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/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "free-space-tree.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
55 #include <asm/cpufeature.h>
58 static const struct extent_io_ops btree_extent_io_ops
;
59 static void end_workqueue_fn(struct btrfs_work
*work
);
60 static void free_fs_root(struct btrfs_root
*root
);
61 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
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_destroy_delalloc_inodes(struct btrfs_root
*root
);
67 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
68 struct extent_io_tree
*dirty_pages
,
70 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
71 struct extent_io_tree
*pinned_extents
);
72 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
73 static void btrfs_error_commit_super(struct btrfs_root
*root
);
76 * btrfs_end_io_wq structs are used to do processing in task context when an IO
77 * is complete. This is used during reads to verify checksums, and it is used
78 * by writes to insert metadata for new file extents after IO is complete.
80 struct btrfs_end_io_wq
{
84 struct btrfs_fs_info
*info
;
86 enum btrfs_wq_endio_type metadata
;
87 struct list_head list
;
88 struct btrfs_work work
;
91 static struct kmem_cache
*btrfs_end_io_wq_cache
;
93 int __init
btrfs_end_io_wq_init(void)
95 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
96 sizeof(struct btrfs_end_io_wq
),
98 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
100 if (!btrfs_end_io_wq_cache
)
105 void btrfs_end_io_wq_exit(void)
107 if (btrfs_end_io_wq_cache
)
108 kmem_cache_destroy(btrfs_end_io_wq_cache
);
112 * async submit bios are used to offload expensive checksumming
113 * onto the worker threads. They checksum file and metadata bios
114 * just before they are sent down the IO stack.
116 struct async_submit_bio
{
119 struct list_head list
;
120 extent_submit_bio_hook_t
*submit_bio_start
;
121 extent_submit_bio_hook_t
*submit_bio_done
;
124 unsigned long bio_flags
;
126 * bio_offset is optional, can be used if the pages in the bio
127 * can't tell us where in the file the bio should go
130 struct btrfs_work work
;
135 * Lockdep class keys for extent_buffer->lock's in this root. For a given
136 * eb, the lockdep key is determined by the btrfs_root it belongs to and
137 * the level the eb occupies in the tree.
139 * Different roots are used for different purposes and may nest inside each
140 * other and they require separate keysets. As lockdep keys should be
141 * static, assign keysets according to the purpose of the root as indicated
142 * by btrfs_root->objectid. This ensures that all special purpose roots
143 * have separate keysets.
145 * Lock-nesting across peer nodes is always done with the immediate parent
146 * node locked thus preventing deadlock. As lockdep doesn't know this, use
147 * subclass to avoid triggering lockdep warning in such cases.
149 * The key is set by the readpage_end_io_hook after the buffer has passed
150 * csum validation but before the pages are unlocked. It is also set by
151 * btrfs_init_new_buffer on freshly allocated blocks.
153 * We also add a check to make sure the highest level of the tree is the
154 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
155 * needs update as well.
157 #ifdef CONFIG_DEBUG_LOCK_ALLOC
158 # if BTRFS_MAX_LEVEL != 8
162 static struct btrfs_lockdep_keyset
{
163 u64 id
; /* root objectid */
164 const char *name_stem
; /* lock name stem */
165 char names
[BTRFS_MAX_LEVEL
+ 1][20];
166 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
167 } btrfs_lockdep_keysets
[] = {
168 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
169 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
170 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
171 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
172 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
173 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
174 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
175 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
176 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
177 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
178 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
179 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
180 { .id
= 0, .name_stem
= "tree" },
183 void __init
btrfs_init_lockdep(void)
187 /* initialize lockdep class names */
188 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
189 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
191 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
192 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
193 "btrfs-%s-%02d", ks
->name_stem
, j
);
197 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
200 struct btrfs_lockdep_keyset
*ks
;
202 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
204 /* find the matching keyset, id 0 is the default entry */
205 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
206 if (ks
->id
== objectid
)
209 lockdep_set_class_and_name(&eb
->lock
,
210 &ks
->keys
[level
], ks
->names
[level
]);
216 * extents on the btree inode are pretty simple, there's one extent
217 * that covers the entire device
219 static struct extent_map
*btree_get_extent(struct inode
*inode
,
220 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
223 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
224 struct extent_map
*em
;
227 read_lock(&em_tree
->lock
);
228 em
= lookup_extent_mapping(em_tree
, start
, len
);
231 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
232 read_unlock(&em_tree
->lock
);
235 read_unlock(&em_tree
->lock
);
237 em
= alloc_extent_map();
239 em
= ERR_PTR(-ENOMEM
);
244 em
->block_len
= (u64
)-1;
246 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
248 write_lock(&em_tree
->lock
);
249 ret
= add_extent_mapping(em_tree
, em
, 0);
250 if (ret
== -EEXIST
) {
252 em
= lookup_extent_mapping(em_tree
, start
, len
);
259 write_unlock(&em_tree
->lock
);
265 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
267 return btrfs_crc32c(seed
, data
, len
);
270 void btrfs_csum_final(u32 crc
, char *result
)
272 put_unaligned_le32(~crc
, result
);
276 * compute the csum for a btree block, and either verify it or write it
277 * into the csum field of the block.
279 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
280 struct extent_buffer
*buf
,
283 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
286 unsigned long cur_len
;
287 unsigned long offset
= BTRFS_CSUM_SIZE
;
289 unsigned long map_start
;
290 unsigned long map_len
;
293 unsigned long inline_result
;
295 len
= buf
->len
- offset
;
297 err
= map_private_extent_buffer(buf
, offset
, 32,
298 &kaddr
, &map_start
, &map_len
);
301 cur_len
= min(len
, map_len
- (offset
- map_start
));
302 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
307 if (csum_size
> sizeof(inline_result
)) {
308 result
= kzalloc(csum_size
, GFP_NOFS
);
312 result
= (char *)&inline_result
;
315 btrfs_csum_final(crc
, result
);
318 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
321 memcpy(&found
, result
, csum_size
);
323 read_extent_buffer(buf
, &val
, 0, csum_size
);
324 btrfs_warn_rl(fs_info
,
325 "%s checksum verify failed on %llu wanted %X found %X "
327 fs_info
->sb
->s_id
, buf
->start
,
328 val
, found
, btrfs_header_level(buf
));
329 if (result
!= (char *)&inline_result
)
334 write_extent_buffer(buf
, result
, 0, csum_size
);
336 if (result
!= (char *)&inline_result
)
342 * we can't consider a given block up to date unless the transid of the
343 * block matches the transid in the parent node's pointer. This is how we
344 * detect blocks that either didn't get written at all or got written
345 * in the wrong place.
347 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
348 struct extent_buffer
*eb
, u64 parent_transid
,
351 struct extent_state
*cached_state
= NULL
;
353 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
355 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
362 btrfs_tree_read_lock(eb
);
363 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
366 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
368 if (extent_buffer_uptodate(eb
) &&
369 btrfs_header_generation(eb
) == parent_transid
) {
373 btrfs_err_rl(eb
->fs_info
,
374 "parent transid verify failed on %llu wanted %llu found %llu",
376 parent_transid
, btrfs_header_generation(eb
));
380 * Things reading via commit roots that don't have normal protection,
381 * like send, can have a really old block in cache that may point at a
382 * block that has been free'd and re-allocated. So don't clear uptodate
383 * if we find an eb that is under IO (dirty/writeback) because we could
384 * end up reading in the stale data and then writing it back out and
385 * making everybody very sad.
387 if (!extent_buffer_under_io(eb
))
388 clear_extent_buffer_uptodate(eb
);
390 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
391 &cached_state
, GFP_NOFS
);
393 btrfs_tree_read_unlock_blocking(eb
);
398 * Return 0 if the superblock checksum type matches the checksum value of that
399 * algorithm. Pass the raw disk superblock data.
401 static int btrfs_check_super_csum(char *raw_disk_sb
)
403 struct btrfs_super_block
*disk_sb
=
404 (struct btrfs_super_block
*)raw_disk_sb
;
405 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
408 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
410 const int csum_size
= sizeof(crc
);
411 char result
[csum_size
];
414 * The super_block structure does not span the whole
415 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
416 * is filled with zeros and is included in the checkum.
418 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
419 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
420 btrfs_csum_final(crc
, result
);
422 if (memcmp(raw_disk_sb
, result
, csum_size
))
426 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
427 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
436 * helper to read a given tree block, doing retries as required when
437 * the checksums don't match and we have alternate mirrors to try.
439 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
440 struct extent_buffer
*eb
,
441 u64 start
, u64 parent_transid
)
443 struct extent_io_tree
*io_tree
;
448 int failed_mirror
= 0;
450 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
451 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
453 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
455 btree_get_extent
, mirror_num
);
457 if (!verify_parent_transid(io_tree
, eb
,
465 * This buffer's crc is fine, but its contents are corrupted, so
466 * there is no reason to read the other copies, they won't be
469 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
472 num_copies
= btrfs_num_copies(root
->fs_info
,
477 if (!failed_mirror
) {
479 failed_mirror
= eb
->read_mirror
;
483 if (mirror_num
== failed_mirror
)
486 if (mirror_num
> num_copies
)
490 if (failed
&& !ret
&& failed_mirror
)
491 repair_eb_io_failure(root
, eb
, failed_mirror
);
497 * checksum a dirty tree block before IO. This has extra checks to make sure
498 * we only fill in the checksum field in the first page of a multi-page block
501 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
503 u64 start
= page_offset(page
);
505 struct extent_buffer
*eb
;
507 eb
= (struct extent_buffer
*)page
->private;
508 if (page
!= eb
->pages
[0])
510 found_start
= btrfs_header_bytenr(eb
);
511 if (WARN_ON(found_start
!= start
|| !PageUptodate(page
)))
513 csum_tree_block(fs_info
, eb
, 0);
517 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
518 struct extent_buffer
*eb
)
520 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
521 u8 fsid
[BTRFS_UUID_SIZE
];
524 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
526 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
530 fs_devices
= fs_devices
->seed
;
535 #define CORRUPT(reason, eb, root, slot) \
536 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
537 "root=%llu, slot=%d", reason, \
538 btrfs_header_bytenr(eb), root->objectid, slot)
540 static noinline
int check_leaf(struct btrfs_root
*root
,
541 struct extent_buffer
*leaf
)
543 struct btrfs_key key
;
544 struct btrfs_key leaf_key
;
545 u32 nritems
= btrfs_header_nritems(leaf
);
551 /* Check the 0 item */
552 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
553 BTRFS_LEAF_DATA_SIZE(root
)) {
554 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
559 * Check to make sure each items keys are in the correct order and their
560 * offsets make sense. We only have to loop through nritems-1 because
561 * we check the current slot against the next slot, which verifies the
562 * next slot's offset+size makes sense and that the current's slot
565 for (slot
= 0; slot
< nritems
- 1; slot
++) {
566 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
567 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
569 /* Make sure the keys are in the right order */
570 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
571 CORRUPT("bad key order", leaf
, root
, slot
);
576 * Make sure the offset and ends are right, remember that the
577 * item data starts at the end of the leaf and grows towards the
580 if (btrfs_item_offset_nr(leaf
, slot
) !=
581 btrfs_item_end_nr(leaf
, slot
+ 1)) {
582 CORRUPT("slot offset bad", leaf
, root
, slot
);
587 * Check to make sure that we don't point outside of the leaf,
588 * just incase all the items are consistent to eachother, but
589 * all point outside of the leaf.
