2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include <asm/cpufeature.h>
55 static struct extent_io_ops btree_extent_io_ops
;
56 static void end_workqueue_fn(struct btrfs_work
*work
);
57 static void free_fs_root(struct btrfs_root
*root
);
58 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
60 static void btrfs_destroy_ordered_operations(struct btrfs_root
*root
);
61 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
62 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
63 struct btrfs_root
*root
);
64 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
66 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
67 struct extent_io_tree
*dirty_pages
,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
70 struct extent_io_tree
*pinned_extents
);
73 * end_io_wq structs are used to do processing in task context when an IO is
74 * complete. This is used during reads to verify checksums, and it is used
75 * by writes to insert metadata for new file extents after IO is complete.
81 struct btrfs_fs_info
*info
;
84 struct list_head list
;
85 struct btrfs_work work
;
89 * async submit bios are used to offload expensive checksumming
90 * onto the worker threads. They checksum file and metadata bios
91 * just before they are sent down the IO stack.
93 struct async_submit_bio
{
96 struct list_head list
;
97 extent_submit_bio_hook_t
*submit_bio_start
;
98 extent_submit_bio_hook_t
*submit_bio_done
;
101 unsigned long bio_flags
;
103 * bio_offset is optional, can be used if the pages in the bio
104 * can't tell us where in the file the bio should go
107 struct btrfs_work work
;
112 * Lockdep class keys for extent_buffer->lock's in this root. For a given
113 * eb, the lockdep key is determined by the btrfs_root it belongs to and
114 * the level the eb occupies in the tree.
116 * Different roots are used for different purposes and may nest inside each
117 * other and they require separate keysets. As lockdep keys should be
118 * static, assign keysets according to the purpose of the root as indicated
119 * by btrfs_root->objectid. This ensures that all special purpose roots
120 * have separate keysets.
122 * Lock-nesting across peer nodes is always done with the immediate parent
123 * node locked thus preventing deadlock. As lockdep doesn't know this, use
124 * subclass to avoid triggering lockdep warning in such cases.
126 * The key is set by the readpage_end_io_hook after the buffer has passed
127 * csum validation but before the pages are unlocked. It is also set by
128 * btrfs_init_new_buffer on freshly allocated blocks.
130 * We also add a check to make sure the highest level of the tree is the
131 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
132 * needs update as well.
134 #ifdef CONFIG_DEBUG_LOCK_ALLOC
135 # if BTRFS_MAX_LEVEL != 8
139 static struct btrfs_lockdep_keyset
{
140 u64 id
; /* root objectid */
141 const char *name_stem
; /* lock name stem */
142 char names
[BTRFS_MAX_LEVEL
+ 1][20];
143 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
144 } btrfs_lockdep_keysets
[] = {
145 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
146 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
147 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
148 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
149 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
150 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
151 { .id
= BTRFS_ORPHAN_OBJECTID
, .name_stem
= "orphan" },
152 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
153 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
154 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
155 { .id
= 0, .name_stem
= "tree" },
158 void __init
btrfs_init_lockdep(void)
162 /* initialize lockdep class names */
163 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
164 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
166 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
167 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
168 "btrfs-%s-%02d", ks
->name_stem
, j
);
172 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
175 struct btrfs_lockdep_keyset
*ks
;
177 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
179 /* find the matching keyset, id 0 is the default entry */
180 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
181 if (ks
->id
== objectid
)
184 lockdep_set_class_and_name(&eb
->lock
,
185 &ks
->keys
[level
], ks
->names
[level
]);
191 * extents on the btree inode are pretty simple, there's one extent
192 * that covers the entire device
194 static struct extent_map
*btree_get_extent(struct inode
*inode
,
195 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
198 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
199 struct extent_map
*em
;
202 read_lock(&em_tree
->lock
);
203 em
= lookup_extent_mapping(em_tree
, start
, len
);
206 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
207 read_unlock(&em_tree
->lock
);
210 read_unlock(&em_tree
->lock
);
212 em
= alloc_extent_map();
214 em
= ERR_PTR(-ENOMEM
);
219 em
->block_len
= (u64
)-1;
221 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
223 write_lock(&em_tree
->lock
);
224 ret
= add_extent_mapping(em_tree
, em
);
225 if (ret
== -EEXIST
) {
227 em
= lookup_extent_mapping(em_tree
, start
, len
);
234 write_unlock(&em_tree
->lock
);
240 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
242 return crc32c(seed
, data
, len
);
245 void btrfs_csum_final(u32 crc
, char *result
)
247 put_unaligned_le32(~crc
, result
);
251 * compute the csum for a btree block, and either verify it or write it
252 * into the csum field of the block.
254 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
257 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
260 unsigned long cur_len
;
261 unsigned long offset
= BTRFS_CSUM_SIZE
;
263 unsigned long map_start
;
264 unsigned long map_len
;
267 unsigned long inline_result
;
269 len
= buf
->len
- offset
;
271 err
= map_private_extent_buffer(buf
, offset
, 32,
272 &kaddr
, &map_start
, &map_len
);
275 cur_len
= min(len
, map_len
- (offset
- map_start
));
276 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
281 if (csum_size
> sizeof(inline_result
)) {
282 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
286 result
= (char *)&inline_result
;
289 btrfs_csum_final(crc
, result
);
292 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
295 memcpy(&found
, result
, csum_size
);
297 read_extent_buffer(buf
, &val
, 0, csum_size
);
298 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
299 "failed on %llu wanted %X found %X "
301 root
->fs_info
->sb
->s_id
,
302 (unsigned long long)buf
->start
, val
, found
,
303 btrfs_header_level(buf
));
304 if (result
!= (char *)&inline_result
)
309 write_extent_buffer(buf
, result
, 0, csum_size
);
311 if (result
!= (char *)&inline_result
)
317 * we can't consider a given block up to date unless the transid of the
318 * block matches the transid in the parent node's pointer. This is how we
319 * detect blocks that either didn't get written at all or got written
320 * in the wrong place.
322 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
323 struct extent_buffer
*eb
, u64 parent_transid
,
326 struct extent_state
*cached_state
= NULL
;
329 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
335 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
337 if (extent_buffer_uptodate(eb
) &&
338 btrfs_header_generation(eb
) == parent_transid
) {
342 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
344 (unsigned long long)eb
->start
,
345 (unsigned long long)parent_transid
,
346 (unsigned long long)btrfs_header_generation(eb
));
348 clear_extent_buffer_uptodate(eb
);
350 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
351 &cached_state
, GFP_NOFS
);
356 * helper to read a given tree block, doing retries as required when
357 * the checksums don't match and we have alternate mirrors to try.
359 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
360 struct extent_buffer
*eb
,
361 u64 start
, u64 parent_transid
)
363 struct extent_io_tree
*io_tree
;
368 int failed_mirror
= 0;
370 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
371 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
373 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
375 btree_get_extent
, mirror_num
);
377 if (!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
,
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 (!PageUptodate(page
)) {
442 csum_tree_block(root
, eb
, 0);
446 static int check_tree_block_fsid(struct btrfs_root
*root
,
447 struct extent_buffer
*eb
)
449 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
450 u8 fsid
[BTRFS_UUID_SIZE
];
453 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
456 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
460 fs_devices
= fs_devices
->seed
;
465 #define CORRUPT(reason, eb, root, slot) \
466 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
467 "root=%llu, slot=%d\n", reason, \
468 (unsigned long long)btrfs_header_bytenr(eb), \
469 (unsigned long long)root->objectid, slot)
471 static noinline
int check_leaf(struct btrfs_root
*root
,
472 struct extent_buffer
*leaf
)
474 struct btrfs_key key
;
475 struct btrfs_key leaf_key
;
476 u32 nritems
= btrfs_header_nritems(leaf
);
482 /* Check the 0 item */
483 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
484 BTRFS_LEAF_DATA_SIZE(root
)) {
485 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
490 * Check to make sure each items keys are in the correct order and their
491 * offsets make sense. We only have to loop through nritems-1 because
492 * we check the current slot against the next slot, which verifies the
493 * next slot's offset+size makes sense and that the current's slot
496 for (slot
= 0; slot
< nritems
- 1; slot
++) {
497 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
498 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
500 /* Make sure the keys are in the right order */
501 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
502 CORRUPT("bad key order", leaf
, root
, slot
);
507 * Make sure the offset and ends are right, remember that the
508 * item data starts at the end of the leaf and grows towards the
511 if (btrfs_item_offset_nr(leaf
, slot
) !=
512 btrfs_item_end_nr(leaf
, slot
+ 1)) {
513 CORRUPT("slot offset bad", leaf
, root
, slot
);
518 * Check to make sure that we don't point outside of the leaf,
519 * just incase all the items are consistent to eachother, but
520 * all point outside of the leaf.
