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 <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
51 #include <asm/cpufeature.h>
54 static struct extent_io_ops btree_extent_io_ops
;
55 static void end_workqueue_fn(struct btrfs_work
*work
);
56 static void free_fs_root(struct btrfs_root
*root
);
57 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
59 static void btrfs_destroy_ordered_operations(struct btrfs_root
*root
);
60 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
61 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
62 struct btrfs_root
*root
);
63 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
);
64 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
65 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
66 struct extent_io_tree
*dirty_pages
,
68 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
69 struct extent_io_tree
*pinned_extents
);
72 * end_io_wq structs are used to do processing in task context when an IO is
73 * complete. This is used during reads to verify checksums, and it is used
74 * by writes to insert metadata for new file extents after IO is complete.
80 struct btrfs_fs_info
*info
;
83 struct list_head list
;
84 struct btrfs_work work
;
88 * async submit bios are used to offload expensive checksumming
89 * onto the worker threads. They checksum file and metadata bios
90 * just before they are sent down the IO stack.
92 struct async_submit_bio
{
95 struct list_head list
;
96 extent_submit_bio_hook_t
*submit_bio_start
;
97 extent_submit_bio_hook_t
*submit_bio_done
;
100 unsigned long bio_flags
;
102 * bio_offset is optional, can be used if the pages in the bio
103 * can't tell us where in the file the bio should go
106 struct btrfs_work work
;
111 * Lockdep class keys for extent_buffer->lock's in this root. For a given
112 * eb, the lockdep key is determined by the btrfs_root it belongs to and
113 * the level the eb occupies in the tree.
115 * Different roots are used for different purposes and may nest inside each
116 * other and they require separate keysets. As lockdep keys should be
117 * static, assign keysets according to the purpose of the root as indicated
118 * by btrfs_root->objectid. This ensures that all special purpose roots
119 * have separate keysets.
121 * Lock-nesting across peer nodes is always done with the immediate parent
122 * node locked thus preventing deadlock. As lockdep doesn't know this, use
123 * subclass to avoid triggering lockdep warning in such cases.
125 * The key is set by the readpage_end_io_hook after the buffer has passed
126 * csum validation but before the pages are unlocked. It is also set by
127 * btrfs_init_new_buffer on freshly allocated blocks.
129 * We also add a check to make sure the highest level of the tree is the
130 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
131 * needs update as well.
133 #ifdef CONFIG_DEBUG_LOCK_ALLOC
134 # if BTRFS_MAX_LEVEL != 8
138 static struct btrfs_lockdep_keyset
{
139 u64 id
; /* root objectid */
140 const char *name_stem
; /* lock name stem */
141 char names
[BTRFS_MAX_LEVEL
+ 1][20];
142 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
143 } btrfs_lockdep_keysets
[] = {
144 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
145 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
146 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
147 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
148 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
149 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
150 { .id
= BTRFS_ORPHAN_OBJECTID
, .name_stem
= "orphan" },
151 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
152 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
153 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
154 { .id
= 0, .name_stem
= "tree" },
157 void __init
btrfs_init_lockdep(void)
161 /* initialize lockdep class names */
162 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
163 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
165 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
166 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
167 "btrfs-%s-%02d", ks
->name_stem
, j
);
171 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
174 struct btrfs_lockdep_keyset
*ks
;
176 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
178 /* find the matching keyset, id 0 is the default entry */
179 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
180 if (ks
->id
== objectid
)
183 lockdep_set_class_and_name(&eb
->lock
,
184 &ks
->keys
[level
], ks
->names
[level
]);
190 * extents on the btree inode are pretty simple, there's one extent
191 * that covers the entire device
193 static struct extent_map
*btree_get_extent(struct inode
*inode
,
194 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
197 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
198 struct extent_map
*em
;
201 read_lock(&em_tree
->lock
);
202 em
= lookup_extent_mapping(em_tree
, start
, len
);
205 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
206 read_unlock(&em_tree
->lock
);
209 read_unlock(&em_tree
->lock
);
211 em
= alloc_extent_map();
213 em
= ERR_PTR(-ENOMEM
);
218 em
->block_len
= (u64
)-1;
220 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
222 write_lock(&em_tree
->lock
);
223 ret
= add_extent_mapping(em_tree
, em
);
224 if (ret
== -EEXIST
) {
226 em
= lookup_extent_mapping(em_tree
, start
, len
);
233 write_unlock(&em_tree
->lock
);
239 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
241 return crc32c(seed
, data
, len
);
244 void btrfs_csum_final(u32 crc
, char *result
)
246 put_unaligned_le32(~crc
, result
);
250 * compute the csum for a btree block, and either verify it or write it
251 * into the csum field of the block.
253 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
256 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
259 unsigned long cur_len
;
260 unsigned long offset
= BTRFS_CSUM_SIZE
;
262 unsigned long map_start
;
263 unsigned long map_len
;
266 unsigned long inline_result
;
268 len
= buf
->len
- offset
;
270 err
= map_private_extent_buffer(buf
, offset
, 32,
271 &kaddr
, &map_start
, &map_len
);
274 cur_len
= min(len
, map_len
- (offset
- map_start
));
275 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
280 if (csum_size
> sizeof(inline_result
)) {
281 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
285 result
= (char *)&inline_result
;
288 btrfs_csum_final(crc
, result
);
291 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
294 memcpy(&found
, result
, csum_size
);
296 read_extent_buffer(buf
, &val
, 0, csum_size
);
297 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
298 "failed on %llu wanted %X found %X "
300 root
->fs_info
->sb
->s_id
,
301 (unsigned long long)buf
->start
, val
, found
,
302 btrfs_header_level(buf
));
303 if (result
!= (char *)&inline_result
)
308 write_extent_buffer(buf
, result
, 0, csum_size
);
310 if (result
!= (char *)&inline_result
)
316 * we can't consider a given block up to date unless the transid of the
317 * block matches the transid in the parent node's pointer. This is how we
318 * detect blocks that either didn't get written at all or got written
319 * in the wrong place.
321 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
322 struct extent_buffer
*eb
, u64 parent_transid
,
325 struct extent_state
*cached_state
= NULL
;
328 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
334 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
336 if (extent_buffer_uptodate(eb
) &&
337 btrfs_header_generation(eb
) == parent_transid
) {
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
343 (unsigned long long)eb
->start
,
344 (unsigned long long)parent_transid
,
345 (unsigned long long)btrfs_header_generation(eb
));
347 clear_extent_buffer_uptodate(eb
);
349 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
350 &cached_state
, GFP_NOFS
);
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
358 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
359 struct extent_buffer
*eb
,
360 u64 start
, u64 parent_transid
)
362 struct extent_io_tree
*io_tree
;
367 int failed_mirror
= 0;
369 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
370 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
372 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
374 btree_get_extent
, mirror_num
);
376 if (!verify_parent_transid(io_tree
, eb
,
384 * This buffer's crc is fine, but its contents are corrupted, so
385 * there is no reason to read the other copies, they won't be
388 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
391 num_copies
= btrfs_num_copies(root
->fs_info
,
396 if (!failed_mirror
) {
398 failed_mirror
= eb
->read_mirror
;
402 if (mirror_num
== failed_mirror
)
405 if (mirror_num
> num_copies
)
409 if (failed
&& !ret
&& failed_mirror
)
410 repair_eb_io_failure(root
, eb
, failed_mirror
);
416 * checksum a dirty tree block before IO. This has extra checks to make sure
417 * we only fill in the checksum field in the first page of a multi-page block
420 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
422 struct extent_io_tree
*tree
;
423 u64 start
= page_offset(page
);
425 struct extent_buffer
*eb
;
427 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
429 eb
= (struct extent_buffer
*)page
->private;
430 if (page
!= eb
->pages
[0])
432 found_start
= btrfs_header_bytenr(eb
);
433 if (found_start
!= start
) {
437 if (!PageUptodate(page
)) {
441 csum_tree_block(root
, eb
, 0);
445 static int check_tree_block_fsid(struct btrfs_root
*root
,
446 struct extent_buffer
*eb
)
448 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
449 u8 fsid
[BTRFS_UUID_SIZE
];
452 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
455 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
459 fs_devices
= fs_devices
->seed
;
464 #define CORRUPT(reason, eb, root, slot) \
465 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
466 "root=%llu, slot=%d\n", reason, \
467 (unsigned long long)btrfs_header_bytenr(eb), \
468 (unsigned long long)root->objectid, slot)
470 static noinline
int check_leaf(struct btrfs_root
*root
,
471 struct extent_buffer
*leaf
)
473 struct btrfs_key key
;
474 struct btrfs_key leaf_key
;
475 u32 nritems
= btrfs_header_nritems(leaf
);
481 /* Check the 0 item */
482 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
483 BTRFS_LEAF_DATA_SIZE(root
)) {
484 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
489 * Check to make sure each items keys are in the correct order and their
490 * offsets make sense. We only have to loop through nritems-1 because
491 * we check the current slot against the next slot, which verifies the
492 * next slot's offset+size makes sense and that the current's slot
495 for (slot
= 0; slot
< nritems
- 1; slot
++) {
496 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
497 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
499 /* Make sure the keys are in the right order */
500 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
501 CORRUPT("bad key order", leaf
, root
, slot
);
506 * Make sure the offset and ends are right, remember that the
507 * item data starts at the end of the leaf and grows towards the
510 if (btrfs_item_offset_nr(leaf
, slot
) !=
511 btrfs_item_end_nr(leaf
, slot
+ 1)) {
512 CORRUPT("slot offset bad", leaf
, root
, slot
);
517 * Check to make sure that we don't point outside of the leaf,
518 * just incase all the items are consistent to eachother, but
519 * all point outside of the leaf.
