2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <linux/semaphore.h>
35 #include <asm/unaligned.h>
39 #include "transaction.h"
40 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "async-thread.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "check-integrity.h"
49 #include "rcu-string.h"
50 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static 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_operations(struct btrfs_transaction
*t
,
63 struct btrfs_root
*root
);
64 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
66 struct btrfs_root
*root
);
67 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
);
68 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
69 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
70 struct extent_io_tree
*dirty_pages
,
72 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
73 struct extent_io_tree
*pinned_extents
);
74 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
75 static void btrfs_error_commit_super(struct btrfs_root
*root
);
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_fs_info
*info
;
89 struct list_head list
;
90 struct btrfs_work work
;
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
98 struct async_submit_bio
{
101 struct list_head list
;
102 extent_submit_bio_hook_t
*submit_bio_start
;
103 extent_submit_bio_hook_t
*submit_bio_done
;
106 unsigned long bio_flags
;
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
112 struct btrfs_work work
;
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 # if BTRFS_MAX_LEVEL != 8
144 static struct btrfs_lockdep_keyset
{
145 u64 id
; /* root objectid */
146 const char *name_stem
; /* lock name stem */
147 char names
[BTRFS_MAX_LEVEL
+ 1][20];
148 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
149 } btrfs_lockdep_keysets
[] = {
150 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
151 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
152 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
153 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
154 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
155 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
156 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
157 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
158 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
159 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
160 { .id
= 0, .name_stem
= "tree" },
163 void __init
btrfs_init_lockdep(void)
167 /* initialize lockdep class names */
168 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
169 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
171 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
172 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
173 "btrfs-%s-%02d", ks
->name_stem
, j
);
177 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
180 struct btrfs_lockdep_keyset
*ks
;
182 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
186 if (ks
->id
== objectid
)
189 lockdep_set_class_and_name(&eb
->lock
,
190 &ks
->keys
[level
], ks
->names
[level
]);
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
199 static struct extent_map
*btree_get_extent(struct inode
*inode
,
200 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
203 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
204 struct extent_map
*em
;
207 read_lock(&em_tree
->lock
);
208 em
= lookup_extent_mapping(em_tree
, start
, len
);
211 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
212 read_unlock(&em_tree
->lock
);
215 read_unlock(&em_tree
->lock
);
217 em
= alloc_extent_map();
219 em
= ERR_PTR(-ENOMEM
);
224 em
->block_len
= (u64
)-1;
226 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
228 write_lock(&em_tree
->lock
);
229 ret
= add_extent_mapping(em_tree
, em
, 0);
230 if (ret
== -EEXIST
) {
232 em
= lookup_extent_mapping(em_tree
, start
, len
);
239 write_unlock(&em_tree
->lock
);
245 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
247 return crc32c(seed
, data
, len
);
250 void btrfs_csum_final(u32 crc
, char *result
)
252 put_unaligned_le32(~crc
, result
);
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
259 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
262 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
265 unsigned long cur_len
;
266 unsigned long offset
= BTRFS_CSUM_SIZE
;
268 unsigned long map_start
;
269 unsigned long map_len
;
272 unsigned long inline_result
;
274 len
= buf
->len
- offset
;
276 err
= map_private_extent_buffer(buf
, offset
, 32,
277 &kaddr
, &map_start
, &map_len
);
280 cur_len
= min(len
, map_len
- (offset
- map_start
));
281 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
286 if (csum_size
> sizeof(inline_result
)) {
287 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
291 result
= (char *)&inline_result
;
294 btrfs_csum_final(crc
, result
);
297 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
300 memcpy(&found
, result
, csum_size
);
302 read_extent_buffer(buf
, &val
, 0, csum_size
);
303 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
304 "failed on %llu wanted %X found %X "
306 root
->fs_info
->sb
->s_id
, buf
->start
,
307 val
, found
, btrfs_header_level(buf
));
308 if (result
!= (char *)&inline_result
)
313 write_extent_buffer(buf
, result
, 0, csum_size
);
315 if (result
!= (char *)&inline_result
)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
327 struct extent_buffer
*eb
, u64 parent_transid
,
330 struct extent_state
*cached_state
= NULL
;
333 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
339 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
341 if (extent_buffer_uptodate(eb
) &&
342 btrfs_header_generation(eb
) == parent_transid
) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 eb
->start
, parent_transid
, btrfs_header_generation(eb
));
350 clear_extent_buffer_uptodate(eb
);
352 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
353 &cached_state
, GFP_NOFS
);
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
361 static int btrfs_check_super_csum(char *raw_disk_sb
)
363 struct btrfs_super_block
*disk_sb
=
364 (struct btrfs_super_block
*)raw_disk_sb
;
365 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
368 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
370 const int csum_size
= sizeof(crc
);
371 char result
[csum_size
];
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
378 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
379 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
380 btrfs_csum_final(crc
, result
);
382 if (memcmp(raw_disk_sb
, result
, csum_size
))
385 if (ret
&& btrfs_super_generation(disk_sb
) < 10) {
386 printk(KERN_WARNING
"btrfs: super block crcs don't match, older mkfs detected\n");
391 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
392 printk(KERN_ERR
"btrfs: unsupported checksum algorithm %u\n",
401 * helper to read a given tree block, doing retries as required when
402 * the checksums don't match and we have alternate mirrors to try.
404 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
405 struct extent_buffer
*eb
,
406 u64 start
, u64 parent_transid
)
408 struct extent_io_tree
*io_tree
;
413 int failed_mirror
= 0;
415 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
416 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
418 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
420 btree_get_extent
, mirror_num
);
422 if (!verify_parent_transid(io_tree
, eb
,
430 * This buffer's crc is fine, but its contents are corrupted, so
431 * there is no reason to read the other copies, they won't be
434 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
437 num_copies
= btrfs_num_copies(root
->fs_info
,
442 if (!failed_mirror
) {
444 failed_mirror
= eb
->read_mirror
;
448 if (mirror_num
== failed_mirror
)
451 if (mirror_num
> num_copies
)
455 if (failed
&& !ret
&& failed_mirror
)
456 repair_eb_io_failure(root
, eb
, failed_mirror
);
462 * checksum a dirty tree block before IO. This has extra checks to make sure
463 * we only fill in the checksum field in the first page of a multi-page block
466 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
468 struct extent_io_tree
*tree
;
469 u64 start
= page_offset(page
);
471 struct extent_buffer
*eb
;
473 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
475 eb
= (struct extent_buffer
*)page
->private;
476 if (page
!= eb
->pages
[0])
478 found_start
= btrfs_header_bytenr(eb
);
479 if (found_start
!= start
) {
483 if (!PageUptodate(page
)) {
487 csum_tree_block(root
, eb
, 0);
491 static int check_tree_block_fsid(struct btrfs_root
*root
,
492 struct extent_buffer
*eb
)
494 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
495 u8 fsid
[BTRFS_UUID_SIZE
];
498 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(eb
), BTRFS_FSID_SIZE
);
500 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
504 fs_devices
= fs_devices
->seed
;
509 #define CORRUPT(reason, eb, root, slot) \
510 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
511 "root=%llu, slot=%d\n", reason, \
512 btrfs_header_bytenr(eb), root->objectid, slot)
514 static noinline
int check_leaf(struct btrfs_root
*root
,
515 struct extent_buffer
*leaf
)
517 struct btrfs_key key
;
518 struct btrfs_key leaf_key
;
519 u32 nritems
= btrfs_header_nritems(leaf
);
525 /* Check the 0 item */
526 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
527 BTRFS_LEAF_DATA_SIZE(root
)) {
528 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
533 * Check to make sure each items keys are in the correct order and their
534 * offsets make sense. We only have to loop through nritems-1 because
535 * we check the current slot against the next slot, which verifies the
536 * next slot's offset+size makes sense and that the current's slot
539 for (slot
= 0; slot
< nritems
- 1; slot
++) {
540 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
541 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
543 /* Make sure the keys are in the right order */
544 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
545 CORRUPT("bad key order", leaf
, root
, slot
);
550 * Make sure the offset and ends are right, remember that the
551 * item data starts at the end of the leaf and grows towards the
554 if (btrfs_item_offset_nr(leaf
, slot
) !=
555 btrfs_item_end_nr(leaf
, slot
+ 1)) {
556 CORRUPT("slot offset bad", leaf
, root
, slot
);
561 * Check to make sure that we don't point outside of the leaf,
562 * just incase all the items are consistent to eachother, but
563 * all point outside of the leaf.
