2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static struct extent_io_ops btree_extent_io_ops
;
58 static void end_workqueue_fn(struct btrfs_work
*work
);
59 static void free_fs_root(struct btrfs_root
*root
);
60 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
62 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
64 struct btrfs_root
*root
);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
66 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
67 struct extent_io_tree
*dirty_pages
,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
70 struct extent_io_tree
*pinned_extents
);
71 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
72 static void btrfs_error_commit_super(struct btrfs_root
*root
);
75 * end_io_wq structs are used to do processing in task context when an IO is
76 * complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
83 struct btrfs_fs_info
*info
;
86 struct list_head list
;
87 struct btrfs_work work
;
91 * async submit bios are used to offload expensive checksumming
92 * onto the worker threads. They checksum file and metadata bios
93 * just before they are sent down the IO stack.
95 struct async_submit_bio
{
98 struct list_head list
;
99 extent_submit_bio_hook_t
*submit_bio_start
;
100 extent_submit_bio_hook_t
*submit_bio_done
;
103 unsigned long bio_flags
;
105 * bio_offset is optional, can be used if the pages in the bio
106 * can't tell us where in the file the bio should go
109 struct btrfs_work work
;
114 * Lockdep class keys for extent_buffer->lock's in this root. For a given
115 * eb, the lockdep key is determined by the btrfs_root it belongs to and
116 * the level the eb occupies in the tree.
118 * Different roots are used for different purposes and may nest inside each
119 * other and they require separate keysets. As lockdep keys should be
120 * static, assign keysets according to the purpose of the root as indicated
121 * by btrfs_root->objectid. This ensures that all special purpose roots
122 * have separate keysets.
124 * Lock-nesting across peer nodes is always done with the immediate parent
125 * node locked thus preventing deadlock. As lockdep doesn't know this, use
126 * subclass to avoid triggering lockdep warning in such cases.
128 * The key is set by the readpage_end_io_hook after the buffer has passed
129 * csum validation but before the pages are unlocked. It is also set by
130 * btrfs_init_new_buffer on freshly allocated blocks.
132 * We also add a check to make sure the highest level of the tree is the
133 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
134 * needs update as well.
136 #ifdef CONFIG_DEBUG_LOCK_ALLOC
137 # if BTRFS_MAX_LEVEL != 8
141 static struct btrfs_lockdep_keyset
{
142 u64 id
; /* root objectid */
143 const char *name_stem
; /* lock name stem */
144 char names
[BTRFS_MAX_LEVEL
+ 1][20];
145 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
146 } btrfs_lockdep_keysets
[] = {
147 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
148 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
149 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
150 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
151 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
152 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
153 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
154 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
155 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
156 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
157 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
158 { .id
= 0, .name_stem
= "tree" },
161 void __init
btrfs_init_lockdep(void)
165 /* initialize lockdep class names */
166 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
167 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
169 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
170 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
171 "btrfs-%s-%02d", ks
->name_stem
, j
);
175 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
178 struct btrfs_lockdep_keyset
*ks
;
180 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
182 /* find the matching keyset, id 0 is the default entry */
183 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
184 if (ks
->id
== objectid
)
187 lockdep_set_class_and_name(&eb
->lock
,
188 &ks
->keys
[level
], ks
->names
[level
]);
194 * extents on the btree inode are pretty simple, there's one extent
195 * that covers the entire device
197 static struct extent_map
*btree_get_extent(struct inode
*inode
,
198 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
201 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
202 struct extent_map
*em
;
205 read_lock(&em_tree
->lock
);
206 em
= lookup_extent_mapping(em_tree
, start
, len
);
209 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
210 read_unlock(&em_tree
->lock
);
213 read_unlock(&em_tree
->lock
);
215 em
= alloc_extent_map();
217 em
= ERR_PTR(-ENOMEM
);
222 em
->block_len
= (u64
)-1;
224 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
226 write_lock(&em_tree
->lock
);
227 ret
= add_extent_mapping(em_tree
, em
, 0);
228 if (ret
== -EEXIST
) {
230 em
= lookup_extent_mapping(em_tree
, start
, len
);
237 write_unlock(&em_tree
->lock
);
243 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
245 return btrfs_crc32c(seed
, data
, len
);
248 void btrfs_csum_final(u32 crc
, char *result
)
250 put_unaligned_le32(~crc
, result
);
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
257 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
260 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
263 unsigned long cur_len
;
264 unsigned long offset
= BTRFS_CSUM_SIZE
;
266 unsigned long map_start
;
267 unsigned long map_len
;
270 unsigned long inline_result
;
272 len
= buf
->len
- offset
;
274 err
= map_private_extent_buffer(buf
, offset
, 32,
275 &kaddr
, &map_start
, &map_len
);
278 cur_len
= min(len
, map_len
- (offset
- map_start
));
279 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
284 if (csum_size
> sizeof(inline_result
)) {
285 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
289 result
= (char *)&inline_result
;
292 btrfs_csum_final(crc
, result
);
295 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
298 memcpy(&found
, result
, csum_size
);
300 read_extent_buffer(buf
, &val
, 0, csum_size
);
301 printk_ratelimited(KERN_INFO
302 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
304 root
->fs_info
->sb
->s_id
, buf
->start
,
305 val
, found
, btrfs_header_level(buf
));
306 if (result
!= (char *)&inline_result
)
311 write_extent_buffer(buf
, result
, 0, csum_size
);
313 if (result
!= (char *)&inline_result
)
319 * we can't consider a given block up to date unless the transid of the
320 * block matches the transid in the parent node's pointer. This is how we
321 * detect blocks that either didn't get written at all or got written
322 * in the wrong place.
324 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
325 struct extent_buffer
*eb
, u64 parent_transid
,
328 struct extent_state
*cached_state
= NULL
;
330 bool need_lock
= (current
->journal_info
==
331 (void *)BTRFS_SEND_TRANS_STUB
);
333 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
340 btrfs_tree_read_lock(eb
);
341 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
344 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
346 if (extent_buffer_uptodate(eb
) &&
347 btrfs_header_generation(eb
) == parent_transid
) {
351 printk_ratelimited(KERN_INFO
"BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
352 eb
->fs_info
->sb
->s_id
, eb
->start
,
353 parent_transid
, btrfs_header_generation(eb
));
357 * Things reading via commit roots that don't have normal protection,
358 * like send, can have a really old block in cache that may point at a
359 * block that has been free'd and re-allocated. So don't clear uptodate
360 * if we find an eb that is under IO (dirty/writeback) because we could
361 * end up reading in the stale data and then writing it back out and
362 * making everybody very sad.
364 if (!extent_buffer_under_io(eb
))
365 clear_extent_buffer_uptodate(eb
);
367 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
368 &cached_state
, GFP_NOFS
);
370 btrfs_tree_read_unlock_blocking(eb
);
375 * Return 0 if the superblock checksum type matches the checksum value of that
376 * algorithm. Pass the raw disk superblock data.
378 static int btrfs_check_super_csum(char *raw_disk_sb
)
380 struct btrfs_super_block
*disk_sb
=
381 (struct btrfs_super_block
*)raw_disk_sb
;
382 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
385 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
387 const int csum_size
= sizeof(crc
);
388 char result
[csum_size
];
391 * The super_block structure does not span the whole
392 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
393 * is filled with zeros and is included in the checkum.
395 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
396 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
397 btrfs_csum_final(crc
, result
);
399 if (memcmp(raw_disk_sb
, result
, csum_size
))
402 if (ret
&& btrfs_super_generation(disk_sb
) < 10) {
404 "BTRFS: super block crcs don't match, older mkfs detected\n");
409 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
410 printk(KERN_ERR
"BTRFS: unsupported checksum algorithm %u\n",
419 * helper to read a given tree block, doing retries as required when
420 * the checksums don't match and we have alternate mirrors to try.
422 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
423 struct extent_buffer
*eb
,
424 u64 start
, u64 parent_transid
)
426 struct extent_io_tree
*io_tree
;
431 int failed_mirror
= 0;
433 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
434 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
436 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
438 btree_get_extent
, mirror_num
);
440 if (!verify_parent_transid(io_tree
, eb
,
448 * This buffer's crc is fine, but its contents are corrupted, so
449 * there is no reason to read the other copies, they won't be
452 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
455 num_copies
= btrfs_num_copies(root
->fs_info
,
460 if (!failed_mirror
) {
462 failed_mirror
= eb
->read_mirror
;
466 if (mirror_num
== failed_mirror
)
469 if (mirror_num
> num_copies
)
473 if (failed
&& !ret
&& failed_mirror
)
474 repair_eb_io_failure(root
, eb
, failed_mirror
);
480 * checksum a dirty tree block before IO. This has extra checks to make sure
481 * we only fill in the checksum field in the first page of a multi-page block
484 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
486 u64 start
= page_offset(page
);
488 struct extent_buffer
*eb
;
490 eb
= (struct extent_buffer
*)page
->private;
491 if (page
!= eb
->pages
[0])
493 found_start
= btrfs_header_bytenr(eb
);
494 if (WARN_ON(found_start
!= start
|| !PageUptodate(page
)))
496 csum_tree_block(root
, eb
, 0);
500 static int check_tree_block_fsid(struct btrfs_root
*root
,
501 struct extent_buffer
*eb
)
503 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
504 u8 fsid
[BTRFS_UUID_SIZE
];
507 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
509 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
513 fs_devices
= fs_devices
->seed
;
518 #define CORRUPT(reason, eb, root, slot) \
519 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
520 "root=%llu, slot=%d", reason, \
521 btrfs_header_bytenr(eb), root->objectid, slot)
523 static noinline
int check_leaf(struct btrfs_root
*root
,
524 struct extent_buffer
*leaf
)
526 struct btrfs_key key
;
527 struct btrfs_key leaf_key
;
528 u32 nritems
= btrfs_header_nritems(leaf
);
534 /* Check the 0 item */
535 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
536 BTRFS_LEAF_DATA_SIZE(root
)) {
537 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
542 * Check to make sure each items keys are in the correct order and their
543 * offsets make sense. We only have to loop through nritems-1 because
544 * we check the current slot against the next slot, which verifies the
545 * next slot's offset+size makes sense and that the current's slot
548 for (slot
= 0; slot
< nritems
- 1; slot
++) {
549 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
550 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
552 /* Make sure the keys are in the right order */
553 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
554 CORRUPT("bad key order", leaf
, root
, slot
);
559 * Make sure the offset and ends are right, remember that the
560 * item data starts at the end of the leaf and grows towards the
563 if (btrfs_item_offset_nr(leaf
, slot
) !=
564 btrfs_item_end_nr(leaf
, slot
+ 1)) {
565 CORRUPT("slot offset bad", leaf
, root
, slot
);
570 * Check to make sure that we don't point outside of the leaf,
571 * just incase all the items are consistent to eachother, but
572 * all point outside of the leaf.