591 if (btrfs_item_end_nr(leaf
, slot
) >
592 BTRFS_LEAF_DATA_SIZE(root
)) {
593 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
601 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
602 u64 phy_offset
, struct page
*page
,
603 u64 start
, u64 end
, int mirror
)
607 struct extent_buffer
*eb
;
608 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
615 eb
= (struct extent_buffer
*)page
->private;
617 /* the pending IO might have been the only thing that kept this buffer
618 * in memory. Make sure we have a ref for all this other checks
620 extent_buffer_get(eb
);
622 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
626 eb
->read_mirror
= mirror
;
627 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
632 found_start
= btrfs_header_bytenr(eb
);
633 if (found_start
!= eb
->start
) {
634 btrfs_err_rl(eb
->fs_info
, "bad tree block start %llu %llu",
635 found_start
, eb
->start
);
639 if (check_tree_block_fsid(root
->fs_info
, eb
)) {
640 btrfs_err_rl(eb
->fs_info
, "bad fsid on block %llu",
645 found_level
= btrfs_header_level(eb
);
646 if (found_level
>= BTRFS_MAX_LEVEL
) {
647 btrfs_err(root
->fs_info
, "bad tree block level %d",
648 (int)btrfs_header_level(eb
));
653 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
656 ret
= csum_tree_block(root
->fs_info
, eb
, 1);
663 * If this is a leaf block and it is corrupt, set the corrupt bit so
664 * that we don't try and read the other copies of this block, just
667 if (found_level
== 0 && check_leaf(root
, eb
)) {
668 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
673 set_extent_buffer_uptodate(eb
);
676 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
677 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
681 * our io error hook is going to dec the io pages
682 * again, we have to make sure it has something
685 atomic_inc(&eb
->io_pages
);
686 clear_extent_buffer_uptodate(eb
);
688 free_extent_buffer(eb
);
693 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
695 struct extent_buffer
*eb
;
696 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
698 eb
= (struct extent_buffer
*)page
->private;
699 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
700 eb
->read_mirror
= failed_mirror
;
701 atomic_dec(&eb
->io_pages
);
702 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
703 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
704 return -EIO
; /* we fixed nothing */
707 static void end_workqueue_bio(struct bio
*bio
)
709 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
710 struct btrfs_fs_info
*fs_info
;
711 struct btrfs_workqueue
*wq
;
712 btrfs_work_func_t func
;
714 fs_info
= end_io_wq
->info
;
715 end_io_wq
->error
= bio
->bi_error
;
717 if (bio
->bi_rw
& REQ_WRITE
) {
718 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
719 wq
= fs_info
->endio_meta_write_workers
;
720 func
= btrfs_endio_meta_write_helper
;
721 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
722 wq
= fs_info
->endio_freespace_worker
;
723 func
= btrfs_freespace_write_helper
;
724 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
725 wq
= fs_info
->endio_raid56_workers
;
726 func
= btrfs_endio_raid56_helper
;
728 wq
= fs_info
->endio_write_workers
;
729 func
= btrfs_endio_write_helper
;
732 if (unlikely(end_io_wq
->metadata
==
733 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
734 wq
= fs_info
->endio_repair_workers
;
735 func
= btrfs_endio_repair_helper
;
736 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
737 wq
= fs_info
->endio_raid56_workers
;
738 func
= btrfs_endio_raid56_helper
;
739 } else if (end_io_wq
->metadata
) {
740 wq
= fs_info
->endio_meta_workers
;
741 func
= btrfs_endio_meta_helper
;
743 wq
= fs_info
->endio_workers
;
744 func
= btrfs_endio_helper
;
748 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
749 btrfs_queue_work(wq
, &end_io_wq
->work
);
752 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
753 enum btrfs_wq_endio_type metadata
)
755 struct btrfs_end_io_wq
*end_io_wq
;
757 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
761 end_io_wq
->private = bio
->bi_private
;
762 end_io_wq
->end_io
= bio
->bi_end_io
;
763 end_io_wq
->info
= info
;
764 end_io_wq
->error
= 0;
765 end_io_wq
->bio
= bio
;
766 end_io_wq
->metadata
= metadata
;
768 bio
->bi_private
= end_io_wq
;
769 bio
->bi_end_io
= end_workqueue_bio
;
773 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
775 unsigned long limit
= min_t(unsigned long,
776 info
->thread_pool_size
,
777 info
->fs_devices
->open_devices
);
781 static void run_one_async_start(struct btrfs_work
*work
)
783 struct async_submit_bio
*async
;
786 async
= container_of(work
, struct async_submit_bio
, work
);
787 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
788 async
->mirror_num
, async
->bio_flags
,
794 static void run_one_async_done(struct btrfs_work
*work
)
796 struct btrfs_fs_info
*fs_info
;
797 struct async_submit_bio
*async
;
800 async
= container_of(work
, struct async_submit_bio
, work
);
801 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
803 limit
= btrfs_async_submit_limit(fs_info
);
804 limit
= limit
* 2 / 3;
807 * atomic_dec_return implies a barrier for waitqueue_active
809 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
810 waitqueue_active(&fs_info
->async_submit_wait
))
811 wake_up(&fs_info
->async_submit_wait
);
813 /* If an error occured we just want to clean up the bio and move on */
815 async
->bio
->bi_error
= async
->error
;
816 bio_endio(async
->bio
);
820 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
821 async
->mirror_num
, async
->bio_flags
,
825 static void run_one_async_free(struct btrfs_work
*work
)
827 struct async_submit_bio
*async
;
829 async
= container_of(work
, struct async_submit_bio
, work
);
833 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
834 int rw
, struct bio
*bio
, int mirror_num
,
835 unsigned long bio_flags
,
837 extent_submit_bio_hook_t
*submit_bio_start
,
838 extent_submit_bio_hook_t
*submit_bio_done
)
840 struct async_submit_bio
*async
;
842 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
846 async
->inode
= inode
;
849 async
->mirror_num
= mirror_num
;
850 async
->submit_bio_start
= submit_bio_start
;
851 async
->submit_bio_done
= submit_bio_done
;
853 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
854 run_one_async_done
, run_one_async_free
);
856 async
->bio_flags
= bio_flags
;
857 async
->bio_offset
= bio_offset
;
861 atomic_inc(&fs_info
->nr_async_submits
);
864 btrfs_set_work_high_priority(&async
->work
);
866 btrfs_queue_work(fs_info
->workers
, &async
->work
);
868 while (atomic_read(&fs_info
->async_submit_draining
) &&
869 atomic_read(&fs_info
->nr_async_submits
)) {
870 wait_event(fs_info
->async_submit_wait
,
871 (atomic_read(&fs_info
->nr_async_submits
) == 0));
877 static int btree_csum_one_bio(struct bio
*bio
)
879 struct bio_vec
*bvec
;
880 struct btrfs_root
*root
;
883 bio_for_each_segment_all(bvec
, bio
, i
) {
884 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
885 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
893 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
894 struct bio
*bio
, int mirror_num
,
895 unsigned long bio_flags
,
899 * when we're called for a write, we're already in the async
900 * submission context. Just jump into btrfs_map_bio
902 return btree_csum_one_bio(bio
);
905 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
906 int mirror_num
, unsigned long bio_flags
,
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
915 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
923 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
925 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
934 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
935 int mirror_num
, unsigned long bio_flags
,
938 int async
= check_async_write(inode
, bio_flags
);
941 if (!(rw
& REQ_WRITE
)) {
943 * called for a read, do the setup so that checksum validation
944 * can happen in the async kernel threads
946 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
947 bio
, BTRFS_WQ_ENDIO_METADATA
);
950 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
953 ret
= btree_csum_one_bio(bio
);
956 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
960 * kthread helpers are used to submit writes so that
961 * checksumming can happen in parallel across all CPUs
963 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
964 inode
, rw
, bio
, mirror_num
, 0,
966 __btree_submit_bio_start
,
967 __btree_submit_bio_done
);
980 #ifdef CONFIG_MIGRATION
981 static int btree_migratepage(struct address_space
*mapping
,
982 struct page
*newpage
, struct page
*page
,
983 enum migrate_mode mode
)
986 * we can't safely write a btree page from here,
987 * we haven't done the locking hook
992 * Buffers may be managed in a filesystem specific way.
993 * We must have no buffers or drop them.
995 if (page_has_private(page
) &&
996 !try_to_release_page(page
, GFP_KERNEL
))
998 return migrate_page(mapping
, newpage
, page
, mode
);
1003 static int btree_writepages(struct address_space
*mapping
,
1004 struct writeback_control
*wbc
)
1006 struct btrfs_fs_info
*fs_info
;
1009 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1011 if (wbc
->for_kupdate
)
1014 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1015 /* this is a bit racy, but that's ok */
1016 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1017 BTRFS_DIRTY_METADATA_THRESH
);
1021 return btree_write_cache_pages(mapping
, wbc
);
1024 static int btree_readpage(struct file
*file
, struct page
*page
)
1026 struct extent_io_tree
*tree
;
1027 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1028 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1031 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1033 if (PageWriteback(page
) || PageDirty(page
))
1036 return try_release_extent_buffer(page
);
1039 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1040 unsigned int length
)
1042 struct extent_io_tree
*tree
;
1043 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1044 extent_invalidatepage(tree
, page
, offset
);
1045 btree_releasepage(page
, GFP_NOFS
);
1046 if (PagePrivate(page
)) {
1047 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1048 "page private not zero on page %llu",
1049 (unsigned long long)page_offset(page
));
1050 ClearPagePrivate(page
);
1051 set_page_private(page
, 0);
1052 page_cache_release(page
);
1056 static int btree_set_page_dirty(struct page
*page
)
1059 struct extent_buffer
*eb
;
1061 BUG_ON(!PagePrivate(page
));
1062 eb
= (struct extent_buffer
*)page
->private;
1064 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1065 BUG_ON(!atomic_read(&eb
->refs
));
1066 btrfs_assert_tree_locked(eb
);
1068 return __set_page_dirty_nobuffers(page
);
1071 static const struct address_space_operations btree_aops
= {
1072 .readpage
= btree_readpage
,
1073 .writepages
= btree_writepages
,
1074 .releasepage
= btree_releasepage
,
1075 .invalidatepage
= btree_invalidatepage
,
1076 #ifdef CONFIG_MIGRATION
1077 .migratepage
= btree_migratepage
,
1079 .set_page_dirty
= btree_set_page_dirty
,
1082 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1084 struct extent_buffer
*buf
= NULL
;
1085 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1087 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1090 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1091 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1092 free_extent_buffer(buf
);
1095 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1096 int mirror_num
, struct extent_buffer
**eb
)
1098 struct extent_buffer
*buf
= NULL
;
1099 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1100 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1103 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1107 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1109 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1110 btree_get_extent
, mirror_num
);
1112 free_extent_buffer(buf
);
1116 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1117 free_extent_buffer(buf
);
1119 } else if (extent_buffer_uptodate(buf
)) {
1122 free_extent_buffer(buf
);
1127 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1130 return find_extent_buffer(fs_info
, bytenr
);
1133 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1136 if (btrfs_test_is_dummy_root(root
))
1137 return alloc_test_extent_buffer(root
->fs_info
, bytenr
);
1138 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1142 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1144 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1145 buf
->start
+ buf
->len
- 1);
1148 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1150 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1151 buf
->start
, buf
->start
+ buf
->len
- 1);
1154 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1157 struct extent_buffer
*buf
= NULL
;
1160 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1162 return ERR_PTR(-ENOMEM
);
1164 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1166 free_extent_buffer(buf
);
1167 return ERR_PTR(ret
);
1173 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1174 struct btrfs_fs_info
*fs_info
,
1175 struct extent_buffer
*buf
)
1177 if (btrfs_header_generation(buf
) ==
1178 fs_info
->running_transaction
->transid
) {
1179 btrfs_assert_tree_locked(buf
);
1181 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1182 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1184 fs_info
->dirty_metadata_batch
);
1185 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1186 btrfs_set_lock_blocking(buf
);
1187 clear_extent_buffer_dirty(buf
);
1192 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1194 struct btrfs_subvolume_writers
*writers
;
1197 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1199 return ERR_PTR(-ENOMEM
);
1201 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1204 return ERR_PTR(ret
);
1207 init_waitqueue_head(&writers
->wait
);
1212 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1214 percpu_counter_destroy(&writers
->counter
);
1218 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1219 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1223 root
->commit_root
= NULL
;
1224 root
->sectorsize
= sectorsize
;
1225 root
->nodesize
= nodesize
;
1226 root
->stripesize
= stripesize
;
1228 root
->orphan_cleanup_state
= 0;
1230 root
->objectid
= objectid
;
1231 root
->last_trans
= 0;
1232 root
->highest_objectid
= 0;
1233 root
->nr_delalloc_inodes
= 0;
1234 root
->nr_ordered_extents
= 0;
1236 root
->inode_tree
= RB_ROOT
;
1237 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1238 root
->block_rsv
= NULL
;
1239 root
->orphan_block_rsv
= NULL
;
1241 INIT_LIST_HEAD(&root
->dirty_list
);
1242 INIT_LIST_HEAD(&root
->root_list
);
1243 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1244 INIT_LIST_HEAD(&root
->delalloc_root
);
