2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
53 #include <asm/cpufeature.h>
56 static struct extent_io_ops btree_extent_io_ops
;
57 static void end_workqueue_fn(struct btrfs_work
*work
);
58 static void free_fs_root(struct btrfs_root
*root
);
59 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
61 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
62 struct btrfs_root
*root
);
63 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
65 struct btrfs_root
*root
);
66 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
68 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
69 struct extent_io_tree
*dirty_pages
,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
72 struct extent_io_tree
*pinned_extents
);
75 * end_io_wq structs are used to do processing in task context when an IO is
76 * complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
83 struct btrfs_fs_info
*info
;
86 struct list_head list
;
87 struct btrfs_work work
;
91 * async submit bios are used to offload expensive checksumming
92 * onto the worker threads. They checksum file and metadata bios
93 * just before they are sent down the IO stack.
95 struct async_submit_bio
{
98 struct list_head list
;
99 extent_submit_bio_hook_t
*submit_bio_start
;
100 extent_submit_bio_hook_t
*submit_bio_done
;
103 unsigned long bio_flags
;
105 * bio_offset is optional, can be used if the pages in the bio
106 * can't tell us where in the file the bio should go
109 struct btrfs_work work
;
114 * Lockdep class keys for extent_buffer->lock's in this root. For a given
115 * eb, the lockdep key is determined by the btrfs_root it belongs to and
116 * the level the eb occupies in the tree.
118 * Different roots are used for different purposes and may nest inside each
119 * other and they require separate keysets. As lockdep keys should be
120 * static, assign keysets according to the purpose of the root as indicated
121 * by btrfs_root->objectid. This ensures that all special purpose roots
122 * have separate keysets.
124 * Lock-nesting across peer nodes is always done with the immediate parent
125 * node locked thus preventing deadlock. As lockdep doesn't know this, use
126 * subclass to avoid triggering lockdep warning in such cases.
128 * The key is set by the readpage_end_io_hook after the buffer has passed
129 * csum validation but before the pages are unlocked. It is also set by
130 * btrfs_init_new_buffer on freshly allocated blocks.
132 * We also add a check to make sure the highest level of the tree is the
133 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
134 * needs update as well.
136 #ifdef CONFIG_DEBUG_LOCK_ALLOC
137 # if BTRFS_MAX_LEVEL != 8
141 static struct btrfs_lockdep_keyset
{
142 u64 id
; /* root objectid */
143 const char *name_stem
; /* lock name stem */
144 char names
[BTRFS_MAX_LEVEL
+ 1][20];
145 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
146 } btrfs_lockdep_keysets
[] = {
147 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
148 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
149 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
150 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
151 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
152 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
153 { .id
= BTRFS_ORPHAN_OBJECTID
, .name_stem
= "orphan" },
154 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
155 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
156 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
157 { .id
= 0, .name_stem
= "tree" },
160 void __init
btrfs_init_lockdep(void)
164 /* initialize lockdep class names */
165 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
166 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
168 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
169 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
170 "btrfs-%s-%02d", ks
->name_stem
, j
);
174 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
177 struct btrfs_lockdep_keyset
*ks
;
179 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
181 /* find the matching keyset, id 0 is the default entry */
182 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
183 if (ks
->id
== objectid
)
186 lockdep_set_class_and_name(&eb
->lock
,
187 &ks
->keys
[level
], ks
->names
[level
]);
193 * extents on the btree inode are pretty simple, there's one extent
194 * that covers the entire device
196 static struct extent_map
*btree_get_extent(struct inode
*inode
,
197 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
200 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
201 struct extent_map
*em
;
204 read_lock(&em_tree
->lock
);
205 em
= lookup_extent_mapping(em_tree
, start
, len
);
208 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
209 read_unlock(&em_tree
->lock
);
212 read_unlock(&em_tree
->lock
);
214 em
= alloc_extent_map();
216 em
= ERR_PTR(-ENOMEM
);
221 em
->block_len
= (u64
)-1;
223 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
225 write_lock(&em_tree
->lock
);
226 ret
= add_extent_mapping(em_tree
, em
, 0);
227 if (ret
== -EEXIST
) {
229 em
= lookup_extent_mapping(em_tree
, start
, len
);
236 write_unlock(&em_tree
->lock
);
242 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
244 return crc32c(seed
, data
, len
);
247 void btrfs_csum_final(u32 crc
, char *result
)
249 put_unaligned_le32(~crc
, result
);
253 * compute the csum for a btree block, and either verify it or write it
254 * into the csum field of the block.
256 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
259 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
262 unsigned long cur_len
;
263 unsigned long offset
= BTRFS_CSUM_SIZE
;
265 unsigned long map_start
;
266 unsigned long map_len
;
269 unsigned long inline_result
;
271 len
= buf
->len
- offset
;
273 err
= map_private_extent_buffer(buf
, offset
, 32,
274 &kaddr
, &map_start
, &map_len
);
277 cur_len
= min(len
, map_len
- (offset
- map_start
));
278 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
283 if (csum_size
> sizeof(inline_result
)) {
284 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
288 result
= (char *)&inline_result
;
291 btrfs_csum_final(crc
, result
);
294 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
297 memcpy(&found
, result
, csum_size
);
299 read_extent_buffer(buf
, &val
, 0, csum_size
);
300 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
301 "failed on %llu wanted %X found %X "
303 root
->fs_info
->sb
->s_id
,
304 (unsigned long long)buf
->start
, val
, found
,
305 btrfs_header_level(buf
));
306 if (result
!= (char *)&inline_result
)
311 write_extent_buffer(buf
, result
, 0, csum_size
);
313 if (result
!= (char *)&inline_result
)
319 * we can't consider a given block up to date unless the transid of the
320 * block matches the transid in the parent node's pointer. This is how we
321 * detect blocks that either didn't get written at all or got written
322 * in the wrong place.
324 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
325 struct extent_buffer
*eb
, u64 parent_transid
,
328 struct extent_state
*cached_state
= NULL
;
331 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
337 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
339 if (extent_buffer_uptodate(eb
) &&
340 btrfs_header_generation(eb
) == parent_transid
) {
344 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
346 (unsigned long long)eb
->start
,
347 (unsigned long long)parent_transid
,
348 (unsigned long long)btrfs_header_generation(eb
));
350 clear_extent_buffer_uptodate(eb
);
352 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
353 &cached_state
, GFP_NOFS
);
358 * helper to read a given tree block, doing retries as required when
359 * the checksums don't match and we have alternate mirrors to try.
361 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
362 struct extent_buffer
*eb
,
363 u64 start
, u64 parent_transid
)
365 struct extent_io_tree
*io_tree
;
370 int failed_mirror
= 0;
372 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
373 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
375 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
377 btree_get_extent
, mirror_num
);
379 if (!verify_parent_transid(io_tree
, eb
,
387 * This buffer's crc is fine, but its contents are corrupted, so
388 * there is no reason to read the other copies, they won't be
391 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
394 num_copies
= btrfs_num_copies(root
->fs_info
,
399 if (!failed_mirror
) {
401 failed_mirror
= eb
->read_mirror
;
405 if (mirror_num
== failed_mirror
)
408 if (mirror_num
> num_copies
)
412 if (failed
&& !ret
&& failed_mirror
)
413 repair_eb_io_failure(root
, eb
, failed_mirror
);
419 * checksum a dirty tree block before IO. This has extra checks to make sure
420 * we only fill in the checksum field in the first page of a multi-page block
423 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
425 struct extent_io_tree
*tree
;
426 u64 start
= page_offset(page
);
428 struct extent_buffer
*eb
;
430 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
432 eb
= (struct extent_buffer
*)page
->private;
433 if (page
!= eb
->pages
[0])
435 found_start
= btrfs_header_bytenr(eb
);
436 if (found_start
!= start
) {
440 if (!PageUptodate(page
)) {
444 csum_tree_block(root
, eb
, 0);
448 static int check_tree_block_fsid(struct btrfs_root
*root
,
449 struct extent_buffer
*eb
)
451 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
452 u8 fsid
[BTRFS_UUID_SIZE
];
455 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
458 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
462 fs_devices
= fs_devices
->seed
;
467 #define CORRUPT(reason, eb, root, slot) \
468 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
469 "root=%llu, slot=%d\n", reason, \
470 (unsigned long long)btrfs_header_bytenr(eb), \
471 (unsigned long long)root->objectid, slot)
473 static noinline
int check_leaf(struct btrfs_root
*root
,
474 struct extent_buffer
*leaf
)
476 struct btrfs_key key
;
477 struct btrfs_key leaf_key
;
478 u32 nritems
= btrfs_header_nritems(leaf
);
484 /* Check the 0 item */
485 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
486 BTRFS_LEAF_DATA_SIZE(root
)) {
487 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
492 * Check to make sure each items keys are in the correct order and their
493 * offsets make sense. We only have to loop through nritems-1 because
494 * we check the current slot against the next slot, which verifies the
495 * next slot's offset+size makes sense and that the current's slot
498 for (slot
= 0; slot
< nritems
- 1; slot
++) {
499 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
500 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
502 /* Make sure the keys are in the right order */
503 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
504 CORRUPT("bad key order", leaf
, root
, slot
);
509 * Make sure the offset and ends are right, remember that the
510 * item data starts at the end of the leaf and grows towards the
513 if (btrfs_item_offset_nr(leaf
, slot
) !=
514 btrfs_item_end_nr(leaf
, slot
+ 1)) {
515 CORRUPT("slot offset bad", leaf
, root
, slot
);
520 * Check to make sure that we don't point outside of the leaf,
521 * just incase all the items are consistent to eachother, but
522 * all point outside of the leaf.