574 if (btrfs_item_end_nr(leaf
, slot
) >
575 BTRFS_LEAF_DATA_SIZE(root
)) {
576 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
584 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
585 u64 phy_offset
, struct page
*page
,
586 u64 start
, u64 end
, int mirror
)
590 struct extent_buffer
*eb
;
591 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
598 eb
= (struct extent_buffer
*)page
->private;
600 /* the pending IO might have been the only thing that kept this buffer
601 * in memory. Make sure we have a ref for all this other checks
603 extent_buffer_get(eb
);
605 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
609 eb
->read_mirror
= mirror
;
610 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
615 found_start
= btrfs_header_bytenr(eb
);
616 if (found_start
!= eb
->start
) {
617 printk_ratelimited(KERN_INFO
"BTRFS (device %s): bad tree block start "
619 eb
->fs_info
->sb
->s_id
, found_start
, eb
->start
);
623 if (check_tree_block_fsid(root
, eb
)) {
624 printk_ratelimited(KERN_INFO
"BTRFS (device %s): bad fsid on block %llu\n",
625 eb
->fs_info
->sb
->s_id
, eb
->start
);
629 found_level
= btrfs_header_level(eb
);
630 if (found_level
>= BTRFS_MAX_LEVEL
) {
631 btrfs_info(root
->fs_info
, "bad tree block level %d",
632 (int)btrfs_header_level(eb
));
637 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
640 ret
= csum_tree_block(root
, eb
, 1);
647 * If this is a leaf block and it is corrupt, set the corrupt bit so
648 * that we don't try and read the other copies of this block, just
651 if (found_level
== 0 && check_leaf(root
, eb
)) {
652 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
657 set_extent_buffer_uptodate(eb
);
660 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
661 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
665 * our io error hook is going to dec the io pages
666 * again, we have to make sure it has something
669 atomic_inc(&eb
->io_pages
);
670 clear_extent_buffer_uptodate(eb
);
672 free_extent_buffer(eb
);
677 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
679 struct extent_buffer
*eb
;
680 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
682 eb
= (struct extent_buffer
*)page
->private;
683 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
684 eb
->read_mirror
= failed_mirror
;
685 atomic_dec(&eb
->io_pages
);
686 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
687 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
688 return -EIO
; /* we fixed nothing */
691 static void end_workqueue_bio(struct bio
*bio
, int err
)
693 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
694 struct btrfs_fs_info
*fs_info
;
695 struct btrfs_workqueue
*wq
;
696 btrfs_work_func_t func
;
698 fs_info
= end_io_wq
->info
;
699 end_io_wq
->error
= err
;
701 if (bio
->bi_rw
& REQ_WRITE
) {
702 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
703 wq
= fs_info
->endio_meta_write_workers
;
704 func
= btrfs_endio_meta_write_helper
;
705 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
706 wq
= fs_info
->endio_freespace_worker
;
707 func
= btrfs_freespace_write_helper
;
708 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
709 wq
= fs_info
->endio_raid56_workers
;
710 func
= btrfs_endio_raid56_helper
;
712 wq
= fs_info
->endio_write_workers
;
713 func
= btrfs_endio_write_helper
;
716 if (unlikely(end_io_wq
->metadata
==
717 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
718 wq
= fs_info
->endio_repair_workers
;
719 func
= btrfs_endio_repair_helper
;
720 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
721 wq
= fs_info
->endio_raid56_workers
;
722 func
= btrfs_endio_raid56_helper
;
723 } else if (end_io_wq
->metadata
) {
724 wq
= fs_info
->endio_meta_workers
;
725 func
= btrfs_endio_meta_helper
;
727 wq
= fs_info
->endio_workers
;
728 func
= btrfs_endio_helper
;
732 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
733 btrfs_queue_work(wq
, &end_io_wq
->work
);
737 * For the metadata arg you want
740 * 1 - if normal metadta
741 * 2 - if writing to the free space cache area
742 * 3 - raid parity work
744 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
747 struct end_io_wq
*end_io_wq
;
749 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
753 end_io_wq
->private = bio
->bi_private
;
754 end_io_wq
->end_io
= bio
->bi_end_io
;
755 end_io_wq
->info
= info
;
756 end_io_wq
->error
= 0;
757 end_io_wq
->bio
= bio
;
758 end_io_wq
->metadata
= metadata
;
760 bio
->bi_private
= end_io_wq
;
761 bio
->bi_end_io
= end_workqueue_bio
;
765 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
767 unsigned long limit
= min_t(unsigned long,
768 info
->thread_pool_size
,
769 info
->fs_devices
->open_devices
);
773 static void run_one_async_start(struct btrfs_work
*work
)
775 struct async_submit_bio
*async
;
778 async
= container_of(work
, struct async_submit_bio
, work
);
779 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
780 async
->mirror_num
, async
->bio_flags
,
786 static void run_one_async_done(struct btrfs_work
*work
)
788 struct btrfs_fs_info
*fs_info
;
789 struct async_submit_bio
*async
;
792 async
= container_of(work
, struct async_submit_bio
, work
);
793 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
795 limit
= btrfs_async_submit_limit(fs_info
);
796 limit
= limit
* 2 / 3;
798 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
799 waitqueue_active(&fs_info
->async_submit_wait
))
800 wake_up(&fs_info
->async_submit_wait
);
802 /* If an error occured we just want to clean up the bio and move on */
804 bio_endio(async
->bio
, async
->error
);
808 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
809 async
->mirror_num
, async
->bio_flags
,
813 static void run_one_async_free(struct btrfs_work
*work
)
815 struct async_submit_bio
*async
;
817 async
= container_of(work
, struct async_submit_bio
, work
);
821 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
822 int rw
, struct bio
*bio
, int mirror_num
,
823 unsigned long bio_flags
,
825 extent_submit_bio_hook_t
*submit_bio_start
,
826 extent_submit_bio_hook_t
*submit_bio_done
)
828 struct async_submit_bio
*async
;
830 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
834 async
->inode
= inode
;
837 async
->mirror_num
= mirror_num
;
838 async
->submit_bio_start
= submit_bio_start
;
839 async
->submit_bio_done
= submit_bio_done
;
841 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
842 run_one_async_done
, run_one_async_free
);
844 async
->bio_flags
= bio_flags
;
845 async
->bio_offset
= bio_offset
;
849 atomic_inc(&fs_info
->nr_async_submits
);
852 btrfs_set_work_high_priority(&async
->work
);
854 btrfs_queue_work(fs_info
->workers
, &async
->work
);
856 while (atomic_read(&fs_info
->async_submit_draining
) &&
857 atomic_read(&fs_info
->nr_async_submits
)) {
858 wait_event(fs_info
->async_submit_wait
,
859 (atomic_read(&fs_info
->nr_async_submits
) == 0));
865 static int btree_csum_one_bio(struct bio
*bio
)
867 struct bio_vec
*bvec
;
868 struct btrfs_root
*root
;
871 bio_for_each_segment_all(bvec
, bio
, i
) {
872 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
873 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
881 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
882 struct bio
*bio
, int mirror_num
,
883 unsigned long bio_flags
,
887 * when we're called for a write, we're already in the async
888 * submission context. Just jump into btrfs_map_bio
890 return btree_csum_one_bio(bio
);
893 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
894 int mirror_num
, unsigned long bio_flags
,
900 * when we're called for a write, we're already in the async
901 * submission context. Just jump into btrfs_map_bio
903 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
909 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
911 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
920 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
921 int mirror_num
, unsigned long bio_flags
,
924 int async
= check_async_write(inode
, bio_flags
);
927 if (!(rw
& REQ_WRITE
)) {
929 * called for a read, do the setup so that checksum validation
930 * can happen in the async kernel threads
932 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
936 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
939 ret
= btree_csum_one_bio(bio
);
942 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
946 * kthread helpers are used to submit writes so that
947 * checksumming can happen in parallel across all CPUs
949 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
950 inode
, rw
, bio
, mirror_num
, 0,
952 __btree_submit_bio_start
,
953 __btree_submit_bio_done
);
963 #ifdef CONFIG_MIGRATION
964 static int btree_migratepage(struct address_space
*mapping
,
965 struct page
*newpage
, struct page
*page
,
966 enum migrate_mode mode
)
969 * we can't safely write a btree page from here,
970 * we haven't done the locking hook
975 * Buffers may be managed in a filesystem specific way.
976 * We must have no buffers or drop them.
978 if (page_has_private(page
) &&
979 !try_to_release_page(page
, GFP_KERNEL
))
981 return migrate_page(mapping
, newpage
, page
, mode
);
986 static int btree_writepages(struct address_space
*mapping
,
987 struct writeback_control
*wbc
)
989 struct btrfs_fs_info
*fs_info
;
992 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
994 if (wbc
->for_kupdate
)
997 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
998 /* this is a bit racy, but that's ok */
999 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1000 BTRFS_DIRTY_METADATA_THRESH
);
1004 return btree_write_cache_pages(mapping
, wbc
);
1007 static int btree_readpage(struct file
*file
, struct page
*page
)
1009 struct extent_io_tree
*tree
;
1010 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1011 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1014 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1016 if (PageWriteback(page
) || PageDirty(page
))
1019 return try_release_extent_buffer(page
);
1022 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1023 unsigned int length
)
1025 struct extent_io_tree
*tree
;
1026 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1027 extent_invalidatepage(tree
, page
, offset
);
1028 btree_releasepage(page
, GFP_NOFS
);
1029 if (PagePrivate(page
)) {
1030 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1031 "page private not zero on page %llu",
1032 (unsigned long long)page_offset(page
));
1033 ClearPagePrivate(page
);
1034 set_page_private(page
, 0);
1035 page_cache_release(page
);
1039 static int btree_set_page_dirty(struct page
*page
)
1042 struct extent_buffer
*eb
;
1044 BUG_ON(!PagePrivate(page
));
1045 eb
= (struct extent_buffer
*)page
->private;
1047 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1048 BUG_ON(!atomic_read(&eb
->refs
));
1049 btrfs_assert_tree_locked(eb
);
1051 return __set_page_dirty_nobuffers(page
);
1054 static const struct address_space_operations btree_aops
= {
1055 .readpage
= btree_readpage
,
1056 .writepages
= btree_writepages
,
1057 .releasepage
= btree_releasepage
,
1058 .invalidatepage
= btree_invalidatepage
,
1059 #ifdef CONFIG_MIGRATION
1060 .migratepage
= btree_migratepage
,
1062 .set_page_dirty
= btree_set_page_dirty
,
1065 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
)
1067 struct extent_buffer
*buf
= NULL
;
1068 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1070 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1073 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1074 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1075 free_extent_buffer(buf
);
1078 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1079 int mirror_num
, struct extent_buffer
**eb
)
1081 struct extent_buffer
*buf
= NULL
;
1082 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1083 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1086 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1090 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1092 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1093 btree_get_extent
, mirror_num
);
1095 free_extent_buffer(buf
);
1099 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1100 free_extent_buffer(buf
);
1102 } else if (extent_buffer_uptodate(buf
)) {
1105 free_extent_buffer(buf
);
1110 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1111 u64 bytenr
, u32 blocksize
)
1113 return find_extent_buffer(root
->fs_info
, bytenr
);
1116 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1117 u64 bytenr
, u32 blocksize
)
1119 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1120 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
)))
1121 return alloc_test_extent_buffer(root
->fs_info
, bytenr
,
1124 return alloc_extent_buffer(root
->fs_info
, bytenr
, blocksize
);
1128 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1130 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1131 buf
->start
+ buf
->len
- 1);
1134 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1136 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1137 buf
->start
, buf
->start
+ buf
->len
- 1);
1140 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1143 struct extent_buffer
*buf
= NULL
;
1146 buf
= btrfs_find_create_tree_block(root
, bytenr
, root
->nodesize
);
1150 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1152 free_extent_buffer(buf
);
1159 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1160 struct extent_buffer
*buf
)
1162 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1164 if (btrfs_header_generation(buf
) ==
1165 fs_info
->running_transaction
->transid
) {
1166 btrfs_assert_tree_locked(buf
);
1168 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1169 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1171 fs_info
->dirty_metadata_batch
);
1172 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1173 btrfs_set_lock_blocking(buf
);
1174 clear_extent_buffer_dirty(buf
);
1179 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1181 struct btrfs_subvolume_writers
*writers
;
1184 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1186 return ERR_PTR(-ENOMEM
);
1188 ret
= percpu_counter_init(&writers
->counter
, 0);
1191 return ERR_PTR(ret
);
1194 init_waitqueue_head(&writers
->wait
);
1199 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1201 percpu_counter_destroy(&writers
