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"
49 static struct extent_io_ops btree_extent_io_ops
;
50 static void end_workqueue_fn(struct btrfs_work
*work
);
51 static void free_fs_root(struct btrfs_root
*root
);
52 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
54 static void btrfs_destroy_ordered_operations(struct btrfs_root
*root
);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
57 struct btrfs_root
*root
);
58 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
);
59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
60 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
61 struct extent_io_tree
*dirty_pages
,
63 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
64 struct extent_io_tree
*pinned_extents
);
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
75 struct btrfs_fs_info
*info
;
78 struct list_head list
;
79 struct btrfs_work work
;
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
87 struct async_submit_bio
{
90 struct list_head list
;
91 extent_submit_bio_hook_t
*submit_bio_start
;
92 extent_submit_bio_hook_t
*submit_bio_done
;
95 unsigned long bio_flags
;
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
101 struct btrfs_work work
;
106 * Lockdep class keys for extent_buffer->lock's in this root. For a given
107 * eb, the lockdep key is determined by the btrfs_root it belongs to and
108 * the level the eb occupies in the tree.
110 * Different roots are used for different purposes and may nest inside each
111 * other and they require separate keysets. As lockdep keys should be
112 * static, assign keysets according to the purpose of the root as indicated
113 * by btrfs_root->objectid. This ensures that all special purpose roots
114 * have separate keysets.
116 * Lock-nesting across peer nodes is always done with the immediate parent
117 * node locked thus preventing deadlock. As lockdep doesn't know this, use
118 * subclass to avoid triggering lockdep warning in such cases.
120 * The key is set by the readpage_end_io_hook after the buffer has passed
121 * csum validation but before the pages are unlocked. It is also set by
122 * btrfs_init_new_buffer on freshly allocated blocks.
124 * We also add a check to make sure the highest level of the tree is the
125 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
126 * needs update as well.
128 #ifdef CONFIG_DEBUG_LOCK_ALLOC
129 # if BTRFS_MAX_LEVEL != 8
133 static struct btrfs_lockdep_keyset
{
134 u64 id
; /* root objectid */
135 const char *name_stem
; /* lock name stem */
136 char names
[BTRFS_MAX_LEVEL
+ 1][20];
137 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
138 } btrfs_lockdep_keysets
[] = {
139 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
140 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
141 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
142 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
143 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
144 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
145 { .id
= BTRFS_ORPHAN_OBJECTID
, .name_stem
= "orphan" },
146 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
147 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
148 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
149 { .id
= 0, .name_stem
= "tree" },
152 void __init
btrfs_init_lockdep(void)
156 /* initialize lockdep class names */
157 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
158 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
160 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
161 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
162 "btrfs-%s-%02d", ks
->name_stem
, j
);
166 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
169 struct btrfs_lockdep_keyset
*ks
;
171 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
173 /* find the matching keyset, id 0 is the default entry */
174 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
175 if (ks
->id
== objectid
)
178 lockdep_set_class_and_name(&eb
->lock
,
179 &ks
->keys
[level
], ks
->names
[level
]);
185 * extents on the btree inode are pretty simple, there's one extent
186 * that covers the entire device
188 static struct extent_map
*btree_get_extent(struct inode
*inode
,
189 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
192 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
193 struct extent_map
*em
;
196 read_lock(&em_tree
->lock
);
197 em
= lookup_extent_mapping(em_tree
, start
, len
);
200 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
201 read_unlock(&em_tree
->lock
);
204 read_unlock(&em_tree
->lock
);
206 em
= alloc_extent_map();
208 em
= ERR_PTR(-ENOMEM
);
213 em
->block_len
= (u64
)-1;
215 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
217 write_lock(&em_tree
->lock
);
218 ret
= add_extent_mapping(em_tree
, em
);
219 if (ret
== -EEXIST
) {
220 u64 failed_start
= em
->start
;
221 u64 failed_len
= em
->len
;
224 em
= lookup_extent_mapping(em_tree
, start
, len
);
228 em
= lookup_extent_mapping(em_tree
, failed_start
,
236 write_unlock(&em_tree
->lock
);
244 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
246 return crc32c(seed
, data
, len
);
249 void btrfs_csum_final(u32 crc
, char *result
)
251 put_unaligned_le32(~crc
, result
);
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
258 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
261 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
264 unsigned long cur_len
;
265 unsigned long offset
= BTRFS_CSUM_SIZE
;
267 unsigned long map_start
;
268 unsigned long map_len
;
271 unsigned long inline_result
;
273 len
= buf
->len
- offset
;
275 err
= map_private_extent_buffer(buf
, offset
, 32,
276 &kaddr
, &map_start
, &map_len
);
279 cur_len
= min(len
, map_len
- (offset
- map_start
));
280 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
285 if (csum_size
> sizeof(inline_result
)) {
286 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
290 result
= (char *)&inline_result
;
293 btrfs_csum_final(crc
, result
);
296 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
299 memcpy(&found
, result
, csum_size
);
301 read_extent_buffer(buf
, &val
, 0, csum_size
);
302 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
305 root
->fs_info
->sb
->s_id
,
306 (unsigned long long)buf
->start
, val
, found
,
307 btrfs_header_level(buf
));
308 if (result
!= (char *)&inline_result
)
313 write_extent_buffer(buf
, result
, 0, csum_size
);
315 if (result
!= (char *)&inline_result
)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
327 struct extent_buffer
*eb
, u64 parent_transid
,
330 struct extent_state
*cached_state
= NULL
;
333 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
339 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
341 if (extent_buffer_uptodate(eb
) &&
342 btrfs_header_generation(eb
) == parent_transid
) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 (unsigned long long)eb
->start
,
349 (unsigned long long)parent_transid
,
350 (unsigned long long)btrfs_header_generation(eb
));
352 clear_extent_buffer_uptodate(eb
);
354 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
355 &cached_state
, GFP_NOFS
);
360 * helper to read a given tree block, doing retries as required when
361 * the checksums don't match and we have alternate mirrors to try.
363 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
364 struct extent_buffer
*eb
,
365 u64 start
, u64 parent_transid
)
367 struct extent_io_tree
*io_tree
;
372 int failed_mirror
= 0;
374 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
375 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
377 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
379 btree_get_extent
, mirror_num
);
380 if (!ret
&& !verify_parent_transid(io_tree
, eb
,
385 * This buffer's crc is fine, but its contents are corrupted, so
386 * there is no reason to read the other copies, they won't be
389 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
392 num_copies
= btrfs_num_copies(&root
->fs_info
->mapping_tree
,
397 if (!failed_mirror
) {
399 failed_mirror
= eb
->read_mirror
;
403 if (mirror_num
== failed_mirror
)
406 if (mirror_num
> num_copies
)
410 if (failed
&& !ret
&& failed_mirror
)
411 repair_eb_io_failure(root
, eb
, failed_mirror
);
417 * checksum a dirty tree block before IO. This has extra checks to make sure
418 * we only fill in the checksum field in the first page of a multi-page block
421 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
423 struct extent_io_tree
*tree
;
424 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
426 struct extent_buffer
*eb
;
428 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
430 eb
= (struct extent_buffer
*)page
->private;
431 if (page
!= eb
->pages
[0])
433 found_start
= btrfs_header_bytenr(eb
);
434 if (found_start
!= start
) {
438 if (eb
->pages
[0] != page
) {
442 if (!PageUptodate(page
)) {
446 csum_tree_block(root
, eb
, 0);
450 static int check_tree_block_fsid(struct btrfs_root
*root
,
451 struct extent_buffer
*eb
)
453 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
454 u8 fsid
[BTRFS_UUID_SIZE
];
457 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
460 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
464 fs_devices
= fs_devices
->seed
;
469 #define CORRUPT(reason, eb, root, slot) \
470 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471 "root=%llu, slot=%d\n", reason, \
472 (unsigned long long)btrfs_header_bytenr(eb), \
473 (unsigned long long)root->objectid, slot)
475 static noinline
int check_leaf(struct btrfs_root
*root
,
476 struct extent_buffer
*leaf
)
478 struct btrfs_key key
;
479 struct btrfs_key leaf_key
;
480 u32 nritems
= btrfs_header_nritems(leaf
);
486 /* Check the 0 item */
487 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
488 BTRFS_LEAF_DATA_SIZE(root
)) {
489 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
494 * Check to make sure each items keys are in the correct order and their
495 * offsets make sense. We only have to loop through nritems-1 because
496 * we check the current slot against the next slot, which verifies the
497 * next slot's offset+size makes sense and that the current's slot
500 for (slot
= 0; slot
< nritems
- 1; slot
++) {
501 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
502 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
504 /* Make sure the keys are in the right order */
505 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
506 CORRUPT("bad key order", leaf
, root
, slot
);
511 * Make sure the offset and ends are right, remember that the
512 * item data starts at the end of the leaf and grows towards the
515 if (btrfs_item_offset_nr(leaf
, slot
) !=
516 btrfs_item_end_nr(leaf
, slot
+ 1)) {
517 CORRUPT("slot offset bad", leaf
, root
, slot
);
522 * Check to make sure that we don't point outside of the leaf,
523 * just incase all the items are consistent to eachother, but
524 * all point outside of the leaf.