524 if (btrfs_item_end_nr(leaf
, slot
) >
525 BTRFS_LEAF_DATA_SIZE(root
)) {
526 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
534 struct extent_buffer
*find_eb_for_page(struct extent_io_tree
*tree
,
535 struct page
*page
, int max_walk
)
537 struct extent_buffer
*eb
;
538 u64 start
= page_offset(page
);
542 if (start
< max_walk
)
545 min_start
= start
- max_walk
;
547 while (start
>= min_start
) {
548 eb
= find_extent_buffer(tree
, start
, 0);
551 * we found an extent buffer and it contains our page
554 if (eb
->start
<= target
&&
555 eb
->start
+ eb
->len
> target
)
558 /* we found an extent buffer that wasn't for us */
559 free_extent_buffer(eb
);
564 start
-= PAGE_CACHE_SIZE
;
569 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
570 struct extent_state
*state
, int mirror
)
572 struct extent_io_tree
*tree
;
575 struct extent_buffer
*eb
;
576 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
583 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
584 eb
= (struct extent_buffer
*)page
->private;
586 /* the pending IO might have been the only thing that kept this buffer
587 * in memory. Make sure we have a ref for all this other checks
589 extent_buffer_get(eb
);
591 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
595 eb
->read_mirror
= mirror
;
596 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
601 found_start
= btrfs_header_bytenr(eb
);
602 if (found_start
!= eb
->start
) {
603 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
605 (unsigned long long)found_start
,
606 (unsigned long long)eb
->start
);
610 if (check_tree_block_fsid(root
, eb
)) {
611 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
612 (unsigned long long)eb
->start
);
616 found_level
= btrfs_header_level(eb
);
617 if (found_level
>= BTRFS_MAX_LEVEL
) {
618 btrfs_info(root
->fs_info
, "bad tree block level %d\n",
619 (int)btrfs_header_level(eb
));
624 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
627 ret
= csum_tree_block(root
, eb
, 1);
634 * If this is a leaf block and it is corrupt, set the corrupt bit so
635 * that we don't try and read the other copies of this block, just
638 if (found_level
== 0 && check_leaf(root
, eb
)) {
639 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
644 set_extent_buffer_uptodate(eb
);
647 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
648 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
652 * our io error hook is going to dec the io pages
653 * again, we have to make sure it has something
656 atomic_inc(&eb
->io_pages
);
657 clear_extent_buffer_uptodate(eb
);
659 free_extent_buffer(eb
);
664 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
666 struct extent_buffer
*eb
;
667 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
669 eb
= (struct extent_buffer
*)page
->private;
670 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
671 eb
->read_mirror
= failed_mirror
;
672 atomic_dec(&eb
->io_pages
);
673 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
674 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
675 return -EIO
; /* we fixed nothing */
678 static void end_workqueue_bio(struct bio
*bio
, int err
)
680 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
681 struct btrfs_fs_info
*fs_info
;
683 fs_info
= end_io_wq
->info
;
684 end_io_wq
->error
= err
;
685 end_io_wq
->work
.func
= end_workqueue_fn
;
686 end_io_wq
->work
.flags
= 0;
688 if (bio
->bi_rw
& REQ_WRITE
) {
689 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
690 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
692 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
693 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
695 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
696 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
699 btrfs_queue_worker(&fs_info
->endio_write_workers
,
702 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
703 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
705 else if (end_io_wq
->metadata
)
706 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
709 btrfs_queue_worker(&fs_info
->endio_workers
,
715 * For the metadata arg you want
718 * 1 - if normal metadta
719 * 2 - if writing to the free space cache area
720 * 3 - raid parity work
722 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
725 struct end_io_wq
*end_io_wq
;
726 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
730 end_io_wq
->private = bio
->bi_private
;
731 end_io_wq
->end_io
= bio
->bi_end_io
;
732 end_io_wq
->info
= info
;
733 end_io_wq
->error
= 0;
734 end_io_wq
->bio
= bio
;
735 end_io_wq
->metadata
= metadata
;
737 bio
->bi_private
= end_io_wq
;
738 bio
->bi_end_io
= end_workqueue_bio
;
742 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
744 unsigned long limit
= min_t(unsigned long,
745 info
->workers
.max_workers
,
746 info
->fs_devices
->open_devices
);
750 static void run_one_async_start(struct btrfs_work
*work
)
752 struct async_submit_bio
*async
;
755 async
= container_of(work
, struct async_submit_bio
, work
);
756 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
757 async
->mirror_num
, async
->bio_flags
,
763 static void run_one_async_done(struct btrfs_work
*work
)
765 struct btrfs_fs_info
*fs_info
;
766 struct async_submit_bio
*async
;
769 async
= container_of(work
, struct async_submit_bio
, work
);
770 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
772 limit
= btrfs_async_submit_limit(fs_info
);
773 limit
= limit
* 2 / 3;
775 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
776 waitqueue_active(&fs_info
->async_submit_wait
))
777 wake_up(&fs_info
->async_submit_wait
);
779 /* If an error occured we just want to clean up the bio and move on */
781 bio_endio(async
->bio
, async
->error
);
785 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
786 async
->mirror_num
, async
->bio_flags
,
790 static void run_one_async_free(struct btrfs_work
*work
)
792 struct async_submit_bio
*async
;
794 async
= container_of(work
, struct async_submit_bio
, work
);
798 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
799 int rw
, struct bio
*bio
, int mirror_num
,
800 unsigned long bio_flags
,
802 extent_submit_bio_hook_t
*submit_bio_start
,
803 extent_submit_bio_hook_t
*submit_bio_done
)
805 struct async_submit_bio
*async
;
807 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
811 async
->inode
= inode
;
814 async
->mirror_num
= mirror_num
;
815 async
->submit_bio_start
= submit_bio_start
;
816 async
->submit_bio_done
= submit_bio_done
;
818 async
->work
.func
= run_one_async_start
;
819 async
->work
.ordered_func
= run_one_async_done
;
820 async
->work
.ordered_free
= run_one_async_free
;
822 async
->work
.flags
= 0;
823 async
->bio_flags
= bio_flags
;
824 async
->bio_offset
= bio_offset
;
828 atomic_inc(&fs_info
->nr_async_submits
);
831 btrfs_set_work_high_prio(&async
->work
);
833 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
835 while (atomic_read(&fs_info
->async_submit_draining
) &&
836 atomic_read(&fs_info
->nr_async_submits
)) {
837 wait_event(fs_info
->async_submit_wait
,
838 (atomic_read(&fs_info
->nr_async_submits
) == 0));
844 static int btree_csum_one_bio(struct bio
*bio
)
846 struct bio_vec
*bvec
= bio
->bi_io_vec
;
848 struct btrfs_root
*root
;
851 WARN_ON(bio
->bi_vcnt
<= 0);
852 while (bio_index
< bio
->bi_vcnt
) {
853 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
854 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
863 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
864 struct bio
*bio
, int mirror_num
,
865 unsigned long bio_flags
,
869 * when we're called for a write, we're already in the async
870 * submission context. Just jump into btrfs_map_bio
872 return btree_csum_one_bio(bio
);
875 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
876 int mirror_num
, unsigned long bio_flags
,
882 * when we're called for a write, we're already in the async
883 * submission context. Just jump into btrfs_map_bio
885 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
891 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
893 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
902 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
903 int mirror_num
, unsigned long bio_flags
,
906 int async
= check_async_write(inode
, bio_flags
);
909 if (!(rw
& REQ_WRITE
)) {
911 * called for a read, do the setup so that checksum validation
912 * can happen in the async kernel threads
914 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
918 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
921 ret
= btree_csum_one_bio(bio
);
924 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
928 * kthread helpers are used to submit writes so that
929 * checksumming can happen in parallel across all CPUs
931 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
932 inode
, rw
, bio
, mirror_num
, 0,
934 __btree_submit_bio_start
,
935 __btree_submit_bio_done
);
945 #ifdef CONFIG_MIGRATION
946 static int btree_migratepage(struct address_space
*mapping
,
947 struct page
*newpage
, struct page
*page
,
948 enum migrate_mode mode
)
951 * we can't safely write a btree page from here,
952 * we haven't done the locking hook
957 * Buffers may be managed in a filesystem specific way.
958 * We must have no buffers or drop them.
960 if (page_has_private(page
) &&
961 !try_to_release_page(page
, GFP_KERNEL
))
963 return migrate_page(mapping
, newpage
, page
, mode
);
968 static int btree_writepages(struct address_space
*mapping
,
969 struct writeback_control
*wbc
)
971 struct extent_io_tree
*tree
;
972 struct btrfs_fs_info
*fs_info
;
975 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
976 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
978 if (wbc
->for_kupdate
)
981 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
982 /* this is a bit racy, but that's ok */
983 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
984 BTRFS_DIRTY_METADATA_THRESH
);
988 return btree_write_cache_pages(mapping
, wbc
);
991 static int btree_readpage(struct file
*file
, struct page
*page
)
993 struct extent_io_tree
*tree
;
994 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
995 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
998 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1000 if (PageWriteback(page
) || PageDirty(page
))
1003 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
1004 * slab allocation from alloc_extent_state down the callchain where
1005 * it'd hit a BUG_ON as those flags are not allowed.