->counter
);
1205 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1206 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1210 root
->commit_root
= NULL
;
1211 root
->sectorsize
= sectorsize
;
1212 root
->nodesize
= nodesize
;
1213 root
->stripesize
= stripesize
;
1215 root
->orphan_cleanup_state
= 0;
1217 root
->objectid
= objectid
;
1218 root
->last_trans
= 0;
1219 root
->highest_objectid
= 0;
1220 root
->nr_delalloc_inodes
= 0;
1221 root
->nr_ordered_extents
= 0;
1223 root
->inode_tree
= RB_ROOT
;
1224 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1225 root
->block_rsv
= NULL
;
1226 root
->orphan_block_rsv
= NULL
;
1228 INIT_LIST_HEAD(&root
->dirty_list
);
1229 INIT_LIST_HEAD(&root
->root_list
);
1230 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1231 INIT_LIST_HEAD(&root
->delalloc_root
);
1232 INIT_LIST_HEAD(&root
->ordered_extents
);
1233 INIT_LIST_HEAD(&root
->ordered_root
);
1234 INIT_LIST_HEAD(&root
->logged_list
[0]);
1235 INIT_LIST_HEAD(&root
->logged_list
[1]);
1236 spin_lock_init(&root
->orphan_lock
);
1237 spin_lock_init(&root
->inode_lock
);
1238 spin_lock_init(&root
->delalloc_lock
);
1239 spin_lock_init(&root
->ordered_extent_lock
);
1240 spin_lock_init(&root
->accounting_lock
);
1241 spin_lock_init(&root
->log_extents_lock
[0]);
1242 spin_lock_init(&root
->log_extents_lock
[1]);
1243 mutex_init(&root
->objectid_mutex
);
1244 mutex_init(&root
->log_mutex
);
1245 mutex_init(&root
->ordered_extent_mutex
);
1246 mutex_init(&root
->delalloc_mutex
);
1247 init_waitqueue_head(&root
->log_writer_wait
);
1248 init_waitqueue_head(&root
->log_commit_wait
[0]);
1249 init_waitqueue_head(&root
->log_commit_wait
[1]);
1250 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1251 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1252 atomic_set(&root
->log_commit
[0], 0);
1253 atomic_set(&root
->log_commit
[1], 0);
1254 atomic_set(&root
->log_writers
, 0);
1255 atomic_set(&root
->log_batch
, 0);
1256 atomic_set(&root
->orphan_inodes
, 0);
1257 atomic_set(&root
->refs
, 1);
1258 atomic_set(&root
->will_be_snapshoted
, 0);
1259 root
->log_transid
= 0;
1260 root
->log_transid_committed
= -1;
1261 root
->last_log_commit
= 0;
1263 extent_io_tree_init(&root
->dirty_log_pages
,
1264 fs_info
->btree_inode
->i_mapping
);
1266 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1267 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1268 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1269 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1271 root
->defrag_trans_start
= fs_info
->generation
;
1273 root
->defrag_trans_start
= 0;
1274 init_completion(&root
->kobj_unregister
);
1275 root
->root_key
.objectid
= objectid
;
1278 spin_lock_init(&root
->root_item_lock
);
1281 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1283 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1285 root
->fs_info
= fs_info
;
1289 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1290 /* Should only be used by the testing infrastructure */
1291 struct btrfs_root
*btrfs_alloc_dummy_root(void)
1293 struct btrfs_root
*root
;
1295 root
= btrfs_alloc_root(NULL
);
1297 return ERR_PTR(-ENOMEM
);
1298 __setup_root(4096, 4096, 4096, root
, NULL
, 1);
1299 set_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
);
1300 root
->alloc_bytenr
= 0;
1306 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1307 struct btrfs_fs_info
*fs_info
,
1310 struct extent_buffer
*leaf
;
1311 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1312 struct btrfs_root
*root
;
1313 struct btrfs_key key
;
1317 root
= btrfs_alloc_root(fs_info
);
1319 return ERR_PTR(-ENOMEM
);
1321 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1322 tree_root
->stripesize
, root
, fs_info
, objectid
);
1323 root
->root_key
.objectid
= objectid
;
1324 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1325 root
->root_key
.offset
= 0;
1327 leaf
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
1328 0, objectid
, NULL
, 0, 0, 0);
1330 ret
= PTR_ERR(leaf
);
1335 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1336 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1337 btrfs_set_header_generation(leaf
, trans
->transid
);
1338 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1339 btrfs_set_header_owner(leaf
, objectid
);
1342 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1344 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1345 btrfs_header_chunk_tree_uuid(leaf
),
1347 btrfs_mark_buffer_dirty(leaf
);
1349 root
->commit_root
= btrfs_root_node(root
);
1350 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1352 root
->root_item
.flags
= 0;
1353 root
->root_item
.byte_limit
= 0;
1354 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1355 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1356 btrfs_set_root_level(&root
->root_item
, 0);
1357 btrfs_set_root_refs(&root
->root_item
, 1);
1358 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1359 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1360 btrfs_set_root_dirid(&root
->root_item
, 0);
1362 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1363 root
->root_item
.drop_level
= 0;
1365 key
.objectid
= objectid
;
1366 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1368 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1372 btrfs_tree_unlock(leaf
);
1378 btrfs_tree_unlock(leaf
);
1379 free_extent_buffer(root
->commit_root
);
1380 free_extent_buffer(leaf
);
1384 return ERR_PTR(ret
);
1387 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1388 struct btrfs_fs_info
*fs_info
)
1390 struct btrfs_root
*root
;
1391 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1392 struct extent_buffer
*leaf
;
1394 root
= btrfs_alloc_root(fs_info
);
1396 return ERR_PTR(-ENOMEM
);
1398 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1399 tree_root
->stripesize
, root
, fs_info
,
1400 BTRFS_TREE_LOG_OBJECTID
);
1402 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1403 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1404 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1407 * DON'T set REF_COWS for log trees
1409 * log trees do not get reference counted because they go away
1410 * before a real commit is actually done. They do store pointers
1411 * to file data extents, and those reference counts still get
1412 * updated (along with back refs to the log tree).
1415 leaf
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
1416 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1420 return ERR_CAST(leaf
);
1423 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1424 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1425 btrfs_set_header_generation(leaf
, trans
->transid
);
1426 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1427 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1430 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1431 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1432 btrfs_mark_buffer_dirty(root
->node
);
1433 btrfs_tree_unlock(root
->node
);
1437 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1438 struct btrfs_fs_info
*fs_info
)
1440 struct btrfs_root
*log_root
;
1442 log_root
= alloc_log_tree(trans
, fs_info
);
1443 if (IS_ERR(log_root
))
1444 return PTR_ERR(log_root
);
1445 WARN_ON(fs_info
->log_root_tree
);
1446 fs_info
->log_root_tree
= log_root
;
1450 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1451 struct btrfs_root
*root
)
1453 struct btrfs_root
*log_root
;
1454 struct btrfs_inode_item
*inode_item
;
1456 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1457 if (IS_ERR(log_root
))
1458 return PTR_ERR(log_root
);
1460 log_root
->last_trans
= trans
->transid
;
1461 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1463 inode_item
= &log_root
->root_item
.inode
;
1464 btrfs_set_stack_inode_generation(inode_item
, 1);
1465 btrfs_set_stack_inode_size(inode_item
, 3);
1466 btrfs_set_stack_inode_nlink(inode_item
, 1);
1467 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1468 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1470 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1472 WARN_ON(root
->log_root
);
1473 root
->log_root
= log_root
;
1474 root
->log_transid
= 0;
1475 root
->log_transid_committed
= -1;
1476 root
->last_log_commit
= 0;
1480 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1481 struct btrfs_key
*key
)
1483 struct btrfs_root
*root
;
1484 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1485 struct btrfs_path
*path
;
1489 path
= btrfs_alloc_path();
1491 return ERR_PTR(-ENOMEM
);
1493 root
= btrfs_alloc_root(fs_info
);
1499 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1500 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1502 ret
= btrfs_find_root(tree_root
, key
, path
,
1503 &root
->root_item
, &root
->root_key
);
1510 generation
= btrfs_root_generation(&root
->root_item
);
1511 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1516 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1520 root
->commit_root
= btrfs_root_node(root
);
1522 btrfs_free_path(path
);
1526 free_extent_buffer(root
->node
);
1530 root
= ERR_PTR(ret
);
1534 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1535 struct btrfs_key
*location
)
1537 struct btrfs_root
*root
;
1539 root
= btrfs_read_tree_root(tree_root
, location
);
1543 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1544 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1545 btrfs_check_and_init_root_item(&root
->root_item
);
1551 int btrfs_init_fs_root(struct btrfs_root
*root
)
1554 struct btrfs_subvolume_writers
*writers
;
1556 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1557 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1559 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1564 writers
= btrfs_alloc_subvolume_writers();
1565 if (IS_ERR(writers
)) {
1566 ret
= PTR_ERR(writers
);
1569 root
->subv_writers
= writers
;
1571 btrfs_init_free_ino_ctl(root
);
1572 spin_lock_init(&root
->ino_cache_lock
);
1573 init_waitqueue_head(&root
->ino_cache_wait
);
1575 ret
= get_anon_bdev(&root
->anon_dev
);
1581 btrfs_free_subvolume_writers(root
->subv_writers
);
1583 kfree(root
->free_ino_ctl
);
1584 kfree(root
->free_ino_pinned
);
1588 static struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1591 struct btrfs_root
*root
;
1593 spin_lock(&fs_info
->fs_roots_radix_lock
);
1594 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1595 (unsigned long)root_id
);
1596 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1600 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1601 struct btrfs_root
*root
)
1605 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1609 spin_lock(&fs_info
->fs_roots_radix_lock
);
1610 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1611 (unsigned long)root
->root_key
.objectid
,
1614 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1615 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1616 radix_tree_preload_end();
1621 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1622 struct btrfs_key
*location
,
1625 struct btrfs_root
*root
;
1628 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1629 return fs_info
->tree_root
;
1630 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1631 return fs_info
->extent_root
;
1632 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1633 return fs_info
->chunk_root
;
1634 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1635 return fs_info
->dev_root
;
1636 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1637 return fs_info
->csum_root
;
1638 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1639 return fs_info
->quota_root
? fs_info
->quota_root
:
1641 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1642 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1645 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1647 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1648 return ERR_PTR(-ENOENT
);
1652 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1656 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1661 ret
= btrfs_init_fs_root(root
);
1665 ret
= btrfs_find_item(fs_info
->tree_root
, NULL
, BTRFS_ORPHAN_OBJECTID
,
1666 location
->objectid
, BTRFS_ORPHAN_ITEM_KEY
, NULL
);
1670 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1672 ret
= btrfs_insert_fs_root(fs_info
, root
);
1674 if (ret
== -EEXIST
) {
1683 return ERR_PTR(ret
);
1686 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1688 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1690 struct btrfs_device
*device
;
1691 struct backing_dev_info
*bdi
;
1694 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1697 bdi
= blk_get_backing_dev_info(device
->bdev
);
1698 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1707 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1711 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1712 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1716 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1717 bdi
->congested_fn
= btrfs_congested_fn
;
1718 bdi
->congested_data
= info
;
1723 * called by the kthread helper functions to finally call the bio end_io
1724 * functions. This is where read checksum verification actually happens
1726 static void end_workqueue_fn(struct btrfs_work
*work
)
1729 struct end_io_wq
*end_io_wq
;
1732 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1733 bio
= end_io_wq
->bio
;
1735 error
= end_io_wq
->error
;
1736 bio
->bi_private
= end_io_wq
->private;
1737 bio
->bi_end_io
= end_io_wq
->end_io
;
1739 bio_endio_nodec(bio
, error
);
1742 static int cleaner_kthread(void *arg
)
1744 struct btrfs_root
*root
= arg
;
1750 /* Make the cleaner go to sleep early. */
1751 if (btrfs_need_cleaner_sleep(root
))
1754 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1758 * Avoid the problem that we change the status of the fs
1759 * during the above check and trylock.
1761 if (btrfs_need_cleaner_sleep(root
)) {
1762 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1766 btrfs_run_delayed_iputs(root
);
1767 btrfs_delete_unused_bgs(root
->fs_info
);
1768 again
= btrfs_clean_one_deleted_snapshot(root
);
1769 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1772 * The defragger has dealt with the R/O remount and umount,
1773 * needn't do anything special here.