522 if (btrfs_item_end_nr(leaf
, slot
) >
523 BTRFS_LEAF_DATA_SIZE(root
)) {
524 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
532 struct extent_buffer
*find_eb_for_page(struct extent_io_tree
*tree
,
533 struct page
*page
, int max_walk
)
535 struct extent_buffer
*eb
;
536 u64 start
= page_offset(page
);
540 if (start
< max_walk
)
543 min_start
= start
- max_walk
;
545 while (start
>= min_start
) {
546 eb
= find_extent_buffer(tree
, start
, 0);
549 * we found an extent buffer and it contains our page
552 if (eb
->start
<= target
&&
553 eb
->start
+ eb
->len
> target
)
556 /* we found an extent buffer that wasn't for us */
557 free_extent_buffer(eb
);
562 start
-= PAGE_CACHE_SIZE
;
567 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
568 struct extent_state
*state
, int mirror
)
570 struct extent_io_tree
*tree
;
573 struct extent_buffer
*eb
;
574 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
581 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
582 eb
= (struct extent_buffer
*)page
->private;
584 /* the pending IO might have been the only thing that kept this buffer
585 * in memory. Make sure we have a ref for all this other checks
587 extent_buffer_get(eb
);
589 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
593 eb
->read_mirror
= mirror
;
594 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
599 found_start
= btrfs_header_bytenr(eb
);
600 if (found_start
!= eb
->start
) {
601 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
603 (unsigned long long)found_start
,
604 (unsigned long long)eb
->start
);
608 if (check_tree_block_fsid(root
, eb
)) {
609 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
610 (unsigned long long)eb
->start
);
614 found_level
= btrfs_header_level(eb
);
616 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
619 ret
= csum_tree_block(root
, eb
, 1);
626 * If this is a leaf block and it is corrupt, set the corrupt bit so
627 * that we don't try and read the other copies of this block, just
630 if (found_level
== 0 && check_leaf(root
, eb
)) {
631 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
636 set_extent_buffer_uptodate(eb
);
638 if (test_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
)) {
639 clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
);
640 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
645 * our io error hook is going to dec the io pages
646 * again, we have to make sure it has something
649 atomic_inc(&eb
->io_pages
);
650 clear_extent_buffer_uptodate(eb
);
652 free_extent_buffer(eb
);
657 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
659 struct extent_buffer
*eb
;
660 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
662 eb
= (struct extent_buffer
*)page
->private;
663 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
664 eb
->read_mirror
= failed_mirror
;
665 atomic_dec(&eb
->io_pages
);
666 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
667 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
668 return -EIO
; /* we fixed nothing */
671 static void end_workqueue_bio(struct bio
*bio
, int err
)
673 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
674 struct btrfs_fs_info
*fs_info
;
676 fs_info
= end_io_wq
->info
;
677 end_io_wq
->error
= err
;
678 end_io_wq
->work
.func
= end_workqueue_fn
;
679 end_io_wq
->work
.flags
= 0;
681 if (bio
->bi_rw
& REQ_WRITE
) {
682 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
683 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
685 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
686 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
688 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
689 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
692 btrfs_queue_worker(&fs_info
->endio_write_workers
,
695 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
696 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
698 else if (end_io_wq
->metadata
)
699 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
702 btrfs_queue_worker(&fs_info
->endio_workers
,
708 * For the metadata arg you want
711 * 1 - if normal metadta
712 * 2 - if writing to the free space cache area
713 * 3 - raid parity work
715 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
718 struct end_io_wq
*end_io_wq
;
719 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
723 end_io_wq
->private = bio
->bi_private
;
724 end_io_wq
->end_io
= bio
->bi_end_io
;
725 end_io_wq
->info
= info
;
726 end_io_wq
->error
= 0;
727 end_io_wq
->bio
= bio
;
728 end_io_wq
->metadata
= metadata
;
730 bio
->bi_private
= end_io_wq
;
731 bio
->bi_end_io
= end_workqueue_bio
;
735 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
737 unsigned long limit
= min_t(unsigned long,
738 info
->workers
.max_workers
,
739 info
->fs_devices
->open_devices
);
743 static void run_one_async_start(struct btrfs_work
*work
)
745 struct async_submit_bio
*async
;
748 async
= container_of(work
, struct async_submit_bio
, work
);
749 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
750 async
->mirror_num
, async
->bio_flags
,
756 static void run_one_async_done(struct btrfs_work
*work
)
758 struct btrfs_fs_info
*fs_info
;
759 struct async_submit_bio
*async
;
762 async
= container_of(work
, struct async_submit_bio
, work
);
763 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
765 limit
= btrfs_async_submit_limit(fs_info
);
766 limit
= limit
* 2 / 3;
768 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
769 waitqueue_active(&fs_info
->async_submit_wait
))
770 wake_up(&fs_info
->async_submit_wait
);
772 /* If an error occured we just want to clean up the bio and move on */
774 bio_endio(async
->bio
, async
->error
);
778 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
779 async
->mirror_num
, async
->bio_flags
,
783 static void run_one_async_free(struct btrfs_work
*work
)
785 struct async_submit_bio
*async
;
787 async
= container_of(work
, struct async_submit_bio
, work
);
791 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
792 int rw
, struct bio
*bio
, int mirror_num
,
793 unsigned long bio_flags
,
795 extent_submit_bio_hook_t
*submit_bio_start
,
796 extent_submit_bio_hook_t
*submit_bio_done
)
798 struct async_submit_bio
*async
;
800 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
804 async
->inode
= inode
;
807 async
->mirror_num
= mirror_num
;
808 async
->submit_bio_start
= submit_bio_start
;
809 async
->submit_bio_done
= submit_bio_done
;
811 async
->work
.func
= run_one_async_start
;
812 async
->work
.ordered_func
= run_one_async_done
;
813 async
->work
.ordered_free
= run_one_async_free
;
815 async
->work
.flags
= 0;
816 async
->bio_flags
= bio_flags
;
817 async
->bio_offset
= bio_offset
;
821 atomic_inc(&fs_info
->nr_async_submits
);
824 btrfs_set_work_high_prio(&async
->work
);
826 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
828 while (atomic_read(&fs_info
->async_submit_draining
) &&
829 atomic_read(&fs_info
->nr_async_submits
)) {
830 wait_event(fs_info
->async_submit_wait
,
831 (atomic_read(&fs_info
->nr_async_submits
) == 0));
837 static int btree_csum_one_bio(struct bio
*bio
)
839 struct bio_vec
*bvec
= bio
->bi_io_vec
;
841 struct btrfs_root
*root
;
844 WARN_ON(bio
->bi_vcnt
<= 0);
845 while (bio_index
< bio
->bi_vcnt
) {
846 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
847 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
856 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
857 struct bio
*bio
, int mirror_num
,
858 unsigned long bio_flags
,
862 * when we're called for a write, we're already in the async
863 * submission context. Just jump into btrfs_map_bio
865 return btree_csum_one_bio(bio
);
868 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
869 int mirror_num
, unsigned long bio_flags
,
875 * when we're called for a write, we're already in the async
876 * submission context. Just jump into btrfs_map_bio
878 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
884 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
886 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
895 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
896 int mirror_num
, unsigned long bio_flags
,
899 int async
= check_async_write(inode
, bio_flags
);
902 if (!(rw
& REQ_WRITE
)) {
904 * called for a read, do the setup so that checksum validation
905 * can happen in the async kernel threads
907 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
911 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
914 ret
= btree_csum_one_bio(bio
);
917 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
921 * kthread helpers are used to submit writes so that
922 * checksumming can happen in parallel across all CPUs
924 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
925 inode
, rw
, bio
, mirror_num
, 0,
927 __btree_submit_bio_start
,
928 __btree_submit_bio_done
);
938 #ifdef CONFIG_MIGRATION
939 static int btree_migratepage(struct address_space
*mapping
,
940 struct page
*newpage
, struct page
*page
,
941 enum migrate_mode mode
)
944 * we can't safely write a btree page from here,
945 * we haven't done the locking hook
950 * Buffers may be managed in a filesystem specific way.
951 * We must have no buffers or drop them.
953 if (page_has_private(page
) &&
954 !try_to_release_page(page
, GFP_KERNEL
))
956 return migrate_page(mapping
, newpage
, page
, mode
);
961 static int btree_writepages(struct address_space
*mapping
,
962 struct writeback_control
*wbc
)
964 struct extent_io_tree
*tree
;
965 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
966 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
967 struct btrfs_root
*root
= BTRFS_I(mapping
->host
)->root
;
969 unsigned long thresh
= 32 * 1024 * 1024;
971 if (wbc
->for_kupdate
)
974 /* this is a bit racy, but that's ok */
975 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
976 if (num_dirty
< thresh
)
979 return btree_write_cache_pages(mapping
, wbc
);
982 static int btree_readpage(struct file
*file
, struct page
*page
)
984 struct extent_io_tree
*tree
;
985 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
986 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
989 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
991 if (PageWriteback(page
) || PageDirty(page
))
994 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
995 * slab allocation from alloc_extent_state down the callchain where
996 * it'd hit a BUG_ON as those flags are not allowed.
998 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
1000 return try_release_extent_buffer(page
, gfp_flags
);
1003 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
1005 struct extent_io_tree
*tree
;
1006 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1007 extent_invalidatepage(tree
, page
, offset
);
1008 btree_releasepage(page
, GFP_NOFS
);
1009 if (PagePrivate(page
)) {
1010 printk(KERN_WARNING
"btrfs warning page private not zero "
1011 "on page %llu\n", (unsigned long long)page_offset(page
));
1012 ClearPagePrivate(page
);
1013 set_page_private(page
, 0);
1014 page_cache_release(page
);
1018 static int btree_set_page_dirty(struct page
*page
)
1021 struct extent_buffer
*eb
;
1023 BUG_ON(!PagePrivate(page
));
1024 eb
= (struct extent_buffer
*)page
->private;
1026 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1027 BUG_ON(!atomic_read(&eb
->refs
));
1028 btrfs_assert_tree_locked(eb
);
1030 return __set_page_dirty_nobuffers(page
);
1033 static const struct address_space_operations btree_aops
= {
1034 .readpage
= btree_readpage
,
1035 .writepages
= btree_writepages
,
1036 .releasepage
= btree_releasepage
,
1037 .invalidatepage
= btree_invalidatepage
,
1038 #ifdef CONFIG_MIGRATION
1039 .migratepage
= btree_migratepage
,
1041 .set_page_dirty
= btree_set_page_dirty
,
1044 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1047 struct extent_buffer
*buf
= NULL
;
1048 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1051 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1054 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1055 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1056 free_extent_buffer(buf
);
1060 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1061 int mirror_num
, struct extent_buffer
**eb
)
1063 struct extent_buffer
*buf
= NULL
;
1064 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1065 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1068 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1072 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1074 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1075 btree_get_extent
, mirror_num
);
1077 free_extent_buffer(buf
);
1081 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1082 free_extent_buffer(buf
);
1084 } else if (extent_buffer_uptodate(buf
)) {
1087 free_extent_buffer(buf
);
1092 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1093 u64 bytenr
, u32 blocksize
)
1095 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1096 struct extent_buffer
*eb
;
1097 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1102 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1103 u64 bytenr
, u32 blocksize
)
1105 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1106 struct extent_buffer
*eb
;
1108 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1114 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1116 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1117 buf
->start
+ buf
->len
- 1);
1120 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1122 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1123 buf
->start
, buf
->start
+ buf
->len
- 1);
1126 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1127 u32 blocksize
, u64 parent_transid
)
1129 struct extent_buffer
*buf
= NULL
;
1132 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1136 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1141 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1142 struct extent_buffer
*buf
)
1144 if (btrfs_header_generation(buf
) ==
1145 root
->fs_info
->running_transaction
->transid
) {
1146 btrfs_assert_tree_locked(buf
);
1148 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1149 spin_lock(&root
->fs_info
->delalloc_lock
);
1150 if (root
->fs_info
->dirty_metadata_bytes
>= buf
->len
)
1151 root
->fs_info
->dirty_metadata_bytes
-= buf
->len
;
1153 spin_unlock(&root
->fs_info
->delalloc_lock
);
1154 btrfs_panic(root
->fs_info
, -EOVERFLOW
,
1155 "Can't clear %lu bytes from "
1156 " dirty_mdatadata_bytes (%llu)",
1158 root
->fs_info
->dirty_metadata_bytes
);
1160 spin_unlock(&root
->fs_info
->delalloc_lock
);