1007 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
1009 return try_release_extent_buffer(page
, gfp_flags
);
1012 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
1014 struct extent_io_tree
*tree
;
1015 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1016 extent_invalidatepage(tree
, page
, offset
);
1017 btree_releasepage(page
, GFP_NOFS
);
1018 if (PagePrivate(page
)) {
1019 printk(KERN_WARNING
"btrfs warning page private not zero "
1020 "on page %llu\n", (unsigned long long)page_offset(page
));
1021 ClearPagePrivate(page
);
1022 set_page_private(page
, 0);
1023 page_cache_release(page
);
1027 static int btree_set_page_dirty(struct page
*page
)
1030 struct extent_buffer
*eb
;
1032 BUG_ON(!PagePrivate(page
));
1033 eb
= (struct extent_buffer
*)page
->private;
1035 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1036 BUG_ON(!atomic_read(&eb
->refs
));
1037 btrfs_assert_tree_locked(eb
);
1039 return __set_page_dirty_nobuffers(page
);
1042 static const struct address_space_operations btree_aops
= {
1043 .readpage
= btree_readpage
,
1044 .writepages
= btree_writepages
,
1045 .releasepage
= btree_releasepage
,
1046 .invalidatepage
= btree_invalidatepage
,
1047 #ifdef CONFIG_MIGRATION
1048 .migratepage
= btree_migratepage
,
1050 .set_page_dirty
= btree_set_page_dirty
,
1053 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1056 struct extent_buffer
*buf
= NULL
;
1057 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1060 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1063 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1064 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1065 free_extent_buffer(buf
);
1069 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1070 int mirror_num
, struct extent_buffer
**eb
)
1072 struct extent_buffer
*buf
= NULL
;
1073 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1074 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1077 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1081 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1083 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1084 btree_get_extent
, mirror_num
);
1086 free_extent_buffer(buf
);
1090 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1091 free_extent_buffer(buf
);
1093 } else if (extent_buffer_uptodate(buf
)) {
1096 free_extent_buffer(buf
);
1101 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1102 u64 bytenr
, u32 blocksize
)
1104 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1105 struct extent_buffer
*eb
;
1106 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1111 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1112 u64 bytenr
, u32 blocksize
)
1114 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1115 struct extent_buffer
*eb
;
1117 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1123 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1125 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1126 buf
->start
+ buf
->len
- 1);
1129 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1131 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1132 buf
->start
, buf
->start
+ buf
->len
- 1);
1135 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1136 u32 blocksize
, u64 parent_transid
)
1138 struct extent_buffer
*buf
= NULL
;
1141 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1145 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1150 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1151 struct extent_buffer
*buf
)
1153 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1155 if (btrfs_header_generation(buf
) ==
1156 fs_info
->running_transaction
->transid
) {
1157 btrfs_assert_tree_locked(buf
);
1159 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1160 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1162 fs_info
->dirty_metadata_batch
);
1163 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1164 btrfs_set_lock_blocking(buf
);
1165 clear_extent_buffer_dirty(buf
);
1170 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1171 u32 stripesize
, struct btrfs_root
*root
,
1172 struct btrfs_fs_info
*fs_info
,
1176 root
->commit_root
= NULL
;
1177 root
->sectorsize
= sectorsize
;
1178 root
->nodesize
= nodesize
;
1179 root
->leafsize
= leafsize
;
1180 root
->stripesize
= stripesize
;
1182 root
->track_dirty
= 0;
1184 root
->orphan_item_inserted
= 0;
1185 root
->orphan_cleanup_state
= 0;
1187 root
->objectid
= objectid
;
1188 root
->last_trans
= 0;
1189 root
->highest_objectid
= 0;
1191 root
->inode_tree
= RB_ROOT
;
1192 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1193 root
->block_rsv
= NULL
;
1194 root
->orphan_block_rsv
= NULL
;
1196 INIT_LIST_HEAD(&root
->dirty_list
);
1197 INIT_LIST_HEAD(&root
->root_list
);
1198 INIT_LIST_HEAD(&root
->logged_list
[0]);
1199 INIT_LIST_HEAD(&root
->logged_list
[1]);
1200 spin_lock_init(&root
->orphan_lock
);
1201 spin_lock_init(&root
->inode_lock
);
1202 spin_lock_init(&root
->accounting_lock
);
1203 spin_lock_init(&root
->log_extents_lock
[0]);
1204 spin_lock_init(&root
->log_extents_lock
[1]);
1205 mutex_init(&root
->objectid_mutex
);
1206 mutex_init(&root
->log_mutex
);
1207 init_waitqueue_head(&root
->log_writer_wait
);
1208 init_waitqueue_head(&root
->log_commit_wait
[0]);
1209 init_waitqueue_head(&root
->log_commit_wait
[1]);
1210 atomic_set(&root
->log_commit
[0], 0);
1211 atomic_set(&root
->log_commit
[1], 0);
1212 atomic_set(&root
->log_writers
, 0);
1213 atomic_set(&root
->log_batch
, 0);
1214 atomic_set(&root
->orphan_inodes
, 0);
1215 root
->log_transid
= 0;
1216 root
->last_log_commit
= 0;
1217 extent_io_tree_init(&root
->dirty_log_pages
,
1218 fs_info
->btree_inode
->i_mapping
);
1220 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1221 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1222 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1223 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1224 root
->defrag_trans_start
= fs_info
->generation
;
1225 init_completion(&root
->kobj_unregister
);
1226 root
->defrag_running
= 0;
1227 root
->root_key
.objectid
= objectid
;
1230 spin_lock_init(&root
->root_item_lock
);
1233 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1234 struct btrfs_fs_info
*fs_info
,
1236 struct btrfs_root
*root
)
1242 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1243 tree_root
->sectorsize
, tree_root
->stripesize
,
1244 root
, fs_info
, objectid
);
1245 ret
= btrfs_find_last_root(tree_root
, objectid
,
1246 &root
->root_item
, &root
->root_key
);
1252 generation
= btrfs_root_generation(&root
->root_item
);
1253 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1254 root
->commit_root
= NULL
;
1255 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1256 blocksize
, generation
);
1257 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1258 free_extent_buffer(root
->node
);
1262 root
->commit_root
= btrfs_root_node(root
);
1266 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1268 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1270 root
->fs_info
= fs_info
;
1274 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1275 struct btrfs_fs_info
*fs_info
,
1278 struct extent_buffer
*leaf
;
1279 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1280 struct btrfs_root
*root
;
1281 struct btrfs_key key
;
1286 root
= btrfs_alloc_root(fs_info
);
1288 return ERR_PTR(-ENOMEM
);
1290 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1291 tree_root
->sectorsize
, tree_root
->stripesize
,
1292 root
, fs_info
, objectid
);
1293 root
->root_key
.objectid
= objectid
;
1294 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1295 root
->root_key
.offset
= 0;
1297 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1298 0, objectid
, NULL
, 0, 0, 0);
1300 ret
= PTR_ERR(leaf
);
1305 bytenr
= leaf
->start
;
1306 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1307 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1308 btrfs_set_header_generation(leaf
, trans
->transid
);
1309 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1310 btrfs_set_header_owner(leaf
, objectid
);
1313 write_extent_buffer(leaf
, fs_info
->fsid
,
1314 (unsigned long)btrfs_header_fsid(leaf
),
1316 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1317 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1319 btrfs_mark_buffer_dirty(leaf
);
1321 root
->commit_root
= btrfs_root_node(root
);
1322 root
->track_dirty
= 1;
1325 root
->root_item
.flags
= 0;
1326 root
->root_item
.byte_limit
= 0;
1327 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1328 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1329 btrfs_set_root_level(&root
->root_item
, 0);
1330 btrfs_set_root_refs(&root
->root_item
, 1);
1331 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1332 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1333 btrfs_set_root_dirid(&root
->root_item
, 0);
1335 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1336 root
->root_item
.drop_level
= 0;
1338 key
.objectid
= objectid
;
1339 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1341 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1345 btrfs_tree_unlock(leaf
);
1351 btrfs_tree_unlock(leaf
);
1352 free_extent_buffer(leaf
);
1356 return ERR_PTR(ret
);
1359 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1360 struct btrfs_fs_info
*fs_info
)
1362 struct btrfs_root
*root
;
1363 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1364 struct extent_buffer
*leaf
;
1366 root
= btrfs_alloc_root(fs_info
);
1368 return ERR_PTR(-ENOMEM
);
1370 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1371 tree_root
->sectorsize
, tree_root
->stripesize
,
1372 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1374 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1375 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1376 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1378 * log trees do not get reference counted because they go away
1379 * before a real commit is actually done. They do store pointers
1380 * to file data extents, and those reference counts still get
1381 * updated (along with back refs to the log tree).
1385 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1386 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1390 return ERR_CAST(leaf
);
1393 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1394 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1395 btrfs_set_header_generation(leaf
, trans
->transid
);
1396 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1397 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1400 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1401 (unsigned long)btrfs_header_fsid(root
->node
),
1403 btrfs_mark_buffer_dirty(root
->node
);
1404 btrfs_tree_unlock(root
->node
);
1408 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1409 struct btrfs_fs_info
*fs_info
)
1411 struct btrfs_root
*log_root
;
1413 log_root
= alloc_log_tree(trans
, fs_info
);
1414 if (IS_ERR(log_root
))
1415 return PTR_ERR(log_root
);
1416 WARN_ON(fs_info
->log_root_tree
);
1417 fs_info
->log_root_tree
= log_root
;
1421 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1422 struct btrfs_root
*root
)
1424 struct btrfs_root
*log_root
;
1425 struct btrfs_inode_item
*inode_item
;
1427 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1428 if (IS_ERR(log_root
))
1429 return PTR_ERR(log_root
);
1431 log_root
->last_trans
= trans
->transid
;
1432 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1434 inode_item
= &log_root
->root_item
.inode
;
1435 inode_item
->generation
= cpu_to_le64(1);
1436 inode_item
->size
= cpu_to_le64(3);
1437 inode_item
->nlink
= cpu_to_le32(1);
1438 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1439 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1441 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1443 WARN_ON(root
->log_root
);
1444 root
->log_root
= log_root
;
1445 root
->log_transid
= 0;
1446 root
->last_log_commit
= 0;
1450 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1451 struct btrfs_key
*location
)
1453 struct btrfs_root
*root
;
1454 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1455 struct btrfs_path
*path
;
1456 struct extent_buffer
*l
;
1462 root
= btrfs_alloc_root(fs_info
);
1464 return ERR_PTR(-ENOMEM
);
1465 if (location
->offset
== (u64
)-1) {
1466 ret
= find_and_setup_root(tree_root
, fs_info
,
1467 location
->objectid
, root
);
1470 return ERR_PTR(ret
);
1475 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1476 tree_root
->sectorsize
, tree_root
->stripesize
,
1477 root
, fs_info
, location
->objectid
);
1479 path
= btrfs_alloc_path();
1482 return ERR_PTR(-ENOMEM
);
1484 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1487 slot
= path
->slots
[0];
1488 btrfs_read_root_item(l
, slot
, &root
->root_item
);
1489 memcpy(&root
->root_key
, location
, sizeof(*location
));
1491 btrfs_free_path(path
);
1496 return ERR_PTR(ret
);
1499 generation
= btrfs_root_generation(&root
->root_item
);
1500 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1501 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1502 blocksize
, generation
);
1503 if (!root
->node
|| !extent_buffer_uptodate(root
->node
)) {
1504 ret
= (!root
->node
) ? -ENOMEM
: -EIO
;
1506 free_extent_buffer(root
->node
);
1508 return ERR_PTR(ret
);
1511 root
->commit_root
= btrfs_root_node(root
);
1512 BUG_ON(!root
->node
); /* -ENOMEM */
1514 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1516 btrfs_check_and_init_root_item(&root
->root_item
);
1522 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1523 struct btrfs_key
*location
)
1525 struct btrfs_root
*root
;
1528 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1529 return fs_info
->tree_root
;
1530 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1531 return fs_info
->extent_root
;
1532 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1533 return fs_info
->chunk_root
;
1534 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1535 return fs_info
->dev_root
;
1536 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1537 return fs_info
->csum_root
;
1538 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1539 return fs_info
->quota_root
? fs_info
->quota_root
:
1542 spin_lock(&fs_info
->fs_roots_radix_lock
);
1543 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1544 (unsigned long)location
->objectid
);
1545 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1549 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1553 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1554 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1556 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1561 btrfs_init_free_ino_ctl(root
);
1562 mutex_init(&root
->fs_commit_mutex
);
1563 spin_lock_init(&root
->cache_lock
);
1564 init_waitqueue_head(&root
->cache_wait
);
1566 ret
= get_anon_bdev(&root
->anon_dev
);
1570 if (btrfs_root_refs(&root
->root_item
) == 0) {
1575 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1579 root
->orphan_item_inserted
= 1;
1581 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1585 spin_lock(&fs_info
->fs_roots_radix_lock
);
1586 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1587 (unsigned long)root
->root_key
.objectid
,
1592 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1593 radix_tree_preload_end();
1595 if (ret
== -EEXIST
) {
1602 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1603 root
->root_key
.objectid
);
1608 return ERR_PTR(ret
);
1611 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1613 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1615 struct btrfs_device
*device
;
1616 struct backing_dev_info
*bdi
;
1619 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1622 bdi
= blk_get_backing_dev_info(device
->bdev
);
1623 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1633 * If this fails, caller must call bdi_destroy() to get rid of the
1636 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1640 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1641 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1645 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1646 bdi
->congested_fn
= btrfs_congested_fn
;
1647 bdi
->congested_data
= info
;
1652 * called by the kthread helper functions to finally call the bio end_io
1653 * functions. This is where read checksum verification actually happens
1655 static void end_workqueue_fn(struct btrfs_work
*work
)
1658 struct end_io_wq
*end_io_wq
;
1659 struct btrfs_fs_info
*fs_info
;
1662 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1663 bio
= end_io_wq
->bio
;
1664 fs_info
= end_io_wq
->info
;
1666 error
= end_io_wq
->error
;
1667 bio
->bi_private
= end_io_wq
->private;
1668 bio
->bi_end_io
= end_io_wq
->end_io
;
1670 bio_endio(bio
, error
);
1673 static int cleaner_kthread(void *arg
)
1675 struct btrfs_root
*root
= arg
;
1680 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1681 down_read_trylock(&root
->fs_info
->sb
->s_umount
)) {
1682 if (mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1683 btrfs_run_delayed_iputs(root
);
1684 again
= btrfs_clean_one_deleted_snapshot(root
);
1685 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1687 btrfs_run_defrag_inodes(root
->fs_info
);
1688 up_read(&root
->fs_info
->sb
->s_umount
);
1691 if (!try_to_freeze() && !again
) {
1692 set_current_state(TASK_INTERRUPTIBLE
);
1693 if (!kthread_should_stop())
1695 __set_current_state(TASK_RUNNING
);
1697 } while (!kthread_should_stop());
1701 static int transaction_kthread(void *arg
)
1703 struct btrfs_root
*root
= arg
;
1704 struct btrfs_trans_handle
*trans
;
1705 struct btrfs_transaction
*cur
;
1708 unsigned long delay
;
1712 cannot_commit
= false;
1714 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1716 spin_lock(&root
->fs_info
->trans_lock
);
1717 cur
= root
->fs_info
->running_transaction
;
1719 spin_unlock(&root
->fs_info
->trans_lock
);
1723 now
= get_seconds();
1724 if (!cur
->blocked
&&
1725 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1726 spin_unlock(&root
->fs_info
->trans_lock
);
1730 transid
= cur
->transid
;
1731 spin_unlock(&root
->fs_info
->trans_lock
);
1733 /* If the file system is aborted, this will always fail. */
1734 trans
= btrfs_attach_transaction(root
);
1735 if (IS_ERR(trans
)) {
1736 if (PTR_ERR(trans
) != -ENOENT
)
1737 cannot_commit
= true;
1740 if (transid
== trans
->transid
) {
1741 btrfs_commit_transaction(trans
, root
);
1743 btrfs_end_transaction(trans
, root
);
1746 wake_up_process(root
->fs_info
->cleaner_kthread
);
1747 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1749 if (!try_to_freeze()) {
1750 set_current_state(TASK_INTERRUPTIBLE
);
1751 if (!kthread_should_stop() &&
1752 (!btrfs_transaction_blocked(root
->fs_info
) ||
1754 schedule_timeout(delay
);
1755 __set_current_state(TASK_RUNNING
);
1757 } while (!kthread_should_stop());
1762 * this will find the highest generation in the array of
1763 * root backups. The index of the highest array is returned,
1764 * or -1 if we can't find anything.