1245 INIT_LIST_HEAD(&root
->ordered_extents
);
1246 INIT_LIST_HEAD(&root
->ordered_root
);
1247 INIT_LIST_HEAD(&root
->logged_list
[0]);
1248 INIT_LIST_HEAD(&root
->logged_list
[1]);
1249 spin_lock_init(&root
->orphan_lock
);
1250 spin_lock_init(&root
->inode_lock
);
1251 spin_lock_init(&root
->delalloc_lock
);
1252 spin_lock_init(&root
->ordered_extent_lock
);
1253 spin_lock_init(&root
->accounting_lock
);
1254 spin_lock_init(&root
->log_extents_lock
[0]);
1255 spin_lock_init(&root
->log_extents_lock
[1]);
1256 mutex_init(&root
->objectid_mutex
);
1257 mutex_init(&root
->log_mutex
);
1258 mutex_init(&root
->ordered_extent_mutex
);
1259 mutex_init(&root
->delalloc_mutex
);
1260 init_waitqueue_head(&root
->log_writer_wait
);
1261 init_waitqueue_head(&root
->log_commit_wait
[0]);
1262 init_waitqueue_head(&root
->log_commit_wait
[1]);
1263 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1264 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1265 atomic_set(&root
->log_commit
[0], 0);
1266 atomic_set(&root
->log_commit
[1], 0);
1267 atomic_set(&root
->log_writers
, 0);
1268 atomic_set(&root
->log_batch
, 0);
1269 atomic_set(&root
->orphan_inodes
, 0);
1270 atomic_set(&root
->refs
, 1);
1271 atomic_set(&root
->will_be_snapshoted
, 0);
1272 atomic_set(&root
->qgroup_meta_rsv
, 0);
1273 root
->log_transid
= 0;
1274 root
->log_transid_committed
= -1;
1275 root
->last_log_commit
= 0;
1277 extent_io_tree_init(&root
->dirty_log_pages
,
1278 fs_info
->btree_inode
->i_mapping
);
1280 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1281 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1282 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1284 root
->defrag_trans_start
= fs_info
->generation
;
1286 root
->defrag_trans_start
= 0;
1287 root
->root_key
.objectid
= objectid
;
1290 spin_lock_init(&root
->root_item_lock
);
1293 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1295 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1297 root
->fs_info
= fs_info
;
1301 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1302 /* Should only be used by the testing infrastructure */
1303 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1305 struct btrfs_root
*root
;
1307 root
= btrfs_alloc_root(NULL
);
1309 return ERR_PTR(-ENOMEM
);
1310 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1311 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1312 root
->alloc_bytenr
= 0;
1318 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1319 struct btrfs_fs_info
*fs_info
,
1322 struct extent_buffer
*leaf
;
1323 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1324 struct btrfs_root
*root
;
1325 struct btrfs_key key
;
1329 root
= btrfs_alloc_root(fs_info
);
1331 return ERR_PTR(-ENOMEM
);
1333 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1334 tree_root
->stripesize
, root
, fs_info
, objectid
);
1335 root
->root_key
.objectid
= objectid
;
1336 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1337 root
->root_key
.offset
= 0;
1339 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1341 ret
= PTR_ERR(leaf
);
1346 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1347 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1348 btrfs_set_header_generation(leaf
, trans
->transid
);
1349 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1350 btrfs_set_header_owner(leaf
, objectid
);
1353 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1355 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1356 btrfs_header_chunk_tree_uuid(leaf
),
1358 btrfs_mark_buffer_dirty(leaf
);
1360 root
->commit_root
= btrfs_root_node(root
);
1361 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1363 root
->root_item
.flags
= 0;
1364 root
->root_item
.byte_limit
= 0;
1365 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1366 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1367 btrfs_set_root_level(&root
->root_item
, 0);
1368 btrfs_set_root_refs(&root
->root_item
, 1);
1369 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1370 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1371 btrfs_set_root_dirid(&root
->root_item
, 0);
1373 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1374 root
->root_item
.drop_level
= 0;
1376 key
.objectid
= objectid
;
1377 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1379 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1383 btrfs_tree_unlock(leaf
);
1389 btrfs_tree_unlock(leaf
);
1390 free_extent_buffer(root
->commit_root
);
1391 free_extent_buffer(leaf
);
1395 return ERR_PTR(ret
);
1398 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1399 struct btrfs_fs_info
*fs_info
)
1401 struct btrfs_root
*root
;
1402 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1403 struct extent_buffer
*leaf
;
1405 root
= btrfs_alloc_root(fs_info
);
1407 return ERR_PTR(-ENOMEM
);
1409 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1410 tree_root
->stripesize
, root
, fs_info
,
1411 BTRFS_TREE_LOG_OBJECTID
);
1413 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1414 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1415 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1418 * DON'T set REF_COWS for log trees
1420 * log trees do not get reference counted because they go away
1421 * before a real commit is actually done. They do store pointers
1422 * to file data extents, and those reference counts still get
1423 * updated (along with back refs to the log tree).
1426 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1430 return ERR_CAST(leaf
);
1433 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1434 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1435 btrfs_set_header_generation(leaf
, trans
->transid
);
1436 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1437 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1440 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1441 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1442 btrfs_mark_buffer_dirty(root
->node
);
1443 btrfs_tree_unlock(root
->node
);
1447 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1448 struct btrfs_fs_info
*fs_info
)
1450 struct btrfs_root
*log_root
;
1452 log_root
= alloc_log_tree(trans
, fs_info
);
1453 if (IS_ERR(log_root
))
1454 return PTR_ERR(log_root
);
1455 WARN_ON(fs_info
->log_root_tree
);
1456 fs_info
->log_root_tree
= log_root
;
1460 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1461 struct btrfs_root
*root
)
1463 struct btrfs_root
*log_root
;
1464 struct btrfs_inode_item
*inode_item
;
1466 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1467 if (IS_ERR(log_root
))
1468 return PTR_ERR(log_root
);
1470 log_root
->last_trans
= trans
->transid
;
1471 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1473 inode_item
= &log_root
->root_item
.inode
;
1474 btrfs_set_stack_inode_generation(inode_item
, 1);
1475 btrfs_set_stack_inode_size(inode_item
, 3);
1476 btrfs_set_stack_inode_nlink(inode_item
, 1);
1477 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1478 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1480 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1482 WARN_ON(root
->log_root
);
1483 root
->log_root
= log_root
;
1484 root
->log_transid
= 0;
1485 root
->log_transid_committed
= -1;
1486 root
->last_log_commit
= 0;
1490 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1491 struct btrfs_key
*key
)
1493 struct btrfs_root
*root
;
1494 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1495 struct btrfs_path
*path
;
1499 path
= btrfs_alloc_path();
1501 return ERR_PTR(-ENOMEM
);
1503 root
= btrfs_alloc_root(fs_info
);
1509 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1510 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1512 ret
= btrfs_find_root(tree_root
, key
, path
,
1513 &root
->root_item
, &root
->root_key
);
1520 generation
= btrfs_root_generation(&root
->root_item
);
1521 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1523 if (IS_ERR(root
->node
)) {
1524 ret
= PTR_ERR(root
->node
);
1526 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1528 free_extent_buffer(root
->node
);
1531 root
->commit_root
= btrfs_root_node(root
);
1533 btrfs_free_path(path
);
1539 root
= ERR_PTR(ret
);
1543 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1544 struct btrfs_key
*location
)
1546 struct btrfs_root
*root
;
1548 root
= btrfs_read_tree_root(tree_root
, location
);
1552 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1553 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1554 btrfs_check_and_init_root_item(&root
->root_item
);
1560 int btrfs_init_fs_root(struct btrfs_root
*root
)
1563 struct btrfs_subvolume_writers
*writers
;
1565 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1566 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1568 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1573 writers
= btrfs_alloc_subvolume_writers();
1574 if (IS_ERR(writers
)) {
1575 ret
= PTR_ERR(writers
);
1578 root
->subv_writers
= writers
;
1580 btrfs_init_free_ino_ctl(root
);
1581 spin_lock_init(&root
->ino_cache_lock
);
1582 init_waitqueue_head(&root
->ino_cache_wait
);
1584 ret
= get_anon_bdev(&root
->anon_dev
);
1590 btrfs_free_subvolume_writers(root
->subv_writers
);
1592 kfree(root
->free_ino_ctl
);
1593 kfree(root
->free_ino_pinned
);
1597 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1600 struct btrfs_root
*root
;
1602 spin_lock(&fs_info
->fs_roots_radix_lock
);
1603 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1604 (unsigned long)root_id
);
1605 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1609 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1610 struct btrfs_root
*root
)
1614 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1618 spin_lock(&fs_info
->fs_roots_radix_lock
);
1619 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1620 (unsigned long)root
->root_key
.objectid
,
1623 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1624 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1625 radix_tree_preload_end();
1630 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1631 struct btrfs_key
*location
,
1634 struct btrfs_root
*root
;
1635 struct btrfs_path
*path
;
1636 struct btrfs_key key
;
1639 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1640 return fs_info
->tree_root
;
1641 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1642 return fs_info
->extent_root
;
1643 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1644 return fs_info
->chunk_root
;
1645 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1646 return fs_info
->dev_root
;
1647 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1648 return fs_info
->csum_root
;
1649 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1650 return fs_info
->quota_root
? fs_info
->quota_root
:
1652 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1653 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1655 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1656 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1659 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1661 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1662 return ERR_PTR(-ENOENT
);
1666 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1670 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1675 ret
= btrfs_init_fs_root(root
);
1679 path
= btrfs_alloc_path();
1684 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1685 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1686 key
.offset
= location
->objectid
;
1688 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1689 btrfs_free_path(path
);
1693 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1695 ret
= btrfs_insert_fs_root(fs_info
, root
);
1697 if (ret
== -EEXIST
) {
1706 return ERR_PTR(ret
);
1709 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1711 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1713 struct btrfs_device
*device
;
1714 struct backing_dev_info
*bdi
;
1717 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1720 bdi
= blk_get_backing_dev_info(device
->bdev
);
1721 if (bdi_congested(bdi
, bdi_bits
)) {
1730 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1734 err
= bdi_setup_and_register(bdi
, "btrfs");
1738 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
;
1739 bdi
->congested_fn
= btrfs_congested_fn
;
1740 bdi
->congested_data
= info
;
1741 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1746 * called by the kthread helper functions to finally call the bio end_io
1747 * functions. This is where read checksum verification actually happens
1749 static void end_workqueue_fn(struct btrfs_work
*work
)
1752 struct btrfs_end_io_wq
*end_io_wq
;
1754 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1755 bio
= end_io_wq
->bio
;
1757 bio
->bi_error
= end_io_wq
->error
;
1758 bio
->bi_private
= end_io_wq
->private;
1759 bio
->bi_end_io
= end_io_wq
->end_io
;
1760 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1764 static int cleaner_kthread(void *arg
)
1766 struct btrfs_root
*root
= arg
;
1768 struct btrfs_trans_handle
*trans
;
1774 /* Make the cleaner go to sleep early. */
1775 if (btrfs_need_cleaner_sleep(root
))
1778 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1782 * Avoid the problem that we change the status of the fs
1783 * during the above check and trylock.
1785 if (btrfs_need_cleaner_sleep(root
)) {
1786 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1790 btrfs_run_delayed_iputs(root
);
1791 again
= btrfs_clean_one_deleted_snapshot(root
);
1792 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1795 * The defragger has dealt with the R/O remount and umount,
1796 * needn't do anything special here.
1798 btrfs_run_defrag_inodes(root
->fs_info
);
1801 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1802 * with relocation (btrfs_relocate_chunk) and relocation
1803 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1804 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1805 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1806 * unused block groups.
1808 btrfs_delete_unused_bgs(root
->fs_info
);
1810 if (!try_to_freeze() && !again
) {
1811 set_current_state(TASK_INTERRUPTIBLE
);
1812 if (!kthread_should_stop())
1814 __set_current_state(TASK_RUNNING
);
1816 } while (!kthread_should_stop());
1819 * Transaction kthread is stopped before us and wakes us up.
1820 * However we might have started a new transaction and COWed some
1821 * tree blocks when deleting unused block groups for example. So
1822 * make sure we commit the transaction we started to have a clean
1823 * shutdown when evicting the btree inode - if it has dirty pages
1824 * when we do the final iput() on it, eviction will trigger a
1825 * writeback for it which will fail with null pointer dereferences
1826 * since work queues and other resources were already released and
1827 * destroyed by the time the iput/eviction/writeback is made.