1162 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1163 btrfs_set_lock_blocking(buf
);
1164 clear_extent_buffer_dirty(buf
);
1169 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1170 u32 stripesize
, struct btrfs_root
*root
,
1171 struct btrfs_fs_info
*fs_info
,
1175 root
->commit_root
= NULL
;
1176 root
->sectorsize
= sectorsize
;
1177 root
->nodesize
= nodesize
;
1178 root
->leafsize
= leafsize
;
1179 root
->stripesize
= stripesize
;
1181 root
->track_dirty
= 0;
1183 root
->orphan_item_inserted
= 0;
1184 root
->orphan_cleanup_state
= 0;
1186 root
->objectid
= objectid
;
1187 root
->last_trans
= 0;
1188 root
->highest_objectid
= 0;
1190 root
->inode_tree
= RB_ROOT
;
1191 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1192 root
->block_rsv
= NULL
;
1193 root
->orphan_block_rsv
= NULL
;
1195 INIT_LIST_HEAD(&root
->dirty_list
);
1196 INIT_LIST_HEAD(&root
->root_list
);
1197 spin_lock_init(&root
->orphan_lock
);
1198 spin_lock_init(&root
->inode_lock
);
1199 spin_lock_init(&root
->accounting_lock
);
1200 mutex_init(&root
->objectid_mutex
);
1201 mutex_init(&root
->log_mutex
);
1202 init_waitqueue_head(&root
->log_writer_wait
);
1203 init_waitqueue_head(&root
->log_commit_wait
[0]);
1204 init_waitqueue_head(&root
->log_commit_wait
[1]);
1205 atomic_set(&root
->log_commit
[0], 0);
1206 atomic_set(&root
->log_commit
[1], 0);
1207 atomic_set(&root
->log_writers
, 0);
1208 atomic_set(&root
->log_batch
, 0);
1209 atomic_set(&root
->orphan_inodes
, 0);
1210 root
->log_transid
= 0;
1211 root
->last_log_commit
= 0;
1212 extent_io_tree_init(&root
->dirty_log_pages
,
1213 fs_info
->btree_inode
->i_mapping
);
1215 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1216 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1217 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1218 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1219 root
->defrag_trans_start
= fs_info
->generation
;
1220 init_completion(&root
->kobj_unregister
);
1221 root
->defrag_running
= 0;
1222 root
->root_key
.objectid
= objectid
;
1225 spin_lock_init(&root
->root_item_lock
);
1228 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1229 struct btrfs_fs_info
*fs_info
,
1231 struct btrfs_root
*root
)
1237 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1238 tree_root
->sectorsize
, tree_root
->stripesize
,
1239 root
, fs_info
, objectid
);
1240 ret
= btrfs_find_last_root(tree_root
, objectid
,
1241 &root
->root_item
, &root
->root_key
);
1247 generation
= btrfs_root_generation(&root
->root_item
);
1248 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1249 root
->commit_root
= NULL
;
1250 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1251 blocksize
, generation
);
1252 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1253 free_extent_buffer(root
->node
);
1257 root
->commit_root
= btrfs_root_node(root
);
1261 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1263 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1265 root
->fs_info
= fs_info
;
1269 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1270 struct btrfs_fs_info
*fs_info
,
1273 struct extent_buffer
*leaf
;
1274 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1275 struct btrfs_root
*root
;
1276 struct btrfs_key key
;
1280 root
= btrfs_alloc_root(fs_info
);
1282 return ERR_PTR(-ENOMEM
);
1284 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1285 tree_root
->sectorsize
, tree_root
->stripesize
,
1286 root
, fs_info
, objectid
);
1287 root
->root_key
.objectid
= objectid
;
1288 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1289 root
->root_key
.offset
= 0;
1291 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1292 0, objectid
, NULL
, 0, 0, 0);
1294 ret
= PTR_ERR(leaf
);
1298 bytenr
= leaf
->start
;
1299 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1300 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1301 btrfs_set_header_generation(leaf
, trans
->transid
);
1302 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1303 btrfs_set_header_owner(leaf
, objectid
);
1306 write_extent_buffer(leaf
, fs_info
->fsid
,
1307 (unsigned long)btrfs_header_fsid(leaf
),
1309 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1310 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1312 btrfs_mark_buffer_dirty(leaf
);
1314 root
->commit_root
= btrfs_root_node(root
);
1315 root
->track_dirty
= 1;
1318 root
->root_item
.flags
= 0;
1319 root
->root_item
.byte_limit
= 0;
1320 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1321 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1322 btrfs_set_root_level(&root
->root_item
, 0);
1323 btrfs_set_root_refs(&root
->root_item
, 1);
1324 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1325 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1326 btrfs_set_root_dirid(&root
->root_item
, 0);
1327 root
->root_item
.drop_level
= 0;
1329 key
.objectid
= objectid
;
1330 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1332 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1336 btrfs_tree_unlock(leaf
);
1340 return ERR_PTR(ret
);
1345 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1346 struct btrfs_fs_info
*fs_info
)
1348 struct btrfs_root
*root
;
1349 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1350 struct extent_buffer
*leaf
;
1352 root
= btrfs_alloc_root(fs_info
);
1354 return ERR_PTR(-ENOMEM
);
1356 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1357 tree_root
->sectorsize
, tree_root
->stripesize
,
1358 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1360 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1361 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1362 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1364 * log trees do not get reference counted because they go away
1365 * before a real commit is actually done. They do store pointers
1366 * to file data extents, and those reference counts still get
1367 * updated (along with back refs to the log tree).
1371 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1372 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1376 return ERR_CAST(leaf
);
1379 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1380 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1381 btrfs_set_header_generation(leaf
, trans
->transid
);
1382 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1383 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1386 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1387 (unsigned long)btrfs_header_fsid(root
->node
),
1389 btrfs_mark_buffer_dirty(root
->node
);
1390 btrfs_tree_unlock(root
->node
);
1394 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1395 struct btrfs_fs_info
*fs_info
)
1397 struct btrfs_root
*log_root
;
1399 log_root
= alloc_log_tree(trans
, fs_info
);
1400 if (IS_ERR(log_root
))
1401 return PTR_ERR(log_root
);
1402 WARN_ON(fs_info
->log_root_tree
);
1403 fs_info
->log_root_tree
= log_root
;
1407 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1408 struct btrfs_root
*root
)
1410 struct btrfs_root
*log_root
;
1411 struct btrfs_inode_item
*inode_item
;
1413 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1414 if (IS_ERR(log_root
))
1415 return PTR_ERR(log_root
);
1417 log_root
->last_trans
= trans
->transid
;
1418 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1420 inode_item
= &log_root
->root_item
.inode
;
1421 inode_item
->generation
= cpu_to_le64(1);
1422 inode_item
->size
= cpu_to_le64(3);
1423 inode_item
->nlink
= cpu_to_le32(1);
1424 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1425 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1427 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1429 WARN_ON(root
->log_root
);
1430 root
->log_root
= log_root
;
1431 root
->log_transid
= 0;
1432 root
->last_log_commit
= 0;
1436 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1437 struct btrfs_key
*location
)
1439 struct btrfs_root
*root
;
1440 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1441 struct btrfs_path
*path
;
1442 struct extent_buffer
*l
;
1448 root
= btrfs_alloc_root(fs_info
);
1450 return ERR_PTR(-ENOMEM
);
1451 if (location
->offset
== (u64
)-1) {
1452 ret
= find_and_setup_root(tree_root
, fs_info
,
1453 location
->objectid
, root
);
1456 return ERR_PTR(ret
);
1461 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1462 tree_root
->sectorsize
, tree_root
->stripesize
,
1463 root
, fs_info
, location
->objectid
);
1465 path
= btrfs_alloc_path();
1468 return ERR_PTR(-ENOMEM
);
1470 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1473 slot
= path
->slots
[0];
1474 btrfs_read_root_item(tree_root
, l
, slot
, &root
->root_item
);
1475 memcpy(&root
->root_key
, location
, sizeof(*location
));
1477 btrfs_free_path(path
);
1482 return ERR_PTR(ret
);
1485 generation
= btrfs_root_generation(&root
->root_item
);
1486 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1487 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1488 blocksize
, generation
);
1489 root
->commit_root
= btrfs_root_node(root
);
1490 BUG_ON(!root
->node
); /* -ENOMEM */
1492 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1494 btrfs_check_and_init_root_item(&root
->root_item
);
1500 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1501 struct btrfs_key
*location
)
1503 struct btrfs_root
*root
;
1506 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1507 return fs_info
->tree_root
;
1508 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1509 return fs_info
->extent_root
;
1510 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1511 return fs_info
->chunk_root
;
1512 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1513 return fs_info
->dev_root
;
1514 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1515 return fs_info
->csum_root
;
1516 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1517 return fs_info
->quota_root
? fs_info
->quota_root
:
1520 spin_lock(&fs_info
->fs_roots_radix_lock
);
1521 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1522 (unsigned long)location
->objectid
);
1523 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1527 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1531 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1532 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1534 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1539 btrfs_init_free_ino_ctl(root
);
1540 mutex_init(&root
->fs_commit_mutex
);
1541 spin_lock_init(&root
->cache_lock
);
1542 init_waitqueue_head(&root
->cache_wait
);
1544 ret
= get_anon_bdev(&root
->anon_dev
);
1548 if (btrfs_root_refs(&root
->root_item
) == 0) {
1553 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1557 root
->orphan_item_inserted
= 1;
1559 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1563 spin_lock(&fs_info
->fs_roots_radix_lock
);
1564 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1565 (unsigned long)root
->root_key
.objectid
,
1570 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1571 radix_tree_preload_end();
1573 if (ret
== -EEXIST
) {
1580 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1581 root
->root_key
.objectid
);
1586 return ERR_PTR(ret
);
1589 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1591 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1593 struct btrfs_device
*device
;
1594 struct backing_dev_info
*bdi
;
1597 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1600 bdi
= blk_get_backing_dev_info(device
->bdev
);
1601 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1611 * If this fails, caller must call bdi_destroy() to get rid of the
1614 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1618 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1619 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1623 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1624 bdi
->congested_fn
= btrfs_congested_fn
;
1625 bdi
->congested_data
= info
;
1630 * called by the kthread helper functions to finally call the bio end_io
1631 * functions. This is where read checksum verification actually happens
1633 static void end_workqueue_fn(struct btrfs_work
*work
)
1636 struct end_io_wq
*end_io_wq
;
1637 struct btrfs_fs_info
*fs_info
;
1640 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1641 bio
= end_io_wq
->bio
;
1642 fs_info
= end_io_wq
->info
;
1644 error
= end_io_wq
->error
;
1645 bio
->bi_private
= end_io_wq
->private;
1646 bio
->bi_end_io
= end_io_wq
->end_io
;
1648 bio_endio(bio
, error
);
1651 static int cleaner_kthread(void *arg
)
1653 struct btrfs_root
*root
= arg
;
1656 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1657 mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1658 btrfs_run_delayed_iputs(root
);
1659 btrfs_clean_old_snapshots(root
);
1660 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1661 btrfs_run_defrag_inodes(root
->fs_info
);
1664 if (!try_to_freeze()) {
1665 set_current_state(TASK_INTERRUPTIBLE
);
1666 if (!kthread_should_stop())
1668 __set_current_state(TASK_RUNNING
);
1670 } while (!kthread_should_stop());
1674 static int transaction_kthread(void *arg
)
1676 struct btrfs_root
*root
= arg
;
1677 struct btrfs_trans_handle
*trans
;
1678 struct btrfs_transaction
*cur
;
1681 unsigned long delay
;
1685 cannot_commit
= false;
1687 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1689 spin_lock(&root
->fs_info
->trans_lock
);
1690 cur
= root
->fs_info
->running_transaction
;
1692 spin_unlock(&root
->fs_info
->trans_lock
);
1696 now
= get_seconds();
1697 if (!cur
->blocked
&&
1698 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1699 spin_unlock(&root
->fs_info
->trans_lock
);
1703 transid
= cur
->transid
;
1704 spin_unlock(&root
->fs_info
->trans_lock
);
1706 /* If the file system is aborted, this will always fail. */
1707 trans
= btrfs_attach_transaction(root
);
1708 if (IS_ERR(trans
)) {
1709 if (PTR_ERR(trans
) != -ENOENT
)
1710 cannot_commit
= true;
1713 if (transid
== trans
->transid
) {
1714 btrfs_commit_transaction(trans
, root
);
1716 btrfs_end_transaction(trans
, root
);
1719 wake_up_process(root
->fs_info
->cleaner_kthread
);
1720 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1722 if (!try_to_freeze()) {
1723 set_current_state(TASK_INTERRUPTIBLE
);
1724 if (!kthread_should_stop() &&
1725 (!btrfs_transaction_blocked(root
->fs_info
) ||
1727 schedule_timeout(delay
);
1728 __set_current_state(TASK_RUNNING
);
1730 } while (!kthread_should_stop());
1735 * this will find the highest generation in the array of
1736 * root backups. The index of the highest array is returned,
1737 * or -1 if we can't find anything.