565 if (btrfs_item_end_nr(leaf
, slot
) >
566 BTRFS_LEAF_DATA_SIZE(root
)) {
567 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
575 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
576 u64 phy_offset
, struct page
*page
,
577 u64 start
, u64 end
, int mirror
)
579 struct extent_io_tree
*tree
;
582 struct extent_buffer
*eb
;
583 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
590 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
591 eb
= (struct extent_buffer
*)page
->private;
593 /* the pending IO might have been the only thing that kept this buffer
594 * in memory. Make sure we have a ref for all this other checks
596 extent_buffer_get(eb
);
598 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
602 eb
->read_mirror
= mirror
;
603 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
608 found_start
= btrfs_header_bytenr(eb
);
609 if (found_start
!= eb
->start
) {
610 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
612 found_start
, eb
->start
);
616 if (check_tree_block_fsid(root
, eb
)) {
617 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
622 found_level
= btrfs_header_level(eb
);
623 if (found_level
>= BTRFS_MAX_LEVEL
) {
624 btrfs_info(root
->fs_info
, "bad tree block level %d\n",
625 (int)btrfs_header_level(eb
));
630 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
633 ret
= csum_tree_block(root
, eb
, 1);
640 * If this is a leaf block and it is corrupt, set the corrupt bit so
641 * that we don't try and read the other copies of this block, just
644 if (found_level
== 0 && check_leaf(root
, eb
)) {
645 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
650 set_extent_buffer_uptodate(eb
);
653 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
654 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
658 * our io error hook is going to dec the io pages
659 * again, we have to make sure it has something
662 atomic_inc(&eb
->io_pages
);
663 clear_extent_buffer_uptodate(eb
);
665 free_extent_buffer(eb
);
670 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
672 struct extent_buffer
*eb
;
673 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
675 eb
= (struct extent_buffer
*)page
->private;
676 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
677 eb
->read_mirror
= failed_mirror
;
678 atomic_dec(&eb
->io_pages
);
679 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
680 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
681 return -EIO
; /* we fixed nothing */
684 static void end_workqueue_bio(struct bio
*bio
, int err
)
686 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
687 struct btrfs_fs_info
*fs_info
;
689 fs_info
= end_io_wq
->info
;
690 end_io_wq
->error
= err
;
691 end_io_wq
->work
.func
= end_workqueue_fn
;
692 end_io_wq
->work
.flags
= 0;
694 if (bio
->bi_rw
& REQ_WRITE
) {
695 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
696 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
698 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
699 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
701 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
702 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
705 btrfs_queue_worker(&fs_info
->endio_write_workers
,
708 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
709 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
711 else if (end_io_wq
->metadata
)
712 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
715 btrfs_queue_worker(&fs_info
->endio_workers
,
721 * For the metadata arg you want
724 * 1 - if normal metadta
725 * 2 - if writing to the free space cache area
726 * 3 - raid parity work
728 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
731 struct end_io_wq
*end_io_wq
;
732 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
736 end_io_wq
->private = bio
->bi_private
;
737 end_io_wq
->end_io
= bio
->bi_end_io
;
738 end_io_wq
->info
= info
;
739 end_io_wq
->error
= 0;
740 end_io_wq
->bio
= bio
;
741 end_io_wq
->metadata
= metadata
;
743 bio
->bi_private
= end_io_wq
;
744 bio
->bi_end_io
= end_workqueue_bio
;
748 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
750 unsigned long limit
= min_t(unsigned long,
751 info
->workers
.max_workers
,
752 info
->fs_devices
->open_devices
);
756 static void run_one_async_start(struct btrfs_work
*work
)
758 struct async_submit_bio
*async
;
761 async
= container_of(work
, struct async_submit_bio
, work
);
762 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
763 async
->mirror_num
, async
->bio_flags
,
769 static void run_one_async_done(struct btrfs_work
*work
)
771 struct btrfs_fs_info
*fs_info
;
772 struct async_submit_bio
*async
;
775 async
= container_of(work
, struct async_submit_bio
, work
);
776 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
778 limit
= btrfs_async_submit_limit(fs_info
);
779 limit
= limit
* 2 / 3;
781 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
782 waitqueue_active(&fs_info
->async_submit_wait
))
783 wake_up(&fs_info
->async_submit_wait
);
785 /* If an error occured we just want to clean up the bio and move on */
787 bio_endio(async
->bio
, async
->error
);
791 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
792 async
->mirror_num
, async
->bio_flags
,
796 static void run_one_async_free(struct btrfs_work
*work
)
798 struct async_submit_bio
*async
;
800 async
= container_of(work
, struct async_submit_bio
, work
);
804 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
805 int rw
, struct bio
*bio
, int mirror_num
,
806 unsigned long bio_flags
,
808 extent_submit_bio_hook_t
*submit_bio_start
,
809 extent_submit_bio_hook_t
*submit_bio_done
)
811 struct async_submit_bio
*async
;
813 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
817 async
->inode
= inode
;
820 async
->mirror_num
= mirror_num
;
821 async
->submit_bio_start
= submit_bio_start
;
822 async
->submit_bio_done
= submit_bio_done
;
824 async
->work
.func
= run_one_async_start
;
825 async
->work
.ordered_func
= run_one_async_done
;
826 async
->work
.ordered_free
= run_one_async_free
;
828 async
->work
.flags
= 0;
829 async
->bio_flags
= bio_flags
;
830 async
->bio_offset
= bio_offset
;
834 atomic_inc(&fs_info
->nr_async_submits
);
837 btrfs_set_work_high_prio(&async
->work
);
839 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
841 while (atomic_read(&fs_info
->async_submit_draining
) &&
842 atomic_read(&fs_info
->nr_async_submits
)) {
843 wait_event(fs_info
->async_submit_wait
,
844 (atomic_read(&fs_info
->nr_async_submits
) == 0));
850 static int btree_csum_one_bio(struct bio
*bio
)
852 struct bio_vec
*bvec
= bio
->bi_io_vec
;
854 struct btrfs_root
*root
;
857 WARN_ON(bio
->bi_vcnt
<= 0);
858 while (bio_index
< bio
->bi_vcnt
) {
859 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
860 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
869 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
870 struct bio
*bio
, int mirror_num
,
871 unsigned long bio_flags
,
875 * when we're called for a write, we're already in the async
876 * submission context. Just jump into btrfs_map_bio
878 return btree_csum_one_bio(bio
);
881 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
882 int mirror_num
, unsigned long bio_flags
,
888 * when we're called for a write, we're already in the async
889 * submission context. Just jump into btrfs_map_bio
891 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
897 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
899 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
908 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
909 int mirror_num
, unsigned long bio_flags
,
912 int async
= check_async_write(inode
, bio_flags
);
915 if (!(rw
& REQ_WRITE
)) {
917 * called for a read, do the setup so that checksum validation
918 * can happen in the async kernel threads
920 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
924 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
927 ret
= btree_csum_one_bio(bio
);
930 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
934 * kthread helpers are used to submit writes so that
935 * checksumming can happen in parallel across all CPUs
937 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
938 inode
, rw
, bio
, mirror_num
, 0,
940 __btree_submit_bio_start
,
941 __btree_submit_bio_done
);
951 #ifdef CONFIG_MIGRATION
952 static int btree_migratepage(struct address_space
*mapping
,
953 struct page
*newpage
, struct page
*page
,
954 enum migrate_mode mode
)
957 * we can't safely write a btree page from here,
958 * we haven't done the locking hook
963 * Buffers may be managed in a filesystem specific way.
964 * We must have no buffers or drop them.
966 if (page_has_private(page
) &&
967 !try_to_release_page(page
, GFP_KERNEL
))
969 return migrate_page(mapping
, newpage
, page
, mode
);
974 static int btree_writepages(struct address_space
*mapping
,
975 struct writeback_control
*wbc
)
977 struct extent_io_tree
*tree
;
978 struct btrfs_fs_info
*fs_info
;
981 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
982 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
984 if (wbc
->for_kupdate
)
987 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
988 /* this is a bit racy, but that's ok */
989 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
990 BTRFS_DIRTY_METADATA_THRESH
);
994 return btree_write_cache_pages(mapping
, wbc
);
997 static int btree_readpage(struct file
*file
, struct page
*page
)
999 struct extent_io_tree
*tree
;
1000 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1001 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1004 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1006 if (PageWriteback(page
) || PageDirty(page
))
1009 return try_release_extent_buffer(page
);
1012 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1013 unsigned int length
)
1015 struct extent_io_tree
*tree
;
1016 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1017 extent_invalidatepage(tree
, page
, offset
);
1018 btree_releasepage(page
, GFP_NOFS
);
1019 if (PagePrivate(page
)) {
1020 printk(KERN_WARNING
"btrfs warning page private not zero "
1021 "on page %llu\n", (unsigned long long)page_offset(page
));
1022 ClearPagePrivate(page
);
1023 set_page_private(page
, 0);
1024 page_cache_release(page
);
1028 static int btree_set_page_dirty(struct page
*page
)
1031 struct extent_buffer
*eb
;
1033 BUG_ON(!PagePrivate(page
));
1034 eb
= (struct extent_buffer
*)page
->private;
1036 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1037 BUG_ON(!atomic_read(&eb
->refs
));
1038 btrfs_assert_tree_locked(eb
);
1040 return __set_page_dirty_nobuffers(page
);
1043 static const struct address_space_operations btree_aops
= {
1044 .readpage
= btree_readpage
,
1045 .writepages
= btree_writepages
,
1046 .releasepage
= btree_releasepage
,
1047 .invalidatepage
= btree_invalidatepage
,
1048 #ifdef CONFIG_MIGRATION
1049 .migratepage
= btree_migratepage
,
1051 .set_page_dirty
= btree_set_page_dirty
,
1054 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1057 struct extent_buffer
*buf
= NULL
;
1058 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1061 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1064 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1065 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1066 free_extent_buffer(buf
);
1070 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1071 int mirror_num
, struct extent_buffer
**eb
)
1073 struct extent_buffer
*buf
= NULL
;
1074 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1075 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1078 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1082 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1084 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1085 btree_get_extent
, mirror_num
);
1087 free_extent_buffer(buf
);
1091 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1092 free_extent_buffer(buf
);
1094 } else if (extent_buffer_uptodate(buf
)) {
1097 free_extent_buffer(buf
);
1102 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1103 u64 bytenr
, u32 blocksize
)
1105 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1106 struct extent_buffer
*eb
;
1107 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1112 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1113 u64 bytenr
, u32 blocksize
)
1115 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1116 struct extent_buffer
*eb
;
1118 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1124 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1126 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1127 buf
->start
+ buf
->len
- 1);
1130 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1132 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1133 buf
->start
, buf
->start
+ buf
->len
- 1);
1136 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1137 u32 blocksize
, u64 parent_transid
)
1139 struct extent_buffer
*buf
= NULL
;
1142 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1146 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1148 free_extent_buffer(buf
);
1155 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1156 struct extent_buffer
*buf
)
1158 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1160 if (btrfs_header_generation(buf
) ==
1161 fs_info
->running_transaction
->transid
) {
1162 btrfs_assert_tree_locked(buf
);
1164 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1165 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1167 fs_info
->dirty_metadata_batch
);
1168 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1169 btrfs_set_lock_blocking(buf
);
1170 clear_extent_buffer_dirty(buf
);
1175 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1176 u32 stripesize
, struct btrfs_root
*root
,
1177 struct btrfs_fs_info
*fs_info
,
1181 root
->commit_root
= NULL
;
1182 root
->sectorsize
= sectorsize
;
1183 root
->nodesize
= nodesize
;
1184 root
->leafsize
= leafsize
;
1185 root
->stripesize
= stripesize
;
1187 root
->track_dirty
= 0;
1189 root
->orphan_item_inserted
= 0;
1190 root
->orphan_cleanup_state
= 0;
1192 root
->objectid
= objectid
;
1193 root
->last_trans
= 0;
1194 root
->highest_objectid
= 0;
1195 root
->nr_delalloc_inodes
= 0;
1196 root
->nr_ordered_extents
= 0;
1198 root
->inode_tree
= RB_ROOT
;
1199 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1200 root
->block_rsv
= NULL
;
1201 root
->orphan_block_rsv
= NULL
;
1203 INIT_LIST_HEAD(&root
->dirty_list
);
1204 INIT_LIST_HEAD(&root
->root_list
);
1205 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1206 INIT_LIST_HEAD(&root
->delalloc_root
);
1207 INIT_LIST_HEAD(&root
->ordered_extents
);
1208 INIT_LIST_HEAD(&root
->ordered_root
);
1209 INIT_LIST_HEAD(&root
->logged_list
[0]);
1210 INIT_LIST_HEAD(&root
->logged_list
[1]);
1211 spin_lock_init(&root
->orphan_lock
);
1212 spin_lock_init(&root
->inode_lock
);
1213 spin_lock_init(&root
->delalloc_lock
);
1214 spin_lock_init(&root
->ordered_extent_lock
);
1215 spin_lock_init(&root
->accounting_lock
);
1216 spin_lock_init(&root
->log_extents_lock
[0]);
1217 spin_lock_init(&root
->log_extents_lock
[1]);
1218 mutex_init(&root
->objectid_mutex
);
1219 mutex_init(&root
->log_mutex
);
1220 init_waitqueue_head(&root
->log_writer_wait
);
1221 init_waitqueue_head(&root
->log_commit_wait
[0]);
1222 init_waitqueue_head(&root
->log_commit_wait
[1]);
1223 atomic_set(&root
->log_commit
[0], 0);
1224 atomic_set(&root
->log_commit
[1], 0);
1225 atomic_set(&root
->log_writers
, 0);
1226 atomic_set(&root
->log_batch
, 0);
1227 atomic_set(&root
->orphan_inodes
, 0);
1228 atomic_set(&root
->refs
, 1);
1229 root
->log_transid
= 0;
1230 root
->last_log_commit
= 0;
1231 extent_io_tree_init(&root
->dirty_log_pages
,
1232 fs_info
->btree_inode
->i_mapping
);
1234 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1235 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1236 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1237 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1238 root
->defrag_trans_start
= fs_info
->generation
;
1239 init_completion(&root
->kobj_unregister
);
1240 root
->defrag_running
= 0;
1241 root
->root_key
.objectid
= objectid
;
1244 spin_lock_init(&root
->root_item_lock
);
1247 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1249 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1251 root
->fs_info
= fs_info
;
1255 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1256 struct btrfs_fs_info
*fs_info
,
1259 struct extent_buffer
*leaf
;
1260 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1261 struct btrfs_root
*root
;
1262 struct btrfs_key key
;
1267 root
= btrfs_alloc_root(fs_info
);
1269 return ERR_PTR(-ENOMEM
);
1271 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1272 tree_root
->sectorsize
, tree_root
->stripesize
,
1273 root
, fs_info
, objectid
);
1274 root
->root_key
.objectid
= objectid
;
1275 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1276 root
->root_key
.offset
= 0;
1278 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1279 0, objectid
, NULL
, 0, 0, 0);
1281 ret
= PTR_ERR(leaf
);
1286 bytenr
= leaf
->start
;
1287 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1288 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1289 btrfs_set_header_generation(leaf
, trans
->transid
);
1290 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1291 btrfs_set_header_owner(leaf
, objectid
);
1294 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(leaf
),
1296 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1297 btrfs_header_chunk_tree_uuid(leaf
),
1299 btrfs_mark_buffer_dirty(leaf
);
1301 root
->commit_root
= btrfs_root_node(root
);
1302 root
->track_dirty
= 1;
1305 root
->root_item
.flags
= 0;
1306 root
->root_item
.byte_limit
= 0;
1307 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1308 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1309 btrfs_set_root_level(&root
->root_item
, 0);
1310 btrfs_set_root_refs(&root
->root_item
, 1);
1311 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1312 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1313 btrfs_set_root_dirid(&root
->root_item
, 0);
1315 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1316 root
->root_item
.drop_level
= 0;
1318 key
.objectid
= objectid
;
1319 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1321 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1325 btrfs_tree_unlock(leaf
);
1331 btrfs_tree_unlock(leaf
);
1332 free_extent_buffer(leaf
);
1336 return ERR_PTR(ret
);
1339 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1340 struct btrfs_fs_info
*fs_info
)
1342 struct btrfs_root
*root
;
1343 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1344 struct extent_buffer
*leaf
;
1346 root
= btrfs_alloc_root(fs_info
);
1348 return ERR_PTR(-ENOMEM
);
1350 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1351 tree_root
->sectorsize
, tree_root
->stripesize
,
1352 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1354 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1355 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1356 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1358 * log trees do not get reference counted because they go away
1359 * before a real commit is actually done. They do store pointers
1360 * to file data extents, and those reference counts still get
1361 * updated (along with back refs to the log tree).