1829 trans
= btrfs_attach_transaction(root
);
1830 if (IS_ERR(trans
)) {
1831 if (PTR_ERR(trans
) != -ENOENT
)
1832 btrfs_err(root
->fs_info
,
1833 "cleaner transaction attach returned %ld",
1838 ret
= btrfs_commit_transaction(trans
, root
);
1840 btrfs_err(root
->fs_info
,
1841 "cleaner open transaction commit returned %d",
1848 static int transaction_kthread(void *arg
)
1850 struct btrfs_root
*root
= arg
;
1851 struct btrfs_trans_handle
*trans
;
1852 struct btrfs_transaction
*cur
;
1855 unsigned long delay
;
1859 cannot_commit
= false;
1860 delay
= HZ
* root
->fs_info
->commit_interval
;
1861 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1863 spin_lock(&root
->fs_info
->trans_lock
);
1864 cur
= root
->fs_info
->running_transaction
;
1866 spin_unlock(&root
->fs_info
->trans_lock
);
1870 now
= get_seconds();
1871 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1872 (now
< cur
->start_time
||
1873 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1874 spin_unlock(&root
->fs_info
->trans_lock
);
1878 transid
= cur
->transid
;
1879 spin_unlock(&root
->fs_info
->trans_lock
);
1881 /* If the file system is aborted, this will always fail. */
1882 trans
= btrfs_attach_transaction(root
);
1883 if (IS_ERR(trans
)) {
1884 if (PTR_ERR(trans
) != -ENOENT
)
1885 cannot_commit
= true;
1888 if (transid
== trans
->transid
) {
1889 btrfs_commit_transaction(trans
, root
);
1891 btrfs_end_transaction(trans
, root
);
1894 wake_up_process(root
->fs_info
->cleaner_kthread
);
1895 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1897 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1898 &root
->fs_info
->fs_state
)))
1899 btrfs_cleanup_transaction(root
);
1900 if (!try_to_freeze()) {
1901 set_current_state(TASK_INTERRUPTIBLE
);
1902 if (!kthread_should_stop() &&
1903 (!btrfs_transaction_blocked(root
->fs_info
) ||
1905 schedule_timeout(delay
);
1906 __set_current_state(TASK_RUNNING
);
1908 } while (!kthread_should_stop());
1913 * this will find the highest generation in the array of
1914 * root backups. The index of the highest array is returned,
1915 * or -1 if we can't find anything.
1917 * We check to make sure the array is valid by comparing the
1918 * generation of the latest root in the array with the generation
1919 * in the super block. If they don't match we pitch it.
1921 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1924 int newest_index
= -1;
1925 struct btrfs_root_backup
*root_backup
;
1928 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1929 root_backup
= info
->super_copy
->super_roots
+ i
;
1930 cur
= btrfs_backup_tree_root_gen(root_backup
);
1931 if (cur
== newest_gen
)
1935 /* check to see if we actually wrapped around */
1936 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1937 root_backup
= info
->super_copy
->super_roots
;
1938 cur
= btrfs_backup_tree_root_gen(root_backup
);
1939 if (cur
== newest_gen
)
1942 return newest_index
;
1947 * find the oldest backup so we know where to store new entries
1948 * in the backup array. This will set the backup_root_index
1949 * field in the fs_info struct
1951 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1954 int newest_index
= -1;
1956 newest_index
= find_newest_super_backup(info
, newest_gen
);
1957 /* if there was garbage in there, just move along */
1958 if (newest_index
== -1) {
1959 info
->backup_root_index
= 0;
1961 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1966 * copy all the root pointers into the super backup array.
1967 * this will bump the backup pointer by one when it is
1970 static void backup_super_roots(struct btrfs_fs_info
*info
)
1973 struct btrfs_root_backup
*root_backup
;
1976 next_backup
= info
->backup_root_index
;
1977 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1978 BTRFS_NUM_BACKUP_ROOTS
;
1981 * just overwrite the last backup if we're at the same generation
1982 * this happens only at umount
1984 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1985 if (btrfs_backup_tree_root_gen(root_backup
) ==
1986 btrfs_header_generation(info
->tree_root
->node
))
1987 next_backup
= last_backup
;
1989 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1992 * make sure all of our padding and empty slots get zero filled
1993 * regardless of which ones we use today
1995 memset(root_backup
, 0, sizeof(*root_backup
));
1997 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1999 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
2000 btrfs_set_backup_tree_root_gen(root_backup
,
2001 btrfs_header_generation(info
->tree_root
->node
));
2003 btrfs_set_backup_tree_root_level(root_backup
,
2004 btrfs_header_level(info
->tree_root
->node
));
2006 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2007 btrfs_set_backup_chunk_root_gen(root_backup
,
2008 btrfs_header_generation(info
->chunk_root
->node
));
2009 btrfs_set_backup_chunk_root_level(root_backup
,
2010 btrfs_header_level(info
->chunk_root
->node
));
2012 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2013 btrfs_set_backup_extent_root_gen(root_backup
,
2014 btrfs_header_generation(info
->extent_root
->node
));
2015 btrfs_set_backup_extent_root_level(root_backup
,
2016 btrfs_header_level(info
->extent_root
->node
));
2019 * we might commit during log recovery, which happens before we set
2020 * the fs_root. Make sure it is valid before we fill it in.
2022 if (info
->fs_root
&& info
->fs_root
->node
) {
2023 btrfs_set_backup_fs_root(root_backup
,
2024 info
->fs_root
->node
->start
);
2025 btrfs_set_backup_fs_root_gen(root_backup
,
2026 btrfs_header_generation(info
->fs_root
->node
));
2027 btrfs_set_backup_fs_root_level(root_backup
,
2028 btrfs_header_level(info
->fs_root
->node
));
2031 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2032 btrfs_set_backup_dev_root_gen(root_backup
,
2033 btrfs_header_generation(info
->dev_root
->node
));
2034 btrfs_set_backup_dev_root_level(root_backup
,
2035 btrfs_header_level(info
->dev_root
->node
));
2037 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2038 btrfs_set_backup_csum_root_gen(root_backup
,
2039 btrfs_header_generation(info
->csum_root
->node
));
2040 btrfs_set_backup_csum_root_level(root_backup
,
2041 btrfs_header_level(info
->csum_root
->node
));
2043 btrfs_set_backup_total_bytes(root_backup
,
2044 btrfs_super_total_bytes(info
->super_copy
));
2045 btrfs_set_backup_bytes_used(root_backup
,
2046 btrfs_super_bytes_used(info
->super_copy
));
2047 btrfs_set_backup_num_devices(root_backup
,
2048 btrfs_super_num_devices(info
->super_copy
));
2051 * if we don't copy this out to the super_copy, it won't get remembered
2052 * for the next commit
2054 memcpy(&info
->super_copy
->super_roots
,
2055 &info
->super_for_commit
->super_roots
,
2056 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2060 * this copies info out of the root backup array and back into
2061 * the in-memory super block. It is meant to help iterate through
2062 * the array, so you send it the number of backups you've already
2063 * tried and the last backup index you used.
2065 * this returns -1 when it has tried all the backups
2067 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2068 struct btrfs_super_block
*super
,
2069 int *num_backups_tried
, int *backup_index
)
2071 struct btrfs_root_backup
*root_backup
;
2072 int newest
= *backup_index
;
2074 if (*num_backups_tried
== 0) {
2075 u64 gen
= btrfs_super_generation(super
);
2077 newest
= find_newest_super_backup(info
, gen
);
2081 *backup_index
= newest
;
2082 *num_backups_tried
= 1;
2083 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2084 /* we've tried all the backups, all done */
2087 /* jump to the next oldest backup */
2088 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2089 BTRFS_NUM_BACKUP_ROOTS
;
2090 *backup_index
= newest
;
2091 *num_backups_tried
+= 1;
2093 root_backup
= super
->super_roots
+ newest
;
2095 btrfs_set_super_generation(super
,
2096 btrfs_backup_tree_root_gen(root_backup
));
2097 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2098 btrfs_set_super_root_level(super
,
2099 btrfs_backup_tree_root_level(root_backup
));
2100 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2103 * fixme: the total bytes and num_devices need to match or we should
2106 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2107 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2111 /* helper to cleanup workers */
2112 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2114 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2115 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2116 btrfs_destroy_workqueue(fs_info
->workers
);
2117 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2118 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2119 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2120 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2121 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2122 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2123 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2124 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2125 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2126 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2127 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2128 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2129 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2130 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2131 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2134 static void free_root_extent_buffers(struct btrfs_root
*root
)
2137 free_extent_buffer(root
->node
);
2138 free_extent_buffer(root
->commit_root
);
2140 root
->commit_root
= NULL
;
2144 /* helper to cleanup tree roots */
2145 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2147 free_root_extent_buffers(info
->tree_root
);
2149 free_root_extent_buffers(info
->dev_root
);
2150 free_root_extent_buffers(info
->extent_root
);
2151 free_root_extent_buffers(info
->csum_root
);
2152 free_root_extent_buffers(info
->quota_root
);
2153 free_root_extent_buffers(info
->uuid_root
);
2155 free_root_extent_buffers(info
->chunk_root
);
2156 free_root_extent_buffers(info
->free_space_root
);
2159 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2162 struct btrfs_root
*gang
[8];
2165 while (!list_empty(&fs_info
->dead_roots
)) {
2166 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2167 struct btrfs_root
, root_list
);
2168 list_del(&gang
[0]->root_list
);
2170 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2171 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2173 free_extent_buffer(gang
[0]->node
);
2174 free_extent_buffer(gang
[0]->commit_root
);
2175 btrfs_put_fs_root(gang
[0]);
2180 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2185 for (i
= 0; i
< ret
; i
++)
2186 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2189 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2190 btrfs_free_log_root_tree(NULL
, fs_info
);
2191 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2192 fs_info
->pinned_extents
);
2196 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2198 mutex_init(&fs_info
->scrub_lock
);
2199 atomic_set(&fs_info
->scrubs_running
, 0);
2200 atomic_set(&fs_info
->scrub_pause_req
, 0);
2201 atomic_set(&fs_info
->scrubs_paused
, 0);
2202 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2203 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2204 fs_info
->scrub_workers_refcnt
= 0;
2207 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2209 spin_lock_init(&fs_info
->balance_lock
);
2210 mutex_init(&fs_info
->balance_mutex
);
2211 atomic_set(&fs_info
->balance_running
, 0);
2212 atomic_set(&fs_info
->balance_pause_req
, 0);
2213 atomic_set(&fs_info
->balance_cancel_req
, 0);
2214 fs_info
->balance_ctl
= NULL
;
2215 init_waitqueue_head(&fs_info
->balance_wait_q
);
2218 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2219 struct btrfs_root
*tree_root
)
2221 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2222 set_nlink(fs_info
->btree_inode
, 1);
2224 * we set the i_size on the btree inode to the max possible int.