1739 * We check to make sure the array is valid by comparing the
1740 * generation of the latest root in the array with the generation
1741 * in the super block. If they don't match we pitch it.
1743 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1746 int newest_index
= -1;
1747 struct btrfs_root_backup
*root_backup
;
1750 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1751 root_backup
= info
->super_copy
->super_roots
+ i
;
1752 cur
= btrfs_backup_tree_root_gen(root_backup
);
1753 if (cur
== newest_gen
)
1757 /* check to see if we actually wrapped around */
1758 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1759 root_backup
= info
->super_copy
->super_roots
;
1760 cur
= btrfs_backup_tree_root_gen(root_backup
);
1761 if (cur
== newest_gen
)
1764 return newest_index
;
1769 * find the oldest backup so we know where to store new entries
1770 * in the backup array. This will set the backup_root_index
1771 * field in the fs_info struct
1773 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1776 int newest_index
= -1;
1778 newest_index
= find_newest_super_backup(info
, newest_gen
);
1779 /* if there was garbage in there, just move along */
1780 if (newest_index
== -1) {
1781 info
->backup_root_index
= 0;
1783 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1788 * copy all the root pointers into the super backup array.
1789 * this will bump the backup pointer by one when it is
1792 static void backup_super_roots(struct btrfs_fs_info
*info
)
1795 struct btrfs_root_backup
*root_backup
;
1798 next_backup
= info
->backup_root_index
;
1799 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1800 BTRFS_NUM_BACKUP_ROOTS
;
1803 * just overwrite the last backup if we're at the same generation
1804 * this happens only at umount
1806 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1807 if (btrfs_backup_tree_root_gen(root_backup
) ==
1808 btrfs_header_generation(info
->tree_root
->node
))
1809 next_backup
= last_backup
;
1811 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1814 * make sure all of our padding and empty slots get zero filled
1815 * regardless of which ones we use today
1817 memset(root_backup
, 0, sizeof(*root_backup
));
1819 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1821 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1822 btrfs_set_backup_tree_root_gen(root_backup
,
1823 btrfs_header_generation(info
->tree_root
->node
));
1825 btrfs_set_backup_tree_root_level(root_backup
,
1826 btrfs_header_level(info
->tree_root
->node
));
1828 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1829 btrfs_set_backup_chunk_root_gen(root_backup
,
1830 btrfs_header_generation(info
->chunk_root
->node
));
1831 btrfs_set_backup_chunk_root_level(root_backup
,
1832 btrfs_header_level(info
->chunk_root
->node
));
1834 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1835 btrfs_set_backup_extent_root_gen(root_backup
,
1836 btrfs_header_generation(info
->extent_root
->node
));
1837 btrfs_set_backup_extent_root_level(root_backup
,
1838 btrfs_header_level(info
->extent_root
->node
));
1841 * we might commit during log recovery, which happens before we set
1842 * the fs_root. Make sure it is valid before we fill it in.
1844 if (info
->fs_root
&& info
->fs_root
->node
) {
1845 btrfs_set_backup_fs_root(root_backup
,
1846 info
->fs_root
->node
->start
);
1847 btrfs_set_backup_fs_root_gen(root_backup
,
1848 btrfs_header_generation(info
->fs_root
->node
));
1849 btrfs_set_backup_fs_root_level(root_backup
,
1850 btrfs_header_level(info
->fs_root
->node
));
1853 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1854 btrfs_set_backup_dev_root_gen(root_backup
,
1855 btrfs_header_generation(info
->dev_root
->node
));
1856 btrfs_set_backup_dev_root_level(root_backup
,
1857 btrfs_header_level(info
->dev_root
->node
));
1859 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1860 btrfs_set_backup_csum_root_gen(root_backup
,
1861 btrfs_header_generation(info
->csum_root
->node
));
1862 btrfs_set_backup_csum_root_level(root_backup
,
1863 btrfs_header_level(info
->csum_root
->node
));
1865 btrfs_set_backup_total_bytes(root_backup
,
1866 btrfs_super_total_bytes(info
->super_copy
));
1867 btrfs_set_backup_bytes_used(root_backup
,
1868 btrfs_super_bytes_used(info
->super_copy
));
1869 btrfs_set_backup_num_devices(root_backup
,
1870 btrfs_super_num_devices(info
->super_copy
));
1873 * if we don't copy this out to the super_copy, it won't get remembered
1874 * for the next commit
1876 memcpy(&info
->super_copy
->super_roots
,
1877 &info
->super_for_commit
->super_roots
,
1878 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1882 * this copies info out of the root backup array and back into
1883 * the in-memory super block. It is meant to help iterate through
1884 * the array, so you send it the number of backups you've already
1885 * tried and the last backup index you used.
1887 * this returns -1 when it has tried all the backups
1889 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1890 struct btrfs_super_block
*super
,
1891 int *num_backups_tried
, int *backup_index
)
1893 struct btrfs_root_backup
*root_backup
;
1894 int newest
= *backup_index
;
1896 if (*num_backups_tried
== 0) {
1897 u64 gen
= btrfs_super_generation(super
);
1899 newest
= find_newest_super_backup(info
, gen
);
1903 *backup_index
= newest
;
1904 *num_backups_tried
= 1;
1905 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1906 /* we've tried all the backups, all done */
1909 /* jump to the next oldest backup */
1910 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1911 BTRFS_NUM_BACKUP_ROOTS
;
1912 *backup_index
= newest
;
1913 *num_backups_tried
+= 1;
1915 root_backup
= super
->super_roots
+ newest
;
1917 btrfs_set_super_generation(super
,
1918 btrfs_backup_tree_root_gen(root_backup
));
1919 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1920 btrfs_set_super_root_level(super
,
1921 btrfs_backup_tree_root_level(root_backup
));
1922 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1925 * fixme: the total bytes and num_devices need to match or we should
1928 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1929 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1933 /* helper to cleanup tree roots */
1934 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1936 free_extent_buffer(info
->tree_root
->node
);
1937 free_extent_buffer(info
->tree_root
->commit_root
);
1938 free_extent_buffer(info
->dev_root
->node
);
1939 free_extent_buffer(info
->dev_root
->commit_root
);
1940 free_extent_buffer(info
->extent_root
->node
);
1941 free_extent_buffer(info
->extent_root
->commit_root
);
1942 free_extent_buffer(info
->csum_root
->node
);
1943 free_extent_buffer(info
->csum_root
->commit_root
);
1944 if (info
->quota_root
) {
1945 free_extent_buffer(info
->quota_root
->node
);
1946 free_extent_buffer(info
->quota_root
->commit_root
);
1949 info
->tree_root
->node
= NULL
;
1950 info
->tree_root
->commit_root
= NULL
;
1951 info
->dev_root
->node
= NULL
;
1952 info
->dev_root
->commit_root
= NULL
;
1953 info
->extent_root
->node
= NULL
;
1954 info
->extent_root
->commit_root
= NULL
;
1955 info
->csum_root
->node
= NULL
;
1956 info
->csum_root
->commit_root
= NULL
;
1957 if (info
->quota_root
) {
1958 info
->quota_root
->node
= NULL
;
1959 info
->quota_root
->commit_root
= NULL
;
1963 free_extent_buffer(info
->chunk_root
->node
);
1964 free_extent_buffer(info
->chunk_root
->commit_root
);
1965 info
->chunk_root
->node
= NULL
;
1966 info
->chunk_root
->commit_root
= NULL
;
1971 int open_ctree(struct super_block
*sb
,
1972 struct btrfs_fs_devices
*fs_devices
,
1982 struct btrfs_key location
;
1983 struct buffer_head
*bh
;
1984 struct btrfs_super_block
*disk_super
;
1985 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
1986 struct btrfs_root
*tree_root
;
1987 struct btrfs_root
*extent_root
;
1988 struct btrfs_root
*csum_root
;
1989 struct btrfs_root
*chunk_root
;
1990 struct btrfs_root
*dev_root
;
1991 struct btrfs_root
*quota_root
;
1992 struct btrfs_root
*log_tree_root
;
1995 int num_backups_tried
= 0;
1996 int backup_index
= 0;
1998 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
1999 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
2000 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
2001 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2002 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
2003 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
2005 if (!tree_root
|| !extent_root
|| !csum_root
||
2006 !chunk_root
|| !dev_root
|| !quota_root
) {
2011 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2017 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2023 fs_info
->btree_inode
= new_inode(sb
);
2024 if (!fs_info
->btree_inode
) {
2029 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2031 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2032 INIT_LIST_HEAD(&fs_info
->trans_list
);
2033 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2034 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2035 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2036 INIT_LIST_HEAD(&fs_info
->ordered_operations
);
2037 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2038 spin_lock_init(&fs_info
->delalloc_lock
);
2039 spin_lock_init(&fs_info
->trans_lock
);
2040 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2041 spin_lock_init(&fs_info
->delayed_iput_lock
);
2042 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2043 spin_lock_init(&fs_info
->free_chunk_lock
);
2044 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2045 rwlock_init(&fs_info
->tree_mod_log_lock
);
2046 mutex_init(&fs_info
->reloc_mutex
);
2048 init_completion(&fs_info
->kobj_unregister
);
2049 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2050 INIT_LIST_HEAD(&fs_info
->space_info
);
2051 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2052 btrfs_mapping_init(&fs_info
->mapping_tree
);
2053 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2054 BTRFS_BLOCK_RSV_GLOBAL
);
2055 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2056 BTRFS_BLOCK_RSV_DELALLOC
);
2057 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2058 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2059 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2060 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2061 BTRFS_BLOCK_RSV_DELOPS
);
2062 atomic_set(&fs_info
->nr_async_submits
, 0);
2063 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2064 atomic_set(&fs_info
->async_submit_draining
, 0);
2065 atomic_set(&fs_info
->nr_async_bios
, 0);
2066 atomic_set(&fs_info
->defrag_running
, 0);
2067 atomic_set(&fs_info
->tree_mod_seq
, 0);
2069 fs_info
->max_inline
= 8192 * 1024;
2070 fs_info
->metadata_ratio
= 0;
2071 fs_info
->defrag_inodes
= RB_ROOT
;
2072 fs_info
->trans_no_join
= 0;
2073 fs_info
->free_chunk_space
= 0;
2074 fs_info
->tree_mod_log
= RB_ROOT
;
2076 /* readahead state */
2077 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2078 spin_lock_init(&fs_info
->reada_lock
);
2080 fs_info
->thread_pool_size
= min_t(unsigned long,
2081 num_online_cpus() + 2, 8);
2083 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2084 spin_lock_init(&fs_info
->ordered_extent_lock
);
2085 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2087 if (!fs_info
->delayed_root
) {
2091 btrfs_init_delayed_root(fs_info
->delayed_root
);
2093 mutex_init(&fs_info
->scrub_lock
);
2094 atomic_set(&fs_info
->scrubs_running
, 0);
2095 atomic_set(&fs_info
->scrub_pause_req
, 0);
2096 atomic_set(&fs_info
->scrubs_paused
, 0);
2097 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2098 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2099 init_rwsem(&fs_info
->scrub_super_lock
);
2100 fs_info
->scrub_workers_refcnt
= 0;
2101 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2102 fs_info
->check_integrity_print_mask
= 0;
2105 spin_lock_init(&fs_info
->balance_lock
);
2106 mutex_init(&fs_info
->balance_mutex
);
2107 atomic_set(&fs_info
->balance_running
, 0);
2108 atomic_set(&fs_info
->balance_pause_req
, 0);
2109 atomic_set(&fs_info
->balance_cancel_req
, 0);
2110 fs_info
->balance_ctl
= NULL
;
2111 init_waitqueue_head(&fs_info
->balance_wait_q
);
2113 sb
->s_blocksize
= 4096;
2114 sb
->s_blocksize_bits
= blksize_bits(4096);
2115 sb
->s_bdi
= &fs_info
->bdi
;
2117 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2118 set_nlink(fs_info
->btree_inode
, 1);
2120 * we set the i_size on the btree inode to the max possible int.