526 if (btrfs_item_end_nr(leaf
, slot
) >
527 BTRFS_LEAF_DATA_SIZE(root
)) {
528 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
536 struct extent_buffer
*find_eb_for_page(struct extent_io_tree
*tree
,
537 struct page
*page
, int max_walk
)
539 struct extent_buffer
*eb
;
540 u64 start
= page_offset(page
);
544 if (start
< max_walk
)
547 min_start
= start
- max_walk
;
549 while (start
>= min_start
) {
550 eb
= find_extent_buffer(tree
, start
, 0);
553 * we found an extent buffer and it contains our page
556 if (eb
->start
<= target
&&
557 eb
->start
+ eb
->len
> target
)
560 /* we found an extent buffer that wasn't for us */
561 free_extent_buffer(eb
);
566 start
-= PAGE_CACHE_SIZE
;
571 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
572 struct extent_state
*state
, int mirror
)
574 struct extent_io_tree
*tree
;
577 struct extent_buffer
*eb
;
578 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
585 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
586 eb
= (struct extent_buffer
*)page
->private;
588 /* the pending IO might have been the only thing that kept this buffer
589 * in memory. Make sure we have a ref for all this other checks
591 extent_buffer_get(eb
);
593 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
597 eb
->read_mirror
= mirror
;
598 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
603 found_start
= btrfs_header_bytenr(eb
);
604 if (found_start
!= eb
->start
) {
605 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
607 (unsigned long long)found_start
,
608 (unsigned long long)eb
->start
);
612 if (check_tree_block_fsid(root
, eb
)) {
613 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
614 (unsigned long long)eb
->start
);
618 found_level
= btrfs_header_level(eb
);
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
623 ret
= csum_tree_block(root
, eb
, 1);
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
634 if (found_level
== 0 && check_leaf(root
, eb
)) {
635 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
640 set_extent_buffer_uptodate(eb
);
642 if (test_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
)) {
643 clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
);
644 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
648 clear_extent_buffer_uptodate(eb
);
649 free_extent_buffer(eb
);
654 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
656 struct extent_buffer
*eb
;
657 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
659 eb
= (struct extent_buffer
*)page
->private;
660 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
661 eb
->read_mirror
= failed_mirror
;
662 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
663 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
664 return -EIO
; /* we fixed nothing */
667 static void end_workqueue_bio(struct bio
*bio
, int err
)
669 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
670 struct btrfs_fs_info
*fs_info
;
672 fs_info
= end_io_wq
->info
;
673 end_io_wq
->error
= err
;
674 end_io_wq
->work
.func
= end_workqueue_fn
;
675 end_io_wq
->work
.flags
= 0;
677 if (bio
->bi_rw
& REQ_WRITE
) {
678 if (end_io_wq
->metadata
== 1)
679 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
681 else if (end_io_wq
->metadata
== 2)
682 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
685 btrfs_queue_worker(&fs_info
->endio_write_workers
,
688 if (end_io_wq
->metadata
)
689 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
692 btrfs_queue_worker(&fs_info
->endio_workers
,
698 * For the metadata arg you want
701 * 1 - if normal metadta
702 * 2 - if writing to the free space cache area
704 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
707 struct end_io_wq
*end_io_wq
;
708 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
712 end_io_wq
->private = bio
->bi_private
;
713 end_io_wq
->end_io
= bio
->bi_end_io
;
714 end_io_wq
->info
= info
;
715 end_io_wq
->error
= 0;
716 end_io_wq
->bio
= bio
;
717 end_io_wq
->metadata
= metadata
;
719 bio
->bi_private
= end_io_wq
;
720 bio
->bi_end_io
= end_workqueue_bio
;
724 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
726 unsigned long limit
= min_t(unsigned long,
727 info
->workers
.max_workers
,
728 info
->fs_devices
->open_devices
);
732 static void run_one_async_start(struct btrfs_work
*work
)
734 struct async_submit_bio
*async
;
737 async
= container_of(work
, struct async_submit_bio
, work
);
738 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
739 async
->mirror_num
, async
->bio_flags
,
745 static void run_one_async_done(struct btrfs_work
*work
)
747 struct btrfs_fs_info
*fs_info
;
748 struct async_submit_bio
*async
;
751 async
= container_of(work
, struct async_submit_bio
, work
);
752 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
754 limit
= btrfs_async_submit_limit(fs_info
);
755 limit
= limit
* 2 / 3;
757 atomic_dec(&fs_info
->nr_async_submits
);
759 if (atomic_read(&fs_info
->nr_async_submits
) < limit
&&
760 waitqueue_active(&fs_info
->async_submit_wait
))
761 wake_up(&fs_info
->async_submit_wait
);
763 /* If an error occured we just want to clean up the bio and move on */
765 bio_endio(async
->bio
, async
->error
);
769 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
770 async
->mirror_num
, async
->bio_flags
,
774 static void run_one_async_free(struct btrfs_work
*work
)
776 struct async_submit_bio
*async
;
778 async
= container_of(work
, struct async_submit_bio
, work
);
782 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
783 int rw
, struct bio
*bio
, int mirror_num
,
784 unsigned long bio_flags
,
786 extent_submit_bio_hook_t
*submit_bio_start
,
787 extent_submit_bio_hook_t
*submit_bio_done
)
789 struct async_submit_bio
*async
;
791 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
795 async
->inode
= inode
;
798 async
->mirror_num
= mirror_num
;
799 async
->submit_bio_start
= submit_bio_start
;
800 async
->submit_bio_done
= submit_bio_done
;
802 async
->work
.func
= run_one_async_start
;
803 async
->work
.ordered_func
= run_one_async_done
;
804 async
->work
.ordered_free
= run_one_async_free
;
806 async
->work
.flags
= 0;
807 async
->bio_flags
= bio_flags
;
808 async
->bio_offset
= bio_offset
;
812 atomic_inc(&fs_info
->nr_async_submits
);
815 btrfs_set_work_high_prio(&async
->work
);
817 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
819 while (atomic_read(&fs_info
->async_submit_draining
) &&
820 atomic_read(&fs_info
->nr_async_submits
)) {
821 wait_event(fs_info
->async_submit_wait
,
822 (atomic_read(&fs_info
->nr_async_submits
) == 0));
828 static int btree_csum_one_bio(struct bio
*bio
)
830 struct bio_vec
*bvec
= bio
->bi_io_vec
;
832 struct btrfs_root
*root
;
835 WARN_ON(bio
->bi_vcnt
<= 0);
836 while (bio_index
< bio
->bi_vcnt
) {
837 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
838 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
847 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
848 struct bio
*bio
, int mirror_num
,
849 unsigned long bio_flags
,
853 * when we're called for a write, we're already in the async
854 * submission context. Just jump into btrfs_map_bio
856 return btree_csum_one_bio(bio
);
859 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
860 int mirror_num
, unsigned long bio_flags
,
864 * when we're called for a write, we're already in the async
865 * submission context. Just jump into btrfs_map_bio
867 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
870 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
871 int mirror_num
, unsigned long bio_flags
,
876 if (!(rw
& REQ_WRITE
)) {
879 * called for a read, do the setup so that checksum validation
880 * can happen in the async kernel threads
882 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
886 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
891 * kthread helpers are used to submit writes so that checksumming
892 * can happen in parallel across all CPUs
894 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
895 inode
, rw
, bio
, mirror_num
, 0,
897 __btree_submit_bio_start
,
898 __btree_submit_bio_done
);
901 #ifdef CONFIG_MIGRATION
902 static int btree_migratepage(struct address_space
*mapping
,
903 struct page
*newpage
, struct page
*page
,
904 enum migrate_mode mode
)
907 * we can't safely write a btree page from here,
908 * we haven't done the locking hook
913 * Buffers may be managed in a filesystem specific way.
914 * We must have no buffers or drop them.
916 if (page_has_private(page
) &&
917 !try_to_release_page(page
, GFP_KERNEL
))
919 return migrate_page(mapping
, newpage
, page
, mode
);
924 static int btree_writepages(struct address_space
*mapping
,
925 struct writeback_control
*wbc
)
927 struct extent_io_tree
*tree
;
928 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
929 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
930 struct btrfs_root
*root
= BTRFS_I(mapping
->host
)->root
;
932 unsigned long thresh
= 32 * 1024 * 1024;
934 if (wbc
->for_kupdate
)
937 /* this is a bit racy, but that's ok */
938 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
939 if (num_dirty
< thresh
)
942 return btree_write_cache_pages(mapping
, wbc
);
945 static int btree_readpage(struct file
*file
, struct page
*page
)
947 struct extent_io_tree
*tree
;
948 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
949 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
952 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
954 if (PageWriteback(page
) || PageDirty(page
))
957 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
958 * slab allocation from alloc_extent_state down the callchain where
959 * it'd hit a BUG_ON as those flags are not allowed.
961 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
963 return try_release_extent_buffer(page
, gfp_flags
);
966 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
968 struct extent_io_tree
*tree
;
969 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
970 extent_invalidatepage(tree
, page
, offset
);
971 btree_releasepage(page
, GFP_NOFS
);
972 if (PagePrivate(page
)) {
973 printk(KERN_WARNING
"btrfs warning page private not zero "
974 "on page %llu\n", (unsigned long long)page_offset(page
));
975 ClearPagePrivate(page
);
976 set_page_private(page
, 0);
977 page_cache_release(page
);
981 static int btree_set_page_dirty(struct page
*page
)
983 struct extent_buffer
*eb
;
985 BUG_ON(!PagePrivate(page
));
986 eb
= (struct extent_buffer
*)page
->private;
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
989 BUG_ON(!atomic_read(&eb
->refs
));
990 btrfs_assert_tree_locked(eb
);
991 return __set_page_dirty_nobuffers(page
);
994 static const struct address_space_operations btree_aops
= {
995 .readpage
= btree_readpage
,
996 .writepages
= btree_writepages
,
997 .releasepage
= btree_releasepage
,
998 .invalidatepage
= btree_invalidatepage
,
999 #ifdef CONFIG_MIGRATION
1000 .migratepage
= btree_migratepage
,
1002 .set_page_dirty
= btree_set_page_dirty
,
1005 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1008 struct extent_buffer
*buf
= NULL
;
1009 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1012 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1016 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1017 free_extent_buffer(buf
);
1021 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1022 int mirror_num
, struct extent_buffer
**eb
)
1024 struct extent_buffer
*buf
= NULL
;
1025 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1026 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1029 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1033 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1035 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1036 btree_get_extent
, mirror_num
);
1038 free_extent_buffer(buf
);
1042 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1043 free_extent_buffer(buf
);
1045 } else if (extent_buffer_uptodate(buf
)) {
1048 free_extent_buffer(buf
);
1053 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1054 u64 bytenr
, u32 blocksize
)
1056 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1057 struct extent_buffer
*eb
;
1058 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1063 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1064 u64 bytenr
, u32 blocksize
)
1066 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1067 struct extent_buffer
*eb
;
1069 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1075 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1077 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1078 buf
->start
+ buf
->len
- 1);
1081 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1083 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1084 buf
->start
, buf
->start
+ buf
->len
- 1);
1087 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1088 u32 blocksize
, u64 parent_transid
)
1090 struct extent_buffer
*buf
= NULL
;
1093 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1097 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1102 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1103 struct extent_buffer
*buf
)
1105 if (btrfs_header_generation(buf
) ==
1106 root
->fs_info
->running_transaction
->transid
) {
1107 btrfs_assert_tree_locked(buf
);
1109 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1110 spin_lock(&root
->fs_info
->delalloc_lock
);
1111 if (root
->fs_info
->dirty_metadata_bytes
>= buf
->len
)
1112 root
->fs_info
->dirty_metadata_bytes
-= buf
->len
;
1114 spin_unlock(&root
->fs_info
->delalloc_lock
);
1115 btrfs_panic(root
->fs_info
, -EOVERFLOW
,
1116 "Can't clear %lu bytes from "
1117 " dirty_mdatadata_bytes (%llu)",
1119 root
->fs_info
->dirty_metadata_bytes
);
1121 spin_unlock(&root
->fs_info
->delalloc_lock
);
1124 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125 btrfs_set_lock_blocking(buf
);
1126 clear_extent_buffer_dirty(buf
);
1130 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1131 u32 stripesize
, struct btrfs_root
*root
,
1132 struct btrfs_fs_info
*fs_info
,
1136 root
->commit_root
= NULL
;
1137 root
->sectorsize
= sectorsize
;
1138 root
->nodesize
= nodesize
;
1139 root
->leafsize
= leafsize
;
1140 root
->stripesize
= stripesize
;
1142 root
->track_dirty