1766 * We check to make sure the array is valid by comparing the
1767 * generation of the latest root in the array with the generation
1768 * in the super block. If they don't match we pitch it.
1770 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1773 int newest_index
= -1;
1774 struct btrfs_root_backup
*root_backup
;
1777 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1778 root_backup
= info
->super_copy
->super_roots
+ i
;
1779 cur
= btrfs_backup_tree_root_gen(root_backup
);
1780 if (cur
== newest_gen
)
1784 /* check to see if we actually wrapped around */
1785 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1786 root_backup
= info
->super_copy
->super_roots
;
1787 cur
= btrfs_backup_tree_root_gen(root_backup
);
1788 if (cur
== newest_gen
)
1791 return newest_index
;
1796 * find the oldest backup so we know where to store new entries
1797 * in the backup array. This will set the backup_root_index
1798 * field in the fs_info struct
1800 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1803 int newest_index
= -1;
1805 newest_index
= find_newest_super_backup(info
, newest_gen
);
1806 /* if there was garbage in there, just move along */
1807 if (newest_index
== -1) {
1808 info
->backup_root_index
= 0;
1810 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1815 * copy all the root pointers into the super backup array.
1816 * this will bump the backup pointer by one when it is
1819 static void backup_super_roots(struct btrfs_fs_info
*info
)
1822 struct btrfs_root_backup
*root_backup
;
1825 next_backup
= info
->backup_root_index
;
1826 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1827 BTRFS_NUM_BACKUP_ROOTS
;
1830 * just overwrite the last backup if we're at the same generation
1831 * this happens only at umount
1833 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1834 if (btrfs_backup_tree_root_gen(root_backup
) ==
1835 btrfs_header_generation(info
->tree_root
->node
))
1836 next_backup
= last_backup
;
1838 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1841 * make sure all of our padding and empty slots get zero filled
1842 * regardless of which ones we use today
1844 memset(root_backup
, 0, sizeof(*root_backup
));
1846 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1848 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1849 btrfs_set_backup_tree_root_gen(root_backup
,
1850 btrfs_header_generation(info
->tree_root
->node
));
1852 btrfs_set_backup_tree_root_level(root_backup
,
1853 btrfs_header_level(info
->tree_root
->node
));
1855 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1856 btrfs_set_backup_chunk_root_gen(root_backup
,
1857 btrfs_header_generation(info
->chunk_root
->node
));
1858 btrfs_set_backup_chunk_root_level(root_backup
,
1859 btrfs_header_level(info
->chunk_root
->node
));
1861 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1862 btrfs_set_backup_extent_root_gen(root_backup
,
1863 btrfs_header_generation(info
->extent_root
->node
));
1864 btrfs_set_backup_extent_root_level(root_backup
,
1865 btrfs_header_level(info
->extent_root
->node
));
1868 * we might commit during log recovery, which happens before we set
1869 * the fs_root. Make sure it is valid before we fill it in.
1871 if (info
->fs_root
&& info
->fs_root
->node
) {
1872 btrfs_set_backup_fs_root(root_backup
,
1873 info
->fs_root
->node
->start
);
1874 btrfs_set_backup_fs_root_gen(root_backup
,
1875 btrfs_header_generation(info
->fs_root
->node
));
1876 btrfs_set_backup_fs_root_level(root_backup
,
1877 btrfs_header_level(info
->fs_root
->node
));
1880 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1881 btrfs_set_backup_dev_root_gen(root_backup
,
1882 btrfs_header_generation(info
->dev_root
->node
));
1883 btrfs_set_backup_dev_root_level(root_backup
,
1884 btrfs_header_level(info
->dev_root
->node
));
1886 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1887 btrfs_set_backup_csum_root_gen(root_backup
,
1888 btrfs_header_generation(info
->csum_root
->node
));
1889 btrfs_set_backup_csum_root_level(root_backup
,
1890 btrfs_header_level(info
->csum_root
->node
));
1892 btrfs_set_backup_total_bytes(root_backup
,
1893 btrfs_super_total_bytes(info
->super_copy
));
1894 btrfs_set_backup_bytes_used(root_backup
,
1895 btrfs_super_bytes_used(info
->super_copy
));
1896 btrfs_set_backup_num_devices(root_backup
,
1897 btrfs_super_num_devices(info
->super_copy
));
1900 * if we don't copy this out to the super_copy, it won't get remembered
1901 * for the next commit
1903 memcpy(&info
->super_copy
->super_roots
,
1904 &info
->super_for_commit
->super_roots
,
1905 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1909 * this copies info out of the root backup array and back into
1910 * the in-memory super block. It is meant to help iterate through
1911 * the array, so you send it the number of backups you've already
1912 * tried and the last backup index you used.
1914 * this returns -1 when it has tried all the backups
1916 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1917 struct btrfs_super_block
*super
,
1918 int *num_backups_tried
, int *backup_index
)
1920 struct btrfs_root_backup
*root_backup
;
1921 int newest
= *backup_index
;
1923 if (*num_backups_tried
== 0) {
1924 u64 gen
= btrfs_super_generation(super
);
1926 newest
= find_newest_super_backup(info
, gen
);
1930 *backup_index
= newest
;
1931 *num_backups_tried
= 1;
1932 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1933 /* we've tried all the backups, all done */
1936 /* jump to the next oldest backup */
1937 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1938 BTRFS_NUM_BACKUP_ROOTS
;
1939 *backup_index
= newest
;
1940 *num_backups_tried
+= 1;
1942 root_backup
= super
->super_roots
+ newest
;
1944 btrfs_set_super_generation(super
,
1945 btrfs_backup_tree_root_gen(root_backup
));
1946 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1947 btrfs_set_super_root_level(super
,
1948 btrfs_backup_tree_root_level(root_backup
));
1949 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1952 * fixme: the total bytes and num_devices need to match or we should
1955 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1956 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1960 /* helper to cleanup workers */
1961 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
1963 btrfs_stop_workers(&fs_info
->generic_worker
);
1964 btrfs_stop_workers(&fs_info
->fixup_workers
);
1965 btrfs_stop_workers(&fs_info
->delalloc_workers
);
1966 btrfs_stop_workers(&fs_info
->workers
);
1967 btrfs_stop_workers(&fs_info
->endio_workers
);
1968 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
1969 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
1970 btrfs_stop_workers(&fs_info
->rmw_workers
);
1971 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
1972 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1973 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
1974 btrfs_stop_workers(&fs_info
->submit_workers
);
1975 btrfs_stop_workers(&fs_info
->delayed_workers
);
1976 btrfs_stop_workers(&fs_info
->caching_workers
);
1977 btrfs_stop_workers(&fs_info
->readahead_workers
);
1978 btrfs_stop_workers(&fs_info
->flush_workers
);
1979 btrfs_stop_workers(&fs_info
->qgroup_rescan_workers
);
1982 /* helper to cleanup tree roots */
1983 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1985 free_extent_buffer(info
->tree_root
->node
);
1986 free_extent_buffer(info
->tree_root
->commit_root
);
1987 free_extent_buffer(info
->dev_root
->node
);
1988 free_extent_buffer(info
->dev_root
->commit_root
);
1989 free_extent_buffer(info
->extent_root
->node
);
1990 free_extent_buffer(info
->extent_root
->commit_root
);
1991 free_extent_buffer(info
->csum_root
->node
);
1992 free_extent_buffer(info
->csum_root
->commit_root
);
1993 if (info
->quota_root
) {
1994 free_extent_buffer(info
->quota_root
->node
);
1995 free_extent_buffer(info
->quota_root
->commit_root
);
1998 info
->tree_root
->node
= NULL
;
1999 info
->tree_root
->commit_root
= NULL
;
2000 info
->dev_root
->node
= NULL
;
2001 info
->dev_root
->commit_root
= NULL
;
2002 info
->extent_root
->node
= NULL
;
2003 info
->extent_root
->commit_root
= NULL
;
2004 info
->csum_root
->node
= NULL
;
2005 info
->csum_root
->commit_root
= NULL
;
2006 if (info
->quota_root
) {
2007 info
->quota_root
->node
= NULL
;
2008 info
->quota_root
->commit_root
= NULL
;
2012 free_extent_buffer(info
->chunk_root
->node
);
2013 free_extent_buffer(info
->chunk_root
->commit_root
);
2014 info
->chunk_root
->node
= NULL
;
2015 info
->chunk_root
->commit_root
= NULL
;
2019 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
2022 struct btrfs_root
*gang
[8];
2025 while (!list_empty(&fs_info
->dead_roots
)) {
2026 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2027 struct btrfs_root
, root_list
);
2028 list_del(&gang
[0]->root_list
);
2030 if (gang
[0]->in_radix
) {
2031 btrfs_free_fs_root(fs_info
, gang
[0]);
2033 free_extent_buffer(gang
[0]->node
);
2034 free_extent_buffer(gang
[0]->commit_root
);
2040 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2045 for (i
= 0; i
< ret
; i
++)
2046 btrfs_free_fs_root(fs_info
, gang
[i
]);
2050 int open_ctree(struct super_block
*sb
,
2051 struct btrfs_fs_devices
*fs_devices
,
2061 struct btrfs_key location
;
2062 struct buffer_head
*bh
;
2063 struct btrfs_super_block
*disk_super
;
2064 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2065 struct btrfs_root
*tree_root
;
2066 struct btrfs_root
*extent_root
;
2067 struct btrfs_root
*csum_root
;
2068 struct btrfs_root
*chunk_root
;
2069 struct btrfs_root
*dev_root
;
2070 struct btrfs_root
*quota_root
;
2071 struct btrfs_root
*log_tree_root
;
2074 int num_backups_tried
= 0;
2075 int backup_index
= 0;
2077 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2078 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
2079 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
2080 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2081 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
2082 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
2084 if (!tree_root
|| !extent_root
|| !csum_root
||
2085 !chunk_root
|| !dev_root
|| !quota_root
) {
2090 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2096 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2102 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2107 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2108 (1 + ilog2(nr_cpu_ids
));
2110 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2113 goto fail_dirty_metadata_bytes
;
2116 fs_info
->btree_inode
= new_inode(sb
);
2117 if (!fs_info
->btree_inode
) {
2119 goto fail_delalloc_bytes
;
2122 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2124 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2125 INIT_LIST_HEAD(&fs_info
->trans_list
);
2126 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2127 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2128 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2129 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2130 spin_lock_init(&fs_info
->delalloc_lock
);
2131 spin_lock_init(&fs_info
->trans_lock
);
2132 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2133 spin_lock_init(&fs_info
->delayed_iput_lock
);
2134 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2135 spin_lock_init(&fs_info
->free_chunk_lock
);
2136 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2137 spin_lock_init(&fs_info
->super_lock
);
2138 rwlock_init(&fs_info
->tree_mod_log_lock
);
2139 mutex_init(&fs_info
->reloc_mutex
);
2140 seqlock_init(&fs_info
->profiles_lock
);
2142 init_completion(&fs_info
->kobj_unregister
);
2143 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2144 INIT_LIST_HEAD(&fs_info
->space_info
);
2145 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2146 btrfs_mapping_init(&fs_info
->mapping_tree
);
2147 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2148 BTRFS_BLOCK_RSV_GLOBAL
);
2149 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2150 BTRFS_BLOCK_RSV_DELALLOC