1775 btrfs_run_defrag_inodes(root
->fs_info
);
1777 if (!try_to_freeze() && !again
) {
1778 set_current_state(TASK_INTERRUPTIBLE
);
1779 if (!kthread_should_stop())
1781 __set_current_state(TASK_RUNNING
);
1783 } while (!kthread_should_stop());
1787 static int transaction_kthread(void *arg
)
1789 struct btrfs_root
*root
= arg
;
1790 struct btrfs_trans_handle
*trans
;
1791 struct btrfs_transaction
*cur
;
1794 unsigned long delay
;
1798 cannot_commit
= false;
1799 delay
= HZ
* root
->fs_info
->commit_interval
;
1800 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1802 spin_lock(&root
->fs_info
->trans_lock
);
1803 cur
= root
->fs_info
->running_transaction
;
1805 spin_unlock(&root
->fs_info
->trans_lock
);
1809 now
= get_seconds();
1810 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1811 (now
< cur
->start_time
||
1812 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1813 spin_unlock(&root
->fs_info
->trans_lock
);
1817 transid
= cur
->transid
;
1818 spin_unlock(&root
->fs_info
->trans_lock
);
1820 /* If the file system is aborted, this will always fail. */
1821 trans
= btrfs_attach_transaction(root
);
1822 if (IS_ERR(trans
)) {
1823 if (PTR_ERR(trans
) != -ENOENT
)
1824 cannot_commit
= true;
1827 if (transid
== trans
->transid
) {
1828 btrfs_commit_transaction(trans
, root
);
1830 btrfs_end_transaction(trans
, root
);
1833 wake_up_process(root
->fs_info
->cleaner_kthread
);
1834 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1836 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1837 &root
->fs_info
->fs_state
)))
1838 btrfs_cleanup_transaction(root
);
1839 if (!try_to_freeze()) {
1840 set_current_state(TASK_INTERRUPTIBLE
);
1841 if (!kthread_should_stop() &&
1842 (!btrfs_transaction_blocked(root
->fs_info
) ||
1844 schedule_timeout(delay
);
1845 __set_current_state(TASK_RUNNING
);
1847 } while (!kthread_should_stop());
1852 * this will find the highest generation in the array of
1853 * root backups. The index of the highest array is returned,
1854 * or -1 if we can't find anything.
1856 * We check to make sure the array is valid by comparing the
1857 * generation of the latest root in the array with the generation
1858 * in the super block. If they don't match we pitch it.
1860 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1863 int newest_index
= -1;
1864 struct btrfs_root_backup
*root_backup
;
1867 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1868 root_backup
= info
->super_copy
->super_roots
+ i
;
1869 cur
= btrfs_backup_tree_root_gen(root_backup
);
1870 if (cur
== newest_gen
)
1874 /* check to see if we actually wrapped around */
1875 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1876 root_backup
= info
->super_copy
->super_roots
;
1877 cur
= btrfs_backup_tree_root_gen(root_backup
);
1878 if (cur
== newest_gen
)
1881 return newest_index
;
1886 * find the oldest backup so we know where to store new entries
1887 * in the backup array. This will set the backup_root_index
1888 * field in the fs_info struct
1890 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1893 int newest_index
= -1;
1895 newest_index
= find_newest_super_backup(info
, newest_gen
);
1896 /* if there was garbage in there, just move along */
1897 if (newest_index
== -1) {
1898 info
->backup_root_index
= 0;
1900 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1905 * copy all the root pointers into the super backup array.
1906 * this will bump the backup pointer by one when it is
1909 static void backup_super_roots(struct btrfs_fs_info
*info
)
1912 struct btrfs_root_backup
*root_backup
;
1915 next_backup
= info
->backup_root_index
;
1916 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1917 BTRFS_NUM_BACKUP_ROOTS
;
1920 * just overwrite the last backup if we're at the same generation
1921 * this happens only at umount
1923 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1924 if (btrfs_backup_tree_root_gen(root_backup
) ==
1925 btrfs_header_generation(info
->tree_root
->node
))
1926 next_backup
= last_backup
;
1928 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1931 * make sure all of our padding and empty slots get zero filled
1932 * regardless of which ones we use today
1934 memset(root_backup
, 0, sizeof(*root_backup
));
1936 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1938 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1939 btrfs_set_backup_tree_root_gen(root_backup
,
1940 btrfs_header_generation(info
->tree_root
->node
));
1942 btrfs_set_backup_tree_root_level(root_backup
,
1943 btrfs_header_level(info
->tree_root
->node
));
1945 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1946 btrfs_set_backup_chunk_root_gen(root_backup
,
1947 btrfs_header_generation(info
->chunk_root
->node
));
1948 btrfs_set_backup_chunk_root_level(root_backup
,
1949 btrfs_header_level(info
->chunk_root
->node
));
1951 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1952 btrfs_set_backup_extent_root_gen(root_backup
,
1953 btrfs_header_generation(info
->extent_root
->node
));
1954 btrfs_set_backup_extent_root_level(root_backup
,
1955 btrfs_header_level(info
->extent_root
->node
));
1958 * we might commit during log recovery, which happens before we set
1959 * the fs_root. Make sure it is valid before we fill it in.
1961 if (info
->fs_root
&& info
->fs_root
->node
) {
1962 btrfs_set_backup_fs_root(root_backup
,
1963 info
->fs_root
->node
->start
);
1964 btrfs_set_backup_fs_root_gen(root_backup
,
1965 btrfs_header_generation(info
->fs_root
->node
));
1966 btrfs_set_backup_fs_root_level(root_backup
,
1967 btrfs_header_level(info
->fs_root
->node
));
1970 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1971 btrfs_set_backup_dev_root_gen(root_backup
,
1972 btrfs_header_generation(info
->dev_root
->node
));
1973 btrfs_set_backup_dev_root_level(root_backup
,
1974 btrfs_header_level(info
->dev_root
->node
));
1976 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1977 btrfs_set_backup_csum_root_gen(root_backup
,
1978 btrfs_header_generation(info
->csum_root
->node
));
1979 btrfs_set_backup_csum_root_level(root_backup
,
1980 btrfs_header_level(info
->csum_root
->node
));
1982 btrfs_set_backup_total_bytes(root_backup
,
1983 btrfs_super_total_bytes(info
->super_copy
));
1984 btrfs_set_backup_bytes_used(root_backup
,
1985 btrfs_super_bytes_used(info
->super_copy
));
1986 btrfs_set_backup_num_devices(root_backup
,
1987 btrfs_super_num_devices(info
->super_copy
));
1990 * if we don't copy this out to the super_copy, it won't get remembered
1991 * for the next commit
1993 memcpy(&info
->super_copy
->super_roots
,
1994 &info
->super_for_commit
->super_roots
,
1995 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1999 * this copies info out of the root backup array and back into
2000 * the in-memory super block. It is meant to help iterate through
2001 * the array, so you send it the number of backups you've already
2002 * tried and the last backup index you used.
2004 * this returns -1 when it has tried all the backups
2006 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2007 struct btrfs_super_block
*super
,
2008 int *num_backups_tried
, int *backup_index
)
2010 struct btrfs_root_backup
*root_backup
;
2011 int newest
= *backup_index
;
2013 if (*num_backups_tried
== 0) {
2014 u64 gen
= btrfs_super_generation(super
);
2016 newest
= find_newest_super_backup(info
, gen
);
2020 *backup_index
= newest
;
2021 *num_backups_tried
= 1;
2022 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2023 /* we've tried all the backups, all done */
2026 /* jump to the next oldest backup */
2027 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2028 BTRFS_NUM_BACKUP_ROOTS
;
2029 *backup_index
= newest
;
2030 *num_backups_tried
+= 1;
2032 root_backup
= super
->super_roots
+ newest
;
2034 btrfs_set_super_generation(super
,
2035 btrfs_backup_tree_root_gen(root_backup
));
2036 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2037 btrfs_set_super_root_level(super
,
2038 btrfs_backup_tree_root_level(root_backup
));
2039 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2042 * fixme: the total bytes and num_devices need to match or we should
2045 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2046 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2050 /* helper to cleanup workers */
2051 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2053 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2054 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2055 btrfs_destroy_workqueue(fs_info
->workers
);
2056 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2057 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2058 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2059 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2060 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2061 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2062 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2063 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2064 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2065 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2066 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2067 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2068 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2069 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2070 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2073 static void free_root_extent_buffers(struct btrfs_root
*root
)
2076 free_extent_buffer(root
->node
);
2077 free_extent_buffer(root
->commit_root
);
2079 root
->commit_root
= NULL
;
2083 /* helper to cleanup tree roots */
2084 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2086 free_root_extent_buffers(info
->tree_root
);
2088 free_root_extent_buffers(info
->dev_root
);
2089 free_root_extent_buffers(info
->extent_root
);
2090 free_root_extent_buffers(info
->csum_root
);
2091 free_root_extent_buffers(info
->quota_root
);
2092 free_root_extent_buffers(info
->uuid_root
);
2094 free_root_extent_buffers(info
->chunk_root
);
2097 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2100 struct btrfs_root
*gang
[8];
2103 while (!list_empty(&fs_info
->dead_roots
)) {
2104 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2105 struct btrfs_root
, root_list
);
2106 list_del(&gang
[0]->root_list
);
2108 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2109 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2111 free_extent_buffer(gang
[0]->node
);
2112 free_extent_buffer(gang
[0]->commit_root
);
2113 btrfs_put_fs_root(gang
[0]);
2118 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2123 for (i
= 0; i
< ret
; i
++)
2124 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2127 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2128 btrfs_free_log_root_tree(NULL
, fs_info
);
2129 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2130 fs_info
->pinned_extents
);
2134 int open_ctree(struct super_block
*sb
,
2135 struct btrfs_fs_devices
*fs_devices
,
2143 struct btrfs_key location
;
2144 struct buffer_head
*bh
;
2145 struct btrfs_super_block
*disk_super
;
2146 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2147 struct btrfs_root
*tree_root
;
2148 struct btrfs_root
*extent_root
;
2149 struct btrfs_root
*csum_root
;
2150 struct btrfs_root
*chunk_root
;
2151 struct btrfs_root
*dev_root
;
2152 struct btrfs_root
*quota_root
;
2153 struct btrfs_root
*uuid_root
;
2154 struct btrfs_root
*log_tree_root
;
2157 int num_backups_tried
= 0;
2158 int backup_index
= 0;
2160 int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2161 bool create_uuid_tree
;
2162 bool check_uuid_tree
;
2164 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2165 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2166 if (!tree_root
|| !chunk_root
) {
2171 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2177 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2183 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2188 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2189 (1 + ilog2(nr_cpu_ids
));
2191 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2194 goto fail_dirty_metadata_bytes
;
2197 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0);
2200 goto fail_delalloc_bytes
;
2203 fs_info
->btree_inode
= new_inode(sb
);
2204 if (!fs_info
->btree_inode
) {
2206 goto fail_bio_counter
;
2209 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2211 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2212 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2213 INIT_LIST_HEAD(&fs_info
->trans_list
);
2214 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2215 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2216 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2217 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2218 spin_lock_init(&fs_info
->delalloc_root_lock
);
2219 spin_lock_init(&fs_info
->trans_lock
);
2220 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2221 spin_lock_init(&fs_info
->delayed_iput_lock
);
2222 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2223 spin_lock_init(&fs_info
->free_chunk_lock
);
2224 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2225 spin_lock_init(&fs_info
->super_lock
);
2226 spin_lock_init(&fs_info
->qgroup_op_lock
);
2227 spin_lock_init(&fs_info
->buffer_lock
);
2228 spin_lock_init(&fs_info
->unused_bgs_lock
);
2229 rwlock_init(&fs_info
->tree_mod_log_lock
);
2230 mutex_init(&fs_info
->reloc_mutex
);
2231 mutex_init(&fs_info
->delalloc_root_mutex
);
2232 seqlock_init(&fs_info
->profiles_lock
);
2234 init_completion(&fs_info
->kobj_unregister
);
2235 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2236 INIT_LIST_HEAD(&fs_info
->space_info
);
2237 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2238 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2239 btrfs_mapping_init(&fs_info
->mapping_tree
);
2240 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2241 BTRFS_BLOCK_RSV_GLOBAL
);
2242 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2243 BTRFS_BLOCK_RSV_DELALLOC
);
2244 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2245 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2246 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2247 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2248 BTRFS_BLOCK_RSV_DELOPS
);
2249 atomic_set(&fs_info
->nr_async_submits
, 0);
2250 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2251 atomic_set(&fs_info
->async_submit_draining
, 0);
2252 atomic_set(&fs_info
->nr_async_bios
, 0);
2253 atomic_set(&fs_info
->defrag_running
, 0);
2254 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2255 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2257 fs_info
->max_inline
= 8192 * 1024;
2258 fs_info
->metadata_ratio
= 0;
2259 fs_info
->defrag_inodes
= RB_ROOT
;
2260 fs_info
->free_chunk_space
= 0;
2261 fs_info
->tree_mod_log
= RB_ROOT
;
2262 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2263 fs_info
->avg_delayed_ref_runtime
= div64_u64(NSEC_PER_SEC
, 64);
2264 /* readahead state */
2265 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2266 spin_lock_init(&fs_info
->reada_lock
);
2268 fs_info
->thread_pool_size
= min_t(unsigned long,
2269 num_online_cpus() + 2, 8);
2271 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2272 spin_lock_init(&fs_info
->ordered_root_lock
);
2273 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2275 if (!fs_info
->delayed_root
) {
2279 btrfs_init_delayed_root(fs_info
->delayed_root
);
2281 mutex_init(&fs_info
->scrub_lock
);
2282 atomic_set(&fs_info
->scrubs_running
, 0);
2283 atomic_set(&fs_info
->scrub_pause_req
, 0);
2284 atomic_set(&fs_info
->scrubs_paused
, 0);
2285 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2286 init_waitqueue_head(&fs_info
->replace_wait
);
2287 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2288 fs_info
->scrub_workers_refcnt
= 0;
2289 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2290 fs_info
->check_integrity_print_mask
= 0;
2293 spin_lock_init(&fs_info
->balance_lock
);
2294 mutex_init(&fs_info
->balance_mutex
);
2295 atomic_set(&fs_info
->balance_running
, 0);
2296 atomic_set(&fs_info
->balance_pause_req
, 0);
2297 atomic_set(&fs_info
->balance_cancel_req
, 0);
2298 fs_info
->balance_ctl
= NULL
;
2299 init_waitqueue_head(&fs_info
->balance_wait_q
);
2300 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2302 sb
->s_blocksize
= 4096;
2303 sb
->s_blocksize_bits
= blksize_bits(4096);
2304 sb
->s_bdi
= &fs_info
->bdi
;
2306 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2307 set_nlink(fs_info
->btree_inode
, 1);
2309 * we set the i_size on the btree inode to the max possible int.