521 if (btrfs_item_end_nr(leaf
, slot
) >
522 BTRFS_LEAF_DATA_SIZE(root
)) {
523 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
531 struct extent_buffer
*find_eb_for_page(struct extent_io_tree
*tree
,
532 struct page
*page
, int max_walk
)
534 struct extent_buffer
*eb
;
535 u64 start
= page_offset(page
);
539 if (start
< max_walk
)
542 min_start
= start
- max_walk
;
544 while (start
>= min_start
) {
545 eb
= find_extent_buffer(tree
, start
, 0);
548 * we found an extent buffer and it contains our page
551 if (eb
->start
<= target
&&
552 eb
->start
+ eb
->len
> target
)
555 /* we found an extent buffer that wasn't for us */
556 free_extent_buffer(eb
);
561 start
-= PAGE_CACHE_SIZE
;
566 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
567 struct extent_state
*state
, int mirror
)
569 struct extent_io_tree
*tree
;
572 struct extent_buffer
*eb
;
573 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
580 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
581 eb
= (struct extent_buffer
*)page
->private;
583 /* the pending IO might have been the only thing that kept this buffer
584 * in memory. Make sure we have a ref for all this other checks
586 extent_buffer_get(eb
);
588 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
592 eb
->read_mirror
= mirror
;
593 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
598 found_start
= btrfs_header_bytenr(eb
);
599 if (found_start
!= eb
->start
) {
600 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
602 (unsigned long long)found_start
,
603 (unsigned long long)eb
->start
);
607 if (check_tree_block_fsid(root
, eb
)) {
608 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
609 (unsigned long long)eb
->start
);
613 found_level
= btrfs_header_level(eb
);
615 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
618 ret
= csum_tree_block(root
, eb
, 1);
625 * If this is a leaf block and it is corrupt, set the corrupt bit so
626 * that we don't try and read the other copies of this block, just
629 if (found_level
== 0 && check_leaf(root
, eb
)) {
630 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
635 set_extent_buffer_uptodate(eb
);
637 if (test_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
)) {
638 clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
);
639 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
643 clear_extent_buffer_uptodate(eb
);
644 free_extent_buffer(eb
);
649 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
651 struct extent_buffer
*eb
;
652 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
654 eb
= (struct extent_buffer
*)page
->private;
655 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
656 eb
->read_mirror
= failed_mirror
;
657 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
658 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
659 return -EIO
; /* we fixed nothing */
662 static void end_workqueue_bio(struct bio
*bio
, int err
)
664 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
665 struct btrfs_fs_info
*fs_info
;
667 fs_info
= end_io_wq
->info
;
668 end_io_wq
->error
= err
;
669 end_io_wq
->work
.func
= end_workqueue_fn
;
670 end_io_wq
->work
.flags
= 0;
672 if (bio
->bi_rw
& REQ_WRITE
) {
673 if (end_io_wq
->metadata
== 1)
674 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
676 else if (end_io_wq
->metadata
== 2)
677 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
680 btrfs_queue_worker(&fs_info
->endio_write_workers
,
683 if (end_io_wq
->metadata
)
684 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
687 btrfs_queue_worker(&fs_info
->endio_workers
,
693 * For the metadata arg you want
696 * 1 - if normal metadta
697 * 2 - if writing to the free space cache area
699 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
702 struct end_io_wq
*end_io_wq
;
703 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
707 end_io_wq
->private = bio
->bi_private
;
708 end_io_wq
->end_io
= bio
->bi_end_io
;
709 end_io_wq
->info
= info
;
710 end_io_wq
->error
= 0;
711 end_io_wq
->bio
= bio
;
712 end_io_wq
->metadata
= metadata
;
714 bio
->bi_private
= end_io_wq
;
715 bio
->bi_end_io
= end_workqueue_bio
;
719 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
721 unsigned long limit
= min_t(unsigned long,
722 info
->workers
.max_workers
,
723 info
->fs_devices
->open_devices
);
727 static void run_one_async_start(struct btrfs_work
*work
)
729 struct async_submit_bio
*async
;
732 async
= container_of(work
, struct async_submit_bio
, work
);
733 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
734 async
->mirror_num
, async
->bio_flags
,
740 static void run_one_async_done(struct btrfs_work
*work
)
742 struct btrfs_fs_info
*fs_info
;
743 struct async_submit_bio
*async
;
746 async
= container_of(work
, struct async_submit_bio
, work
);
747 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
749 limit
= btrfs_async_submit_limit(fs_info
);
750 limit
= limit
* 2 / 3;
752 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
753 waitqueue_active(&fs_info
->async_submit_wait
))
754 wake_up(&fs_info
->async_submit_wait
);
756 /* If an error occured we just want to clean up the bio and move on */
758 bio_endio(async
->bio
, async
->error
);
762 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
763 async
->mirror_num
, async
->bio_flags
,
767 static void run_one_async_free(struct btrfs_work
*work
)
769 struct async_submit_bio
*async
;
771 async
= container_of(work
, struct async_submit_bio
, work
);
775 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
776 int rw
, struct bio
*bio
, int mirror_num
,
777 unsigned long bio_flags
,
779 extent_submit_bio_hook_t
*submit_bio_start
,
780 extent_submit_bio_hook_t
*submit_bio_done
)
782 struct async_submit_bio
*async
;
784 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
788 async
->inode
= inode
;
791 async
->mirror_num
= mirror_num
;
792 async
->submit_bio_start
= submit_bio_start
;
793 async
->submit_bio_done
= submit_bio_done
;
795 async
->work
.func
= run_one_async_start
;
796 async
->work
.ordered_func
= run_one_async_done
;
797 async
->work
.ordered_free
= run_one_async_free
;
799 async
->work
.flags
= 0;
800 async
->bio_flags
= bio_flags
;
801 async
->bio_offset
= bio_offset
;
805 atomic_inc(&fs_info
->nr_async_submits
);
808 btrfs_set_work_high_prio(&async
->work
);
810 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
812 while (atomic_read(&fs_info
->async_submit_draining
) &&
813 atomic_read(&fs_info
->nr_async_submits
)) {
814 wait_event(fs_info
->async_submit_wait
,
815 (atomic_read(&fs_info
->nr_async_submits
) == 0));
821 static int btree_csum_one_bio(struct bio
*bio
)
823 struct bio_vec
*bvec
= bio
->bi_io_vec
;
825 struct btrfs_root
*root
;
828 WARN_ON(bio
->bi_vcnt
<= 0);
829 while (bio_index
< bio
->bi_vcnt
) {
830 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
831 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
840 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
841 struct bio
*bio
, int mirror_num
,
842 unsigned long bio_flags
,
846 * when we're called for a write, we're already in the async
847 * submission context. Just jump into btrfs_map_bio
849 return btree_csum_one_bio(bio
);
852 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
853 int mirror_num
, unsigned long bio_flags
,
859 * when we're called for a write, we're already in the async
860 * submission context. Just jump into btrfs_map_bio
862 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
868 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
870 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
879 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
880 int mirror_num
, unsigned long bio_flags
,
883 int async
= check_async_write(inode
, bio_flags
);
886 if (!(rw
& REQ_WRITE
)) {
888 * called for a read, do the setup so that checksum validation
889 * can happen in the async kernel threads
891 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
895 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
898 ret
= btree_csum_one_bio(bio
);
901 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
905 * kthread helpers are used to submit writes so that
906 * checksumming can happen in parallel across all CPUs
908 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
909 inode
, rw
, bio
, mirror_num
, 0,
911 __btree_submit_bio_start
,
912 __btree_submit_bio_done
);
922 #ifdef CONFIG_MIGRATION
923 static int btree_migratepage(struct address_space
*mapping
,
924 struct page
*newpage
, struct page
*page
,
925 enum migrate_mode mode
)
928 * we can't safely write a btree page from here,
929 * we haven't done the locking hook
934 * Buffers may be managed in a filesystem specific way.
935 * We must have no buffers or drop them.
937 if (page_has_private(page
) &&
938 !try_to_release_page(page
, GFP_KERNEL
))
940 return migrate_page(mapping
, newpage
, page
, mode
);
945 static int btree_writepages(struct address_space
*mapping
,
946 struct writeback_control
*wbc
)
948 struct extent_io_tree
*tree
;
949 struct btrfs_fs_info
*fs_info
;
952 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
953 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
955 if (wbc
->for_kupdate
)
958 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
959 /* this is a bit racy, but that's ok */
960 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
961 BTRFS_DIRTY_METADATA_THRESH
);
965 return btree_write_cache_pages(mapping
, wbc
);
968 static int btree_readpage(struct file
*file
, struct page
*page
)
970 struct extent_io_tree
*tree
;
971 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
972 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
975 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
977 if (PageWriteback(page
) || PageDirty(page
))
980 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
981 * slab allocation from alloc_extent_state down the callchain where
982 * it'd hit a BUG_ON as those flags are not allowed.
984 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
986 return try_release_extent_buffer(page
, gfp_flags
);
989 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
991 struct extent_io_tree
*tree
;
992 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
993 extent_invalidatepage(tree
, page
, offset
);
994 btree_releasepage(page
, GFP_NOFS
);
995 if (PagePrivate(page
)) {
996 printk(KERN_WARNING
"btrfs warning page private not zero "
997 "on page %llu\n", (unsigned long long)page_offset(page
));
998 ClearPagePrivate(page
);
999 set_page_private(page
, 0);
1000 page_cache_release(page
);
1004 static int btree_set_page_dirty(struct page
*page
)
1007 struct extent_buffer
*eb
;
1009 BUG_ON(!PagePrivate(page
));
1010 eb
= (struct extent_buffer
*)page
->private;
1012 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1013 BUG_ON(!atomic_read(&eb
->refs
));
1014 btrfs_assert_tree_locked(eb
);
1016 return __set_page_dirty_nobuffers(page
);
1019 static const struct address_space_operations btree_aops
= {
1020 .readpage
= btree_readpage
,
1021 .writepages
= btree_writepages
,
1022 .releasepage
= btree_releasepage
,
1023 .invalidatepage
= btree_invalidatepage
,
1024 #ifdef CONFIG_MIGRATION
1025 .migratepage
= btree_migratepage
,
1027 .set_page_dirty
= btree_set_page_dirty
,
1030 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1033 struct extent_buffer
*buf
= NULL
;
1034 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1037 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1040 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1041 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1042 free_extent_buffer(buf
);
1046 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1047 int mirror_num
, struct extent_buffer
**eb
)
1049 struct extent_buffer
*buf
= NULL
;
1050 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1051 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1054 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1058 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1060 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1061 btree_get_extent
, mirror_num
);
1063 free_extent_buffer(buf
);
1067 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1068 free_extent_buffer(buf
);
1070 } else if (extent_buffer_uptodate(buf
)) {
1073 free_extent_buffer(buf
);
1078 struct extent_buffer
*btrfs_find_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
;
1083 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1088 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1089 u64 bytenr
, u32 blocksize
)
1091 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1092 struct extent_buffer
*eb
;
1094 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1100 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1102 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1103 buf
->start
+ buf
->len
- 1);
1106 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1108 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1109 buf
->start
, buf
->start
+ buf
->len
- 1);
1112 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1113 u32 blocksize
, u64 parent_transid
)
1115 struct extent_buffer
*buf
= NULL
;
1118 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1122 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1127 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1128 struct extent_buffer
*buf
)
1130 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1132 if (btrfs_header_generation(buf
) ==
1133 fs_info
->running_transaction
->transid
) {
1134 btrfs_assert_tree_locked(buf
);
1136 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1137 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1139 fs_info
->dirty_metadata_batch
);
1140 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1141 btrfs_set_lock_blocking(buf
);
1142 clear_extent_buffer_dirty(buf
);
1147 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1148 u32 stripesize
, struct btrfs_root
*root
,
1149 struct btrfs_fs_info
*fs_info
,
1153 root
->commit_root
= NULL
;
1154 root
->sectorsize
= sectorsize
;
1155 root
->nodesize
= nodesize
;
1156 root
->leafsize
= leafsize
;
1157 root
->stripesize
= stripesize
;
1159 root
->track_dirty
= 0;
1161 root
->orphan_item_inserted
= 0;
1162 root