2225 * the real end of the address space is determined by all of
2226 * the devices in the system
2228 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2229 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2231 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2232 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2233 fs_info
->btree_inode
->i_mapping
);
2234 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2235 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2237 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2239 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2240 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2241 sizeof(struct btrfs_key
));
2242 set_bit(BTRFS_INODE_DUMMY
,
2243 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2244 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2247 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2249 fs_info
->dev_replace
.lock_owner
= 0;
2250 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2251 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2252 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2253 mutex_init(&fs_info
->dev_replace
.lock
);
2254 init_waitqueue_head(&fs_info
->replace_wait
);
2257 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2259 spin_lock_init(&fs_info
->qgroup_lock
);
2260 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2261 fs_info
->qgroup_tree
= RB_ROOT
;
2262 fs_info
->qgroup_op_tree
= RB_ROOT
;
2263 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2264 fs_info
->qgroup_seq
= 1;
2265 fs_info
->quota_enabled
= 0;
2266 fs_info
->pending_quota_state
= 0;
2267 fs_info
->qgroup_ulist
= NULL
;
2268 mutex_init(&fs_info
->qgroup_rescan_lock
);
2271 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2272 struct btrfs_fs_devices
*fs_devices
)
2274 int max_active
= fs_info
->thread_pool_size
;
2275 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2278 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2281 fs_info
->delalloc_workers
=
2282 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2284 fs_info
->flush_workers
=
2285 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2287 fs_info
->caching_workers
=
2288 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2291 * a higher idle thresh on the submit workers makes it much more
2292 * likely that bios will be send down in a sane order to the
2295 fs_info
->submit_workers
=
2296 btrfs_alloc_workqueue("submit", flags
,
2297 min_t(u64
, fs_devices
->num_devices
,
2300 fs_info
->fixup_workers
=
2301 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2304 * endios are largely parallel and should have a very
2307 fs_info
->endio_workers
=
2308 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2309 fs_info
->endio_meta_workers
=
2310 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2311 fs_info
->endio_meta_write_workers
=
2312 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2313 fs_info
->endio_raid56_workers
=
2314 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2315 fs_info
->endio_repair_workers
=
2316 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2317 fs_info
->rmw_workers
=
2318 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2319 fs_info
->endio_write_workers
=
2320 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2321 fs_info
->endio_freespace_worker
=
2322 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2323 fs_info
->delayed_workers
=
2324 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2325 fs_info
->readahead_workers
=
2326 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2327 fs_info
->qgroup_rescan_workers
=
2328 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2329 fs_info
->extent_workers
=
2330 btrfs_alloc_workqueue("extent-refs", flags
,
2331 min_t(u64
, fs_devices
->num_devices
,
2334 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2335 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2336 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2337 fs_info
->endio_meta_write_workers
&&
2338 fs_info
->endio_repair_workers
&&
2339 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2340 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2341 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2342 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2343 fs_info
->extent_workers
&&
2344 fs_info
->qgroup_rescan_workers
)) {
2351 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2352 struct btrfs_fs_devices
*fs_devices
)
2355 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2356 struct btrfs_root
*log_tree_root
;
2357 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2358 u64 bytenr
= btrfs_super_log_root(disk_super
);
2360 if (fs_devices
->rw_devices
== 0) {
2361 btrfs_warn(fs_info
, "log replay required on RO media");
2365 log_tree_root
= btrfs_alloc_root(fs_info
);
2369 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2370 tree_root
->stripesize
, log_tree_root
, fs_info
,
2371 BTRFS_TREE_LOG_OBJECTID
);
2373 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2374 fs_info
->generation
+ 1);
2375 if (IS_ERR(log_tree_root
->node
)) {
2376 btrfs_warn(fs_info
, "failed to read log tree");
2377 ret
= PTR_ERR(log_tree_root
->node
);
2378 kfree(log_tree_root
);
2380 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2381 btrfs_err(fs_info
, "failed to read log tree");
2382 free_extent_buffer(log_tree_root
->node
);
2383 kfree(log_tree_root
);
2386 /* returns with log_tree_root freed on success */
2387 ret
= btrfs_recover_log_trees(log_tree_root
);
2389 btrfs_std_error(tree_root
->fs_info
, ret
,
2390 "Failed to recover log tree");
2391 free_extent_buffer(log_tree_root
->node
);
2392 kfree(log_tree_root
);
2396 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2397 ret
= btrfs_commit_super(tree_root
);
2405 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2406 struct btrfs_root
*tree_root
)
2408 struct btrfs_root
*root
;
2409 struct btrfs_key location
;
2412 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2413 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2414 location
.offset
= 0;
2416 root
= btrfs_read_tree_root(tree_root
, &location
);
2418 return PTR_ERR(root
);
2419 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2420 fs_info
->extent_root
= root
;
2422 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2423 root
= btrfs_read_tree_root(tree_root
, &location
);
2425 return PTR_ERR(root
);
2426 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2427 fs_info
->dev_root
= root
;
2428 btrfs_init_devices_late(fs_info
);
2430 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2431 root
= btrfs_read_tree_root(tree_root
, &location
);
2433 return PTR_ERR(root
);
2434 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2435 fs_info
->csum_root
= root
;
2437 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2438 root
= btrfs_read_tree_root(tree_root
, &location
);
2439 if (!IS_ERR(root
)) {
2440 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2441 fs_info
->quota_enabled
= 1;
2442 fs_info
->pending_quota_state
= 1;
2443 fs_info
->quota_root
= root
;
2446 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2447 root
= btrfs_read_tree_root(tree_root
, &location
);
2449 ret
= PTR_ERR(root
);
2453 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2454 fs_info
->uuid_root
= root
;
2457 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2458 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2459 root
= btrfs_read_tree_root(tree_root
, &location
);
2461 return PTR_ERR(root
);
2462 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2463 fs_info
->free_space_root
= root
;
2469 int open_ctree(struct super_block
*sb
,
2470 struct btrfs_fs_devices
*fs_devices
,
2478 struct btrfs_key location
;
2479 struct buffer_head
*bh
;
2480 struct btrfs_super_block
*disk_super
;
2481 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2482 struct btrfs_root
*tree_root
;
2483 struct btrfs_root
*chunk_root
;
2486 int num_backups_tried
= 0;
2487 int backup_index
= 0;
2490 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2491 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2492 if (!tree_root
|| !chunk_root
) {
2497 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2503 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2509 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2514 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2515 (1 + ilog2(nr_cpu_ids
));
2517 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2520 goto fail_dirty_metadata_bytes
;
2523 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2526 goto fail_delalloc_bytes
;
2529 fs_info
->btree_inode
= new_inode(sb
);
2530 if (!fs_info
->btree_inode
) {
2532 goto fail_bio_counter
;
2535 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2537 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2538 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2539 INIT_LIST_HEAD(&fs_info
->trans_list
);
2540 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2541 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2542 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2543 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2544 spin_lock_init(&fs_info
->delalloc_root_lock
);
2545 spin_lock_init(&fs_info
->trans_lock
);
2546 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2547 spin_lock_init(&fs_info
->delayed_iput_lock
);
2548 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2549 spin_lock_init(&fs_info
->free_chunk_lock
);
2550 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2551 spin_lock_init(&fs_info
->super_lock
);
2552 spin_lock_init(&fs_info
->qgroup_op_lock
);
2553 spin_lock_init(&fs_info
->buffer_lock
);
2554 spin_lock_init(&fs_info
->unused_bgs_lock
);
2555 rwlock_init(&fs_info
->tree_mod_log_lock
);
2556 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2557 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2558 mutex_init(&fs_info
->reloc_mutex
);
2559 mutex_init(&fs_info
->delalloc_root_mutex
);
2560 seqlock_init(&fs_info
->profiles_lock
);
2561 init_rwsem(&fs_info
->delayed_iput_sem
);
2563 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2564 INIT_LIST_HEAD(&fs_info
->space_info
);
2565 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2566 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2567 btrfs_mapping_init(&fs_info
->mapping_tree
);
2568 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2569 BTRFS_BLOCK_RSV_GLOBAL
);
2570 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2571 BTRFS_BLOCK_RSV_DELALLOC
);
2572 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2573 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2574 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2575 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2576 BTRFS_BLOCK_RSV_DELOPS
);
2577 atomic_set(&fs_info
->nr_async_submits
, 0);
2578 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2579 atomic_set(&fs_info
->async_submit_draining
, 0);
2580 atomic_set(&fs_info
->nr_async_bios
, 0);
2581 atomic_set(&fs_info
->defrag_running
, 0);
2582 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2583 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2585 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2586 fs_info
->metadata_ratio
= 0;
2587 fs_info
->defrag_inodes
= RB_ROOT
;
2588 fs_info
->free_chunk_space
= 0;
2589 fs_info
->tree_mod_log
= RB_ROOT
;
2590 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2591 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2592 /* readahead state */
2593 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2594 spin_lock_init(&fs_info
->reada_lock
);
2596 fs_info
->thread_pool_size
= min_t(unsigned long,
2597 num_online_cpus() + 2, 8);
2599 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2600 spin_lock_init(&fs_info
->ordered_root_lock
);
2601 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2603 if (!fs_info
->delayed_root
) {
2607 btrfs_init_delayed_root(fs_info
->delayed_root
);
2609 btrfs_init_scrub(fs_info
);
2610 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2611 fs_info
->check_integrity_print_mask
= 0;
2613 btrfs_init_balance(fs_info
);
2614 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2616 sb
->s_blocksize
= 4096;
2617 sb
->s_blocksize_bits
= blksize_bits(4096);
2618 sb
->s_bdi
= &fs_info
->bdi
;
2620 btrfs_init_btree_inode(fs_info
, tree_root
);
2622 spin_lock_init(&fs_info
->block_group_cache_lock
);
2623 fs_info
->block_group_cache_tree
= RB_ROOT
;
2624 fs_info
->first_logical_byte
= (u64
)-1;
2626 extent_io_tree_init(&fs_info
->freed_extents
[0],
2627 fs_info
->btree_inode
->i_mapping
);
2628 extent_io_tree_init(&fs_info
->freed_extents
[1],
2629 fs_info
->btree_inode
->i_mapping
);
2630 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2631 fs_info
->do_barriers
= 1;
2634 mutex_init(&fs_info
->ordered_operations_mutex
);
2635 mutex_init(&fs_info
->tree_log_mutex
);
2636 mutex_init(&fs_info
->chunk_mutex
);
2637 mutex_init(&fs_info
->transaction_kthread_mutex
);
2638 mutex_init(&fs_info
->cleaner_mutex
);
2639 mutex_init(&fs_info
->volume_mutex
);
2640 mutex_init(&fs_info
->ro_block_group_mutex
);
2641 init_rwsem(&fs_info
->commit_root_sem
);
2642 init_rwsem(&fs_info
->cleanup_work_sem
);
2643 init_rwsem(&fs_info
->subvol_sem
);
2644 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2646 btrfs_init_dev_replace_locks(fs_info
);
2647 btrfs_init_qgroup(fs_info
);
2649 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2650 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2652 init_waitqueue_head(&fs_info
->transaction_throttle
);
2653 init_waitqueue_head(&fs_info
->transaction_wait
);
2654 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2655 init_waitqueue_head(&fs_info
->async_submit_wait
);
2657 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2659 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2665 __setup_root(4096, 4096, 4096, tree_root
,
2666 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2668 invalidate_bdev(fs_devices
->latest_bdev
);
2671 * Read super block and check the signature bytes only
2673 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2680 * We want to check superblock checksum, the type is stored inside.
2681 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2683 if (btrfs_check_super_csum(bh
->b_data
)) {
2684 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2691 * super_copy is zeroed at allocation time and we never touch the
2692 * following bytes up to INFO_SIZE, the checksum is calculated from
2693 * the whole block of INFO_SIZE
2695 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2696 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2697 sizeof(*fs_info
->super_for_commit
));
2700 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2702 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2704 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2709 disk_super
= fs_info
->super_copy
;
2710 if (!btrfs_super_root(disk_super
))
2713 /* check FS state, whether FS is broken. */
2714 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2715 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2718 * run through our array of backup supers and setup
2719 * our ring pointer to the oldest one
2721 generation
= btrfs_super_generation(disk_super
);
2722 find_oldest_super_backup(fs_info
, generation
);
2725 * In the long term, we'll store the compression type in the super
2726 * block, and it'll be used for per file compression control.
2728 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2730 ret
= btrfs_parse_options(tree_root
, options
);
2736 features
= btrfs_super_incompat_flags(disk_super
) &
2737 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2739 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2740 "unsupported optional features (%Lx).\n",
2747 * Leafsize and nodesize were always equal, this is only a sanity check.