2310 * the real end of the address space is determined by all of
2311 * the devices in the system
2313 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2314 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2315 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2317 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2318 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2319 fs_info
->btree_inode
->i_mapping
);
2320 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2321 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2323 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2325 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2326 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2327 sizeof(struct btrfs_key
));
2328 set_bit(BTRFS_INODE_DUMMY
,
2329 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2330 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2332 spin_lock_init(&fs_info
->block_group_cache_lock
);
2333 fs_info
->block_group_cache_tree
= RB_ROOT
;
2334 fs_info
->first_logical_byte
= (u64
)-1;
2336 extent_io_tree_init(&fs_info
->freed_extents
[0],
2337 fs_info
->btree_inode
->i_mapping
);
2338 extent_io_tree_init(&fs_info
->freed_extents
[1],
2339 fs_info
->btree_inode
->i_mapping
);
2340 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2341 fs_info
->do_barriers
= 1;
2344 mutex_init(&fs_info
->ordered_operations_mutex
);
2345 mutex_init(&fs_info
->ordered_extent_flush_mutex
);
2346 mutex_init(&fs_info
->tree_log_mutex
);
2347 mutex_init(&fs_info
->chunk_mutex
);
2348 mutex_init(&fs_info
->transaction_kthread_mutex
);
2349 mutex_init(&fs_info
->cleaner_mutex
);
2350 mutex_init(&fs_info
->volume_mutex
);
2351 init_rwsem(&fs_info
->commit_root_sem
);
2352 init_rwsem(&fs_info
->cleanup_work_sem
);
2353 init_rwsem(&fs_info
->subvol_sem
);
2354 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2355 fs_info
->dev_replace
.lock_owner
= 0;
2356 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2357 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2358 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2359 mutex_init(&fs_info
->dev_replace
.lock
);
2361 spin_lock_init(&fs_info
->qgroup_lock
);
2362 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2363 fs_info
->qgroup_tree
= RB_ROOT
;
2364 fs_info
->qgroup_op_tree
= RB_ROOT
;
2365 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2366 fs_info
->qgroup_seq
= 1;
2367 fs_info
->quota_enabled
= 0;
2368 fs_info
->pending_quota_state
= 0;
2369 fs_info
->qgroup_ulist
= NULL
;
2370 mutex_init(&fs_info
->qgroup_rescan_lock
);
2372 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2373 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2375 init_waitqueue_head(&fs_info
->transaction_throttle
);
2376 init_waitqueue_head(&fs_info
->transaction_wait
);
2377 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2378 init_waitqueue_head(&fs_info
->async_submit_wait
);
2380 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2386 __setup_root(4096, 4096, 4096, tree_root
,
2387 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2389 invalidate_bdev(fs_devices
->latest_bdev
);
2392 * Read super block and check the signature bytes only
2394 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2401 * We want to check superblock checksum, the type is stored inside.
2402 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2404 if (btrfs_check_super_csum(bh
->b_data
)) {
2405 printk(KERN_ERR
"BTRFS: superblock checksum mismatch\n");
2411 * super_copy is zeroed at allocation time and we never touch the
2412 * following bytes up to INFO_SIZE, the checksum is calculated from
2413 * the whole block of INFO_SIZE
2415 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2416 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2417 sizeof(*fs_info
->super_for_commit
));
2420 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2422 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2424 printk(KERN_ERR
"BTRFS: superblock contains fatal errors\n");
2429 disk_super
= fs_info
->super_copy
;
2430 if (!btrfs_super_root(disk_super
))
2433 /* check FS state, whether FS is broken. */
2434 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2435 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2438 * run through our array of backup supers and setup
2439 * our ring pointer to the oldest one
2441 generation
= btrfs_super_generation(disk_super
);
2442 find_oldest_super_backup(fs_info
, generation
);
2445 * In the long term, we'll store the compression type in the super
2446 * block, and it'll be used for per file compression control.
2448 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2450 ret
= btrfs_parse_options(tree_root
, options
);
2456 features
= btrfs_super_incompat_flags(disk_super
) &
2457 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2459 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2460 "unsupported optional features (%Lx).\n",
2467 * Leafsize and nodesize were always equal, this is only a sanity check.
2469 if (le32_to_cpu(disk_super
->__unused_leafsize
) !=
2470 btrfs_super_nodesize(disk_super
)) {
2471 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2472 "blocksizes don't match. node %d leaf %d\n",
2473 btrfs_super_nodesize(disk_super
),
2474 le32_to_cpu(disk_super
->__unused_leafsize
));
2478 if (btrfs_super_nodesize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2479 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2480 "blocksize (%d) was too large\n",
2481 btrfs_super_nodesize(disk_super
));
2486 features
= btrfs_super_incompat_flags(disk_super
);
2487 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2488 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2489 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2491 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2492 printk(KERN_ERR
"BTRFS: has skinny extents\n");
2495 * flag our filesystem as having big metadata blocks if
2496 * they are bigger than the page size
2498 if (btrfs_super_nodesize(disk_super
) > PAGE_CACHE_SIZE
) {
2499 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2500 printk(KERN_INFO
"BTRFS: flagging fs with big metadata feature\n");
2501 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2504 nodesize
= btrfs_super_nodesize(disk_super
);
2505 sectorsize
= btrfs_super_sectorsize(disk_super
);
2506 stripesize
= btrfs_super_stripesize(disk_super
);
2507 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2508 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2511 * mixed block groups end up with duplicate but slightly offset
2512 * extent buffers for the same range. It leads to corruptions
2514 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2515 (sectorsize
!= nodesize
)) {
2516 printk(KERN_WARNING
"BTRFS: unequal leaf/node/sector sizes "
2517 "are not allowed for mixed block groups on %s\n",
2523 * Needn't use the lock because there is no other task which will
2526 btrfs_set_super_incompat_flags(disk_super
, features
);
2528 features
= btrfs_super_compat_ro_flags(disk_super
) &
2529 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2530 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2531 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2532 "unsupported option features (%Lx).\n",
2538 max_active
= fs_info
->thread_pool_size
;
2541 btrfs_alloc_workqueue("worker", flags
| WQ_HIGHPRI
,
2544 fs_info
->delalloc_workers
=
2545 btrfs_alloc_workqueue("delalloc", flags
, max_active
, 2);
2547 fs_info
->flush_workers
=
2548 btrfs_alloc_workqueue("flush_delalloc", flags
, max_active
, 0);
2550 fs_info
->caching_workers
=
2551 btrfs_alloc_workqueue("cache", flags
, max_active
, 0);
2554 * a higher idle thresh on the submit workers makes it much more
2555 * likely that bios will be send down in a sane order to the
2558 fs_info
->submit_workers
=
2559 btrfs_alloc_workqueue("submit", flags
,
2560 min_t(u64
, fs_devices
->num_devices
,
2563 fs_info
->fixup_workers
=
2564 btrfs_alloc_workqueue("fixup", flags
, 1, 0);
2567 * endios are largely parallel and should have a very
2570 fs_info
->endio_workers
=
2571 btrfs_alloc_workqueue("endio", flags
, max_active
, 4);
2572 fs_info
->endio_meta_workers
=
2573 btrfs_alloc_workqueue("endio-meta", flags
, max_active
, 4);
2574 fs_info
->endio_meta_write_workers
=
2575 btrfs_alloc_workqueue("endio-meta-write", flags
, max_active
, 2);
2576 fs_info
->endio_raid56_workers
=
2577 btrfs_alloc_workqueue("endio-raid56", flags
, max_active
, 4);
2578 fs_info
->endio_repair_workers
=
2579 btrfs_alloc_workqueue("endio-repair", flags
, 1, 0);
2580 fs_info
->rmw_workers
=
2581 btrfs_alloc_workqueue("rmw", flags
, max_active
, 2);
2582 fs_info
->endio_write_workers
=
2583 btrfs_alloc_workqueue("endio-write", flags
, max_active
, 2);
2584 fs_info
->endio_freespace_worker
=
2585 btrfs_alloc_workqueue("freespace-write", flags
, max_active
, 0);
2586 fs_info
->delayed_workers
=
2587 btrfs_alloc_workqueue("delayed-meta", flags
, max_active
, 0);
2588 fs_info
->readahead_workers
=
2589 btrfs_alloc_workqueue("readahead", flags
, max_active
, 2);
2590 fs_info
->qgroup_rescan_workers
=
2591 btrfs_alloc_workqueue("qgroup-rescan", flags
, 1, 0);
2592 fs_info
->extent_workers
=
2593 btrfs_alloc_workqueue("extent-refs", flags
,
2594 min_t(u64
, fs_devices
->num_devices
,
2597 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2598 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2599 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2600 fs_info
->endio_meta_write_workers
&&
2601 fs_info
->endio_repair_workers
&&
2602 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2603 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2604 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2605 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2606 fs_info
->extent_workers
&&
2607 fs_info
->qgroup_rescan_workers
)) {
2609 goto fail_sb_buffer
;
2612 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2613 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2614 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2616 tree_root
->nodesize
= nodesize
;
2617 tree_root
->sectorsize
= sectorsize
;
2618 tree_root
->stripesize
= stripesize
;
2620 sb
->s_blocksize
= sectorsize
;
2621 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2623 if (btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
2624 printk(KERN_INFO
"BTRFS: valid FS not found on %s\n", sb
->s_id
);
2625 goto fail_sb_buffer
;
2628 if (sectorsize
!= PAGE_SIZE
) {
2629 printk(KERN_WARNING
"BTRFS: Incompatible sector size(%lu) "
2630 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2631 goto fail_sb_buffer
;
2634 mutex_lock(&fs_info
->chunk_mutex
);
2635 ret
= btrfs_read_sys_array(tree_root
);
2636 mutex_unlock(&fs_info
->chunk_mutex
);
2638 printk(KERN_WARNING
"BTRFS: failed to read the system "
2639 "array on %s\n", sb
->s_id
);
2640 goto fail_sb_buffer
;
2643 generation
= btrfs_super_chunk_root_generation(disk_super
);
2645 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2646 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2648 chunk_root