->orphan_cleanup_state
= 0;
1164 root
->objectid
= objectid
;
1165 root
->last_trans
= 0;
1166 root
->highest_objectid
= 0;
1168 root
->inode_tree
= RB_ROOT
;
1169 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1170 root
->block_rsv
= NULL
;
1171 root
->orphan_block_rsv
= NULL
;
1173 INIT_LIST_HEAD(&root
->dirty_list
);
1174 INIT_LIST_HEAD(&root
->root_list
);
1175 INIT_LIST_HEAD(&root
->logged_list
[0]);
1176 INIT_LIST_HEAD(&root
->logged_list
[1]);
1177 spin_lock_init(&root
->orphan_lock
);
1178 spin_lock_init(&root
->inode_lock
);
1179 spin_lock_init(&root
->accounting_lock
);
1180 spin_lock_init(&root
->log_extents_lock
[0]);
1181 spin_lock_init(&root
->log_extents_lock
[1]);
1182 mutex_init(&root
->objectid_mutex
);
1183 mutex_init(&root
->log_mutex
);
1184 init_waitqueue_head(&root
->log_writer_wait
);
1185 init_waitqueue_head(&root
->log_commit_wait
[0]);
1186 init_waitqueue_head(&root
->log_commit_wait
[1]);
1187 atomic_set(&root
->log_commit
[0], 0);
1188 atomic_set(&root
->log_commit
[1], 0);
1189 atomic_set(&root
->log_writers
, 0);
1190 atomic_set(&root
->log_batch
, 0);
1191 atomic_set(&root
->orphan_inodes
, 0);
1192 root
->log_transid
= 0;
1193 root
->last_log_commit
= 0;
1194 extent_io_tree_init(&root
->dirty_log_pages
,
1195 fs_info
->btree_inode
->i_mapping
);
1197 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1198 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1199 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1200 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1201 root
->defrag_trans_start
= fs_info
->generation
;
1202 init_completion(&root
->kobj_unregister
);
1203 root
->defrag_running
= 0;
1204 root
->root_key
.objectid
= objectid
;
1207 spin_lock_init(&root
->root_item_lock
);
1210 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1211 struct btrfs_fs_info
*fs_info
,
1213 struct btrfs_root
*root
)
1219 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1220 tree_root
->sectorsize
, tree_root
->stripesize
,
1221 root
, fs_info
, objectid
);
1222 ret
= btrfs_find_last_root(tree_root
, objectid
,
1223 &root
->root_item
, &root
->root_key
);
1229 generation
= btrfs_root_generation(&root
->root_item
);
1230 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1231 root
->commit_root
= NULL
;
1232 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1233 blocksize
, generation
);
1234 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1235 free_extent_buffer(root
->node
);
1239 root
->commit_root
= btrfs_root_node(root
);
1243 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1245 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1247 root
->fs_info
= fs_info
;
1251 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1252 struct btrfs_fs_info
*fs_info
,
1255 struct extent_buffer
*leaf
;
1256 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1257 struct btrfs_root
*root
;
1258 struct btrfs_key key
;
1262 root
= btrfs_alloc_root(fs_info
);
1264 return ERR_PTR(-ENOMEM
);
1266 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1267 tree_root
->sectorsize
, tree_root
->stripesize
,
1268 root
, fs_info
, objectid
);
1269 root
->root_key
.objectid
= objectid
;
1270 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1271 root
->root_key
.offset
= 0;
1273 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1274 0, objectid
, NULL
, 0, 0, 0);
1276 ret
= PTR_ERR(leaf
);
1280 bytenr
= leaf
->start
;
1281 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1282 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1283 btrfs_set_header_generation(leaf
, trans
->transid
);
1284 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1285 btrfs_set_header_owner(leaf
, objectid
);
1288 write_extent_buffer(leaf
, fs_info
->fsid
,
1289 (unsigned long)btrfs_header_fsid(leaf
),
1291 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1292 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1294 btrfs_mark_buffer_dirty(leaf
);
1296 root
->commit_root
= btrfs_root_node(root
);
1297 root
->track_dirty
= 1;
1300 root
->root_item
.flags
= 0;
1301 root
->root_item
.byte_limit
= 0;
1302 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1303 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1304 btrfs_set_root_level(&root
->root_item
, 0);
1305 btrfs_set_root_refs(&root
->root_item
, 1);
1306 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1307 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1308 btrfs_set_root_dirid(&root
->root_item
, 0);
1309 root
->root_item
.drop_level
= 0;
1311 key
.objectid
= objectid
;
1312 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1314 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1318 btrfs_tree_unlock(leaf
);
1322 return ERR_PTR(ret
);
1327 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1328 struct btrfs_fs_info
*fs_info
)
1330 struct btrfs_root
*root
;
1331 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1332 struct extent_buffer
*leaf
;
1334 root
= btrfs_alloc_root(fs_info
);
1336 return ERR_PTR(-ENOMEM
);
1338 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1339 tree_root
->sectorsize
, tree_root
->stripesize
,
1340 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1342 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1343 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1344 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1346 * log trees do not get reference counted because they go away
1347 * before a real commit is actually done. They do store pointers
1348 * to file data extents, and those reference counts still get
1349 * updated (along with back refs to the log tree).
1353 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1354 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1358 return ERR_CAST(leaf
);
1361 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1362 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1363 btrfs_set_header_generation(leaf
, trans
->transid
);
1364 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1365 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1368 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1369 (unsigned long)btrfs_header_fsid(root
->node
),
1371 btrfs_mark_buffer_dirty(root
->node
);
1372 btrfs_tree_unlock(root
->node
);
1376 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1377 struct btrfs_fs_info
*fs_info
)
1379 struct btrfs_root
*log_root
;
1381 log_root
= alloc_log_tree(trans
, fs_info
);
1382 if (IS_ERR(log_root
))
1383 return PTR_ERR(log_root
);
1384 WARN_ON(fs_info
->log_root_tree
);
1385 fs_info
->log_root_tree
= log_root
;
1389 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1390 struct btrfs_root
*root
)
1392 struct btrfs_root
*log_root
;
1393 struct btrfs_inode_item
*inode_item
;
1395 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1396 if (IS_ERR(log_root
))
1397 return PTR_ERR(log_root
);
1399 log_root
->last_trans
= trans
->transid
;
1400 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1402 inode_item
= &log_root
->root_item
.inode
;
1403 inode_item
->generation
= cpu_to_le64(1);
1404 inode_item
->size
= cpu_to_le64(3);
1405 inode_item
->nlink
= cpu_to_le32(1);
1406 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1407 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1409 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1411 WARN_ON(root
->log_root
);
1412 root
->log_root
= log_root
;
1413 root
->log_transid
= 0;
1414 root
->last_log_commit
= 0;
1418 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1419 struct btrfs_key
*location
)
1421 struct btrfs_root
*root
;
1422 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1423 struct btrfs_path
*path
;
1424 struct extent_buffer
*l
;
1430 root
= btrfs_alloc_root(fs_info
);
1432 return ERR_PTR(-ENOMEM
);
1433 if (location
->offset
== (u64
)-1) {
1434 ret
= find_and_setup_root(tree_root
, fs_info
,
1435 location
->objectid
, root
);
1438 return ERR_PTR(ret
);
1443 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1444 tree_root
->sectorsize
, tree_root
->stripesize
,
1445 root
, fs_info
, location
->objectid
);
1447 path
= btrfs_alloc_path();
1450 return ERR_PTR(-ENOMEM
);
1452 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1455 slot
= path
->slots
[0];
1456 btrfs_read_root_item(tree_root
, l
, slot
, &root
->root_item
);
1457 memcpy(&root
->root_key
, location
, sizeof(*location
));
1459 btrfs_free_path(path
);
1464 return ERR_PTR(ret
);
1467 generation
= btrfs_root_generation(&root
->root_item
);
1468 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1469 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1470 blocksize
, generation
);
1471 root
->commit_root
= btrfs_root_node(root
);
1472 BUG_ON(!root
->node
); /* -ENOMEM */
1474 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1476 btrfs_check_and_init_root_item(&root
->root_item
);
1482 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1483 struct btrfs_key
*location
)
1485 struct btrfs_root
*root
;
1488 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1489 return fs_info
->tree_root
;
1490 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1491 return fs_info
->extent_root
;
1492 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1493 return fs_info
->chunk_root
;
1494 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1495 return fs_info
->dev_root
;
1496 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1497 return fs_info
->csum_root
;
1498 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1499 return fs_info
->quota_root
? fs_info
->quota_root
:
1502 spin_lock(&fs_info
->fs_roots_radix_lock
);
1503 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1504 (unsigned long)location
->objectid
);
1505 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1509 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1513 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1514 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1516 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1521 btrfs_init_free_ino_ctl(root
);
1522 mutex_init(&root
->fs_commit_mutex
);
1523 spin_lock_init(&root
->cache_lock
);
1524 init_waitqueue_head(&root
->cache_wait
);
1526 ret
= get_anon_bdev(&root
->anon_dev
);
1530 if (btrfs_root_refs(&root
->root_item
) == 0) {
1535 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1539 root
->orphan_item_inserted
= 1;
1541 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1545 spin_lock(&fs_info
->fs_roots_radix_lock
);
1546 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1547 (unsigned long)root
->root_key
.objectid
,
1552 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1553 radix_tree_preload_end();
1555 if (ret
== -EEXIST
) {
1562 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1563 root
->root_key
.objectid
);
1568 return ERR_PTR(ret
);
1571 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1573 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1575 struct btrfs_device
*device
;
1576 struct backing_dev_info
*bdi
;
1579 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1582 bdi
= blk_get_backing_dev_info(device
->bdev
);
1583 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1593 * If this fails, caller must call bdi_destroy() to get rid of the
1596 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1600 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1601 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1605 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1606 bdi
->congested_fn
= btrfs_congested_fn
;
1607 bdi
->congested_data
= info
;
1612 * called by the kthread helper functions to finally call the bio end_io
1613 * functions. This is where read checksum verification actually happens
1615 static void end_workqueue_fn(struct btrfs_work
*work
)
1618 struct end_io_wq
*end_io_wq
;
1619 struct btrfs_fs_info
*fs_info
;
1622 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1623 bio
= end_io_wq
->bio
;
1624 fs_info
= end_io_wq
->info
;
1626 error
= end_io_wq
->error
;
1627 bio
->bi_private
= end_io_wq
->private;
1628 bio
->bi_end_io
= end_io_wq
->end_io
;
1630 bio_endio(bio
, error
);
1633 static int cleaner_kthread(void *arg
)
1635 struct btrfs_root
*root
= arg
;
1638 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1639 mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1640 btrfs_run_delayed_iputs(root
);
1641 btrfs_clean_old_snapshots(root
);
1642 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1643 btrfs_run_defrag_inodes(root
->fs_info
);
1646 if (!try_to_freeze()) {
1647 set_current_state(TASK_INTERRUPTIBLE
);
1648 if (!kthread_should_stop())
1650 __set_current_state(TASK_RUNNING
);
1652 } while (!kthread_should_stop());
1656 static int transaction_kthread(void *arg
)
1658 struct btrfs_root
*root
= arg
;
1659 struct btrfs_trans_handle
*trans
;
1660 struct btrfs_transaction
*cur
;
1663 unsigned long delay
;
1667 cannot_commit
= false;
1669 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1671 spin_lock(&root
->fs_info
->trans_lock
);
1672 cur
= root
->fs_info
->running_transaction
;
1674 spin_unlock(&root
->fs_info
->trans_lock
);
1678 now
= get_seconds();
1679 if (!cur
->blocked
&&
1680 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1681 spin_unlock(&root
->fs_info
->trans_lock
);
1685 transid
= cur
->transid
;
1686 spin_unlock(&root
->fs_info
->trans_lock
);
1688 /* If the file system is aborted, this will always fail. */
1689 trans
= btrfs_attach_transaction(root
);
1690 if (IS_ERR(trans
)) {
1691 if (PTR_ERR(trans
) != -ENOENT
)
1692 cannot_commit
= true;
1695 if (transid
== trans
->transid
) {
1696 btrfs_commit_transaction(trans
, root
);
1698 btrfs_end_transaction(trans
, root
);
1701 wake_up_process(root
->fs_info
->cleaner_kthread
);
1702 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1704 if (!try_to_freeze()) {
1705 set_current_state(TASK_INTERRUPTIBLE
);
1706 if (!kthread_should_stop() &&
1707 (!btrfs_transaction_blocked(root
->fs_info
) ||
1709 schedule_timeout(delay
);
1710 __set_current_state(TASK_RUNNING
);
1712 } while (!kthread_should_stop());
1717 * this will find the highest generation in the array of
1718 * root backups. The index of the highest array is returned,
1719 * or -1 if we can't find anything.