);
2151 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2152 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2153 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2154 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2155 BTRFS_BLOCK_RSV_DELOPS
);
2156 atomic_set(&fs_info
->nr_async_submits
, 0);
2157 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2158 atomic_set(&fs_info
->async_submit_draining
, 0);
2159 atomic_set(&fs_info
->nr_async_bios
, 0);
2160 atomic_set(&fs_info
->defrag_running
, 0);
2161 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2163 fs_info
->max_inline
= 8192 * 1024;
2164 fs_info
->metadata_ratio
= 0;
2165 fs_info
->defrag_inodes
= RB_ROOT
;
2166 fs_info
->trans_no_join
= 0;
2167 fs_info
->free_chunk_space
= 0;
2168 fs_info
->tree_mod_log
= RB_ROOT
;
2170 /* readahead state */
2171 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2172 spin_lock_init(&fs_info
->reada_lock
);
2174 fs_info
->thread_pool_size
= min_t(unsigned long,
2175 num_online_cpus() + 2, 8);
2177 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2178 spin_lock_init(&fs_info
->ordered_extent_lock
);
2179 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2181 if (!fs_info
->delayed_root
) {
2185 btrfs_init_delayed_root(fs_info
->delayed_root
);
2187 mutex_init(&fs_info
->scrub_lock
);
2188 atomic_set(&fs_info
->scrubs_running
, 0);
2189 atomic_set(&fs_info
->scrub_pause_req
, 0);
2190 atomic_set(&fs_info
->scrubs_paused
, 0);
2191 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2192 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2193 init_rwsem(&fs_info
->scrub_super_lock
);
2194 fs_info
->scrub_workers_refcnt
= 0;
2195 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2196 fs_info
->check_integrity_print_mask
= 0;
2199 spin_lock_init(&fs_info
->balance_lock
);
2200 mutex_init(&fs_info
->balance_mutex
);
2201 atomic_set(&fs_info
->balance_running
, 0);
2202 atomic_set(&fs_info
->balance_pause_req
, 0);
2203 atomic_set(&fs_info
->balance_cancel_req
, 0);
2204 fs_info
->balance_ctl
= NULL
;
2205 init_waitqueue_head(&fs_info
->balance_wait_q
);
2207 sb
->s_blocksize
= 4096;
2208 sb
->s_blocksize_bits
= blksize_bits(4096);
2209 sb
->s_bdi
= &fs_info
->bdi
;
2211 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2212 set_nlink(fs_info
->btree_inode
, 1);
2214 * we set the i_size on the btree inode to the max possible int.
2215 * the real end of the address space is determined by all of
2216 * the devices in the system
2218 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2219 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2220 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2222 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2223 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2224 fs_info
->btree_inode
->i_mapping
);
2225 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2226 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2228 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2230 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2231 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2232 sizeof(struct btrfs_key
));
2233 set_bit(BTRFS_INODE_DUMMY
,
2234 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2235 insert_inode_hash(fs_info
->btree_inode
);
2237 spin_lock_init(&fs_info
->block_group_cache_lock
);
2238 fs_info
->block_group_cache_tree
= RB_ROOT
;
2239 fs_info
->first_logical_byte
= (u64
)-1;
2241 extent_io_tree_init(&fs_info
->freed_extents
[0],
2242 fs_info
->btree_inode
->i_mapping
);
2243 extent_io_tree_init(&fs_info
->freed_extents
[1],
2244 fs_info
->btree_inode
->i_mapping
);
2245 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2246 fs_info
->do_barriers
= 1;
2249 mutex_init(&fs_info
->ordered_operations_mutex
);
2250 mutex_init(&fs_info
->tree_log_mutex
);
2251 mutex_init(&fs_info
->chunk_mutex
);
2252 mutex_init(&fs_info
->transaction_kthread_mutex
);
2253 mutex_init(&fs_info
->cleaner_mutex
);
2254 mutex_init(&fs_info
->volume_mutex
);
2255 init_rwsem(&fs_info
->extent_commit_sem
);
2256 init_rwsem(&fs_info
->cleanup_work_sem
);
2257 init_rwsem(&fs_info
->subvol_sem
);
2258 fs_info
->dev_replace
.lock_owner
= 0;
2259 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2260 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2261 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2262 mutex_init(&fs_info
->dev_replace
.lock
);
2264 spin_lock_init(&fs_info
->qgroup_lock
);
2265 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2266 fs_info
->qgroup_tree
= RB_ROOT
;
2267 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2268 fs_info
->qgroup_seq
= 1;
2269 fs_info
->quota_enabled
= 0;
2270 fs_info
->pending_quota_state
= 0;
2271 mutex_init(&fs_info
->qgroup_rescan_lock
);
2273 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2274 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2276 init_waitqueue_head(&fs_info
->transaction_throttle
);
2277 init_waitqueue_head(&fs_info
->transaction_wait
);
2278 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2279 init_waitqueue_head(&fs_info
->async_submit_wait
);
2281 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2287 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2288 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2290 invalidate_bdev(fs_devices
->latest_bdev
);
2291 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2297 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2298 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2299 sizeof(*fs_info
->super_for_commit
));
2302 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2304 disk_super
= fs_info
->super_copy
;
2305 if (!btrfs_super_root(disk_super
))
2308 /* check FS state, whether FS is broken. */
2309 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2310 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2312 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2314 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2320 * run through our array of backup supers and setup
2321 * our ring pointer to the oldest one
2323 generation
= btrfs_super_generation(disk_super
);
2324 find_oldest_super_backup(fs_info
, generation
);
2327 * In the long term, we'll store the compression type in the super
2328 * block, and it'll be used for per file compression control.
2330 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2332 ret
= btrfs_parse_options(tree_root
, options
);
2338 features
= btrfs_super_incompat_flags(disk_super
) &
2339 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2341 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2342 "unsupported optional features (%Lx).\n",
2343 (unsigned long long)features
);
2348 if (btrfs_super_leafsize(disk_super
) !=
2349 btrfs_super_nodesize(disk_super
)) {
2350 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2351 "blocksizes don't match. node %d leaf %d\n",
2352 btrfs_super_nodesize(disk_super
),
2353 btrfs_super_leafsize(disk_super
));
2357 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2358 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2359 "blocksize (%d) was too large\n",
2360 btrfs_super_leafsize(disk_super
));
2365 features
= btrfs_super_incompat_flags(disk_super
);
2366 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2367 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2368 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2370 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2371 printk(KERN_ERR
"btrfs: has skinny extents\n");
2374 * flag our filesystem as having big metadata blocks if
2375 * they are bigger than the page size
2377 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2378 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2379 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2380 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2383 nodesize
= btrfs_super_nodesize(disk_super
);
2384 leafsize
= btrfs_super_leafsize(disk_super
);
2385 sectorsize
= btrfs_super_sectorsize(disk_super
);
2386 stripesize
= btrfs_super_stripesize(disk_super
);
2387 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2388 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2391 * mixed block groups end up with duplicate but slightly offset
2392 * extent buffers for the same range. It leads to corruptions
2394 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2395 (sectorsize
!= leafsize
)) {
2396 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2397 "are not allowed for mixed block groups on %s\n",
2403 * Needn't use the lock because there is no other task which will
2406 btrfs_set_super_incompat_flags(disk_super
, features
);
2408 features
= btrfs_super_compat_ro_flags(disk_super
) &
2409 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2410 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2411 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2412 "unsupported option features (%Lx).\n",
2413 (unsigned long long)features
);
2418 btrfs_init_workers(&fs_info
->generic_worker
,
2419 "genwork", 1, NULL
);
2421 btrfs_init_workers(&fs_info
->workers
, "worker",
2422 fs_info
->thread_pool_size
,
2423 &fs_info
->generic_worker
);
2425 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2426 fs_info
->thread_pool_size
,
2427 &fs_info
->generic_worker
);
2429 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2430 fs_info
->thread_pool_size
,
2431 &fs_info
->generic_worker
);
2433 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2434 min_t(u64
, fs_devices
->num_devices
,
2435 fs_info
->thread_pool_size
),
2436 &fs_info
->generic_worker
);
2438 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2439 2, &fs_info
->generic_worker
);
2441 /* a higher idle thresh on the submit workers makes it much more
2442 * likely that bios will be send down in a sane order to the
2445 fs_info
->submit_workers
.idle_thresh
= 64;
2447 fs_info
->workers
.idle_thresh
= 16;
2448 fs_info
->workers
.ordered
= 1;
2450 fs_info
->delalloc_workers
.idle_thresh
= 2;
2451 fs_info
->delalloc_workers
.ordered
= 1;
2453 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2454 &fs_info
->generic_worker
);
2455 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2456 fs_info
->thread_pool_size
,
2457 &fs_info
->generic_worker
);
2458 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2459 fs_info
->thread_pool_size
,
2460 &fs_info
->generic_worker
);
2461 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2462 "endio-meta-write", fs_info
->thread_pool_size
,
2463 &fs_info
->generic_worker
);
2464 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2465 "endio-raid56", fs_info
->thread_pool_size
,
2466 &fs_info
->generic_worker
);
2467 btrfs_init_workers(&fs_info
->rmw_workers
,
2468 "rmw", fs_info
->thread_pool_size
,
2469 &fs_info
->generic_worker
);
2470 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2471 fs_info
->thread_pool_size
,
2472 &fs_info
->generic_worker
);
2473 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2474 1, &fs_info
->generic_worker
);
2475 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2476 fs_info
->thread_pool_size
,
2477 &fs_info
->generic_worker
);
2478 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2479 fs_info
->thread_pool_size
,
2480 &fs_info
->generic_worker
);
2481 btrfs_init_workers(&fs_info
->qgroup_rescan_workers
, "qgroup-rescan", 1,
2482 &fs_info
->generic_worker
);
2485 * endios are largely parallel and should have a very
2488 fs_info
->endio_workers
.idle_thresh
= 4;
2489 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2490 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2491 fs_info
->rmw_workers
.idle_thresh
= 2;
2493 fs_info
->endio_write_workers
.idle_thresh
= 2;
2494 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2495 fs_info
->readahead_workers
.idle_thresh
= 2;
2498 * btrfs_start_workers can really only fail because of ENOMEM so just
2499 * return -ENOMEM if any of these fail.