= 0;
1144 root
->orphan_item_inserted
= 0;
1145 root
->orphan_cleanup_state
= 0;
1147 root
->objectid
= objectid
;
1148 root
->last_trans
= 0;
1149 root
->highest_objectid
= 0;
1151 root
->inode_tree
= RB_ROOT
;
1152 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1153 root
->block_rsv
= NULL
;
1154 root
->orphan_block_rsv
= NULL
;
1156 INIT_LIST_HEAD(&root
->dirty_list
);
1157 INIT_LIST_HEAD(&root
->root_list
);
1158 spin_lock_init(&root
->orphan_lock
);
1159 spin_lock_init(&root
->inode_lock
);
1160 spin_lock_init(&root
->accounting_lock
);
1161 mutex_init(&root
->objectid_mutex
);
1162 mutex_init(&root
->log_mutex
);
1163 init_waitqueue_head(&root
->log_writer_wait
);
1164 init_waitqueue_head(&root
->log_commit_wait
[0]);
1165 init_waitqueue_head(&root
->log_commit_wait
[1]);
1166 atomic_set(&root
->log_commit
[0], 0);
1167 atomic_set(&root
->log_commit
[1], 0);
1168 atomic_set(&root
->log_writers
, 0);
1169 atomic_set(&root
->orphan_inodes
, 0);
1170 root
->log_batch
= 0;
1171 root
->log_transid
= 0;
1172 root
->last_log_commit
= 0;
1173 extent_io_tree_init(&root
->dirty_log_pages
,
1174 fs_info
->btree_inode
->i_mapping
);
1176 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1177 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1178 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1179 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1180 root
->defrag_trans_start
= fs_info
->generation
;
1181 init_completion(&root
->kobj_unregister
);
1182 root
->defrag_running
= 0;
1183 root
->root_key
.objectid
= objectid
;
1186 spin_lock_init(&root
->root_times_lock
);
1189 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1190 struct btrfs_fs_info
*fs_info
,
1192 struct btrfs_root
*root
)
1198 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1199 tree_root
->sectorsize
, tree_root
->stripesize
,
1200 root
, fs_info
, objectid
);
1201 ret
= btrfs_find_last_root(tree_root
, objectid
,
1202 &root
->root_item
, &root
->root_key
);
1208 generation
= btrfs_root_generation(&root
->root_item
);
1209 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1210 root
->commit_root
= NULL
;
1211 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1212 blocksize
, generation
);
1213 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1214 free_extent_buffer(root
->node
);
1218 root
->commit_root
= btrfs_root_node(root
);
1222 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1224 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1226 root
->fs_info
= fs_info
;
1230 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1231 struct btrfs_fs_info
*fs_info
,
1234 struct extent_buffer
*leaf
;
1235 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1236 struct btrfs_root
*root
;
1237 struct btrfs_key key
;
1241 root
= btrfs_alloc_root(fs_info
);
1243 return ERR_PTR(-ENOMEM
);
1245 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1246 tree_root
->sectorsize
, tree_root
->stripesize
,
1247 root
, fs_info
, objectid
);
1248 root
->root_key
.objectid
= objectid
;
1249 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1250 root
->root_key
.offset
= 0;
1252 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1253 0, objectid
, NULL
, 0, 0, 0);
1255 ret
= PTR_ERR(leaf
);
1259 bytenr
= leaf
->start
;
1260 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1261 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1262 btrfs_set_header_generation(leaf
, trans
->transid
);
1263 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1264 btrfs_set_header_owner(leaf
, objectid
);
1267 write_extent_buffer(leaf
, fs_info
->fsid
,
1268 (unsigned long)btrfs_header_fsid(leaf
),
1270 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1271 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1273 btrfs_mark_buffer_dirty(leaf
);
1275 root
->commit_root
= btrfs_root_node(root
);
1276 root
->track_dirty
= 1;
1279 root
->root_item
.flags
= 0;
1280 root
->root_item
.byte_limit
= 0;
1281 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1282 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1283 btrfs_set_root_level(&root
->root_item
, 0);
1284 btrfs_set_root_refs(&root
->root_item
, 1);
1285 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1286 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1287 btrfs_set_root_dirid(&root
->root_item
, 0);
1288 root
->root_item
.drop_level
= 0;
1290 key
.objectid
= objectid
;
1291 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1293 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1297 btrfs_tree_unlock(leaf
);
1301 return ERR_PTR(ret
);
1306 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1307 struct btrfs_fs_info
*fs_info
)
1309 struct btrfs_root
*root
;
1310 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1311 struct extent_buffer
*leaf
;
1313 root
= btrfs_alloc_root(fs_info
);
1315 return ERR_PTR(-ENOMEM
);
1317 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1318 tree_root
->sectorsize
, tree_root
->stripesize
,
1319 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1321 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1322 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1323 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1325 * log trees do not get reference counted because they go away
1326 * before a real commit is actually done. They do store pointers
1327 * to file data extents, and those reference counts still get
1328 * updated (along with back refs to the log tree).
1332 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1333 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1337 return ERR_CAST(leaf
);
1340 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1341 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1342 btrfs_set_header_generation(leaf
, trans
->transid
);
1343 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1344 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1347 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1348 (unsigned long)btrfs_header_fsid(root
->node
),
1350 btrfs_mark_buffer_dirty(root
->node
);
1351 btrfs_tree_unlock(root
->node
);
1355 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1356 struct btrfs_fs_info
*fs_info
)
1358 struct btrfs_root
*log_root
;
1360 log_root
= alloc_log_tree(trans
, fs_info
);
1361 if (IS_ERR(log_root
))
1362 return PTR_ERR(log_root
);
1363 WARN_ON(fs_info
->log_root_tree
);
1364 fs_info
->log_root_tree
= log_root
;
1368 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1369 struct btrfs_root
*root
)
1371 struct btrfs_root
*log_root
;
1372 struct btrfs_inode_item
*inode_item
;
1374 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1375 if (IS_ERR(log_root
))
1376 return PTR_ERR(log_root
);
1378 log_root
->last_trans
= trans
->transid
;
1379 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1381 inode_item
= &log_root
->root_item
.inode
;
1382 inode_item
->generation
= cpu_to_le64(1);
1383 inode_item
->size
= cpu_to_le64(3);
1384 inode_item
->nlink
= cpu_to_le32(1);
1385 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1386 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1388 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1390 WARN_ON(root
->log_root
);
1391 root
->log_root
= log_root
;
1392 root
->log_transid
= 0;
1393 root
->last_log_commit
= 0;
1397 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1398 struct btrfs_key
*location
)
1400 struct btrfs_root
*root
;
1401 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1402 struct btrfs_path
*path
;
1403 struct extent_buffer
*l
;
1409 root
= btrfs_alloc_root(fs_info
);
1411 return ERR_PTR(-ENOMEM
);
1412 if (location
->offset
== (u64
)-1) {
1413 ret
= find_and_setup_root(tree_root
, fs_info
,
1414 location
->objectid
, root
);
1417 return ERR_PTR(ret
);
1422 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1423 tree_root
->sectorsize
, tree_root
->stripesize
,
1424 root
, fs_info
, location
->objectid
);
1426 path
= btrfs_alloc_path();
1429 return ERR_PTR(-ENOMEM
);
1431 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1434 slot
= path
->slots
[0];
1435 btrfs_read_root_item(tree_root
, l
, slot
, &root
->root_item
);
1436 memcpy(&root
->root_key
, location
, sizeof(*location
));
1438 btrfs_free_path(path
);
1443 return ERR_PTR(ret
);
1446 generation
= btrfs_root_generation(&root
->root_item
);
1447 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1448 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1449 blocksize
, generation
);
1450 root
->commit_root
= btrfs_root_node(root
);
1451 BUG_ON(!root
->node
); /* -ENOMEM */
1453 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1455 btrfs_check_and_init_root_item(&root
->root_item
);
1461 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1462 struct btrfs_key
*location
)
1464 struct btrfs_root
*root
;
1467 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1468 return fs_info
->tree_root
;
1469 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1470 return fs_info
->extent_root
;
1471 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1472 return fs_info
->chunk_root
;
1473 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1474 return fs_info
->dev_root
;
1475 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1476 return fs_info
->csum_root
;
1477 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1478 return fs_info
->quota_root
? fs_info
->quota_root
:
1481 spin_lock(&fs_info
->fs_roots_radix_lock
);
1482 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1483 (unsigned long)location
->objectid
);
1484 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1488 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1492 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1493 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1495 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1500 btrfs_init_free_ino_ctl(root
);
1501 mutex_init(&root
->fs_commit_mutex
);
1502 spin_lock_init(&root
->cache_lock
);
1503 init_waitqueue_head(&root
->cache_wait
);
1505 ret
= get_anon_bdev(&root
->anon_dev
);
1509 if (btrfs_root_refs(&root
->root_item
) == 0) {
1514 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1518 root
->orphan_item_inserted
= 1;
1520 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1524 spin_lock(&fs_info
->fs_roots_radix_lock
);
1525 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1526 (unsigned long)root
->root_key
.objectid
,
1531 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1532 radix_tree_preload_end();
1534 if (ret
== -EEXIST
) {
1541 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1542 root
->root_key
.objectid
);
1547 return ERR_PTR(ret
);
1550 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1552 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1554 struct btrfs_device
*device
;
1555 struct backing_dev_info
*bdi
;
1558 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1561 bdi
= blk_get_backing_dev_info(device
->bdev
);
1562 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1572 * If this fails, caller must call bdi_destroy() to get rid of the
1575 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1579 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1580 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1584 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1585 bdi
->congested_fn
= btrfs_congested_fn
;
1586 bdi
->congested_data
= info
;
1591 * called by the kthread helper functions to finally call the bio end_io
1592 * functions. This is where read checksum verification actually happens
1594 static void end_workqueue_fn(struct btrfs_work
*work
)
1597 struct end_io_wq
*end_io_wq
;
1598 struct btrfs_fs_info
*fs_info
;
1601 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1602 bio
= end_io_wq
->bio
;
1603 fs_info
= end_io_wq
->info
;
1605 error
= end_io_wq
->error
;
1606 bio
->bi_private
= end_io_wq
->private;
1607 bio
->bi_end_io
= end_io_wq
->end_io
;
1609 bio_endio(bio
, error
);
1612 static int cleaner_kthread(void *arg
)
1614 struct btrfs_root
*root
= arg
;
1617 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1619 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1620 mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1621 btrfs_run_delayed_iputs(root
);
1622 btrfs_clean_old_snapshots(root
);
1623 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1624 btrfs_run_defrag_inodes(root
->fs_info
);
1627 if (!try_to_freeze()) {
1628 set_current_state(TASK_INTERRUPTIBLE
);
1629 if (!kthread_should_stop())
1631 __set_current_state(TASK_RUNNING
);
1633 } while (!kthread_should_stop());
1637 static int transaction_kthread(void *arg
)
1639 struct btrfs_root
*root
= arg
;
1640 struct btrfs_trans_handle
*trans
;
1641 struct btrfs_transaction
*cur
;
1644 unsigned long delay
;
1648 cannot_commit
= false;
1650 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1651 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1653 spin_lock(&root
->fs_info
->trans_lock
);
1654 cur
= root
->fs_info
->running_transaction
;
1656 spin_unlock(&root
->fs_info
->trans_lock
);
1660 now
= get_seconds();
1661 if (!cur
->blocked
&&
1662 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1663 spin_unlock(&root
->fs_info
->trans_lock
);
1667 transid
= cur
->transid
;
1668 spin_unlock(&root
->fs_info
->trans_lock
);
1670 /* If the file system is aborted, this will always fail. */
1671 trans
= btrfs_join_transaction(root
);
1672 if (IS_ERR(trans
)) {
1673 cannot_commit
= true;
1676 if (transid
== trans
->transid
) {
1677 btrfs_commit_transaction(trans
, root
);
1679 btrfs_end_transaction(trans
, root
);
1682 wake_up_process(root
->fs_info
->cleaner_kthread
);
1683 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1685 if (!try_to_freeze()) {
1686 set_current_state(TASK_INTERRUPTIBLE
);
1687 if (!kthread_should_stop() &&
1688 (!btrfs_transaction_blocked(root
->fs_info
) ||
1690 schedule_timeout(delay
);
1691 __set_current_state(TASK_RUNNING
);
1693 } while (!kthread_should_stop());
1698 * this will find the highest generation in the array of
1699 * root backups. The index of the highest array is returned,
1700 * or -1 if we can't find anything.