1721 * We check to make sure the array is valid by comparing the
1722 * generation of the latest root in the array with the generation
1723 * in the super block. If they don't match we pitch it.
1725 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1728 int newest_index
= -1;
1729 struct btrfs_root_backup
*root_backup
;
1732 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1733 root_backup
= info
->super_copy
->super_roots
+ i
;
1734 cur
= btrfs_backup_tree_root_gen(root_backup
);
1735 if (cur
== newest_gen
)
1739 /* check to see if we actually wrapped around */
1740 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1741 root_backup
= info
->super_copy
->super_roots
;
1742 cur
= btrfs_backup_tree_root_gen(root_backup
);
1743 if (cur
== newest_gen
)
1746 return newest_index
;
1751 * find the oldest backup so we know where to store new entries
1752 * in the backup array. This will set the backup_root_index
1753 * field in the fs_info struct
1755 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1758 int newest_index
= -1;
1760 newest_index
= find_newest_super_backup(info
, newest_gen
);
1761 /* if there was garbage in there, just move along */
1762 if (newest_index
== -1) {
1763 info
->backup_root_index
= 0;
1765 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1770 * copy all the root pointers into the super backup array.
1771 * this will bump the backup pointer by one when it is
1774 static void backup_super_roots(struct btrfs_fs_info
*info
)
1777 struct btrfs_root_backup
*root_backup
;
1780 next_backup
= info
->backup_root_index
;
1781 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1782 BTRFS_NUM_BACKUP_ROOTS
;
1785 * just overwrite the last backup if we're at the same generation
1786 * this happens only at umount
1788 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1789 if (btrfs_backup_tree_root_gen(root_backup
) ==
1790 btrfs_header_generation(info
->tree_root
->node
))
1791 next_backup
= last_backup
;
1793 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1796 * make sure all of our padding and empty slots get zero filled
1797 * regardless of which ones we use today
1799 memset(root_backup
, 0, sizeof(*root_backup
));
1801 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1803 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1804 btrfs_set_backup_tree_root_gen(root_backup
,
1805 btrfs_header_generation(info
->tree_root
->node
));
1807 btrfs_set_backup_tree_root_level(root_backup
,
1808 btrfs_header_level(info
->tree_root
->node
));
1810 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1811 btrfs_set_backup_chunk_root_gen(root_backup
,
1812 btrfs_header_generation(info
->chunk_root
->node
));
1813 btrfs_set_backup_chunk_root_level(root_backup
,
1814 btrfs_header_level(info
->chunk_root
->node
));
1816 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1817 btrfs_set_backup_extent_root_gen(root_backup
,
1818 btrfs_header_generation(info
->extent_root
->node
));
1819 btrfs_set_backup_extent_root_level(root_backup
,
1820 btrfs_header_level(info
->extent_root
->node
));
1823 * we might commit during log recovery, which happens before we set
1824 * the fs_root. Make sure it is valid before we fill it in.
1826 if (info
->fs_root
&& info
->fs_root
->node
) {
1827 btrfs_set_backup_fs_root(root_backup
,
1828 info
->fs_root
->node
->start
);
1829 btrfs_set_backup_fs_root_gen(root_backup
,
1830 btrfs_header_generation(info
->fs_root
->node
));
1831 btrfs_set_backup_fs_root_level(root_backup
,
1832 btrfs_header_level(info
->fs_root
->node
));
1835 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1836 btrfs_set_backup_dev_root_gen(root_backup
,
1837 btrfs_header_generation(info
->dev_root
->node
));
1838 btrfs_set_backup_dev_root_level(root_backup
,
1839 btrfs_header_level(info
->dev_root
->node
));
1841 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1842 btrfs_set_backup_csum_root_gen(root_backup
,
1843 btrfs_header_generation(info
->csum_root
->node
));
1844 btrfs_set_backup_csum_root_level(root_backup
,
1845 btrfs_header_level(info
->csum_root
->node
));
1847 btrfs_set_backup_total_bytes(root_backup
,
1848 btrfs_super_total_bytes(info
->super_copy
));
1849 btrfs_set_backup_bytes_used(root_backup
,
1850 btrfs_super_bytes_used(info
->super_copy
));
1851 btrfs_set_backup_num_devices(root_backup
,
1852 btrfs_super_num_devices(info
->super_copy
));
1855 * if we don't copy this out to the super_copy, it won't get remembered
1856 * for the next commit
1858 memcpy(&info
->super_copy
->super_roots
,
1859 &info
->super_for_commit
->super_roots
,
1860 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1864 * this copies info out of the root backup array and back into
1865 * the in-memory super block. It is meant to help iterate through
1866 * the array, so you send it the number of backups you've already
1867 * tried and the last backup index you used.
1869 * this returns -1 when it has tried all the backups
1871 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1872 struct btrfs_super_block
*super
,
1873 int *num_backups_tried
, int *backup_index
)
1875 struct btrfs_root_backup
*root_backup
;
1876 int newest
= *backup_index
;
1878 if (*num_backups_tried
== 0) {
1879 u64 gen
= btrfs_super_generation(super
);
1881 newest
= find_newest_super_backup(info
, gen
);
1885 *backup_index
= newest
;
1886 *num_backups_tried
= 1;
1887 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1888 /* we've tried all the backups, all done */
1891 /* jump to the next oldest backup */
1892 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1893 BTRFS_NUM_BACKUP_ROOTS
;
1894 *backup_index
= newest
;
1895 *num_backups_tried
+= 1;
1897 root_backup
= super
->super_roots
+ newest
;
1899 btrfs_set_super_generation(super
,
1900 btrfs_backup_tree_root_gen(root_backup
));
1901 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1902 btrfs_set_super_root_level(super
,
1903 btrfs_backup_tree_root_level(root_backup
));
1904 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1907 * fixme: the total bytes and num_devices need to match or we should
1910 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1911 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1915 /* helper to cleanup tree roots */
1916 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1918 free_extent_buffer(info
->tree_root
->node
);
1919 free_extent_buffer(info
->tree_root
->commit_root
);
1920 free_extent_buffer(info
->dev_root
->node
);
1921 free_extent_buffer(info
->dev_root
->commit_root
);
1922 free_extent_buffer(info
->extent_root
->node
);
1923 free_extent_buffer(info
->extent_root
->commit_root
);
1924 free_extent_buffer(info
->csum_root
->node
);
1925 free_extent_buffer(info
->csum_root
->commit_root
);
1926 if (info
->quota_root
) {
1927 free_extent_buffer(info
->quota_root
->node
);
1928 free_extent_buffer(info
->quota_root
->commit_root
);
1931 info
->tree_root
->node
= NULL
;
1932 info
->tree_root
->commit_root
= NULL
;
1933 info
->dev_root
->node
= NULL
;
1934 info
->dev_root
->commit_root
= NULL
;
1935 info
->extent_root
->node
= NULL
;
1936 info
->extent_root
->commit_root
= NULL
;
1937 info
->csum_root
->node
= NULL
;
1938 info
->csum_root
->commit_root
= NULL
;
1939 if (info
->quota_root
) {
1940 info
->quota_root
->node
= NULL
;
1941 info
->quota_root
->commit_root
= NULL
;
1945 free_extent_buffer(info
->chunk_root
->node
);
1946 free_extent_buffer(info
->chunk_root
->commit_root
);
1947 info
->chunk_root
->node
= NULL
;
1948 info
->chunk_root
->commit_root
= NULL
;
1953 int open_ctree(struct super_block
*sb
,
1954 struct btrfs_fs_devices
*fs_devices
,
1964 struct btrfs_key location
;
1965 struct buffer_head
*bh
;
1966 struct btrfs_super_block
*disk_super
;
1967 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
1968 struct btrfs_root
*tree_root
;
1969 struct btrfs_root
*extent_root
;
1970 struct btrfs_root
*csum_root
;
1971 struct btrfs_root
*chunk_root
;
1972 struct btrfs_root
*dev_root
;
1973 struct btrfs_root
*quota_root
;
1974 struct btrfs_root
*log_tree_root
;
1977 int num_backups_tried
= 0;
1978 int backup_index
= 0;
1980 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
1981 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
1982 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
1983 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
1984 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
1985 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
1987 if (!tree_root
|| !extent_root
|| !csum_root
||
1988 !chunk_root
|| !dev_root
|| !quota_root
) {
1993 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
1999 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2005 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2010 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2011 (1 + ilog2(nr_cpu_ids
));
2013 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2016 goto fail_dirty_metadata_bytes
;
2019 fs_info
->btree_inode
= new_inode(sb
);
2020 if (!fs_info
->btree_inode
) {
2022 goto fail_delalloc_bytes
;
2025 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2027 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2028 INIT_LIST_HEAD(&fs_info
->trans_list
);
2029 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2030 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2031 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2032 INIT_LIST_HEAD(&fs_info
->ordered_operations
);
2033 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2034 spin_lock_init(&fs_info
->delalloc_lock
);
2035 spin_lock_init(&fs_info
->trans_lock
);
2036 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2037 spin_lock_init(&fs_info
->delayed_iput_lock
);
2038 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2039 spin_lock_init(&fs_info
->free_chunk_lock
);
2040 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2041 rwlock_init(&fs_info
->tree_mod_log_lock
);
2042 mutex_init(&fs_info
->reloc_mutex
);
2043 seqlock_init(&fs_info
->profiles_lock
);
2045 init_completion(&fs_info
->kobj_unregister
);
2046 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2047 INIT_LIST_HEAD(&fs_info
->space_info
);
2048 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2049 btrfs_mapping_init(&fs_info
->mapping_tree
);
2050 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2051 BTRFS_BLOCK_RSV_GLOBAL
);
2052 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2053 BTRFS_BLOCK_RSV_DELALLOC
);
2054 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2055 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2056 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2057 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2058 BTRFS_BLOCK_RSV_DELOPS
);
2059 atomic_set(&fs_info
->nr_async_submits
, 0);
2060 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2061 atomic_set(&fs_info
->async_submit_draining
, 0);
2062 atomic_set(&fs_info
->nr_async_bios
, 0);
2063 atomic_set(&fs_info
->defrag_running
, 0);
2064 atomic_set(&fs_info
->tree_mod_seq
, 0);
2066 fs_info
->max_inline
= 8192 * 1024;
2067 fs_info
->metadata_ratio
= 0;
2068 fs_info
->defrag_inodes
= RB_ROOT
;
2069 fs_info
->trans_no_join
= 0;
2070 fs_info
->free_chunk_space
= 0;
2071 fs_info
->tree_mod_log
= RB_ROOT
;
2073 /* readahead state */
2074 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2075 spin_lock_init(&fs_info
->reada_lock
);
2077 fs_info
->thread_pool_size
= min_t(unsigned long,
2078 num_online_cpus() + 2, 8);
2080 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2081 spin_lock_init(&fs_info
->ordered_extent_lock
);
2082 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2084 if (!fs_info
->delayed_root
) {
2088 btrfs_init_delayed_root(fs_info
->delayed_root
);
2090 mutex_init(&fs_info
->scrub_lock
);
2091 atomic_set(&fs_info
->scrubs_running
, 0);
2092 atomic_set(&fs_info
->scrub_pause_req
, 0);
2093 atomic_set(&fs_info
->scrubs_paused
, 0);
2094 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2095 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2096 init_rwsem(&fs_info
->scrub_super_lock
);
2097 fs_info
->scrub_workers_refcnt
= 0;
2098 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2099 fs_info
->check_integrity_print_mask
= 0;
2102 spin_lock_init(&fs_info
->balance_lock
);
2103 mutex_init(&fs_info
->balance_mutex
);
2104 atomic_set(&fs_info
->balance_running
, 0);
2105 atomic_set(&fs_info
->balance_pause_req
, 0);
2106 atomic_set(&fs_info
->balance_cancel_req
, 0);
2107 fs_info
->balance_ctl
= NULL
;
2108 init_waitqueue_head(&fs_info
->balance_wait_q
);
2110 sb
->s_blocksize
= 4096;
2111 sb
->s_blocksize_bits
= blksize_bits(4096);
2112 sb
->s_bdi
= &fs_info
->bdi
;
2114 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2115 set_nlink(fs_info
->btree_inode
, 1);
2117 * we set the i_size on the btree inode to the max possible int.
