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 "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static const struct extent_io_ops btree_extent_io_ops
;
58 static void end_workqueue_fn(struct btrfs_work
*work
);
59 static void free_fs_root(struct btrfs_root
*root
);
60 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
62 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
64 struct btrfs_root
*root
);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
66 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
67 struct extent_io_tree
*dirty_pages
,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
70 struct extent_io_tree
*pinned_extents
);
71 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
72 static void btrfs_error_commit_super(struct btrfs_root
*root
);
75 * btrfs_end_io_wq structs are used to do processing in task context when an IO
76 * is complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
79 struct btrfs_end_io_wq
{
83 struct btrfs_fs_info
*info
;
85 enum btrfs_wq_endio_type metadata
;
86 struct list_head list
;
87 struct btrfs_work work
;
90 static struct kmem_cache
*btrfs_end_io_wq_cache
;
92 int __init
btrfs_end_io_wq_init(void)
94 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
95 sizeof(struct btrfs_end_io_wq
),
97 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
99 if (!btrfs_end_io_wq_cache
)
104 void btrfs_end_io_wq_exit(void)
106 if (btrfs_end_io_wq_cache
)
107 kmem_cache_destroy(btrfs_end_io_wq_cache
);
111 * async submit bios are used to offload expensive checksumming
112 * onto the worker threads. They checksum file and metadata bios
113 * just before they are sent down the IO stack.
115 struct async_submit_bio
{
118 struct list_head list
;
119 extent_submit_bio_hook_t
*submit_bio_start
;
120 extent_submit_bio_hook_t
*submit_bio_done
;
123 unsigned long bio_flags
;
125 * bio_offset is optional, can be used if the pages in the bio
126 * can't tell us where in the file the bio should go
129 struct btrfs_work work
;
134 * Lockdep class keys for extent_buffer->lock's in this root. For a given
135 * eb, the lockdep key is determined by the btrfs_root it belongs to and
136 * the level the eb occupies in the tree.
138 * Different roots are used for different purposes and may nest inside each
139 * other and they require separate keysets. As lockdep keys should be
140 * static, assign keysets according to the purpose of the root as indicated
141 * by btrfs_root->objectid. This ensures that all special purpose roots
142 * have separate keysets.
144 * Lock-nesting across peer nodes is always done with the immediate parent
145 * node locked thus preventing deadlock. As lockdep doesn't know this, use
146 * subclass to avoid triggering lockdep warning in such cases.
148 * The key is set by the readpage_end_io_hook after the buffer has passed
149 * csum validation but before the pages are unlocked. It is also set by
150 * btrfs_init_new_buffer on freshly allocated blocks.
152 * We also add a check to make sure the highest level of the tree is the
153 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
154 * needs update as well.
156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
157 # if BTRFS_MAX_LEVEL != 8
161 static struct btrfs_lockdep_keyset
{
162 u64 id
; /* root objectid */
163 const char *name_stem
; /* lock name stem */
164 char names
[BTRFS_MAX_LEVEL
+ 1][20];
165 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
166 } btrfs_lockdep_keysets
[] = {
167 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
168 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
169 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
170 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
171 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
172 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
173 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
174 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
175 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
176 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
177 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
178 { .id
= 0, .name_stem
= "tree" },
181 void __init
btrfs_init_lockdep(void)
185 /* initialize lockdep class names */
186 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
187 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
189 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
190 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
191 "btrfs-%s-%02d", ks
->name_stem
, j
);
195 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
198 struct btrfs_lockdep_keyset
*ks
;
200 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
202 /* find the matching keyset, id 0 is the default entry */
203 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
204 if (ks
->id
== objectid
)
207 lockdep_set_class_and_name(&eb
->lock
,
208 &ks
->keys
[level
], ks
->names
[level
]);
214 * extents on the btree inode are pretty simple, there's one extent
215 * that covers the entire device
217 static struct extent_map
*btree_get_extent(struct inode
*inode
,
218 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
221 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
222 struct extent_map
*em
;
225 read_lock(&em_tree
->lock
);
226 em
= lookup_extent_mapping(em_tree
, start
, len
);
229 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
230 read_unlock(&em_tree
->lock
);
233 read_unlock(&em_tree
->lock
);
235 em
= alloc_extent_map();
237 em
= ERR_PTR(-ENOMEM
);
242 em
->block_len
= (u64
)-1;
244 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
246 write_lock(&em_tree
->lock
);
247 ret
= add_extent_mapping(em_tree
, em
, 0);
248 if (ret
== -EEXIST
) {
250 em
= lookup_extent_mapping(em_tree
, start
, len
);
257 write_unlock(&em_tree
->lock
);
263 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
265 return btrfs_crc32c(seed
, data
, len
);
268 void btrfs_csum_final(u32 crc
, char *result
)
270 put_unaligned_le32(~crc
, result
);
274 * compute the csum for a btree block, and either verify it or write it
275 * into the csum field of the block.
277 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
278 struct extent_buffer
*buf
,
281 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
284 unsigned long cur_len
;
285 unsigned long offset
= BTRFS_CSUM_SIZE
;
287 unsigned long map_start
;
288 unsigned long map_len
;
291 unsigned long inline_result
;
293 len
= buf
->len
- offset
;
295 err
= map_private_extent_buffer(buf
, offset
, 32,
296 &kaddr
, &map_start
, &map_len
);
299 cur_len
= min(len
, map_len
- (offset
- map_start
));
300 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
305 if (csum_size
> sizeof(inline_result
)) {
306 result
= kzalloc(csum_size
, GFP_NOFS
);
310 result
= (char *)&inline_result
;
313 btrfs_csum_final(crc
, result
);
316 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
319 memcpy(&found
, result
, csum_size
);
321 read_extent_buffer(buf
, &val
, 0, csum_size
);
322 printk_ratelimited(KERN_WARNING
323 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
325 fs_info
->sb
->s_id
, buf
->start
,
326 val
, found
, btrfs_header_level(buf
));
327 if (result
!= (char *)&inline_result
)
332 write_extent_buffer(buf
, result
, 0, csum_size
);
334 if (result
!= (char *)&inline_result
)
340 * we can't consider a given block up to date unless the transid of the
341 * block matches the transid in the parent node's pointer. This is how we
342 * detect blocks that either didn't get written at all or got written
343 * in the wrong place.
345 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
346 struct extent_buffer
*eb
, u64 parent_transid
,
349 struct extent_state
*cached_state
= NULL
;
351 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
353 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
360 btrfs_tree_read_lock(eb
);
361 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
364 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
366 if (extent_buffer_uptodate(eb
) &&
367 btrfs_header_generation(eb
) == parent_transid
) {
371 printk_ratelimited(KERN_ERR
372 "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
373 eb
->fs_info
->sb
->s_id
, eb
->start
,
374 parent_transid
, btrfs_header_generation(eb
));
378 * Things reading via commit roots that don't have normal protection,
379 * like send, can have a really old block in cache that may point at a
380 * block that has been free'd and re-allocated. So don't clear uptodate
381 * if we find an eb that is under IO (dirty/writeback) because we could
382 * end up reading in the stale data and then writing it back out and
383 * making everybody very sad.
385 if (!extent_buffer_under_io(eb
))
386 clear_extent_buffer_uptodate(eb
);
388 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
389 &cached_state
, GFP_NOFS
);
391 btrfs_tree_read_unlock_blocking(eb
);
396 * Return 0 if the superblock checksum type matches the checksum value of that
397 * algorithm. Pass the raw disk superblock data.
399 static int btrfs_check_super_csum(char *raw_disk_sb
)
401 struct btrfs_super_block
*disk_sb
=
402 (struct btrfs_super_block
*)raw_disk_sb
;
403 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
406 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
408 const int csum_size
= sizeof(crc
);
409 char result
[csum_size
];
412 * The super_block structure does not span the whole
413 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 * is filled with zeros and is included in the checkum.
416 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
417 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
418 btrfs_csum_final(crc
, result
);
420 if (memcmp(raw_disk_sb
, result
, csum_size
))
423 if (ret
&& btrfs_super_generation(disk_sb
) < 10) {
425 "BTRFS: super block crcs don't match, older mkfs detected\n");
430 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
431 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
443 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
444 struct extent_buffer
*eb
,
445 u64 start
, u64 parent_transid
)
447 struct extent_io_tree
*io_tree
;
452 int failed_mirror
= 0;
454 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
455 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
457 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
459 btree_get_extent
, mirror_num
);
461 if (!verify_parent_transid(io_tree
, eb
,
469 * This buffer's crc is fine, but its contents are corrupted, so
470 * there is no reason to read the other copies, they won't be
473 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
476 num_copies
= btrfs_num_copies(root
->fs_info
,
481 if (!failed_mirror
) {
483 failed_mirror
= eb
->read_mirror
;
487 if (mirror_num
== failed_mirror
)
490 if (mirror_num
> num_copies
)
494 if (failed
&& !ret
&& failed_mirror
)
495 repair_eb_io_failure(root
, eb
, failed_mirror
);
501 * checksum a dirty tree block before IO. This has extra checks to make sure
502 * we only fill in the checksum field in the first page of a multi-page block
505 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
507 u64 start
= page_offset(page
);
509 struct extent_buffer
*eb
;
511 eb
= (struct extent_buffer
*)page
->private;
512 if (page
!= eb
->pages
[0])
514 found_start
= btrfs_header_bytenr(eb
);
515 if (WARN_ON(found_start
!= start
|| !PageUptodate(page
)))
517 csum_tree_block(fs_info
, eb
, 0);
521 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
522 struct extent_buffer
*eb
)
524 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
525 u8 fsid
[BTRFS_UUID_SIZE
];
528 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
530 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
534 fs_devices
= fs_devices
->seed
;
539 #define CORRUPT(reason, eb, root, slot) \
540 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
541 "root=%llu, slot=%d", reason, \
542 btrfs_header_bytenr(eb), root->objectid, slot)
544 static noinline
int check_leaf(struct btrfs_root
*root
,
545 struct extent_buffer
*leaf
)
547 struct btrfs_key key
;
548 struct btrfs_key leaf_key
;
549 u32 nritems
= btrfs_header_nritems(leaf
);
555 /* Check the 0 item */
556 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
557 BTRFS_LEAF_DATA_SIZE(root
)) {
558 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
563 * Check to make sure each items keys are in the correct order and their
564 * offsets make sense. We only have to loop through nritems-1 because
565 * we check the current slot against the next slot, which verifies the
566 * next slot's offset+size makes sense and that the current's slot
569 for (slot
= 0; slot
< nritems
- 1; slot
++) {
570 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
571 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
573 /* Make sure the keys are in the right order */
574 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
575 CORRUPT("bad key order", leaf
, root
, slot
);
580 * Make sure the offset and ends are right, remember that the
581 * item data starts at the end of the leaf and grows towards the
584 if (btrfs_item_offset_nr(leaf
, slot
) !=
585 btrfs_item_end_nr(leaf
, slot
+ 1)) {
586 CORRUPT("slot offset bad", leaf
, root
, slot
);
591 * Check to make sure that we don't point outside of the leaf,
592 * just incase all the items are consistent to eachother, but
593 * all point outside of the leaf.