2501 ret
= btrfs_start_workers(&fs_info
->workers
);
2502 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2503 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2504 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2505 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2506 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2507 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2508 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2509 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2510 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2511 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2512 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2513 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2514 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2515 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2516 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2517 ret
|= btrfs_start_workers(&fs_info
->qgroup_rescan_workers
);
2520 goto fail_sb_buffer
;
2523 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2524 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2525 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2527 tree_root
->nodesize
= nodesize
;
2528 tree_root
->leafsize
= leafsize
;
2529 tree_root
->sectorsize
= sectorsize
;
2530 tree_root
->stripesize
= stripesize
;
2532 sb
->s_blocksize
= sectorsize
;
2533 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2535 if (disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2536 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2537 goto fail_sb_buffer
;
2540 if (sectorsize
!= PAGE_SIZE
) {
2541 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2542 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2543 goto fail_sb_buffer
;
2546 mutex_lock(&fs_info
->chunk_mutex
);
2547 ret
= btrfs_read_sys_array(tree_root
);
2548 mutex_unlock(&fs_info
->chunk_mutex
);
2550 printk(KERN_WARNING
"btrfs: failed to read the system "
2551 "array on %s\n", sb
->s_id
);
2552 goto fail_sb_buffer
;
2555 blocksize
= btrfs_level_size(tree_root
,
2556 btrfs_super_chunk_root_level(disk_super
));
2557 generation
= btrfs_super_chunk_root_generation(disk_super
);
2559 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2560 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2562 chunk_root
->node
= read_tree_block(chunk_root
,
2563 btrfs_super_chunk_root(disk_super
),
2564 blocksize
, generation
);
2565 if (!chunk_root
->node
||
2566 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2567 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2569 goto fail_tree_roots
;
2571 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2572 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2574 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2575 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2578 ret
= btrfs_read_chunk_tree(chunk_root
);
2580 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2582 goto fail_tree_roots
;
2586 * keep the device that is marked to be the target device for the
2587 * dev_replace procedure
2589 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2591 if (!fs_devices
->latest_bdev
) {
2592 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2594 goto fail_tree_roots
;
2598 blocksize
= btrfs_level_size(tree_root
,
2599 btrfs_super_root_level(disk_super
));
2600 generation
= btrfs_super_generation(disk_super
);
2602 tree_root
->node
= read_tree_block(tree_root
,
2603 btrfs_super_root(disk_super
),
2604 blocksize
, generation
);
2605 if (!tree_root
->node
||
2606 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2607 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2610 goto recovery_tree_root
;
2613 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2614 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2616 ret
= find_and_setup_root(tree_root
, fs_info
,
2617 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2619 goto recovery_tree_root
;
2620 extent_root
->track_dirty
= 1;
2622 ret
= find_and_setup_root(tree_root
, fs_info
,
2623 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2625 goto recovery_tree_root
;
2626 dev_root
->track_dirty
= 1;
2628 ret
= find_and_setup_root(tree_root
, fs_info
,
2629 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2631 goto recovery_tree_root
;
2632 csum_root
->track_dirty
= 1;
2634 ret
= find_and_setup_root(tree_root
, fs_info
,
2635 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2638 quota_root
= fs_info
->quota_root
= NULL
;
2640 quota_root
->track_dirty
= 1;
2641 fs_info
->quota_enabled
= 1;
2642 fs_info
->pending_quota_state
= 1;
2645 fs_info
->generation
= generation
;
2646 fs_info
->last_trans_committed
= generation
;
2648 ret
= btrfs_recover_balance(fs_info
);
2650 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2651 goto fail_block_groups
;
2654 ret
= btrfs_init_dev_stats(fs_info
);
2656 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2658 goto fail_block_groups
;
2661 ret
= btrfs_init_dev_replace(fs_info
);
2663 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2664 goto fail_block_groups
;
2667 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2669 ret
= btrfs_init_space_info(fs_info
);
2671 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2672 goto fail_block_groups
;
2675 ret
= btrfs_read_block_groups(extent_root
);
2677 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2678 goto fail_block_groups
;
2680 fs_info
->num_tolerated_disk_barrier_failures
=
2681 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2682 if (fs_info
->fs_devices
->missing_devices
>
2683 fs_info
->num_tolerated_disk_barrier_failures
&&
2684 !(sb
->s_flags
& MS_RDONLY
)) {
2686 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2687 goto fail_block_groups
;
2690 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2692 if (IS_ERR(fs_info
->cleaner_kthread
))
2693 goto fail_block_groups
;
2695 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2697 "btrfs-transaction");
2698 if (IS_ERR(fs_info
->transaction_kthread
))
2701 if (!btrfs_test_opt(tree_root
, SSD
) &&
2702 !btrfs_test_opt(tree_root
, NOSSD
) &&
2703 !fs_info
->fs_devices
->rotating
) {
2704 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2706 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2709 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2710 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2711 ret
= btrfsic_mount(tree_root
, fs_devices
,
2712 btrfs_test_opt(tree_root
,
2713 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2715 fs_info
->check_integrity_print_mask
);
2717 printk(KERN_WARNING
"btrfs: failed to initialize"
2718 " integrity check module %s\n", sb
->s_id
);
2721 ret
= btrfs_read_qgroup_config(fs_info
);
2723 goto fail_trans_kthread
;
2725 /* do not make disk changes in broken FS */
2726 if (btrfs_super_log_root(disk_super
) != 0) {
2727 u64 bytenr
= btrfs_super_log_root(disk_super
);
2729 if (fs_devices
->rw_devices
== 0) {
2730 printk(KERN_WARNING
"Btrfs log replay required "
2736 btrfs_level_size(tree_root
,
2737 btrfs_super_log_root_level(disk_super
));
2739 log_tree_root
= btrfs_alloc_root(fs_info
);
2740 if (!log_tree_root
) {
2745 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2746 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2748 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2751 if (!log_tree_root
->node
||
2752 !extent_buffer_uptodate(log_tree_root
->node
)) {
2753 printk(KERN_ERR
"btrfs: failed to read log tree\n");
2754 free_extent_buffer(log_tree_root
->node
);
2755 kfree(log_tree_root
);
2756 goto fail_trans_kthread
;
2758 /* returns with log_tree_root freed on success */
2759 ret
= btrfs_recover_log_trees(log_tree_root
);
2761 btrfs_error(tree_root
->fs_info
, ret
,
2762 "Failed to recover log tree");
2763 free_extent_buffer(log_tree_root
->node
);
2764 kfree(log_tree_root
);
2765 goto fail_trans_kthread
;
2768 if (sb
->s_flags
& MS_RDONLY
) {
2769 ret
= btrfs_commit_super(tree_root
);
2771 goto fail_trans_kthread
;
2775 ret
= btrfs_find_orphan_roots(tree_root
);
2777 goto fail_trans_kthread
;
2779 if (!(sb
->s_flags
& MS_RDONLY
)) {
2780 ret
= btrfs_cleanup_fs_roots(fs_info
);
2782 goto fail_trans_kthread
;
2784 ret
= btrfs_recover_relocation(tree_root
);
2787 "btrfs: failed to recover relocation\n");
2793 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2794 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2795 location
.offset
= (u64
)-1;
2797 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2798 if (!fs_info
->fs_root
)
2800 if (IS_ERR(fs_info
->fs_root
)) {
2801 err
= PTR_ERR(fs_info
->fs_root
);
2805 if (sb
->s_flags
& MS_RDONLY
)
2808 down_read(&fs_info
->cleanup_work_sem
);
2809 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2810 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2811 up_read(&fs_info
->cleanup_work_sem
);
2812 close_ctree(tree_root
);
2815 up_read(&fs_info
->cleanup_work_sem
);
2817 ret
= btrfs_resume_balance_async(fs_info
);
2819 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2820 close_ctree(tree_root
);
2824 ret
= btrfs_resume_dev_replace_async(fs_info
);
2826 pr_warn("btrfs: failed to resume dev_replace\n");
2827 close_ctree(tree_root
);
2834 btrfs_free_qgroup_config(fs_info
);
2836 kthread_stop(fs_info
->transaction_kthread
);
2837 del_fs_roots(fs_info
);
2838 btrfs_cleanup_transaction(fs_info
->tree_root
);
2840 kthread_stop(fs_info
->cleaner_kthread
);
2843 * make sure we're done with the btree inode before we stop our
2846 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2849 btrfs_put_block_group_cache(fs_info
);
2850 btrfs_free_block_groups(fs_info
);
2853 free_root_pointers(fs_info
, 1);
2854 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2857 btrfs_stop_all_workers(fs_info
);
2860 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2862 iput(fs_info
->btree_inode
);
2863 fail_delalloc_bytes
:
2864 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2865 fail_dirty_metadata_bytes
:
2866 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2868 bdi_destroy(&fs_info
->bdi
);
2870 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2872 btrfs_free_stripe_hash_table(fs_info
);
2873 btrfs_close_devices(fs_info
->fs_devices
);
2877 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2878 goto fail_tree_roots
;
2880 free_root_pointers(fs_info
, 0);
2882 /* don't use the log in recovery mode, it won't be valid */
2883 btrfs_set_super_log_root(disk_super
, 0);
2885 /* we can't trust the free space cache either */
2886 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2888 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2889 &num_backups_tried
, &backup_index
);
2891 goto fail_block_groups
;
2892 goto retry_root_backup
;
2895 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2898 set_buffer_uptodate(bh
);
2900 struct btrfs_device
*device
= (struct btrfs_device
*)
2903 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2904 "I/O error on %s\n",
2905 rcu_str_deref(device
->name
));
2906 /* note, we dont' set_buffer_write_io_error because we have
2907 * our own ways of dealing with the IO errors
2909 clear_buffer_uptodate(bh
);
2910 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2916 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2918 struct buffer_head
*bh
;
2919 struct buffer_head
*latest
= NULL
;
2920 struct btrfs_super_block
*super
;
2925 /* we would like to check all the supers, but that would make
2926 * a btrfs mount succeed after a mkfs from a different FS.