1365 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1366 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1370 return ERR_CAST(leaf
);
1373 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1374 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1375 btrfs_set_header_generation(leaf
, trans
->transid
);
1376 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1377 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1380 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1381 btrfs_header_fsid(root
->node
), BTRFS_FSID_SIZE
);
1382 btrfs_mark_buffer_dirty(root
->node
);
1383 btrfs_tree_unlock(root
->node
);
1387 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1388 struct btrfs_fs_info
*fs_info
)
1390 struct btrfs_root
*log_root
;
1392 log_root
= alloc_log_tree(trans
, fs_info
);
1393 if (IS_ERR(log_root
))
1394 return PTR_ERR(log_root
);
1395 WARN_ON(fs_info
->log_root_tree
);
1396 fs_info
->log_root_tree
= log_root
;
1400 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1401 struct btrfs_root
*root
)
1403 struct btrfs_root
*log_root
;
1404 struct btrfs_inode_item
*inode_item
;
1406 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1407 if (IS_ERR(log_root
))
1408 return PTR_ERR(log_root
);
1410 log_root
->last_trans
= trans
->transid
;
1411 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1413 inode_item
= &log_root
->root_item
.inode
;
1414 btrfs_set_stack_inode_generation(inode_item
, 1);
1415 btrfs_set_stack_inode_size(inode_item
, 3);
1416 btrfs_set_stack_inode_nlink(inode_item
, 1);
1417 btrfs_set_stack_inode_nbytes(inode_item
, root
->leafsize
);
1418 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1420 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1422 WARN_ON(root
->log_root
);
1423 root
->log_root
= log_root
;
1424 root
->log_transid
= 0;
1425 root
->last_log_commit
= 0;
1429 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1430 struct btrfs_key
*key
)
1432 struct btrfs_root
*root
;
1433 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1434 struct btrfs_path
*path
;
1439 path
= btrfs_alloc_path();
1441 return ERR_PTR(-ENOMEM
);
1443 root
= btrfs_alloc_root(fs_info
);
1449 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1450 tree_root
->sectorsize
, tree_root
->stripesize
,
1451 root
, fs_info
, key
->objectid
);
1453 ret
= btrfs_find_root(tree_root
, key
, path
,
1454 &root
->root_item
, &root
->root_key
);
1461 generation
= btrfs_root_generation(&root
->root_item
);
1462 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1463 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1464 blocksize
, generation
);
1468 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1472 root
->commit_root
= btrfs_root_node(root
);
1474 btrfs_free_path(path
);
1478 free_extent_buffer(root
->node
);
1482 root
= ERR_PTR(ret
);
1486 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1487 struct btrfs_key
*location
)
1489 struct btrfs_root
*root
;
1491 root
= btrfs_read_tree_root(tree_root
, location
);
1495 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1497 btrfs_check_and_init_root_item(&root
->root_item
);
1503 int btrfs_init_fs_root(struct btrfs_root
*root
)
1507 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1508 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1510 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1515 btrfs_init_free_ino_ctl(root
);
1516 mutex_init(&root
->fs_commit_mutex
);
1517 spin_lock_init(&root
->cache_lock
);
1518 init_waitqueue_head(&root
->cache_wait
);
1520 ret
= get_anon_bdev(&root
->anon_dev
);
1525 kfree(root
->free_ino_ctl
);
1526 kfree(root
->free_ino_pinned
);
1530 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1533 struct btrfs_root
*root
;
1535 spin_lock(&fs_info
->fs_roots_radix_lock
);
1536 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1537 (unsigned long)root_id
);
1538 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1542 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1543 struct btrfs_root
*root
)
1547 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1551 spin_lock(&fs_info
->fs_roots_radix_lock
);
1552 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1553 (unsigned long)root
->root_key
.objectid
,
1557 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1558 radix_tree_preload_end();
1563 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1564 struct btrfs_key
*location
)
1566 struct btrfs_root
*root
;
1569 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1570 return fs_info
->tree_root
;
1571 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1572 return fs_info
->extent_root
;
1573 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1574 return fs_info
->chunk_root
;
1575 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1576 return fs_info
->dev_root
;
1577 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1578 return fs_info
->csum_root
;
1579 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1580 return fs_info
->quota_root
? fs_info
->quota_root
:
1582 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1583 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1586 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1588 if (btrfs_root_refs(&root
->root_item
) == 0)
1589 return ERR_PTR(-ENOENT
);
1593 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1597 if (btrfs_root_refs(&root
->root_item
) == 0) {
1602 ret
= btrfs_init_fs_root(root
);
1606 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1610 root
->orphan_item_inserted
= 1;
1612 ret
= btrfs_insert_fs_root(fs_info
, root
);
1614 if (ret
== -EEXIST
) {
1623 return ERR_PTR(ret
);
1626 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1628 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1630 struct btrfs_device
*device
;
1631 struct backing_dev_info
*bdi
;
1634 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1637 bdi
= blk_get_backing_dev_info(device
->bdev
);
1638 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1648 * If this fails, caller must call bdi_destroy() to get rid of the
1651 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1655 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1656 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1660 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1661 bdi
->congested_fn
= btrfs_congested_fn
;
1662 bdi
->congested_data
= info
;
1667 * called by the kthread helper functions to finally call the bio end_io
1668 * functions. This is where read checksum verification actually happens
1670 static void end_workqueue_fn(struct btrfs_work
*work
)
1673 struct end_io_wq
*end_io_wq
;
1674 struct btrfs_fs_info
*fs_info
;
1677 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1678 bio
= end_io_wq
->bio
;
1679 fs_info
= end_io_wq
->info
;
1681 error
= end_io_wq
->error
;
1682 bio
->bi_private
= end_io_wq
->private;
1683 bio
->bi_end_io
= end_io_wq
->end_io
;
1685 bio_endio(bio
, error
);
1688 static int cleaner_kthread(void *arg
)
1690 struct btrfs_root
*root
= arg
;
1696 /* Make the cleaner go to sleep early. */
1697 if (btrfs_need_cleaner_sleep(root
))
1700 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1704 * Avoid the problem that we change the status of the fs
1705 * during the above check and trylock.
1707 if (btrfs_need_cleaner_sleep(root
)) {
1708 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1712 btrfs_run_delayed_iputs(root
);
1713 again
= btrfs_clean_one_deleted_snapshot(root
);
1714 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1717 * The defragger has dealt with the R/O remount and umount,
1718 * needn't do anything special here.
1720 btrfs_run_defrag_inodes(root
->fs_info
);
1722 if (!try_to_freeze() && !again
) {
1723 set_current_state(TASK_INTERRUPTIBLE
);
1724 if (!kthread_should_stop())
1726 __set_current_state(TASK_RUNNING
);
1728 } while (!kthread_should_stop());
1732 static int transaction_kthread(void *arg
)
1734 struct btrfs_root
*root
= arg
;
1735 struct btrfs_trans_handle
*trans
;
1736 struct btrfs_transaction
*cur
;
1739 unsigned long delay
;
1743 cannot_commit
= false;
1744 delay
= HZ
* root
->fs_info
->commit_interval
;
1745 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1747 spin_lock(&root
->fs_info
->trans_lock
);
1748 cur
= root
->fs_info
->running_transaction
;
1750 spin_unlock(&root
->fs_info
->trans_lock
);
1754 now
= get_seconds();
1755 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1756 (now
< cur
->start_time
||
1757 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1758 spin_unlock(&root
->fs_info
->trans_lock
);
1762 transid
= cur
->transid
;
1763 spin_unlock(&root
->fs_info
->trans_lock
);
1765 /* If the file system is aborted, this will always fail. */
1766 trans
= btrfs_attach_transaction(root
);
1767 if (IS_ERR(trans
)) {
1768 if (PTR_ERR(trans
) != -ENOENT
)
1769 cannot_commit
= true;
1772 if (transid
== trans
->transid
) {
1773 btrfs_commit_transaction(trans
, root
);
1775 btrfs_end_transaction(trans
, root
);
1778 wake_up_process(root
->fs_info
->cleaner_kthread
);
1779 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1781 if (!try_to_freeze()) {
1782 set_current_state(TASK_INTERRUPTIBLE
);
1783 if (!kthread_should_stop() &&
1784 (!btrfs_transaction_blocked(root
->fs_info
) ||
1786 schedule_timeout(delay
);
1787 __set_current_state(TASK_RUNNING
);
1789 } while (!kthread_should_stop());
1794 * this will find the highest generation in the array of
1795 * root backups. The index of the highest array is returned,
1796 * or -1 if we can't find anything.
1798 * We check to make sure the array is valid by comparing the
1799 * generation of the latest root in the array with the generation
1800 * in the super block. If they don't match we pitch it.