2118 * the real end of the address space is determined by all of
2119 * the devices in the system
2121 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2122 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2123 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2125 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2126 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2127 fs_info
->btree_inode
->i_mapping
);
2128 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2129 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2131 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2133 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2134 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2135 sizeof(struct btrfs_key
));
2136 set_bit(BTRFS_INODE_DUMMY
,
2137 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2138 insert_inode_hash(fs_info
->btree_inode
);
2140 spin_lock_init(&fs_info
->block_group_cache_lock
);
2141 fs_info
->block_group_cache_tree
= RB_ROOT
;
2142 fs_info
->first_logical_byte
= (u64
)-1;
2144 extent_io_tree_init(&fs_info
->freed_extents
[0],
2145 fs_info
->btree_inode
->i_mapping
);
2146 extent_io_tree_init(&fs_info
->freed_extents
[1],
2147 fs_info
->btree_inode
->i_mapping
);
2148 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2149 fs_info
->do_barriers
= 1;
2152 mutex_init(&fs_info
->ordered_operations_mutex
);
2153 mutex_init(&fs_info
->tree_log_mutex
);
2154 mutex_init(&fs_info
->chunk_mutex
);
2155 mutex_init(&fs_info
->transaction_kthread_mutex
);
2156 mutex_init(&fs_info
->cleaner_mutex
);
2157 mutex_init(&fs_info
->volume_mutex
);
2158 init_rwsem(&fs_info
->extent_commit_sem
);
2159 init_rwsem(&fs_info
->cleanup_work_sem
);
2160 init_rwsem(&fs_info
->subvol_sem
);
2161 fs_info
->dev_replace
.lock_owner
= 0;
2162 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2163 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2164 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2165 mutex_init(&fs_info
->dev_replace
.lock
);
2167 spin_lock_init(&fs_info
->qgroup_lock
);
2168 fs_info
->qgroup_tree
= RB_ROOT
;
2169 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2170 fs_info
->qgroup_seq
= 1;
2171 fs_info
->quota_enabled
= 0;
2172 fs_info
->pending_quota_state
= 0;
2174 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2175 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2177 init_waitqueue_head(&fs_info
->transaction_throttle
);
2178 init_waitqueue_head(&fs_info
->transaction_wait
);
2179 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2180 init_waitqueue_head(&fs_info
->async_submit_wait
);
2182 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2183 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2185 invalidate_bdev(fs_devices
->latest_bdev
);
2186 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2192 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2193 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2194 sizeof(*fs_info
->super_for_commit
));
2197 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2199 disk_super
= fs_info
->super_copy
;
2200 if (!btrfs_super_root(disk_super
))
2203 /* check FS state, whether FS is broken. */
2204 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2205 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2207 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2209 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2215 * run through our array of backup supers and setup
2216 * our ring pointer to the oldest one
2218 generation
= btrfs_super_generation(disk_super
);
2219 find_oldest_super_backup(fs_info
, generation
);
2222 * In the long term, we'll store the compression type in the super
2223 * block, and it'll be used for per file compression control.
2225 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2227 ret
= btrfs_parse_options(tree_root
, options
);
2233 features
= btrfs_super_incompat_flags(disk_super
) &
2234 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2236 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2237 "unsupported optional features (%Lx).\n",
2238 (unsigned long long)features
);
2243 if (btrfs_super_leafsize(disk_super
) !=
2244 btrfs_super_nodesize(disk_super
)) {
2245 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2246 "blocksizes don't match. node %d leaf %d\n",
2247 btrfs_super_nodesize(disk_super
),
2248 btrfs_super_leafsize(disk_super
));
2252 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2253 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2254 "blocksize (%d) was too large\n",
2255 btrfs_super_leafsize(disk_super
));
2260 features
= btrfs_super_incompat_flags(disk_super
);
2261 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2262 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2263 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2266 * flag our filesystem as having big metadata blocks if
2267 * they are bigger than the page size
2269 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2270 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2271 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2272 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2275 nodesize
= btrfs_super_nodesize(disk_super
);
2276 leafsize
= btrfs_super_leafsize(disk_super
);
2277 sectorsize
= btrfs_super_sectorsize(disk_super
);
2278 stripesize
= btrfs_super_stripesize(disk_super
);
2279 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2280 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2283 * mixed block groups end up with duplicate but slightly offset
2284 * extent buffers for the same range. It leads to corruptions
2286 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2287 (sectorsize
!= leafsize
)) {
2288 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2289 "are not allowed for mixed block groups on %s\n",
2294 btrfs_set_super_incompat_flags(disk_super
, features
);
2296 features
= btrfs_super_compat_ro_flags(disk_super
) &
2297 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2298 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2299 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2300 "unsupported option features (%Lx).\n",
2301 (unsigned long long)features
);
2306 btrfs_init_workers(&fs_info
->generic_worker
,
2307 "genwork", 1, NULL
);
2309 btrfs_init_workers(&fs_info
->workers
, "worker",
2310 fs_info
->thread_pool_size
,
2311 &fs_info
->generic_worker
);
2313 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2314 fs_info
->thread_pool_size
,
2315 &fs_info
->generic_worker
);
2317 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2318 fs_info
->thread_pool_size
,
2319 &fs_info
->generic_worker
);
2321 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2322 min_t(u64
, fs_devices
->num_devices
,
2323 fs_info
->thread_pool_size
),
2324 &fs_info
->generic_worker
);
2326 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2327 2, &fs_info
->generic_worker
);
2329 /* a higher idle thresh on the submit workers makes it much more
2330 * likely that bios will be send down in a sane order to the
2333 fs_info
->submit_workers
.idle_thresh
= 64;
2335 fs_info
->workers
.idle_thresh
= 16;
2336 fs_info
->workers
.ordered
= 1;
2338 fs_info
->delalloc_workers
.idle_thresh
= 2;
2339 fs_info
->delalloc_workers
.ordered
= 1;
2341 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2342 &fs_info
->generic_worker
);
2343 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2344 fs_info
->thread_pool_size
,
2345 &fs_info
->generic_worker
);
2346 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2347 fs_info
->thread_pool_size
,
2348 &fs_info
->generic_worker
);
2349 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2350 "endio-meta-write", fs_info
->thread_pool_size
,
2351 &fs_info
->generic_worker
);
2352 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2353 fs_info
->thread_pool_size
,
2354 &fs_info
->generic_worker
);
2355 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2356 1, &fs_info
->generic_worker
);
2357 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2358 fs_info
->thread_pool_size
,
2359 &fs_info
->generic_worker
);
2360 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2361 fs_info
->thread_pool_size
,
2362 &fs_info
->generic_worker
);
2365 * endios are largely parallel and should have a very
2368 fs_info
->endio_workers
.idle_thresh
= 4;
2369 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2371 fs_info
->endio_write_workers
.idle_thresh
= 2;
2372 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2373 fs_info
->readahead_workers
.idle_thresh
= 2;
2376 * btrfs_start_workers can really only fail because of ENOMEM so just
2377 * return -ENOMEM if any of these fail.