2749 if (le32_to_cpu(disk_super
->__unused_leafsize
) !=
2750 btrfs_super_nodesize(disk_super
)) {
2751 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2752 "blocksizes don't match. node %d leaf %d\n",
2753 btrfs_super_nodesize(disk_super
),
2754 le32_to_cpu(disk_super
->__unused_leafsize
));
2758 if (btrfs_super_nodesize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2759 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2760 "blocksize (%d) was too large\n",
2761 btrfs_super_nodesize(disk_super
));
2766 features
= btrfs_super_incompat_flags(disk_super
);
2767 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2768 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2769 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2771 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2772 printk(KERN_INFO
"BTRFS: has skinny extents\n");
2775 * flag our filesystem as having big metadata blocks if
2776 * they are bigger than the page size
2778 if (btrfs_super_nodesize(disk_super
) > PAGE_CACHE_SIZE
) {
2779 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2780 printk(KERN_INFO
"BTRFS: flagging fs with big metadata feature\n");
2781 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2784 nodesize
= btrfs_super_nodesize(disk_super
);
2785 sectorsize
= btrfs_super_sectorsize(disk_super
);
2786 stripesize
= btrfs_super_stripesize(disk_super
);
2787 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2788 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2791 * mixed block groups end up with duplicate but slightly offset
2792 * extent buffers for the same range. It leads to corruptions
2794 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2795 (sectorsize
!= nodesize
)) {
2796 printk(KERN_ERR
"BTRFS: unequal leaf/node/sector sizes "
2797 "are not allowed for mixed block groups on %s\n",
2803 * Needn't use the lock because there is no other task which will
2806 btrfs_set_super_incompat_flags(disk_super
, features
);
2808 features
= btrfs_super_compat_ro_flags(disk_super
) &
2809 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2810 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2811 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2812 "unsupported option features (%Lx).\n",
2818 max_active
= fs_info
->thread_pool_size
;
2820 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2823 goto fail_sb_buffer
;
2826 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2827 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2828 SZ_4M
/ PAGE_CACHE_SIZE
);
2830 tree_root
->nodesize
= nodesize
;
2831 tree_root
->sectorsize
= sectorsize
;
2832 tree_root
->stripesize
= stripesize
;
2834 sb
->s_blocksize
= sectorsize
;
2835 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2837 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2838 printk(KERN_ERR
"BTRFS: valid FS not found on %s\n", sb
->s_id
);
2839 goto fail_sb_buffer
;
2842 if (sectorsize
!= PAGE_SIZE
) {
2843 printk(KERN_ERR
"BTRFS: incompatible sector size (%lu) "
2844 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2845 goto fail_sb_buffer
;
2848 mutex_lock(&fs_info
->chunk_mutex
);
2849 ret
= btrfs_read_sys_array(tree_root
);
2850 mutex_unlock(&fs_info
->chunk_mutex
);
2852 printk(KERN_ERR
"BTRFS: failed to read the system "
2853 "array on %s\n", sb
->s_id
);
2854 goto fail_sb_buffer
;
2857 generation
= btrfs_super_chunk_root_generation(disk_super
);
2859 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2860 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2862 chunk_root
->node
= read_tree_block(chunk_root
,
2863 btrfs_super_chunk_root(disk_super
),
2865 if (IS_ERR(chunk_root
->node
) ||
2866 !extent_buffer_uptodate(chunk_root
->node
)) {
2867 printk(KERN_ERR
"BTRFS: failed to read chunk root on %s\n",
2869 if (!IS_ERR(chunk_root
->node
))
2870 free_extent_buffer(chunk_root
->node
);
2871 chunk_root
->node
= NULL
;
2872 goto fail_tree_roots
;
2874 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2875 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2877 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2878 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2880 ret
= btrfs_read_chunk_tree(chunk_root
);
2882 printk(KERN_ERR
"BTRFS: failed to read chunk tree on %s\n",
2884 goto fail_tree_roots
;
2888 * keep the device that is marked to be the target device for the
2889 * dev_replace procedure
2891 btrfs_close_extra_devices(fs_devices
, 0);
2893 if (!fs_devices
->latest_bdev
) {
2894 printk(KERN_ERR
"BTRFS: failed to read devices on %s\n",
2896 goto fail_tree_roots
;
2900 generation
= btrfs_super_generation(disk_super
);
2902 tree_root
->node
= read_tree_block(tree_root
,
2903 btrfs_super_root(disk_super
),
2905 if (IS_ERR(tree_root
->node
) ||
2906 !extent_buffer_uptodate(tree_root
->node
)) {
2907 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2909 if (!IS_ERR(tree_root
->node
))
2910 free_extent_buffer(tree_root
->node
);
2911 tree_root
->node
= NULL
;
2912 goto recovery_tree_root
;
2915 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2916 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2917 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2919 ret
= btrfs_read_roots(fs_info
, tree_root
);
2921 goto recovery_tree_root
;
2923 fs_info
->generation
= generation
;
2924 fs_info
->last_trans_committed
= generation
;
2926 ret
= btrfs_recover_balance(fs_info
);
2928 printk(KERN_ERR
"BTRFS: failed to recover balance\n");
2929 goto fail_block_groups
;
2932 ret
= btrfs_init_dev_stats(fs_info
);
2934 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2936 goto fail_block_groups
;
2939 ret
= btrfs_init_dev_replace(fs_info
);
2941 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2942 goto fail_block_groups
;
2945 btrfs_close_extra_devices(fs_devices
, 1);
2947 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
2949 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret
);
2950 goto fail_block_groups
;
2953 ret
= btrfs_sysfs_add_device(fs_devices
);
2955 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret
);
2956 goto fail_fsdev_sysfs
;
2959 ret
= btrfs_sysfs_add_mounted(fs_info
);
2961 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2962 goto fail_fsdev_sysfs
;
2965 ret
= btrfs_init_space_info(fs_info
);
2967 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2971 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2973 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
2976 fs_info
->num_tolerated_disk_barrier_failures
=
2977 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2978 if (fs_info
->fs_devices
->missing_devices
>
2979 fs_info
->num_tolerated_disk_barrier_failures
&&
2980 !(sb
->s_flags
& MS_RDONLY
)) {
2981 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2982 fs_info
->fs_devices
->missing_devices
,
2983 fs_info
->num_tolerated_disk_barrier_failures
);
2987 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2989 if (IS_ERR(fs_info
->cleaner_kthread
))
2992 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2994 "btrfs-transaction");
2995 if (IS_ERR(fs_info
->transaction_kthread
))
2998 if (!btrfs_test_opt(tree_root
, SSD
) &&
2999 !btrfs_test_opt(tree_root
, NOSSD
) &&
3000 !fs_info
->fs_devices
->rotating
) {
3001 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
3003 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
3007 * Mount does not set all options immediatelly, we can do it now and do
3008 * not have to wait for transaction commit
3010 btrfs_apply_pending_changes(fs_info
);
3012 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3013 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
3014 ret
= btrfsic_mount(tree_root
, fs_devices
,
3015 btrfs_test_opt(tree_root
,
3016 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3018 fs_info
->check_integrity_print_mask
);
3020 printk(KERN_WARNING
"BTRFS: failed to initialize"
3021 " integrity check module %s\n", sb
->s_id
);
3024 ret
= btrfs_read_qgroup_config(fs_info
);
3026 goto fail_trans_kthread
;
3028 /* do not make disk changes in broken FS */
3029 if (btrfs_super_log_root(disk_super
) != 0) {
3030 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3037 ret
= btrfs_find_orphan_roots(tree_root
);
3041 if (!(sb
->s_flags
& MS_RDONLY
)) {
3042 ret
= btrfs_cleanup_fs_roots(fs_info
);
3046 mutex_lock(&fs_info
->cleaner_mutex
);
3047 ret
= btrfs_recover_relocation(tree_root
);
3048 mutex_unlock(&fs_info
->cleaner_mutex
);
3051 "BTRFS: failed to recover relocation\n");
3057 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3058 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3059 location
.offset
= 0;
3061 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3062 if (IS_ERR(fs_info
->fs_root
)) {
3063 err
= PTR_ERR(fs_info
->fs_root
);
3067 if (sb
->s_flags
& MS_RDONLY
)
3070 if (btrfs_test_opt(tree_root
, FREE_SPACE_TREE
) &&
3071 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3072 pr_info("BTRFS: creating free space tree\n");
3073 ret
= btrfs_create_free_space_tree(fs_info
);
3075 pr_warn("BTRFS: failed to create free space tree %d\n",
3077 close_ctree(tree_root
);
3082 down_read(&fs_info
->cleanup_work_sem
);
3083 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3084 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3085 up_read(&fs_info
->cleanup_work_sem
);
3086 close_ctree(tree_root
);
3089 up_read(&fs_info
->cleanup_work_sem
);
3091 ret
= btrfs_resume_balance_async(fs_info
);
3093 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
3094 close_ctree(tree_root
);
3098 ret
= btrfs_resume_dev_replace_async(fs_info
);
3100 pr_warn("BTRFS: failed to resume dev_replace\n");
3101 close_ctree(tree_root
);
3105 btrfs_qgroup_rescan_resume(fs_info
);
3107 if (btrfs_test_opt(tree_root
, CLEAR_CACHE
) &&
3108 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3109 pr_info("BTRFS: clearing free space tree\n");
3110 ret
= btrfs_clear_free_space_tree(fs_info
);
3112 pr_warn("BTRFS: failed to clear free space tree %d\n",
3114 close_ctree(tree_root
);
3119 if (!fs_info
->uuid_root
) {
3120 pr_info("BTRFS: creating UUID tree\n");
3121 ret
= btrfs_create_uuid_tree(fs_info
);
3123 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3125 close_ctree(tree_root
);
3128 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3129 fs_info
->generation
!=
3130 btrfs_super_uuid_tree_generation(disk_super
)) {
3131 pr_info("BTRFS: checking UUID tree\n");
3132 ret
= btrfs_check_uuid_tree(fs_info
);
3134 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3136 close_ctree(tree_root
);
3140 fs_info
->update_uuid_tree_gen
= 1;
3148 btrfs_free_qgroup_config(fs_info
);
3150 kthread_stop(fs_info
->transaction_kthread
);
3151 btrfs_cleanup_transaction(fs_info
->tree_root
);
3152 btrfs_free_fs_roots(fs_info
);
3154 kthread_stop(fs_info
->cleaner_kthread
);
3157 * make sure we're done with the btree inode before we stop our
3160 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3163 btrfs_sysfs_remove_mounted(fs_info
);
3166 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3169 btrfs_put_block_group_cache(fs_info
);
3170 btrfs_free_block_groups(fs_info
);
3173 free_root_pointers(fs_info
, 1);
3174 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3177 btrfs_stop_all_workers(fs_info
);
3180 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3182 iput(fs_info
->btree_inode
);
3184 percpu_counter_destroy(&fs_info
->bio_counter
);
3185 fail_delalloc_bytes
:
3186 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3187 fail_dirty_metadata_bytes
:
3188 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3190 bdi_destroy(&fs_info
->bdi
);
3192 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3194 btrfs_free_stripe_hash_table(fs_info
);
3195 btrfs_close_devices(fs_info
->fs_devices
);
3199 if (!btrfs_test_opt(tree_root
, RECOVERY
))
3200 goto fail_tree_roots
;
3202 free_root_pointers(fs_info
, 0);
3204 /* don't use the log in recovery mode, it won't be valid */
3205 btrfs_set_super_log_root(disk_super
, 0);
3207 /* we can't trust the free space cache either */
3208 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3210 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3211 &num_backups_tried
, &backup_index
);
3213 goto fail_block_groups
;
3214 goto retry_root_backup
;
3217 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3220 set_buffer_uptodate(bh
);
3222 struct btrfs_device
*device
= (struct btrfs_device
*)
3225 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3226 "lost page write due to IO error on %s",
3227 rcu_str_deref(device
->name
));
3228 /* note, we dont' set_buffer_write_io_error because we have
3229 * our own ways of dealing with the IO errors
3231 clear_buffer_uptodate(bh
);
3232 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3238 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3239 struct buffer_head
**bh_ret
)
3241 struct buffer_head
*bh
;
3242 struct btrfs_super_block
*super
;
3245 bytenr
= btrfs_sb_offset(copy_num
);
3246 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3249 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3251 * If we fail to read from the underlying devices, as of now
3252 * the best option we have is to mark it EIO.
3257 super
= (struct btrfs_super_block
*)bh
->b_data
;
3258 if (btrfs_super_bytenr(super
) != bytenr
||
3259 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3269 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3271 struct buffer_head
*bh
;
3272 struct buffer_head
*latest
= NULL
;
3273 struct btrfs_super_block
*super
;
3278 /* we would like to check all the supers, but that would make
3279 * a btrfs mount succeed after a mkfs from a different FS.
3280 * So, we need to add a special mount option to scan for
3281 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3283 for (i
= 0; i
< 1; i
++) {
3284 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3288 super
= (struct btrfs_super_block
*)bh
->b_data
;
3290 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3293 transid
= btrfs_super_generation(super
);
3300 return ERR_PTR(ret
);
3306 * this should be called twice, once with wait == 0 and
3307 * once with wait == 1. When wait == 0 is done, all the buffer heads
3308 * we write are pinned.
3310 * They are released when wait == 1 is done.