2927 * So, we need to add a special mount option to scan for
2928 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2930 for (i
= 0; i
< 1; i
++) {
2931 bytenr
= btrfs_sb_offset(i
);
2932 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2934 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2938 super
= (struct btrfs_super_block
*)bh
->b_data
;
2939 if (btrfs_super_bytenr(super
) != bytenr
||
2940 super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2945 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2948 transid
= btrfs_super_generation(super
);
2957 * this should be called twice, once with wait == 0 and
2958 * once with wait == 1. When wait == 0 is done, all the buffer heads
2959 * we write are pinned.
2961 * They are released when wait == 1 is done.
2962 * max_mirrors must be the same for both runs, and it indicates how
2963 * many supers on this one device should be written.
2965 * max_mirrors == 0 means to write them all.
2967 static int write_dev_supers(struct btrfs_device
*device
,
2968 struct btrfs_super_block
*sb
,
2969 int do_barriers
, int wait
, int max_mirrors
)
2971 struct buffer_head
*bh
;
2978 if (max_mirrors
== 0)
2979 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2981 for (i
= 0; i
< max_mirrors
; i
++) {
2982 bytenr
= btrfs_sb_offset(i
);
2983 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2987 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2988 BTRFS_SUPER_INFO_SIZE
);
2991 if (!buffer_uptodate(bh
))
2994 /* drop our reference */
2997 /* drop the reference from the wait == 0 run */
3001 btrfs_set_super_bytenr(sb
, bytenr
);
3004 crc
= btrfs_csum_data((char *)sb
+
3005 BTRFS_CSUM_SIZE
, crc
,
3006 BTRFS_SUPER_INFO_SIZE
-
3008 btrfs_csum_final(crc
, sb
->csum
);
3011 * one reference for us, and we leave it for the
3014 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3015 BTRFS_SUPER_INFO_SIZE
);
3016 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3018 /* one reference for submit_bh */
3021 set_buffer_uptodate(bh
);
3023 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3024 bh
->b_private
= device
;
3028 * we fua the first super. The others we allow
3031 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3035 return errors
< i
? 0 : -1;
3039 * endio for the write_dev_flush, this will wake anyone waiting
3040 * for the barrier when it is done
3042 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3045 if (err
== -EOPNOTSUPP
)
3046 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3047 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3049 if (bio
->bi_private
)
3050 complete(bio
->bi_private
);
3055 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3056 * sent down. With wait == 1, it waits for the previous flush.
3058 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3061 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3066 if (device
->nobarriers
)
3070 bio
= device
->flush_bio
;
3074 wait_for_completion(&device
->flush_wait
);
3076 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3077 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3078 rcu_str_deref(device
->name
));
3079 device
->nobarriers
= 1;
3080 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3082 btrfs_dev_stat_inc_and_print(device
,
3083 BTRFS_DEV_STAT_FLUSH_ERRS
);
3086 /* drop the reference from the wait == 0 run */
3088 device
->flush_bio
= NULL
;
3094 * one reference for us, and we leave it for the
3097 device
->flush_bio
= NULL
;
3098 bio
= bio_alloc(GFP_NOFS
, 0);
3102 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3103 bio
->bi_bdev
= device
->bdev
;
3104 init_completion(&device
->flush_wait
);
3105 bio
->bi_private
= &device
->flush_wait
;
3106 device
->flush_bio
= bio
;
3109 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3115 * send an empty flush down to each device in parallel,
3116 * then wait for them
3118 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3120 struct list_head
*head
;
3121 struct btrfs_device
*dev
;
3122 int errors_send
= 0;
3123 int errors_wait
= 0;
3126 /* send down all the barriers */
3127 head
= &info
->fs_devices
->devices
;
3128 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3133 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3136 ret
= write_dev_flush(dev
, 0);
3141 /* wait for all the barriers */
3142 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3147 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3150 ret
= write_dev_flush(dev
, 1);
3154 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3155 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3160 int btrfs_calc_num_tolerated_disk_barrier_failures(
3161 struct btrfs_fs_info
*fs_info
)
3163 struct btrfs_ioctl_space_info space
;
3164 struct btrfs_space_info
*sinfo
;
3165 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3166 BTRFS_BLOCK_GROUP_SYSTEM
,
3167 BTRFS_BLOCK_GROUP_METADATA
,
3168 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3172 int num_tolerated_disk_barrier_failures
=
3173 (int)fs_info
->fs_devices
->num_devices
;
3175 for (i
= 0; i
< num_types
; i
++) {
3176 struct btrfs_space_info
*tmp
;
3180 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3181 if (tmp
->flags
== types
[i
]) {
3191 down_read(&sinfo
->groups_sem
);
3192 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3193 if (!list_empty(&sinfo
->block_groups
[c
])) {
3196 btrfs_get_block_group_info(
3197 &sinfo
->block_groups
[c
], &space
);
3198 if (space
.total_bytes
== 0 ||
3199 space
.used_bytes
== 0)
3201 flags
= space
.flags
;
3204 * 0: if dup, single or RAID0 is configured for
3205 * any of metadata, system or data, else
3206 * 1: if RAID5 is configured, or if RAID1 or
3207 * RAID10 is configured and only two mirrors
3209 * 2: if RAID6 is configured, else
3210 * num_mirrors - 1: if RAID1 or RAID10 is
3211 * configured and more than
3212 * 2 mirrors are used.
3214 if (num_tolerated_disk_barrier_failures
> 0 &&
3215 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3216 BTRFS_BLOCK_GROUP_RAID0
)) ||
3217 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3219 num_tolerated_disk_barrier_failures
= 0;
3220 else if (num_tolerated_disk_barrier_failures
> 1) {
3221 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3222 BTRFS_BLOCK_GROUP_RAID5
|
3223 BTRFS_BLOCK_GROUP_RAID10
)) {
3224 num_tolerated_disk_barrier_failures
= 1;
3226 BTRFS_BLOCK_GROUP_RAID5
) {
3227 num_tolerated_disk_barrier_failures
= 2;
3232 up_read(&sinfo
->groups_sem
);
3235 return num_tolerated_disk_barrier_failures
;
3238 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3240 struct list_head
*head
;
3241 struct btrfs_device
*dev
;
3242 struct btrfs_super_block
*sb
;
3243 struct btrfs_dev_item
*dev_item
;
3247 int total_errors
= 0;
3250 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3251 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3252 backup_super_roots(root
->fs_info
);
3254 sb
= root
->fs_info
->super_for_commit
;
3255 dev_item
= &sb
->dev_item
;
3257 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3258 head
= &root
->fs_info
->fs_devices
->devices
;
3261 ret
= barrier_all_devices(root
->fs_info
);
3264 &root
->fs_info
->fs_devices
->device_list_mutex
);
3265 btrfs_error(root
->fs_info
, ret
,
3266 "errors while submitting device barriers.");
3271 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3276 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3279 btrfs_set_stack_device_generation(dev_item
, 0);
3280 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3281 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3282 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3283 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3284 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3285 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3286 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3287 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3288 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3290 flags
= btrfs_super_flags(sb
);
3291 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3293 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3297 if (total_errors
> max_errors
) {
3298 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3301 /* This shouldn't happen. FUA is masked off if unsupported */
3306 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3309 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3312 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3316 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3317 if (total_errors
> max_errors
) {
3318 btrfs_error(root
->fs_info
, -EIO
,
3319 "%d errors while writing supers", total_errors
);
3325 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3326 struct btrfs_root
*root
, int max_mirrors
)
3330 ret
= write_all_supers(root
, max_mirrors
);
3334 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3336 spin_lock(&fs_info
->fs_roots_radix_lock
);
3337 radix_tree_delete(&fs_info
->fs_roots_radix
,
3338 (unsigned long)root
->root_key
.objectid
);
3339 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3341 if (btrfs_root_refs(&root
->root_item
) == 0)
3342 synchronize_srcu(&fs_info
->subvol_srcu
);
3344 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3345 btrfs_free_log(NULL
, root
);
3346 btrfs_free_log_root_tree(NULL
, fs_info
);
3349 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3350 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3354 static void free_fs_root(struct btrfs_root
*root
)
3356 iput(root
->cache_inode
);
3357 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3359 free_anon_bdev(root
->anon_dev
);
3360 free_extent_buffer(root
->node
);
3361 free_extent_buffer(root
->commit_root
);
3362 kfree(root
->free_ino_ctl
);
3363 kfree(root
->free_ino_pinned
);
3368 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3370 u64 root_objectid
= 0;
3371 struct btrfs_root
*gang
[8];
3376 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3377 (void **)gang
, root_objectid
,
3382 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3383 for (i
= 0; i
< ret
; i
++) {
3386 root_objectid
= gang
[i
]->root_key
.objectid
;
3387 err
= btrfs_orphan_cleanup(gang
[i
]);
3396 int btrfs_commit_super(struct btrfs_root
*root
)
3398 struct btrfs_trans_handle
*trans
;
3401 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3402 btrfs_run_delayed_iputs(root
);
3403 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3404 wake_up_process(root
->fs_info
->cleaner_kthread
);
3406 /* wait until ongoing cleanup work done */
3407 down_write(&root
->fs_info
->cleanup_work_sem
);
3408 up_write(&root
->fs_info
->cleanup_work_sem
);
3410 trans
= btrfs_join_transaction(root
);
3412 return PTR_ERR(trans
);
3413 ret
= btrfs_commit_transaction(trans
, root
);
3416 /* run commit again to drop the original snapshot */
3417 trans
= btrfs_join_transaction(root
);
3419 return PTR_ERR(trans
);
3420 ret
= btrfs_commit_transaction(trans
, root
);
3423 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3425 btrfs_error(root
->fs_info
, ret
,
3426 "Failed to sync btree inode to disk.");
3430 ret
= write_ctree_super(NULL
, root
, 0);
3434 int close_ctree(struct btrfs_root
*root
)
3436 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3439 fs_info
->closing
= 1;
3442 /* pause restriper - we want to resume on mount */
3443 btrfs_pause_balance(fs_info
);
3445 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3447 btrfs_scrub_cancel(fs_info
);
3449 /* wait for any defraggers to finish */
3450 wait_event(fs_info
->transaction_wait
,
3451 (atomic_read(&fs_info
->defrag_running
) == 0));
3453 /* clear out the rbtree of defraggable inodes */
3454 btrfs_cleanup_defrag_inodes(fs_info
);