1802 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1805 int newest_index
= -1;
1806 struct btrfs_root_backup
*root_backup
;
1809 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1810 root_backup
= info
->super_copy
->super_roots
+ i
;
1811 cur
= btrfs_backup_tree_root_gen(root_backup
);
1812 if (cur
== newest_gen
)
1816 /* check to see if we actually wrapped around */
1817 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1818 root_backup
= info
->super_copy
->super_roots
;
1819 cur
= btrfs_backup_tree_root_gen(root_backup
);
1820 if (cur
== newest_gen
)
1823 return newest_index
;
1828 * find the oldest backup so we know where to store new entries
1829 * in the backup array. This will set the backup_root_index
1830 * field in the fs_info struct
1832 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1835 int newest_index
= -1;
1837 newest_index
= find_newest_super_backup(info
, newest_gen
);
1838 /* if there was garbage in there, just move along */
1839 if (newest_index
== -1) {
1840 info
->backup_root_index
= 0;
1842 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1847 * copy all the root pointers into the super backup array.
1848 * this will bump the backup pointer by one when it is
1851 static void backup_super_roots(struct btrfs_fs_info
*info
)
1854 struct btrfs_root_backup
*root_backup
;
1857 next_backup
= info
->backup_root_index
;
1858 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1859 BTRFS_NUM_BACKUP_ROOTS
;
1862 * just overwrite the last backup if we're at the same generation
1863 * this happens only at umount
1865 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1866 if (btrfs_backup_tree_root_gen(root_backup
) ==
1867 btrfs_header_generation(info
->tree_root
->node
))
1868 next_backup
= last_backup
;
1870 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1873 * make sure all of our padding and empty slots get zero filled
1874 * regardless of which ones we use today
1876 memset(root_backup
, 0, sizeof(*root_backup
));
1878 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1880 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1881 btrfs_set_backup_tree_root_gen(root_backup
,
1882 btrfs_header_generation(info
->tree_root
->node
));
1884 btrfs_set_backup_tree_root_level(root_backup
,
1885 btrfs_header_level(info
->tree_root
->node
));
1887 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1888 btrfs_set_backup_chunk_root_gen(root_backup
,
1889 btrfs_header_generation(info
->chunk_root
->node
));
1890 btrfs_set_backup_chunk_root_level(root_backup
,
1891 btrfs_header_level(info
->chunk_root
->node
));
1893 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1894 btrfs_set_backup_extent_root_gen(root_backup
,
1895 btrfs_header_generation(info
->extent_root
->node
));
1896 btrfs_set_backup_extent_root_level(root_backup
,
1897 btrfs_header_level(info
->extent_root
->node
));
1900 * we might commit during log recovery, which happens before we set
1901 * the fs_root. Make sure it is valid before we fill it in.
1903 if (info
->fs_root
&& info
->fs_root
->node
) {
1904 btrfs_set_backup_fs_root(root_backup
,
1905 info
->fs_root
->node
->start
);
1906 btrfs_set_backup_fs_root_gen(root_backup
,
1907 btrfs_header_generation(info
->fs_root
->node
));
1908 btrfs_set_backup_fs_root_level(root_backup
,
1909 btrfs_header_level(info
->fs_root
->node
));
1912 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1913 btrfs_set_backup_dev_root_gen(root_backup
,
1914 btrfs_header_generation(info
->dev_root
->node
));
1915 btrfs_set_backup_dev_root_level(root_backup
,
1916 btrfs_header_level(info
->dev_root
->node
));
1918 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1919 btrfs_set_backup_csum_root_gen(root_backup
,
1920 btrfs_header_generation(info
->csum_root
->node
));
1921 btrfs_set_backup_csum_root_level(root_backup
,
1922 btrfs_header_level(info
->csum_root
->node
));
1924 btrfs_set_backup_total_bytes(root_backup
,
1925 btrfs_super_total_bytes(info
->super_copy
));
1926 btrfs_set_backup_bytes_used(root_backup
,
1927 btrfs_super_bytes_used(info
->super_copy
));
1928 btrfs_set_backup_num_devices(root_backup
,
1929 btrfs_super_num_devices(info
->super_copy
));
1932 * if we don't copy this out to the super_copy, it won't get remembered
1933 * for the next commit
1935 memcpy(&info
->super_copy
->super_roots
,
1936 &info
->super_for_commit
->super_roots
,
1937 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1941 * this copies info out of the root backup array and back into
1942 * the in-memory super block. It is meant to help iterate through
1943 * the array, so you send it the number of backups you've already
1944 * tried and the last backup index you used.
1946 * this returns -1 when it has tried all the backups
1948 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1949 struct btrfs_super_block
*super
,
1950 int *num_backups_tried
, int *backup_index
)
1952 struct btrfs_root_backup
*root_backup
;
1953 int newest
= *backup_index
;
1955 if (*num_backups_tried
== 0) {
1956 u64 gen
= btrfs_super_generation(super
);
1958 newest
= find_newest_super_backup(info
, gen
);
1962 *backup_index
= newest
;
1963 *num_backups_tried
= 1;
1964 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1965 /* we've tried all the backups, all done */
1968 /* jump to the next oldest backup */
1969 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1970 BTRFS_NUM_BACKUP_ROOTS
;
1971 *backup_index
= newest
;
1972 *num_backups_tried
+= 1;
1974 root_backup
= super
->super_roots
+ newest
;
1976 btrfs_set_super_generation(super
,
1977 btrfs_backup_tree_root_gen(root_backup
));
1978 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1979 btrfs_set_super_root_level(super
,
1980 btrfs_backup_tree_root_level(root_backup
));
1981 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1984 * fixme: the total bytes and num_devices need to match or we should
1987 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1988 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1992 /* helper to cleanup workers */
1993 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
1995 btrfs_stop_workers(&fs_info
->generic_worker
);
1996 btrfs_stop_workers(&fs_info
->fixup_workers
);
1997 btrfs_stop_workers(&fs_info
->delalloc_workers
);
1998 btrfs_stop_workers(&fs_info
->workers
);
1999 btrfs_stop_workers(&fs_info
->endio_workers
);
2000 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2001 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
2002 btrfs_stop_workers(&fs_info
->rmw_workers
);
2003 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2004 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2005 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2006 btrfs_stop_workers(&fs_info
->submit_workers
);
2007 btrfs_stop_workers(&fs_info
->delayed_workers
);
2008 btrfs_stop_workers(&fs_info
->caching_workers
);
2009 btrfs_stop_workers(&fs_info
->readahead_workers
);
2010 btrfs_stop_workers(&fs_info
->flush_workers
);
2011 btrfs_stop_workers(&fs_info
->qgroup_rescan_workers
);
2014 /* helper to cleanup tree roots */
2015 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2017 free_extent_buffer(info
->tree_root
->node
);
2018 free_extent_buffer(info
->tree_root
->commit_root
);
2019 info
->tree_root
->node
= NULL
;
2020 info
->tree_root
->commit_root
= NULL
;
2022 if (info
->dev_root
) {
2023 free_extent_buffer(info
->dev_root
->node
);
2024 free_extent_buffer(info
->dev_root
->commit_root
);
2025 info
->dev_root
->node
= NULL
;
2026 info
->dev_root
->commit_root
= NULL
;
2028 if (info
->extent_root
) {
2029 free_extent_buffer(info
->extent_root
->node
);
2030 free_extent_buffer(info
->extent_root
->commit_root
);
2031 info
->extent_root
->node
= NULL
;
2032 info
->extent_root
->commit_root
= NULL
;
2034 if (info
->csum_root
) {
2035 free_extent_buffer(info
->csum_root
->node
);
2036 free_extent_buffer(info
->csum_root
->commit_root
);
2037 info
->csum_root
->node
= NULL
;
2038 info
->csum_root
->commit_root
= NULL
;
2040 if (info
->quota_root
) {
2041 free_extent_buffer(info
->quota_root
->node
);
2042 free_extent_buffer(info
->quota_root
->commit_root
);
2043 info
->quota_root
->node
= NULL
;
2044 info
->quota_root
->commit_root
= NULL
;
2046 if (info
->uuid_root
) {
2047 free_extent_buffer(info
->uuid_root
->node
);
2048 free_extent_buffer(info
->uuid_root
->commit_root
);
2049 info
->uuid_root
->node
= NULL
;
2050 info
->uuid_root
->commit_root
= NULL
;
2053 free_extent_buffer(info
->chunk_root
->node
);
2054 free_extent_buffer(info
->chunk_root
->commit_root
);
2055 info
->chunk_root
->node
= NULL
;
2056 info
->chunk_root
->commit_root
= NULL
;
2060 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
2063 struct btrfs_root
*gang
[8];
2066 while (!list_empty(&fs_info
->dead_roots
)) {
2067 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2068 struct btrfs_root
, root_list
);
2069 list_del(&gang
[0]->root_list
);
2071 if (gang
[0]->in_radix
) {
2072 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2074 free_extent_buffer(gang
[0]->node
);
2075 free_extent_buffer(gang
[0]->commit_root
);
2076 btrfs_put_fs_root(gang
[0]);
2081 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2086 for (i
= 0; i
< ret
; i
++)
2087 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2091 int open_ctree(struct super_block
*sb
,
2092 struct btrfs_fs_devices
*fs_devices
,
2102 struct btrfs_key location
;
2103 struct buffer_head
*bh
;
2104 struct btrfs_super_block
*disk_super
;
2105 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2106 struct btrfs_root
*tree_root
;
2107 struct btrfs_root
*extent_root
;
2108 struct btrfs_root
*csum_root
;
2109 struct btrfs_root
*chunk_root
;
2110 struct btrfs_root
*dev_root
;
2111 struct btrfs_root
*quota_root
;
2112 struct btrfs_root
*uuid_root
;
2113 struct btrfs_root
*log_tree_root
;
2116 int num_backups_tried
= 0;
2117 int backup_index
= 0;
2118 bool create_uuid_tree
;
2119 bool check_uuid_tree
;
2121 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2122 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2123 if (!tree_root
|| !chunk_root
) {
2128 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2134 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2140 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2145 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2146 (1 + ilog2(nr_cpu_ids
));
2148 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2151 goto fail_dirty_metadata_bytes
;
2154 fs_info
->btree_inode
= new_inode(sb
);
2155 if (!fs_info
->btree_inode
) {
2157 goto fail_delalloc_bytes
;
2160 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2162 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2163 INIT_LIST_HEAD(&fs_info
->trans_list
);
2164 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2165 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2166 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2167 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2168 spin_lock_init(&fs_info
->delalloc_root_lock
);
2169 spin_lock_init(&fs_info
->trans_lock
);
2170 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2171 spin_lock_init(&fs_info
->delayed_iput_lock
);
2172 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2173 spin_lock_init(&fs_info
->free_chunk_lock
);
2174 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2175 spin_lock_init(&fs_info
->super_lock
);
2176 rwlock_init(&fs_info
->tree_mod_log_lock
);
2177 mutex_init(&fs_info
->reloc_mutex
);
2178 seqlock_init(&fs_info
->profiles_lock
);
2180 init_completion(&fs_info
->kobj_unregister
);
2181 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2182 INIT_LIST_HEAD(&fs_info
->space_info
);
2183 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2184 btrfs_mapping_init(&fs_info
->mapping_tree
);
2185 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2186 BTRFS_BLOCK_RSV_GLOBAL
);
2187 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2188 BTRFS_BLOCK_RSV_DELALLOC
);
2189 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2190 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2191 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2192 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2193 BTRFS_BLOCK_RSV_DELOPS
);
2194 atomic_set(&fs_info
->nr_async_submits
, 0);
2195 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2196 atomic_set(&fs_info
->async_submit_draining
, 0);
2197 atomic_set(&fs_info
->nr_async_bios
, 0);
2198 atomic_set(&fs_info
->defrag_running
, 0);
2199 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2201 fs_info
->max_inline
= 8192 * 1024;
2202 fs_info
->metadata_ratio
= 0;
2203 fs_info
->defrag_inodes
= RB_ROOT
;
2204 fs_info
->free_chunk_space
= 0;
2205 fs_info
->tree_mod_log
= RB_ROOT
;
2206 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2208 /* readahead state */
2209 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2210 spin_lock_init(&fs_info
->reada_lock
);
2212 fs_info
->thread_pool_size
= min_t(unsigned long,
2213 num_online_cpus() + 2, 8);
2215 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2216 spin_lock_init(&fs_info
->ordered_root_lock
);
2217 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2219 if (!fs_info
->delayed_root
) {
2223 btrfs_init_delayed_root(fs_info
->delayed_root
);
2225 mutex_init(&fs_info
->scrub_lock
);
2226 atomic_set(&fs_info
->scrubs_running
, 0);
2227 atomic_set(&fs_info
->scrub_pause_req
, 0);
2228 atomic_set(&fs_info
->scrubs_paused
, 0);
2229 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2230 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2231 init_rwsem(&fs_info
->scrub_super_lock
);
2232 fs_info
->scrub_workers_refcnt
= 0;
2233 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2234 fs_info
->check_integrity_print_mask
= 0;
2237 spin_lock_init(&fs_info
->balance_lock
);
2238 mutex_init(&fs_info
->balance_mutex
);
2239 atomic_set(&fs_info
->balance_running
, 0);
2240 atomic_set(&fs_info
->balance_pause_req
, 0);
2241 atomic_set(&fs_info
->balance_cancel_req
, 0);
2242 fs_info
->balance_ctl
= NULL
;
2243 init_waitqueue_head(&fs_info
->balance_wait_q
);
2245 sb
->s_blocksize
= 4096;
2246 sb
->s_blocksize_bits
= blksize_bits(4096);
2247 sb
->s_bdi
= &fs_info
->bdi
;
2249 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2250 set_nlink(fs_info
->btree_inode
, 1);
2252 * we set the i_size on the btree inode to the max possible int.