->node
= read_tree_block(chunk_root
,
2649 btrfs_super_chunk_root(disk_super
),
2651 if (!chunk_root
->node
||
2652 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2653 printk(KERN_WARNING
"BTRFS: failed to read chunk root on %s\n",
2655 goto fail_tree_roots
;
2657 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2658 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2660 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2661 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2663 ret
= btrfs_read_chunk_tree(chunk_root
);
2665 printk(KERN_WARNING
"BTRFS: failed to read chunk tree on %s\n",
2667 goto fail_tree_roots
;
2671 * keep the device that is marked to be the target device for the
2672 * dev_replace procedure
2674 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2676 if (!fs_devices
->latest_bdev
) {
2677 printk(KERN_CRIT
"BTRFS: failed to read devices on %s\n",
2679 goto fail_tree_roots
;
2683 generation
= btrfs_super_generation(disk_super
);
2685 tree_root
->node
= read_tree_block(tree_root
,
2686 btrfs_super_root(disk_super
),
2688 if (!tree_root
->node
||
2689 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2690 printk(KERN_WARNING
"BTRFS: failed to read tree root on %s\n",
2693 goto recovery_tree_root
;
2696 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2697 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2698 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2700 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2701 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2702 location
.offset
= 0;
2704 extent_root
= btrfs_read_tree_root(tree_root
, &location
);
2705 if (IS_ERR(extent_root
)) {
2706 ret
= PTR_ERR(extent_root
);
2707 goto recovery_tree_root
;
2709 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &extent_root
->state
);
2710 fs_info
->extent_root
= extent_root
;
2712 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2713 dev_root
= btrfs_read_tree_root(tree_root
, &location
);
2714 if (IS_ERR(dev_root
)) {
2715 ret
= PTR_ERR(dev_root
);
2716 goto recovery_tree_root
;
2718 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &dev_root
->state
);
2719 fs_info
->dev_root
= dev_root
;
2720 btrfs_init_devices_late(fs_info
);
2722 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2723 csum_root
= btrfs_read_tree_root(tree_root
, &location
);
2724 if (IS_ERR(csum_root
)) {
2725 ret
= PTR_ERR(csum_root
);
2726 goto recovery_tree_root
;
2728 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &csum_root
->state
);
2729 fs_info
->csum_root
= csum_root
;
2731 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2732 quota_root
= btrfs_read_tree_root(tree_root
, &location
);
2733 if (!IS_ERR(quota_root
)) {
2734 set_bit(BTRFS_ROOT_TRACK_DIRTY
, "a_root
->state
);
2735 fs_info
->quota_enabled
= 1;
2736 fs_info
->pending_quota_state
= 1;
2737 fs_info
->quota_root
= quota_root
;
2740 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2741 uuid_root
= btrfs_read_tree_root(tree_root
, &location
);
2742 if (IS_ERR(uuid_root
)) {
2743 ret
= PTR_ERR(uuid_root
);
2745 goto recovery_tree_root
;
2746 create_uuid_tree
= true;
2747 check_uuid_tree
= false;
2749 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &uuid_root
->state
);
2750 fs_info
->uuid_root
= uuid_root
;
2751 create_uuid_tree
= false;
2753 generation
!= btrfs_super_uuid_tree_generation(disk_super
);
2756 fs_info
->generation
= generation
;
2757 fs_info
->last_trans_committed
= generation
;
2759 ret
= btrfs_recover_balance(fs_info
);
2761 printk(KERN_WARNING
"BTRFS: failed to recover balance\n");
2762 goto fail_block_groups
;
2765 ret
= btrfs_init_dev_stats(fs_info
);
2767 printk(KERN_ERR
"BTRFS: failed to init dev_stats: %d\n",
2769 goto fail_block_groups
;
2772 ret
= btrfs_init_dev_replace(fs_info
);
2774 pr_err("BTRFS: failed to init dev_replace: %d\n", ret
);
2775 goto fail_block_groups
;
2778 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2780 ret
= btrfs_sysfs_add_one(fs_info
);
2782 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret
);
2783 goto fail_block_groups
;
2786 ret
= btrfs_init_space_info(fs_info
);
2788 printk(KERN_ERR
"BTRFS: Failed to initial space info: %d\n", ret
);
2792 ret
= btrfs_read_block_groups(extent_root
);
2794 printk(KERN_ERR
"BTRFS: Failed to read block groups: %d\n", ret
);
2797 fs_info
->num_tolerated_disk_barrier_failures
=
2798 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2799 if (fs_info
->fs_devices
->missing_devices
>
2800 fs_info
->num_tolerated_disk_barrier_failures
&&
2801 !(sb
->s_flags
& MS_RDONLY
)) {
2802 printk(KERN_WARNING
"BTRFS: "
2803 "too many missing devices, writeable mount is not allowed\n");
2807 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2809 if (IS_ERR(fs_info
->cleaner_kthread
))
2812 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2814 "btrfs-transaction");
2815 if (IS_ERR(fs_info
->transaction_kthread
))
2818 if (!btrfs_test_opt(tree_root
, SSD
) &&
2819 !btrfs_test_opt(tree_root
, NOSSD
) &&
2820 !fs_info
->fs_devices
->rotating
) {
2821 printk(KERN_INFO
"BTRFS: detected SSD devices, enabling SSD "
2823 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2826 /* Set the real inode map cache flag */
2827 if (btrfs_test_opt(tree_root
, CHANGE_INODE_CACHE
))
2828 btrfs_set_opt(tree_root
->fs_info
->mount_opt
, INODE_MAP_CACHE
);
2830 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2831 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2832 ret
= btrfsic_mount(tree_root
, fs_devices
,
2833 btrfs_test_opt(tree_root
,
2834 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2836 fs_info
->check_integrity_print_mask
);
2838 printk(KERN_WARNING
"BTRFS: failed to initialize"
2839 " integrity check module %s\n", sb
->s_id
);
2842 ret
= btrfs_read_qgroup_config(fs_info
);
2844 goto fail_trans_kthread
;
2846 /* do not make disk changes in broken FS */
2847 if (btrfs_super_log_root(disk_super
) != 0) {
2848 u64 bytenr
= btrfs_super_log_root(disk_super
);
2850 if (fs_devices
->rw_devices
== 0) {
2851 printk(KERN_WARNING
"BTRFS: log replay required "
2857 log_tree_root
= btrfs_alloc_root(fs_info
);
2858 if (!log_tree_root
) {
2863 __setup_root(nodesize
, sectorsize
, stripesize
,
2864 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2866 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2868 if (!log_tree_root
->node
||
2869 !extent_buffer_uptodate(log_tree_root
->node
)) {
2870 printk(KERN_ERR
"BTRFS: failed to read log tree\n");
2871 free_extent_buffer(log_tree_root
->node
);
2872 kfree(log_tree_root
);
2875 /* returns with log_tree_root freed on success */
2876 ret
= btrfs_recover_log_trees(log_tree_root
);
2878 btrfs_error(tree_root
->fs_info
, ret
,
2879 "Failed to recover log tree");
2880 free_extent_buffer(log_tree_root
->node
);
2881 kfree(log_tree_root
);
2885 if (sb
->s_flags
& MS_RDONLY
) {
2886 ret
= btrfs_commit_super(tree_root
);
2892 ret
= btrfs_find_orphan_roots(tree_root
);
2896 if (!(sb
->s_flags
& MS_RDONLY
)) {
2897 ret
= btrfs_cleanup_fs_roots(fs_info
);
2901 mutex_lock(&fs_info
->cleaner_mutex
);
2902 ret
= btrfs_recover_relocation(tree_root
);
2903 mutex_unlock(&fs_info
->cleaner_mutex
);
2906 "BTRFS: failed to recover relocation\n");
2912 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2913 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2914 location
.offset
= 0;
2916 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2917 if (IS_ERR(fs_info
->fs_root
)) {
2918 err
= PTR_ERR(fs_info
->fs_root
);
2922 if (sb
->s_flags
& MS_RDONLY
)
2925 down_read(&fs_info
->cleanup_work_sem
);
2926 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2927 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2928 up_read(&fs_info
->cleanup_work_sem
);
2929 close_ctree(tree_root
);
2932 up_read(&fs_info
->cleanup_work_sem
);
2934 ret
= btrfs_resume_balance_async(fs_info
);
2936 printk(KERN_WARNING
"BTRFS: failed to resume balance\n");
2937 close_ctree(tree_root
);
2941 ret
= btrfs_resume_dev_replace_async(fs_info
);
2943 pr_warn("BTRFS: failed to resume dev_replace\n");
2944 close_ctree(tree_root
);
2948 btrfs_qgroup_rescan_resume(fs_info
);
2950 if (create_uuid_tree
) {
2951 pr_info("BTRFS: creating UUID tree\n");
2952 ret
= btrfs_create_uuid_tree(fs_info
);
2954 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2956 close_ctree(tree_root
);
2959 } else if (check_uuid_tree
||
2960 btrfs_test_opt(tree_root
, RESCAN_UUID_TREE
)) {
2961 pr_info("BTRFS: checking UUID tree\n");
2962 ret
= btrfs_check_uuid_tree(fs_info
);
2964 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2966 close_ctree(tree_root
);
2970 fs_info
->update_uuid_tree_gen
= 1;
2978 btrfs_free_qgroup_config(fs_info
);
2980 kthread_stop(fs_info
->transaction_kthread
);
2981 btrfs_cleanup_transaction(fs_info
->tree_root
);
2982 btrfs_free_fs_roots(fs_info
);
2984 kthread_stop(fs_info
->cleaner_kthread
);
2987 * make sure we're done with the btree inode before we stop our
2990 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2993 btrfs_sysfs_remove_one(fs_info
);
2996 btrfs_put_block_group_cache(fs_info
);
2997 btrfs_free_block_groups(fs_info
);
3000 free_root_pointers(fs_info
, 1);
3001 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3004 btrfs_stop_all_workers(fs_info
);
3007 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3009 iput(fs_info
->btree_inode
);
3011 percpu_counter_destroy(&fs_info
->bio_counter
);
3012 fail_delalloc_bytes
:
3013 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3014 fail_dirty_metadata_bytes
:
3015 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3017 bdi_destroy(&fs_info
->bdi
);
3019 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3021 btrfs_free_stripe_hash_table(fs_info
);
3022 btrfs_close_devices(fs_info
->fs_devices
);
3026 if (!btrfs_test_opt(tree_root
, RECOVERY
))
3027 goto fail_tree_roots
;
3029 free_root_pointers(fs_info
, 0);
3031 /* don't use the log in recovery mode, it won't be valid */
3032 btrfs_set_super_log_root(disk_super
, 0);
3034 /* we can't trust the free space cache either */
3035 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3037 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3038 &num_backups_tried
, &backup_index
);
3040 goto fail_block_groups
;
3041 goto retry_root_backup
;
3044 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3047 set_buffer_uptodate(bh
);
3049 struct btrfs_device
*device
= (struct btrfs_device
*)
3052 printk_ratelimited_in_rcu(KERN_WARNING
"BTRFS: lost page write due to "
3053 "I/O error on %s\n",
3054 rcu_str_deref(device
->name
));
3055 /* note, we dont' set_buffer_write_io_error because we have
3056 * our own ways of dealing with the IO errors
3058 clear_buffer_uptodate(bh
);
3059 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3065 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3067 struct buffer_head
*bh
;
3068 struct buffer_head
*latest
= NULL
;
3069 struct btrfs_super_block
*super
;
3074 /* we would like to check all the supers, but that would make
3075 * a btrfs mount succeed after a mkfs from a different FS.
