= NULL
;
1975 info
->quota_root
->commit_root
= NULL
;
1979 free_extent_buffer(info
->chunk_root
->node
);
1980 free_extent_buffer(info
->chunk_root
->commit_root
);
1981 info
->chunk_root
->node
= NULL
;
1982 info
->chunk_root
->commit_root
= NULL
;
1986 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
1989 struct btrfs_root
*gang
[8];
1992 while (!list_empty(&fs_info
->dead_roots
)) {
1993 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
1994 struct btrfs_root
, root_list
);
1995 list_del(&gang
[0]->root_list
);
1997 if (gang
[0]->in_radix
) {
1998 btrfs_free_fs_root(fs_info
, gang
[0]);
2000 free_extent_buffer(gang
[0]->node
);
2001 free_extent_buffer(gang
[0]->commit_root
);
2007 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2012 for (i
= 0; i
< ret
; i
++)
2013 btrfs_free_fs_root(fs_info
, gang
[i
]);
2017 int open_ctree(struct super_block
*sb
,
2018 struct btrfs_fs_devices
*fs_devices
,
2028 struct btrfs_key location
;
2029 struct buffer_head
*bh
;
2030 struct btrfs_super_block
*disk_super
;
2031 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2032 struct btrfs_root
*tree_root
;
2033 struct btrfs_root
*extent_root
;
2034 struct btrfs_root
*csum_root
;
2035 struct btrfs_root
*chunk_root
;
2036 struct btrfs_root
*dev_root
;
2037 struct btrfs_root
*quota_root
;
2038 struct btrfs_root
*log_tree_root
;
2041 int num_backups_tried
= 0;
2042 int backup_index
= 0;
2044 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2045 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
2046 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
2047 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2048 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
2049 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
2051 if (!tree_root
|| !extent_root
|| !csum_root
||
2052 !chunk_root
|| !dev_root
|| !quota_root
) {
2057 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2063 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2069 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2074 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2075 (1 + ilog2(nr_cpu_ids
));
2077 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2080 goto fail_dirty_metadata_bytes
;
2083 fs_info
->btree_inode
= new_inode(sb
);
2084 if (!fs_info
->btree_inode
) {
2086 goto fail_delalloc_bytes
;
2089 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2091 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2092 INIT_LIST_HEAD(&fs_info
->trans_list
);
2093 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2094 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2095 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2096 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2097 spin_lock_init(&fs_info
->delalloc_lock
);
2098 spin_lock_init(&fs_info
->trans_lock
);
2099 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2100 spin_lock_init(&fs_info
->delayed_iput_lock
);
2101 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2102 spin_lock_init(&fs_info
->free_chunk_lock
);
2103 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2104 spin_lock_init(&fs_info
->super_lock
);
2105 rwlock_init(&fs_info
->tree_mod_log_lock
);
2106 mutex_init(&fs_info
->reloc_mutex
);
2107 seqlock_init(&fs_info
->profiles_lock
);
2109 init_completion(&fs_info
->kobj_unregister
);
2110 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2111 INIT_LIST_HEAD(&fs_info
->space_info
);
2112 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2113 btrfs_mapping_init(&fs_info
->mapping_tree
);
2114 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2115 BTRFS_BLOCK_RSV_GLOBAL
);
2116 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2117 BTRFS_BLOCK_RSV_DELALLOC
);
2118 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2119 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2120 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2121 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2122 BTRFS_BLOCK_RSV_DELOPS
);
2123 atomic_set(&fs_info
->nr_async_submits
, 0);
2124 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2125 atomic_set(&fs_info
->async_submit_draining
, 0);
2126 atomic_set(&fs_info
->nr_async_bios
, 0);
2127 atomic_set(&fs_info
->defrag_running
, 0);
2128 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2130 fs_info
->max_inline
= 8192 * 1024;
2131 fs_info
->metadata_ratio
= 0;
2132 fs_info
->defrag_inodes
= RB_ROOT
;
2133 fs_info
->trans_no_join
= 0;
2134 fs_info
->free_chunk_space
= 0;
2135 fs_info
->tree_mod_log
= RB_ROOT
;
2137 /* readahead state */
2138 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2139 spin_lock_init(&fs_info
->reada_lock
);
2141 fs_info
->thread_pool_size
= min_t(unsigned long,
2142 num_online_cpus() + 2, 8);
2144 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2145 spin_lock_init(&fs_info
->ordered_extent_lock
);
2146 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2148 if (!fs_info
->delayed_root
) {
2152 btrfs_init_delayed_root(fs_info
->delayed_root
);
2154 mutex_init(&fs_info
->scrub_lock
);
2155 atomic_set(&fs_info
->scrubs_running
, 0);
2156 atomic_set(&fs_info
->scrub_pause_req
, 0);
2157 atomic_set(&fs_info
->scrubs_paused
, 0);
2158 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2159 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2160 init_rwsem(&fs_info
->scrub_super_lock
);
2161 fs_info
->scrub_workers_refcnt
= 0;
2162 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2163 fs_info
->check_integrity_print_mask
= 0;
2166 spin_lock_init(&fs_info
->balance_lock
);
2167 mutex_init(&fs_info
->balance_mutex
);
2168 atomic_set(&fs_info
->balance_running
, 0);
2169 atomic_set(&fs_info
->balance_pause_req
, 0);
2170 atomic_set(&fs_info
->balance_cancel_req
, 0);
2171 fs_info
->balance_ctl
= NULL
;
2172 init_waitqueue_head(&fs_info
->balance_wait_q
);
2174 sb
->s_blocksize
= 4096;
2175 sb
->s_blocksize_bits
= blksize_bits(4096);
2176 sb
->s_bdi
= &fs_info
->bdi
;
2178 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2179 set_nlink(fs_info
->btree_inode
, 1);
2181 * we set the i_size on the btree inode to the max possible int.