2253 * the real end of the address space is determined by all of
2254 * the devices in the system
2256 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2257 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2258 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2260 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2261 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2262 fs_info
->btree_inode
->i_mapping
);
2263 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2264 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2266 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2268 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2269 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2270 sizeof(struct btrfs_key
));
2271 set_bit(BTRFS_INODE_DUMMY
,
2272 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2273 insert_inode_hash(fs_info
->btree_inode
);
2275 spin_lock_init(&fs_info
->block_group_cache_lock
);
2276 fs_info
->block_group_cache_tree
= RB_ROOT
;
2277 fs_info
->first_logical_byte
= (u64
)-1;
2279 extent_io_tree_init(&fs_info
->freed_extents
[0],
2280 fs_info
->btree_inode
->i_mapping
);
2281 extent_io_tree_init(&fs_info
->freed_extents
[1],
2282 fs_info
->btree_inode
->i_mapping
);
2283 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2284 fs_info
->do_barriers
= 1;
2287 mutex_init(&fs_info
->ordered_operations_mutex
);
2288 mutex_init(&fs_info
->ordered_extent_flush_mutex
);
2289 mutex_init(&fs_info
->tree_log_mutex
);
2290 mutex_init(&fs_info
->chunk_mutex
);
2291 mutex_init(&fs_info
->transaction_kthread_mutex
);
2292 mutex_init(&fs_info
->cleaner_mutex
);
2293 mutex_init(&fs_info
->volume_mutex
);
2294 init_rwsem(&fs_info
->extent_commit_sem
);
2295 init_rwsem(&fs_info
->cleanup_work_sem
);
2296 init_rwsem(&fs_info
->subvol_sem
);
2297 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2298 fs_info
->dev_replace
.lock_owner
= 0;
2299 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2300 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2301 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2302 mutex_init(&fs_info
->dev_replace
.lock
);
2304 spin_lock_init(&fs_info
->qgroup_lock
);
2305 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2306 fs_info
->qgroup_tree
= RB_ROOT
;
2307 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2308 fs_info
->qgroup_seq
= 1;
2309 fs_info
->quota_enabled
= 0;
2310 fs_info
->pending_quota_state
= 0;
2311 fs_info
->qgroup_ulist
= NULL
;
2312 mutex_init(&fs_info
->qgroup_rescan_lock
);
2314 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2315 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2317 init_waitqueue_head(&fs_info
->transaction_throttle
);
2318 init_waitqueue_head(&fs_info
->transaction_wait
);
2319 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2320 init_waitqueue_head(&fs_info
->async_submit_wait
);
2322 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2328 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2329 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2331 invalidate_bdev(fs_devices
->latest_bdev
);
2334 * Read super block and check the signature bytes only
2336 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2343 * We want to check superblock checksum, the type is stored inside.
2344 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2346 if (btrfs_check_super_csum(bh
->b_data
)) {
2347 printk(KERN_ERR
"btrfs: superblock checksum mismatch\n");
2353 * super_copy is zeroed at allocation time and we never touch the
2354 * following bytes up to INFO_SIZE, the checksum is calculated from
2355 * the whole block of INFO_SIZE
2357 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2358 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2359 sizeof(*fs_info
->super_for_commit
));
2362 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2364 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2366 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2371 disk_super
= fs_info
->super_copy
;
2372 if (!btrfs_super_root(disk_super
))
2375 /* check FS state, whether FS is broken. */
2376 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2377 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2380 * run through our array of backup supers and setup
2381 * our ring pointer to the oldest one
2383 generation
= btrfs_super_generation(disk_super
);
2384 find_oldest_super_backup(fs_info
, generation
);
2387 * In the long term, we'll store the compression type in the super
2388 * block, and it'll be used for per file compression control.
2390 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2392 ret
= btrfs_parse_options(tree_root
, options
);
2398 features
= btrfs_super_incompat_flags(disk_super
) &
2399 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2401 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2402 "unsupported optional features (%Lx).\n",
2408 if (btrfs_super_leafsize(disk_super
) !=
2409 btrfs_super_nodesize(disk_super
)) {
2410 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2411 "blocksizes don't match. node %d leaf %d\n",
2412 btrfs_super_nodesize(disk_super
),
2413 btrfs_super_leafsize(disk_super
));
2417 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2418 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2419 "blocksize (%d) was too large\n",
2420 btrfs_super_leafsize(disk_super
));
2425 features
= btrfs_super_incompat_flags(disk_super
);
2426 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2427 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2428 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2430 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2431 printk(KERN_ERR
"btrfs: has skinny extents\n");
2434 * flag our filesystem as having big metadata blocks if
2435 * they are bigger than the page size
2437 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2438 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2439 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2440 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2443 nodesize
= btrfs_super_nodesize(disk_super
);
2444 leafsize
= btrfs_super_leafsize(disk_super
);
2445 sectorsize
= btrfs_super_sectorsize(disk_super
);
2446 stripesize
= btrfs_super_stripesize(disk_super
);
2447 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2448 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2451 * mixed block groups end up with duplicate but slightly offset
2452 * extent buffers for the same range. It leads to corruptions
2454 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2455 (sectorsize
!= leafsize
)) {
2456 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2457 "are not allowed for mixed block groups on %s\n",
2463 * Needn't use the lock because there is no other task which will
2466 btrfs_set_super_incompat_flags(disk_super
, features
);
2468 features
= btrfs_super_compat_ro_flags(disk_super
) &
2469 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2470 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2471 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2472 "unsupported option features (%Lx).\n",
2478 btrfs_init_workers(&fs_info
->generic_worker
,
2479 "genwork", 1, NULL
);
2481 btrfs_init_workers(&fs_info
->workers
, "worker",
2482 fs_info
->thread_pool_size
,
2483 &fs_info
->generic_worker
);
2485 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2486 fs_info
->thread_pool_size
, NULL
);
2488 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2489 fs_info
->thread_pool_size
, NULL
);
2491 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2492 min_t(u64
, fs_devices
->num_devices
,
2493 fs_info
->thread_pool_size
), NULL
);
2495 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2496 fs_info
->thread_pool_size
, NULL
);
2498 /* a higher idle thresh on the submit workers makes it much more
2499 * likely that bios will be send down in a sane order to the
2502 fs_info
->submit_workers
.idle_thresh
= 64;
2504 fs_info
->workers
.idle_thresh
= 16;
2505 fs_info
->workers
.ordered
= 1;
2507 fs_info
->delalloc_workers
.idle_thresh
= 2;
2508 fs_info
->delalloc_workers
.ordered
= 1;
2510 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2511 &fs_info
->generic_worker
);
2512 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2513 fs_info
->thread_pool_size
,
2514 &fs_info
->generic_worker
);
2515 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2516 fs_info
->thread_pool_size
,
2517 &fs_info
->generic_worker
);
2518 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2519 "endio-meta-write", fs_info
->thread_pool_size
,
2520 &fs_info
->generic_worker
);
2521 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2522 "endio-raid56", fs_info
->thread_pool_size
,
2523 &fs_info
->generic_worker
);
2524 btrfs_init_workers(&fs_info
->rmw_workers
,
2525 "rmw", fs_info
->thread_pool_size
,
2526 &fs_info
->generic_worker
);
2527 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2528 fs_info
->thread_pool_size
,
2529 &fs_info
->generic_worker
);
2530 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2531 1, &fs_info
->generic_worker
);
2532 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2533 fs_info
->thread_pool_size
,
2534 &fs_info
->generic_worker
);
2535 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2536 fs_info
->thread_pool_size
,
2537 &fs_info
->generic_worker
);
2538 btrfs_init_workers(&fs_info
->qgroup_rescan_workers
, "qgroup-rescan", 1,
2539 &fs_info
->generic_worker
);
2542 * endios are largely parallel and should have a very
2545 fs_info
->endio_workers
.idle_thresh
= 4;
2546 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2547 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2548 fs_info
->rmw_workers
.idle_thresh
= 2;
2550 fs_info
->endio_write_workers
.idle_thresh
= 2;
2551 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2552 fs_info
->readahead_workers
.idle_thresh
= 2;
2555 * btrfs_start_workers can really only fail because of ENOMEM so just
2556 * return -ENOMEM if any of these fail.