3076 * So, we need to add a special mount option to scan for
3077 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3079 for (i
= 0; i
< 1; i
++) {
3080 bytenr
= btrfs_sb_offset(i
);
3081 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3082 i_size_read(bdev
->bd_inode
))
3084 bh
= __bread(bdev
, bytenr
/ 4096,
3085 BTRFS_SUPER_INFO_SIZE
);
3089 super
= (struct btrfs_super_block
*)bh
->b_data
;
3090 if (btrfs_super_bytenr(super
) != bytenr
||
3091 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3096 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3099 transid
= btrfs_super_generation(super
);
3108 * this should be called twice, once with wait == 0 and
3109 * once with wait == 1. When wait == 0 is done, all the buffer heads
3110 * we write are pinned.
3112 * They are released when wait == 1 is done.
3113 * max_mirrors must be the same for both runs, and it indicates how
3114 * many supers on this one device should be written.
3116 * max_mirrors == 0 means to write them all.
3118 static int write_dev_supers(struct btrfs_device
*device
,
3119 struct btrfs_super_block
*sb
,
3120 int do_barriers
, int wait
, int max_mirrors
)
3122 struct buffer_head
*bh
;
3129 if (max_mirrors
== 0)
3130 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3132 for (i
= 0; i
< max_mirrors
; i
++) {
3133 bytenr
= btrfs_sb_offset(i
);
3134 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3135 device
->commit_total_bytes
)
3139 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3140 BTRFS_SUPER_INFO_SIZE
);
3146 if (!buffer_uptodate(bh
))
3149 /* drop our reference */
3152 /* drop the reference from the wait == 0 run */
3156 btrfs_set_super_bytenr(sb
, bytenr
);
3159 crc
= btrfs_csum_data((char *)sb
+
3160 BTRFS_CSUM_SIZE
, crc
,
3161 BTRFS_SUPER_INFO_SIZE
-
3163 btrfs_csum_final(crc
, sb
->csum
);
3166 * one reference for us, and we leave it for the
3169 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3170 BTRFS_SUPER_INFO_SIZE
);
3172 printk(KERN_ERR
"BTRFS: couldn't get super "
3173 "buffer head for bytenr %Lu\n", bytenr
);
3178 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3180 /* one reference for submit_bh */
3183 set_buffer_uptodate(bh
);
3185 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3186 bh
->b_private
= device
;
3190 * we fua the first super. The others we allow
3194 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3196 ret
= btrfsic_submit_bh(WRITE_SYNC
, bh
);
3200 return errors
< i
? 0 : -1;
3204 * endio for the write_dev_flush, this will wake anyone waiting
3205 * for the barrier when it is done
3207 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3210 if (err
== -EOPNOTSUPP
)
3211 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3212 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3214 if (bio
->bi_private
)
3215 complete(bio
->bi_private
);
3220 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3221 * sent down. With wait == 1, it waits for the previous flush.
3223 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3226 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3231 if (device
->nobarriers
)
3235 bio
= device
->flush_bio
;
3239 wait_for_completion(&device
->flush_wait
);
3241 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3242 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3243 rcu_str_deref(device
->name
));
3244 device
->nobarriers
= 1;
3245 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3247 btrfs_dev_stat_inc_and_print(device
,
3248 BTRFS_DEV_STAT_FLUSH_ERRS
);
3251 /* drop the reference from the wait == 0 run */
3253 device
->flush_bio
= NULL
;
3259 * one reference for us, and we leave it for the
3262 device
->flush_bio
= NULL
;
3263 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3267 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3268 bio
->bi_bdev
= device
->bdev
;
3269 init_completion(&device
->flush_wait
);
3270 bio
->bi_private
= &device
->flush_wait
;
3271 device
->flush_bio
= bio
;
3274 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3280 * send an empty flush down to each device in parallel,
3281 * then wait for them
3283 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3285 struct list_head
*head
;
3286 struct btrfs_device
*dev
;
3287 int errors_send
= 0;
3288 int errors_wait
= 0;
3291 /* send down all the barriers */
3292 head
= &info
->fs_devices
->devices
;
3293 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3300 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3303 ret
= write_dev_flush(dev
, 0);
3308 /* wait for all the barriers */
3309 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3316 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3319 ret
= write_dev_flush(dev
, 1);
3323 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3324 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3329 int btrfs_calc_num_tolerated_disk_barrier_failures(
3330 struct btrfs_fs_info
*fs_info
)
3332 struct btrfs_ioctl_space_info space
;
3333 struct btrfs_space_info
*sinfo
;
3334 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3335 BTRFS_BLOCK_GROUP_SYSTEM
,
3336 BTRFS_BLOCK_GROUP_METADATA
,
3337 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3341 int num_tolerated_disk_barrier_failures
=
3342 (int)fs_info
->fs_devices
->num_devices
;
3344 for (i
= 0; i
< num_types
; i
++) {
3345 struct btrfs_space_info
*tmp
;
3349 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3350 if (tmp
->flags
== types
[i
]) {
3360 down_read(&sinfo
->groups_sem
);
3361 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3362 if (!list_empty(&sinfo
->block_groups
[c
])) {
3365 btrfs_get_block_group_info(
3366 &sinfo
->block_groups
[c
], &space
);
3367 if (space
.total_bytes
== 0 ||
3368 space
.used_bytes
== 0)
3370 flags
= space
.flags
;
3373 * 0: if dup, single or RAID0 is configured for
3374 * any of metadata, system or data, else
3375 * 1: if RAID5 is configured, or if RAID1 or
3376 * RAID10 is configured and only two mirrors
3378 * 2: if RAID6 is configured, else
3379 * num_mirrors - 1: if RAID1 or RAID10 is
3380 * configured and more than
3381 * 2 mirrors are used.
3383 if (num_tolerated_disk_barrier_failures
> 0 &&
3384 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3385 BTRFS_BLOCK_GROUP_RAID0
)) ||
3386 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3388 num_tolerated_disk_barrier_failures
= 0;
3389 else if (num_tolerated_disk_barrier_failures
> 1) {
3390 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3391 BTRFS_BLOCK_GROUP_RAID5
|
3392 BTRFS_BLOCK_GROUP_RAID10
)) {
3393 num_tolerated_disk_barrier_failures
= 1;
3395 BTRFS_BLOCK_GROUP_RAID6
) {
3396 num_tolerated_disk_barrier_failures
= 2;
3401 up_read(&sinfo
->groups_sem
);
3404 return num_tolerated_disk_barrier_failures
;
3407 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3409 struct list_head
*head
;
3410 struct btrfs_device
*dev
;
3411 struct btrfs_super_block
*sb
;
3412 struct btrfs_dev_item
*dev_item
;
3416 int total_errors
= 0;
3419 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3420 backup_super_roots(root
->fs_info
);
3422 sb
= root
->fs_info
->super_for_commit
;
3423 dev_item
= &sb
->dev_item
;
3425 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3426 head
= &root
->fs_info
->fs_devices
->devices
;
3427 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3430 ret
= barrier_all_devices(root
->fs_info
);
3433 &root
->fs_info
->fs_devices
->device_list_mutex
);
3434 btrfs_error(root
->fs_info
, ret
,
3435 "errors while submitting device barriers.");
3440 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3445 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3448 btrfs_set_stack_device_generation(dev_item
, 0);
3449 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3450 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3451 btrfs_set_stack_device_total_bytes(dev_item
,
3452 dev
->commit_total_bytes
);
3453 btrfs_set_stack_device_bytes_used(dev_item
,
3454 dev
->commit_bytes_used
);
3455 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3456 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3457 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3458 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3459 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3461 flags
= btrfs_super_flags(sb
);
3462 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3464 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3468 if (total_errors
> max_errors
) {
3469 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3471 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3473 /* FUA is masked off if unsupported and can't be the reason */
3474 btrfs_error(root
->fs_info
, -EIO
,
3475 "%d errors while writing supers", total_errors
);
3480 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3483 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3486 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3490 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3491 if (total_errors
> max_errors
) {
3492 btrfs_error(root
->fs_info
, -EIO
,
3493 "%d errors while writing supers", total_errors
);
3499 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3500 struct btrfs_root
*root
, int max_mirrors
)
3502 return write_all_supers(root
, max_mirrors
);
3505 /* Drop a fs root from the radix tree and free it. */
3506 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3507 struct btrfs_root
*root
)
3509 spin_lock(&fs_info
->fs_roots_radix_lock
);
3510 radix_tree_delete(&fs_info
->fs_roots_radix
,
3511 (unsigned long)root
->root_key
.objectid
);
3512 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3514 if (btrfs_root_refs(&root
->root_item
) == 0)
3515 synchronize_srcu(&fs_info
->subvol_srcu
);
3517 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3518 btrfs_free_log(NULL
, root
);
3520 if (root
->free_ino_pinned
)
3521 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3522 if (root
->free_ino_ctl
)
3523 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3527 static void free_fs_root(struct btrfs_root
*root
)
3529 iput(root
->ino_cache_inode
);
3530 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3531 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3532 root
->orphan_block_rsv
= NULL
;
3534 free_anon_bdev(root
->anon_dev
);
3535 if (root
->subv_writers
)
3536 btrfs_free_subvolume_writers(root
->subv_writers
);
3537 free_extent_buffer(root
->node
);
3538 free_extent_buffer(root
->commit_root
);
3539 kfree(root
->free_ino_ctl
);
3540 kfree(root
->free_ino_pinned
);
3542 btrfs_put_fs_root(root
);
3545 void btrfs_free_fs_root(struct btrfs_root
*root
)
3550 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3552 u64 root_objectid
= 0;
3553 struct btrfs_root
*gang
[8];
3556 unsigned int ret
= 0;
3560 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3561 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3562 (void **)gang
, root_objectid
,
3565 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3568 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3570 for (i
= 0; i
< ret
; i
++) {
3571 /* Avoid to grab roots in dead_roots */
3572 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3576 /* grab all the search result for later use */
3577 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3579 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3581 for (i
= 0; i
< ret
; i
++) {
3584 root_objectid
= gang
[i
]->root_key
.objectid
;
3585 err
= btrfs_orphan_cleanup(gang
[i
]);
3588 btrfs_put_fs_root(gang
[i
]);
3593 /* release the uncleaned roots due to error */
3594 for (; i
< ret
; i
++) {
3596 btrfs_put_fs_root(gang
[i
]);
3601 int btrfs_commit_super(struct btrfs_root
*root
)
3603 struct btrfs_trans_handle
*trans
;
3605 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3606 btrfs_run_delayed_iputs(root
);
3607 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3608 wake_up_process(root
->fs_info
->cleaner_kthread
);
3610 /* wait until ongoing cleanup work done */
3611 down_write(&root
->fs_info
->cleanup_work_sem
);
3612 up_write(&root
->fs_info
->cleanup_work_sem
);
3614 trans
= btrfs_join_transaction(root
);
3616 return PTR_ERR(trans
);
3617 return btrfs_commit_transaction(trans
, root
);
3620 void close_ctree(struct btrfs_root
*root
)
3622 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3625 fs_info
->closing
= 1;
3628 /* wait for the uuid_scan task to finish */
3629 down(&fs_info
->uuid_tree_rescan_sem
);
3630 /* avoid complains from lockdep et al., set sem back to initial state */
3631 up(&fs_info
->uuid_tree_rescan_sem
);
3633 /* pause restriper - we want to resume on mount */
3634 btrfs_pause_balance(fs_info
);
3636 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3638 btrfs_scrub_cancel(fs_info
);
3640 /* wait for any defraggers to finish */
3641 wait_event(fs_info
->transaction_wait
,
3642 (atomic_read(&fs_info
->defrag_running
) == 0));
3644 /* clear out the rbtree of defraggable inodes */
3645 btrfs_cleanup_defrag_inodes(fs_info
);
3647 cancel_work_sync(&fs_info
->async_reclaim_work
);
3649 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3650 ret
= btrfs_commit_super(root
);
3652 btrfs_err(root
->fs_info
, "commit super ret %d", ret
);
3655 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3656 btrfs_error_commit_super(root
);
3658 kthread_stop(fs_info
->transaction_kthread
);
3659 kthread_stop(fs_info
->cleaner_kthread
);
3661 fs_info
->closing
= 2;
3664 btrfs_free_qgroup_config(root
->fs_info
);
3666 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3667 btrfs_info(root
->fs_info
, "at unmount delalloc count %lld",
3668 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3671 btrfs_sysfs_remove_one(fs_info
);
3673 btrfs_free_fs_roots(fs_info
);
3675 btrfs_put_block_group_cache(fs_info
);
3677 btrfs_free_block_groups(fs_info
);
3680 * we must make sure there is not any read request to
3681 * submit after we stopping all workers.