2182 * the real end of the address space is determined by all of
2183 * the devices in the system
2185 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2186 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2187 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2189 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2190 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2191 fs_info
->btree_inode
->i_mapping
);
2192 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2193 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2195 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2197 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2198 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2199 sizeof(struct btrfs_key
));
2200 set_bit(BTRFS_INODE_DUMMY
,
2201 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2202 insert_inode_hash(fs_info
->btree_inode
);
2204 spin_lock_init(&fs_info
->block_group_cache_lock
);
2205 fs_info
->block_group_cache_tree
= RB_ROOT
;
2206 fs_info
->first_logical_byte
= (u64
)-1;
2208 extent_io_tree_init(&fs_info
->freed_extents
[0],
2209 fs_info
->btree_inode
->i_mapping
);
2210 extent_io_tree_init(&fs_info
->freed_extents
[1],
2211 fs_info
->btree_inode
->i_mapping
);
2212 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2213 fs_info
->do_barriers
= 1;
2216 mutex_init(&fs_info
->ordered_operations_mutex
);
2217 mutex_init(&fs_info
->tree_log_mutex
);
2218 mutex_init(&fs_info
->chunk_mutex
);
2219 mutex_init(&fs_info
->transaction_kthread_mutex
);
2220 mutex_init(&fs_info
->cleaner_mutex
);
2221 mutex_init(&fs_info
->volume_mutex
);
2222 init_rwsem(&fs_info
->extent_commit_sem
);
2223 init_rwsem(&fs_info
->cleanup_work_sem
);
2224 init_rwsem(&fs_info
->subvol_sem
);
2225 fs_info
->dev_replace
.lock_owner
= 0;
2226 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2227 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2228 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2229 mutex_init(&fs_info
->dev_replace
.lock
);
2231 spin_lock_init(&fs_info
->qgroup_lock
);
2232 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2233 fs_info
->qgroup_tree
= RB_ROOT
;
2234 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2235 fs_info
->qgroup_seq
= 1;
2236 fs_info
->quota_enabled
= 0;
2237 fs_info
->pending_quota_state
= 0;
2238 mutex_init(&fs_info
->qgroup_rescan_lock
);
2240 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2241 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2243 init_waitqueue_head(&fs_info
->transaction_throttle
);
2244 init_waitqueue_head(&fs_info
->transaction_wait
);
2245 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2246 init_waitqueue_head(&fs_info
->async_submit_wait
);
2248 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2254 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2255 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2257 invalidate_bdev(fs_devices
->latest_bdev
);
2258 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2264 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2265 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2266 sizeof(*fs_info
->super_for_commit
));
2269 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2271 disk_super
= fs_info
->super_copy
;
2272 if (!btrfs_super_root(disk_super
))
2275 /* check FS state, whether FS is broken. */
2276 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2277 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2279 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2281 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2287 * run through our array of backup supers and setup
2288 * our ring pointer to the oldest one
2290 generation
= btrfs_super_generation(disk_super
);
2291 find_oldest_super_backup(fs_info
, generation
);
2294 * In the long term, we'll store the compression type in the super
2295 * block, and it'll be used for per file compression control.
2297 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2299 ret
= btrfs_parse_options(tree_root
, options
);
2305 features
= btrfs_super_incompat_flags(disk_super
) &
2306 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2308 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2309 "unsupported optional features (%Lx).\n",
2310 (unsigned long long)features
);
2315 if (btrfs_super_leafsize(disk_super
) !=
2316 btrfs_super_nodesize(disk_super
)) {
2317 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2318 "blocksizes don't match. node %d leaf %d\n",
2319 btrfs_super_nodesize(disk_super
),
2320 btrfs_super_leafsize(disk_super
));
2324 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2325 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2326 "blocksize (%d) was too large\n",
2327 btrfs_super_leafsize(disk_super
));
2332 features
= btrfs_super_incompat_flags(disk_super
);
2333 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2334 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2335 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2337 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2338 printk(KERN_ERR
"btrfs: has skinny extents\n");
2341 * flag our filesystem as having big metadata blocks if
2342 * they are bigger than the page size
2344 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2345 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2346 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2347 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2350 nodesize
= btrfs_super_nodesize(disk_super
);
2351 leafsize
= btrfs_super_leafsize(disk_super
);
2352 sectorsize
= btrfs_super_sectorsize(disk_super
);
2353 stripesize
= btrfs_super_stripesize(disk_super
);
2354 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2355 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2358 * mixed block groups end up with duplicate but slightly offset
2359 * extent buffers for the same range. It leads to corruptions
2361 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2362 (sectorsize
!= leafsize
)) {
2363 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2364 "are not allowed for mixed block groups on %s\n",
2370 * Needn't use the lock because there is no other task which will
2373 btrfs_set_super_incompat_flags(disk_super
, features
);
2375 features
= btrfs_super_compat_ro_flags(disk_super
) &
2376 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2377 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2378 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2379 "unsupported option features (%Lx).\n",
2380 (unsigned long long)features
);
2385 btrfs_init_workers(&fs_info
->generic_worker
,
2386 "genwork", 1, NULL
);
2388 btrfs_init_workers(&fs_info
->workers
, "worker",
2389 fs_info
->thread_pool_size
,
2390 &fs_info
->generic_worker
);
2392 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2393 fs_info
->thread_pool_size
,
2394 &fs_info
->generic_worker
);
2396 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2397 fs_info
->thread_pool_size
,
2398 &fs_info
->generic_worker
);