2558 ret
= btrfs_start_workers(&fs_info
->workers
);
2559 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2560 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2561 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2562 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2563 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2564 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2565 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2566 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2567 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2568 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2569 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2570 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2571 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2572 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2573 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2574 ret
|= btrfs_start_workers(&fs_info
->qgroup_rescan_workers
);
2577 goto fail_sb_buffer
;
2580 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2581 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2582 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2584 tree_root
->nodesize
= nodesize
;
2585 tree_root
->leafsize
= leafsize
;
2586 tree_root
->sectorsize
= sectorsize
;
2587 tree_root
->stripesize
= stripesize
;
2589 sb
->s_blocksize
= sectorsize
;
2590 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2592 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2593 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2594 goto fail_sb_buffer
;
2597 if (sectorsize
!= PAGE_SIZE
) {
2598 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2599 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2600 goto fail_sb_buffer
;
2603 mutex_lock(&fs_info
->chunk_mutex
);
2604 ret
= btrfs_read_sys_array(tree_root
);
2605 mutex_unlock(&fs_info
->chunk_mutex
);
2607 printk(KERN_WARNING
"btrfs: failed to read the system "
2608 "array on %s\n", sb
->s_id
);
2609 goto fail_sb_buffer
;
2612 blocksize
= btrfs_level_size(tree_root
,
2613 btrfs_super_chunk_root_level(disk_super
));
2614 generation
= btrfs_super_chunk_root_generation(disk_super
);
2616 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2617 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2619 chunk_root
->node
= read_tree_block(chunk_root
,
2620 btrfs_super_chunk_root(disk_super
),
2621 blocksize
, generation
);
2622 if (!chunk_root
->node
||
2623 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2624 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2626 goto fail_tree_roots
;
2628 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2629 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2631 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2632 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2634 ret
= btrfs_read_chunk_tree(chunk_root
);
2636 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2638 goto fail_tree_roots
;
2642 * keep the device that is marked to be the target device for the
2643 * dev_replace procedure
2645 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2647 if (!fs_devices
->latest_bdev
) {
2648 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2650 goto fail_tree_roots
;
2654 blocksize
= btrfs_level_size(tree_root
,
2655 btrfs_super_root_level(disk_super
));
2656 generation
= btrfs_super_generation(disk_super
);
2658 tree_root
->node
= read_tree_block(tree_root
,
2659 btrfs_super_root(disk_super
),
2660 blocksize
, generation
);
2661 if (!tree_root
->node
||
2662 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2663 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2666 goto recovery_tree_root
;
2669 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2670 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2672 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2673 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2674 location
.offset
= 0;
2676 extent_root
= btrfs_read_tree_root(tree_root
, &location
);
2677 if (IS_ERR(extent_root
)) {
2678 ret
= PTR_ERR(extent_root
);
2679 goto recovery_tree_root
;
2681 extent_root
->track_dirty
= 1;
2682 fs_info
->extent_root
= extent_root
;
2684 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2685 dev_root
= btrfs_read_tree_root(tree_root
, &location
);
2686 if (IS_ERR(dev_root
)) {
2687 ret
= PTR_ERR(dev_root
);
2688 goto recovery_tree_root
;
2690 dev_root
->track_dirty
= 1;
2691 fs_info
->dev_root
= dev_root
;
2692 btrfs_init_devices_late(fs_info
);
2694 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2695 csum_root
= btrfs_read_tree_root(tree_root
, &location
);
2696 if (IS_ERR(csum_root
)) {
2697 ret
= PTR_ERR(csum_root
);
2698 goto recovery_tree_root
;
2700 csum_root
->track_dirty
= 1;
2701 fs_info
->csum_root
= csum_root
;
2703 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2704 quota_root
= btrfs_read_tree_root(tree_root
, &location
);
2705 if (!IS_ERR(quota_root
)) {
2706 quota_root
->track_dirty
= 1;
2707 fs_info
->quota_enabled
= 1;
2708 fs_info
->pending_quota_state
= 1;
2709 fs_info
->quota_root
= quota_root
;
2712 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2713 uuid_root
= btrfs_read_tree_root(tree_root
, &location
);
2714 if (IS_ERR(uuid_root
)) {
2715 ret
= PTR_ERR(uuid_root
);
2717 goto recovery_tree_root
;
2718 create_uuid_tree
= true;
2719 check_uuid_tree
= false;
2721 uuid_root
->track_dirty
= 1;
2722 fs_info
->uuid_root
= uuid_root
;
2723 create_uuid_tree
= false;
2725 generation
!= btrfs_super_uuid_tree_generation(disk_super
);
2728 fs_info
->generation
= generation
;
2729 fs_info
->last_trans_committed
= generation
;
2731 ret
= btrfs_recover_balance(fs_info
);
2733 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2734 goto fail_block_groups
;
2737 ret
= btrfs_init_dev_stats(fs_info
);
2739 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2741 goto fail_block_groups
;
2744 ret
= btrfs_init_dev_replace(fs_info
);
2746 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2747 goto fail_block_groups
;
2750 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2752 ret
= btrfs_init_space_info(fs_info
);
2754 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2755 goto fail_block_groups
;
2758 ret
= btrfs_read_block_groups(extent_root
);
2760 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2761 goto fail_block_groups
;
2763 fs_info
->num_tolerated_disk_barrier_failures
=
2764 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2765 if (fs_info
->fs_devices
->missing_devices
>
2766 fs_info
->num_tolerated_disk_barrier_failures
&&
2767 !(sb
->s_flags
& MS_RDONLY
)) {
2769 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2770 goto fail_block_groups
;
2773 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2775 if (IS_ERR(fs_info
->cleaner_kthread
))
2776 goto fail_block_groups
;
2778 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2780 "btrfs-transaction");
2781 if (IS_ERR(fs_info
->transaction_kthread
))
2784 if (!btrfs_test_opt(tree_root
, SSD
) &&
2785 !btrfs_test_opt(tree_root
, NOSSD
) &&
2786 !fs_info
->fs_devices
->rotating
) {
2787 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2789 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2792 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2793 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2794 ret
= btrfsic_mount(tree_root
, fs_devices
,
2795 btrfs_test_opt(tree_root
,
2796 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2798 fs_info
->check_integrity_print_mask
);
2800 printk(KERN_WARNING
"btrfs: failed to initialize"
2801 " integrity check module %s\n", sb
->s_id
);
2804 ret
= btrfs_read_qgroup_config(fs_info
);
2806 goto fail_trans_kthread
;
2808 /* do not make disk changes in broken FS */
2809 if (btrfs_super_log_root(disk_super
) != 0) {
2810 u64 bytenr
= btrfs_super_log_root(disk_super
);
2812 if (fs_devices
->rw_devices
== 0) {
2813 printk(KERN_WARNING
"Btrfs log replay required "
2819 btrfs_level_size(tree_root
,
2820 btrfs_super_log_root_level(disk_super
));
2822 log_tree_root
= btrfs_alloc_root(fs_info
);
2823 if (!log_tree_root
) {
2828 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2829 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2831 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2834 if (!log_tree_root
->node
||
2835 !extent_buffer_uptodate(log_tree_root
->node
)) {
2836 printk(KERN_ERR
"btrfs: failed to read log tree\n");
2837 free_extent_buffer(log_tree_root
->node
);
2838 kfree(log_tree_root
);
2839 goto fail_trans_kthread
;
2841 /* returns with log_tree_root freed on success */
2842 ret
= btrfs_recover_log_trees(log_tree_root
);
2844 btrfs_error(tree_root
->fs_info
, ret
,
2845 "Failed to recover log tree");
2846 free_extent_buffer(log_tree_root
->node
);
2847 kfree(log_tree_root
);
2848 goto fail_trans_kthread
;
2851 if (sb
->s_flags
& MS_RDONLY
) {
2852 ret
= btrfs_commit_super(tree_root
);
2854 goto fail_trans_kthread
;
2858 ret
= btrfs_find_orphan_roots(tree_root
);
2860 goto fail_trans_kthread
;
2862 if (!(sb
->s_flags
& MS_RDONLY
)) {
2863 ret
= btrfs_cleanup_fs_roots(fs_info
);
2865 goto fail_trans_kthread
;
2867 ret
= btrfs_recover_relocation(tree_root
);
2870 "btrfs: failed to recover relocation\n");
2876 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2877 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2878 location
.offset
= 0;
2880 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2881 if (IS_ERR(fs_info
->fs_root
)) {
2882 err
= PTR_ERR(fs_info
->fs_root
);
2886 if (sb
->s_flags
& MS_RDONLY
)
2889 down_read(&fs_info
->cleanup_work_sem
);
2890 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2891 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2892 up_read(&fs_info
->cleanup_work_sem
);
2893 close_ctree(tree_root
);
2896 up_read(&fs_info
->cleanup_work_sem
);
2898 ret
= btrfs_resume_balance_async(fs_info
);
2900 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2901 close_ctree(tree_root
);
2905 ret
= btrfs_resume_dev_replace_async(fs_info
);
2907 pr_warn("btrfs: failed to resume dev_replace\n");
2908 close_ctree(tree_root
);
2912 btrfs_qgroup_rescan_resume(fs_info
);
2914 if (create_uuid_tree
) {
2915 pr_info("btrfs: creating UUID tree\n");
2916 ret
= btrfs_create_uuid_tree(fs_info
);
2918 pr_warn("btrfs: failed to create the UUID tree %d\n",
2920 close_ctree(tree_root
);
2923 } else if (check_uuid_tree
||
2924 btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
)) {
2925 pr_info("btrfs: checking UUID tree\n");
2926 ret
= btrfs_check_uuid_tree(fs_info
);
2928 pr_warn("btrfs: failed to check the UUID tree %d\n",
2930 close_ctree(tree_root
);
2934 fs_info
->update_uuid_tree_gen
= 1;
2940 btrfs_free_qgroup_config(fs_info
);
2942 kthread_stop(fs_info
->transaction_kthread
);
2943 btrfs_cleanup_transaction(fs_info
->tree_root
);
2944 del_fs_roots(fs_info
);
2946 kthread_stop(fs_info
->cleaner_kthread
);
2949 * make sure we're done with the btree inode before we stop our
2952 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2955 btrfs_put_block_group_cache(fs_info
);
2956 btrfs_free_block_groups(fs_info
);
2959 free_root_pointers(fs_info
, 1);
2960 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2963 btrfs_stop_all_workers(fs_info
);
2966 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2968 iput(fs_info
->btree_inode
);
2969 fail_delalloc_bytes
:
2970 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2971 fail_dirty_metadata_bytes
:
2972 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2974 bdi_destroy(&fs_info
->bdi
);
2976 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2978 btrfs_free_stripe_hash_table(fs_info
);
2979 btrfs_close_devices(fs_info
->fs_devices
);
2983 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2984 goto fail_tree_roots
;
2986 free_root_pointers(fs_info
, 0);
2988 /* don't use the log in recovery mode, it won't be valid */
2989 btrfs_set_super_log_root(disk_super
, 0);
2991 /* we can't trust the free space cache either */
2992 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2994 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2995 &num_backups_tried
, &backup_index
);
2997 goto fail_block_groups
;
2998 goto retry_root_backup
;
3001 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3004 set_buffer_uptodate(bh
);
3006 struct btrfs_device
*device
= (struct btrfs_device
*)
3009 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
3010 "I/O error on %s\n",
3011 rcu_str_deref(device
->name
));
3012 /* note, we dont' set_buffer_write_io_error because we have
3013 * our own ways of dealing with the IO errors
3015 clear_buffer_uptodate(bh
);
3016 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3022 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3024 struct buffer_head
*bh
;
3025 struct buffer_head
*latest
= NULL
;
3026 struct btrfs_super_block
*super
;
3031 /* we would like to check all the supers, but that would make
3032 * a btrfs mount succeed after a mkfs from a different FS.
3033 * So, we need to add a special mount option to scan for
3034 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3036 for (i
= 0; i
< 1; i
++) {
3037 bytenr
= btrfs_sb_offset(i
);
3038 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3039 i_size_read(bdev
->bd_inode
))
3041 bh
= __bread(bdev
, bytenr
/ 4096,
3042 BTRFS_SUPER_INFO_SIZE
);
3046 super
= (struct btrfs_super_block
*)bh
->b_data
;
3047 if (btrfs_super_bytenr(super
) != bytenr
||
3048 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3053 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3056 transid
= btrfs_super_generation(super
);
3065 * this should be called twice, once with wait == 0 and
3066 * once with wait == 1. When wait == 0 is done, all the buffer heads
3067 * we write are pinned.
3069 * They are released when wait == 1 is done.
3070 * max_mirrors must be the same for both runs, and it indicates how
3071 * many supers on this one device should be written.