595 if (btrfs_item_end_nr(leaf
, slot
) >
596 BTRFS_LEAF_DATA_SIZE(root
)) {
597 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
605 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
606 u64 phy_offset
, struct page
*page
,
607 u64 start
, u64 end
, int mirror
)
611 struct extent_buffer
*eb
;
612 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
619 eb
= (struct extent_buffer
*)page
->private;
621 /* the pending IO might have been the only thing that kept this buffer
622 * in memory. Make sure we have a ref for all this other checks
624 extent_buffer_get(eb
);
626 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
630 eb
->read_mirror
= mirror
;
631 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
636 found_start
= btrfs_header_bytenr(eb
);
637 if (found_start
!= eb
->start
) {
638 printk_ratelimited(KERN_ERR
"BTRFS (device %s): bad tree block start "
640 eb
->fs_info
->sb
->s_id
, found_start
, eb
->start
);
644 if (check_tree_block_fsid(root
->fs_info
, eb
)) {
645 printk_ratelimited(KERN_ERR
"BTRFS (device %s): bad fsid on block %llu\n",
646 eb
->fs_info
->sb
->s_id
, eb
->start
);
650 found_level
= btrfs_header_level(eb
);
651 if (found_level
>= BTRFS_MAX_LEVEL
) {
652 btrfs_err(root
->fs_info
, "bad tree block level %d",
653 (int)btrfs_header_level(eb
));
658 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
661 ret
= csum_tree_block(root
->fs_info
, eb
, 1);
668 * If this is a leaf block and it is corrupt, set the corrupt bit so
669 * that we don't try and read the other copies of this block, just
672 if (found_level
== 0 && check_leaf(root
, eb
)) {
673 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
678 set_extent_buffer_uptodate(eb
);
681 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
682 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
686 * our io error hook is going to dec the io pages
687 * again, we have to make sure it has something
690 atomic_inc(&eb
->io_pages
);
691 clear_extent_buffer_uptodate(eb
);
693 free_extent_buffer(eb
);
698 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
700 struct extent_buffer
*eb
;
701 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
703 eb
= (struct extent_buffer
*)page
->private;
704 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
705 eb
->read_mirror
= failed_mirror
;
706 atomic_dec(&eb
->io_pages
);
707 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
708 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
709 return -EIO
; /* we fixed nothing */
712 static void end_workqueue_bio(struct bio
*bio
, int err
)
714 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
715 struct btrfs_fs_info
*fs_info
;
716 struct btrfs_workqueue
*wq
;
717 btrfs_work_func_t func
;
719 fs_info
= end_io_wq
->info
;
720 end_io_wq
->error
= err
;
722 if (bio
->bi_rw
& REQ_WRITE
) {
723 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
724 wq
= fs_info
->endio_meta_write_workers
;
725 func
= btrfs_endio_meta_write_helper
;
726 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
727 wq
= fs_info
->endio_freespace_worker
;
728 func
= btrfs_freespace_write_helper
;
729 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
730 wq
= fs_info
->endio_raid56_workers
;
731 func
= btrfs_endio_raid56_helper
;
733 wq
= fs_info
->endio_write_workers
;
734 func
= btrfs_endio_write_helper
;
737 if (unlikely(end_io_wq
->metadata
==
738 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
739 wq
= fs_info
->endio_repair_workers
;
740 func
= btrfs_endio_repair_helper
;
741 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
742 wq
= fs_info
->endio_raid56_workers
;
743 func
= btrfs_endio_raid56_helper
;
744 } else if (end_io_wq
->metadata
) {
745 wq
= fs_info
->endio_meta_workers
;
746 func
= btrfs_endio_meta_helper
;
748 wq
= fs_info
->endio_workers
;
749 func
= btrfs_endio_helper
;
753 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
754 btrfs_queue_work(wq
, &end_io_wq
->work
);
757 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
758 enum btrfs_wq_endio_type metadata
)
760 struct btrfs_end_io_wq
*end_io_wq
;
762 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
766 end_io_wq
->private = bio
->bi_private
;
767 end_io_wq
->end_io
= bio
->bi_end_io
;
768 end_io_wq
->info
= info
;
769 end_io_wq
->error
= 0;
770 end_io_wq
->bio
= bio
;
771 end_io_wq
->metadata
= metadata
;
773 bio
->bi_private
= end_io_wq
;
774 bio
->bi_end_io
= end_workqueue_bio
;
778 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
780 unsigned long limit
= min_t(unsigned long,
781 info
->thread_pool_size
,
782 info
->fs_devices
->open_devices
);
786 static void run_one_async_start(struct btrfs_work
*work
)
788 struct async_submit_bio
*async
;
791 async
= container_of(work
, struct async_submit_bio
, work
);
792 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
793 async
->mirror_num
, async
->bio_flags
,
799 static void run_one_async_done(struct btrfs_work
*work
)
801 struct btrfs_fs_info
*fs_info
;
802 struct async_submit_bio
*async
;
805 async
= container_of(work
, struct async_submit_bio
, work
);
806 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
808 limit
= btrfs_async_submit_limit(fs_info
);
809 limit
= limit
* 2 / 3;
811 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
812 waitqueue_active(&fs_info
->async_submit_wait
))
813 wake_up(&fs_info
->async_submit_wait
);
815 /* If an error occured we just want to clean up the bio and move on */
817 bio_endio(async
->bio
, async
->error
);
821 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
822 async
->mirror_num
, async
->bio_flags
,
826 static void run_one_async_free(struct btrfs_work
*work
)
828 struct async_submit_bio
*async
;
830 async
= container_of(work
, struct async_submit_bio
, work
);
834 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
835 int rw
, struct bio
*bio
, int mirror_num
,
836 unsigned long bio_flags
,
838 extent_submit_bio_hook_t
*submit_bio_start
,
839 extent_submit_bio_hook_t
*submit_bio_done
)
841 struct async_submit_bio
*async
;
843 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
847 async
->inode
= inode
;
850 async
->mirror_num
= mirror_num
;
851 async
->submit_bio_start
= submit_bio_start
;
852 async
->submit_bio_done
= submit_bio_done
;
854 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
855 run_one_async_done
, run_one_async_free
);
857 async
->bio_flags
= bio_flags
;
858 async
->bio_offset
= bio_offset
;
862 atomic_inc(&fs_info
->nr_async_submits
);
865 btrfs_set_work_high_priority(&async
->work
);
867 btrfs_queue_work(fs_info
->workers
, &async
->work
);
869 while (atomic_read(&fs_info
->async_submit_draining
) &&
870 atomic_read(&fs_info
->nr_async_submits
)) {
871 wait_event(fs_info
->async_submit_wait
,
872 (atomic_read(&fs_info
->nr_async_submits
) == 0));
878 static int btree_csum_one_bio(struct bio
*bio
)
880 struct bio_vec
*bvec
;
881 struct btrfs_root
*root
;
884 bio_for_each_segment_all(bvec
, bio
, i
) {
885 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
886 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
894 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
895 struct bio
*bio
, int mirror_num
,
896 unsigned long bio_flags
,
900 * when we're called for a write, we're already in the async
901 * submission context. Just jump into btrfs_map_bio
903 return btree_csum_one_bio(bio
);
906 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
907 int mirror_num
, unsigned long bio_flags
,
913 * when we're called for a write, we're already in the async
914 * submission context. Just jump into btrfs_map_bio
916 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
922 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
924 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
933 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
934 int mirror_num
, unsigned long bio_flags
,
937 int async
= check_async_write(inode
, bio_flags
);
940 if (!(rw
& REQ_WRITE
)) {
942 * called for a read, do the setup so that checksum validation
943 * can happen in the async kernel threads
945 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
946 bio
, BTRFS_WQ_ENDIO_METADATA
);
949 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
952 ret
= btree_csum_one_bio(bio
);
955 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
959 * kthread helpers are used to submit writes so that
960 * checksumming can happen in parallel across all CPUs
962 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
963 inode
, rw
, bio
, mirror_num
, 0,
965 __btree_submit_bio_start
,
966 __btree_submit_bio_done
);
976 #ifdef CONFIG_MIGRATION
977 static int btree_migratepage(struct address_space
*mapping
,
978 struct page
*newpage
, struct page
*page
,
979 enum migrate_mode mode
)
982 * we can't safely write a btree page from here,
983 * we haven't done the locking hook
988 * Buffers may be managed in a filesystem specific way.
989 * We must have no buffers or drop them.
991 if (page_has_private(page
) &&
992 !try_to_release_page(page
, GFP_KERNEL
))
994 return migrate_page(mapping
, newpage
, page
, mode
);
999 static int btree_writepages(struct address_space
*mapping
,
1000 struct writeback_control
*wbc
)
1002 struct btrfs_fs_info
*fs_info
;
1005 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1007 if (wbc
->for_kupdate
)
1010 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1011 /* this is a bit racy, but that's ok */
1012 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1013 BTRFS_DIRTY_METADATA_THRESH
);
1017 return btree_write_cache_pages(mapping
, wbc
);
1020 static int btree_readpage(struct file
*file
, struct page
*page
)
1022 struct extent_io_tree
*tree
;
1023 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1024 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1027 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1029 if (PageWriteback(page
) || PageDirty(page
))
1032 return try_release_extent_buffer(page
);
1035 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1036 unsigned int length
)
1038 struct extent_io_tree
*tree
;
1039 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1040 extent_invalidatepage(tree
, page
, offset
);
1041 btree_releasepage(page
, GFP_NOFS
);
1042 if (PagePrivate(page
)) {
1043 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1044 "page private not zero on page %llu",
1045 (unsigned long long)page_offset(page
));
1046 ClearPagePrivate(page
);
1047 set_page_private(page
, 0);
1048 page_cache_release(page
);
1052 static int btree_set_page_dirty(struct page
*page
)
1055 struct extent_buffer
*eb
;
1057 BUG_ON(!PagePrivate(page
));
1058 eb
= (struct extent_buffer
*)page
->private;
1060 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1061 BUG_ON(!atomic_read(&eb
->refs
));
1062 btrfs_assert_tree_locked(eb
);
1064 return __set_page_dirty_nobuffers(page
);
1067 static const struct address_space_operations btree_aops
= {
1068 .readpage
= btree_readpage
,
1069 .writepages
= btree_writepages
,
1070 .releasepage
= btree_releasepage
,
1071 .invalidatepage
= btree_invalidatepage
,
1072 #ifdef CONFIG_MIGRATION
1073 .migratepage
= btree_migratepage
,
1075 .set_page_dirty
= btree_set_page_dirty
,
1078 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1080 struct extent_buffer
*buf
= NULL
;
1081 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1083 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1086 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1087 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1088 free_extent_buffer(buf
);
1091 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1092 int mirror_num
, struct extent_buffer
**eb
)
1094 struct extent_buffer
*buf
= NULL
;
1095 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1096 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1099 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1103 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1105 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1106 btree_get_extent
, mirror_num
);
1108 free_extent_buffer(buf
);
1112 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1113 free_extent_buffer(buf
);
1115 } else if (extent_buffer_uptodate(buf
)) {
1118 free_extent_buffer(buf
);
1123 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1126 return find_extent_buffer(fs_info
, bytenr
);
1129 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1132 if (btrfs_test_is_dummy_root(root
))
1133 return alloc_test_extent_buffer(root
->fs_info
, bytenr
);
1134 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1138 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1140 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1141 buf
->start
+ buf
->len
- 1);
1144 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1146 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1147 buf
->start
, buf
->start
+ buf
->len
- 1);
1150 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1153 struct extent_buffer
*buf
= NULL
;
1156 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1160 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1162 free_extent_buffer(buf
);
1169 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1170 struct btrfs_fs_info
*fs_info
,
1171 struct extent_buffer
*buf
)
1173 if (btrfs_header_generation(buf
) ==
1174 fs_info
->running_transaction
->transid
) {
1175 btrfs_assert_tree_locked(buf
);
1177 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1178 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1180 fs_info
->dirty_metadata_batch
);
1181 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1182 btrfs_set_lock_blocking(buf
);
1183 clear_extent_buffer_dirty(buf
);
1188 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1190 struct btrfs_subvolume_writers
*writers
;
1193 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1195 return ERR_PTR(-ENOMEM
);
1197 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1200 return ERR_PTR(ret
);
1203 init_waitqueue_head(&writers
->wait
);
1208 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1210 percpu_counter_destroy(&writers
->counter
);
1214 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1215 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1219 root