1702 * We check to make sure the array is valid by comparing the
1703 * generation of the latest root in the array with the generation
1704 * in the super block. If they don't match we pitch it.
1706 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1709 int newest_index
= -1;
1710 struct btrfs_root_backup
*root_backup
;
1713 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1714 root_backup
= info
->super_copy
->super_roots
+ i
;
1715 cur
= btrfs_backup_tree_root_gen(root_backup
);
1716 if (cur
== newest_gen
)
1720 /* check to see if we actually wrapped around */
1721 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1722 root_backup
= info
->super_copy
->super_roots
;
1723 cur
= btrfs_backup_tree_root_gen(root_backup
);
1724 if (cur
== newest_gen
)
1727 return newest_index
;
1732 * find the oldest backup so we know where to store new entries
1733 * in the backup array. This will set the backup_root_index
1734 * field in the fs_info struct
1736 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1739 int newest_index
= -1;
1741 newest_index
= find_newest_super_backup(info
, newest_gen
);
1742 /* if there was garbage in there, just move along */
1743 if (newest_index
== -1) {
1744 info
->backup_root_index
= 0;
1746 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1751 * copy all the root pointers into the super backup array.
1752 * this will bump the backup pointer by one when it is
1755 static void backup_super_roots(struct btrfs_fs_info
*info
)
1758 struct btrfs_root_backup
*root_backup
;
1761 next_backup
= info
->backup_root_index
;
1762 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1763 BTRFS_NUM_BACKUP_ROOTS
;
1766 * just overwrite the last backup if we're at the same generation
1767 * this happens only at umount
1769 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1770 if (btrfs_backup_tree_root_gen(root_backup
) ==
1771 btrfs_header_generation(info
->tree_root
->node
))
1772 next_backup
= last_backup
;
1774 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1777 * make sure all of our padding and empty slots get zero filled
1778 * regardless of which ones we use today
1780 memset(root_backup
, 0, sizeof(*root_backup
));
1782 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1784 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1785 btrfs_set_backup_tree_root_gen(root_backup
,
1786 btrfs_header_generation(info
->tree_root
->node
));
1788 btrfs_set_backup_tree_root_level(root_backup
,
1789 btrfs_header_level(info
->tree_root
->node
));
1791 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1792 btrfs_set_backup_chunk_root_gen(root_backup
,
1793 btrfs_header_generation(info
->chunk_root
->node
));
1794 btrfs_set_backup_chunk_root_level(root_backup
,
1795 btrfs_header_level(info
->chunk_root
->node
));
1797 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1798 btrfs_set_backup_extent_root_gen(root_backup
,
1799 btrfs_header_generation(info
->extent_root
->node
));
1800 btrfs_set_backup_extent_root_level(root_backup
,
1801 btrfs_header_level(info
->extent_root
->node
));
1804 * we might commit during log recovery, which happens before we set
1805 * the fs_root. Make sure it is valid before we fill it in.
1807 if (info
->fs_root
&& info
->fs_root
->node
) {
1808 btrfs_set_backup_fs_root(root_backup
,
1809 info
->fs_root
->node
->start
);
1810 btrfs_set_backup_fs_root_gen(root_backup
,
1811 btrfs_header_generation(info
->fs_root
->node
));
1812 btrfs_set_backup_fs_root_level(root_backup
,
1813 btrfs_header_level(info
->fs_root
->node
));
1816 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1817 btrfs_set_backup_dev_root_gen(root_backup
,
1818 btrfs_header_generation(info
->dev_root
->node
));
1819 btrfs_set_backup_dev_root_level(root_backup
,
1820 btrfs_header_level(info
->dev_root
->node
));
1822 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1823 btrfs_set_backup_csum_root_gen(root_backup
,
1824 btrfs_header_generation(info
->csum_root
->node
));
1825 btrfs_set_backup_csum_root_level(root_backup
,
1826 btrfs_header_level(info
->csum_root
->node
));
1828 btrfs_set_backup_total_bytes(root_backup
,
1829 btrfs_super_total_bytes(info
->super_copy
));
1830 btrfs_set_backup_bytes_used(root_backup
,
1831 btrfs_super_bytes_used(info
->super_copy
));
1832 btrfs_set_backup_num_devices(root_backup
,
1833 btrfs_super_num_devices(info
->super_copy
));
1836 * if we don't copy this out to the super_copy, it won't get remembered
1837 * for the next commit
1839 memcpy(&info
->super_copy
->super_roots
,
1840 &info
->super_for_commit
->super_roots
,
1841 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1845 * this copies info out of the root backup array and back into
1846 * the in-memory super block. It is meant to help iterate through
1847 * the array, so you send it the number of backups you've already
1848 * tried and the last backup index you used.
1850 * this returns -1 when it has tried all the backups
1852 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1853 struct btrfs_super_block
*super
,
1854 int *num_backups_tried
, int *backup_index
)
1856 struct btrfs_root_backup
*root_backup
;
1857 int newest
= *backup_index
;
1859 if (*num_backups_tried
== 0) {
1860 u64 gen
= btrfs_super_generation(super
);
1862 newest
= find_newest_super_backup(info
, gen
);
1866 *backup_index
= newest
;
1867 *num_backups_tried
= 1;
1868 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1869 /* we've tried all the backups, all done */
1872 /* jump to the next oldest backup */
1873 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1874 BTRFS_NUM_BACKUP_ROOTS
;
1875 *backup_index
= newest
;
1876 *num_backups_tried
+= 1;
1878 root_backup
= super
->super_roots
+ newest
;
1880 btrfs_set_super_generation(super
,
1881 btrfs_backup_tree_root_gen(root_backup
));
1882 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1883 btrfs_set_super_root_level(super
,
1884 btrfs_backup_tree_root_level(root_backup
));
1885 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1888 * fixme: the total bytes and num_devices need to match or we should
1891 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1892 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1896 /* helper to cleanup tree roots */
1897 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1899 free_extent_buffer(info
->tree_root
->node
);
1900 free_extent_buffer(info
->tree_root
->commit_root
);
1901 free_extent_buffer(info
->dev_root
->node
);
1902 free_extent_buffer(info
->dev_root
->commit_root
);
1903 free_extent_buffer(info
->extent_root
->node
);
1904 free_extent_buffer(info
->extent_root
->commit_root
);
1905 free_extent_buffer(info
->csum_root
->node
);
1906 free_extent_buffer(info
->csum_root
->commit_root
);
1907 if (info
->quota_root
) {
1908 free_extent_buffer(info
->quota_root
->node
);
1909 free_extent_buffer(info
->quota_root
->commit_root
);
1912 info
->tree_root
->node
= NULL
;
1913 info
->tree_root
->commit_root
= NULL
;
1914 info
->dev_root
->node
= NULL
;
1915 info
->dev_root
->commit_root
= NULL
;
1916 info
->extent_root
->node
= NULL
;
1917 info
->extent_root
->commit_root
= NULL
;
1918 info
->csum_root
->node
= NULL
;
1919 info
->csum_root
->commit_root
= NULL
;
1920 if (info
->quota_root
) {
1921 info
->quota_root
->node
= NULL
;
1922 info
->quota_root
->commit_root
= NULL
;
1926 free_extent_buffer(info
->chunk_root
->node
);
1927 free_extent_buffer(info
->chunk_root
->commit_root
);
1928 info
->chunk_root
->node
= NULL
;
1929 info
->chunk_root
->commit_root
= NULL
;
1934 int open_ctree(struct super_block
*sb
,
1935 struct btrfs_fs_devices
*fs_devices
,
1945 struct btrfs_key location
;
1946 struct buffer_head
*bh
;
1947 struct btrfs_super_block
*disk_super
;
1948 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
1949 struct btrfs_root
*tree_root
;
1950 struct btrfs_root
*extent_root
;
1951 struct btrfs_root
*csum_root
;
1952 struct btrfs_root
*chunk_root
;
1953 struct btrfs_root
*dev_root
;
1954 struct btrfs_root
*quota_root
;
1955 struct btrfs_root
*log_tree_root
;
1958 int num_backups_tried
= 0;
1959 int backup_index
= 0;
1961 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
1962 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
1963 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
1964 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
1965 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
1966 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
1968 if (!tree_root
|| !extent_root
|| !csum_root
||
1969 !chunk_root
|| !dev_root
|| !quota_root
) {
1974 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
1980 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
1986 fs_info
->btree_inode
= new_inode(sb
);
1987 if (!fs_info
->btree_inode
) {
1992 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1994 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
1995 INIT_LIST_HEAD(&fs_info
->trans_list
);
1996 INIT_LIST_HEAD(&fs_info
->dead_roots
);
1997 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
1998 INIT_LIST_HEAD(&fs_info
->hashers
);
1999 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2000 INIT_LIST_HEAD(&fs_info
->ordered_operations
);
2001 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2002 spin_lock_init(&fs_info
->delalloc_lock
);
2003 spin_lock_init(&fs_info
->trans_lock
);
2004 spin_lock_init(&fs_info
->ref_cache_lock
);
2005 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2006 spin_lock_init(&fs_info
->delayed_iput_lock
);
2007 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2008 spin_lock_init(&fs_info
->free_chunk_lock
);
2009 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2010 rwlock_init(&fs_info
->tree_mod_log_lock
);
2011 mutex_init(&fs_info
->reloc_mutex
);
2013 init_completion(&fs_info
->kobj_unregister
);
2014 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2015 INIT_LIST_HEAD(&fs_info
->space_info
);
2016 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2017 btrfs_mapping_init(&fs_info
->mapping_tree
);
2018 btrfs_init_block_rsv(&fs_info
->global_block_rsv
);
2019 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
);
2020 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
);
2021 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
);
2022 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
);
2023 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
);
2024 atomic_set(&fs_info
->nr_async_submits
, 0);
2025 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2026 atomic_set(&fs_info
->async_submit_draining
, 0);
2027 atomic_set(&fs_info
->nr_async_bios
, 0);
2028 atomic_set(&fs_info
->defrag_running
, 0);
2029 atomic_set(&fs_info
->tree_mod_seq
, 0);
2031 fs_info
->max_inline
= 8192 * 1024;
2032 fs_info
->metadata_ratio
= 0;
2033 fs_info
->defrag_inodes
= RB_ROOT
;
2034 fs_info
->trans_no_join
= 0;
2035 fs_info
->free_chunk_space
= 0;
2036 fs_info
->tree_mod_log
= RB_ROOT
;
2038 init_waitqueue_head(&fs_info
->tree_mod_seq_wait
);
2040 /* readahead state */
2041 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2042 spin_lock_init(&fs_info
->reada_lock
);
2044 fs_info
->thread_pool_size
= min_t(unsigned long,
2045 num_online_cpus() + 2, 8);
2047 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2048 spin_lock_init(&fs_info
->ordered_extent_lock
);
2049 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2051 if (!fs_info
->delayed_root
) {
2055 btrfs_init_delayed_root(fs_info
->delayed_root
);
2057 mutex_init(&fs_info
->scrub_lock
);
2058 atomic_set(&fs_info
->scrubs_running
, 0);
2059 atomic_set(&fs_info
->scrub_pause_req
, 0);
2060 atomic_set(&fs_info
->scrubs_paused
, 0);
2061 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2062 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2063 init_rwsem(&fs_info
->scrub_super_lock
);
2064 fs_info
->scrub_workers_refcnt
= 0;
2065 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2066 fs_info
->check_integrity_print_mask
= 0;
2069 spin_lock_init(&fs_info
->balance_lock
);
2070 mutex_init(&fs_info
->balance_mutex
);
2071 atomic_set(&fs_info
->balance_running
, 0);
2072 atomic_set(&fs_info
->balance_pause_req
, 0);
2073 atomic_set(&fs_info
->balance_cancel_req
, 0);
2074 fs_info
->balance_ctl
= NULL
;
2075 init_waitqueue_head(&fs_info
->balance_wait_q
);
2077 sb
->s_blocksize
= 4096;
2078 sb
->s_blocksize_bits
= blksize_bits(4096);
2079 sb
->s_bdi
= &fs_info
->bdi
;
2081 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2082 set_nlink(fs_info
->btree_inode
, 1);
2084 * we set the i_size on the btree inode to the max possible int.