3073 * max_mirrors == 0 means to write them all.
3075 static int write_dev_supers(struct btrfs_device
*device
,
3076 struct btrfs_super_block
*sb
,
3077 int do_barriers
, int wait
, int max_mirrors
)
3079 struct buffer_head
*bh
;
3086 if (max_mirrors
== 0)
3087 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3089 for (i
= 0; i
< max_mirrors
; i
++) {
3090 bytenr
= btrfs_sb_offset(i
);
3091 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
3095 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3096 BTRFS_SUPER_INFO_SIZE
);
3102 if (!buffer_uptodate(bh
))
3105 /* drop our reference */
3108 /* drop the reference from the wait == 0 run */
3112 btrfs_set_super_bytenr(sb
, bytenr
);
3115 crc
= btrfs_csum_data((char *)sb
+
3116 BTRFS_CSUM_SIZE
, crc
,
3117 BTRFS_SUPER_INFO_SIZE
-
3119 btrfs_csum_final(crc
, sb
->csum
);
3122 * one reference for us, and we leave it for the
3125 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3126 BTRFS_SUPER_INFO_SIZE
);
3128 printk(KERN_ERR
"btrfs: couldn't get super "
3129 "buffer head for bytenr %Lu\n", bytenr
);
3134 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3136 /* one reference for submit_bh */
3139 set_buffer_uptodate(bh
);
3141 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3142 bh
->b_private
= device
;
3146 * we fua the first super. The others we allow
3149 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3153 return errors
< i
? 0 : -1;
3157 * endio for the write_dev_flush, this will wake anyone waiting
3158 * for the barrier when it is done
3160 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3163 if (err
== -EOPNOTSUPP
)
3164 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3165 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3167 if (bio
->bi_private
)
3168 complete(bio
->bi_private
);
3173 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3174 * sent down. With wait == 1, it waits for the previous flush.
3176 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3179 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3184 if (device
->nobarriers
)
3188 bio
= device
->flush_bio
;
3192 wait_for_completion(&device
->flush_wait
);
3194 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3195 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3196 rcu_str_deref(device
->name
));
3197 device
->nobarriers
= 1;
3198 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3200 btrfs_dev_stat_inc_and_print(device
,
3201 BTRFS_DEV_STAT_FLUSH_ERRS
);
3204 /* drop the reference from the wait == 0 run */
3206 device
->flush_bio
= NULL
;
3212 * one reference for us, and we leave it for the
3215 device
->flush_bio
= NULL
;
3216 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3220 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3221 bio
->bi_bdev
= device
->bdev
;
3222 init_completion(&device
->flush_wait
);
3223 bio
->bi_private
= &device
->flush_wait
;
3224 device
->flush_bio
= bio
;
3227 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3233 * send an empty flush down to each device in parallel,
3234 * then wait for them
3236 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3238 struct list_head
*head
;
3239 struct btrfs_device
*dev
;
3240 int errors_send
= 0;
3241 int errors_wait
= 0;
3244 /* send down all the barriers */
3245 head
= &info
->fs_devices
->devices
;
3246 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3251 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3254 ret
= write_dev_flush(dev
, 0);
3259 /* wait for all the barriers */
3260 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3265 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3268 ret
= write_dev_flush(dev
, 1);
3272 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3273 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3278 int btrfs_calc_num_tolerated_disk_barrier_failures(
3279 struct btrfs_fs_info
*fs_info
)
3281 struct btrfs_ioctl_space_info space
;
3282 struct btrfs_space_info
*sinfo
;
3283 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3284 BTRFS_BLOCK_GROUP_SYSTEM
,
3285 BTRFS_BLOCK_GROUP_METADATA
,
3286 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3290 int num_tolerated_disk_barrier_failures
=
3291 (int)fs_info
->fs_devices
->num_devices
;
3293 for (i
= 0; i
< num_types
; i
++) {
3294 struct btrfs_space_info
*tmp
;
3298 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3299 if (tmp
->flags
== types
[i
]) {
3309 down_read(&sinfo
->groups_sem
);
3310 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3311 if (!list_empty(&sinfo
->block_groups
[c
])) {
3314 btrfs_get_block_group_info(
3315 &sinfo
->block_groups
[c
], &space
);
3316 if (space
.total_bytes
== 0 ||
3317 space
.used_bytes
== 0)
3319 flags
= space
.flags
;
3322 * 0: if dup, single or RAID0 is configured for
3323 * any of metadata, system or data, else
3324 * 1: if RAID5 is configured, or if RAID1 or
3325 * RAID10 is configured and only two mirrors
3327 * 2: if RAID6 is configured, else
3328 * num_mirrors - 1: if RAID1 or RAID10 is
3329 * configured and more than
3330 * 2 mirrors are used.
3332 if (num_tolerated_disk_barrier_failures
> 0 &&
3333 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3334 BTRFS_BLOCK_GROUP_RAID0
)) ||
3335 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3337 num_tolerated_disk_barrier_failures
= 0;
3338 else if (num_tolerated_disk_barrier_failures
> 1) {
3339 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3340 BTRFS_BLOCK_GROUP_RAID5
|
3341 BTRFS_BLOCK_GROUP_RAID10
)) {
3342 num_tolerated_disk_barrier_failures
= 1;
3344 BTRFS_BLOCK_GROUP_RAID6
) {
3345 num_tolerated_disk_barrier_failures
= 2;
3350 up_read(&sinfo
->groups_sem
);
3353 return num_tolerated_disk_barrier_failures
;
3356 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3358 struct list_head
*head
;
3359 struct btrfs_device
*dev
;
3360 struct btrfs_super_block
*sb
;
3361 struct btrfs_dev_item
*dev_item
;
3365 int total_errors
= 0;
3368 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3369 backup_super_roots(root
->fs_info
);
3371 sb
= root
->fs_info
->super_for_commit
;
3372 dev_item
= &sb
->dev_item
;
3374 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3375 head
= &root
->fs_info
->fs_devices
->devices
;
3376 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3379 ret
= barrier_all_devices(root
->fs_info
);
3382 &root
->fs_info
->fs_devices
->device_list_mutex
);
3383 btrfs_error(root
->fs_info
, ret
,
3384 "errors while submitting device barriers.");
3389 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3394 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3397 btrfs_set_stack_device_generation(dev_item
, 0);
3398 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3399 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3400 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3401 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3402 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3403 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3404 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3405 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3406 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3408 flags
= btrfs_super_flags(sb
);
3409 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3411 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3415 if (total_errors
> max_errors
) {
3416 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3419 /* FUA is masked off if unsupported and can't be the reason */
3420 btrfs_error(root
->fs_info
, -EIO
,
3421 "%d errors while writing supers", total_errors
);
3426 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3429 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3432 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3436 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3437 if (total_errors
> max_errors
) {
3438 btrfs_error(root
->fs_info
, -EIO
,
3439 "%d errors while writing supers", total_errors
);
3445 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3446 struct btrfs_root
*root
, int max_mirrors
)
3450 ret
= write_all_supers(root
, max_mirrors
);
3454 /* Drop a fs root from the radix tree and free it. */
3455 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3456 struct btrfs_root
*root
)
3458 spin_lock(&fs_info
->fs_roots_radix_lock
);
3459 radix_tree_delete(&fs_info
->fs_roots_radix
,
3460 (unsigned long)root
->root_key
.objectid
);
3461 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3463 if (btrfs_root_refs(&root
->root_item
) == 0)
3464 synchronize_srcu(&fs_info
->subvol_srcu
);
3466 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3467 btrfs_free_log(NULL
, root
);
3468 btrfs_free_log_root_tree(NULL
, fs_info
);
3471 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3472 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3476 static void free_fs_root(struct btrfs_root
*root
)
3478 iput(root
->cache_inode
);
3479 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3480 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3481 root
->orphan_block_rsv
= NULL
;
3483 free_anon_bdev(root
->anon_dev
);
3484 free_extent_buffer(root
->node
);
3485 free_extent_buffer(root
->commit_root
);
3486 kfree(root
->free_ino_ctl
);
3487 kfree(root
->free_ino_pinned
);
3489 btrfs_put_fs_root(root
);
3492 void btrfs_free_fs_root(struct btrfs_root
*root
)
3497 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3499 u64 root_objectid
= 0;
3500 struct btrfs_root
*gang
[8];
3505 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3506 (void **)gang
, root_objectid
,
3511 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3512 for (i
= 0; i
< ret
; i
++) {
3515 root_objectid
= gang
[i
]->root_key
.objectid
;
3516 err
= btrfs_orphan_cleanup(gang
[i
]);
3525 int btrfs_commit_super(struct btrfs_root
*root
)
3527 struct btrfs_trans_handle
*trans
;
3530 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3531 btrfs_run_delayed_iputs(root
);
3532 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3533 wake_up_process(root
->fs_info
->cleaner_kthread
);
3535 /* wait until ongoing cleanup work done */
3536 down_write(&root
->fs_info
->cleanup_work_sem
);
3537 up_write(&root
->fs_info
->cleanup_work_sem
);
3539 trans
= btrfs_join_transaction(root
);
3541 return PTR_ERR(trans
);
3542 ret
= btrfs_commit_transaction(trans
, root
);
3545 /* run commit again to drop the original snapshot */
3546 trans
= btrfs_join_transaction(root
);
3548 return PTR_ERR(trans
);
3549 ret
= btrfs_commit_transaction(trans
, root
);
3552 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3554 btrfs_error(root
->fs_info
, ret
,
3555 "Failed to sync btree inode to disk.");
3559 ret
= write_ctree_super(NULL
, root
, 0);
3563 int close_ctree(struct btrfs_root
*root
)
3565 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3568 fs_info
->closing
= 1;
3571 /* wait for the uuid_scan task to finish */
3572 down(&fs_info
->uuid_tree_rescan_sem
);
3573 /* avoid complains from lockdep et al., set sem back to initial state */
3574 up(&fs_info
->uuid_tree_rescan_sem
);
3576 /* pause restriper - we want to resume on mount */
3577 btrfs_pause_balance(fs_info
);
3579 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3581 btrfs_scrub_cancel(fs_info
);
3583 /* wait for any defraggers to finish */
3584 wait_event(fs_info
->transaction_wait
,
3585 (atomic_read(&fs_info
->defrag_running
) == 0));
3587 /* clear out the rbtree of defraggable inodes */
3588 btrfs_cleanup_defrag_inodes(fs_info
);
3590 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3591 ret
= btrfs_commit_super(root
);
3593 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3596 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3597 btrfs_error_commit_super(root
);
3599 btrfs_put_block_group_cache(fs_info
);
3601 kthread_stop(fs_info
->transaction_kthread
);
3602 kthread_stop(fs_info
->cleaner_kthread
);
3604 fs_info
->closing
= 2;
3607 btrfs_free_qgroup_config(root
->fs_info
);
3609 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3610 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3611 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3614 btrfs_free_block_groups(fs_info
);
3616 btrfs_stop_all_workers(fs_info
);
3618 del_fs_roots(fs_info
);
3620 free_root_pointers(fs_info
, 1);
3622 iput(fs_info
->btree_inode
);
3624 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3625 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3626 btrfsic_unmount(root
, fs_info
->fs_devices
);
3629 btrfs_close_devices(fs_info
->fs_devices
);
3630 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3632 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3633 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3634 bdi_destroy(&fs_info
->bdi
);
3635 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3637 btrfs_free_stripe_hash_table(fs_info
);
3639 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3640 root
->orphan_block_rsv
= NULL
;
3645 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3649 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3651 ret
= extent_buffer_uptodate(buf
);
3655 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3656 parent_transid
, atomic
);
3662 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3664 return set_extent_buffer_uptodate(buf
);
3667 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3669 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3670 u64 transid
= btrfs_header_generation(buf
);
3673 btrfs_assert_tree_locked(buf
);
3674 if (transid
!= root
->fs_info
->generation
)
3675 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3676 "found %llu running %llu\n",
3677 buf
->start
, transid
, root
->fs_info
->generation
);
3678 was_dirty
= set_extent_buffer_dirty(buf
);
3680 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3682 root
->fs_info
->dirty_metadata_batch
);
3685 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3689 * looks as though older kernels can get into trouble with
3690 * this code, they end up stuck in balance_dirty_pages forever
3694 if (current
->flags
& PF_MEMALLOC
)
3698 btrfs_balance_delayed_items(root
);
3700 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3701 BTRFS_DIRTY_METADATA_THRESH
);
3703 balance_dirty_pages_ratelimited(
3704 root
->fs_info
->btree_inode
->i_mapping
);
3709 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3711 __btrfs_btree_balance_dirty(root
, 1);
3714 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3716 __btrfs_btree_balance_dirty(root
, 0);
3719 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3721 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3722 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3725 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3729 * Placeholder for checks
3734 static void btrfs_error_commit_super(struct btrfs_root
*root
)
3736 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3737 btrfs_run_delayed_iputs(root
);
3738 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3740 down_write(&root
->fs_info
->cleanup_work_sem
);
3741 up_write(&root
->fs_info
->cleanup_work_sem
);
3743 /* cleanup FS via transaction */
3744 btrfs_cleanup_transaction(root
);
3747 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
3748 struct btrfs_root
*root
)
3750 struct btrfs_inode
*btrfs_inode
;
3751 struct list_head splice
;
3753 INIT_LIST_HEAD(&splice
);
3755 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3756 spin_lock(&root
->fs_info
->ordered_root_lock
);
3758 list_splice_init(&t
->ordered_operations
, &splice
);
3759 while (!list_empty(&splice
)) {
3760 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3761 ordered_operations
);
3763 list_del_init(&btrfs_inode
->ordered_operations
);
3764 spin_unlock(&root
->fs_info
->ordered_root_lock
);
3766 btrfs_invalidate_inodes(btrfs_inode
->root
);
3768 spin_lock(&root
->fs_info
->ordered_root_lock
);
3771 spin_unlock(&root
->fs_info
->ordered_root_lock
);
3772 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3775 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3777 struct btrfs_ordered_extent
*ordered
;
3779 spin_lock(&root
->ordered_extent_lock
);
3781 * This will just short circuit the ordered completion stuff which will
3782 * make sure the ordered extent gets properly cleaned up.