->commit_root
= NULL
;
1220 root
->sectorsize
= sectorsize
;
1221 root
->nodesize
= nodesize
;
1222 root
->stripesize
= stripesize
;
1224 root
->orphan_cleanup_state
= 0;
1226 root
->objectid
= objectid
;
1227 root
->last_trans
= 0;
1228 root
->highest_objectid
= 0;
1229 root
->nr_delalloc_inodes
= 0;
1230 root
->nr_ordered_extents
= 0;
1232 root
->inode_tree
= RB_ROOT
;
1233 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1234 root
->block_rsv
= NULL
;
1235 root
->orphan_block_rsv
= NULL
;
1237 INIT_LIST_HEAD(&root
->dirty_list
);
1238 INIT_LIST_HEAD(&root
->root_list
);
1239 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1240 INIT_LIST_HEAD(&root
->delalloc_root
);
1241 INIT_LIST_HEAD(&root
->ordered_extents
);
1242 INIT_LIST_HEAD(&root
->ordered_root
);
1243 INIT_LIST_HEAD(&root
->logged_list
[0]);
1244 INIT_LIST_HEAD(&root
->logged_list
[1]);
1245 spin_lock_init(&root
->orphan_lock
);
1246 spin_lock_init(&root
->inode_lock
);
1247 spin_lock_init(&root
->delalloc_lock
);
1248 spin_lock_init(&root
->ordered_extent_lock
);
1249 spin_lock_init(&root
->accounting_lock
);
1250 spin_lock_init(&root
->log_extents_lock
[0]);
1251 spin_lock_init(&root
->log_extents_lock
[1]);
1252 mutex_init(&root
->objectid_mutex
);
1253 mutex_init(&root
->log_mutex
);
1254 mutex_init(&root
->ordered_extent_mutex
);
1255 mutex_init(&root
->delalloc_mutex
);
1256 init_waitqueue_head(&root
->log_writer_wait
);
1257 init_waitqueue_head(&root
->log_commit_wait
[0]);
1258 init_waitqueue_head(&root
->log_commit_wait
[1]);
1259 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1260 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1261 atomic_set(&root
->log_commit
[0], 0);
1262 atomic_set(&root
->log_commit
[1], 0);
1263 atomic_set(&root
->log_writers
, 0);
1264 atomic_set(&root
->log_batch
, 0);
1265 atomic_set(&root
->orphan_inodes
, 0);
1266 atomic_set(&root
->refs
, 1);
1267 atomic_set(&root
->will_be_snapshoted
, 0);
1268 root
->log_transid
= 0;
1269 root
->log_transid_committed
= -1;
1270 root
->last_log_commit
= 0;
1272 extent_io_tree_init(&root
->dirty_log_pages
,
1273 fs_info
->btree_inode
->i_mapping
);
1275 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1276 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1277 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1279 root
->defrag_trans_start
= fs_info
->generation
;
1281 root
->defrag_trans_start
= 0;
1282 root
->root_key
.objectid
= objectid
;
1285 spin_lock_init(&root
->root_item_lock
);
1288 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1290 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1292 root
->fs_info
= fs_info
;
1296 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1297 /* Should only be used by the testing infrastructure */
1298 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1300 struct btrfs_root
*root
;
1302 root
= btrfs_alloc_root(NULL
);
1304 return ERR_PTR(-ENOMEM
);
1305 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1306 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1307 root
->alloc_bytenr
= 0;
1313 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1314 struct btrfs_fs_info
*fs_info
,
1317 struct extent_buffer
*leaf
;
1318 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1319 struct btrfs_root
*root
;
1320 struct btrfs_key key
;
1324 root
= btrfs_alloc_root(fs_info
);
1326 return ERR_PTR(-ENOMEM
);
1328 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1329 tree_root
->stripesize
, root
, fs_info
, objectid
);
1330 root
->root_key
.objectid
= objectid
;
1331 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1332 root
->root_key
.offset
= 0;
1334 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1336 ret
= PTR_ERR(leaf
);
1341 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1342 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1343 btrfs_set_header_generation(leaf
, trans
->transid
);
1344 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1345 btrfs_set_header_owner(leaf
, objectid
);
1348 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1350 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1351 btrfs_header_chunk_tree_uuid(leaf
),
1353 btrfs_mark_buffer_dirty(leaf
);
1355 root
->commit_root
= btrfs_root_node(root
);
1356 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1358 root
->root_item
.flags
= 0;
1359 root
->root_item
.byte_limit
= 0;
1360 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1361 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1362 btrfs_set_root_level(&root
->root_item
, 0);
1363 btrfs_set_root_refs(&root
->root_item
, 1);
1364 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1365 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1366 btrfs_set_root_dirid(&root
->root_item
, 0);
1368 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1369 root
->root_item
.drop_level
= 0;
1371 key
.objectid
= objectid
;
1372 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1374 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1378 btrfs_tree_unlock(leaf
);
1384 btrfs_tree_unlock(leaf
);
1385 free_extent_buffer(root
->commit_root
);
1386 free_extent_buffer(leaf
);
1390 return ERR_PTR(ret
);
1393 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1394 struct btrfs_fs_info
*fs_info
)
1396 struct btrfs_root
*root
;
1397 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1398 struct extent_buffer
*leaf
;
1400 root
= btrfs_alloc_root(fs_info
);
1402 return ERR_PTR(-ENOMEM
);
1404 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1405 tree_root
->stripesize
, root
, fs_info
,
1406 BTRFS_TREE_LOG_OBJECTID
);
1408 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1409 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1410 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1413 * DON'T set REF_COWS for log trees
1415 * log trees do not get reference counted because they go away
1416 * before a real commit is actually done. They do store pointers
1417 * to file data extents, and those reference counts still get
1418 * updated (along with back refs to the log tree).
1421 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1425 return ERR_CAST(leaf
);
1428 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1429 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1430 btrfs_set_header_generation(leaf
, trans
->transid
);
1431 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1432 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1435 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1436 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1437 btrfs_mark_buffer_dirty(root
->node
);
1438 btrfs_tree_unlock(root
->node
);
1442 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1443 struct btrfs_fs_info
*fs_info
)
1445 struct btrfs_root
*log_root
;
1447 log_root
= alloc_log_tree(trans
, fs_info
);
1448 if (IS_ERR(log_root
))
1449 return PTR_ERR(log_root
);
1450 WARN_ON(fs_info
->log_root_tree
);
1451 fs_info
->log_root_tree
= log_root
;
1455 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1456 struct btrfs_root
*root
)
1458 struct btrfs_root
*log_root
;
1459 struct btrfs_inode_item
*inode_item
;
1461 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1462 if (IS_ERR(log_root
))
1463 return PTR_ERR(log_root
);
1465 log_root
->last_trans
= trans
->transid
;
1466 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1468 inode_item
= &log_root
->root_item
.inode
;
1469 btrfs_set_stack_inode_generation(inode_item
, 1);
1470 btrfs_set_stack_inode_size(inode_item
, 3);
1471 btrfs_set_stack_inode_nlink(inode_item
, 1);
1472 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1473 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1475 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1477 WARN_ON(root
->log_root
);
1478 root
->log_root
= log_root
;
1479 root
->log_transid
= 0;
1480 root
->log_transid_committed
= -1;
1481 root
->last_log_commit
= 0;
1485 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1486 struct btrfs_key
*key
)
1488 struct btrfs_root
*root
;
1489 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1490 struct btrfs_path
*path
;
1494 path
= btrfs_alloc_path();
1496 return ERR_PTR(-ENOMEM
);
1498 root
= btrfs_alloc_root(fs_info
);
1504 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1505 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1507 ret
= btrfs_find_root(tree_root
, key
, path
,
1508 &root
->root_item
, &root
->root_key
);
1515 generation
= btrfs_root_generation(&root
->root_item
);
1516 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1521 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1525 root
->commit_root
= btrfs_root_node(root
);
1527 btrfs_free_path(path
);
1531 free_extent_buffer(root
->node
);
1535 root
= ERR_PTR(ret
);
1539 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1540 struct btrfs_key
*location
)
1542 struct btrfs_root
*root
;
1544 root
= btrfs_read_tree_root(tree_root
, location
);
1548 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1549 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1550 btrfs_check_and_init_root_item(&root
->root_item
);
1556 int btrfs_init_fs_root(struct btrfs_root
*root
)
1559 struct btrfs_subvolume_writers
*writers
;
1561 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1562 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1564 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1569 writers
= btrfs_alloc_subvolume_writers();
1570 if (IS_ERR(writers
)) {
1571 ret
= PTR_ERR(writers
);
1574 root
->subv_writers
= writers
;
1576 btrfs_init_free_ino_ctl(root
);
1577 spin_lock_init(&root
->ino_cache_lock
);
1578 init_waitqueue_head(&root
->ino_cache_wait
);
1580 ret
= get_anon_bdev(&root
->anon_dev
);
1586 btrfs_free_subvolume_writers(root
->subv_writers
);
1588 kfree(root
->free_ino_ctl
);
1589 kfree(root
->free_ino_pinned
);
1593 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1596 struct btrfs_root
*root
;
1598 spin_lock(&fs_info
->fs_roots_radix_lock
);
1599 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1600 (unsigned long)root_id
);
1601 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1605 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1606 struct btrfs_root
*root
)
1610 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1614 spin_lock(&fs_info
->fs_roots_radix_lock
);
1615 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1616 (unsigned long)root
->root_key
.objectid
,
1619 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1620 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1621 radix_tree_preload_end();
1626 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1627 struct btrfs_key
*location
,
1630 struct btrfs_root
*root
;
1631 struct btrfs_path
*path
;
1632 struct btrfs_key key
;
1635 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1636 return fs_info
->tree_root
;
1637 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1638 return fs_info
->extent_root
;
1639 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1640 return fs_info
->chunk_root
;
1641 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1642 return fs_info
->dev_root
;
1643 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1644 return fs_info
->csum_root
;
1645 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1646 return fs_info
->quota_root
? fs_info
->quota_root
:
1648 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1649 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1652 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1654 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1655 return ERR_PTR(-ENOENT
);
1659 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1663 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1668 ret
= btrfs_init_fs_root(root
);
1672 path
= btrfs_alloc_path();
1677 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1678 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1679 key
.offset
= location
->objectid
;
1681 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1682 btrfs_free_path(path
);
1686 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1688 ret
= btrfs_insert_fs_root(fs_info
, root
);
1690 if (ret
== -EEXIST
) {
1699 return ERR_PTR(ret
);
1702 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1704 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1706 struct btrfs_device
*device
;
1707 struct backing_dev_info
*bdi
;
1710 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1713 bdi
= blk_get_backing_dev_info(device
->bdev
);
1714 if (bdi_congested(bdi
, bdi_bits
)) {
1723 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1727 err
= bdi_setup_and_register(bdi
, "btrfs");
1731 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
;
1732 bdi
->congested_fn
= btrfs_congested_fn
;
1733 bdi
->congested_data
= info
;
1738 * called by the kthread helper functions to finally call the bio end_io
1739 * functions. This is where read checksum verification actually happens
1741 static void end_workqueue_fn(struct btrfs_work
*work
)
1744 struct btrfs_end_io_wq
*end_io_wq
;
1747 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1748 bio
= end_io_wq
->bio
;
1750 error
= end_io_wq
->error
;
1751 bio
->bi_private
= end_io_wq
->private;
1752 bio
->bi_end_io
= end_io_wq
->end_io
;
1753 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1754 bio_endio_nodec(bio
, error
);
1757 static int cleaner_kthread(void *arg
)
1759 struct btrfs_root
*root
= arg
;
1765 /* Make the cleaner go to sleep early. */
1766 if (btrfs_need_cleaner_sleep(root
))
1769 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1773 * Avoid the problem that we change the status of the fs
1774 * during the above check and trylock.
1776 if (btrfs_need_cleaner_sleep(root
)) {
1777 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1781 btrfs_run_delayed_iputs(root
);
1782 btrfs_delete_unused_bgs(root
->fs_info
);
1783 again
= btrfs_clean_one_deleted_snapshot(root
);
1784 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1787 * The defragger has dealt with the R/O remount and umount,
1788 * needn't do anything special here.