3456 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3457 ret
= btrfs_commit_super(root
);
3459 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3462 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3463 btrfs_error_commit_super(root
);
3465 btrfs_put_block_group_cache(fs_info
);
3467 kthread_stop(fs_info
->transaction_kthread
);
3468 kthread_stop(fs_info
->cleaner_kthread
);
3470 fs_info
->closing
= 2;
3473 btrfs_free_qgroup_config(root
->fs_info
);
3475 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3476 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3477 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3480 free_root_pointers(fs_info
, 1);
3482 btrfs_free_block_groups(fs_info
);
3484 del_fs_roots(fs_info
);
3486 iput(fs_info
->btree_inode
);
3488 btrfs_stop_all_workers(fs_info
);
3490 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3491 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3492 btrfsic_unmount(root
, fs_info
->fs_devices
);
3495 btrfs_close_devices(fs_info
->fs_devices
);
3496 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3498 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3499 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3500 bdi_destroy(&fs_info
->bdi
);
3501 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3503 btrfs_free_stripe_hash_table(fs_info
);
3508 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3512 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3514 ret
= extent_buffer_uptodate(buf
);
3518 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3519 parent_transid
, atomic
);
3525 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3527 return set_extent_buffer_uptodate(buf
);
3530 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3532 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3533 u64 transid
= btrfs_header_generation(buf
);
3536 btrfs_assert_tree_locked(buf
);
3537 if (transid
!= root
->fs_info
->generation
)
3538 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3539 "found %llu running %llu\n",
3540 (unsigned long long)buf
->start
,
3541 (unsigned long long)transid
,
3542 (unsigned long long)root
->fs_info
->generation
);
3543 was_dirty
= set_extent_buffer_dirty(buf
);
3545 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3547 root
->fs_info
->dirty_metadata_batch
);
3550 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3554 * looks as though older kernels can get into trouble with
3555 * this code, they end up stuck in balance_dirty_pages forever
3559 if (current
->flags
& PF_MEMALLOC
)
3563 btrfs_balance_delayed_items(root
);
3565 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3566 BTRFS_DIRTY_METADATA_THRESH
);
3568 balance_dirty_pages_ratelimited(
3569 root
->fs_info
->btree_inode
->i_mapping
);
3574 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3576 __btrfs_btree_balance_dirty(root
, 1);
3579 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3581 __btrfs_btree_balance_dirty(root
, 0);
3584 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3586 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3587 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3590 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3593 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3594 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3604 void btrfs_error_commit_super(struct btrfs_root
*root
)
3606 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3607 btrfs_run_delayed_iputs(root
);
3608 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3610 down_write(&root
->fs_info
->cleanup_work_sem
);
3611 up_write(&root
->fs_info
->cleanup_work_sem
);
3613 /* cleanup FS via transaction */
3614 btrfs_cleanup_transaction(root
);
3617 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
3618 struct btrfs_root
*root
)
3620 struct btrfs_inode
*btrfs_inode
;
3621 struct list_head splice
;
3623 INIT_LIST_HEAD(&splice
);
3625 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3626 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3628 list_splice_init(&t
->ordered_operations
, &splice
);
3629 while (!list_empty(&splice
)) {
3630 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3631 ordered_operations
);
3633 list_del_init(&btrfs_inode
->ordered_operations
);
3635 btrfs_invalidate_inodes(btrfs_inode
->root
);
3638 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3639 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3642 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3644 struct btrfs_ordered_extent
*ordered
;
3646 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3648 * This will just short circuit the ordered completion stuff which will
3649 * make sure the ordered extent gets properly cleaned up.
3651 list_for_each_entry(ordered
, &root
->fs_info
->ordered_extents
,
3653 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3654 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3657 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3658 struct btrfs_root
*root
)
3660 struct rb_node
*node
;
3661 struct btrfs_delayed_ref_root
*delayed_refs
;
3662 struct btrfs_delayed_ref_node
*ref
;
3665 delayed_refs
= &trans
->delayed_refs
;
3667 spin_lock(&delayed_refs
->lock
);
3668 if (delayed_refs
->num_entries
== 0) {
3669 spin_unlock(&delayed_refs
->lock
);
3670 printk(KERN_INFO
"delayed_refs has NO entry\n");
3674 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3675 struct btrfs_delayed_ref_head
*head
= NULL
;
3677 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3678 atomic_set(&ref
->refs
, 1);
3679 if (btrfs_delayed_ref_is_head(ref
)) {
3681 head
= btrfs_delayed_node_to_head(ref
);
3682 if (!mutex_trylock(&head
->mutex
)) {
3683 atomic_inc(&ref
->refs
);
3684 spin_unlock(&delayed_refs
->lock
);
3686 /* Need to wait for the delayed ref to run */
3687 mutex_lock(&head
->mutex
);
3688 mutex_unlock(&head
->mutex
);
3689 btrfs_put_delayed_ref(ref
);
3691 spin_lock(&delayed_refs
->lock
);
3695 if (head
->must_insert_reserved
)
3696 btrfs_pin_extent(root
, ref
->bytenr
,
3698 btrfs_free_delayed_extent_op(head
->extent_op
);
3699 delayed_refs
->num_heads
--;
3700 if (list_empty(&head
->cluster
))
3701 delayed_refs
->num_heads_ready
--;
3702 list_del_init(&head
->cluster
);
3706 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3707 delayed_refs
->num_entries
--;
3709 mutex_unlock(&head
->mutex
);
3710 spin_unlock(&delayed_refs
->lock
);
3711 btrfs_put_delayed_ref(ref
);
3714 spin_lock(&delayed_refs
->lock
);
3717 spin_unlock(&delayed_refs
->lock
);
3722 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
)
3724 struct btrfs_pending_snapshot
*snapshot
;
3725 struct list_head splice
;
3727 INIT_LIST_HEAD(&splice
);
3729 list_splice_init(&t
->pending_snapshots
, &splice
);
3731 while (!list_empty(&splice
)) {
3732 snapshot
= list_entry(splice
.next
,
3733 struct btrfs_pending_snapshot
,
3735 snapshot
->error
= -ECANCELED
;
3736 list_del_init(&snapshot
->list
);
3740 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3742 struct btrfs_inode
*btrfs_inode
;
3743 struct list_head splice
;
3745 INIT_LIST_HEAD(&splice
);
3747 spin_lock(&root
->fs_info
->delalloc_lock
);
3748 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3750 while (!list_empty(&splice
)) {
3751 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3754 list_del_init(&btrfs_inode
->delalloc_inodes
);
3755 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3756 &btrfs_inode
->runtime_flags
);
3758 btrfs_invalidate_inodes(btrfs_inode
->root
);
3761 spin_unlock(&root
->fs_info
->delalloc_lock
);
3764 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3765 struct extent_io_tree
*dirty_pages
,
3769 struct extent_buffer
*eb
;
3774 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3779 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3780 while (start
<= end
) {
3781 eb
= btrfs_find_tree_block(root
, start
,
3786 wait_on_extent_buffer_writeback(eb
);
3788 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3790 clear_extent_buffer_dirty(eb
);
3791 free_extent_buffer_stale(eb
);
3798 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3799 struct extent_io_tree
*pinned_extents
)
3801 struct extent_io_tree
*unpin
;
3807 unpin
= pinned_extents
;
3810 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3811 EXTENT_DIRTY
, NULL
);
3816 if (btrfs_test_opt(root
, DISCARD
))
3817 ret
= btrfs_error_discard_extent(root
, start
,
3821 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3822 btrfs_error_unpin_extent_range(root
, start
, end
);
3827 if (unpin
== &root
->fs_info
->freed_extents
[0])
3828 unpin
= &root
->fs_info
->freed_extents
[1];
3830 unpin
= &root
->fs_info
->freed_extents
[0];
3838 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3839 struct btrfs_root
*root
)
3841 btrfs_destroy_delayed_refs(cur_trans
, root
);
3842 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3843 cur_trans
->dirty_pages
.dirty_bytes
);
3845 /* FIXME: cleanup wait for commit */
3846 cur_trans
->in_commit
= 1;
3847 cur_trans
->blocked
= 1;
3848 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3850 btrfs_evict_pending_snapshots(cur_trans
);
3852 cur_trans
->blocked
= 0;
3853 wake_up(&root
->fs_info
->transaction_wait
);
3855 cur_trans
->commit_done
= 1;
3856 wake_up(&cur_trans
->commit_wait
);
3858 btrfs_destroy_delayed_inodes(root
);
3859 btrfs_assert_delayed_root_empty(root
);
3861 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3863 btrfs_destroy_pinned_extent(root
,
3864 root
->fs_info
->pinned_extents
);
3867 memset(cur_trans, 0, sizeof(*cur_trans));
3868 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3872 int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3874 struct btrfs_transaction
*t
;
3877 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3879 spin_lock(&root
->fs_info
->trans_lock
);
3880 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3881 root
->fs_info
->trans_no_join
= 1;
3882 spin_unlock(&root
->fs_info
->trans_lock
);
3884 while (!list_empty(&list
)) {
3885 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3887 btrfs_destroy_ordered_operations(t
, root
);
3889 btrfs_destroy_ordered_extents(root
);
3891 btrfs_destroy_delayed_refs(t
, root
);
3893 /* FIXME: cleanup wait for commit */
3897 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3898 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3900 btrfs_evict_pending_snapshots(t
);
3904 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3905 wake_up(&root
->fs_info
->transaction_wait
);
3909 if (waitqueue_active(&t
->commit_wait
))
3910 wake_up(&t
->commit_wait
);
3912 btrfs_destroy_delayed_inodes(root
);
3913 btrfs_assert_delayed_root_empty(root
);
3915 btrfs_destroy_delalloc_inodes(root
);
3917 spin_lock(&root
->fs_info
->trans_lock
);
3918 root
->fs_info
->running_transaction
= NULL
;
3919 spin_unlock(&root
->fs_info
->trans_lock
);
3921 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3924 btrfs_destroy_pinned_extent(root
,
3925 root
->fs_info
->pinned_extents
);
3927 atomic_set(&t
->use_count
, 0);
3928 list_del_init(&t
->list
);
3929 memset(t
, 0, sizeof(*t
));
3930 kmem_cache_free(btrfs_transaction_cachep
, t
);
3933 spin_lock(&root
->fs_info
->trans_lock
);
3934 root
->fs_info
->trans_no_join
= 0;
3935 spin_unlock(&root
->fs_info
->trans_lock
);
3936 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3941 static struct extent_io_ops btree_extent_io_ops
= {
3942 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3943 .readpage_io_failed_hook
= btree_io_failed_hook
,
3944 .submit_bio_hook
= btree_submit_bio_hook
,
3945 /* note we're sharing with inode.c for the merge bio hook */
3946 .merge_bio_hook
= btrfs_merge_bio_hook
,