2400 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2401 min_t(u64
, fs_devices
->num_devices
,
2402 fs_info
->thread_pool_size
),
2403 &fs_info
->generic_worker
);
2405 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2406 2, &fs_info
->generic_worker
);
2408 /* a higher idle thresh on the submit workers makes it much more
2409 * likely that bios will be send down in a sane order to the
2412 fs_info
->submit_workers
.idle_thresh
= 64;
2414 fs_info
->workers
.idle_thresh
= 16;
2415 fs_info
->workers
.ordered
= 1;
2417 fs_info
->delalloc_workers
.idle_thresh
= 2;
2418 fs_info
->delalloc_workers
.ordered
= 1;
2420 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2421 &fs_info
->generic_worker
);
2422 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2423 fs_info
->thread_pool_size
,
2424 &fs_info
->generic_worker
);
2425 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2426 fs_info
->thread_pool_size
,
2427 &fs_info
->generic_worker
);
2428 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2429 "endio-meta-write", fs_info
->thread_pool_size
,
2430 &fs_info
->generic_worker
);
2431 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2432 "endio-raid56", fs_info
->thread_pool_size
,
2433 &fs_info
->generic_worker
);
2434 btrfs_init_workers(&fs_info
->rmw_workers
,
2435 "rmw", fs_info
->thread_pool_size
,
2436 &fs_info
->generic_worker
);
2437 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2438 fs_info
->thread_pool_size
,
2439 &fs_info
->generic_worker
);
2440 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2441 1, &fs_info
->generic_worker
);
2442 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2443 fs_info
->thread_pool_size
,
2444 &fs_info
->generic_worker
);
2445 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2446 fs_info
->thread_pool_size
,
2447 &fs_info
->generic_worker
);
2448 btrfs_init_workers(&fs_info
->qgroup_rescan_workers
, "qgroup-rescan", 1,
2449 &fs_info
->generic_worker
);
2452 * endios are largely parallel and should have a very
2455 fs_info
->endio_workers
.idle_thresh
= 4;
2456 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2457 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2458 fs_info
->rmw_workers
.idle_thresh
= 2;
2460 fs_info
->endio_write_workers
.idle_thresh
= 2;
2461 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2462 fs_info
->readahead_workers
.idle_thresh
= 2;
2465 * btrfs_start_workers can really only fail because of ENOMEM so just
2466 * return -ENOMEM if any of these fail.
2468 ret
= btrfs_start_workers(&fs_info
->workers
);
2469 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2470 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2471 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2472 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2473 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2474 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2475 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2476 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2477 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2478 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2479 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2480 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2481 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2482 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2483 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2484 ret
|= btrfs_start_workers(&fs_info
->qgroup_rescan_workers
);
2487 goto fail_sb_buffer
;
2490 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2491 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2492 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2494 tree_root
->nodesize
= nodesize
;
2495 tree_root
->leafsize
= leafsize
;
2496 tree_root
->sectorsize
= sectorsize
;
2497 tree_root
->stripesize
= stripesize
;
2499 sb
->s_blocksize
= sectorsize
;
2500 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2502 if (disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2503 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2504 goto fail_sb_buffer
;
2507 if (sectorsize
!= PAGE_SIZE
) {
2508 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2509 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2510 goto fail_sb_buffer
;
2513 mutex_lock(&fs_info
->chunk_mutex
);
2514 ret
= btrfs_read_sys_array(tree_root
);
2515 mutex_unlock(&fs_info
->chunk_mutex
);
2517 printk(KERN_WARNING
"btrfs: failed to read the system "
2518 "array on %s\n", sb
->s_id
);
2519 goto fail_sb_buffer
;
2522 blocksize
= btrfs_level_size(tree_root
,
2523 btrfs_super_chunk_root_level(disk_super
));
2524 generation
= btrfs_super_chunk_root_generation(disk_super
);
2526 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2527 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2529 chunk_root
->node
= read_tree_block(chunk_root
,
2530 btrfs_super_chunk_root(disk_super
),
2531 blocksize
, generation
);
2532 if (!chunk_root
->node
||
2533 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2534 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2536 goto fail_tree_roots
;
2538 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2539 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2541 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2542 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2545 ret
= btrfs_read_chunk_tree(chunk_root
);
2547 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2549 goto fail_tree_roots
;
2553 * keep the device that is marked to be the target device for the
2554 * dev_replace procedure
2556 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2558 if (!fs_devices
->latest_bdev
) {
2559 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2561 goto fail_tree_roots
;
2565 blocksize
= btrfs_level_size(tree_root
,
2566 btrfs_super_root_level(disk_super
));
2567 generation
= btrfs_super_generation(disk_super
);
2569 tree_root
->node
= read_tree_block(tree_root
,
2570 btrfs_super_root(disk_super
),
2571 blocksize
, generation
);
2572 if (!tree_root
->node
||
2573 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2574 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2577 goto recovery_tree_root
;
2580 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2581 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2583 ret
= find_and_setup_root(tree_root
, fs_info
,
2584 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2586 goto recovery_tree_root
;
2587 extent_root
->track_dirty
= 1;
2589 ret
= find_and_setup_root(tree_root
, fs_info
,
2590 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2592 goto recovery_tree_root
;
2593 dev_root
->track_dirty
= 1;
2595 ret
= find_and_setup_root(tree_root
, fs_info
,
2596 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2598 goto recovery_tree_root
;
2599 csum_root
->track_dirty
= 1;
2601 ret
= find_and_setup_root(tree_root
, fs_info
,
2602 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2605 quota_root
= fs_info
->quota_root
= NULL