1790 btrfs_run_defrag_inodes(root
->fs_info
);
1792 if (!try_to_freeze() && !again
) {
1793 set_current_state(TASK_INTERRUPTIBLE
);
1794 if (!kthread_should_stop())
1796 __set_current_state(TASK_RUNNING
);
1798 } while (!kthread_should_stop());
1802 static int transaction_kthread(void *arg
)
1804 struct btrfs_root
*root
= arg
;
1805 struct btrfs_trans_handle
*trans
;
1806 struct btrfs_transaction
*cur
;
1809 unsigned long delay
;
1813 cannot_commit
= false;
1814 delay
= HZ
* root
->fs_info
->commit_interval
;
1815 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1817 spin_lock(&root
->fs_info
->trans_lock
);
1818 cur
= root
->fs_info
->running_transaction
;
1820 spin_unlock(&root
->fs_info
->trans_lock
);
1824 now
= get_seconds();
1825 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1826 (now
< cur
->start_time
||
1827 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1828 spin_unlock(&root
->fs_info
->trans_lock
);
1832 transid
= cur
->transid
;
1833 spin_unlock(&root
->fs_info
->trans_lock
);
1835 /* If the file system is aborted, this will always fail. */
1836 trans
= btrfs_attach_transaction(root
);
1837 if (IS_ERR(trans
)) {
1838 if (PTR_ERR(trans
) != -ENOENT
)
1839 cannot_commit
= true;
1842 if (transid
== trans
->transid
) {
1843 btrfs_commit_transaction(trans
, root
);
1845 btrfs_end_transaction(trans
, root
);
1848 wake_up_process(root
->fs_info
->cleaner_kthread
);
1849 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1851 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1852 &root
->fs_info
->fs_state
)))
1853 btrfs_cleanup_transaction(root
);
1854 if (!try_to_freeze()) {
1855 set_current_state(TASK_INTERRUPTIBLE
);
1856 if (!kthread_should_stop() &&
1857 (!btrfs_transaction_blocked(root
->fs_info
) ||
1859 schedule_timeout(delay
);
1860 __set_current_state(TASK_RUNNING
);
1862 } while (!kthread_should_stop());
1867 * this will find the highest generation in the array of
1868 * root backups. The index of the highest array is returned,
1869 * or -1 if we can't find anything.
1871 * We check to make sure the array is valid by comparing the
1872 * generation of the latest root in the array with the generation
1873 * in the super block. If they don't match we pitch it.
1875 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1878 int newest_index
= -1;
1879 struct btrfs_root_backup
*root_backup
;
1882 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1883 root_backup
= info
->super_copy
->super_roots
+ i
;
1884 cur
= btrfs_backup_tree_root_gen(root_backup
);
1885 if (cur
== newest_gen
)
1889 /* check to see if we actually wrapped around */
1890 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1891 root_backup
= info
->super_copy
->super_roots
;
1892 cur
= btrfs_backup_tree_root_gen(root_backup
);
1893 if (cur
== newest_gen
)
1896 return newest_index
;
1901 * find the oldest backup so we know where to store new entries
1902 * in the backup array. This will set the backup_root_index
1903 * field in the fs_info struct
1905 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1908 int newest_index
= -1;
1910 newest_index
= find_newest_super_backup(info
, newest_gen
);
1911 /* if there was garbage in there, just move along */
1912 if (newest_index
== -1) {
1913 info
->backup_root_index
= 0;
1915 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1920 * copy all the root pointers into the super backup array.
1921 * this will bump the backup pointer by one when it is
1924 static void backup_super_roots(struct btrfs_fs_info
*info
)
1927 struct btrfs_root_backup
*root_backup
;
1930 next_backup
= info
->backup_root_index
;
1931 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1932 BTRFS_NUM_BACKUP_ROOTS
;
1935 * just overwrite the last backup if we're at the same generation
1936 * this happens only at umount
1938 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1939 if (btrfs_backup_tree_root_gen(root_backup
) ==
1940 btrfs_header_generation(info
->tree_root
->node
))
1941 next_backup
= last_backup
;
1943 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1946 * make sure all of our padding and empty slots get zero filled
1947 * regardless of which ones we use today
1949 memset(root_backup
, 0, sizeof(*root_backup
));
1951 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1953 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1954 btrfs_set_backup_tree_root_gen(root_backup
,
1955 btrfs_header_generation(info
->tree_root
->node
));
1957 btrfs_set_backup_tree_root_level(root_backup
,
1958 btrfs_header_level(info
->tree_root
->node
));
1960 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1961 btrfs_set_backup_chunk_root_gen(root_backup
,
1962 btrfs_header_generation(info
->chunk_root
->node
));
1963 btrfs_set_backup_chunk_root_level(root_backup
,
1964 btrfs_header_level(info
->chunk_root
->node
));
1966 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1967 btrfs_set_backup_extent_root_gen(root_backup
,
1968 btrfs_header_generation(info
->extent_root
->node
));
1969 btrfs_set_backup_extent_root_level(root_backup
,
1970 btrfs_header_level(info
->extent_root
->node
));
1973 * we might commit during log recovery, which happens before we set
1974 * the fs_root. Make sure it is valid before we fill it in.
1976 if (info
->fs_root
&& info
->fs_root
->node
) {
1977 btrfs_set_backup_fs_root(root_backup
,
1978 info
->fs_root
->node
->start
);
1979 btrfs_set_backup_fs_root_gen(root_backup
,
1980 btrfs_header_generation(info
->fs_root
->node
));
1981 btrfs_set_backup_fs_root_level(root_backup
,
1982 btrfs_header_level(info
->fs_root
->node
));
1985 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1986 btrfs_set_backup_dev_root_gen(root_backup
,
1987 btrfs_header_generation(info
->dev_root
->node
));
1988 btrfs_set_backup_dev_root_level(root_backup
,
1989 btrfs_header_level(info
->dev_root
->node
));
1991 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1992 btrfs_set_backup_csum_root_gen(root_backup
,
1993 btrfs_header_generation(info
->csum_root
->node
));
1994 btrfs_set_backup_csum_root_level(root_backup
,
1995 btrfs_header_level(info
->csum_root
->node
));
1997 btrfs_set_backup_total_bytes(root_backup
,
1998 btrfs_super_total_bytes(info
->super_copy
));
1999 btrfs_set_backup_bytes_used(root_backup
,
2000 btrfs_super_bytes_used(info
->super_copy
));
2001 btrfs_set_backup_num_devices(root_backup
,
2002 btrfs_super_num_devices(info
->super_copy
));
2005 * if we don't copy this out to the super_copy, it won't get remembered
2006 * for the next commit
2008 memcpy(&info
->super_copy
->super_roots
,
2009 &info
->super_for_commit
->super_roots
,
2010 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2014 * this copies info out of the root backup array and back into
2015 * the in-memory super block. It is meant to help iterate through
2016 * the array, so you send it the number of backups you've already
2017 * tried and the last backup index you used.
2019 * this returns -1 when it has tried all the backups
2021 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2022 struct btrfs_super_block
*super
,
2023 int *num_backups_tried
, int *backup_index
)
2025 struct btrfs_root_backup
*root_backup
;
2026 int newest
= *backup_index
;
2028 if (*num_backups_tried
== 0) {
2029 u64 gen
= btrfs_super_generation(super
);
2031 newest
= find_newest_super_backup(info
, gen
);
2035 *backup_index
= newest
;
2036 *num_backups_tried
= 1;
2037 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2038 /* we've tried all the backups, all done */
2041 /* jump to the next oldest backup */
2042 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2043 BTRFS_NUM_BACKUP_ROOTS
;
2044 *backup_index
= newest
;
2045 *num_backups_tried
+= 1;
2047 root_backup
= super
->super_roots
+ newest
;
2049 btrfs_set_super_generation(super
,
2050 btrfs_backup_tree_root_gen(root_backup
));
2051 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2052 btrfs_set_super_root_level(super
,
2053 btrfs_backup_tree_root_level(root_backup
));
2054 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2057 * fixme: the total bytes and num_devices need to match or we should
2060 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2061 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2065 /* helper to cleanup workers */
2066 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2068 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2069 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2070 btrfs_destroy_workqueue(fs_info
->workers
);
2071 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2072 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2073 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2074 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2075 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2076 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2077 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2078 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2079 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2080 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2081 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2082 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2083 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2084 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2085 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2088 static void free_root_extent_buffers(struct btrfs_root
*root
)
2091 free_extent_buffer(root
->node
);
2092 free_extent_buffer(root
->commit_root
);
2094 root
->commit_root
= NULL
;
2098 /* helper to cleanup tree roots */
2099 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2101 free_root_extent_buffers(info
->tree_root
);
2103 free_root_extent_buffers(info
->dev_root
);
2104 free_root_extent_buffers(info
->extent_root
);
2105 free_root_extent_buffers(info
->csum_root
);
2106 free_root_extent_buffers(info
->quota_root
);
2107 free_root_extent_buffers(info
->uuid_root
);
2109 free_root_extent_buffers(info
->chunk_root
);
2112 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2115 struct btrfs_root
*gang
[8];
2118 while (!list_empty(&fs_info
->dead_roots
)) {
2119 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2120 struct btrfs_root
, root_list
);
2121 list_del(&gang
[0]->root_list
);
2123 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2124 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2126 free_extent_buffer(gang
[0]->node
);
2127 free_extent_buffer(gang
[0]->commit_root
);
2128 btrfs_put_fs_root(gang
[0]);
2133 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2138 for (i
= 0; i
< ret
; i
++)
2139 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2142 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2143 btrfs_free_log_root_tree(NULL
, fs_info
);
2144 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2145 fs_info
->pinned_extents
);
2149 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2151 mutex_init(&fs_info
->scrub_lock
);
2152 atomic_set(&fs_info
->scrubs_running
, 0);
2153 atomic_set(&fs_info
->scrub_pause_req
, 0);
2154 atomic_set(&fs_info
->scrubs_paused
, 0);
2155 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2156 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2157 fs_info
->scrub_workers_refcnt
= 0;
2160 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2162 spin_lock_init(&fs_info
->balance_lock
);
2163 mutex_init(&fs_info
->balance_mutex
);
2164 atomic_set(&fs_info
->balance_running
, 0);
2165 atomic_set(&fs_info
->balance_pause_req
, 0);
2166 atomic_set(&fs_info
->balance_cancel_req
, 0);
2167 fs_info
->balance_ctl
= NULL
;
2168 init_waitqueue_head(&fs_info
->balance_wait_q
);
2171 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2172 struct btrfs_root
*tree_root
)
2174 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2175 set_nlink(fs_info
->btree_inode
, 1);
2177 * we set the i_size on the btree inode to the max possible int.