2379 ret
= btrfs_start_workers(&fs_info
->workers
);
2380 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2381 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2382 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2383 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2384 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2385 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2386 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2387 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2388 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2389 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2390 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2391 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2392 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2395 goto fail_sb_buffer
;
2398 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2399 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2400 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2402 tree_root
->nodesize
= nodesize
;
2403 tree_root
->leafsize
= leafsize
;
2404 tree_root
->sectorsize
= sectorsize
;
2405 tree_root
->stripesize
= stripesize
;
2407 sb
->s_blocksize
= sectorsize
;
2408 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2410 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
2411 sizeof(disk_super
->magic
))) {
2412 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2413 goto fail_sb_buffer
;
2416 if (sectorsize
!= PAGE_SIZE
) {
2417 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2418 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2419 goto fail_sb_buffer
;
2422 mutex_lock(&fs_info
->chunk_mutex
);
2423 ret
= btrfs_read_sys_array(tree_root
);
2424 mutex_unlock(&fs_info
->chunk_mutex
);
2426 printk(KERN_WARNING
"btrfs: failed to read the system "
2427 "array on %s\n", sb
->s_id
);
2428 goto fail_sb_buffer
;
2431 blocksize
= btrfs_level_size(tree_root
,
2432 btrfs_super_chunk_root_level(disk_super
));
2433 generation
= btrfs_super_chunk_root_generation(disk_super
);
2435 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2436 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2438 chunk_root
->node
= read_tree_block(chunk_root
,
2439 btrfs_super_chunk_root(disk_super
),
2440 blocksize
, generation
);
2441 BUG_ON(!chunk_root
->node
); /* -ENOMEM */
2442 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2443 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2445 goto fail_tree_roots
;
2447 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2448 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2450 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2451 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2454 ret
= btrfs_read_chunk_tree(chunk_root
);
2456 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2458 goto fail_tree_roots
;
2462 * keep the device that is marked to be the target device for the
2463 * dev_replace procedure
2465 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2467 if (!fs_devices
->latest_bdev
) {
2468 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2470 goto fail_tree_roots
;
2474 blocksize
= btrfs_level_size(tree_root
,
2475 btrfs_super_root_level(disk_super
));
2476 generation
= btrfs_super_generation(disk_super
);
2478 tree_root
->node
= read_tree_block(tree_root
,
2479 btrfs_super_root(disk_super
),
2480 blocksize
, generation
);
2481 if (!tree_root
->node
||
2482 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2483 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2486 goto recovery_tree_root
;
2489 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2490 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2492 ret
= find_and_setup_root(tree_root
, fs_info
,
2493 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2495 goto recovery_tree_root
;
2496 extent_root
->track_dirty
= 1;
2498 ret
= find_and_setup_root(tree_root
, fs_info
,
2499 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2501 goto recovery_tree_root
;
2502 dev_root
->track_dirty
= 1;
2504 ret
= find_and_setup_root(tree_root
, fs_info
,
2505 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2507 goto recovery_tree_root
;
2508 csum_root
->track_dirty
= 1;
2510 ret
= find_and_setup_root(tree_root
, fs_info
,
2511 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2514 quota_root
= fs_info
->quota_root
= NULL
;
2516 quota_root
->track_dirty
= 1;
2517 fs_info
->quota_enabled
= 1;
2518 fs_info
->pending_quota_state
= 1;
2521 fs_info
->generation
= generation
;
2522 fs_info
->last_trans_committed
= generation
;
2524 ret
= btrfs_recover_balance(fs_info
);
2526 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2527 goto fail_block_groups
;
2530 ret
= btrfs_init_dev_stats(fs_info
);
2532 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2534 goto fail_block_groups
;
2537 ret
= btrfs_init_dev_replace(fs_info
);
2539 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2540 goto fail_block_groups
;
2543 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2545 ret
= btrfs_init_space_info(fs_info
);
2547 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2548 goto fail_block_groups
;
2551 ret
= btrfs_read_block_groups(extent_root
);
2553 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2554 goto fail_block_groups
;
2556 fs_info
->num_tolerated_disk_barrier_failures
=
2557 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2558 if (fs_info
->fs_devices
->missing_devices
>
2559 fs_info
->num_tolerated_disk_barrier_failures
&&
2560 !(sb
->s_flags
& MS_RDONLY
)) {
2562 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2563 goto fail_block_groups
;
2566 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2568 if (IS_ERR(fs_info
->cleaner_kthread
))
2569 goto fail_block_groups
;
2571 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2573 "btrfs-transaction");
2574 if (IS_ERR(fs_info
->transaction_kthread
))
2577 if (!btrfs_test_opt(tree_root
, SSD
) &&
2578 !btrfs_test_opt(tree_root
, NOSSD
) &&
2579 !fs_info
->fs_devices
->rotating
) {
2580 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2582 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2585 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2586 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2587 ret
= btrfsic_mount(tree_root
, fs_devices
,
2588 btrfs_test_opt(tree_root
,
2589 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2591 fs_info
->check_integrity_print_mask
);
2593 printk(KERN_WARNING
"btrfs: failed to initialize"
2594 " integrity check module %s\n", sb
->s_id
);
2597 ret
= btrfs_read_qgroup_config(fs_info
);
2599 goto fail_trans_kthread
;
2601 /* do not make disk changes in broken FS */
2602 if (btrfs_super_log_root(disk_super
) != 0) {
2603 u64 bytenr
= btrfs_super_log_root(disk_super
);
2605 if (fs_devices
->rw_devices
== 0) {
2606 printk(KERN_WARNING
"Btrfs log replay required "
2612 btrfs_level_size(tree_root
,
2613 btrfs_super_log_root_level(disk_super
));
2615 log_tree_root
= btrfs_alloc_root(fs_info
);
2616 if (!log_tree_root
) {
2621 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2622 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2624 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2627 /* returns with log_tree_root freed on success */
2628 ret
= btrfs_recover_log_trees(log_tree_root
);
2630 btrfs_error(tree_root
->fs_info
, ret
,
2631 "Failed to recover log tree");
2632 free_extent_buffer(log_tree_root
->node
);
2633 kfree(log_tree_root
);
2634 goto fail_trans_kthread
;
2637 if (sb
->s_flags
& MS_RDONLY
) {
2638 ret
= btrfs_commit_super(tree_root
);
2640 goto fail_trans_kthread
;
2644 ret
= btrfs_find_orphan_roots(tree_root
);
2646 goto fail_trans_kthread
;
2648 if (!(sb
->s_flags
& MS_RDONLY
)) {
2649 ret
= btrfs_cleanup_fs_roots(fs_info
);
2651 goto fail_trans_kthread
;
2653 ret
= btrfs_recover_relocation(tree_root
);
2656 "btrfs: failed to recover relocation\n");
2662 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2663 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2664 location
.offset
= (u64
)-1;
2666 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2667 if (!fs_info
->fs_root
)
2669 if (IS_ERR(fs_info
->fs_root
)) {
2670 err
= PTR_ERR(fs_info
->fs_root
);
2674 if (sb
->s_flags
& MS_RDONLY
)
2677 down_read(&fs_info
->cleanup_work_sem
);
2678 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2679 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2680 up_read(&fs_info
->cleanup_work_sem
);
2681 close_ctree(tree_root
);
2684 up_read(&fs_info
->cleanup_work_sem
);
2686 ret
= btrfs_resume_balance_async(fs_info
);
2688 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2689 close_ctree(tree_root
);
2693 ret
= btrfs_resume_dev_replace_async(fs_info
);
2695 pr_warn("btrfs: failed to resume dev_replace\n");
2696 close_ctree(tree_root
);
2703 btrfs_free_qgroup_config(fs_info
);
2705 kthread_stop(fs_info
->transaction_kthread
);
2707 kthread_stop(fs_info
->cleaner_kthread
);
2710 * make sure we're done with the btree inode before we stop our
2713 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2714 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2717 btrfs_free_block_groups(fs_info
);
2720 free_root_pointers(fs_info
, 1);
2723 btrfs_stop_workers(&fs_info
->generic_worker
);
2724 btrfs_stop_workers(&fs_info
->readahead_workers
);
2725 btrfs_stop_workers(&fs_info
->fixup_workers
);
2726 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2727 btrfs_stop_workers(&fs_info
->workers
);
2728 btrfs_stop_workers(&fs_info
->endio_workers
);
2729 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2730 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2731 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2732 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2733 btrfs_stop_workers(&fs_info
->submit_workers
);
2734 btrfs_stop_workers(&fs_info
->delayed_workers
);
2735 btrfs_stop_workers(&fs_info
->caching_workers
);
2736 btrfs_stop_workers(&fs_info
->flush_workers
);
2739 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2741 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2742 iput(fs_info
->btree_inode
);
2743 fail_delalloc_bytes
:
2744 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2745 fail_dirty_metadata_bytes
:
2746 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2748 bdi_destroy(&fs_info
->bdi
);
2750 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2752 btrfs_close_devices(fs_info
->fs_devices
);
2756 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2757 goto fail_tree_roots
;
2759 free_root_pointers(fs_info
, 0);
2761 /* don't use the log in recovery mode, it won't be valid */
2762 btrfs_set_super_log_root(disk_super
, 0);
2764 /* we can't trust the free space cache either */
2765 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2767 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2768 &num_backups_tried
, &backup_index
);
2770 goto fail_block_groups
;
2771 goto retry_root_backup
;
2774 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2777 set_buffer_uptodate(bh
);
2779 struct btrfs_device
*device
= (struct btrfs_device
*)
2782 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2783 "I/O error on %s\n",
2784 rcu_str_deref(device
->name
));
2785 /* note, we dont' set_buffer_write_io_error because we have
2786 * our own ways of dealing with the IO errors
2788 clear_buffer_uptodate(bh
);
2789 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2795 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2797 struct buffer_head
*bh
;
2798 struct buffer_head
*latest
= NULL
;
2799 struct btrfs_super_block
*super
;
2804 /* we would like to check all the supers, but that would make
2805 * a btrfs mount succeed after a mkfs from a different FS.
2806 * So, we need to add a special mount option to scan for
2807 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2809 for (i
= 0; i
< 1; i
++) {
2810 bytenr
= btrfs_sb_offset(i
);
2811 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2813 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2817 super
= (struct btrfs_super_block
*)bh
->b_data
;
2818 if (btrfs_super_bytenr(super
) != bytenr
||
2819 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
2820 sizeof(super
->magic
))) {
2825 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2828 transid
= btrfs_super_generation(super
);
2837 * this should be called twice, once with wait == 0 and
2838 * once with wait == 1. When wait == 0 is done, all the buffer heads
2839 * we write are pinned.
2841 * They are released when wait == 1 is done.
2842 * max_mirrors must be the same for both runs, and it indicates how
2843 * many supers on this one device should be written.
2845 * max_mirrors == 0 means to write them all.
2847 static int write_dev_supers(struct btrfs_device
*device
,
2848 struct btrfs_super_block
*sb
,
2849 int do_barriers
, int wait
, int max_mirrors
)
2851 struct buffer_head
*bh
;
2858 if (max_mirrors
== 0)
2859 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2861 for (i
= 0; i
< max_mirrors
; i
++) {
2862 bytenr
= btrfs_sb_offset(i
);
2863 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2867 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2868 BTRFS_SUPER_INFO_SIZE
);
2871 if (!buffer_uptodate(bh
))
2874 /* drop our reference */
2877 /* drop the reference from the wait == 0 run */
2881 btrfs_set_super_bytenr(sb
, bytenr
);
2884 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2885 BTRFS_CSUM_SIZE
, crc
,
2886 BTRFS_SUPER_INFO_SIZE
-
2888 btrfs_csum_final(crc
, sb
->csum
);
2891 * one reference for us, and we leave it for the
2894 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2895 BTRFS_SUPER_INFO_SIZE
);
2896 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2898 /* one reference for submit_bh */
2901 set_buffer_uptodate(bh
);
2903 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2904 bh
->b_private
= device
;
2908 * we fua the first super. The others we allow
2911 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
2915 return errors
< i
? 0 : -1;
2919 * endio for the write_dev_flush, this will wake anyone waiting
2920 * for the barrier when it is done
2922 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
2925 if (err
== -EOPNOTSUPP
)
2926 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
2927 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2929 if (bio
->bi_private
)
2930 complete(bio
->bi_private
);
2935 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2936 * sent down. With wait == 1, it waits for the previous flush.