2121 * the real end of the address space is determined by all of
2122 * the devices in the system
2124 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2125 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2126 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2128 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2129 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2130 fs_info
->btree_inode
->i_mapping
);
2131 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2132 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2134 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2136 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2137 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2138 sizeof(struct btrfs_key
));
2139 set_bit(BTRFS_INODE_DUMMY
,
2140 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2141 insert_inode_hash(fs_info
->btree_inode
);
2143 spin_lock_init(&fs_info
->block_group_cache_lock
);
2144 fs_info
->block_group_cache_tree
= RB_ROOT
;
2146 extent_io_tree_init(&fs_info
->freed_extents
[0],
2147 fs_info
->btree_inode
->i_mapping
);
2148 extent_io_tree_init(&fs_info
->freed_extents
[1],
2149 fs_info
->btree_inode
->i_mapping
);
2150 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2151 fs_info
->do_barriers
= 1;
2154 mutex_init(&fs_info
->ordered_operations_mutex
);
2155 mutex_init(&fs_info
->tree_log_mutex
);
2156 mutex_init(&fs_info
->chunk_mutex
);
2157 mutex_init(&fs_info
->transaction_kthread_mutex
);
2158 mutex_init(&fs_info
->cleaner_mutex
);
2159 mutex_init(&fs_info
->volume_mutex
);
2160 init_rwsem(&fs_info
->extent_commit_sem
);
2161 init_rwsem(&fs_info
->cleanup_work_sem
);
2162 init_rwsem(&fs_info
->subvol_sem
);
2163 fs_info
->dev_replace
.lock_owner
= 0;
2164 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2165 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2166 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2167 mutex_init(&fs_info
->dev_replace
.lock
);
2169 spin_lock_init(&fs_info
->qgroup_lock
);
2170 fs_info
->qgroup_tree
= RB_ROOT
;
2171 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2172 fs_info
->qgroup_seq
= 1;
2173 fs_info
->quota_enabled
= 0;
2174 fs_info
->pending_quota_state
= 0;
2176 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2177 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2179 init_waitqueue_head(&fs_info
->transaction_throttle
);
2180 init_waitqueue_head(&fs_info
->transaction_wait
);
2181 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2182 init_waitqueue_head(&fs_info
->async_submit_wait
);
2184 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2190 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2191 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2193 invalidate_bdev(fs_devices
->latest_bdev
);
2194 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2200 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2201 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2202 sizeof(*fs_info
->super_for_commit
));
2205 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2207 disk_super
= fs_info
->super_copy
;
2208 if (!btrfs_super_root(disk_super
))
2211 /* check FS state, whether FS is broken. */
2212 fs_info
->fs_state
|= btrfs_super_flags(disk_super
);
2214 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2216 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2222 * run through our array of backup supers and setup
2223 * our ring pointer to the oldest one
2225 generation
= btrfs_super_generation(disk_super
);
2226 find_oldest_super_backup(fs_info
, generation
);
2229 * In the long term, we'll store the compression type in the super
2230 * block, and it'll be used for per file compression control.
2232 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2234 ret
= btrfs_parse_options(tree_root
, options
);
2240 features
= btrfs_super_incompat_flags(disk_super
) &
2241 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2243 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2244 "unsupported optional features (%Lx).\n",
2245 (unsigned long long)features
);
2250 if (btrfs_super_leafsize(disk_super
) !=
2251 btrfs_super_nodesize(disk_super
)) {
2252 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2253 "blocksizes don't match. node %d leaf %d\n",
2254 btrfs_super_nodesize(disk_super
),
2255 btrfs_super_leafsize(disk_super
));
2259 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2260 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2261 "blocksize (%d) was too large\n",
2262 btrfs_super_leafsize(disk_super
));
2267 features
= btrfs_super_incompat_flags(disk_super
);
2268 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2269 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2270 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2273 * flag our filesystem as having big metadata blocks if
2274 * they are bigger than the page size
2276 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2277 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2278 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2279 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2282 nodesize
= btrfs_super_nodesize(disk_super
);
2283 leafsize
= btrfs_super_leafsize(disk_super
);
2284 sectorsize
= btrfs_super_sectorsize(disk_super
);
2285 stripesize
= btrfs_super_stripesize(disk_super
);
2288 * mixed block groups end up with duplicate but slightly offset
2289 * extent buffers for the same range. It leads to corruptions
2291 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2292 (sectorsize
!= leafsize
)) {
2293 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2294 "are not allowed for mixed block groups on %s\n",
2299 btrfs_set_super_incompat_flags(disk_super
, features
);
2301 features
= btrfs_super_compat_ro_flags(disk_super
) &
2302 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2303 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2304 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2305 "unsupported option features (%Lx).\n",
2306 (unsigned long long)features
);
2311 btrfs_init_workers(&fs_info
->generic_worker
,
2312 "genwork", 1, NULL
);
2314 btrfs_init_workers(&fs_info
->workers
, "worker",
2315 fs_info
->thread_pool_size
,
2316 &fs_info
->generic_worker
);
2318 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2319 fs_info
->thread_pool_size
,
2320 &fs_info
->generic_worker
);
2322 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2323 fs_info
->thread_pool_size
,
2324 &fs_info
->generic_worker
);
2326 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2327 min_t(u64
, fs_devices
->num_devices
,
2328 fs_info
->thread_pool_size
),
2329 &fs_info
->generic_worker
);
2331 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2332 2, &fs_info
->generic_worker
);
2334 /* a higher idle thresh on the submit workers makes it much more
2335 * likely that bios will be send down in a sane order to the
2338 fs_info
->submit_workers
.idle_thresh
= 64;
2340 fs_info
->workers
.idle_thresh
= 16;
2341 fs_info
->workers
.ordered
= 1;
2343 fs_info
->delalloc_workers
.idle_thresh
= 2;
2344 fs_info
->delalloc_workers
.ordered
= 1;
2346 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2347 &fs_info
->generic_worker
);
2348 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2349 fs_info
->thread_pool_size
,
2350 &fs_info
->generic_worker
);
2351 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2352 fs_info
->thread_pool_size
,
2353 &fs_info
->generic_worker
);
2354 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2355 "endio-meta-write", fs_info
->thread_pool_size
,
2356 &fs_info
->generic_worker
);
2357 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2358 "endio-raid56", fs_info
->thread_pool_size
,
2359 &fs_info
->generic_worker
);
2360 btrfs_init_workers(&fs_info
->rmw_workers
,
2361 "rmw", fs_info
->thread_pool_size
,
2362 &fs_info
->generic_worker
);
2363 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2364 fs_info
->thread_pool_size
,
2365 &fs_info
->generic_worker
);
2366 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2367 1, &fs_info
->generic_worker
);
2368 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2369 fs_info
->thread_pool_size
,
2370 &fs_info
->generic_worker
);
2371 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2372 fs_info
->thread_pool_size
,
2373 &fs_info
->generic_worker
);
2376 * endios are largely parallel and should have a very
2379 fs_info
->endio_workers
.idle_thresh
= 4;
2380 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2381 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2382 fs_info
->rmw_workers
.idle_thresh
= 2;
2384 fs_info
->endio_write_workers
.idle_thresh
= 2;
2385 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2386 fs_info
->readahead_workers
.idle_thresh
= 2;
2389 * btrfs_start_workers can really only fail because of ENOMEM so just
2390 * return -ENOMEM if any of these fail.