2178 * the real end of the address space is determined by all of
2179 * the devices in the system
2181 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2182 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2184 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2185 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2186 fs_info
->btree_inode
->i_mapping
);
2187 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2188 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2190 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2192 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2193 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2194 sizeof(struct btrfs_key
));
2195 set_bit(BTRFS_INODE_DUMMY
,
2196 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2197 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2200 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2202 fs_info
->dev_replace
.lock_owner
= 0;
2203 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2204 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2205 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2206 mutex_init(&fs_info
->dev_replace
.lock
);
2207 init_waitqueue_head(&fs_info
->replace_wait
);
2210 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2212 spin_lock_init(&fs_info
->qgroup_lock
);
2213 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2214 fs_info
->qgroup_tree
= RB_ROOT
;
2215 fs_info
->qgroup_op_tree
= RB_ROOT
;
2216 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2217 fs_info
->qgroup_seq
= 1;
2218 fs_info
->quota_enabled
= 0;
2219 fs_info
->pending_quota_state
= 0;
2220 fs_info
->qgroup_ulist
= NULL
;
2221 mutex_init(&fs_info
->qgroup_rescan_lock
);
2224 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2225 struct btrfs_fs_devices
*fs_devices
)
2227 int max_active
= fs_info
->thread_pool_size
;
2228 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2231 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2234 fs_info
->delalloc_workers
=
2235 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2237 fs_info
->flush_workers
=
2238 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2240 fs_info
->caching_workers
=
2241 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2244 * a higher idle thresh on the submit workers makes it much more
2245 * likely that bios will be send down in a sane order to the
2248 fs_info
->submit_workers
=
2249 btrfs_alloc_workqueue("submit", flags
,
2250 min_t(u64
, fs_devices
->num_devices
,
2253 fs_info
->fixup_workers
=
2254 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2257 * endios are largely parallel and should have a very
2260 fs_info
->endio_workers
=
2261 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2262 fs_info
->endio_meta_workers
=
2263 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2264 fs_info
->endio_meta_write_workers
=
2265 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2266 fs_info
->endio_raid56_workers
=
2267 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2268 fs_info
->endio_repair_workers
=
2269 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2270 fs_info
->rmw_workers
=
2271 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2272 fs_info
->endio_write_workers
=
2273 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2274 fs_info
->endio_freespace_worker
=
2275 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2276 fs_info
->delayed_workers
=
2277 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2278 fs_info
->readahead_workers
=
2279 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2280 fs_info
->qgroup_rescan_workers
=
2281 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2282 fs_info
->extent_workers
=
2283 btrfs_alloc_workqueue("extent-refs", flags
,
2284 min_t(u64
, fs_devices
->num_devices
,
2287 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2288 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2289 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2290 fs_info
->endio_meta_write_workers
&&
2291 fs_info
->endio_repair_workers
&&
2292 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2293 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2294 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2295 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2296 fs_info
->extent_workers
&&
2297 fs_info
->qgroup_rescan_workers
)) {
2304 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2305 struct btrfs_fs_devices
*fs_devices
)
2308 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2309 struct btrfs_root
*log_tree_root
;
2310 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2311 u64 bytenr
= btrfs_super_log_root(disk_super
);
2313 if (fs_devices
->rw_devices
== 0) {
2314 printk(KERN_WARNING
"BTRFS: log replay required "
2319 log_tree_root
= btrfs_alloc_root(fs_info
);
2323 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2324 tree_root
->stripesize
, log_tree_root
, fs_info
,
2325 BTRFS_TREE_LOG_OBJECTID
);
2327 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2328 fs_info
->generation
+ 1);
2329 if (!log_tree_root
->node
||
2330 !extent_buffer_uptodate(log_tree_root
->node
)) {
2331 printk(KERN_ERR
"BTRFS: failed to read log tree\n");
2332 free_extent_buffer(log_tree_root
->node
);
2333 kfree(log_tree_root
);
2336 /* returns with log_tree_root freed on success */
2337 ret
= btrfs_recover_log_trees(log_tree_root
);
2339 btrfs_error(tree_root
->fs_info
, ret
,
2340 "Failed to recover log tree");
2341 free_extent_buffer(log_tree_root
->node
);
2342 kfree(log_tree_root
);
2346 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2347 ret
= btrfs_commit_super(tree_root
);
2355 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2356 struct btrfs_root
*tree_root
)
2358 struct btrfs_root
*root
;
2359 struct btrfs_key location
;
2362 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2363 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2364 location
.offset
= 0;
2366 root
= btrfs_read_tree_root(tree_root
, &location
);
2368 return PTR_ERR(root
);
2369 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2370 fs_info
->extent_root
= root
;
2372 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2373 root
= btrfs_read_tree_root(tree_root
, &location
);
2375 return PTR_ERR(root
);
2376 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2377 fs_info
->dev_root
= root
;
2378 btrfs_init_devices_late(fs_info
);
2380 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2381 root
= btrfs_read_tree_root(tree_root
, &location
);
2383 return PTR_ERR(root
);
2384 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2385 fs_info
->csum_root
= root
;
2387 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2388 root
= btrfs_read_tree_root(tree_root
, &location
);
2389 if (!IS_ERR(root
)) {
2390 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2391 fs_info
->quota_enabled
= 1;
2392 fs_info
->pending_quota_state
= 1;
2393 fs_info
->quota_root
= root
;
2396 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2397 root
= btrfs_read_tree_root(tree_root
, &location
);
2399 ret
= PTR_ERR(root
);
2403 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2404 fs_info
->uuid_root
= root
;
2410 int open_ctree(struct super_block
*sb
,
2411 struct btrfs_fs_devices
*fs_devices
,
2419 struct btrfs_key location
;
2420 struct buffer_head
*bh
;
2421 struct btrfs_super_block
*disk_super
;
2422 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2423 struct btrfs_root
*tree_root
;
2424 struct btrfs_root
*chunk_root
;
2427 int num_backups_tried
= 0;
2428 int backup_index
= 0;
2431 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2432 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2433 if (!tree_root
|| !chunk_root
) {
2438 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2444 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2450 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2455 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2456 (1 + ilog2(nr_cpu_ids
));
2458 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2461 goto fail_dirty_metadata_bytes
;
2464 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2467 goto fail_delalloc_bytes
;
2470 fs_info
->btree_inode
= new_inode(sb
);
2471 if (!fs_info
->btree_inode
) {
2473 goto fail_bio_counter
;
2476 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2478 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2479 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2480 INIT_LIST_HEAD(&fs_info
->trans_list
);
2481 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2482 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2483 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2484 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2485 spin_lock_init(&fs_info
->delalloc_root_lock
);
2486 spin_lock_init(&fs_info
->trans_lock
);
2487 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2488 spin_lock_init(&fs_info
->delayed_iput_lock
);
2489 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2490 spin_lock_init(&fs_info
->free_chunk_lock
);
2491 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2492 spin_lock_init(&fs_info
->super_lock
);
2493 spin_lock_init(&fs_info
->qgroup_op_lock
);
2494 spin_lock_init(&fs_info
->buffer_lock
);
2495 spin_lock_init(&fs_info
->unused_bgs_lock
);
2496 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2497 rwlock_init(&fs_info
->tree_mod_log_lock
);
2498 mutex_init(&fs_info
->reloc_mutex
);
2499 mutex_init(&fs_info
->delalloc_root_mutex
);
2500 seqlock_init(&fs_info
->profiles_lock
);
2502 init_completion(&fs_info
->kobj_unregister
);
2503 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2504 INIT_LIST_HEAD(&fs_info
->space_info
);
2505 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2506 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2507 btrfs_mapping_init(&fs_info
->mapping_tree
);
2508 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2509 BTRFS_BLOCK_RSV_GLOBAL
);
2510 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2511 BTRFS_BLOCK_RSV_DELALLOC
);
2512 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2513 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2514 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2515 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2516 BTRFS_BLOCK_RSV_DELOPS
);
2517 atomic_set(&fs_info
->nr_async_submits
, 0);
2518 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2519 atomic_set(&fs_info
->async_submit_draining
, 0);
2520 atomic_set(&fs_info
->nr_async_bios
, 0);
2521 atomic_set(&fs_info
->defrag_running
, 0);
2522 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2523 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2525 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2526 fs_info
->metadata_ratio
= 0;
2527 fs_info
->defrag_inodes
= RB_ROOT
;
2528 fs_info
->free_chunk_space
= 0;
2529 fs_info
->tree_mod_log
= RB_ROOT
;
2530 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2531 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2532 /* readahead state */
2533 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2534 spin_lock_init(&fs_info
->reada_lock
);
2536 fs_info
->thread_pool_size
= min_t(unsigned long,
2537 num_online_cpus() + 2, 8);
2539 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2540 spin_lock_init(&fs_info
->ordered_root_lock
);
2541 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2543 if (!fs_info
->delayed_root
) {
2547 btrfs_init_delayed_root(fs_info
->delayed_root
);
2549 btrfs_init_scrub(fs_info
);
2550 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2551 fs_info
->check_integrity_print_mask
= 0;
2553 btrfs_init_balance(fs_info
);
2554 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2556 sb
->s_blocksize
= 4096;
2557 sb
->s_blocksize_bits
= blksize_bits(4096);
2558 sb
->s_bdi
= &fs_info
->bdi
;
2560 btrfs_init_btree_inode(fs_info
, tree_root
);
2562 spin_lock_init(&fs_info
->block_group_cache_lock
);
2563 fs_info
->block_group_cache_tree
= RB_ROOT
;
2564 fs_info
->first_logical_byte
= (u64
)-1;
2566 extent_io_tree_init(&fs_info
->freed_extents
[0],
2567 fs_info
->btree_inode
->i_mapping
);
2568 extent_io_tree_init(&fs_info
->freed_extents
[1],
2569 fs_info
->btree_inode
->i_mapping
);
2570 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2571 fs_info
->do_barriers
= 1;
2574 mutex_init(&fs_info
->ordered_operations_mutex
);
2575 mutex_init(&fs_info
->ordered_extent_flush_mutex
);
2576 mutex_init(&fs_info
->tree_log_mutex
);
2577 mutex_init(&fs_info
->chunk_mutex
);
2578 mutex_init(&fs_info
->transaction_kthread_mutex
);
2579 mutex_init(&fs_info
->cleaner_mutex
);
2580 mutex_init(&fs_info
->volume_mutex
);
2581 init_rwsem(&fs_info
->commit_root_sem
);
2582 init_rwsem(&fs_info
->cleanup_work_sem
);
2583 init_rwsem(&fs_info
->subvol_sem
);
2584 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2586 btrfs_init_dev_replace_locks(fs_info
);
2587 btrfs_init_qgroup(fs_info
);
2589 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2590 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2592 init_waitqueue_head(&fs_info
->transaction_throttle
);
2593 init_waitqueue_head(&fs_info
->transaction_wait
);
2594 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2595 init_waitqueue_head(&fs_info
->async_submit_wait
);
2597 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2599 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2605 __setup_root(4096, 4096, 4096, tree_root
,
2606 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2608 invalidate_bdev(fs_devices
->latest_bdev
);
2611 * Read super block and check the signature bytes only
2613 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2620 * We want to check superblock checksum, the type is stored inside.
2621 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2623 if (btrfs_check_super_csum(bh
->b_data
)) {
2624 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2630 * super_copy is zeroed at allocation time and we never touch the
2631 * following bytes up to INFO_SIZE, the checksum is calculated from
2632 * the whole block of INFO_SIZE
2634 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2635 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2636 sizeof(*fs_info
->super_for_commit
));
2639 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2641 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2643 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2648 disk_super
= fs_info
->super_copy
;
2649 if (!btrfs_super_root(disk_super
))
2652 /* check FS state, whether FS is broken. */
2653 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2654 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2657 * run through our array of backup supers and setup
2658 * our ring pointer to the oldest one
2660 generation
= btrfs_super_generation(disk_super
);
2661 find_oldest_super_backup(fs_info
, generation
);
2664 * In the long term, we'll store the compression type in the super
2665 * block, and it'll be used for per file compression control.
2667 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2669 ret
= btrfs_parse_options(tree_root
, options
);
2675 features
= btrfs_super_incompat_flags(disk_super
) &
2676 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2678 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2679 "unsupported optional features (%Lx).\n",
2686 * Leafsize and nodesize were always equal, this is only a sanity check.