2085 * the real end of the address space is determined by all of
2086 * the devices in the system
2088 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2089 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2090 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2092 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2093 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2094 fs_info
->btree_inode
->i_mapping
);
2095 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2096 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2098 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2100 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2101 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2102 sizeof(struct btrfs_key
));
2103 set_bit(BTRFS_INODE_DUMMY
,
2104 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2105 insert_inode_hash(fs_info
->btree_inode
);
2107 spin_lock_init(&fs_info
->block_group_cache_lock
);
2108 fs_info
->block_group_cache_tree
= RB_ROOT
;
2110 extent_io_tree_init(&fs_info
->freed_extents
[0],
2111 fs_info
->btree_inode
->i_mapping
);
2112 extent_io_tree_init(&fs_info
->freed_extents
[1],
2113 fs_info
->btree_inode
->i_mapping
);
2114 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2115 fs_info
->do_barriers
= 1;
2118 mutex_init(&fs_info
->ordered_operations_mutex
);
2119 mutex_init(&fs_info
->tree_log_mutex
);
2120 mutex_init(&fs_info
->chunk_mutex
);
2121 mutex_init(&fs_info
->transaction_kthread_mutex
);
2122 mutex_init(&fs_info
->cleaner_mutex
);
2123 mutex_init(&fs_info
->volume_mutex
);
2124 init_rwsem(&fs_info
->extent_commit_sem
);
2125 init_rwsem(&fs_info
->cleanup_work_sem
);
2126 init_rwsem(&fs_info
->subvol_sem
);
2128 spin_lock_init(&fs_info
->qgroup_lock
);
2129 fs_info
->qgroup_tree
= RB_ROOT
;
2130 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2131 fs_info
->qgroup_seq
= 1;
2132 fs_info
->quota_enabled
= 0;
2133 fs_info
->pending_quota_state
= 0;
2135 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2136 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2138 init_waitqueue_head(&fs_info
->transaction_throttle
);
2139 init_waitqueue_head(&fs_info
->transaction_wait
);
2140 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2141 init_waitqueue_head(&fs_info
->async_submit_wait
);
2143 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2144 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2146 invalidate_bdev(fs_devices
->latest_bdev
);
2147 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2153 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2154 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2155 sizeof(*fs_info
->super_for_commit
));
2158 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2160 disk_super
= fs_info
->super_copy
;
2161 if (!btrfs_super_root(disk_super
))
2164 /* check FS state, whether FS is broken. */
2165 fs_info
->fs_state
|= btrfs_super_flags(disk_super
);
2167 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2169 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2175 * run through our array of backup supers and setup
2176 * our ring pointer to the oldest one
2178 generation
= btrfs_super_generation(disk_super
);
2179 find_oldest_super_backup(fs_info
, generation
);
2182 * In the long term, we'll store the compression type in the super
2183 * block, and it'll be used for per file compression control.
2185 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2187 ret
= btrfs_parse_options(tree_root
, options
);
2193 features
= btrfs_super_incompat_flags(disk_super
) &
2194 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2196 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2197 "unsupported optional features (%Lx).\n",
2198 (unsigned long long)features
);
2203 if (btrfs_super_leafsize(disk_super
) !=
2204 btrfs_super_nodesize(disk_super
)) {
2205 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2206 "blocksizes don't match. node %d leaf %d\n",
2207 btrfs_super_nodesize(disk_super
),
2208 btrfs_super_leafsize(disk_super
));
2212 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2213 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2214 "blocksize (%d) was too large\n",
2215 btrfs_super_leafsize(disk_super
));
2220 features
= btrfs_super_incompat_flags(disk_super
);
2221 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2222 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2223 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2226 * flag our filesystem as having big metadata blocks if
2227 * they are bigger than the page size
2229 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2230 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2231 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2232 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2235 nodesize
= btrfs_super_nodesize(disk_super
);
2236 leafsize
= btrfs_super_leafsize(disk_super
);
2237 sectorsize
= btrfs_super_sectorsize(disk_super
);
2238 stripesize
= btrfs_super_stripesize(disk_super
);
2241 * mixed block groups end up with duplicate but slightly offset
2242 * extent buffers for the same range. It leads to corruptions
2244 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2245 (sectorsize
!= leafsize
)) {
2246 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2247 "are not allowed for mixed block groups on %s\n",
2252 btrfs_set_super_incompat_flags(disk_super
, features
);
2254 features
= btrfs_super_compat_ro_flags(disk_super
) &
2255 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2256 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2257 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2258 "unsupported option features (%Lx).\n",
2259 (unsigned long long)features
);
2264 btrfs_init_workers(&fs_info
->generic_worker
,
2265 "genwork", 1, NULL
);
2267 btrfs_init_workers(&fs_info
->workers
, "worker",
2268 fs_info
->thread_pool_size
,
2269 &fs_info
->generic_worker
);
2271 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2272 fs_info
->thread_pool_size
,
2273 &fs_info
->generic_worker
);
2275 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2276 min_t(u64
, fs_devices
->num_devices
,
2277 fs_info
->thread_pool_size
),
2278 &fs_info
->generic_worker
);
2280 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2281 2, &fs_info
->generic_worker
);
2283 /* a higher idle thresh on the submit workers makes it much more
2284 * likely that bios will be send down in a sane order to the
2287 fs_info
->submit_workers
.idle_thresh
= 64;
2289 fs_info
->workers
.idle_thresh
= 16;
2290 fs_info
->workers
.ordered
= 1;
2292 fs_info
->delalloc_workers
.idle_thresh
= 2;
2293 fs_info
->delalloc_workers
.ordered
= 1;
2295 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2296 &fs_info
->generic_worker
);
2297 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2298 fs_info
->thread_pool_size
,
2299 &fs_info
->generic_worker
);
2300 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2301 fs_info
->thread_pool_size
,
2302 &fs_info
->generic_worker
);
2303 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2304 "endio-meta-write", fs_info
->thread_pool_size
,
2305 &fs_info
->generic_worker
);
2306 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2307 fs_info
->thread_pool_size
,
2308 &fs_info
->generic_worker
);
2309 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2310 1, &fs_info
->generic_worker
);
2311 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2312 fs_info
->thread_pool_size
,
2313 &fs_info
->generic_worker
);
2314 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2315 fs_info
->thread_pool_size
,
2316 &fs_info
->generic_worker
);
2319 * endios are largely parallel and should have a very
2322 fs_info
->endio_workers
.idle_thresh
= 4;
2323 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2325 fs_info
->endio_write_workers
.idle_thresh
= 2;
2326 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2327 fs_info
->readahead_workers
.idle_thresh
= 2;
2330 * btrfs_start_workers can really only fail because of ENOMEM so just
2331 * return -ENOMEM if any of these fail.