2938 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2941 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
2946 if (device
->nobarriers
)
2950 bio
= device
->flush_bio
;
2954 wait_for_completion(&device
->flush_wait
);
2956 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
2957 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2958 rcu_str_deref(device
->name
));
2959 device
->nobarriers
= 1;
2960 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
2962 btrfs_dev_stat_inc_and_print(device
,
2963 BTRFS_DEV_STAT_FLUSH_ERRS
);
2966 /* drop the reference from the wait == 0 run */
2968 device
->flush_bio
= NULL
;
2974 * one reference for us, and we leave it for the
2977 device
->flush_bio
= NULL
;
2978 bio
= bio_alloc(GFP_NOFS
, 0);
2982 bio
->bi_end_io
= btrfs_end_empty_barrier
;
2983 bio
->bi_bdev
= device
->bdev
;
2984 init_completion(&device
->flush_wait
);
2985 bio
->bi_private
= &device
->flush_wait
;
2986 device
->flush_bio
= bio
;
2989 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
2995 * send an empty flush down to each device in parallel,
2996 * then wait for them
2998 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3000 struct list_head
*head
;
3001 struct btrfs_device
*dev
;
3002 int errors_send
= 0;
3003 int errors_wait
= 0;
3006 /* send down all the barriers */
3007 head
= &info
->fs_devices
->devices
;
3008 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3013 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3016 ret
= write_dev_flush(dev
, 0);
3021 /* wait for all the barriers */
3022 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3027 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3030 ret
= write_dev_flush(dev
, 1);
3034 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3035 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3040 int btrfs_calc_num_tolerated_disk_barrier_failures(
3041 struct btrfs_fs_info
*fs_info
)
3043 struct btrfs_ioctl_space_info space
;
3044 struct btrfs_space_info
*sinfo
;
3045 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3046 BTRFS_BLOCK_GROUP_SYSTEM
,
3047 BTRFS_BLOCK_GROUP_METADATA
,
3048 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3052 int num_tolerated_disk_barrier_failures
=
3053 (int)fs_info
->fs_devices
->num_devices
;
3055 for (i
= 0; i
< num_types
; i
++) {
3056 struct btrfs_space_info
*tmp
;
3060 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3061 if (tmp
->flags
== types
[i
]) {
3071 down_read(&sinfo
->groups_sem
);
3072 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3073 if (!list_empty(&sinfo
->block_groups
[c
])) {
3076 btrfs_get_block_group_info(
3077 &sinfo
->block_groups
[c
], &space
);
3078 if (space
.total_bytes
== 0 ||
3079 space
.used_bytes
== 0)
3081 flags
= space
.flags
;
3084 * 0: if dup, single or RAID0 is configured for
3085 * any of metadata, system or data, else
3086 * 1: if RAID5 is configured, or if RAID1 or
3087 * RAID10 is configured and only two mirrors
3089 * 2: if RAID6 is configured, else
3090 * num_mirrors - 1: if RAID1 or RAID10 is
3091 * configured and more than
3092 * 2 mirrors are used.
3094 if (num_tolerated_disk_barrier_failures
> 0 &&
3095 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3096 BTRFS_BLOCK_GROUP_RAID0
)) ||
3097 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3099 num_tolerated_disk_barrier_failures
= 0;
3100 else if (num_tolerated_disk_barrier_failures
> 1
3102 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3103 BTRFS_BLOCK_GROUP_RAID10
)))
3104 num_tolerated_disk_barrier_failures
= 1;
3107 up_read(&sinfo
->groups_sem
);
3110 return num_tolerated_disk_barrier_failures
;
3113 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3115 struct list_head
*head
;
3116 struct btrfs_device
*dev
;
3117 struct btrfs_super_block
*sb
;
3118 struct btrfs_dev_item
*dev_item
;
3122 int total_errors
= 0;
3125 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3126 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3127 backup_super_roots(root
->fs_info
);
3129 sb
= root
->fs_info
->super_for_commit
;
3130 dev_item
= &sb
->dev_item
;
3132 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3133 head
= &root
->fs_info
->fs_devices
->devices
;
3136 ret
= barrier_all_devices(root
->fs_info
);
3139 &root
->fs_info
->fs_devices
->device_list_mutex
);
3140 btrfs_error(root
->fs_info
, ret
,
3141 "errors while submitting device barriers.");
3146 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3151 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3154 btrfs_set_stack_device_generation(dev_item
, 0);
3155 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3156 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3157 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3158 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3159 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3160 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3161 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3162 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3163 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3165 flags
= btrfs_super_flags(sb
);
3166 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3168 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3172 if (total_errors
> max_errors
) {
3173 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3176 /* This shouldn't happen. FUA is masked off if unsupported */
3181 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3184 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3187 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3191 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3192 if (total_errors
> max_errors
) {
3193 btrfs_error(root
->fs_info
, -EIO
,
3194 "%d errors while writing supers", total_errors
);
3200 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3201 struct btrfs_root
*root
, int max_mirrors
)
3205 ret
= write_all_supers(root
, max_mirrors
);
3209 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3211 spin_lock(&fs_info
->fs_roots_radix_lock
);
3212 radix_tree_delete(&fs_info
->fs_roots_radix
,
3213 (unsigned long)root
->root_key
.objectid
);
3214 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3216 if (btrfs_root_refs(&root
->root_item
) == 0)
3217 synchronize_srcu(&fs_info
->subvol_srcu
);
3219 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3220 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3224 static void free_fs_root(struct btrfs_root
*root
)
3226 iput(root
->cache_inode
);
3227 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3229 free_anon_bdev(root
->anon_dev
);
3230 free_extent_buffer(root
->node
);
3231 free_extent_buffer(root
->commit_root
);
3232 kfree(root
->free_ino_ctl
);
3233 kfree(root
->free_ino_pinned
);
3238 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
3241 struct btrfs_root
*gang
[8];
3244 while (!list_empty(&fs_info
->dead_roots
)) {
3245 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
3246 struct btrfs_root
, root_list
);
3247 list_del(&gang
[0]->root_list
);
3249 if (gang
[0]->in_radix
) {
3250 btrfs_free_fs_root(fs_info
, gang
[0]);
3252 free_extent_buffer(gang
[0]->node
);
3253 free_extent_buffer(gang
[0]->commit_root
);
3259 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3264 for (i
= 0; i
< ret
; i
++)
3265 btrfs_free_fs_root(fs_info
, gang
[i
]);
3269 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3271 u64 root_objectid
= 0;
3272 struct btrfs_root
*gang
[8];
3277 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3278 (void **)gang
, root_objectid
,
3283 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3284 for (i
= 0; i
< ret
; i
++) {
3287 root_objectid
= gang
[i
]->root_key
.objectid
;
3288 err
= btrfs_orphan_cleanup(gang
[i
]);
3297 int btrfs_commit_super(struct btrfs_root
*root
)
3299 struct btrfs_trans_handle
*trans
;
3302 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3303 btrfs_run_delayed_iputs(root
);
3304 btrfs_clean_old_snapshots(root
);
3305 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3307 /* wait until ongoing cleanup work done */
3308 down_write(&root
->fs_info
->cleanup_work_sem
);
3309 up_write(&root
->fs_info
->cleanup_work_sem
);
3311 trans
= btrfs_join_transaction(root
);
3313 return PTR_ERR(trans
);
3314 ret
= btrfs_commit_transaction(trans
, root
);
3317 /* run commit again to drop the original snapshot */
3318 trans
= btrfs_join_transaction(root
);
3320 return PTR_ERR(trans
);
3321 ret
= btrfs_commit_transaction(trans
, root
);
3324 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3326 btrfs_error(root
->fs_info
, ret
,
3327 "Failed to sync btree inode to disk.");
3331 ret
= write_ctree_super(NULL
, root
, 0);
3335 int close_ctree(struct btrfs_root
*root
)
3337 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3340 fs_info
->closing
= 1;
3343 /* pause restriper - we want to resume on mount */
3344 btrfs_pause_balance(fs_info
);
3346 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3348 btrfs_scrub_cancel(fs_info
);
3350 /* wait for any defraggers to finish */
3351 wait_event(fs_info
->transaction_wait
,
3352 (atomic_read(&fs_info
->defrag_running
) == 0));
3354 /* clear out the rbtree of defraggable inodes */
3355 btrfs_cleanup_defrag_inodes(fs_info
);
3357 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3358 ret
= btrfs_commit_super(root
);
3360 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3363 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3364 btrfs_error_commit_super(root
);
3366 btrfs_put_block_group_cache(fs_info
);
3368 kthread_stop(fs_info
->transaction_kthread
);
3369 kthread_stop(fs_info
->cleaner_kthread
);
3371 fs_info
->closing
= 2;
3374 btrfs_free_qgroup_config(root
->fs_info
);
3376 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3377 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3378 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3381 free_extent_buffer(fs_info
->extent_root
->node
);
3382 free_extent_buffer(fs_info
->extent_root
->commit_root
);
3383 free_extent_buffer(fs_info
->tree_root
->node
);
3384 free_extent_buffer(fs_info
->tree_root
->commit_root
);
3385 free_extent_buffer(fs_info
->chunk_root
->node
);
3386 free_extent_buffer(fs_info
->chunk_root
->commit_root
);
3387 free_extent_buffer(fs_info
->dev_root
->node
);
3388 free_extent_buffer(fs_info
->dev_root
->commit_root
);
3389 free_extent_buffer(fs_info
->csum_root
->node
);
3390 free_extent_buffer(fs_info
->csum_root
->commit_root
);
3391 if (fs_info
->quota_root
) {
3392 free_extent_buffer(fs_info
->quota_root
->node
);
3393 free_extent_buffer(fs_info
->quota_root
->commit_root
);
3396 btrfs_free_block_groups(fs_info
);
3398 del_fs_roots(fs_info
);
3400 iput(fs_info
->btree_inode
);
3402 btrfs_stop_workers(&fs_info
->generic_worker
);
3403 btrfs_stop_workers(&fs_info
->fixup_workers
);
3404 btrfs_stop_workers(&fs_info
->delalloc_workers
);
3405 btrfs_stop_workers(&fs_info
->workers
);
3406 btrfs_stop_workers(&fs_info
->endio_workers
);
3407 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
3408 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
3409 btrfs_stop_workers(&fs_info
->endio_write_workers
);
3410 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
3411 btrfs_stop_workers(&fs_info
->submit_workers
);
3412 btrfs_stop_workers(&fs_info
->delayed_workers
);
3413 btrfs_stop_workers(&fs_info
->caching_workers
);
3414 btrfs_stop_workers(&fs_info
->readahead_workers
);
3415 btrfs_stop_workers(&fs_info
->flush_workers
);
3417 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3418 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3419 btrfsic_unmount(root
, fs_info
->fs_devices
);