2688 if (le32_to_cpu(disk_super
->__unused_leafsize
) !=
2689 btrfs_super_nodesize(disk_super
)) {
2690 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2691 "blocksizes don't match. node %d leaf %d\n",
2692 btrfs_super_nodesize(disk_super
),
2693 le32_to_cpu(disk_super
->__unused_leafsize
));
2697 if (btrfs_super_nodesize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2698 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2699 "blocksize (%d) was too large\n",
2700 btrfs_super_nodesize(disk_super
));
2705 features
= btrfs_super_incompat_flags(disk_super
);
2706 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2707 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2708 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2710 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2711 printk(KERN_INFO
"BTRFS: has skinny extents\n");
2714 * flag our filesystem as having big metadata blocks if
2715 * they are bigger than the page size
2717 if (btrfs_super_nodesize(disk_super
) > PAGE_CACHE_SIZE
) {
2718 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2719 printk(KERN_INFO
"BTRFS: flagging fs with big metadata feature\n");
2720 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2723 nodesize
= btrfs_super_nodesize(disk_super
);
2724 sectorsize
= btrfs_super_sectorsize(disk_super
);
2725 stripesize
= btrfs_super_stripesize(disk_super
);
2726 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2727 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2730 * mixed block groups end up with duplicate but slightly offset
2731 * extent buffers for the same range. It leads to corruptions
2733 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2734 (sectorsize
!= nodesize
)) {
2735 printk(KERN_ERR
"BTRFS: unequal leaf/node/sector sizes "
2736 "are not allowed for mixed block groups on %s\n",
2742 * Needn't use the lock because there is no other task which will
2745 btrfs_set_super_incompat_flags(disk_super
, features
);
2747 features
= btrfs_super_compat_ro_flags(disk_super
) &
2748 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2749 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2750 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2751 "unsupported option features (%Lx).\n",
2757 max_active
= fs_info
->thread_pool_size
;
2759 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2762 goto fail_sb_buffer
;
2765 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2766 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2767 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2769 tree_root
->nodesize
= nodesize
;
2770 tree_root
->sectorsize
= sectorsize
;
2771 tree_root
->stripesize
= stripesize
;
2773 sb
->s_blocksize
= sectorsize
;
2774 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2776 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2777 printk(KERN_ERR
"BTRFS: valid FS not found on %s\n", sb
->s_id
);
2778 goto fail_sb_buffer
;
2781 if (sectorsize
!= PAGE_SIZE
) {
2782 printk(KERN_ERR
"BTRFS: incompatible sector size (%lu) "
2783 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2784 goto fail_sb_buffer
;
2787 mutex_lock(&fs_info
->chunk_mutex
);
2788 ret
= btrfs_read_sys_array(tree_root
);
2789 mutex_unlock(&fs_info
->chunk_mutex
);
2791 printk(KERN_ERR
"BTRFS: failed to read the system "
2792 "array on %s\n", sb
->s_id
);
2793 goto fail_sb_buffer
;
2796 generation
= btrfs_super_chunk_root_generation(disk_super
);
2798 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2799 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2801 chunk_root
->node
= read_tree_block(chunk_root
,
2802 btrfs_super_chunk_root(disk_super
),
2804 if (!chunk_root
->node
||
2805 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2806 printk(KERN_ERR
"BTRFS: failed to read chunk root on %s\n",
2808 goto fail_tree_roots
;
2810 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2811 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2813 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2814 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2816 ret
= btrfs_read_chunk_tree(chunk_root
);
2818 printk(KERN_ERR
"BTRFS: failed to read chunk tree on %s\n",
2820 goto fail_tree_roots
;
2824 * keep the device that is marked to be the target device for the
2825 * dev_replace procedure
2827 btrfs_close_extra_devices(fs_devices
, 0);
2829 if (!fs_devices
->latest_bdev
) {
2830 printk(KERN_ERR
"BTRFS: failed to read devices on %s\n",
2832 goto fail_tree_roots
;
2836 generation
= btrfs_super_generation(disk_super
);
2838 tree_root
->node
= read_tree_block(tree_root
,
2839 btrfs_super_root(disk_super
),
2841 if (!tree_root
->node
||
2842 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2843 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2846 goto recovery_tree_root
;
2849 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2850 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2851 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2853 ret
= btrfs_read_roots(fs_info
, tree_root
);
2855 goto recovery_tree_root
;
2857 fs_info
->generation
= generation
;
2858 fs_info
->last_trans_committed
= generation
;
2860 ret
= btrfs_recover_balance(fs_info
);
2862 printk(KERN_ERR
"BTRFS: failed to recover balance\n");
2863 goto fail_block_groups
;
2866 ret
= btrfs_init_dev_stats(fs_info
);
2868 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2870 goto fail_block_groups
;
2873 ret
= btrfs_init_dev_replace(fs_info
);
2875 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2876 goto fail_block_groups
;
2879 btrfs_close_extra_devices(fs_devices
, 1);
2881 ret
= btrfs_sysfs_add_one(fs_info
);
2883 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2884 goto fail_block_groups
;
2887 ret
= btrfs_init_space_info(fs_info
);
2889 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2893 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
2895 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
2898 fs_info
->num_tolerated_disk_barrier_failures
=
2899 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2900 if (fs_info
->fs_devices
->missing_devices
>
2901 fs_info
->num_tolerated_disk_barrier_failures
&&
2902 !(sb
->s_flags
& MS_RDONLY
)) {
2903 printk(KERN_WARNING
"BTRFS: "
2904 "too many missing devices, writeable mount is not allowed\n");
2908 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2910 if (IS_ERR(fs_info
->cleaner_kthread
))
2913 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2915 "btrfs-transaction");
2916 if (IS_ERR(fs_info
->transaction_kthread
))
2919 if (!btrfs_test_opt(tree_root
, SSD
) &&
2920 !btrfs_test_opt(tree_root
, NOSSD
) &&
2921 !fs_info
->fs_devices
->rotating
) {
2922 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
2924 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2928 * Mount does not set all options immediatelly, we can do it now and do
2929 * not have to wait for transaction commit
2931 btrfs_apply_pending_changes(fs_info
);
2933 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2934 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2935 ret
= btrfsic_mount(tree_root
, fs_devices
,
2936 btrfs_test_opt(tree_root
,
2937 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2939 fs_info
->check_integrity_print_mask
);
2941 printk(KERN_WARNING
"BTRFS: failed to initialize"
2942 " integrity check module %s\n", sb
->s_id
);
2945 ret
= btrfs_read_qgroup_config(fs_info
);
2947 goto fail_trans_kthread
;
2949 /* do not make disk changes in broken FS */
2950 if (btrfs_super_log_root(disk_super
) != 0) {
2951 ret
= btrfs_replay_log(fs_info
, fs_devices
);
2958 ret
= btrfs_find_orphan_roots(tree_root
);
2962 if (!(sb
->s_flags
& MS_RDONLY
)) {
2963 ret
= btrfs_cleanup_fs_roots(fs_info
);
2967 mutex_lock(&fs_info
->cleaner_mutex
);
2968 ret
= btrfs_recover_relocation(tree_root
);
2969 mutex_unlock(&fs_info
->cleaner_mutex
);
2972 "BTRFS: failed to recover relocation\n");
2978 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2979 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2980 location
.offset
= 0;
2982 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2983 if (IS_ERR(fs_info
->fs_root
)) {
2984 err
= PTR_ERR(fs_info
->fs_root
);
2988 if (sb
->s_flags
& MS_RDONLY
)
2991 down_read(&fs_info
->cleanup_work_sem
);
2992 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2993 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2994 up_read(&fs_info
->cleanup_work_sem
);
2995 close_ctree(tree_root
);
2998 up_read(&fs_info
->cleanup_work_sem
);
3000 ret
= btrfs_resume_balance_async(fs_info
);
3002 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
3003 close_ctree(tree_root
);
3007 ret
= btrfs_resume_dev_replace_async(fs_info
);
3009 pr_warn("BTRFS: failed to resume dev_replace\n");
3010 close_ctree(tree_root
);
3014 btrfs_qgroup_rescan_resume(fs_info
);
3016 if (!fs_info
->uuid_root
) {
3017 pr_info("BTRFS: creating UUID tree\n");
3018 ret
= btrfs_create_uuid_tree(fs_info
);
3020 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3022 close_ctree(tree_root
);
3025 } else if (btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
) ||
3026 fs_info
->generation
!=
3027 btrfs_super_uuid_tree_generation(disk_super
)) {
3028 pr_info("BTRFS: checking UUID tree\n");
3029 ret
= btrfs_check_uuid_tree(fs_info
);
3031 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3033 close_ctree(tree_root
);
3037 fs_info
->update_uuid_tree_gen
= 1;
3045 btrfs_free_qgroup_config(fs_info
);
3047 kthread_stop(fs_info
->transaction_kthread
);
3048 btrfs_cleanup_transaction(fs_info
->tree_root
);
3049 btrfs_free_fs_roots(fs_info
);
3051 kthread_stop(fs_info
->cleaner_kthread
);
3054 * make sure we're done with the btree inode before we stop our
3057 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3060 btrfs_sysfs_remove_one(fs_info
);
3063 btrfs_put_block_group_cache(fs_info
);
3064 btrfs_free_block_groups(fs_info
);
3067 free_root_pointers(fs_info
, 1);
3068 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3071 btrfs_stop_all_workers(fs_info
);
3074 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3076 iput(fs_info
->btree_inode
);
3078 percpu_counter_destroy(&fs_info
->bio_counter
);
3079 fail_delalloc_bytes
:
3080 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3081 fail_dirty_metadata_bytes
:
3082 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3084 bdi_destroy(&fs_info
->bdi
);
3086 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3088 btrfs_free_stripe_hash_table(fs_info
);
3089 btrfs_close_devices(fs_info
->fs_devices
);
3093 if (!btrfs_test_opt(tree_root
, RECOVERY
))
3094 goto fail_tree_roots
;
3096 free_root_pointers(fs_info
, 0);
3098 /* don't use the log in recovery mode, it won't be valid */
3099 btrfs_set_super_log_root(disk_super
, 0);
3101 /* we can't trust the free space cache either */
3102 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3104 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3105 &num_backups_tried
, &backup_index
);
3107 goto fail_block_groups
;
3108 goto retry_root_backup
;
3111 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3114 set_buffer_uptodate(bh
);
3116 struct btrfs_device
*device
= (struct btrfs_device
*)
3119 printk_ratelimited_in_rcu(KERN_WARNING
"BTRFS: lost page write due to "
3120 "I/O error on %s\n",
3121 rcu_str_deref(device
->name
));
3122 /* note, we dont' set_buffer_write_io_error because we have
3123 * our own ways of dealing with the IO errors
3125 clear_buffer_uptodate(bh
);
3126 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3132 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3134 struct buffer_head
*bh
;
3135 struct buffer_head
*latest
= NULL
;
3136 struct btrfs_super_block
*super
;
3141 /* we would like to check all the supers, but that would make
3142 * a btrfs mount succeed after a mkfs from a different FS.
3143 * So, we need to add a special mount option to scan for
3144 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3146 for (i
= 0; i
< 1; i
++) {
3147 bytenr
= btrfs_sb_offset(i
);
3148 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3149 i_size_read(bdev
->bd_inode
))
3151 bh
= __bread(bdev
, bytenr
/ 4096,
3152 BTRFS_SUPER_INFO_SIZE
);
3156 super
= (struct btrfs_super_block
*)bh
->b_data
;
3157 if (btrfs_super_bytenr(super
) != bytenr
||
3158 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3163 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3166 transid
= btrfs_super_generation(super
);
3175 * this should be called twice, once with wait == 0 and
3176 * once with wait == 1. When wait == 0 is done, all the buffer heads
3177 * we write are pinned.
3179 * They are released when wait == 1 is done.
3180 * max_mirrors must be the same for both runs, and it indicates how
3181 * many supers on this one device should be written.
3183 * max_mirrors == 0 means to write them all.
3185 static int write_dev_supers(struct btrfs_device
*device
,
3186 struct btrfs_super_block
*sb
,
3187 int do_barriers
, int wait
, int max_mirrors
)
3189 struct buffer_head
*bh
;
3196 if (max_mirrors
== 0)
3197 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3199 for (i
= 0; i
< max_mirrors
; i
++) {
3200 bytenr
= btrfs_sb_offset(i
);
3201 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3202 device
->commit_total_bytes
)
3206 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3207 BTRFS_SUPER_INFO_SIZE
);
3213 if (!buffer_uptodate(bh
))
3216 /* drop our reference */
3219 /* drop the reference from the wait == 0 run */
3223 btrfs_set_super_bytenr(sb
, bytenr
);
3226 crc
= btrfs_csum_data((char *)sb
+
3227 BTRFS_CSUM_SIZE
, crc
,
3228 BTRFS_SUPER_INFO_SIZE
-
3230 btrfs_csum_final(crc
, sb
->csum
);
3233 * one reference for us, and we leave it for the
3236 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3237 BTRFS_SUPER_INFO_SIZE
);
3239 printk(KERN_ERR
"BTRFS: couldn't get super "
3240 "buffer head for bytenr %Lu\n", bytenr
);
3245 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3247 /* one reference for submit_bh */
3250 set_buffer_uptodate(bh
);
3252 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3253 bh
->b_private
= device
;
3257 * we fua the first super. The others we allow
3261 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3263 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3267 return errors
< i
? 0 : -1;
3271 * endio for the write_dev_flush, this will wake anyone waiting
3272 * for the barrier when it is done
3274 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3277 if (err
== -EOPNOTSUPP
)
3278 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3279 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3281 if (bio
->bi_private
)
3282 complete(bio
->bi_private
);
3287 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3288 * sent down. With wait == 1, it waits for the previous flush.