3311 * max_mirrors must be the same for both runs, and it indicates how
3312 * many supers on this one device should be written.
3314 * max_mirrors == 0 means to write them all.
3316 static int write_dev_supers(struct btrfs_device
*device
,
3317 struct btrfs_super_block
*sb
,
3318 int do_barriers
, int wait
, int max_mirrors
)
3320 struct buffer_head
*bh
;
3327 if (max_mirrors
== 0)
3328 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3330 for (i
= 0; i
< max_mirrors
; i
++) {
3331 bytenr
= btrfs_sb_offset(i
);
3332 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3333 device
->commit_total_bytes
)
3337 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3338 BTRFS_SUPER_INFO_SIZE
);
3344 if (!buffer_uptodate(bh
))
3347 /* drop our reference */
3350 /* drop the reference from the wait == 0 run */
3354 btrfs_set_super_bytenr(sb
, bytenr
);
3357 crc
= btrfs_csum_data((char *)sb
+
3358 BTRFS_CSUM_SIZE
, crc
,
3359 BTRFS_SUPER_INFO_SIZE
-
3361 btrfs_csum_final(crc
, sb
->csum
);
3364 * one reference for us, and we leave it for the
3367 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3368 BTRFS_SUPER_INFO_SIZE
);
3370 btrfs_err(device
->dev_root
->fs_info
,
3371 "couldn't get super buffer head for bytenr %llu",
3377 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3379 /* one reference for submit_bh */
3382 set_buffer_uptodate(bh
);
3384 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3385 bh
->b_private
= device
;
3389 * we fua the first super. The others we allow
3393 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3395 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3399 return errors
< i
? 0 : -1;
3403 * endio for the write_dev_flush, this will wake anyone waiting
3404 * for the barrier when it is done
3406 static void btrfs_end_empty_barrier(struct bio
*bio
)
3408 if (bio
->bi_private
)
3409 complete(bio
->bi_private
);
3414 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3415 * sent down. With wait == 1, it waits for the previous flush.
3417 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3420 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3425 if (device
->nobarriers
)
3429 bio
= device
->flush_bio
;
3433 wait_for_completion(&device
->flush_wait
);
3435 if (bio
->bi_error
) {
3436 ret
= bio
->bi_error
;
3437 btrfs_dev_stat_inc_and_print(device
,
3438 BTRFS_DEV_STAT_FLUSH_ERRS
);
3441 /* drop the reference from the wait == 0 run */
3443 device
->flush_bio
= NULL
;
3449 * one reference for us, and we leave it for the
3452 device
->flush_bio
= NULL
;
3453 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3457 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3458 bio
->bi_bdev
= device
->bdev
;
3459 init_completion(&device
->flush_wait
);
3460 bio
->bi_private
= &device
->flush_wait
;
3461 device
->flush_bio
= bio
;
3464 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3470 * send an empty flush down to each device in parallel,
3471 * then wait for them
3473 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3475 struct list_head
*head
;
3476 struct btrfs_device
*dev
;
3477 int errors_send
= 0;
3478 int errors_wait
= 0;
3481 /* send down all the barriers */
3482 head
= &info
->fs_devices
->devices
;
3483 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3490 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3493 ret
= write_dev_flush(dev
, 0);
3498 /* wait for all the barriers */
3499 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3506 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3509 ret
= write_dev_flush(dev
, 1);
3513 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3514 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3519 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3522 int min_tolerated
= INT_MAX
;
3524 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3525 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3526 min_tolerated
= min(min_tolerated
,
3527 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3528 tolerated_failures
);
3530 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3531 if (raid_type
== BTRFS_RAID_SINGLE
)
3533 if (!(flags
& btrfs_raid_group
[raid_type
]))
3535 min_tolerated
= min(min_tolerated
,
3536 btrfs_raid_array
[raid_type
].
3537 tolerated_failures
);
3540 if (min_tolerated
== INT_MAX
) {
3541 pr_warn("BTRFS: unknown raid flag: %llu\n", flags
);
3545 return min_tolerated
;
3548 int btrfs_calc_num_tolerated_disk_barrier_failures(
3549 struct btrfs_fs_info
*fs_info
)
3551 struct btrfs_ioctl_space_info space
;
3552 struct btrfs_space_info
*sinfo
;
3553 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3554 BTRFS_BLOCK_GROUP_SYSTEM
,
3555 BTRFS_BLOCK_GROUP_METADATA
,
3556 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3559 int num_tolerated_disk_barrier_failures
=
3560 (int)fs_info
->fs_devices
->num_devices
;
3562 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3563 struct btrfs_space_info
*tmp
;
3567 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3568 if (tmp
->flags
== types
[i
]) {
3578 down_read(&sinfo
->groups_sem
);
3579 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3582 if (list_empty(&sinfo
->block_groups
[c
]))
3585 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3587 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3589 flags
= space
.flags
;
3591 num_tolerated_disk_barrier_failures
= min(
3592 num_tolerated_disk_barrier_failures
,
3593 btrfs_get_num_tolerated_disk_barrier_failures(
3596 up_read(&sinfo
->groups_sem
);
3599 return num_tolerated_disk_barrier_failures
;
3602 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3604 struct list_head
*head
;
3605 struct btrfs_device
*dev
;
3606 struct btrfs_super_block
*sb
;
3607 struct btrfs_dev_item
*dev_item
;
3611 int total_errors
= 0;
3614 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3615 backup_super_roots(root
->fs_info
);
3617 sb
= root
->fs_info
->super_for_commit
;
3618 dev_item
= &sb
->dev_item
;
3620 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3621 head
= &root
->fs_info
->fs_devices
->devices
;
3622 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3625 ret
= barrier_all_devices(root
->fs_info
);
3628 &root
->fs_info
->fs_devices
->device_list_mutex
);
3629 btrfs_std_error(root
->fs_info
, ret
,
3630 "errors while submitting device barriers.");
3635 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3640 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3643 btrfs_set_stack_device_generation(dev_item
, 0);
3644 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3645 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3646 btrfs_set_stack_device_total_bytes(dev_item
,
3647 dev
->commit_total_bytes
);
3648 btrfs_set_stack_device_bytes_used(dev_item
,
3649 dev
->commit_bytes_used
);
3650 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3651 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3652 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3653 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3654 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3656 flags
= btrfs_super_flags(sb
);
3657 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3659 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3663 if (total_errors
> max_errors
) {
3664 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3666 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3668 /* FUA is masked off if unsupported and can't be the reason */
3669 btrfs_std_error(root
->fs_info
, -EIO
,
3670 "%d errors while writing supers", total_errors
);
3675 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3678 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3681 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3685 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3686 if (total_errors
> max_errors
) {
3687 btrfs_std_error(root
->fs_info
, -EIO
,
3688 "%d errors while writing supers", total_errors
);
3694 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3695 struct btrfs_root
*root
, int max_mirrors
)
3697 return write_all_supers(root
, max_mirrors
);
3700 /* Drop a fs root from the radix tree and free it. */
3701 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3702 struct btrfs_root
*root
)
3704 spin_lock(&fs_info
->fs_roots_radix_lock
);
3705 radix_tree_delete(&fs_info
->fs_roots_radix
,
3706 (unsigned long)root
->root_key
.objectid
);
3707 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3709 if (btrfs_root_refs(&root
->root_item
) == 0)
3710 synchronize_srcu(&fs_info
->subvol_srcu
);
3712 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3713 btrfs_free_log(NULL
, root
);
3715 if (root
->free_ino_pinned
)
3716 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3717 if (root
->free_ino_ctl
)
3718 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3722 static void free_fs_root(struct btrfs_root
*root
)
3724 iput(root
->ino_cache_inode
);
3725 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3726 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3727 root
->orphan_block_rsv
= NULL
;
3729 free_anon_bdev(root
->anon_dev
);
3730 if (root
->subv_writers
)
3731 btrfs_free_subvolume_writers(root
->subv_writers
);
3732 free_extent_buffer(root
->node
);
3733 free_extent_buffer(root
->commit_root
);
3734 kfree(root
->free_ino_ctl
);
3735 kfree(root
->free_ino_pinned
);
3737 btrfs_put_fs_root(root
);
3740 void btrfs_free_fs_root(struct btrfs_root
*root
)
3745 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3747 u64 root_objectid
= 0;
3748 struct btrfs_root
*gang
[8];
3751 unsigned int ret
= 0;
3755 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3756 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3757 (void **)gang
, root_objectid
,
3760 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3763 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3765 for (i
= 0; i
< ret
; i
++) {
3766 /* Avoid to grab roots in dead_roots */
3767 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3771 /* grab all the search result for later use */
3772 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3774 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3776 for (i
= 0; i
< ret
; i
++) {
3779 root_objectid
= gang
[i
]->root_key
.objectid
;
3780 err
= btrfs_orphan_cleanup(gang
[i
]);
3783 btrfs_put_fs_root(gang
[i
]);
3788 /* release the uncleaned roots due to error */
3789 for (; i
< ret
; i
++) {
3791 btrfs_put_fs_root(gang
[i
]);
3796 int btrfs_commit_super(struct btrfs_root
*root
)
3798 struct btrfs_trans_handle
*trans
;
3800 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3801 btrfs_run_delayed_iputs(root
);
3802 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3803 wake_up_process(root
->fs_info
->cleaner_kthread
);
3805 /* wait until ongoing cleanup work done */
3806 down_write(&root
->fs_info
->cleanup_work_sem
);
3807 up_write(&root
->fs_info
->cleanup_work_sem
);
3809 trans
= btrfs_join_transaction(root
);
3811 return PTR_ERR(trans
);
3812 return btrfs_commit_transaction(trans
, root
);
3815 void close_ctree(struct btrfs_root
*root
)
3817 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3820 fs_info
->closing
= 1;
3823 /* wait for the qgroup rescan worker to stop */
3824 btrfs_qgroup_wait_for_completion(fs_info
);
3826 /* wait for the uuid_scan task to finish */
3827 down(&fs_info
->uuid_tree_rescan_sem
);
3828 /* avoid complains from lockdep et al., set sem back to initial state */
3829 up(&fs_info
->uuid_tree_rescan_sem
);
3831 /* pause restriper - we want to resume on mount */
3832 btrfs_pause_balance(fs_info
);
3834 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3836 btrfs_scrub_cancel(fs_info
);
3838 /* wait for any defraggers to finish */
3839 wait_event(fs_info
->transaction_wait
,
3840 (atomic_read(&fs_info
->defrag_running
) == 0));
3842 /* clear out the rbtree of defraggable inodes */
3843 btrfs_cleanup_defrag_inodes(fs_info
);
3845 cancel_work_sync(&fs_info
->async_reclaim_work
);
3847 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3849 * If the cleaner thread is stopped and there are
3850 * block groups queued for removal, the deletion will be
3851 * skipped when we quit the cleaner thread.
3853 btrfs_delete_unused_bgs(root
->fs_info
);
3855 ret
= btrfs_commit_super(root
);
3857 btrfs_err(fs_info
, "commit super ret %d", ret
);
3860 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3861 btrfs_error_commit_super(root
);
3863 kthread_stop(fs_info
->transaction_kthread
);
3864 kthread_stop(fs_info
->cleaner_kthread
);
3866 fs_info
->closing
= 2;
3869 btrfs_free_qgroup_config(fs_info
);
3871 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3872 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3873 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3876 btrfs_sysfs_remove_mounted(fs_info
);
3877 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3879 btrfs_free_fs_roots(fs_info
);
3881 btrfs_put_block_group_cache(fs_info
);
3883 btrfs_free_block_groups(fs_info
);
3886 * we must make sure there is not any read request to
3887 * submit after we stopping all workers.
3889 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3890 btrfs_stop_all_workers(fs_info
);
3893 free_root_pointers(fs_info
, 1);
3895 iput(fs_info
->btree_inode
);
3897 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3898 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3899 btrfsic_unmount(root
, fs_info
->fs_devices
);
3902 btrfs_close_devices(fs_info
->fs_devices
);
3903 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3905 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3906 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3907 percpu_counter_destroy(&fs_info
->bio_counter
);
3908 bdi_destroy(&fs_info
->bdi
);
3909 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3911 btrfs_free_stripe_hash_table(fs_info
);
3913 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
3914 root
->orphan_block_rsv
= NULL
;
3917 while (!list_empty(&fs_info
->pinned_chunks
)) {
3918 struct extent_map
*em
;
3920 em
= list_first_entry(&fs_info
->pinned_chunks
,
3921 struct extent_map
, list
);
3922 list_del_init(&em
->list
);
3923 free_extent_map(em
);
3925 unlock_chunks(root
);
3928 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3932 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3934 ret
= extent_buffer_uptodate(buf
);
3938 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3939 parent_transid
, atomic
);
3945 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3947 struct btrfs_root
*root
;
3948 u64 transid
= btrfs_header_generation(buf
);
3951 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3953 * This is a fast path so only do this check if we have sanity tests
3954 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3955 * outside of the sanity tests.