2333 ret
= btrfs_start_workers(&fs_info
->workers
);
2334 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2335 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2336 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2337 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2338 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2339 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2340 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2341 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2342 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2343 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2344 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2345 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2348 goto fail_sb_buffer
;
2351 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2352 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2353 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2355 tree_root
->nodesize
= nodesize
;
2356 tree_root
->leafsize
= leafsize
;
2357 tree_root
->sectorsize
= sectorsize
;
2358 tree_root
->stripesize
= stripesize
;
2360 sb
->s_blocksize
= sectorsize
;
2361 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2363 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
2364 sizeof(disk_super
->magic
))) {
2365 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2366 goto fail_sb_buffer
;
2369 if (sectorsize
!= PAGE_SIZE
) {
2370 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2371 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2372 goto fail_sb_buffer
;
2375 mutex_lock(&fs_info
->chunk_mutex
);
2376 ret
= btrfs_read_sys_array(tree_root
);
2377 mutex_unlock(&fs_info
->chunk_mutex
);
2379 printk(KERN_WARNING
"btrfs: failed to read the system "
2380 "array on %s\n", sb
->s_id
);
2381 goto fail_sb_buffer
;
2384 blocksize
= btrfs_level_size(tree_root
,
2385 btrfs_super_chunk_root_level(disk_super
));
2386 generation
= btrfs_super_chunk_root_generation(disk_super
);
2388 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2389 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2391 chunk_root
->node
= read_tree_block(chunk_root
,
2392 btrfs_super_chunk_root(disk_super
),
2393 blocksize
, generation
);
2394 BUG_ON(!chunk_root
->node
); /* -ENOMEM */
2395 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2396 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2398 goto fail_tree_roots
;
2400 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2401 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2403 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2404 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2407 ret
= btrfs_read_chunk_tree(chunk_root
);
2409 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2411 goto fail_tree_roots
;
2414 btrfs_close_extra_devices(fs_devices
);
2416 if (!fs_devices
->latest_bdev
) {
2417 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2419 goto fail_tree_roots
;
2423 blocksize
= btrfs_level_size(tree_root
,
2424 btrfs_super_root_level(disk_super
));
2425 generation
= btrfs_super_generation(disk_super
);
2427 tree_root
->node
= read_tree_block(tree_root
,
2428 btrfs_super_root(disk_super
),
2429 blocksize
, generation
);
2430 if (!tree_root
->node
||
2431 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2432 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2435 goto recovery_tree_root
;
2438 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2439 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2441 ret
= find_and_setup_root(tree_root
, fs_info
,
2442 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2444 goto recovery_tree_root
;
2445 extent_root
->track_dirty
= 1;
2447 ret
= find_and_setup_root(tree_root
, fs_info
,
2448 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2450 goto recovery_tree_root
;
2451 dev_root
->track_dirty
= 1;
2453 ret
= find_and_setup_root(tree_root
, fs_info
,
2454 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2456 goto recovery_tree_root
;
2457 csum_root
->track_dirty
= 1;
2459 ret
= find_and_setup_root(tree_root
, fs_info
,
2460 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2463 quota_root
= fs_info
->quota_root
= NULL
;
2465 quota_root
->track_dirty
= 1;
2466 fs_info
->quota_enabled
= 1;
2467 fs_info
->pending_quota_state
= 1;
2470 fs_info
->generation
= generation
;
2471 fs_info
->last_trans_committed
= generation
;
2473 ret
= btrfs_recover_balance(fs_info
);
2475 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2476 goto fail_block_groups
;
2479 ret
= btrfs_init_dev_stats(fs_info
);
2481 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2483 goto fail_block_groups
;
2486 ret
= btrfs_init_space_info(fs_info
);
2488 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2489 goto fail_block_groups
;
2492 ret
= btrfs_read_block_groups(extent_root
);
2494 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2495 goto fail_block_groups
;
2498 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2500 if (IS_ERR(fs_info
->cleaner_kthread
))
2501 goto fail_block_groups
;
2503 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2505 "btrfs-transaction");
2506 if (IS_ERR(fs_info
->transaction_kthread
))
2509 if (!btrfs_test_opt(tree_root
, SSD
) &&
2510 !btrfs_test_opt(tree_root
, NOSSD
) &&
2511 !fs_info
->fs_devices
->rotating
) {
2512 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2514 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2517 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2518 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2519 ret
= btrfsic_mount(tree_root
, fs_devices
,
2520 btrfs_test_opt(tree_root
,
2521 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2523 fs_info
->check_integrity_print_mask
);
2525 printk(KERN_WARNING
"btrfs: failed to initialize"
2526 " integrity check module %s\n", sb
->s_id
);
2529 ret
= btrfs_read_qgroup_config(fs_info
);
2531 goto fail_trans_kthread
;
2533 /* do not make disk changes in broken FS */
2534 if (btrfs_super_log_root(disk_super
) != 0 &&
2535 !(fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)) {
2536 u64 bytenr
= btrfs_super_log_root(disk_super
);
2538 if (fs_devices
->rw_devices
== 0) {
2539 printk(KERN_WARNING
"Btrfs log replay required "
2545 btrfs_level_size(tree_root
,
2546 btrfs_super_log_root_level(disk_super
));
2548 log_tree_root
= btrfs_alloc_root(fs_info
);
2549 if (!log_tree_root
) {
2554 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2555 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2557 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2560 /* returns with log_tree_root freed on success */
2561 ret
= btrfs_recover_log_trees(log_tree_root
);
2563 btrfs_error(tree_root
->fs_info
, ret
,
2564 "Failed to recover log tree");
2565 free_extent_buffer(log_tree_root
->node
);
2566 kfree(log_tree_root
);
2567 goto fail_trans_kthread
;
2570 if (sb
->s_flags
& MS_RDONLY
) {
2571 ret
= btrfs_commit_super(tree_root
);
2573 goto fail_trans_kthread
;
2577 ret
= btrfs_find_orphan_roots(tree_root
);
2579 goto fail_trans_kthread
;
2581 if (!(sb
->s_flags
& MS_RDONLY
)) {
2582 ret
= btrfs_cleanup_fs_roots(fs_info
);
2584 goto fail_trans_kthread
;
2586 ret
= btrfs_recover_relocation(tree_root
);
2589 "btrfs: failed to recover relocation\n");
2595 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2596 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2597 location
.offset
= (u64
)-1;
2599 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2600 if (!fs_info
->fs_root
)
2602 if (IS_ERR(fs_info
->fs_root
)) {
2603 err
= PTR_ERR(fs_info
->fs_root
);
2607 if (sb
->s_flags
& MS_RDONLY
)
2610 down_read(&fs_info
->cleanup_work_sem
);
2611 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2612 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2613 up_read(&fs_info
->cleanup_work_sem
);
2614 close_ctree(tree_root
);
2617 up_read(&fs_info
->cleanup_work_sem
);
2619 ret
= btrfs_resume_balance_async(fs_info
);
2621 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2622 close_ctree(tree_root
);
2629 btrfs_free_qgroup_config(fs_info
);
2631 kthread_stop(fs_info
->transaction_kthread
);
2633 kthread_stop(fs_info
->cleaner_kthread
);
2636 * make sure we're done with the btree inode before we stop our
2639 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2640 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2643 btrfs_free_block_groups(fs_info
);
2646 free_root_pointers(fs_info
, 1);
2649 btrfs_stop_workers(&fs_info
->generic_worker
);
2650 btrfs_stop_workers(&fs_info
->readahead_workers
);
2651 btrfs_stop_workers(&fs_info
->fixup_workers
);
2652 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2653 btrfs_stop_workers(&fs_info
->workers
);
2654 btrfs_stop_workers(&fs_info
->endio_workers
);
2655 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2656 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2657 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2658 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2659 btrfs_stop_workers(&fs_info
->submit_workers
);
2660 btrfs_stop_workers(&fs_info
->delayed_workers
);
2661 btrfs_stop_workers(&fs_info
->caching_workers
);
2664 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2666 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2667 iput(fs_info
->btree_inode
);
2669 bdi_destroy(&fs_info
->bdi
);
2671 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2673 btrfs_close_devices(fs_info
->fs_devices
);
2677 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2678 goto fail_tree_roots
;
2680 free_root_pointers(fs_info
, 0);
2682 /* don't use the log in recovery mode, it won't be valid */
2683 btrfs_set_super_log_root(disk_super
, 0);
2685 /* we can't trust the free space cache either */
2686 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2688 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2689 &num_backups_tried
, &backup_index
);
2691 goto fail_block_groups
;
2692 goto retry_root_backup
;
2695 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2698 set_buffer_uptodate(bh
);
2700 struct btrfs_device
*device
= (struct btrfs_device
*)
2703 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2704 "I/O error on %s\n",
2705 rcu_str_deref(device
->name
));
2706 /* note, we dont' set_buffer_write_io_error because we have
2707 * our own ways of dealing with the IO errors
2709 clear_buffer_uptodate(bh
);
2710 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2716 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2718 struct buffer_head
*bh
;
2719 struct buffer_head
*latest
= NULL
;
2720 struct btrfs_super_block
*super
;
2725 /* we would like to check all the supers, but that would make
2726 * a btrfs mount succeed after a mkfs from a different FS.
2727 * So, we need to add a special mount option to scan for
2728 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2730 for (i
= 0; i
< 1; i
++) {
2731 bytenr
= btrfs_sb_offset(i
);
2732 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2734 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2738 super
= (struct btrfs_super_block
*)bh
->b_data
;
2739 if (btrfs_super_bytenr(super
) != bytenr
||
2740 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
2741 sizeof(super
->magic
))) {
2746 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2749 transid
= btrfs_super_generation(super
);
2758 * this should be called twice, once with wait == 0 and
2759 * once with wait == 1. When wait == 0 is done, all the buffer heads
2760 * we write are pinned.
2762 * They are released when wait == 1 is done.
2763 * max_mirrors must be the same for both runs, and it indicates how
2764 * many supers on this one device should be written.
2766 * max_mirrors == 0 means to write them all.
2768 static int write_dev_supers(struct btrfs_device
*device
,
2769 struct btrfs_super_block
*sb
,
2770 int do_barriers
, int wait
, int max_mirrors
)
2772 struct buffer_head
*bh
;
2779 if (max_mirrors
== 0)
2780 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2782 for (i
= 0; i
< max_mirrors
; i
++) {
2783 bytenr
= btrfs_sb_offset(i
);
2784 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2788 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2789 BTRFS_SUPER_INFO_SIZE
);
2792 if (!buffer_uptodate(bh
))
2795 /* drop our reference */
2798 /* drop the reference from the wait == 0 run */
2802 btrfs_set_super_bytenr(sb
, bytenr
);
2805 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2806 BTRFS_CSUM_SIZE
, crc
,
2807 BTRFS_SUPER_INFO_SIZE
-
2809 btrfs_csum_final(crc
, sb
->csum
);
2812 * one reference for us, and we leave it for the
2815 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2816 BTRFS_SUPER_INFO_SIZE
);
2817 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2819 /* one reference for submit_bh */
2822 set_buffer_uptodate(bh
);
2824 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2825 bh
->b_private
= device
;
2829 * we fua the first super. The others we allow
2832 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
2836 return errors
< i
? 0 : -1;
2840 * endio for the write_dev_flush, this will wake anyone waiting
2841 * for the barrier when it is done
2843 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
2846 if (err
== -EOPNOTSUPP
)
2847 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
2848 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2850 if (bio
->bi_private
)
2851 complete(bio
->bi_private
);
2856 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2857 * sent down. With wait == 1, it waits for the previous flush.