3290 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3293 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3298 if (device
->nobarriers
)
3302 bio
= device
->flush_bio
;
3306 wait_for_completion(&device
->flush_wait
);
3308 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3309 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3310 rcu_str_deref(device
->name
));
3311 device
->nobarriers
= 1;
3312 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3314 btrfs_dev_stat_inc_and_print(device
,
3315 BTRFS_DEV_STAT_FLUSH_ERRS
);
3318 /* drop the reference from the wait == 0 run */
3320 device
->flush_bio
= NULL
;
3326 * one reference for us, and we leave it for the
3329 device
->flush_bio
= NULL
;
3330 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3334 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3335 bio
->bi_bdev
= device
->bdev
;
3336 init_completion(&device
->flush_wait
);
3337 bio
->bi_private
= &device
->flush_wait
;
3338 device
->flush_bio
= bio
;
3341 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3347 * send an empty flush down to each device in parallel,
3348 * then wait for them
3350 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3352 struct list_head
*head
;
3353 struct btrfs_device
*dev
;
3354 int errors_send
= 0;
3355 int errors_wait
= 0;
3358 /* send down all the barriers */
3359 head
= &info
->fs_devices
->devices
;
3360 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3367 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3370 ret
= write_dev_flush(dev
, 0);
3375 /* wait for all the barriers */
3376 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3383 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3386 ret
= write_dev_flush(dev
, 1);
3390 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3391 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3396 int btrfs_calc_num_tolerated_disk_barrier_failures(
3397 struct btrfs_fs_info
*fs_info
)
3399 struct btrfs_ioctl_space_info space
;
3400 struct btrfs_space_info
*sinfo
;
3401 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3402 BTRFS_BLOCK_GROUP_SYSTEM
,
3403 BTRFS_BLOCK_GROUP_METADATA
,
3404 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3408 int num_tolerated_disk_barrier_failures
=
3409 (int)fs_info
->fs_devices
->num_devices
;
3411 for (i
= 0; i
< num_types
; i
++) {
3412 struct btrfs_space_info
*tmp
;
3416 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3417 if (tmp
->flags
== types
[i
]) {
3427 down_read(&sinfo
->groups_sem
);
3428 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3429 if (!list_empty(&sinfo
->block_groups
[c
])) {
3432 btrfs_get_block_group_info(
3433 &sinfo
->block_groups
[c
], &space
);
3434 if (space
.total_bytes
== 0 ||
3435 space
.used_bytes
== 0)
3437 flags
= space
.flags
;
3440 * 0: if dup, single or RAID0 is configured for
3441 * any of metadata, system or data, else
3442 * 1: if RAID5 is configured, or if RAID1 or
3443 * RAID10 is configured and only two mirrors
3445 * 2: if RAID6 is configured, else
3446 * num_mirrors - 1: if RAID1 or RAID10 is
3447 * configured and more than
3448 * 2 mirrors are used.
3450 if (num_tolerated_disk_barrier_failures
> 0 &&
3451 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3452 BTRFS_BLOCK_GROUP_RAID0
)) ||
3453 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3455 num_tolerated_disk_barrier_failures
= 0;
3456 else if (num_tolerated_disk_barrier_failures
> 1) {
3457 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3458 BTRFS_BLOCK_GROUP_RAID5
|
3459 BTRFS_BLOCK_GROUP_RAID10
)) {
3460 num_tolerated_disk_barrier_failures
= 1;
3462 BTRFS_BLOCK_GROUP_RAID6
) {
3463 num_tolerated_disk_barrier_failures
= 2;
3468 up_read(&sinfo
->groups_sem
);
3471 return num_tolerated_disk_barrier_failures
;
3474 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3476 struct list_head
*head
;
3477 struct btrfs_device
*dev
;
3478 struct btrfs_super_block
*sb
;
3479 struct btrfs_dev_item
*dev_item
;
3483 int total_errors
= 0;
3486 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3487 backup_super_roots(root
->fs_info
);
3489 sb
= root
->fs_info
->super_for_commit
;
3490 dev_item
= &sb
->dev_item
;
3492 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3493 head
= &root
->fs_info
->fs_devices
->devices
;
3494 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3497 ret
= barrier_all_devices(root
->fs_info
);
3500 &root
->fs_info
->fs_devices
->device_list_mutex
);
3501 btrfs_error(root
->fs_info
, ret
,
3502 "errors while submitting device barriers.");
3507 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3512 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3515 btrfs_set_stack_device_generation(dev_item
, 0);
3516 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3517 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3518 btrfs_set_stack_device_total_bytes(dev_item
,
3519 dev
->commit_total_bytes
);
3520 btrfs_set_stack_device_bytes_used(dev_item
,
3521 dev
->commit_bytes_used
);
3522 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3523 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3524 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3525 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3526 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3528 flags
= btrfs_super_flags(sb
);
3529 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3531 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3535 if (total_errors
> max_errors
) {
3536 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3538 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3540 /* FUA is masked off if unsupported and can't be the reason */
3541 btrfs_error(root
->fs_info
, -EIO
,
3542 "%d errors while writing supers", total_errors
);
3547 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3550 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3553 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3557 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3558 if (total_errors
> max_errors
) {
3559 btrfs_error(root
->fs_info
, -EIO
,
3560 "%d errors while writing supers", total_errors
);
3566 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3567 struct btrfs_root
*root
, int max_mirrors
)
3569 return write_all_supers(root
, max_mirrors
);
3572 /* Drop a fs root from the radix tree and free it. */
3573 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3574 struct btrfs_root
*root
)
3576 spin_lock(&fs_info
->fs_roots_radix_lock
);
3577 radix_tree_delete(&fs_info
->fs_roots_radix
,
3578 (unsigned long)root
->root_key
.objectid
);
3579 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3581 if (btrfs_root_refs(&root
->root_item
) == 0)
3582 synchronize_srcu(&fs_info
->subvol_srcu
);
3584 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3585 btrfs_free_log(NULL
, root
);
3587 if (root
->free_ino_pinned
)
3588 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3589 if (root
->free_ino_ctl
)
3590 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3594 static void free_fs_root(struct btrfs_root
*root
)
3596 iput(root
->ino_cache_inode
);
3597 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3598 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3599 root
->orphan_block_rsv
= NULL
;
3601 free_anon_bdev(root
->anon_dev
);
3602 if (root
->subv_writers
)
3603 btrfs_free_subvolume_writers(root
->subv_writers
);
3604 free_extent_buffer(root
->node
);
3605 free_extent_buffer(root
->commit_root
);
3606 kfree(root
->free_ino_ctl
);
3607 kfree(root
->free_ino_pinned
);
3609 btrfs_put_fs_root(root
);
3612 void btrfs_free_fs_root(struct btrfs_root
*root
)
3617 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3619 u64 root_objectid
= 0;
3620 struct btrfs_root
*gang
[8];
3623 unsigned int ret
= 0;
3627 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3628 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3629 (void **)gang
, root_objectid
,
3632 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3635 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3637 for (i
= 0; i
< ret
; i
++) {
3638 /* Avoid to grab roots in dead_roots */
3639 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3643 /* grab all the search result for later use */
3644 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3646 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3648 for (i
= 0; i
< ret
; i
++) {
3651 root_objectid
= gang
[i
]->root_key
.objectid
;
3652 err
= btrfs_orphan_cleanup(gang
[i
]);
3655 btrfs_put_fs_root(gang
[i
]);
3660 /* release the uncleaned roots due to error */
3661 for (; i
< ret
; i
++) {
3663 btrfs_put_fs_root(gang
[i
]);
3668 int btrfs_commit_super(struct btrfs_root
*root
)
3670 struct btrfs_trans_handle
*trans
;
3672 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3673 btrfs_run_delayed_iputs(root
);
3674 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3675 wake_up_process(root
->fs_info
->cleaner_kthread
);
3677 /* wait until ongoing cleanup work done */
3678 down_write(&root
->fs_info
->cleanup_work_sem
);
3679 up_write(&root
->fs_info
->cleanup_work_sem
);
3681 trans
= btrfs_join_transaction(root
);
3683 return PTR_ERR(trans
);
3684 return btrfs_commit_transaction(trans
, root
);
3687 void close_ctree(struct btrfs_root
*root
)
3689 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3692 fs_info
->closing
= 1;
3695 /* wait for the uuid_scan task to finish */
3696 down(&fs_info
->uuid_tree_rescan_sem
);
3697 /* avoid complains from lockdep et al., set sem back to initial state */
3698 up(&fs_info
->uuid_tree_rescan_sem
);
3700 /* pause restriper - we want to resume on mount */
3701 btrfs_pause_balance(fs_info
);
3703 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3705 btrfs_scrub_cancel(fs_info
);
3707 /* wait for any defraggers to finish */
3708 wait_event(fs_info
->transaction_wait
,
3709 (atomic_read(&fs_info
->defrag_running
) == 0));
3711 /* clear out the rbtree of defraggable inodes */
3712 btrfs_cleanup_defrag_inodes(fs_info
);
3714 cancel_work_sync(&fs_info
->async_reclaim_work
);
3716 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3717 ret
= btrfs_commit_super(root
);
3719 btrfs_err(fs_info
, "commit super ret %d", ret
);
3722 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3723 btrfs_error_commit_super(root
);
3725 kthread_stop(fs_info
->transaction_kthread
);
3726 kthread_stop(fs_info
->cleaner_kthread
);
3728 fs_info
->closing
= 2;
3731 btrfs_free_qgroup_config(fs_info
);
3733 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3734 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3735 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3738 btrfs_sysfs_remove_one(fs_info
);
3740 btrfs_free_fs_roots(fs_info
);
3742 btrfs_put_block_group_cache(fs_info
);
3744 btrfs_free_block_groups(fs_info
);
3747 * we must make sure there is not any read request to
3748 * submit after we stopping all workers.
3750 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3751 btrfs_stop_all_workers(fs_info
);
3754 free_root_pointers(fs_info
, 1);
3756 iput(fs_info
->btree_inode
);
3758 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3759 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3760 btrfsic_unmount(root
, fs_info
->fs_devices
);
3763 btrfs_close_devices(fs_info
->fs_devices
);
3764 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3766 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3767 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3768 percpu_counter_destroy(&fs_info
->bio_counter
);
3769 bdi_destroy(&fs_info
->bdi
);
3770 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3772 btrfs_free_stripe_hash_table(fs_info
);
3774 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3775 root
->orphan_block_rsv
= NULL
;
3778 while (!list_empty(&fs_info
->pinned_chunks
)) {
3779 struct extent_map
*em
;
3781 em
= list_first_entry(&fs_info
->pinned_chunks
,
3782 struct extent_map
, list
);
3783 list_del_init(&em
->list
);
3784 free_extent_map(em
);
3786 unlock_chunks(root
);
3789 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3793 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3795 ret
= extent_buffer_uptodate(buf
);
3799 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3800 parent_transid
, atomic
);
3806 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3808 return set_extent_buffer_uptodate(buf
);
3811 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3813 struct btrfs_root
*root
;
3814 u64 transid
= btrfs_header_generation(buf
);
3817 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3819 * This is a fast path so only do this check if we have sanity tests
3820 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3821 * outside of the sanity tests.
3823 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3826 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3827 btrfs_assert_tree_locked(buf
);
3828 if (transid
!= root
->fs_info
->generation
)
3829 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3830 "found %llu running %llu\n",
3831 buf
->start
, transid
, root
->fs_info
->generation
);
3832 was_dirty
= set_extent_buffer_dirty(buf
);
3834 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3836 root
->fs_info
->dirty_metadata_batch
);
3837 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3838 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3839 btrfs_print_leaf(root
, buf
);
3845 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3849 * looks as though older kernels can get into trouble with
3850 * this code, they end up stuck in balance_dirty_pages forever
3854 if (current
->flags
& PF_MEMALLOC
)
3858 btrfs_balance_delayed_items(root
);
3860 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3861 BTRFS_DIRTY_METADATA_THRESH
);
3863 balance_dirty_pages_ratelimited(
3864 root
->fs_info
->btree_inode
->i_mapping
);
3869 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3871 __btrfs_btree_balance_dirty(root
, 1);
3874 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3876 __btrfs_btree_balance_dirty(root
, 0);
3879 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3881 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3882 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3885 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3888 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
3891 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3892 printk(KERN_ERR
"BTRFS: tree_root level too big: %d >= %d\n",
3893 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
3896 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3897 printk(KERN_ERR
"BTRFS: chunk_root level too big: %d >= %d\n",
3898 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
3901 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
3902 printk(KERN_ERR
"BTRFS: log_root level too big: %d >= %d\n",
3903 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
3908 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3909 * items yet, they'll be verified later. Issue just a warning.