2392 ret
= btrfs_start_workers(&fs_info
->workers
);
2393 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2394 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2395 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2396 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2397 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2398 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2399 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2400 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2401 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2402 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2403 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2404 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2405 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2406 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2407 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2410 goto fail_sb_buffer
;
2413 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2414 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2415 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2417 tree_root
->nodesize
= nodesize
;
2418 tree_root
->leafsize
= leafsize
;
2419 tree_root
->sectorsize
= sectorsize
;
2420 tree_root
->stripesize
= stripesize
;
2422 sb
->s_blocksize
= sectorsize
;
2423 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2425 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
2426 sizeof(disk_super
->magic
))) {
2427 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2428 goto fail_sb_buffer
;
2431 if (sectorsize
!= PAGE_SIZE
) {
2432 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2433 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2434 goto fail_sb_buffer
;
2437 mutex_lock(&fs_info
->chunk_mutex
);
2438 ret
= btrfs_read_sys_array(tree_root
);
2439 mutex_unlock(&fs_info
->chunk_mutex
);
2441 printk(KERN_WARNING
"btrfs: failed to read the system "
2442 "array on %s\n", sb
->s_id
);
2443 goto fail_sb_buffer
;
2446 blocksize
= btrfs_level_size(tree_root
,
2447 btrfs_super_chunk_root_level(disk_super
));
2448 generation
= btrfs_super_chunk_root_generation(disk_super
);
2450 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2451 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2453 chunk_root
->node
= read_tree_block(chunk_root
,
2454 btrfs_super_chunk_root(disk_super
),
2455 blocksize
, generation
);
2456 BUG_ON(!chunk_root
->node
); /* -ENOMEM */
2457 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2458 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2460 goto fail_tree_roots
;
2462 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2463 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2465 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2466 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2469 ret
= btrfs_read_chunk_tree(chunk_root
);
2471 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2473 goto fail_tree_roots
;
2477 * keep the device that is marked to be the target device for the
2478 * dev_replace procedure
2480 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2482 if (!fs_devices
->latest_bdev
) {
2483 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2485 goto fail_tree_roots
;
2489 blocksize
= btrfs_level_size(tree_root
,
2490 btrfs_super_root_level(disk_super
));
2491 generation
= btrfs_super_generation(disk_super
);
2493 tree_root
->node
= read_tree_block(tree_root
,
2494 btrfs_super_root(disk_super
),
2495 blocksize
, generation
);
2496 if (!tree_root
->node
||
2497 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2498 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2501 goto recovery_tree_root
;
2504 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2505 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2507 ret
= find_and_setup_root(tree_root
, fs_info
,
2508 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2510 goto recovery_tree_root
;
2511 extent_root
->track_dirty
= 1;
2513 ret
= find_and_setup_root(tree_root
, fs_info
,
2514 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2516 goto recovery_tree_root
;
2517 dev_root
->track_dirty
= 1;
2519 ret
= find_and_setup_root(tree_root
, fs_info
,
2520 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2522 goto recovery_tree_root
;
2523 csum_root
->track_dirty
= 1;
2525 ret
= find_and_setup_root(tree_root
, fs_info
,
2526 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2529 quota_root
= fs_info
->quota_root
= NULL
;
2531 quota_root
->track_dirty
= 1;
2532 fs_info
->quota_enabled
= 1;
2533 fs_info
->pending_quota_state
= 1;
2536 fs_info
->generation
= generation
;
2537 fs_info
->last_trans_committed
= generation
;
2539 ret
= btrfs_recover_balance(fs_info
);
2541 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2542 goto fail_block_groups
;
2545 ret
= btrfs_init_dev_stats(fs_info
);
2547 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2549 goto fail_block_groups
;
2552 ret
= btrfs_init_dev_replace(fs_info
);
2554 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2555 goto fail_block_groups
;
2558 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2560 ret
= btrfs_init_space_info(fs_info
);
2562 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2563 goto fail_block_groups
;
2566 ret
= btrfs_read_block_groups(extent_root
);
2568 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2569 goto fail_block_groups
;
2571 fs_info
->num_tolerated_disk_barrier_failures
=
2572 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2573 if (fs_info
->fs_devices
->missing_devices
>
2574 fs_info
->num_tolerated_disk_barrier_failures
&&
2575 !(sb
->s_flags
& MS_RDONLY
)) {
2577 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2578 goto fail_block_groups
;
2581 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2583 if (IS_ERR(fs_info
->cleaner_kthread
))
2584 goto fail_block_groups
;
2586 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2588 "btrfs-transaction");
2589 if (IS_ERR(fs_info
->transaction_kthread
))
2592 if (!btrfs_test_opt(tree_root
, SSD
) &&
2593 !btrfs_test_opt(tree_root
, NOSSD
) &&
2594 !fs_info
->fs_devices
->rotating
) {
2595 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2597 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2600 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2601 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2602 ret
= btrfsic_mount(tree_root
, fs_devices
,
2603 btrfs_test_opt(tree_root
,
2604 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2606 fs_info
->check_integrity_print_mask
);
2608 printk(KERN_WARNING
"btrfs: failed to initialize"
2609 " integrity check module %s\n", sb
->s_id
);
2612 ret
= btrfs_read_qgroup_config(fs_info
);
2614 goto fail_trans_kthread
;
2616 /* do not make disk changes in broken FS */
2617 if (btrfs_super_log_root(disk_super
) != 0) {
2618 u64 bytenr
= btrfs_super_log_root(disk_super
);
2620 if (fs_devices
->rw_devices
== 0) {
2621 printk(KERN_WARNING
"Btrfs log replay required "
2627 btrfs_level_size(tree_root
,
2628 btrfs_super_log_root_level(disk_super
));
2630 log_tree_root
= btrfs_alloc_root(fs_info
);
2631 if (!log_tree_root
) {
2636 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2637 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2639 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2642 /* returns with log_tree_root freed on success */
2643 ret
= btrfs_recover_log_trees(log_tree_root
);
2645 btrfs_error(tree_root
->fs_info
, ret
,
2646 "Failed to recover log tree");
2647 free_extent_buffer(log_tree_root
->node
);
2648 kfree(log_tree_root
);
2649 goto fail_trans_kthread
;
2652 if (sb
->s_flags
& MS_RDONLY
) {
2653 ret
= btrfs_commit_super(tree_root
);
2655 goto fail_trans_kthread
;
2659 ret
= btrfs_find_orphan_roots(tree_root
);
2661 goto fail_trans_kthread
;
2663 if (!(sb
->s_flags
& MS_RDONLY
)) {
2664 ret
= btrfs_cleanup_fs_roots(fs_info
);
2666 goto fail_trans_kthread
;
2668 ret
= btrfs_recover_relocation(tree_root
);
2671 "btrfs: failed to recover relocation\n");
2677 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2678 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2679 location
.offset
= (u64
)-1;
2681 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2682 if (!fs_info
->fs_root
)
2684 if (IS_ERR(fs_info
->fs_root
)) {
2685 err
= PTR_ERR(fs_info
->fs_root
);
2689 if (sb
->s_flags
& MS_RDONLY
)
2692 down_read(&fs_info
->cleanup_work_sem
);
2693 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2694 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2695 up_read(&fs_info
->cleanup_work_sem
);
2696 close_ctree(tree_root
);
2699 up_read(&fs_info
->cleanup_work_sem
);
2701 ret
= btrfs_resume_balance_async(fs_info
);
2703 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2704 close_ctree(tree_root
);
2708 ret
= btrfs_resume_dev_replace_async(fs_info
);
2710 pr_warn("btrfs: failed to resume dev_replace\n");
2711 close_ctree(tree_root
);
2718 btrfs_free_qgroup_config(fs_info
);
2720 kthread_stop(fs_info
->transaction_kthread
);
2722 kthread_stop(fs_info
->cleaner_kthread
);
2725 * make sure we're done with the btree inode before we stop our
2728 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2729 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2732 btrfs_free_block_groups(fs_info
);
2735 free_root_pointers(fs_info
, 1);
2738 btrfs_stop_workers(&fs_info
->generic_worker
);
2739 btrfs_stop_workers(&fs_info
->readahead_workers
);
2740 btrfs_stop_workers(&fs_info
->fixup_workers
);
2741 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2742 btrfs_stop_workers(&fs_info
->workers
);
2743 btrfs_stop_workers(&fs_info
->endio_workers
);
2744 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2745 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
2746 btrfs_stop_workers(&fs_info
->rmw_workers
);
2747 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2748 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2749 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2750 btrfs_stop_workers(&fs_info
->submit_workers
);
2751 btrfs_stop_workers(&fs_info
->delayed_workers
);
2752 btrfs_stop_workers(&fs_info
->caching_workers
);
2753 btrfs_stop_workers(&fs_info
->flush_workers
);
2756 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2758 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2759 iput(fs_info
->btree_inode
);
2761 bdi_destroy(&fs_info
->bdi
);
2763 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2765 btrfs_free_stripe_hash_table(fs_info
);
2766 btrfs_close_devices(fs_info
->fs_devices
);
2770 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2771 goto fail_tree_roots
;
2773 free_root_pointers(fs_info
, 0);
2775 /* don't use the log in recovery mode, it won't be valid */
2776 btrfs_set_super_log_root(disk_super
, 0);
2778 /* we can't trust the free space cache either */
2779 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2781 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2782 &num_backups_tried
, &backup_index
);
2784 goto fail_block_groups
;
2785 goto retry_root_backup
;
2788 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2791 set_buffer_uptodate(bh
);
2793 struct btrfs_device
*device
= (struct btrfs_device
*)
2796 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2797 "I/O error on %s\n",
2798 rcu_str_deref(device
->name
));
2799 /* note, we dont' set_buffer_write_io_error because we have
2800 * our own ways of dealing with the IO errors
2802 clear_buffer_uptodate(bh
);
2803 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2809 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2811 struct buffer_head
*bh
;
2812 struct buffer_head
*latest
= NULL
;
2813 struct btrfs_super_block
*super
;
2818 /* we would like to check all the supers, but that would make
2819 * a btrfs mount succeed after a mkfs from a different FS.
2820 * So, we need to add a special mount option to scan for
2821 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2823 for (i
= 0; i
< 1; i
++) {
2824 bytenr
= btrfs_sb_offset(i
);
2825 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2827 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2831 super
= (struct btrfs_super_block
*)bh
->b_data
;
2832 if (btrfs_super_bytenr(super
) != bytenr
||
2833 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
2834 sizeof(super
->magic
))) {
2839 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2842 transid
= btrfs_super_generation(super
);
2851 * this should be called twice, once with wait == 0 and
2852 * once with wait == 1. When wait == 0 is done, all the buffer heads
2853 * we write are pinned.
2855 * They are released when wait == 1 is done.
2856 * max_mirrors must be the same for both runs, and it indicates how
2857 * many supers on this one device should be written.
2859 * max_mirrors == 0 means to write them all.
2861 static int write_dev_supers(struct btrfs_device
*device
,
2862 struct btrfs_super_block
*sb
,
2863 int do_barriers
, int wait
, int max_mirrors
)
2865 struct buffer_head
*bh
;
2872 if (max_mirrors
== 0)
2873 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2875 for (i
= 0; i
< max_mirrors
; i
++) {
2876 bytenr
= btrfs_sb_offset(i
);
2877 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2881 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2882 BTRFS_SUPER_INFO_SIZE
);
2885 if (!buffer_uptodate(bh
))
2888 /* drop our reference */
2891 /* drop the reference from the wait == 0 run */
2895 btrfs_set_super_bytenr(sb
, bytenr
);
2898 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2899 BTRFS_CSUM_SIZE
, crc
,
2900 BTRFS_SUPER_INFO_SIZE
-
2902 btrfs_csum_final(crc
, sb
->csum
);
2905 * one reference for us, and we leave it for the
2908 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2909 BTRFS_SUPER_INFO_SIZE
);
2910 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2912 /* one reference for submit_bh */
2915 set_buffer_uptodate(bh
);
2917 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2918 bh
->b_private
= device
;
2922 * we fua the first super. The others we allow
2925 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
2929 return errors
< i
? 0 : -1;
2933 * endio for the write_dev_flush, this will wake anyone waiting
2934 * for the barrier when it is done
2936 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
2939 if (err
== -EOPNOTSUPP
)
2940 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
2941 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2943 if (bio
->bi_private
)
2944 complete(bio
->bi_private
);
2949 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2950 * sent down. With wait == 1, it waits for the previous flush.