2859 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2862 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
2867 if (device
->nobarriers
)
2871 bio
= device
->flush_bio
;
2875 wait_for_completion(&device
->flush_wait
);
2877 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
2878 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2879 rcu_str_deref(device
->name
));
2880 device
->nobarriers
= 1;
2882 if (!bio_flagged(bio
, BIO_UPTODATE
)) {
2884 if (!bio_flagged(bio
, BIO_EOPNOTSUPP
))
2885 btrfs_dev_stat_inc_and_print(device
,
2886 BTRFS_DEV_STAT_FLUSH_ERRS
);
2889 /* drop the reference from the wait == 0 run */
2891 device
->flush_bio
= NULL
;
2897 * one reference for us, and we leave it for the
2900 device
->flush_bio
= NULL
;
2901 bio
= bio_alloc(GFP_NOFS
, 0);
2905 bio
->bi_end_io
= btrfs_end_empty_barrier
;
2906 bio
->bi_bdev
= device
->bdev
;
2907 init_completion(&device
->flush_wait
);
2908 bio
->bi_private
= &device
->flush_wait
;
2909 device
->flush_bio
= bio
;
2912 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
2918 * send an empty flush down to each device in parallel,
2919 * then wait for them
2921 static int barrier_all_devices(struct btrfs_fs_info
*info
)
2923 struct list_head
*head
;
2924 struct btrfs_device
*dev
;
2928 /* send down all the barriers */
2929 head
= &info
->fs_devices
->devices
;
2930 list_for_each_entry_rcu(dev
, head
, dev_list
) {
2935 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2938 ret
= write_dev_flush(dev
, 0);
2943 /* wait for all the barriers */
2944 list_for_each_entry_rcu(dev
, head
, dev_list
) {
2949 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2952 ret
= write_dev_flush(dev
, 1);
2961 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
2963 struct list_head
*head
;
2964 struct btrfs_device
*dev
;
2965 struct btrfs_super_block
*sb
;
2966 struct btrfs_dev_item
*dev_item
;
2970 int total_errors
= 0;
2973 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
2974 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
2975 backup_super_roots(root
->fs_info
);
2977 sb
= root
->fs_info
->super_for_commit
;
2978 dev_item
= &sb
->dev_item
;
2980 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2981 head
= &root
->fs_info
->fs_devices
->devices
;
2984 barrier_all_devices(root
->fs_info
);
2986 list_for_each_entry_rcu(dev
, head
, dev_list
) {
2991 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2994 btrfs_set_stack_device_generation(dev_item
, 0);
2995 btrfs_set_stack_device_type(dev_item
, dev
->type
);
2996 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
2997 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
2998 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
2999 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3000 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3001 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3002 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3003 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3005 flags
= btrfs_super_flags(sb
);
3006 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3008 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3012 if (total_errors
> max_errors
) {
3013 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3016 /* This shouldn't happen. FUA is masked off if unsupported */
3021 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3024 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3027 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3031 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3032 if (total_errors
> max_errors
) {
3033 btrfs_error(root
->fs_info
, -EIO
,
3034 "%d errors while writing supers", total_errors
);
3040 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3041 struct btrfs_root
*root
, int max_mirrors
)
3045 ret
= write_all_supers(root
, max_mirrors
);
3049 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3051 spin_lock(&fs_info
->fs_roots_radix_lock
);
3052 radix_tree_delete(&fs_info
->fs_roots_radix
,
3053 (unsigned long)root
->root_key
.objectid
);
3054 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3056 if (btrfs_root_refs(&root
->root_item
) == 0)
3057 synchronize_srcu(&fs_info
->subvol_srcu
);
3059 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3060 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3064 static void free_fs_root(struct btrfs_root
*root
)
3066 iput(root
->cache_inode
);
3067 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3069 free_anon_bdev(root
->anon_dev
);
3070 free_extent_buffer(root
->node
);
3071 free_extent_buffer(root
->commit_root
);
3072 kfree(root
->free_ino_ctl
);
3073 kfree(root
->free_ino_pinned
);
3078 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
3081 struct btrfs_root
*gang
[8];
3084 while (!list_empty(&fs_info
->dead_roots
)) {
3085 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
3086 struct btrfs_root
, root_list
);
3087 list_del(&gang
[0]->root_list
);
3089 if (gang
[0]->in_radix
) {
3090 btrfs_free_fs_root(fs_info
, gang
[0]);
3092 free_extent_buffer(gang
[0]->node
);
3093 free_extent_buffer(gang
[0]->commit_root
);
3099 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3104 for (i
= 0; i
< ret
; i
++)
3105 btrfs_free_fs_root(fs_info
, gang
[i
]);
3109 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3111 u64 root_objectid
= 0;
3112 struct btrfs_root
*gang
[8];
3117 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3118 (void **)gang
, root_objectid
,
3123 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3124 for (i
= 0; i
< ret
; i
++) {
3127 root_objectid
= gang
[i
]->root_key
.objectid
;
3128 err
= btrfs_orphan_cleanup(gang
[i
]);
3137 int btrfs_commit_super(struct btrfs_root
*root
)
3139 struct btrfs_trans_handle
*trans
;
3142 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3143 btrfs_run_delayed_iputs(root
);
3144 btrfs_clean_old_snapshots(root
);
3145 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3147 /* wait until ongoing cleanup work done */
3148 down_write(&root
->fs_info
->cleanup_work_sem
);
3149 up_write(&root
->fs_info
->cleanup_work_sem
);
3151 trans
= btrfs_join_transaction(root
);
3153 return PTR_ERR(trans
);
3154 ret
= btrfs_commit_transaction(trans
, root
);
3157 /* run commit again to drop the original snapshot */
3158 trans
= btrfs_join_transaction(root
);
3160 return PTR_ERR(trans
);
3161 ret
= btrfs_commit_transaction(trans
, root
);
3164 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3166 btrfs_error(root
->fs_info
, ret
,
3167 "Failed to sync btree inode to disk.");
3171 ret
= write_ctree_super(NULL
, root
, 0);
3175 int close_ctree(struct btrfs_root
*root
)
3177 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3180 fs_info
->closing
= 1;
3183 /* pause restriper - we want to resume on mount */
3184 btrfs_pause_balance(root
->fs_info
);
3186 btrfs_scrub_cancel(root
);
3188 /* wait for any defraggers to finish */
3189 wait_event(fs_info
->transaction_wait
,
3190 (atomic_read(&fs_info
->defrag_running
) == 0));
3192 /* clear out the rbtree of defraggable inodes */
3193 btrfs_run_defrag_inodes(fs_info
);
3196 * Here come 2 situations when btrfs is broken to flip readonly:
3198 * 1. when btrfs flips readonly somewhere else before
3199 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3200 * and btrfs will skip to write sb directly to keep
3201 * ERROR state on disk.
3203 * 2. when btrfs flips readonly just in btrfs_commit_super,
3204 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3205 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3206 * btrfs will cleanup all FS resources first and write sb then.
3208 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3209 ret
= btrfs_commit_super(root
);
3211 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3214 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
3215 ret
= btrfs_error_commit_super(root
);
3217 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3220 btrfs_put_block_group_cache(fs_info
);
3222 kthread_stop(fs_info
->transaction_kthread
);
3223 kthread_stop(fs_info
->cleaner_kthread
);
3225 fs_info
->closing
= 2;
3228 btrfs_free_qgroup_config(root
->fs_info
);
3230 if (fs_info
->delalloc_bytes
) {
3231 printk(KERN_INFO
"btrfs: at unmount delalloc count %llu\n",
3232 (unsigned long long)fs_info
->delalloc_bytes
);
3234 if (fs_info
->total_ref_cache_size
) {
3235 printk(KERN_INFO
"btrfs: at umount reference cache size %llu\n",
3236 (unsigned long long)fs_info
->total_ref_cache_size
);
3239 free_extent_buffer(fs_info
->extent_root
->node
);
3240 free_extent_buffer(fs_info
->extent_root
->commit_root
);
3241 free_extent_buffer(fs_info
->tree_root
->node
);
3242 free_extent_buffer(fs_info
->tree_root
->commit_root
);
3243 free_extent_buffer(fs_info
->chunk_root
->node
);
3244 free_extent_buffer(fs_info
->chunk_root
->commit_root
);
3245 free_extent_buffer(fs_info
->dev_root
->node
);
3246 free_extent_buffer(fs_info
->dev_root
->commit_root
);
3247 free_extent_buffer(fs_info
->csum_root
->node
);
3248 free_extent_buffer(fs_info
->csum_root
->commit_root
);
3249 if (fs_info
->quota_root
) {
3250 free_extent_buffer(fs_info
->quota_root
->node
);
3251 free_extent_buffer(fs_info
->quota_root
->commit_root
);
3254 btrfs_free_block_groups(fs_info
);
3256 del_fs_roots(fs_info
);
3258 iput(fs_info
->btree_inode
);
3260 btrfs_stop_workers(&fs_info
->generic_worker
);
3261 btrfs_stop_workers(&fs_info
->fixup_workers
);
3262 btrfs_stop_workers(&fs_info
->delalloc_workers
);
3263 btrfs_stop_workers(&fs_info
->workers
);
3264 btrfs_stop_workers(&fs_info
->endio_workers
);
3265 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
3266 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
3267 btrfs_stop_workers(&fs_info
->endio_write_workers
);
3268 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
3269 btrfs_stop_workers(&fs_info
->submit_workers
);
3270 btrfs_stop_workers(&fs_info
->delayed_workers
);
3271 btrfs_stop_workers(&fs_info
->caching_workers
);
3272 btrfs_stop_workers(&fs_info
->readahead_workers
);
3274 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3275 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3276 btrfsic_unmount(root
, fs_info
->fs_devices
);
3279 btrfs_close_devices(fs_info
->fs_devices
);
3280 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3282 bdi_destroy(&fs_info
->bdi
);
3283 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3288 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3292 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3294 ret
= extent_buffer_uptodate(buf
);
3298 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3299 parent_transid
, atomic
);
3305 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3307 return set_extent_buffer_uptodate(buf
);
3310 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3312 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3313 u64 transid
= btrfs_header_generation(buf
);
3316 btrfs_assert_tree_locked(buf
);
3317 if (transid
!= root
->fs_info
->generation
) {
3318 printk(KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3319 "found %llu running %llu\n",
3320 (unsigned long long)buf
->start
,
3321 (unsigned long long)transid
,
3322 (unsigned long long)root
->fs_info
->generation
);
3325 was_dirty
= set_extent_buffer_dirty(buf
);
3327 spin_lock(&root
->fs_info
->delalloc_lock
);
3328 root
->fs_info
->dirty_metadata_bytes
+= buf
->len
;
3329 spin_unlock(&root
->fs_info
->delalloc_lock
);
3333 void btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
3336 * looks as though older kernels can get into trouble with