3911 if (!IS_ALIGNED(btrfs_super_root(sb
), 4096))
3912 printk(KERN_WARNING
"BTRFS: tree_root block unaligned: %llu\n",
3913 btrfs_super_root(sb
));
3914 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), 4096))
3915 printk(KERN_WARNING
"BTRFS: chunk_root block unaligned: %llu\n",
3916 btrfs_super_chunk_root(sb
));
3917 if (!IS_ALIGNED(btrfs_super_log_root(sb
), 4096))
3918 printk(KERN_WARNING
"BTRFS: log_root block unaligned: %llu\n",
3919 btrfs_super_log_root(sb
));
3922 * Check the lower bound, the alignment and other constraints are
3925 if (btrfs_super_nodesize(sb
) < 4096) {
3926 printk(KERN_ERR
"BTRFS: nodesize too small: %u < 4096\n",
3927 btrfs_super_nodesize(sb
));
3930 if (btrfs_super_sectorsize(sb
) < 4096) {
3931 printk(KERN_ERR
"BTRFS: sectorsize too small: %u < 4096\n",
3932 btrfs_super_sectorsize(sb
));
3936 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
3937 printk(KERN_ERR
"BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3938 fs_info
->fsid
, sb
->dev_item
.fsid
);
3943 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3946 if (btrfs_super_num_devices(sb
) > (1UL << 31))
3947 printk(KERN_WARNING
"BTRFS: suspicious number of devices: %llu\n",
3948 btrfs_super_num_devices(sb
));
3949 if (btrfs_super_num_devices(sb
) == 0) {
3950 printk(KERN_ERR
"BTRFS: number of devices is 0\n");
3954 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
3955 printk(KERN_ERR
"BTRFS: super offset mismatch %llu != %u\n",
3956 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
3961 * Obvious sys_chunk_array corruptions, it must hold at least one key
3964 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
3965 printk(KERN_ERR
"BTRFS: system chunk array too big %u > %u\n",
3966 btrfs_super_sys_array_size(sb
),
3967 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
3970 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
3971 + sizeof(struct btrfs_chunk
)) {
3972 printk(KERN_ERR
"BTRFS: system chunk array too small %u < %zu\n",
3973 btrfs_super_sys_array_size(sb
),
3974 sizeof(struct btrfs_disk_key
)
3975 + sizeof(struct btrfs_chunk
));
3980 * The generation is a global counter, we'll trust it more than the others
3981 * but it's still possible that it's the one that's wrong.
3983 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
3985 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3986 btrfs_super_generation(sb
), btrfs_super_chunk_root_generation(sb
));
3987 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
3988 && btrfs_super_cache_generation(sb
) != (u64
)-1)
3990 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3991 btrfs_super_generation(sb
), btrfs_super_cache_generation(sb
));
3996 static void btrfs_error_commit_super(struct btrfs_root
*root
)
3998 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3999 btrfs_run_delayed_iputs(root
);
4000 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4002 down_write(&root
->fs_info
->cleanup_work_sem
);
4003 up_write(&root
->fs_info
->cleanup_work_sem
);
4005 /* cleanup FS via transaction */
4006 btrfs_cleanup_transaction(root
);
4009 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4011 struct btrfs_ordered_extent
*ordered
;
4013 spin_lock(&root
->ordered_extent_lock
);
4015 * This will just short circuit the ordered completion stuff which will
4016 * make sure the ordered extent gets properly cleaned up.
4018 list_for_each_entry(ordered
, &root
->ordered_extents
,
4020 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4021 spin_unlock(&root
->ordered_extent_lock
);
4024 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4026 struct btrfs_root
*root
;
4027 struct list_head splice
;
4029 INIT_LIST_HEAD(&splice
);
4031 spin_lock(&fs_info
->ordered_root_lock
);
4032 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4033 while (!list_empty(&splice
)) {
4034 root
= list_first_entry(&splice
, struct btrfs_root
,
4036 list_move_tail(&root
->ordered_root
,
4037 &fs_info
->ordered_roots
);
4039 spin_unlock(&fs_info
->ordered_root_lock
);
4040 btrfs_destroy_ordered_extents(root
);
4043 spin_lock(&fs_info
->ordered_root_lock
);
4045 spin_unlock(&fs_info
->ordered_root_lock
);
4048 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4049 struct btrfs_root
*root
)
4051 struct rb_node
*node
;
4052 struct btrfs_delayed_ref_root
*delayed_refs
;
4053 struct btrfs_delayed_ref_node
*ref
;
4056 delayed_refs
= &trans
->delayed_refs
;
4058 spin_lock(&delayed_refs
->lock
);
4059 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4060 spin_unlock(&delayed_refs
->lock
);
4061 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4065 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4066 struct btrfs_delayed_ref_head
*head
;
4067 bool pin_bytes
= false;
4069 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4071 if (!mutex_trylock(&head
->mutex
)) {
4072 atomic_inc(&head
->node
.refs
);
4073 spin_unlock(&delayed_refs
->lock
);
4075 mutex_lock(&head
->mutex
);
4076 mutex_unlock(&head
->mutex
);
4077 btrfs_put_delayed_ref(&head
->node
);
4078 spin_lock(&delayed_refs
->lock
);
4081 spin_lock(&head
->lock
);
4082 while ((node
= rb_first(&head
->ref_root
)) != NULL
) {
4083 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
4086 rb_erase(&ref
->rb_node
, &head
->ref_root
);
4087 atomic_dec(&delayed_refs
->num_entries
);
4088 btrfs_put_delayed_ref(ref
);
4090 if (head
->must_insert_reserved
)
4092 btrfs_free_delayed_extent_op(head
->extent_op
);
4093 delayed_refs
->num_heads
--;
4094 if (head
->processing
== 0)
4095 delayed_refs
->num_heads_ready
--;
4096 atomic_dec(&delayed_refs
->num_entries
);
4097 head
->node
.in_tree
= 0;
4098 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4099 spin_unlock(&head
->lock
);
4100 spin_unlock(&delayed_refs
->lock
);
4101 mutex_unlock(&head
->mutex
);
4104 btrfs_pin_extent(root
, head
->node
.bytenr
,
4105 head
->node
.num_bytes
, 1);
4106 btrfs_put_delayed_ref(&head
->node
);
4108 spin_lock(&delayed_refs
->lock
);
4111 spin_unlock(&delayed_refs
->lock
);
4116 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4118 struct btrfs_inode
*btrfs_inode
;
4119 struct list_head splice
;
4121 INIT_LIST_HEAD(&splice
);
4123 spin_lock(&root
->delalloc_lock
);
4124 list_splice_init(&root
->delalloc_inodes
, &splice
);
4126 while (!list_empty(&splice
)) {
4127 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4130 list_del_init(&btrfs_inode
->delalloc_inodes
);
4131 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4132 &btrfs_inode
->runtime_flags
);
4133 spin_unlock(&root
->delalloc_lock
);
4135 btrfs_invalidate_inodes(btrfs_inode
->root
);
4137 spin_lock(&root
->delalloc_lock
);
4140 spin_unlock(&root
->delalloc_lock
);
4143 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4145 struct btrfs_root
*root
;
4146 struct list_head splice
;
4148 INIT_LIST_HEAD(&splice
);
4150 spin_lock(&fs_info
->delalloc_root_lock
);
4151 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4152 while (!list_empty(&splice
)) {
4153 root
= list_first_entry(&splice
, struct btrfs_root
,
4155 list_del_init(&root
->delalloc_root
);
4156 root
= btrfs_grab_fs_root(root
);
4158 spin_unlock(&fs_info
->delalloc_root_lock
);
4160 btrfs_destroy_delalloc_inodes(root
);
4161 btrfs_put_fs_root(root
);
4163 spin_lock(&fs_info
->delalloc_root_lock
);
4165 spin_unlock(&fs_info
->delalloc_root_lock
);
4168 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4169 struct extent_io_tree
*dirty_pages
,
4173 struct extent_buffer
*eb
;
4178 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4183 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4184 while (start
<= end
) {
4185 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4186 start
+= root
->nodesize
;
4189 wait_on_extent_buffer_writeback(eb
);
4191 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4193 clear_extent_buffer_dirty(eb
);
4194 free_extent_buffer_stale(eb
);
4201 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4202 struct extent_io_tree
*pinned_extents
)
4204 struct extent_io_tree
*unpin
;
4210 unpin
= pinned_extents
;
4213 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4214 EXTENT_DIRTY
, NULL
);
4218 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4219 btrfs_error_unpin_extent_range(root
, start
, end
);
4224 if (unpin
== &root
->fs_info
->freed_extents
[0])
4225 unpin
= &root
->fs_info
->freed_extents
[1];
4227 unpin
= &root
->fs_info
->freed_extents
[0];
4235 static void btrfs_free_pending_ordered(struct btrfs_transaction
*cur_trans
,
4236 struct btrfs_fs_info
*fs_info
)
4238 struct btrfs_ordered_extent
*ordered
;
4240 spin_lock(&fs_info
->trans_lock
);
4241 while (!list_empty(&cur_trans
->pending_ordered
)) {
4242 ordered
= list_first_entry(&cur_trans
->pending_ordered
,
4243 struct btrfs_ordered_extent
,
4245 list_del_init(&ordered
->trans_list
);
4246 spin_unlock(&fs_info
->trans_lock
);
4248 btrfs_put_ordered_extent(ordered
);
4249 spin_lock(&fs_info
->trans_lock
);
4251 spin_unlock(&fs_info
->trans_lock
);
4254 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4255 struct btrfs_root
*root
)
4257 btrfs_destroy_delayed_refs(cur_trans
, root
);
4259 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4260 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4262 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4263 wake_up(&root
->fs_info
->transaction_wait
);
4265 btrfs_free_pending_ordered(cur_trans
, root
->fs_info
);
4266 btrfs_destroy_delayed_inodes(root
);
4267 btrfs_assert_delayed_root_empty(root
);
4269 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4271 btrfs_destroy_pinned_extent(root
,
4272 root
->fs_info
->pinned_extents
);
4274 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4275 wake_up(&cur_trans
->commit_wait
);
4278 memset(cur_trans, 0, sizeof(*cur_trans));
4279 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4283 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4285 struct btrfs_transaction
*t
;
4287 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4289 spin_lock(&root
->fs_info
->trans_lock
);
4290 while (!list_empty(&root
->fs_info
->trans_list
)) {
4291 t
= list_first_entry(&root
->fs_info
->trans_list
,
4292 struct btrfs_transaction
, list
);
4293 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4294 atomic_inc(&t
->use_count
);
4295 spin_unlock(&root
->fs_info
->trans_lock
);
4296 btrfs_wait_for_commit(root
, t
->transid
);
4297 btrfs_put_transaction(t
);
4298 spin_lock(&root
->fs_info
->trans_lock
);
4301 if (t
== root
->fs_info
->running_transaction
) {
4302 t
->state
= TRANS_STATE_COMMIT_DOING
;
4303 spin_unlock(&root
->fs_info
->trans_lock
);
4305 * We wait for 0 num_writers since we don't hold a trans
4306 * handle open currently for this transaction.
4308 wait_event(t
->writer_wait
,
4309 atomic_read(&t
->num_writers
) == 0);
4311 spin_unlock(&root
->fs_info
->trans_lock
);
4313 btrfs_cleanup_one_transaction(t
, root
);
4315 spin_lock(&root
->fs_info
->trans_lock
);
4316 if (t
== root
->fs_info
->running_transaction
)
4317 root
->fs_info
->running_transaction
= NULL
;
4318 list_del_init(&t
->list
);
4319 spin_unlock(&root
->fs_info
->trans_lock
);
4321 btrfs_put_transaction(t
);
4322 trace_btrfs_transaction_commit(root
);
4323 spin_lock(&root
->fs_info
->trans_lock
);
4325 spin_unlock(&root
->fs_info
->trans_lock
);
4326 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4327 btrfs_destroy_delayed_inodes(root
);
4328 btrfs_assert_delayed_root_empty(root
);
4329 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4330 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4331 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4336 static const struct extent_io_ops btree_extent_io_ops
= {
4337 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4338 .readpage_io_failed_hook
= btree_io_failed_hook
,
4339 .submit_bio_hook
= btree_submit_bio_hook
,
4340 /* note we're sharing with inode.c for the merge bio hook */
4341 .merge_bio_hook
= btrfs_merge_bio_hook
,