;
2607 quota_root
->track_dirty
= 1;
2608 fs_info
->quota_enabled
= 1;
2609 fs_info
->pending_quota_state
= 1;
2612 fs_info
->generation
= generation
;
2613 fs_info
->last_trans_committed
= generation
;
2615 ret
= btrfs_recover_balance(fs_info
);
2617 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2618 goto fail_block_groups
;
2621 ret
= btrfs_init_dev_stats(fs_info
);
2623 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2625 goto fail_block_groups
;
2628 ret
= btrfs_init_dev_replace(fs_info
);
2630 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2631 goto fail_block_groups
;
2634 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2636 ret
= btrfs_init_space_info(fs_info
);
2638 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2639 goto fail_block_groups
;
2642 ret
= btrfs_read_block_groups(extent_root
);
2644 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2645 goto fail_block_groups
;
2647 fs_info
->num_tolerated_disk_barrier_failures
=
2648 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2649 if (fs_info
->fs_devices
->missing_devices
>
2650 fs_info
->num_tolerated_disk_barrier_failures
&&
2651 !(sb
->s_flags
& MS_RDONLY
)) {
2653 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2654 goto fail_block_groups
;
2657 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2659 if (IS_ERR(fs_info
->cleaner_kthread
))
2660 goto fail_block_groups
;
2662 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2664 "btrfs-transaction");
2665 if (IS_ERR(fs_info
->transaction_kthread
))
2668 if (!btrfs_test_opt(tree_root
, SSD
) &&
2669 !btrfs_test_opt(tree_root
, NOSSD
) &&
2670 !fs_info
->fs_devices
->rotating
) {
2671 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2673 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2676 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2677 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2678 ret
= btrfsic_mount(tree_root
, fs_devices
,
2679 btrfs_test_opt(tree_root
,
2680 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2682 fs_info
->check_integrity_print_mask
);
2684 printk(KERN_WARNING
"btrfs: failed to initialize"
2685 " integrity check module %s\n", sb
->s_id
);
2688 ret
= btrfs_read_qgroup_config(fs_info
);
2690 goto fail_trans_kthread
;
2692 /* do not make disk changes in broken FS */
2693 if (btrfs_super_log_root(disk_super
) != 0) {
2694 u64 bytenr
= btrfs_super_log_root(disk_super
);
2696 if (fs_devices
->rw_devices
== 0) {
2697 printk(KERN_WARNING
"Btrfs log replay required "
2703 btrfs_level_size(tree_root
,
2704 btrfs_super_log_root_level(disk_super
));
2706 log_tree_root
= btrfs_alloc_root(fs_info
);
2707 if (!log_tree_root
) {
2712 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2713 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2715 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2718 if (!log_tree_root
->node
||
2719 !extent_buffer_uptodate(log_tree_root
->node
)) {
2720 printk(KERN_ERR
"btrfs: failed to read log tree\n");
2721 free_extent_buffer(log_tree_root
->node
);
2722 kfree(log_tree_root
);
2723 goto fail_trans_kthread
;
2725 /* returns with log_tree_root freed on success */
2726 ret
= btrfs_recover_log_trees(log_tree_root
);
2728 btrfs_error(tree_root
->fs_info
, ret
,
2729 "Failed to recover log tree");
2730 free_extent_buffer(log_tree_root
->node
);
2731 kfree(log_tree_root
);
2732 goto fail_trans_kthread
;
2735 if (sb
->s_flags
& MS_RDONLY
) {
2736 ret
= btrfs_commit_super(tree_root
);
2738 goto fail_trans_kthread
;
2742 ret
= btrfs_find_orphan_roots(tree_root
);
2744 goto fail_trans_kthread
;
2746 if (!(sb
->s_flags
& MS_RDONLY
)) {
2747 ret
= btrfs_cleanup_fs_roots(fs_info
);
2749 goto fail_trans_kthread
;
2751 ret
= btrfs_recover_relocation(tree_root
);
2754 "btrfs: failed to recover relocation\n");
2760 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2761 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2762 location
.offset
= (u64
)-1;
2764 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2765 if (!fs_info
->fs_root
)
2767 if (IS_ERR(fs_info
->fs_root
)) {
2768 err
= PTR_ERR(fs_info
->fs_root
);
2772 if (sb
->s_flags
& MS_RDONLY
)
2775 down_read(&fs_info
->cleanup_work_sem
);
2776 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2777 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2778 up_read(&fs_info
->cleanup_work_sem
);
2779 close_ctree(tree_root
);
2782 up_read(&fs_info
->cleanup_work_sem
);
2784 ret
= btrfs_resume_balance_async(fs_info
);
2786 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2787 close_ctree(tree_root
);
2791 ret
= btrfs_resume_dev_replace_async(fs_info
);
2793 pr_warn("btrfs: failed to resume dev_replace\n");
2794 close_ctree(tree_root
);
2801 btrfs_free_qgroup_config(fs_info
);
2803 kthread_stop(fs_info
->transaction_kthread
);
2804 del_fs_roots(fs_info
);
2805 btrfs_cleanup_transaction(fs_info
->tree_root
);
2807 kthread_stop(fs_info
->cleaner_kthread
);
2810 * make sure we're done with the btree inode before we stop our
2813 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2816 btrfs_put_block_group_cache(fs_info
);
2817 btrfs_free_block_groups(fs_info
);
2820 free_root_pointers(fs_info
, 1);
2821 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2824 btrfs_stop_all_workers(fs_info
);
2827 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2829 iput(fs_info
->btree_inode
);
2830 fail_delalloc_bytes
:
2831 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2832 fail_dirty_metadata_bytes
:
2833 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2835 bdi_destroy(&fs_info
->bdi
);
2837 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2839 btrfs_free_stripe_hash_table(fs_info
);
2840 btrfs_close_devices(fs_info
->fs_devices
);
2844 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2845 goto fail_tree_roots
;
2847 free_root_pointers(fs_info
, 0);
2849 /* don't use the log in recovery mode, it won't be valid */
2850 btrfs_set_super_log_root(disk_super
, 0);
2852 /* we can't trust the free space cache either */
2853 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2855 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2856 &num_backups_tried
, &backup_index
);
2858 goto fail_block_groups
;
2859 goto retry_root_backup
;
2862 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2865 set_buffer_uptodate(bh
);
2867 struct btrfs_device
*device
= (struct btrfs_device
*)
2870 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2871 "I/O error on %s\n",
2872 rcu_str_deref(device
->name
));
2873 /* note, we dont' set_buffer_write_io_error because we have
2874 * our own ways of dealing with the IO errors
2876 clear_buffer_uptodate(bh
);
2877 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2883 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2885 struct buffer_head
*bh
;
2886 struct buffer_head
*latest
= NULL
;
2887 struct btrfs_super_block
*super
;
2892 /* we would like to check all the supers, but that would make
2893 * a btrfs mount succeed after a mkfs from a different FS.