3422 btrfs_close_devices(fs_info
->fs_devices
);
3423 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3425 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3426 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3427 bdi_destroy(&fs_info
->bdi
);
3428 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3433 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3437 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3439 ret
= extent_buffer_uptodate(buf
);
3443 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3444 parent_transid
, atomic
);
3450 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3452 return set_extent_buffer_uptodate(buf
);
3455 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3457 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3458 u64 transid
= btrfs_header_generation(buf
);
3461 btrfs_assert_tree_locked(buf
);
3462 if (transid
!= root
->fs_info
->generation
)
3463 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3464 "found %llu running %llu\n",
3465 (unsigned long long)buf
->start
,
3466 (unsigned long long)transid
,
3467 (unsigned long long)root
->fs_info
->generation
);
3468 was_dirty
= set_extent_buffer_dirty(buf
);
3470 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3472 root
->fs_info
->dirty_metadata_batch
);
3475 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3479 * looks as though older kernels can get into trouble with
3480 * this code, they end up stuck in balance_dirty_pages forever
3484 if (current
->flags
& PF_MEMALLOC
)
3488 btrfs_balance_delayed_items(root
);
3490 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3491 BTRFS_DIRTY_METADATA_THRESH
);
3493 balance_dirty_pages_ratelimited_nr(
3494 root
->fs_info
->btree_inode
->i_mapping
, 1);
3499 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3501 __btrfs_btree_balance_dirty(root
, 1);
3504 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3506 __btrfs_btree_balance_dirty(root
, 0);
3509 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3511 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3512 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3515 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3518 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3519 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3529 void btrfs_error_commit_super(struct btrfs_root
*root
)
3531 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3532 btrfs_run_delayed_iputs(root
);
3533 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3535 down_write(&root
->fs_info
->cleanup_work_sem
);
3536 up_write(&root
->fs_info
->cleanup_work_sem
);
3538 /* cleanup FS via transaction */
3539 btrfs_cleanup_transaction(root
);
3542 static void btrfs_destroy_ordered_operations(struct btrfs_root
*root
)
3544 struct btrfs_inode
*btrfs_inode
;
3545 struct list_head splice
;
3547 INIT_LIST_HEAD(&splice
);
3549 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3550 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3552 list_splice_init(&root
->fs_info
->ordered_operations
, &splice
);
3553 while (!list_empty(&splice
)) {
3554 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3555 ordered_operations
);
3557 list_del_init(&btrfs_inode
->ordered_operations
);
3559 btrfs_invalidate_inodes(btrfs_inode
->root
);
3562 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3563 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3566 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3568 struct list_head splice
;
3569 struct btrfs_ordered_extent
*ordered
;
3570 struct inode
*inode
;
3572 INIT_LIST_HEAD(&splice
);
3574 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3576 list_splice_init(&root
->fs_info
->ordered_extents
, &splice
);
3577 while (!list_empty(&splice
)) {
3578 ordered
= list_entry(splice
.next
, struct btrfs_ordered_extent
,
3581 list_del_init(&ordered
->root_extent_list
);
3582 atomic_inc(&ordered
->refs
);
3584 /* the inode may be getting freed (in sys_unlink path). */
3585 inode
= igrab(ordered
->inode
);
3587 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3591 atomic_set(&ordered
->refs
, 1);
3592 btrfs_put_ordered_extent(ordered
);
3594 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3597 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3600 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3601 struct btrfs_root
*root
)
3603 struct rb_node
*node
;
3604 struct btrfs_delayed_ref_root
*delayed_refs
;
3605 struct btrfs_delayed_ref_node
*ref
;
3608 delayed_refs
= &trans
->delayed_refs
;
3610 spin_lock(&delayed_refs
->lock
);
3611 if (delayed_refs
->num_entries
== 0) {
3612 spin_unlock(&delayed_refs
->lock
);
3613 printk(KERN_INFO
"delayed_refs has NO entry\n");
3617 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3618 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3620 atomic_set(&ref
->refs
, 1);
3621 if (btrfs_delayed_ref_is_head(ref
)) {
3622 struct btrfs_delayed_ref_head
*head
;
3624 head
= btrfs_delayed_node_to_head(ref
);
3625 if (!mutex_trylock(&head
->mutex
)) {
3626 atomic_inc(&ref
->refs
);
3627 spin_unlock(&delayed_refs
->lock
);
3629 /* Need to wait for the delayed ref to run */
3630 mutex_lock(&head
->mutex
);
3631 mutex_unlock(&head
->mutex
);
3632 btrfs_put_delayed_ref(ref
);
3634 spin_lock(&delayed_refs
->lock
);
3638 btrfs_free_delayed_extent_op(head
->extent_op
);
3639 delayed_refs
->num_heads
--;
3640 if (list_empty(&head
->cluster
))
3641 delayed_refs
->num_heads_ready
--;
3642 list_del_init(&head
->cluster
);
3645 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3646 delayed_refs
->num_entries
--;
3648 spin_unlock(&delayed_refs
->lock
);
3649 btrfs_put_delayed_ref(ref
);
3652 spin_lock(&delayed_refs
->lock
);
3655 spin_unlock(&delayed_refs
->lock
);
3660 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
)
3662 struct btrfs_pending_snapshot
*snapshot
;
3663 struct list_head splice
;
3665 INIT_LIST_HEAD(&splice
);
3667 list_splice_init(&t
->pending_snapshots
, &splice
);
3669 while (!list_empty(&splice
)) {
3670 snapshot
= list_entry(splice
.next
,
3671 struct btrfs_pending_snapshot
,
3674 list_del_init(&snapshot
->list
);
3680 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3682 struct btrfs_inode
*btrfs_inode
;
3683 struct list_head splice
;
3685 INIT_LIST_HEAD(&splice
);
3687 spin_lock(&root
->fs_info
->delalloc_lock
);
3688 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3690 while (!list_empty(&splice
)) {
3691 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3694 list_del_init(&btrfs_inode
->delalloc_inodes
);
3695 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3696 &btrfs_inode
->runtime_flags
);
3698 btrfs_invalidate_inodes(btrfs_inode
->root
);
3701 spin_unlock(&root
->fs_info
->delalloc_lock
);
3704 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3705 struct extent_io_tree
*dirty_pages
,
3710 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
3711 struct extent_buffer
*eb
;
3715 unsigned long index
;
3718 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3723 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3724 while (start
<= end
) {
3725 index
= start
>> PAGE_CACHE_SHIFT
;
3726 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
3727 page
= find_get_page(btree_inode
->i_mapping
, index
);
3730 offset
= page_offset(page
);
3732 spin_lock(&dirty_pages
->buffer_lock
);
3733 eb
= radix_tree_lookup(
3734 &(&BTRFS_I(page
->mapping
->host
)->io_tree
)->buffer
,
3735 offset
>> PAGE_CACHE_SHIFT
);
3736 spin_unlock(&dirty_pages
->buffer_lock
);
3738 ret
= test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3740 if (PageWriteback(page
))
3741 end_page_writeback(page
);
3744 if (PageDirty(page
)) {
3745 clear_page_dirty_for_io(page
);
3746 spin_lock_irq(&page
->mapping
->tree_lock
);
3747 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3749 PAGECACHE_TAG_DIRTY
);
3750 spin_unlock_irq(&page
->mapping
->tree_lock
);
3754 page_cache_release(page
);
3761 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3762 struct extent_io_tree
*pinned_extents
)
3764 struct extent_io_tree
*unpin
;
3770 unpin
= pinned_extents
;
3773 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3774 EXTENT_DIRTY
, NULL
);
3779 if (btrfs_test_opt(root
, DISCARD
))
3780 ret
= btrfs_error_discard_extent(root
, start
,
3784 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3785 btrfs_error_unpin_extent_range(root
, start
, end
);
3790 if (unpin
== &root
->fs_info
->freed_extents
[0])
3791 unpin
= &root
->fs_info
->freed_extents
[1];
3793 unpin
= &root
->fs_info
->freed_extents
[0];
3801 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3802 struct btrfs_root
*root
)
3804 btrfs_destroy_delayed_refs(cur_trans
, root
);
3805 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3806 cur_trans
->dirty_pages
.dirty_bytes
);
3808 /* FIXME: cleanup wait for commit */
3809 cur_trans
->in_commit
= 1;
3810 cur_trans
->blocked
= 1;
3811 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3813 cur_trans
->blocked
= 0;
3814 wake_up(&root
->fs_info
->transaction_wait
);
3816 cur_trans
->commit_done
= 1;
3817 wake_up(&cur_trans
->commit_wait
);
3819 btrfs_destroy_delayed_inodes(root
);
3820 btrfs_assert_delayed_root_empty(root
);
3822 btrfs_destroy_pending_snapshots(cur_trans
);
3824 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3826 btrfs_destroy_pinned_extent(root
,
3827 root
->fs_info
->pinned_extents
);
3830 memset(cur_trans, 0, sizeof(*cur_trans));
3831 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3835 int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3837 struct btrfs_transaction
*t
;
3840 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3842 spin_lock(&root
->fs_info
->trans_lock
);
3843 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3844 root
->fs_info
->trans_no_join
= 1;
3845 spin_unlock(&root
->fs_info
->trans_lock
);
3847 while (!list_empty(&list
)) {
3848 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3852 btrfs_destroy_ordered_operations(root
);
3854 btrfs_destroy_ordered_extents(root
);
3856 btrfs_destroy_delayed_refs(t
, root
);
3858 btrfs_block_rsv_release(root
,
3859 &root
->fs_info
->trans_block_rsv
,
3860 t
->dirty_pages
.dirty_bytes
);
3862 /* FIXME: cleanup wait for commit */
3866 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3867 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3871 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3872 wake_up(&root
->fs_info
->transaction_wait
);
3876 if (waitqueue_active(&t
->commit_wait
))
3877 wake_up(&t
->commit_wait
);
3879 btrfs_destroy_delayed_inodes(root
);
3880 btrfs_assert_delayed_root_empty(root
);
3882 btrfs_destroy_pending_snapshots(t
);
3884 btrfs_destroy_delalloc_inodes(root
);
3886 spin_lock(&root
->fs_info
->trans_lock
);
3887 root
->fs_info
->running_transaction
= NULL
;
3888 spin_unlock(&root
->fs_info
->trans_lock
);
3890 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3893 btrfs_destroy_pinned_extent(root
,
3894 root
->fs_info
->pinned_extents
);
3896 atomic_set(&t
->use_count
, 0);
3897 list_del_init(&t
->list
);
3898 memset(t
, 0, sizeof(*t
));
3899 kmem_cache_free(btrfs_transaction_cachep
, t
);
3902 spin_lock(&root
->fs_info
->trans_lock
);
3903 root
->fs_info
->trans_no_join
= 0;
3904 spin_unlock(&root
->fs_info
->trans_lock
);
3905 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3910 static struct extent_io_ops btree_extent_io_ops
= {
3911 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3912 .readpage_io_failed_hook
= btree_io_failed_hook
,
3913 .submit_bio_hook
= btree_submit_bio_hook
,
3914 /* note we're sharing with inode.c for the merge bio hook */
3915 .merge_bio_hook
= btrfs_merge_bio_hook
,