3337 * this code, they end up stuck in balance_dirty_pages forever
3340 unsigned long thresh
= 32 * 1024 * 1024;
3342 if (current
->flags
& PF_MEMALLOC
)
3345 btrfs_balance_delayed_items(root
);
3347 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
3349 if (num_dirty
> thresh
) {
3350 balance_dirty_pages_ratelimited_nr(
3351 root
->fs_info
->btree_inode
->i_mapping
, 1);
3356 void __btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
3359 * looks as though older kernels can get into trouble with
3360 * this code, they end up stuck in balance_dirty_pages forever
3363 unsigned long thresh
= 32 * 1024 * 1024;
3365 if (current
->flags
& PF_MEMALLOC
)
3368 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
3370 if (num_dirty
> thresh
) {
3371 balance_dirty_pages_ratelimited_nr(
3372 root
->fs_info
->btree_inode
->i_mapping
, 1);
3377 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3379 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3380 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3383 int btree_lock_page_hook(struct page
*page
, void *data
,
3384 void (*flush_fn
)(void *))
3386 struct inode
*inode
= page
->mapping
->host
;
3387 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3388 struct extent_buffer
*eb
;
3391 * We culled this eb but the page is still hanging out on the mapping,
3394 if (!PagePrivate(page
))
3397 eb
= (struct extent_buffer
*)page
->private;
3402 if (page
!= eb
->pages
[0])
3405 if (!btrfs_try_tree_write_lock(eb
)) {
3407 btrfs_tree_lock(eb
);
3409 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
3411 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
3412 spin_lock(&root
->fs_info
->delalloc_lock
);
3413 if (root
->fs_info
->dirty_metadata_bytes
>= eb
->len
)
3414 root
->fs_info
->dirty_metadata_bytes
-= eb
->len
;
3417 spin_unlock(&root
->fs_info
->delalloc_lock
);
3420 btrfs_tree_unlock(eb
);
3422 if (!trylock_page(page
)) {
3429 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3432 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3433 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3440 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
3441 printk(KERN_WARNING
"warning: mount fs with errors, "
3442 "running btrfsck is recommended\n");
3448 int btrfs_error_commit_super(struct btrfs_root
*root
)
3452 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3453 btrfs_run_delayed_iputs(root
);
3454 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3456 down_write(&root
->fs_info
->cleanup_work_sem
);
3457 up_write(&root
->fs_info
->cleanup_work_sem
);
3459 /* cleanup FS via transaction */
3460 btrfs_cleanup_transaction(root
);
3462 ret
= write_ctree_super(NULL
, root
, 0);
3467 static void btrfs_destroy_ordered_operations(struct btrfs_root
*root
)
3469 struct btrfs_inode
*btrfs_inode
;
3470 struct list_head splice
;
3472 INIT_LIST_HEAD(&splice
);
3474 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3475 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3477 list_splice_init(&root
->fs_info
->ordered_operations
, &splice
);
3478 while (!list_empty(&splice
)) {
3479 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3480 ordered_operations
);
3482 list_del_init(&btrfs_inode
->ordered_operations
);
3484 btrfs_invalidate_inodes(btrfs_inode
->root
);
3487 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3488 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3491 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3493 struct list_head splice
;
3494 struct btrfs_ordered_extent
*ordered
;
3495 struct inode
*inode
;
3497 INIT_LIST_HEAD(&splice
);
3499 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3501 list_splice_init(&root
->fs_info
->ordered_extents
, &splice
);
3502 while (!list_empty(&splice
)) {
3503 ordered
= list_entry(splice
.next
, struct btrfs_ordered_extent
,
3506 list_del_init(&ordered
->root_extent_list
);
3507 atomic_inc(&ordered
->refs
);
3509 /* the inode may be getting freed (in sys_unlink path). */
3510 inode
= igrab(ordered
->inode
);
3512 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3516 atomic_set(&ordered
->refs
, 1);
3517 btrfs_put_ordered_extent(ordered
);
3519 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3522 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3525 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3526 struct btrfs_root
*root
)
3528 struct rb_node
*node
;
3529 struct btrfs_delayed_ref_root
*delayed_refs
;
3530 struct btrfs_delayed_ref_node
*ref
;
3533 delayed_refs
= &trans
->delayed_refs
;
3535 spin_lock(&delayed_refs
->lock
);
3536 if (delayed_refs
->num_entries
== 0) {
3537 spin_unlock(&delayed_refs
->lock
);
3538 printk(KERN_INFO
"delayed_refs has NO entry\n");
3542 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3543 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3545 atomic_set(&ref
->refs
, 1);
3546 if (btrfs_delayed_ref_is_head(ref
)) {
3547 struct btrfs_delayed_ref_head
*head
;
3549 head
= btrfs_delayed_node_to_head(ref
);
3550 if (!mutex_trylock(&head
->mutex
)) {
3551 atomic_inc(&ref
->refs
);
3552 spin_unlock(&delayed_refs
->lock
);
3554 /* Need to wait for the delayed ref to run */
3555 mutex_lock(&head
->mutex
);
3556 mutex_unlock(&head
->mutex
);
3557 btrfs_put_delayed_ref(ref
);
3559 spin_lock(&delayed_refs
->lock
);
3563 kfree(head
->extent_op
);
3564 delayed_refs
->num_heads
--;
3565 if (list_empty(&head
->cluster
))
3566 delayed_refs
->num_heads_ready
--;
3567 list_del_init(&head
->cluster
);
3570 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3571 delayed_refs
->num_entries
--;
3573 spin_unlock(&delayed_refs
->lock
);
3574 btrfs_put_delayed_ref(ref
);
3577 spin_lock(&delayed_refs
->lock
);
3580 spin_unlock(&delayed_refs
->lock
);
3585 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
)
3587 struct btrfs_pending_snapshot
*snapshot
;
3588 struct list_head splice
;
3590 INIT_LIST_HEAD(&splice
);
3592 list_splice_init(&t
->pending_snapshots
, &splice
);
3594 while (!list_empty(&splice
)) {
3595 snapshot
= list_entry(splice
.next
,
3596 struct btrfs_pending_snapshot
,
3599 list_del_init(&snapshot
->list
);
3605 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3607 struct btrfs_inode
*btrfs_inode
;
3608 struct list_head splice
;
3610 INIT_LIST_HEAD(&splice
);
3612 spin_lock(&root
->fs_info
->delalloc_lock
);
3613 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3615 while (!list_empty(&splice
)) {
3616 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3619 list_del_init(&btrfs_inode
->delalloc_inodes
);
3621 btrfs_invalidate_inodes(btrfs_inode
->root
);
3624 spin_unlock(&root
->fs_info
->delalloc_lock
);
3627 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3628 struct extent_io_tree
*dirty_pages
,
3633 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
3634 struct extent_buffer
*eb
;
3638 unsigned long index
;
3641 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3646 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3647 while (start
<= end
) {
3648 index
= start
>> PAGE_CACHE_SHIFT
;
3649 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
3650 page
= find_get_page(btree_inode
->i_mapping
, index
);
3653 offset
= page_offset(page
);
3655 spin_lock(&dirty_pages
->buffer_lock
);
3656 eb
= radix_tree_lookup(
3657 &(&BTRFS_I(page
->mapping
->host
)->io_tree
)->buffer
,
3658 offset
>> PAGE_CACHE_SHIFT
);
3659 spin_unlock(&dirty_pages
->buffer_lock
);
3661 ret
= test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3663 if (PageWriteback(page
))
3664 end_page_writeback(page
);
3667 if (PageDirty(page
)) {
3668 clear_page_dirty_for_io(page
);
3669 spin_lock_irq(&page
->mapping
->tree_lock
);
3670 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3672 PAGECACHE_TAG_DIRTY
);
3673 spin_unlock_irq(&page
->mapping
->tree_lock
);
3677 page_cache_release(page
);
3684 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3685 struct extent_io_tree
*pinned_extents
)
3687 struct extent_io_tree
*unpin
;
3693 unpin
= pinned_extents
;
3696 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3702 if (btrfs_test_opt(root
, DISCARD
))
3703 ret
= btrfs_error_discard_extent(root
, start
,
3707 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3708 btrfs_error_unpin_extent_range(root
, start
, end
);
3713 if (unpin
== &root
->fs_info
->freed_extents
[0])
3714 unpin
= &root
->fs_info
->freed_extents
[1];
3716 unpin
= &root
->fs_info
->freed_extents
[0];
3724 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3725 struct btrfs_root
*root
)
3727 btrfs_destroy_delayed_refs(cur_trans
, root
);
3728 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3729 cur_trans
->dirty_pages
.dirty_bytes
);
3731 /* FIXME: cleanup wait for commit */
3732 cur_trans
->in_commit
= 1;
3733 cur_trans
->blocked
= 1;
3734 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3736 cur_trans
->blocked
= 0;
3737 wake_up(&root
->fs_info
->transaction_wait
);
3739 cur_trans
->commit_done
= 1;
3740 wake_up(&cur_trans
->commit_wait
);
3742 btrfs_destroy_delayed_inodes(root
);
3743 btrfs_assert_delayed_root_empty(root
);
3745 btrfs_destroy_pending_snapshots(cur_trans
);
3747 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3749 btrfs_destroy_pinned_extent(root
,
3750 root
->fs_info
->pinned_extents
);
3753 memset(cur_trans, 0, sizeof(*cur_trans));
3754 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3758 int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3760 struct btrfs_transaction
*t
;
3763 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3765 spin_lock(&root
->fs_info
->trans_lock
);
3766 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3767 root
->fs_info
->trans_no_join
= 1;
3768 spin_unlock(&root
->fs_info
->trans_lock
);
3770 while (!list_empty(&list
)) {
3771 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3775 btrfs_destroy_ordered_operations(root
);
3777 btrfs_destroy_ordered_extents(root
);
3779 btrfs_destroy_delayed_refs(t
, root
);
3781 btrfs_block_rsv_release(root
,
3782 &root
->fs_info
->trans_block_rsv
,
3783 t
->dirty_pages
.dirty_bytes
);
3785 /* FIXME: cleanup wait for commit */
3788 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3789 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3792 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3793 wake_up(&root
->fs_info
->transaction_wait
);
3796 if (waitqueue_active(&t
->commit_wait
))
3797 wake_up(&t
->commit_wait
);
3799 btrfs_destroy_delayed_inodes(root
);
3800 btrfs_assert_delayed_root_empty(root
);
3802 btrfs_destroy_pending_snapshots(t
);
3804 btrfs_destroy_delalloc_inodes(root
);
3806 spin_lock(&root
->fs_info
->trans_lock
);
3807 root
->fs_info
->running_transaction
= NULL
;
3808 spin_unlock(&root
->fs_info
->trans_lock
);
3810 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3813 btrfs_destroy_pinned_extent(root
,
3814 root
->fs_info
->pinned_extents
);
3816 atomic_set(&t
->use_count
, 0);
3817 list_del_init(&t
->list
);
3818 memset(t
, 0, sizeof(*t
));
3819 kmem_cache_free(btrfs_transaction_cachep
, t
);
3822 spin_lock(&root
->fs_info
->trans_lock
);
3823 root
->fs_info
->trans_no_join
= 0;
3824 spin_unlock(&root
->fs_info
->trans_lock
);
3825 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3830 static struct extent_io_ops btree_extent_io_ops
= {
3831 .write_cache_pages_lock_hook
= btree_lock_page_hook
,
3832 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3833 .readpage_io_failed_hook
= btree_io_failed_hook
,
3834 .submit_bio_hook
= btree_submit_bio_hook
,
3835 /* note we're sharing with inode.c for the merge bio hook */
3836 .merge_bio_hook
= btrfs_merge_bio_hook
,