2894 * So, we need to add a special mount option to scan for
2895 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2897 for (i
= 0; i
< 1; i
++) {
2898 bytenr
= btrfs_sb_offset(i
);
2899 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2901 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2905 super
= (struct btrfs_super_block
*)bh
->b_data
;
2906 if (btrfs_super_bytenr(super
) != bytenr
||
2907 super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2912 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2915 transid
= btrfs_super_generation(super
);
2924 * this should be called twice, once with wait == 0 and
2925 * once with wait == 1. When wait == 0 is done, all the buffer heads
2926 * we write are pinned.
2928 * They are released when wait == 1 is done.
2929 * max_mirrors must be the same for both runs, and it indicates how
2930 * many supers on this one device should be written.
2932 * max_mirrors == 0 means to write them all.
2934 static int write_dev_supers(struct btrfs_device
*device
,
2935 struct btrfs_super_block
*sb
,
2936 int do_barriers
, int wait
, int max_mirrors
)
2938 struct buffer_head
*bh
;
2945 if (max_mirrors
== 0)
2946 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2948 for (i
= 0; i
< max_mirrors
; i
++) {
2949 bytenr
= btrfs_sb_offset(i
);
2950 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2954 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2955 BTRFS_SUPER_INFO_SIZE
);
2961 if (!buffer_uptodate(bh
))
2964 /* drop our reference */
2967 /* drop the reference from the wait == 0 run */
2971 btrfs_set_super_bytenr(sb
, bytenr
);
2974 crc
= btrfs_csum_data((char *)sb
+
2975 BTRFS_CSUM_SIZE
, crc
,
2976 BTRFS_SUPER_INFO_SIZE
-
2978 btrfs_csum_final(crc
, sb
->csum
);
2981 * one reference for us, and we leave it for the
2984 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2985 BTRFS_SUPER_INFO_SIZE
);
2987 printk(KERN_ERR
"btrfs: couldn't get super "
2988 "buffer head for bytenr %Lu\n", bytenr
);
2993 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2995 /* one reference for submit_bh */
2998 set_buffer_uptodate(bh
);
3000 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3001 bh
->b_private
= device
;
3005 * we fua the first super. The others we allow
3008 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3012 return errors
< i
? 0 : -1;
3016 * endio for the write_dev_flush, this will wake anyone waiting
3017 * for the barrier when it is done
3019 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3022 if (err
== -EOPNOTSUPP
)
3023 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3024 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3026 if (bio
->bi_private
)
3027 complete(bio
->bi_private
);
3032 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3033 * sent down. With wait == 1, it waits for the previous flush.
3035 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3038 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3043 if (device
->nobarriers
)
3047 bio
= device
->flush_bio
;
3051 wait_for_completion(&device
->flush_wait
);
3053 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3054 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3055 rcu_str_deref(device
->name
));
3056 device
->nobarriers
= 1;
3057 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3059 btrfs_dev_stat_inc_and_print(device
,
3060 BTRFS_DEV_STAT_FLUSH_ERRS
);
3063 /* drop the reference from the wait == 0 run */
3065 device
->flush_bio
= NULL
;
3071 * one reference for us, and we leave it for the
3074 device
->flush_bio
= NULL
;
3075 bio
= bio_alloc(GFP_NOFS
, 0);
3079 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3080 bio
->bi_bdev
= device
->bdev
;
3081 init_completion(&device
->flush_wait
);
3082 bio
->bi_private
= &device
->flush_wait
;
3083 device
->flush_bio
= bio
;
3086 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3092 * send an empty flush down to each device in parallel,
3093 * then wait for them
3095 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3097 struct list_head
*head
;
3098 struct btrfs_device
*dev
;
3099 int errors_send
= 0;
3100 int errors_wait
= 0;
3103 /* send down all the barriers */
3104 head
= &info
->fs_devices
->devices
;
3105 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3110 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3113 ret
= write_dev_flush(dev
, 0);
3118 /* wait for all the barriers */
3119 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3124 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3127 ret
= write_dev_flush(dev
, 1);
3131 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3132 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3137 int btrfs_calc_num_tolerated_disk_barrier_failures(
3138 struct btrfs_fs_info
*fs_info
)
3140 struct btrfs_ioctl_space_info space
;
3141 struct btrfs_space_info
*sinfo
;
3142 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3143 BTRFS_BLOCK_GROUP_SYSTEM
,
3144 BTRFS_BLOCK_GROUP_METADATA
,
3145 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3149 int num_tolerated_disk_barrier_failures
=
3150 (int)fs_info
->fs_devices
->num_devices
;
3152 for (i
= 0; i
< num_types
; i
++) {
3153 struct btrfs_space_info
*tmp
;
3157 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3158 if (tmp
->flags
== types
[i
]) {
3168 down_read(&sinfo
->groups_sem
);
3169 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3170 if (!list_empty(&sinfo
->block_groups
[c
])) {
3173 btrfs_get_block_group_info(
3174 &sinfo
->block_groups
[c
], &space
);
3175 if (space
.total_bytes
== 0 ||
3176 space
.used_bytes
== 0)
3178 flags
= space
.flags
;
3181 * 0: if dup, single or RAID0 is configured for
3182 * any of metadata, system or data, else
3183 * 1: if RAID5 is configured, or if RAID1 or
3184 * RAID10 is configured and only two mirrors
3186 * 2: if RAID6 is configured, else
3187 * num_mirrors - 1: if RAID1 or RAID10 is
3188 * configured and more than
3189 * 2 mirrors are used.