3683 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3684 btrfs_stop_all_workers(fs_info
);
3687 free_root_pointers(fs_info
, 1);
3689 iput(fs_info
->btree_inode
);
3691 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3692 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3693 btrfsic_unmount(root
, fs_info
->fs_devices
);
3696 btrfs_close_devices(fs_info
->fs_devices
);
3697 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3699 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3700 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3701 percpu_counter_destroy(&fs_info
->bio_counter
);
3702 bdi_destroy(&fs_info
->bdi
);
3703 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3705 btrfs_free_stripe_hash_table(fs_info
);
3707 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3708 root
->orphan_block_rsv
= NULL
;
3711 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3715 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3717 ret
= extent_buffer_uptodate(buf
);
3721 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3722 parent_transid
, atomic
);
3728 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3730 return set_extent_buffer_uptodate(buf
);
3733 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3735 struct btrfs_root
*root
;
3736 u64 transid
= btrfs_header_generation(buf
);
3739 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3741 * This is a fast path so only do this check if we have sanity tests
3742 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3743 * outside of the sanity tests.
3745 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
3748 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3749 btrfs_assert_tree_locked(buf
);
3750 if (transid
!= root
->fs_info
->generation
)
3751 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3752 "found %llu running %llu\n",
3753 buf
->start
, transid
, root
->fs_info
->generation
);
3754 was_dirty
= set_extent_buffer_dirty(buf
);
3756 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3758 root
->fs_info
->dirty_metadata_batch
);
3759 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3760 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
3761 btrfs_print_leaf(root
, buf
);
3767 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3771 * looks as though older kernels can get into trouble with
3772 * this code, they end up stuck in balance_dirty_pages forever
3776 if (current
->flags
& PF_MEMALLOC
)
3780 btrfs_balance_delayed_items(root
);
3782 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3783 BTRFS_DIRTY_METADATA_THRESH
);
3785 balance_dirty_pages_ratelimited(
3786 root
->fs_info
->btree_inode
->i_mapping
);
3791 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3793 __btrfs_btree_balance_dirty(root
, 1);
3796 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3798 __btrfs_btree_balance_dirty(root
, 0);
3801 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3803 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3804 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3807 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3811 * Placeholder for checks
3816 static void btrfs_error_commit_super(struct btrfs_root
*root
)
3818 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3819 btrfs_run_delayed_iputs(root
);
3820 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3822 down_write(&root
->fs_info
->cleanup_work_sem
);
3823 up_write(&root
->fs_info
->cleanup_work_sem
);
3825 /* cleanup FS via transaction */
3826 btrfs_cleanup_transaction(root
);
3829 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3831 struct btrfs_ordered_extent
*ordered
;
3833 spin_lock(&root
->ordered_extent_lock
);
3835 * This will just short circuit the ordered completion stuff which will
3836 * make sure the ordered extent gets properly cleaned up.
3838 list_for_each_entry(ordered
, &root
->ordered_extents
,
3840 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3841 spin_unlock(&root
->ordered_extent_lock
);
3844 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
3846 struct btrfs_root
*root
;
3847 struct list_head splice
;
3849 INIT_LIST_HEAD(&splice
);
3851 spin_lock(&fs_info
->ordered_root_lock
);
3852 list_splice_init(&fs_info
->ordered_roots
, &splice
);
3853 while (!list_empty(&splice
)) {
3854 root
= list_first_entry(&splice
, struct btrfs_root
,
3856 list_move_tail(&root
->ordered_root
,
3857 &fs_info
->ordered_roots
);
3859 spin_unlock(&fs_info
->ordered_root_lock
);
3860 btrfs_destroy_ordered_extents(root
);
3863 spin_lock(&fs_info
->ordered_root_lock
);
3865 spin_unlock(&fs_info
->ordered_root_lock
);
3868 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3869 struct btrfs_root
*root
)
3871 struct rb_node
*node
;
3872 struct btrfs_delayed_ref_root
*delayed_refs
;
3873 struct btrfs_delayed_ref_node
*ref
;
3876 delayed_refs
= &trans
->delayed_refs
;
3878 spin_lock(&delayed_refs
->lock
);
3879 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
3880 spin_unlock(&delayed_refs
->lock
);
3881 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
3885 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
3886 struct btrfs_delayed_ref_head
*head
;
3887 bool pin_bytes
= false;
3889 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
3891 if (!mutex_trylock(&head
->mutex
)) {
3892 atomic_inc(&head
->node
.refs
);
3893 spin_unlock(&delayed_refs
->lock
);
3895 mutex_lock(&head
->mutex
);
3896 mutex_unlock(&head
->mutex
);
3897 btrfs_put_delayed_ref(&head
->node
);
3898 spin_lock(&delayed_refs
->lock
);
3901 spin_lock(&head
->lock
);
3902 while ((node
= rb_first(&head
->ref_root
)) != NULL
) {
3903 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
3906 rb_erase(&ref
->rb_node
, &head
->ref_root
);
3907 atomic_dec(&delayed_refs
->num_entries
);
3908 btrfs_put_delayed_ref(ref
);
3910 if (head
->must_insert_reserved
)
3912 btrfs_free_delayed_extent_op(head
->extent_op
);
3913 delayed_refs
->num_heads
--;
3914 if (head
->processing
== 0)
3915 delayed_refs
->num_heads_ready
--;
3916 atomic_dec(&delayed_refs
->num_entries
);
3917 head
->node
.in_tree
= 0;
3918 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
3919 spin_unlock(&head
->lock
);
3920 spin_unlock(&delayed_refs
->lock
);
3921 mutex_unlock(&head
->mutex
);
3924 btrfs_pin_extent(root
, head
->node
.bytenr
,
3925 head
->node
.num_bytes
, 1);
3926 btrfs_put_delayed_ref(&head
->node
);
3928 spin_lock(&delayed_refs
->lock
);
3931 spin_unlock(&delayed_refs
->lock
);
3936 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3938 struct btrfs_inode
*btrfs_inode
;
3939 struct list_head splice
;
3941 INIT_LIST_HEAD(&splice
);
3943 spin_lock(&root
->delalloc_lock
);
3944 list_splice_init(&root
->delalloc_inodes
, &splice
);
3946 while (!list_empty(&splice
)) {
3947 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
3950 list_del_init(&btrfs_inode
->delalloc_inodes
);
3951 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3952 &btrfs_inode
->runtime_flags
);
3953 spin_unlock(&root
->delalloc_lock
);
3955 btrfs_invalidate_inodes(btrfs_inode
->root
);
3957 spin_lock(&root
->delalloc_lock
);
3960 spin_unlock(&root
->delalloc_lock
);
3963 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
3965 struct btrfs_root
*root
;
3966 struct list_head splice
;
3968 INIT_LIST_HEAD(&splice
);
3970 spin_lock(&fs_info
->delalloc_root_lock
);
3971 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
3972 while (!list_empty(&splice
)) {
3973 root
= list_first_entry(&splice
, struct btrfs_root
,
3975 list_del_init(&root
->delalloc_root
);
3976 root
= btrfs_grab_fs_root(root
);
3978 spin_unlock(&fs_info
->delalloc_root_lock
);
3980 btrfs_destroy_delalloc_inodes(root
);
3981 btrfs_put_fs_root(root
);
3983 spin_lock(&fs_info
->delalloc_root_lock
);
3985 spin_unlock(&fs_info
->delalloc_root_lock
);
3988 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3989 struct extent_io_tree
*dirty_pages
,
3993 struct extent_buffer
*eb
;
3998 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4003 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
4004 while (start
<= end
) {
4005 eb
= btrfs_find_tree_block(root
, start
,
4007 start
+= root
->nodesize
;
4010 wait_on_extent_buffer_writeback(eb
);
4012 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4014 clear_extent_buffer_dirty(eb
);
4015 free_extent_buffer_stale(eb
);
4022 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4023 struct extent_io_tree
*pinned_extents
)
4025 struct extent_io_tree
*unpin
;
4031 unpin
= pinned_extents
;
4034 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4035 EXTENT_DIRTY
, NULL
);
4040 if (btrfs_test_opt(root
, DISCARD
))
4041 ret
= btrfs_error_discard_extent(root
, start
,
4045 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4046 btrfs_error_unpin_extent_range(root
, start
, end
);
4051 if (unpin
== &root
->fs_info
->freed_extents
[0])
4052 unpin
= &root
->fs_info
->freed_extents
[1];
4054 unpin
= &root
->fs_info
->freed_extents
[0];
4062 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4063 struct btrfs_root
*root
)
4065 btrfs_destroy_delayed_refs(cur_trans
, root
);
4067 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4068 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4070 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4071 wake_up(&root
->fs_info
->transaction_wait
);
4073 btrfs_destroy_delayed_inodes(root
);
4074 btrfs_assert_delayed_root_empty(root
);
4076 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4078 btrfs_destroy_pinned_extent(root
,
4079 root
->fs_info
->pinned_extents
);
4081 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4082 wake_up(&cur_trans
->commit_wait
);
4085 memset(cur_trans, 0, sizeof(*cur_trans));
4086 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4090 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4092 struct btrfs_transaction
*t
;
4094 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4096 spin_lock(&root
->fs_info
->trans_lock
);
4097 while (!list_empty(&root
->fs_info
->trans_list
)) {
4098 t
= list_first_entry(&root
->fs_info
->trans_list
,
4099 struct btrfs_transaction
, list
);
4100 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4101 atomic_inc(&t
->use_count
);
4102 spin_unlock(&root
->fs_info
->trans_lock
);
4103 btrfs_wait_for_commit(root
, t
->transid
);
4104 btrfs_put_transaction(t
);
4105 spin_lock(&root
->fs_info
->trans_lock
);
4108 if (t
== root
->fs_info
->running_transaction
) {
4109 t
->state
= TRANS_STATE_COMMIT_DOING
;
4110 spin_unlock(&root
->fs_info
->trans_lock
);
4112 * We wait for 0 num_writers since we don't hold a trans
4113 * handle open currently for this transaction.
4115 wait_event(t
->writer_wait
,
4116 atomic_read(&t
->num_writers
) == 0);
4118 spin_unlock(&root
->fs_info
->trans_lock
);
4120 btrfs_cleanup_one_transaction(t
, root
);
4122 spin_lock(&root
->fs_info
->trans_lock
);
4123 if (t
== root
->fs_info
->running_transaction
)
4124 root
->fs_info
->running_transaction
= NULL
;
4125 list_del_init(&t
->list
);
4126 spin_unlock(&root
->fs_info
->trans_lock
);
4128 btrfs_put_transaction(t
);
4129 trace_btrfs_transaction_commit(root
);
4130 spin_lock(&root
->fs_info
->trans_lock
);
4132 spin_unlock(&root
->fs_info
->trans_lock
);
4133 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4134 btrfs_destroy_delayed_inodes(root
);
4135 btrfs_assert_delayed_root_empty(root
);
4136 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4137 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4138 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4143 static struct extent_io_ops btree_extent_io_ops
= {
4144 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4145 .readpage_io_failed_hook
= btree_io_failed_hook
,
4146 .submit_bio_hook
= btree_submit_bio_hook
,
4147 /* note we're sharing with inode.c for the merge bio hook */
4148 .merge_bio_hook
= btrfs_merge_bio_hook
,