3957 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3960 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3961 btrfs_assert_tree_locked(buf
);
3962 if (transid
!= root
->fs_info
->generation
)
3963 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3964 "found %llu running %llu\n",
3965 buf
->start
, transid
, root
->fs_info
->generation
);
3966 was_dirty
= set_extent_buffer_dirty(buf
);
3968 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3970 root
->fs_info
->dirty_metadata_batch
);
3971 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3972 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3973 btrfs_print_leaf(root
, buf
);
3979 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3983 * looks as though older kernels can get into trouble with
3984 * this code, they end up stuck in balance_dirty_pages forever
3988 if (current
->flags
& PF_MEMALLOC
)
3992 btrfs_balance_delayed_items(root
);
3994 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3995 BTRFS_DIRTY_METADATA_THRESH
);
3997 balance_dirty_pages_ratelimited(
3998 root
->fs_info
->btree_inode
->i_mapping
);
4002 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
4004 __btrfs_btree_balance_dirty(root
, 1);
4007 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
4009 __btrfs_btree_balance_dirty(root
, 0);
4012 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
4014 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4015 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
4018 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
4021 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
4024 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4025 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
4026 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
4029 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4030 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
4031 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
4034 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4035 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
4036 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
4041 * The common minimum, we don't know if we can trust the nodesize/sectorsize
4042 * items yet, they'll be verified later. Issue just a warning.
4044 if (!IS_ALIGNED(btrfs_super_root(sb
), 4096))
4045 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
4046 btrfs_super_root(sb
));
4047 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), 4096))
4048 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
4049 btrfs_super_chunk_root(sb
));
4050 if (!IS_ALIGNED(btrfs_super_log_root(sb
), 4096))
4051 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
4052 btrfs_super_log_root(sb
));
4055 * Check the lower bound, the alignment and other constraints are
4058 if (btrfs_super_nodesize(sb
) < 4096) {
4059 printk(KERN_ERR
"BTRFS: nodesize too small: %u < 4096\n",
4060 btrfs_super_nodesize(sb
));
4063 if (btrfs_super_sectorsize(sb
) < 4096) {
4064 printk(KERN_ERR
"BTRFS: sectorsize too small: %u < 4096\n",
4065 btrfs_super_sectorsize(sb
));
4069 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4070 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4071 fs_info
->fsid
, sb
->dev_item
.fsid
);
4076 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4079 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4080 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
4081 btrfs_super_num_devices(sb
));
4082 if (btrfs_super_num_devices(sb
) == 0) {
4083 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
4087 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4088 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
4089 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4094 * Obvious sys_chunk_array corruptions, it must hold at least one key
4097 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4098 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
4099 btrfs_super_sys_array_size(sb
),
4100 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4103 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4104 + sizeof(struct btrfs_chunk
)) {
4105 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
4106 btrfs_super_sys_array_size(sb
),
4107 sizeof(struct btrfs_disk_key
)
4108 + sizeof(struct btrfs_chunk
));
4113 * The generation is a global counter, we'll trust it more than the others
4114 * but it's still possible that it's the one that's wrong.
4116 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4118 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4119 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
4120 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4121 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4123 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4124 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
4129 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4131 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4132 btrfs_run_delayed_iputs(root
);
4133 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4135 down_write(&root
->fs_info
->cleanup_work_sem
);
4136 up_write(&root
->fs_info
->cleanup_work_sem
);
4138 /* cleanup FS via transaction */
4139 btrfs_cleanup_transaction(root
);
4142 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4144 struct btrfs_ordered_extent
*ordered
;
4146 spin_lock(&root
->ordered_extent_lock
);
4148 * This will just short circuit the ordered completion stuff which will
4149 * make sure the ordered extent gets properly cleaned up.
4151 list_for_each_entry(ordered
, &root
->ordered_extents
,
4153 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4154 spin_unlock(&root
->ordered_extent_lock
);
4157 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4159 struct btrfs_root
*root
;
4160 struct list_head splice
;
4162 INIT_LIST_HEAD(&splice
);
4164 spin_lock(&fs_info
->ordered_root_lock
);
4165 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4166 while (!list_empty(&splice
)) {
4167 root
= list_first_entry(&splice
, struct btrfs_root
,
4169 list_move_tail(&root
->ordered_root
,
4170 &fs_info
->ordered_roots
);
4172 spin_unlock(&fs_info
->ordered_root_lock
);
4173 btrfs_destroy_ordered_extents(root
);
4176 spin_lock(&fs_info
->ordered_root_lock
);
4178 spin_unlock(&fs_info
->ordered_root_lock
);
4181 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4182 struct btrfs_root
*root
)
4184 struct rb_node
*node
;
4185 struct btrfs_delayed_ref_root
*delayed_refs
;
4186 struct btrfs_delayed_ref_node
*ref
;
4189 delayed_refs
= &trans
->delayed_refs
;
4191 spin_lock(&delayed_refs
->lock
);
4192 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4193 spin_unlock(&delayed_refs
->lock
);
4194 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4198 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4199 struct btrfs_delayed_ref_head
*head
;
4200 struct btrfs_delayed_ref_node
*tmp
;
4201 bool pin_bytes
= false;
4203 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4205 if (!mutex_trylock(&head
->mutex
)) {
4206 atomic_inc(&head
->node
.refs
);
4207 spin_unlock(&delayed_refs
->lock
);
4209 mutex_lock(&head
->mutex
);
4210 mutex_unlock(&head
->mutex
);
4211 btrfs_put_delayed_ref(&head
->node
);
4212 spin_lock(&delayed_refs
->lock
);
4215 spin_lock(&head
->lock
);
4216 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4219 list_del(&ref
->list
);
4220 atomic_dec(&delayed_refs
->num_entries
);
4221 btrfs_put_delayed_ref(ref
);
4223 if (head
->must_insert_reserved
)
4225 btrfs_free_delayed_extent_op(head
->extent_op
);
4226 delayed_refs
->num_heads
--;
4227 if (head
->processing
== 0)
4228 delayed_refs
->num_heads_ready
--;
4229 atomic_dec(&delayed_refs
->num_entries
);
4230 head
->node
.in_tree
= 0;
4231 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4232 spin_unlock(&head
->lock
);
4233 spin_unlock(&delayed_refs
->lock
);
4234 mutex_unlock(&head
->mutex
);
4237 btrfs_pin_extent(root
, head
->node
.bytenr
,
4238 head
->node
.num_bytes
, 1);
4239 btrfs_put_delayed_ref(&head
->node
);
4241 spin_lock(&delayed_refs
->lock
);
4244 spin_unlock(&delayed_refs
->lock
);
4249 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4251 struct btrfs_inode
*btrfs_inode
;
4252 struct list_head splice
;
4254 INIT_LIST_HEAD(&splice
);
4256 spin_lock(&root
->delalloc_lock
);
4257 list_splice_init(&root
->delalloc_inodes
, &splice
);
4259 while (!list_empty(&splice
)) {
4260 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4263 list_del_init(&btrfs_inode
->delalloc_inodes
);
4264 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4265 &btrfs_inode
->runtime_flags
);
4266 spin_unlock(&root
->delalloc_lock
);
4268 btrfs_invalidate_inodes(btrfs_inode
->root
);
4270 spin_lock(&root
->delalloc_lock
);
4273 spin_unlock(&root
->delalloc_lock
);
4276 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4278 struct btrfs_root
*root
;
4279 struct list_head splice
;
4281 INIT_LIST_HEAD(&splice
);
4283 spin_lock(&fs_info
->delalloc_root_lock
);
4284 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4285 while (!list_empty(&splice
)) {
4286 root
= list_first_entry(&splice
, struct btrfs_root
,
4288 list_del_init(&root
->delalloc_root
);
4289 root
= btrfs_grab_fs_root(root
);
4291 spin_unlock(&fs_info
->delalloc_root_lock
);
4293 btrfs_destroy_delalloc_inodes(root
);
4294 btrfs_put_fs_root(root
);
4296 spin_lock(&fs_info
->delalloc_root_lock
);
4298 spin_unlock(&fs_info
->delalloc_root_lock
);
4301 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4302 struct extent_io_tree
*dirty_pages
,
4306 struct extent_buffer
*eb
;
4311 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4316 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4317 while (start
<= end
) {
4318 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4319 start
+= root
->nodesize
;
4322 wait_on_extent_buffer_writeback(eb
);
4324 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4326 clear_extent_buffer_dirty(eb
);
4327 free_extent_buffer_stale(eb
);
4334 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4335 struct extent_io_tree
*pinned_extents
)
4337 struct extent_io_tree
*unpin
;
4343 unpin
= pinned_extents
;
4346 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4347 EXTENT_DIRTY
, NULL
);
4351 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4352 btrfs_error_unpin_extent_range(root
, start
, end
);
4357 if (unpin
== &root
->fs_info
->freed_extents
[0])
4358 unpin
= &root
->fs_info
->freed_extents
[1];
4360 unpin
= &root
->fs_info
->freed_extents
[0];
4368 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4369 struct btrfs_root
*root
)
4371 btrfs_destroy_delayed_refs(cur_trans
, root
);
4373 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4374 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4376 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4377 wake_up(&root
->fs_info
->transaction_wait
);
4379 btrfs_destroy_delayed_inodes(root
);
4380 btrfs_assert_delayed_root_empty(root
);
4382 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4384 btrfs_destroy_pinned_extent(root
,
4385 root
->fs_info
->pinned_extents
);
4387 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4388 wake_up(&cur_trans
->commit_wait
);
4391 memset(cur_trans, 0, sizeof(*cur_trans));
4392 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4396 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4398 struct btrfs_transaction
*t
;
4400 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4402 spin_lock(&root
->fs_info
->trans_lock
);
4403 while (!list_empty(&root
->fs_info
->trans_list
)) {
4404 t
= list_first_entry(&root
->fs_info
->trans_list
,
4405 struct btrfs_transaction
, list
);
4406 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4407 atomic_inc(&t
->use_count
);
4408 spin_unlock(&root
->fs_info
->trans_lock
);
4409 btrfs_wait_for_commit(root
, t
->transid
);
4410 btrfs_put_transaction(t
);
4411 spin_lock(&root
->fs_info
->trans_lock
);
4414 if (t
== root
->fs_info
->running_transaction
) {
4415 t
->state
= TRANS_STATE_COMMIT_DOING
;
4416 spin_unlock(&root
->fs_info
->trans_lock
);
4418 * We wait for 0 num_writers since we don't hold a trans
4419 * handle open currently for this transaction.
4421 wait_event(t
->writer_wait
,
4422 atomic_read(&t
->num_writers
) == 0);
4424 spin_unlock(&root
->fs_info
->trans_lock
);
4426 btrfs_cleanup_one_transaction(t
, root
);
4428 spin_lock(&root
->fs_info
->trans_lock
);
4429 if (t
== root
->fs_info
->running_transaction
)
4430 root
->fs_info
->running_transaction
= NULL
;
4431 list_del_init(&t
->list
);
4432 spin_unlock(&root
->fs_info
->trans_lock
);
4434 btrfs_put_transaction(t
);
4435 trace_btrfs_transaction_commit(root
);
4436 spin_lock(&root
->fs_info
->trans_lock
);
4438 spin_unlock(&root
->fs_info
->trans_lock
);
4439 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4440 btrfs_destroy_delayed_inodes(root
);
4441 btrfs_assert_delayed_root_empty(root
);
4442 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4443 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4444 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4449 static const struct extent_io_ops btree_extent_io_ops
= {
4450 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4451 .readpage_io_failed_hook
= btree_io_failed_hook
,
4452 .submit_bio_hook
= btree_submit_bio_hook
,
4453 /* note we're sharing with inode.c for the merge bio hook */
4454 .merge_bio_hook
= btrfs_merge_bio_hook
,