3191 if (num_tolerated_disk_barrier_failures
> 0 &&
3192 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3193 BTRFS_BLOCK_GROUP_RAID0
)) ||
3194 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3196 num_tolerated_disk_barrier_failures
= 0;
3197 else if (num_tolerated_disk_barrier_failures
> 1) {
3198 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3199 BTRFS_BLOCK_GROUP_RAID5
|
3200 BTRFS_BLOCK_GROUP_RAID10
)) {
3201 num_tolerated_disk_barrier_failures
= 1;
3203 BTRFS_BLOCK_GROUP_RAID5
) {
3204 num_tolerated_disk_barrier_failures
= 2;
3209 up_read(&sinfo
->groups_sem
);
3212 return num_tolerated_disk_barrier_failures
;
3215 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3217 struct list_head
*head
;
3218 struct btrfs_device
*dev
;
3219 struct btrfs_super_block
*sb
;
3220 struct btrfs_dev_item
*dev_item
;
3224 int total_errors
= 0;
3227 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3228 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3229 backup_super_roots(root
->fs_info
);
3231 sb
= root
->fs_info
->super_for_commit
;
3232 dev_item
= &sb
->dev_item
;
3234 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3235 head
= &root
->fs_info
->fs_devices
->devices
;
3238 ret
= barrier_all_devices(root
->fs_info
);
3241 &root
->fs_info
->fs_devices
->device_list_mutex
);
3242 btrfs_error(root
->fs_info
, ret
,
3243 "errors while submitting device barriers.");
3248 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3253 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3256 btrfs_set_stack_device_generation(dev_item
, 0);
3257 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3258 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3259 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3260 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3261 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3262 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3263 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3264 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3265 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3267 flags
= btrfs_super_flags(sb
);
3268 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3270 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3274 if (total_errors
> max_errors
) {
3275 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3278 /* This shouldn't happen. FUA is masked off if unsupported */
3283 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3286 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3289 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3293 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3294 if (total_errors
> max_errors
) {
3295 btrfs_error(root
->fs_info
, -EIO
,
3296 "%d errors while writing supers", total_errors
);
3302 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3303 struct btrfs_root
*root
, int max_mirrors
)
3307 ret
= write_all_supers(root
, max_mirrors
);
3311 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3313 spin_lock(&fs_info
->fs_roots_radix_lock
);
3314 radix_tree_delete(&fs_info
->fs_roots_radix
,
3315 (unsigned long)root
->root_key
.objectid
);
3316 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3318 if (btrfs_root_refs(&root
->root_item
) == 0)
3319 synchronize_srcu(&fs_info
->subvol_srcu
);
3321 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3322 btrfs_free_log(NULL
, root
);
3323 btrfs_free_log_root_tree(NULL
, fs_info
);
3326 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3327 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3331 static void free_fs_root(struct btrfs_root
*root
)
3333 iput(root
->cache_inode
);
3334 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3336 free_anon_bdev(root
->anon_dev
);
3337 free_extent_buffer(root
->node
);
3338 free_extent_buffer(root
->commit_root
);
3339 kfree(root
->free_ino_ctl
);
3340 kfree(root
->free_ino_pinned
);
3345 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3347 u64 root_objectid
= 0;
3348 struct btrfs_root
*gang
[8];
3353 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3354 (void **)gang
, root_objectid
,
3359 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3360 for (i
= 0; i
< ret
; i
++) {
3363 root_objectid
= gang
[i
]->root_key
.objectid
;
3364 err
= btrfs_orphan_cleanup(gang
[i
]);
3373 int btrfs_commit_super(struct btrfs_root
*root
)
3375 struct btrfs_trans_handle
*trans
;
3378 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3379 btrfs_run_delayed_iputs(root
);
3380 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3381 wake_up_process(root
->fs_info
->cleaner_kthread
);
3383 /* wait until ongoing cleanup work done */
3384 down_write(&root
->fs_info
->cleanup_work_sem
);
3385 up_write(&root
->fs_info
->cleanup_work_sem
);
3387 trans
= btrfs_join_transaction(root
);
3389 return PTR_ERR(trans
);
3390 ret
= btrfs_commit_transaction(trans
, root
);
3393 /* run commit again to drop the original snapshot */
3394 trans
= btrfs_join_transaction(root
);
3396 return PTR_ERR(trans
);
3397 ret
= btrfs_commit_transaction(trans
, root
);
3400 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3402 btrfs_error(root
->fs_info
, ret
,
3403 "Failed to sync btree inode to disk.");
3407 ret
= write_ctree_super(NULL
, root
, 0);
3411 int close_ctree(struct btrfs_root