2952 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2955 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
2960 if (device
->nobarriers
)
2964 bio
= device
->flush_bio
;
2968 wait_for_completion(&device
->flush_wait
);
2970 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
2971 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2972 rcu_str_deref(device
->name
));
2973 device
->nobarriers
= 1;
2974 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
2976 btrfs_dev_stat_inc_and_print(device
,
2977 BTRFS_DEV_STAT_FLUSH_ERRS
);
2980 /* drop the reference from the wait == 0 run */
2982 device
->flush_bio
= NULL
;
2988 * one reference for us, and we leave it for the
2991 device
->flush_bio
= NULL
;
2992 bio
= bio_alloc(GFP_NOFS
, 0);
2996 bio
->bi_end_io
= btrfs_end_empty_barrier
;
2997 bio
->bi_bdev
= device
->bdev
;
2998 init_completion(&device
->flush_wait
);
2999 bio
->bi_private
= &device
->flush_wait
;
3000 device
->flush_bio
= bio
;
3003 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3009 * send an empty flush down to each device in parallel,
3010 * then wait for them
3012 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3014 struct list_head
*head
;
3015 struct btrfs_device
*dev
;
3016 int errors_send
= 0;
3017 int errors_wait
= 0;
3020 /* send down all the barriers */
3021 head
= &info
->fs_devices
->devices
;
3022 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3027 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3030 ret
= write_dev_flush(dev
, 0);
3035 /* wait for all the barriers */
3036 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3041 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3044 ret
= write_dev_flush(dev
, 1);
3048 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3049 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3054 int btrfs_calc_num_tolerated_disk_barrier_failures(
3055 struct btrfs_fs_info
*fs_info
)
3057 struct btrfs_ioctl_space_info space
;
3058 struct btrfs_space_info
*sinfo
;
3059 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3060 BTRFS_BLOCK_GROUP_SYSTEM
,
3061 BTRFS_BLOCK_GROUP_METADATA
,
3062 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3066 int num_tolerated_disk_barrier_failures
=
3067 (int)fs_info
->fs_devices
->num_devices
;
3069 for (i
= 0; i
< num_types
; i
++) {
3070 struct btrfs_space_info
*tmp
;
3074 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3075 if (tmp
->flags
== types
[i
]) {
3085 down_read(&sinfo
->groups_sem
);
3086 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3087 if (!list_empty(&sinfo
->block_groups
[c
])) {
3090 btrfs_get_block_group_info(
3091 &sinfo
->block_groups
[c
], &space
);
3092 if (space
.total_bytes
== 0 ||
3093 space
.used_bytes
== 0)
3095 flags
= space
.flags
;
3098 * 0: if dup, single or RAID0 is configured for
3099 * any of metadata, system or data, else
3100 * 1: if RAID5 is configured, or if RAID1 or
3101 * RAID10 is configured and only two mirrors
3103 * 2: if RAID6 is configured, else
3104 * num_mirrors - 1: if RAID1 or RAID10 is
3105 * configured and more than
3106 * 2 mirrors are used.
3108 if (num_tolerated_disk_barrier_failures
> 0 &&
3109 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3110 BTRFS_BLOCK_GROUP_RAID0
)) ||
3111 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3113 num_tolerated_disk_barrier_failures
= 0;
3114 else if (num_tolerated_disk_barrier_failures
> 1) {
3115 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3116 BTRFS_BLOCK_GROUP_RAID5
|
3117 BTRFS_BLOCK_GROUP_RAID10
)) {
3118 num_tolerated_disk_barrier_failures
= 1;
3120 BTRFS_BLOCK_GROUP_RAID5
) {
3121 num_tolerated_disk_barrier_failures
= 2;
3126 up_read(&sinfo
->groups_sem
);
3129 return num_tolerated_disk_barrier_failures
;
3132 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3134 struct list_head
*head
;
3135 struct btrfs_device
*dev
;
3136 struct btrfs_super_block
*sb
;
3137 struct btrfs_dev_item
*dev_item
;
3141 int total_errors
= 0;
3144 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3145 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3146 backup_super_roots(root
->fs_info
);
3148 sb
= root
->fs_info
->super_for_commit
;
3149 dev_item
= &sb
->dev_item
;
3151 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3152 head
= &root
->fs_info
->fs_devices
->devices
;
3155 ret
= barrier_all_devices(root
->fs_info
);
3158 &root
->fs_info
->fs_devices
->device_list_mutex
);
3159 btrfs_error(root
->fs_info
, ret
,
3160 "errors while submitting device barriers.");
3165 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3170 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3173 btrfs_set_stack_device_generation(dev_item
, 0);
3174 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3175 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3176 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3177 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3178 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3179 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3180 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3181 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3182 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3184 flags
= btrfs_super_flags(sb
);
3185 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3187 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3191 if (total_errors
> max_errors
) {
3192 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3195 /* This shouldn't happen. FUA is masked off if unsupported */
3200 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3203 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3206 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3210 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3211 if (total_errors
> max_errors
) {
3212 btrfs_error(root
->fs_info
, -EIO
,
3213 "%d errors while writing supers", total_errors
);
3219 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3220 struct btrfs_root
*root
, int max_mirrors
)
3224 ret
= write_all_supers(root
, max_mirrors
);
3228 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3230 spin_lock(&fs_info
->fs_roots_radix_lock
);
3231 radix_tree_delete(&fs_info
->fs_roots_radix
,
3232 (unsigned long)root
->root_key
.objectid
);
3233 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3235 if (btrfs_root_refs(&root
->root_item
) == 0)
3236 synchronize_srcu(&fs_info
->subvol_srcu
);
3238 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3239 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3243 static void free_fs_root(struct btrfs_root
*root
)
3245 iput(root
->cache_inode
);
3246 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3248 free_anon_bdev(root
->anon_dev
);
3249 free_extent_buffer(root
->node
);
3250 free_extent_buffer(root
->commit_root
);
3251 kfree(root
->free_ino_ctl
);
3252 kfree(root
->free_ino_pinned
);
3257 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
3260 struct btrfs_root
*gang
[8];
3263 while (!list_empty(&fs_info
->dead_roots
)) {
3264 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
3265 struct btrfs_root
, root_list
);
3266 list_del(&gang
[0]->root_list
);
3268 if (gang
[0]->in_radix
) {
3269 btrfs_free_fs_root(fs_info
, gang
[0]);
3271 free_extent_buffer(gang
[0]->node
);
3272 free_extent_buffer(gang
[0]->commit_root
);
3278 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3283 for (i
= 0; i
< ret
; i
++)
3284 btrfs_free_fs_root(fs_info
, gang
[i
]);
3288 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3290 u64 root_objectid
= 0;
3291 struct btrfs_root
*gang
[8];
3296 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3297 (void **)gang
, root_objectid
,
3302 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3303 for (i
= 0; i
< ret
; i
++) {
3306 root_objectid
= gang
[i
]->root_key
.objectid
;
3307 err
= btrfs_orphan_cleanup(gang
[i
]);
3316 int btrfs_commit_super(struct btrfs_root
*root
)
3318 struct btrfs_trans_handle
*trans
;
3321 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3322 btrfs_run_delayed_iputs(root
);
3323 btrfs_clean_old_snapshots(root
);
3324 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3326 /* wait until ongoing cleanup work done */
3327 down_write(&root
->fs_info
->cleanup_work_sem
);
3328 up_write(&root
->fs_info
->cleanup_work_sem
);
3330 trans
= btrfs_join_transaction(root
);
3332 return PTR_ERR(trans
);
3333 ret
= btrfs_commit_transaction(trans
, root
);
3336 /* run commit again to drop the original snapshot */
3337 trans
= btrfs_join_transaction(root
);
3339 return PTR_ERR(trans
);
3340 ret
= btrfs_commit_transaction(trans
, root
);
3343 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3345 btrfs_error(root
->fs_info
, ret
,
3346 "Failed to sync btree inode to disk.");
3350 ret
= write_ctree_super(NULL
, root
, 0);
3354 int close_ctree(struct btrfs_root
*root
)
3356 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3359 fs_info
->closing
= 1;
3362 /* pause restriper - we want to resume on mount */
3363 btrfs_pause_balance(fs_info
);
3365 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3367 btrfs_scrub_cancel(fs_info
);
3369 /* wait for any defraggers to finish */
3370 wait_event(fs_info
->transaction_wait
,
3371 (atomic_read(&fs_info
->defrag_running
) == 0));
3373 /* clear out the rbtree of defraggable inodes */
3374 btrfs_cleanup_defrag_inodes(fs_info
);
3376 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3377 ret
= btrfs_commit_super(root
);
3379 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3382 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
3383 btrfs_error_commit_super(root
);
3385 btrfs_put_block_group_cache(fs_info
);
3387 kthread_stop(fs_info
->transaction_kthread
);
3388 kthread_stop(fs_info
->cleaner_kthread
);
3390 fs_info
->closing
= 2;
3393 btrfs_free_qgroup_config(root
->fs_info
);
3395 if (fs_info
->delalloc_bytes
) {
3396 printk(KERN_INFO
"btrfs: at unmount delalloc count %llu\n",
3397 (unsigned long long)fs_info
->delalloc_bytes
);
3400 free_extent_buffer(fs_info
->extent_root
->node
);
3401 free_extent_buffer(fs_info
->extent_root
->commit_root
);
3402 free_extent_buffer(fs_info
->tree_root
->node
);
3403 free_extent_buffer(fs_info
->tree_root
->commit_root
);
3404 free_extent_buffer(fs_info
->chunk_root
->node
);
3405 free_extent_buffer(fs_info
->chunk_root
->commit_root
);
3406 free_extent_buffer(fs_info
->dev_root
->node
);
3407 free_extent_buffer(fs_info
->dev_root
->commit_root
);
3408 free_extent_buffer(fs_info
->csum_root
->node
);
3409 free_extent_buffer(fs_info
->csum_root
->commit_root
);
3410 if (fs_info
->quota_root
) {
3411 free_extent_buffer(fs_info
->quota_root
->node
);
3412 free_extent_buffer(fs_info
->quota_root
->commit_root
);
3415 btrfs_free_block_groups(fs_info
);
3417 del_fs_roots(fs_info
);
3419 iput(fs_info
->btree_inode
);
3421 btrfs_stop_workers(&fs_info
->generic_worker
);
3422 btrfs_stop_workers(&fs_info
->fixup_workers
);
3423 btrfs_stop_workers(&fs_info
->delalloc_workers
);
3424 btrfs_stop_workers(&fs_info
->workers
);
3425 btrfs_stop_workers(&fs_info
->endio_workers
);
3426 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
3427 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
3428 btrfs_stop_workers(&fs_info
->rmw_workers
);
3429 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
3430 btrfs_stop_workers(&fs_info
->endio_write_workers
);
3431 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
3432 btrfs_stop_workers(&fs_info
->submit_workers
);
3433 btrfs_stop_workers(&fs_info
->delayed_workers
);
3434 btrfs_stop_workers(&fs_info
->caching_workers
);
3435 btrfs_stop_workers(&fs_info
->readahead_workers
);
3436 btrfs_stop_workers(&fs_info
->flush_workers
);