3784 list_for_each_entry(ordered
, &root
->ordered_extents
,
3786 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3787 spin_unlock(&root
->ordered_extent_lock
);
3790 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
3792 struct btrfs_root
*root
;
3793 struct list_head splice
;
3795 INIT_LIST_HEAD(&splice
);
3797 spin_lock(&fs_info
->ordered_root_lock
);
3798 list_splice_init(&fs_info
->ordered_roots
, &splice
);
3799 while (!list_empty(&splice
)) {
3800 root
= list_first_entry(&splice
, struct btrfs_root
,
3802 list_del_init(&root
->ordered_root
);
3804 btrfs_destroy_ordered_extents(root
);
3806 cond_resched_lock(&fs_info
->ordered_root_lock
);
3808 spin_unlock(&fs_info
->ordered_root_lock
);
3811 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3812 struct btrfs_root
*root
)
3814 struct rb_node
*node
;
3815 struct btrfs_delayed_ref_root
*delayed_refs
;
3816 struct btrfs_delayed_ref_node
*ref
;
3819 delayed_refs
= &trans
->delayed_refs
;
3821 spin_lock(&delayed_refs
->lock
);
3822 if (delayed_refs
->num_entries
== 0) {
3823 spin_unlock(&delayed_refs
->lock
);
3824 printk(KERN_INFO
"delayed_refs has NO entry\n");
3828 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3829 struct btrfs_delayed_ref_head
*head
= NULL
;
3830 bool pin_bytes
= false;
3832 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3833 atomic_set(&ref
->refs
, 1);
3834 if (btrfs_delayed_ref_is_head(ref
)) {
3836 head
= btrfs_delayed_node_to_head(ref
);
3837 if (!mutex_trylock(&head
->mutex
)) {
3838 atomic_inc(&ref
->refs
);
3839 spin_unlock(&delayed_refs
->lock
);
3841 /* Need to wait for the delayed ref to run */
3842 mutex_lock(&head
->mutex
);
3843 mutex_unlock(&head
->mutex
);
3844 btrfs_put_delayed_ref(ref
);
3846 spin_lock(&delayed_refs
->lock
);
3850 if (head
->must_insert_reserved
)
3852 btrfs_free_delayed_extent_op(head
->extent_op
);
3853 delayed_refs
->num_heads
--;
3854 if (list_empty(&head
->cluster
))
3855 delayed_refs
->num_heads_ready
--;
3856 list_del_init(&head
->cluster
);
3860 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3861 delayed_refs
->num_entries
--;
3862 spin_unlock(&delayed_refs
->lock
);
3865 btrfs_pin_extent(root
, ref
->bytenr
,
3867 mutex_unlock(&head
->mutex
);
3869 btrfs_put_delayed_ref(ref
);
3872 spin_lock(&delayed_refs
->lock
);
3875 spin_unlock(&delayed_refs
->lock
);
3880 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
)
3882 struct btrfs_pending_snapshot
*snapshot
;
3883 struct list_head splice
;
3885 INIT_LIST_HEAD(&splice
);
3887 list_splice_init(&t
->pending_snapshots
, &splice
);
3889 while (!list_empty(&splice
)) {
3890 snapshot
= list_entry(splice
.next
,
3891 struct btrfs_pending_snapshot
,
3893 snapshot
->error
= -ECANCELED
;
3894 list_del_init(&snapshot
->list
);
3898 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3900 struct btrfs_inode
*btrfs_inode
;
3901 struct list_head splice
;
3903 INIT_LIST_HEAD(&splice
);
3905 spin_lock(&root
->delalloc_lock
);
3906 list_splice_init(&root
->delalloc_inodes
, &splice
);
3908 while (!list_empty(&splice
)) {
3909 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
3912 list_del_init(&btrfs_inode
->delalloc_inodes
);
3913 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3914 &btrfs_inode
->runtime_flags
);
3915 spin_unlock(&root
->delalloc_lock
);
3917 btrfs_invalidate_inodes(btrfs_inode
->root
);
3919 spin_lock(&root
->delalloc_lock
);
3922 spin_unlock(&root
->delalloc_lock
);
3925 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
3927 struct btrfs_root
*root
;
3928 struct list_head splice
;
3930 INIT_LIST_HEAD(&splice
);
3932 spin_lock(&fs_info
->delalloc_root_lock
);
3933 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
3934 while (!list_empty(&splice
)) {
3935 root
= list_first_entry(&splice
, struct btrfs_root
,
3937 list_del_init(&root
->delalloc_root
);
3938 root
= btrfs_grab_fs_root(root
);
3940 spin_unlock(&fs_info
->delalloc_root_lock
);
3942 btrfs_destroy_delalloc_inodes(root
);
3943 btrfs_put_fs_root(root
);
3945 spin_lock(&fs_info
->delalloc_root_lock
);
3947 spin_unlock(&fs_info
->delalloc_root_lock
);
3950 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3951 struct extent_io_tree
*dirty_pages
,
3955 struct extent_buffer
*eb
;
3960 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3965 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3966 while (start
<= end
) {
3967 eb
= btrfs_find_tree_block(root
, start
,
3969 start
+= root
->leafsize
;
3972 wait_on_extent_buffer_writeback(eb
);
3974 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3976 clear_extent_buffer_dirty(eb
);
3977 free_extent_buffer_stale(eb
);
3984 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3985 struct extent_io_tree
*pinned_extents
)
3987 struct extent_io_tree
*unpin
;
3993 unpin
= pinned_extents
;
3996 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3997 EXTENT_DIRTY
, NULL
);
4002 if (btrfs_test_opt(root
, DISCARD
))
4003 ret
= btrfs_error_discard_extent(root
, start
,
4007 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4008 btrfs_error_unpin_extent_range(root
, start
, end
);
4013 if (unpin
== &root
->fs_info
->freed_extents
[0])
4014 unpin
= &root
->fs_info
->freed_extents
[1];
4016 unpin
= &root
->fs_info
->freed_extents
[0];
4024 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4025 struct btrfs_root
*root
)
4027 btrfs_destroy_delayed_refs(cur_trans
, root
);
4028 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
4029 cur_trans
->dirty_pages
.dirty_bytes
);
4031 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4032 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4034 btrfs_evict_pending_snapshots(cur_trans
);
4036 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4037 wake_up(&root
->fs_info
->transaction_wait
);
4039 btrfs_destroy_delayed_inodes(root
);
4040 btrfs_assert_delayed_root_empty(root
);
4042 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4044 btrfs_destroy_pinned_extent(root
,
4045 root
->fs_info
->pinned_extents
);
4047 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4048 wake_up(&cur_trans
->commit_wait
);
4051 memset(cur_trans, 0, sizeof(*cur_trans));
4052 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4056 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4058 struct btrfs_transaction
*t
;
4061 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4063 spin_lock(&root
->fs_info
->trans_lock
);
4064 list_splice_init(&root
->fs_info
->trans_list
, &list
);
4065 root
->fs_info
->running_transaction
= NULL
;
4066 spin_unlock(&root
->fs_info
->trans_lock
);
4068 while (!list_empty(&list
)) {
4069 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
4071 btrfs_destroy_ordered_operations(t
, root
);
4073 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4075 btrfs_destroy_delayed_refs(t
, root
);
4078 * FIXME: cleanup wait for commit
4079 * We needn't acquire the lock here, because we are during
4080 * the umount, there is no other task which will change it.
4082 t
->state
= TRANS_STATE_COMMIT_START
;
4084 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
4085 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4087 btrfs_evict_pending_snapshots(t
);
4089 t
->state
= TRANS_STATE_UNBLOCKED
;
4091 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
4092 wake_up(&root
->fs_info
->transaction_wait
);
4094 btrfs_destroy_delayed_inodes(root
);
4095 btrfs_assert_delayed_root_empty(root
);
4097 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4099 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
4102 btrfs_destroy_pinned_extent(root
,
4103 root
->fs_info
->pinned_extents
);
4105 t
->state
= TRANS_STATE_COMPLETED
;
4107 if (waitqueue_active(&t
->commit_wait
))
4108 wake_up(&t
->commit_wait
);
4110 atomic_set(&t
->use_count
, 0);
4111 list_del_init(&t
->list
);
4112 memset(t
, 0, sizeof(*t
));
4113 kmem_cache_free(btrfs_transaction_cachep
, t
);
4116 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4121 static struct extent_io_ops btree_extent_io_ops
= {
4122 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4123 .readpage_io_failed_hook
= btree_io_failed_hook
,
4124 .submit_bio_hook
= btree_submit_bio_hook
,
4125 /* note we're sharing with inode.c for the merge bio hook */
4126 .merge_bio_hook
= btrfs_merge_bio_hook
,