*root
)
3413 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3416 fs_info
->closing
= 1;
3419 /* pause restriper - we want to resume on mount */
3420 btrfs_pause_balance(fs_info
);
3422 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3424 btrfs_scrub_cancel(fs_info
);
3426 /* wait for any defraggers to finish */
3427 wait_event(fs_info
->transaction_wait
,
3428 (atomic_read(&fs_info
->defrag_running
) == 0));
3430 /* clear out the rbtree of defraggable inodes */
3431 btrfs_cleanup_defrag_inodes(fs_info
);
3433 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3434 ret
= btrfs_commit_super(root
);
3436 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3439 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3440 btrfs_error_commit_super(root
);
3442 btrfs_put_block_group_cache(fs_info
);
3444 kthread_stop(fs_info
->transaction_kthread
);
3445 kthread_stop(fs_info
->cleaner_kthread
);
3447 fs_info
->closing
= 2;
3450 btrfs_free_qgroup_config(root
->fs_info
);
3452 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3453 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3454 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3457 free_root_pointers(fs_info
, 1);
3459 btrfs_free_block_groups(fs_info
);
3461 del_fs_roots(fs_info
);
3463 iput(fs_info
->btree_inode
);
3465 btrfs_stop_all_workers(fs_info
);
3467 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3468 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3469 btrfsic_unmount(root
, fs_info
->fs_devices
);
3472 btrfs_close_devices(fs_info
->fs_devices
);
3473 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3475 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3476 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3477 bdi_destroy(&fs_info
->bdi
);
3478 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3480 btrfs_free_stripe_hash_table(fs_info
);
3485 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3489 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3491 ret
= extent_buffer_uptodate(buf
);
3495 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3496 parent_transid
, atomic
);
3502 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3504 return set_extent_buffer_uptodate(buf
);
3507 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3509 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3510 u64 transid
= btrfs_header_generation(buf
);
3513 btrfs_assert_tree_locked(buf
);
3514 if (transid
!= root
->fs_info
->generation
)
3515 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3516 "found %llu running %llu\n",
3517 (unsigned long long)buf
->start
,
3518 (unsigned long long)transid
,
3519 (unsigned long long)root
->fs_info
->generation
);
3520 was_dirty
= set_extent_buffer_dirty(buf
);
3522 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3524 root
->fs_info
->dirty_metadata_batch
);
3527 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3531 * looks as though older kernels can get into trouble with
3532 * this code, they end up stuck in balance_dirty_pages forever
3536 if (current
->flags
& PF_MEMALLOC
)
3540 btrfs_balance_delayed_items(root
);
3542 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3543 BTRFS_DIRTY_METADATA_THRESH
);
3545 balance_dirty_pages_ratelimited(
3546 root
->fs_info
->btree_inode
->i_mapping
);
3551 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3553 __btrfs_btree_balance_dirty(root
, 1);
3556 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3558 __btrfs_btree_balance_dirty(root
, 0);
3561 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3563 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3564 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3567 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3570 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3571 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3581 static void btrfs_error_commit_super(struct btrfs_root
*root
)
3583 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3584 btrfs_run_delayed_iputs(root
);
3585 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3587 down_write(&root
->fs_info
->cleanup_work_sem
);
3588 up_write(&root
->fs_info
->cleanup_work_sem
);
3590 /* cleanup FS via transaction */
3591 btrfs_cleanup_transaction(root
);
3594 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
3595 struct btrfs_root
*root
)
3597 struct btrfs_inode
*btrfs_inode
;
3598 struct list_head splice
;
3600 INIT_LIST_HEAD(&splice
);
3602 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3603 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3605 list_splice_init(&t
->ordered_operations
, &splice
);
3606 while (!list_empty(&splice
)) {
3607 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3608 ordered_operations
);
3610 list_del_init(&btrfs_inode
->ordered_operations
);
3612 btrfs_invalidate_inodes(btrfs_inode
->root
);
3615 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3616 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3619 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3621 struct btrfs_ordered_extent
*ordered
;
3623 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3625 * This will just short circuit the ordered completion stuff which will
3626 * make sure the ordered extent gets properly cleaned up.