3438 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3439 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3440 btrfsic_unmount(root
, fs_info
->fs_devices
);
3443 btrfs_close_devices(fs_info
->fs_devices
);
3444 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3446 bdi_destroy(&fs_info
->bdi
);
3447 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3449 btrfs_free_stripe_hash_table(fs_info
);
3454 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3458 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3460 ret
= extent_buffer_uptodate(buf
);
3464 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3465 parent_transid
, atomic
);
3471 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3473 return set_extent_buffer_uptodate(buf
);
3476 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3478 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3479 u64 transid
= btrfs_header_generation(buf
);
3482 btrfs_assert_tree_locked(buf
);
3483 if (transid
!= root
->fs_info
->generation
)
3484 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3485 "found %llu running %llu\n",
3486 (unsigned long long)buf
->start
,
3487 (unsigned long long)transid
,
3488 (unsigned long long)root
->fs_info
->generation
);
3489 was_dirty
= set_extent_buffer_dirty(buf
);
3491 spin_lock(&root
->fs_info
->delalloc_lock
);
3492 root
->fs_info
->dirty_metadata_bytes
+= buf
->len
;
3493 spin_unlock(&root
->fs_info
->delalloc_lock
);
3497 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3501 * looks as though older kernels can get into trouble with
3502 * this code, they end up stuck in balance_dirty_pages forever
3505 unsigned long thresh
= 32 * 1024 * 1024;
3507 if (current
->flags
& PF_MEMALLOC
)
3511 btrfs_balance_delayed_items(root
);
3513 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
3515 if (num_dirty
> thresh
) {
3516 balance_dirty_pages_ratelimited_nr(
3517 root
->fs_info
->btree_inode
->i_mapping
, 1);
3522 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3524 __btrfs_btree_balance_dirty(root
, 1);
3527 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3529 __btrfs_btree_balance_dirty(root
, 0);
3532 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3534 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3535 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3538 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3541 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3542 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3552 void btrfs_error_commit_super(struct btrfs_root
*root
)
3554 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3555 btrfs_run_delayed_iputs(root
);
3556 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3558 down_write(&root
->fs_info
->cleanup_work_sem
);
3559 up_write(&root
->fs_info
->cleanup_work_sem
);
3561 /* cleanup FS via transaction */
3562 btrfs_cleanup_transaction(root
);
3565 static void btrfs_destroy_ordered_operations(struct btrfs_root
*root
)
3567 struct btrfs_inode
*btrfs_inode
;
3568 struct list_head splice
;
3570 INIT_LIST_HEAD(&splice
);
3572 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3573 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3575 list_splice_init(&root
->fs_info
->ordered_operations
, &splice
);
3576 while (!list_empty(&splice
)) {
3577 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3578 ordered_operations
);
3580 list_del_init(&btrfs_inode
->ordered_operations
);
3582 btrfs_invalidate_inodes(btrfs_inode
->root
);
3585 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3586 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3589 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3591 struct list_head splice
;
3592 struct btrfs_ordered_extent
*ordered
;
3593 struct inode
*inode
;
3595 INIT_LIST_HEAD(&splice
);
3597 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3599 list_splice_init(&root
->fs_info
->ordered_extents
, &splice
);
3600 while (!list_empty(&splice
)) {
3601 ordered
= list_entry(splice
.next
, struct btrfs_ordered_extent
,
3604 list_del_init(&ordered
->root_extent_list
);
3605 atomic_inc(&ordered
->refs
);
3607 /* the inode may be getting freed (in sys_unlink path). */
3608 inode
= igrab(ordered
->inode
);
3610 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3614 atomic_set(&ordered
->refs
, 1);
3615 btrfs_put_ordered_extent(ordered
);
3617 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3620 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3623 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3624 struct btrfs_root
*root
)
3626 struct rb_node
*node
;
3627 struct btrfs_delayed_ref_root
*delayed_refs
;
3628 struct btrfs_delayed_ref_node
*ref
;
3631 delayed_refs
= &trans
->delayed_refs
;
3633 spin_lock(&delayed_refs
->lock
);
3634 if (delayed_refs
->num_entries
== 0) {
3635 spin_unlock(&delayed_refs
->lock
);
3636 printk(KERN_INFO
"delayed_refs has NO entry\n");
3640 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3641 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3643 atomic_set(&ref
->refs
, 1);
3644 if (btrfs_delayed_ref_is_head(ref
)) {
3645 struct btrfs_delayed_ref_head
*head
;
3647 head
= btrfs_delayed_node_to_head(ref
);
3648 if (!mutex_trylock(&head
->mutex
)) {
3649 atomic_inc(&ref
->refs
);
3650 spin_unlock(&delayed_refs
->lock
);
3652 /* Need to wait for the delayed ref to run */
3653 mutex_lock(&head
->mutex
);
3654 mutex_unlock(&head
->mutex
);
3655 btrfs_put_delayed_ref(ref
);
3657 spin_lock(&delayed_refs
->lock
);
3661 kfree(head
->extent_op
);
3662 delayed_refs
->num_heads
--;
3663 if (list_empty(&head
->cluster
))
3664 delayed_refs
->num_heads_ready
--;
3665 list_del_init(&head
->cluster
);
3668 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3669 delayed_refs
->num_entries
--;
3671 spin_unlock(&delayed_refs
->lock
);
3672 btrfs_put_delayed_ref(ref
);
3675 spin_lock(&delayed_refs
->lock
);
3678 spin_unlock(&delayed_refs
->lock
);
3683 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
)
3685 struct btrfs_pending_snapshot
*snapshot
;
3686 struct list_head splice
;
3688 INIT_LIST_HEAD(&splice
);
3690 list_splice_init(&t
->pending_snapshots
, &splice
);
3692 while (!list_empty(&splice
)) {
3693 snapshot
= list_entry(splice
.next
,
3694 struct btrfs_pending_snapshot
,
3697 list_del_init(&snapshot
->list
);
3703 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3705 struct btrfs_inode
*btrfs_inode
;
3706 struct list_head splice
;
3708 INIT_LIST_HEAD(&splice
);
3710 spin_lock(&root
->fs_info
->delalloc_lock
);
3711 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3713 while (!list_empty(&splice
)) {
3714 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3717 list_del_init(&btrfs_inode
->delalloc_inodes
);
3719 btrfs_invalidate_inodes(btrfs_inode
->root
);
3722 spin_unlock(&root
->fs_info
->delalloc_lock
);
3725 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3726 struct extent_io_tree
*dirty_pages
,
3731 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
3732 struct extent_buffer
*eb
;
3736 unsigned long index
;
3739 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3744 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3745 while (start
<= end
) {
3746 index
= start
>> PAGE_CACHE_SHIFT
;
3747 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
3748 page
= find_get_page(btree_inode
->i_mapping
, index
);
3751 offset
= page_offset(page
);
3753 spin_lock(&dirty_pages
->buffer_lock
);
3754 eb
= radix_tree_lookup(
3755 &(&BTRFS_I(page
->mapping
->host
)->io_tree
)->buffer
,
3756 offset
>> PAGE_CACHE_SHIFT
);
3757 spin_unlock(&dirty_pages
->buffer_lock
);
3759 ret
= test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3761 if (PageWriteback(page
))
3762 end_page_writeback(page
);
3765 if (PageDirty(page
)) {
3766 clear_page_dirty_for_io(page
);
3767 spin_lock_irq(&page
->mapping
->tree_lock
);
3768 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3770 PAGECACHE_TAG_DIRTY
);
3771 spin_unlock_irq(&page
->mapping
->tree_lock
);
3775 page_cache_release(page
);
3782 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3783 struct extent_io_tree
*pinned_extents
)
3785 struct extent_io_tree
*unpin
;
3791 unpin
= pinned_extents
;
3794 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3795 EXTENT_DIRTY
, NULL
);
3800 if (btrfs_test_opt(root
, DISCARD
))
3801 ret
= btrfs_error_discard_extent(root
, start
,
3805 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3806 btrfs_error_unpin_extent_range(root
, start
, end
);
3811 if (unpin
== &root
->fs_info
->freed_extents
[0])
3812 unpin
= &root
->fs_info
->freed_extents
[1];
3814 unpin
= &root
->fs_info
->freed_extents
[0];
3822 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3823 struct btrfs_root
*root
)
3825 btrfs_destroy_delayed_refs(cur_trans
, root
);
3826 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3827 cur_trans
->dirty_pages
.dirty_bytes
);
3829 /* FIXME: cleanup wait for commit */
3830 cur_trans
->in_commit
= 1;
3831 cur_trans
->blocked
= 1;
3832 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3834 cur_trans
->blocked
= 0;
3835 wake_up(&root
->fs_info
->transaction_wait
);
3837 cur_trans
->commit_done
= 1;
3838 wake_up(&cur_trans
->commit_wait
);
3840 btrfs_destroy_delayed_inodes(root
);
3841 btrfs_assert_delayed_root_empty(root
);
3843 btrfs_destroy_pending_snapshots(cur_trans
);
3845 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3847 btrfs_destroy_pinned_extent(root
,
3848 root
->fs_info
->pinned_extents
);
3851 memset(cur_trans, 0, sizeof(*cur_trans));
3852 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3856 int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3858 struct btrfs_transaction
*t
;
3861 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3863 spin_lock(&root
->fs_info
->trans_lock
);
3864 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3865 root
->fs_info
->trans_no_join
= 1;
3866 spin_unlock(&root
->fs_info
->trans_lock
);
3868 while (!list_empty(&list
)) {
3869 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3873 btrfs_destroy_ordered_operations(root
);
3875 btrfs_destroy_ordered_extents(root
);
3877 btrfs_destroy_delayed_refs(t
, root
);
3879 btrfs_block_rsv_release(root
,
3880 &root
->fs_info
->trans_block_rsv
,
3881 t
->dirty_pages
.dirty_bytes
);
3883 /* FIXME: cleanup wait for commit */
3887 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3888 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3892 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3893 wake_up(&root
->fs_info
->transaction_wait
);
3897 if (waitqueue_active(&t
->commit_wait
))
3898 wake_up(&t
->commit_wait
);
3900 btrfs_destroy_delayed_inodes(root
);
3901 btrfs_assert_delayed_root_empty(root
);
3903 btrfs_destroy_pending_snapshots(t
);
3905 btrfs_destroy_delalloc_inodes(root
);
3907 spin_lock(&root
->fs_info
->trans_lock
);
3908 root
->fs_info
->running_transaction
= NULL
;
3909 spin_unlock(&root
->fs_info
->trans_lock
);
3911 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3914 btrfs_destroy_pinned_extent(root
,
3915 root
->fs_info
->pinned_extents
);
3917 atomic_set(&t
->use_count
, 0);
3918 list_del_init(&t
->list
);
3919 memset(t
, 0, sizeof(*t
));
3920 kmem_cache_free(btrfs_transaction_cachep
, t
);
3923 spin_lock(&root
->fs_info
->trans_lock
);
3924 root
->fs_info
->trans_no_join
= 0;
3925 spin_unlock(&root
->fs_info
->trans_lock
);
3926 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3931 static struct extent_io_ops btree_extent_io_ops
= {
3932 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3933 .readpage_io_failed_hook
= btree_io_failed_hook
,
3934 .submit_bio_hook
= btree_submit_bio_hook
,
3935 /* note we're sharing with inode.c for the merge bio hook */
3936 .merge_bio_hook
= btrfs_merge_bio_hook
,