3628 list_for_each_entry(ordered
, &root
->fs_info
->ordered_extents
,
3630 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3631 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3634 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3635 struct btrfs_root
*root
)
3637 struct rb_node
*node
;
3638 struct btrfs_delayed_ref_root
*delayed_refs
;
3639 struct btrfs_delayed_ref_node
*ref
;
3642 delayed_refs
= &trans
->delayed_refs
;
3644 spin_lock(&delayed_refs
->lock
);
3645 if (delayed_refs
->num_entries
== 0) {
3646 spin_unlock(&delayed_refs
->lock
);
3647 printk(KERN_INFO
"delayed_refs has NO entry\n");
3651 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3652 struct btrfs_delayed_ref_head
*head
= NULL
;
3654 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3655 atomic_set(&ref
->refs
, 1);
3656 if (btrfs_delayed_ref_is_head(ref
)) {
3658 head
= btrfs_delayed_node_to_head(ref
);
3659 if (!mutex_trylock(&head
->mutex
)) {
3660 atomic_inc(&ref
->refs
);
3661 spin_unlock(&delayed_refs
->lock
);
3663 /* Need to wait for the delayed ref to run */
3664 mutex_lock(&head
->mutex
);
3665 mutex_unlock(&head
->mutex
);
3666 btrfs_put_delayed_ref(ref
);
3668 spin_lock(&delayed_refs
->lock
);
3672 if (head
->must_insert_reserved
)
3673 btrfs_pin_extent(root
, ref
->bytenr
,
3675 btrfs_free_delayed_extent_op(head
->extent_op
);
3676 delayed_refs
->num_heads
--;
3677 if (list_empty(&head
->cluster
))
3678 delayed_refs
->num_heads_ready
--;
3679 list_del_init(&head
->cluster
);
3683 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3684 delayed_refs
->num_entries
--;
3686 mutex_unlock(&head
->mutex
);
3687 spin_unlock(&delayed_refs
->lock
);
3688 btrfs_put_delayed_ref(ref
);
3691 spin_lock(&delayed_refs
->lock
);
3694 spin_unlock(&delayed_refs
->lock
);
3699 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
)
3701 struct btrfs_pending_snapshot
*snapshot
;
3702 struct list_head splice
;
3704 INIT_LIST_HEAD(&splice
);
3706 list_splice_init(&t
->pending_snapshots
, &splice
);
3708 while (!list_empty(&splice
)) {
3709 snapshot
= list_entry(splice
.next
,
3710 struct btrfs_pending_snapshot
,
3712 snapshot
->error
= -ECANCELED
;
3713 list_del_init(&snapshot
->list
);
3717 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3719 struct btrfs_inode
*btrfs_inode
;
3720 struct list_head splice
;
3722 INIT_LIST_HEAD(&splice
);
3724 spin_lock(&root
->fs_info
->delalloc_lock
);
3725 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3727 while (!list_empty(&splice
)) {
3728 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3731 list_del_init(&btrfs_inode
->delalloc_inodes
);
3732 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3733 &btrfs_inode
->runtime_flags
);
3735 btrfs_invalidate_inodes(btrfs_inode
->root
);
3738 spin_unlock(&root
->fs_info
->delalloc_lock
);
3741 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3742 struct extent_io_tree
*dirty_pages
,
3746 struct extent_buffer
*eb
;
3751 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3756 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3757 while (start
<= end
) {
3758 eb
= btrfs_find_tree_block(root
, start
,
3763 wait_on_extent_buffer_writeback(eb
);
3765 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3767 clear_extent_buffer_dirty(eb
);
3768 free_extent_buffer_stale(eb
);
3775 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3776 struct extent_io_tree
*pinned_extents
)
3778 struct extent_io_tree
*unpin
;
3784 unpin
= pinned_extents
;
3787 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3788 EXTENT_DIRTY
, NULL
);
3793 if (btrfs_test_opt(root
, DISCARD
))
3794 ret
= btrfs_error_discard_extent(root
, start
,
3798 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3799 btrfs_error_unpin_extent_range(root
, start
, end
);
3804 if (unpin
== &root
->fs_info
->freed_extents
[0])
3805 unpin
= &root
->fs_info
->freed_extents
[1];
3807 unpin
= &root
->fs_info
->freed_extents
[0];
3815 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3816 struct btrfs_root
*root
)
3818 btrfs_destroy_delayed_refs(cur_trans
, root
);
3819 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3820 cur_trans
->dirty_pages
.dirty_bytes
);
3822 /* FIXME: cleanup wait for commit */
3823 cur_trans
->in_commit
= 1;
3824 cur_trans
->blocked
= 1;
3825 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3827 btrfs_evict_pending_snapshots(cur_trans
);
3829 cur_trans
->blocked
= 0;
3830 wake_up(&root
->fs_info
->transaction_wait
);
3832 cur_trans
->commit_done
= 1;
3833 wake_up(&cur_trans
->commit_wait
);
3835 btrfs_destroy_delayed_inodes(root
);
3836 btrfs_assert_delayed_root_empty(root
);
3838 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3840 btrfs_destroy_pinned_extent(root
,
3841 root
->fs_info
->pinned_extents
);
3844 memset(cur_trans, 0, sizeof(*cur_trans));
3845 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3849 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3851 struct btrfs_transaction
*t
;
3854 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3856 spin_lock(&root
->fs_info
->trans_lock
);
3857 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3858 root
->fs_info
->trans_no_join
= 1;
3859 spin_unlock(&root
->fs_info
->trans_lock
);
3861 while (!list_empty(&list
)) {
3862 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3864 btrfs_destroy_ordered_operations(t
, root
);
3866 btrfs_destroy_ordered_extents(root
);
3868 btrfs_destroy_delayed_refs(t
, root
);
3870 /* FIXME: cleanup wait for commit */
3874 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3875 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3877 btrfs_evict_pending_snapshots(t
);
3881 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3882 wake_up(&root
->fs_info
->transaction_wait
);
3886 if (waitqueue_active(&t
->commit_wait
))
3887 wake_up(&t
->commit_wait
);
3889 btrfs_destroy_delayed_inodes(root
);
3890 btrfs_assert_delayed_root_empty(root
);
3892 btrfs_destroy_delalloc_inodes(root
);
3894 spin_lock(&root
->fs_info
->trans_lock
);
3895 root
->fs_info
->running_transaction
= NULL
;
3896 spin_unlock(&root
->fs_info
->trans_lock
);
3898 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3901 btrfs_destroy_pinned_extent(root
,
3902 root
->fs_info
->pinned_extents
);
3904 atomic_set(&t
->use_count
, 0);
3905 list_del_init(&t
->list
);
3906 memset(t
, 0, sizeof(*t
));
3907 kmem_cache_free(btrfs_transaction_cachep
, t
);
3910 spin_lock(&root
->fs_info
->trans_lock
);
3911 root
->fs_info
->trans_no_join
= 0;
3912 spin_unlock(&root
->fs_info
->trans_lock
);
3913 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3918 static struct extent_io_ops btree_extent_io_ops
= {
3919 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3920 .readpage_io_failed_hook
= btree_io_failed_hook
,
3921 .submit_bio_hook
= btree_submit_bio_hook
,
3922 /* note we're sharing with inode.c for the merge bio hook */
3923 .merge_bio_hook
= btrfs_merge_bio_hook
,